Home | History | Annotate | Line # | Download | only in rumpvfs
rump_vfs.c revision 1.12.2.1 
      1 /*	$NetBSD: rump_vfs.c,v 1.12.2.1 2009/05/13 17:22:58 jym Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 2008 Antti Kantee.  All Rights Reserved.
      5  *
      6  * Development of this software was supported by the
      7  * Finnish Cultural Foundation.
      8  *
      9  * Redistribution and use in source and binary forms, with or without
     10  * modification, are permitted provided that the following conditions
     11  * are met:
     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 the
     16  *    documentation and/or other materials provided with the distribution.
     17  *
     18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
     19  * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
     20  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
     21  * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
     22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
     24  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     28  * SUCH DAMAGE.
     29  */
     30 
     31 #include <sys/cdefs.h>
     32 __KERNEL_RCSID(0, "$NetBSD");
     33 
     34 #include <sys/param.h>
     35 #include <sys/buf.h>
     36 #include <sys/conf.h>
     37 #include <sys/evcnt.h>
     38 #include <sys/filedesc.h>
     39 #include <sys/lockf.h>
     40 #include <sys/kthread.h>
     41 #include <sys/module.h>
     42 #include <sys/namei.h>
     43 #include <sys/queue.h>
     44 #include <sys/vfs_syscalls.h>
     45 #include <sys/vnode.h>
     46 #include <sys/wapbl.h>
     47 
     48 #include <miscfs/specfs/specdev.h>
     49 #include <miscfs/syncfs/syncfs.h>
     50 
     51 #include <rump/rump.h>
     52 #include <rump/rumpuser.h>
     53 
     54 #include "rump_private.h"
     55 #include "rump_vfs_private.h"
     56 
     57 struct fakeblk {
     58 	char path[MAXPATHLEN];
     59 	LIST_ENTRY(fakeblk) entries;
     60 };
     61 static LIST_HEAD(, fakeblk) fakeblks = LIST_HEAD_INITIALIZER(fakeblks);
     62 
     63 static struct cwdinfo rump_cwdi;
     64 
     65 static void rump_rcvp_lwpset(struct vnode *, struct vnode *, struct lwp *);
     66 
     67 static void
     68 pvfs_init(struct proc *p)
     69 {
     70 
     71 	p->p_cwdi = cwdinit();
     72 }
     73 
     74 static void
     75 pvfs_rele(struct proc *p)
     76 {
     77 
     78 	cwdfree(p->p_cwdi);
     79 }
     80 
     81 void
     82 rump_vfs_init(void)
     83 {
     84 	char buf[64];
     85 	int error;
     86 
     87 	dovfsusermount = 1;
     88 
     89 	if (rumpuser_getenv("RUMP_NVNODES", buf, sizeof(buf), &error) == 0) {
     90 		desiredvnodes = strtoul(buf, NULL, 10);
     91 	} else {
     92 		desiredvnodes = 1<<16;
     93 	}
     94 
     95 	rumpblk_init();
     96 
     97 	cache_cpu_init(&rump_cpu);
     98 	vfsinit();
     99 	bufinit();
    100 	wapbl_init();
    101 	cwd_sys_init();
