Home | History | Annotate | Line # | Download | only in rumpvfs
rumpfs.c revision 1.50
      1 /*	$NetBSD: rumpfs.c,v 1.50 2010/05/11 16:59:42 pooka Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 2009  Antti Kantee.  All Rights Reserved.
      5  *
      6  * Redistribution and use in source and binary forms, with or without
      7  * modification, are permitted provided that the following conditions
      8  * are met:
      9  * 1. Redistributions of source code must retain the above copyright
     10  *    notice, this list of conditions and the following disclaimer.
     11  * 2. Redistributions in binary form must reproduce the above copyright
     12  *    notice, this list of conditions and the following disclaimer in the
     13  *    documentation and/or other materials provided with the distribution.
     14  *
     15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
     16  * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
     17  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
     18  * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
     19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
     21  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     25  * SUCH DAMAGE.
     26  */
     27 
     28 #include <sys/cdefs.h>
     29 __KERNEL_RCSID(0, "$NetBSD: rumpfs.c,v 1.50 2010/05/11 16:59:42 pooka Exp $");
     30 
     31 #include <sys/param.h>
     32 #include <sys/atomic.h>
     33 #include <sys/dirent.h>
     34 #include <sys/errno.h>
     35 #include <sys/filedesc.h>
     36 #include <sys/fcntl.h>
     37 #include <sys/kauth.h>
     38 #include <sys/malloc.h>
     39 #include <sys/module.h>
     40 #include <sys/mount.h>
     41 #include <sys/namei.h>
     42 #include <sys/lock.h>
     43 #include <sys/lockf.h>
     44 #include <sys/queue.h>
     45 #include <sys/stat.h>
     46 #include <sys/syscallargs.h>
     47 #include <sys/vnode.h>
     48 
     49 #include <miscfs/fifofs/fifo.h>
     50 #include <miscfs/specfs/specdev.h>
     51 #include <miscfs/genfs/genfs.h>
     52 
     53 #include <rump/rumpuser.h>
     54 
     55 #include "rump_private.h"
     56 #include "rump_vfs_private.h"
     57 
     58 static int rump_vop_lookup(void *);
     59 static int rump_vop_getattr(void *);
     60 static int rump_vop_mkdir(void *);
     61 static int rump_vop_mknod(void *);
     62 static int rump_vop_create(void *);
     63 static int rump_vop_inactive(void *);
     64 static int rump_vop_reclaim(void *);
     65 static int rump_vop_success(void *);
     66 static int rump_vop_readdir(void *);
     67 static int rump_vop_spec(void *);
     68 static int rump_vop_read(void *);
     69 static int rump_vop_write(void *);
     70 static int rump_vop_open(void *);
     71 
     72 int (**fifo_vnodeop_p)(void *);
     73 const struct vnodeopv_entry_desc fifo_vnodeop_entries[] = {
     74 	{ &vop_default_desc, vn_default_error },
     75 	{ NULL, NULL }
     76 };
     77 const struct vnodeopv_desc fifo_vnodeop_opv_desc =
     78 	{ &fifo_vnodeop_p, fifo_vnodeop_entries };
     79 
     80 int (**rump_vnodeop_p)(void *);
     81 const struct vnodeopv_entry_desc rump_vnodeop_entries[] = {
     82 	{ &vop_default_desc, vn_default_error },
     83 	{ &vop_lookup_desc, rump_vop_lookup },
     84 	{ &vop_getattr_desc, rump_vop_getattr },
     85 	{ &vop_mkdir_desc, rump_vop_mkdir },
     86 	{ &vop_mknod_desc, rump_vop_mknod },
     87 	{ &vop_create_desc, rump_vop_create },
     88 	{ &vop_symlink_desc, genfs_eopnotsupp },
     89 	{ &vop_access_desc, rump_vop_success },
     90 	{ &vop_readdir_desc, rump_vop_readdir },
     91 	{ &vop_read_desc, rump_vop_read },
     92 	{ &vop_write_desc, rump_vop_write },
     93 	{ &vop_open_desc, rump_vop_open },
     94 	{ &vop_putpages_desc, genfs_null_putpages },
     95 	{ &vop_fsync_desc, rump_vop_success },
     96 	{ &vop_lock_desc, genfs_lock },
     97 	{ &vop_unlock_desc, genfs_unlock },
     98 	{ &vop_islocked_desc, genfs_islocked },
