hammer2 - update documentation, begin working on callback I/O
[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 212 hammer2_inode_t *ip,
bca9f8e6 213 const hammer2_inode_data_t *ripdata,
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,
bca9f8e6 219 const hammer2_inode_data_t *ripdata,
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,
bca9f8e6 226 const hammer2_inode_data_t *ripdata,
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,
bca9f8e6 234 const hammer2_inode_data_t *ripdata,
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 *
bca9f8e6 329hammer2_pfsalloc(const hammer2_inode_data_t *ripdata, hammer2_tid_t alloc_tid)
50456506
MD
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;
bca9f8e6
MD
344 if (ripdata) {
345 pmp->inode_tid = ripdata->pfs_inum + 1;
346 pmp->pfs_clid = ripdata->pfs_clid;
50456506
MD
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;
bca9f8e6 392 const hammer2_inode_data_t *ripdata;
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 */
bca9f8e6 633 ripdata = &hammer2_chain_rdata(schain)->ipdata;
50456506 634 KKASSERT(schain->pmp == spmp);
bca9f8e6 635 spmp->pfs_clid = ripdata->pfs_clid;
50456506
MD
636
637 /*
7bed8d7e
MD
638 * NOTE: inode_get sucks up schain's lock.
639 */
278ab2b2 640 cluster = hammer2_cluster_from_chain(schain);
50456506
MD
641 spmp->iroot = hammer2_inode_get(spmp, NULL, cluster);
642 spmp->spmp_hmp = hmp;
643 hammer2_inode_ref(spmp->iroot);
644 hammer2_inode_unlock_ex(spmp->iroot, cluster);
a5913bdf 645 schain = NULL;
50456506 646 /* leave spmp->iroot with one ref */
10136ab6
MD
647
648 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
649 error = hammer2_recovery(hmp);
650 /* XXX do something with error */
651 }
063639ba
MD
652 ++hmp->pmp_count;
653
654 /*
655 * XXX RDONLY stuff is totally broken FIXME XXX
656 *
657 * Automatic LNK_CONN
658 * Automatic handling of received LNK_SPAN
659 * Automatic handling of received LNK_CIRC
660 * No automatic LNK_SPAN generation - we do this ourselves
661 * No automatic LNK_CIRC generation - we do this ourselves
662 */
663 kdmsg_iocom_init(&hmp->iocom, hmp,
664 KDMSG_IOCOMF_AUTOCONN |
da6f36f4 665 KDMSG_IOCOMF_AUTORXSPAN,
063639ba
MD
666 hmp->mchain, hammer2_rcvdmsg);
667
668 /*
669 * Ref the cluster management messaging descriptor. The mount
670 * program deals with the other end of the communications pipe.
671 */
672 fp = holdfp(curproc->p_fd, info.cluster_fd, -1);
673 if (fp) {
674 hammer2_cluster_reconnect(hmp, fp);
675 } else {
676 kprintf("hammer2_mount: bad cluster_fd!\n");
677 }
50456506
MD
678 } else {
679 spmp = hmp->spmp;
063639ba 680 ++hmp->pmp_count;
50e4f8f4 681 }
5c23d7f1 682
10252dc7 683 /*
7bed8d7e 684 * Lookup mount point under the media-localized super-root.
50456506
MD
685 *
686 * cluster->pmp will incorrectly point to spmp and must be fixed
687 * up later on.
10252dc7 688 */
50456506 689 cparent = hammer2_inode_lock_ex(spmp->iroot);
e4e20f48 690 lhc = hammer2_dirhash(label, strlen(label));
278ab2b2 691 cluster = hammer2_cluster_lookup(cparent, &key_next,
c667909f 692 lhc, lhc + HAMMER2_DIRHASH_LOMASK,
278ab2b2
MD
693 0, &ddflag);
694 while (cluster) {
695 if (hammer2_cluster_type(cluster) == HAMMER2_BREF_TYPE_INODE &&
696 strcmp(label,
bca9f8e6 697 hammer2_cluster_rdata(cluster)->ipdata.filename) == 0) {
7cfa8da5
MD
698 break;
699 }
278ab2b2 700 cluster = hammer2_cluster_next(cparent, cluster, &key_next,
1897c66e 701 key_next,
278ab2b2 702 lhc + HAMMER2_DIRHASH_LOMASK, 0);
7cfa8da5 703 }
50456506 704 hammer2_inode_unlock_ex(spmp->iroot, cparent);
7bed8d7e 705
278ab2b2 706 if (cluster == NULL) {
e4e20f48 707 kprintf("hammer2_mount: PFS label not found\n");
623d43d4
MD
708 hammer2_vfs_unmount_hmp1(mp, hmp);
709 hammer2_vfs_unmount_hmp2(mp, hmp);
58e43599 710 lockmgr(&hammer2_mntlk, LK_RELEASE);
9c2e0de0 711 hammer2_vfs_unmount(mp, MNT_FORCE);
7cfa8da5
MD
712 return EINVAL;
713 }
278ab2b2
MD
714
715 for (i = 0; i < cluster->nchains; ++i) {
716 rchain = cluster->array[i];
717 if (rchain->flags & HAMMER2_CHAIN_MOUNTED) {
718 kprintf("hammer2_mount: PFS label already mounted!\n");
719 hammer2_cluster_unlock(cluster);
720 hammer2_vfs_unmount_hmp1(mp, hmp);
721 hammer2_vfs_unmount_hmp2(mp, hmp);
58e43599 722 lockmgr(&hammer2_mntlk, LK_RELEASE);
278ab2b2
MD
723 hammer2_vfs_unmount(mp, MNT_FORCE);
724 return EBUSY;
725 }
da6f36f4 726 KKASSERT(rchain->pmp == NULL);
9b6b3df4 727#if 0
278ab2b2
MD
728 if (rchain->flags & HAMMER2_CHAIN_RECYCLE) {
729 kprintf("hammer2_mount: PFS label is recycling\n");
730 hammer2_cluster_unlock(cluster);
731 hammer2_vfs_unmount_hmp1(mp, hmp);
732 hammer2_vfs_unmount_hmp2(mp, hmp);
58e43599 733 lockmgr(&hammer2_mntlk, LK_RELEASE);
278ab2b2
MD
734 hammer2_vfs_unmount(mp, MNT_FORCE);
735 return EBUSY;
736 }
9b6b3df4 737#endif
278ab2b2
MD
738 }
739
58e43599
MD
740 /*
741 * Check to see if the cluster id is already mounted at the mount
742 * point. If it is, add us to the cluster.
743 */
bca9f8e6 744 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
50456506 745 hammer2_cluster_bref(cluster, &bref);
58e43599 746 TAILQ_FOREACH(pmp, &hammer2_pfslist, mntentry) {
50456506 747 if (pmp->spmp_hmp == NULL &&
bca9f8e6 748 bcmp(&pmp->pfs_clid, &ripdata->pfs_clid,
50456506 749 sizeof(pmp->pfs_clid)) == 0) {
58e43599 750 break;
50456506 751 }
58e43599 752 }
50456506 753
58e43599
MD
754 if (pmp) {
755 int i;
756 int j;
757
758 hammer2_inode_ref(pmp->iroot);
759 ccms_thread_lock(&pmp->iroot->topo_cst, CCMS_STATE_EXCLUSIVE);
760
761 if (pmp->iroot->cluster.nchains + cluster->nchains >
762 HAMMER2_MAXCLUSTER) {
763 kprintf("hammer2_mount: cluster full!\n");
764
765 ccms_thread_unlock(&pmp->iroot->topo_cst);
766 hammer2_inode_drop(pmp->iroot);
767
768 hammer2_cluster_unlock(cluster);
769 hammer2_vfs_unmount_hmp1(mp, hmp);
770 hammer2_vfs_unmount_hmp2(mp, hmp);
771 lockmgr(&hammer2_mntlk, LK_RELEASE);
772 hammer2_vfs_unmount(mp, MNT_FORCE);
773 return EBUSY;
774 }
775 kprintf("hammer2_vfs_mount: Adding pfs to existing cluster\n");
776 j = pmp->iroot->cluster.nchains;
777 for (i = 0; i < cluster->nchains; ++i) {
50456506
MD
778 rchain = cluster->array[i];
779 KKASSERT(rchain->pmp == NULL);
780 rchain->pmp = pmp;
58e43599
MD
781 hammer2_chain_ref(cluster->array[i]);
782 pmp->iroot->cluster.array[j] = cluster->array[i];
783 ++j;
784 }
785 pmp->iroot->cluster.nchains = j;
786 ccms_thread_unlock(&pmp->iroot->topo_cst);
787 hammer2_inode_drop(pmp->iroot);
788 hammer2_cluster_unlock(cluster);
789 lockmgr(&hammer2_mntlk, LK_RELEASE);
790
791 kprintf("ok\n");
b53e7c4f 792 hammer2_inode_install_hidden(pmp);
58e43599
MD
793
794 return ERANGE;
795 }
796
797 /*
798 * Block device opened successfully, finish initializing the
799 * mount structure.
800 *
801 * From this point on we have to call hammer2_unmount() on failure.
802 */
bca9f8e6 803 pmp = hammer2_pfsalloc(ripdata, bref.mirror_tid);
50456506
MD
804 kprintf("PMP mirror_tid is %016jx\n", bref.mirror_tid);
805 for (i = 0; i < cluster->nchains; ++i) {
806 rchain = cluster->array[i];
807 KKASSERT(rchain->pmp == NULL);
808 rchain->pmp = pmp;
809 atomic_set_int(&rchain->flags, HAMMER2_CHAIN_MOUNTED);
810 }
811 cluster->pmp = pmp;
58e43599 812
58e43599
MD
813 ccms_domain_init(&pmp->ccms_dom);
814 TAILQ_INSERT_TAIL(&hammer2_pfslist, pmp, mntentry);
815 lockmgr(&hammer2_mntlk, LK_RELEASE);
816
817 kprintf("hammer2_mount hmp=%p pmp=%p pmpcnt=%d\n",
818 hmp, pmp, hmp->pmp_count);
819
820 mp->mnt_flag = MNT_LOCAL;
821 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE; /* all entry pts are SMP */
822 mp->mnt_kern_flag |= MNTK_THR_SYNC; /* new vsyncscan semantics */
823
824 /*
825 * required mount structure initializations
826 */
827 mp->mnt_stat.f_iosize = HAMMER2_PBUFSIZE;
828 mp->mnt_stat.f_bsize = HAMMER2_PBUFSIZE;
829
830 mp->mnt_vstat.f_frsize = HAMMER2_PBUFSIZE;
831 mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
832
833 /*
834 * Optional fields
835 */
836 mp->mnt_iosize_max = MAXPHYS;
837 mp->mnt_data = (qaddr_t)pmp;
838 pmp->mp = mp;
839
623d43d4
MD
840 /*
841 * After this point hammer2_vfs_unmount() has visibility on hmp
842 * and manual hmp1/hmp2 calls are not needed on fatal errors.
843 */
278ab2b2 844 pmp->iroot = hammer2_inode_get(pmp, NULL, cluster);
476d2aad 845 hammer2_inode_ref(pmp->iroot); /* ref for pmp->iroot */
278ab2b2 846 hammer2_inode_unlock_ex(pmp->iroot, cluster);
609a8021 847
065f4046
MD
848 /*
849 * The logical file buffer bio write thread handles things
850 * like physical block assignment and compression.
50456506
MD
851 *
852 * (only applicable to pfs mounts, not applicable to spmp)
065f4046 853 */
065f4046
MD
854 pmp->wthread_destroy = 0;
855 lwkt_create(hammer2_write_thread, pmp,
856 &pmp->wthread_td, NULL, 0, -1, "hwrite-%s", label);
857
044541cd
MD
858 /*
859 * With the cluster operational install ihidden.
50456506 860 * (only applicable to pfs mounts, not applicable to spmp)
044541cd
MD
861 */
862 hammer2_inode_install_hidden(pmp);
863
bfc3a7b1
MD
864 /*
865 * Finish setup
866 */
f0206a67
VS
867 vfs_getnewfsid(mp);
868 vfs_add_vnodeops(mp, &hammer2_vnode_vops, &mp->mnt_vn_norm_ops);
869 vfs_add_vnodeops(mp, &hammer2_spec_vops, &mp->mnt_vn_spec_ops);
870 vfs_add_vnodeops(mp, &hammer2_fifo_vops, &mp->mnt_vn_fifo_ops);
703720e4 871
54f522df 872 copyinstr(info.volume, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
703720e4
MD
873 bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
874 bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
875 copyinstr(path, mp->mnt_stat.f_mntonname,
876 sizeof(mp->mnt_stat.f_mntonname) - 1,
877 &size);
878
bfc3a7b1
MD
879 /*
880 * Initial statfs to prime mnt_stat.
