1178479Sjb/* 2178479Sjb * CDDL HEADER START 3178479Sjb * 4178479Sjb * The contents of this file are subject to the terms of the 5178479Sjb * Common Development and Distribution License (the "License"). 6178479Sjb * You may not use this file except in compliance with the License. 7178479Sjb * 8178479Sjb * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9178479Sjb * or http://www.opensolaris.org/os/licensing. 10178479Sjb * See the License for the specific language governing permissions 11178479Sjb * and limitations under the License. 12178479Sjb * 13178479Sjb * When distributing Covered Code, include this CDDL HEADER in each 14178479Sjb * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15178479Sjb * If applicable, add the following below this CDDL HEADER, with the 16178479Sjb * fields enclosed by brackets "[]" replaced with your own identifying 17178479Sjb * information: Portions Copyright [yyyy] [name of copyright owner] 18178479Sjb * 19178479Sjb * CDDL HEADER END 20178479Sjb */ 21178479Sjb 22178479Sjb/* 23210767Srpaulo * Copyright 2010 Sun Microsystems, Inc. All rights reserved. 24178479Sjb * Use is subject to license terms. 25178479Sjb */ 26178479Sjb 27178479Sjb/* 28178479Sjb * DTrace Process Control 29178479Sjb * 30178479Sjb * This file provides a set of routines that permit libdtrace and its clients 31178479Sjb * to create and grab process handles using libproc, and to share these handles 32178479Sjb * between library mechanisms that need libproc access, such as ustack(), and 33178479Sjb * client mechanisms that need libproc access, such as dtrace(1M) -c and -p. 34178479Sjb * The library provides several mechanisms in the libproc control layer: 35178479Sjb * 36178479Sjb * Reference Counting: The library code and client code can independently grab 37178479Sjb * the same process handles without interfering with one another. Only when 38178479Sjb * the reference count drops to zero and the handle is not being cached (see 39178479Sjb * below for more information on caching) will Prelease() be called on it. 40178479Sjb * 41178479Sjb * Handle Caching: If a handle is grabbed PGRAB_RDONLY (e.g. by ustack()) and 42178479Sjb * the reference count drops to zero, the handle is not immediately released. 43178479Sjb * Instead, libproc handles are maintained on dph_lrulist in order from most- 44178479Sjb * recently accessed to least-recently accessed. Idle handles are maintained 45178479Sjb * until a pre-defined LRU cache limit is exceeded, permitting repeated calls 46178479Sjb * to ustack() to avoid the overhead of releasing and re-grabbing processes. 47178479Sjb * 48178479Sjb * Process Control: For processes that are grabbed for control (~PGRAB_RDONLY) 49178479Sjb * or created by dt_proc_create(), a control thread is created to provide 50178479Sjb * callbacks on process exit and symbol table caching on dlopen()s. 51178479Sjb * 52178479Sjb * MT-Safety: Libproc is not MT-Safe, so dt_proc_lock() and dt_proc_unlock() 53178479Sjb * are provided to synchronize access to the libproc handle between libdtrace 54178479Sjb * code and client code and the control thread's use of the ps_prochandle. 55178479Sjb * 56178479Sjb * NOTE: MT-Safety is NOT provided for libdtrace itself, or for use of the 57178479Sjb * dtrace_proc_grab/dtrace_proc_create mechanisms. Like all exported libdtrace 58178479Sjb * calls, these are assumed to be MT-Unsafe. MT-Safety is ONLY provided for 59178479Sjb * synchronization between libdtrace control threads and the client thread. 60178479Sjb * 61178479Sjb * The ps_prochandles themselves are maintained along with a dt_proc_t struct 62178479Sjb * in a hash table indexed by PID. This provides basic locking and reference 63178479Sjb * counting. The dt_proc_t is also maintained in LRU order on dph_lrulist. 64178479Sjb * The dph_lrucnt and dph_lrulim count the number of cacheable processes and 65178479Sjb * the current limit on the number of actively cached entries. 