     99 	{ &vop_inactive_desc, rump_vop_inactive },
    100 	{ &vop_reclaim_desc, rump_vop_reclaim },
    101 	{ NULL, NULL }
    102 };
    103 const struct vnodeopv_desc rump_vnodeop_opv_desc =
    104 	{ &rump_vnodeop_p, rump_vnodeop_entries };
    105 
    106 int (**rump_specop_p)(void *);
    107 const struct vnodeopv_entry_desc rump_specop_entries[] = {
    108 	{ &vop_default_desc, rump_vop_spec },
    109 	{ NULL, NULL }
    110 };
    111 const struct vnodeopv_desc rump_specop_opv_desc =
    112 	{ &rump_specop_p, rump_specop_entries };
    113 
    114 const struct vnodeopv_desc * const rump_opv_descs[] = {
    115 	&rump_vnodeop_opv_desc,
    116 	&rump_specop_opv_desc,
    117 	NULL
    118 };
    119 
    120 struct rumpfs_dent {
    121 	char *rd_name;
    122 	int rd_namelen;
    123 	struct rumpfs_node *rd_node;
    124 
    125 	LIST_ENTRY(rumpfs_dent) rd_entries;
    126 };
    127 
    128 struct rumpfs_node {
    129 	struct vattr rn_va;
    130 	struct vnode *rn_vp;
    131 	char *rn_hostpath;
    132 	int rn_flags;
    133 
    134 	union {
    135 		struct {		/* VREG */
    136 			int readfd;
    137 			int writefd;
    138 			uint64_t offset;
    139 		} reg;
    140 		struct {		/* VDIR */
    141 			LIST_HEAD(, rumpfs_dent) dents;
    142 			int flags;
    143 		} dir;
    144 	} rn_u;
    145 };
    146 #define rn_readfd	rn_u.reg.readfd
    147 #define rn_writefd	rn_u.reg.writefd
    148 #define rn_offset	rn_u.reg.offset
    149 #define rn_dir		rn_u.dir.dents
    150 
    151 #define RUMPNODE_CANRECLAIM	0x01
    152 #define RUMPNODE_DIR_ET		0x02
    153 #define RUMPNODE_DIR_ETSUBS	0x04
    154 
    155 struct rumpfs_mount {
    156 	struct vnode *rfsmp_rvp;
    157 };
    158 
    159 static struct rumpfs_node *makeprivate(enum vtype, dev_t, off_t);
    160 
    161 /*
    162  * Extra Terrestrial stuff.  We map a given key (pathname) to a file on
    163  * the host FS.  ET phones home only from the root node of rumpfs.
    164  *
    165  * When an etfs node is removed, a vnode potentially behind it is not
    166  * immediately recycled.
    167  */
    168 
    169 struct etfs {
    170 	char et_key[MAXPATHLEN];
    171 	size_t et_keylen;
    172 	bool et_prefixkey;
    173 
    174 	LIST_ENTRY(etfs) et_entries;
    175 
    176 	struct rumpfs_node *et_rn;
    177 };
    178 static kmutex_t etfs_lock;
    179 static LIST_HEAD(, etfs) etfs_list = LIST_HEAD_INITIALIZER(etfs_list);
    180 
    181 static enum vtype
    182 ettype_to_vtype(enum rump_etfs_type et)
    183 {
    184 	enum vtype vt;
    185 
    186 	switch (et) {
    187 	case RUMP_ETFS_REG:
    188 		vt = VREG;
    189 		break;
    190 	case RUMP_ETFS_BLK:
    191 		vt = VBLK;
    192 		break;
    193 	case RUMP_ETFS_CHR:
    194 		vt = VCHR;
    195 		break;
    196 	case RUMP_ETFS_DIR:
    197 		vt = VDIR;
    198 		break;
    199 	case RUMP_ETFS_DIR_SUBDIRS:
    200 		vt = VDIR;
    201 		break;
    202 	default:
    203 		panic("invalid et type: %d", et);
    204 	}
    205 
    206 	return vt;
    207 }
    208 
    209 static enum vtype
    210 hft_to_vtype(int hft)
    211 {
    212 	enum vtype vt;
    213 
    214 	switch (hft) {
    215 	case RUMPUSER_FT_OTHER:
    216 		vt = VNON;
    217 		break;
    218 	case RUMPUSER_FT_DIR:
    219 		vt = VDIR;
    220 		break;
    221 	case RUMPUSER_FT_REG:
    222 		vt = VREG;
    223 		break;
    224 	case RUMPUSER_FT_BLK:
    225 		vt = VBLK;
    226 		break;
    227 	case RUMPUSER_FT_CHR:
    228 		vt = VCHR;
    229 		break;
    230 	default:
    231 		vt = VNON;
    232 		break;
    233 	}
    234 
    235 	return vt;
    236 }
    237 
    238 static bool
    239 etfs_find(const char *key, struct etfs **etp, bool forceprefix)
    240 {
    241 	struct etfs *et;
    242 	size_t keylen = strlen(key);
    243 
    244 	KASSERT(mutex_owned(&etfs_lock));
    245 
    246 	LIST_FOREACH(et, &etfs_list, et_entries) {
    247 		if ((keylen == et->et_keylen || et->et_prefixkey || forceprefix)
