2 * Copyright (c) 2011-2013 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
6 * by Daniel Flores (GSOC 2013 - mentored by Matthew Dillon, compression)
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in
16 * the documentation and/or other materials provided with the
18 * 3. Neither the name of The DragonFly Project nor the names of its
19 * contributors may be used to endorse or promote products derived
20 * from this software without specific, prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
25 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
26 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
27 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
28 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
29 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
30 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
31 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
32 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/kernel.h>
38 #include <sys/nlookup.h>
39 #include <sys/vnode.h>
40 #include <sys/mount.h>
41 #include <sys/fcntl.h>
44 #include <sys/vfsops.h>
45 #include <sys/sysctl.h>
46 #include <sys/socket.h>
47 #include <sys/objcache.h>
50 #include <sys/namei.h>
51 #include <sys/mountctl.h>
52 #include <sys/dirent.h>
55 #include <sys/mutex.h>
56 #include <sys/mutex2.h>
59 #include "hammer2_disk.h"
60 #include "hammer2_mount.h"
63 #include "hammer2_lz4.h"
65 #include "zlib/hammer2_zlib.h"
67 #define REPORT_REFS_ERRORS 1 /* XXX remove me */
69 MALLOC_DEFINE(M_OBJCACHE, "objcache", "Object Cache");
71 struct hammer2_sync_info {
72 hammer2_trans_t trans;
77 TAILQ_HEAD(hammer2_mntlist, hammer2_mount);
78 static struct hammer2_mntlist hammer2_mntlist;
79 static struct lock hammer2_mntlk;
82 int hammer2_cluster_enable = 1;
83 int hammer2_hardlink_enable = 1;
84 long hammer2_iod_file_read;
85 long hammer2_iod_meta_read;
86 long hammer2_iod_indr_read;
87 long hammer2_iod_fmap_read;
88 long hammer2_iod_volu_read;
89 long hammer2_iod_file_write;
90 long hammer2_iod_meta_write;
91 long hammer2_iod_indr_write;
92 long hammer2_iod_fmap_write;
93 long hammer2_iod_volu_write;
94 long hammer2_ioa_file_read;
95 long hammer2_ioa_meta_read;
96 long hammer2_ioa_indr_read;
97 long hammer2_ioa_fmap_read;
98 long hammer2_ioa_volu_read;
99 long hammer2_ioa_fmap_write;
100 long hammer2_ioa_file_write;
101 long hammer2_ioa_meta_write;
102 long hammer2_ioa_indr_write;
103 long hammer2_ioa_volu_write;
105 MALLOC_DECLARE(C_BUFFER);
106 MALLOC_DEFINE(C_BUFFER, "compbuffer", "Buffer used for compression.");
108 MALLOC_DECLARE(D_BUFFER);
109 MALLOC_DEFINE(D_BUFFER, "decompbuffer", "Buffer used for decompression.");
111 MALLOC_DECLARE(W_BIOQUEUE);
112 MALLOC_DEFINE(W_BIOQUEUE, "wbioqueue", "Writing bio queue.");
114 MALLOC_DECLARE(W_MTX);
115 MALLOC_DEFINE(W_MTX, "wmutex", "Mutex for write thread.");
117 SYSCTL_NODE(_vfs, OID_AUTO, hammer2, CTLFLAG_RW, 0, "HAMMER2 filesystem");
119 SYSCTL_INT(_vfs_hammer2, OID_AUTO, debug, CTLFLAG_RW,
120 &hammer2_debug, 0, "");
121 SYSCTL_INT(_vfs_hammer2, OID_AUTO, cluster_enable, CTLFLAG_RW,
122 &hammer2_cluster_enable, 0, "");
123 SYSCTL_INT(_vfs_hammer2, OID_AUTO, hardlink_enable, CTLFLAG_RW,
124 &hammer2_hardlink_enable, 0, "");
126 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_read, CTLFLAG_RW,
127 &hammer2_iod_file_read, 0, "");
128 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_read, CTLFLAG_RW,
129 &hammer2_iod_meta_read, 0, "");
130 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_read, CTLFLAG_RW,
131 &hammer2_iod_indr_read, 0, "");
132 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_fmap_read, CTLFLAG_RW,
133 &hammer2_iod_fmap_read, 0, "");
134 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_read, CTLFLAG_RW,
135 &hammer2_iod_volu_read, 0, "");
137 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_write, CTLFLAG_RW,
138 &hammer2_iod_file_write, 0, "");
139 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_write, CTLFLAG_RW,
140 &hammer2_iod_meta_write, 0, "");
141 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_write, CTLFLAG_RW,
142 &hammer2_iod_indr_write, 0, "");
143 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_fmap_write, CTLFLAG_RW,
144 &hammer2_iod_fmap_write, 0, "");
145 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_write, CTLFLAG_RW,
146 &hammer2_iod_volu_write, 0, "");
148 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_read, CTLFLAG_RW,
149 &hammer2_ioa_file_read, 0, "");
150 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_read, CTLFLAG_RW,
151 &hammer2_ioa_meta_read, 0, "");
152 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_read, CTLFLAG_RW,
153 &hammer2_ioa_indr_read, 0, "");
154 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_fmap_read, CTLFLAG_RW,
155 &hammer2_ioa_fmap_read, 0, "");
156 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_volu_read, CTLFLAG_RW,
157 &hammer2_ioa_volu_read, 0, "");
159 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_write, CTLFLAG_RW,
160 &hammer2_ioa_file_write, 0, "");
161 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_write, CTLFLAG_RW,
162 &hammer2_ioa_meta_write, 0, "");
163 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_write, CTLFLAG_RW,
164 &hammer2_ioa_indr_write, 0, "");
165 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_fmap_write, CTLFLAG_RW,
166 &hammer2_ioa_fmap_write, 0, "");
167 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_volu_write, CTLFLAG_RW,
168 &hammer2_ioa_volu_write, 0, "");
170 static int hammer2_vfs_init(struct vfsconf *conf);
171 static int hammer2_vfs_uninit(struct vfsconf *vfsp);
172 static int hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
174 static int hammer2_remount(struct mount *, char *, struct vnode *,
176 static int hammer2_vfs_unmount(struct mount *mp, int mntflags);
177 static int hammer2_vfs_root(struct mount *mp, struct vnode **vpp);
178 static int hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp,
180 static int hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
182 static int hammer2_vfs_sync(struct mount *mp, int waitfor);
183 static int hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
184 ino_t ino, struct vnode **vpp);
185 static int hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
186 struct fid *fhp, struct vnode **vpp);
187 static int hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp);
188 static int hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
189 int *exflagsp, struct ucred **credanonp);
191 static int hammer2_install_volume_header(hammer2_mount_t *hmp);
192 static int hammer2_sync_scan1(struct mount *mp, struct vnode *vp, void *data);
193 static int hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
195 static void hammer2_write_thread(void *arg);
198 * Functions for compression in threads,
199 * from hammer2_vnops.c
201 static void hammer2_write_file_core(struct buf *bp, hammer2_trans_t *trans,
203 hammer2_inode_data_t *ipdata,
204 hammer2_chain_t **parentp,
205 hammer2_key_t lbase, int ioflag, int pblksize,
207 static void hammer2_compress_and_write(struct buf *bp, hammer2_trans_t *trans,
209 hammer2_inode_data_t *ipdata,
210 hammer2_chain_t **parentp,
211 hammer2_key_t lbase, int ioflag,
212 int pblksize, int *errorp, int comp_algo);
213 static void hammer2_zero_check_and_write(struct buf *bp,
214 hammer2_trans_t *trans, hammer2_inode_t *ip,
215 hammer2_inode_data_t *ipdata,
216 hammer2_chain_t **parentp,
218 int ioflag, int pblksize, int *errorp);
219 static int test_block_zeros(const char *buf, size_t bytes);
220 static void zero_write(struct buf *bp, hammer2_trans_t *trans,
222 hammer2_inode_data_t *ipdata,
223 hammer2_chain_t **parentp,
226 static void hammer2_write_bp(hammer2_chain_t *chain, struct buf *bp,
227 int ioflag, int pblksize, int *errorp);
229 static int hammer2_rcvdmsg(kdmsg_msg_t *msg);
230 static void hammer2_autodmsg(kdmsg_msg_t *msg);
234 * HAMMER2 vfs operations.
