1 /* 2 * Copyright (c) 1993, 1995 Jan-Simon Pendry 3 * Copyright (c) 1993, 1995 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_vnops.c 8.18 (Berkeley) 5/21/95 34 * 35 * $FreeBSD: src/sys/miscfs/procfs/procfs_vnops.c,v 1.76.2.7 2002/01/22 17:22:59 nectar Exp $ 36 */ 37 38 /* 39 * procfs vnode interface 40 */ 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/time.h> 45 #include <sys/kernel.h> 46 #include <sys/lock.h> 47 #include <sys/fcntl.h> 48 #include <sys/proc.h> 49 #include <sys/caps.h> 50 #include <sys/signalvar.h> 51 #include <sys/vnode.h> 52 #include <sys/uio.h> 53 #include <sys/mount.h> 54 #include <sys/namei.h> 55 #include <sys/dirent.h> 56 #include <sys/malloc.h> 57 #include <sys/reg.h> 58 #include <vm/vm_zone.h> 59 #include <vfs/procfs/procfs.h> 60 #include <sys/pioctl.h> 61 62 #include <sys/spinlock2.h> 63 64 #include <machine/limits.h> 65 66 static int procfs_access (struct vop_access_args *); 67 static int procfs_badop (struct vop_generic_args *); 68 static int procfs_bmap (struct vop_bmap_args *); 69 static int procfs_close (struct vop_close_args *); 70 static int procfs_getattr (struct vop_getattr_args *); 71 static int procfs_inactive (struct vop_inactive_args *); 72 static int procfs_ioctl (struct vop_ioctl_args *); 73 static int procfs_lookup (struct vop_old_lookup_args *); 74 static int procfs_open (struct vop_open_args *); 75 static int procfs_print (struct vop_print_args *); 76 static int procfs_readdir (struct vop_readdir_args *); 77 static int procfs_readlink (struct vop_readlink_args *); 78 static int procfs_reclaim (struct vop_reclaim_args *); 79 static int procfs_setattr (struct vop_setattr_args *); 80 81 static int procfs_readdir_proc(struct vop_readdir_args *); 82 static int procfs_readdir_root(struct vop_readdir_args *); 83 84 /* 85 * procfs vnode operations. 86 */ 87 struct vop_ops procfs_vnode_vops = { 88 .vop_default = vop_defaultop, 89 .vop_access = procfs_access, 90 .vop_advlock = (void *)procfs_badop, 91 .vop_bmap = procfs_bmap, 92 .vop_close = procfs_close, 93 .vop_old_create = (void *)procfs_badop, 94 .vop_getattr = procfs_getattr, 95 .vop_inactive = procfs_inactive, 96 .vop_old_link = (void *)procfs_badop, 97 .vop_old_lookup = procfs_lookup, 98 .vop_old_mkdir = (void *)procfs_badop, 99 .vop_old_mknod = (void *)procfs_badop, 100 .vop_open = procfs_open, 101 .vop_pathconf = vop_stdpathconf, 102 .vop_print = procfs_print, 103 .vop_read = procfs_rw, 104 .vop_readdir = procfs_readdir, 105 .vop_readlink = procfs_readlink, 106 .vop_reclaim = procfs_reclaim, 107 .vop_old_remove = (void *)procfs_badop, 108 .vop_old_rename = (void *)procfs_badop, 109 .vop_old_rmdir = (void *)procfs_badop, 110 .vop_setattr = procfs_setattr, 111 .vop_old_symlink = (void *)procfs_badop, 112 .vop_write = (void *)procfs_rw, 113 .vop_ioctl = procfs_ioctl 114 }; 115 116 117 /* 118 * This is a list of the valid names in the 119 * process-specific sub-directories. It is 120 * used in procfs_lookup and procfs_readdir 121 */ 122 static struct proc_target { 123 u_char pt_type; 124 u_char pt_namlen; 125 char *pt_name; 126 pfstype pt_pfstype; 127 int (*pt_valid) (struct lwp *p); 128 } proc_targets[] = { 129 #define N(s) sizeof(s)-1, s 130 /* name type validp */ 131 { DT_DIR, N("."), Pproc, NULL }, 132 { DT_DIR, N(".."), Proot, NULL }, 133 { DT_REG, N("mem"), Pmem, NULL }, 134 { DT_REG, N("regs"), Pregs, procfs_validregs }, 135 { DT_REG, N("fpregs"), Pfpregs, procfs_validfpregs }, 136 { DT_REG, N("dbregs"), Pdbregs, procfs_validdbregs }, 137 { DT_REG, N("ctl"), Pctl, NULL }, 138 { DT_REG, N("status"), Pstatus, NULL }, 139 { DT_REG, N("note"), Pnote, NULL }, 140 { DT_REG, N("notepg"), Pnotepg, NULL }, 141 { DT_REG, N("map"), Pmap, procfs_validmap }, 142 { DT_REG, N("etype"), Ptype, procfs_validtype }, 143 { DT_REG, N("cmdline"), Pcmdline, NULL }, 144 { DT_REG, N("rlimit"), Prlimit, NULL }, 145 { DT_LNK, N("file"), Pfile, NULL }, 146 { DT_LNK, N("exe"), Pfile, NULL }, 147 #undef N 148 }; 149 static const int nproc_targets = NELEM(proc_targets); 150 151 static pid_t atopid (const char *, u_int); 152 153 /* 154 * set things up for doing i/o on 155 * the pfsnode (vp). (vp) is locked 156 * on entry, and should be left locked 157 * on exit. 158 * 159 * for procfs we don't need to do anything 160 * in particular for i/o. all that is done 161 * is to support exclusive open on process 162 * memory images. 163 * 164 * procfs_open(struct vnode *a_vp, int a_mode, struct ucred *a_cred, 165 * struct file *a_fp) 166 */ 167 static int 168 procfs_open(struct vop_open_args *ap) 169 { 170 struct pfsnode *pfs = VTOPFS(ap->a_vp); 171 struct proc *p1, *p2; 172 int error; 173 174 p2 = pfs_pfind(pfs->pfs_pid); 175 if (p2 == NULL) 176 return (ENOENT); 177 if (pfs->pfs_pid && !PRISON_CHECK(ap->a_cred, p2->p_ucred)) { 178 error = ENOENT; 179 goto done; 180 } 181 182 switch (pfs->pfs_type) { 183 case Pmem: 184 if (((pfs->pfs_flags & FWRITE) && (ap->a_mode & O_EXCL)) || 185 ((pfs->pfs_flags & O_EXCL) && (ap->a_mode & FWRITE))) { 186 error = EBUSY; 187 goto done; 188 } 189 190 p1 = curproc; 191 KKASSERT(p1); 192 /* Can't trace a process that's currently exec'ing. */ 193 if ((p2->p_flags & P_INEXEC) != 0) { 194 error = EAGAIN; 195 goto done; 196 } 197 if (!CHECKIO(p1, p2) || p_trespass(ap->a_cred, p2->p_ucred)) { 198 error = EPERM; 199 goto done; 200 } 201 202 if (ap->a_mode & FWRITE) 203 pfs->pfs_flags = ap->a_mode & (FWRITE|O_EXCL); 204 205 break; 206 207 default: 208 break; 209 } 210 error = vop_stdopen(ap); 211 done: 212 pfs_pdone(p2); 213 return error; 214 } 215 216 /* 217 * close the pfsnode (vp) after doing i/o. 218 * (vp) is not locked on entry or exit. 219 * 220 * nothing to do for procfs other than undo 221 * any exclusive open flag (see _open above). 222 * 223 * procfs_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred) 224 */ 225 static int 226 procfs_close(struct vop_close_args *ap) 227 { 228 struct pfsnode *pfs = VTOPFS(ap->a_vp); 229 struct proc *p; 230 231 /* 232 * Make sure the lock is exclusive for opencount tests 233 */ 234 vn_lock(ap->a_vp, LK_UPGRADE | LK_RETRY); 235 236 switch (pfs->pfs_type) { 237 case Pmem: 238 if ((ap->a_fflag & FWRITE) && (pfs->pfs_flags & O_EXCL)) 239 pfs->pfs_flags &= ~(FWRITE|O_EXCL); 240 /* 241 * v_opencount determines the last real close on the vnode. 242 * 243 * If this is the last close, then it checks to see if 244 * the target process has PF_LINGER set in p_pfsflags, 245 * if this is *not* the case, then the process' stop flags 246 * are cleared, and the process is woken up. This is 247 * to help prevent the case where a process has been 248 * told to stop on an event, but then the requesting process 249 * has gone away or forgotten about it. 250 */ 251 p = NULL; 252 if ((ap->a_vp->v_opencount < 2) 253 && ((p = pfs_pfind(pfs->pfs_pid)) != NULL || 254 (p = pfs_zpfind(pfs->pfs_pid)) != NULL) 255 && !(p->p_pfsflags & PF_LINGER)) { 256 spin_lock(&p->p_spin); 257 p->p_stops = 0; 258 p->p_step = 0; 259 spin_unlock(&p->p_spin); 260 wakeup(&p->p_stype); 261 wakeup(&p->p_step); 262 } 263 pfs_pdone(p); 264 break; 265 default: 266 break; 267 } 268 269 return (vop_stdclose(ap)); 270 } 271 272 /* 273 * do an ioctl operation on a pfsnode (vp). 