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/cdefs.h> 66 #include <sys/param.h> 67 #include <sys/capsicum.h> 68 #include <sys/fcntl.h> 69 #include <sys/file.h> 70 #include <sys/filedesc.h> 71 #include <sys/kernel.h> 72 #include <sys/lock.h> 73 #include <sys/mutex.h> 74 #include <sys/poll.h> 75 #include <sys/proc.h> 76 #include <sys/procdesc.h> 77 #include <sys/resourcevar.h> 78 #include <sys/stat.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 99 static struct fileops procdesc_ops = { 100 .fo_read = invfo_rdwr, 101 .fo_write = invfo_rdwr, 102 .fo_truncate = invfo_truncate, 103 .fo_ioctl = invfo_ioctl, 104 .fo_poll = procdesc_poll, 105 .fo_kqfilter = procdesc_kqfilter, 106 .fo_stat = procdesc_stat, 107 .fo_close = procdesc_close, 108 .fo_chmod = invfo_chmod, 109 .fo_chown = invfo_chown, 110 .fo_sendfile = invfo_sendfile, 111 .fo_fill_kinfo = procdesc_fill_kinfo, 112 .fo_flags = DFLAG_PASSABLE, 113 }; 114 115 /* 116 * Return a locked process given a process descriptor, or ESRCH if it has 117 * died. 118 */ 119 int 120 procdesc_find(struct thread *td, int fd, cap_rights_t *rightsp, 121 struct proc **p) 122 { 123 struct procdesc *pd; 124 struct file *fp; 125 int error; 126 127 error = fget(td, fd, rightsp, &fp); 128 if (error) 129 return (error); 130 if (fp->f_type != DTYPE_PROCDESC) { 131 error = EBADF; 132 goto out; 133 } 134 pd = fp->f_data; 135 sx_slock(&proctree_lock); 136 if (pd->pd_proc != NULL) { 137 *p = pd->pd_proc; 138 PROC_LOCK(*p); 139 } else 140 error = ESRCH; 141 sx_sunlock(&proctree_lock); 142 out: 143 fdrop(fp, td); 144 return (error); 145 } 146 147 /* 148 * Function to be used by procstat(1) sysctls when returning procdesc 149 * information. 150 */ 151 pid_t 152 procdesc_pid(struct file *fp_procdesc) 153 { 154 struct procdesc *pd; 155 156 KASSERT(fp_procdesc->f_type == DTYPE_PROCDESC, 157 ("procdesc_pid: !procdesc")); 158 159 pd = fp_procdesc->f_data; 160 return (pd->pd_pid); 161 } 162 163 /* 164 * Retrieve the PID associated with a process descriptor. 165 */ 166 int 167 kern_pdgetpid(struct thread *td, int fd, cap_rights_t *rightsp, pid_t *pidp) 168 { 169 struct file *fp; 170 int error; 171 172 error = fget(td, fd, rightsp, &fp); 173 if (error) 174 return (error); 175 if (fp->f_type != DTYPE_PROCDESC) { 176 error = EBADF; 177 goto out; 178 } 179 *pidp = procdesc_pid(fp); 180 out: 181 fdrop(fp, td); 182 return (error); 183 } 184 185 /* 186 * System call to return the pid of a process given its process descriptor. 187 */ 188 int 189 sys_pdgetpid(struct thread *td, struct pdgetpid_args *uap) 190 { 191 pid_t pid; 192 int error; 193 194 AUDIT_ARG_FD(uap->fd); 195 error = kern_pdgetpid(td, uap->fd, &cap_pdgetpid_rights, &pid); 196 if (error == 0) 197 error = copyout(&pid, uap->pidp, sizeof(pid)); 198 return (error); 199 } 200 201 /* 202 * When a new process is forked by pdfork(), a file descriptor is allocated 203 * by the fork code first, then the process is forked, and then we get a 204 * chance to set up the process descriptor. Failure is not permitted at this 205 * point, so procdesc_new() must succeed. 