1 /* 2 * Copyright (c) 1982, 1986, 1989, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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 #include <sys/param.h> 31 #include <sys/systm.h> 32 #include <sys/kernel.h> 33 #include <sys/sysctl.h> 34 #include <sys/malloc.h> 35 #include <sys/proc.h> 36 #include <sys/vnode.h> 37 #include <sys/jail.h> 38 #include <sys/filedesc.h> 39 #include <sys/tty.h> 40 #include <sys/dsched.h> 41 #include <sys/signalvar.h> 42 #include <sys/spinlock.h> 43 #include <sys/random.h> 44 #include <sys/vnode.h> 45 #include <sys/exec.h> 46 #include <vm/vm.h> 47 #include <sys/lock.h> 48 #include <vm/pmap.h> 49 #include <vm/vm_map.h> 50 #include <sys/user.h> 51 #include <machine/smp.h> 52 53 #include <sys/refcount.h> 54 #include <sys/spinlock2.h> 55 56 /* 57 * Hash table size must be a power of two and is not currently dynamically 58 * sized. There is a trade-off between the linear scans which must iterate 59 * all HSIZE elements and the number of elements which might accumulate 60 * within each hash chain. 61 */ 62 #define ALLPROC_HSIZE 256 63 #define ALLPROC_HMASK (ALLPROC_HSIZE - 1) 64 #define ALLPROC_HASH(pid) (pid & ALLPROC_HMASK) 65 #define PGRP_HASH(pid) (pid & ALLPROC_HMASK) 66 #define SESS_HASH(pid) (pid & ALLPROC_HMASK) 67 68 /* 69 * pid_doms[] management, used to control how quickly a PID can be recycled. 70 * Must be a multiple of ALLPROC_HSIZE for the proc_makepid() inner loops. 71 * 72 * WARNING! PIDDOM_DELAY should not be defined > 20 or so unless you change 73 * the array from int8_t's to int16_t's. 74 */ 75 #define PIDDOM_COUNT 10 /* 10 pids per domain - reduce array size */ 76 #define PIDDOM_DELAY 10 /* min 10 seconds after exit before reuse */ 77 #define PIDDOM_SCALE 10 /* (10,000*SCALE)/sec performance guarantee */ 78 #define PIDSEL_DOMAINS (PID_MAX * PIDDOM_SCALE / PIDDOM_COUNT / \ 79 ALLPROC_HSIZE * ALLPROC_HSIZE) 80 81 /* Used by libkvm */ 82 int allproc_hsize = ALLPROC_HSIZE; 83 84 LIST_HEAD(pidhashhead, proc); 85 86 static MALLOC_DEFINE(M_PGRP, "pgrp", "process group header"); 87 MALLOC_DEFINE(M_SESSION, "session", "session header"); 88 MALLOC_DEFINE(M_PROC, "proc", "Proc structures"); 89 MALLOC_DEFINE(M_LWP, "lwp", "lwp structures"); 90 MALLOC_DEFINE(M_SUBPROC, "subproc", "Proc sub-structures"); 91 92 int ps_showallprocs = 1; 93 static int ps_showallthreads = 1; 94 SYSCTL_INT(_security, OID_AUTO, ps_showallprocs, CTLFLAG_RW, 95 &ps_showallprocs, 0, 96 "Unprivileged processes can see processes with different UID/GID"); 97 SYSCTL_INT(_security, OID_AUTO, ps_showallthreads, CTLFLAG_RW, 98 &ps_showallthreads, 0, 99 "Unprivileged processes can see kernel threads"); 100 static u_int pid_domain_skips; 101 SYSCTL_UINT(_kern, OID_AUTO, pid_domain_skips, CTLFLAG_RW, 102 &pid_domain_skips, 0, 103 "Number of pid_doms[] skipped"); 104 static u_int pid_inner_skips; 105 SYSCTL_UINT(_kern, OID_AUTO, pid_inner_skips, CTLFLAG_RW, 106 &pid_inner_skips, 0, 107 "Number of pid_doms[] skipped"); 108 109 static void orphanpg(struct pgrp *pg); 110 static void proc_makepid(struct proc *p, int random_offset); 111 112 /* 113 * Process related lists (for proc_token, allproc, allpgrp, and allsess) 114 */ 115 typedef struct procglob procglob_t; 116 117 static procglob_t procglob[ALLPROC_HSIZE]; 118 119 /* 120 * We try our best to avoid recycling a PID too quickly. We do this by 121 * storing (uint8_t)time_second in the related pid domain on-reap and then 122 * using that to skip-over the domain on-allocate. 123 * 124 * This array has to be fairly large to support a high fork/exec rate. 125 * A ~100,000 entry array will support a 10-second reuse latency at 126 * 10,000 execs/second, worst case. Best-case multiply by PIDDOM_COUNT 127 * (approximately 100,000 execs/second). 128 * 129 * Currently we allocate around a megabyte, making the worst-case fork 130 * rate around 100,000/second. 131 */ 132 static uint8_t *pid_doms; 133 134 /* 135 * Random component to nextpid generation. We mix in a random factor to make 136 * it a little harder to predict. We sanity check the modulus value to avoid 137 * doing it in critical paths. Don't let it be too small or we pointlessly 138 * waste randomness entropy, and don't let it be impossibly large. Using a 139 * modulus that is too big causes a LOT more process table scans and slows 140 * down fork processing as the pidchecked caching is defeated. 141 */ 142 static int randompid = 0; 143 144 static __inline 145 struct ucred * 146 pcredcache(struct ucred *cr, struct proc *p) 147 { 148 if (cr != p->p_ucred) { 149 if (cr) 150 crfree(cr); 151 spin_lock(&p->p_spin); 152 if ((cr = p->p_ucred) != NULL) 153 crhold(cr); 154 spin_unlock(&p->p_spin); 155 } 156 return cr; 157 } 158 159 /* 160 * No requirements. 161 */ 162 static int 163 sysctl_kern_randompid(SYSCTL_HANDLER_ARGS) 164 { 165 int error, pid; 166 167 pid = randompid; 168 error = sysctl_handle_int(oidp, &pid, 0, req); 169 if (error || !req->newptr) 170 return (error); 171 if (pid < 0 || pid > PID_MAX - 100) /* out of range */ 172 pid = PID_MAX - 100; 173 else if (pid < 2) /* NOP */ 174 pid = 0; 175 else if (pid < 100) /* Make it reasonable */ 176 pid = 100; 177 randompid = pid; 178 return (error); 179 } 180 181 SYSCTL_PROC(_kern, OID_AUTO, randompid, CTLTYPE_INT|CTLFLAG_RW, 182 0, 0, sysctl_kern_randompid, "I", "Random PID modulus"); 183 184 /* 185 * Initialize global process hashing structures. 186 * 187 * These functions are ONLY called from the low level boot code and do 188 * not lock their operations. 189 */ 190 void 191 procinit(void) 192 { 193 u_long i; 194 195 /* 196 * Allocate dynamically. This array can be large (~1MB) so don't 197 * waste boot loader space. 198 */ 199 pid_doms = kmalloc(sizeof(pid_doms[0]) * PIDSEL_DOMAINS, 200 M_PROC, M_WAITOK | M_ZERO); 201 202 /* 203 * Avoid unnecessary stalls due to pid_doms[] values all being 204 * the same. Make sure that the allocation of pid 1 and pid 2 205 * succeeds. 