    102 	lf_init();
    103 
    104 	rumpuser_bioinit(rump_biodone);
    105 	rumpfs_init();
    106 
    107 	rump_proc_vfs_init = pvfs_init;
    108 	rump_proc_vfs_release = pvfs_rele;
    109 
    110 	/* bootstrap cwdi */
    111 	rw_init(&rump_cwdi.cwdi_lock);
    112 	rump_cwdi.cwdi_cdir = rootvnode;
    113 	vref(rump_cwdi.cwdi_cdir);
    114 	proc0.p_cwdi = &rump_cwdi;
    115 	proc0.p_cwdi = cwdinit();
    116 
    117 	if (rump_threads) {
    118 		int rv;
    119 
    120 		if ((rv = kthread_create(PRI_IOFLUSH, KTHREAD_MPSAFE, NULL,
    121 		    sched_sync, NULL, NULL, "ioflush")) != 0)
    122 			panic("syncer thread create failed: %d", rv);
    123 	} else {
    124 		syncdelay = 0;
    125 	}
    126 }
    127 
    128 struct mount *
    129 rump_mnt_init(struct vfsops *vfsops, int mntflags)
    130 {
    131 	struct mount *mp;
    132 
    133 	mp = kmem_zalloc(sizeof(struct mount), KM_SLEEP);
    134 
    135 	mp->mnt_op = vfsops;
    136 	mp->mnt_flag = mntflags;
    137 	TAILQ_INIT(&mp->mnt_vnodelist);
    138 	rw_init(&mp->mnt_unmounting);
    139 	mutex_init(&mp->mnt_updating, MUTEX_DEFAULT, IPL_NONE);
    140 	mutex_init(&mp->mnt_renamelock, MUTEX_DEFAULT, IPL_NONE);
    141 	mp->mnt_refcnt = 1;
    142 	mp->mnt_vnodecovered = rootvnode;
    143 
    144 	mount_initspecific(mp);
    145 
    146 	return mp;
    147 }
    148 
    149 int
    150 rump_mnt_mount(struct mount *mp, const char *path, void *data, size_t *dlen)
    151 {
    152 	struct vnode *rvp;
    153 	int rv;
    154 
    155 	rv = VFS_MOUNT(mp, path, data, dlen);
    156 	if (rv)
    157 		return rv;
    158 
    159 	(void) VFS_STATVFS(mp, &mp->mnt_stat);
    160 	rv = VFS_START(mp, 0);
    161 	if (rv) {
    162 		VFS_UNMOUNT(mp, MNT_FORCE);
    163 		return rv;
    164 	}
    165 
    166 	/*
    167 	 * XXX: set a root for lwp0.  This is strictly not correct,
    168 	 * but makes things work for single fs case without having
    169 	 * to manually call rump_rcvp_set().
    170 	 */
    171 	VFS_ROOT(mp, &rvp);
    172 	rump_rcvp_lwpset(rvp, rvp, &lwp0);
    173 	vput(rvp);
    174 
    175 	return rv;
    176 }
    177 
    178 void
    179 rump_mnt_destroy(struct mount *mp)
    180 {
    181 
    182 	/* See rcvp XXX above */
    183 	rump_cwdi.cwdi_rdir = NULL;
    184 	vref(rootvnode);
    185 	rump_cwdi.cwdi_cdir = rootvnode;
    186 
    187 	mount_finispecific(mp);
    188 	kmem_free(mp, sizeof(*mp));
    189 }
    190 
    191 struct componentname *
    192 rump_makecn(u_long nameiop, u_long flags, const char *name, size_t namelen,
    193 	kauth_cred_t creds, struct lwp *l)
    194 {
    195 	struct componentname *cnp;
    196 	const char *cp = NULL;
    197 
    198 	cnp = kmem_zalloc(sizeof(struct componentname), KM_SLEEP);
    199 
    200 	cnp->cn_nameiop = nameiop;
    201 	cnp->cn_flags = flags | HASBUF;
    202 
    203 	cnp->cn_pnbuf = PNBUF_GET();
    204 	strcpy(cnp->cn_pnbuf, name);