236 static struct vfsops hammer2_vfsops = {
237 .vfs_init = hammer2_vfs_init,
238 .vfs_uninit = hammer2_vfs_uninit,
239 .vfs_sync = hammer2_vfs_sync,
240 .vfs_mount = hammer2_vfs_mount,
241 .vfs_unmount = hammer2_vfs_unmount,
242 .vfs_root = hammer2_vfs_root,
243 .vfs_statfs = hammer2_vfs_statfs,
244 .vfs_statvfs = hammer2_vfs_statvfs,
245 .vfs_vget = hammer2_vfs_vget,
246 .vfs_vptofh = hammer2_vfs_vptofh,
247 .vfs_fhtovp = hammer2_vfs_fhtovp,
248 .vfs_checkexp = hammer2_vfs_checkexp
251 MALLOC_DEFINE(M_HAMMER2, "HAMMER2-mount", "");
253 VFS_SET(hammer2_vfsops, hammer2, 0);
254 MODULE_VERSION(hammer2, 1);
258 hammer2_vfs_init(struct vfsconf *conf)
260 static struct objcache_malloc_args margs_read;
261 static struct objcache_malloc_args margs_write;
267 if (HAMMER2_BLOCKREF_BYTES != sizeof(struct hammer2_blockref))
269 if (HAMMER2_INODE_BYTES != sizeof(struct hammer2_inode_data))
271 if (HAMMER2_VOLUME_BYTES != sizeof(struct hammer2_volume_data))
275 kprintf("HAMMER2 structure size mismatch; cannot continue.\n");
277 margs_read.objsize = 65536;
278 margs_read.mtype = D_BUFFER;
280 margs_write.objsize = 32768;
281 margs_write.mtype = C_BUFFER;
283 cache_buffer_read = objcache_create(margs_read.mtype->ks_shortdesc,
284 0, 1, NULL, NULL, NULL, objcache_malloc_alloc,
285 objcache_malloc_free, &margs_read);
286 cache_buffer_write = objcache_create(margs_write.mtype->ks_shortdesc,
287 0, 1, NULL, NULL, NULL, objcache_malloc_alloc,
288 objcache_malloc_free, &margs_write);
290 lockinit(&hammer2_mntlk, "mntlk", 0, 0);
291 TAILQ_INIT(&hammer2_mntlist);
298 hammer2_vfs_uninit(struct vfsconf *vfsp __unused)
300 objcache_destroy(cache_buffer_read);
301 objcache_destroy(cache_buffer_write);
306 * Mount or remount HAMMER2 fileystem from physical media
309 * mp mount point structure
315 * mp mount point structure
316 * path path to mount point
317 * data pointer to argument structure in user space
318 * volume volume path (device@LABEL form)
319 * hflags user mount flags
320 * cred user credentials
327 hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
330 struct hammer2_mount_info info;
331 hammer2_pfsmount_t *pmp;
332 hammer2_mount_t *hmp;
333 hammer2_key_t key_next;
334 hammer2_key_t key_dummy;
337 struct nlookupdata nd;
338 hammer2_chain_t *parent;
339 hammer2_chain_t *schain;
340 hammer2_chain_t *rchain;
342 char devstr[MNAMELEN];
358 kprintf("hammer2_mount\n");
364 bzero(&info, sizeof(info));
365 info.cluster_fd = -1;
369 * Non-root mount or updating a mount
371 error = copyin(data, &info, sizeof(info));
375 error = copyinstr(info.volume, devstr, MNAMELEN - 1, &done);
379 /* Extract device and label */
381 label = strchr(devstr, '@');
383 ((label + 1) - dev) > done) {
391 if (mp->mnt_flag & MNT_UPDATE) {
393 /* HAMMER2 implements NFS export via mountctl */
396 error = hammer2_remount(mp, path, devvp, cred);
404 * Lookup name and verify it refers to a block device.
406 error = nlookup_init(&nd, dev, UIO_SYSSPACE, NLC_FOLLOW);
408 error = nlookup(&nd);
410 error = cache_vref(&nd.nl_nch, nd.nl_cred, &devvp);
414 if (vn_isdisk(devvp, &error))
415 error = vfs_mountedon(devvp);
419 * Determine if the device has already been mounted. After this
420 * check hmp will be non-NULL if we are doing the second or more
421 * hammer2 mounts from the same device.
423 lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
424 TAILQ_FOREACH(hmp, &hammer2_mntlist, mntentry) {
425 if (hmp->devvp == devvp)
430 * Open the device if this isn't a secondary mount and construct
431 * the H2 device mount (hmp).
434 if (error == 0 && vcount(devvp) > 0)
438 * Now open the device
441 ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
442 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
443 error = vinvalbuf(devvp, V_SAVE, 0, 0);
445 error = VOP_OPEN(devvp,
446 ronly ? FREAD : FREAD | FWRITE,
451 if (error && devvp) {
456 lockmgr(&hammer2_mntlk, LK_RELEASE);
459 hmp = kmalloc(sizeof(*hmp), M_HAMMER2, M_WAITOK | M_ZERO);
462 kmalloc_create(&hmp->mchain, "HAMMER2-chains");
463 TAILQ_INSERT_TAIL(&hammer2_mntlist, hmp, mntentry);
465 lockinit(&hmp->alloclk, "h2alloc", 0, 0);
466 lockinit(&hmp->voldatalk, "voldata", 0, LK_CANRECURSE);
467 TAILQ_INIT(&hmp->transq);
470 * vchain setup. vchain.data is embedded.
471 * vchain.refs is initialized and will never drop to 0.
473 hmp->vchain.hmp = hmp;
474 hmp->vchain.refs = 1;
475 hmp->vchain.data = (void *)&hmp->voldata;
476 hmp->vchain.bref.type = HAMMER2_BREF_TYPE_VOLUME;
477 hmp->vchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
478 hmp->vchain.delete_tid = HAMMER2_MAX_TID;
479 hammer2_chain_core_alloc(NULL, &hmp->vchain, NULL);
480 /* hmp->vchain.u.xxx is left NULL */
483 * fchain setup. fchain.data is embedded.
484 * fchain.refs is initialized and will never drop to 0.
486 * The data is not used but needs to be initialized to
487 * pass assertion muster. We use this chain primarily
488 * as a placeholder for the freemap's top-level RBTREE
489 * so it does not interfere with the volume's topology
492 hmp->fchain.hmp = hmp;
493 hmp->fchain.refs = 1;
494 hmp->fchain.data = (void *)&hmp->voldata.freemap_blockset;
495 hmp->fchain.bref.type = HAMMER2_BREF_TYPE_FREEMAP;
496 hmp->fchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
497 hmp->fchain.bref.methods =
498 HAMMER2_ENC_CHECK(HAMMER2_CHECK_FREEMAP) |
499 HAMMER2_ENC_COMP(HAMMER2_COMP_NONE);
500 hmp->fchain.delete_tid = HAMMER2_MAX_TID;
502 hammer2_chain_core_alloc(NULL, &hmp->fchain, NULL);
503 /* hmp->fchain.u.xxx is left NULL */
506 * Install the volume header
508 error = hammer2_install_volume_header(hmp);
510 hammer2_vfs_unmount(mp, MNT_FORCE);
515 * First locate the super-root inode, which is key 0
516 * relative to the volume header's blockset.
518 * Then locate the root inode by scanning the directory keyspace
519 * represented by the label.
