Home | History | Annotate | Line # | Download | only in procfs
procfs_subr.c revision 1.45
      1 /*	$NetBSD: procfs_subr.c,v 1.45 2003/02/03 22:29:07 jdolecek Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1994 Christopher G. Demetriou.  All rights reserved.
      5  * Copyright (c) 1993 Jan-Simon Pendry
      6  * Copyright (c) 1993
      7  *	The Regents of the University of California.  All rights reserved.
      8  *
      9  * This code is derived from software contributed to Berkeley by
     10  * Jan-Simon Pendry.
     11  *
     12  * Redistribution and use in source and binary forms, with or without
     13  * modification, are permitted provided that the following conditions
     14  * are met:
     15  * 1. Redistributions of source code must retain the above copyright
     16  *    notice, this list of conditions and the following disclaimer.
     17  * 2. Redistributions in binary form must reproduce the above copyright
     18  *    notice, this list of conditions and the following disclaimer in the
     19  *    documentation and/or other materials provided with the distribution.
     20  * 3. All advertising materials mentioning features or use of this software
     21  *    must display the following acknowledgement:
     22  *	This product includes software developed by the University of
     23  *	California, Berkeley and its contributors.
     24  * 4. Neither the name of the University nor the names of its contributors
     25  *    may be used to endorse or promote products derived from this software
     26  *    without specific prior written permission.
     27  *
     28  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     29  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     30  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     31  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     32  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     33  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     34  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     35  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     36  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     37  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     38  * SUCH DAMAGE.
     39  *
     40  *	@(#)procfs_subr.c	8.6 (Berkeley) 5/14/95
     41  */
     42 
     43 #include <sys/cdefs.h>
     44 __KERNEL_RCSID(0, "$NetBSD: procfs_subr.c,v 1.45 2003/02/03 22:29:07 jdolecek Exp $");
     45 
     46 #include <sys/param.h>
     47 #include <sys/systm.h>
     48 #include <sys/time.h>
     49 #include <sys/kernel.h>
     50 #include <sys/proc.h>
     51 #include <sys/vnode.h>
     52 #include <sys/malloc.h>
     53 #include <sys/stat.h>
     54 #include <sys/file.h>
     55 #include <sys/filedesc.h>
     56 
     57 #include <miscfs/procfs/procfs.h>
     58 
     59 void procfs_hashins __P((struct pfsnode *));
     60 void procfs_hashrem __P((struct pfsnode *));
     61 struct vnode *procfs_hashget __P((pid_t, pfstype, int, struct mount *));
     62 
     63 LIST_HEAD(pfs_hashhead, pfsnode) *pfs_hashtbl;
     64 u_long	pfs_ihash;	/* size of hash table - 1 */
     65 #define PFSPIDHASH(pid)	((pid) & pfs_ihash)
     66 
     67 struct lock pfs_hashlock;
     68 struct simplelock pfs_hash_slock;
     69 
     70 #define	ISSET(t, f)	((t) & (f))
     71 
     72 /*
     73  * allocate a pfsnode/vnode pair.  the vnode is
     74  * referenced, and locked.
     75  *
     76  * the pid, pfs_type, and mount point uniquely
     77  * identify a pfsnode.  the mount point is needed
     78  * because someone might mount this filesystem
     79  * twice.
     80  *
     81  * all pfsnodes are maintained on a singly-linked
     82  * list.  new nodes are only allocated when they cannot
     83  * be found on this list.  entries on the list are
     84  * removed when the vfs reclaim entry is called.
     85  *
     86  * a single lock is kept for the entire list.  this is
     87  * needed because the getnewvnode() function can block
     88  * waiting for a vnode to become free, in which case there
     89  * may be more than one process trying to get the same
     90  * vnode.  this lock is only taken if we are going to
     91  * call getnewvnode, since the kernel itself is single-threaded.
     92  *
     93  * if an entry is found on the list, then call vget() to
     94  * take a reference.  this is done because there may be
     95  * zero references to it and so it needs to removed from
     96  * the vnode free list.