    248 		    && strncmp(key, et->et_key, et->et_keylen) == 0) {
    249 			if (etp)
    250 				*etp = et;
    251 			return true;
    252 		}
    253 	}
    254 
    255 	return false;
    256 }
    257 
    258 #define REGDIR(ftype) \
    259     ((ftype) == RUMP_ETFS_DIR || (ftype) == RUMP_ETFS_DIR_SUBDIRS)
    260 static int
    261 doregister(const char *key, const char *hostpath,
    262 	enum rump_etfs_type ftype, uint64_t begin, uint64_t size)
    263 {
    264 	struct etfs *et;
    265 	struct rumpfs_node *rn;
    266 	uint64_t fsize;
    267 	dev_t rdev = NODEV;
    268 	devminor_t dmin;
    269 	int hft, error;
    270 
    271 	if (rumpuser_getfileinfo(hostpath, &fsize, &hft, &error))
    272 		return error;
    273 
    274 	/* etfs directory requires a directory on the host */
    275 	if (REGDIR(ftype)) {
    276 		if (hft != RUMPUSER_FT_DIR)
    277 			return ENOTDIR;
    278 		if (begin != 0)
    279 			return EISDIR;
    280 		if (size != RUMP_ETFS_SIZE_ENDOFF)
    281 			return EISDIR;
    282 		size = fsize;
    283 	} else {
    284 		if (begin > fsize)
    285 			return EINVAL;
    286 		if (size == RUMP_ETFS_SIZE_ENDOFF)
    287 			size = fsize - begin;
    288 		if (begin + size > fsize)
    289 			return EINVAL;
    290 	}
    291 
    292 	if (ftype == RUMP_ETFS_BLK || ftype == RUMP_ETFS_CHR) {
    293 		error = rumpblk_register(hostpath, &dmin, begin, size);
    294 		if (error != 0) {
    295 			return error;
    296 		}
    297 		rdev = makedev(RUMPBLK_DEVMAJOR, dmin);
    298 	}
    299 
    300 	et = kmem_alloc(sizeof(*et), KM_SLEEP);
    301 	strcpy(et->et_key, key);
    302 	et->et_keylen = strlen(et->et_key);
    303 	et->et_rn = rn = makeprivate(ettype_to_vtype(ftype), rdev, size);
    304 
    305 	if (ftype == RUMP_ETFS_REG || REGDIR(ftype)) {
    306 		size_t len = strlen(hostpath)+1;
    307 
    308 		rn->rn_hostpath = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
    309 		memcpy(rn->rn_hostpath, hostpath, len);
    310 		rn->rn_offset = begin;
    311 	}
    312 
    313 	if (REGDIR(ftype)) {
    314 		rn->rn_flags |= RUMPNODE_DIR_ET;
    315 		et->et_prefixkey = true;
    316 	} else {
    317 		et->et_prefixkey = false;
    318 	}
    319 
    320 	if (ftype == RUMP_ETFS_DIR_SUBDIRS)
    321 		rn->rn_flags |= RUMPNODE_DIR_ETSUBS;
    322 
    323 	mutex_enter(&etfs_lock);
    324 	if (etfs_find(key, NULL, REGDIR(ftype))) {
    325 		mutex_exit(&etfs_lock);
    326 		kmem_free(et, sizeof(*et));
    327 		/* XXX: rumpblk_deregister(hostpath); */
    328 		return EEXIST;
    329 	}
    330 	LIST_INSERT_HEAD(&etfs_list, et, et_entries);
    331 	mutex_exit(&etfs_lock);
    332 
    333 	return 0;
    334 }
    335 #undef REGDIR
    336 
    337 int
    338 rump_etfs_register(const char *key, const char *hostpath,
    339 	enum rump_etfs_type ftype)
    340 {
    341 
    342 	return doregister(key, hostpath, ftype, 0, RUMP_ETFS_SIZE_ENDOFF);
    343 }
    344 
    345 int
    346 rump_etfs_register_withsize(const char *key, const char *hostpath,
    347 	enum rump_etfs_type ftype, uint64_t begin, uint64_t size)
    348 {
    349 
    350 	/*
    351 	 * Check that we're mapping at block offsets.  I guess this
    352 	 * is not technically necessary except for BLK/CHR backends
    353 	 * (i.e. what getfileinfo() returns, not ftype) and can be
    354 	 * removed later if there are problems.
    355 	 */
    356 	if ((begin & (DEV_BSIZE-1)) != 0)
    357 		return EINVAL;
    358 	if (size != RUMP_ETFS_SIZE_ENDOFF && (size & (DEV_BSIZE-1)) != 0)
    359 		return EINVAL;
    360 
    361 	return doregister(key, hostpath, ftype, begin, size);
    362 }
    363 
    364 int
    365 rump_etfs_remove(const char *key)
    366 {
    367 	struct etfs *et;
    368 	size_t keylen = strlen(key);
    369 
    370 	mutex_enter(&etfs_lock);
    371 	LIST_FOREACH(et, &etfs_list, et_entries) {
    372 		if (keylen == et->et_keylen && strcmp(et->et_key, key) == 0) {
    373 			LIST_REMOVE(et, et_entries);