521 parent = hammer2_chain_lookup_init(&hmp->vchain, 0);
522 schain = hammer2_chain_lookup(&parent, &key_dummy,
523 HAMMER2_SROOT_KEY, HAMMER2_SROOT_KEY,
525 hammer2_chain_lookup_done(parent);
526 if (schain == NULL) {
527 kprintf("hammer2_mount: invalid super-root\n");
528 hammer2_vfs_unmount(mp, MNT_FORCE);
531 hammer2_chain_ref(schain); /* for hmp->schain */
532 hmp->schain = schain; /* left locked for inode_get */
533 hmp->sroot = hammer2_inode_get(NULL, NULL, schain);
534 hammer2_inode_ref(hmp->sroot); /* for hmp->sroot */
535 hammer2_inode_unlock_ex(hmp->sroot, schain);
538 mtx_init(&hmp->wthread_mtx);
539 bioq_init(&hmp->wthread_bioq);
540 hmp->wthread_destroy = 0;
545 lwkt_create(hammer2_write_thread, hmp,
546 NULL, NULL, 0, -1, "hammer2-write");
550 * Block device opened successfully, finish initializing the
553 * From this point on we have to call hammer2_unmount() on failure.
555 pmp = kmalloc(sizeof(*pmp), M_HAMMER2, M_WAITOK | M_ZERO);
556 pmp->mount_cluster = kmalloc(sizeof(hammer2_cluster_t), M_HAMMER2,
558 pmp->cluster = pmp->mount_cluster;
560 kmalloc_create(&pmp->minode, "HAMMER2-inodes");
561 kmalloc_create(&pmp->mmsg, "HAMMER2-pfsmsg");
563 pmp->mount_cluster->hmp = hmp;
564 spin_init(&pmp->inum_spin);
565 RB_INIT(&pmp->inum_tree);
567 kdmsg_iocom_init(&pmp->iocom, pmp,
568 KDMSG_IOCOMF_AUTOCONN |
569 KDMSG_IOCOMF_AUTOSPAN |
570 KDMSG_IOCOMF_AUTOCIRC,
571 pmp->mmsg, hammer2_rcvdmsg);
573 ccms_domain_init(&pmp->ccms_dom);
575 lockmgr(&hammer2_mntlk, LK_RELEASE);
576 kprintf("hammer2_mount hmp=%p pmp=%p pmpcnt=%d\n",
577 hmp, pmp, hmp->pmp_count);
579 mp->mnt_flag = MNT_LOCAL;
580 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE; /* all entry pts are SMP */
583 * required mount structure initializations
585 mp->mnt_stat.f_iosize = HAMMER2_PBUFSIZE;
586 mp->mnt_stat.f_bsize = HAMMER2_PBUFSIZE;
588 mp->mnt_vstat.f_frsize = HAMMER2_PBUFSIZE;
589 mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
594 mp->mnt_iosize_max = MAXPHYS;
595 mp->mnt_data = (qaddr_t)pmp;
599 * schain only has 1 ref now for its hmp->schain assignment.
600 * Setup for lookup (which will lock it).
602 parent = hammer2_chain_lookup_init(hmp->schain, 0);
603 lhc = hammer2_dirhash(label, strlen(label));
604 rchain = hammer2_chain_lookup(&parent, &key_next,
605 lhc, lhc + HAMMER2_DIRHASH_LOMASK,
608 if (rchain->bref.type == HAMMER2_BREF_TYPE_INODE &&
609 strcmp(label, rchain->data->ipdata.filename) == 0) {
612 rchain = hammer2_chain_next(&parent, rchain, &key_next,
614 lhc + HAMMER2_DIRHASH_LOMASK,
617 hammer2_chain_lookup_done(parent);
618 if (rchain == NULL) {
619 kprintf("hammer2_mount: PFS label not found\n");
620 hammer2_vfs_unmount(mp, MNT_FORCE);
623 if (rchain->flags & HAMMER2_CHAIN_MOUNTED) {
624 hammer2_chain_unlock(rchain);
625 kprintf("hammer2_mount: PFS label already mounted!\n");
626 hammer2_vfs_unmount(mp, MNT_FORCE);
629 if (rchain->flags & HAMMER2_CHAIN_RECYCLE) {
630 kprintf("hammer2_mount: PFS label currently recycling\n");
631 hammer2_vfs_unmount(mp, MNT_FORCE);
635 atomic_set_int(&rchain->flags, HAMMER2_CHAIN_MOUNTED);
638 * NOTE: *_get() integrates chain's lock into the inode lock.
640 hammer2_chain_ref(rchain); /* for pmp->rchain */
641 pmp->mount_cluster->rchain = rchain; /* left held & unlocked */
642 pmp->iroot = hammer2_inode_get(pmp, NULL, rchain);
643 hammer2_inode_ref(pmp->iroot); /* ref for pmp->iroot */
645 KKASSERT(rchain->pmp == NULL); /* tracking pmp for rchain */
647 atomic_add_long(&pmp->inmem_chains, 1);
649 hammer2_inode_unlock_ex(pmp->iroot, rchain);
651 kprintf("iroot %p\n", pmp->iroot);
654 * Ref the cluster management messaging descriptor. The mount
655 * program deals with the other end of the communications pipe.
657 fp = holdfp(curproc->p_fd, info.cluster_fd, -1);
659 kprintf("hammer2_mount: bad cluster_fd!\n");
660 hammer2_vfs_unmount(mp, MNT_FORCE);
663 hammer2_cluster_reconnect(pmp, fp);
669 vfs_add_vnodeops(mp, &hammer2_vnode_vops, &mp->mnt_vn_norm_ops);
670 vfs_add_vnodeops(mp, &hammer2_spec_vops, &mp->mnt_vn_spec_ops);
671 vfs_add_vnodeops(mp, &hammer2_fifo_vops, &mp->mnt_vn_fifo_ops);
673 copyinstr(info.volume, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
674 bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
675 bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
676 copyinstr(path, mp->mnt_stat.f_mntonname,
677 sizeof(mp->mnt_stat.f_mntonname) - 1,
681 * Initial statfs to prime mnt_stat.
683 hammer2_vfs_statfs(mp, &mp->mnt_stat, cred);
689 * Handle bioq for strategy write
693 hammer2_write_thread(void *arg)
695 hammer2_mount_t* hmp;
698 hammer2_trans_t trans;
700 hammer2_inode_t *last_ip;
702 hammer2_chain_t *parent;
703 hammer2_chain_t **parentp;
704 hammer2_inode_data_t *ipdata;
712 mtx_lock(&hmp->wthread_mtx);
713 while (hmp->wthread_destroy == 0) {
714 if (bioq_first(&hmp->wthread_bioq) == NULL) {
715 mtxsleep(&hmp->wthread_bioq, &hmp->wthread_mtx,
722 while ((bio = bioq_takefirst(&hmp->wthread_bioq)) != NULL) {
723 mtx_unlock(&hmp->wthread_mtx);
731 * Cache transaction for multi-buffer flush efficiency.
732 * Lock the ip separately for each buffer to allow
733 * interleaving with frontend writes.
737 hammer2_trans_done(&trans);
738 hammer2_trans_init(&trans, ip->pmp,
739 HAMMER2_TRANS_BUFCACHE);
742 parent = hammer2_inode_lock_ex(ip);
745 * Inode is modified, flush size and mtime changes
746 * to ensure that the file size remains consistent
747 * with the buffers being flushed.
749 if (ip->flags & (HAMMER2_INODE_RESIZED |
750 HAMMER2_INODE_MTIME)) {
751 hammer2_inode_fsync(&trans, ip, parentp);
753 ipdata = hammer2_chain_modify_ip(&trans, ip,
755 lblksize = hammer2_calc_logical(ip, bio->bio_offset,
757 pblksize = hammer2_calc_physical(ip, lbase);
758 hammer2_write_file_core(bp, &trans, ip, ipdata,
762 hammer2_inode_unlock_ex(ip, parent);
764 kprintf("hammer2: error in buffer write\n");
765 bp->b_flags |= B_ERROR;
769 mtx_lock(&hmp->wthread_mtx);
773 * Clean out transaction cache
776 hammer2_trans_done(&trans);
778 hmp->wthread_destroy = -1;
779 wakeup(&hmp->wthread_destroy);
781 mtx_unlock(&hmp->wthread_mtx);
785 * Return a chain suitable for I/O, creating the chain if necessary
786 * and assigning its physical block.
