Home | History | Annotate | Line # | Download | only in libpuffs
puffs.c revision 1.92.4.3 
      1 /*	$NetBSD: puffs.c,v 1.92.4.3 2009/10/18 12:46:07 sborrill Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 2005, 2006, 2007  Antti Kantee.  All Rights Reserved.
      5  *
      6  * Development of this software was supported by the
      7  * Google Summer of Code program and the Ulla Tuominen Foundation.
      8  * The Google SoC project was mentored by Bill Studenmund.
      9  *
     10  * Redistribution and use in source and binary forms, with or without
     11  * modification, are permitted provided that the following conditions
     12  * are met:
     13  * 1. Redistributions of source code must retain the above copyright
     14  *    notice, this list of conditions and the following disclaimer.
     15  * 2. Redistributions in binary form must reproduce the above copyright
     16  *    notice, this list of conditions and the following disclaimer in the
     17  *    documentation and/or other materials provided with the distribution.
     18  *
     19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
     20  * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
     21  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
     22  * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
     23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
     25  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     29  * SUCH DAMAGE.
     30  */
     31 
     32 #include <sys/cdefs.h>
     33 #if !defined(lint)
     34 __RCSID("$NetBSD: puffs.c,v 1.92.4.3 2009/10/18 12:46:07 sborrill Exp $");
     35 #endif /* !lint */
     36 
     37 #include <sys/param.h>
     38 #include <sys/mount.h>
     39 
     40 #include <assert.h>
     41 #include <err.h>
     42 #include <errno.h>
     43 #include <fcntl.h>
     44 #include <mntopts.h>
     45 #include <paths.h>
     46 #include <puffs.h>
     47 #include <stdio.h>
     48 #include <stdlib.h>
     49 #include <string.h>
     50 #include <syslog.h>
     51 #include <unistd.h>
     52 
     53 #include "puffs_priv.h"
     54 
     55 /* Most file systems want this for opts, so just give it to them */
     56 const struct mntopt puffsmopts[] = {
     57 	MOPT_STDOPTS,
     58 	PUFFSMOPT_STD,
     59 	MOPT_NULL,
     60 };
     61 
     62 #ifdef PUFFS_WITH_THREADS
     63 #include <pthread.h>
     64 pthread_mutex_t pu_lock = PTHREAD_MUTEX_INITIALIZER;
     65 #endif
     66 
     67 #define FILLOP(lower, upper)						\
     68 do {									\
     69 	if (pops->puffs_node_##lower)					\
     70 		opmask[PUFFS_VN_##upper] = 1;				\
     71 } while (/*CONSTCOND*/0)
     72 static void
     73 fillvnopmask(struct puffs_ops *pops, uint8_t *opmask)
     74 {
     75 
     76 	memset(opmask, 0, PUFFS_VN_MAX);
     77 
     78 	FILLOP(create,   CREATE);
     79 	FILLOP(mknod,    MKNOD);
     80 	FILLOP(open,     OPEN);
     81 	FILLOP(close,    CLOSE);
     82 	FILLOP(access,   ACCESS);
     83 	FILLOP(getattr,  GETATTR);
     84 	FILLOP(setattr,  SETATTR);
     85 	FILLOP(poll,     POLL); /* XXX: not ready in kernel */
     86 	FILLOP(mmap,     MMAP);
     87 	FILLOP(fsync,    FSYNC);
     88 	FILLOP(seek,     SEEK);
     89 	FILLOP(remove,   REMOVE);
     90 	FILLOP(link,     LINK);
     91 	FILLOP(rename,   RENAME);
     92 	FILLOP(mkdir,    MKDIR);
     93 	FILLOP(rmdir,    RMDIR);
     94 	FILLOP(symlink,  SYMLINK);
     95 	FILLOP(readdir,  READDIR);
     96 	FILLOP(readlink, READLINK);
     97 	FILLOP(reclaim,  RECLAIM);
     98 	FILLOP(inactive, INACTIVE);
     99 	FILLOP(print,    PRINT);
    100 	FILLOP(read,     READ);
    101 	FILLOP(write,    WRITE);
    102 	FILLOP(abortop,  ABORTOP);
    103 }
    104 #undef FILLOP
    105 
    106 /*
    107  * Go over all framev entries and write everything we can.  This is
    108  * mostly for the benefit of delivering "unmount" to the kernel.
