kern_proc.c revision 1.94.4.1
1 /* $NetBSD: kern_proc.c,v 1.94.4.1 2006/09/11 18:19:09 ad Exp $ */ 2 3 /*- 4 * Copyright (c) 1999 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility, 9 * NASA Ames Research Center. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 3. All advertising materials mentioning features or use of this software 20 * must display the following acknowledgement: 21 * This product includes software developed by the NetBSD 22 * Foundation, Inc. and its contributors. 23 * 4. Neither the name of The NetBSD Foundation nor the names of its 24 * contributors may be used to endorse or promote products derived 25 * from this software without specific prior written permission. 26 * 27 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 28 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 29 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 30 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 31 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 32 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 33 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 34 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 35 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 36 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 37 * POSSIBILITY OF SUCH DAMAGE. 38 */ 39 40 /* 41 * Copyright (c) 1982, 1986, 1989, 1991, 1993 42 * The Regents of the University of California. All rights reserved. 43 * 44 * Redistribution and use in source and binary forms, with or without 45 * modification, are permitted provided that the following conditions 46 * are met: 47 * 1. Redistributions of source code must retain the above copyright 48 * notice, this list of conditions and the following disclaimer. 49 * 2. Redistributions in binary form must reproduce the above copyright 50 * notice, this list of conditions and the following disclaimer in the 51 * documentation and/or other materials provided with the distribution. 52 * 3. Neither the name of the University nor the names of its contributors 53 * may be used to endorse or promote products derived from this software 54 * without specific prior written permission. 55 * 56 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 57 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 58 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 59 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 60 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 61 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 62 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 63 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 64 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 65 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 66 * SUCH DAMAGE. 67 * 68 * @(#)kern_proc.c 8.7 (Berkeley) 2/14/95 69 */ 70 71 #include <sys/cdefs.h> 72 __KERNEL_RCSID(0, "$NetBSD: kern_proc.c,v 1.94.4.1 2006/09/11 18:19:09 ad Exp $"); 73 74 #include "opt_kstack.h" 75 #include "opt_maxuprc.h" 76 #include "opt_multiprocessor.h" 77 #include "opt_lockdebug.h" 78 79 #include <sys/param.h> 80 #include <sys/systm.h> 81 #include <sys/kernel.h> 82 #include <sys/proc.h> 83 #include <sys/resourcevar.h> 84 #include <sys/buf.h> 85 #include <sys/acct.h> 86 #include <sys/wait.h> 87 #include <sys/file.h> 88 #include <ufs/ufs/quota.h> 89 #include <sys/uio.h> 90 #include <sys/malloc.h> 91 #include <sys/pool.h> 92 #include <sys/mbuf.h> 93 #include <sys/ioctl.h> 94 #include <sys/tty.h> 95 #include <sys/signalvar.h> 96 #include <sys/ras.h> 97 #include <sys/sa.h> 98 #include <sys/savar.h> 99 #include <sys/filedesc.h> 100 #include <sys/kauth.h> 101 #include <sys/turnstile.h> 102 103 #include <uvm/uvm.h> 104 #include <uvm/uvm_extern.h> 105 106 /* 107 * Other process lists 108 */ 109 110 struct proclist allproc; 111 struct proclist zombproc; /* resources have been freed */ 112 113 /* 114 * There are three locks on global process state. 115 * 116 * 1. proclist_lock is a reader/writer lock and is used when modifying or 117 * examining process state from a process context. It protects our internal 118 * tables, all of the process lists, and a number of members of struct lwp 119 * and struct proc. 120 121 * 2. proclist_mutex is used when allproc must be traversed from an 122 * interrupt context, or when we must signal processes from an interrupt 123 * context. The proclist_lock should always be used in preference. 