790 hammer2_assign_physical(hammer2_trans_t *trans,
791 hammer2_inode_t *ip, hammer2_chain_t **parentp,
792 hammer2_key_t lbase, int pblksize, int *errorp)
794 hammer2_chain_t *parent;
795 hammer2_chain_t *chain;
797 hammer2_key_t key_dummy;
798 int pradix = hammer2_getradix(pblksize);
799 int cache_index = -1;
802 * Locate the chain associated with lbase, return a locked chain.
803 * However, do not instantiate any data reference (which utilizes a
804 * device buffer) because we will be using direct IO via the
805 * logical buffer cache buffer.
808 KKASSERT(pblksize >= HAMMER2_MIN_ALLOC);
811 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS); /* extra lock */
812 chain = hammer2_chain_lookup(&parent, &key_dummy,
814 &cache_index, HAMMER2_LOOKUP_NODATA);
818 * We found a hole, create a new chain entry.
820 * NOTE: DATA chains are created without device backing
821 * store (nor do we want any).
823 *errorp = hammer2_chain_create(trans, &parent, &chain,
824 lbase, HAMMER2_PBUFRADIX,
825 HAMMER2_BREF_TYPE_DATA,
828 hammer2_chain_lookup_done(parent);
829 panic("hammer2_chain_create: par=%p error=%d\n",
834 pbase = chain->bref.data_off & ~HAMMER2_OFF_MASK_RADIX;
835 /*ip->delta_dcount += pblksize;*/
837 switch (chain->bref.type) {
838 case HAMMER2_BREF_TYPE_INODE:
840 * The data is embedded in the inode. The
841 * caller is responsible for marking the inode
842 * modified and copying the data to the embedded
847 case HAMMER2_BREF_TYPE_DATA:
848 if (chain->bytes != pblksize) {
849 hammer2_chain_resize(trans, ip,
852 HAMMER2_MODIFY_OPTDATA);
854 hammer2_chain_modify(trans, &chain,
855 HAMMER2_MODIFY_OPTDATA);
856 pbase = chain->bref.data_off & ~HAMMER2_OFF_MASK_RADIX;
859 panic("hammer2_assign_physical: bad type");
867 * Cleanup. If chain wound up being the inode (i.e. DIRECTDATA),
868 * we might have to replace *parentp.
870 hammer2_chain_lookup_done(parent);
872 if (*parentp != chain &&
873 (*parentp)->core == chain->core) {
875 *parentp = chain; /* eats lock */
876 hammer2_chain_unlock(parent);
877 hammer2_chain_lock(chain, 0); /* need another */
879 /* else chain already locked for return */
885 * From hammer2_vnops.c.
886 * The core write function which determines which path to take
887 * depending on compression settings.
891 hammer2_write_file_core(struct buf *bp, hammer2_trans_t *trans,
892 hammer2_inode_t *ip, hammer2_inode_data_t *ipdata,
893 hammer2_chain_t **parentp,
894 hammer2_key_t lbase, int ioflag, int pblksize,
897 hammer2_chain_t *chain;
899 switch(HAMMER2_DEC_COMP(ipdata->comp_algo)) {
900 case HAMMER2_COMP_NONE:
902 * We have to assign physical storage to the buffer
903 * we intend to dirty or write now to avoid deadlocks
904 * in the strategy code later.
906 * This can return NOOFFSET for inode-embedded data.
907 * The strategy code will take care of it in that case.
909 chain = hammer2_assign_physical(trans, ip, parentp,
912 hammer2_write_bp(chain, bp, ioflag, pblksize, errorp);
914 hammer2_chain_unlock(chain);
916 case HAMMER2_COMP_AUTOZERO:
918 * Check for zero-fill only
920 hammer2_zero_check_and_write(bp, trans, ip,
921 ipdata, parentp, lbase,
922 ioflag, pblksize, errorp);
924 case HAMMER2_COMP_LZ4:
925 case HAMMER2_COMP_ZLIB:
928 * Check for zero-fill and attempt compression.
930 hammer2_compress_and_write(bp, trans, ip,
937 ipdata = &ip->chain->data->ipdata; /* reload */
941 * From hammer2_vnops.c
942 * Generic function that will perform the compression in compression
943 * write path. The compression algorithm is determined by the settings
944 * obtained from inode.
948 hammer2_compress_and_write(struct buf *bp, hammer2_trans_t *trans,
949 hammer2_inode_t *ip, hammer2_inode_data_t *ipdata,
950 hammer2_chain_t **parentp,
951 hammer2_key_t lbase, int ioflag, int pblksize,
952 int *errorp, int comp_algo)
954 hammer2_chain_t *chain;
959 if (test_block_zeros(bp->b_data, pblksize)) {
960 zero_write(bp, trans, ip, ipdata, parentp, lbase, errorp);
967 KKASSERT(pblksize / 2 <= 32768);
969 if (ip->comp_heuristic < 8 || (ip->comp_heuristic & 7) == 0) {
970 z_stream strm_compress;
974 switch(HAMMER2_DEC_COMP(comp_algo)) {
975 case HAMMER2_COMP_LZ4:
976 comp_buffer = objcache_get(cache_buffer_write,
978 comp_size = LZ4_compress_limitedOutput(
980 &comp_buffer[sizeof(int)],
982 pblksize / 2 - sizeof(int));
984 * We need to prefix with the size, LZ4
985 * doesn't do it for us. Add the related
988 *(int *)comp_buffer = comp_size;
990 comp_size += sizeof(int);
992 case HAMMER2_COMP_ZLIB:
993 comp_level = HAMMER2_DEC_LEVEL(comp_algo);
995 comp_level = 6; /* default zlib compression */
996 else if (comp_level < 6)
998 else if (comp_level > 9)
1000 ret = deflateInit(&strm_compress, comp_level);
1002 kprintf("HAMMER2 ZLIB: fatal error "
1003 "on deflateInit.\n");
1006 comp_buffer = objcache_get(cache_buffer_write,
1008 strm_compress.next_in = bp->b_data;
1009 strm_compress.avail_in = pblksize;
1010 strm_compress.next_out = comp_buffer;
1011 strm_compress.avail_out = pblksize / 2;
1012 ret = deflate(&strm_compress, Z_FINISH);
1013 if (ret == Z_STREAM_END) {
1014 comp_size = pblksize / 2 -
1015 strm_compress.avail_out;
1019 ret = deflateEnd(&strm_compress);
1022 kprintf("Error: Unknown compression method.\n");
1023 kprintf("Comp_method = %d.\n", comp_algo);
1028 if (comp_size == 0) {
1030 * compression failed or turned off
1032 comp_block_size = pblksize; /* safety */
1033 if (++ip->comp_heuristic > 128)
1034 ip->comp_heuristic = 8;
1037 * compression succeeded
1039 ip->comp_heuristic = 0;
1040 if (comp_size <= 1024) {
1041 comp_block_size = 1024;
1042 } else if (comp_size <= 2048) {
1043 comp_block_size = 2048;
1044 } else if (comp_size <= 4096) {
1045 comp_block_size = 4096;
1046 } else if (comp_size <= 8192) {
1047 comp_block_size = 8192;
1048 } else if (comp_size <= 16384) {
1049 comp_block_size = 16384;
1050 } else if (comp_size <= 32768) {
1051 comp_block_size = 32768;
1053 panic("hammer2: WRITE PATH: "
1054 "Weird comp_size value.");
1056 comp_block_size = pblksize;
1060 chain = hammer2_assign_physical(trans, ip, parentp,
1061 lbase, comp_block_size,
1063 ipdata = &ip->chain->data->ipdata; /* RELOAD */
1066 kprintf("WRITE PATH: An error occurred while "
1067 "assigning physical space.\n");
1068 KKASSERT(chain == NULL);
1070 /* Get device offset */
1071 hammer2_off_t pbase;
1072 hammer2_off_t pmask;
1079 KKASSERT(chain->flags & HAMMER2_CHAIN_MODIFIED);
1081 switch(chain->bref.type) {
1082 case HAMMER2_BREF_TYPE_INODE:
1083 KKASSERT(chain->data->ipdata.op_flags &
1084 HAMMER2_OPFLAG_DIRECTDATA);
1085 KKASSERT(bp->b_loffset == 0);
1086 bcopy(bp->b_data, chain->data->ipdata.u.data,
1087 HAMMER2_EMBEDDED_BYTES);
1089 case HAMMER2_BREF_TYPE_DATA:
1090 psize = hammer2_devblksize(chain->bytes);
1091 pmask = (hammer2_off_t)psize - 1;
1092 pbase = chain->bref.data_off & ~pmask;
1093 boff = chain->bref.data_off &
1094 (HAMMER2_OFF_MASK & pmask);
1095 peof = (pbase + HAMMER2_SEGMASK64) &
1097 temp_check = HAMMER2_DEC_CHECK(chain->bref.methods);
1100 * Optimize out the read-before-write
1103 if (comp_block_size == psize) {
1104 dbp = getblk(chain->hmp->devvp, pbase,
1107 *errorp = bread(chain->hmp->devvp,
1108 pbase, psize, &dbp);
1110 kprintf("hammer2: WRITE PATH: "
1111 "dbp bread error\n");
1117 * When loading the block make sure we don't
1118 * leave garbage after the compressed data.