274 * (vp) is not locked on entry or exit. 275 */ 276 static int 277 procfs_ioctl(struct vop_ioctl_args *ap) 278 { 279 struct pfsnode *pfs = VTOPFS(ap->a_vp); 280 struct proc *procp; 281 struct proc *p; 282 int error; 283 int signo; 284 struct procfs_status *psp; 285 unsigned char flags; 286 287 procp = pfs_pfind(pfs->pfs_pid); 288 if (procp == NULL) 289 return ENOTTY; 290 p = curproc; 291 if (p == NULL) { 292 error = EINVAL; 293 goto done; 294 } 295 296 /* Can't trace a process that's currently exec'ing. */ 297 if ((procp->p_flags & P_INEXEC) != 0) { 298 error = EAGAIN; 299 goto done; 300 } 301 if (!CHECKIO(p, procp) || p_trespass(ap->a_cred, procp->p_ucred)) { 302 error = EPERM; 303 goto done; 304 } 305 306 switch (ap->a_command) { 307 case PIOCBIS: 308 spin_lock(&procp->p_spin); 309 procp->p_stops |= *(unsigned int*)ap->a_data; 310 spin_unlock(&procp->p_spin); 311 break; 312 case PIOCBIC: 313 spin_lock(&procp->p_spin); 314 procp->p_stops &= ~*(unsigned int*)ap->a_data; 315 spin_unlock(&procp->p_spin); 316 break; 317 case PIOCSFL: 318 /* 319 * NFLAGS is "non-suser_xxx flags" -- currently, only 320 * PFS_ISUGID ("ignore set u/g id"); 321 */ 322 #define NFLAGS (PF_ISUGID) 323 flags = (unsigned char)*(unsigned int*)ap->a_data; 324 if (flags & NFLAGS && 325 (error = caps_priv_check(ap->a_cred, SYSCAP_RESTRICTEDROOT))) 326 { 327 goto done; 328 } 329 procp->p_pfsflags = flags; 330 break; 331 case PIOCGFL: 332 *(unsigned int*)ap->a_data = (unsigned int)procp->p_pfsflags; 333 break; 334 case PIOCSTATUS: 335 /* 336 * NOTE: syscall entry deals with stopevents and may run without 337 * the MP lock. 338 */ 339 psp = (struct procfs_status *)ap->a_data; 340 psp->flags = procp->p_pfsflags; 341 psp->events = procp->p_stops; 342 spin_lock(&procp->p_spin); 343 if (procp->p_step) { 344 psp->state = 0; 345 psp->why = procp->p_stype; 346 psp->val = procp->p_xstat; 347 spin_unlock(&procp->p_spin); 348 } else { 349 psp->state = 1; 350 spin_unlock(&procp->p_spin); 351 psp->why = 0; /* Not defined values */ 352 psp->val = 0; /* Not defined values */ 353 } 354 break; 355 case PIOCWAIT: 356 /* 357 * NOTE: syscall entry deals with stopevents and may run without 358 * the MP lock. 359 */ 360 psp = (struct procfs_status *)ap->a_data; 361 spin_lock(&procp->p_spin); 362 while (procp->p_step == 0) { 363 tsleep_interlock(&procp->p_stype, PCATCH); 364 spin_unlock(&procp->p_spin); 365 if (procp->p_stops == 0) { 366 error = 0; 367 goto done; 368 } 369 if (procp->p_flags & P_POSTEXIT) { 370 error = EINVAL; 371 goto done; 372 } 373 if (procp->p_flags & P_INEXEC) { 374 error = EAGAIN; 375 goto done; 376 } 377 error = tsleep(&procp->p_stype, PCATCH | PINTERLOCKED, 378 "piocwait", 0); 379 if (error) 380 goto done; 381 spin_lock(&procp->p_spin); 382 } 383 spin_unlock(&procp->p_spin); 384 psp->state = 1; /* It stopped */ 385 psp->flags = procp->p_pfsflags; 386 psp->events = procp->p_stops; 387 psp->why = procp->p_stype; /* why it stopped */ 388 psp->val = procp->p_xstat; /* any extra info */ 389 break; 390 case PIOCCONT: /* Restart a proc */ 391 /* 392 * NOTE: syscall entry deals with stopevents and may run without 393 * the MP lock. However, the caller is presumably interlocked 394 * by having waited. 395 */ 396 if (procp->p_step == 0) { 397 error = EINVAL; /* Can only start a stopped process */ 398 goto done; 399 } 400 if ((signo = *(int*)ap->a_data) != 0) { 401 if (signo >= NSIG || signo <= 0) { 402 error = EINVAL; 403 goto done; 404 } 405 ksignal(procp, signo); 406 } 407 procp->p_step = 0; 408 wakeup(&procp->p_step); 409 break; 410 default: 411 error = ENOTTY; 412 goto done; 413 } 414 error = 0; 415 done: 416 pfs_pdone(procp); 417 return error; 418 } 419 420 /* 421 * do block mapping for pfsnode (vp). 