1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 3 * 4 * Copyright (c) 2009, 2016 Robert N. M. Watson 5 * All rights reserved. 6 * 7 * This software was developed at the University of Cambridge Computer 8 * Laboratory with support from a grant from Google, Inc. 9 * 10 * Portions of this software were developed by BAE Systems, the University of 11 * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL 12 * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent 13 * Computing (TC) research program. 14 * 15 * Redistribution and use in source and binary forms, with or without 16 * modification, are permitted provided that the following conditions 17 * are met: 18 * 1. Redistributions of source code must retain the above copyright 19 * notice, this list of conditions and the following disclaimer. 20 * 2. Redistributions in binary form must reproduce the above copyright 21 * notice, this list of conditions and the following disclaimer in the 22 * documentation and/or other materials provided with the distribution. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 */ 36 37 /*- 38 * FreeBSD process descriptor facility. 39 * 40 * Some processes are represented by a file descriptor, which will be used in 41 * preference to signaling and pids for the purposes of process management, 42 * and is, in effect, a form of capability. When a process descriptor is 43 * used with a process, it ceases to be visible to certain traditional UNIX 44 * process facilities, such as waitpid(2). 45 * 46 * Some semantics: 47 * 48 * - At most one process descriptor will exist for any process, although 49 * references to that descriptor may be held from many processes (or even 50 * be in flight between processes over a local domain socket). 51 * - Last close on the process descriptor will terminate the process using 52 * SIGKILL and reparent it to init so that there's a process to reap it 53 * when it's done exiting. 54 * - If the process exits before the descriptor is closed, it will not 55 * generate SIGCHLD on termination, or be picked up by waitpid(). 56 * - The pdkill(2) system call may be used to deliver a signal to the process 57 * using its process descriptor. 58 * 59 * Open questions: 60 * 61 * - Will we want to add a pidtoprocdesc(2) system call to allow process 62 * descriptors to be created for processes without pdfork(2)? 63 */ 64 65 #include <sys/param.h> 66 #include <sys/capsicum.h> 67 #include <sys/fcntl.h> 68 #include <sys/file.h> 69 #include <sys/filedesc.h> 70 #include <sys/kernel.h> 71 #include <sys/lock.h> 72 #include <sys/mutex.h> 73 #include <sys/poll.h> 74 #include <sys/proc.h> 75 #include <sys/procdesc.h> 76 #include <sys/resourcevar.h> 77 #include <sys/stat.h> 78 #include <sys/syscallsubr.h> 79 #include <sys/sysproto.h> 80 #include <sys/sysctl.h> 81 #include <sys/systm.h> 82 #include <sys/ucred.h> 83 #include <sys/user.h> 84 85 #include <security/audit/audit.h> 86 87 #include <vm/uma.h> 88 89 FEATURE(process_descriptors, "Process Descriptors"); 90 91 MALLOC_DEFINE(M_PROCDESC, "procdesc", "process descriptors"); 92 93 static fo_poll_t procdesc_poll; 94 static fo_kqfilter_t procdesc_kqfilter; 95 static fo_stat_t procdesc_stat; 