881 */
066e00cc 882 hammer2_vfs_statfs(mp, &mp->mnt_stat, cred);
355d67fc 883
47902fef 884 return 0;
703720e4
MD
885}
886
355d67fc
MD
887/*
888 * Handle bioq for strategy write
889 */
890static
891void
892hammer2_write_thread(void *arg)
893{
065f4046 894 hammer2_pfsmount_t *pmp;
355d67fc
MD
895 struct bio *bio;
896 struct buf *bp;
897 hammer2_trans_t trans;
898 struct vnode *vp;
355d67fc 899 hammer2_inode_t *ip;
278ab2b2 900 hammer2_cluster_t *cparent;
6a5f4fe6 901 hammer2_inode_data_t *wipdata;
355d67fc
MD
902 hammer2_key_t lbase;
903 int lblksize;
904 int pblksize;
905 int error;
906
065f4046 907 pmp = arg;
355d67fc 908
065f4046
MD
909 mtx_lock(&pmp->wthread_mtx);
910 while (pmp->wthread_destroy == 0) {
911 if (bioq_first(&pmp->wthread_bioq) == NULL) {
912 mtxsleep(&pmp->wthread_bioq, &pmp->wthread_mtx,
355d67fc
MD
913 0, "h2bioqw", 0);
914 }
278ab2b2 915 cparent = NULL;
355d67fc 916
50456506 917 hammer2_trans_init(&trans, pmp, HAMMER2_TRANS_BUFCACHE);
a7720be7 918
065f4046 919 while ((bio = bioq_takefirst(&pmp->wthread_bioq)) != NULL) {
a7720be7 920 /*
a4dc31e0
MD
921 * dummy bio for synchronization. The transaction
922 * must be reinitialized.
a7720be7
MD
923 */
924 if (bio->bio_buf == NULL) {
925 bio->bio_flags |= BIO_DONE;
926 wakeup(bio);
a4dc31e0 927 hammer2_trans_done(&trans);
50456506 928 hammer2_trans_init(&trans, pmp,
a4dc31e0 929 HAMMER2_TRANS_BUFCACHE);
a7720be7
MD
930 continue;
931 }
932
933 /*
934 * else normal bio processing
935 */
065f4046 936 mtx_unlock(&pmp->wthread_mtx);
3f5b8b3b
MD
937
938 hammer2_lwinprog_drop(pmp);
355d67fc
MD
939
940 error = 0;
941 bp = bio->bio_buf;
942 vp = bp->b_vp;
943 ip = VTOI(vp);
944
355d67fc
MD
945 /*
946 * Inode is modified, flush size and mtime changes
947 * to ensure that the file size remains consistent
948 * with the buffers being flushed.
72ebfa75
MD
949 *
950 * NOTE: The inode_fsync() call only flushes the
951 * inode's meta-data state, it doesn't try
952 * to flush underlying buffers or chains.
355d67fc 953 */
278ab2b2 954 cparent = hammer2_inode_lock_ex(ip);
355d67fc
MD
955 if (ip->flags & (HAMMER2_INODE_RESIZED |
956 HAMMER2_INODE_MTIME)) {
278ab2b2 957 hammer2_inode_fsync(&trans, ip, cparent);
355d67fc 958 }
6a5f4fe6 959 wipdata = hammer2_cluster_modify_ip(&trans, ip,
278ab2b2 960 cparent, 0);
355d67fc
MD
961 lblksize = hammer2_calc_logical(ip, bio->bio_offset,
962 &lbase, NULL);
6a5f4fe6
MD
963 pblksize = hammer2_calc_physical(ip, wipdata, lbase);
964 hammer2_write_file_core(bp, &trans, ip, wipdata,
278ab2b2 965 cparent,
355d67fc
MD
966 lbase, IO_ASYNC,
967 pblksize, &error);
6a5f4fe6 968 hammer2_cluster_modsync(cparent);
278ab2b2 969 hammer2_inode_unlock_ex(ip, cparent);
355d67fc 970 if (error) {
1e5c08ba
MD
971 kprintf("hammer2: error in buffer write\n");
972 bp->b_flags |= B_ERROR;
973 bp->b_error = EIO;
355d67fc
MD
974 }
975 biodone(bio);
065f4046 976 mtx_lock(&pmp->wthread_mtx);
355d67fc 977 }
a7720be7 978 hammer2_trans_done(&trans);
355d67fc 979 }
065f4046
MD
980 pmp->wthread_destroy = -1;
981 wakeup(&pmp->wthread_destroy);
355d67fc 982
065f4046 983 mtx_unlock(&pmp->wthread_mtx);
355d67fc
MD
984}
985
a7720be7
MD
986void
987hammer2_bioq_sync(hammer2_pfsmount_t *pmp)
988{
989 struct bio sync_bio;
990
991 bzero(&sync_bio, sizeof(sync_bio)); /* dummy with no bio_buf */
992 mtx_lock(&pmp->wthread_mtx);
72ebfa75
MD
993 if (pmp->wthread_destroy == 0 &&
994 TAILQ_FIRST(&pmp->wthread_bioq.queue)) {
995 bioq_insert_tail(&pmp->wthread_bioq, &sync_bio);
a7720be7
MD
996 while ((sync_bio.bio_flags & BIO_DONE) == 0)
997 mtxsleep(&sync_bio, &pmp->wthread_mtx, 0, "h2bioq", 0);
998 }
999 mtx_unlock(&pmp->wthread_mtx);
1000}
1001
355d67fc 1002/*
51a0d27c
MD
1003 * Return a chain suitable for I/O, creating the chain if necessary
1004 * and assigning its physical block.
355d67fc
MD
1005 */
1006static
278ab2b2 1007hammer2_cluster_t *
355d67fc 1008hammer2_assign_physical(hammer2_trans_t *trans,
278ab2b2 1009 hammer2_inode_t *ip, hammer2_cluster_t *cparent,
355d67fc
MD
1010 hammer2_key_t lbase, int pblksize, int *errorp)
1011{
278ab2b2 1012 hammer2_cluster_t *cluster;
84e47819 1013 hammer2_cluster_t *dparent;
1897c66e 1014 hammer2_key_t key_dummy;
355d67fc 1015 int pradix = hammer2_getradix(pblksize);
278ab2b2 1016 int ddflag;
355d67fc
MD
1017
1018 /*
1019 * Locate the chain associated with lbase, return a locked chain.
1020 * However, do not instantiate any data reference (which utilizes a
1021 * device buffer) because we will be using direct IO via the
1022 * logical buffer cache buffer.
1023 */
1024 *errorp = 0;
50456506 1025 KKASSERT(pblksize >= HAMMER2_ALLOC_MIN);
355d67fc 1026retry:
84e47819
MD
1027 dparent = hammer2_cluster_lookup_init(cparent, 0);
1028 cluster = hammer2_cluster_lookup(dparent, &key_dummy,
355d67fc 1029 lbase, lbase,
278ab2b2 1030 HAMMER2_LOOKUP_NODATA, &ddflag);
355d67fc 1031
278ab2b2 1032 if (cluster == NULL) {
355d67fc
MD
1033 /*
1034 * We found a hole, create a new chain entry.
1035 *
1036 * NOTE: DATA chains are created without device backing
1037 * store (nor do we want any).
1038 */
84e47819 1039 *errorp = hammer2_cluster_create(trans, dparent, &cluster,
355d67fc
MD
1040 lbase, HAMMER2_PBUFRADIX,
1041 HAMMER2_BREF_TYPE_DATA,
b3659de2 1042 pblksize, 0);
278ab2b2 1043 if (cluster == NULL) {
84e47819 1044 hammer2_cluster_lookup_done(dparent);
278ab2b2 1045 panic("hammer2_cluster_create: par=%p error=%d\n",
84e47819 1046 dparent->focus, *errorp);
355d67fc
MD
1047 goto retry;
1048 }
355d67fc
MD
1049 /*ip->delta_dcount += pblksize;*/
1050 } else {
278ab2b2 1051 switch (hammer2_cluster_type(cluster)) {
355d67fc
MD
1052 case HAMMER2_BREF_TYPE_INODE:
1053 /*
1054 * The data is embedded in the inode. The
1055 * caller is responsible for marking the inode
1056 * modified and copying the data to the embedded
1057 * area.
1058 */
355d67fc
MD
1059 break;
1060 case HAMMER2_BREF_TYPE_DATA:
05dd26e4 1061 if (hammer2_cluster_need_resize(cluster, pblksize)) {
278ab2b2 1062 hammer2_cluster_resize(trans, ip,
84e47819 1063 dparent, cluster,
355d67fc
MD
1064 pradix,
1065 HAMMER2_MODIFY_OPTDATA);
1066 }
05dd26e4
MD
1067
1068 /*
1069 * DATA buffers must be marked modified whether the
1070 * data is in a logical buffer or not. We also have
1071 * to make this call to fixup the chain data pointers
1072 * after resizing in case this is an encrypted or
1073 * compressed buffer.
1074 */
278ab2b2 1075 hammer2_cluster_modify(trans, cluster,
05dd26e4 1076 HAMMER2_MODIFY_OPTDATA);
355d67fc
MD
1077 break;
1078 default:
1079 panic("hammer2_assign_physical: bad type");
1080 /* NOT REACHED */
355d67fc
MD
1081 break;
1082 }
1083 }
1084
1085 /*
84e47819
MD
1086 * Cleanup. If cluster wound up being the inode itself, i.e.
1087 * the DIRECTDATA case for offset 0, then we need to update cparent.
1088 * The caller expects cparent to not become stale.
355d67fc 1089 */
84e47819
MD
1090 hammer2_cluster_lookup_done(dparent);
1091 /* dparent = NULL; safety */
1092 if (cluster && ddflag)
278ab2b2 1093 hammer2_cluster_replace_locked(cparent, cluster);
278ab2b2 1094 return (cluster);
355d67fc
MD
1095}
1096
1097/*
a71db85d
MD
1098 * bio queued from hammer2_vnops.c.
1099 *
355d67fc 1100 * The core write function which determines which path to take
a71db85d
MD
1101 * depending on compression settings. We also have to locate the
1102 * related clusters so we can calculate and set the check data for
1103 * the blockref.
355d67fc
MD
1104 */
1105static
1106void
1e5c08ba 1107hammer2_write_file_core(struct buf *bp, hammer2_trans_t *trans,
bca9f8e6
MD
1108 hammer2_inode_t *ip,
1109 const hammer2_inode_data_t *ripdata,
278ab2b2 1110 hammer2_cluster_t *cparent,
355d67fc
MD
1111 hammer2_key_t lbase, int ioflag, int pblksize,
1112 int *errorp)
1113{
278ab2b2 1114 hammer2_cluster_t *cluster;
f481450f 1115
bca9f8e6 1116 switch(HAMMER2_DEC_ALGO(ripdata->comp_algo)) {
f481450f 1117 case HAMMER2_COMP_NONE:
355d67fc
MD
1118 /*
1119 * We have to assign physical storage to the buffer
1120 * we intend to dirty or write now to avoid deadlocks
1121 * in the strategy code later.
1122 *
1123 * This can return NOOFFSET for inode-embedded data.
1124 * The strategy code will take care of it in that case.
1125 */
278ab2b2 1126 cluster = hammer2_assign_physical(trans, ip, cparent,
355d67fc
MD
1127 lbase, pblksize,
1128 errorp);
e07becf8 1129 hammer2_write_bp(cluster, bp, ioflag, pblksize, errorp,
bca9f8e6 1130 ripdata->check_algo);
278ab2b2
MD
1131 if (cluster)
1132 hammer2_cluster_unlock(cluster);
f481450f
MD
1133 break;
1134 case HAMMER2_COMP_AUTOZERO:
1135 /*
1136 * Check for zero-fill only
1137 */
1138 hammer2_zero_check_and_write(bp, trans, ip,
bca9f8e6 1139 ripdata, cparent, lbase,
e07becf8 1140 ioflag, pblksize, errorp,
bca9f8e6 1141 ripdata->check_algo);
f481450f
MD
1142 break;
1143 case HAMMER2_COMP_LZ4:
1144 case HAMMER2_COMP_ZLIB:
1145 default:
1146 /*
1147 * Check for zero-fill and attempt compression.
1148 */
1149 hammer2_compress_and_write(bp, trans, ip,
bca9f8e6 1150 ripdata, cparent,
f481450f
MD
1151 lbase, ioflag,
1152 pblksize, errorp,
bca9f8e6
MD
1153 ripdata->comp_algo,
1154 ripdata->check_algo);
f481450f 1155 break;
355d67fc 1156 }
355d67fc
MD
1157}
1158
1159/*
355d67fc
MD
1160 * Generic function that will perform the compression in compression
1161 * write path. The compression algorithm is determined by the settings
1162 * obtained from inode.