206 */ 207 for (i = 0; i < PIDSEL_DOMAINS; ++i) 208 pid_doms[i] = (int8_t)i - (int8_t)(PIDDOM_DELAY + 1); 209 210 /* 211 * Other misc init. 212 */ 213 for (i = 0; i < ALLPROC_HSIZE; ++i) { 214 procglob_t *prg = &procglob[i]; 215 LIST_INIT(&prg->allproc); 216 LIST_INIT(&prg->allsess); 217 LIST_INIT(&prg->allpgrp); 218 lwkt_token_init(&prg->proc_token, "allproc"); 219 } 220 uihashinit(); 221 } 222 223 void 224 procinsertinit(struct proc *p) 225 { 226 LIST_INSERT_HEAD(&procglob[ALLPROC_HASH(p->p_pid)].allproc, 227 p, p_list); 228 } 229 230 void 231 pgrpinsertinit(struct pgrp *pg) 232 { 233 LIST_INSERT_HEAD(&procglob[ALLPROC_HASH(pg->pg_id)].allpgrp, 234 pg, pg_list); 235 } 236 237 void 238 sessinsertinit(struct session *sess) 239 { 240 LIST_INSERT_HEAD(&procglob[ALLPROC_HASH(sess->s_sid)].allsess, 241 sess, s_list); 242 } 243 244 /* 245 * Process hold/release support functions. Called via the PHOLD(), 246 * PRELE(), and PSTALL() macros. 247 * 248 * p->p_lock is a simple hold count with a waiting interlock. No wakeup() 249 * is issued unless someone is actually waiting for the process. 250 * 251 * Most holds are short-term, allowing a process scan or other similar 252 * operation to access a proc structure without it getting ripped out from 253 * under us. procfs and process-list sysctl ops also use the hold function 254 * interlocked with various p_flags to keep the vmspace intact when reading 255 * or writing a user process's address space. 256 * 257 * There are two situations where a hold count can be longer. Exiting lwps 258 * hold the process until the lwp is reaped, and the parent will hold the 259 * child during vfork()/exec() sequences while the child is marked P_PPWAIT. 260 * 261 * The kernel waits for the hold count to drop to 0 (or 1 in some cases) at 262 * various critical points in the fork/exec and exit paths before proceeding. 263 */ 264 #define PLOCK_ZOMB 0x20000000 265 #define PLOCK_WAITING 0x40000000 266 #define PLOCK_MASK 0x1FFFFFFF 267 268 void 269 pstall(struct proc *p, const char *wmesg, int count) 270 { 271 int o; 272 int n; 273 274 for (;;) { 275 o = p->p_lock; 276 cpu_ccfence(); 277 if ((o & PLOCK_MASK) <= count) 278 break; 279 n = o | PLOCK_WAITING; 280 tsleep_interlock(&p->p_lock, 0); 281 282 /* 283 * If someone is trying to single-step the process during 284 * an exec or an exit they can deadlock us because procfs 285 * sleeps with the process held. 286 */ 287 if (p->p_stops) { 288 if (p->p_flags & P_INEXEC) { 289 wakeup(&p->p_stype); 290 } else if (p->p_flags & P_POSTEXIT) { 291 spin_lock(&p->p_spin); 292 p->p_stops = 0; 293 p->p_step = 0; 294 spin_unlock(&p->p_spin); 295 wakeup(&p->p_stype); 296 } 297 } 298 299 if (atomic_cmpset_int(&p->p_lock, o, n)) { 300 tsleep(&p->p_lock, PINTERLOCKED, wmesg, 0); 301 } 302 } 303 } 304 305 void 306 phold(struct proc *p) 307 { 308 atomic_add_int(&p->p_lock, 1); 309 } 310 311 /* 312 * WARNING! On last release (p) can become instantly invalid due to 313 * MP races. 314 */ 315 void 316 prele(struct proc *p) 317 { 318 int o; 319 int n; 320 321 /* 322 * Fast path 323 */ 324 if (atomic_cmpset_int(&p->p_lock, 1, 0)) 325 return; 326 327 /* 328 * Slow path 329 */ 330 for (;;) { 331 o = p->p_lock; 332 KKASSERT((o & PLOCK_MASK) > 0); 333 cpu_ccfence(); 334 n = (o - 1) & ~PLOCK_WAITING; 335 if (atomic_cmpset_int(&p->p_lock, o, n)) { 336 if (o & PLOCK_WAITING) 337 wakeup(&p->p_lock); 338 break; 339 } 340 } 341 } 342 343 /* 344 * Hold and flag serialized for zombie reaping purposes. 345 * 346 * This function will fail if it has to block, returning non-zero with 347 * neither the flag set or the hold count bumped. Note that (p) may 348 * not be valid in this case if the caller does not have some other 349 * reference on (p). 350 * 351 * This function does not block on other PHOLD()s, only on other 352 * PHOLDZOMB()s. 353 * 354 * Zero is returned on success. The hold count will be incremented and 355 * the serialization flag acquired. Note that serialization is only against 356 * other pholdzomb() calls, not against phold() calls. 357 */ 358 int 359 pholdzomb(struct proc *p) 360 { 361 int o; 362 int n; 363 364 /* 365 * Fast path 366 */ 367 if (atomic_cmpset_int(&p->p_lock, 0, PLOCK_ZOMB | 1)) 368 return(0); 369 370 /* 371 * Slow path 372 */ 373 for (;;) { 374 o = p->p_lock; 375 cpu_ccfence(); 376 if ((o & PLOCK_ZOMB) == 0) { 377 n = (o + 1) | PLOCK_ZOMB; 378 if (atomic_cmpset_int(&p->p_lock, o, n)) 379 return(0); 380 } else { 381 KKASSERT((o & PLOCK_MASK) > 0); 382 n = o | PLOCK_WAITING; 383 tsleep_interlock(&p->p_lock, 0); 384 if (atomic_cmpset_int(&p->p_lock, o, n)) { 385 tsleep(&p->p_lock, PINTERLOCKED, "phldz", 0); 386 /* (p) can be ripped out at this point */ 387 return(1); 388 } 389 } 390 } 391 } 392 393 /* 394 * Release PLOCK_ZOMB and the hold count, waking up any waiters. 395 * 396 * WARNING! On last release (p) can become instantly invalid due to 397 * MP races. 398 */ 399 void 400 prelezomb(struct proc *p) 401 { 402 int o; 403 int n; 404 405 /* 406 * Fast path 407 */ 408 if (atomic_cmpset_int(&p->p_lock, PLOCK_ZOMB | 1, 0)) 409 return; 410 411 /* 412 * Slow path 413 */ 414 KKASSERT(p->p_lock & PLOCK_ZOMB); 415 for (;;) { 416 o = p->p_lock; 417 KKASSERT((o & PLOCK_MASK) > 0); 418 cpu_ccfence(); 419 n = (o - 1) & ~(PLOCK_ZOMB | PLOCK_WAITING); 420 if (atomic_cmpset_int(&p->p_lock, o, n)) { 421 if (o & PLOCK_WAITING) 422 wakeup(&p->p_lock); 423 break; 424 } 425 } 426 } 427 428 /* 429 * Is p an inferior of the current process? 430 * 431 * No requirements. 432 */ 433 int 434 inferior(struct proc *p) 435 { 436 struct proc *p2; 437 438 PHOLD(p); 439 lwkt_gettoken_shared(&p->p_token); 440 while (p != curproc) { 441 if (p->p_pid == 0) { 442 lwkt_reltoken(&p->p_token); 443 return (0); 444 } 445 p2 = p->p_pptr; 446 PHOLD(p2); 447 lwkt_reltoken(&p->p_token); 448 PRELE(p); 449 lwkt_gettoken_shared(&p2->p_token); 450 p = p2; 451 } 452 lwkt_reltoken(&p->p_token); 453 PRELE(p); 454 455 return (1); 456 } 457 458 /* 459 * Locate a process by number. The returned process will be referenced and 460 * must be released with PRELE(). 461 * 462 * No requirements. 463 */ 464 struct proc * 465 pfind(pid_t pid) 466 { 467 struct proc *p = curproc; 468 procglob_t *prg; 469 int n; 470 471 /* 472 * Shortcut the current process 473 */ 474 if (p && p->p_pid == pid) { 475 PHOLD(p); 476 return (p); 477 } 478 479 /* 480 * Otherwise find it in the hash table. 481 */ 482 n = ALLPROC_HASH(pid); 483 prg = &procglob[n]; 484 485 lwkt_gettoken_shared(&prg->proc_token); 486 LIST_FOREACH(p, &prg->allproc, p_list) { 487 if (p->p_stat == SZOMB) 488 continue; 489 if (p->p_pid == pid) { 490 PHOLD(p); 491 lwkt_reltoken(&prg->proc_token); 492 return (p); 493 } 494 } 495 lwkt_reltoken(&prg->proc_token); 496 497 return (NULL); 498 } 499 500 /* 501 * Locate a process by number. The returned process is NOT referenced. 