    109  */
    110 static void
    111 finalpush(struct puffs_usermount *pu)
    112 {
    113 	struct puffs_fctrl_io *fio;
    114 
    115 	LIST_FOREACH(fio, &pu->pu_ios, fio_entries) {
    116 		if (fio->stat & FIO_WRGONE)
    117 			continue;
    118 
    119 		puffs__framev_output(pu, fio->fctrl, fio);
    120 	}
    121 }
    122 
    123 /*ARGSUSED*/
    124 static void
    125 puffs_defaulterror(struct puffs_usermount *pu, uint8_t type,
    126 	int error, const char *str, puffs_cookie_t cookie)
    127 {
    128 
    129 	fprintf(stderr, "abort: type %d, error %d, cookie %p (%s)\n",
    130 	    type, error, cookie, str);
    131 	abort();
    132 }
    133 
    134 int
    135 puffs_getselectable(struct puffs_usermount *pu)
    136 {
    137 
    138 	return pu->pu_fd;
    139 }
    140 
    141 uint64_t
    142 puffs__nextreq(struct puffs_usermount *pu)
    143 {
    144 	uint64_t rv;
    145 
    146 	PU_LOCK();
    147 	rv = pu->pu_nextreq++;
    148 	PU_UNLOCK();
    149 
    150 	return rv;
    151 }
    152 
    153 int
    154 puffs_setblockingmode(struct puffs_usermount *pu, int mode)
    155 {
    156 	int rv, x;
    157 
    158 	assert(puffs_getstate(pu) == PUFFS_STATE_RUNNING);
    159 
    160 	if (mode != PUFFSDEV_BLOCK && mode != PUFFSDEV_NONBLOCK) {
    161 		errno = EINVAL;
    162 		return -1;
    163 	}
    164 
    165 	x = mode;
    166 	rv = ioctl(pu->pu_fd, FIONBIO, &x);
    167 
    168 	if (rv == 0) {
    169 		if (mode == PUFFSDEV_BLOCK)
    170 			pu->pu_state &= ~PU_ASYNCFD;
    171 		else
    172 			pu->pu_state |= PU_ASYNCFD;
    173 	}
    174 
    175 	return rv;
    176 }
    177 
    178 int
    179 puffs_getstate(struct puffs_usermount *pu)
    180 {
    181 
    182 	return pu->pu_state & PU_STATEMASK;
    183 }
    184 
    185 void
    186 puffs_setstacksize(struct puffs_usermount *pu, size_t ss)
    187 {
    188 	long psize, minsize;
    189 	int stackshift;
    190 	int bonus;
    191 
    192 	assert(puffs_getstate(pu) == PUFFS_STATE_BEFOREMOUNT);
    193 
    194 	psize = sysconf(_SC_PAGESIZE);
    195 	minsize = 4*psize;
    196 	if (ss < minsize || ss == PUFFS_STACKSIZE_MIN) {
    197 		if (ss != PUFFS_STACKSIZE_MIN)
    198 			fprintf(stderr, "puffs_setstacksize: adjusting "
    199 			    "stacksize to minimum %ld\n", minsize);
    200 		ss = 4*psize;
    201 	}
    202 
    203 	stackshift = -1;
    204 	bonus = 0;
    205 	while (ss) {
    206 		if (ss & 0x1)
    207 			bonus++;
    208 		ss >>= 1;
    209 		stackshift++;
    210 	}
    211 	if (bonus > 1) {
    212 		stackshift++;
    213 		fprintf(stderr, "puffs_setstacksize: using next power of two: "
    214 		    "%d\n", 1<<stackshift);
    215 	}
    216 
    217 	pu->pu_cc_stackshift = stackshift;
    218 }
    219 
    220 struct puffs_pathobj *
    221 puffs_getrootpathobj(struct puffs_usermount *pu)
    222 {
    223 	struct puffs_node *pnr;
    224 
    225 	pnr = pu->pu_pn_root;
    226 	if (pnr == NULL) {
    227 		errno = ENOENT;
    228 		return NULL;
    229 	}
    230 
    231 	return &pnr->pn_po;
    232 }
    233 
    234 void
    235 puffs_setroot(struct puffs_usermount *pu, struct puffs_node *pn)
    236 {
    237 
    238 	pu->pu_pn_root = pn;
    239 }
    240 
    241 struct puffs_node *
    242 puffs_getroot(struct puffs_usermount *pu)
    243 {
    244 
    245 	return pu->pu_pn_root;
    246 }
    247 
