1/*- 2 * Copyright (c) 2009, 2016 Robert N. M. Watson 3 * All rights reserved. 4 * 5 * This software was developed at the University of Cambridge Computer 6 * Laboratory with support from a grant from Google, Inc. 7 * 8 * Portions of this software were developed by BAE Systems, the University of 9 * Cambridge Computer Laboratory, and Memorial University under DARPA/AFRL 10 * contract FA8650-15-C-7558 ("CADETS"), as part of the DARPA Transparent 11 * Computing (TC) research program. 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 1. Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 2. Redistributions in binary form must reproduce the above copyright 19 * notice, this list of conditions and the following disclaimer in the 20 * documentation and/or other materials provided with the distribution. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 */ 34 35/*- 36 * FreeBSD process descriptor facility. 37 * 38 * Some processes are represented by a file descriptor, which will be used in 39 * preference to signaling and pids for the purposes of process management, 40 * and is, in effect, a form of capability. When a process descriptor is 41 * used with a process, it ceases to be visible to certain traditional UNIX 42 * process facilities, such as waitpid(2). 43 * 44 * Some semantics: 45 * 46 * - At most one process descriptor will exist for any process, although 47 * references to that descriptor may be held from many processes (or even 48 * be in flight between processes over a local domain socket). 49 * - Last close on the process descriptor will terminate the process using 50 * SIGKILL and reparent it to init so that there's a process to reap it 51 * when it's done exiting. 52 * - If the process exits before the descriptor is closed, it will not 53 * generate SIGCHLD on termination, or be picked up by waitpid(). 54 * - The pdkill(2) system call may be used to deliver a signal to the process 55 * using its process descriptor. 56 * - The pdwait4(2) system call may be used to block (or not) on a process 57 * descriptor to collect termination information. 58 * 59 * Open questions: 60 * 61 * - How to handle ptrace(2)? 62 * - Will we want to add a pidtoprocdesc(2) system call to allow process 63 * descriptors to be created for processes without pdfork(2)? 64 */ 65 66#include <sys/cdefs.h> 67__FBSDID("$FreeBSD: stable/11/sys/kern/sys_procdesc.c 330964 2018-03-15 02:20:06Z eadler $"); 68 69#include <sys/param.h> 70#include <sys/capsicum.h> 71#include <sys/fcntl.h> 72#include <sys/file.h> 73#include <sys/filedesc.h> 74#include <sys/kernel.h> 75#include <sys/lock.h> 76#include <sys/mutex.h> 77#include <sys/poll.h> 78#include <sys/proc.h> 79#include <sys/procdesc.h> 80#include <sys/resourcevar.h> 81#include <sys/stat.h> 82#include <sys/sysproto.h> 83#include <sys/sysctl.h> 84#include <sys/systm.h> 85#include <sys/ucred.h> 86#include <sys/user.h> 87 88#include <security/audit/audit.h> 89 90#include <vm/uma.h> 91 92FEATURE(process_descriptors, "Process Descriptors"); 93 94static uma_zone_t procdesc_zone; 95 96static fo_poll_t procdesc_poll; 97static fo_kqfilter_t procdesc_kqfilter; 98static fo_stat_t procdesc_stat; 99static fo_close_t procdesc_close; 100static fo_fill_kinfo_t procdesc_fill_kinfo; 101 102static struct fileops procdesc_ops = { 103 .fo_read = invfo_rdwr, 104 .fo_write = invfo_rdwr, 105 .fo_truncate = invfo_truncate, 106 .fo_ioctl = invfo_ioctl, 107 .fo_poll = procdesc_poll, 108 .fo_kqfilter = procdesc_kqfilter, 109 .fo_stat = procdesc_stat, 110 .fo_close = procdesc_close, 111 .fo_chmod = invfo_chmod, 112 .fo_chown = invfo_chown, 113 .fo_sendfile = invfo_sendfile, 114 .fo_fill_kinfo = procdesc_fill_kinfo, 115 .fo_flags = DFLAG_PASSABLE, 116}; 117 118/* 119 * Initialize with VFS so that process descriptors are available along with 120 * other file descriptor types. As long as it runs before init(8) starts, 121 * there shouldn't be a problem. 