1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 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 * - The pdwait4(2) system call may be used to block (or not) on a process 59 * descriptor to collect termination information. 60 * 61 * Open questions: 62 * 63 * - Will we want to add a pidtoprocdesc(2) system call to allow process 64 * descriptors to be created for processes without pdfork(2)? 65 */ 66 67 #include <sys/cdefs.h> 68 __FBSDID("$FreeBSD$"); 69 70 #include <sys/param.h> 71 #include <sys/capsicum.h> 72 #include <sys/fcntl.h> 73 #include <sys/file.h> 74 #include <sys/filedesc.h> 75 #include <sys/kernel.h> 76 #include <sys/lock.h> 77 #include <sys/mutex.h> 78 #include <sys/poll.h> 79 #include <sys/proc.h> 80 #include <sys/procdesc.h> 81 #include <sys/resourcevar.h> 82 #include <sys/stat.h> 83 #include <sys/sysproto.h> 84 #include <sys/sysctl.h> 85 #include <sys/systm.h> 86 #include <sys/ucred.h> 87 #include <sys/user.h> 88 89 #include <security/audit/audit.h> 90 91 #include <vm/uma.h> 92 93 FEATURE(process_descriptors, "Process Descriptors"); 94 95 static uma_zone_t procdesc_zone; 96 97 static fo_poll_t procdesc_poll; 98 static fo_kqfilter_t procdesc_kqfilter; 99 static fo_stat_t procdesc_stat; 100 static fo_close_t procdesc_close; 101 static fo_fill_kinfo_t procdesc_fill_kinfo; 102 103 static struct fileops procdesc_ops = { 104 .fo_read = invfo_rdwr, 105 .fo_write = invfo_rdwr, 106 .fo_truncate = invfo_truncate, 107 .fo_ioctl = invfo_ioctl, 108 .fo_poll = procdesc_poll, 109 .fo_kqfilter = procdesc_kqfilter, 110 .fo_stat = procdesc_stat, 111 .fo_close = procdesc_close, 112 .fo_chmod = invfo_chmod, 113 .fo_chown = invfo_chown, 114 .fo_sendfile = invfo_sendfile, 115 .fo_fill_kinfo = procdesc_fill_kinfo, 116 .fo_flags = DFLAG_PASSABLE, 117 }; 118 119 /* 120 * Initialize with VFS so that process descriptors are available along with 121 * other file descriptor types. As long as it runs before init(8) starts, 122 * there shouldn't be a problem. 123 */ 124 static void 125 procdesc_init(void *dummy __unused) 126 { 127 128 procdesc_zone = uma_zcreate("procdesc", sizeof(struct procdesc), 129 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 130 if (procdesc_zone == NULL) 131 panic("procdesc_init: procdesc_zone not initialized"); 132 } 133 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_ANY, procdesc_init, NULL); 134 135 /* 136 * Return a locked process given a process descriptor, or ESRCH if it has 137 * died. 138 */ 139 int 140 procdesc_find(struct thread *td, int fd, cap_rights_t *rightsp, 141 struct proc **p) 142 { 143 struct procdesc *pd; 144 struct file *fp; 145 int error; 146 147 error = fget(td, fd, rightsp, &fp); 148 if (error) 149 return (error); 150 if (fp->f_type != DTYPE_PROCDESC) { 151 error = EBADF; 152 goto out; 153 } 154 pd = fp->f_data; 155 sx_slock(&proctree_lock); 156 if (pd->pd_proc != NULL) { 157 *p = pd->pd_proc; 158 PROC_LOCK(*p); 159 } else 160 error = ESRCH; 161 sx_sunlock(&proctree_lock); 162 out: 163 fdrop(fp, td); 164 return (error); 165 } 166 167 /* 168 * Function to be used by procstat(1) sysctls when returning procdesc 169 * information. 170 */ 171 pid_t 172 procdesc_pid(struct file *fp_procdesc) 173 { 174 struct procdesc *pd; 175 176 KASSERT(fp_procdesc->f_type == DTYPE_PROCDESC, 177 ("procdesc_pid: !procdesc")); 178 179 pd = fp_procdesc->f_data; 180 return (pd->pd_pid); 181 } 182 183 /* 184 * Retrieve the PID associated with a process descriptor. 