     97  */
     98 int
     99 procfs_allocvp(mp, vpp, pid, pfs_type, fd)
    100 	struct mount *mp;
    101 	struct vnode **vpp;
    102 	pid_t pid;
    103 	pfstype pfs_type;
    104 	int fd;
    105 {
    106 	struct pfsnode *pfs;
    107 	struct vnode *vp;
    108 	int error;
    109 
    110 	do {
    111 		if ((*vpp = procfs_hashget(pid, pfs_type, fd, mp)) != NULL)
    112 			return (0);
    113 	} while (lockmgr(&pfs_hashlock, LK_EXCLUSIVE|LK_SLEEPFAIL, 0));
    114 
    115 	if ((error = getnewvnode(VT_PROCFS, mp, procfs_vnodeop_p, &vp)) != 0) {
    116 		*vpp = NULL;
    117 		lockmgr(&pfs_hashlock, LK_RELEASE, NULL);
    118 		return (error);
    119 	}
    120 
    121 	MALLOC(pfs, void *, sizeof(struct pfsnode), M_TEMP, M_WAITOK);
    122 	vp->v_data = pfs;
    123 
    124 	pfs->pfs_pid = pid;
    125 	pfs->pfs_type = pfs_type;
    126 	pfs->pfs_vnode = vp;
    127 	pfs->pfs_flags = 0;
    128 	pfs->pfs_fileno = PROCFS_FILENO(pid, pfs_type, fd);
    129 	pfs->pfs_fd = fd;
    130 
    131 	switch (pfs_type) {
    132 	case Proot:	/* /proc = dr-xr-xr-x */
    133 		pfs->pfs_mode = S_IRUSR|S_IXUSR|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH;
    134 		vp->v_type = VDIR;
    135 		vp->v_flag = VROOT;
    136 		break;
    137 
    138 	case Pcurproc:	/* /proc/curproc = lr-xr-xr-x */
    139 	case Pself:	/* /proc/self    = lr-xr-xr-x */
    140 		pfs->pfs_mode = S_IRUSR|S_IXUSR|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH;
    141 		vp->v_type = VLNK;
    142 		break;
    143 
    144 	case Pproc:	/* /proc/N = dr-xr-xr-x */
    145 	case Pfd:
    146 		if (fd == -1) {	/* /proc/N/fd = dr-xr-xr-x */
    147 			pfs->pfs_mode =
    148 			    S_IRUSR|S_IXUSR|S_IRGRP|S_IXGRP|S_IROTH|S_IXOTH;
    149 			vp->v_type = VDIR;
    150 		} else {	/* /proc/N/fd/M = [ps-]rw------- */
    151 			struct file *fp;
    152 			struct vnode *vxp;
    153 
    154 			/* XXX can procfs_getfp() ever fail here? */
    155 			if ((error = procfs_getfp(pfs, &fp)) != 0)
    156 				goto bad;
    157 
    158 			pfs->pfs_mode = S_IRUSR|S_IWUSR;
    159 			switch (fp->f_type) {
    160 			case DTYPE_VNODE:
    161 				vxp = (struct vnode *)fp->f_data;
    162 				vp->v_type = vxp->v_type;
    163 				break;
    164 			case DTYPE_PIPE:
    165 				vp->v_type = VFIFO;
    166 				break;
    167 			case DTYPE_SOCKET:
    168 				vp->v_type = VSOCK;
    169 				break;
    170 			default:
    171 				error = EOPNOTSUPP;
    172 				goto bad;
    173 			}
    174 		}
    175 		break;
    176 
    177 	case Pfile:	/* /proc/N/file = -rw------- */
    178 	case Pmem:	/* /proc/N/mem = -rw------- */
    179 	case Pregs:	/* /proc/N/regs = -rw------- */
    180 	case Pfpregs:	/* /proc/N/fpregs = -rw------- */
    181 		pfs->pfs_mode = S_IRUSR|S_IWUSR;
    182 		vp->v_type = VREG;
    183 		break;
    184 
    185 	case Pctl:	/* /proc/N/ctl = --w------ */
    186 	case Pnote:	/* /proc/N/note = --w------ */
    187 	case Pnotepg:	/* /proc/N/notepg = --w------ */
    188 		pfs->pfs_mode = S_IWUSR;
    189 		vp->v_type = VREG;
    190 		break;
    191 
    192 	case Pmap:	/* /proc/N/map = -r--r--r-- */
    193 	case Pmaps:	/* /proc/N/maps = -r--r--r-- */
    194 	case Pstatus:	/* /proc/N/status = -r--r--r-- */
    195 	case Pcmdline:	/* /proc/N/cmdline = -r--r--r-- */
    196 	case Pmeminfo:	/* /proc/meminfo = -r--r--r-- */
    197 	case Pcpuinfo:	/* /proc/cpuinfo = -r--r--r-- */
    198 		pfs->pfs_mode = S_IRUSR|S_IRGRP|S_IROTH;
    199 		vp->v_type = VREG;
    200 		break;