66178479Sjb * 67178479Sjb * The control thread for a process establishes breakpoints at the rtld_db 68178479Sjb * locations of interest, updates mappings and symbol tables at these points, 69178479Sjb * and handles exec and fork (by always following the parent). The control 70178479Sjb * thread automatically exits when the process dies or control is lost. 71178479Sjb * 72178479Sjb * A simple notification mechanism is provided for libdtrace clients using 73178479Sjb * dtrace_handle_proc() for notification of PS_UNDEAD or PS_LOST events. If 74178479Sjb * such an event occurs, the dt_proc_t itself is enqueued on a notification 75178479Sjb * list and the control thread broadcasts to dph_cv. dtrace_sleep() will wake 76178479Sjb * up using this condition and will then call the client handler as necessary. 77178479Sjb */ 78178479Sjb 79178479Sjb#include <sys/wait.h> 80277300Ssmh#ifdef illumos 81178479Sjb#include <sys/lwp.h> 82178565Sjb#endif 83178479Sjb#include <strings.h> 84178479Sjb#include <signal.h> 85178479Sjb#include <assert.h> 86178479Sjb#include <errno.h> 87178479Sjb 88178479Sjb#include <dt_proc.h> 89178479Sjb#include <dt_pid.h> 90178479Sjb#include <dt_impl.h> 91178479Sjb 92277300Ssmh#ifndef illumos 93211554Srpaulo#include <sys/syscall.h> 94211554Srpaulo#include <libproc_compat.h> 95211554Srpaulo#define SYS_forksys SYS_fork 96211554Srpaulo#endif 97211554Srpaulo 98210767Srpaulo#define IS_SYS_EXEC(w) (w == SYS_execve) 99210767Srpaulo#define IS_SYS_FORK(w) (w == SYS_vfork || w == SYS_forksys) 100178479Sjb 101178479Sjbstatic dt_bkpt_t * 102178479Sjbdt_proc_bpcreate(dt_proc_t *dpr, uintptr_t addr, dt_bkpt_f *func, void *data) 103178479Sjb{ 104178479Sjb struct ps_prochandle *P = dpr->dpr_proc; 105178479Sjb dt_bkpt_t *dbp; 106178479Sjb 107210775Srpaulo assert(DT_MUTEX_HELD(&dpr->dpr_lock)); 108178479Sjb 109178479Sjb if ((dbp = dt_zalloc(dpr->dpr_hdl, sizeof (dt_bkpt_t))) != NULL) { 110178479Sjb dbp->dbp_func = func; 111178479Sjb dbp->dbp_data = data; 112178479Sjb dbp->dbp_addr = addr; 113178479Sjb 114178479Sjb if (Psetbkpt(P, dbp->dbp_addr, &dbp->dbp_instr) == 0) 115178479Sjb dbp->dbp_active = B_TRUE; 116178479Sjb 117178479Sjb dt_list_append(&dpr->dpr_bps, dbp); 118178479Sjb } 119178479Sjb 120178479Sjb return (dbp); 121178479Sjb} 122178479Sjb 123178479Sjbstatic void 124178479Sjbdt_proc_bpdestroy(dt_proc_t *dpr, int delbkpts) 125178479Sjb{ 126178479Sjb int state = Pstate(dpr->dpr_proc); 127178479Sjb dt_bkpt_t *dbp, *nbp; 128178479Sjb 129210775Srpaulo assert(DT_MUTEX_HELD(&dpr->dpr_lock)); 130178479Sjb 131178479Sjb for (dbp = dt_list_next(&dpr->dpr_bps); dbp != NULL; dbp = nbp) { 132178479Sjb if (delbkpts && dbp->dbp_active && 133178479Sjb state != PS_LOST && state != PS_UNDEAD) { 134178479Sjb (void) Pdelbkpt(dpr->dpr_proc, 135178479Sjb dbp->dbp_addr, dbp->dbp_instr); 136178479Sjb } 137178479Sjb nbp = dt_list_next(dbp); 138178479Sjb dt_list_delete(&dpr->dpr_bps, dbp); 139178479Sjb dt_free(dpr->dpr_hdl, dbp); 140178479Sjb } 141178479Sjb} 142178479Sjb 143178479Sjbstatic void 144178479Sjbdt_proc_bpmatch(dtrace_hdl_t *dtp, dt_proc_t *dpr) 145178479Sjb{ 146277300Ssmh#ifdef illumos 147178479Sjb const lwpstatus_t *psp = &Pstatus(dpr->dpr_proc)->pr_lwp; 148211554Srpaulo#else 149211554Srpaulo unsigned long pc; 150211554Srpaulo#endif 151178479Sjb dt_bkpt_t *dbp; 152178479Sjb 153210775Srpaulo assert(DT_MUTEX_HELD(&dpr->dpr_lock)); 154178479Sjb 155277300Ssmh#ifndef illumos 156211554Srpaulo proc_regget(dpr->dpr_proc, REG_PC, &pc); 157211554Srpaulo proc_bkptregadj(&pc); 158211554Srpaulo#endif 159211554Srpaulo 160178479Sjb for (dbp = dt_list_next(&dpr->dpr_bps); 161178479Sjb dbp != NULL; dbp = dt_list_next(dbp)) { 162277300Ssmh#ifdef illumos 163178479Sjb if (psp->pr_reg[R_PC] == dbp->dbp_addr) 164178479Sjb break; 165211554Srpaulo#else 166211554Srpaulo if (pc == dbp->dbp_addr) 167211554Srpaulo break; 168211554Srpaulo#endif 169178479Sjb } 170178479Sjb 171178479Sjb if (dbp == NULL) { 172178479Sjb dt_dprintf("pid %d: spurious breakpoint wakeup for %lx\n", 173277300Ssmh#ifdef illumos 174178479Sjb (int)dpr->dpr_pid, (ulong_t)psp->pr_reg[R_PC]); 175211554Srpaulo#else 176211554Srpaulo (int)dpr->dpr_pid, pc); 177211554Srpaulo#endif 178178479Sjb return; 179178479Sjb } 180178479Sjb 181178479Sjb dt_dprintf("pid %d: hit breakpoint at %lx (%lu)\n", 182178479Sjb (int)dpr->dpr_pid, (ulong_t)dbp->dbp_addr, ++dbp->dbp_hits); 183178479Sjb 184178479Sjb dbp->dbp_func(dtp, dpr, dbp->dbp_data); 185178479Sjb (void) Pxecbkpt(dpr->dpr_proc, dbp->dbp_instr); 186178479Sjb} 187178479Sjb 188178479Sjbstatic void 189178479Sjbdt_proc_bpenable(dt_proc_t *dpr) 190178479Sjb{ 191178479Sjb dt_bkpt_t *dbp; 192178479Sjb 193210775Srpaulo assert(DT_MUTEX_HELD(&dpr->dpr_lock)); 194178479Sjb 195178479Sjb for (dbp = dt_list_next(&dpr->dpr_bps); 196178479Sjb dbp != NULL; dbp = dt_list_next(dbp)) { 197178479Sjb if (!dbp->dbp_active && Psetbkpt(dpr->dpr_proc, 198178479Sjb dbp->dbp_addr, &dbp->dbp_instr) == 0) 199178479Sjb dbp->dbp_active = B_TRUE; 200178479Sjb } 201178479Sjb 202178479Sjb dt_dprintf("breakpoints enabled\n"); 203178479Sjb} 204178479Sjb 205178479Sjbstatic void 206178479Sjbdt_proc_bpdisable(dt_proc_t *dpr) 207178479Sjb{ 208178479Sjb dt_bkpt_t *dbp; 209178479Sjb 210210775Srpaulo assert(DT_MUTEX_HELD(&dpr->dpr_lock)); 211178479Sjb 212178479Sjb for (dbp = dt_list_next(&dpr->dpr_bps); 213178479Sjb dbp != NULL; dbp = dt_list_next(dbp)) { 214178479Sjb if (dbp->dbp_active && Pdelbkpt(dpr->dpr_proc, 215178479Sjb dbp->dbp_addr, dbp->dbp_instr) == 0) 216178479Sjb dbp->dbp_active = B_FALSE; 217178479Sjb } 218178479Sjb 219178479Sjb dt_dprintf("breakpoints disabled\n"); 220178479Sjb} 221178479Sjb 222178479Sjbstatic void 223178479Sjbdt_proc_notify(dtrace_hdl_t *dtp, dt_proc_hash_t *dph, dt_proc_t *dpr, 224178479Sjb const char *msg) 225178479Sjb{ 226178479Sjb dt_proc_notify_t *dprn = dt_alloc(dtp, sizeof (dt_proc_notify_t)); 227178479Sjb 228178479Sjb if (dprn == NULL) { 229178479Sjb dt_dprintf("failed to allocate notification for %d %s\n", 230178479Sjb (int)dpr->dpr_pid, msg); 231178479Sjb } else { 232178479Sjb dprn->dprn_dpr = dpr; 233178479Sjb if (msg == NULL) 234178479Sjb dprn->dprn_errmsg[0] = '\0'; 235178479Sjb else 236178479Sjb (void) strlcpy(dprn->dprn_errmsg, msg, 237178479Sjb sizeof (dprn->dprn_errmsg)); 238178479Sjb 239178479Sjb (void) pthread_mutex_lock(&dph->dph_lock); 240178479Sjb 241178479Sjb dprn->dprn_next = dph->dph_notify; 242178479Sjb dph->dph_notify = dprn; 243178479Sjb 244178479Sjb (void) pthread_cond_broadcast(&dph->dph_cv); 245178479Sjb (void) pthread_mutex_unlock(&dph->dph_lock); 246178479Sjb } 247178479Sjb} 248178479Sjb 249178479Sjb/* 250178479Sjb * Check to see if the control thread was requested to stop when the victim 251178479Sjb * process reached a particular event (why) rather than continuing the victim. 252178479Sjb * If 'why' is set in the stop mask, we wait on dpr_cv for dt_proc_continue(). 253178479Sjb * If 'why' is not set, this function returns immediately and does nothing. 254178479Sjb */ 255178479Sjbstatic void 256178479Sjbdt_proc_stop(dt_proc_t *dpr, uint8_t why) 257178479Sjb{ 258210775Srpaulo assert(DT_MUTEX_HELD(&dpr->dpr_lock)); 259178479Sjb assert(why != DT_PROC_STOP_IDLE); 260178479Sjb 261178479Sjb if (dpr->dpr_stop & why) { 262178479Sjb dpr->dpr_stop |= DT_PROC_STOP_IDLE; 263178479Sjb dpr->dpr_stop &= ~why; 264178479Sjb 265178479Sjb (void) pthread_cond_broadcast(&dpr->dpr_cv); 266178479Sjb 267178479Sjb /* 268178479Sjb * We disable breakpoints while stopped to preserve the 269178479Sjb * integrity of the program text for both our own disassembly 270178479Sjb * and that of the kernel. 