    248 void
    249 puffs_setrootinfo(struct puffs_usermount *pu, enum vtype vt,
    250 	vsize_t vsize, dev_t rdev)
    251 {
    252 	struct puffs_kargs *pargs = pu->pu_kargp;
    253 
    254 	if (puffs_getstate(pu) != PUFFS_STATE_BEFOREMOUNT) {
    255 		warnx("puffs_setrootinfo: call has effect only "
    256 		    "before mount\n");
    257 		return;
    258 	}
    259 
    260 	pargs->pa_root_vtype = vt;
    261 	pargs->pa_root_vsize = vsize;
    262 	pargs->pa_root_rdev = rdev;
    263 }
    264 
    265 void *
    266 puffs_getspecific(struct puffs_usermount *pu)
    267 {
    268 
    269 	return pu->pu_privdata;
    270 }
    271 
    272 void
    273 puffs_setspecific(struct puffs_usermount *pu, void *privdata)
    274 {
    275 
    276 	pu->pu_privdata = privdata;
    277 }
    278 
    279 size_t
    280 puffs_getmaxreqlen(struct puffs_usermount *pu)
    281 {
    282 
    283 	return pu->pu_maxreqlen;
    284 }
    285 
    286 void
    287 puffs_setmaxreqlen(struct puffs_usermount *pu, size_t reqlen)
    288 {
    289 
    290 	if (puffs_getstate(pu) != PUFFS_STATE_BEFOREMOUNT)
    291 		warnx("puffs_setmaxreqlen: call has effect only "
    292 		    "before mount\n");
    293 
    294 	pu->pu_kargp->pa_maxmsglen = reqlen;
    295 }
    296 
    297 void
    298 puffs_setfhsize(struct puffs_usermount *pu, size_t fhsize, int flags)
    299 {
    300 
    301 	if (puffs_getstate(pu) != PUFFS_STATE_BEFOREMOUNT)
    302 		warnx("puffs_setfhsize: call has effect only before mount\n");
    303 
    304 	pu->pu_kargp->pa_fhsize = fhsize;
    305 	pu->pu_kargp->pa_fhflags = flags;
    306 }
    307 
    308 void
    309 puffs_setncookiehash(struct puffs_usermount *pu, int nhash)
    310 {
    311 
    312 	if (puffs_getstate(pu) != PUFFS_STATE_BEFOREMOUNT)
    313 		warnx("puffs_setfhsize: call has effect only before mount\n");
    314 
    315 	pu->pu_kargp->pa_nhashbuckets = nhash;
    316 }
    317 
    318 void
    319 puffs_set_pathbuild(struct puffs_usermount *pu, pu_pathbuild_fn fn)
    320 {
    321 
    322 	pu->pu_pathbuild = fn;
    323 }
    324 
    325 void
    326 puffs_set_pathtransform(struct puffs_usermount *pu, pu_pathtransform_fn fn)
    327 {
    328 
    329 	pu->pu_pathtransform = fn;
    330 }
    331 
    332 void
    333 puffs_set_pathcmp(struct puffs_usermount *pu, pu_pathcmp_fn fn)
    334 {
    335 
    336 	pu->pu_pathcmp = fn;
    337 }
    338 
    339 void
    340 puffs_set_pathfree(struct puffs_usermount *pu, pu_pathfree_fn fn)
    341 {
    342 
    343 	pu->pu_pathfree = fn;
    344 }
    345 
    346 void
    347 puffs_set_namemod(struct puffs_usermount *pu, pu_namemod_fn fn)
    348 {
    349 
    350 	pu->pu_namemod = fn;
    351 }
    352 
    353 void
    354 puffs_set_errnotify(struct puffs_usermount *pu, pu_errnotify_fn fn)
    355 {
    356 
    357 	pu->pu_errnotify = fn;
    358 }
    359 
    360 void
    361 puffs_set_cmap(struct puffs_usermount *pu, pu_cmap_fn fn)
    362 {
    363 
    364 	pu->pu_cmap = fn;
    365 }
    366 
    367 void
    368 puffs_ml_setloopfn(struct puffs_usermount *pu, puffs_ml_loop_fn lfn)
    369 {
    370 
    371 	pu->pu_ml_lfn = lfn;
    372 }
    373 
    374 void
    375 puffs_ml_settimeout(struct puffs_usermount *pu, struct timespec *ts)
    376 {
    377 
    378 	if (ts == NULL) {
    379 		pu->pu_ml_timep = NULL;
    380 	} else {