    201 
    202 #ifdef __HAVE_PROCFS_MACHDEP
    203 	PROCFS_MACHDEP_NODETYPE_CASES
    204 		procfs_machdep_allocvp(vp);
    205 		break;
    206 #endif
    207 
    208 	default:
    209 		panic("procfs_allocvp");
    210 	}
    211 
    212 	procfs_hashins(pfs);
    213 	uvm_vnp_setsize(vp, 0);
    214 	lockmgr(&pfs_hashlock, LK_RELEASE, NULL);
    215 
    216 	*vpp = vp;
    217 	return (0);
    218 
    219  bad:
    220 	FREE(pfs, M_TEMP);
    221 	ungetnewvnode(vp);
    222 	return (error);
    223 }
    224 
    225 int
    226 procfs_freevp(vp)
    227 	struct vnode *vp;
    228 {
    229 	struct pfsnode *pfs = VTOPFS(vp);
    230 
    231 	procfs_hashrem(pfs);
    232 
    233 	FREE(vp->v_data, M_TEMP);
    234 	vp->v_data = 0;
    235 	return (0);
    236 }
    237 
    238 int
    239 procfs_rw(v)
    240 	void *v;
    241 {
    242 	struct vop_read_args *ap = v;
    243 	struct vnode *vp = ap->a_vp;
    244 	struct uio *uio = ap->a_uio;
    245 	struct proc *curp = uio->uio_procp;
    246 	struct pfsnode *pfs = VTOPFS(vp);
    247 	struct lwp *l;
    248 	struct proc *p;
    249 
    250 	p = PFIND(pfs->pfs_pid);
    251 	if (p == 0)
    252 		return (EINVAL);
    253 
    254 	/* XXX NJWLWP
    255 	 * The entire procfs interface needs work to be useful to
    256 	 * a process with multiple LWPs. For the moment, we'll
    257 	 * just kluge this and fail on others.
    258 	 */
    259 	l = proc_representative_lwp(p);
    260 
    261 	switch (pfs->pfs_type) {
    262 	case Pregs:
    263 	case Pfpregs:
    264 	case Pmem:
    265 #if defined(__HAVE_PROCFS_MACHDEP) && defined(PROCFS_MACHDEP_PROTECT_CASES)
    266 	PROCFS_MACHDEP_PROTECT_CASES
    267 #endif
    268 		/*
    269 		 * Do not allow init to be modified while in secure mode; it
    270 		 * could be duped into changing the security level.
    271 		 */
    272 		if (uio->uio_rw == UIO_WRITE &&
    273 		    p == initproc && securelevel > -1)
    274 			return (EPERM);
    275 		break;
    276 
    277 	default:
    278 		break;
    279 	}
    280 
    281 	switch (pfs->pfs_type) {
    282 	case Pnote:
    283 	case Pnotepg:
    284 		return (procfs_donote(curp, p, pfs, uio));
    285 
    286 	case Pregs:
    287 		return (procfs_doregs(curp, l, pfs, uio));
    288 
    289 	case Pfpregs:
    290 		return (procfs_dofpregs(curp, l, pfs, uio));
    291 
    292 	case Pctl:
    293 		return (procfs_doctl(curp, l, pfs, uio));
    294 
    295 	case Pstatus:
    296 		return (procfs_dostatus(curp, l, pfs, uio));
    297 
    298 	case Pmap:
    299 		return (procfs_domap(curp, p, pfs, uio, 0));
    300 
    301 	case Pmaps:
    302 		return (procfs_domap(curp, p, pfs, uio, 1));
    303 
    304 	case Pmem:
    305 		return (procfs_domem(curp, p, pfs, uio));
    306 
    307 	case Pcmdline:
    308 		return (procfs_docmdline(curp, p, pfs, uio));
    309 
    310 	case Pmeminfo:
    311 		return (procfs_domeminfo(curp, p, pfs, uio));
    312 
    313 	case Pcpuinfo:
    314 		return (procfs_docpuinfo(curp, p, pfs, uio));
    315 
    316 	case Pfd:
    317 		return (procfs_dofd(curp, p, pfs, uio));
    318 
    319 #ifdef __HAVE_PROCFS_MACHDEP
    320 	PROCFS_MACHDEP_NODETYPE_CASES
    321 		return (procfs_machdep_rw(curp, l, pfs, uio));
    322 #endif
    323 
    324 	default:
    325 		return (EOPNOTSUPP);
    326 	}
    327 }
    328 
    329 /*
    330  * Get a string from userland into (buf).  Strip a trailing
    331  * nl character (to allow easy access from the shell).