271178479Sjb */ 272178479Sjb dt_proc_bpdisable(dpr); 273178479Sjb 274178479Sjb while (dpr->dpr_stop & DT_PROC_STOP_IDLE) 275178479Sjb (void) pthread_cond_wait(&dpr->dpr_cv, &dpr->dpr_lock); 276178479Sjb 277178479Sjb dt_proc_bpenable(dpr); 278178479Sjb } 279178479Sjb} 280178479Sjb 281178479Sjb/*ARGSUSED*/ 282178479Sjbstatic void 283178479Sjbdt_proc_bpmain(dtrace_hdl_t *dtp, dt_proc_t *dpr, const char *fname) 284178479Sjb{ 285178479Sjb dt_dprintf("pid %d: breakpoint at %s()\n", (int)dpr->dpr_pid, fname); 286178479Sjb dt_proc_stop(dpr, DT_PROC_STOP_MAIN); 287178479Sjb} 288178479Sjb 289178479Sjbstatic void 290178479Sjbdt_proc_rdevent(dtrace_hdl_t *dtp, dt_proc_t *dpr, const char *evname) 291178479Sjb{ 292178479Sjb rd_event_msg_t rdm; 293178479Sjb rd_err_e err; 294178479Sjb 295178479Sjb if ((err = rd_event_getmsg(dpr->dpr_rtld, &rdm)) != RD_OK) { 296178479Sjb dt_dprintf("pid %d: failed to get %s event message: %s\n", 297178479Sjb (int)dpr->dpr_pid, evname, rd_errstr(err)); 298178479Sjb return; 299178479Sjb } 300178479Sjb 301178479Sjb dt_dprintf("pid %d: rtld event %s type=%d state %d\n", 302178479Sjb (int)dpr->dpr_pid, evname, rdm.type, rdm.u.state); 303178479Sjb 304178479Sjb switch (rdm.type) { 305178479Sjb case RD_DLACTIVITY: 306178479Sjb if (rdm.u.state != RD_CONSISTENT) 307178479Sjb break; 308178479Sjb 309178479Sjb Pupdate_syms(dpr->dpr_proc); 310178479Sjb if (dt_pid_create_probes_module(dtp, dpr) != 0) 311178479Sjb dt_proc_notify(dtp, dtp->dt_procs, dpr, 312178479Sjb dpr->dpr_errmsg); 313178479Sjb 314178479Sjb break; 315178479Sjb case RD_PREINIT: 316178479Sjb Pupdate_syms(dpr->dpr_proc); 317178479Sjb dt_proc_stop(dpr, DT_PROC_STOP_PREINIT); 318178479Sjb break; 319178479Sjb case RD_POSTINIT: 320178479Sjb Pupdate_syms(dpr->dpr_proc); 321178479Sjb dt_proc_stop(dpr, DT_PROC_STOP_POSTINIT); 322178479Sjb break; 323178479Sjb } 324178479Sjb} 325178479Sjb 326178479Sjbstatic void 327178479Sjbdt_proc_rdwatch(dt_proc_t *dpr, rd_event_e event, const char *evname) 328178479Sjb{ 329178479Sjb rd_notify_t rdn; 330178479Sjb rd_err_e err; 331178479Sjb 332178479Sjb if ((err = rd_event_addr(dpr->dpr_rtld, event, &rdn)) != RD_OK) { 333178479Sjb dt_dprintf("pid %d: failed to get event address for %s: %s\n", 334178479Sjb (int)dpr->dpr_pid, evname, rd_errstr(err)); 335178479Sjb return; 336178479Sjb } 337178479Sjb 338178479Sjb if (rdn.type != RD_NOTIFY_BPT) { 339178479Sjb dt_dprintf("pid %d: event %s has unexpected type %d\n", 340178479Sjb (int)dpr->dpr_pid, evname, rdn.type); 341178479Sjb return; 342178479Sjb } 343178479Sjb 344178479Sjb (void) dt_proc_bpcreate(dpr, rdn.u.bptaddr, 345277300Ssmh#ifdef illumos 346178479Sjb (dt_bkpt_f *)dt_proc_rdevent, (void *)evname); 347211554Srpaulo#else 348211554Srpaulo /* XXX ugly */ 349211554Srpaulo (dt_bkpt_f *)dt_proc_rdevent, __DECONST(void *, evname)); 350211554Srpaulo#endif 351178479Sjb} 352178479Sjb 353178479Sjb/* 354178479Sjb * Common code for enabling events associated with the run-time linker after 355178479Sjb * attaching to a process or after a victim process completes an exec(2). 356178479Sjb */ 357178479Sjbstatic void 358178479Sjbdt_proc_attach(dt_proc_t *dpr, int exec) 359178479Sjb{ 360277300Ssmh#ifdef illumos 361178479Sjb const pstatus_t *psp = Pstatus(dpr->dpr_proc); 362211554Srpaulo#endif 363178479Sjb rd_err_e err; 364178479Sjb GElf_Sym sym; 365178479Sjb 366210775Srpaulo assert(DT_MUTEX_HELD(&dpr->dpr_lock)); 367178479Sjb 368178479Sjb if (exec) { 369277300Ssmh#ifdef illumos 370178479Sjb if (psp->pr_lwp.pr_errno != 0) 371178479Sjb return; /* exec failed: nothing needs to be done */ 372211554Srpaulo#endif 373178479Sjb 374178479Sjb dt_proc_bpdestroy(dpr, B_FALSE); 375277300Ssmh#ifdef illumos 376178479Sjb Preset_maps(dpr->dpr_proc); 377211554Srpaulo#endif 378178479Sjb } 379178479Sjb if ((dpr->dpr_rtld = Prd_agent(dpr->dpr_proc)) != NULL && 380178479Sjb (err = rd_event_enable(dpr->dpr_rtld, B_TRUE)) == RD_OK) { 381277300Ssmh#ifdef illumos 382178479Sjb dt_proc_rdwatch(dpr, RD_PREINIT, "RD_PREINIT"); 383211554Srpaulo#endif 384178479Sjb dt_proc_rdwatch(dpr, RD_POSTINIT, "RD_POSTINIT"); 385277300Ssmh#ifdef illumos 386178479Sjb dt_proc_rdwatch(dpr, RD_DLACTIVITY, "RD_DLACTIVITY"); 387211554Srpaulo#endif 388178479Sjb } else { 389178479Sjb dt_dprintf("pid %d: failed to enable rtld events: %s\n", 390178479Sjb (int)dpr->dpr_pid, dpr->dpr_rtld ? rd_errstr(err) : 391178479Sjb "rtld_db agent initialization failed"); 392178479Sjb } 393178479Sjb 394178479Sjb Pupdate_maps(dpr->dpr_proc); 395178479Sjb 396178479Sjb if (Pxlookup_by_name(dpr->dpr_proc, LM_ID_BASE, 397178479Sjb "a.out", "main", &sym, NULL) == 0) { 398178479Sjb (void) dt_proc_bpcreate(dpr, (uintptr_t)sym.st_value, 399178479Sjb (dt_bkpt_f *)dt_proc_bpmain, "a.out`main"); 400178479Sjb } else { 401178479Sjb dt_dprintf("pid %d: failed to find a.out`main: %s\n", 402178479Sjb (int)dpr->dpr_pid, strerror(errno)); 403178479Sjb } 404178479Sjb} 405178479Sjb 406178479Sjb/* 407178479Sjb * Wait for a stopped process to be set running again by some other debugger. 408178479Sjb * This is typically not required by /proc-based debuggers, since the usual 409178479Sjb * model is that one debugger controls one victim. But DTrace, as usual, has 410178479Sjb * its own needs: the stop() action assumes that prun(1) or some other tool 411178479Sjb * will be applied to resume the victim process. This could be solved by 412178479Sjb * adding a PCWRUN directive to /proc, but that seems like overkill unless 413178479Sjb * other debuggers end up needing this functionality, so we implement a cheap 414178479Sjb * equivalent to PCWRUN using the set of existing kernel mechanisms. 415178479Sjb * 416178479Sjb * Our intent is really not just to wait for the victim to run, but rather to 417178479Sjb * wait for it to run and then stop again for a reason other than the current 418178479Sjb * PR_REQUESTED stop. Since PCWSTOP/Pstopstatus() can be applied repeatedly 419178479Sjb * to a stopped process and will return the same result without affecting the 420178479Sjb * victim, we can just perform these operations repeatedly until Pstate() 421178479Sjb * changes, the representative LWP ID changes, or the stop timestamp advances. 422178479Sjb * dt_proc_control() will then rediscover the new state and continue as usual. 423178479Sjb * When the process is still stopped in the same exact state, we sleep for a 424178479Sjb * brief interval before waiting again so as not to spin consuming CPU cycles. 425178479Sjb */ 426178479Sjbstatic void 427178479Sjbdt_proc_waitrun(dt_proc_t *dpr) 428178479Sjb{ 429326302Semaste printf("%s:%s(%d): not implemented\n", __FUNCTION__, __FILE__, 430326302Semaste __LINE__); 431211554Srpaulo#ifdef DOODAD 432178479Sjb struct ps_prochandle *P = dpr->dpr_proc; 433178479Sjb const lwpstatus_t *psp = &Pstatus(P)->pr_lwp; 434178479Sjb 435178479Sjb int krflag = psp->pr_flags & (PR_KLC | PR_RLC); 436178479Sjb timestruc_t tstamp = psp->pr_tstamp; 437178479Sjb lwpid_t lwpid = psp->pr_lwpid; 438178479Sjb 439178479Sjb const long wstop = PCWSTOP; 440178479Sjb int pfd = Pctlfd(P); 441178479Sjb 442210775Srpaulo assert(DT_MUTEX_HELD(&dpr->dpr_lock)); 443178479Sjb assert(psp->pr_flags & PR_STOPPED); 444178479Sjb assert(Pstate(P) == PS_STOP); 445178479Sjb 446178479Sjb /* 447178479Sjb * While we are waiting for the victim to run, clear PR_KLC and PR_RLC 448178479Sjb * so that if the libdtrace client is killed, the victim stays stopped. 449178479Sjb * dt_proc_destroy() will also observe this and perform PRELEASE_HANG. 450178479Sjb */ 451178479Sjb (void) Punsetflags(P, krflag); 452178479Sjb Psync(P); 453178479Sjb 454178479Sjb (void) pthread_mutex_unlock(&dpr->dpr_lock); 455178479Sjb 456178479Sjb while (!dpr->dpr_quit) { 457178479Sjb if (write(pfd, &wstop, sizeof (wstop)) == -1 && errno == EINTR) 458178479Sjb continue; /* check dpr_quit and continue waiting */ 459178479Sjb 460178479Sjb (void) pthread_mutex_lock(&dpr->dpr_lock); 461178479Sjb (void) Pstopstatus(P, PCNULL, 0); 462178479Sjb psp = &Pstatus(P)->pr_lwp; 463178479Sjb 464178479Sjb /* 465178479Sjb * If we've reached a new state, found a new representative, or 466178479Sjb * the stop timestamp has changed, restore PR_KLC/PR_RLC to its 467178479Sjb * original