96 static fo_close_t procdesc_close; 97 static fo_fill_kinfo_t procdesc_fill_kinfo; 98 static fo_cmp_t procdesc_cmp; 99 100 static const struct fileops procdesc_ops = { 101 .fo_read = invfo_rdwr, 102 .fo_write = invfo_rdwr, 103 .fo_truncate = invfo_truncate, 104 .fo_ioctl = invfo_ioctl, 105 .fo_poll = procdesc_poll, 106 .fo_kqfilter = procdesc_kqfilter, 107 .fo_stat = procdesc_stat, 108 .fo_close = procdesc_close, 109 .fo_chmod = invfo_chmod, 110 .fo_chown = invfo_chown, 111 .fo_sendfile = invfo_sendfile, 112 .fo_fill_kinfo = procdesc_fill_kinfo, 113 .fo_cmp = procdesc_cmp, 114 .fo_flags = DFLAG_PASSABLE, 115 }; 116 117 /* 118 * Return a locked process given a process descriptor, or ESRCH if it has 119 * died. 120 */ 121 int 122 procdesc_find(struct thread *td, int fd, const cap_rights_t *rightsp, 123 struct proc **p) 124 { 125 struct procdesc *pd; 126 struct file *fp; 127 int error; 128 129 error = fget(td, fd, rightsp, &fp); 130 if (error) 131 return (error); 132 if (fp->f_type != DTYPE_PROCDESC) { 133 error = EINVAL; 134 goto out; 135 } 136 pd = fp->f_data; 137 sx_slock(&proctree_lock); 138 if (pd->pd_proc != NULL) { 139 *p = pd->pd_proc; 140 PROC_LOCK(*p); 141 } else 142 error = ESRCH; 143 sx_sunlock(&proctree_lock); 144 out: 145 fdrop(fp, td); 146 return (error); 147 } 148 149 /* 150 * Function to be used by procstat(1) sysctls when returning procdesc 151 * information. 152 */ 153 pid_t 154 procdesc_pid(struct file *fp_procdesc) 155 { 156 struct procdesc *pd; 157 158 KASSERT(fp_procdesc->f_type == DTYPE_PROCDESC, 159 ("procdesc_pid: !procdesc")); 160 161 pd = fp_procdesc->f_data; 162 return (pd->pd_pid); 163 } 164 165 /* 166 * Retrieve the PID associated with a process descriptor. 167 */ 168 int 169 kern_pdgetpid(struct thread *td, int fd, const cap_rights_t *rightsp, 170 pid_t *pidp) 171 { 172 struct file *fp; 173 int error; 174 175 error = fget(td, fd, rightsp, &fp); 176 if (error) 177 return (error); 178 if (fp->f_type != DTYPE_PROCDESC) { 179 error = EBADF; 180 goto out; 181 } 182 *pidp = procdesc_pid(fp); 183 out: 184 fdrop(fp, td); 185 return (error); 186 } 187 188 /* 189 * System call to return the pid of a process given its process descriptor. 190 */ 191 int 192 sys_pdgetpid(struct thread *td, struct pdgetpid_args *uap) 193 { 194 pid_t pid; 195 int error; 196 197 AUDIT_ARG_FD(uap->fd); 198 error = kern_pdgetpid(td, uap->fd, &cap_pdgetpid_rights, &pid); 199 if (error == 0) 200 error = copyout(&pid, uap->pidp, sizeof(pid)); 201 return (error); 202 } 203 204 /* 205 * When a new process is forked by pdfork(), a file descriptor is allocated 206 * by the fork code first, then the process is forked, and then we get a 207 * chance to set up the process descriptor. Failure is not permitted at this 208 * point, so procdesc_new() must succeed. 