124 * 125 * 3. alllwp_mutex protects the "alllwp" list. 126 * 127 * proclist_lock proclist_mutex alllwp_mutex structure 128 * --------------- --------------- --------------- ----------------- 129 * x zombproc 130 * x pid_table 131 * x proc::p_pptr 132 * x proc::p_sibling 133 * x proc::p_children 134 * x x allproc 135 * x x proc::p_pgrp 136 * x x proc::p_pglist 137 * x x proc::p_session 138 * x x proc::p_list 139 * x alllwp 140 * x lwp::l_list 141 */ 142 krwlock_t proclist_lock; 143 kmutex_t proclist_mutex; 144 145 /* 146 * pid to proc lookup is done by indexing the pid_table array. 147 * Since pid numbers are only allocated when an empty slot 148 * has been found, there is no need to search any lists ever. 149 * (an orphaned pgrp will lock the slot, a session will lock 150 * the pgrp with the same number.) 151 * If the table is too small it is reallocated with twice the 152 * previous size and the entries 'unzipped' into the two halves. 153 * A linked list of free entries is passed through the pt_proc 154 * field of 'free' items - set odd to be an invalid ptr. 155 */ 156 157 struct pid_table { 158 struct proc *pt_proc; 159 struct pgrp *pt_pgrp; 160 }; 161 #if 1 /* strongly typed cast - should be a noop */ 162 static inline uint p2u(struct proc *p) { return (uint)(uintptr_t)p; } 163 #else 164 #define p2u(p) ((uint)p) 165 #endif 166 #define P_VALID(p) (!(p2u(p) & 1)) 167 #define P_NEXT(p) (p2u(p) >> 1) 168 #define P_FREE(pid) ((struct proc *)(uintptr_t)((pid) << 1 | 1)) 169 170 #define INITIAL_PID_TABLE_SIZE (1 << 5) 171 static struct pid_table *pid_table; 172 static uint pid_tbl_mask = INITIAL_PID_TABLE_SIZE - 1; 173 static uint pid_alloc_lim; /* max we allocate before growing table */ 174 static uint pid_alloc_cnt; /* number of allocated pids */ 175 176 /* links through free slots - never empty! */ 177 static uint next_free_pt, last_free_pt; 178 static pid_t pid_max = PID_MAX; /* largest value we allocate */ 179 180 /* Components of the first process -- never freed. */ 181 struct session session0; 182 struct pgrp pgrp0; 183 struct proc proc0; 184 struct lwp lwp0; 185 kauth_cred_t cred0; 186 struct filedesc0 filedesc0; 187 struct cwdinfo cwdi0; 188 struct plimit limit0; 189 struct pstats pstat0; 190 struct vmspace vmspace0; 191 struct sigacts sigacts0; 192 struct turnstile turnstile0; 193 194 extern struct user *proc0paddr; 195 196 extern const struct emul emul_netbsd; /* defined in kern_exec.c */ 197 198 int nofile = NOFILE; 199 int maxuprc = MAXUPRC; 200 int cmask = CMASK; 201 202 POOL_INIT(proc_pool, sizeof(struct proc), 0, 0, 0, "procpl", 203 &pool_allocator_nointr); 204 POOL_INIT(lwp_pool, sizeof(struct lwp), 0, 0, 0, "lwppl", 205 &pool_allocator_nointr); 206 POOL_INIT(lwp_uc_pool, sizeof(ucontext_t), 0, 0, 0, "lwpucpl", 207 &pool_allocator_nointr); 208 POOL_INIT(pgrp_pool, sizeof(struct pgrp), 0, 0, 0, "pgrppl", 209 &pool_allocator_nointr); 210 POOL_INIT(plimit_pool, sizeof(struct plimit), 0, 0, 0, "plimitpl", 211 &pool_allocator_nointr); 212 POOL_INIT(pstats_pool, sizeof(struct pstats), 0, 0, 0, "pstatspl", 213 &pool_allocator_nointr); 214 POOL_INIT(rusage_pool, sizeof(struct rusage), 0, 0, 0, "rusgepl", 215 &pool_allocator_nointr); 216 POOL_INIT(ras_pool, sizeof(struct ras), 0, 0, 0, "raspl", 217 &pool_allocator_nointr); 218 POOL_INIT(session_pool, sizeof(struct session), 0, 0, 0, "sessionpl", 219 &pool_allocator_nointr); 220 221 MALLOC_DEFINE(M_EMULDATA, "emuldata", "Per-process emulation data"); 222 MALLOC_DEFINE(M_PROC, "proc", "Proc structures"); 223 MALLOC_DEFINE(M_SUBPROC, "subproc", "Proc sub-structures"); 224 225 /* 226 * The process list descriptors, used during pid allocation and 227 * by sysctl. No locking on this data structure is needed since 228 * it is completely static. 229 */ 230 const struct proclist_desc proclists[] = { 231 { &allproc }, 232 { &zombproc }, 233 { NULL }, 234 }; 235 236 static void orphanpg(struct pgrp *); 237 static void pg_delete(pid_t); 238 239 /* 240 * Initialize global process hashing structures. 241 */ 242 void 243 procinit(void) 244 { 245 const struct proclist_desc *pd; 246 int i; 247 #define LINK_EMPTY ((PID_MAX + INITIAL_PID_TABLE_SIZE) & ~(INITIAL_PID_TABLE_SIZE - 1)) 248 249 for (pd = proclists; pd->pd_list != NULL; pd++) 250 LIST_INIT(pd->pd_list); 251 252 rw_init(&proclist_lock); 253 mutex_init(&proclist_mutex, MUTEX_SPIN, IPL_SCHED); 254 mutex_init(&alllwp_mutex, MUTEX_SPIN, IPL_SCHED); 255 256 pid_table = malloc(INITIAL_PID_TABLE_SIZE * sizeof *pid_table, 257 M_PROC, M_WAITOK); 258 /* Set free list running through table... 