2 * Copyright (c) 2011-2014 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 TAILQ_HEAD(hammer2_pfslist, hammer2_pfsmount);
79 static struct hammer2_mntlist hammer2_mntlist;
80 static struct hammer2_pfslist hammer2_pfslist;
81 static struct lock hammer2_mntlk;
84 int hammer2_cluster_enable = 1;
85 int hammer2_hardlink_enable = 1;
86 int hammer2_flush_pipe = 100;
87 int hammer2_synchronous_flush = 1;
88 int hammer2_dio_count;
89 long hammer2_limit_dirty_chains;
90 long hammer2_iod_file_read;
91 long hammer2_iod_meta_read;
92 long hammer2_iod_indr_read;
93 long hammer2_iod_fmap_read;
94 long hammer2_iod_volu_read;
95 long hammer2_iod_file_write;
96 long hammer2_iod_meta_write;
97 long hammer2_iod_indr_write;
98 long hammer2_iod_fmap_write;
99 long hammer2_iod_volu_write;
100 long hammer2_ioa_file_read;
101 long hammer2_ioa_meta_read;
102 long hammer2_ioa_indr_read;
103 long hammer2_ioa_fmap_read;
104 long hammer2_ioa_volu_read;
105 long hammer2_ioa_fmap_write;
106 long hammer2_ioa_file_write;
107 long hammer2_ioa_meta_write;
108 long hammer2_ioa_indr_write;
109 long hammer2_ioa_volu_write;
111 MALLOC_DECLARE(C_BUFFER);
112 MALLOC_DEFINE(C_BUFFER, "compbuffer", "Buffer used for compression.");
114 MALLOC_DECLARE(D_BUFFER);
115 MALLOC_DEFINE(D_BUFFER, "decompbuffer", "Buffer used for decompression.");
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, "");
125 SYSCTL_INT(_vfs_hammer2, OID_AUTO, flush_pipe, CTLFLAG_RW,
126 &hammer2_flush_pipe, 0, "");
127 SYSCTL_INT(_vfs_hammer2, OID_AUTO, synchronous_flush, CTLFLAG_RW,
128 &hammer2_synchronous_flush, 0, "");
129 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, limit_dirty_chains, CTLFLAG_RW,
130 &hammer2_limit_dirty_chains, 0, "");
131 SYSCTL_INT(_vfs_hammer2, OID_AUTO, dio_count, CTLFLAG_RD,
132 &hammer2_dio_count, 0, "");
134 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_read, CTLFLAG_RW,
135 &hammer2_iod_file_read, 0, "");
136 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_read, CTLFLAG_RW,
137 &hammer2_iod_meta_read, 0, "");
138 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_read, CTLFLAG_RW,
139 &hammer2_iod_indr_read, 0, "");
140 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_fmap_read, CTLFLAG_RW,
141 &hammer2_iod_fmap_read, 0, "");
142 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_read, CTLFLAG_RW,
143 &hammer2_iod_volu_read, 0, "");
145 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_file_write, CTLFLAG_RW,
146 &hammer2_iod_file_write, 0, "");
147 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_meta_write, CTLFLAG_RW,
148 &hammer2_iod_meta_write, 0, "");
149 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_indr_write, CTLFLAG_RW,
150 &hammer2_iod_indr_write, 0, "");
151 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_fmap_write, CTLFLAG_RW,
152 &hammer2_iod_fmap_write, 0, "");
153 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, iod_volu_write, CTLFLAG_RW,
154 &hammer2_iod_volu_write, 0, "");
156 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_read, CTLFLAG_RW,
157 &hammer2_ioa_file_read, 0, "");
158 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_read, CTLFLAG_RW,
159 &hammer2_ioa_meta_read, 0, "");
160 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_read, CTLFLAG_RW,
161 &hammer2_ioa_indr_read, 0, "");
162 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_fmap_read, CTLFLAG_RW,
163 &hammer2_ioa_fmap_read, 0, "");
164 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_volu_read, CTLFLAG_RW,
165 &hammer2_ioa_volu_read, 0, "");
167 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_file_write, CTLFLAG_RW,
168 &hammer2_ioa_file_write, 0, "");
169 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_meta_write, CTLFLAG_RW,
170 &hammer2_ioa_meta_write, 0, "");
171 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_indr_write, CTLFLAG_RW,
172 &hammer2_ioa_indr_write, 0, "");
173 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_fmap_write, CTLFLAG_RW,
174 &hammer2_ioa_fmap_write, 0, "");
175 SYSCTL_LONG(_vfs_hammer2, OID_AUTO, ioa_volu_write, CTLFLAG_RW,
176 &hammer2_ioa_volu_write, 0, "");
178 static int hammer2_vfs_init(struct vfsconf *conf);
179 static int hammer2_vfs_uninit(struct vfsconf *vfsp);
180 static int hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
182 static int hammer2_remount(hammer2_mount_t *, struct mount *, char *,
183 struct vnode *, struct ucred *);
184 static int hammer2_recovery(hammer2_mount_t *hmp);
185 static int hammer2_vfs_unmount(struct mount *mp, int mntflags);
186 static int hammer2_vfs_root(struct mount *mp, struct vnode **vpp);
187 static int hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp,
189 static int hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp,
191 static int hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
192 ino_t ino, struct vnode **vpp);
193 static int hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
194 struct fid *fhp, struct vnode **vpp);
195 static int hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp);
196 static int hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
197 int *exflagsp, struct ucred **credanonp);
199 static int hammer2_install_volume_header(hammer2_mount_t *hmp);
200 static int hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data);
202 static void hammer2_write_thread(void *arg);
204 static void hammer2_vfs_unmount_hmp1(struct mount *mp, hammer2_mount_t *hmp);
205 static void hammer2_vfs_unmount_hmp2(struct mount *mp, hammer2_mount_t *hmp);
208 * Functions for compression in threads,
209 * from hammer2_vnops.c
211 static void hammer2_write_file_core(struct buf *bp, hammer2_trans_t *trans,
213 hammer2_inode_data_t *ipdata,
214 hammer2_cluster_t *cparent,
215 hammer2_key_t lbase, int ioflag, int pblksize,
217 static void hammer2_compress_and_write(struct buf *bp, hammer2_trans_t *trans,
219 const hammer2_inode_data_t *ipdata,
220 hammer2_cluster_t *cparent,
221 hammer2_key_t lbase, int ioflag,
222 int pblksize, int *errorp, int comp_algo);
223 static void hammer2_zero_check_and_write(struct buf *bp,
224 hammer2_trans_t *trans, hammer2_inode_t *ip,
225 const hammer2_inode_data_t *ipdata,
226 hammer2_cluster_t *cparent,
228 int ioflag, int pblksize, int *errorp);
229 static int test_block_zeros(const char *buf, size_t bytes);
230 static void zero_write(struct buf *bp, hammer2_trans_t *trans,
232 const hammer2_inode_data_t *ipdata,
233 hammer2_cluster_t *cparent,
236 static void hammer2_write_bp(hammer2_cluster_t *cluster, struct buf *bp,
237 int ioflag, int pblksize, int *errorp);
239 static int hammer2_rcvdmsg(kdmsg_msg_t *msg);
240 static void hammer2_autodmsg(kdmsg_msg_t *msg);
241 static int hammer2_lnk_span_reply(kdmsg_state_t *state, kdmsg_msg_t *msg);
245 * HAMMER2 vfs operations.
247 static struct vfsops hammer2_vfsops = {
248 .vfs_init = hammer2_vfs_init,
249 .vfs_uninit = hammer2_vfs_uninit,
250 .vfs_sync = hammer2_vfs_sync,
251 .vfs_mount = hammer2_vfs_mount,
252 .vfs_unmount = hammer2_vfs_unmount,
253 .vfs_root = hammer2_vfs_root,
254 .vfs_statfs = hammer2_vfs_statfs,
255 .vfs_statvfs = hammer2_vfs_statvfs,
256 .vfs_vget = hammer2_vfs_vget,
257 .vfs_vptofh = hammer2_vfs_vptofh,
258 .vfs_fhtovp = hammer2_vfs_fhtovp,
259 .vfs_checkexp = hammer2_vfs_checkexp
262 MALLOC_DEFINE(M_HAMMER2, "HAMMER2-mount", "");
264 VFS_SET(hammer2_vfsops, hammer2, 0);
265 MODULE_VERSION(hammer2, 1);
269 hammer2_vfs_init(struct vfsconf *conf)
271 static struct objcache_malloc_args margs_read;
272 static struct objcache_malloc_args margs_write;
278 if (HAMMER2_BLOCKREF_BYTES != sizeof(struct hammer2_blockref))
280 if (HAMMER2_INODE_BYTES != sizeof(struct hammer2_inode_data))
282 if (HAMMER2_VOLUME_BYTES != sizeof(struct hammer2_volume_data))
286 kprintf("HAMMER2 structure size mismatch; cannot continue.\n");
288 margs_read.objsize = 65536;
289 margs_read.mtype = D_BUFFER;
291 margs_write.objsize = 32768;
292 margs_write.mtype = C_BUFFER;
294 cache_buffer_read = objcache_create(margs_read.mtype->ks_shortdesc,
295 0, 1, NULL, NULL, NULL, objcache_malloc_alloc,
296 objcache_malloc_free, &margs_read);
297 cache_buffer_write = objcache_create(margs_write.mtype->ks_shortdesc,
298 0, 1, NULL, NULL, NULL, objcache_malloc_alloc,
299 objcache_malloc_free, &margs_write);
301 lockinit(&hammer2_mntlk, "mntlk", 0, 0);
302 TAILQ_INIT(&hammer2_mntlist);
303 TAILQ_INIT(&hammer2_pfslist);
305 hammer2_limit_dirty_chains = desiredvnodes / 10;
307 hammer2_trans_manage_init();
314 hammer2_vfs_uninit(struct vfsconf *vfsp __unused)
316 objcache_destroy(cache_buffer_read);
317 objcache_destroy(cache_buffer_write);
322 * Core PFS allocator. Used to allocate the pmp structure for PFS cluster
323 * mounts and the spmp structure for media (hmp) structures.