1163 */
1164static
1165void
1e5c08ba 1166hammer2_compress_and_write(struct buf *bp, hammer2_trans_t *trans,
bca9f8e6 1167 hammer2_inode_t *ip, const hammer2_inode_data_t *ripdata,
278ab2b2 1168 hammer2_cluster_t *cparent,
355d67fc 1169 hammer2_key_t lbase, int ioflag, int pblksize,
e07becf8 1170 int *errorp, int comp_algo, int check_algo)
355d67fc 1171{
278ab2b2 1172 hammer2_cluster_t *cluster;
355d67fc 1173 hammer2_chain_t *chain;
f481450f
MD
1174 int comp_size;
1175 int comp_block_size;
278ab2b2 1176 int i;
f481450f 1177 char *comp_buffer;
355d67fc 1178
f481450f 1179 if (test_block_zeros(bp->b_data, pblksize)) {
bca9f8e6 1180 zero_write(bp, trans, ip, ripdata, cparent, lbase, errorp);
f481450f
MD
1181 return;
1182 }
1e5c08ba 1183
f481450f
MD
1184 comp_size = 0;
1185 comp_buffer = NULL;
355d67fc 1186
f481450f 1187 KKASSERT(pblksize / 2 <= 32768);
355d67fc 1188
f481450f
MD
1189 if (ip->comp_heuristic < 8 || (ip->comp_heuristic & 7) == 0) {
1190 z_stream strm_compress;
1191 int comp_level;
1192 int ret;
1193
e07becf8 1194 switch(HAMMER2_DEC_ALGO(comp_algo)) {
f481450f
MD
1195 case HAMMER2_COMP_LZ4:
1196 comp_buffer = objcache_get(cache_buffer_write,
1197 M_INTWAIT);
1198 comp_size = LZ4_compress_limitedOutput(
1199 bp->b_data,
1200 &comp_buffer[sizeof(int)],
1201 pblksize,
1202 pblksize / 2 - sizeof(int));
1203 /*
1204 * We need to prefix with the size, LZ4
1205 * doesn't do it for us. Add the related
1206 * overhead.
1207 */
1208 *(int *)comp_buffer = comp_size;
1209 if (comp_size)
1210 comp_size += sizeof(int);
1211 break;
1212 case HAMMER2_COMP_ZLIB:
1213 comp_level = HAMMER2_DEC_LEVEL(comp_algo);
1214 if (comp_level == 0)
1215 comp_level = 6; /* default zlib compression */
1216 else if (comp_level < 6)
1217 comp_level = 6;
1218 else if (comp_level > 9)
1219 comp_level = 9;
1220 ret = deflateInit(&strm_compress, comp_level);
1221 if (ret != Z_OK) {
1222 kprintf("HAMMER2 ZLIB: fatal error "
1223 "on deflateInit.\n");
1224 }
1225
1226 comp_buffer = objcache_get(cache_buffer_write,
1227 M_INTWAIT);
1228 strm_compress.next_in = bp->b_data;
1229 strm_compress.avail_in = pblksize;
1230 strm_compress.next_out = comp_buffer;
1231 strm_compress.avail_out = pblksize / 2;
1232 ret = deflate(&strm_compress, Z_FINISH);
1233 if (ret == Z_STREAM_END) {
1234 comp_size = pblksize / 2 -
1235 strm_compress.avail_out;
1e5c08ba 1236 } else {
f481450f 1237 comp_size = 0;
355d67fc 1238 }
f481450f
MD
1239 ret = deflateEnd(&strm_compress);
1240 break;
1241 default:
1242 kprintf("Error: Unknown compression method.\n");
1243 kprintf("Comp_method = %d.\n", comp_algo);
1244 break;
355d67fc 1245 }
f481450f 1246 }
1e5c08ba 1247
f481450f
MD
1248 if (comp_size == 0) {
1249 /*
1250 * compression failed or turned off
1251 */
1252 comp_block_size = pblksize; /* safety */
1253 if (++ip->comp_heuristic > 128)
1254 ip->comp_heuristic = 8;
1255 } else {
1256 /*
1257 * compression succeeded
1258 */
1259 ip->comp_heuristic = 0;
1260 if (comp_size <= 1024) {
1261 comp_block_size = 1024;
1262 } else if (comp_size <= 2048) {
1263 comp_block_size = 2048;
1264 } else if (comp_size <= 4096) {
1265 comp_block_size = 4096;
1266 } else if (comp_size <= 8192) {
1267 comp_block_size = 8192;
1268 } else if (comp_size <= 16384) {
1269 comp_block_size = 16384;
1270 } else if (comp_size <= 32768) {
1271 comp_block_size = 32768;
355d67fc 1272 } else {
f481450f
MD
1273 panic("hammer2: WRITE PATH: "
1274 "Weird comp_size value.");
1275 /* NOT REACHED */
1276 comp_block_size = pblksize;
1277 }
1278 }
1279
278ab2b2
MD
1280 cluster = hammer2_assign_physical(trans, ip, cparent,
1281 lbase, comp_block_size,
1282 errorp);
bca9f8e6 1283 ripdata = NULL;
f481450f
MD
1284
1285 if (*errorp) {
1286 kprintf("WRITE PATH: An error occurred while "
1287 "assigning physical space.\n");
278ab2b2
MD
1288 KKASSERT(cluster == NULL);
1289 goto done;
1290 }
1291
1292 for (i = 0; i < cluster->nchains; ++i) {
bca9f8e6 1293 hammer2_inode_data_t *wipdata;
fdf62707
MD
1294 hammer2_io_t *dio;
1295 char *bdata;
f481450f 1296
bca9f8e6 1297 chain = cluster->array[i]; /* XXX */
f481450f
MD
1298 KKASSERT(chain->flags & HAMMER2_CHAIN_MODIFIED);
1299
1300 switch(chain->bref.type) {
1301 case HAMMER2_BREF_TYPE_INODE:
bca9f8e6
MD
1302 wipdata = &hammer2_chain_wdata(chain)->ipdata;
1303 KKASSERT(wipdata->op_flags & HAMMER2_OPFLAG_DIRECTDATA);
f481450f 1304 KKASSERT(bp->b_loffset == 0);
bca9f8e6 1305 bcopy(bp->b_data, wipdata->u.data,
f481450f
MD
1306 HAMMER2_EMBEDDED_BYTES);
1307 break;
1308 case HAMMER2_BREF_TYPE_DATA:
1e5c08ba 1309 /*
f481450f
MD
1310 * Optimize out the read-before-write
1311 * if possible.
1e5c08ba 1312 */
fdf62707
MD
1313 *errorp = hammer2_io_newnz(chain->hmp,
1314 chain->bref.data_off,
1315 chain->bytes,
1316 &dio);
1317 if (*errorp) {
1318 hammer2_io_brelse(&dio);
1319 kprintf("hammer2: WRITE PATH: "
1320 "dbp bread error\n");
1321 break;
355d67fc 1322 }
fdf62707 1323 bdata = hammer2_io_data(dio, chain->bref.data_off);
355d67fc 1324
f481450f
MD
1325 /*
1326 * When loading the block make sure we don't
1327 * leave garbage after the compressed data.
1328 */
1329 if (comp_size) {
1330 chain->bref.methods =
1331 HAMMER2_ENC_COMP(comp_algo) +
e07becf8 1332 HAMMER2_ENC_CHECK(check_algo);
fdf62707 1333 bcopy(comp_buffer, bdata, comp_size);
f481450f 1334 if (comp_size != comp_block_size) {
fdf62707
MD
1335 bzero(bdata + comp_size,
1336 comp_block_size - comp_size);
355d67fc 1337 }
f481450f
MD
1338 } else {
1339 chain->bref.methods =
1340 HAMMER2_ENC_COMP(
1341 HAMMER2_COMP_NONE) +
e07becf8 1342 HAMMER2_ENC_CHECK(check_algo);
fdf62707 1343 bcopy(bp->b_data, bdata, pblksize);
f481450f 1344 }
355d67fc 1345
a71db85d
MD
1346 /*
1347 * The flush code doesn't calculate check codes for
1348 * file data (doing so can result in excessive I/O),
1349 * so we do it here.
1350 */
1351 hammer2_chain_setcheck(chain, bdata);
1352
f481450f 1353 /*
da6f36f4
MD
1354 * Device buffer is now valid, chain is no longer in
1355 * the initial state.
1356 *
1357 * (No blockref table worries with file data)
f481450f 1358 */
fdf62707 1359 atomic_clear_int(&chain->flags, HAMMER2_CHAIN_INITIAL);
355d67fc 1360
f481450f
MD
1361 /* Now write the related bdp. */
1362 if (ioflag & IO_SYNC) {
355d67fc 1363 /*
f481450f 1364 * Synchronous I/O requested.
355d67fc 1365 */
fdf62707 1366 hammer2_io_bwrite(&dio);
f481450f
MD
1367 /*
1368 } else if ((ioflag & IO_DIRECT) &&
1369 loff + n == pblksize) {
fdf62707 1370 hammer2_io_bdwrite(&dio);
f481450f
MD
1371 */
1372 } else if (ioflag & IO_ASYNC) {
fdf62707 1373 hammer2_io_bawrite(&dio);
f481450f 1374 } else {
fdf62707 1375 hammer2_io_bdwrite(&dio);
355d67fc 1376 }
f481450f
MD
1377 break;
1378 default:
1379 panic("hammer2_write_bp: bad chain type %d\n",
1380 chain->bref.type);
1381 /* NOT REACHED */
1382 break;
355d67fc 1383 }
355d67fc 1384 }
278ab2b2 1385done:
837bd39b
MD
1386 if (cluster)
1387 hammer2_cluster_unlock(cluster);
f481450f
MD
1388 if (comp_buffer)
1389 objcache_put(cache_buffer_write, comp_buffer);
355d67fc
MD
1390}
1391
1392/*
1393 * Function that performs zero-checking and writing without compression,
1394 * it corresponds to default zero-checking path.
1395 */
1396static
1397void
1e5c08ba 1398hammer2_zero_check_and_write(struct buf *bp, hammer2_trans_t *trans,
bca9f8e6 1399 hammer2_inode_t *ip, const hammer2_inode_data_t *ripdata,
278ab2b2 1400 hammer2_cluster_t *cparent,
e07becf8
MD
1401 hammer2_key_t lbase, int ioflag, int pblksize, int *errorp,
1402 int check_algo)
355d67fc 1403{
278ab2b2 1404 hammer2_cluster_t *cluster;
355d67fc 1405
f481450f 1406 if (test_block_zeros(bp->b_data, pblksize)) {
bca9f8e6 1407 zero_write(bp, trans, ip, ripdata, cparent, lbase, errorp);
f481450f 1408 } else {
278ab2b2
MD
1409 cluster = hammer2_assign_physical(trans, ip, cparent,
1410 lbase, pblksize, errorp);
e07becf8
MD
1411 hammer2_write_bp(cluster, bp, ioflag, pblksize, errorp,
1412 check_algo);
278ab2b2
MD
1413 if (cluster)
1414 hammer2_cluster_unlock(cluster);
355d67fc
MD
1415 }
1416}
1417
1418/*
1419 * A function to test whether a block of data contains only zeros,
f481450f 1420 * returns TRUE (non-zero) if the block is all zeros.
355d67fc
MD
1421 */
1422static
1423int
f481450f 1424test_block_zeros(const char *buf, size_t bytes)
355d67fc
MD
1425{
1426 size_t i;
1427
1428 for (i = 0; i < bytes; i += sizeof(long)) {
f481450f
MD
1429 if (*(const long *)(buf + i) != 0)
1430 return (0);
355d67fc 1431 }
f481450f 1432 return (1);
355d67fc
MD
1433}
1434
1435/*
1436 * Function to "write" a block that contains only zeros.
1437 */
1438static
1439void
6a5f4fe6 1440zero_write(struct buf *bp, hammer2_trans_t *trans,
bca9f8e6 1441 hammer2_inode_t *ip, const hammer2_inode_data_t *ripdata,
da6f36f4
MD
1442 hammer2_cluster_t *cparent,
1443 hammer2_key_t lbase, int *errorp __unused)
355d67fc 1444{
278ab2b2
MD
1445 hammer2_cluster_t *cluster;
1446 hammer2_media_data_t *data;
1897c66e 1447 hammer2_key_t key_dummy;
278ab2b2 1448 int ddflag;
355d67fc 1449
278ab2b2
MD
1450 cparent = hammer2_cluster_lookup_init(cparent, 0);
1451 cluster = hammer2_cluster_lookup(cparent, &key_dummy, lbase, lbase,
1452 HAMMER2_LOOKUP_NODATA, &ddflag);
1453 if (cluster) {
6a5f4fe6 1454 data = hammer2_cluster_wdata(cluster);
355d67fc 1455
278ab2b2 1456 if (ddflag) {
6a5f4fe6
MD
1457 KKASSERT(cluster->focus->flags &
1458 HAMMER2_CHAIN_MODIFIED);
278ab2b2 1459 bzero(data->ipdata.u.data, HAMMER2_EMBEDDED_BYTES);
6a5f4fe6 1460 hammer2_cluster_modsync(cluster);
355d67fc 1461 } else {
da6f36f4
MD
1462 hammer2_cluster_delete(trans, cparent, cluster,
1463 HAMMER2_DELETE_PERMANENT);
355d67fc 1464 }
278ab2b2 1465 hammer2_cluster_unlock(cluster);
355d67fc 1466 }
278ab2b2 1467 hammer2_cluster_lookup_done(cparent);
355d67fc
MD
1468}
1469
1470/*
1471 * Function to write the data as it is, without performing any sort of
1472 * compression. This function is used in path without compression and
1473 * default zero-checking path.