1121 chain->bref.methods =
1122 HAMMER2_ENC_COMP(comp_algo) +
1123 HAMMER2_ENC_CHECK(temp_check);
1124 bcopy(comp_buffer, dbp->b_data + boff,
1126 if (comp_size != comp_block_size) {
1127 bzero(dbp->b_data + boff +
1133 chain->bref.methods =
1135 HAMMER2_COMP_NONE) +
1136 HAMMER2_ENC_CHECK(temp_check);
1137 bcopy(bp->b_data, dbp->b_data + boff,
1142 * Device buffer is now valid, chain is no
1143 * longer in the initial state.
1145 atomic_clear_int(&chain->flags,
1146 HAMMER2_CHAIN_INITIAL);
1148 /* Now write the related bdp. */
1149 if (ioflag & IO_SYNC) {
1151 * Synchronous I/O requested.
1155 } else if ((ioflag & IO_DIRECT) &&
1156 loff + n == pblksize) {
1159 } else if (ioflag & IO_ASYNC) {
1161 } else if (hammer2_cluster_enable) {
1162 cluster_write(dbp, peof,
1170 panic("hammer2_write_bp: bad chain type %d\n",
1176 hammer2_chain_unlock(chain);
1179 objcache_put(cache_buffer_write, comp_buffer);
1183 * Function that performs zero-checking and writing without compression,
1184 * it corresponds to default zero-checking path.
1188 hammer2_zero_check_and_write(struct buf *bp, hammer2_trans_t *trans,
1189 hammer2_inode_t *ip, hammer2_inode_data_t *ipdata,
1190 hammer2_chain_t **parentp,
1191 hammer2_key_t lbase, int ioflag, int pblksize, int *errorp)
1193 hammer2_chain_t *chain;
1195 if (test_block_zeros(bp->b_data, pblksize)) {
1196 zero_write(bp, trans, ip, ipdata, parentp, lbase, errorp);
1198 chain = hammer2_assign_physical(trans, ip, parentp,
1199 lbase, pblksize, errorp);
1200 hammer2_write_bp(chain, bp, ioflag, pblksize, errorp);
1202 hammer2_chain_unlock(chain);
1207 * A function to test whether a block of data contains only zeros,
1208 * returns TRUE (non-zero) if the block is all zeros.
1212 test_block_zeros(const char *buf, size_t bytes)
1216 for (i = 0; i < bytes; i += sizeof(long)) {
1217 if (*(const long *)(buf + i) != 0)
1224 * Function to "write" a block that contains only zeros.
1228 zero_write(struct buf *bp, hammer2_trans_t *trans, hammer2_inode_t *ip,
1229 hammer2_inode_data_t *ipdata, hammer2_chain_t **parentp,
1230 hammer2_key_t lbase, int *errorp __unused)
1232 hammer2_chain_t *parent;
1233 hammer2_chain_t *chain;
1234 hammer2_key_t key_dummy;
1235 int cache_index = -1;
1237 parent = hammer2_chain_lookup_init(*parentp, 0);
1239 chain = hammer2_chain_lookup(&parent, &key_dummy, lbase, lbase,
1240 &cache_index, HAMMER2_LOOKUP_NODATA);
1242 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE) {
1243 bzero(chain->data->ipdata.u.data,
1244 HAMMER2_EMBEDDED_BYTES);
1246 hammer2_chain_delete(trans, chain, 0);
1248 hammer2_chain_unlock(chain);
1250 hammer2_chain_lookup_done(parent);
1254 * Function to write the data as it is, without performing any sort of
1255 * compression. This function is used in path without compression and
1256 * default zero-checking path.
1260 hammer2_write_bp(hammer2_chain_t *chain, struct buf *bp, int ioflag,
1261 int pblksize, int *errorp)
1263 hammer2_off_t pbase;
1264 hammer2_off_t pmask;
1270 int temp_check = HAMMER2_DEC_CHECK(chain->bref.methods);
1272 KKASSERT(chain->flags & HAMMER2_CHAIN_MODIFIED);
1274 switch(chain->bref.type) {
1275 case HAMMER2_BREF_TYPE_INODE:
1276 KKASSERT(chain->data->ipdata.op_flags &
1277 HAMMER2_OPFLAG_DIRECTDATA);
1278 KKASSERT(bp->b_loffset == 0);
1279 bcopy(bp->b_data, chain->data->ipdata.u.data,
1280 HAMMER2_EMBEDDED_BYTES);
1283 case HAMMER2_BREF_TYPE_DATA:
1284 psize = hammer2_devblksize(chain->bytes);
1285 pmask = (hammer2_off_t)psize - 1;
1286 pbase = chain->bref.data_off & ~pmask;
1287 boff = chain->bref.data_off & (HAMMER2_OFF_MASK & pmask);
1288 peof = (pbase + HAMMER2_SEGMASK64) & ~HAMMER2_SEGMASK64;
1290 if (psize == pblksize) {
1291 dbp = getblk(chain->hmp->devvp, pbase,
1295 error = bread(chain->hmp->devvp, pbase, psize, &dbp);
1297 kprintf("hammer2: WRITE PATH: "
1298 "dbp bread error\n");
1303 chain->bref.methods = HAMMER2_ENC_COMP(HAMMER2_COMP_NONE) +
1304 HAMMER2_ENC_CHECK(temp_check);
1305 bcopy(bp->b_data, dbp->b_data + boff, chain->bytes);
1308 * Device buffer is now valid, chain is no
1309 * longer in the initial state.
1311 atomic_clear_int(&chain->flags, HAMMER2_CHAIN_INITIAL);
1313 if (ioflag & IO_SYNC) {
1315 * Synchronous I/O requested.
1319 } else if ((ioflag & IO_DIRECT) && loff + n == pblksize) {
1322 } else if (ioflag & IO_ASYNC) {
1324 } else if (hammer2_cluster_enable) {
1325 cluster_write(dbp, peof, HAMMER2_PBUFSIZE, 4/*XXX*/);
1331 panic("hammer2_write_bp: bad chain type %d\n",
1342 hammer2_remount(struct mount *mp, char *path, struct vnode *devvp,
1350 hammer2_vfs_unmount(struct mount *mp, int mntflags)
1352 hammer2_pfsmount_t *pmp;
1353 hammer2_mount_t *hmp;
1354 hammer2_cluster_t *cluster;
1357 int ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
1359 struct vnode *devvp;
1362 cluster = pmp->mount_cluster;
1366 if (mntflags & MNT_FORCE)
1367 flags |= FORCECLOSE;
1369 hammer2_mount_exlock(hmp);
1372 * If mount initialization proceeded far enough we must flush
1376 error = vflush(mp, 0, flags);
1379 hammer2_mount_unlock(hmp);
1383 lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
1385 kprintf("hammer2_unmount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
1388 * Flush any left over chains. The voldata lock is only used
1389 * to synchronize against HAMMER2_CHAIN_MODIFIED_AUX.