206 */ 207 void 208 procdesc_new(struct proc *p, int flags) 209 { 210 struct procdesc *pd; 211 212 pd = malloc(sizeof(*pd), M_PROCDESC, M_WAITOK | M_ZERO); 213 pd->pd_proc = p; 214 pd->pd_pid = p->p_pid; 215 p->p_procdesc = pd; 216 pd->pd_flags = 0; 217 if (flags & PD_DAEMON) 218 pd->pd_flags |= PDF_DAEMON; 219 PROCDESC_LOCK_INIT(pd); 220 knlist_init_mtx(&pd->pd_selinfo.si_note, &pd->pd_lock); 221 222 /* 223 * Process descriptors start out with two references: one from their 224 * struct file, and the other from their struct proc. 225 */ 226 refcount_init(&pd->pd_refcount, 2); 227 } 228 229 /* 230 * Create a new process decriptor for the process that refers to it. 231 */ 232 int 233 procdesc_falloc(struct thread *td, struct file **resultfp, int *resultfd, 234 int flags, struct filecaps *fcaps) 235 { 236 int fflags; 237 238 fflags = 0; 239 if (flags & PD_CLOEXEC) 240 fflags = O_CLOEXEC; 241 242 return (falloc_caps(td, resultfp, resultfd, fflags, fcaps)); 243 } 244 245 /* 246 * Initialize a file with a process descriptor. 247 */ 248 void 249 procdesc_finit(struct procdesc *pdp, struct file *fp) 250 { 251 252 finit(fp, FREAD | FWRITE, DTYPE_PROCDESC, pdp, &procdesc_ops); 253 } 254 255 static void 256 procdesc_free(struct procdesc *pd) 257 { 258 259 /* 260 * When the last reference is released, we assert that the descriptor 261 * has been closed, but not that the process has exited, as we will 262 * detach the descriptor before the process dies if the descript is 263 * closed, as we can't wait synchronously. 264 */ 265 if (refcount_release(&pd->pd_refcount)) { 266 KASSERT(pd->pd_proc == NULL, 267 ("procdesc_free: pd_proc != NULL")); 268 KASSERT((pd->pd_flags & PDF_CLOSED), 269 ("procdesc_free: !PDF_CLOSED")); 270 271 knlist_destroy(&pd->pd_selinfo.si_note); 272 PROCDESC_LOCK_DESTROY(pd); 273 free(pd, M_PROCDESC); 274 } 275 } 276 277 /* 278 * procdesc_exit() - notify a process descriptor that its process is exiting. 279 * We use the proctree_lock to ensure that process exit either happens 280 * strictly before or strictly after a concurrent call to procdesc_close(). 281 */ 282 int 283 procdesc_exit(struct proc *p) 284 { 285 struct procdesc *pd; 286 287 sx_assert(&proctree_lock, SA_XLOCKED); 288 PROC_LOCK_ASSERT(p, MA_OWNED); 289 KASSERT(p->p_procdesc != NULL, ("procdesc_exit: p_procdesc NULL")); 290 291 pd = p->p_procdesc; 292 293 PROCDESC_LOCK(pd); 294 KASSERT((pd->pd_flags & PDF_CLOSED) == 0 || p->p_pptr == p->p_reaper, 295 ("procdesc_exit: closed && parent not reaper")); 296 297 pd->pd_flags |= PDF_EXITED; 298 pd->pd_xstat = KW_EXITCODE(p->p_xexit, p->p_xsig); 299 300 /* 301 * If the process descriptor has been closed, then we have nothing 302 * to do; return 1 so that init will get SIGCHLD and do the reaping. 303 * Clean up the procdesc now rather than letting it happen during 304 * that reap. 305 */ 306 if (pd->pd_flags & PDF_CLOSED) { 307 PROCDESC_UNLOCK(pd); 308 pd->pd_proc = NULL; 309 p->p_procdesc = NULL; 310 procdesc_free(pd); 311 return (1); 312 } 313 if (pd->pd_flags & PDF_SELECTED) { 314 pd->pd_flags &= ~PDF_SELECTED; 315 selwakeup(&pd->pd_selinfo); 316 } 317 KNOTE_LOCKED(&pd->pd_selinfo.si_note, NOTE_EXIT); 318 PROCDESC_UNLOCK(pd); 319 return (0); 320 } 321 322 /* 323 * When a process descriptor is reaped, perhaps as a result of close(), release 324 * the process's reference on the process descriptor. 