setting and then return with dpr_lock held. 468178479Sjb */ 469178479Sjb if (Pstate(P) != PS_STOP || psp->pr_lwpid != lwpid || 470178479Sjb bcmp(&psp->pr_tstamp, &tstamp, sizeof (tstamp)) != 0) { 471178479Sjb (void) Psetflags(P, krflag); 472178479Sjb Psync(P); 473178479Sjb return; 474178479Sjb } 475178479Sjb 476178479Sjb (void) pthread_mutex_unlock(&dpr->dpr_lock); 477178479Sjb (void) poll(NULL, 0, MILLISEC / 2); 478178479Sjb } 479178479Sjb 480178479Sjb (void) pthread_mutex_lock(&dpr->dpr_lock); 481211554Srpaulo#endif 482178479Sjb} 483178479Sjb 484178479Sjbtypedef struct dt_proc_control_data { 485178479Sjb dtrace_hdl_t *dpcd_hdl; /* DTrace handle */ 486178479Sjb dt_proc_t *dpcd_proc; /* proccess to control */ 487178479Sjb} dt_proc_control_data_t; 488178479Sjb 489178479Sjb/* 490178479Sjb * Main loop for all victim process control threads. We initialize all the 491178479Sjb * appropriate /proc control mechanisms, and then enter a loop waiting for 492178479Sjb * the process to stop on an event or die. We process any events by calling 493178479Sjb * appropriate subroutines, and exit when the victim dies or we lose control. 494178479Sjb * 495178479Sjb * The control thread synchronizes the use of dpr_proc with other libdtrace 496178479Sjb * threads using dpr_lock. We hold the lock for all of our operations except 497178479Sjb * waiting while the process is running: this is accomplished by writing a 498178479Sjb * PCWSTOP directive directly to the underlying /proc/<pid>/ctl file. If the 499178479Sjb * libdtrace client wishes to exit or abort our wait, SIGCANCEL can be used. 500178479Sjb */ 501178479Sjbstatic void * 502178479Sjbdt_proc_control(void *arg) 503178479Sjb{ 504178479Sjb dt_proc_control_data_t *datap = arg; 505178479Sjb dtrace_hdl_t *dtp = datap->dpcd_hdl; 506178479Sjb dt_proc_t *dpr = datap->dpcd_proc; 507249573Spfg dt_proc_hash_t *dph = dpr->dpr_hdl->dt_procs; 508178479Sjb struct ps_prochandle *P = dpr->dpr_proc; 509178565Sjb int pid = dpr->dpr_pid; 510178479Sjb 511277300Ssmh#ifdef illumos 512178479Sjb int pfd = Pctlfd(P); 513178479Sjb 514178479Sjb const long wstop = PCWSTOP; 515178565Sjb#endif 516178479Sjb int notify = B_FALSE; 517178479Sjb 518178479Sjb /* 519178479Sjb * We disable the POSIX thread cancellation mechanism so that the 520178479Sjb * client program using libdtrace can't accidentally cancel our thread. 521178479Sjb * dt_proc_destroy() uses SIGCANCEL explicitly to simply poke us out 522178479Sjb * of PCWSTOP with EINTR, at which point we will see dpr_quit and exit. 523178479Sjb */ 524178479Sjb (void) pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, NULL); 525178479Sjb 526178479Sjb /* 527178479Sjb * Set up the corresponding process for tracing by libdtrace. We want 528178479Sjb * to be able to catch breakpoints and efficiently single-step over 529178479Sjb * them, and we need to enable librtld_db to watch libdl activity. 530178479Sjb */ 531178479Sjb (void) pthread_mutex_lock(&dpr->dpr_lock); 532178479Sjb 533277300Ssmh#ifdef illumos 534178479Sjb (void) Punsetflags(P, PR_ASYNC); /* require synchronous mode */ 535178479Sjb (void) Psetflags(P, PR_BPTADJ); /* always adjust eip on x86 */ 536178479Sjb (void) Punsetflags(P, PR_FORK); /* do not inherit on fork */ 537178479Sjb 538178479Sjb (void) Pfault(P, FLTBPT, B_TRUE); /* always trace breakpoints */ 539178479Sjb (void) Pfault(P, FLTTRACE, B_TRUE); /* always trace single-step */ 540178479Sjb 541178479Sjb /* 542178479Sjb * We must trace exit from exec() system calls so that if the exec is 543178479Sjb * successful, we can reset our breakpoints and re-initialize libproc. 544178479Sjb */ 545178479Sjb (void) Psysexit(P, SYS_execve, B_TRUE); 546178479Sjb 547178479Sjb /* 548178479Sjb * We must trace entry and exit for fork() system calls in order to 549178479Sjb * disable our breakpoints temporarily during the fork. We do not set 550178479Sjb * the PR_FORK flag, so if fork succeeds the child begins executing and 551178479Sjb * does not inherit any other tracing behaviors or a control thread. 