    374 			kmem_free(et, sizeof(*et));
    375 			break;
    376 		}
    377 	}
    378 	mutex_exit(&etfs_lock);
    379 
    380 	if (!et)
    381 		return ENOENT;
    382 	return 0;
    383 }
    384 
    385 /*
    386  * rumpfs
    387  */
    388 
    389 static int lastino = 1;
    390 static kmutex_t reclock;
    391 
    392 static struct rumpfs_node *
    393 makeprivate(enum vtype vt, dev_t rdev, off_t size)
    394 {
    395 	struct rumpfs_node *rn;
    396 	struct vattr *va;
    397 	struct timespec ts;
    398 
    399 	rn = kmem_zalloc(sizeof(*rn), KM_SLEEP);
    400 
    401 	switch (vt) {
    402 	case VDIR:
    403 		LIST_INIT(&rn->rn_dir);
    404 		break;
    405 	case VREG:
    406 		rn->rn_readfd = -1;
    407 		rn->rn_writefd = -1;
    408 		break;
    409 	default:
    410 		break;
    411 	}
    412 
    413 	nanotime(&ts);
    414 
    415 	va = &rn->rn_va;
    416 	va->va_type = vt;
    417 	va->va_mode = 0755;
    418 	if (vt == VDIR)
    419 		va->va_nlink = 2;
    420 	else
    421 		va->va_nlink = 1;
    422 	va->va_uid = 0;
    423 	va->va_gid = 0;
    424 	va->va_fsid =
    425 	va->va_fileid = atomic_inc_uint_nv(&lastino);
    426 	va->va_size = size;
    427 	va->va_blocksize = 512;
    428 	va->va_atime = ts;
    429 	va->va_mtime = ts;
    430 	va->va_ctime = ts;
    431 	va->va_birthtime = ts;
    432 	va->va_gen = 0;
    433 	va->va_flags = 0;
    434 	va->va_rdev = rdev;
    435 	va->va_bytes = 512;
    436 	va->va_filerev = 0;
    437 	va->va_vaflags = 0;
    438 
    439 	return rn;
    440 }
    441 
    442 static int
    443 makevnode(struct mount *mp, struct rumpfs_node *rn, struct vnode **vpp)
    444 {
    445 	struct vnode *vp;
    446 	int (**vpops)(void *);
    447 	struct vattr *va = &rn->rn_va;
    448 	int rv;
    449 
    450 	KASSERT(!mutex_owned(&reclock));
    451 
    452 	if (va->va_type == VCHR || va->va_type == VBLK) {
    453 		vpops = rump_specop_p;
    454 	} else {
    455 		vpops = rump_vnodeop_p;
    456 	}
    457 	if (vpops != rump_specop_p && va->va_type != VDIR
    458 	    && !(va->va_type == VREG && rn->rn_hostpath != NULL)
    459 	    && va->va_type != VSOCK)
    460 		return EOPNOTSUPP;
    461 
    462 	rv = getnewvnode(VT_RUMP, mp, vpops, &vp);
    463 	if (rv)
    464 		return rv;
    465 
    466 	vp->v_size = vp->v_writesize = va->va_size;
    467 	vp->v_type = va->va_type;
    468 
    469 	if (vpops == rump_specop_p) {
    470 		spec_node_init(vp, va->va_rdev);
    471 	}
    472 	vp->v_data = rn;
    473 
    474 	vn_lock(vp, LK_RETRY | LK_EXCLUSIVE);
    475 	mutex_enter(&reclock);
    476 	rn->rn_vp = vp;
    477 	mutex_exit(&reclock);
    478 
    479 	*vpp = vp;
    480 
    481 	return 0;
    482 }
    483 
    484 
    485 static void
    486 makedir(struct rumpfs_node *rnd,
    487 	struct componentname *cnp, struct rumpfs_node *rn)
    488 {
    489 	struct rumpfs_dent *rdent;
    490 
    491 	rdent = kmem_alloc(sizeof(*rdent), KM_SLEEP);
    492 	rdent->rd_name = kmem_alloc(cnp->cn_namelen+1, KM_SLEEP);
    493 	rdent->rd_node = rn;
    494 	strlcpy(rdent->rd_name, cnp->cn_nameptr, cnp->cn_namelen+1);
    495 	rdent->rd_namelen = strlen(rdent->rd_name);
    496 
    497 	LIST_INSERT_HEAD(&rnd->rn_dir, rdent, rd_entries);
    498 }
    499 
    500 /*
    501  * Simple lookup for faking lookup of device entry for rump file systems
    502  * and for locating/creating directories.  Yes, this will panic if you
    503  * call it with the wrong arguments.
    504  *
    505  * uhm, this is twisted.  C F C C, hope of C C F C looming
    506  */
    507 static int
    508 rump_vop_lookup(void *v)
    509 {
    510 	struct vop_lookup_args /* {
    511 		struct vnode *a_dvp;
    512 		struct vnode **a_vpp;
    513 		struct componentname *a_cnp;
    514 	}; */ *ap = v;
    515 	struct componentname *cnp = ap->a_cnp;
    516 	struct vnode *dvp = ap->a_dvp;
    517 	struct vnode **vpp = ap->a_vpp;
    518 	struct vnode *vp;
    519 	struct rumpfs_node *rnd = dvp->v_data, *rn;