185 */ 186 int 187 kern_pdgetpid(struct thread *td, int fd, cap_rights_t *rightsp, pid_t *pidp) 188 { 189 struct file *fp; 190 int error; 191 192 error = fget(td, fd, rightsp, &fp); 193 if (error) 194 return (error); 195 if (fp->f_type != DTYPE_PROCDESC) { 196 error = EBADF; 197 goto out; 198 } 199 *pidp = procdesc_pid(fp); 200 out: 201 fdrop(fp, td); 202 return (error); 203 } 204 205 /* 206 * System call to return the pid of a process given its process descriptor. 207 */ 208 int 209 sys_pdgetpid(struct thread *td, struct pdgetpid_args *uap) 210 { 211 pid_t pid; 212 int error; 213 214 AUDIT_ARG_FD(uap->fd); 215 error = kern_pdgetpid(td, uap->fd, &cap_pdgetpid_rights, &pid); 216 if (error == 0) 217 error = copyout(&pid, uap->pidp, sizeof(pid)); 218 return (error); 219 } 220 221 /* 222 * When a new process is forked by pdfork(), a file descriptor is allocated 223 * by the fork code first, then the process is forked, and then we get a 224 * chance to set up the process descriptor. Failure is not permitted at this 225 * point, so procdesc_new() must succeed. 226 */ 227 void 228 procdesc_new(struct proc *p, int flags) 229 { 230 struct procdesc *pd; 231 232 pd = uma_zalloc(procdesc_zone, M_WAITOK | M_ZERO); 233 pd->pd_proc = p; 234 pd->pd_pid = p->p_pid; 235 p->p_procdesc = pd; 236 pd->pd_flags = 0; 237 if (flags & PD_DAEMON) 238 pd->pd_flags |= PDF_DAEMON; 239 PROCDESC_LOCK_INIT(pd); 240 knlist_init_mtx(&pd->pd_selinfo.si_note, &pd->pd_lock); 241 242 /* 243 * Process descriptors start out with two references: one from their 244 * struct file, and the other from their struct proc. 245 */ 246 refcount_init(&pd->pd_refcount, 2); 247 } 248 249 /* 250 * Create a new process decriptor for the process that refers to it. 251 */ 252 int 253 procdesc_falloc(struct thread *td, struct file **resultfp, int *resultfd, 254 int flags, struct filecaps *fcaps) 255 { 256 int fflags; 257 258 fflags = 0; 259 if (flags & PD_CLOEXEC) 260 fflags = O_CLOEXEC; 261 262 return (falloc_caps(td, resultfp, resultfd, fflags, fcaps)); 263 } 264 265 /* 266 * Initialize a file with a process descriptor. 267 */ 268 void 269 procdesc_finit(struct procdesc *pdp, struct file *fp) 270 { 271 272 finit(fp, FREAD | FWRITE, DTYPE_PROCDESC, pdp, &procdesc_ops); 273 } 274 275 static void 276 procdesc_free(struct procdesc *pd) 277 { 278 279 /* 280 * When the last reference is released, we assert that the descriptor 281 * has been closed, but not that the process has exited, as we will 282 * detach the descriptor before the process dies if the descript is 283 * closed, as we can't wait synchronously. 284 */ 285 if (refcount_release(&pd->pd_refcount)) { 286 KASSERT(pd->pd_proc == NULL, 287 ("procdesc_free: pd_proc != NULL")); 288 KASSERT((pd->pd_flags & PDF_CLOSED), 289 ("procdesc_free: !PDF_CLOSED")); 290 291 knlist_destroy(&pd->pd_selinfo.si_note); 292 PROCDESC_LOCK_DESTROY(pd); 293 uma_zfree(procdesc_zone, pd); 294 } 295 } 296 297 /* 298 * procdesc_exit() - notify a process descriptor that its process is exiting. 299 * We use the proctree_lock to ensure that process exit either happens 300 * strictly before or strictly after a concurrent call to procdesc_close(). 301 */ 302 int 303 procdesc_exit(struct proc *p) 304 { 305 struct procdesc *pd; 306 307 sx_assert(&proctree_lock, SA_XLOCKED); 308 PROC_LOCK_ASSERT(p, MA_OWNED); 309 KASSERT(p->p_procdesc != NULL, ("procdesc_exit: p_procdesc NULL")); 310 311 pd = p->p_procdesc; 312 313 PROCDESC_LOCK(pd); 314 KASSERT((pd->pd_flags & PDF_CLOSED) == 0 || p->p_pptr == p->p_reaper, 315 ("procdesc_exit: closed && parent not reaper")); 316 317 pd->pd_flags |= PDF_EXITED; 318 pd->pd_xstat = KW_EXITCODE(p->p_xexit, p->p_xsig); 319 320 /* 321 * If the process descriptor has been closed, then we have nothing 322 * to do; return 1 so that init will get SIGCHLD and do the reaping. 