1391 hammer2_voldata_lock(hmp);
1392 if ((hmp->vchain.flags | hmp->fchain.flags) &
1393 (HAMMER2_CHAIN_MODIFIED | HAMMER2_CHAIN_SUBMODIFIED)) {
1394 hammer2_voldata_unlock(hmp, 0);
1395 hammer2_vfs_sync(mp, MNT_WAIT);
1396 hammer2_vfs_sync(mp, MNT_WAIT);
1398 hammer2_voldata_unlock(hmp, 0);
1400 if (hmp->pmp_count == 0) {
1401 if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
1402 HAMMER2_CHAIN_SUBMODIFIED)) {
1403 kprintf("hammer2_unmount: chains left over after "
1405 if (hammer2_debug & 0x0010)
1406 Debugger("entered debugger");
1411 * Cleanup the root and super-root chain elements (which should be
1415 #if REPORT_REFS_ERRORS
1416 if (pmp->iroot->refs != 1)
1417 kprintf("PMP->IROOT %p REFS WRONG %d\n",
1418 pmp->iroot, pmp->iroot->refs);
1420 KKASSERT(pmp->iroot->refs == 1);
1422 hammer2_inode_drop(pmp->iroot); /* ref for pmp->iroot */
1425 if (cluster->rchain) {
1426 atomic_clear_int(&cluster->rchain->flags,
1427 HAMMER2_CHAIN_MOUNTED);
1428 #if REPORT_REFS_ERRORS
1429 if (cluster->rchain->refs != 1)
1430 kprintf("PMP->RCHAIN %p REFS WRONG %d\n",
1431 cluster->rchain, cluster->rchain->refs);
1433 KKASSERT(cluster->rchain->refs == 1);
1435 hammer2_chain_drop(cluster->rchain);
1436 cluster->rchain = NULL;
1438 ccms_domain_uninit(&pmp->ccms_dom);
1441 * Kill cluster controller
1443 kdmsg_iocom_uninit(&pmp->iocom);
1446 * If no PFS's left drop the master hammer2_mount for the device.
1448 if (hmp->pmp_count == 0) {
1450 hammer2_inode_drop(hmp->sroot);
1454 #if REPORT_REFS_ERRORS
1455 if (hmp->schain->refs != 1)
1456 kprintf("HMP->SCHAIN %p REFS WRONG %d\n",
1457 hmp->schain, hmp->schain->refs);
1459 KKASSERT(hmp->schain->refs == 1);
1461 hammer2_chain_drop(hmp->schain);
1466 * Finish up with the device vnode
1468 if ((devvp = hmp->devvp) != NULL) {
1469 vinvalbuf(devvp, (ronly ? 0 : V_SAVE), 0, 0);
1471 VOP_CLOSE(devvp, (ronly ? FREAD : FREAD|FWRITE));
1477 * Final drop of embedded freemap root chain to clean up
1478 * fchain.core (fchain structure is not flagged ALLOCATED
1479 * so it is cleaned out and then left to rot).
1481 hammer2_chain_drop(&hmp->fchain);
1484 * Final drop of embedded volume root chain to clean up
1485 * vchain.core (vchain structure is not flagged ALLOCATED
1486 * so it is cleaned out and then left to rot).
1489 hammer2_dump_chain(&hmp->vchain, 0, &dumpcnt);
1490 hammer2_mount_unlock(hmp);
1491 hammer2_chain_drop(&hmp->vchain);
1493 hammer2_mount_unlock(hmp);
1497 mp->mnt_data = NULL;
1499 pmp->mount_cluster = NULL;
1500 pmp->cluster = NULL; /* XXX */
1502 kmalloc_destroy(&pmp->mmsg);
1503 kmalloc_destroy(&pmp->minode);
1505 cluster->hmp = NULL;
1507 kfree(cluster, M_HAMMER2);
1508 kfree(pmp, M_HAMMER2);
1509 if (hmp->pmp_count == 0) {
1510 mtx_lock(&hmp->wthread_mtx);
1511 hmp->wthread_destroy = 1;
1512 wakeup(&hmp->wthread_bioq);
1513 while (hmp->wthread_destroy != -1) {
1514 mtxsleep(&hmp->wthread_destroy, &hmp->wthread_mtx, 0,
1517 mtx_unlock(&hmp->wthread_mtx);
1519 TAILQ_REMOVE(&hammer2_mntlist, hmp, mntentry);
1520 kmalloc_destroy(&hmp->mchain);
1521 kfree(hmp, M_HAMMER2);
1523 lockmgr(&hammer2_mntlk, LK_RELEASE);
1530 hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
1531 ino_t ino, struct vnode **vpp)
1533 kprintf("hammer2_vget\n");
1534 return (EOPNOTSUPP);
1539 hammer2_vfs_root(struct mount *mp, struct vnode **vpp)
1541 hammer2_pfsmount_t *pmp;
1542 hammer2_chain_t *parent;
1547 if (pmp->iroot == NULL) {
1551 parent = hammer2_inode_lock_sh(pmp->iroot);
1552 vp = hammer2_igetv(pmp->iroot, &error);
1553 hammer2_inode_unlock_sh(pmp->iroot, parent);
1556 kprintf("vnodefail\n");
1565 * XXX incorporate ipdata->inode_quota and data_quota
1569 hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
1571 hammer2_pfsmount_t *pmp;
1572 hammer2_mount_t *hmp;
1577 mp->mnt_stat.f_files = pmp->inode_count;
1578 mp->mnt_stat.f_ffree = 0;
1579 mp->mnt_stat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
1580 mp->mnt_stat.f_bfree = hmp->voldata.allocator_free / HAMMER2_PBUFSIZE;
1581 mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
1583 *sbp = mp->mnt_stat;
1589 hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
1591 hammer2_pfsmount_t *pmp;
1592 hammer2_mount_t *hmp;
1597 mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
1598 mp->mnt_vstat.f_files = pmp->inode_count;
1599 mp->mnt_vstat.f_ffree = 0;
1600 mp->mnt_vstat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
1601 mp->mnt_vstat.f_bfree = hmp->voldata.allocator_free / HAMMER2_PBUFSIZE;
1602 mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
1604 *sbp = mp->mnt_vstat;
1609 * Sync the entire filesystem; this is called from the filesystem syncer
1610 * process periodically and whenever a user calls sync(1) on the hammer
1613 * Currently is actually called from the syncer! \o/
1615 * This task will have to snapshot the state of the dirty inode chain.
1616 * From that, it will have to make sure all of the inodes on the dirty
1617 * chain have IO initiated. We make sure that io is initiated for the root
1620 * If waitfor is set, we wait for media to acknowledge the new rootblock.
1622 * THINKS: side A vs side B, to have sync not stall all I/O?
1626 hammer2_vfs_sync(struct mount *mp, int waitfor)
1628 struct hammer2_sync_info info;
1629 hammer2_pfsmount_t *pmp;
1630 hammer2_cluster_t *cluster;
1631 hammer2_mount_t *hmp;
1639 * We can't acquire locks on existing vnodes while in a transaction
1640 * without risking a deadlock. This assumes that vfsync() can be
1641 * called without the vnode locked (which it can in DragonFly).
1642 * Otherwise we'd have to implement a multi-pass or flag the lock
1643 * failures and retry.
1645 /*flags = VMSC_GETVP;*/
1647 if (waitfor & MNT_LAZY)
1648 flags |= VMSC_ONEPASS;
1650 hammer2_trans_init(&info.trans, pmp, HAMMER2_TRANS_ISFLUSH);
1653 info.waitfor = MNT_NOWAIT;
1654 vmntvnodescan(mp, flags | VMSC_NOWAIT,
1656 hammer2_sync_scan2, &info);
1657 if (info.error == 0 && (waitfor & MNT_WAIT)) {
1658 info.waitfor = waitfor;
1659 vmntvnodescan(mp, flags,
1661 hammer2_sync_scan2, &info);
1665 if (waitfor == MNT_WAIT) {
1672 cluster = pmp->cluster;
1675 hammer2_chain_lock(&hmp->vchain, HAMMER2_RESOLVE_ALWAYS);
1676 if (hmp->vchain.flags & (HAMMER2_CHAIN_MODIFIED |
1677 HAMMER2_CHAIN_SUBMODIFIED)) {
1678 hammer2_chain_flush(&info.trans, &hmp->vchain);
1680 hammer2_chain_unlock(&hmp->vchain);
1684 * Rollup flush. The fsyncs above basically just flushed
1685 * data blocks. The flush below gets all the meta-data.