    520 	struct rumpfs_dent *rd = NULL;
    521 	struct etfs *et;
    522 	int rv;
    523 
    524 	/* we handle only some "non-special" cases */
    525 	if (!(((cnp->cn_flags & ISLASTCN) == 0)
    526 	    || (cnp->cn_nameiop == LOOKUP || cnp->cn_nameiop == CREATE)))
    527 		return EOPNOTSUPP;
    528 	if (!((cnp->cn_flags & ISDOTDOT) == 0))
    529 		return EOPNOTSUPP;
    530 
    531 	/* check for dot, return directly if the case */
    532 	if (cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.') {
    533 		vref(dvp);
    534 		*vpp = dvp;
    535 		return 0;
    536 	}
    537 
    538 	/* check for etfs */
    539 	if (dvp == rootvnode && cnp->cn_nameiop == LOOKUP) {
    540 		bool found;
    541 		mutex_enter(&etfs_lock);
    542 		found = etfs_find(cnp->cn_pnbuf, &et, false);
    543 		mutex_exit(&etfs_lock);
    544 
    545 		if (found) {
    546 			char *offset;
    547 
    548 			offset = strstr(cnp->cn_pnbuf, et->et_key);
    549 			KASSERT(offset);
    550 
    551 			rn = et->et_rn;
    552 			cnp->cn_consume += et->et_keylen
    553 			    - (cnp->cn_nameptr - offset) - cnp->cn_namelen;
    554 			if (rn->rn_va.va_type != VDIR)
    555 				cnp->cn_flags &= ~REQUIREDIR;
    556 			goto getvnode;
    557 		}
    558 	}
    559 
    560 	if (rnd->rn_flags & RUMPNODE_DIR_ET) {
    561 		uint64_t fsize;
    562 		char *newpath;
    563 		size_t newpathlen;
    564 		int hft, error;
    565 
    566 		newpathlen = strlen(rnd->rn_hostpath) + 1 + cnp->cn_namelen + 1;
    567 		newpath = malloc(newpathlen, M_TEMP, M_WAITOK);
    568 
    569 		strlcpy(newpath, rnd->rn_hostpath, newpathlen);
    570 		strlcat(newpath, "/", newpathlen);
    571 		strlcat(newpath, cnp->cn_nameptr, newpathlen);
    572 
    573 		if (rumpuser_getfileinfo(newpath, &fsize, &hft, &error)) {
    574 			free(newpath, M_TEMP);
    575 			return error;
    576 		}
    577 
    578 		/* allow only dirs and regular files */
    579 		if (hft != RUMPUSER_FT_REG && hft != RUMPUSER_FT_DIR) {
    580 			free(newpath, M_TEMP);
    581 			return ENOENT;
    582 		}
    583 
    584 		rn = makeprivate(hft_to_vtype(hft), NODEV, fsize);
    585 		rn->rn_flags |= RUMPNODE_CANRECLAIM;
    586 		if (rnd->rn_flags & RUMPNODE_DIR_ETSUBS) {
    587 			rn->rn_flags |= RUMPNODE_DIR_ET | RUMPNODE_DIR_ETSUBS;
    588 		}
    589 		rn->rn_hostpath = newpath;
    590 
    591 		goto getvnode;
    592 	} else {
    593 		LIST_FOREACH(rd, &rnd->rn_dir, rd_entries) {
    594 			if (rd->rd_namelen == cnp->cn_namelen &&
    595 			    strncmp(rd->rd_name, cnp->cn_nameptr,
    596 			      cnp->cn_namelen) == 0)
    597 				break;
    598 		}
    599 	}
    600 
    601 	if (!rd && ((cnp->cn_flags & ISLASTCN) == 0||cnp->cn_nameiop != CREATE))
    602 		return ENOENT;
    603 
    604 	if (!rd && (cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) {
    605 		cnp->cn_flags |= SAVENAME;
    606 		return EJUSTRETURN;
    607 	}
    608 	rn = rd->rd_node;
    609 
    610  getvnode:
    611 	KASSERT(rn);
    612 	mutex_enter(&reclock);
    613 	if ((vp = rn->rn_vp)) {
    614 		mutex_enter(&vp->v_interlock);
    615 		mutex_exit(&reclock);
    616 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK))
    617 			goto getvnode;
    618 		*vpp = vp;
    619 	} else {
    620 		mutex_exit(&reclock);
    621 		rv = makevnode(dvp->v_mount, rn, vpp);
    622 		if (rv)
    623 			return rv;
    624 	}
    625 
    626 	return 0;
    627 }
    628 
    629 static int
    630 rump_vop_getattr(void *v)
    631 {
    632 	struct vop_getattr_args /* {
    633 		struct vnode *a_vp;
    634 		struct vattr *a_vap;
    635 		kauth_cred_t a_cred;
    636 	} */ *ap = v;
    637 	struct rumpfs_node *rn = ap->a_vp->v_data;
    638 
    639 	memcpy(ap->a_vap, &rn->rn_va, sizeof(struct vattr));
    640 	return 0;
    641 }
    642 
    643 static int
    644 rump_vop_mkdir(void *v)
    645 {
    646 	struct vop_mkdir_args /* {
    647 		struct vnode *a_dvp;
    648 		struct vnode **a_vpp;