259 Preset 'use count' above PID_MAX so we allocate pid 1 next. */ 260 for (i = 0; i <= pid_tbl_mask; i++) { 261 pid_table[i].pt_proc = P_FREE(LINK_EMPTY + i + 1); 262 pid_table[i].pt_pgrp = 0; 263 } 264 /* slot 0 is just grabbed */ 265 next_free_pt = 1; 266 /* Need to fix last entry. */ 267 last_free_pt = pid_tbl_mask; 268 pid_table[last_free_pt].pt_proc = P_FREE(LINK_EMPTY); 269 /* point at which we grow table - to avoid reusing pids too often */ 270 pid_alloc_lim = pid_tbl_mask - 1; 271 #undef LINK_EMPTY 272 273 LIST_INIT(&alllwp); 274 275 uihashtbl = 276 hashinit(maxproc / 16, HASH_LIST, M_PROC, M_WAITOK, &uihash); 277 } 278 279 /* 280 * Initialize process 0. 281 */ 282 void 283 proc0_init(void) 284 { 285 struct proc *p; 286 struct pgrp *pg; 287 struct session *sess; 288 struct lwp *l; 289 u_int i; 290 rlim_t lim; 291 292 p = &proc0; 293 pg = &pgrp0; 294 sess = &session0; 295 l = &lwp0; 296 297 simple_lock_init(&p->p_lock); 298 mutex_init(&p->p_crmutex, MUTEX_DEFAULT, IPL_NONE); 299 LIST_INIT(&p->p_lwps); 300 LIST_INSERT_HEAD(&p->p_lwps, l, l_sibling); 301 p->p_nlwps = 1; 302 simple_lock_init(&p->p_sigctx.ps_silock); 303 CIRCLEQ_INIT(&p->p_sigctx.ps_siginfo); 304 305 pid_table[0].pt_proc = p; 306 LIST_INSERT_HEAD(&allproc, p, p_list); 307 LIST_INSERT_HEAD(&alllwp, l, l_list); 308 309 p->p_pgrp = pg; 310 pid_table[0].pt_pgrp = pg; 311 LIST_INIT(&pg->pg_members); 312 LIST_INSERT_HEAD(&pg->pg_members, p, p_pglist); 313 314 pg->pg_session = sess; 315 sess->s_count = 1; 316 sess->s_sid = 0; 317 sess->s_leader = p; 318 319 /* 320 * Set P_NOCLDWAIT so that kernel threads are reparented to 321 * init(8) when they exit. init(8) can easily wait them out 322 * for us. 323 */ 324 p->p_flag = P_SYSTEM | P_NOCLDWAIT; 325 p->p_stat = SACTIVE; 326 p->p_nice = NZERO; 327 p->p_emul = &emul_netbsd; 328 #ifdef __HAVE_SYSCALL_INTERN 329 (*p->p_emul->e_syscall_intern)(p); 330 #endif 331 strncpy(p->p_comm, "swapper", MAXCOMLEN); 332 333 l->l_flag = L_INMEM; 334 l->l_stat = LSONPROC; 335 l->l_ts = &turnstile0; 336 p->p_nrlwps = 1; 337 338 callout_init(&l->l_tsleep_ch); 339 340 /* Create credentials. */ 341 cred0 = kauth_cred_alloc(); 342 p->p_cred = cred0; 343 kauth_cred_hold(cred0); 344 l->l_cred = cred0; 345 346 /* Create the CWD info. */ 347 p->p_cwdi = &cwdi0; 348 cwdi0.cwdi_cmask = cmask; 349 cwdi0.cwdi_refcnt = 1; 350 simple_lock_init(&cwdi0.cwdi_slock); 351 352 /* Create the limits structures. */ 353 p->p_limit = &limit0; 354 simple_lock_init(&limit0.p_slock); 355 for (i = 0; i < sizeof(p->p_rlimit)/sizeof(p->p_rlimit[0]); i++) 356 limit0.pl_rlimit[i].rlim_cur = 357 limit0.pl_rlimit[i].rlim_max = RLIM_INFINITY; 358 359 limit0.pl_rlimit[RLIMIT_NOFILE].rlim_max = maxfiles; 360 limit0.pl_rlimit[RLIMIT_NOFILE].rlim_cur = 361 maxfiles < nofile ? maxfiles : nofile; 362 363 limit0.pl_rlimit[RLIMIT_NPROC].rlim_max = maxproc; 364 limit0.pl_rlimit[RLIMIT_NPROC].rlim_cur = 365 maxproc < maxuprc ? maxproc : maxuprc; 366 367 lim = ptoa(uvmexp.free); 368 limit0.pl_rlimit[RLIMIT_RSS].rlim_max = lim; 369 limit0.pl_rlimit[RLIMIT_MEMLOCK].rlim_max = lim; 370 limit0.pl_rlimit[RLIMIT_MEMLOCK].rlim_cur = lim / 3; 371 limit0.pl_corename = defcorename; 372 limit0.p_refcnt = 1; 373 374 /* Configure virtual memory system, set vm rlimits. */ 375 uvm_init_limits(p); 376 377 /* Initialize file descriptor table for proc0. */ 378 p->p_fd = &filedesc0.fd_fd; 379 fdinit1(&filedesc0); 380 381 /* 382 * Initialize proc0's vmspace, which uses the kernel pmap. 383 * All kernel processes (which never have user space mappings) 384 * share proc0's vmspace, and thus, the kernel pmap. 385 */ 386 uvmspace_init(&vmspace0, pmap_kernel(), round_page(VM_MIN_ADDRESS), 387 trunc_page(VM_MAX_ADDRESS)); 388 p->p_vmspace = &vmspace0; 389 390 l->l_addr = proc0paddr; /* XXX */ 391 392 p->p_stats = &pstat0; 393 394 /* Initialize signal state for proc0. */ 395 p->p_sigacts = &sigacts0; 396 siginit(p); 397 } 398 399 /* 400 * Check that the specified process group is in the session of the 401 * specified process. 402 * Treats -ve ids as process ids. 403 * Used to validate TIOCSPGRP requests. 404 */ 405 int 406 pgid_in_session(struct proc *p, pid_t pg_id) 407 { 408 struct pgrp *pgrp; 409 410 if (pg_id < 0) { 411 struct proc *p1 = pfind(-pg_id); 412 if (p1 == NULL) 413 return EINVAL; 414 pgrp = p1->p_pgrp; 415 } else { 416 pgrp = pgfind(pg_id); 417 if (pgrp == NULL) 418 return EINVAL; 419 } 420 if (pgrp->pg_session != p->p_pgrp->pg_session) 421 return EPERM; 422 return 0; 423 } 424 425 /* 426 * Is p an inferior of q? 427 * 428 * Call with the proclist_lock held. 429 */ 430 int 431 inferior(struct proc *p, struct proc *q) 432 { 433 434 for (; p != q; p = p->p_pptr) 435 if (p->p_pid == 0) 436 return 0; 437 return 1; 438 } 439 440 /* 441 * Locate a process by number 442 */ 443 struct proc * 444 p_find(pid_t pid, uint flags) 445 { 446 struct proc *p; 447 char stat; 448 449 if (!