1/*- 2 * Copyright (c) 2009 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 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 */ 29 30/*- 31 * FreeBSD process descriptor facility. 32 * 33 * Some processes are represented by a file descriptor, which will be used in 34 * preference to signaling and pids for the purposes of process management, 35 * and is, in effect, a form of capability. When a process descriptor is 36 * used with a process, it ceases to be visible to certain traditional UNIX 37 * process facilities, such as waitpid(2). 38 * 39 * Some semantics: 40 * 41 * - At most one process descriptor will exist for any process, although 42 * references to that descriptor may be held from many processes (or even 43 * be in flight between processes over a local domain socket). 44 * - Last close on the process descriptor will terminate the process using 45 * SIGKILL and reparent it to init so that there's a process to reap it 46 * when it's done exiting. 47 * - If the process exits before the descriptor is closed, it will not 48 * generate SIGCHLD on termination, or be picked up by waitpid(). 49 * - The pdkill(2) system call may be used to deliver a signal to the process 50 * using its process descriptor. 51 * - The pdwait4(2) system call may be used to block (or not) on a process 52 * descriptor to collect termination information. 53 * 54 * Open questions: 55 * 56 * - How to handle ptrace(2)? 57 * - Will we want to add a pidtoprocdesc(2) system call to allow process 58 * descriptors to be created for processes without pdfork(2)? 59 */ 60 61#include <sys/cdefs.h> 62__FBSDID("$FreeBSD: releng/11.0/sys/kern/sys_procdesc.c 301573 2016-06-08 02:09:14Z oshogbo $"); 63 64#include <sys/param.h> 65#include <sys/capsicum.h> 66#include <sys/fcntl.h> 67#include <sys/file.h> 68#include <sys/filedesc.h> 69#include <sys/kernel.h> 70#include <sys/lock.h> 71#include <sys/mutex.h> 72#include <sys/poll.h> 73#include <sys/proc.h> 74#include <sys/procdesc.h> 75#include <sys/resourcevar.h> 76#include <sys/stat.h> 77#include <sys/sysproto.h> 78#include <sys/sysctl.h> 79#include <sys/systm.h> 80#include <sys/ucred.h> 81#include <sys/user.h> 82 83#include <security/audit/audit.h> 84 85#include <vm/uma.h> 86 87FEATURE(process_descriptors, "Process Descriptors"); 88 89static uma_zone_t procdesc_zone; 90 91static fo_poll_t procdesc_poll; 92static fo_kqfilter_t procdesc_kqfilter; 93static fo_stat_t procdesc_stat; 94static fo_close_t procdesc_close; 95static fo_fill_kinfo_t procdesc_fill_kinfo; 96 97static struct fileops procdesc_ops = { 98 .fo_read = invfo_rdwr, 99 .fo_write = invfo_rdwr, 100 .fo_truncate = invfo_truncate, 101 .fo_ioctl = invfo_ioctl, 102 .fo_poll = procdesc_poll, 103 .fo_kqfilter = procdesc_kqfilter, 104 .fo_stat = procdesc_stat, 105 .fo_close = procdesc_close, 106 .fo_chmod = invfo_chmod, 107 .fo_chown = invfo_chown, 108 .fo_sendfile = invfo_sendfile, 109 .fo_fill_kinfo = procdesc_fill_kinfo, 110 .fo_flags = DFLAG_PASSABLE, 111}; 112 113/* 114 * Initialize with VFS so that process descriptors are available along with 115 * other file descriptor types. As long as it runs before init(8) starts, 116 * there shouldn't be a problem. 