    649 		struct componentname *a_cnp;
    650 		struct vattr *a_vap;
    651 	}; */ *ap = v;
    652 	struct vnode *dvp = ap->a_dvp;
    653 	struct vnode **vpp = ap->a_vpp;
    654 	struct componentname *cnp = ap->a_cnp;
    655 	struct rumpfs_node *rnd = dvp->v_data, *rn;
    656 	int rv = 0;
    657 
    658 	rn = makeprivate(VDIR, NODEV, DEV_BSIZE);
    659 	rv = makevnode(dvp->v_mount, rn, vpp);
    660 	if (rv)
    661 		goto out;
    662 
    663 	makedir(rnd, cnp, rn);
    664 
    665  out:
    666 	vput(dvp);
    667 	return rv;
    668 }
    669 
    670 static int
    671 rump_vop_mknod(void *v)
    672 {
    673 	struct vop_mknod_args /* {
    674 		struct vnode *a_dvp;
    675 		struct vnode **a_vpp;
    676 		struct componentname *a_cnp;
    677 		struct vattr *a_vap;
    678 	}; */ *ap = v;
    679 	struct vnode *dvp = ap->a_dvp;
    680 	struct vnode **vpp = ap->a_vpp;
    681 	struct componentname *cnp = ap->a_cnp;
    682 	struct vattr *va = ap->a_vap;
    683 	struct rumpfs_node *rnd = dvp->v_data, *rn;
    684 	int rv;
    685 
    686 	rn = makeprivate(va->va_type, va->va_rdev, DEV_BSIZE);
    687 	rv = makevnode(dvp->v_mount, rn, vpp);
    688 	if (rv)
    689 		goto out;
    690 
    691 	makedir(rnd, cnp, rn);
    692 
    693  out:
    694 	vput(dvp);
    695 	return rv;
    696 }
    697 
    698 static int
    699 rump_vop_create(void *v)
    700 {
    701 	struct vop_create_args /* {
    702 		struct vnode *a_dvp;
    703 		struct vnode **a_vpp;
    704 		struct componentname *a_cnp;
    705 		struct vattr *a_vap;
    706 	}; */ *ap = v;
    707 	struct vnode *dvp = ap->a_dvp;
    708 	struct vnode **vpp = ap->a_vpp;
    709 	struct componentname *cnp = ap->a_cnp;
    710 	struct vattr *va = ap->a_vap;
    711 	struct rumpfs_node *rnd = dvp->v_data, *rn;
    712 	int rv;
    713 
    714 	if (va->va_type != VSOCK) {
    715 		rv = EOPNOTSUPP;
    716 		goto out;
    717 	}
    718 	rn = makeprivate(VSOCK, NODEV, DEV_BSIZE);
    719 	rv = makevnode(dvp->v_mount, rn, vpp);
    720 	if (rv)
    721 		goto out;
    722 
    723 	makedir(rnd, cnp, rn);
    724 
    725  out:
    726 	vput(dvp);
    727 	return rv;
    728 }
    729 
    730 static int
    731 rump_vop_open(void *v)
    732 {
    733 	struct vop_open_args /* {
    734 		struct vnode *a_vp;
    735 		int a_mode;
    736 		kauth_cred_t a_cred;
    737 	} */ *ap = v;
    738 	struct vnode *vp = ap->a_vp;
    739 	struct rumpfs_node *rn = vp->v_data;
    740 	int mode = ap->a_mode;
    741 	int error = EINVAL;
    742 
    743 	if (vp->v_type != VREG)
    744 		return 0;
    745 
    746 	if (mode & FREAD) {
    747 		if (rn->rn_readfd != -1)
    748 			return 0;
    749 		rn->rn_readfd = rumpuser_open(rn->rn_hostpath,
    750 		    O_RDONLY, &error);
    751 	} else if (mode & FWRITE) {
    752 		if (rn->rn_writefd != -1)
    753 			return 0;
    754 		rn->rn_writefd = rumpuser_open(rn->rn_hostpath,
    755 		    O_WRONLY, &error);
    756 	}
    757 
    758 	return error;
    759 }
    760 
    761 /* simple readdir.  event omits dotstuff and periods */
    762 static int
    763 rump_vop_readdir(void *v)
    764 {
    765 	struct vop_readdir_args /* {
    766 		struct vnode *a_vp;
    767 		struct uio *a_uio;
    768 		kauth_cred_t a_cred;
    769 		int *a_eofflag;
    770 		off_t **a_cookies;
    771 		int *a_ncookies;
    772 	} */ *ap = v;
    773 	struct vnode *vp = ap->a_vp;
    774 	struct uio *uio = ap->a_uio;
    775 	struct rumpfs_node *rnd = vp->v_data;
    776 	struct rumpfs_dent *rdent;
    777 	unsigned i;
    778 	int rv = 0;
    779 
    780 	/* seek to current entry */
    781 	for (i = 0, rdent = LIST_FIRST(&rnd->rn_dir);
    782 	    (i < uio->uio_offset) && rdent;
    783 	    i++, rdent = LIST_NEXT(rdent, rd_entries))
    784 		continue;
    785 	if (!rdent)
    786 		goto out;
    787 
    788 	/* copy entries */
    789 	for (; rdent && uio->uio_resid > 0;
    790 	    rdent = LIST_NEXT(rdent, rd_entries), i++) {