(flags & PFIND_LOCKED)) 450 rw_enter(&proclist_lock, RW_READER); 451 452 p = pid_table[pid & pid_tbl_mask].pt_proc; 453 /* Only allow live processes to be found by pid. */ 454 if (P_VALID(p) && p->p_pid == pid && 455 ((stat = p->p_stat) == SACTIVE || stat == SSTOP 456 || (stat == SZOMB && (flags & PFIND_ZOMBIE)))) { 457 if (flags & PFIND_UNLOCK_OK) 458 rw_exit(&proclist_lock); 459 return p; 460 } 461 if (flags & PFIND_UNLOCK_FAIL) 462 rw_exit(&proclist_lock); 463 return NULL; 464 } 465 466 467 /* 468 * Locate a process group by number 469 */ 470 struct pgrp * 471 pg_find(pid_t pgid, uint flags) 472 { 473 struct pgrp *pg; 474 475 if (!(flags & PFIND_LOCKED)) 476 rw_enter(&proclist_lock, RW_READER); 477 pg = pid_table[pgid & pid_tbl_mask].pt_pgrp; 478 /* 479 * Can't look up a pgrp that only exists because the session 480 * hasn't died yet (traditional) 481 */ 482 if (pg == NULL || pg->pg_id != pgid || LIST_EMPTY(&pg->pg_members)) { 483 if (flags & PFIND_UNLOCK_FAIL) 484 rw_exit(&proclist_lock); 485 return NULL; 486 } 487 488 if (flags & PFIND_UNLOCK_OK) 489 rw_exit(&proclist_lock); 490 return pg; 491 } 492 493 static void 494 expand_pid_table(void) 495 { 496 uint pt_size = pid_tbl_mask + 1; 497 struct pid_table *n_pt, *new_pt; 498 struct proc *proc; 499 struct pgrp *pgrp; 500 int i; 501 pid_t pid; 502 503 new_pt = malloc(pt_size * 2 * sizeof *new_pt, M_PROC, M_WAITOK); 504 505 rw_enter(&proclist_lock, RW_WRITER); 506 if (pt_size != pid_tbl_mask + 1) { 507 /* Another process beat us to it... */ 508 rw_exit(&proclist_lock); 509 FREE(new_pt, M_PROC); 510 return; 511 } 512 513 /* 514 * Copy entries from old table into new one. 515 * If 'pid' is 'odd' we need to place in the upper half, 516 * even pid's to the lower half. 517 * Free items stay in the low half so we don't have to 518 * fixup the reference to them. 519 * We stuff free items on the front of the freelist 520 * because we can't write to unmodified entries. 521 * Processing the table backwards maintains a semblance 522 * of issueing pid numbers that increase with time. 523 */ 524 i = pt_size - 1; 525 n_pt = new_pt + i; 526 for (; ; i--, n_pt--) { 527 proc = pid_table[i].pt_proc; 528 pgrp = pid_table[i].pt_pgrp; 529 if (!P_VALID(proc)) { 530 /* Up 'use count' so that link is valid */ 531 pid = (P_NEXT(proc) + pt_size) & ~pt_size; 532 proc = P_FREE(pid); 533 if (pgrp) 534 pid = pgrp->pg_id; 535 } else 536 pid = proc->p_pid; 537 538 /* Save entry in appropriate half of table */ 539 n_pt[pid & pt_size].pt_proc = proc; 540 n_pt[pid & pt_size].pt_pgrp = pgrp; 541 542 /* Put other piece on start of free list */ 543 pid = (pid ^ pt_size) & ~pid_tbl_mask; 544 n_pt[pid & pt_size].pt_proc = 545 P_FREE((pid & ~pt_size) | next_free_pt); 546 n_pt[pid & pt_size].pt_pgrp = 0; 547 next_free_pt = i | (pid & pt_size); 548 if (i == 0) 549 break; 550 } 551 552 /* Switch tables */ 553 n_pt = pid_table; 554 pid_table = new_pt; 555 pid_tbl_mask = pt_size * 2 - 1; 556 557 /* 558 * pid_max starts as PID_MAX (= 30000), once we have 16384 559 * allocated pids we need it to be larger! 560 */ 561 if (pid_tbl_mask > PID_MAX) { 562 pid_max = pid_tbl_mask * 2 + 1; 563 pid_alloc_lim |= pid_alloc_lim << 1; 564 } else 565 pid_alloc_lim <<= 1; /* doubles number of free slots... */ 566 567 rw_exit(&proclist_lock); 568 FREE(n_pt, M_PROC); 569 } 570 571 struct proc * 572 proc_alloc(void) 573 { 574 struct proc *p; 575 int nxt; 576 pid_t pid; 577 struct pid_table *pt; 578 579 p = pool_get(&proc_pool, PR_WAITOK); 580 p->p_stat = SIDL; /* protect against others */ 581 582 /* allocate next free pid */ 583 584 for (;;expand_pid_table()) { 585 if (__predict_false(pid_alloc_cnt >= pid_alloc_lim)) 586 /* ensure pids cycle through 2000+ values */ 587 continue; 588 rw_enter(&proclist_lock, RW_WRITER); 589 pt = &pid_table[next_free_pt]; 590 #ifdef DIAGNOSTIC 591 if (__predict_false(P_VALID(pt->pt_proc) || pt->pt_pgrp)) 592 panic("proc_alloc: slot busy"); 593 #endif 594 nxt = P_NEXT(pt->pt_proc); 595 if (nxt & pid_tbl_mask) 596 break; 597 /* Table full - expand (NB last entry not used....) */ 598 rw_exit(&proclist_lock); 599 } 600 601 /* pid is 'saved use count' + 'size' + entry */ 602 pid = (nxt & ~pid_tbl_mask) + pid_tbl_mask + 1 + next_free_pt; 603 if ((uint)pid > (uint)pid_max) 604 pid &= pid_tbl_mask; 605 p->p_pid = pid; 606 next_free_pt = nxt & pid_tbl_mask; 607 608 /* Grab table slot */ 609 pt->pt_proc = p; 610 pid_alloc_cnt++; 611 612 rw_exit(&proclist_lock); 613 614 return p; 615 } 616 617 /* 618 * Free last resources of a process - called from proc_free (in kern_exit.c) 619 * 620 * Called with the proclist_lock write held, and releases upon exit. 621 */ 622 void 623 proc_free_mem(struct proc *p) 624 { 625 pid_t pid = p->p_pid; 626 struct pid_table *pt; 627 628 LOCK_ASSERT(rw_write_held(&proclist_lock)); 629 630 pt = &pid_table[pid & pid_tbl_mask]; 631 #ifdef DIAGNOSTIC 632 if (__predict_false(pt->pt_proc != p)) 633 panic("proc_free: pid_table mismatch, pid %x, proc %p", 634 pid, p); 635 #endif 636 /* save pid use count in slot */ 637 pt->pt_proc = P_FREE(pid & ~pid_tbl_mask); 638 639 if (pt->pt_pgrp == NULL) { 640 /* link last freed entry onto ours */ 641 pid &= pid_tbl_mask; 642 pt = &pid_table[last_free_pt]; 643 pt->pt_proc = P_FREE(P_NEXT(pt->pt_proc) | pid); 644 last_free_pt = pid; 645 pid_alloc_cnt--; 646 } 647 648 nprocs--; 649 rw_exit(&proclist_lock); 650 651 pool_put(&proc_pool, p); 652 } 653 654 /* 655 * Move p to a new or existing process group (and session) 656 * 657 * If we are creating a new pgrp, the pgid should equal 658 * the calling process' pid. 659 * If is only valid to enter a process group that is in the session 660 * of the process. 661 * Also mksess should only be set if we are creating a process group 662 * 663 * Only called from sys_setsid, sys_setpgid/sys_setpgrp and the 664 * SYSV setpgrp support for hpux. 665 */ 666 int 667 enterpgrp(struct proc *curp, pid_t pid, pid_t pgid, int mksess) 668 { 669 struct pgrp *new_pgrp, *pgrp; 670 struct session *sess; 671 struct proc *p; 672 int rval; 673 pid_t pg_id = NO_PGID; 674 675 /* Allocate data areas we might need before doing any validity checks */ 676 rw_enter(&proclist_lock, RW_READER); /* Because pid_table might change */ 677 if (pid_table[pgid & pid_tbl_mask].pt_pgrp == 0) { 678 rw_exit(&proclist_lock); 679 new_pgrp = pool_get(&pgrp_pool, PR_WAITOK); 680 } else { 681 rw_exit(&proclist_lock); 682 new_pgrp = NULL; 683 } 684 if (mksess) 685 sess = pool_get(&session_pool, M_WAITOK); 686 else 687 sess = NULL; 688 689 rw_enter(&proclist_lock, RW_WRITER); 690 rval = EPERM; /* most common error (to save typing) */ 691 692 /* Check pgrp exists or can be created */ 693 pgrp = pid_table[pgid & pid_tbl_mask].pt_pgrp; 694 if (pgrp != NULL && pgrp->pg_id != pgid) 695 goto done; 696 697 /* Can only set another process under restricted circumstances. */ 698 if (pid != curp->p_pid) { 699 /* must exist and be one of our children... */ 700 if ((p = p_find(pid, PFIND_LOCKED)) == NULL || 701 !inferior(p, curp)) { 702 rval = ESRCH; 703 goto done; 704 } 705 /* ... in the same session... */ 706 if (sess != NULL || p->p_session != curp->p_session) 707 goto done; 708 /* ... existing pgid must be in same session ... */ 709 if (pgrp != NULL && pgrp->pg_session != p->p_session) 710 goto done; 711 /* ... and not done an exec. */ 712 if (p->p_flag & P_EXEC) { 713 rval = EACCES; 714 goto done; 715 } 716 } else { 717 /* ... setsid() cannot re-enter a pgrp */ 718 if (mksess && (curp->p_pgid == curp->p_pid || 719 pg_find(curp->p_pid, PFIND_LOCKED))) 720 goto done; 721 p = curp; 722 } 723 724 /* Changing the process group/session of a session 725 leader is definitely off limits. */ 726 if (SESS_LEADER(p)) { 727 if (sess == NULL && p->p_pgrp == pgrp) 728 /* unless it's a definite noop */ 729 rval = 0; 730 goto done; 731 } 732 733 /* Can only create a process group with id of process */ 734 if (pgrp == NULL && pgid != pid) 735 goto done; 736 737 /* Can only create a session if creating pgrp */ 738 if (sess != NULL && pgrp != NULL) 739 goto done; 740 741 /* Check we allocated memory for a pgrp... */ 742 if (pgrp == NULL && new_pgrp == NULL) 743 goto done; 744 745 /* Don't attach to 'zombie' pgrp */ 746 if (pgrp != NULL && LIST_EMPTY(&pgrp->pg_members)) 747 goto done; 748 749 /* Expect to succeed now */ 750 rval = 0; 751 752 if (pgrp == p->p_pgrp) 753 /* nothing to do */ 754 goto done; 755 756 /* Ok all setup, link up required structures */ 757 if (pgrp == NULL) { 758 pgrp = new_pgrp; 759 new_pgrp = 0; 760 if (sess != NULL) { 761 sess->s_sid = p->p_pid; 762 sess->s_leader = p; 763 sess->s_count = 1; 764 sess->s_ttyvp = NULL; 765 sess->s_ttyp = NULL; 766 sess->s_flags = p->p_session->s_flags & ~S_LOGIN_SET; 767 memcpy(sess->s_login, p->p_session->s_login, 768 sizeof(sess->s_login)); 769 p->p_flag &= ~P_CONTROLT; 770 } else { 771 sess = p->p_pgrp->pg_session; 772 SESSHOLD(sess); 773 } 774 pgrp->pg_session = sess; 775 sess = 0; 776 777 pgrp->pg_id = pgid; 778 LIST_INIT(&pgrp->pg_members); 779 #ifdef DIAGNOSTIC 780 if (__predict_false(pid_table[pgid & pid_tbl_mask].pt_pgrp)) 781 panic("enterpgrp: pgrp table slot in use"); 782 if (__predict_false(mksess && p != curp)) 783 panic("enterpgrp: mksession and p != curproc"); 784 #endif 785 pid_table[pgid & pid_tbl_mask].pt_pgrp = pgrp; 786 pgrp->pg_jobc = 0; 787 } 788 789 #ifdef notyet 790 /* 791 * If there's a controlling terminal for the current session, we 792 * have to interlock with it. See ttread(). 793 */ 794 if (p->p_session->s_ttyvp != NULL) { 795 tp = p->p_session->s_ttyp; 796 mutex_enter(&tp->t_mutex); 797 } else 798 tp = NULL; 799 #endif 800 801 /* 802 * Adjust eligibility of affected pgrps to participate in job control. 803 * Increment eligibility counts before decrementing, otherwise we 804 * could reach 0 spuriously during the first call. 805 */ 806 fixjobc(p, pgrp, 1); 807 fixjobc(p, p->p_pgrp, 0); 808 809 /* Move process to requested group. */ 810 mutex_enter(&proclist_mutex); 811 LIST_REMOVE(p, p_pglist); 812 if (LIST_EMPTY(&p->p_pgrp->pg_members)) 813 /* defer delete until we've dumped the lock */ 814 pg_id = p->p_pgrp->pg_id; 815 p->p_pgrp = pgrp; 816 LIST_INSERT_HEAD(&pgrp->pg_members, p, p_pglist); 817 mutex_exit(&proclist_mutex); 818 819 #ifdef notyet 820 /* Done with the swap; we can release the tty mutex. */ 821 if (tp != NULL) 822 mutex_exit(&tp->t_mutex); 823 #endif 824 825 done: 826 rw_exit(&proclist_lock); 827 if (sess != NULL) 828 pool_put(&session_pool, sess); 829 if (new_pgrp != NULL) 830 pool_put(&pgrp_pool, new_pgrp); 831 if (pg_id != NO_PGID) 832 pg_delete(pg_id); 833 #ifdef DEBUG_PGRP 834 if (__predict_false(rval)) 835 printf("enterpgrp(%d,%d,%d), curproc %d, rval %d\n", 836 pid, pgid, mksess, curp->p_pid, rval); 837 #endif 838 return rval; 839 } 840 841 /* 842 * Remove a process from its process group. 843 */ 844 void 845 leavepgrp(struct proc *p) 846 { 847 struct pgrp *pgrp; 848 849 /* 850 * If there's a controlling terminal for the session, we have to 851 * interlock with it. See ttread(). 852 */ 853 rw_enter(&proclist_lock, RW_WRITER); 854 mutex_enter(&proclist_mutex); 855 856 #ifdef notyet 857 if (p_>p_session->s_ttyvp != NULL) { 858 tp = p->p_session->s_ttyp; 859 mutex_enter(&tp->t_mutex); 860 } else 861 tp = NULL; 862 #endif 863 864 pgrp = p->p_pgrp; 865 LIST_REMOVE(p, p_pglist); 866 p->p_pgrp = NULL; 867 868 #ifdef notyet 869 if (tp != NULL) 870 mutex_exit(&tp->t_mutex); 871 #endif 872 mutex_exit(&proclist_mutex); 873 rw_exit(&proclist_lock); 874 875 if (LIST_EMPTY(&pgrp->pg_members)) 876 pg_delete(pgrp->pg_id); 877 } 878 879 static void 880 pg_free(pid_t pg_id) 881 { 882 struct pgrp *pgrp; 883 struct pid_table *pt; 884 885 rw_enter(&proclist_lock, RW_WRITER); 886 pt = &pid_table[pg_id & pid_tbl_mask]; 887 pgrp = pt->pt_pgrp; 888 #ifdef DIAGNOSTIC 889 if (__predict_false(!pgrp || pgrp->pg_id != pg_id 890 || !LIST_EMPTY(&pgrp->pg_members))) 891 panic("pg_free: process group absent or has members"); 892 #endif 893 pt->pt_pgrp = 0; 894 895 if (!P_VALID(pt->pt_proc)) { 896 /* orphaned pgrp, put slot onto free list */ 897 #ifdef DIAGNOSTIC 898 if (__predict_false(P_NEXT(pt->pt_proc) & pid_tbl_mask)) 899 panic("pg_free: process slot on free list"); 900 #endif 901 902 pg_id &= pid_tbl_mask; 903 pt = &pid_table[last_free_pt]; 904 pt->pt_proc = P_FREE(P_NEXT(pt->pt_proc) | pg_id); 905 last_free_pt = pg_id; 906 pid_alloc_cnt--; 907 } 908 rw_exit(&proclist_lock); 909 910 pool_put(&pgrp_pool, pgrp); 911 } 912 913 /* 914 * delete a process group 915 */ 916 static void 917 pg_delete(pid_t pg_id) 918 { 919 struct pgrp *pgrp; 920 struct tty *ttyp; 921 struct session *ss; 922 int is_pgrp_leader; 923 924 rw_enter(&proclist_lock, RW_WRITER); 925 pgrp = pid_table[pg_id & pid_tbl_mask].pt_pgrp; 926 if (pgrp == NULL || pgrp->pg_id != pg_id || 927 !LIST_EMPTY(&pgrp->pg_members)) { 928 rw_exit(&proclist_lock); 929 return; 930 } 931 932 ss = pgrp->pg_session; 933 934 /* Remove reference (if any) from tty to this process group */ 935 ttyp = ss->s_ttyp; 936 if (ttyp != NULL && ttyp->t_pgrp == pgrp) { 937 ttyp->t_pgrp = NULL; 938 #ifdef DIAGNOSTIC 939 if (ttyp->t_session != ss) 940 panic("pg_delete: wrong session on terminal"); 941 #endif 942 } 943 944 /* 945 * The leading process group in a session is freed 946 * by sessdelete() if last reference. 