325 static hammer2_pfsmount_t *
326 hammer2_pfsalloc(const hammer2_inode_data_t *ipdata, hammer2_tid_t alloc_tid)
328 hammer2_pfsmount_t *pmp;
330 pmp = kmalloc(sizeof(*pmp), M_HAMMER2, M_WAITOK | M_ZERO);
331 kmalloc_create(&pmp->minode, "HAMMER2-inodes");
332 kmalloc_create(&pmp->mmsg, "HAMMER2-pfsmsg");
333 lockinit(&pmp->lock, "pfslk", 0, 0);
334 spin_init(&pmp->inum_spin);
335 RB_INIT(&pmp->inum_tree);
336 TAILQ_INIT(&pmp->unlinkq);
337 spin_init(&pmp->unlinkq_spin);
339 pmp->alloc_tid = alloc_tid + 1; /* our first media transaction id */
340 pmp->flush_tid = pmp->alloc_tid;
342 pmp->inode_tid = ipdata->pfs_inum + 1;
343 pmp->pfs_clid = ipdata->pfs_clid;
345 mtx_init(&pmp->wthread_mtx);
346 bioq_init(&pmp->wthread_bioq);
352 * Mount or remount HAMMER2 fileystem from physical media
355 * mp mount point structure
361 * mp mount point structure
362 * path path to mount point
363 * data pointer to argument structure in user space
364 * volume volume path (device@LABEL form)
365 * hflags user mount flags
366 * cred user credentials
373 hammer2_vfs_mount(struct mount *mp, char *path, caddr_t data,
376 struct hammer2_mount_info info;
377 hammer2_pfsmount_t *pmp;
378 hammer2_pfsmount_t *spmp;
379 hammer2_mount_t *hmp;
380 hammer2_key_t key_next;
381 hammer2_key_t key_dummy;
384 struct nlookupdata nd;
385 hammer2_chain_t *parent;
386 hammer2_chain_t *rchain;
387 hammer2_cluster_t *cluster;
388 hammer2_cluster_t *cparent;
389 const hammer2_inode_data_t *ipdata;
390 hammer2_blockref_t bref;
392 char devstr[MNAMELEN];
410 kprintf("hammer2_mount\n");
416 bzero(&info, sizeof(info));
417 info.cluster_fd = -1;
421 * Non-root mount or updating a mount
423 error = copyin(data, &info, sizeof(info));
427 error = copyinstr(info.volume, devstr, MNAMELEN - 1, &done);
431 /* Extract device and label */
433 label = strchr(devstr, '@');
435 ((label + 1) - dev) > done) {
443 if (mp->mnt_flag & MNT_UPDATE) {
445 /* HAMMER2 implements NFS export via mountctl */
447 for (i = 0; i < pmp->iroot->cluster.nchains; ++i) {
448 hmp = pmp->iroot->cluster.array[i]->hmp;
450 error = hammer2_remount(hmp, mp, path,
455 /*hammer2_inode_install_hidden(pmp);*/
464 * Lookup name and verify it refers to a block device.
466 error = nlookup_init(&nd, dev, UIO_SYSSPACE, NLC_FOLLOW);
468 error = nlookup(&nd);
470 error = cache_vref(&nd.nl_nch, nd.nl_cred, &devvp);
474 if (vn_isdisk(devvp, &error))
475 error = vfs_mountedon(devvp);
479 * Determine if the device has already been mounted. After this
480 * check hmp will be non-NULL if we are doing the second or more
481 * hammer2 mounts from the same device.
483 lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
484 TAILQ_FOREACH(hmp, &hammer2_mntlist, mntentry) {
485 if (hmp->devvp == devvp)
490 * Open the device if this isn't a secondary mount and construct
491 * the H2 device mount (hmp).
494 hammer2_chain_t *schain;
497 if (error == 0 && vcount(devvp) > 0)
501 * Now open the device
504 ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
505 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
506 error = vinvalbuf(devvp, V_SAVE, 0, 0);
508 error = VOP_OPEN(devvp,
509 ronly ? FREAD : FREAD | FWRITE,
514 if (error && devvp) {
519 lockmgr(&hammer2_mntlk, LK_RELEASE);
522 hmp = kmalloc(sizeof(*hmp), M_HAMMER2, M_WAITOK | M_ZERO);
525 kmalloc_create(&hmp->mchain, "HAMMER2-chains");
526 TAILQ_INSERT_TAIL(&hammer2_mntlist, hmp, mntentry);
527 RB_INIT(&hmp->iotree);
529 lockinit(&hmp->vollk, "h2vol", 0, 0);
532 * vchain setup. vchain.data is embedded.
533 * vchain.refs is initialized and will never drop to 0.
535 * NOTE! voldata is not yet loaded.
537 hmp->vchain.hmp = hmp;
538 hmp->vchain.refs = 1;
539 hmp->vchain.data = (void *)&hmp->voldata;
540 hmp->vchain.bref.type = HAMMER2_BREF_TYPE_VOLUME;
541 hmp->vchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
542 hmp->vchain.bref.mirror_tid = hmp->voldata.mirror_tid;
543 hmp->vchain.delete_xid = HAMMER2_XID_MAX;
545 hammer2_chain_core_alloc(NULL, &hmp->vchain, NULL);
546 /* hmp->vchain.u.xxx is left NULL */
549 * fchain setup. fchain.data is embedded.
550 * fchain.refs is initialized and will never drop to 0.
552 * The data is not used but needs to be initialized to
553 * pass assertion muster. We use this chain primarily
554 * as a placeholder for the freemap's top-level RBTREE
555 * so it does not interfere with the volume's topology
558 hmp->fchain.hmp = hmp;
559 hmp->fchain.refs = 1;
560 hmp->fchain.data = (void *)&hmp->voldata.freemap_blockset;
561 hmp->fchain.bref.type = HAMMER2_BREF_TYPE_FREEMAP;
562 hmp->fchain.bref.data_off = 0 | HAMMER2_PBUFRADIX;
563 hmp->fchain.bref.mirror_tid = hmp->voldata.freemap_tid;
564 hmp->fchain.bref.methods =
565 HAMMER2_ENC_CHECK(HAMMER2_CHECK_FREEMAP) |
566 HAMMER2_ENC_COMP(HAMMER2_COMP_NONE);
567 hmp->fchain.delete_xid = HAMMER2_XID_MAX;
569 hammer2_chain_core_alloc(NULL, &hmp->fchain, NULL);
570 /* hmp->fchain.u.xxx is left NULL */
573 * Install the volume header and initialize fields from
576 error = hammer2_install_volume_header(hmp);
579 hammer2_vfs_unmount_hmp1(mp, hmp);
580 hammer2_vfs_unmount_hmp2(mp, hmp);
581 lockmgr(&hammer2_mntlk, LK_RELEASE);
582 hammer2_vfs_unmount(mp, MNT_FORCE);
587 * Really important to get these right or flush will get
590 hmp->spmp = hammer2_pfsalloc(NULL, hmp->voldata.mirror_tid);
591 kprintf("alloc spmp %p tid %016jx\n",
592 hmp->spmp, hmp->voldata.mirror_tid);
597 hmp->vchain.bref.mirror_tid = hmp->voldata.mirror_tid;
598 hmp->vchain.bref.modify_tid = hmp->vchain.bref.mirror_tid;
599 hmp->vchain.modify_xid = xid;
600 hmp->vchain.update_xlo = xid;
601 hmp->vchain.update_xhi = xid;
602 hmp->vchain.pmp = spmp;
603 hmp->fchain.bref.mirror_tid = hmp->voldata.freemap_tid;
604 hmp->fchain.bref.modify_tid = hmp->fchain.bref.mirror_tid;
605 hmp->fchain.modify_xid = xid;
606 hmp->fchain.update_xlo = xid;
607 hmp->fchain.update_xhi = xid;
608 hmp->fchain.pmp = spmp;
611 * First locate the super-root inode, which is key 0
612 * relative to the volume header's blockset.
614 * Then locate the root inode by scanning the directory keyspace
615 * represented by the label.
617 parent = hammer2_chain_lookup_init(&hmp->vchain, 0);
618 schain = hammer2_chain_lookup(&parent, &key_dummy,
619 HAMMER2_SROOT_KEY, HAMMER2_SROOT_KEY,
620 &cache_index, 0, &ddflag);
621 hammer2_chain_lookup_done(parent);
622 if (schain == NULL) {
623 kprintf("hammer2_mount: invalid super-root\n");
625 hammer2_vfs_unmount_hmp1(mp, hmp);
626 hammer2_vfs_unmount_hmp2(mp, hmp);
627 lockmgr(&hammer2_mntlk, LK_RELEASE);
628 hammer2_vfs_unmount(mp, MNT_FORCE);
633 * Sanity-check schain's pmp, finish initializing spmp.
635 KKASSERT(schain->pmp == spmp);
636 spmp->pfs_clid = schain->data->ipdata.pfs_clid;
639 * NOTE: The CHAIN_PFSROOT is not set on the super-root inode.
640 * NOTE: inode_get sucks up schain's lock.
642 cluster = hammer2_cluster_from_chain(schain);
643 spmp->iroot = hammer2_inode_get(spmp, NULL, cluster);
644 spmp->spmp_hmp = hmp;
645 hammer2_inode_ref(spmp->iroot);
646 hammer2_inode_unlock_ex(spmp->iroot, cluster);
648 /* leave spmp->iroot with one ref */
650 if ((mp->mnt_flag & MNT_RDONLY) == 0) {
651 error = hammer2_recovery(hmp);
652 /* XXX do something with error */
657 * XXX RDONLY stuff is totally broken FIXME XXX
660 * Automatic handling of received LNK_SPAN
661 * Automatic handling of received LNK_CIRC
662 * No automatic LNK_SPAN generation - we do this ourselves
663 * No automatic LNK_CIRC generation - we do this ourselves
665 kdmsg_iocom_init(&hmp->iocom, hmp,
666 KDMSG_IOCOMF_AUTOCONN |
667 KDMSG_IOCOMF_AUTORXSPAN |
668 KDMSG_IOCOMF_AUTORXCIRC,
669 hmp->mchain, hammer2_rcvdmsg);
672 * Ref the cluster management messaging descriptor. The mount
673 * program deals with the other end of the communications pipe.
675 fp = holdfp(curproc->p_fd, info.cluster_fd, -1);
677 hammer2_cluster_reconnect(hmp, fp);
679 kprintf("hammer2_mount: bad cluster_fd!\n");
687 * Lookup mount point under the media-localized super-root.
689 * cluster->pmp will incorrectly point to spmp and must be fixed
692 cparent = hammer2_inode_lock_ex(spmp->iroot);
693 lhc = hammer2_dirhash(label, strlen(label));
694 cluster = hammer2_cluster_lookup(cparent, &key_next,
695 lhc, lhc + HAMMER2_DIRHASH_LOMASK,
698 if (hammer2_cluster_type(cluster) == HAMMER2_BREF_TYPE_INODE &&
700 hammer2_cluster_data(cluster)->ipdata.filename) == 0) {
703 cluster = hammer2_cluster_next(cparent, cluster, &key_next,
705 lhc + HAMMER2_DIRHASH_LOMASK, 0);
707 hammer2_inode_unlock_ex(spmp->iroot, cparent);
709 if (cluster == NULL) {
710 kprintf("hammer2_mount: PFS label not found\n");
711 hammer2_vfs_unmount_hmp1(mp, hmp);
712 hammer2_vfs_unmount_hmp2(mp, hmp);
713 lockmgr(&hammer2_mntlk, LK_RELEASE);
714 hammer2_vfs_unmount(mp, MNT_FORCE);
718 for (i = 0; i < cluster->nchains; ++i) {
719 rchain = cluster->array[i];
720 KKASSERT(rchain->pmp == NULL);
721 if (rchain->flags & HAMMER2_CHAIN_MOUNTED) {
722 kprintf("hammer2_mount: PFS label already mounted!\n");
723 hammer2_cluster_unlock(cluster);
724 hammer2_vfs_unmount_hmp1(mp, hmp);
725 hammer2_vfs_unmount_hmp2(mp, hmp);
726 lockmgr(&hammer2_mntlk, LK_RELEASE);
727 hammer2_vfs_unmount(mp, MNT_FORCE);
731 if (rchain->flags & HAMMER2_CHAIN_RECYCLE) {
732 kprintf("hammer2_mount: PFS label is recycling\n");
733 hammer2_cluster_unlock(cluster);
734 hammer2_vfs_unmount_hmp1(mp, hmp);
735 hammer2_vfs_unmount_hmp2(mp, hmp);
736 lockmgr(&hammer2_mntlk, LK_RELEASE);
737 hammer2_vfs_unmount(mp, MNT_FORCE);
744 * Check to see if the cluster id is already mounted at the mount
745 * point. If it is, add us to the cluster.
