1 /* 2 * Copyright (c) 1993 Jan-Simon Pendry 3 * Copyright (c) 1993 4 * The Regents of the University of California. All rights reserved. 5 * 6 * This code is derived from software contributed to Berkeley by 7 * Jan-Simon Pendry. 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 * 3. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)procfs_subr.c 8.6 (Berkeley) 5/14/95 34 * 35 * $FreeBSD: src/sys/miscfs/procfs/procfs_subr.c,v 1.26.2.3 2002/02/18 21:28:04 des Exp $ 36 */ 37 38 #include <sys/param.h> 39 #include <sys/systm.h> 40 #include <sys/sysctl.h> 41 #include <sys/proc.h> 42 #include <sys/mount.h> 43 #include <sys/vnode.h> 44 #include <sys/malloc.h> 45 #include <sys/thread2.h> 46 47 #include <vfs/procfs/procfs.h> 48 49 #define PFS_HSIZE 256 50 #define PFS_HMASK (PFS_HSIZE - 1) 51 52 static struct pfsnode *pfshead[PFS_HSIZE]; 53 static int pfsvplock; 54 55 #define PFSHASH(pid) &pfshead[(pid) & PFS_HMASK] 56 57 /* 58 * Allocate a pfsnode/vnode pair. If no error occurs the returned vnode 59 * will be referenced and exclusively locked. 60 * 61 * The pid, pfs_type, and mount point uniquely identify a pfsnode. 62 * The mount point is needed because someone might mount this filesystem 63 * twice. 64 * 65 * All pfsnodes are maintained on a singly-linked list. new nodes are 66 * only allocated when they cannot be found on this list. entries on 67 * the list are removed when the vfs reclaim entry is called. 68 * 69 * A single lock is kept for the entire list. this is needed because the 70 * getnewvnode() function can block waiting for a vnode to become free, 71 * in which case there may be more than one process trying to get the same 72 * vnode. this lock is only taken if we are going to call getnewvnode, 73 * since the kernel itself is single-threaded. 74 * 75 * If an entry is found on the list, then call vget() to take a reference 76 * and obtain the lock. This will properly re-reference the vnode if it 77 * had gotten onto the free list. 78 */ 79 int 80 procfs_allocvp(struct mount *mp, struct vnode **vpp, long pid, pfstype pfs_type) 81 { 82 struct pfsnode *pfs; 83 struct vnode *vp; 84 struct pfsnode **pp; 85 int error; 86 87 pp = PFSHASH(pid); 88 loop: 89 for (pfs = *pp; pfs; pfs = pfs->pfs_next) { 90 if (pfs->pfs_pid == pid && pfs->pfs_type == pfs_type && 91 PFSTOV(pfs)->v_mount == mp) { 92 vp = PFSTOV(pfs); 93 vhold_interlocked(vp); 94 if (vget(vp, LK_EXCLUSIVE)) { 95 vdrop(vp); 96 goto loop; 97 } 98 99 /* 100 * Make sure the vnode is still in the cache after 101 * getting the interlock to avoid racing a free. 102 */ 103 for (pfs = *pp; pfs; pfs = pfs->pfs_next) { 104 if (PFSTOV(pfs) == vp && 105 pfs->pfs_pid == pid && 106 pfs->pfs_type == pfs_type && 107 PFSTOV(pfs)->v_mount == mp) { 108 break; 109 } 110 } 111 vdrop(vp); 112 if (pfs == NULL || PFSTOV(pfs) != vp) { 113 vput(vp); 114 goto loop; 115 116 } 117 KKASSERT(vp->v_data == pfs); 118 *vpp = vp; 119 return (0); 120 } 121 } 122 123 /* 124 * otherwise lock the vp list while we call getnewvnode 125 * since that can block. 