947 */ 948 is_pgrp_leader = (ss->s_sid == pgrp->pg_id); 949 rw_exit(&proclist_lock); 950 SESSRELE(ss); 951 952 if (is_pgrp_leader) 953 return; 954 955 pg_free(pg_id); 956 } 957 958 /* 959 * Delete session - called from SESSRELE when s_count becomes zero. 960 */ 961 void 962 sessdelete(struct session *ss) 963 { 964 /* 965 * We keep the pgrp with the same id as the session in 966 * order to stop a process being given the same pid. 967 * Since the pgrp holds a reference to the session, it 968 * must be a 'zombie' pgrp by now. 969 */ 970 971 pg_free(ss->s_sid); 972 973 pool_put(&session_pool, ss); 974 } 975 976 /* 977 * Adjust pgrp jobc counters when specified process changes process group. 978 * We count the number of processes in each process group that "qualify" 979 * the group for terminal job control (those with a parent in a different 980 * process group of the same session). If that count reaches zero, the 981 * process group becomes orphaned. Check both the specified process' 982 * process group and that of its children. 983 * entering == 0 => p is leaving specified group. 984 * entering == 1 => p is entering specified group. 985 * 986 * Call with proclist_lock held. 987 */ 988 void 989 fixjobc(struct proc *p, struct pgrp *pgrp, int entering) 990 { 991 struct pgrp *hispgrp; 992 struct session *mysession = pgrp->pg_session; 993 struct proc *child; 994 995 /* 996 * Check p's parent to see whether p qualifies its own process 997 * group; if so, adjust count for p's process group. 998 */ 999 hispgrp = p->p_pptr->p_pgrp; 1000 if (hispgrp != pgrp && hispgrp->pg_session == mysession) { 1001 if (entering) 1002 pgrp->pg_jobc++; 1003 else if (--pgrp->pg_jobc == 0) 1004 orphanpg(pgrp); 1005 } 1006 1007 /* 1008 * Check this process' children to see whether they qualify 1009 * their process groups; if so, adjust counts for children's 1010 * process groups. 1011 */ 1012 LIST_FOREACH(child, &p->p_children, p_sibling) { 1013 hispgrp = child->p_pgrp; 1014 if (hispgrp != pgrp && hispgrp->pg_session == mysession && 1015 !P_ZOMBIE(child)) { 1016 if (entering) 1017 hispgrp->pg_jobc++; 1018 else if (--hispgrp->pg_jobc == 0) 1019 orphanpg(hispgrp); 1020 } 1021 } 1022 } 1023 1024 /* 1025 * A process group has become orphaned; 1026 * if there are any stopped processes in the group, 1027 * hang-up all process in that group. 1028 * 1029 * Call with proclist_lock held. 1030 */ 1031 static void 1032 orphanpg(struct pgrp *pg) 1033 { 1034 struct proc *p; 1035 1036 LIST_FOREACH(p, &pg->pg_members, p_pglist) { 1037 if (p->p_stat == SSTOP) { 1038 LIST_FOREACH(p, &pg->pg_members, p_pglist) { 1039 psignal(p, SIGHUP); 1040 psignal(p, SIGCONT); 1041 } 1042 return; 1043 } 1044 } 1045 } 1046 1047 /* mark process as suid/sgid, reset some values to defaults */ 1048 void 1049 p_sugid(struct proc *p) 1050 { 1051 struct plimit *lim; 1052 char *cn; 1053 1054 p->p_flag |= P_SUGID; 1055 /* reset what needs to be reset in plimit */ 1056 lim = p->p_limit; 1057 if (lim->pl_corename != defcorename) { 1058 if (lim->p_refcnt > 1 && 1059 (lim->p_lflags & PL_SHAREMOD) == 0) { 1060 p->p_limit = limcopy(lim); 1061 limfree(lim); 1062 lim = p->p_limit; 1063 } 1064 simple_lock(&lim->p_slock); 1065 cn = lim->pl_corename; 1066 lim->pl_corename = defcorename; 1067 simple_unlock(&lim->p_slock); 1068 if (cn != defcorename) 1069 free(cn, M_TEMP); 1070 } 1071 } 1072 1073 #ifdef DDB 1074 #include <ddb/db_output.h> 1075 void pidtbl_dump(void); 1076 void 1077 pidtbl_dump(void) 1078 { 1079 struct pid_table *pt; 1080 struct proc *p; 1081 struct pgrp *pgrp; 1082 int id; 1083 1084 db_printf("pid table %p size %x, next %x, last %x\n", 1085 pid_table, pid_tbl_mask+1, 1086 next_free_pt, last_free_pt); 1087 for (pt = pid_table, id = 0; id <= pid_tbl_mask; id++, pt++) { 1088 p = pt->pt_proc; 1089 if (!P_VALID(p) && !pt->pt_pgrp) 1090 continue; 1091 db_printf(" id %x: ", id); 1092 if (P_VALID(p)) 1093 db_printf("proc %p id %d (0x%x) %s\n", 1094 p, p->p_pid, p->p_pid, p->p_comm); 1095 else 1096 db_printf("next %x use %x\n", 1097 P_NEXT(p) & pid_tbl_mask, 1098 P_NEXT(p) & ~pid_tbl_mask); 1099 if ((pgrp = pt->pt_pgrp)) { 1100 db_printf("\tsession %p, sid %d, count %d, login %s\n", 1101 pgrp->pg_session, pgrp->pg_session->s_sid, 1102 pgrp->pg_session->s_count, 1103 pgrp->pg_session->s_login); 1104 db_printf("\tpgrp %p, pg_id %d, pg_jobc %d, members %p\n", 1105 