747 ipdata = &hammer2_cluster_data(cluster)->ipdata;
748 hammer2_cluster_bref(cluster, &bref);
749 TAILQ_FOREACH(pmp, &hammer2_pfslist, mntentry) {
750 if (pmp->spmp_hmp == NULL &&
751 bcmp(&pmp->pfs_clid, &ipdata->pfs_clid,
752 sizeof(pmp->pfs_clid)) == 0) {
761 hammer2_inode_ref(pmp->iroot);
762 ccms_thread_lock(&pmp->iroot->topo_cst, CCMS_STATE_EXCLUSIVE);
764 if (pmp->iroot->cluster.nchains + cluster->nchains >
765 HAMMER2_MAXCLUSTER) {
766 kprintf("hammer2_mount: cluster full!\n");
768 ccms_thread_unlock(&pmp->iroot->topo_cst);
769 hammer2_inode_drop(pmp->iroot);
771 hammer2_cluster_unlock(cluster);
772 hammer2_vfs_unmount_hmp1(mp, hmp);
773 hammer2_vfs_unmount_hmp2(mp, hmp);
774 lockmgr(&hammer2_mntlk, LK_RELEASE);
775 hammer2_vfs_unmount(mp, MNT_FORCE);
778 kprintf("hammer2_vfs_mount: Adding pfs to existing cluster\n");
779 j = pmp->iroot->cluster.nchains;
780 for (i = 0; i < cluster->nchains; ++i) {
781 rchain = cluster->array[i];
782 KKASSERT(rchain->pmp == NULL);
784 hammer2_chain_ref(cluster->array[i]);
785 pmp->iroot->cluster.array[j] = cluster->array[i];
788 pmp->iroot->cluster.nchains = j;
789 ccms_thread_unlock(&pmp->iroot->topo_cst);
790 hammer2_inode_drop(pmp->iroot);
791 hammer2_cluster_unlock(cluster);
792 lockmgr(&hammer2_mntlk, LK_RELEASE);
795 hammer2_inode_install_hidden(pmp);
801 * Block device opened successfully, finish initializing the
804 * From this point on we have to call hammer2_unmount() on failure.
806 pmp = hammer2_pfsalloc(ipdata, bref.mirror_tid);
807 kprintf("PMP mirror_tid is %016jx\n", bref.mirror_tid);
808 for (i = 0; i < cluster->nchains; ++i) {
809 rchain = cluster->array[i];
810 KKASSERT(rchain->pmp == NULL);
812 atomic_set_int(&rchain->flags, HAMMER2_CHAIN_MOUNTED);
816 ccms_domain_init(&pmp->ccms_dom);
817 TAILQ_INSERT_TAIL(&hammer2_pfslist, pmp, mntentry);
818 lockmgr(&hammer2_mntlk, LK_RELEASE);
820 kprintf("hammer2_mount hmp=%p pmp=%p pmpcnt=%d\n",
821 hmp, pmp, hmp->pmp_count);
823 mp->mnt_flag = MNT_LOCAL;
824 mp->mnt_kern_flag |= MNTK_ALL_MPSAFE; /* all entry pts are SMP */
825 mp->mnt_kern_flag |= MNTK_THR_SYNC; /* new vsyncscan semantics */
828 * required mount structure initializations
830 mp->mnt_stat.f_iosize = HAMMER2_PBUFSIZE;
831 mp->mnt_stat.f_bsize = HAMMER2_PBUFSIZE;
833 mp->mnt_vstat.f_frsize = HAMMER2_PBUFSIZE;
834 mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
839 mp->mnt_iosize_max = MAXPHYS;
840 mp->mnt_data = (qaddr_t)pmp;
844 * After this point hammer2_vfs_unmount() has visibility on hmp
845 * and manual hmp1/hmp2 calls are not needed on fatal errors.
847 pmp->iroot = hammer2_inode_get(pmp, NULL, cluster);
848 hammer2_inode_ref(pmp->iroot); /* ref for pmp->iroot */
849 hammer2_inode_unlock_ex(pmp->iroot, cluster);
852 * The logical file buffer bio write thread handles things
853 * like physical block assignment and compression.
855 * (only applicable to pfs mounts, not applicable to spmp)
857 pmp->wthread_destroy = 0;
858 lwkt_create(hammer2_write_thread, pmp,
859 &pmp->wthread_td, NULL, 0, -1, "hwrite-%s", label);
862 * With the cluster operational install ihidden.
863 * (only applicable to pfs mounts, not applicable to spmp)
865 hammer2_inode_install_hidden(pmp);
871 vfs_add_vnodeops(mp, &hammer2_vnode_vops, &mp->mnt_vn_norm_ops);
872 vfs_add_vnodeops(mp, &hammer2_spec_vops, &mp->mnt_vn_spec_ops);
873 vfs_add_vnodeops(mp, &hammer2_fifo_vops, &mp->mnt_vn_fifo_ops);
875 copyinstr(info.volume, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, &size);
876 bzero(mp->mnt_stat.f_mntfromname + size, MNAMELEN - size);
877 bzero(mp->mnt_stat.f_mntonname, sizeof(mp->mnt_stat.f_mntonname));
878 copyinstr(path, mp->mnt_stat.f_mntonname,
879 sizeof(mp->mnt_stat.f_mntonname) - 1,
883 * Initial statfs to prime mnt_stat.
885 hammer2_vfs_statfs(mp, &mp->mnt_stat, cred);
891 * Handle bioq for strategy write
895 hammer2_write_thread(void *arg)
897 hammer2_pfsmount_t *pmp;
900 hammer2_trans_t trans;
903 hammer2_cluster_t *cparent;
904 hammer2_inode_data_t *wipdata;
912 mtx_lock(&pmp->wthread_mtx);
913 while (pmp->wthread_destroy == 0) {
914 if (bioq_first(&pmp->wthread_bioq) == NULL) {
915 mtxsleep(&pmp->wthread_bioq, &pmp->wthread_mtx,
920 hammer2_trans_init(&trans, pmp, HAMMER2_TRANS_BUFCACHE);
922 while ((bio = bioq_takefirst(&pmp->wthread_bioq)) != NULL) {
924 * dummy bio for synchronization. The transaction
925 * must be reinitialized.
927 if (bio->bio_buf == NULL) {
928 bio->bio_flags |= BIO_DONE;
930 hammer2_trans_done(&trans);
931 hammer2_trans_init(&trans, pmp,
932 HAMMER2_TRANS_BUFCACHE);
937 * else normal bio processing
939 mtx_unlock(&pmp->wthread_mtx);
941 hammer2_lwinprog_drop(pmp);
949 * Inode is modified, flush size and mtime changes
950 * to ensure that the file size remains consistent
951 * with the buffers being flushed.
953 * NOTE: The inode_fsync() call only flushes the
954 * inode's meta-data state, it doesn't try
955 * to flush underlying buffers or chains.
957 cparent = hammer2_inode_lock_ex(ip);
958 if (ip->flags & (HAMMER2_INODE_RESIZED |
959 HAMMER2_INODE_MTIME)) {
960 hammer2_inode_fsync(&trans, ip, cparent);
962 wipdata = hammer2_cluster_modify_ip(&trans, ip,
964 lblksize = hammer2_calc_logical(ip, bio->bio_offset,
966 pblksize = hammer2_calc_physical(ip, wipdata, lbase);
967 hammer2_write_file_core(bp, &trans, ip, wipdata,
971 hammer2_cluster_modsync(cparent);
972 hammer2_inode_unlock_ex(ip, cparent);
974 kprintf("hammer2: error in buffer write\n");
975 bp->b_flags |= B_ERROR;
979 mtx_lock(&pmp->wthread_mtx);
981 hammer2_trans_done(&trans);
983 pmp->wthread_destroy = -1;
984 wakeup(&pmp->wthread_destroy);
986 mtx_unlock(&pmp->wthread_mtx);
990 hammer2_bioq_sync(hammer2_pfsmount_t *pmp)
994 bzero(&sync_bio, sizeof(sync_bio)); /* dummy with no bio_buf */
995 mtx_lock(&pmp->wthread_mtx);
996 if (pmp->wthread_destroy == 0 &&
997 TAILQ_FIRST(&pmp->wthread_bioq.queue)) {
998 bioq_insert_tail(&pmp->wthread_bioq, &sync_bio);
999 while ((sync_bio.bio_flags & BIO_DONE) == 0)
1000 mtxsleep(&sync_bio, &pmp->wthread_mtx, 0, "h2bioq", 0);
1002 mtx_unlock(&pmp->wthread_mtx);
1006 * Return a chain suitable for I/O, creating the chain if necessary
1007 * and assigning its physical block.
1011 hammer2_assign_physical(hammer2_trans_t *trans,
1012 hammer2_inode_t *ip, hammer2_cluster_t *cparent,
1013 hammer2_key_t lbase, int pblksize, int *errorp)
1015 hammer2_cluster_t *cluster;
1016 hammer2_cluster_t *dparent;
1017 hammer2_key_t key_dummy;
1018 int pradix = hammer2_getradix(pblksize);
1022 * Locate the chain associated with lbase, return a locked chain.
1023 * However, do not instantiate any data reference (which utilizes a
1024 * device buffer) because we will be using direct IO via the
1025 * logical buffer cache buffer.
1028 KKASSERT(pblksize >= HAMMER2_ALLOC_MIN);
1030 dparent = hammer2_cluster_lookup_init(cparent, 0);
1031 cluster = hammer2_cluster_lookup(dparent, &key_dummy,
1033 HAMMER2_LOOKUP_NODATA, &ddflag);
1035 if (cluster == NULL) {
1037 * We found a hole, create a new chain entry.
1039 * NOTE: DATA chains are created without device backing
1040 * store (nor do we want any).
1042 *errorp = hammer2_cluster_create(trans, dparent, &cluster,
1043 lbase, HAMMER2_PBUFRADIX,
1044 HAMMER2_BREF_TYPE_DATA,
1046 if (cluster == NULL) {
1047 hammer2_cluster_lookup_done(dparent);
1048 panic("hammer2_cluster_create: par=%p error=%d\n",
1049 dparent->focus, *errorp);
1052 /*ip->delta_dcount += pblksize;*/
1054 switch (hammer2_cluster_type(cluster)) {
1055 case HAMMER2_BREF_TYPE_INODE:
1057 * The data is embedded in the inode. The
1058 * caller is responsible for marking the inode
1059 * modified and copying the data to the embedded
1063 case HAMMER2_BREF_TYPE_DATA:
1064 if (hammer2_cluster_bytes(cluster) != pblksize) {
1065 hammer2_cluster_resize(trans, ip,
1068 HAMMER2_MODIFY_OPTDATA);
1070 hammer2_cluster_modify(trans, cluster,
1071 HAMMER2_MODIFY_OPTDATA);
1074 panic("hammer2_assign_physical: bad type");
1081 * Cleanup. If cluster wound up being the inode itself, i.e.
1082 * the DIRECTDATA case for offset 0, then we need to update cparent.
1083 * The caller expects cparent to not become stale.
1085 hammer2_cluster_lookup_done(dparent);
1086 /* dparent = NULL; safety */
1087 if (cluster && ddflag)
1088 hammer2_cluster_replace_locked(cparent, cluster);
1093 * From hammer2_vnops.c.
1094 * The core write function which determines which path to take
1095 * depending on compression settings.
1099 hammer2_write_file_core(struct buf *bp, hammer2_trans_t *trans,
1100 hammer2_inode_t *ip, hammer2_inode_data_t *ipdata,
1101 hammer2_cluster_t *cparent,
1102 hammer2_key_t lbase, int ioflag, int pblksize,
1105 hammer2_cluster_t *cluster;
1107 switch(HAMMER2_DEC_COMP(ipdata->comp_algo)) {
1108 case HAMMER2_COMP_NONE:
1110 * We have to assign physical storage to the buffer
1111 * we intend to dirty or write now to avoid deadlocks
1112 * in the strategy code later.