502 * The result will not be stable and is typically only used to validate 503 * against a process that the caller has in-hand. 504 * 505 * No requirements. 506 */ 507 struct proc * 508 pfindn(pid_t pid) 509 { 510 struct proc *p = curproc; 511 procglob_t *prg; 512 int n; 513 514 /* 515 * Shortcut the current process 516 */ 517 if (p && p->p_pid == pid) 518 return (p); 519 520 /* 521 * Otherwise find it in the hash table. 522 */ 523 n = ALLPROC_HASH(pid); 524 prg = &procglob[n]; 525 526 lwkt_gettoken_shared(&prg->proc_token); 527 LIST_FOREACH(p, &prg->allproc, p_list) { 528 if (p->p_stat == SZOMB) 529 continue; 530 if (p->p_pid == pid) { 531 lwkt_reltoken(&prg->proc_token); 532 return (p); 533 } 534 } 535 lwkt_reltoken(&prg->proc_token); 536 537 return (NULL); 538 } 539 540 /* 541 * Locate a process on the zombie list. Return a process or NULL. 542 * The returned process will be referenced and the caller must release 543 * it with PRELE(). 544 * 545 * No other requirements. 546 */ 547 struct proc * 548 zpfind(pid_t pid) 549 { 550 struct proc *p = curproc; 551 procglob_t *prg; 552 int n; 553 554 /* 555 * Shortcut the current process 556 */ 557 if (p && p->p_pid == pid) { 558 PHOLD(p); 559 return (p); 560 } 561 562 /* 563 * Otherwise find it in the hash table. 564 */ 565 n = ALLPROC_HASH(pid); 566 prg = &procglob[n]; 567 568 lwkt_gettoken_shared(&prg->proc_token); 569 LIST_FOREACH(p, &prg->allproc, p_list) { 570 if (p->p_stat != SZOMB) 571 continue; 572 if (p->p_pid == pid) { 573 PHOLD(p); 574 lwkt_reltoken(&prg->proc_token); 575 return (p); 576 } 577 } 578 lwkt_reltoken(&prg->proc_token); 579 580 return (NULL); 581 } 582 583 584 void 585 pgref(struct pgrp *pgrp) 586 { 587 refcount_acquire(&pgrp->pg_refs); 588 } 589 590 void 591 pgrel(struct pgrp *pgrp) 592 { 593 procglob_t *prg; 594 int count; 595 int n; 596 597 n = PGRP_HASH(pgrp->pg_id); 598 prg = &procglob[n]; 599 600 for (;;) { 601 count = pgrp->pg_refs; 602 cpu_ccfence(); 603 KKASSERT(count > 0); 604 if (count == 1) { 605 lwkt_gettoken(&prg->proc_token); 606 if (atomic_cmpset_int(&pgrp->pg_refs, 1, 0)) 607 break; 608 lwkt_reltoken(&prg->proc_token); 609 /* retry */ 610 } else { 611 if (atomic_cmpset_int(&pgrp->pg_refs, count, count - 1)) 612 return; 613 /* retry */ 614 } 615 } 616 617 /* 618 * Successful 1->0 transition, pghash_spin is held. 619 */ 620 LIST_REMOVE(pgrp, pg_list); 621 if (pid_doms[pgrp->pg_id % PIDSEL_DOMAINS] != (uint8_t)time_second) 622 pid_doms[pgrp->pg_id % PIDSEL_DOMAINS] = (uint8_t)time_second; 623 624 /* 625 * Reset any sigio structures pointing to us as a result of 626 * F_SETOWN with our pgid. 627 */ 628 funsetownlst(&pgrp->pg_sigiolst); 629 630 if (pgrp->pg_session->s_ttyp != NULL && 631 pgrp->pg_session->s_ttyp->t_pgrp == pgrp) { 632 pgrp->pg_session->s_ttyp->t_pgrp = NULL; 633 } 634 lwkt_reltoken(&prg->proc_token); 635 636 sess_rele(pgrp->pg_session); 637 kfree(pgrp, M_PGRP); 638 } 639 640 /* 641 * Locate a process group by number. The returned process group will be 642 * referenced w/pgref() and must be released with pgrel() (or assigned 643 * somewhere if you wish to keep the reference). 644 * 645 * No requirements. 646 */ 647 struct pgrp * 648 pgfind(pid_t pgid) 649 { 650 struct pgrp *pgrp; 651 procglob_t *prg; 652 int n; 653 654 n = PGRP_HASH(pgid); 655 prg = &procglob[n]; 656 lwkt_gettoken_shared(&prg->proc_token); 657 658 LIST_FOREACH(pgrp, &prg->allpgrp, pg_list) { 659 if (pgrp->pg_id == pgid) { 660 refcount_acquire(&pgrp->pg_refs); 661 lwkt_reltoken(&prg->proc_token); 662 return (pgrp); 663 } 664 } 665 lwkt_reltoken(&prg->proc_token); 666 return (NULL); 667 } 668 669 /* 670 * Move p to a new or existing process group (and session) 671 * 672 * No requirements. 673 */ 674 int 675 enterpgrp(struct proc *p, pid_t pgid, int mksess) 676 { 677 struct pgrp *pgrp; 678 struct pgrp *opgrp; 679 int error; 680 681 pgrp = pgfind(pgid); 682 683 KASSERT(pgrp == NULL || !mksess, 684 ("enterpgrp: setsid into non-empty pgrp")); 685 KASSERT(!SESS_LEADER(p), 686 ("enterpgrp: session leader attempted setpgrp")); 687 688 if (pgrp == NULL) { 689 pid_t savepid = p->p_pid; 690 struct proc *np; 691 procglob_t *prg; 692 int n; 693 694 /* 695 * new process group 696 */ 697 KASSERT(p->p_pid == pgid, 698 ("enterpgrp: new pgrp and pid != pgid")); 699 pgrp = kmalloc(sizeof(struct pgrp), M_PGRP, M_WAITOK | M_ZERO); 700 pgrp->pg_id = pgid; 701 LIST_INIT(&pgrp->pg_members); 702 pgrp->pg_jobc = 0; 703 SLIST_INIT(&pgrp->pg_sigiolst); 704 lwkt_token_init(&pgrp->pg_token, "pgrp_token"); 705 refcount_init(&pgrp->pg_refs, 1); 706 lockinit(&pgrp->pg_lock, "pgwt", 0, 0); 707 708 n = PGRP_HASH(pgid); 709 prg = &procglob[n]; 710 711 if ((np = pfindn(savepid)) == NULL || np != p) { 712 lwkt_reltoken(&prg->proc_token); 713 error = ESRCH; 714 kfree(pgrp, M_PGRP); 715 goto fatal; 716 } 717 718 lwkt_gettoken(&prg->proc_token); 719 if (mksess) { 720 struct session *sess; 721 722 /* 723 * new session 724 */ 725 sess = kmalloc(sizeof(struct session), M_SESSION, 726 M_WAITOK | M_ZERO); 727 lwkt_gettoken(&p->p_token); 728 sess->s_prg = prg; 729 sess->s_leader = p; 730 sess->s_sid = p->p_pid; 731 sess->s_count = 1; 732 sess->s_ttyvp = NULL; 733 sess->s_ttyp = NULL; 734 bcopy(p->p_session->s_login, sess->s_login, 735 sizeof(sess->s_login)); 736 pgrp->pg_session = sess; 737 KASSERT(p == curproc, 738 ("enterpgrp: mksession and p != curproc")); 739 p->p_flags &= ~P_CONTROLT; 740 LIST_INSERT_HEAD(&prg->allsess, sess, s_list); 741 lwkt_reltoken(&p->p_token); 742 } else { 743 lwkt_gettoken(&p->p_token); 744 pgrp->pg_session = p->p_session; 745 sess_hold(pgrp->pg_session); 746 lwkt_reltoken(&p->p_token); 747 } 748 LIST_INSERT_HEAD(&prg->allpgrp, pgrp, pg_list); 749 750 lwkt_reltoken(&prg->proc_token); 751 } else if (pgrp == p->p_pgrp) { 752 pgrel(pgrp); 753 goto done; 754 } /* else pgfind() referenced the pgrp */ 755 756 lwkt_gettoken(&pgrp->pg_token); 757 lwkt_gettoken(&p->p_token); 758 759 /* 760 * Replace p->p_pgrp, handling any races that occur. 761 */ 762 while ((opgrp = p->p_pgrp) != NULL) { 763 pgref(opgrp); 764 lwkt_gettoken(&opgrp->pg_token); 765 if (opgrp != p->p_pgrp) { 766 lwkt_reltoken(&opgrp->pg_token); 767 pgrel(opgrp); 768 continue; 769 } 770 LIST_REMOVE(p, p_pglist); 771 break; 772 } 773 p->p_pgrp = pgrp; 774 LIST_INSERT_HEAD(&pgrp->pg_members, p, p_pglist); 775 776 /* 777 * Adjust eligibility of affected pgrps to participate in job control. 