117 */ 118static void 119procdesc_init(void *dummy __unused) 120{ 121 122 procdesc_zone = uma_zcreate("procdesc", sizeof(struct procdesc), 123 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); 124 if (procdesc_zone == NULL) 125 panic("procdesc_init: procdesc_zone not initialized"); 126} 127SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_ANY, procdesc_init, NULL); 128 129/* 130 * Return a locked process given a process descriptor, or ESRCH if it has 131 * died. 132 */ 133int 134procdesc_find(struct thread *td, int fd, cap_rights_t *rightsp, 135 struct proc **p) 136{ 137 struct procdesc *pd; 138 struct file *fp; 139 int error; 140 141 error = fget(td, fd, rightsp, &fp); 142 if (error) 143 return (error); 144 if (fp->f_type != DTYPE_PROCDESC) { 145 error = EBADF; 146 goto out; 147 } 148 pd = fp->f_data; 149 sx_slock(&proctree_lock); 150 if (pd->pd_proc != NULL) { 151 *p = pd->pd_proc; 152 PROC_LOCK(*p); 153 } else 154 error = ESRCH; 155 sx_sunlock(&proctree_lock); 156out: 157 fdrop(fp, td); 158 return (error); 159} 160 161/* 162 * Function to be used by procstat(1) sysctls when returning procdesc 163 * information. 164 */ 165pid_t 166procdesc_pid(struct file *fp_procdesc) 167{ 168 struct procdesc *pd; 169 170 KASSERT(fp_procdesc->f_type == DTYPE_PROCDESC, 171 ("procdesc_pid: !procdesc")); 172 173 pd = fp_procdesc->f_data; 174 return (pd->pd_pid); 175} 176 177/* 178 * Retrieve the PID associated with a process descriptor. 179 */ 180int 181kern_pdgetpid(struct thread *td, int fd, cap_rights_t *rightsp, pid_t *pidp) 182{ 183 struct file *fp; 184 int error; 185 186 error = fget(td, fd, rightsp, &fp); 187 if (error) 188 return (error); 189 if (fp->f_type != DTYPE_PROCDESC) { 190 error = EBADF; 191 goto out; 192 } 193 *pidp = procdesc_pid(fp); 194out: 195 fdrop(fp, td); 196 return (error); 197} 198 199/* 200 * System call to return the pid of a process given its process descriptor. 201 */ 202int 203sys_pdgetpid(struct thread *td, struct pdgetpid_args *uap) 204{ 205 cap_rights_t rights; 206 pid_t pid; 207 int error; 208 209 AUDIT_ARG_FD(uap->fd); 210 error = kern_pdgetpid(td, uap->fd, 211 cap_rights_init(&rights, CAP_PDGETPID), &pid); 212 if (error == 0) 213 error = copyout(&pid, uap->pidp, sizeof(pid)); 214 return (error); 215} 216 217/* 218 * When a new process is forked by pdfork(), a file descriptor is allocated 219 * by the fork code first, then the process is forked, and then we get a 220 * chance to set up the process descriptor. Failure is not permitted at this 221 * point, so procdesc_new() must succeed. 222 */ 223void 224procdesc_new(struct proc *p, int flags) 225{ 226 struct procdesc *pd; 227 228 pd = uma_zalloc(procdesc_zone, M_WAITOK | M_ZERO); 229 pd->pd_proc = p; 230 pd->pd_pid = p->p_pid; 231 p->p_procdesc = pd; 232 pd->pd_flags = 0; 233 if (flags & PD_DAEMON) 234 pd->pd_flags |= PDF_DAEMON; 235 PROCDESC_LOCK_INIT(pd); 236 knlist_init_mtx(&pd->pd_selinfo.si_note, &pd->pd_lock); 237 238 /* 239 * Process descriptors start out with two references: one from their 240 * struct file, and the other from their struct proc. 241 */ 242 refcount_init(&pd->pd_refcount, 2); 243} 244 245/* 246 * Create a new process decriptor for the process that refers to it. 247 */ 248int 249procdesc_falloc(struct thread *td, struct file **resultfp, int *resultfd, 250 int flags, struct filecaps *fcaps) 251{ 252 int fflags; 253 254 fflags = 0; 255 if (flags & PD_CLOEXEC) 256 fflags = O_CLOEXEC; 257 258 return (falloc_caps(td, resultfp, resultfd, fflags, fcaps)); 259} 260 261/* 262 * Initialize a file with a process descriptor. 263 */ 264void 265procdesc_finit(struct procdesc *pdp, struct file *fp) 266{ 267 268 finit(fp, FREAD | FWRITE, DTYPE_PROCDESC, pdp, &procdesc_ops); 269} 270 271static void 272procdesc_free(struct procdesc *pd) 273{ 274 275 /* 276 * When the last reference is released, we assert that the descriptor 277 * has been closed, but not that the process has exited, as we will 278 * detach the descriptor before the process dies if the descript is 279 * closed, as we can't wait synchronously. 