325 */ 326 void 327 procdesc_reap(struct proc *p) 328 { 329 struct procdesc *pd; 330 331 sx_assert(&proctree_lock, SA_XLOCKED); 332 KASSERT(p->p_procdesc != NULL, ("procdesc_reap: p_procdesc == NULL")); 333 334 pd = p->p_procdesc; 335 pd->pd_proc = NULL; 336 p->p_procdesc = NULL; 337 procdesc_free(pd); 338 } 339 340 /* 341 * procdesc_close() - last close on a process descriptor. If the process is 342 * still running, terminate with SIGKILL (unless PDF_DAEMON is set) and let 343 * its reaper clean up the mess; if not, we have to clean up the zombie 344 * ourselves. 345 */ 346 static int 347 procdesc_close(struct file *fp, struct thread *td) 348 { 349 struct procdesc *pd; 350 struct proc *p; 351 352 KASSERT(fp->f_type == DTYPE_PROCDESC, ("procdesc_close: !procdesc")); 353 354 pd = fp->f_data; 355 fp->f_ops = &badfileops; 356 fp->f_data = NULL; 357 358 sx_xlock(&proctree_lock); 359 PROCDESC_LOCK(pd); 360 pd->pd_flags |= PDF_CLOSED; 361 PROCDESC_UNLOCK(pd); 362 p = pd->pd_proc; 363 if (p == NULL) { 364 /* 365 * This is the case where process' exit status was already 366 * collected and procdesc_reap() was already called. 367 */ 368 sx_xunlock(&proctree_lock); 369 } else { 370 PROC_LOCK(p); 371 AUDIT_ARG_PROCESS(p); 372 if (p->p_state == PRS_ZOMBIE) { 373 /* 374 * If the process is already dead and just awaiting 375 * reaping, do that now. This will release the 376 * process's reference to the process descriptor when it 377 * calls back into procdesc_reap(). 378 */ 379 proc_reap(curthread, p, NULL, 0); 380 } else { 381 /* 382 * If the process is not yet dead, we need to kill it, 383 * but we can't wait around synchronously for it to go 384 * away, as that path leads to madness (and deadlocks). 385 * First, detach the process from its descriptor so that 386 * its exit status will be reported normally. 387 */ 388 pd->pd_proc = NULL; 389 p->p_procdesc = NULL; 390 procdesc_free(pd); 391 392 /* 393 * Next, reparent it to its reaper (usually init(8)) so 394 * that there's someone to pick up the pieces; finally, 395 * terminate with prejudice. 396 */ 397 p->p_sigparent = SIGCHLD; 398 if ((p->p_flag & P_TRACED) == 0) { 399 proc_reparent(p, p->p_reaper, true); 400 } else { 401 proc_clear_orphan(p); 402 p->p_oppid = p->p_reaper->p_pid; 403 proc_add_orphan(p, p->p_reaper); 404 } 405 if ((pd->pd_flags & PDF_DAEMON) == 0) 406 kern_psignal(p, SIGKILL); 407 PROC_UNLOCK(p); 408 sx_xunlock(&proctree_lock); 409 } 410 } 411 412 /* 413 * Release the file descriptor's reference on the process descriptor. 