422 * since we don't use the buffer cache 423 * for procfs this function should never 424 * be called. in any case, it's not clear 425 * what part of the kernel ever makes use 426 * of this function. for sanity, this is the 427 * usual no-op bmap, although returning 428 * (EIO) would be a reasonable alternative. 429 * 430 * XXX mmap assumes buffer cache operation 431 * 432 * procfs_bmap(struct vnode *a_vp, off_t a_loffset, 433 * off_t *a_doffsetp, int *a_runp, int *a_runb) 434 */ 435 static int 436 procfs_bmap(struct vop_bmap_args *ap) 437 { 438 if (ap->a_doffsetp != NULL) 439 *ap->a_doffsetp = ap->a_loffset; 440 if (ap->a_runp != NULL) 441 *ap->a_runp = 0; 442 if (ap->a_runb != NULL) 443 *ap->a_runb = 0; 444 return (0); 445 } 446 447 /* 448 * procfs_inactive is called when the pfsnode 449 * is vrele'd and the reference count goes 450 * to zero. (vp) will be on the vnode free 451 * list, so to get it back vget() must be 452 * used. 453 * 454 * (vp) is locked on entry, but must be unlocked on exit. 455 * 456 * procfs_inactive(struct vnode *a_vp) 457 */ 458 static int 459 procfs_inactive(struct vop_inactive_args *ap) 460 { 461 struct pfsnode *pfs = VTOPFS(ap->a_vp); 462 463 if (pfs->pfs_pid & PFS_DEAD) 464 vrecycle(ap->a_vp); 465 return (0); 466 } 467 468 /* 469 * _reclaim is called when getnewvnode() 470 * wants to make use of an entry on the vnode 471 * free list. at this time the filesystem needs 472 * to free any private data and remove the node 473 * from any private lists. 474 * 475 * procfs_reclaim(struct vnode *a_vp) 476 */ 477 static int 478 procfs_reclaim(struct vop_reclaim_args *ap) 479 { 480 return (procfs_freevp(ap->a_vp)); 481 } 482 483 /* 484 * _print is used for debugging. 485 * just print a readable description 486 * of (vp). 487 * 488 * procfs_print(struct vnode *a_vp) 489 */ 490 static int 491 procfs_print(struct vop_print_args *ap) 492 { 493 struct pfsnode *pfs = VTOPFS(ap->a_vp); 494 495 kprintf("tag VT_PROCFS, type %d, pid %ld, mode %x, flags %lx\n", 496 pfs->pfs_type, (long)pfs->pfs_pid, pfs->pfs_mode, pfs->pfs_flags); 497 return (0); 498 } 499 500 /* 501 * generic entry point for unsupported operations 502 */ 503 static int 504 procfs_badop(struct vop_generic_args *ap) 505 { 506 return (EIO); 507 } 508 509 /* 510 * Invent attributes for pfsnode (vp) and store 511 * them in (vap). 512 * Directories lengths are returned as zero since 513 * any real length would require the genuine size 514 * to be computed, and nothing cares anyway. 515 * 516 * this is relatively minimal for procfs. 517 * 518 * procfs_getattr(struct vnode *a_vp, struct vattr *a_vap) 519 */ 520 static int 521 procfs_getattr(struct vop_getattr_args *ap) 522 { 523 struct pfsnode *pfs = VTOPFS(ap->a_vp); 524 struct vattr *vap = ap->a_vap; 525 struct proc *procp; 526 int error; 527 528 /* 529 * First make sure that the process and its credentials 530 * still exist. 531 */ 532 switch (pfs->pfs_type) { 533 case Proot: 534 case Pcurproc: 535 procp = NULL; 536 break; 537 default: 538 procp = pfs_pfind(pfs->pfs_pid); 539 if (procp == NULL || procp->p_ucred == NULL) { 540 error = ENOENT; 541 goto done; 542 } 543 break; 544 } 545 546 error = 0; 547 548 /* start by zeroing out the attributes */ 549 VATTR_NULL(vap); 550 551 /* next do all the common fields */ 552 vap->va_type = ap->a_vp->v_type; 553 vap->va_mode = pfs->pfs_mode; 554 vap->va_fileid = pfs->pfs_fileno; 555 vap->va_flags = 0; 556 vap->va_blocksize = PAGE_SIZE; 557 vap->va_bytes = vap->va_size = 0; 558 vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0]; 559 560 /* 561 * Make all times be current TOD. 562 * It would be possible to get the process start 563 * time from the p_stat structure, but there's 564 * no "file creation" time stamp anyway, and the 565 * p_stat structure is not addressible if u. gets 566 * swapped out for that process. 