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