    205 	cnp->cn_nameptr = cnp->cn_pnbuf;
    206 	cnp->cn_namelen = namelen;
    207 	cnp->cn_hash = namei_hash(name, &cp);
    208 
    209 	cnp->cn_cred = creds;
    210 
    211 	return cnp;
    212 }
    213 
    214 void
    215 rump_freecn(struct componentname *cnp, int flags)
    216 {
    217 
    218 	if (flags & RUMPCN_FREECRED)
    219 		rump_cred_put(cnp->cn_cred);
    220 
    221 	if (cnp->cn_flags & SAVENAME) {
    222 		if (flags & RUMPCN_ISLOOKUP || cnp->cn_flags & SAVESTART)
    223 			PNBUF_PUT(cnp->cn_pnbuf);
    224 	} else {
    225 		PNBUF_PUT(cnp->cn_pnbuf);
    226 	}
    227 	kmem_free(cnp, sizeof(*cnp));
    228 }
    229 
    230 /* hey baby, what's your namei? */
    231 int
    232 rump_namei(uint32_t op, uint32_t flags, const char *namep,
    233 	struct vnode **dvpp, struct vnode **vpp, struct componentname **cnpp)
    234 {
    235 	struct nameidata nd;
    236 	int rv;
    237 
    238 	NDINIT(&nd, op, flags, UIO_SYSSPACE, namep);
    239 	rv = namei(&nd);
    240 	if (rv)
    241 		return rv;
    242 
    243 	if (dvpp) {
    244 		KASSERT(flags & LOCKPARENT);
    245 		*dvpp = nd.ni_dvp;
    246 	} else {
    247 		KASSERT((flags & LOCKPARENT) == 0);
    248 	}
    249 
    250 	if (vpp) {
    251 		*vpp = nd.ni_vp;
    252 	} else {
    253 		if (nd.ni_vp) {
    254 			if (flags & LOCKLEAF)
    255 				vput(nd.ni_vp);
    256 			else
    257 				vrele(nd.ni_vp);
    258 		}
    259 	}
    260 
    261 	if (cnpp) {
    262 		struct componentname *cnp;
    263 
    264 		cnp = kmem_alloc(sizeof(*cnp), KM_SLEEP);
    265 		memcpy(cnp, &nd.ni_cnd, sizeof(*cnp));
    266 		*cnpp = cnp;
    267 	} else if (nd.ni_cnd.cn_flags & HASBUF) {
    268 		panic("%s: pathbuf mismatch", __func__);
    269 	}
    270 
    271 	return rv;
    272 }
    273 
    274 static struct fakeblk *
    275 _rump_fakeblk_find(const char *path)
    276 {
    277 	char buf[MAXPATHLEN];
    278 	struct fakeblk *fblk;
    279 	int error;
    280 
    281 	if (rumpuser_realpath(path, buf, &error) == NULL)
    282 		return NULL;
    283 
    284 	LIST_FOREACH(fblk, &fakeblks, entries)
    285 		if (strcmp(fblk->path, buf) == 0)
    286 			return fblk;
    287 
    288 	return NULL;
    289 }
    290 
    291 int
    292 rump_fakeblk_register(const char *path)
    293 {
    294 	char buf[MAXPATHLEN];
    295 	struct fakeblk *fblk;
    296 	int error;
    297 
    298 	if (_rump_fakeblk_find(path))
    299 		return EEXIST;
    300 
    301 	if (rumpuser_realpath(path, buf, &error) == NULL)
    302 		return error;
    303 
    304 	fblk = kmem_alloc(sizeof(struct fakeblk), KM_NOSLEEP);
    305 	if (fblk == NULL)
    306 		return ENOMEM;
    307 
    308 	strlcpy(fblk->path, buf, MAXPATHLEN);
    309 	LIST_INSERT_HEAD(&fakeblks, fblk, entries);
    310 
    311 	return 0;
    312 }
    313 
    314 int
    315 rump_fakeblk_find(const char *path)
    316 {
    317 
    318 	return _rump_fakeblk_find(path) != NULL;