778 * Increment eligibility counts before decrementing, otherwise we 779 * could reach 0 spuriously during the first call. 780 */ 781 fixjobc(p, pgrp, 1); 782 if (opgrp) { 783 fixjobc(p, opgrp, 0); 784 lwkt_reltoken(&opgrp->pg_token); 785 pgrel(opgrp); /* manual pgref */ 786 pgrel(opgrp); /* p->p_pgrp ref */ 787 } 788 lwkt_reltoken(&p->p_token); 789 lwkt_reltoken(&pgrp->pg_token); 790 done: 791 error = 0; 792 fatal: 793 return (error); 794 } 795 796 /* 797 * Remove process from process group 798 * 799 * No requirements. 800 */ 801 int 802 leavepgrp(struct proc *p) 803 { 804 struct pgrp *pg = p->p_pgrp; 805 806 lwkt_gettoken(&p->p_token); 807 while ((pg = p->p_pgrp) != NULL) { 808 pgref(pg); 809 lwkt_gettoken(&pg->pg_token); 810 if (p->p_pgrp != pg) { 811 lwkt_reltoken(&pg->pg_token); 812 pgrel(pg); 813 continue; 814 } 815 p->p_pgrp = NULL; 816 LIST_REMOVE(p, p_pglist); 817 lwkt_reltoken(&pg->pg_token); 818 pgrel(pg); /* manual pgref */ 819 pgrel(pg); /* p->p_pgrp ref */ 820 break; 821 } 822 lwkt_reltoken(&p->p_token); 823 824 return (0); 825 } 826 827 /* 828 * Adjust the ref count on a session structure. When the ref count falls to 829 * zero the tty is disassociated from the session and the session structure 830 * is freed. Note that tty assocation is not itself ref-counted. 831 * 832 * No requirements. 833 */ 834 void 835 sess_hold(struct session *sp) 836 { 837 atomic_add_int(&sp->s_count, 1); 838 } 839 840 /* 841 * No requirements. 842 */ 843 void 844 sess_rele(struct session *sess) 845 { 846 procglob_t *prg; 847 struct tty *tp; 848 int count; 849 int n; 850 851 n = SESS_HASH(sess->s_sid); 852 prg = &procglob[n]; 853 854 for (;;) { 855 count = sess->s_count; 856 cpu_ccfence(); 857 KKASSERT(count > 0); 858 if (count == 1) { 859 lwkt_gettoken(&prg->proc_token); 860 if (atomic_cmpset_int(&sess->s_count, 1, 0)) 861 break; 862 lwkt_reltoken(&prg->proc_token); 863 /* retry */ 864 } else { 865 if (atomic_cmpset_int(&sess->s_count, count, count - 1)) 866 return; 867 /* retry */ 868 } 869 } 870 871 /* 872 * Successful 1->0 transition and prg->proc_token is held. 873 */ 874 LIST_REMOVE(sess, s_list); 875 if (pid_doms[sess->s_sid % PIDSEL_DOMAINS] != (uint8_t)time_second) 876 pid_doms[sess->s_sid % PIDSEL_DOMAINS] = (uint8_t)time_second; 877 878 if (sess->s_ttyp && sess->s_ttyp->t_session) { 879 #ifdef TTY_DO_FULL_CLOSE 880 /* FULL CLOSE, see ttyclearsession() */ 881 KKASSERT(sess->s_ttyp->t_session == sess); 882 sess->s_ttyp->t_session = NULL; 883 #else 884 /* HALF CLOSE, see ttyclearsession() */ 885 if (sess->s_ttyp->t_session == sess) 886 sess->s_ttyp->t_session = NULL; 887 #endif 888 } 889 if ((tp = sess->s_ttyp) != NULL) { 890 sess->s_ttyp = NULL; 891 ttyunhold(tp); 892 } 893 lwkt_reltoken(&prg->proc_token); 894 895 kfree(sess, M_SESSION); 896 } 897 898 /* 899 * Adjust pgrp jobc counters when specified process changes process group. 900 * We count the number of processes in each process group that "qualify" 901 * the group for terminal job control (those with a parent in a different 902 * process group of the same session). If that count reaches zero, the 903 * process group becomes orphaned. Check both the specified process' 904 * process group and that of its children. 905 * entering == 0 => p is leaving specified group. 906 * entering == 1 => p is entering specified group. 907 * 908 * No requirements. 909 */ 910 void 911 fixjobc(struct proc *p, struct pgrp *pgrp, int entering) 912 { 913 struct pgrp *hispgrp; 914 struct session *mysession; 915 struct proc *np; 916 917 /* 918 * Check p's parent to see whether p qualifies its own process 919 * group; if so, adjust count for p's process group. 920 */ 921 lwkt_gettoken(&p->p_token); /* p_children scan */ 922 lwkt_gettoken(&pgrp->pg_token); 923 924 mysession = pgrp->pg_session; 925 if ((hispgrp = p->p_pptr->p_pgrp) != pgrp && 926 hispgrp->pg_session == mysession) { 927 if (entering) 928 pgrp->pg_jobc++; 929 else if (--pgrp->pg_jobc == 0) 930 orphanpg(pgrp); 931 } 932 933 /* 934 * Check this process' children to see whether they qualify 935 * their process groups; if so, adjust counts for children's 936 * process groups. 937 */ 938 LIST_FOREACH(np, &p->p_children, p_sibling) { 939 PHOLD(np); 940 lwkt_gettoken(&np->p_token); 941 if ((hispgrp = np->p_pgrp) != pgrp && 942 hispgrp->pg_session == mysession && 943 np->p_stat != SZOMB) { 944 pgref(hispgrp); 945 lwkt_gettoken(&hispgrp->pg_token); 946 if (entering) 947 hispgrp->pg_jobc++; 948 else if (--hispgrp->pg_jobc == 0) 949 orphanpg(hispgrp); 950 lwkt_reltoken(&hispgrp->pg_token); 951 pgrel(hispgrp); 952 } 953 lwkt_reltoken(&np->p_token); 954 PRELE(np); 955 } 956 KKASSERT(pgrp->pg_refs > 0); 957 lwkt_reltoken(&pgrp->pg_token); 958 lwkt_reltoken(&p->p_token); 959 } 960 961 /* 962 * A process group has become orphaned; 963 * if there are any stopped processes in the group, 964 * hang-up all process in that group. 965 * 966 * The caller must hold pg_token. 967 */ 968 static void 969 orphanpg(struct pgrp *pg) 970 { 971 struct proc *p; 972 973 LIST_FOREACH(p, &pg->pg_members, p_pglist) { 974 if (p->p_stat == SSTOP) { 975 LIST_FOREACH(p, &pg->pg_members, p_pglist) { 976 ksignal(p, SIGHUP); 977 ksignal(p, SIGCONT); 978 } 979 return; 980 } 981 } 982 } 983 984 /* 985 * Add a new process to the allproc list and the PID hash. This 986 * also assigns a pid to the new process. 987 * 988 * No requirements. 989 */ 990 void 991 proc_add_allproc(struct proc *p) 992 { 993 int random_offset; 994 995 if ((random_offset = randompid) != 0) { 996 read_random(&random_offset, sizeof(random_offset)); 997 random_offset = (random_offset & 0x7FFFFFFF) % randompid; 998 } 999 proc_makepid(p, random_offset); 1000 } 1001 1002 /* 1003 * Calculate a new process pid. This function is integrated into 1004 * proc_add_allproc() to guarentee that the new pid is not reused before 1005 * the new process can be added to the allproc list. 1006 * 1007 * p_pid is assigned and the process is added to the allproc hash table 1008 * 1009 * WARNING! We need to allocate PIDs sequentially during early boot. 1010 * In particular, init needs to have a pid of 1. 1011 */ 1012 static 1013 void 1014 proc_makepid(struct proc *p, int random_offset) 1015 { 1016 static pid_t nextpid = 1; /* heuristic, allowed to race */ 1017 procglob_t *prg; 1018 struct pgrp *pg; 1019 struct proc *ps; 1020 struct session *sess; 1021 pid_t base; 1022 int8_t delta8; 1023 int retries; 1024 int n; 1025 1026 /* 1027 * Select the next pid base candidate. 