    381 		pu->pu_ml_timeout = *ts;
    382 		pu->pu_ml_timep = &pu->pu_ml_timeout;
    383 	}
    384 }
    385 
    386 void
    387 puffs_set_prepost(struct puffs_usermount *pu,
    388 	pu_prepost_fn pre, pu_prepost_fn pst)
    389 {
    390 
    391 	pu->pu_oppre = pre;
    392 	pu->pu_oppost = pst;
    393 }
    394 
    395 void
    396 puffs_setback(struct puffs_cc *pcc, int whatback)
    397 {
    398 	struct puffs_req *preq = puffs__framebuf_getdataptr(pcc->pcc_pb);
    399 
    400 	assert(PUFFSOP_OPCLASS(preq->preq_opclass) == PUFFSOP_VN && (
    401 	    preq->preq_optype == PUFFS_VN_OPEN ||
    402 	    preq->preq_optype == PUFFS_VN_MMAP ||
    403 	    preq->preq_optype == PUFFS_VN_REMOVE ||
    404 	    preq->preq_optype == PUFFS_VN_RMDIR ||
    405 	    preq->preq_optype == PUFFS_VN_INACTIVE));
    406 
    407 	preq->preq_setbacks |= whatback & PUFFS_SETBACK_MASK;
    408 }
    409 
    410 int
    411 puffs_daemon(struct puffs_usermount *pu, int nochdir, int noclose)
    412 {
    413 	ssize_t n;
    414 	int parent, value, fd;
    415 
    416 	if (pipe(pu->pu_dpipe) == -1)
    417 		return -1;
    418 
    419 	switch (fork()) {
    420 	case -1:
    421 		return -1;
    422 	case 0:
    423 		parent = 0;
    424 		break;
    425 	default:
    426 		parent = 1;
    427 		break;
    428 	}
    429 	pu->pu_state |= PU_PUFFSDAEMON;
    430 
    431 	if (parent) {
    432 		n = read(pu->pu_dpipe[0], &value, sizeof(int));
    433 		if (n == -1)
    434 			err(1, "puffs_daemon");
    435 		assert(n == sizeof(value));
    436 		if (value) {
    437 			errno = value;
    438 			err(1, "puffs_daemon");
    439 		}
    440 		exit(0);
    441 	} else {
    442 		if (setsid() == -1)
    443 			goto fail;
    444 
    445 		if (!nochdir)
    446 			chdir("/");
    447 
    448 		if (!noclose) {
    449 			fd = open(_PATH_DEVNULL, O_RDWR, 0);
    450 			if (fd == -1)
    451 				goto fail;
    452 			dup2(fd, STDIN_FILENO);
    453 			dup2(fd, STDOUT_FILENO);
    454 			dup2(fd, STDERR_FILENO);
    455 			if (fd > STDERR_FILENO)
    456 				close(fd);
    457 		}
    458 		return 0;
    459 	}
    460 
    461  fail:
    462 	n = write(pu->pu_dpipe[1], &errno, sizeof(int));
    463 	assert(n == 4);
    464 	return -1;
    465 }
    466 
    467 static void
    468 shutdaemon(struct puffs_usermount *pu, int error)
    469 {
    470 	ssize_t n;
    471 
    472 	n = write(pu->pu_dpipe[1], &error, sizeof(int));
    473 	assert(n == 4);
    474 	close(pu->pu_dpipe[0]);
    475 	close(pu->pu_dpipe[1]);
    476 	pu->pu_state &= ~PU_PUFFSDAEMON;
    477 }
    478 
    479 int
    480 puffs_mount(struct puffs_usermount *pu, const char *dir, int mntflags,
    481 	puffs_cookie_t cookie)
    482 {
    483 	char rp[MAXPATHLEN];
    484 	int rv, fd, sverrno;
    485 	char *comfd;
    486 
    487 	pu->pu_kargp->pa_root_cookie = cookie;
    488 
    489 	/* XXXkludgehere */
    490 	/* kauth doesn't provide this service any longer */
    491 	if (geteuid() != 0)
    492 		mntflags |= MNT_NOSUID | MNT_NODEV;
    493 
    494 	if (realpath(dir, rp) == NULL) {
    495 		rv = -1;
    496 		goto out;
    497 	}
    498 
    499 	if (strcmp(dir, rp) != 0) {
    500 		warnx("puffs_mount: \"%s\" is a relative path.", dir);
    501 		warnx("puffs_mount: using \"%s\" instead.", rp);
    502 	}
    503 
    504 	/*
    505 	 * Undocumented...  Well, documented only here.