567 */ 568 vfs_timestamp(&vap->va_ctime); 569 vap->va_atime = vap->va_mtime = vap->va_ctime; 570 571 /* 572 * If the process has exercised some setuid or setgid 573 * privilege, then rip away read/write permission so 574 * that only root can gain access. 575 */ 576 switch (pfs->pfs_type) { 577 case Pctl: 578 case Pregs: 579 case Pfpregs: 580 case Pdbregs: 581 case Pmem: 582 if (procp->p_flags & P_SUGID) { 583 vap->va_mode &= ~((VREAD|VWRITE)| 584 ((VREAD|VWRITE)>>3)| 585 ((VREAD|VWRITE)>>6)); 586 } 587 break; 588 default: 589 break; 590 } 591 592 /* 593 * now do the object specific fields 594 * 595 * The size could be set from struct reg, but it's hardly 596 * worth the trouble, and it puts some (potentially) machine 597 * dependent data into this machine-independent code. If it 598 * becomes important then this function should break out into 599 * a per-file stat function in the corresponding .c file. 600 */ 601 602 vap->va_nlink = 1; 603 if (procp) { 604 if (procp->p_ucred) { 605 vap->va_uid = procp->p_ucred->cr_uid; 606 vap->va_gid = procp->p_ucred->cr_gid; 607 } else { 608 vap->va_uid = -1; 609 vap->va_gid = -1; 610 } 611 } 612 613 switch (pfs->pfs_type) { 614 case Proot: 615 /* 616 * Set nlink to 1 to tell fts(3) we don't actually know. 617 */ 618 vap->va_nlink = 1; 619 vap->va_uid = 0; 620 vap->va_gid = 0; 621 vap->va_size = vap->va_bytes = DEV_BSIZE; 622 break; 623 624 case Pcurproc: { 625 char buf[16]; /* should be enough */ 626 627 vap->va_uid = 0; 628 vap->va_gid = 0; 629 vap->va_size = ksnprintf(buf, sizeof(buf), 630 "%ld", (long)curproc->p_pid); 631 vap->va_bytes = vap->va_size; 632 break; 633 } 634 635 case Pproc: 636 vap->va_nlink = nproc_targets; 637 vap->va_size = vap->va_bytes = DEV_BSIZE; 638 break; 639 640 case Pfile: { 641 char *fullpath, *freepath; 642 643 if (procp->p_textnch.ncp) { 644 struct nchandle nch; 645 646 cache_copy(&procp->p_textnch, &nch); 647 error = cache_fullpath(procp, &nch, NULL, 648 &fullpath, &freepath, 0); 649 cache_drop(&nch); 650 } else { 651 error = EINVAL; 652 } 653 654 if (error == 0) { 655 vap->va_size = strlen(fullpath); 656 kfree(freepath, M_TEMP); 657 } else { 658 vap->va_size = sizeof("unknown") - 1; 659 error = 0; 660 } 661 vap->va_bytes = vap->va_size; 662 break; 663 } 664 665 case Pmem: 666 /* 667 * If we denied owner access earlier, then we have to 668 * change the owner to root - otherwise 'ps' and friends 669 * will break even though they are setgid kmem. *SIGH* 670 */ 671 if (procp->p_flags & P_SUGID) 672 vap->va_uid = 0; 673 else if (procp->p_ucred) 674 vap->va_uid = procp->p_ucred->cr_uid; 675 else 676 vap->va_uid = -1; 677 break; 678 679 case Pregs: 680 vap->va_bytes = vap->va_size = sizeof(struct reg); 681 break; 682 683 case Pfpregs: 684 vap->va_bytes = vap->va_size = sizeof(struct fpreg); 685 break; 686 687 case Pdbregs: 688 vap->va_bytes = vap->va_size = sizeof(struct dbreg); 689 break; 690 691 case Ptype: 692 case Pmap: 693 case Pctl: 694 case Pstatus: 695 case Pnote: 696 case Pnotepg: 697 case Pcmdline: 698 case Prlimit: 699 break; 700 701 default: 702 panic("procfs_getattr"); 703 } 704 done: 705 pfs_pdone(procp); 706 return (error); 707 } 708 709 /* 710 * procfs_setattr(struct vnode *a_vp, struct vattr *a_vap, 711 * struct ucred *a_cred) 712 */ 713 static int 714 procfs_setattr(struct vop_setattr_args *ap) 715 { 716 if (ap->a_vap->va_flags != VNOVAL) 717 return (EOPNOTSUPP); 718 719 /* 720 * just fake out attribute setting 721 * it's not good to generate an error 722 * return, otherwise things like creat() 723 * will fail when they try to set the 724 * file length to 0. worse, this means 725 * that echo $note > /proc/$pid/note will fail. 726 */ 727 728 return (0); 729 } 730 731 /* 732 * implement access checking. 