1028 * 1029 * Check cyclement, do not allow a pid < 100. 1030 */ 1031 retries = 0; 1032 retry: 1033 base = atomic_fetchadd_int(&nextpid, 1) + random_offset; 1034 if (base <= 0 || base >= PID_MAX) { 1035 base = base % PID_MAX; 1036 if (base < 0) 1037 base = 100; 1038 if (base < 100) 1039 base += 100; 1040 nextpid = base; /* reset (SMP race ok) */ 1041 } 1042 1043 /* 1044 * Do not allow a base pid to be selected from a domain that has 1045 * recently seen a pid/pgid/sessid reap. Sleep a little if we looped 1046 * through all available domains. 1047 * 1048 * WARNING: We want the early pids to be allocated linearly, 1049 * particularly pid 1 and pid 2. 1050 */ 1051 if (++retries >= PIDSEL_DOMAINS) 1052 tsleep(&nextpid, 0, "makepid", 1); 1053 if (base >= 100) { 1054 delta8 = (int8_t)time_second - 1055 (int8_t)pid_doms[base % PIDSEL_DOMAINS]; 1056 if (delta8 >= 0 && delta8 <= PIDDOM_DELAY) { 1057 ++pid_domain_skips; 1058 goto retry; 1059 } 1060 } 1061 1062 /* 1063 * Calculate a hash index and find an unused process id within 1064 * the table, looping if we cannot find one. 1065 * 1066 * The inner loop increments by ALLPROC_HSIZE which keeps the 1067 * PID at the same pid_doms[] index as well as the same hash index. 1068 */ 1069 n = ALLPROC_HASH(base); 1070 prg = &procglob[n]; 1071 lwkt_gettoken(&prg->proc_token); 1072 1073 restart1: 1074 LIST_FOREACH(ps, &prg->allproc, p_list) { 1075 if (ps->p_pid == base) { 1076 base += ALLPROC_HSIZE; 1077 if (base >= PID_MAX) { 1078 lwkt_reltoken(&prg->proc_token); 1079 goto retry; 1080 } 1081 ++pid_inner_skips; 1082 goto restart1; 1083 } 1084 } 1085 LIST_FOREACH(pg, &prg->allpgrp, pg_list) { 1086 if (pg->pg_id == base) { 1087 base += ALLPROC_HSIZE; 1088 if (base >= PID_MAX) { 1089 lwkt_reltoken(&prg->proc_token); 1090 goto retry; 1091 } 1092 ++pid_inner_skips; 1093 goto restart1; 1094 } 1095 } 1096 LIST_FOREACH(sess, &prg->allsess, s_list) { 1097 if (sess->s_sid == base) { 1098 base += ALLPROC_HSIZE; 1099 if (base >= PID_MAX) { 1100 lwkt_reltoken(&prg->proc_token); 1101 goto retry; 1102 } 1103 ++pid_inner_skips; 1104 goto restart1; 1105 } 1106 } 1107 1108 /* 1109 * Assign the pid and insert the process. 1110 */ 1111 p->p_pid = base; 1112 LIST_INSERT_HEAD(&prg->allproc, p, p_list); 1113 lwkt_reltoken(&prg->proc_token); 1114 } 1115 1116 /* 1117 * Called from exit1 to place the process into a zombie state. 1118 * The process is removed from the pid hash and p_stat is set 1119 * to SZOMB. Normal pfind[n]() calls will not find it any more. 1120 * 1121 * Caller must hold p->p_token. We are required to wait until p_lock 1122 * becomes zero before we can manipulate the list, allowing allproc 1123 * scans to guarantee consistency during a list scan. 1124 */ 1125 void 1126 proc_move_allproc_zombie(struct proc *p) 1127 { 1128 procglob_t *prg; 1129 int n; 1130 1131 n = ALLPROC_HASH(p->p_pid); 1132 prg = &procglob[n]; 1133 PSTALL(p, "reap1", 0); 1134 lwkt_gettoken(&prg->proc_token); 1135 1136 PSTALL(p, "reap1a", 0); 1137 p->p_stat = SZOMB; 1138 1139 lwkt_reltoken(&prg->proc_token); 1140 dsched_exit_proc(p); 1141 } 1142 1143 /* 1144 * This routine is called from kern_wait() and will remove the process 1145 * from the zombie list and the sibling list. This routine will block 1146 * if someone has a lock on the proces (p_lock). 1147 * 1148 * Caller must hold p->p_token. We are required to wait until p_lock 1149 * becomes one before we can manipulate the list, allowing allproc 1150 * scans to guarantee consistency during a list scan. 1151 * 1152 * Assumes caller has one ref. 1153 */ 1154 void 1155 proc_remove_zombie(struct proc *p) 1156 { 1157 procglob_t *prg; 1158 int n; 1159 1160 n = ALLPROC_HASH(p->p_pid); 1161 prg = &procglob[n]; 1162 1163 PSTALL(p, "reap2", 1); 1164 lwkt_gettoken(&prg->proc_token); 1165 PSTALL(p, "reap2a", 1); 1166 LIST_REMOVE(p, p_list); /* from remove master list */ 1167 LIST_REMOVE(p, p_sibling); /* and from sibling list */ 1168 p->p_pptr = NULL; 1169 p->p_ppid = 0; 1170 if (pid_doms[p->p_pid % PIDSEL_DOMAINS] != (uint8_t)time_second) 1171 pid_doms[p->p_pid % PIDSEL_DOMAINS] = (uint8_t)time_second; 1172 lwkt_reltoken(&prg->proc_token); 1173 } 1174 1175 /* 1176 * Handle various requirements prior to returning to usermode. Called from 1177 * platform trap and system call code. 1178 */ 1179 void 1180 lwpuserret(struct lwp *lp) 1181 { 1182 struct proc *p = lp->lwp_proc; 1183 1184 if (lp->lwp_mpflags & LWP_MP_VNLRU) { 1185 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_VNLRU); 1186 allocvnode_gc(); 1187 } 1188 if (lp->lwp_mpflags & LWP_MP_WEXIT) { 1189 lwkt_gettoken(&p->p_token); 1190 lwp_exit(0, NULL); 1191 lwkt_reltoken(&p->p_token); /* NOT REACHED */ 1192 } 1193 } 1194 1195 /* 1196 * Kernel threads run from user processes can also accumulate deferred 1197 * actions which need to be acted upon. Callers include: 1198 * 1199 * nfsd - Can allocate lots of vnodes 1200 */ 1201 void 1202 lwpkthreaddeferred(void) 1203 { 1204 struct lwp *lp = curthread->td_lwp; 1205 1206 if (lp) { 1207 if (lp->lwp_mpflags & LWP_MP_VNLRU) { 1208 atomic_clear_int(&lp->lwp_mpflags, LWP_MP_VNLRU); 1209 allocvnode_gc(); 1210 } 1211 } 1212 } 1213 1214 void 1215 proc_usermap(struct proc *p, int invfork) 1216 { 1217 struct sys_upmap *upmap; 1218 1219 lwkt_gettoken(&p->p_token); 1220 upmap = kmalloc(roundup2(sizeof(*upmap), PAGE_SIZE), M_PROC, 1221 M_WAITOK | M_ZERO); 1222 if (p->p_upmap == NULL) { 1223 upmap->header[0].type = UKPTYPE_VERSION; 1224 upmap->header[0].offset = offsetof(struct sys_upmap, version); 1225 upmap->header[1].type = UPTYPE_RUNTICKS; 1226 upmap->header[1].offset = offsetof(struct sys_upmap, runticks); 1227 upmap->header[2].type = UPTYPE_FORKID; 1228 upmap->header[2].offset = offsetof(struct sys_upmap, forkid); 1229 upmap->header[3].type = UPTYPE_PID; 1230 upmap->header[3].offset = offsetof(struct sys_upmap, pid); 1231 upmap->header[4].type = UPTYPE_PROC_TITLE; 1232 upmap->header[4].offset = offsetof(struct sys_upmap,proc_title); 1233 upmap->header[5].type = UPTYPE_INVFORK; 1234 upmap->header[5].offset = offsetof(struct sys_upmap, invfork); 1235 1236 upmap->version = UPMAP_VERSION; 1237 upmap->pid = p->p_pid; 1238 upmap->forkid = p->p_forkid; 1239 upmap->invfork = invfork; 1240 p->p_upmap = upmap; 1241 } else { 1242 kfree(upmap, M_PROC); 1243 } 1244 lwkt_reltoken(&p->p_token); 1245 } 1246 1247 void 1248 proc_userunmap(struct proc *p) 1249 { 1250 struct sys_upmap *upmap; 1251 1252 lwkt_gettoken(&p->p_token); 1253 if ((upmap = p->p_upmap) != NULL) { 1254 p->p_upmap = NULL; 1255 kfree(upmap, M_PROC); 1256 } 1257 lwkt_reltoken(&p->p_token); 1258 } 1259 1260 /* 1261 * Scan all processes on the allproc list. The process is automatically 1262 * held for the callback. A return value of -1 terminates the loop. 1263 * Zombie procs are skipped. 