    506 	 *
    507 	 * This is used for imaginative purposes.  If the env variable is
    508 	 * set, puffs_mount() doesn't do the regular mount procedure.
    509 	 * Rather, it crams the mount data down the comfd and sets comfd as
    510 	 * the puffs descriptor.
    511 	 *
    512 	 * This shouldn't be used unless you can read my mind ( ... or write
    513 	 * it, not to mention execute it, but that's starting to get silly).
    514 	 */
    515 	if ((comfd = getenv("PUFFS_COMFD")) != NULL) {
    516 		size_t len;
    517 
    518 		if (sscanf(comfd, "%d", &pu->pu_fd) != 1) {
    519 			errno = EINVAL;
    520 			rv = -1;
    521 			goto out;
    522 		}
    523 		/* check that what we got at least resembles an fd */
    524 		if (fcntl(pu->pu_fd, F_GETFL) == -1) {
    525 			rv = -1;
    526 			goto out;
    527 		}
    528 
    529 		len = strlen(dir)+1;
    530 
    531 #define allwrite(buf, len)						\
    532 do {									\
    533 	ssize_t al_rv;							\
    534 	al_rv = write(pu->pu_fd, buf, len);				\
    535 	if (al_rv != len) {						\
    536 		if (al_rv != -1)					\
    537 			errno = EIO;					\
    538 		rv = -1;						\
    539 		abort();\
    540 		goto out;						\
    541 	}								\
    542 } while (/*CONSTCOND*/0)
    543 		allwrite(&len, sizeof(len));
    544 		allwrite(dir, len);
    545 		len = strlen(pu->pu_kargp->pa_mntfromname)+1;
    546 		allwrite(&len, sizeof(len));
    547 		allwrite(pu->pu_kargp->pa_mntfromname, len);
    548 		allwrite(&mntflags, sizeof(mntflags));
    549 		allwrite(pu->pu_kargp, sizeof(*pu->pu_kargp));
    550 		allwrite(&pu->pu_flags, sizeof(pu->pu_flags));
    551 #undef allwrite
    552 
    553 		rv = 0;
    554 	} else {
    555 		fd = open(_PATH_PUFFS, O_RDWR);
    556 		if (fd == -1) {
    557 			warnx("puffs_mount: cannot open %s", _PATH_PUFFS);
    558 			rv = -1;
    559 			goto out;
    560 		}
    561 		if (fd <= 2)
    562 			warnx("puffs_mount: device fd %d (<= 2), sure this is "
    563 			    "what you want?", fd);
    564 
    565 		pu->pu_kargp->pa_fd = pu->pu_fd = fd;
    566 		if ((rv = mount(MOUNT_PUFFS, rp, mntflags,
    567 		    pu->pu_kargp, sizeof(struct puffs_kargs))) == -1)
    568 			goto out;
    569 	}
    570 
    571 	PU_SETSTATE(pu, PUFFS_STATE_RUNNING);
    572 
    573  out:
    574 	if (rv != 0)
    575 		sverrno = errno;
    576 	else
    577 		sverrno = 0;
    578 	free(pu->pu_kargp);
    579 	pu->pu_kargp = NULL;
    580 
    581 	if (pu->pu_state & PU_PUFFSDAEMON)
    582 		shutdaemon(pu, sverrno);
    583 
    584 	errno = sverrno;
    585 	return rv;
    586 }
    587 
    588 /*ARGSUSED*/
    589 struct puffs_usermount *
    590 _puffs_init(int dummy, struct puffs_ops *pops, const char *mntfromname,
    591 	const char *puffsname, void *priv, uint32_t pflags)
    592 {
    593 	struct puffs_usermount *pu;
    594 	struct puffs_kargs *pargs;
    595 	int sverrno;
    596 
    597 	pu = malloc(sizeof(struct puffs_usermount));
    598 	if (pu == NULL)
    599 		goto failfree;
    600 	memset(pu, 0, sizeof(struct puffs_usermount));
    601 
    602 	pargs = pu->pu_kargp = malloc(sizeof(struct puffs_kargs));
    603 	if (pargs == NULL)
    604 		goto failfree;
    605 	memset(pargs, 0, sizeof(struct puffs_kargs));
    606 