126 */ 127 if (pfsvplock & PROCFS_LOCKED) { 128 pfsvplock |= PROCFS_WANT; 129 (void) tsleep((caddr_t) &pfsvplock, 0, "pfsavp", 0); 130 goto loop; 131 } 132 pfsvplock |= PROCFS_LOCKED; 133 134 /* 135 * Do the MALLOC before the getnewvnode since doing so afterward 136 * might cause a bogus v_data pointer to get dereferenced 137 * elsewhere if MALLOC should block. 138 * 139 * XXX this may not matter anymore since getnewvnode now returns 140 * a VX locked vnode. 141 */ 142 pfs = kmalloc(sizeof(struct pfsnode), M_TEMP, M_WAITOK); 143 144 error = getnewvnode(VT_PROCFS, mp, vpp, 0, 0); 145 if (error) { 146 kfree(pfs, M_TEMP); 147 goto out; 148 } 149 vp = *vpp; 150 151 vp->v_data = pfs; 152 153 pfs->pfs_next = 0; 154 pfs->pfs_pid = (pid_t) pid; 155 pfs->pfs_type = pfs_type; 156 pfs->pfs_vnode = vp; 157 pfs->pfs_flags = 0; 158 pfs->pfs_lockowner = 0; 159 pfs->pfs_fileno = PROCFS_FILENO(pid, pfs_type); 160 161 switch (pfs_type) { 162 case Proot: /* /proc = dr-xr-xr-x */ 163 pfs->pfs_mode = (VREAD|VEXEC) | 164 (VREAD|VEXEC) >> 3 | 165 (VREAD|VEXEC) >> 6; 166 vp->v_type = VDIR; 167 vp->v_flag = VROOT; 168 break; 169 170 case Pcurproc: /* /proc/curproc = lr--r--r-- */ 171 pfs->pfs_mode = (VREAD) | 172 (VREAD >> 3) | 173 (VREAD >> 6); 174 vp->v_type = VLNK; 175 break; 176 177 case Pproc: 178 pfs->pfs_mode = (VREAD|VEXEC) | 179 (VREAD|VEXEC) >> 3 | 180 (VREAD|VEXEC) >> 6; 181 vp->v_type = VDIR; 182 break; 183 184 case Pfile: 185 pfs->pfs_mode = (VREAD|VEXEC) | 186 (VREAD|VEXEC) >> 3 | 187 (VREAD|VEXEC) >> 6; 188 vp->v_type = VLNK; 189 break; 190 191 case Pmem: 192 pfs->pfs_mode = (VREAD|VWRITE); 193 vp->v_type = VREG; 194 break; 195 196 case Pregs: 197 case Pfpregs: 198 case Pdbregs: 199 pfs->pfs_mode = (VREAD|VWRITE); 200 vp->v_type = VREG; 201 break; 202 203 case Pctl: 204 case Pnote: 205 case Pnotepg: 206 pfs->pfs_mode = (VWRITE); 207 vp->v_type = VREG; 208 break; 209 210 case Ptype: 211 case Pmap: 212 case Pstatus: 213 case Pcmdline: 214 case Prlimit: 215 pfs->pfs_mode = (VREAD) | 216 (VREAD >> 3) | 217 (VREAD >> 6); 218 vp->v_type = VREG; 219 break; 220 221 default: 222 panic("procfs_allocvp"); 223 } 224 225 /* add to procfs vnode list */ 226 pfs->pfs_next = *pp; 227 *pp = pfs; 228 229 out: 230 pfsvplock &= ~PROCFS_LOCKED; 231 232 if (pfsvplock & PROCFS_WANT) { 233 pfsvplock &= ~PROCFS_WANT; 234 wakeup((caddr_t) &pfsvplock); 235 } 236 237 return (error); 238 } 239 240 int 241 procfs_freevp(struct vnode *vp) 242 { 243 struct pfsnode **pfspp; 244 struct pfsnode *pfs; 245 246 pfs = VTOPFS(vp); 247 vp->v_data = NULL; 248 249 pfspp = PFSHASH(pfs->pfs_pid); 250 while (*pfspp != pfs && *pfspp) 251 pfspp = &(*pfspp)->pfs_next; 252 KKASSERT(*pfspp); 253 *pfspp = pfs->pfs_next; 254 pfs->pfs_next = NULL; 255 pfs->pfs_vnode = NULL; 256 kfree(pfs, M_TEMP); 257 return (0); 258 } 259 260 /* 261 * Try to find the calling pid. Note that pfind() 262 * now references the proc structure to be returned 263 * and needs to be released later with PRELE(). 