pgrp, pgrp->pg_id, pgrp->pg_jobc, 1106 pgrp->pg_members.lh_first); 1107 for (p = pgrp->pg_members.lh_first; p != 0; 1108 p = p->p_pglist.le_next) { 1109 db_printf("\t\tpid %d addr %p pgrp %p %s\n", 1110 p->p_pid, p, p->p_pgrp, p->p_comm); 1111 } 1112 } 1113 } 1114 } 1115 #endif /* DDB */ 1116 1117 #ifdef KSTACK_CHECK_MAGIC 1118 #include <sys/user.h> 1119 1120 #define KSTACK_MAGIC 0xdeadbeaf 1121 1122 /* XXX should be per process basis? */ 1123 int kstackleftmin = KSTACK_SIZE; 1124 int kstackleftthres = KSTACK_SIZE / 8; /* warn if remaining stack is 1125 less than this */ 1126 1127 void 1128 kstack_setup_magic(const struct lwp *l) 1129 { 1130 uint32_t *ip; 1131 uint32_t const *end; 1132 1133 KASSERT(l != NULL); 1134 KASSERT(l != &lwp0); 1135 1136 /* 1137 * fill all the stack with magic number 1138 * so that later modification on it can be detected. 1139 */ 1140 ip = (uint32_t *)KSTACK_LOWEST_ADDR(l); 1141 end = (uint32_t *)((caddr_t)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE); 1142 for (; ip < end; ip++) { 1143 *ip = KSTACK_MAGIC; 1144 } 1145 } 1146 1147 void 1148 kstack_check_magic(const struct lwp *l) 1149 { 1150 uint32_t const *ip, *end; 1151 int stackleft; 1152 1153 KASSERT(l != NULL); 1154 1155 /* don't check proc0 */ /*XXX*/ 1156 if (l == &lwp0) 1157 return; 1158 1159 #ifdef __MACHINE_STACK_GROWS_UP 1160 /* stack grows upwards (eg. hppa) */ 1161 ip = (uint32_t *)((caddr_t)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE); 1162 end = (uint32_t *)KSTACK_LOWEST_ADDR(l); 1163 for (ip--; ip >= end; ip--) 1164 if (*ip != KSTACK_MAGIC) 1165 break; 1166 1167 stackleft = (caddr_t)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE - (caddr_t)ip; 1168 #else /* __MACHINE_STACK_GROWS_UP */ 1169 /* stack grows downwards (eg. i386) */ 1170 ip = (uint32_t *)KSTACK_LOWEST_ADDR(l); 1171 end = (uint32_t *)((caddr_t)KSTACK_LOWEST_ADDR(l) + KSTACK_SIZE); 1172 for (; ip < end; ip++) 1173 if (*ip != KSTACK_MAGIC) 1174 break; 1175 1176 stackleft = ((const char *)ip) - (const char *)KSTACK_LOWEST_ADDR(l); 1177 #endif /* __MACHINE_STACK_GROWS_UP */ 1178 1179 if (kstackleftmin > stackleft) { 1180 kstackleftmin = stackleft; 1181 if (stackleft < kstackleftthres) 1182 printf("warning: kernel stack left %d bytes" 1183 "(pid %u:lid %u)\n", stackleft, 1184 (u_int)l->l_proc->p_pid, (u_int)l->l_lid); 1185 } 1186 1187 if (stackleft <= 0) { 1188 panic("magic on the top of kernel stack changed for " 1189 "pid %u, lid %u: maybe kernel stack overflow", 1190 (u_int)l->l_proc->p_pid, (u_int)l->l_lid); 1191 } 1192 } 1193 #endif /* KSTACK_CHECK_MAGIC */ 1194 1195 /* 1196 * XXXSMP this is bust, it grabs a read lock and then messes about 1197 * with allproc. 1198 */ 1199 int 1200 proclist_foreach_call(struct proclist *list, 1201 int (*callback)(struct proc *, void *arg), void *arg) 1202 { 1203 struct proc marker; 1204 struct proc *p; 1205 struct lwp * const l = curlwp; 1206 int ret = 0; 1207 1208 marker.p_flag = P_MARKER; 1209 PHOLD(l); 1210 rw_enter(&proclist_lock, RW_READER); 1211 for (p = LIST_FIRST(list); ret == 0 && p != NULL;) { 1212 if (p->p_flag & P_MARKER) { 1213 p = LIST_NEXT(p, p_list); 1214 continue; 1215 } 1216 LIST_INSERT_AFTER(p, &marker, p_list); 1217 ret = (*callback)(p, arg); 1218 KASSERT(rw_read_held(&proclist_lock)); 1219 p = LIST_NEXT(&marker, p_list); 1220 LIST_REMOVE(&marker, p_list); 1221 } 1222 rw_exit(&proclist_lock); 1223 PRELE(l); 1224 1225 return ret; 1226 } 1227 1228 int 1229 proc_vmspace_getref(struct proc *p, struct vmspace **vm) 1230 { 1231 1232 /* XXXCDC: how should locking work here? */ 1233 1234 /* curproc exception is for coredump. */ 1235 1236 if ((p != curproc && (p->p_flag & P_WEXIT) != 0) || 1237 (p->p_vmspace->vm_refcnt < 1)) { /* XXX */ 1238 return EFAULT; 1239 } 1240 1241 uvmspace_addref(p->p_vmspace); 1242 *vm = p->p_vmspace; 1243 1244 return 0; 1245 } 1246 1247 /* 1248 * Acquire a write lock on the process credential. 1249 */ 1250 void 1251 proc_crmod_enter(struct proc *p) 1252 { 1253 1254 mutex_enter(&p->p_crmutex); 1255 } 1256 1257 /* 1258 * Block out readers, set in a new process credential, and drop the write 1259 * lock. The credential must have a reference already. Optionally, free a 1260 * no-longer required credential. 1261 */ 1262 void 1263 proc_crmod_leave(struct proc *p, kauth_cred_t scred, kauth_cred_t fcred) 1264 { 1265 1266 p->p_cred = scred; 1267 mutex_exit(&p->p_crmutex); 1268 if (fcred != NULL) 1269 kauth_cred_free(fcred); 1270 } 1271
Indexes created Sun Sep 21 20:09:37 GMT 2025