    607 	pargs->pa_vers = PUFFSDEVELVERS | PUFFSVERSION;
    608 	pargs->pa_flags = PUFFS_FLAG_KERN(pflags);
    609 	fillvnopmask(pops, pargs->pa_vnopmask);
    610 	(void)strlcpy(pargs->pa_typename, puffsname,
    611 	    sizeof(pargs->pa_typename));
    612 	(void)strlcpy(pargs->pa_mntfromname, mntfromname,
    613 	    sizeof(pargs->pa_mntfromname));
    614 
    615 	puffs_zerostatvfs(&pargs->pa_svfsb);
    616 	pargs->pa_root_cookie = NULL;
    617 	pargs->pa_root_vtype = VDIR;
    618 	pargs->pa_root_vsize = 0;
    619 	pargs->pa_root_rdev = 0;
    620 	pargs->pa_maxmsglen = 0;
    621 
    622 	pu->pu_flags = pflags;
    623 	pu->pu_ops = *pops;
    624 	free(pops); /* XXX */
    625 
    626 	pu->pu_privdata = priv;
    627 	pu->pu_cc_stackshift = PUFFS_CC_STACKSHIFT_DEFAULT;
    628 	LIST_INIT(&pu->pu_pnodelst);
    629 	LIST_INIT(&pu->pu_ios);
    630 	LIST_INIT(&pu->pu_ios_rmlist);
    631 	LIST_INIT(&pu->pu_ccmagazin);
    632 	TAILQ_INIT(&pu->pu_sched);
    633 
    634 	pu->pu_framectrl[PU_FRAMECTRL_FS].rfb = puffs__fsframe_read;
    635 	pu->pu_framectrl[PU_FRAMECTRL_FS].wfb = puffs__fsframe_write;
    636 	pu->pu_framectrl[PU_FRAMECTRL_FS].cmpfb = puffs__fsframe_cmp;
    637 	pu->pu_framectrl[PU_FRAMECTRL_FS].gotfb = puffs__fsframe_gotframe;
    638 	pu->pu_framectrl[PU_FRAMECTRL_FS].fdnotfn = puffs_framev_unmountonclose;
    639 
    640 	/* defaults for some user-settable translation functions */
    641 	pu->pu_cmap = NULL; /* identity translation */
    642 
    643 	pu->pu_pathbuild = puffs_stdpath_buildpath;
    644 	pu->pu_pathfree = puffs_stdpath_freepath;
    645 	pu->pu_pathcmp = puffs_stdpath_cmppath;
    646 	pu->pu_pathtransform = NULL;
    647 	pu->pu_namemod = NULL;
    648 
    649 	pu->pu_errnotify = puffs_defaulterror;
    650 
    651 	PU_SETSTATE(pu, PUFFS_STATE_BEFOREMOUNT);
    652 
    653 	return pu;
    654 
    655  failfree:
    656 	/* can't unmount() from here for obvious reasons */
    657 	sverrno = errno;
    658 	free(pu);
    659 	errno = sverrno;
    660 	return NULL;
    661 }
    662 
    663 void
    664 puffs_cancel(struct puffs_usermount *pu, int error)
    665 {
    666 
    667 	assert(puffs_getstate(pu) < PUFFS_STATE_RUNNING);
    668 	shutdaemon(pu, error);
    669 	free(pu);
    670 }
    671 
    672 /*
    673  * XXX: there's currently no clean way to request unmount from
    674  * within the user server, so be very brutal about it.
    675  */
    676 /*ARGSUSED1*/
    677 int
    678 puffs_exit(struct puffs_usermount *pu, int force)
    679 {
    680 	struct puffs_node *pn;
    681 
    682 	force = 1; /* currently */
    683 	assert((pu->pu_state & PU_PUFFSDAEMON) == 0);
    684 
    685 	if (pu->pu_fd)
    686 		close(pu->pu_fd);
    687 
    688 	while ((pn = LIST_FIRST(&pu->pu_pnodelst)) != NULL)
    689 		puffs_pn_put(pn);
    690 
    691 	finalpush(pu);
    692 	puffs__framev_exit(pu);
    693 	puffs__cc_exit(pu);
    694 	if (pu->pu_state & PU_HASKQ)
    695 		close(pu->pu_kq);
    696 	free(pu);
    697 
    698 	return 0; /* always succesful for now, WILL CHANGE */
    699 }
    700 
    701 /*
    702  * Actual mainloop.  This is called from a context which can block.
    703  * It is called either from puffs_mainloop (indirectly, via
    704  * puffs_cc_continue() or from puffs_cc_yield()).