1474 */
1475static
1476void
278ab2b2 1477hammer2_write_bp(hammer2_cluster_t *cluster, struct buf *bp, int ioflag,
e07becf8 1478 int pblksize, int *errorp, int check_algo)
355d67fc 1479{
278ab2b2 1480 hammer2_chain_t *chain;
bca9f8e6 1481 hammer2_inode_data_t *wipdata;
fdf62707
MD
1482 hammer2_io_t *dio;
1483 char *bdata;
355d67fc 1484 int error;
278ab2b2 1485 int i;
355d67fc 1486
278ab2b2 1487 error = 0; /* XXX TODO below */
355d67fc 1488
278ab2b2 1489 for (i = 0; i < cluster->nchains; ++i) {
bca9f8e6 1490 chain = cluster->array[i]; /* XXX */
278ab2b2
MD
1491 KKASSERT(chain->flags & HAMMER2_CHAIN_MODIFIED);
1492
1493 switch(chain->bref.type) {
1494 case HAMMER2_BREF_TYPE_INODE:
bca9f8e6
MD
1495 wipdata = &hammer2_chain_wdata(chain)->ipdata;
1496 KKASSERT(wipdata->op_flags & HAMMER2_OPFLAG_DIRECTDATA);
278ab2b2 1497 KKASSERT(bp->b_loffset == 0);
bca9f8e6 1498 bcopy(bp->b_data, wipdata->u.data,
278ab2b2
MD
1499 HAMMER2_EMBEDDED_BYTES);
1500 error = 0;
fdf62707 1501 break;
278ab2b2
MD
1502 case HAMMER2_BREF_TYPE_DATA:
1503 error = hammer2_io_newnz(chain->hmp,
1504 chain->bref.data_off,
1505 chain->bytes, &dio);
1506 if (error) {
1507 hammer2_io_bqrelse(&dio);
1508 kprintf("hammer2: WRITE PATH: "
1509 "dbp bread error\n");
1510 break;
1511 }
1512 bdata = hammer2_io_data(dio, chain->bref.data_off);
355d67fc 1513
278ab2b2
MD
1514 chain->bref.methods = HAMMER2_ENC_COMP(
1515 HAMMER2_COMP_NONE) +
e07becf8 1516 HAMMER2_ENC_CHECK(check_algo);
278ab2b2 1517 bcopy(bp->b_data, bdata, chain->bytes);
355d67fc 1518
a71db85d
MD
1519 /*
1520 * The flush code doesn't calculate check codes for
1521 * file data (doing so can result in excessive I/O),
1522 * so we do it here.
1523 */
1524 hammer2_chain_setcheck(chain, bdata);
1525
355d67fc 1526 /*
da6f36f4
MD
1527 * Device buffer is now valid, chain is no longer in
1528 * the initial state.
1529 *
1530 * (No blockref table worries with file data)
355d67fc 1531 */
278ab2b2
MD
1532 atomic_clear_int(&chain->flags, HAMMER2_CHAIN_INITIAL);
1533
1534 if (ioflag & IO_SYNC) {
1535 /*
1536 * Synchronous I/O requested.
1537 */
1538 hammer2_io_bwrite(&dio);
1539 /*
1540 } else if ((ioflag & IO_DIRECT) &&
1541 loff + n == pblksize) {
1542 hammer2_io_bdwrite(&dio);
1543 */
1544 } else if (ioflag & IO_ASYNC) {
1545 hammer2_io_bawrite(&dio);
1546 } else {
1547 hammer2_io_bdwrite(&dio);
1548 }
1549 break;
1550 default:
1551 panic("hammer2_write_bp: bad chain type %d\n",
1552 chain->bref.type);
1553 /* NOT REACHED */
1554 error = 0;
1555 break;
355d67fc 1556 }
278ab2b2 1557 KKASSERT(error == 0); /* XXX TODO */
355d67fc 1558 }
1e5c08ba 1559 *errorp = error;
355d67fc
MD
1560}
1561
e118c14f
MD
1562static
1563int
10136ab6
MD
1564hammer2_remount(hammer2_mount_t *hmp, struct mount *mp, char *path,
1565 struct vnode *devvp, struct ucred *cred)
703720e4 1566{
10136ab6
MD
1567 int error;
1568
1569 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
1570 error = hammer2_recovery(hmp);
1571 } else {
1572 error = 0;
1573 }
1574 return error;
703720e4
MD
1575}
1576
e118c14f
MD
1577static
1578int
066e00cc 1579hammer2_vfs_unmount(struct mount *mp, int mntflags)
703720e4 1580{
e4e20f48 1581 hammer2_pfsmount_t *pmp;
54eb943b 1582 hammer2_mount_t *hmp;
7bed8d7e 1583 hammer2_chain_t *rchain;
58e43599 1584 hammer2_cluster_t *cluster;
703720e4 1585 int flags;
50e4f8f4 1586 int error = 0;
7bed8d7e 1587 int i;
703720e4 1588
e4e20f48 1589 pmp = MPTOPMP(mp);
703720e4 1590
623d43d4
MD
1591 if (pmp == NULL)
1592 return(0);
1593
9b6b3df4 1594 lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
58e43599 1595 TAILQ_REMOVE(&hammer2_pfslist, pmp, mntentry);
9b6b3df4 1596
065f4046
MD
1597 /*
1598 * If mount initialization proceeded far enough we must flush
1599 * its vnodes.
1600 */
1601 if (mntflags & MNT_FORCE)
1602 flags = FORCECLOSE;
1603 else
7bed8d7e 1604 flags = 0;
065f4046
MD
1605 if (pmp->iroot) {
1606 error = vflush(mp, 0, flags);
1607 if (error)
1608 goto failed;
1609 }
50e4f8f4 1610
eae2ed61 1611 ccms_domain_uninit(&pmp->ccms_dom);
eae2ed61 1612
065f4046
MD
1613 if (pmp->wthread_td) {
1614 mtx_lock(&pmp->wthread_mtx);
1615 pmp->wthread_destroy = 1;
1616 wakeup(&pmp->wthread_bioq);
1617 while (pmp->wthread_destroy != -1) {
1618 mtxsleep(&pmp->wthread_destroy,
1619 &pmp->wthread_mtx, 0,
1620 "umount-sleep", 0);
1621 }
1622 mtx_unlock(&pmp->wthread_mtx);
1623 pmp->wthread_td = NULL;
1624 }
e4e20f48 1625
044541cd
MD
1626 /*
1627 * Cleanup our reference on ihidden.
1628 */
1629 if (pmp->ihidden) {
1630 hammer2_inode_drop(pmp->ihidden);
1631 pmp->ihidden = NULL;
1632 }
7bed8d7e 1633
044541cd
MD
1634 /*
1635 * Cleanup our reference on iroot. iroot is (should) not be needed
1636 * by the flush code.
1637 */
1638 if (pmp->iroot) {
58e43599
MD
1639 cluster = &pmp->iroot->cluster;
1640 for (i = 0; i < pmp->iroot->cluster.nchains; ++i) {
1641 rchain = pmp->iroot->cluster.array[i];
1642 if (rchain == NULL)
1643 continue;
1644 hmp = rchain->hmp;
1645 hammer2_vfs_unmount_hmp1(mp, hmp);
7bed8d7e 1646
7bed8d7e 1647 atomic_clear_int(&rchain->flags, HAMMER2_CHAIN_MOUNTED);
0dea3156 1648#if REPORT_REFS_ERRORS
7bed8d7e
MD
1649 if (rchain->refs != 1)
1650 kprintf("PMP->RCHAIN %p REFS WRONG %d\n",
1651 rchain, rchain->refs);
0dea3156 1652#else
7bed8d7e 1653 KKASSERT(rchain->refs == 1);
0dea3156 1654#endif
7bed8d7e 1655 hammer2_chain_drop(rchain);
58e43599
MD
1656 cluster->array[i] = NULL;
1657 hammer2_vfs_unmount_hmp2(mp, hmp);
e4e20f48 1658 }
58e43599 1659 cluster->focus = NULL;
222d9e22 1660
58e43599
MD
1661#if REPORT_REFS_ERRORS
1662 if (pmp->iroot->refs != 1)
1663 kprintf("PMP->IROOT %p REFS WRONG %d\n",
1664 pmp->iroot, pmp->iroot->refs);
1665#else
1666 KKASSERT(pmp->iroot->refs == 1);
1667#endif
1668 /* ref for pmp->iroot */
1669 hammer2_inode_drop(pmp->iroot);
1670 pmp->iroot = NULL;
54eb943b 1671 }
703720e4 1672
e4e20f48 1673 pmp->mp = NULL;
54eb943b 1674 mp->mnt_data = NULL;
703720e4 1675
26bf1a36 1676 kmalloc_destroy(&pmp->mmsg);
99da41ea 1677 kmalloc_destroy(&pmp->minode);
26bf1a36 1678
e4e20f48 1679 kfree(pmp, M_HAMMER2);
9b6b3df4
MD
1680 error = 0;
1681
1682failed:
e4e20f48 1683 lockmgr(&hammer2_mntlk, LK_RELEASE);
a5913bdf 1684
703720e4
MD
1685 return (error);
1686}
1687
623d43d4
MD
1688static
1689void
1690hammer2_vfs_unmount_hmp1(struct mount *mp, hammer2_mount_t *hmp)
1691{
1692 hammer2_mount_exlock(hmp);
1693 --hmp->pmp_count;
1694
1695 kprintf("hammer2_unmount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
1696
063639ba
MD
1697 kdmsg_iocom_uninit(&hmp->iocom); /* XXX chain depend deadlck? */
1698
623d43d4 1699 /*
850687d2
MD
1700 * Cycle the volume data lock as a safety (probably not needed any
1701 * more). To ensure everything is out we need to flush at least
1702 * three times. (1) The running of the unlinkq can dirty the
1703 * filesystem, (2) A normal flush can dirty the freemap, and
1704 * (3) ensure that the freemap is fully synchronized.
623d43d4 1705 *
850687d2
MD
1706 * The next mount's recovery scan can clean everything up but we want
1707 * to leave the filesystem in a 100% clean state on a normal unmount.
623d43d4
MD
1708 */
1709 hammer2_voldata_lock(hmp);
850687d2 1710 hammer2_voldata_unlock(hmp);
18e8ab5f
MD
1711 if (mp->mnt_data) {
1712 hammer2_vfs_sync(mp, MNT_WAIT);
1713 hammer2_vfs_sync(mp, MNT_WAIT);
1714 hammer2_vfs_sync(mp, MNT_WAIT);
1715 }
850687d2 1716
623d43d4 1717 if (hmp->pmp_count == 0) {
da6f36f4
MD
1718 if ((hmp->vchain.flags | hmp->fchain.flags) &
1719 HAMMER2_CHAIN_FLUSH_MASK) {
623d43d4
MD
1720 kprintf("hammer2_unmount: chains left over "
1721 "after final sync\n");
da6f36f4
MD
1722 kprintf(" vchain %08x\n", hmp->vchain.flags);
1723 kprintf(" fchain %08x\n", hmp->fchain.flags);
0924b3f8 1724
623d43d4
MD
1725 if (hammer2_debug & 0x0010)
1726 Debugger("entered debugger");
1727 }
1728 }
1729}
1730
1731static
1732void
1733hammer2_vfs_unmount_hmp2(struct mount *mp, hammer2_mount_t *hmp)
1734{
50456506 1735 hammer2_pfsmount_t *spmp;
623d43d4
MD
1736 struct vnode *devvp;
1737 int dumpcnt;
1738 int ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
1739
1740 /*
1741 * If no PFS's left drop the master hammer2_mount for the
1742 * device.
1743 */
1744 if (hmp->pmp_count == 0) {
50456506
MD
1745 /*
1746 * Clean up SPMP and the super-root inode
1747 */
1748 spmp = hmp->spmp;
1749 if (spmp) {
1750 if (spmp->iroot) {
1751 hammer2_inode_drop(spmp->iroot);
1752 spmp->iroot = NULL;
1753 }
1754 hmp->spmp = NULL;
837bd39b
MD
1755 kmalloc_destroy(&spmp->mmsg);
1756 kmalloc_destroy(&spmp->minode);
50456506 1757 kfree(spmp, M_HAMMER2);
623d43d4
MD
1758 }
1759
1760 /*
1761 * Finish up with the device vnode
1762 */
1763 if ((devvp = hmp->devvp) != NULL) {
1764 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1765 vinvalbuf(devvp, (ronly ? 0 : V_SAVE), 0, 0);
1766 hmp->devvp = NULL;
8138a154 1767 VOP_CLOSE(devvp, (ronly ? FREAD : FREAD|FWRITE), NULL);
623d43d4
MD
1768 vn_unlock(devvp);
1769 vrele(devvp);
1770 devvp = NULL;
1771 }
1772
50456506
MD
1773 /*
1774 * Clear vchain/fchain flags that might prevent final cleanup
1775 * of these chains.