1264 * 1265 * The callback is made with the process held and proc_token held. 1266 * 1267 * We limit the scan to the number of processes as-of the start of 1268 * the scan so as not to get caught up in an endless loop if new processes 1269 * are created more quickly than we can scan the old ones. Add a little 1270 * slop to try to catch edge cases since nprocs can race. 1271 * 1272 * No requirements. 1273 */ 1274 void 1275 allproc_scan(int (*callback)(struct proc *, void *), void *data, int segmented) 1276 { 1277 int limit = nprocs + ncpus; 1278 struct proc *p; 1279 int ns; 1280 int ne; 1281 int r; 1282 int n; 1283 1284 if (segmented) { 1285 int id = mycpu->gd_cpuid; 1286 ns = id * ALLPROC_HSIZE / ncpus; 1287 ne = (id + 1) * ALLPROC_HSIZE / ncpus; 1288 } else { 1289 ns = 0; 1290 ne = ALLPROC_HSIZE; 1291 } 1292 1293 /* 1294 * prg->proc_token protects the allproc list and PHOLD() prevents the 1295 * process from being removed from the allproc list or the zombproc 1296 * list. 1297 */ 1298 for (n = ns; n < ne; ++n) { 1299 procglob_t *prg = &procglob[n]; 1300 if (LIST_FIRST(&prg->allproc) == NULL) 1301 continue; 1302 lwkt_gettoken(&prg->proc_token); 1303 LIST_FOREACH(p, &prg->allproc, p_list) { 1304 if (p->p_stat == SZOMB) 1305 continue; 1306 PHOLD(p); 1307 r = callback(p, data); 1308 PRELE(p); 1309 if (r < 0) 1310 break; 1311 if (--limit < 0) 1312 break; 1313 } 1314 lwkt_reltoken(&prg->proc_token); 1315 1316 /* 1317 * Check if asked to stop early 1318 */ 1319 if (p) 1320 break; 1321 } 1322 } 1323 1324 /* 1325 * Scan all lwps of processes on the allproc list. The lwp is automatically 1326 * held for the callback. A return value of -1 terminates the loop. 1327 * 1328 * The callback is made with the proces and lwp both held, and proc_token held. 1329 * 1330 * No requirements. 1331 */ 1332 void 1333 alllwp_scan(int (*callback)(struct lwp *, void *), void *data, int segmented) 1334 { 1335 struct proc *p; 1336 struct lwp *lp; 1337 int ns; 1338 int ne; 1339 int r = 0; 1340 int n; 1341 1342 if (segmented) { 1343 int id = mycpu->gd_cpuid; 1344 ns = id * ALLPROC_HSIZE / ncpus; 1345 ne = (id + 1) * ALLPROC_HSIZE / ncpus; 1346 } else { 1347 ns = 0; 1348 ne = ALLPROC_HSIZE; 1349 } 1350 1351 for (n = ns; n < ne; ++n) { 1352 procglob_t *prg = &procglob[n]; 1353 1354 if (LIST_FIRST(&prg->allproc) == NULL) 1355 continue; 1356 lwkt_gettoken(&prg->proc_token); 1357 LIST_FOREACH(p, &prg->allproc, p_list) { 1358 if (p->p_stat == SZOMB) 1359 continue; 1360 PHOLD(p); 1361 lwkt_gettoken(&p->p_token); 1362 FOREACH_LWP_IN_PROC(lp, p) { 1363 LWPHOLD(lp); 1364 r = callback(lp, data); 1365 LWPRELE(lp); 1366 } 1367 lwkt_reltoken(&p->p_token); 1368 PRELE(p); 1369 if (r < 0) 1370 break; 1371 } 1372 lwkt_reltoken(&prg->proc_token); 1373 1374 /* 1375 * Asked to exit early 1376 */ 1377 if (p) 1378 break; 1379 } 1380 } 1381 1382 /* 1383 * Scan all processes on the zombproc list. The process is automatically 1384 * held for the callback. A return value of -1 terminates the loop. 1385 * 1386 * No requirements. 1387 * The callback is made with the proces held and proc_token held. 1388 */ 1389 void 1390 zombproc_scan(int (*callback)(struct proc *, void *), void *data) 1391 { 1392 struct proc *p; 1393 int r; 1394 int n; 1395 1396 /* 1397 * prg->proc_token protects the allproc list and PHOLD() prevents the 1398 * process from being removed from the allproc list or the zombproc 1399 * list. 1400 */ 1401 for (n = 0; n < ALLPROC_HSIZE; ++n) { 1402 procglob_t *prg = &procglob[n]; 1403 1404 if (LIST_FIRST(&prg->allproc) == NULL) 1405 continue; 1406 lwkt_gettoken(&prg->proc_token); 1407 LIST_FOREACH(p, &prg->allproc, p_list) { 1408 if (p->p_stat != SZOMB) 1409 continue; 1410 PHOLD(p); 1411 r = callback(p, data); 1412 PRELE(p); 1413 if (r < 0) 1414 break; 1415 } 1416 lwkt_reltoken(&prg->proc_token); 1417 1418 /* 1419 * Check if asked to stop early 1420 */ 1421 if (p) 1422 break; 1423 } 1424 } 1425 1426 #include "opt_ddb.h" 1427 #ifdef DDB 1428 #include <ddb/ddb.h> 1429 1430 /* 1431 * Debugging only 1432 */ 1433 DB_SHOW_COMMAND(pgrpdump, pgrpdump) 1434 { 1435 struct pgrp *pgrp; 1436 struct proc *p; 1437 procglob_t *prg; 1438 int i; 1439 1440 for (i = 0; i < ALLPROC_HSIZE; ++i) { 1441 prg = &procglob[i]; 1442 1443 if (LIST_EMPTY(&prg->allpgrp)) 1444 continue; 1445 kprintf("\tindx %d\n", i); 1446 LIST_FOREACH(pgrp, &prg->allpgrp, pg_list) { 1447 kprintf("\tpgrp %p, pgid %ld, sess %p, " 1448 "sesscnt %d, mem %p\n", 1449 (void *)pgrp, (long)pgrp->pg_id, 1450 (void *)pgrp->pg_session, 1451 pgrp->pg_session->s_count, 1452 (void *)LIST_FIRST(&pgrp->pg_members)); 1453 LIST_FOREACH(p, &pgrp->pg_members, p_pglist) { 1454 kprintf("\t\tpid %ld addr %p pgrp %p\n", 1455 (long)p->p_pid, (void *)p, 1456 (void *)p->p_pgrp); 1457 } 1458 } 1459 } 1460 } 1461 #endif /* DDB */ 1462 1463 /* 1464 * The caller must hold proc_token. 1465 */ 1466 static int 1467 sysctl_out_proc(struct proc *p, struct sysctl_req *req, int flags) 1468 { 1469 struct kinfo_proc ki; 1470 struct lwp *lp; 1471 int skp = 0, had_output = 0; 1472 int error; 1473 1474 bzero(&ki, sizeof(ki)); 1475 lwkt_gettoken_shared(&p->p_token); 1476 fill_kinfo_proc(p, &ki); 1477 if ((flags & KERN_PROC_FLAG_LWP) == 0) 1478 skp = 1; 1479 error = 0; 1480 FOREACH_LWP_IN_PROC(lp, p) { 1481 LWPHOLD(lp); 1482 fill_kinfo_lwp(lp, &ki.kp_lwp); 1483 had_output = 1; 1484 error = SYSCTL_OUT(req, &ki, sizeof(ki)); 1485 LWPRELE(lp); 1486 if (error) 1487 break; 1488 if (skp) 1489 break; 1490 } 1491 lwkt_reltoken(&p->p_token); 1492 /* We need to output at least the proc, even if there is no lwp. */ 1493 if (had_output == 0) { 1494 error = SYSCTL_OUT(req, &ki, sizeof(ki)); 1495 } 1496 return (error); 1497 } 1498 1499 /* 1500 * The caller must hold proc_token. 1501 */ 1502 static int 1503 sysctl_out_proc_kthread(struct thread *td, struct sysctl_req *req) 1504 { 1505 struct kinfo_proc ki; 1506 int error; 1507 1508 fill_kinfo_proc_kthread(td, &ki); 1509 error = SYSCTL_OUT(req, &ki, sizeof(ki)); 1510 if (error) 1511 return error; 1512 return(0); 1513 } 1514 1515 /* 1516 * No requirements. 