    705  */
    706 void
    707 puffs__theloop(struct puffs_cc *pcc)
    708 {
    709 	struct puffs_usermount *pu = pcc->pcc_pu;
    710 	struct puffs_framectrl *pfctrl;
    711 	struct puffs_fctrl_io *fio;
    712 	struct kevent *curev;
    713 	size_t nchanges;
    714 	int ndone;
    715 
    716 	while (puffs_getstate(pu) != PUFFS_STATE_UNMOUNTED) {
    717 		/*
    718 		 * Schedule existing requests.
    719 		 */
    720 		while ((pcc = TAILQ_FIRST(&pu->pu_sched)) != NULL) {
    721 			TAILQ_REMOVE(&pu->pu_sched, pcc, pcc_schedent);
    722 			puffs__goto(pcc);
    723 		}
    724 
    725 		if (pu->pu_ml_lfn)
    726 			pu->pu_ml_lfn(pu);
    727 
    728 		/* XXX: can we still do these optimizations? */
    729 #if 0
    730 		/*
    731 		 * Do this here, because:
    732 		 *  a) loopfunc might generate some results
    733 		 *  b) it's still "after" event handling (except for round 1)
    734 		 */
    735 		if (puffs_req_putput(ppr) == -1)
    736 			goto out;
    737 		puffs_req_resetput(ppr);
    738 
    739 		/* micro optimization: skip kevent syscall if possible */
    740 		if (pu->pu_nfds == 1 && pu->pu_ml_timep == NULL
    741 		    && (pu->pu_state & PU_ASYNCFD) == 0) {
    742 			pfctrl = XXX->fctrl;
    743 			puffs_framev_input(pu, pfctrl, XXX);
    744 			continue;
    745 		}
    746 #endif
    747 
    748 		/* else: do full processing */
    749 		/* Don't bother worrying about O(n) for now */
    750 		LIST_FOREACH(fio, &pu->pu_ios, fio_entries) {
    751 			if (fio->stat & FIO_WRGONE)
    752 				continue;
    753 
    754 			pfctrl = fio->fctrl;
    755 
    756 			/*
    757 			 * Try to write out everything to avoid the
    758 			 * need for enabling EVFILT_WRITE.  The likely
    759 			 * case is that we can fit everything into the
    760 			 * socket buffer.
    761 			 */
    762 			puffs__framev_output(pu, pfctrl, fio);
    763 		}
    764 
    765 		/*
    766 		 * Build list of which to enable/disable in writecheck.
    767 		 */
    768 		nchanges = 0;
    769 		LIST_FOREACH(fio, &pu->pu_ios, fio_entries) {
    770 			if (fio->stat & FIO_WRGONE)
    771 				continue;
    772 
    773 			/* en/disable write checks for kqueue as needed */
    774 			assert((FIO_EN_WRITE(fio) && FIO_RM_WRITE(fio)) == 0);
    775 			if (FIO_EN_WRITE(fio)) {
    776 				EV_SET(&pu->pu_evs[nchanges], fio->io_fd,
    777 				    EVFILT_WRITE, EV_ENABLE, 0, 0,
    778 				    (uintptr_t)fio);
    779 				fio->stat |= FIO_WR;
    780 				nchanges++;
    781 			}
    782 			if (FIO_RM_WRITE(fio)) {
    783 				EV_SET(&pu->pu_evs[nchanges], fio->io_fd,
    784 				    EVFILT_WRITE, EV_DISABLE, 0, 0,
    785 				    (uintptr_t)fio);
    786 				fio->stat &= ~FIO_WR;
    787 				nchanges++;
    788 			}
    789 			assert(nchanges <= pu->pu_nfds);
    790 		}
    791 
    792 		ndone = kevent(pu->pu_kq, pu->pu_evs, nchanges,
    793 		    pu->pu_evs, 2*pu->pu_nfds, pu->pu_ml_timep);
    794 
    795 		if (ndone == -1) {
    796 			if (errno != EINTR)
    797 				break;
    798 			else
    799 				continue;
    800 		}
    801 
    802 		/* uoptimize */
    803 		if (ndone == 0)
    804 			continue;
    805 
    806 		/* iterate over the results */
    807 		for (curev = pu->pu_evs; ndone--; curev++) {
    808 			int what;
    809 
    810 #if 0
    811 			/* get & possibly dispatch events from kernel */
    812 			if (curev->ident == puffsfd) {
    813 				if (puffs_req_handle(pgr, ppr, 0) == -1)
    814 					goto out;
    815 				continue;
    816 			}
    817 #endif
    818 
    819 			fio = (void *)curev->udata;
    820 			pfctrl = fio->fctrl;
    821 			if (curev->flags & EV_ERROR) {
    822 				assert(curev->filter == EVFILT_WRITE);
    823 				fio->stat &= ~FIO_WR;
    824 
    825 				/* XXX: how to know if it's a transient error */
    826 				puffs__framev_writeclose(pu, fio,
    827 				    (int)curev->data);
    828 				puffs__framev_notify(fio, PUFFS_FBIO_ERROR);
    829 				continue;
    830 			}
    831 
    832 			what = 0;
    833 			if (curev->filter == EVFILT_READ) {
    834 				puffs__framev_input(pu, pfctrl, fio);
    835 				what |= PUFFS_FBIO_READ;
    836 			}
    837 
    838 			else if (curev->filter == EVFILT_WRITE) {
    839 				puffs__framev_output(pu, pfctrl, fio);
    840 				what |= PUFFS_FBIO_WRITE;
    841 			}
    842 			if (what)
    843 				puffs__framev_notify(fio, what);
    844 		}
    845 
    846 		/*
    847 		 * Really free fd's now that we don't have references
    848 		 * to them.