552178479Sjb */ 553178479Sjb (void) Psysentry(P, SYS_vfork, B_TRUE); 554178479Sjb (void) Psysexit(P, SYS_vfork, B_TRUE); 555178479Sjb (void) Psysentry(P, SYS_forksys, B_TRUE); 556178479Sjb (void) Psysexit(P, SYS_forksys, B_TRUE); 557178479Sjb 558178479Sjb Psync(P); /* enable all /proc changes */ 559211554Srpaulo#endif 560178479Sjb dt_proc_attach(dpr, B_FALSE); /* enable rtld breakpoints */ 561178479Sjb 562178479Sjb /* 563178479Sjb * If PR_KLC is set, we created the process; otherwise we grabbed it. 564178479Sjb * Check for an appropriate stop request and wait for dt_proc_continue. 565178479Sjb */ 566277300Ssmh#ifdef illumos 567178479Sjb if (Pstatus(P)->pr_flags & PR_KLC) 568211554Srpaulo#else 569211554Srpaulo if (proc_getflags(P) & PR_KLC) 570211554Srpaulo#endif 571178479Sjb dt_proc_stop(dpr, DT_PROC_STOP_CREATE); 572178479Sjb else 573178479Sjb dt_proc_stop(dpr, DT_PROC_STOP_GRAB); 574178479Sjb 575178479Sjb if (Psetrun(P, 0, 0) == -1) { 576178479Sjb dt_dprintf("pid %d: failed to set running: %s\n", 577178479Sjb (int)dpr->dpr_pid, strerror(errno)); 578178479Sjb } 579178479Sjb 580178479Sjb (void) pthread_mutex_unlock(&dpr->dpr_lock); 581178479Sjb 582178479Sjb /* 583178479Sjb * Wait for the process corresponding to this control thread to stop, 584178479Sjb * process the event, and then set it running again. We want to sleep 585178479Sjb * with dpr_lock *unheld* so that other parts of libdtrace can use the 586178479Sjb * ps_prochandle in the meantime (e.g. ustack()). To do this, we write 587178479Sjb * a PCWSTOP directive directly to the underlying /proc/<pid>/ctl file. 588178479Sjb * Once the process stops, we wake up, grab dpr_lock, and then call 589178479Sjb * Pwait() (which will return immediately) and do our processing. 590178479Sjb */ 591178479Sjb while (!dpr->dpr_quit) { 592178479Sjb const lwpstatus_t *psp; 593178479Sjb 594277300Ssmh#ifdef illumos 595178479Sjb if (write(pfd, &wstop, sizeof (wstop)) == -1 && errno == EINTR) 596178479Sjb continue; /* check dpr_quit and continue waiting */ 597178565Sjb#else 598178565Sjb /* Wait for the process to report status. */ 599210695Srpaulo proc_wstatus(P); 600211554Srpaulo if (errno == EINTR) 601211554Srpaulo continue; /* check dpr_quit and continue waiting */ 602178565Sjb#endif 603178479Sjb 604178479Sjb (void) pthread_mutex_lock(&dpr->dpr_lock); 605178565Sjb 606277300Ssmh#ifdef illumos 607178479Sjbpwait_locked: 608178479Sjb if (Pstopstatus(P, PCNULL, 0) == -1 && errno == EINTR) { 609178479Sjb (void) pthread_mutex_unlock(&dpr->dpr_lock); 610178479Sjb continue; /* check dpr_quit and continue waiting */ 611178479Sjb } 612178565Sjb#endif 613178479Sjb 614211554Srpaulo switch (Pstate(P)) { 615211554Srpaulo case PS_STOP: 616277300Ssmh#ifdef illumos 617211554Srpaulo psp = &Pstatus(P)->pr_lwp; 618178565Sjb#else 619211554Srpaulo psp = proc_getlwpstatus(P); 620178565Sjb#endif 621178479Sjb 622178479Sjb dt_dprintf("pid %d: proc stopped showing %d/%d\n", 623178479Sjb pid, psp->pr_why, psp->pr_what); 624178479Sjb 625178479Sjb /* 626178479Sjb * If the process stops showing PR_REQUESTED, then the 627178479Sjb * DTrace stop() action was applied to it or another 628178479Sjb * debugging utility (e.g. pstop(1)) asked it to stop. 629178479Sjb * In either case, the user's intention is for the 630178479Sjb * process to remain stopped until another external 631178479Sjb * mechanism (e.g. prun(1)) is applied. So instead of 632178479Sjb * setting the process running ourself, we wait for 633178479Sjb * someone else to do so. Once that happens, we return 634178479Sjb * to our normal loop waiting for an event of interest. 635178479Sjb */ 636178479Sjb if (psp->pr_why == PR_REQUESTED) { 637178479Sjb dt_proc_waitrun(dpr); 638178479Sjb (void) pthread_mutex_unlock(&dpr->dpr_lock); 639178479Sjb continue; 640178479Sjb } 641178479Sjb 642178479Sjb /* 643178479Sjb * If the process stops showing one of the events that 644178479Sjb * we are tracing, perform the appropriate response. 