209 */ 210 void 211 procdesc_new(struct proc *p, int flags) 212 { 213 struct procdesc *pd; 214 215 pd = malloc(sizeof(*pd), M_PROCDESC, M_WAITOK | M_ZERO); 216 pd->pd_proc = p; 217 pd->pd_pid = p->p_pid; 218 p->p_procdesc = pd; 219 pd->pd_flags = 0; 220 if (flags & PD_DAEMON) 221 pd->pd_flags |= PDF_DAEMON; 222 PROCDESC_LOCK_INIT(pd); 223 knlist_init_mtx(&pd->pd_selinfo.si_note, &pd->pd_lock); 224 225 /* 226 * Process descriptors start out with two references: one from their 227 * struct file, and the other from their struct proc. 228 */ 229 refcount_init(&pd->pd_refcount, 2); 230 } 231 232 /* 233 * Create a new process decriptor for the process that refers to it. 234 */ 235 int 236 procdesc_falloc(struct thread *td, struct file **resultfp, int *resultfd, 237 int flags, struct filecaps *fcaps) 238 { 239 int fflags; 240 241 fflags = 0; 242 if (flags & PD_CLOEXEC) 243 fflags = O_CLOEXEC; 244 245 return (falloc_caps(td, resultfp, resultfd, fflags, fcaps)); 246 } 247 248 /* 249 * Initialize a file with a process descriptor. 250 */ 251 void 252 procdesc_finit(struct procdesc *pdp, struct file *fp) 253 { 254 255 finit(fp, FREAD | FWRITE, DTYPE_PROCDESC, pdp, &procdesc_ops); 256 } 257 258 static void 259 procdesc_free(struct procdesc *pd) 260 { 261 262 /* 263 * When the last reference is released, we assert that the descriptor 264 * has been closed, but not that the process has exited, as we will 265 * detach the descriptor before the process dies if the descript is 266 * closed, as we can't wait synchronously. 267 */ 268 if (refcount_release(&pd->pd_refcount)) { 269 KASSERT(pd->pd_proc == NULL, 270 ("procdesc_free: pd_proc != NULL")); 271 KASSERT((pd->pd_flags & PDF_CLOSED), 272 ("procdesc_free: !PDF_CLOSED")); 273 274 if (pd->pd_pid != -1) 275 proc_id_clear(PROC_ID_PID, pd->pd_pid); 276 277 seldrain(&pd->pd_selinfo); 278 knlist_destroy(&pd->pd_selinfo.si_note); 279 PROCDESC_LOCK_DESTROY(pd); 280 free(pd, M_PROCDESC); 281 } 282 } 283 284 /* 285 * procdesc_exit() - notify a process descriptor that its process is exiting. 286 * We use the proctree_lock to ensure that process exit either happens 287 * strictly before or strictly after a concurrent call to procdesc_close(). 288 */ 289 int 290 procdesc_exit(struct proc *p) 291 { 292 struct procdesc *pd; 293 294 sx_assert(&proctree_lock, SA_XLOCKED); 295 PROC_LOCK_ASSERT(p, MA_OWNED); 296 KASSERT(p->p_procdesc != NULL, ("procdesc_exit: p_procdesc NULL")); 297 298 pd = p->p_procdesc; 299 300 PROCDESC_LOCK(pd); 301 KASSERT((pd->pd_flags & PDF_CLOSED) == 0 || p->p_pptr == p->p_reaper, 302 ("procdesc_exit: closed && parent not reaper")); 303 304 pd->pd_flags |= PDF_EXITED; 305 pd->pd_xstat = KW_EXITCODE(p->p_xexit, p->p_xsig); 306 307 /* 308 * If the process descriptor has been closed, then we have nothing 309 * to do; return 1 so that init will get SIGCHLD and do the reaping. 310 * Clean up the procdesc now rather than letting it happen during 311 * that reap. 312 */ 313 if (pd->pd_flags & PDF_CLOSED) { 314 PROCDESC_UNLOCK(pd); 315 pd->pd_proc = NULL; 316 p->p_procdesc = NULL; 317 procdesc_free(pd); 318 return (1); 319 } 320 selwakeup(&pd->pd_selinfo); 321 KNOTE_LOCKED(&pd->pd_selinfo.si_note, NOTE_EXIT); 322 PROCDESC_UNLOCK(pd); 323 324 /* Wakeup all waiters for this procdesc' process exit. */ 325 wakeup(&p->p_procdesc); 326 return (0); 327 } 328 329 /* 330 * When a process descriptor is reaped, perhaps as a result of close(), release 331 * the process's reference on the process descriptor. 