280 */ 281 if (refcount_release(&pd->pd_refcount)) { 282 KASSERT(pd->pd_proc == NULL, 283 ("procdesc_free: pd_proc != NULL")); 284 KASSERT((pd->pd_flags & PDF_CLOSED), 285 ("procdesc_free: !PDF_CLOSED")); 286 287 knlist_destroy(&pd->pd_selinfo.si_note); 288 PROCDESC_LOCK_DESTROY(pd); 289 uma_zfree(procdesc_zone, pd); 290 } 291} 292 293/* 294 * procdesc_exit() - notify a process descriptor that its process is exiting. 295 * We use the proctree_lock to ensure that process exit either happens 296 * strictly before or strictly after a concurrent call to procdesc_close(). 297 */ 298int 299procdesc_exit(struct proc *p) 300{ 301 struct procdesc *pd; 302 303 sx_assert(&proctree_lock, SA_XLOCKED); 304 PROC_LOCK_ASSERT(p, MA_OWNED); 305 KASSERT(p->p_procdesc != NULL, ("procdesc_exit: p_procdesc NULL")); 306 307 pd = p->p_procdesc; 308 309 PROCDESC_LOCK(pd); 310 KASSERT((pd->pd_flags & PDF_CLOSED) == 0 || p->p_pptr == initproc, 311 ("procdesc_exit: closed && parent not init")); 312 313 pd->pd_flags |= PDF_EXITED; 314 pd->pd_xstat = KW_EXITCODE(p->p_xexit, p->p_xsig); 315 316 /* 317 * If the process descriptor has been closed, then we have nothing 318 * to do; return 1 so that init will get SIGCHLD and do the reaping. 319 * Clean up the procdesc now rather than letting it happen during 320 * that reap. 321 */ 322 if (pd->pd_flags & PDF_CLOSED) { 323 PROCDESC_UNLOCK(pd); 324 pd->pd_proc = NULL; 325 p->p_procdesc = NULL; 326 procdesc_free(pd); 327 return (1); 328 } 329 if (pd->pd_flags & PDF_SELECTED) { 330 pd->pd_flags &= ~PDF_SELECTED; 331 selwakeup(&pd->pd_selinfo); 332 } 333 KNOTE_LOCKED(&pd->pd_selinfo.si_note, NOTE_EXIT); 334 PROCDESC_UNLOCK(pd); 335 return (0); 336} 337 338/* 339 * When a process descriptor is reaped, perhaps as a result of close() or 340 * pdwait4(), release the process's reference on the process descriptor. 341 */ 342void 343procdesc_reap(struct proc *p) 344{ 345 struct procdesc *pd; 346 347 sx_assert(&proctree_lock, SA_XLOCKED); 348 KASSERT(p->p_procdesc != NULL, ("procdesc_reap: p_procdesc == NULL")); 349 350 pd = p->p_procdesc; 351 pd->pd_proc = NULL; 352 p->p_procdesc = NULL; 353 procdesc_free(pd); 354} 355 356/* 357 * procdesc_close() - last close on a process descriptor. If the process is 358 * still running, terminate with SIGKILL (unless PDF_DAEMON is set) and let 359 * init(8) clean up the mess; if not, we have to clean up the zombie ourselves. 360 */ 361static int 362procdesc_close(struct file *fp, struct thread *td) 363{ 364 struct procdesc *pd; 365 struct proc *p; 366 367 KASSERT(fp->f_type == DTYPE_PROCDESC, ("procdesc_close: !procdesc")); 368 369 pd = fp->f_data; 370 fp->f_ops = &badfileops; 371 fp->f_data = NULL; 372 373 sx_xlock(&proctree_lock); 374 PROCDESC_LOCK(pd); 375 pd->pd_flags |= PDF_CLOSED; 376 PROCDESC_UNLOCK(pd); 377 p = pd->pd_proc; 378 if (p == NULL) { 379 /* 380 * This is the case where process' exit status was already 381 * collected and procdesc_reap() was already called. 382 */ 383 sx_xunlock(&proctree_lock); 384 } else { 385 PROC_LOCK(p); 386 if (p->p_state == PRS_ZOMBIE) { 387 /* 388 * If the process is already dead and just awaiting 389 * reaping, do that now. This will release the 390 * process's reference to the process descriptor when it 391 * calls back into procdesc_reap(). 392 */ 393 PROC_SLOCK(p); 394 proc_reap(curthread, p, NULL, 0); 395 } else { 396 /* 397 * If the process is not yet dead, we need to kill it, 398 * but we can't wait around synchronously for it to go 399 * away, as that path leads to madness (and deadlocks). 400 * First, detach the process from its descriptor so that 401 * its exit status will be reported normally. 402 */ 403 pd->pd_proc = NULL; 404 p->p_procdesc = NULL; 405 procdesc_free(pd); 406 407 /* 408 * Next, reparent it to init(8) so that there's someone 409 * to pick up the pieces; finally, terminate with 410 * prejudice. 