1687 hammer2_chain_lock(&hmp->fchain, HAMMER2_RESOLVE_ALWAYS);
1688 if (hmp->fchain.flags & (HAMMER2_CHAIN_MODIFIED |
1689 HAMMER2_CHAIN_SUBMODIFIED)) {
1690 /* this will modify vchain as a side effect */
1691 hammer2_chain_flush(&info.trans, &hmp->fchain);
1693 hammer2_chain_unlock(&hmp->fchain);
1700 * We can't safely flush the volume header until we have
1701 * flushed any device buffers which have built up.
1703 * XXX this isn't being incremental
1705 vn_lock(hmp->devvp, LK_EXCLUSIVE | LK_RETRY);
1706 error = VOP_FSYNC(hmp->devvp, MNT_WAIT, 0);
1707 vn_unlock(hmp->devvp);
1710 * The flush code sets CHAIN_VOLUMESYNC to indicate that the
1711 * volume header needs synchronization via hmp->volsync.
1713 * XXX synchronize the flag & data with only this flush XXX
1715 if (error == 0 && (hmp->vchain.flags & HAMMER2_CHAIN_VOLUMESYNC)) {
1719 * Synchronize the disk before flushing the volume
1723 bp->b_bio1.bio_offset = 0;
1726 bp->b_cmd = BUF_CMD_FLUSH;
1727 bp->b_bio1.bio_done = biodone_sync;
1728 bp->b_bio1.bio_flags |= BIO_SYNC;
1729 vn_strategy(hmp->devvp, &bp->b_bio1);
1730 biowait(&bp->b_bio1, "h2vol");
1734 * Then we can safely flush the version of the volume header
1735 * synchronized by the flush code.
1737 i = hmp->volhdrno + 1;
1738 if (i >= HAMMER2_NUM_VOLHDRS)
1740 if (i * HAMMER2_ZONE_BYTES64 + HAMMER2_SEGSIZE >
1741 hmp->volsync.volu_size) {
1744 kprintf("sync volhdr %d %jd\n",
1745 i, (intmax_t)hmp->volsync.volu_size);
1746 bp = getblk(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
1747 HAMMER2_PBUFSIZE, 0, 0);
1748 atomic_clear_int(&hmp->vchain.flags, HAMMER2_CHAIN_VOLUMESYNC);
1749 bcopy(&hmp->volsync, bp->b_data, HAMMER2_PBUFSIZE);
1753 hammer2_trans_done(&info.trans);
1760 * NOTE: We don't test SUBMODIFIED or MOVED here because the fsync code
1761 * won't flush on those flags. The syncer code above will do a
1762 * general meta-data flush globally that will catch these flags.
1765 hammer2_sync_scan1(struct mount *mp, struct vnode *vp, void *data)
1767 hammer2_inode_t *ip;
1770 if (vp->v_type == VNON || ip == NULL ||
1771 ((ip->flags & HAMMER2_INODE_MODIFIED) == 0 &&
1772 RB_EMPTY(&vp->v_rbdirty_tree))) {
1779 hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
1781 struct hammer2_sync_info *info = data;
1782 hammer2_inode_t *ip;
1783 hammer2_chain_t *parent;
1787 if (vp->v_type == VNON || vp->v_type == VBAD ||
1788 ((ip->flags & HAMMER2_INODE_MODIFIED) == 0 &&
1789 RB_EMPTY(&vp->v_rbdirty_tree))) {
1794 * VOP_FSYNC will start a new transaction so replicate some code
1795 * here to do it inline (see hammer2_vop_fsync()).
1797 * WARNING: The vfsync interacts with the buffer cache and might
1798 * block, we can't hold the inode lock at that time.
1800 atomic_clear_int(&ip->flags, HAMMER2_INODE_MODIFIED);
1802 vfsync(ip->vp, MNT_NOWAIT, 1, NULL, NULL);
1803 parent = hammer2_inode_lock_ex(ip);
1804 hammer2_chain_flush(&info->trans, parent);
1805 hammer2_inode_unlock_ex(ip, parent);
1808 error = VOP_FSYNC(vp, MNT_NOWAIT, 0);
1811 info->error = error;
1817 hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp)
1824 hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
1825 struct fid *fhp, struct vnode **vpp)
1832 hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
1833 int *exflagsp, struct ucred **credanonp)
1839 * Support code for hammer2_mount(). Read, verify, and install the volume
1840 * header into the HMP
1842 * XXX read four volhdrs and use the one with the highest TID whos CRC
1847 * XXX For filesystems w/ less than 4 volhdrs, make sure to not write to
1848 * nonexistant locations.
1850 * XXX Record selected volhdr and ring updates to each of 4 volhdrs
1854 hammer2_install_volume_header(hammer2_mount_t *hmp)
1856 hammer2_volume_data_t *vd;
1858 hammer2_crc32_t crc0, crc, bcrc0, bcrc;
1870 * There are up to 4 copies of the volume header (syncs iterate
1871 * between them so there is no single master). We don't trust the
1872 * volu_size field so we don't know precisely how large the filesystem
1873 * is, so depend on the OS to return an error if we go beyond the
1874 * block device's EOF.
1876 for (i = 0; i < HAMMER2_NUM_VOLHDRS; i++) {
1877 error = bread(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
1878 HAMMER2_VOLUME_BYTES, &bp);
1885 vd = (struct hammer2_volume_data *) bp->b_data;
1886 if ((vd->magic != HAMMER2_VOLUME_ID_HBO) &&
1887 (vd->magic != HAMMER2_VOLUME_ID_ABO)) {
1893 if (vd->magic == HAMMER2_VOLUME_ID_ABO) {
1894 /* XXX: Reversed-endianness filesystem */
1895 kprintf("hammer2: reverse-endian filesystem detected");
1901 crc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT0];
1902 crc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC0_OFF,
1903 HAMMER2_VOLUME_ICRC0_SIZE);
1904 bcrc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT1];
1905 bcrc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC1_OFF,
1906 HAMMER2_VOLUME_ICRC1_SIZE);
1907 if ((crc0 != crc) || (bcrc0 != bcrc)) {
1908 kprintf("hammer2 volume header crc "
1909 "mismatch copy #%d %08x/%08x\n",
1916 if (valid == 0 || hmp->voldata.mirror_tid < vd->mirror_tid) {
1925 hmp->volsync = hmp->voldata;
1927 if (error_reported || bootverbose || 1) { /* 1/DEBUG */
1928 kprintf("hammer2: using volume header #%d\n",
1933 kprintf("hammer2: no valid volume headers found!\n");
1939 * Reconnect using the passed file pointer. The caller must ref the
1943 hammer2_cluster_reconnect(hammer2_pfsmount_t *pmp, struct file *fp)
1945 hammer2_inode_data_t *ipdata;
1946 hammer2_chain_t *parent;
1947 hammer2_mount_t *hmp;
1950 hmp = pmp->mount_cluster->hmp;
1953 * Closes old comm descriptor, kills threads, cleans up
1954 * states, then installs the new descriptor and creates
1957 kdmsg_iocom_reconnect(&pmp->iocom, fp, "hammer2");
1960 * Setup LNK_CONN fields for autoinitiated state machine
1962 parent = hammer2_inode_lock_ex(pmp->iroot);
1963 ipdata = &parent->data->ipdata;
1964 pmp->iocom.auto_lnk_conn.pfs_clid = ipdata->pfs_clid;
1965 pmp->iocom.auto_lnk_conn.pfs_fsid = ipdata->pfs_fsid;
1966 pmp->iocom.auto_lnk_conn.pfs_type = ipdata->pfs_type;
1967 pmp->iocom.auto_lnk_conn.proto_version = DMSG_SPAN_PROTO_1;
1968 pmp->iocom.auto_lnk_conn.peer_type = hmp->voldata.peer_type;
1971 * Filter adjustment. Clients do not need visibility into other
1972 * clients (otherwise millions of clients would present a serious
1973 * problem). The fs_label also serves to restrict the namespace.