1776 */
1777 if (hmp->vchain.flags & HAMMER2_CHAIN_MODIFIED) {
1778 atomic_clear_int(&hmp->vchain.flags,
1779 HAMMER2_CHAIN_MODIFIED);
da6f36f4 1780 hammer2_pfs_memory_wakeup(hmp->vchain.pmp);
50456506
MD
1781 hammer2_chain_drop(&hmp->vchain);
1782 }
da6f36f4 1783 if (hmp->vchain.flags & HAMMER2_CHAIN_UPDATE) {
50456506 1784 atomic_clear_int(&hmp->vchain.flags,
da6f36f4 1785 HAMMER2_CHAIN_UPDATE);
50456506
MD
1786 hammer2_chain_drop(&hmp->vchain);
1787 }
1788
1789 if (hmp->fchain.flags & HAMMER2_CHAIN_MODIFIED) {
1790 atomic_clear_int(&hmp->fchain.flags,
1791 HAMMER2_CHAIN_MODIFIED);
da6f36f4 1792 hammer2_pfs_memory_wakeup(hmp->fchain.pmp);
50456506
MD
1793 hammer2_chain_drop(&hmp->fchain);
1794 }
da6f36f4 1795 if (hmp->fchain.flags & HAMMER2_CHAIN_UPDATE) {
50456506 1796 atomic_clear_int(&hmp->fchain.flags,
da6f36f4 1797 HAMMER2_CHAIN_UPDATE);
50456506
MD
1798 hammer2_chain_drop(&hmp->fchain);
1799 }
1800
623d43d4
MD
1801 /*
1802 * Final drop of embedded freemap root chain to
1803 * clean up fchain.core (fchain structure is not
1804 * flagged ALLOCATED so it is cleaned out and then
1805 * left to rot).
1806 */
1807 hammer2_chain_drop(&hmp->fchain);
1808
1809 /*
1810 * Final drop of embedded volume root chain to clean
1811 * up vchain.core (vchain structure is not flagged
1812 * ALLOCATED so it is cleaned out and then left to
1813 * rot).
1814 */
1815 dumpcnt = 50;
8138a154 1816 hammer2_dump_chain(&hmp->vchain, 0, &dumpcnt, 'v');
623d43d4 1817 dumpcnt = 50;
8138a154 1818 hammer2_dump_chain(&hmp->fchain, 0, &dumpcnt, 'f');
623d43d4
MD
1819 hammer2_mount_unlock(hmp);
1820 hammer2_chain_drop(&hmp->vchain);
1821
1822 hammer2_io_cleanup(hmp, &hmp->iotree);
1823 if (hmp->iofree_count) {
1824 kprintf("io_cleanup: %d I/O's left hanging\n",
1825 hmp->iofree_count);
1826 }
1827
1828 TAILQ_REMOVE(&hammer2_mntlist, hmp, mntentry);
1829 kmalloc_destroy(&hmp->mchain);
1830 kfree(hmp, M_HAMMER2);
1831 } else {
1832 hammer2_mount_unlock(hmp);
1833 }
1834}
1835
e118c14f
MD
1836static
1837int
066e00cc 1838hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
703720e4
MD
1839 ino_t ino, struct vnode **vpp)
1840{
1841 kprintf("hammer2_vget\n");
1842 return (EOPNOTSUPP);
1843}
1844
e118c14f
MD
1845static
1846int
066e00cc 1847hammer2_vfs_root(struct mount *mp, struct vnode **vpp)
703720e4 1848{
e4e20f48 1849 hammer2_pfsmount_t *pmp;
278ab2b2 1850 hammer2_cluster_t *cparent;
703720e4
MD
1851 int error;
1852 struct vnode *vp;
1853
e4e20f48 1854 pmp = MPTOPMP(mp);
e4e20f48 1855 if (pmp->iroot == NULL) {
703720e4
MD
1856 *vpp = NULL;
1857 error = EINVAL;
1858 } else {
278ab2b2 1859 cparent = hammer2_inode_lock_sh(pmp->iroot);
84e47819 1860 vp = hammer2_igetv(pmp->iroot, cparent, &error);
278ab2b2 1861 hammer2_inode_unlock_sh(pmp->iroot, cparent);
703720e4
MD
1862 *vpp = vp;
1863 if (vp == NULL)
1864 kprintf("vnodefail\n");
1865 }
703720e4
MD
1866
1867 return (error);
1868}
1869
28ee5f14
MD
1870/*
1871 * Filesystem status
1872 *
476d2aad 1873 * XXX incorporate ipdata->inode_quota and data_quota
28ee5f14 1874 */
e118c14f
MD
1875static
1876int
066e00cc 1877hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
703720e4 1878{
28ee5f14 1879 hammer2_pfsmount_t *pmp;
54eb943b 1880 hammer2_mount_t *hmp;
703720e4 1881
28ee5f14 1882 pmp = MPTOPMP(mp);
58e43599
MD
1883 KKASSERT(pmp->iroot->cluster.nchains >= 1);
1884 hmp = pmp->iroot->cluster.focus->hmp; /* XXX */
703720e4 1885
476d2aad 1886 mp->mnt_stat.f_files = pmp->inode_count;
28ee5f14
MD
1887 mp->mnt_stat.f_ffree = 0;
1888 mp->mnt_stat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
9f604b01 1889 mp->mnt_stat.f_bfree = hmp->voldata.allocator_free / HAMMER2_PBUFSIZE;
50e4f8f4 1890 mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
703720e4 1891
50e4f8f4 1892 *sbp = mp->mnt_stat;
703720e4
MD
1893 return (0);
1894}
1895
e118c14f
MD
1896static
1897int
066e00cc 1898hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
703720e4 1899{
28ee5f14 1900 hammer2_pfsmount_t *pmp;
50e4f8f4
MD
1901 hammer2_mount_t *hmp;
1902
28ee5f14 1903 pmp = MPTOPMP(mp);
58e43599
MD
1904 KKASSERT(pmp->iroot->cluster.nchains >= 1);
1905 hmp = pmp->iroot->cluster.focus->hmp; /* XXX */
50e4f8f4 1906
88a032af 1907 mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
476d2aad 1908 mp->mnt_vstat.f_files = pmp->inode_count;
28ee5f14
MD
1909 mp->mnt_vstat.f_ffree = 0;
1910 mp->mnt_vstat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
51f0b4b7 1911 mp->mnt_vstat.f_bfree = hmp->voldata.allocator_free / HAMMER2_PBUFSIZE;
28ee5f14 1912 mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
50e4f8f4
MD
1913
1914 *sbp = mp->mnt_vstat;
1915 return (0);
703720e4
MD
1916}
1917
10136ab6
MD
1918/*
1919 * Mount-time recovery (RW mounts)
1920 *
1921 * Updates to the free block table are allowed to lag flushes by one
1922 * transaction. In case of a crash, then on a fresh mount we must do an
50456506
MD
1923 * incremental scan of the last committed transaction id and make sure that
1924 * all related blocks have been marked allocated.
10136ab6 1925 *
50456506
MD
1926 * The super-root topology and each PFS has its own transaction id domain,
1927 * so we must track PFS boundary transitions.
10136ab6
MD
1928 */
1929struct hammer2_recovery_elm {
1930 TAILQ_ENTRY(hammer2_recovery_elm) entry;
1931 hammer2_chain_t *chain;
50456506 1932 hammer2_tid_t sync_tid;
10136ab6
MD
1933};
1934
1935TAILQ_HEAD(hammer2_recovery_list, hammer2_recovery_elm);
1936
50456506
MD
1937struct hammer2_recovery_info {
1938 struct hammer2_recovery_list list;
1939 int depth;
1940};
1941
10136ab6
MD
1942static int hammer2_recovery_scan(hammer2_trans_t *trans, hammer2_mount_t *hmp,
1943 hammer2_chain_t *parent,
50456506
MD
1944 struct hammer2_recovery_info *info,
1945 hammer2_tid_t sync_tid);
10136ab6
MD
1946
1947#define HAMMER2_RECOVERY_MAXDEPTH 10
1948
1949static
1950int
1951hammer2_recovery(hammer2_mount_t *hmp)
1952{
1953 hammer2_trans_t trans;
50456506 1954 struct hammer2_recovery_info info;
10136ab6
MD
1955 struct hammer2_recovery_elm *elm;
1956 hammer2_chain_t *parent;
50456506 1957 hammer2_tid_t sync_tid;
10136ab6
MD
1958 int error;
1959 int cumulative_error = 0;
1960
50456506 1961 hammer2_trans_init(&trans, hmp->spmp, 0);
10136ab6 1962
50456506
MD
1963 sync_tid = 0;
1964 TAILQ_INIT(&info.list);
1965 info.depth = 0;
10136ab6 1966 parent = hammer2_chain_lookup_init(&hmp->vchain, 0);
50456506
MD
1967 cumulative_error = hammer2_recovery_scan(&trans, hmp, parent,
1968 &info, sync_tid);
10136ab6
MD
1969 hammer2_chain_lookup_done(parent);
1970
50456506
MD
1971 while ((elm = TAILQ_FIRST(&info.list)) != NULL) {
1972 TAILQ_REMOVE(&info.list, elm, entry);
10136ab6 1973 parent = elm->chain;
50456506 1974 sync_tid = elm->sync_tid;
10136ab6
MD
1975 kfree(elm, M_HAMMER2);
1976
1977 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS |
1978 HAMMER2_RESOLVE_NOREF);
50456506
MD
1979 error = hammer2_recovery_scan(&trans, hmp, parent,
1980 &info, sync_tid);
10136ab6
MD
1981 hammer2_chain_unlock(parent);
1982 if (error)
1983 cumulative_error = error;
1984 }
1985 hammer2_trans_done(&trans);
1986
1987 return cumulative_error;
1988}
1989
1990static
1991int
1992hammer2_recovery_scan(hammer2_trans_t *trans, hammer2_mount_t *hmp,
1993 hammer2_chain_t *parent,
50456506
MD
1994 struct hammer2_recovery_info *info,
1995 hammer2_tid_t sync_tid)
10136ab6 1996{
bca9f8e6 1997 const hammer2_inode_data_t *ripdata;
10136ab6
MD
1998 hammer2_chain_t *chain;
1999 int cache_index;
2000 int cumulative_error = 0;
50456506 2001 int pfs_boundary = 0;
10136ab6
MD
2002 int error;
2003
10136ab6
MD
2004 /*
2005 * Adjust freemap to ensure that the block(s) are marked allocated.
2006 */
2007 if (parent->bref.type != HAMMER2_BREF_TYPE_VOLUME) {
2008 hammer2_freemap_adjust(trans, hmp, &parent->bref,
2009 HAMMER2_FREEMAP_DORECOVER);
2010 }
2011
2012 /*
2013 * Check type for recursive scan
2014 */
2015 switch(parent->bref.type) {
2016 case HAMMER2_BREF_TYPE_VOLUME:
2017 /* data already instantiated */
2018 break;
2019 case HAMMER2_BREF_TYPE_INODE:
2020 /*
2021 * Must instantiate data for DIRECTDATA test and also
2022 * for recursion.
2023 */
2024 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS);
bca9f8e6
MD
2025 ripdata = &hammer2_chain_rdata(parent)->ipdata;
2026 if (ripdata->op_flags & HAMMER2_OPFLAG_DIRECTDATA) {
10136ab6 2027 /* not applicable to recovery scan */
50456506 2028 hammer2_chain_unlock(parent);
10136ab6
MD
2029 return 0;
2030 }
bca9f8e6 2031 if ((ripdata->op_flags & HAMMER2_OPFLAG_PFSROOT) &&
50456506
MD
2032 info->depth != 0) {
2033 pfs_boundary = 1;
2034 sync_tid = parent->bref.mirror_tid - 1;
2035 }
2036 hammer2_chain_unlock(parent);
10136ab6
MD
2037 break;
2038 case HAMMER2_BREF_TYPE_INDIRECT:
2039 /*
2040 * Must instantiate data for recursion
2041 */
2042 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS);
2043 hammer2_chain_unlock(parent);
2044 break;
2045 case HAMMER2_BREF_TYPE_DATA:
2046 case HAMMER2_BREF_TYPE_FREEMAP:
2047 case HAMMER2_BREF_TYPE_FREEMAP_NODE:
2048 case HAMMER2_BREF_TYPE_FREEMAP_LEAF:
2049 /* not applicable to recovery scan */
2050 return 0;
2051 break;
2052 default:
2053 return EDOM;
2054 }
2055
50456506
MD
2056 /*
2057 * Defer operation if depth limit reached or if we are crossing a
2058 * PFS boundary.
2059 */
2060 if (info->depth >= HAMMER2_RECOVERY_MAXDEPTH || pfs_boundary) {
2061 struct hammer2_recovery_elm *elm;
2062
2063 elm = kmalloc(sizeof(*elm), M_HAMMER2, M_ZERO | M_WAITOK);
2064 elm->chain = parent;
2065 elm->sync_tid = sync_tid;
2066 hammer2_chain_ref(parent);
2067 TAILQ_INSERT_TAIL(&info->list, elm, entry);
2068 /* unlocked by caller */
2069
2070 return(0);
2071 }
2072
2073
10136ab6
MD
2074 /*
2075 * Recursive scan of the last flushed transaction only. We are
2076 * doing this without pmp assignments so don't leave the chains
2077 * hanging around after we are done with them.
2078 */
2079 cache_index = 0;
2080 chain = hammer2_chain_scan(parent, NULL, &cache_index,
2081 HAMMER2_LOOKUP_NODATA);
2082 while (chain) {
2083 atomic_set_int(&chain->flags, HAMMER2_CHAIN_RELEASE);
50456506
MD
2084 if (chain->bref.mirror_tid >= sync_tid) {
2085 ++info->depth;
10136ab6 2086 error = hammer2_recovery_scan(trans, hmp, chain,
50456506
MD
2087 info, sync_tid);
2088 --info->depth;
10136ab6
MD
2089 if (error)
2090 cumulative_error = error;
2091 }
2092 chain = hammer2_chain_scan(parent, chain, &cache_index,
2093 HAMMER2_LOOKUP_NODATA);
2094 }
2095
2096 return cumulative_error;
2097}
2098
703720e4
MD
2099/*
2100 * Sync the entire filesystem; this is called from the filesystem syncer
2101 * process periodically and whenever a user calls sync(1) on the hammer
2102 * mountpoint.