733 * 734 * procfs_access(struct vnode *a_vp, int a_mode, struct ucred *a_cred) 735 */ 736 static int 737 procfs_access(struct vop_access_args *ap) 738 { 739 struct vattr vattr; 740 int error; 741 742 error = VOP_GETATTR(ap->a_vp, &vattr); 743 if (!error) 744 error = vop_helper_access(ap, vattr.va_uid, vattr.va_gid, 745 vattr.va_mode, 0); 746 return (error); 747 } 748 749 /* 750 * lookup. this is incredibly complicated in the general case, however 751 * for most pseudo-filesystems very little needs to be done. 752 * 753 * procfs_lookup(struct vnode *a_dvp, struct vnode **a_vpp, 754 * struct componentname *a_cnp) 755 */ 756 static int 757 procfs_lookup(struct vop_old_lookup_args *ap) 758 { 759 struct componentname *cnp = ap->a_cnp; 760 struct vnode **vpp = ap->a_vpp; 761 struct vnode *dvp = ap->a_dvp; 762 char *pname = cnp->cn_nameptr; 763 /* struct proc *curp = cnp->cn_proc; */ 764 struct proc_target *pt; 765 pid_t pid; 766 struct pfsnode *pfs; 767 struct proc *p; 768 struct lwp *lp; 769 int i; 770 int error; 771 772 *vpp = NULL; 773 774 if (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME) 775 return (EROFS); 776 777 p = NULL; 778 error = 0; 779 if (cnp->cn_namelen == 1 && *pname == '.') { 780 *vpp = dvp; 781 vref(*vpp); 782 goto out; 783 } 784 785 pfs = VTOPFS(dvp); 786 switch (pfs->pfs_type) { 787 case Proot: 788 if (cnp->cn_flags & CNP_ISDOTDOT) 789 return (EIO); 790 791 if (CNEQ(cnp, "curproc", 7) || CNEQ(cnp, "self", 4)) { 792 error = procfs_allocvp(dvp->v_mount, vpp, 0, Pcurproc); 793 goto out; 794 } 795 796 pid = atopid(pname, cnp->cn_namelen); 797 if (pid == NO_PID) 798 break; 799 800 p = pfs_pfind(pid); 801 if (p == NULL) 802 break; 803 804 if (!PRISON_CHECK(ap->a_cnp->cn_cred, p->p_ucred)) 805 break; 806 807 if (ps_showallprocs == 0 && ap->a_cnp->cn_cred->cr_uid != 0 && 808 ap->a_cnp->cn_cred->cr_uid != p->p_ucred->cr_uid) 809 break; 810 811 error = procfs_allocvp(dvp->v_mount, vpp, pid, Pproc); 812 goto out; 813 814 case Pproc: 815 if (cnp->cn_flags & CNP_ISDOTDOT) { 816 error = procfs_root(dvp->v_mount, vpp); 817 goto out; 818 } 819 820 p = pfs_pfind(pfs->pfs_pid); 821 if (p == NULL) 822 break; 823 /* XXX lwp */ 824 lp = FIRST_LWP_IN_PROC(p); 825 if (lp == NULL) 826 break; 827 828 if (!PRISON_CHECK(ap->a_cnp->cn_cred, p->p_ucred)) 829 break; 830 831 if (ps_showallprocs == 0 && ap->a_cnp->cn_cred->cr_uid != 0 && 832 ap->a_cnp->cn_cred->cr_uid != p->p_ucred->cr_uid) 833 break; 834 835 for (pt = proc_targets, i = 0; i < nproc_targets; pt++, i++) { 836 if (cnp->cn_namelen == pt->pt_namlen && 837 bcmp(pt->pt_name, pname, cnp->cn_namelen) == 0 && 838 (pt->pt_valid == NULL || (*pt->pt_valid)(lp))) 839 goto found; 840 } 841 break; 842 found: 843 error = procfs_allocvp(dvp->v_mount, vpp, pfs->pfs_pid, 844 pt->pt_pfstype); 845 goto out; 846 847 default: 848 error = ENOTDIR; 849 goto out; 850 } 851 if (cnp->cn_nameiop == NAMEI_LOOKUP) 852 error = ENOENT; 853 else 854 error = EROFS; 855 /* 856 * If no error occured *vpp will hold a referenced locked vnode. 857 * dvp was passed to us locked and *vpp must be returned locked. 858 * If *vpp != dvp then we should unlock dvp if (1) this is not the 859 * last component or (2) CNP_LOCKPARENT is not set. 860 */ 861 out: 862 if (error == 0 && *vpp != dvp) { 863 if ((cnp->cn_flags & CNP_LOCKPARENT) == 0) { 864 cnp->cn_flags |= CNP_PDIRUNLOCK; 865 vn_unlock(dvp); 866 } 867 } 868 pfs_pdone(p); 869 return (error); 870 } 871 872 /* 873 * Does this process have a text file? 874 */ 875 int 876 procfs_validfile(struct lwp *lp) 877 { 878 return (procfs_findtextvp(lp->lwp_proc) != NULLVP); 879 } 880 881 /* 882 * readdir() returns directory entries from pfsnode (vp). 