122 */ 123static void 124procdesc_init(void *dummy __unused) 125{ 126 127 procdesc_zone = uma_zcreate("procdesc", sizeof(struct procdesc), 128 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 129 if (procdesc_zone == NULL) 130 panic("procdesc_init: procdesc_zone not initialized"); 131} 132SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_ANY, procdesc_init, NULL); 133 134/* 135 * Return a locked process given a process descriptor, or ESRCH if it has 136 * died. 137 */ 138int 139procdesc_find(struct thread *td, int fd, cap_rights_t *rightsp, 140 struct proc **p) 141{ 142 struct procdesc *pd; 143 struct file *fp; 144 int error; 145 146 error = fget(td, fd, rightsp, &fp); 147 if (error) 148 return (error); 149 if (fp->f_type != DTYPE_PROCDESC) { 150 error = EBADF; 151 goto out; 152 } 153 pd = fp->f_data; 154 sx_slock(&proctree_lock); 155 if (pd->pd_proc != NULL) { 156 *p = pd->pd_proc; 157 PROC_LOCK(*p); 158 } else 159 error = ESRCH; 160 sx_sunlock(&proctree_lock); 161out: 162 fdrop(fp, td); 163 return (error); 164} 165 166/* 167 * Function to be used by procstat(1) sysctls when returning procdesc 168 * information. 169 */ 170pid_t 171procdesc_pid(struct file *fp_procdesc) 172{ 173 struct procdesc *pd; 174 175 KASSERT(fp_procdesc->f_type == DTYPE_PROCDESC, 176 ("procdesc_pid: !procdesc")); 177 178 pd = fp_procdesc->f_data; 179 return (pd->pd_pid); 180} 181 182/* 183 * Retrieve the PID associated with a process descriptor. 184 */ 185int 186kern_pdgetpid(struct thread *td, int fd, cap_rights_t *rightsp, pid_t *pidp) 187{ 188 struct file *fp; 189 int error; 190 191 error = fget(td, fd, rightsp, &fp); 192 if (error) 193 return (error); 194 if (fp->f_type != DTYPE_PROCDESC) { 195 error = EBADF; 196 goto out; 197 } 198 *pidp = procdesc_pid(fp); 199out: 200 fdrop(fp, td); 201 return (error); 202} 203 204/* 205 * System call to return the pid of a process given its process descriptor. 206 */ 207int 208sys_pdgetpid(struct thread *td, struct pdgetpid_args *uap) 209{ 210 cap_rights_t rights; 211 pid_t pid; 212 int error; 213 214 AUDIT_ARG_FD(uap->fd); 215 error = kern_pdgetpid(td, uap->fd, 216 cap_rights_init(&rights, CAP_PDGETPID), &pid); 217 if (error == 0) 218 error = copyout(&pid, uap->pidp, sizeof(pid)); 219 return (error); 220} 221 222/* 223 * When a new process is forked by pdfork(), a file descriptor is allocated 224 * by the fork code first, then the process is forked, and then we get a 225 * chance to set up the process descriptor. Failure is not permitted at this 226 * point, so procdesc_new() must succeed. 227 */ 228void 229procdesc_new(struct proc *p, int flags) 230{ 231 struct procdesc *pd; 232 233 pd = uma_zalloc(procdesc_zone, M_WAITOK | M_ZERO); 234 pd->pd_proc = p; 235 pd->pd_pid = p->p_pid; 236 p->p_procdesc = pd; 237 pd->pd_flags = 0; 238 if (flags & PD_DAEMON) 239 pd->pd_flags |= PDF_DAEMON; 240 PROCDESC_LOCK_INIT(pd); 241 knlist_init_mtx(&pd->pd_selinfo.si_note, &pd->pd_lock); 242 243 /* 244 * Process descriptors start out with two references: one from their 245 * struct file, and the other from their struct proc. 246 */ 247 refcount_init(&pd->pd_refcount, 2); 248} 249 250/* 251 * Create a new process decriptor for the process that refers to it. 252 */ 253int 254procdesc_falloc(struct thread *td, struct file **resultfp, int *resultfd, 255 int flags, struct filecaps *fcaps) 256{ 257 int fflags; 258 259 fflags = 0; 260 if (flags & PD_CLOEXEC) 261 fflags = O_CLOEXEC; 262 263 return (falloc_caps(td, resultfp, resultfd, fflags, fcaps)); 264} 265 266/* 267 * Initialize a file with a process descriptor. 