1114 * This can return NOOFFSET for inode-embedded data.
1115 * The strategy code will take care of it in that case.
1117 cluster = hammer2_assign_physical(trans, ip, cparent,
1120 hammer2_write_bp(cluster, bp, ioflag, pblksize, errorp);
1122 hammer2_cluster_unlock(cluster);
1124 case HAMMER2_COMP_AUTOZERO:
1126 * Check for zero-fill only
1128 hammer2_zero_check_and_write(bp, trans, ip,
1129 ipdata, cparent, lbase,
1130 ioflag, pblksize, errorp);
1132 case HAMMER2_COMP_LZ4:
1133 case HAMMER2_COMP_ZLIB:
1136 * Check for zero-fill and attempt compression.
1138 hammer2_compress_and_write(bp, trans, ip,
1148 * Generic function that will perform the compression in compression
1149 * write path. The compression algorithm is determined by the settings
1150 * obtained from inode.
1154 hammer2_compress_and_write(struct buf *bp, hammer2_trans_t *trans,
1155 hammer2_inode_t *ip, const hammer2_inode_data_t *ipdata,
1156 hammer2_cluster_t *cparent,
1157 hammer2_key_t lbase, int ioflag, int pblksize,
1158 int *errorp, int comp_algo)
1160 hammer2_cluster_t *cluster;
1161 hammer2_chain_t *chain;
1163 int comp_block_size;
1167 if (test_block_zeros(bp->b_data, pblksize)) {
1168 zero_write(bp, trans, ip, ipdata, cparent, lbase, errorp);
1175 KKASSERT(pblksize / 2 <= 32768);
1177 if (ip->comp_heuristic < 8 || (ip->comp_heuristic & 7) == 0) {
1178 z_stream strm_compress;
1182 switch(HAMMER2_DEC_COMP(comp_algo)) {
1183 case HAMMER2_COMP_LZ4:
1184 comp_buffer = objcache_get(cache_buffer_write,
1186 comp_size = LZ4_compress_limitedOutput(
1188 &comp_buffer[sizeof(int)],
1190 pblksize / 2 - sizeof(int));
1192 * We need to prefix with the size, LZ4
1193 * doesn't do it for us. Add the related
1196 *(int *)comp_buffer = comp_size;
1198 comp_size += sizeof(int);
1200 case HAMMER2_COMP_ZLIB:
1201 comp_level = HAMMER2_DEC_LEVEL(comp_algo);
1202 if (comp_level == 0)
1203 comp_level = 6; /* default zlib compression */
1204 else if (comp_level < 6)
1206 else if (comp_level > 9)
1208 ret = deflateInit(&strm_compress, comp_level);
1210 kprintf("HAMMER2 ZLIB: fatal error "
1211 "on deflateInit.\n");
1214 comp_buffer = objcache_get(cache_buffer_write,
1216 strm_compress.next_in = bp->b_data;
1217 strm_compress.avail_in = pblksize;
1218 strm_compress.next_out = comp_buffer;
1219 strm_compress.avail_out = pblksize / 2;
1220 ret = deflate(&strm_compress, Z_FINISH);
1221 if (ret == Z_STREAM_END) {
1222 comp_size = pblksize / 2 -
1223 strm_compress.avail_out;
1227 ret = deflateEnd(&strm_compress);
1230 kprintf("Error: Unknown compression method.\n");
1231 kprintf("Comp_method = %d.\n", comp_algo);
1236 if (comp_size == 0) {
1238 * compression failed or turned off
1240 comp_block_size = pblksize; /* safety */
1241 if (++ip->comp_heuristic > 128)
1242 ip->comp_heuristic = 8;
1245 * compression succeeded
1247 ip->comp_heuristic = 0;
1248 if (comp_size <= 1024) {
1249 comp_block_size = 1024;
1250 } else if (comp_size <= 2048) {
1251 comp_block_size = 2048;
1252 } else if (comp_size <= 4096) {
1253 comp_block_size = 4096;
1254 } else if (comp_size <= 8192) {
1255 comp_block_size = 8192;
1256 } else if (comp_size <= 16384) {
1257 comp_block_size = 16384;
1258 } else if (comp_size <= 32768) {
1259 comp_block_size = 32768;
1261 panic("hammer2: WRITE PATH: "
1262 "Weird comp_size value.");
1264 comp_block_size = pblksize;
1268 cluster = hammer2_assign_physical(trans, ip, cparent,
1269 lbase, comp_block_size,
1271 ipdata = &hammer2_cluster_data(cparent)->ipdata;
1274 kprintf("WRITE PATH: An error occurred while "
1275 "assigning physical space.\n");
1276 KKASSERT(cluster == NULL);
1280 for (i = 0; i < cluster->nchains; ++i) {
1285 chain = cluster->array[i];
1286 KKASSERT(chain->flags & HAMMER2_CHAIN_MODIFIED);
1288 switch(chain->bref.type) {
1289 case HAMMER2_BREF_TYPE_INODE:
1290 KKASSERT(chain->data->ipdata.op_flags &
1291 HAMMER2_OPFLAG_DIRECTDATA);
1292 KKASSERT(bp->b_loffset == 0);
1293 bcopy(bp->b_data, chain->data->ipdata.u.data,
1294 HAMMER2_EMBEDDED_BYTES);
1296 case HAMMER2_BREF_TYPE_DATA:
1297 temp_check = HAMMER2_DEC_CHECK(chain->bref.methods);
1300 * Optimize out the read-before-write
1303 *errorp = hammer2_io_newnz(chain->hmp,
1304 chain->bref.data_off,
1308 hammer2_io_brelse(&dio);
1309 kprintf("hammer2: WRITE PATH: "
1310 "dbp bread error\n");
1313 bdata = hammer2_io_data(dio, chain->bref.data_off);
1316 * When loading the block make sure we don't
1317 * leave garbage after the compressed data.
1320 chain->bref.methods =
1321 HAMMER2_ENC_COMP(comp_algo) +
1322 HAMMER2_ENC_CHECK(temp_check);
1323 bcopy(comp_buffer, bdata, comp_size);
1324 if (comp_size != comp_block_size) {
1325 bzero(bdata + comp_size,
1326 comp_block_size - comp_size);
1329 chain->bref.methods =
1331 HAMMER2_COMP_NONE) +
1332 HAMMER2_ENC_CHECK(temp_check);
1333 bcopy(bp->b_data, bdata, pblksize);
1337 * Device buffer is now valid, chain is no
1338 * longer in the initial state.
1340 atomic_clear_int(&chain->flags, HAMMER2_CHAIN_INITIAL);
1342 /* Now write the related bdp. */
1343 if (ioflag & IO_SYNC) {
1345 * Synchronous I/O requested.
1347 hammer2_io_bwrite(&dio);
1349 } else if ((ioflag & IO_DIRECT) &&
1350 loff + n == pblksize) {
1351 hammer2_io_bdwrite(&dio);
1353 } else if (ioflag & IO_ASYNC) {
1354 hammer2_io_bawrite(&dio);
1356 hammer2_io_bdwrite(&dio);
1360 panic("hammer2_write_bp: bad chain type %d\n",
1368 hammer2_cluster_unlock(cluster);
1370 objcache_put(cache_buffer_write, comp_buffer);
1374 * Function that performs zero-checking and writing without compression,
1375 * it corresponds to default zero-checking path.
1379 hammer2_zero_check_and_write(struct buf *bp, hammer2_trans_t *trans,
1380 hammer2_inode_t *ip, const hammer2_inode_data_t *ipdata,
1381 hammer2_cluster_t *cparent,
1382 hammer2_key_t lbase, int ioflag, int pblksize, int *errorp)
1384 hammer2_cluster_t *cluster;
1386 if (test_block_zeros(bp->b_data, pblksize)) {
1387 zero_write(bp, trans, ip, ipdata, cparent, lbase, errorp);
1389 cluster = hammer2_assign_physical(trans, ip, cparent,
1390 lbase, pblksize, errorp);
1391 hammer2_write_bp(cluster, bp, ioflag, pblksize, errorp);
1393 hammer2_cluster_unlock(cluster);
1398 * A function to test whether a block of data contains only zeros,
1399 * returns TRUE (non-zero) if the block is all zeros.
1403 test_block_zeros(const char *buf, size_t bytes)
1407 for (i = 0; i < bytes; i += sizeof(long)) {
1408 if (*(const long *)(buf + i) != 0)
1415 * Function to "write" a block that contains only zeros.
1419 zero_write(struct buf *bp, hammer2_trans_t *trans,
1420 hammer2_inode_t *ip, const hammer2_inode_data_t *ipdata,
1421 hammer2_cluster_t *cparent,
1422 hammer2_key_t lbase, int *errorp __unused)
1424 hammer2_cluster_t *cluster;
1425 hammer2_media_data_t *data;
1426 hammer2_key_t key_dummy;
1429 cparent = hammer2_cluster_lookup_init(cparent, 0);
1430 cluster = hammer2_cluster_lookup(cparent, &key_dummy, lbase, lbase,
1431 HAMMER2_LOOKUP_NODATA, &ddflag);
1433 data = hammer2_cluster_wdata(cluster);
1436 KKASSERT(cluster->focus->flags &
1437 HAMMER2_CHAIN_MODIFIED);
1438 bzero(data->ipdata.u.data, HAMMER2_EMBEDDED_BYTES);
1439 hammer2_cluster_modsync(cluster);
1441 hammer2_cluster_delete(trans, cluster, 0);
1443 hammer2_cluster_unlock(cluster);
1445 hammer2_cluster_lookup_done(cparent);
1449 * Function to write the data as it is, without performing any sort of
1450 * compression. This function is used in path without compression and
1451 * default zero-checking path.
1455 hammer2_write_bp(hammer2_cluster_t *cluster, struct buf *bp, int ioflag,
1456 int pblksize, int *errorp)
1458 hammer2_chain_t *chain;
1465 error = 0; /* XXX TODO below */
1467 for (i = 0; i < cluster->nchains; ++i) {
1468 chain = cluster->array[i];
1470 temp_check = HAMMER2_DEC_CHECK(chain->bref.methods);
1472 KKASSERT(chain->flags & HAMMER2_CHAIN_MODIFIED);
1474 switch(chain->bref.type) {
1475 case HAMMER2_BREF_TYPE_INODE:
1476 KKASSERT(chain->data->ipdata.op_flags &
1477 HAMMER2_OPFLAG_DIRECTDATA);
1478 KKASSERT(bp->b_loffset == 0);
1479 bcopy(bp->b_data, chain->data->ipdata.u.data,
1480 HAMMER2_EMBEDDED_BYTES);
1483 case HAMMER2_BREF_TYPE_DATA:
1484 error = hammer2_io_newnz(chain->hmp,
1485 chain->bref.data_off,
1486 chain->bytes, &dio);
1488 hammer2_io_bqrelse(&dio);
1489 kprintf("hammer2: WRITE PATH: "
1490 "dbp bread error\n");
1493 bdata = hammer2_io_data(dio, chain->bref.data_off);
1495 chain->bref.methods = HAMMER2_ENC_COMP(
1496 HAMMER2_COMP_NONE) +
1497 HAMMER2_ENC_CHECK(temp_check);
1498 bcopy(bp->b_data, bdata, chain->bytes);
1501 * Device buffer is now valid, chain is no
1502 * longer in the initial state.