2103 *
2104 * Currently is actually called from the syncer! \o/
2105 *
2106 * This task will have to snapshot the state of the dirty inode chain.
2107 * From that, it will have to make sure all of the inodes on the dirty
2108 * chain have IO initiated. We make sure that io is initiated for the root
2109 * block.
2110 *
2111 * If waitfor is set, we wait for media to acknowledge the new rootblock.
2112 *
2113 * THINKS: side A vs side B, to have sync not stall all I/O?
2114 */
e118c14f 2115int
066e00cc 2116hammer2_vfs_sync(struct mount *mp, int waitfor)
703720e4 2117{
b7926f31 2118 struct hammer2_sync_info info;
50456506 2119 hammer2_inode_t *iroot;
a7720be7 2120 hammer2_chain_t *chain;
50456506 2121 hammer2_chain_t *parent;
a5913bdf 2122 hammer2_pfsmount_t *pmp;
54eb943b 2123 hammer2_mount_t *hmp;
b7926f31
MD
2124 int flags;
2125 int error;
7bed8d7e 2126 int total_error;
a7720be7 2127 int force_fchain;
3fc4c63d 2128 int i;
03188ed8 2129 int j;
703720e4 2130
a5913bdf 2131 pmp = MPTOPMP(mp);
50456506
MD
2132 iroot = pmp->iroot;
2133 KKASSERT(iroot);
2134 KKASSERT(iroot->pmp == pmp);
703720e4 2135
355d67fc
MD
2136 /*
2137 * We can't acquire locks on existing vnodes while in a transaction
2138 * without risking a deadlock. This assumes that vfsync() can be
2139 * called without the vnode locked (which it can in DragonFly).
2140 * Otherwise we'd have to implement a multi-pass or flag the lock
2141 * failures and retry.
c057466c
MD
2142 *
2143 * The reclamation code interlocks with the sync list's token
2144 * (by removing the vnode from the scan list) before unlocking
2145 * the inode, giving us time to ref the inode.
355d67fc
MD
2146 */
2147 /*flags = VMSC_GETVP;*/
2148 flags = 0;
b7926f31
MD
2149 if (waitfor & MNT_LAZY)
2150 flags |= VMSC_ONEPASS;
2151
d7bfb2cb 2152 /*
052e0aa0
MD
2153 * Start our flush transaction. This does not return until all
2154 * concurrent transactions have completed and will prevent any
2155 * new transactions from running concurrently, except for the
2156 * buffer cache transactions.
2157 *
2158 * For efficiency do an async pass before making sure with a
2159 * synchronous pass on all related buffer cache buffers. It
2160 * should theoretically not be possible for any new file buffers
2161 * to be instantiated during this sequence.
d7bfb2cb 2162 */
50456506
MD
2163 hammer2_trans_init(&info.trans, pmp, HAMMER2_TRANS_ISFLUSH |
2164 HAMMER2_TRANS_PREFLUSH);
eae2ed61 2165 hammer2_run_unlinkq(&info.trans, pmp);
50456506 2166
b7926f31
MD
2167 info.error = 0;
2168 info.waitfor = MNT_NOWAIT;
eddc656a 2169 vsyncscan(mp, flags | VMSC_NOWAIT, hammer2_sync_scan2, &info);
052e0aa0
MD
2170 info.waitfor = MNT_WAIT;
2171 vsyncscan(mp, flags, hammer2_sync_scan2, &info);
a7720be7 2172
052e0aa0
MD
2173 /*
2174 * Clear PREFLUSH. This prevents (or asserts on) any new logical
2175 * buffer cache flushes which occur during the flush. Device buffers
2176 * are not affected.
2177 */
2178
2179#if 0
b7926f31
MD
2180 if (info.error == 0 && (waitfor & MNT_WAIT)) {
2181 info.waitfor = waitfor;
eddc656a 2182 vsyncscan(mp, flags, hammer2_sync_scan2, &info);
b7926f31
MD
2183
2184 }
052e0aa0 2185#endif
a4dc31e0 2186 hammer2_bioq_sync(info.trans.pmp);
052e0aa0 2187 atomic_clear_int(&info.trans.flags, HAMMER2_TRANS_PREFLUSH);
a7720be7 2188
50456506
MD
2189 total_error = 0;
2190
2191 /*
2192 * Flush all storage elements making up the cluster
2193 *
2194 * We must also flush any deleted siblings because the super-root
2195 * flush won't do it for us. They all must be staged or the
2196 * super-root flush will not be able to update its block table
2197 * properly.
2198 *
2199 * XXX currently done serially instead of concurrently
2200 */
2201 for (i = 0; iroot && i < iroot->cluster.nchains; ++i) {
2202 chain = iroot->cluster.array[i];
2203 if (chain) {
2204 hammer2_chain_lock(chain, HAMMER2_RESOLVE_ALWAYS);
da6f36f4 2205 hammer2_flush(&info.trans, chain);
50456506
MD
2206 hammer2_chain_unlock(chain);
2207 }
50456506 2208 }
052e0aa0 2209#if 0
50456506
MD
2210 hammer2_trans_done(&info.trans);
2211#endif
2212
a7720be7 2213 /*
50456506
MD
2214 * Flush all volume roots to synchronize PFS flushes with the
2215 * storage media. Use a super-root transaction for each one.
2216 *
2217 * The flush code will detect super-root -> pfs-root chain
2218 * transitions using the last pfs-root flush.
a7720be7 2219 */
50456506
MD
2220 for (i = 0; iroot && i < iroot->cluster.nchains; ++i) {
2221 chain = iroot->cluster.array[i];
2222 if (chain == NULL)
2223 continue;
2224
2225 hmp = chain->hmp;
03188ed8
MD
2226
2227 /*
2228 * We only have to flush each hmp once
2229 */
2230 for (j = i - 1; j >= 0; --j) {
2231 if (iroot->cluster.array[j] &&
2232 iroot->cluster.array[j]->hmp == hmp)
2233 break;
2234 }
2235 if (j >= 0)
2236 continue;
2237 hammer2_trans_spmp(&info.trans, hmp->spmp);
a5913bdf 2238
50456506
MD
2239 /*
2240 * Force an update of the XID from the PFS root to the
2241 * topology root. We couldn't do this from the PFS
2242 * transaction because a SPMP transaction is needed.
2243 * This does not modify blocks, instead what it does is
2244 * allow the flush code to find the transition point and
2245 * then update on the way back up.
2246 */
da6f36f4 2247 parent = chain->parent;
50456506 2248 KKASSERT(chain->pmp != parent->pmp);
da6f36f4 2249 hammer2_chain_setflush(&info.trans, parent);
a5913bdf 2250
d7bfb2cb
MD
2251 /*
2252 * Media mounts have two 'roots', vchain for the topology
2253 * and fchain for the free block table. Flush both.
2254 *
2255 * Note that the topology and free block table are handled
2256 * independently, so the free block table can wind up being
2257 * ahead of the topology. We depend on the bulk free scan
2258 * code to deal with any loose ends.
2259 */
50456506 2260 hammer2_chain_lock(&hmp->vchain, HAMMER2_RESOLVE_ALWAYS);
10136ab6 2261 hammer2_chain_lock(&hmp->fchain, HAMMER2_RESOLVE_ALWAYS);
da6f36f4 2262 if (hmp->fchain.flags & HAMMER2_CHAIN_FLUSH_MASK) {
50456506
MD
2263 /*
2264 * This will also modify vchain as a side effect,
2265 * mark vchain as modified now.
2266 */
2267 hammer2_voldata_modify(hmp);
10136ab6 2268 chain = &hmp->fchain;
da6f36f4 2269 hammer2_flush(&info.trans, chain);
10136ab6
MD
2270 KKASSERT(chain == &hmp->fchain);
2271 }
2272 hammer2_chain_unlock(&hmp->fchain);
50456506 2273 hammer2_chain_unlock(&hmp->vchain);
10136ab6 2274
7bed8d7e 2275 hammer2_chain_lock(&hmp->vchain, HAMMER2_RESOLVE_ALWAYS);
da6f36f4 2276 if (hmp->vchain.flags & HAMMER2_CHAIN_FLUSH_MASK) {
a7720be7 2277 chain = &hmp->vchain;
da6f36f4 2278 hammer2_flush(&info.trans, chain);
a7720be7
MD
2279 KKASSERT(chain == &hmp->vchain);
2280 force_fchain = 1;
2281 } else {
2282 force_fchain = 0;
7bed8d7e
MD
2283 }
2284 hammer2_chain_unlock(&hmp->vchain);
1a7cfe5a 2285
10136ab6 2286#if 0
7bed8d7e 2287 hammer2_chain_lock(&hmp->fchain, HAMMER2_RESOLVE_ALWAYS);
da6f36f4 2288 if ((hmp->fchain.flags & HAMMER2_CHAIN_FLUSH_MASK) ||
a7720be7 2289 force_fchain) {
d7bfb2cb 2290 /* this will also modify vchain as a side effect */
a7720be7 2291 chain = &hmp->fchain;
da6f36f4 2292 hammer2_flush(&info.trans, chain);
a7720be7 2293 KKASSERT(chain == &hmp->fchain);
7bed8d7e
MD
2294 }
2295 hammer2_chain_unlock(&hmp->fchain);
10136ab6 2296#endif
1a7cfe5a 2297
7bed8d7e 2298 error = 0;
b7926f31 2299
2910a90c 2300 /*
7bed8d7e
MD
2301 * We can't safely flush the volume header until we have
2302 * flushed any device buffers which have built up.
2303 *
2304 * XXX this isn't being incremental
2910a90c 2305 */
7bed8d7e
MD
2306 vn_lock(hmp->devvp, LK_EXCLUSIVE | LK_RETRY);
2307 error = VOP_FSYNC(hmp->devvp, MNT_WAIT, 0);
2308 vn_unlock(hmp->devvp);
b7926f31 2309
2910a90c 2310 /*
7bed8d7e
MD
2311 * The flush code sets CHAIN_VOLUMESYNC to indicate that the
2312 * volume header needs synchronization via hmp->volsync.
2313 *
2314 * XXX synchronize the flag & data with only this flush XXX
2910a90c 2315 */
7bed8d7e
MD
2316 if (error == 0 &&
2317 (hmp->vchain.flags & HAMMER2_CHAIN_VOLUMESYNC)) {
2318 struct buf *bp;
2319
2320 /*
2321 * Synchronize the disk before flushing the volume
2322 * header.
2323 */
2324 bp = getpbuf(NULL);
2325 bp->b_bio1.bio_offset = 0;
2326 bp->b_bufsize = 0;
2327 bp->b_bcount = 0;
2328 bp->b_cmd = BUF_CMD_FLUSH;
2329 bp->b_bio1.bio_done = biodone_sync;
2330 bp->b_bio1.bio_flags |= BIO_SYNC;
2331 vn_strategy(hmp->devvp, &bp->b_bio1);
2332 biowait(&bp->b_bio1, "h2vol");
2333 relpbuf(bp, NULL);
2334
2335 /*
2336 * Then we can safely flush the version of the
2337 * volume header synchronized by the flush code.
2338 */
2339 i = hmp->volhdrno + 1;
2340 if (i >= HAMMER2_NUM_VOLHDRS)
2341 i = 0;
2342 if (i * HAMMER2_ZONE_BYTES64 + HAMMER2_SEGSIZE >
2343 hmp->volsync.volu_size) {
2344 i = 0;
2345 }
2346 kprintf("sync volhdr %d %jd\n",
2347 i, (intmax_t)hmp->volsync.volu_size);
2348 bp = getblk(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
2349 HAMMER2_PBUFSIZE, 0, 0);
2350 atomic_clear_int(&hmp->vchain.flags,
2351 HAMMER2_CHAIN_VOLUMESYNC);
2352 bcopy(&hmp->volsync, bp->b_data, HAMMER2_PBUFSIZE);
2353 bawrite(bp);
2354 hmp->volhdrno = i;
3fc4c63d 2355 }
7bed8d7e
MD
2356 if (error)
2357 total_error = error;
50456506
MD
2358
2359#if 0
2360 hammer2_trans_done(&info.trans);
2361#endif
b7926f31 2362 }
d001f460 2363 hammer2_trans_done(&info.trans);
a4dc31e0 2364
7bed8d7e 2365 return (total_error);
b7926f31 2366}
703720e4 2367
214f4a77
MD
2368/*
2369 * Sync passes.