    319 }
    320 
    321 void
    322 rump_fakeblk_deregister(const char *path)
    323 {
    324 	struct fakeblk *fblk;
    325 
    326 	fblk = _rump_fakeblk_find(path);
    327 	if (fblk == NULL)
    328 		return;
    329 
    330 	LIST_REMOVE(fblk, entries);
    331 	kmem_free(fblk, sizeof(*fblk));
    332 }
    333 
    334 void
    335 rump_getvninfo(struct vnode *vp, enum vtype *vtype, voff_t *vsize, dev_t *vdev)
    336 {
    337 
    338 	*vtype = vp->v_type;
    339 	*vsize = vp->v_size;
    340 	if (vp->v_specnode)
    341 		*vdev = vp->v_rdev;
    342 	else
    343 		*vdev = 0;
    344 }
    345 
    346 struct vfsops *
    347 rump_vfslist_iterate(struct vfsops *ops)
    348 {
    349 
    350 	if (ops == NULL)
    351 		return LIST_FIRST(&vfs_list);
    352 	else
    353 		return LIST_NEXT(ops, vfs_list);
    354 }
    355 
    356 struct vfsops *
    357 rump_vfs_getopsbyname(const char *name)
    358 {
    359 
    360 	return vfs_getopsbyname(name);
    361 }
    362 
    363 int
    364 rump_vfs_getmp(const char *path, struct mount **mpp)
    365 {
    366 	struct nameidata nd;
    367 	struct vnode *vp;
    368 	int rv;
    369 
    370 	NDINIT(&nd, LOOKUP, FOLLOW | TRYEMULROOT, UIO_USERSPACE, path);
    371 	if ((rv = namei(&nd)) != 0)
    372 		return rv;
    373 	vp = nd.ni_vp;
    374 
    375 	*mpp = vp->v_mount;
    376 	vrele(vp);
    377 	return 0;
    378 }
    379 
    380 struct vattr*
    381 rump_vattr_init(void)
    382 {
    383 	struct vattr *vap;
    384 
    385 	vap = kmem_alloc(sizeof(struct vattr), KM_SLEEP);
    386 	vattr_null(vap);
    387 
    388 	return vap;
    389 }
    390 
    391 void
    392 rump_vattr_settype(struct vattr *vap, enum vtype vt)
    393 {
    394 
    395 	vap->va_type = vt;
    396 }
    397 
    398 void
    399 rump_vattr_setmode(struct vattr *vap, mode_t mode)
    400 {
    401 
    402 	vap->va_mode = mode;
    403 }
    404 
    405 void
    406 rump_vattr_setrdev(struct vattr *vap, dev_t dev)
    407 {
    408 
    409 	vap->va_rdev = dev;
    410 }
    411 
    412 void
    413 rump_vattr_free(struct vattr *vap)
    414 {
    415 
    416 	kmem_free(vap, sizeof(*vap));
    417 }
    418 
    419 void
    420 rump_vp_incref(struct vnode *vp)
    421 {
    422 
    423 	mutex_enter(&vp->v_interlock);
    424 	++vp->v_usecount;
    425 	mutex_exit(&vp->v_interlock);
    426 }
    427 
    428 int
    429 rump_vp_getref(struct vnode *vp)
    430 {
    431 
    432 	return vp->v_usecount;
    433 }
    434 
    435 void
    436 rump_vp_decref(struct vnode *vp)
    437 {
    438 
    439 	mutex_enter(&vp->v_interlock);
    440 	--vp->v_usecount;
    441 	mutex_exit(&vp->v_interlock);
    442 }
    443 
    444 /*
    445  * Really really recycle with a cherry on top.  We should be
    446  * extra-sure we can do this.  For example with p2k there is
    447  * no problem, since puffs in the kernel takes care of refcounting
    448  * for us.