1504 atomic_clear_int(&chain->flags, HAMMER2_CHAIN_INITIAL);
1506 if (ioflag & IO_SYNC) {
1508 * Synchronous I/O requested.
1510 hammer2_io_bwrite(&dio);
1512 } else if ((ioflag & IO_DIRECT) &&
1513 loff + n == pblksize) {
1514 hammer2_io_bdwrite(&dio);
1516 } else if (ioflag & IO_ASYNC) {
1517 hammer2_io_bawrite(&dio);
1519 hammer2_io_bdwrite(&dio);
1523 panic("hammer2_write_bp: bad chain type %d\n",
1529 KKASSERT(error == 0); /* XXX TODO */
1536 hammer2_remount(hammer2_mount_t *hmp, struct mount *mp, char *path,
1537 struct vnode *devvp, struct ucred *cred)
1541 if (hmp->ronly && (mp->mnt_kern_flag & MNTK_WANTRDWR)) {
1542 error = hammer2_recovery(hmp);
1551 hammer2_vfs_unmount(struct mount *mp, int mntflags)
1553 hammer2_pfsmount_t *pmp;
1554 hammer2_mount_t *hmp;
1555 hammer2_chain_t *rchain;
1556 hammer2_cluster_t *cluster;
1566 lockmgr(&hammer2_mntlk, LK_EXCLUSIVE);
1567 TAILQ_REMOVE(&hammer2_pfslist, pmp, mntentry);
1570 * If mount initialization proceeded far enough we must flush
1573 if (mntflags & MNT_FORCE)
1578 error = vflush(mp, 0, flags);
1583 ccms_domain_uninit(&pmp->ccms_dom);
1585 if (pmp->wthread_td) {
1586 mtx_lock(&pmp->wthread_mtx);
1587 pmp->wthread_destroy = 1;
1588 wakeup(&pmp->wthread_bioq);
1589 while (pmp->wthread_destroy != -1) {
1590 mtxsleep(&pmp->wthread_destroy,
1591 &pmp->wthread_mtx, 0,
1594 mtx_unlock(&pmp->wthread_mtx);
1595 pmp->wthread_td = NULL;
1599 * Cleanup our reference on ihidden.
1602 hammer2_inode_drop(pmp->ihidden);
1603 pmp->ihidden = NULL;
1607 * Cleanup our reference on iroot. iroot is (should) not be needed
1608 * by the flush code.
1611 cluster = &pmp->iroot->cluster;
1612 for (i = 0; i < pmp->iroot->cluster.nchains; ++i) {
1613 rchain = pmp->iroot->cluster.array[i];
1617 hammer2_vfs_unmount_hmp1(mp, hmp);
1619 atomic_clear_int(&rchain->flags, HAMMER2_CHAIN_MOUNTED);
1620 #if REPORT_REFS_ERRORS
1621 if (rchain->refs != 1)
1622 kprintf("PMP->RCHAIN %p REFS WRONG %d\n",
1623 rchain, rchain->refs);
1625 KKASSERT(rchain->refs == 1);
1627 hammer2_chain_drop(rchain);
1628 cluster->array[i] = NULL;
1629 hammer2_vfs_unmount_hmp2(mp, hmp);
1631 cluster->focus = NULL;
1633 #if REPORT_REFS_ERRORS
1634 if (pmp->iroot->refs != 1)
1635 kprintf("PMP->IROOT %p REFS WRONG %d\n",
1636 pmp->iroot, pmp->iroot->refs);
1638 KKASSERT(pmp->iroot->refs == 1);
1640 /* ref for pmp->iroot */
1641 hammer2_inode_drop(pmp->iroot);
1646 mp->mnt_data = NULL;
1648 kmalloc_destroy(&pmp->mmsg);
1649 kmalloc_destroy(&pmp->minode);
1651 kfree(pmp, M_HAMMER2);
1655 lockmgr(&hammer2_mntlk, LK_RELEASE);
1662 hammer2_vfs_unmount_hmp1(struct mount *mp, hammer2_mount_t *hmp)
1664 hammer2_mount_exlock(hmp);
1667 kprintf("hammer2_unmount hmp=%p pmpcnt=%d\n", hmp, hmp->pmp_count);
1669 kdmsg_iocom_uninit(&hmp->iocom); /* XXX chain depend deadlck? */
1672 * Flush any left over chains. The voldata lock is only used
1673 * to synchronize against HAMMER2_CHAIN_MODIFIED_AUX.
1675 * Flush twice to ensure that the freemap is completely
1676 * synchronized. If we only do it once the next mount's
1677 * recovery scan will have to do some fixups (which isn't
1678 * bad, but we don't want it to have to do it except when
1679 * recovering from a crash).
1681 hammer2_voldata_lock(hmp);
1682 if (((hmp->vchain.flags | hmp->fchain.flags) &
1683 HAMMER2_CHAIN_MODIFIED) ||
1684 hmp->vchain.update_xhi > hmp->vchain.update_xlo ||
1685 hmp->fchain.update_xhi > hmp->fchain.update_xlo) {
1686 hammer2_voldata_unlock(hmp);
1687 hammer2_vfs_sync(mp, MNT_WAIT);
1688 /*hammer2_vfs_sync(mp, MNT_WAIT);*/
1690 hammer2_voldata_unlock(hmp);
1692 if (hmp->pmp_count == 0) {
1693 if (((hmp->vchain.flags | hmp->fchain.flags) &
1694 HAMMER2_CHAIN_MODIFIED) ||
1695 hmp->vchain.update_xhi > hmp->vchain.update_xlo ||
1696 hmp->fchain.update_xhi > hmp->fchain.update_xlo) {
1697 kprintf("hammer2_unmount: chains left over "
1698 "after final sync\n");
1699 kprintf(" vchain %08x update_xlo/hi %08x/%08x\n",
1701 hmp->vchain.update_xlo,
1702 hmp->vchain.update_xhi);
1703 kprintf(" fchain %08x update_xhi/hi %08x/%08x\n",
1705 hmp->fchain.update_xlo,
1706 hmp->fchain.update_xhi);
1708 if (hammer2_debug & 0x0010)
1709 Debugger("entered debugger");
1716 hammer2_vfs_unmount_hmp2(struct mount *mp, hammer2_mount_t *hmp)
1718 hammer2_pfsmount_t *spmp;
1719 struct vnode *devvp;
1721 int ronly = ((mp->mnt_flag & MNT_RDONLY) != 0);
1724 * If no PFS's left drop the master hammer2_mount for the
1727 if (hmp->pmp_count == 0) {
1729 * Clean up SPMP and the super-root inode
1734 hammer2_inode_drop(spmp->iroot);
1738 kmalloc_destroy(&spmp->mmsg);
1739 kmalloc_destroy(&spmp->minode);
1740 kfree(spmp, M_HAMMER2);
1744 * Finish up with the device vnode
1746 if ((devvp = hmp->devvp) != NULL) {
1747 vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1748 vinvalbuf(devvp, (ronly ? 0 : V_SAVE), 0, 0);
1750 VOP_CLOSE(devvp, (ronly ? FREAD : FREAD|FWRITE), NULL);
1757 * Clear vchain/fchain flags that might prevent final cleanup
1760 if (hmp->vchain.flags & HAMMER2_CHAIN_MODIFIED) {
1761 atomic_clear_int(&hmp->vchain.flags,
1762 HAMMER2_CHAIN_MODIFIED);
1763 hammer2_chain_drop(&hmp->vchain);
1765 if (hmp->vchain.flags & HAMMER2_CHAIN_FLUSH_CREATE) {
1766 atomic_clear_int(&hmp->vchain.flags,
1767 HAMMER2_CHAIN_FLUSH_CREATE);
1768 hammer2_chain_drop(&hmp->vchain);
1770 if (hmp->vchain.flags & HAMMER2_CHAIN_FLUSH_DELETE) {
1771 atomic_clear_int(&hmp->vchain.flags,
1772 HAMMER2_CHAIN_FLUSH_DELETE);
1773 hammer2_chain_drop(&hmp->vchain);
1776 if (hmp->fchain.flags & HAMMER2_CHAIN_MODIFIED) {
1777 atomic_clear_int(&hmp->fchain.flags,
1778 HAMMER2_CHAIN_MODIFIED);
1779 hammer2_chain_drop(&hmp->fchain);
1781 if (hmp->fchain.flags & HAMMER2_CHAIN_FLUSH_CREATE) {
1782 atomic_clear_int(&hmp->fchain.flags,
1783 HAMMER2_CHAIN_FLUSH_CREATE);
1784 hammer2_chain_drop(&hmp->fchain);
1786 if (hmp->fchain.flags & HAMMER2_CHAIN_FLUSH_DELETE) {
1787 atomic_clear_int(&hmp->fchain.flags,
1788 HAMMER2_CHAIN_FLUSH_DELETE);
1789 hammer2_chain_drop(&hmp->fchain);
1793 * Final drop of embedded freemap root chain to
1794 * clean up fchain.core (fchain structure is not
1795 * flagged ALLOCATED so it is cleaned out and then
1798 hammer2_chain_drop(&hmp->fchain);
1801 * Final drop of embedded volume root chain to clean
1802 * up vchain.core (vchain structure is not flagged
1803 * ALLOCATED so it is cleaned out and then left to
1807 hammer2_dump_chain(&hmp->vchain, 0, &dumpcnt, 'v');
1809 hammer2_dump_chain(&hmp->fchain, 0, &dumpcnt, 'f');
1810 hammer2_mount_unlock(hmp);
1811 hammer2_chain_drop(&hmp->vchain);
1813 hammer2_io_cleanup(hmp, &hmp->iotree);
1814 if (hmp->iofree_count) {
1815 kprintf("io_cleanup: %d I/O's left hanging\n",
1819 TAILQ_REMOVE(&hammer2_mntlist, hmp, mntentry);
1820 kmalloc_destroy(&hmp->mchain);
1821 kfree(hmp, M_HAMMER2);
1823 hammer2_mount_unlock(hmp);
1829 hammer2_vfs_vget(struct mount *mp, struct vnode *dvp,
1830 ino_t ino, struct vnode **vpp)
1832 kprintf("hammer2_vget\n");
1833 return (EOPNOTSUPP);
1838 hammer2_vfs_root(struct mount *mp, struct vnode **vpp)
1840 hammer2_pfsmount_t *pmp;
1841 hammer2_cluster_t *cparent;
1846 if (pmp->iroot == NULL) {
1850 cparent = hammer2_inode_lock_sh(pmp->iroot);
1851 vp = hammer2_igetv(pmp->iroot, cparent, &error);
1852 hammer2_inode_unlock_sh(pmp->iroot, cparent);
1855 kprintf("vnodefail\n");
1864 * XXX incorporate ipdata->inode_quota and data_quota
1868 hammer2_vfs_statfs(struct mount *mp, struct statfs *sbp, struct ucred *cred)
1870 hammer2_pfsmount_t *pmp;
1871 hammer2_mount_t *hmp;
1874 KKASSERT(pmp->iroot->cluster.nchains >= 1);
1875 hmp = pmp->iroot->cluster.focus->hmp; /* XXX */
1877 mp->mnt_stat.f_files = pmp->inode_count;
1878 mp->mnt_stat.f_ffree = 0;
1879 mp->mnt_stat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
1880 mp->mnt_stat.f_bfree = hmp->voldata.allocator_free / HAMMER2_PBUFSIZE;
1881 mp->mnt_stat.f_bavail = mp->mnt_stat.f_bfree;
1883 *sbp = mp->mnt_stat;
1889 hammer2_vfs_statvfs(struct mount *mp, struct statvfs *sbp, struct ucred *cred)
1891 hammer2_pfsmount_t *pmp;
1892 hammer2_mount_t *hmp;
1895 KKASSERT(pmp->iroot->cluster.nchains >= 1);
1896 hmp = pmp->iroot->cluster.focus->hmp; /* XXX */
1898 mp->mnt_vstat.f_bsize = HAMMER2_PBUFSIZE;
1899 mp->mnt_vstat.f_files = pmp->inode_count;
1900 mp->mnt_vstat.f_ffree = 0;
1901 mp->mnt_vstat.f_blocks = hmp->voldata.allocator_size / HAMMER2_PBUFSIZE;
1902 mp->mnt_vstat.f_bfree = hmp->voldata.allocator_free / HAMMER2_PBUFSIZE;
1903 mp->mnt_vstat.f_bavail = mp->mnt_vstat.f_bfree;
1905 *sbp = mp->mnt_vstat;
1910 * Mount-time recovery (RW mounts)
1912 * Updates to the free block table are allowed to lag flushes by one
1913 * transaction. In case of a crash, then on a fresh mount we must do an
1914 * incremental scan of the last committed transaction id and make sure that
1915 * all related blocks have been marked allocated.