214f4a77 2370 */
b7926f31
MD
2371static int
2372hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
2373{
2374 struct hammer2_sync_info *info = data;
2375 hammer2_inode_t *ip;
2376 int error;
2377
c057466c
MD
2378 /*
2379 *
2380 */
b7926f31 2381 ip = VTOI(vp);
eddc656a
MD
2382 if (ip == NULL)
2383 return(0);
2384 if (vp->v_type == VNON || vp->v_type == VBAD) {
2385 vclrisdirty(vp);
2386 return(0);
2387 }
2388 if ((ip->flags & HAMMER2_INODE_MODIFIED) == 0 &&
2389 RB_EMPTY(&vp->v_rbdirty_tree)) {
2390 vclrisdirty(vp);
b7926f31
MD
2391 return(0);
2392 }
d001f460
MD
2393
2394 /*
2395 * VOP_FSYNC will start a new transaction so replicate some code
2396 * here to do it inline (see hammer2_vop_fsync()).
355d67fc
MD
2397 *
2398 * WARNING: The vfsync interacts with the buffer cache and might
2399 * block, we can't hold the inode lock at that time.
c057466c
MD
2400 * However, we MUST ref ip before blocking to ensure that
2401 * it isn't ripped out from under us (since we do not
2402 * hold a lock on the vnode).
d001f460 2403 */
c057466c 2404 hammer2_inode_ref(ip);
9596b8c4 2405 atomic_clear_int(&ip->flags, HAMMER2_INODE_MODIFIED);
c057466c
MD
2406 if (vp)
2407 vfsync(vp, MNT_NOWAIT, 1, NULL, NULL);
a7720be7 2408
c057466c 2409 hammer2_inode_drop(ip);
052e0aa0 2410#if 1
d001f460 2411 error = 0;
b7926f31
MD
2412 if (error)
2413 info->error = error;
052e0aa0 2414#endif
b7926f31 2415 return(0);
703720e4
MD
2416}
2417
e118c14f
MD
2418static
2419int
066e00cc 2420hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp)
703720e4
MD
2421{
2422 return (0);
2423}
2424
e118c14f
MD
2425static
2426int
066e00cc 2427hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
703720e4
MD
2428 struct fid *fhp, struct vnode **vpp)
2429{
2430 return (0);
2431}
2432
e118c14f
MD
2433static
2434int
066e00cc 2435hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
703720e4
MD
2436 int *exflagsp, struct ucred **credanonp)
2437{
2438 return (0);
2439}
50e4f8f4
MD
2440
2441/*
2442 * Support code for hammer2_mount(). Read, verify, and install the volume
2443 * header into the HMP
2444 *
2445 * XXX read four volhdrs and use the one with the highest TID whos CRC
2446 * matches.
2447 *
2448 * XXX check iCRCs.
57381c9e
VS
2449 *
2450 * XXX For filesystems w/ less than 4 volhdrs, make sure to not write to
2451 * nonexistant locations.
2452 *
2453 * XXX Record selected volhdr and ring updates to each of 4 volhdrs
50e4f8f4
MD
2454 */
2455static
2456int
2457hammer2_install_volume_header(hammer2_mount_t *hmp)
2458{
2459 hammer2_volume_data_t *vd;
99924359 2460 struct buf *bp;
60fbd5f4 2461 hammer2_crc32_t crc0, crc, bcrc0, bcrc;
99924359
MD
2462 int error_reported;
2463 int error;
0b3147ba
VS
2464 int valid;
2465 int i;
50e4f8f4 2466
99924359 2467 error_reported = 0;
0b3147ba
VS
2468 error = 0;
2469 valid = 0;
99924359 2470 bp = NULL;
0b3147ba 2471
99924359
MD
2472 /*
2473 * There are up to 4 copies of the volume header (syncs iterate
2474 * between them so there is no single master). We don't trust the
2475 * volu_size field so we don't know precisely how large the filesystem
2476 * is, so depend on the OS to return an error if we go beyond the
2477 * block device's EOF.
2478 */
0b3147ba 2479 for (i = 0; i < HAMMER2_NUM_VOLHDRS; i++) {
62efe6ec 2480 error = bread(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
99924359 2481 HAMMER2_VOLUME_BYTES, &bp);
0b3147ba 2482 if (error) {
99924359
MD
2483 brelse(bp);
2484 bp = NULL;
0b3147ba
VS
2485 continue;
2486 }
2487
88a032af
VS
2488 vd = (struct hammer2_volume_data *) bp->b_data;
2489 if ((vd->magic != HAMMER2_VOLUME_ID_HBO) &&
2490 (vd->magic != HAMMER2_VOLUME_ID_ABO)) {
2491 brelse(bp);
2492 bp = NULL;
0b3147ba 2493 continue;
88a032af
VS
2494 }
2495
2496 if (vd->magic == HAMMER2_VOLUME_ID_ABO) {
2497 /* XXX: Reversed-endianness filesystem */
2498 kprintf("hammer2: reverse-endian filesystem detected");
2499 brelse(bp);
2500 bp = NULL;
2501 continue;
2502 }
0b3147ba
VS
2503
2504 crc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT0];
60fbd5f4 2505 crc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC0_OFF,
99924359 2506 HAMMER2_VOLUME_ICRC0_SIZE);
60fbd5f4
VS
2507 bcrc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT1];
2508 bcrc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC1_OFF,
2509 HAMMER2_VOLUME_ICRC1_SIZE);
2510 if ((crc0 != crc) || (bcrc0 != bcrc)) {
99924359 2511 kprintf("hammer2 volume header crc "
3fc4c63d 2512 "mismatch copy #%d %08x/%08x\n",
60fbd5f4 2513 i, crc0, crc);
99924359
MD
2514 error_reported = 1;
2515 brelse(bp);
2516 bp = NULL;
0b3147ba
VS
2517 continue;
2518 }
4d5318eb 2519 if (valid == 0 || hmp->voldata.mirror_tid < vd->mirror_tid) {
99924359
MD
2520 valid = 1;
2521 hmp->voldata = *vd;
3fc4c63d 2522 hmp->volhdrno = i;
0b3147ba 2523 }
99924359
MD
2524 brelse(bp);
2525 bp = NULL;
50e4f8f4 2526 }
0b3147ba 2527 if (valid) {
3fc4c63d 2528 hmp->volsync = hmp->voldata;
57381c9e 2529 error = 0;
3fc4c63d
MD
2530 if (error_reported || bootverbose || 1) { /* 1/DEBUG */
2531 kprintf("hammer2: using volume header #%d\n",
2532 hmp->volhdrno);
2533 }
0b3147ba
VS
2534 } else {
2535 error = EINVAL;
99924359 2536 kprintf("hammer2: no valid volume headers found!\n");
0b3147ba 2537 }
0b3147ba 2538 return (error);
50e4f8f4 2539}
0b3147ba 2540
1a34728c
MD
2541/*
2542 * Reconnect using the passed file pointer. The caller must ref the
2543 * fp for us.
2544 */
2545void
063639ba 2546hammer2_cluster_reconnect(hammer2_mount_t *hmp, struct file *fp)
1a34728c 2547{
3a5aa68f 2548 size_t name_len;
063639ba 2549 const char *name = "disk-volume";
a5913bdf 2550
70c3c3b7 2551 /*
3a5aa68f
MD
2552 * Closes old comm descriptor, kills threads, cleans up
2553 * states, then installs the new descriptor and creates
2554 * new threads.
1a34728c 2555 */
063639ba 2556 kdmsg_iocom_reconnect(&hmp->iocom, fp, "hammer2");
bfc3a7b1 2557
9b8b748f 2558 /*
063639ba
MD
2559 * Setup LNK_CONN fields for autoinitiated state machine. We
2560 * will use SPANs to advertise multiple PFSs so only pass the
2561 * fsid and HAMMER2_PFSTYPE_SUPROOT for the AUTOCONN.
2562 *
2563 * We are not initiating a LNK_SPAN so we do not have to set-up
2564 * iocom.auto_lnk_span.
537d97bc 2565 */
063639ba
MD
2566 bzero(&hmp->iocom.auto_lnk_conn.pfs_clid,
2567 sizeof(hmp->iocom.auto_lnk_conn.pfs_clid));
2568 hmp->iocom.auto_lnk_conn.pfs_fsid = hmp->voldata.fsid;
2569 hmp->iocom.auto_lnk_conn.pfs_type = HAMMER2_PFSTYPE_SUPROOT;
2570 hmp->iocom.auto_lnk_conn.proto_version = DMSG_SPAN_PROTO_1;
2571#if 0
2572 hmp->iocom.auto_lnk_conn.peer_type = hmp->voldata.peer_type;
2573#endif
2574 hmp->iocom.auto_lnk_conn.peer_type = DMSG_PEER_HAMMER2;
537d97bc
MD
2575
2576 /*
2577 * Filter adjustment. Clients do not need visibility into other
2578 * clients (otherwise millions of clients would present a serious
2579 * problem). The fs_label also serves to restrict the namespace.
9b8b748f 2580 */
063639ba
MD
2581 hmp->iocom.auto_lnk_conn.peer_mask = 1LLU << DMSG_PEER_HAMMER2;
2582 hmp->iocom.auto_lnk_conn.pfs_mask = (uint64_t)-1;
2583
2584#if 0
476d2aad 2585 switch (ipdata->pfs_type) {
537d97bc 2586 case DMSG_PFSTYPE_CLIENT:
063639ba 2587 hmp->iocom.auto_lnk_conn.peer_mask &=
537d97bc
MD
2588 ~(1LLU << DMSG_PFSTYPE_CLIENT);
2589 break;
2590 default:
2591 break;
2592 }
063639ba 2593#endif
537d97bc 2594
063639ba
MD
2595 name_len = strlen(name);
2596 if (name_len >= sizeof(hmp->iocom.auto_lnk_conn.fs_label))
2597 name_len = sizeof(hmp->iocom.auto_lnk_conn.fs_label) - 1;
2598 bcopy(name, hmp->iocom.auto_lnk_conn.fs_label, name_len);
2599 hmp->iocom.auto_lnk_conn.fs_label[name_len] = 0;
537d97bc 2600
063639ba 2601 kdmsg_iocom_autoinitiate(&hmp->iocom, hammer2_autodmsg);
70c3c3b7
MD
2602}
2603
9b8b748f 2604static int
537d97bc 2605hammer2_rcvdmsg(kdmsg_msg_t *msg)
9b8b748f 2606{
063639ba
MD
2607 kprintf("RCVMSG %08x\n", msg->tcmd);
2608
2609 switch(msg->tcmd) {
537d97bc 2610 case DMSG_DBG_SHELL:
10c86c4e 2611 /*
ea7c725f 2612 * (non-transaction)
537d97bc 2613 * Execute shell command (not supported atm)
10c86c4e 2614 */
063639ba 2615 kdmsg_msg_result(msg, DMSG_ERR_NOSUPP);
8c280d5d 2616 break;
537d97bc 2617 case DMSG_DBG_SHELL | DMSGF_REPLY:
ea7c725f
MD
2618 /*
2619 * (non-transaction)
2620 */
537d97bc
MD
2621 if (msg->aux_data) {
2622 msg->aux_data[msg->aux_size - 1] = 0;
2623 kprintf("HAMMER2 DBG: %s\n", msg->aux_data);
2624 }
8c280d5d
MD
2625 break;
2626 default:
3b76886b 2627 /*
ea7c725f 2628 * Unsupported message received. We only need to
3b76886b 2629 * reply if it's a transaction in order to close our end.
063639ba 2630 * Ignore any one-way messages or any further messages
3b76886b
MD
2631 * associated with the transaction.
2632 *
2633 * NOTE: This case also includes DMSG_LNK_ERROR messages
2634 * which might be one-way, replying to those would
2635 * cause an infinite ping-pong.
2636 */
2637 if (msg->any.head.cmd & DMSGF_CREATE)
2638 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
8c280d5d
MD
2639 break;
2640 }
2641 return(0);
2642}
2643
2644/*
537d97bc
MD
2645 * This function is called after KDMSG has automatically handled processing
2646 * of a LNK layer message (typically CONN, SPAN, or CIRC).
8c280d5d 2647 *
537d97bc
MD
2648 * We tag off the LNK_CONN to trigger our LNK_VOLCONF messages which
2649 * advertises all available hammer2 super-root volumes.
8c280d5d 2650 */
da6f36f4
MD
2651static void hammer2_update_spans(hammer2_mount_t *hmp, kdmsg_state_t *state);
2652
537d97bc
MD
2653static void
2654hammer2_autodmsg(kdmsg_msg_t *msg)
8c280d5d 2655{
da6f36f4 2656 hammer2_mount_t *hmp = msg->state->iocom->handle;
1a34728c 2657 int copyid;
8c280d5d 2658
063639ba 2659 kprintf("RCAMSG %08x\n", msg->tcmd);
537d97bc 2660
063639ba
MD
2661 switch(msg->tcmd) {
2662 case DMSG_LNK_CONN | DMSGF_CREATE | DMSGF_REPLY:
2663 case DMSG_LNK_CONN | DMSGF_CREATE | DMSGF_DELETE | DMSGF_REPLY:
2664 if (msg->any.head.cmd & DMSGF_CREATE) {
2665 kprintf("HAMMER2: VOLDATA DUMP\n");
70c3c3b7 2666
063639ba
MD
2667 /*
2668 * Dump the configuration stored in the volume header.
2669 * This will typically be import/export access rights,
2670 * master encryption keys (encrypted), etc.