    332  * The buffer should be *buflenp + 1 chars long.  vfs_getuserstr
    333  * will automatically add a nul char at the end.
    334  *
    335  * Returns 0 on success or the following errors
    336  *
    337  * EINVAL:    file offset is non-zero.
    338  * EMSGSIZE:  message is longer than kernel buffer
    339  * EFAULT:    user i/o buffer is not addressable
    340  */
    341 int
    342 vfs_getuserstr(uio, buf, buflenp)
    343 	struct uio *uio;
    344 	char *buf;
    345 	int *buflenp;
    346 {
    347 	int xlen;
    348 	int error;
    349 
    350 	if (uio->uio_offset != 0)
    351 		return (EINVAL);
    352 
    353 	xlen = *buflenp;
    354 
    355 	/* must be able to read the whole string in one go */
    356 	if (xlen < uio->uio_resid)
    357 		return (EMSGSIZE);
    358 	xlen = uio->uio_resid;
    359 
    360 	if ((error = uiomove(buf, xlen, uio)) != 0)
    361 		return (error);
    362 
    363 	/* allow multiple writes without seeks */
    364 	uio->uio_offset = 0;
    365 
    366 	/* cleanup string and remove trailing newline */
    367 	buf[xlen] = '\0';
    368 	xlen = strlen(buf);
    369 	if (xlen > 0 && buf[xlen-1] == '\n')
    370 		buf[--xlen] = '\0';
    371 	*buflenp = xlen;
    372 
    373 	return (0);
    374 }
    375 
    376 const vfs_namemap_t *
    377 vfs_findname(nm, buf, buflen)
    378 	const vfs_namemap_t *nm;
    379 	const char *buf;
    380 	int buflen;
    381 {
    382 
    383 	for (; nm->nm_name; nm++)
    384 		if (memcmp(buf, nm->nm_name, buflen+1) == 0)
    385 			return (nm);
    386 
    387 	return (0);
    388 }
    389 
    390 /*
    391  * Initialize pfsnode hash table.
    392  */
    393 void
    394 procfs_hashinit()
    395 {
    396 	lockinit(&pfs_hashlock, PINOD, "pfs_hashlock", 0, 0);
    397 	pfs_hashtbl = hashinit(desiredvnodes / 4, HASH_LIST, M_UFSMNT,
    398 	    M_WAITOK, &pfs_ihash);
    399 	simple_lock_init(&pfs_hash_slock);
    400 }
    401 
    402 void
    403 procfs_hashreinit()
    404 {
    405 	struct pfsnode *pp;
    406 	struct pfs_hashhead *oldhash, *hash;
    407 	u_long i, oldmask, mask, val;
    408 
    409 	hash = hashinit(desiredvnodes / 4, HASH_LIST, M_UFSMNT, M_WAITOK,
    410 	    &mask);
    411 
    412 	simple_lock(&pfs_hash_slock);
    413 	oldhash = pfs_hashtbl;
    414 	oldmask = pfs_ihash;
    415 	pfs_hashtbl = hash;
    416 	pfs_ihash = mask;
    417 	for (i = 0; i <= oldmask; i++) {
    418 		while ((pp = LIST_FIRST(&oldhash[i])) != NULL) {
    419 			LIST_REMOVE(pp, pfs_hash);
    420 			val = PFSPIDHASH(pp->pfs_pid);
    421 			LIST_INSERT_HEAD(&hash[val], pp, pfs_hash);
    422 		}
    423 	}
    424 	simple_unlock(&pfs_hash_slock);
    425 	hashdone(oldhash, M_UFSMNT);
    426 }
    427 
    428 /*
    429  * Free pfsnode hash table.