332 */ 333 void 334 procdesc_reap(struct proc *p) 335 { 336 struct procdesc *pd; 337 338 sx_assert(&proctree_lock, SA_XLOCKED); 339 KASSERT(p->p_procdesc != NULL, ("procdesc_reap: p_procdesc == NULL")); 340 341 pd = p->p_procdesc; 342 pd->pd_proc = NULL; 343 p->p_procdesc = NULL; 344 procdesc_free(pd); 345 } 346 347 /* 348 * procdesc_close() - last close on a process descriptor. If the process is 349 * still running, terminate with SIGKILL (unless PDF_DAEMON is set) and let 350 * its reaper clean up the mess; if not, we have to clean up the zombie 351 * ourselves. 352 */ 353 static int 354 procdesc_close(struct file *fp, struct thread *td) 355 { 356 struct procdesc *pd; 357 struct proc *p; 358 359 KASSERT(fp->f_type == DTYPE_PROCDESC, ("procdesc_close: !procdesc")); 360 361 pd = fp->f_data; 362 fp->f_ops = &badfileops; 363 fp->f_data = NULL; 364 365 sx_xlock(&proctree_lock); 366 PROCDESC_LOCK(pd); 367 pd->pd_flags |= PDF_CLOSED; 368 PROCDESC_UNLOCK(pd); 369 p = pd->pd_proc; 370 if (p == NULL) { 371 /* 372 * This is the case where process' exit status was already 373 * collected and procdesc_reap() was already called. 374 */ 375 sx_xunlock(&proctree_lock); 376 } else { 377 PROC_LOCK(p); 378 AUDIT_ARG_PROCESS(p); 379 if (p->p_state == PRS_ZOMBIE) { 380 /* 381 * If the process is already dead and just awaiting 382 * reaping, do that now. This will release the 383 * process's reference to the process descriptor when it 384 * calls back into procdesc_reap(). 385 */ 386 proc_reap(curthread, p, NULL, 0); 387 } else { 388 /* 389 * If the process is not yet dead, we need to kill it, 390 * but we can't wait around synchronously for it to go 391 * away, as that path leads to madness (and deadlocks). 392 * First, detach the process from its descriptor so that 393 * its exit status will be reported normally. 394 */ 395 pd->pd_proc = NULL; 396 p->p_procdesc = NULL; 397 pd->pd_pid = -1; 398 procdesc_free(pd); 399 400 /* 401 * Next, reparent it to its reaper (usually init(8)) so 402 * that there's someone to pick up the pieces; finally, 403 * terminate with prejudice. 404 */ 405 p->p_sigparent = SIGCHLD; 406 if ((p->p_flag & P_TRACED) == 0) { 407 proc_reparent(p, p->p_reaper, true); 408 } else { 409 proc_clear_orphan(p); 410 p->p_oppid = p->p_reaper->p_pid; 411 proc_add_orphan(p, p->p_reaper); 412 } 413 if ((pd->pd_flags & PDF_DAEMON) == 0) 414 kern_psignal(p, SIGKILL); 415 PROC_UNLOCK(p); 416 sx_xunlock(&proctree_lock); 417 } 418 } 419 420 /* 421 * Release the file descriptor's reference on the process descriptor. 422 */ 423 procdesc_free(pd); 424 return (0); 425 } 426 427 static int 428 procdesc_poll(struct file *fp, int events, struct ucred *active_cred, 429 struct thread *td) 430 { 431 struct procdesc *pd; 432 int revents; 433 434 revents = 0; 435 pd = fp->f_data; 436 PROCDESC_LOCK(pd); 437 if (pd->pd_flags & PDF_EXITED) 438 revents |= POLLHUP; 439 else 440 selrecord(td, &pd->pd_selinfo); 441 PROCDESC_UNLOCK(pd); 442 return (revents); 443 } 444 445 static void 446 procdesc_kqops_detach(struct knote *kn) 447 { 448 struct procdesc *pd; 449 450 pd = kn->kn_fp->f_data; 451 knlist_remove(&pd->pd_selinfo.si_note, kn, 0); 452 } 453 454 static int 455 procdesc_kqops_event(struct knote *kn, long hint) 456 { 457 struct procdesc *pd; 458 u_int event; 459 460 pd = kn->kn_fp->f_data; 461 if (hint == 0) { 462 /* 463 * Initial test after registration. Generate a NOTE_EXIT in 464 * case the process already terminated before registration. 