    449  */
    450 void
    451 rump_vp_recycle_nokidding(struct vnode *vp)
    452 {
    453 
    454 	mutex_enter(&vp->v_interlock);
    455 	vp->v_usecount = 1;
    456 	/*
    457 	 * XXX: NFS holds a reference to the root vnode, so don't clean
    458 	 * it out.  This is very wrong, but fixing it properly would
    459 	 * take too much effort for now
    460 	 */
    461 	if (vp->v_tag == VT_NFS && vp->v_vflag & VV_ROOT) {
    462 		mutex_exit(&vp->v_interlock);
    463 		return;
    464 	}
    465 	vclean(vp, DOCLOSE);
    466 	vrelel(vp, 0);
    467 }
    468 
    469 void
    470 rump_vp_rele(struct vnode *vp)
    471 {
    472 
    473 	vrele(vp);
    474 }
    475 
    476 void
    477 rump_vp_interlock(struct vnode *vp)
    478 {
    479 
    480 	mutex_enter(&vp->v_interlock);
    481 }
    482 
    483 int
    484 rump_vfs_unmount(struct mount *mp, int mntflags)
    485 {
    486 #if 0
    487 	struct evcnt *ev;
    488 
    489 	printf("event counters:\n");
    490 	TAILQ_FOREACH(ev, &allevents, ev_list)
    491 		printf("%s: %llu\n", ev->ev_name, ev->ev_count);
    492 #endif
    493 
    494 	return VFS_UNMOUNT(mp, mntflags);
    495 }
    496 
    497 int
    498 rump_vfs_root(struct mount *mp, struct vnode **vpp, int lock)
    499 {
    500 	int rv;
    501 
    502 	rv = VFS_ROOT(mp, vpp);
    503 	if (rv)
    504 		return rv;
    505 
    506 	if (!lock)
    507 		VOP_UNLOCK(*vpp, 0);
    508 
    509 	return 0;
    510 }
    511 
    512 int
    513 rump_vfs_statvfs(struct mount *mp, struct statvfs *sbp)
    514 {
    515 
    516 	return VFS_STATVFS(mp, sbp);
    517 }
    518 
    519 int
    520 rump_vfs_sync(struct mount *mp, int wait, kauth_cred_t cred)
    521 {
    522 
    523 	return VFS_SYNC(mp, wait ? MNT_WAIT : MNT_NOWAIT, cred);
    524 }
    525 
    526 int
    527 rump_vfs_fhtovp(struct mount *mp, struct fid *fid, struct vnode **vpp)
    528 {
    529 
    530 	return VFS_FHTOVP(mp, fid, vpp);
    531 }
    532 
    533 int
    534 rump_vfs_vptofh(struct vnode *vp, struct fid *fid, size_t *fidsize)
    535 {
    536 
    537 	return VFS_VPTOFH(vp, fid, fidsize);
    538 }
    539 
    540 /*ARGSUSED*/
    541 void
    542 rump_vfs_syncwait(struct mount *mp)
    543 {
    544 	int n;
    545 
    546 	n = buf_syncwait();
    547 	if (n)
    548 		printf("syncwait: unsynced buffers: %d\n", n);
    549 }
    550 
    551 void
    552 rump_biodone(void *arg, size_t count, int error)
    553 {
    554 	struct buf *bp = arg;
    555 
    556 	bp->b_resid = bp->b_bcount - count;
    557 	KASSERT(bp->b_resid >= 0);
    558 	bp->b_error = error;
    559 
    560 	biodone(bp);
    561 }
    562 
    563 static void
    564 rump_rcvp_lwpset(struct vnode *rvp, struct vnode *cvp, struct lwp *l)
    565 {
    566 	struct cwdinfo *cwdi = l->l_proc->p_cwdi;
    567 
    568 	KASSERT(cvp);
    569 
    570 	rw_enter(&cwdi->cwdi_lock, RW_WRITER);
    571 	if (cwdi->cwdi_rdir)
    572 		vrele(cwdi->cwdi_rdir);
    573 	if (rvp)
    574 		vref(rvp);
    575 	cwdi->cwdi_rdir = rvp;
    576 
    577 	vrele(cwdi->cwdi_cdir);
    578 	vref(cvp);
    579 	cwdi->cwdi_cdir = cvp;
    580 	rw_exit(&cwdi->cwdi_lock);
    581 }
    582 
    583 void
    584 rump_rcvp_set(struct vnode *rvp, struct vnode *cvp)
    585 {
    586 
    587 	rump_rcvp_lwpset(rvp, cvp, curlwp);
    588 }
    589 
    590 struct vnode *
    591 rump_cdir_get(void)
    592 {
    593 	struct vnode *vp;
    594 	struct cwdinfo *cwdi = curlwp->l_proc->p_cwdi;
    595 
    596 	rw_enter(&cwdi->cwdi_lock, RW_READER);
    597 	vp = cwdi->cwdi_cdir;
    598 	rw_exit(&cwdi->cwdi_lock);
    599 	vref(vp);
    600 
    601 	return vp;
    602 }
    603