1975 pmp->iocom.auto_lnk_conn.peer_mask = 1LLU << HAMMER2_PEER_HAMMER2;
1976 pmp->iocom.auto_lnk_conn.pfs_mask = (uint64_t)-1;
1977 switch (ipdata->pfs_type) {
1978 case DMSG_PFSTYPE_CLIENT:
1979 pmp->iocom.auto_lnk_conn.peer_mask &=
1980 ~(1LLU << DMSG_PFSTYPE_CLIENT);
1986 name_len = ipdata->name_len;
1987 if (name_len >= sizeof(pmp->iocom.auto_lnk_conn.fs_label))
1988 name_len = sizeof(pmp->iocom.auto_lnk_conn.fs_label) - 1;
1989 bcopy(ipdata->filename,
1990 pmp->iocom.auto_lnk_conn.fs_label,
1992 pmp->iocom.auto_lnk_conn.fs_label[name_len] = 0;
1995 * Setup LNK_SPAN fields for autoinitiated state machine
1997 pmp->iocom.auto_lnk_span.pfs_clid = ipdata->pfs_clid;
1998 pmp->iocom.auto_lnk_span.pfs_fsid = ipdata->pfs_fsid;
1999 pmp->iocom.auto_lnk_span.pfs_type = ipdata->pfs_type;
2000 pmp->iocom.auto_lnk_span.peer_type = hmp->voldata.peer_type;
2001 pmp->iocom.auto_lnk_span.proto_version = DMSG_SPAN_PROTO_1;
2002 name_len = ipdata->name_len;
2003 if (name_len >= sizeof(pmp->iocom.auto_lnk_span.fs_label))
2004 name_len = sizeof(pmp->iocom.auto_lnk_span.fs_label) - 1;
2005 bcopy(ipdata->filename,
2006 pmp->iocom.auto_lnk_span.fs_label,
2008 pmp->iocom.auto_lnk_span.fs_label[name_len] = 0;
2009 hammer2_inode_unlock_ex(pmp->iroot, parent);
2011 kdmsg_iocom_autoinitiate(&pmp->iocom, hammer2_autodmsg);
2015 hammer2_rcvdmsg(kdmsg_msg_t *msg)
2017 switch(msg->any.head.cmd & DMSGF_TRANSMASK) {
2018 case DMSG_DBG_SHELL:
2021 * Execute shell command (not supported atm)
2023 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
2025 case DMSG_DBG_SHELL | DMSGF_REPLY:
2029 if (msg->aux_data) {
2030 msg->aux_data[msg->aux_size - 1] = 0;
2031 kprintf("HAMMER2 DBG: %s\n", msg->aux_data);
2036 * Unsupported message received. We only need to
2037 * reply if it's a transaction in order to close our end.
2038 * Ignore any one-way messages are any further messages
2039 * associated with the transaction.
2041 * NOTE: This case also includes DMSG_LNK_ERROR messages
2042 * which might be one-way, replying to those would
2043 * cause an infinite ping-pong.
2045 if (msg->any.head.cmd & DMSGF_CREATE)
2046 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
2053 * This function is called after KDMSG has automatically handled processing
2054 * of a LNK layer message (typically CONN, SPAN, or CIRC).
2056 * We tag off the LNK_CONN to trigger our LNK_VOLCONF messages which
2057 * advertises all available hammer2 super-root volumes.
2060 hammer2_autodmsg(kdmsg_msg_t *msg)
2062 hammer2_pfsmount_t *pmp = msg->iocom->handle;
2063 hammer2_mount_t *hmp = pmp->mount_cluster->hmp;
2067 * We only care about replies to our LNK_CONN auto-request. kdmsg
2068 * has already processed the reply, we use this calback as a shim
2069 * to know when we can advertise available super-root volumes.
2071 if ((msg->any.head.cmd & DMSGF_TRANSMASK) !=
2072 (DMSG_LNK_CONN | DMSGF_CREATE | DMSGF_REPLY) ||
2073 msg->state == NULL) {
2077 kprintf("LNK_CONN REPLY RECEIVED CMD %08x\n", msg->any.head.cmd);
2079 if (msg->any.head.cmd & DMSGF_CREATE) {
2080 kprintf("HAMMER2: VOLDATA DUMP\n");
2083 * Dump the configuration stored in the volume header
2085 hammer2_voldata_lock(hmp);
2086 for (copyid = 0; copyid < HAMMER2_COPYID_COUNT; ++copyid) {
2087 if (hmp->voldata.copyinfo[copyid].copyid == 0)
2089 hammer2_volconf_update(pmp, copyid);
2091 hammer2_voldata_unlock(hmp, 0);
2093 if ((msg->any.head.cmd & DMSGF_DELETE) &&
2094 msg->state && (msg->state->txcmd & DMSGF_DELETE) == 0) {
2095 kprintf("HAMMER2: CONN WAS TERMINATED\n");
2100 * Volume configuration updates are passed onto the userland service
2101 * daemon via the open LNK_CONN transaction.
2104 hammer2_volconf_update(hammer2_pfsmount_t *pmp, int index)
2106 hammer2_mount_t *hmp = pmp->mount_cluster->hmp;
2109 /* XXX interlock against connection state termination */
2110 kprintf("volconf update %p\n", pmp->iocom.conn_state);
2111 if (pmp->iocom.conn_state) {
2112 kprintf("TRANSMIT VOLCONF VIA OPEN CONN TRANSACTION\n");
2113 msg = kdmsg_msg_alloc_state(pmp->iocom.conn_state,
2114 DMSG_LNK_VOLCONF, NULL, NULL);
2115 msg->any.lnk_volconf.copy = hmp->voldata.copyinfo[index];
2116 msg->any.lnk_volconf.mediaid = hmp->voldata.fsid;
2117 msg->any.lnk_volconf.index = index;
2118 kdmsg_msg_write(msg);
2123 hammer2_dump_chain(hammer2_chain_t *chain, int tab, int *countp)
2125 hammer2_chain_layer_t *layer;
2126 hammer2_chain_t *scan;
2127 hammer2_chain_t *first_parent;
2131 kprintf("%*.*s...\n", tab, tab, "");
2136 first_parent = chain->core ? TAILQ_FIRST(&chain->core->ownerq) : NULL;
2137 kprintf("%*.*schain %p.%d [%08x][core=%p fp=%p] (%s) np=%p dt=%s refs=%d",
2139 chain, chain->bref.type, chain->flags,
2142 ((chain->bref.type == HAMMER2_BREF_TYPE_INODE &&
2143 chain->data) ? (char *)chain->data->ipdata.filename : "?"),
2144 (first_parent ? TAILQ_NEXT(chain, core_entry) : NULL),
2145 (chain->delete_tid == HAMMER2_MAX_TID ? "max" : "fls"),
2148 kprintf(" [fpflags %08x fprefs %d\n",
2149 first_parent->flags,
2150 first_parent->refs);
2151 if (chain->core == NULL || TAILQ_EMPTY(&chain->core->layerq))
2155 TAILQ_FOREACH(layer, &chain->core->layerq, entry) {
2156 RB_FOREACH(scan, hammer2_chain_tree, &layer->rbtree) {
2157 hammer2_dump_chain(scan, tab + 4, countp);
2160 if (chain->core && !TAILQ_EMPTY(&chain->core->layerq)) {
2161 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE && chain->data)
2162 kprintf("%*.*s}(%s)\n", tab, tab, "",
2163 chain->data->ipdata.filename);
2165 kprintf("%*.*s}\n", tab, tab, "");