465 */ 466 event = pd->pd_flags & PDF_EXITED ? NOTE_EXIT : 0; 467 } else { 468 /* Mask off extra data. */ 469 event = (u_int)hint & NOTE_PCTRLMASK; 470 } 471 472 /* If the user is interested in this event, record it. */ 473 if (kn->kn_sfflags & event) 474 kn->kn_fflags |= event; 475 476 /* Process is gone, so flag the event as finished. */ 477 if (event == NOTE_EXIT) { 478 kn->kn_flags |= EV_EOF | EV_ONESHOT; 479 if (kn->kn_fflags & NOTE_EXIT) 480 kn->kn_data = pd->pd_xstat; 481 if (kn->kn_fflags == 0) 482 kn->kn_flags |= EV_DROP; 483 return (1); 484 } 485 486 return (kn->kn_fflags != 0); 487 } 488 489 static const struct filterops procdesc_kqops = { 490 .f_isfd = 1, 491 .f_detach = procdesc_kqops_detach, 492 .f_event = procdesc_kqops_event, 493 .f_copy = knote_triv_copy, 494 }; 495 496 static int 497 procdesc_kqfilter(struct file *fp, struct knote *kn) 498 { 499 struct procdesc *pd; 500 501 pd = fp->f_data; 502 switch (kn->kn_filter) { 503 case EVFILT_PROCDESC: 504 kn->kn_fop = &procdesc_kqops; 505 kn->kn_flags |= EV_CLEAR; 506 knlist_add(&pd->pd_selinfo.si_note, kn, 0); 507 return (0); 508 default: 509 return (EINVAL); 510 } 511 } 512 513 static int 514 procdesc_stat(struct file *fp, struct stat *sb, struct ucred *active_cred) 515 { 516 struct procdesc *pd; 517 struct timeval pstart, boottime; 518 519 /* 520 * XXXRW: Perhaps we should cache some more information from the 521 * process so that we can return it reliably here even after it has 522 * died. For example, caching its credential data. 523 */ 524 bzero(sb, sizeof(*sb)); 525 pd = fp->f_data; 526 sx_slock(&proctree_lock); 527 if (pd->pd_proc != NULL) { 528 PROC_LOCK(pd->pd_proc); 529 AUDIT_ARG_PROCESS(pd->pd_proc); 530 531 /* Set birth and [acm] times to process start time. */ 532 pstart = pd->pd_proc->p_stats->p_start; 533 getboottime(&boottime); 534 timevaladd(&pstart, &boottime); 535 TIMEVAL_TO_TIMESPEC(&pstart, &sb->st_birthtim); 536 sb->st_atim = sb->st_birthtim; 537 sb->st_ctim = sb->st_birthtim; 538 sb->st_mtim = sb->st_birthtim; 539 if (pd->pd_proc->p_state != PRS_ZOMBIE) 540 sb->st_mode = S_IFREG | S_IRWXU; 541 else 542 sb->st_mode = S_IFREG; 543 sb->st_uid = pd->pd_proc->p_ucred->cr_ruid; 544 sb->st_gid = pd->pd_proc->p_ucred->cr_rgid; 545 PROC_UNLOCK(pd->pd_proc); 546 } else 547 sb->st_mode = S_IFREG; 548 sx_sunlock(&proctree_lock); 549 return (0); 550 } 551 552 static int 553 procdesc_fill_kinfo(struct file *fp, struct kinfo_file *kif, 554 struct filedesc *fdp) 555 { 556 struct procdesc *pdp; 557 558 kif->kf_type = KF_TYPE_PROCDESC; 559 pdp = fp->f_data; 560 kif->kf_un.kf_proc.kf_pid = pdp->pd_pid; 561 return (0); 562 } 563 564 static int 565 procdesc_cmp(struct file *fp1, struct file *fp2, struct thread *td) 566 { 567 struct procdesc *pdp1, *pdp2; 568 569 if (fp2->f_type != DTYPE_PROCDESC) 570 return (3); 571 pdp1 = fp1->f_data; 572 pdp2 = fp2->f_data; 573 return (kcmp_cmp((uintptr_t)pdp1->pd_pid, (uintptr_t)pdp2->pd_pid)); 574 } 575