411 */ 412 p->p_sigparent = SIGCHLD; 413 proc_reparent(p, initproc); 414 if ((pd->pd_flags & PDF_DAEMON) == 0) 415 kern_psignal(p, SIGKILL); 416 PROC_UNLOCK(p); 417 sx_xunlock(&proctree_lock); 418 } 419 } 420 421 /* 422 * Release the file descriptor's reference on the process descriptor. 423 */ 424 procdesc_free(pd); 425 return (0); 426} 427 428static int 429procdesc_poll(struct file *fp, int events, struct ucred *active_cred, 430 struct thread *td) 431{ 432 struct procdesc *pd; 433 int revents; 434 435 revents = 0; 436 pd = fp->f_data; 437 PROCDESC_LOCK(pd); 438 if (pd->pd_flags & PDF_EXITED) 439 revents |= POLLHUP; 440 if (revents == 0) { 441 selrecord(td, &pd->pd_selinfo); 442 pd->pd_flags |= PDF_SELECTED; 443 } 444 PROCDESC_UNLOCK(pd); 445 return (revents); 446} 447 448static void 449procdesc_kqops_detach(struct knote *kn) 450{ 451 struct procdesc *pd; 452 453 pd = kn->kn_fp->f_data; 454 knlist_remove(&pd->pd_selinfo.si_note, kn, 0); 455} 456 457static int 458procdesc_kqops_event(struct knote *kn, long hint) 459{ 460 struct procdesc *pd; 461 u_int event; 462 463 pd = kn->kn_fp->f_data; 464 if (hint == 0) { 465 /* 466 * Initial test after registration. Generate a NOTE_EXIT in 467 * case the process already terminated before registration. 468 */ 469 event = pd->pd_flags & PDF_EXITED ? NOTE_EXIT : 0; 470 } else { 471 /* Mask off extra data. */ 472 event = (u_int)hint & NOTE_PCTRLMASK; 473 } 474 475 /* If the user is interested in this event, record it. */ 476 if (kn->kn_sfflags & event) 477 kn->kn_fflags |= event; 478 479 /* Process is gone, so flag the event as finished. */ 480 if (event == NOTE_EXIT) { 481 kn->kn_flags |= EV_EOF | EV_ONESHOT; 482 if (kn->kn_fflags & NOTE_EXIT) 483 kn->kn_data = pd->pd_xstat; 484 if (kn->kn_fflags == 0) 485 kn->kn_flags |= EV_DROP; 486 return (1); 487 } 488 489 return (kn->kn_fflags != 0); 490} 491 492static struct filterops procdesc_kqops = { 493 .f_isfd = 1, 494 .f_detach = procdesc_kqops_detach, 495 .f_event = procdesc_kqops_event, 496}; 497 498static int 499procdesc_kqfilter(struct file *fp, struct knote *kn) 500{ 501 struct procdesc *pd; 502 503 pd = fp->f_data; 504 switch (kn->kn_filter) { 505 case EVFILT_PROCDESC: 506 kn->kn_fop = &procdesc_kqops; 507 kn->kn_flags |= EV_CLEAR; 508 knlist_add(&pd->pd_selinfo.si_note, kn, 0); 509 return (0); 510 default: 511 return (EINVAL); 512 } 513} 514 515static int 516procdesc_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 517 struct thread *td) 518{ 519 struct procdesc *pd; 520 struct timeval pstart; 521 522 /* 523 * XXXRW: Perhaps we should cache some more information from the 524 * process so that we can return it reliably here even after it has 525 * died. For example, caching its credential data. 526 */ 527 bzero(sb, sizeof(*sb)); 528 pd = fp->f_data; 529 sx_slock(&proctree_lock); 530 if (pd->pd_proc != NULL) { 531 PROC_LOCK(pd->pd_proc); 532 533 /* Set birth and [acm] times to process start time. */ 534 pstart = pd->pd_proc->p_stats->p_start; 535 timevaladd(&pstart, &boottime); 536 TIMEVAL_TO_TIMESPEC(&pstart, &sb->st_birthtim); 537 sb->st_atim = sb->st_birthtim; 538 sb->st_ctim = sb->st_birthtim; 539 sb->st_mtim = sb->st_birthtim; 540 if (pd->pd_proc->p_state != PRS_ZOMBIE) 541 sb->st_mode = S_IFREG | S_IRWXU; 542 else 543 sb->st_mode = S_IFREG; 544 sb->st_uid = pd->pd_proc->p_ucred->cr_ruid; 545 sb->st_gid = pd->pd_proc->p_ucred->cr_rgid; 546 PROC_UNLOCK(pd->pd_proc); 547 } else 548 sb->st_mode = S_IFREG; 549 sx_sunlock(&proctree_lock); 550 return (0); 551} 552 553static int 554procdesc_fill_kinfo(struct file *fp, struct kinfo_file *kif, 555 struct filedesc *fdp) 556{ 557 struct procdesc *pdp; 558 559 kif->kf_type = KF_TYPE_PROCDESC; 560 pdp = fp->f_data; 561 kif->kf_un.kf_proc.kf_pid = pdp->pd_pid; 562 return (0); 563} 564