1517 */ 1518 static int 1519 sysctl_kern_proc(SYSCTL_HANDLER_ARGS) 1520 { 1521 int *name = (int *)arg1; 1522 int oid = oidp->oid_number; 1523 u_int namelen = arg2; 1524 struct proc *p; 1525 struct thread *td; 1526 struct thread *marker; 1527 int flags = 0; 1528 int error = 0; 1529 int n; 1530 int origcpu; 1531 struct ucred *cr1 = curproc->p_ucred; 1532 struct ucred *crcache = NULL; 1533 1534 flags = oid & KERN_PROC_FLAGMASK; 1535 oid &= ~KERN_PROC_FLAGMASK; 1536 1537 if ((oid == KERN_PROC_ALL && namelen != 0) || 1538 (oid != KERN_PROC_ALL && namelen != 1)) { 1539 return (EINVAL); 1540 } 1541 1542 /* 1543 * proc_token protects the allproc list and PHOLD() prevents the 1544 * process from being removed from the allproc list or the zombproc 1545 * list. 1546 */ 1547 if (oid == KERN_PROC_PID) { 1548 p = pfind((pid_t)name[0]); 1549 if (p) { 1550 crcache = pcredcache(crcache, p); 1551 if (PRISON_CHECK(cr1, crcache)) 1552 error = sysctl_out_proc(p, req, flags); 1553 PRELE(p); 1554 } 1555 goto post_threads; 1556 } 1557 p = NULL; 1558 1559 if (!req->oldptr) { 1560 /* overestimate by 5 procs */ 1561 error = SYSCTL_OUT(req, 0, sizeof (struct kinfo_proc) * 5); 1562 if (error) 1563 goto post_threads; 1564 } 1565 1566 for (n = 0; n < ALLPROC_HSIZE; ++n) { 1567 procglob_t *prg = &procglob[n]; 1568 1569 if (LIST_EMPTY(&prg->allproc)) 1570 continue; 1571 lwkt_gettoken_shared(&prg->proc_token); 1572 LIST_FOREACH(p, &prg->allproc, p_list) { 1573 /* 1574 * Show a user only their processes. 1575 */ 1576 if (ps_showallprocs == 0) { 1577 crcache = pcredcache(crcache, p); 1578 if (crcache == NULL || 1579 p_trespass(cr1, crcache)) { 1580 continue; 1581 } 1582 } 1583 1584 /* 1585 * Skip embryonic processes. 1586 */ 1587 if (p->p_stat == SIDL) 1588 continue; 1589 /* 1590 * TODO - make more efficient (see notes below). 1591 * do by session. 1592 */ 1593 switch (oid) { 1594 case KERN_PROC_PGRP: 1595 /* could do this by traversing pgrp */ 1596 if (p->p_pgrp == NULL || 1597 p->p_pgrp->pg_id != (pid_t)name[0]) 1598 continue; 1599 break; 1600 1601 case KERN_PROC_TTY: 1602 if ((p->p_flags & P_CONTROLT) == 0 || 1603 p->p_session == NULL || 1604 p->p_session->s_ttyp == NULL || 1605 dev2udev(p->p_session->s_ttyp->t_dev) != 1606 (udev_t)name[0]) 1607 continue; 1608 break; 1609 1610 case KERN_PROC_UID: 1611 crcache = pcredcache(crcache, p); 1612 if (crcache == NULL || 1613 crcache->cr_uid != (uid_t)name[0]) { 1614 continue; 1615 } 1616 break; 1617 1618 case KERN_PROC_RUID: 1619 crcache = pcredcache(crcache, p); 1620 if (crcache == NULL || 1621 crcache->cr_ruid != (uid_t)name[0]) { 1622 continue; 1623 } 1624 break; 1625 } 1626 1627 crcache = pcredcache(crcache, p); 1628 if (!PRISON_CHECK(cr1, crcache)) 1629 continue; 1630 PHOLD(p); 1631 error = sysctl_out_proc(p, req, flags); 1632 PRELE(p); 1633 if (error) { 1634 lwkt_reltoken(&prg->proc_token); 1635 goto post_threads; 1636 } 1637 } 1638 lwkt_reltoken(&prg->proc_token); 1639 } 1640 1641 /* 1642 * Iterate over all active cpus and scan their thread list. Start 1643 * with the next logical cpu and end with our original cpu. We 1644 * migrate our own thread to each target cpu in order to safely scan 1645 * its thread list. In the last loop we migrate back to our original 1646 * cpu. 1647 */ 1648 origcpu = mycpu->gd_cpuid; 1649 if (!ps_showallthreads || jailed(cr1)) 1650 goto post_threads; 1651 1652 marker = kmalloc(sizeof(struct thread), M_TEMP, M_WAITOK|M_ZERO); 1653 marker->td_flags = TDF_MARKER; 1654 error = 0; 1655 1656 for (n = 1; n <= ncpus; ++n) { 1657 globaldata_t rgd; 1658 int nid; 1659 1660 nid = (origcpu + n) % ncpus; 1661 if (CPUMASK_TESTBIT(smp_active_mask, nid) == 0) 1662 continue; 1663 rgd = globaldata_find(nid); 1664 lwkt_setcpu_self(rgd); 1665 1666 crit_enter(); 1667 TAILQ_INSERT_TAIL(&rgd->gd_tdallq, marker, td_allq); 1668 1669 while ((td = TAILQ_PREV(marker, lwkt_queue, td_allq)) != NULL) { 1670 TAILQ_REMOVE(&rgd->gd_tdallq, marker, td_allq); 1671 TAILQ_INSERT_BEFORE(td, marker, td_allq); 1672 if (td->td_flags & TDF_MARKER) 1673 continue; 1674 if (td->td_proc) 1675 continue; 1676 1677 lwkt_hold(td); 1678 crit_exit(); 1679 1680 switch (oid) { 1681 case KERN_PROC_PGRP: 1682 case KERN_PROC_TTY: 1683 case KERN_PROC_UID: 1684 case KERN_PROC_RUID: 1685 break; 1686 default: 1687 error = sysctl_out_proc_kthread(td, req); 1688 break; 1689 } 1690 lwkt_rele(td); 1691 crit_enter(); 1692 if (error) 1693 break; 1694 } 1695 TAILQ_REMOVE(&rgd->gd_tdallq, marker, td_allq); 1696 crit_exit(); 1697 1698 if (error) 1699 break; 1700 } 1701 1702 /* 1703 * Userland scheduler expects us to return on the same cpu we 1704 * started on. 1705 */ 1706 if (mycpu->gd_cpuid != origcpu) 1707 lwkt_setcpu_self(globaldata_find(origcpu)); 1708 1709 kfree(marker, M_TEMP); 1710 1711 post_threads: 1712 if (crcache) 1713 crfree(crcache); 1714 return (error); 1715 } 1716 1717 /* 1718 * This sysctl allows a process to retrieve the argument list or process 1719 * title for another process without groping around in the address space 1720 * of the other process. It also allow a process to set its own "process 1721 * title to a string of its own choice. 1722 * 1723 * No requirements. 1724 */ 1725 static int 1726 sysctl_kern_proc_args(SYSCTL_HANDLER_ARGS) 1727 { 1728 int *name = (int*) arg1; 1729 u_int namelen = arg2; 1730 struct proc *p; 1731 struct pargs *opa; 1732 struct pargs *pa; 1733 int error = 0; 1734 struct ucred *cr1 = curproc->p_ucred; 1735 1736 if (namelen != 1) 1737 return (EINVAL); 1738 1739 p = pfind((pid_t)name[0]); 1740 if (p == NULL) 1741 goto done; 1742 lwkt_gettoken(&p->p_token); 1743 1744 if ((!ps_argsopen) && p_trespass(cr1, p->p_ucred)) 1745 goto done; 1746 1747 if (req->newptr && curproc != p) { 1748 error = EPERM; 1749 goto done; 1750 } 1751 if (req->oldptr) { 1752 if (p->p_upmap != NULL && p->p_upmap->proc_title[0]) { 1753 /* 1754 * Args set via writable user process mmap. 1755 * We must calculate the string length manually 1756 * because the user data can change at any time. 1757 */ 1758 size_t n; 1759 char *base; 1760 1761 base = p->p_upmap->proc_title; 1762 for (n = 0; n < UPMAP_MAXPROCTITLE - 1; ++n) { 1763 if (base[n] == 0) 1764 break; 1765 } 1766 error = SYSCTL_OUT(req, base, n); 1767 if (error == 0) 1768 error = SYSCTL_OUT(req, "", 1); 1769 } else if ((pa = p->p_args) != NULL) { 1770 /* 1771 * Args set by setproctitle() sysctl. 