    849 		 */
    850 		while ((fio = LIST_FIRST(&pu->pu_ios_rmlist)) != NULL) {
    851 			LIST_REMOVE(fio, fio_entries);
    852 			free(fio);
    853 		}
    854 	}
    855 
    856 	if (puffs__cc_restoremain(pu) == -1)
    857 		warn("cannot restore main context.  impending doom");
    858 }
    859 
    860 int
    861 puffs_mainloop(struct puffs_usermount *pu)
    862 {
    863 	struct puffs_fctrl_io *fio;
    864 	struct puffs_cc *pcc;
    865 	struct kevent *curev;
    866 	int sverrno;
    867 
    868 	assert(puffs_getstate(pu) >= PUFFS_STATE_RUNNING);
    869 
    870 	pu->pu_kq = kqueue();
    871 	if (pu->pu_kq == -1)
    872 		goto out;
    873 	pu->pu_state |= PU_HASKQ;
    874 
    875 	puffs_setblockingmode(pu, PUFFSDEV_NONBLOCK);
    876 	if (puffs__framev_addfd_ctrl(pu, puffs_getselectable(pu),
    877 	    PUFFS_FBIO_READ | PUFFS_FBIO_WRITE,
    878 	    &pu->pu_framectrl[PU_FRAMECTRL_FS]) == -1)
    879 		goto out;
    880 
    881 	curev = realloc(pu->pu_evs, (2*pu->pu_nfds)*sizeof(struct kevent));
    882 	if (curev == NULL)
    883 		goto out;
    884 	pu->pu_evs = curev;
    885 
    886 	LIST_FOREACH(fio, &pu->pu_ios, fio_entries) {
    887 		EV_SET(curev, fio->io_fd, EVFILT_READ, EV_ADD,
    888 		    0, 0, (uintptr_t)fio);
    889 		curev++;
    890 		EV_SET(curev, fio->io_fd, EVFILT_WRITE, EV_ADD | EV_DISABLE,
    891 		    0, 0, (uintptr_t)fio);
    892 		curev++;
    893 	}
    894 	if (kevent(pu->pu_kq, pu->pu_evs, 2*pu->pu_nfds, NULL, 0, NULL) == -1)
    895 		goto out;
    896 
    897 	pu->pu_state |= PU_INLOOP;
    898 
    899 	/*
    900 	 * Create alternate execution context and jump to it.  Note
    901 	 * that we come "out" of savemain twice.  Where we come out
    902 	 * of it depends on the architecture.  If the return address is
    903 	 * stored on the stack, we jump out from puffs_cc_continue(),
    904 	 * for a register return address from puffs__cc_savemain().
    905 	 * PU_MAINRESTORE makes sure we DTRT in both cases.
    906 	 */
    907 	if (puffs__cc_create(pu, puffs__theloop, &pcc) == -1) {
    908 		goto out;
    909 	}
    910 	if (puffs__cc_savemain(pu) == -1) {
    911 		goto out;
    912 	}
    913 	if ((pu->pu_state & PU_MAINRESTORE) == 0)
    914 		puffs_cc_continue(pcc);
    915 
    916 	finalpush(pu);
    917 	errno = 0;
    918 
    919  out:
    920 	/* store the real error for a while */
    921 	sverrno = errno;
    922 
    923 	errno = sverrno;
    924 	if (errno)
    925 		return -1;
    926 	else
    927 		return 0;
    928 }
    929