414 */ 415 procdesc_free(pd); 416 return (0); 417 } 418 419 static int 420 procdesc_poll(struct file *fp, int events, struct ucred *active_cred, 421 struct thread *td) 422 { 423 struct procdesc *pd; 424 int revents; 425 426 revents = 0; 427 pd = fp->f_data; 428 PROCDESC_LOCK(pd); 429 if (pd->pd_flags & PDF_EXITED) 430 revents |= POLLHUP; 431 if (revents == 0) { 432 selrecord(td, &pd->pd_selinfo); 433 pd->pd_flags |= PDF_SELECTED; 434 } 435 PROCDESC_UNLOCK(pd); 436 return (revents); 437 } 438 439 static void 440 procdesc_kqops_detach(struct knote *kn) 441 { 442 struct procdesc *pd; 443 444 pd = kn->kn_fp->f_data; 445 knlist_remove(&pd->pd_selinfo.si_note, kn, 0); 446 } 447 448 static int 449 procdesc_kqops_event(struct knote *kn, long hint) 450 { 451 struct procdesc *pd; 452 u_int event; 453 454 pd = kn->kn_fp->f_data; 455 if (hint == 0) { 456 /* 457 * Initial test after registration. Generate a NOTE_EXIT in 458 * case the process already terminated before registration. 459 */ 460 event = pd->pd_flags & PDF_EXITED ? NOTE_EXIT : 0; 461 } else { 462 /* Mask off extra data. */ 463 event = (u_int)hint & NOTE_PCTRLMASK; 464 } 465 466 /* If the user is interested in this event, record it. */ 467 if (kn->kn_sfflags & event) 468 kn->kn_fflags |= event; 469 470 /* Process is gone, so flag the event as finished. */ 471 if (event == NOTE_EXIT) { 472 kn->kn_flags |= EV_EOF | EV_ONESHOT; 473 if (kn->kn_fflags & NOTE_EXIT) 474 kn->kn_data = pd->pd_xstat; 475 if (kn->kn_fflags == 0) 476 kn->kn_flags |= EV_DROP; 477 return (1); 478 } 479 480 return (kn->kn_fflags != 0); 481 } 482 483 static struct filterops procdesc_kqops = { 484 .f_isfd = 1, 485 .f_detach = procdesc_kqops_detach, 486 .f_event = procdesc_kqops_event, 487 }; 488 489 static int 490 procdesc_kqfilter(struct file *fp, struct knote *kn) 491 { 492 struct procdesc *pd; 493 494 pd = fp->f_data; 495 switch (kn->kn_filter) { 496 case EVFILT_PROCDESC: 497 kn->kn_fop = &procdesc_kqops; 498 kn->kn_flags |= EV_CLEAR; 499 knlist_add(&pd->pd_selinfo.si_note, kn, 0); 500 return (0); 501 default: 502 return (EINVAL); 503 } 504 } 505 506 static int 507 procdesc_stat(struct file *fp, struct stat *sb, struct ucred *active_cred) 508 { 509 struct procdesc *pd; 510 struct timeval pstart, boottime; 511 512 /* 513 * XXXRW: Perhaps we should cache some more information from the 514 * process so that we can return it reliably here even after it has 515 * died. For example, caching its credential data. 516 */ 517 bzero(sb, sizeof(*sb)); 518 pd = fp->f_data; 519 sx_slock(&proctree_lock); 520 if (pd->pd_proc != NULL) { 521 PROC_LOCK(pd->pd_proc); 522 AUDIT_ARG_PROCESS(pd->pd_proc); 523 524 /* Set birth and [acm] times to process start time. */ 525 pstart = pd->pd_proc->p_stats->p_start; 526 getboottime(&boottime); 527 timevaladd(&pstart, &boottime); 528 TIMEVAL_TO_TIMESPEC(&pstart, &sb->st_birthtim); 529 sb->st_atim = sb->st_birthtim; 530 sb->st_ctim = sb->st_birthtim; 531 sb->st_mtim = sb->st_birthtim; 532 if (pd->pd_proc->p_state != PRS_ZOMBIE) 533 sb->st_mode = S_IFREG | S_IRWXU; 534 else 535 sb->st_mode = S_IFREG; 536 sb->st_uid = pd->pd_proc->p_ucred->cr_ruid; 537 sb->st_gid = pd->pd_proc->p_ucred->cr_rgid; 538 PROC_UNLOCK(pd->pd_proc); 539 } else 540 sb->st_mode = S_IFREG; 541 sx_sunlock(&proctree_lock); 542 return (0); 543 } 544 545 static int 546 procdesc_fill_kinfo(struct file *fp, struct kinfo_file *kif, 547 struct filedesc *fdp) 548 { 549 struct procdesc *pdp; 550 551 kif->kf_type = KF_TYPE_PROCDESC; 552 pdp = fp->f_data; 553 kif->kf_un.kf_proc.kf_pid = pdp->pd_pid; 554 return (0); 555 } 556