268 */ 269void 270procdesc_finit(struct procdesc *pdp, struct file *fp) 271{ 272 273 finit(fp, FREAD | FWRITE, DTYPE_PROCDESC, pdp, &procdesc_ops); 274} 275 276static void 277procdesc_free(struct procdesc *pd) 278{ 279 280 /* 281 * When the last reference is released, we assert that the descriptor 282 * has been closed, but not that the process has exited, as we will 283 * detach the descriptor before the process dies if the descript is 284 * closed, as we can't wait synchronously. 285 */ 286 if (refcount_release(&pd->pd_refcount)) { 287 KASSERT(pd->pd_proc == NULL, 288 ("procdesc_free: pd_proc != NULL")); 289 KASSERT((pd->pd_flags & PDF_CLOSED), 290 ("procdesc_free: !PDF_CLOSED")); 291 292 knlist_destroy(&pd->pd_selinfo.si_note); 293 PROCDESC_LOCK_DESTROY(pd); 294 uma_zfree(procdesc_zone, pd); 295 } 296} 297 298/* 299 * procdesc_exit() - notify a process descriptor that its process is exiting. 300 * We use the proctree_lock to ensure that process exit either happens 301 * strictly before or strictly after a concurrent call to procdesc_close(). 302 */ 303int 304procdesc_exit(struct proc *p) 305{ 306 struct procdesc *pd; 307 308 sx_assert(&proctree_lock, SA_XLOCKED); 309 PROC_LOCK_ASSERT(p, MA_OWNED); 310 KASSERT(p->p_procdesc != NULL, ("procdesc_exit: p_procdesc NULL")); 311 312 pd = p->p_procdesc; 313 314 PROCDESC_LOCK(pd); 315 KASSERT((pd->pd_flags & PDF_CLOSED) == 0 || p->p_pptr == initproc, 316 ("procdesc_exit: closed && parent not init")); 317 318 pd->pd_flags |= PDF_EXITED; 319 pd->pd_xstat = KW_EXITCODE(p->p_xexit, p->p_xsig); 320 321 /* 322 * If the process descriptor has been closed, then we have nothing 323 * to do; return 1 so that init will get SIGCHLD and do the reaping. 324 * Clean up the procdesc now rather than letting it happen during 325 * that reap. 326 */ 327 if (pd->pd_flags & PDF_CLOSED) { 328 PROCDESC_UNLOCK(pd); 329 pd->pd_proc = NULL; 330 p->p_procdesc = NULL; 331 procdesc_free(pd); 332 return (1); 333 } 334 if (pd->pd_flags & PDF_SELECTED) { 335 pd->pd_flags &= ~PDF_SELECTED; 336 selwakeup(&pd->pd_selinfo); 337 } 338 KNOTE_LOCKED(&pd->pd_selinfo.si_note, NOTE_EXIT); 339 PROCDESC_UNLOCK(pd); 340 return (0); 341} 342 343/* 344 * When a process descriptor is reaped, perhaps as a result of close() or 345 * pdwait4(), release the process's reference on the process descriptor. 346 */ 347void 348procdesc_reap(struct proc *p) 349{ 350 struct procdesc *pd; 351 352 sx_assert(&proctree_lock, SA_XLOCKED); 353 KASSERT(p->p_procdesc != NULL, ("procdesc_reap: p_procdesc == NULL")); 354 355 pd = p->p_procdesc; 356 pd->pd_proc = NULL; 357 p->p_procdesc = NULL; 358 procdesc_free(pd); 359} 360 361/* 362 * procdesc_close() - last close on a process descriptor. If the process is 363 * still running, terminate with SIGKILL (unless PDF_DAEMON is set) and let 364 * init(8) clean up the mess; if not, we have to clean up the zombie ourselves. 365 */ 366static int 367procdesc_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_SLOCK(p); 400 proc_reap(curthread, p, NULL, 0); 401 } else { 402 /* 403 * If the process is not yet dead, we need to kill it, 404 * but we can't wait around synchronously for it to go 405 * away, as that path leads to madness (and deadlocks). 406 * First, detach the process from its descriptor so that 407 * its exit status will be reported normally. 408 */ 409 pd->pd_proc = NULL; 410 p->p_procdesc = NULL; 411 procdesc_free(pd); 412 413 /* 414 * Next, reparent it to init(8) so that there's someone 415 * to pick up the pieces; finally, terminate with 416 * prejudice. 