264 */ 265 struct proc * 266 pfs_pfind(pid_t pfs_pid) 267 { 268 struct proc *p = NULL; 269 270 if (pfs_pid == 0) { 271 p = &proc0; 272 PHOLD(p); 273 } else { 274 p = pfind(pfs_pid); 275 } 276 277 /* 278 * Make sure the process is not in the middle of exiting (where 279 * a lot of its structural members may wind up being NULL). If it 280 * is we give up on it. 281 */ 282 if (p) { 283 lwkt_gettoken(&p->p_token); 284 if (p->p_flags & P_POSTEXIT) { 285 lwkt_reltoken(&p->p_token); 286 PRELE(p); 287 p = NULL; 288 } 289 } 290 return p; 291 } 292 293 struct proc * 294 pfs_zpfind(pid_t pfs_pid) 295 { 296 struct proc *p = NULL; 297 298 if (pfs_pid == 0) { 299 p = &proc0; 300 PHOLD(p); 301 } else { 302 p = zpfind(pfs_pid); 303 } 304 305 /* 306 * Make sure the process is not in the middle of exiting (where 307 * a lot of its structural members may wind up being NULL). If it 308 * is we give up on it. 309 */ 310 if (p) { 311 lwkt_gettoken(&p->p_token); 312 if (p->p_flags & P_POSTEXIT) { 313 lwkt_reltoken(&p->p_token); 314 PRELE(p); 315 p = NULL; 316 } 317 } 318 return p; 319 } 320 321 void 322 pfs_pdone(struct proc *p) 323 { 324 if (p) { 325 lwkt_reltoken(&p->p_token); 326 PRELE(p); 327 } 328 } 329 330 int 331 procfs_rw(struct vop_read_args *ap) 332 { 333 struct vnode *vp = ap->a_vp; 334 struct uio *uio = ap->a_uio; 335 struct thread *curtd = uio->uio_td; 336 struct proc *curp; 337 struct pfsnode *pfs = VTOPFS(vp); 338 struct proc *p; 339 struct lwp *lp; 340 int rtval; 341 342 if (curtd == NULL) 343 return (EINVAL); 344 if ((curp = curtd->td_proc) == NULL) /* XXX */ 345 return (EINVAL); 346 347 lwkt_gettoken(&proc_token); 348 p = pfs_pfind(pfs->pfs_pid); 349 if (p == NULL) { 350 rtval = EINVAL; 351 goto out; 352 } 353 if (p->p_pid == 1 && securelevel > 0 && uio->uio_rw == UIO_WRITE) { 354 rtval = EACCES; 355 goto out; 356 } 357 /* XXX lwp */ 358 lp = FIRST_LWP_IN_PROC(p); 359 LWPHOLD(lp); 360 361 while (pfs->pfs_lockowner) { 362 tsleep(&pfs->pfs_lockowner, 0, "pfslck", 0); 363 } 364 pfs->pfs_lockowner = curproc->p_pid; 365 366 switch (pfs->pfs_type) { 367 case Pnote: 368 case Pnotepg: 369 rtval = procfs_donote(curp, lp, pfs, uio); 370 break; 371 372 case Pregs: 373 rtval = procfs_doregs(curp, lp, pfs, uio); 374 break; 375 376 case Pfpregs: 377 rtval = procfs_dofpregs(curp, lp, pfs, uio); 378 break; 379 380 case Pdbregs: 381 rtval = procfs_dodbregs(curp, lp, pfs, uio); 382 break; 383 384 case Pctl: 385 rtval = procfs_doctl(curp, lp, pfs, uio); 386 break; 387 388 case Pstatus: 389 rtval = procfs_dostatus(curp, lp, pfs, uio); 390 break; 391 392 case Pmap: 393 rtval = procfs_domap(curp, lp, pfs, uio); 394 break; 395 396 case Pmem: 397 rtval = procfs_domem(curp, lp, pfs, uio); 398 break; 399 400 case Ptype: 401 rtval = procfs_dotype(curp, lp, pfs, uio); 402 break; 403 404 case Pcmdline: 405 rtval = procfs_docmdline(curp, lp, pfs, uio); 