1772 */ 1773 refcount_acquire(&pa->ar_ref); 1774 error = SYSCTL_OUT(req, pa->ar_args, pa->ar_length); 1775 if (refcount_release(&pa->ar_ref)) 1776 kfree(pa, M_PARGS); 1777 } 1778 } 1779 if (req->newptr == NULL) 1780 goto done; 1781 1782 if (req->newlen + sizeof(struct pargs) > ps_arg_cache_limit) { 1783 goto done; 1784 } 1785 1786 pa = kmalloc(sizeof(struct pargs) + req->newlen, M_PARGS, M_WAITOK); 1787 refcount_init(&pa->ar_ref, 1); 1788 pa->ar_length = req->newlen; 1789 error = SYSCTL_IN(req, pa->ar_args, req->newlen); 1790 if (error) { 1791 kfree(pa, M_PARGS); 1792 goto done; 1793 } 1794 1795 1796 /* 1797 * Replace p_args with the new pa. p_args may have previously 1798 * been NULL. 1799 */ 1800 opa = p->p_args; 1801 p->p_args = pa; 1802 1803 if (opa) { 1804 KKASSERT(opa->ar_ref > 0); 1805 if (refcount_release(&opa->ar_ref)) { 1806 kfree(opa, M_PARGS); 1807 /* opa = NULL; */ 1808 } 1809 } 1810 done: 1811 if (p) { 1812 lwkt_reltoken(&p->p_token); 1813 PRELE(p); 1814 } 1815 return (error); 1816 } 1817 1818 static int 1819 sysctl_kern_proc_cwd(SYSCTL_HANDLER_ARGS) 1820 { 1821 int *name = (int*) arg1; 1822 u_int namelen = arg2; 1823 struct proc *p; 1824 int error = 0; 1825 char *fullpath, *freepath; 1826 struct ucred *cr1 = curproc->p_ucred; 1827 1828 if (namelen != 1) 1829 return (EINVAL); 1830 1831 p = pfind((pid_t)name[0]); 1832 if (p == NULL) 1833 goto done; 1834 lwkt_gettoken_shared(&p->p_token); 1835 1836 /* 1837 * If we are not allowed to see other args, we certainly shouldn't 1838 * get the cwd either. Also check the usual trespassing. 1839 */ 1840 if ((!ps_argsopen) && p_trespass(cr1, p->p_ucred)) 1841 goto done; 1842 1843 if (req->oldptr && p->p_fd != NULL && p->p_fd->fd_ncdir.ncp) { 1844 struct nchandle nch; 1845 1846 cache_copy(&p->p_fd->fd_ncdir, &nch); 1847 error = cache_fullpath(p, &nch, NULL, 1848 &fullpath, &freepath, 0); 1849 cache_drop(&nch); 1850 if (error) 1851 goto done; 1852 error = SYSCTL_OUT(req, fullpath, strlen(fullpath) + 1); 1853 kfree(freepath, M_TEMP); 1854 } 1855 1856 done: 1857 if (p) { 1858 lwkt_reltoken(&p->p_token); 1859 PRELE(p); 1860 } 1861 return (error); 1862 } 1863 1864 /* 1865 * This sysctl allows a process to retrieve the path of the executable for 1866 * itself or another process. 1867 */ 1868 static int 1869 sysctl_kern_proc_pathname(SYSCTL_HANDLER_ARGS) 1870 { 1871 pid_t *pidp = (pid_t *)arg1; 1872 unsigned int arglen = arg2; 1873 struct proc *p; 1874 char *retbuf, *freebuf; 1875 int error = 0; 1876 struct nchandle nch; 1877 1878 if (arglen != 1) 1879 return (EINVAL); 1880 if (*pidp == -1) { /* -1 means this process */ 1881 p = curproc; 1882 } else { 1883 p = pfind(*pidp); 1884 if (p == NULL) 1885 return (ESRCH); 1886 } 1887 1888 cache_copy(&p->p_textnch, &nch); 1889 error = cache_fullpath(p, &nch, NULL, &retbuf, &freebuf, 0); 1890 cache_drop(&nch); 1891 if (error) 1892 goto done; 1893 error = SYSCTL_OUT(req, retbuf, strlen(retbuf) + 1); 1894 kfree(freebuf, M_TEMP); 1895 done: 1896 if (*pidp != -1) 1897 PRELE(p); 1898 1899 return (error); 1900 } 1901 1902 static int 1903 sysctl_kern_proc_sigtramp(SYSCTL_HANDLER_ARGS) 1904 { 1905 /*int *name = (int *)arg1;*/ 1906 u_int namelen = arg2; 1907 struct kinfo_sigtramp kst; 1908 const struct sysentvec *sv; 1909 int error; 1910 1911 if (namelen > 1) 1912 return (EINVAL); 1913 /* ignore pid if passed in (freebsd compatibility) */ 1914 1915 sv = curproc->p_sysent; 1916 bzero(&kst, sizeof(kst)); 1917 if (sv->sv_szsigcode) { 1918 intptr_t sigbase; 1919 1920 sigbase = trunc_page64((intptr_t)PS_STRINGS - 1921 *sv->sv_szsigcode); 1922 sigbase -= SZSIGCODE_EXTRA_BYTES; 1923 1924 kst.ksigtramp_start = (void *)sigbase; 1925 kst.ksigtramp_end = (void *)(sigbase + *sv->sv_szsigcode); 1926 } 1927 error = SYSCTL_OUT(req, &kst, sizeof(kst)); 1928 1929 return (error); 1930 } 1931 1932 SYSCTL_NODE(_kern, KERN_PROC, proc, CTLFLAG_RD, 0, "Process table"); 1933 1934 SYSCTL_PROC(_kern_proc, KERN_PROC_ALL, all, 1935 CTLFLAG_RD | CTLTYPE_STRUCT | CTLFLAG_NOLOCK, 1936 0, 0, sysctl_kern_proc, "S,proc", "Return entire process table"); 1937 1938 SYSCTL_NODE(_kern_proc, KERN_PROC_PGRP, pgrp, 1939 CTLFLAG_RD | CTLFLAG_NOLOCK, 1940 sysctl_kern_proc, "Process table"); 1941 1942 SYSCTL_NODE(_kern_proc, KERN_PROC_TTY, tty, 1943 CTLFLAG_RD | CTLFLAG_NOLOCK, 1944 sysctl_kern_proc, "Process table"); 1945 1946 SYSCTL_NODE(_kern_proc, KERN_PROC_UID, uid, 1947 CTLFLAG_RD | CTLFLAG_NOLOCK, 1948 sysctl_kern_proc, "Process table"); 1949 1950 SYSCTL_NODE(_kern_proc, KERN_PROC_RUID, ruid, 1951 CTLFLAG_RD | CTLFLAG_NOLOCK, 1952 sysctl_kern_proc, "Process table"); 1953 1954 SYSCTL_NODE(_kern_proc, KERN_PROC_PID, pid, 1955 CTLFLAG_RD | CTLFLAG_NOLOCK, 1956 sysctl_kern_proc, "Process table"); 1957 1958 SYSCTL_NODE(_kern_proc, (KERN_PROC_ALL | KERN_PROC_FLAG_LWP), all_lwp, 1959 CTLFLAG_RD | CTLFLAG_NOLOCK, 1960 sysctl_kern_proc, "Process table"); 1961 1962 SYSCTL_NODE(_kern_proc, (KERN_PROC_PGRP | KERN_PROC_FLAG_LWP), pgrp_lwp, 1963 CTLFLAG_RD | CTLFLAG_NOLOCK, 1964 sysctl_kern_proc, "Process table"); 1965 1966 SYSCTL_NODE(_kern_proc, (KERN_PROC_TTY | KERN_PROC_FLAG_LWP), tty_lwp, 1967 CTLFLAG_RD | CTLFLAG_NOLOCK, 1968 sysctl_kern_proc, "Process table"); 1969 1970 SYSCTL_NODE(_kern_proc, (KERN_PROC_UID | KERN_PROC_FLAG_LWP), uid_lwp, 1971 CTLFLAG_RD | CTLFLAG_NOLOCK, 1972 sysctl_kern_proc, "Process table"); 1973 1974 SYSCTL_NODE(_kern_proc, (KERN_PROC_RUID | KERN_PROC_FLAG_LWP), ruid_lwp, 1975 CTLFLAG_RD | CTLFLAG_NOLOCK, 1976 sysctl_kern_proc, "Process table"); 1977 1978 SYSCTL_NODE(_kern_proc, (KERN_PROC_PID | KERN_PROC_FLAG_LWP), pid_lwp, 1979 CTLFLAG_RD | CTLFLAG_NOLOCK, 1980 sysctl_kern_proc, "Process table"); 1981 1982 SYSCTL_NODE(_kern_proc, KERN_PROC_ARGS, args, 1983 CTLFLAG_RW | CTLFLAG_ANYBODY | CTLFLAG_NOLOCK, 1984 sysctl_kern_proc_args, "Process argument list"); 1985 1986 SYSCTL_NODE(_kern_proc, KERN_PROC_CWD, cwd, 1987 CTLFLAG_RD | CTLFLAG_ANYBODY | CTLFLAG_NOLOCK, 1988 sysctl_kern_proc_cwd, "Process argument list"); 1989 1990 static SYSCTL_NODE(_kern_proc, KERN_PROC_PATHNAME, pathname, 1991 CTLFLAG_RD | CTLFLAG_NOLOCK, 1992 sysctl_kern_proc_pathname, "Process executable path"); 1993 1994 SYSCTL_PROC(_kern_proc, KERN_PROC_SIGTRAMP, sigtramp, 1995 CTLFLAG_RD | CTLTYPE_STRUCT | CTLFLAG_NOLOCK, 1996 0, 0, sysctl_kern_proc_sigtramp, "S,sigtramp", 1997 "Return sigtramp address range"); 1998