    791 		struct dirent dent;
    792 
    793 		dent.d_fileno = rdent->rd_node->rn_va.va_fileid;
    794 		strlcpy(dent.d_name, rdent->rd_name, sizeof(dent.d_name));
    795 		dent.d_namlen = strlen(dent.d_name);
    796 		dent.d_type = vtype2dt(rdent->rd_node->rn_va.va_type);
    797 		dent.d_reclen = _DIRENT_RECLEN(&dent, dent.d_namlen);
    798 
    799 		if (uio->uio_resid < dent.d_reclen) {
    800 			i--;
    801 			break;
    802 		}
    803 
    804 		rv = uiomove(&dent, dent.d_reclen, uio);
    805 		if (rv) {
    806 			i--;
    807 			break;
    808 		}
    809 	}
    810 
    811  out:
    812 	if (ap->a_cookies) {
    813 		*ap->a_ncookies = 0;
    814 		*ap->a_cookies = NULL;
    815 	}
    816 	if (rdent)
    817 		*ap->a_eofflag = 0;
    818 	else
    819 		*ap->a_eofflag = 1;
    820 	uio->uio_offset = i;
    821 
    822 	return rv;
    823 }
    824 
    825 static int
    826 rump_vop_read(void *v)
    827 {
    828 	struct vop_read_args /* {
    829 		struct vnode *a_vp;
    830 		struct uio *a_uio;
    831 		int ioflags a_ioflag;
    832 		kauth_cred_t a_cred;
    833 	}; */ *ap = v;
    834 	struct vnode *vp = ap->a_vp;
    835 	struct rumpfs_node *rn = vp->v_data;
    836 	struct uio *uio = ap->a_uio;
    837 	uint8_t *buf;
    838 	size_t bufsize;
    839 	int error = 0;
    840 
    841 	bufsize = uio->uio_resid;
    842 	buf = kmem_alloc(bufsize, KM_SLEEP);
    843 	if (rumpuser_pread(rn->rn_readfd, buf, bufsize,
    844 	    uio->uio_offset + rn->rn_offset, &error) == -1)
    845 		goto out;
    846 	error = uiomove(buf, bufsize, uio);
    847 
    848  out:
    849 	kmem_free(buf, bufsize);
    850 	return error;
    851 }
    852 
    853 static int
    854 rump_vop_write(void *v)
    855 {
    856 	struct vop_read_args /* {
    857 		struct vnode *a_vp;
    858 		struct uio *a_uio;
    859 		int ioflags a_ioflag;
    860 		kauth_cred_t a_cred;
    861 	}; */ *ap = v;
    862 	struct vnode *vp = ap->a_vp;
    863 	struct rumpfs_node *rn = vp->v_data;
    864 	struct uio *uio = ap->a_uio;
    865 	uint8_t *buf;
    866 	size_t bufsize;
    867 	int error = 0;
    868 
    869 	bufsize = uio->uio_resid;
    870 	buf = kmem_alloc(bufsize, KM_SLEEP);
    871 	error = uiomove(buf, bufsize, uio);
    872 	if (error)
    873 		goto out;
    874 	KASSERT(uio->uio_resid == 0);
    875 	rumpuser_pwrite(rn->rn_writefd, buf, bufsize,
    876 	    uio->uio_offset + rn->rn_offset, &error);
    877 
    878  out:
    879 	kmem_free(buf, bufsize);
    880 	return error;
    881 }
    882 
    883 static int
    884 rump_vop_success(void *v)
    885 {
    886 
    887 	return 0;
    888 }
    889 
    890 static int
    891 rump_vop_inactive(void *v)
    892 {
    893 	struct vop_inactive_args *ap = v;
    894 	struct vnode *vp = ap->a_vp;
    895 	struct rumpfs_node *rn = vp->v_data;
    896 	int error;
    897 
    898 	if (vp->v_type == VREG) {
    899 		if (rn->rn_readfd != -1) {
    900 			rumpuser_close(rn->rn_readfd, &error);
    901 			rn->rn_readfd = -1;
    902 		}
    903 		if (rn->rn_writefd != -1) {
    904 			rumpuser_close(rn->rn_writefd, &error);
    905 			rn->rn_writefd = -1;
    906 		}
    907 	}
    908 
    909 	VOP_UNLOCK(vp, 0);
    910 	return 0;
    911 }
    912 
    913 static int
    914 rump_vop_reclaim(void *v)
    915 {
    916 	struct vop_reclaim_args /* {
    917 		struct vnode *a_vp;
    918 	} */ *ap = v;
    919 	struct vnode *vp = ap->a_vp;
    920 	struct rumpfs_node *rn = vp->v_data;
    921 
    922 	mutex_enter(&reclock);
    923 	rn->rn_vp = NULL;
    924 	mutex_exit(&reclock);
    925 	vp->v_data = NULL;
    926 
    927 	if (rn->rn_flags & RUMPNODE_CANRECLAIM) {
    928 		if (rn->rn_hostpath)
    929 			free(rn->rn_hostpath, M_TEMP);
    930 		kmem_free(rn, sizeof(*rn));
    931 	}
    932 
    933 	return 0;
    934 }
    935 
    936 static int
    937 rump_vop_spec(void *v)
    938 {
    939 	struct vop_generic_args *ap = v;
    940 	int (**opvec)(void *);
    941 