323 * Clean up the procdesc now rather than letting it happen during 324 * that reap. 325 */ 326 if (pd->pd_flags & PDF_CLOSED) { 327 PROCDESC_UNLOCK(pd); 328 pd->pd_proc = NULL; 329 p->p_procdesc = NULL; 330 procdesc_free(pd); 331 return (1); 332 } 333 if (pd->pd_flags & PDF_SELECTED) { 334 pd->pd_flags &= ~PDF_SELECTED; 335 selwakeup(&pd->pd_selinfo); 336 } 337 KNOTE_LOCKED(&pd->pd_selinfo.si_note, NOTE_EXIT); 338 PROCDESC_UNLOCK(pd); 339 return (0); 340 } 341 342 /* 343 * When a process descriptor is reaped, perhaps as a result of close() or 344 * pdwait4(), release the process's reference on the process descriptor. 345 */ 346 void 347 procdesc_reap(struct proc *p) 348 { 349 struct procdesc *pd; 350 351 sx_assert(&proctree_lock, SA_XLOCKED); 352 KASSERT(p->p_procdesc != NULL, ("procdesc_reap: p_procdesc == NULL")); 353 354 pd = p->p_procdesc; 355 pd->pd_proc = NULL; 356 p->p_procdesc = NULL; 357 procdesc_free(pd); 358 } 359 360 /* 361 * procdesc_close() - last close on a process descriptor. If the process is 362 * still running, terminate with SIGKILL (unless PDF_DAEMON is set) and let 363 * its reaper clean up the mess; if not, we have to clean up the zombie 364 * ourselves. 365 */ 366 static int 367 procdesc_close(struct file *fp, struct thread *td) 368 { 369 struct procdesc *pd; 370 struct proc *p; 371 372 KASSERT(fp->f_type == DTYPE_PROCDESC, ("procdesc_close: !procdesc")); 373 374 pd = fp->f_data; 375 fp->f_ops = &badfileops; 376 fp->f_data = NULL; 377 378 sx_xlock(&proctree_lock); 379 PROCDESC_LOCK(pd); 380 pd->pd_flags |= PDF_CLOSED; 381 PROCDESC_UNLOCK(pd); 382 p = pd->pd_proc; 383 if (p == NULL) { 384 /* 385 * This is the case where process' exit status was already 386 * collected and procdesc_reap() was already called. 387 */ 388 sx_xunlock(&proctree_lock); 389 } else { 390 PROC_LOCK(p); 391 AUDIT_ARG_PROCESS(p); 392 if (p->p_state == PRS_ZOMBIE) { 393 /* 394 * If the process is already dead and just awaiting 395 * reaping, do that now. This will release the 396 * process's reference to the process descriptor when it 397 * calls back into procdesc_reap(). 398 */ 399 proc_reap(curthread, p, NULL, 0); 400 } else { 401 /* 402 * If the process is not yet dead, we need to kill it, 403 * but we can't wait around synchronously for it to go 404 * away, as that path leads to madness (and deadlocks). 405 * First, detach the process from its descriptor so that 406 * its exit status will be reported normally. 407 */ 408 pd->pd_proc = NULL; 409 p->p_procdesc = NULL; 410 procdesc_free(pd); 411 412 /* 413 * Next, reparent it to its reaper (usually init(8)) so 414 * that there's someone to pick up the pieces; finally, 415 * terminate with prejudice. 416 */ 417 p->p_sigparent = SIGCHLD; 418 if ((p->p_flag & P_TRACED) == 0) { 419 proc_reparent(p, p->p_reaper, true); 420 } else { 421 proc_clear_orphan(p); 422 p->p_oppid = p->p_reaper->p_pid; 423 proc_add_orphan(p, p->p_reaper); 424 } 425 if ((pd->pd_flags & PDF_DAEMON) == 0) 426 kern_psignal(p, SIGKILL); 427 PROC_UNLOCK(p); 428 sx_xunlock(&proctree_lock); 429 } 430 } 431 432 /* 433 * Release the file descriptor's reference on the process descriptor. 