    430  */
    431 void
    432 procfs_hashdone()
    433 {
    434 	hashdone(pfs_hashtbl, M_UFSMNT);
    435 }
    436 
    437 struct vnode *
    438 procfs_hashget(pid, type, fd, mp)
    439 	pid_t pid;
    440 	pfstype type;
    441 	int fd;
    442 	struct mount *mp;
    443 {
    444 	struct pfs_hashhead *ppp;
    445 	struct pfsnode *pp;
    446 	struct vnode *vp;
    447 
    448 loop:
    449 	simple_lock(&pfs_hash_slock);
    450 	ppp = &pfs_hashtbl[PFSPIDHASH(pid)];
    451 	LIST_FOREACH(pp, ppp, pfs_hash) {
    452 		vp = PFSTOV(pp);
    453 		if (pid == pp->pfs_pid && pp->pfs_type == type &&
    454 		    pp->pfs_fd == fd && vp->v_mount == mp) {
    455 			simple_lock(&vp->v_interlock);
    456 			simple_unlock(&pfs_hash_slock);
    457 			if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK))
    458 				goto loop;
    459 			return (vp);
    460 		}
    461 	}
    462 	simple_unlock(&pfs_hash_slock);
    463 	return (NULL);
    464 }
    465 
    466 /*
    467  * Insert the pfsnode into the hash table and lock it.
    468  */
    469 void
    470 procfs_hashins(pp)
    471 	struct pfsnode *pp;
    472 {
    473 	struct pfs_hashhead *ppp;
    474 
    475 	/* lock the pfsnode, then put it on the appropriate hash list */
    476 	lockmgr(&pp->pfs_vnode->v_lock, LK_EXCLUSIVE, (struct simplelock *)0);
    477 
    478 	simple_lock(&pfs_hash_slock);
    479 	ppp = &pfs_hashtbl[PFSPIDHASH(pp->pfs_pid)];
    480 	LIST_INSERT_HEAD(ppp, pp, pfs_hash);
    481 	simple_unlock(&pfs_hash_slock);
    482 }
    483 
    484 /*
    485  * Remove the pfsnode from the hash table.
    486  */
    487 void
    488 procfs_hashrem(pp)
    489 	struct pfsnode *pp;
    490 {
    491 	simple_lock(&pfs_hash_slock);
    492 	LIST_REMOVE(pp, pfs_hash);
    493 	simple_unlock(&pfs_hash_slock);
    494 }
    495 
    496 void
    497 procfs_revoke_vnodes(p, arg)
    498 	struct proc *p;
    499 	void *arg;
    500 {
    501 	struct pfsnode *pfs, *pnext;
    502 	struct vnode *vp;
    503 	struct mount *mp = (struct mount *)arg;
    504 	struct pfs_hashhead *ppp;
    505 
    506 	if (!(p->p_flag & P_SUGID))
    507 		return;
    508 
    509 	ppp = &pfs_hashtbl[PFSPIDHASH(p->p_pid)];
    510 	for (pfs = LIST_FIRST(ppp); pfs; pfs = pnext) {
    511 		vp = PFSTOV(pfs);
    512 		pnext = LIST_NEXT(pfs, pfs_hash);
    513 		if (vp->v_usecount > 0 && pfs->pfs_pid == p->p_pid &&
    514 		    vp->v_mount == mp)
    515 			VOP_REVOKE(vp, REVOKEALL);
    516 	}
    517 }
    518 
    519 int
    520 procfs_getfp(pfs, fp)
    521 	struct pfsnode *pfs;
    522 	struct file **fp;
    523 {
    524 	struct proc *p = PFIND(pfs->pfs_pid);
    525 
    526 	if (p == NULL)
    527 		return ESRCH;
    528 
    529 	if (pfs->pfs_fd == -1)
    530 		return EINVAL;
    531 
    532 	if ((*fp = p->p_fd->fd_ofiles[pfs->pfs_fd]) == NULL)
    533 		return EBADF;
    534 	return 0;
    535 }
    536