    942 	switch (ap->a_desc->vdesc_offset) {
    943 	case VOP_ACCESS_DESCOFFSET:
    944 	case VOP_GETATTR_DESCOFFSET:
    945 	case VOP_LOCK_DESCOFFSET:
    946 	case VOP_UNLOCK_DESCOFFSET:
    947 	case VOP_RECLAIM_DESCOFFSET:
    948 		opvec = rump_vnodeop_p;
    949 		break;
    950 	default:
    951 		opvec = spec_vnodeop_p;
    952 		break;
    953 	}
    954 
    955 	return VOCALL(opvec, ap->a_desc->vdesc_offset, v);
    956 }
    957 
    958 /*
    959  * Begin vfs-level stuff
    960  */
    961 
    962 VFS_PROTOS(rumpfs);
    963 struct vfsops rumpfs_vfsops = {
    964 	.vfs_name =		MOUNT_RUMPFS,
    965 	.vfs_min_mount_data = 	0,
    966 	.vfs_mount =		rumpfs_mount,
    967 	.vfs_start =		(void *)nullop,
    968 	.vfs_unmount = 		rumpfs_unmount,
    969 	.vfs_root =		rumpfs_root,
    970 	.vfs_quotactl =		(void *)eopnotsupp,
    971 	.vfs_statvfs =		genfs_statvfs,
    972 	.vfs_sync =		(void *)nullop,
    973 	.vfs_vget =		rumpfs_vget,
    974 	.vfs_fhtovp =		(void *)eopnotsupp,
    975 	.vfs_vptofh =		(void *)eopnotsupp,
    976 	.vfs_init =		rumpfs_init,
    977 	.vfs_reinit =		NULL,
    978 	.vfs_done =		rumpfs_done,
    979 	.vfs_mountroot =	rumpfs_mountroot,
    980 	.vfs_snapshot =		(void *)eopnotsupp,
    981 	.vfs_extattrctl =	(void *)eopnotsupp,
    982 	.vfs_suspendctl =	(void *)eopnotsupp,
    983 	.vfs_opv_descs =	rump_opv_descs,
    984 	/* vfs_refcount */
    985 	/* vfs_list */
    986 };
    987 
    988 int
    989 rumpfs_mount(struct mount *mp, const char *mntpath, void *arg, size_t *alen)
    990 {
    991 
    992 	return EOPNOTSUPP;
    993 }
    994 
    995 int
    996 rumpfs_unmount(struct mount *mp, int flags)
    997 {
    998 
    999 	/* if going for it, just lie about it */
   1000 	if (panicstr)
   1001 		return 0;
   1002 
   1003 	return EOPNOTSUPP; /* ;) */
   1004 }
   1005 
   1006 int
   1007 rumpfs_root(struct mount *mp, struct vnode **vpp)
   1008 {
   1009 	struct rumpfs_mount *rfsmp = mp->mnt_data;
   1010 
   1011 	vget(rfsmp->rfsmp_rvp, LK_EXCLUSIVE | LK_RETRY);
   1012 	*vpp = rfsmp->rfsmp_rvp;
   1013 	return 0;
   1014 }
   1015 
   1016 int
   1017 rumpfs_vget(struct mount *mp, ino_t ino, struct vnode **vpp)
   1018 {
   1019 
   1020 	return EOPNOTSUPP;
   1021 }
   1022 
   1023 void
   1024 rumpfs_init()
   1025 {
   1026 
   1027 	CTASSERT(RUMP_ETFS_SIZE_ENDOFF == RUMPBLK_SIZENOTSET);
   1028 
   1029 	mutex_init(&reclock, MUTEX_DEFAULT, IPL_NONE);
   1030 	mutex_init(&etfs_lock, MUTEX_DEFAULT, IPL_NONE);
   1031 }
   1032 
   1033 void
   1034 rumpfs_done()
   1035 {
   1036 
   1037 	mutex_destroy(&reclock);
   1038 	mutex_destroy(&etfs_lock);
   1039 }
   1040 
   1041 int
   1042 rumpfs_mountroot()
   1043 {
   1044 	struct mount *mp;
   1045 	struct rumpfs_mount *rfsmp;
   1046 	struct rumpfs_node *rn;
   1047 	int error;
   1048 
   1049 	if ((error = vfs_rootmountalloc(MOUNT_RUMPFS, "rootdev", &mp)) != 0) {
   1050 		vrele(rootvp);
   1051 		return error;
   1052 	}
   1053 
   1054 	rfsmp = kmem_alloc(sizeof(*rfsmp), KM_SLEEP);
   1055 
   1056 	rn = makeprivate(VDIR, NODEV, DEV_BSIZE);
   1057 	error = makevnode(mp, rn, &rfsmp->rfsmp_rvp);
   1058 	if (error)
   1059 		panic("could not create root vnode: %d", error);
   1060 	rfsmp->rfsmp_rvp->v_vflag |= VV_ROOT;
   1061 	VOP_UNLOCK(rfsmp->rfsmp_rvp, 0);
   1062 
   1063 	mutex_enter(&mountlist_lock);
   1064 	CIRCLEQ_INSERT_TAIL(&mountlist, mp, mnt_list);
   1065 	mutex_exit(&mountlist_lock);
   1066 
   1067 	mp->mnt_data = rfsmp;
   1068 	mp->mnt_stat.f_namemax = MAXNAMLEN;
   1069 	mp->mnt_stat.f_iosize = 512;
   1070 	mp->mnt_flag |= MNT_LOCAL;
   1071 	mp->mnt_iflag |= IMNT_MPSAFE;
   1072 	vfs_getnewfsid(mp);
   1073 
   1074 	error = set_statvfs_info("/", UIO_SYSSPACE, "rumpfs", UIO_SYSSPACE,
   1075 	    mp->mnt_op->vfs_name, mp, curlwp);
   1076 	if (error)
   1077 		panic("set statvfsinfo for rootfs failed");
   1078 
   1079 	vfs_unbusy(mp, false, NULL);
   1080 
   1081 	return 0;
   1082 }
   1083