417 */ 418 p->p_sigparent = SIGCHLD; 419 proc_reparent(p, initproc); 420 if ((pd->pd_flags & PDF_DAEMON) == 0) 421 kern_psignal(p, SIGKILL); 422 PROC_UNLOCK(p); 423 sx_xunlock(&proctree_lock); 424 } 425 } 426 427 /* 428 * Release the file descriptor's reference on the process descriptor. 429 */ 430 procdesc_free(pd); 431 return (0); 432} 433 434static int 435procdesc_poll(struct file *fp, int events, struct ucred *active_cred, 436 struct thread *td) 437{ 438 struct procdesc *pd; 439 int revents; 440 441 revents = 0; 442 pd = fp->f_data; 443 PROCDESC_LOCK(pd); 444 if (pd->pd_flags & PDF_EXITED) 445 revents |= POLLHUP; 446 if (revents == 0) { 447 selrecord(td, &pd->pd_selinfo); 448 pd->pd_flags |= PDF_SELECTED; 449 } 450 PROCDESC_UNLOCK(pd); 451 return (revents); 452} 453 454static void 455procdesc_kqops_detach(struct knote *kn) 456{ 457 struct procdesc *pd; 458 459 pd = kn->kn_fp->f_data; 460 knlist_remove(&pd->pd_selinfo.si_note, kn, 0); 461} 462 463static int 464procdesc_kqops_event(struct knote *kn, long hint) 465{ 466 struct procdesc *pd; 467 u_int event; 468 469 pd = kn->kn_fp->f_data; 470 if (hint == 0) { 471 /* 472 * Initial test after registration. Generate a NOTE_EXIT in 473 * case the process already terminated before registration. 474 */ 475 event = pd->pd_flags & PDF_EXITED ? NOTE_EXIT : 0; 476 } else { 477 /* Mask off extra data. */ 478 event = (u_int)hint & NOTE_PCTRLMASK; 479 } 480 481 /* If the user is interested in this event, record it. */ 482 if (kn->kn_sfflags & event) 483 kn->kn_fflags |= event; 484 485 /* Process is gone, so flag the event as finished. */ 486 if (event == NOTE_EXIT) { 487 kn->kn_flags |= EV_EOF | EV_ONESHOT; 488 if (kn->kn_fflags & NOTE_EXIT) 489 kn->kn_data = pd->pd_xstat; 490 if (kn->kn_fflags == 0) 491 kn->kn_flags |= EV_DROP; 492 return (1); 493 } 494 495 return (kn->kn_fflags != 0); 496} 497 498static struct filterops procdesc_kqops = { 499 .f_isfd = 1, 500 .f_detach = procdesc_kqops_detach, 501 .f_event = procdesc_kqops_event, 502}; 503 504static int 505procdesc_kqfilter(struct file *fp, struct knote *kn) 506{ 507 struct procdesc *pd; 508 509 pd = fp->f_data; 510 switch (kn->kn_filter) { 511 case EVFILT_PROCDESC: 512 kn->kn_fop = &procdesc_kqops; 513 kn->kn_flags |= EV_CLEAR; 514 knlist_add(&pd->pd_selinfo.si_note, kn, 0); 515 return (0); 516 default: 517 return (EINVAL); 518 } 519} 520 521static int 522procdesc_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 523 struct thread *td) 524{ 525 struct procdesc *pd; 526 struct timeval pstart, boottime; 527 528 /* 529 * XXXRW: Perhaps we should cache some more information from the 530 * process so that we can return it reliably here even after it has 531 * died. For example, caching its credential data. 532 */ 533 bzero(sb, sizeof(*sb)); 534 pd = fp->f_data; 535 sx_slock(&proctree_lock); 536 if (pd->pd_proc != NULL) { 537 PROC_LOCK(pd->pd_proc); 538 AUDIT_ARG_PROCESS(pd->pd_proc); 539 540 /* Set birth and [acm] times to process start time. */ 541 pstart = pd->pd_proc->p_stats->p_start; 542 getboottime(&boottime); 543 timevaladd(&pstart, &boottime); 544 TIMEVAL_TO_TIMESPEC(&pstart, &sb->st_birthtim); 545 sb->st_atim = sb->st_birthtim; 546 sb->st_ctim = sb->st_birthtim; 547 sb->st_mtim = sb->st_birthtim; 548 if (pd->pd_proc->p_state != PRS_ZOMBIE) 549 sb->st_mode = S_IFREG | S_IRWXU; 550 else 551 sb->st_mode = S_IFREG; 552 sb->st_uid = pd->pd_proc->p_ucred->cr_ruid; 553 sb->st_gid = pd->pd_proc->p_ucred->cr_rgid; 554 PROC_UNLOCK(pd->pd_proc); 555 } else 556 sb->st_mode = S_IFREG; 557 sx_sunlock(&proctree_lock); 558 return (0); 559} 560 561static int 562procdesc_fill_kinfo(struct file *fp, struct kinfo_file *kif, 563 struct filedesc *fdp) 564{ 565 struct procdesc *pdp; 566 567 kif->kf_type = KF_TYPE_PROCDESC; 568 pdp = fp->f_data; 569 kif->kf_un.kf_proc.kf_pid = pdp->pd_pid; 570 return (0); 571} 572