406 break; 407 408 case Prlimit: 409 rtval = procfs_dorlimit(curp, lp, pfs, uio); 410 break; 411 412 default: 413 rtval = EOPNOTSUPP; 414 break; 415 } 416 LWPRELE(lp); 417 418 pfs->pfs_lockowner = 0; 419 wakeup(&pfs->pfs_lockowner); 420 421 out: 422 pfs_pdone(p); 423 lwkt_reltoken(&proc_token); 424 425 return rtval; 426 } 427 428 /* 429 * Get a string from userland into (buf). Strip a trailing 430 * nl character (to allow easy access from the shell). 431 * The buffer should be *buflenp + 1 chars long. vfs_getuserstr 432 * will automatically add a nul char at the end. 433 * 434 * Returns 0 on success or the following errors 435 * 436 * EINVAL: file offset is non-zero. 437 * EMSGSIZE: message is longer than kernel buffer 438 * EFAULT: user i/o buffer is not addressable 439 */ 440 int 441 vfs_getuserstr(struct uio *uio, char *buf, int *buflenp) 442 { 443 int xlen; 444 int error; 445 446 if (uio->uio_offset != 0) 447 return (EINVAL); 448 449 xlen = *buflenp; 450 451 /* must be able to read the whole string in one go */ 452 if (xlen < uio->uio_resid) 453 return (EMSGSIZE); 454 xlen = uio->uio_resid; 455 456 if ((error = uiomove(buf, xlen, uio)) != 0) 457 return (error); 458 459 /* allow multiple writes without seeks */ 460 uio->uio_offset = 0; 461 462 /* cleanup string and remove trailing newline */ 463 buf[xlen] = '\0'; 464 xlen = strlen(buf); 465 if (xlen > 0 && buf[xlen-1] == '\n') 466 buf[--xlen] = '\0'; 467 *buflenp = xlen; 468 469 return (0); 470 } 471 472 vfs_namemap_t * 473 vfs_findname(vfs_namemap_t *nm, char *buf, int buflen) 474 { 475 476 for (; nm->nm_name; nm++) 477 if (bcmp(buf, nm->nm_name, buflen+1) == 0) 478 return (nm); 479 480 return (0); 481 } 482 483 void 484 procfs_exit(struct thread *td) 485 { 486 struct pfsnode *pfs; 487 struct vnode *vp; 488 pid_t pid; 489 490 KKASSERT(td->td_proc); 491 pid = td->td_proc->p_pid; 492 493 /* 494 * NOTE: We can't just vgone() the vnode any more, not while 495 * it may potentially still be active. This will clean 496 * the vp and clear the mount and cause the new VOP subsystem 497 * to assert or panic when someone tries to do an operation 498 * on an open (exited) procfs descriptor. 499 * 500 * Prevent further operations on this pid by setting pfs_pid to -1. 501 * Note that a pfs_pid of 0 is used for nodes which do not track 502 * any particular pid. 503 * 504 * Use vx_get() to properly ref/lock a vp which may not have any 505 * refs and which may or may not already be reclaimed. vx_put() 506 * will then properly deactivate it and cause it to be recycled. 507 * 508 * The hash table can also get ripped out from under us when 509 * we block so take the easy way out and restart the scan. 510 */ 511 again: 512 pfs = *PFSHASH(pid); 513 while (pfs) { 514 if (pfs->pfs_pid == pid) { 515 vp = PFSTOV(pfs); 516 vx_get(vp); 517 pfs->pfs_pid |= PFS_DEAD; /* does not effect hash */ 518 vx_put(vp); 519 goto again; 520 } 521 pfs = pfs->pfs_next; 522 } 523 } 524 525