1917 * The super-root topology and each PFS has its own transaction id domain,
1918 * so we must track PFS boundary transitions.
1920 struct hammer2_recovery_elm {
1921 TAILQ_ENTRY(hammer2_recovery_elm) entry;
1922 hammer2_chain_t *chain;
1923 hammer2_tid_t sync_tid;
1926 TAILQ_HEAD(hammer2_recovery_list, hammer2_recovery_elm);
1928 struct hammer2_recovery_info {
1929 struct hammer2_recovery_list list;
1933 static int hammer2_recovery_scan(hammer2_trans_t *trans, hammer2_mount_t *hmp,
1934 hammer2_chain_t *parent,
1935 struct hammer2_recovery_info *info,
1936 hammer2_tid_t sync_tid);
1938 #define HAMMER2_RECOVERY_MAXDEPTH 10
1942 hammer2_recovery(hammer2_mount_t *hmp)
1944 hammer2_trans_t trans;
1945 struct hammer2_recovery_info info;
1946 struct hammer2_recovery_elm *elm;
1947 hammer2_chain_t *parent;
1948 hammer2_tid_t sync_tid;
1950 int cumulative_error = 0;
1952 hammer2_trans_init(&trans, hmp->spmp, 0);
1955 TAILQ_INIT(&info.list);
1957 parent = hammer2_chain_lookup_init(&hmp->vchain, 0);
1958 cumulative_error = hammer2_recovery_scan(&trans, hmp, parent,
1960 hammer2_chain_lookup_done(parent);
1962 while ((elm = TAILQ_FIRST(&info.list)) != NULL) {
1963 TAILQ_REMOVE(&info.list, elm, entry);
1964 parent = elm->chain;
1965 sync_tid = elm->sync_tid;
1966 kfree(elm, M_HAMMER2);
1968 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS |
1969 HAMMER2_RESOLVE_NOREF);
1970 error = hammer2_recovery_scan(&trans, hmp, parent,
1972 hammer2_chain_unlock(parent);
1974 cumulative_error = error;
1976 hammer2_trans_done(&trans);
1978 return cumulative_error;
1983 hammer2_recovery_scan(hammer2_trans_t *trans, hammer2_mount_t *hmp,
1984 hammer2_chain_t *parent,
1985 struct hammer2_recovery_info *info,
1986 hammer2_tid_t sync_tid)
1988 hammer2_chain_t *chain;
1990 int cumulative_error = 0;
1991 int pfs_boundary = 0;
1995 * Adjust freemap to ensure that the block(s) are marked allocated.
1997 if (parent->bref.type != HAMMER2_BREF_TYPE_VOLUME) {
1998 hammer2_freemap_adjust(trans, hmp, &parent->bref,
1999 HAMMER2_FREEMAP_DORECOVER);
2003 * Check type for recursive scan
2005 switch(parent->bref.type) {
2006 case HAMMER2_BREF_TYPE_VOLUME:
2007 /* data already instantiated */
2009 case HAMMER2_BREF_TYPE_INODE:
2011 * Must instantiate data for DIRECTDATA test and also
2014 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS);
2015 if (parent->data->ipdata.op_flags & HAMMER2_OPFLAG_DIRECTDATA) {
2016 /* not applicable to recovery scan */
2017 hammer2_chain_unlock(parent);
2020 if ((parent->data->ipdata.op_flags & HAMMER2_OPFLAG_PFSROOT) &&
2023 sync_tid = parent->bref.mirror_tid - 1;
2025 hammer2_chain_unlock(parent);
2027 case HAMMER2_BREF_TYPE_INDIRECT:
2029 * Must instantiate data for recursion
2031 hammer2_chain_lock(parent, HAMMER2_RESOLVE_ALWAYS);
2032 hammer2_chain_unlock(parent);
2034 case HAMMER2_BREF_TYPE_DATA:
2035 case HAMMER2_BREF_TYPE_FREEMAP:
2036 case HAMMER2_BREF_TYPE_FREEMAP_NODE:
2037 case HAMMER2_BREF_TYPE_FREEMAP_LEAF:
2038 /* not applicable to recovery scan */
2046 * Defer operation if depth limit reached or if we are crossing a
2049 if (info->depth >= HAMMER2_RECOVERY_MAXDEPTH || pfs_boundary) {
2050 struct hammer2_recovery_elm *elm;
2052 elm = kmalloc(sizeof(*elm), M_HAMMER2, M_ZERO | M_WAITOK);
2053 elm->chain = parent;
2054 elm->sync_tid = sync_tid;
2055 hammer2_chain_ref(parent);
2056 TAILQ_INSERT_TAIL(&info->list, elm, entry);
2057 /* unlocked by caller */
2064 * Recursive scan of the last flushed transaction only. We are
2065 * doing this without pmp assignments so don't leave the chains
2066 * hanging around after we are done with them.
2069 chain = hammer2_chain_scan(parent, NULL, &cache_index,
2070 HAMMER2_LOOKUP_NODATA);
2072 atomic_set_int(&chain->flags, HAMMER2_CHAIN_RELEASE);
2073 if (chain->bref.mirror_tid >= sync_tid) {
2075 error = hammer2_recovery_scan(trans, hmp, chain,
2079 cumulative_error = error;
2081 chain = hammer2_chain_scan(parent, chain, &cache_index,
2082 HAMMER2_LOOKUP_NODATA);
2085 return cumulative_error;
2089 * Sync the entire filesystem; this is called from the filesystem syncer
2090 * process periodically and whenever a user calls sync(1) on the hammer
2093 * Currently is actually called from the syncer! \o/
2095 * This task will have to snapshot the state of the dirty inode chain.
2096 * From that, it will have to make sure all of the inodes on the dirty
2097 * chain have IO initiated. We make sure that io is initiated for the root
2100 * If waitfor is set, we wait for media to acknowledge the new rootblock.
2102 * THINKS: side A vs side B, to have sync not stall all I/O?
2105 hammer2_vfs_sync(struct mount *mp, int waitfor)
2107 struct hammer2_sync_info info;
2108 hammer2_inode_t *iroot;
2109 hammer2_chain_t *chain;
2110 hammer2_chain_t *parent;
2111 hammer2_pfsmount_t *pmp;
2112 hammer2_mount_t *hmp;
2123 KKASSERT(iroot->pmp == pmp);
2126 * We can't acquire locks on existing vnodes while in a transaction
2127 * without risking a deadlock. This assumes that vfsync() can be
2128 * called without the vnode locked (which it can in DragonFly).
2129 * Otherwise we'd have to implement a multi-pass or flag the lock
2130 * failures and retry.
2132 * The reclamation code interlocks with the sync list's token
2133 * (by removing the vnode from the scan list) before unlocking
2134 * the inode, giving us time to ref the inode.
2136 /*flags = VMSC_GETVP;*/
2138 if (waitfor & MNT_LAZY)
2139 flags |= VMSC_ONEPASS;
2142 * Start our flush transaction. This does not return until all
2143 * concurrent transactions have completed and will prevent any
2144 * new transactions from running concurrently, except for the
2145 * buffer cache transactions.
2147 * For efficiency do an async pass before making sure with a
2148 * synchronous pass on all related buffer cache buffers. It
2149 * should theoretically not be possible for any new file buffers
2150 * to be instantiated during this sequence.
2152 hammer2_trans_init(&info.trans, pmp, HAMMER2_TRANS_ISFLUSH |
2153 HAMMER2_TRANS_PREFLUSH);
2154 hammer2_run_unlinkq(&info.trans, pmp);
2157 info.waitfor = MNT_NOWAIT;
2158 vsyncscan(mp, flags | VMSC_NOWAIT, hammer2_sync_scan2, &info);
2159 info.waitfor = MNT_WAIT;
2160 vsyncscan(mp, flags, hammer2_sync_scan2, &info);
2163 * Clear PREFLUSH. This prevents (or asserts on) any new logical
2164 * buffer cache flushes which occur during the flush. Device buffers
2169 if (info.error == 0 && (waitfor & MNT_WAIT)) {
2170 info.waitfor = waitfor;
2171 vsyncscan(mp, flags, hammer2_sync_scan2, &info);
2175 hammer2_bioq_sync(info.trans.pmp);
2176 atomic_clear_int(&info.trans.flags, HAMMER2_TRANS_PREFLUSH);
2181 * Flush all storage elements making up the cluster
2183 * We must also flush any deleted siblings because the super-root
2184 * flush won't do it for us. They all must be staged or the
2185 * super-root flush will not be able to update its block table
2188 * XXX currently done serially instead of concurrently
2190 for (i = 0; iroot && i < iroot->cluster.nchains; ++i) {
2191 chain = iroot->cluster.array[i];
2193 hammer2_chain_lock(chain, HAMMER2_RESOLVE_ALWAYS);
2194 hammer2_flush(&info.trans, &chain);
2195 hammer2_chain_unlock(chain);
2198 hammer2_chain_t *nchain;
2199 chain = TAILQ_FIRST(&chain->core->ownerq);
2200 hammer2_chain_ref(chain);
2202 hammer2_chain_lock(chain,
2203 HAMMER2_RESOLVE_ALWAYS);
2204 hammer2_flush(&info.trans, &chain);
2205 hammer2_chain_unlock(chain);
2206 nchain = TAILQ_NEXT(chain, core_entry);
2208 hammer2_chain_ref(nchain);
2209 hammer2_chain_drop(chain);
2215 hammer2_trans_done(&info.trans);
2219 * Flush all volume roots to synchronize PFS flushes with the
2220 * storage media. Use a super-root transaction for each one.
2222 * The flush code will detect super-root -> pfs-root chain
2223 * transitions using the last pfs-root flush.
2225 for (i = 0; iroot && i < iroot->cluster.nchains; ++i) {
2226 chain = iroot->cluster.array[i];
2233 * We only have to flush each hmp once
2235 for (j = i - 1; j >= 0; --j) {
2236 if (iroot->cluster.array[j] &&
2237 iroot->cluster.array[j]->hmp == hmp)
2242 hammer2_trans_spmp(&info.trans, hmp->spmp);
2245 * Force an update of the XID from the PFS root to the
2246 * topology root. We couldn't do this from the PFS
2247 * transaction because a SPMP transaction is needed.