2671 */
2672 hammer2_voldata_lock(hmp);
2673 copyid = 0;
2674 while (copyid < HAMMER2_COPYID_COUNT) {
2675 if (hmp->voldata.copyinfo[copyid].copyid)
2676 hammer2_volconf_update(hmp, copyid);
2677 ++copyid;
2678 }
2679 hammer2_voldata_unlock(hmp);
1a34728c 2680
063639ba 2681 kprintf("HAMMER2: INITIATE SPANs\n");
da6f36f4 2682 hammer2_update_spans(hmp, msg->state);
1a34728c 2683 }
063639ba
MD
2684 if ((msg->any.head.cmd & DMSGF_DELETE) &&
2685 msg->state && (msg->state->txcmd & DMSGF_DELETE) == 0) {
2686 kprintf("HAMMER2: CONN WAS TERMINATED\n");
2687 }
2688 break;
2689 default:
2690 break;
8c280d5d 2691 }
063639ba
MD
2692}
2693
2694/*
2695 * Update LNK_SPAN state
2696 */
da6f36f4
MD
2697static void
2698hammer2_update_spans(hammer2_mount_t *hmp, kdmsg_state_t *state)
063639ba 2699{
bca9f8e6 2700 const hammer2_inode_data_t *ripdata;
063639ba
MD
2701 hammer2_cluster_t *cparent;
2702 hammer2_cluster_t *cluster;
2703 hammer2_pfsmount_t *spmp;
2704 hammer2_key_t key_next;
2705 kdmsg_msg_t *rmsg;
2706 size_t name_len;
2707 int ddflag;
2708
2709 /*
2710 * Lookup mount point under the media-localized super-root.
2711 *
2712 * cluster->pmp will incorrectly point to spmp and must be fixed
2713 * up later on.
2714 */
2715 spmp = hmp->spmp;
2716 cparent = hammer2_inode_lock_ex(spmp->iroot);
2717 cluster = hammer2_cluster_lookup(cparent, &key_next,
2718 HAMMER2_KEY_MIN,
2719 HAMMER2_KEY_MAX,
2720 0, &ddflag);
2721 while (cluster) {
2722 if (hammer2_cluster_type(cluster) != HAMMER2_BREF_TYPE_INODE)
2723 continue;
bca9f8e6
MD
2724 ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
2725 kprintf("UPDATE SPANS: %s\n", ripdata->filename);
063639ba 2726
da6f36f4 2727 rmsg = kdmsg_msg_alloc(state, DMSG_LNK_SPAN | DMSGF_CREATE,
063639ba 2728 hammer2_lnk_span_reply, NULL);
bca9f8e6
MD
2729 rmsg->any.lnk_span.pfs_clid = ripdata->pfs_clid;
2730 rmsg->any.lnk_span.pfs_fsid = ripdata->pfs_fsid;
2731 rmsg->any.lnk_span.pfs_type = ripdata->pfs_type;
063639ba
MD
2732 rmsg->any.lnk_span.peer_type = DMSG_PEER_HAMMER2;
2733 rmsg->any.lnk_span.proto_version = DMSG_SPAN_PROTO_1;
bca9f8e6 2734 name_len = ripdata->name_len;
063639ba
MD
2735 if (name_len >= sizeof(rmsg->any.lnk_span.fs_label))
2736 name_len = sizeof(rmsg->any.lnk_span.fs_label) - 1;
bca9f8e6 2737 bcopy(ripdata->filename, rmsg->any.lnk_span.fs_label, name_len);
063639ba
MD
2738
2739 kdmsg_msg_write(rmsg);
2740
2741 cluster = hammer2_cluster_next(cparent, cluster,
2742 &key_next,
2743 key_next,
2744 HAMMER2_KEY_MAX,
2745 0);
8c280d5d 2746 }
063639ba
MD
2747 hammer2_inode_unlock_ex(spmp->iroot, cparent);
2748}
2749
2750static
2751int
2752hammer2_lnk_span_reply(kdmsg_state_t *state, kdmsg_msg_t *msg)
2753{
2754 if ((state->txcmd & DMSGF_DELETE) == 0 &&
2755 (msg->any.head.cmd & DMSGF_DELETE)) {
2756 kdmsg_msg_reply(msg, 0);
2757 }
2758 return 0;
9b8b748f 2759}
1a34728c
MD
2760
2761/*
2762 * Volume configuration updates are passed onto the userland service
2763 * daemon via the open LNK_CONN transaction.
2764 */
2765void
063639ba 2766hammer2_volconf_update(hammer2_mount_t *hmp, int index)
1a34728c 2767{
3a5aa68f 2768 kdmsg_msg_t *msg;
1a34728c
MD
2769
2770 /* XXX interlock against connection state termination */
063639ba
MD
2771 kprintf("volconf update %p\n", hmp->iocom.conn_state);
2772 if (hmp->iocom.conn_state) {
1a34728c 2773 kprintf("TRANSMIT VOLCONF VIA OPEN CONN TRANSACTION\n");
da6f36f4
MD
2774 msg = kdmsg_msg_alloc(hmp->iocom.conn_state,
2775 DMSG_LNK_HAMMER2_VOLCONF,
2776 NULL, NULL);
1f179146
MD
2777 H2_LNK_VOLCONF(msg)->copy = hmp->voldata.copyinfo[index];
2778 H2_LNK_VOLCONF(msg)->mediaid = hmp->voldata.fsid;
2779 H2_LNK_VOLCONF(msg)->index = index;
3a5aa68f 2780 kdmsg_msg_write(msg);
1a34728c
MD
2781 }
2782}
9797e933 2783
3f5b8b3b
MD
2784/*
2785 * This handles hysteresis on regular file flushes. Because the BIOs are
2786 * routed to a thread it is possible for an excessive number to build up
2787 * and cause long front-end stalls long before the runningbuffspace limit
2788 * is hit, so we implement hammer2_flush_pipe to control the
2789 * hysteresis.
2790 *
2791 * This is a particular problem when compression is used.
2792 */
2793void
2794hammer2_lwinprog_ref(hammer2_pfsmount_t *pmp)
2795{
2796 atomic_add_int(&pmp->count_lwinprog, 1);
2797}
2798
2799void
2800hammer2_lwinprog_drop(hammer2_pfsmount_t *pmp)
2801{
2802 int lwinprog;
2803
2804 lwinprog = atomic_fetchadd_int(&pmp->count_lwinprog, -1);
2805 if ((lwinprog & HAMMER2_LWINPROG_WAITING) &&
2806 (lwinprog & HAMMER2_LWINPROG_MASK) <= hammer2_flush_pipe * 2 / 3) {
2807 atomic_clear_int(&pmp->count_lwinprog,
2808 HAMMER2_LWINPROG_WAITING);
2809 wakeup(&pmp->count_lwinprog);
2810 }
2811}
2812
2813void
2814hammer2_lwinprog_wait(hammer2_pfsmount_t *pmp)
2815{
2816 int lwinprog;
2817
2818 for (;;) {
2819 lwinprog = pmp->count_lwinprog;
2820 cpu_ccfence();
2821 if ((lwinprog & HAMMER2_LWINPROG_MASK) < hammer2_flush_pipe)
2822 break;
2823 tsleep_interlock(&pmp->count_lwinprog, 0);
2824 atomic_set_int(&pmp->count_lwinprog, HAMMER2_LWINPROG_WAITING);
2825 lwinprog = pmp->count_lwinprog;
2826 if ((lwinprog & HAMMER2_LWINPROG_MASK) < hammer2_flush_pipe)
2827 break;
2828 tsleep(&pmp->count_lwinprog, PINTERLOCKED, "h2wpipe", hz);
2829 }
2830}
2831
278ab2b2
MD
2832/*
2833 * Manage excessive memory resource use for chain and related
2834 * structures.
2835 */
2836void
2837hammer2_pfs_memory_wait(hammer2_pfsmount_t *pmp)
2838{
da6f36f4
MD
2839 uint32_t waiting;
2840 uint32_t count;
2841 uint32_t limit;
278ab2b2
MD
2842#if 0
2843 static int zzticks;
2844#endif
2845
2846 /*
2847 * Atomic check condition and wait. Also do an early speedup of
2848 * the syncer to try to avoid hitting the wait.
2849 */
2850 for (;;) {
2851 waiting = pmp->inmem_dirty_chains;
2852 cpu_ccfence();
2853 count = waiting & HAMMER2_DIRTYCHAIN_MASK;
2854
2855 limit = pmp->mp->mnt_nvnodelistsize / 10;
2856 if (limit < hammer2_limit_dirty_chains)
2857 limit = hammer2_limit_dirty_chains;
2858 if (limit < 1000)
2859 limit = 1000;
2860
2861#if 0
2862 if ((int)(ticks - zzticks) > hz) {
2863 zzticks = ticks;
2864 kprintf("count %ld %ld\n", count, limit);
2865 }
2866#endif
2867
2868 /*
2869 * Block if there are too many dirty chains present, wait
2870 * for the flush to clean some out.
2871 */
2872 if (count > limit) {
2873 tsleep_interlock(&pmp->inmem_dirty_chains, 0);
da6f36f4 2874 if (atomic_cmpset_int(&pmp->inmem_dirty_chains,
278ab2b2
MD
2875 waiting,
2876 waiting | HAMMER2_DIRTYCHAIN_WAITING)) {
2877 speedup_syncer(pmp->mp);
2878 tsleep(&pmp->inmem_dirty_chains, PINTERLOCKED,
2879 "chnmem", hz);
2880 }
2881 continue; /* loop on success or fail */
2882 }
2883
2884 /*
2885 * Try to start an early flush before we are forced to block.
2886 */
2887 if (count > limit * 7 / 10)
2888 speedup_syncer(pmp->mp);
2889 break;
2890 }
2891}
2892
2893void
2894hammer2_pfs_memory_inc(hammer2_pfsmount_t *pmp)
2895{
da6f36f4
MD
2896 if (pmp) {
2897 atomic_add_int(&pmp->inmem_dirty_chains, 1);
2898 }
278ab2b2
MD
2899}
2900
2901void
2902hammer2_pfs_memory_wakeup(hammer2_pfsmount_t *pmp)
2903{
da6f36f4 2904 uint32_t waiting;
278ab2b2
MD
2905
2906 if (pmp == NULL)
2907 return;
2908
2909 for (;;) {
2910 waiting = pmp->inmem_dirty_chains;
2911 cpu_ccfence();
da6f36f4 2912 if (atomic_cmpset_int(&pmp->inmem_dirty_chains,
278ab2b2
MD
2913 waiting,
2914 (waiting - 1) &
2915 ~HAMMER2_DIRTYCHAIN_WAITING)) {
2916 break;
2917 }
2918 }
2919
2920 if (waiting & HAMMER2_DIRTYCHAIN_WAITING)
2921 wakeup(&pmp->inmem_dirty_chains);
2922}
2923
2924/*
2925 * Debugging
2926 */
9797e933 2927void
8138a154 2928hammer2_dump_chain(hammer2_chain_t *chain, int tab, int *countp, char pfx)
9797e933
MD
2929{
2930 hammer2_chain_t *scan;
da6f36f4 2931 hammer2_chain_t *parent;
9797e933 2932
a02dfba1
MD
2933 --*countp;
2934 if (*countp == 0) {
2935 kprintf("%*.*s...\n", tab, tab, "");
2936 return;
2937 }
2938 if (*countp < 0)
2939 return;
8138a154
MD
2940 kprintf("%*.*s%c-chain %p.%d %016jx/%d mir=%016jx\n",
2941 tab, tab, "", pfx,
a7720be7
MD
2942 chain, chain->bref.type,
2943 chain->bref.key, chain->bref.keybits,
2944 chain->bref.mirror_tid);
2945
da6f36f4 2946 kprintf("%*.*s [%08x] (%s) refs=%d\n",
a7720be7
MD
2947 tab, tab, "",
2948 chain->flags,
9797e933
MD
2949 ((chain->bref.type == HAMMER2_BREF_TYPE_INODE &&
2950 chain->data) ? (char *)chain->data->ipdata.filename : "?"),
8853dfb5 2951 chain->refs);
a7720be7 2952
da6f36f4 2953 kprintf("%*.*s core [%08x]",
a7720be7 2954 tab, tab, "",
da6f36f4 2955 chain->core.flags);
a7720be7 2956
da6f36f4
MD
2957 parent = chain->parent;
2958 if (parent)
2959 kprintf("\n%*.*s p=%p [pflags %08x prefs %d",
8138a154 2960 tab, tab, "",
da6f36f4
MD
2961 parent, parent->flags, parent->refs);
2962 if (RB_EMPTY(&chain->core.rbtree)) {
9797e933 2963 kprintf("\n");
da6f36f4 2964 } else {
9797e933 2965 kprintf(" {\n");
da6f36f4 2966 RB_FOREACH(scan, hammer2_chain_tree, &chain->core.rbtree)
8138a154 2967 hammer2_dump_chain(scan, tab + 4, countp, 'a');
9797e933
MD
2968 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE && chain->data)
2969 kprintf("%*.*s}(%s)\n", tab, tab, "",
2970 chain->data->ipdata.filename);
2971 else
2972 kprintf("%*.*s}\n", tab, tab, "");
2973 }
2974}