883 * 884 * We generate just one directory entry at a time, as it would probably 885 * not pay off to buffer several entries locally to save uiomove calls. 886 * 887 * procfs_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred, 888 * int *a_eofflag, int *a_ncookies, off_t **a_cookies) 889 */ 890 static int 891 procfs_readdir(struct vop_readdir_args *ap) 892 { 893 struct pfsnode *pfs; 894 int error; 895 896 if (ap->a_uio->uio_offset < 0 || ap->a_uio->uio_offset > INT_MAX) 897 return (EINVAL); 898 error = vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY | LK_FAILRECLAIM); 899 if (error) 900 return (error); 901 pfs = VTOPFS(ap->a_vp); 902 903 switch (pfs->pfs_type) { 904 case Pproc: 905 /* 906 * this is for the process-specific sub-directories. 907 * all that is needed to is copy out all the entries 908 * from the procent[] table (top of this file). 909 */ 910 error = procfs_readdir_proc(ap); 911 break; 912 case Proot: 913 /* 914 * this is for the root of the procfs filesystem 915 * what is needed is a special entry for "curproc" 916 * followed by an entry for each process on allproc 917 */ 918 error = procfs_readdir_root(ap); 919 break; 920 default: 921 error = ENOTDIR; 922 break; 923 } 924 925 vn_unlock(ap->a_vp); 926 return (error); 927 } 928 929 static int 930 procfs_readdir_proc(struct vop_readdir_args *ap) 931 { 932 struct pfsnode *pfs; 933 int error, i, retval; 934 struct proc *p; 935 struct lwp *lp; 936 struct proc_target *pt; 937 struct uio *uio = ap->a_uio; 938 939 pfs = VTOPFS(ap->a_vp); 940 p = pfs_pfind(pfs->pfs_pid); 941 if (p == NULL) 942 return(0); 943 if (!PRISON_CHECK(ap->a_cred, p->p_ucred)) { 944 error = 0; 945 goto done; 946 } 947 /* XXX lwp, not MPSAFE */ 948 lp = FIRST_LWP_IN_PROC(p); 949 if (lp == NULL) { 950 error = EINVAL; 951 goto done; 952 } 953 954 error = 0; 955 i = (int)uio->uio_offset; 956 if (i < 0) { 957 error = EINVAL; 958 goto done; 959 } 960 961 for (pt = &proc_targets[i]; 962 !error && uio->uio_resid > 0 && i < nproc_targets; pt++, i++) { 963 if (pt->pt_valid && (*pt->pt_valid)(lp) == 0) 964 continue; 965 966 retval = vop_write_dirent(&error, uio, 967 PROCFS_FILENO(pfs->pfs_pid, pt->pt_pfstype), pt->pt_type, 968 pt->pt_namlen, pt->pt_name); 969 if (retval) 970 break; 971 } 972 973 uio->uio_offset = (off_t)i; 974 error = 0; 975 done: 976 pfs_pdone(p); 977 return error; 978 } 979 980 struct procfs_readdir_root_info { 981 int error; 982 int i; 983 int pcnt; 984 struct uio *uio; 985 struct ucred *cred; 986 }; 987 988 static int procfs_readdir_root_callback(struct proc *p, void *data); 989 990 static int 991 procfs_readdir_root(struct vop_readdir_args *ap) 992 { 993 struct procfs_readdir_root_info info; 994 struct uio *uio = ap->a_uio; 995 int res; 996 997 res = 0; 998 info.error = 0; 999 info.i = (int)uio->uio_offset; 1000 1001 if (info.i < 0) 1002 return (EINVAL); 1003 1004 info.pcnt = 0; 1005 info.uio = uio; 1006 info.cred = ap->a_cred; 1007 while (info.pcnt < 4) { 1008 res = procfs_readdir_root_callback(NULL, &info); 1009 if (res < 0) 1010 break; 1011 } 1012 if (res >= 0) 1013 allproc_scan(procfs_readdir_root_callback, &info, 0); 1014 uio->uio_offset = (off_t)info.i; 1015 1016 return (info.error); 1017 } 1018 1019 static int 1020 procfs_readdir_root_callback(struct proc *p, void *data) 1021 { 1022 struct procfs_readdir_root_info *info = data; 1023 struct uio *uio; 1024 int retval; 1025 ino_t d_ino; 1026 const char *d_name; 1027 char d_name_pid[20]; 1028 size_t d_namlen; 1029 uint8_t d_type; 1030 1031 uio = info->uio; 1032 1033 if (uio->uio_resid <= 0 || info->error) 1034 