2248 * This does not modify blocks, instead what it does is
2249 * allow the flush code to find the transition point and
2250 * then update on the way back up.
2252 parent = TAILQ_LAST(&chain->above->ownerq, h2_core_list);
2253 KKASSERT(chain->pmp != parent->pmp);
2254 hammer2_chain_setsubmod(&info.trans, parent);
2257 * Media mounts have two 'roots', vchain for the topology
2258 * and fchain for the free block table. Flush both.
2260 * Note that the topology and free block table are handled
2261 * independently, so the free block table can wind up being
2262 * ahead of the topology. We depend on the bulk free scan
2263 * code to deal with any loose ends.
2265 hammer2_chain_lock(&hmp->vchain, HAMMER2_RESOLVE_ALWAYS);
2266 hammer2_chain_lock(&hmp->fchain, HAMMER2_RESOLVE_ALWAYS);
2267 if ((hmp->fchain.flags & HAMMER2_CHAIN_MODIFIED) ||
2268 hmp->fchain.update_xhi > hmp->fchain.update_xlo) {
2270 * This will also modify vchain as a side effect,
2271 * mark vchain as modified now.
2273 hammer2_voldata_modify(hmp);
2274 chain = &hmp->fchain;
2275 hammer2_flush(&info.trans, &chain);
2276 KKASSERT(chain == &hmp->fchain);
2278 hammer2_chain_unlock(&hmp->fchain);
2279 hammer2_chain_unlock(&hmp->vchain);
2281 hammer2_chain_lock(&hmp->vchain, HAMMER2_RESOLVE_ALWAYS);
2282 if ((hmp->vchain.flags & HAMMER2_CHAIN_MODIFIED) ||
2283 hmp->vchain.update_xhi > hmp->vchain.update_xlo) {
2284 chain = &hmp->vchain;
2285 hammer2_flush(&info.trans, &chain);
2286 KKASSERT(chain == &hmp->vchain);
2291 hammer2_chain_unlock(&hmp->vchain);
2294 hammer2_chain_lock(&hmp->fchain, HAMMER2_RESOLVE_ALWAYS);
2295 if ((hmp->fchain.flags & HAMMER2_CHAIN_MODIFIED) ||
2296 hmp->fchain.update_xhi > hmp->fchain.update_xlo ||
2298 /* this will also modify vchain as a side effect */
2299 chain = &hmp->fchain;
2300 hammer2_flush(&info.trans, &chain);
2301 KKASSERT(chain == &hmp->fchain);
2303 hammer2_chain_unlock(&hmp->fchain);
2309 * We can't safely flush the volume header until we have
2310 * flushed any device buffers which have built up.
2312 * XXX this isn't being incremental
2314 vn_lock(hmp->devvp, LK_EXCLUSIVE | LK_RETRY);
2315 error = VOP_FSYNC(hmp->devvp, MNT_WAIT, 0);
2316 vn_unlock(hmp->devvp);
2319 * The flush code sets CHAIN_VOLUMESYNC to indicate that the
2320 * volume header needs synchronization via hmp->volsync.
2322 * XXX synchronize the flag & data with only this flush XXX
2325 (hmp->vchain.flags & HAMMER2_CHAIN_VOLUMESYNC)) {
2329 * Synchronize the disk before flushing the volume
2333 bp->b_bio1.bio_offset = 0;
2336 bp->b_cmd = BUF_CMD_FLUSH;
2337 bp->b_bio1.bio_done = biodone_sync;
2338 bp->b_bio1.bio_flags |= BIO_SYNC;
2339 vn_strategy(hmp->devvp, &bp->b_bio1);
2340 biowait(&bp->b_bio1, "h2vol");
2344 * Then we can safely flush the version of the
2345 * volume header synchronized by the flush code.
2347 i = hmp->volhdrno + 1;
2348 if (i >= HAMMER2_NUM_VOLHDRS)
2350 if (i * HAMMER2_ZONE_BYTES64 + HAMMER2_SEGSIZE >
2351 hmp->volsync.volu_size) {
2354 kprintf("sync volhdr %d %jd\n",
2355 i, (intmax_t)hmp->volsync.volu_size);
2356 bp = getblk(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
2357 HAMMER2_PBUFSIZE, 0, 0);
2358 atomic_clear_int(&hmp->vchain.flags,
2359 HAMMER2_CHAIN_VOLUMESYNC);
2360 bcopy(&hmp->volsync, bp->b_data, HAMMER2_PBUFSIZE);
2365 total_error = error;
2368 hammer2_trans_done(&info.trans);
2371 hammer2_trans_done(&info.trans);
2373 return (total_error);
2380 hammer2_sync_scan2(struct mount *mp, struct vnode *vp, void *data)
2382 struct hammer2_sync_info *info = data;
2383 hammer2_inode_t *ip;
2392 if (vp->v_type == VNON || vp->v_type == VBAD) {
2396 if ((ip->flags & HAMMER2_INODE_MODIFIED) == 0 &&
2397 RB_EMPTY(&vp->v_rbdirty_tree)) {
2403 * VOP_FSYNC will start a new transaction so replicate some code
2404 * here to do it inline (see hammer2_vop_fsync()).
2406 * WARNING: The vfsync interacts with the buffer cache and might
2407 * block, we can't hold the inode lock at that time.
2408 * However, we MUST ref ip before blocking to ensure that
2409 * it isn't ripped out from under us (since we do not
2410 * hold a lock on the vnode).
2412 hammer2_inode_ref(ip);
2413 atomic_clear_int(&ip->flags, HAMMER2_INODE_MODIFIED);
2415 vfsync(vp, MNT_NOWAIT, 1, NULL, NULL);
2417 hammer2_inode_drop(ip);
2421 info->error = error;
2428 hammer2_vfs_vptofh(struct vnode *vp, struct fid *fhp)
2435 hammer2_vfs_fhtovp(struct mount *mp, struct vnode *rootvp,
2436 struct fid *fhp, struct vnode **vpp)
2443 hammer2_vfs_checkexp(struct mount *mp, struct sockaddr *nam,
2444 int *exflagsp, struct ucred **credanonp)
2450 * Support code for hammer2_mount(). Read, verify, and install the volume
2451 * header into the HMP
2453 * XXX read four volhdrs and use the one with the highest TID whos CRC
2458 * XXX For filesystems w/ less than 4 volhdrs, make sure to not write to
2459 * nonexistant locations.
2461 * XXX Record selected volhdr and ring updates to each of 4 volhdrs
2465 hammer2_install_volume_header(hammer2_mount_t *hmp)
2467 hammer2_volume_data_t *vd;
2469 hammer2_crc32_t crc0, crc, bcrc0, bcrc;
2481 * There are up to 4 copies of the volume header (syncs iterate
2482 * between them so there is no single master). We don't trust the
2483 * volu_size field so we don't know precisely how large the filesystem
2484 * is, so depend on the OS to return an error if we go beyond the
2485 * block device's EOF.
2487 for (i = 0; i < HAMMER2_NUM_VOLHDRS; i++) {
2488 error = bread(hmp->devvp, i * HAMMER2_ZONE_BYTES64,
2489 HAMMER2_VOLUME_BYTES, &bp);
2496 vd = (struct hammer2_volume_data *) bp->b_data;
2497 if ((vd->magic != HAMMER2_VOLUME_ID_HBO) &&
2498 (vd->magic != HAMMER2_VOLUME_ID_ABO)) {
2504 if (vd->magic == HAMMER2_VOLUME_ID_ABO) {
2505 /* XXX: Reversed-endianness filesystem */
2506 kprintf("hammer2: reverse-endian filesystem detected");
2512 crc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT0];
2513 crc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC0_OFF,
2514 HAMMER2_VOLUME_ICRC0_SIZE);
2515 bcrc = vd->icrc_sects[HAMMER2_VOL_ICRC_SECT1];
2516 bcrc0 = hammer2_icrc32(bp->b_data + HAMMER2_VOLUME_ICRC1_OFF,
2517 HAMMER2_VOLUME_ICRC1_SIZE);
2518 if ((crc0 != crc) || (bcrc0 != bcrc)) {
2519 kprintf("hammer2 volume header crc "
2520 "mismatch copy #%d %08x/%08x\n",
2527 if (valid == 0 || hmp->voldata.mirror_tid < vd->mirror_tid) {
2536 hmp->volsync = hmp->voldata;
2538 if (error_reported || bootverbose || 1) { /* 1/DEBUG */
2539 kprintf("hammer2: using volume header #%d\n",
2544 kprintf("hammer2: no valid volume headers found!\n");
2550 * Reconnect using the passed file pointer. The caller must ref the
2554 hammer2_cluster_reconnect(hammer2_mount_t *hmp, struct file *fp)
2557 const char *name = "disk-volume";
2560 * Closes old comm descriptor, kills threads, cleans up
2561 * states, then installs the new descriptor and creates
2564 kdmsg_iocom_reconnect(&hmp->iocom, fp, "hammer2");
2567 * Setup LNK_CONN fields for autoinitiated state machine. We
2568 * will use SPANs to advertise multiple PFSs so only pass the
2569 * fsid and HAMMER2_PFSTYPE_SUPROOT for the AUTOCONN.
2571 * We are not initiating a LNK_SPAN so we do not have to set-up
2572 * iocom.auto_lnk_span.
2574 bzero(&hmp->iocom.auto_lnk_conn.pfs_clid,
2575 sizeof(hmp->iocom.auto_lnk_conn.pfs_clid));
2576 hmp->iocom.auto_lnk_conn.pfs_fsid = hmp->voldata.fsid;
2577 hmp->iocom.auto_lnk_conn.pfs_type = HAMMER2_PFSTYPE_SUPROOT;
2578 hmp->iocom.auto_lnk_conn.proto_version = DMSG_SPAN_PROTO_1;
2580 hmp->iocom.auto_lnk_conn.peer_type = hmp->voldata.peer_type;
2582 hmp->iocom.auto_lnk_conn.peer_type = DMSG_PEER_HAMMER2;
2585 * Filter adjustment. Clients do not need visibility into other
2586 * clients (otherwise millions of clients would present a serious
2587 * problem). The fs_label also serves to restrict the namespace.
2589 hmp->iocom.auto_lnk_conn.peer_mask = 1LLU << DMSG_PEER_HAMMER2;
2590 hmp->iocom.auto_lnk_conn.pfs_mask = (uint64_t)-1;
2593 switch (ipdata->pfs_type) {
2594 case DMSG_PFSTYPE_CLIENT:
2595 hmp->iocom.auto_lnk_conn.peer_mask &=
2596 ~(1LLU << DMSG_PFSTYPE_CLIENT);
2603 name_len = strlen(name);
2604 if (name_len >= sizeof(hmp->iocom.auto_lnk_conn.fs_label))
2605 name_len = sizeof(hmp->iocom.auto_lnk_conn.fs_label) - 1;
2606 bcopy(name, hmp->iocom.auto_lnk_conn.fs_label, name_len);
2607 hmp->iocom.auto_lnk_conn.fs_label[name_len] = 0;
2609 kdmsg_iocom_autoinitiate(&hmp->iocom, hammer2_autodmsg);
2613 hammer2_rcvdmsg(kdmsg_msg_t *msg)
2615 kprintf("RCVMSG %08x\n", msg->tcmd);
2618 case DMSG_DBG_SHELL:
2621 * Execute shell command (not supported atm)
2623 kdmsg_msg_result(msg, DMSG_ERR_NOSUPP);
2625 case DMSG_DBG_SHELL | DMSGF_REPLY:
2629 if (msg->aux_data) {
2630 msg->aux_data[msg->aux_size - 1] = 0;
2631 kprintf("HAMMER2 DBG: %s\n", msg->aux_data);
2636 * Unsupported message received. We only need to
2637 * reply if it's a transaction in order to close our end.