434 */ 435 procdesc_free(pd); 436 return (0); 437 } 438 439 static int 440 procdesc_poll(struct file *fp, int events, struct ucred *active_cred, 441 struct thread *td) 442 { 443 struct procdesc *pd; 444 int revents; 445 446 revents = 0; 447 pd = fp->f_data; 448 PROCDESC_LOCK(pd); 449 if (pd->pd_flags & PDF_EXITED) 450 revents |= POLLHUP; 451 if (revents == 0) { 452 selrecord(td, &pd->pd_selinfo); 453 pd->pd_flags |= PDF_SELECTED; 454 } 455 PROCDESC_UNLOCK(pd); 456 return (revents); 457 } 458 459 static void 460 procdesc_kqops_detach(struct knote *kn) 461 { 462 struct procdesc *pd; 463 464 pd = kn->kn_fp->f_data; 465 knlist_remove(&pd->pd_selinfo.si_note, kn, 0); 466 } 467 468 static int 469 procdesc_kqops_event(struct knote *kn, long hint) 470 { 471 struct procdesc *pd; 472 u_int event; 473 474 pd = kn->kn_fp->f_data; 475 if (hint == 0) { 476 /* 477 * Initial test after registration. Generate a NOTE_EXIT in 478 * case the process already terminated before registration. 479 */ 480 event = pd->pd_flags & PDF_EXITED ? NOTE_EXIT : 0; 481 } else { 482 /* Mask off extra data. */ 483 event = (u_int)hint & NOTE_PCTRLMASK; 484 } 485 486 /* If the user is interested in this event, record it. */ 487 if (kn->kn_sfflags & event) 488 kn->kn_fflags |= event; 489 490 /* Process is gone, so flag the event as finished. */ 491 if (event == NOTE_EXIT) { 492 kn->kn_flags |= EV_EOF | EV_ONESHOT; 493 if (kn->kn_fflags & NOTE_EXIT) 494 kn->kn_data = pd->pd_xstat; 495 if (kn->kn_fflags == 0) 496 kn->kn_flags |= EV_DROP; 497 return (1); 498 } 499 500 return (kn->kn_fflags != 0); 501 } 502 503 static struct filterops procdesc_kqops = { 504 .f_isfd = 1, 505 .f_detach = procdesc_kqops_detach, 506 .f_event = procdesc_kqops_event, 507 }; 508 509 static int 510 procdesc_kqfilter(struct file *fp, struct knote *kn) 511 { 512 struct procdesc *pd; 513 514 pd = fp->f_data; 515 switch (kn->kn_filter) { 516 case EVFILT_PROCDESC: 517 kn->kn_fop = &procdesc_kqops; 518 kn->kn_flags |= EV_CLEAR; 519 knlist_add(&pd->pd_selinfo.si_note, kn, 0); 520 return (0); 521 default: 522 return (EINVAL); 523 } 524 } 525 526 static int 527 procdesc_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 528 struct thread *td) 529 { 530 struct procdesc *pd; 531 struct timeval pstart, boottime; 532 533 /* 534 * XXXRW: Perhaps we should cache some more information from the 535 * process so that we can return it reliably here even after it has 536 * died. For example, caching its credential data. 537 */ 538 bzero(sb, sizeof(*sb)); 539 pd = fp->f_data; 540 sx_slock(&proctree_lock); 541 if (pd->pd_proc != NULL) { 542 PROC_LOCK(pd->pd_proc); 543 AUDIT_ARG_PROCESS(pd->pd_proc); 544 545 /* Set birth and [acm] times to process start time. */ 546 pstart = pd->pd_proc->p_stats->p_start; 547 getboottime(&boottime); 548 timevaladd(&pstart, &boottime); 549 TIMEVAL_TO_TIMESPEC(&pstart, &sb->st_birthtim); 550 sb->st_atim = sb->st_birthtim; 551 sb->st_ctim = sb->st_birthtim; 552 sb->st_mtim = sb->st_birthtim; 553 if (pd->pd_proc->p_state != PRS_ZOMBIE) 554 sb->st_mode = S_IFREG | S_IRWXU; 555 else 556 sb->st_mode = S_IFREG; 557 sb->st_uid = pd->pd_proc->p_ucred->cr_ruid; 558 sb->st_gid = pd->pd_proc->p_ucred->cr_rgid; 559 PROC_UNLOCK(pd->pd_proc); 560 } else 561 sb->st_mode = S_IFREG; 562 sx_sunlock(&proctree_lock); 563 return (0); 564 } 565 566 static int 567 procdesc_fill_kinfo(struct file *fp, struct kinfo_file *kif, 568 struct filedesc *fdp) 569 { 570 struct procdesc *pdp; 571 572 kif->kf_type = KF_TYPE_PROCDESC; 573 pdp = fp->f_data; 574 kif->kf_un.kf_proc.kf_pid = pdp->pd_pid; 575 return (0); 576 } 577