return(-1); 1035 1036 switch (info->pcnt) { 1037 case 0: /* `.' */ 1038 d_ino = PROCFS_FILENO(0, Proot); 1039 d_name = "."; 1040 d_namlen = 1; 1041 d_type = DT_DIR; 1042 break; 1043 case 1: /* `..' */ 1044 d_ino = PROCFS_FILENO(0, Proot); 1045 d_name = ".."; 1046 d_namlen = 2; 1047 d_type = DT_DIR; 1048 break; 1049 1050 case 2: 1051 d_ino = PROCFS_FILENO(0, Pcurproc); 1052 d_namlen = 7; 1053 d_name = "curproc"; 1054 d_type = DT_LNK; 1055 break; 1056 1057 case 3: 1058 d_ino = PROCFS_FILENO(0, Pcurproc); 1059 d_namlen = 4; 1060 d_name = "self"; 1061 d_type = DT_LNK; 1062 break; 1063 1064 default: 1065 if (!PRISON_CHECK(info->cred, p->p_ucred)) 1066 return(0); 1067 if (ps_showallprocs == 0 && 1068 info->cred->cr_uid != 0 && 1069 info->cred->cr_uid != p->p_ucred->cr_uid) { 1070 return(0); 1071 } 1072 1073 /* 1074 * Skip entries we have already returned (optimization) 1075 */ 1076 if (info->pcnt < info->i) { 1077 ++info->pcnt; 1078 return(0); 1079 } 1080 1081 d_ino = PROCFS_FILENO(p->p_pid, Pproc); 1082 d_namlen = ksnprintf(d_name_pid, sizeof(d_name_pid), 1083 "%ld", (long)p->p_pid); 1084 d_name = d_name_pid; 1085 d_type = DT_DIR; 1086 break; 1087 } 1088 1089 /* 1090 * Skip entries we have already returned (optimization) 1091 */ 1092 if (info->pcnt < info->i) { 1093 ++info->pcnt; 1094 return(0); 1095 } 1096 1097 retval = vop_write_dirent(&info->error, uio, 1098 d_ino, d_type, d_namlen, d_name); 1099 if (retval) 1100 return(-1); 1101 ++info->pcnt; 1102 ++info->i; 1103 return(0); 1104 } 1105 1106 /* 1107 * readlink reads the link of `curproc' or `file' 1108 */ 1109 static int 1110 procfs_readlink(struct vop_readlink_args *ap) 1111 { 1112 char buf[16]; /* should be enough */ 1113 struct proc *procp; 1114 struct vnode *vp = ap->a_vp; 1115 struct pfsnode *pfs = VTOPFS(vp); 1116 char *fullpath, *freepath; 1117 int error, len; 1118 1119 switch (pfs->pfs_type) { 1120 case Pcurproc: 1121 if (pfs->pfs_fileno != PROCFS_FILENO(0, Pcurproc)) 1122 return (EINVAL); 1123 1124 len = ksnprintf(buf, sizeof(buf), "%ld", (long)curproc->p_pid); 1125 1126 return (uiomove(buf, len, ap->a_uio)); 1127 case Pfile: 1128 /* 1129 * procfs's directory topology is somewhat asynchronous from 1130 * reality so it is possible for pid requests to race exiting 1131 * processes. In this situation, bit 31 is set in 1132 * pfs->pfs_pid which guarantees that pfs_pfind() will return 1133 * NULL. 1134 * 1135 * It is also possible to catch a process in the middle of 1136 * an exit sequence so various fields might wind up being 1137 * NULL that are not normally NULL. 1138 */ 1139 procp = pfs_pfind(pfs->pfs_pid); 1140 if (procp == NULL || procp->p_ucred == NULL) { 1141 pfs_pdone(procp); 1142 return (uiomove("unknown", sizeof("unknown") - 1, 1143 ap->a_uio)); 1144 } 1145 if (procp->p_textnch.ncp) { 1146 struct nchandle nch; 1147 1148 cache_copy(&procp->p_textnch, &nch); 1149 error = cache_fullpath(procp, &nch, NULL, 1150 &fullpath, &freepath, 0); 1151 cache_drop(&nch); 1152 } else { 1153 error = EINVAL; 1154 } 1155 1156 if (error != 0) { 1157 pfs_pdone(procp); 1158 return (uiomove("unknown", sizeof("unknown") - 1, 1159 ap->a_uio)); 1160 } 1161 error = uiomove(fullpath, strlen(fullpath), ap->a_uio); 1162 kfree(freepath, M_TEMP); 1163 pfs_pdone(procp); 1164 return (error); 1165 default: 1166 return (EINVAL); 1167 } 1168 } 1169 1170 /* 1171 * convert decimal ascii to pid_t 1172 */ 1173 static pid_t 1174 atopid(const char *b, u_int len) 1175 { 1176 pid_t p = 0; 1177 1178 while (len--) { 1179 char c = *b++; 1180 if (c < '0' || c > '9') 1181 return (NO_PID); 1182 p = 10 * p + (c - '0'); 1183 if (p > PID_MAX) 1184 return (NO_PID); 1185 } 1186 1187 return (p); 1188 } 1189 1190