2638 * Ignore any one-way messages or any further messages
2639 * associated with the transaction.
2641 * NOTE: This case also includes DMSG_LNK_ERROR messages
2642 * which might be one-way, replying to those would
2643 * cause an infinite ping-pong.
2645 if (msg->any.head.cmd & DMSGF_CREATE)
2646 kdmsg_msg_reply(msg, DMSG_ERR_NOSUPP);
2653 * This function is called after KDMSG has automatically handled processing
2654 * of a LNK layer message (typically CONN, SPAN, or CIRC).
2656 * We tag off the LNK_CONN to trigger our LNK_VOLCONF messages which
2657 * advertises all available hammer2 super-root volumes.
2660 hammer2_autodmsg(kdmsg_msg_t *msg)
2662 hammer2_mount_t *hmp = msg->iocom->handle;
2665 kprintf("RCAMSG %08x\n", msg->tcmd);
2668 case DMSG_LNK_CONN | DMSGF_CREATE | DMSGF_REPLY:
2669 case DMSG_LNK_CONN | DMSGF_CREATE | DMSGF_DELETE | DMSGF_REPLY:
2670 if (msg->any.head.cmd & DMSGF_CREATE) {
2671 kprintf("HAMMER2: VOLDATA DUMP\n");
2674 * Dump the configuration stored in the volume header.
2675 * This will typically be import/export access rights,
2676 * master encryption keys (encrypted), etc.
2678 hammer2_voldata_lock(hmp);
2680 while (copyid < HAMMER2_COPYID_COUNT) {
2681 if (hmp->voldata.copyinfo[copyid].copyid)
2682 hammer2_volconf_update(hmp, copyid);
2685 hammer2_voldata_unlock(hmp);
2687 kprintf("HAMMER2: INITIATE SPANs\n");
2688 hammer2_update_spans(hmp);
2690 if ((msg->any.head.cmd & DMSGF_DELETE) &&
2691 msg->state && (msg->state->txcmd & DMSGF_DELETE) == 0) {
2692 kprintf("HAMMER2: CONN WAS TERMINATED\n");
2701 * Update LNK_SPAN state
2704 hammer2_update_spans(hammer2_mount_t *hmp)
2706 const hammer2_inode_data_t *ipdata;
2707 hammer2_cluster_t *cparent;
2708 hammer2_cluster_t *cluster;
2709 hammer2_pfsmount_t *spmp;
2710 hammer2_key_t key_next;
2716 * Lookup mount point under the media-localized super-root.
2718 * cluster->pmp will incorrectly point to spmp and must be fixed
2722 cparent = hammer2_inode_lock_ex(spmp->iroot);
2723 cluster = hammer2_cluster_lookup(cparent, &key_next,
2728 if (hammer2_cluster_type(cluster) != HAMMER2_BREF_TYPE_INODE)
2730 ipdata = &hammer2_cluster_data(cluster)->ipdata;
2731 kprintf("UPDATE SPANS: %s\n", ipdata->filename);
2733 rmsg = kdmsg_msg_alloc(&hmp->iocom, NULL,
2734 DMSG_LNK_SPAN | DMSGF_CREATE,
2735 hammer2_lnk_span_reply, NULL);
2736 rmsg->any.lnk_span.pfs_clid = ipdata->pfs_clid;
2737 rmsg->any.lnk_span.pfs_fsid = ipdata->pfs_fsid;
2738 rmsg->any.lnk_span.pfs_type = ipdata->pfs_type;
2739 rmsg->any.lnk_span.peer_type = DMSG_PEER_HAMMER2;
2740 rmsg->any.lnk_span.proto_version = DMSG_SPAN_PROTO_1;
2741 name_len = ipdata->name_len;
2742 if (name_len >= sizeof(rmsg->any.lnk_span.fs_label))
2743 name_len = sizeof(rmsg->any.lnk_span.fs_label) - 1;
2744 bcopy(ipdata->filename, rmsg->any.lnk_span.fs_label, name_len);
2746 kdmsg_msg_write(rmsg);
2748 cluster = hammer2_cluster_next(cparent, cluster,
2754 hammer2_inode_unlock_ex(spmp->iroot, cparent);
2759 hammer2_lnk_span_reply(kdmsg_state_t *state, kdmsg_msg_t *msg)
2761 if ((state->txcmd & DMSGF_DELETE) == 0 &&
2762 (msg->any.head.cmd & DMSGF_DELETE)) {
2763 kdmsg_msg_reply(msg, 0);
2769 * Volume configuration updates are passed onto the userland service
2770 * daemon via the open LNK_CONN transaction.
2773 hammer2_volconf_update(hammer2_mount_t *hmp, int index)
2777 /* XXX interlock against connection state termination */
2778 kprintf("volconf update %p\n", hmp->iocom.conn_state);
2779 if (hmp->iocom.conn_state) {
2780 kprintf("TRANSMIT VOLCONF VIA OPEN CONN TRANSACTION\n");
2781 msg = kdmsg_msg_alloc_state(hmp->iocom.conn_state,
2782 DMSG_LNK_HAMMER2_VOLCONF,
2784 H2_LNK_VOLCONF(msg)->copy = hmp->voldata.copyinfo[index];
2785 H2_LNK_VOLCONF(msg)->mediaid = hmp->voldata.fsid;
2786 H2_LNK_VOLCONF(msg)->index = index;
2787 kdmsg_msg_write(msg);
2792 * This handles hysteresis on regular file flushes. Because the BIOs are
2793 * routed to a thread it is possible for an excessive number to build up
2794 * and cause long front-end stalls long before the runningbuffspace limit
2795 * is hit, so we implement hammer2_flush_pipe to control the
2798 * This is a particular problem when compression is used.
2801 hammer2_lwinprog_ref(hammer2_pfsmount_t *pmp)
2803 atomic_add_int(&pmp->count_lwinprog, 1);
2807 hammer2_lwinprog_drop(hammer2_pfsmount_t *pmp)
2811 lwinprog = atomic_fetchadd_int(&pmp->count_lwinprog, -1);
2812 if ((lwinprog & HAMMER2_LWINPROG_WAITING) &&
2813 (lwinprog & HAMMER2_LWINPROG_MASK) <= hammer2_flush_pipe * 2 / 3) {
2814 atomic_clear_int(&pmp->count_lwinprog,
2815 HAMMER2_LWINPROG_WAITING);
2816 wakeup(&pmp->count_lwinprog);
2821 hammer2_lwinprog_wait(hammer2_pfsmount_t *pmp)
2826 lwinprog = pmp->count_lwinprog;
2828 if ((lwinprog & HAMMER2_LWINPROG_MASK) < hammer2_flush_pipe)
2830 tsleep_interlock(&pmp->count_lwinprog, 0);
2831 atomic_set_int(&pmp->count_lwinprog, HAMMER2_LWINPROG_WAITING);
2832 lwinprog = pmp->count_lwinprog;
2833 if ((lwinprog & HAMMER2_LWINPROG_MASK) < hammer2_flush_pipe)
2835 tsleep(&pmp->count_lwinprog, PINTERLOCKED, "h2wpipe", hz);
2840 * Manage excessive memory resource use for chain and related
2844 hammer2_pfs_memory_wait(hammer2_pfsmount_t *pmp)
2854 * Atomic check condition and wait. Also do an early speedup of
2855 * the syncer to try to avoid hitting the wait.
2858 waiting = pmp->inmem_dirty_chains;
2860 count = waiting & HAMMER2_DIRTYCHAIN_MASK;
2862 limit = pmp->mp->mnt_nvnodelistsize / 10;
2863 if (limit < hammer2_limit_dirty_chains)
2864 limit = hammer2_limit_dirty_chains;
2869 if ((int)(ticks - zzticks) > hz) {
2871 kprintf("count %ld %ld\n", count, limit);
2876 * Block if there are too many dirty chains present, wait
2877 * for the flush to clean some out.
2879 if (count > limit) {
2880 tsleep_interlock(&pmp->inmem_dirty_chains, 0);
2881 if (atomic_cmpset_long(&pmp->inmem_dirty_chains,
2883 waiting | HAMMER2_DIRTYCHAIN_WAITING)) {
2884 speedup_syncer(pmp->mp);
2885 tsleep(&pmp->inmem_dirty_chains, PINTERLOCKED,
2888 continue; /* loop on success or fail */
2892 * Try to start an early flush before we are forced to block.
2894 if (count > limit * 7 / 10)
2895 speedup_syncer(pmp->mp);
2901 hammer2_pfs_memory_inc(hammer2_pfsmount_t *pmp)
2904 atomic_add_long(&pmp->inmem_dirty_chains, 1);
2908 hammer2_pfs_memory_wakeup(hammer2_pfsmount_t *pmp)
2916 waiting = pmp->inmem_dirty_chains;
2918 if (atomic_cmpset_long(&pmp->inmem_dirty_chains,
2921 ~HAMMER2_DIRTYCHAIN_WAITING)) {
2926 if (waiting & HAMMER2_DIRTYCHAIN_WAITING)
2927 wakeup(&pmp->inmem_dirty_chains);
2934 hammer2_dump_chain(hammer2_chain_t *chain, int tab, int *countp, char pfx)
2936 hammer2_chain_t *scan;
2937 hammer2_chain_t *first_parent;
2941 kprintf("%*.*s...\n", tab, tab, "");
2946 first_parent = chain->core ? TAILQ_FIRST(&chain->core->ownerq) : NULL;
2947 kprintf("%*.*s%c-chain %p.%d %016jx/%d mir=%016jx\n",
2949 chain, chain->bref.type,
2950 chain->bref.key, chain->bref.keybits,
2951 chain->bref.mirror_tid);
2953 kprintf("%*.*s [%08x] (%s) mod=%08x del=%08x "
2954 "lo=%08x hi=%08x refs=%d\n",
2957 ((chain->bref.type == HAMMER2_BREF_TYPE_INODE &&
2958 chain->data) ? (char *)chain->data->ipdata.filename : "?"),
2965 kprintf("%*.*s core %p [%08x]",
2967 chain->core, (chain->core ? chain->core->flags : 0));
2970 kprintf("\n%*.*s fp=%p np=%p [fpflags %08x fprefs %d",
2973 (first_parent ? TAILQ_NEXT(first_parent, core_entry) :
2975 first_parent->flags,
2976 first_parent->refs);
2977 if (chain->core == NULL || RB_EMPTY(&chain->core->rbtree))
2982 RB_FOREACH(scan, hammer2_chain_tree, &chain->core->rbtree)
2983 hammer2_dump_chain(scan, tab + 4, countp, 'a');
2984 RB_FOREACH(scan, hammer2_chain_tree, &chain->core->dbtree)
2985 hammer2_dump_chain(scan, tab + 4, countp, 'r');
2986 TAILQ_FOREACH(scan, &chain->core->dbq, db_entry)
2987 hammer2_dump_chain(scan, tab + 4, countp, 'd');
2989 if (chain->core && !RB_EMPTY(&chain->core->rbtree)) {
2990 if (chain->bref.type == HAMMER2_BREF_TYPE_INODE && chain->data)
2991 kprintf("%*.*s}(%s)\n", tab, tab, "",
2992 chain->data->ipdata.filename);
2994 kprintf("%*.*s}\n", tab, tab, "");