645178479Sjb * Note that we ignore PR_SUSPENDED, PR_CHECKPOINT, and 646178479Sjb * PR_JOBCONTROL by design: if one of these conditions 647178479Sjb * occurs, we will fall through to Psetrun() but the 648178479Sjb * process will remain stopped in the kernel by the 649178479Sjb * corresponding mechanism (e.g. job control stop). 650178479Sjb */ 651178479Sjb if (psp->pr_why == PR_FAULTED && psp->pr_what == FLTBPT) 652178479Sjb dt_proc_bpmatch(dtp, dpr); 653178479Sjb else if (psp->pr_why == PR_SYSENTRY && 654178479Sjb IS_SYS_FORK(psp->pr_what)) 655178479Sjb dt_proc_bpdisable(dpr); 656178479Sjb else if (psp->pr_why == PR_SYSEXIT && 657178479Sjb IS_SYS_FORK(psp->pr_what)) 658178479Sjb dt_proc_bpenable(dpr); 659178479Sjb else if (psp->pr_why == PR_SYSEXIT && 660178479Sjb IS_SYS_EXEC(psp->pr_what)) 661178479Sjb dt_proc_attach(dpr, B_TRUE); 662178479Sjb break; 663178479Sjb 664178479Sjb case PS_LOST: 665277300Ssmh#ifdef illumos 666178479Sjb if (Preopen(P) == 0) 667178479Sjb goto pwait_locked; 668178565Sjb#endif 669178479Sjb 670178479Sjb dt_dprintf("pid %d: proc lost: %s\n", 671178479Sjb pid, strerror(errno)); 672178479Sjb 673178479Sjb dpr->dpr_quit = B_TRUE; 674178479Sjb notify = B_TRUE; 675178479Sjb break; 676178479Sjb 677178479Sjb case PS_UNDEAD: 678178479Sjb dt_dprintf("pid %d: proc died\n", pid); 679178479Sjb dpr->dpr_quit = B_TRUE; 680178479Sjb notify = B_TRUE; 681178479Sjb break; 682178479Sjb } 683178479Sjb 684178479Sjb if (Pstate(P) != PS_UNDEAD && Psetrun(P, 0, 0) == -1) { 685178479Sjb dt_dprintf("pid %d: failed to set running: %s\n", 686178479Sjb (int)dpr->dpr_pid, strerror(errno)); 687178479Sjb } 688178479Sjb 689178479Sjb (void) pthread_mutex_unlock(&dpr->dpr_lock); 690178479Sjb } 691178479Sjb 692178479Sjb /* 693178479Sjb * If the control thread detected PS_UNDEAD or PS_LOST, then enqueue 694178479Sjb * the dt_proc_t structure on the dt_proc_hash_t notification list. 695178479Sjb */ 696178479Sjb if (notify) 697178479Sjb dt_proc_notify(dtp, dph, dpr, NULL); 698178479Sjb 699178479Sjb /* 700178479Sjb * Destroy and remove any remaining breakpoints, set dpr_done and clear 701178479Sjb * dpr_tid to indicate the control thread has exited, and notify any 702178479Sjb * waiting thread in dt_proc_destroy() that we have succesfully exited. 703178479Sjb */ 704178479Sjb (void) pthread_mutex_lock(&dpr->dpr_lock); 705178479Sjb 706178479Sjb dt_proc_bpdestroy(dpr, B_TRUE); 707178479Sjb dpr->dpr_done = B_TRUE; 708178479Sjb dpr->dpr_tid = 0; 709178479Sjb 710178479Sjb (void) pthread_cond_broadcast(&dpr->dpr_cv); 711178479Sjb (void) pthread_mutex_unlock(&dpr->dpr_lock); 712178479Sjb 713178479Sjb return (NULL); 714178479Sjb} 715178479Sjb 716178479Sjb/*PRINTFLIKE3*/ 717178479Sjbstatic struct ps_prochandle * 718178479Sjbdt_proc_error(dtrace_hdl_t *dtp, dt_proc_t *dpr, const char *format, ...) 719178479Sjb{ 720178479Sjb va_list ap; 721178479Sjb 722178479Sjb va_start(ap, format); 723178479Sjb dt_set_errmsg(dtp, NULL, NULL, NULL, 0, format, ap); 724178479Sjb va_end(ap); 725178479Sjb 726178479Sjb if (dpr->dpr_proc != NULL) 727178479Sjb Prelease(dpr->dpr_proc, 0); 728178479Sjb 729178479Sjb dt_free(dtp, dpr); 730178479Sjb (void) dt_set_errno(dtp, EDT_COMPILER); 731178479Sjb return (NULL); 732178479Sjb} 733178479Sjb 734178479Sjbdt_proc_t * 735178479Sjbdt_proc_lookup(dtrace_hdl_t *dtp, struct ps_prochandle *P, int remove) 736178479Sjb{ 737178479Sjb dt_proc_hash_t *dph = dtp->dt_procs; 738277300Ssmh#ifdef illumos 739178479Sjb pid_t pid = Pstatus(P)->pr_pid; 740178565Sjb#else 741178565Sjb pid_t pid = proc_getpid(P); 742178565Sjb#endif 743178479Sjb dt_proc_t *dpr, **dpp = &dph->dph_hash[pid & (dph->dph_hashlen - 1)]; 744178479Sjb 745178479Sjb for (dpr = *dpp; dpr != NULL; dpr = dpr->dpr_hash) { 746178479Sjb if (dpr->dpr_pid == pid) 747178479Sjb break; 748178479Sjb else 749178479Sjb dpp = &dpr->dpr_hash; 750178479Sjb } 751178479Sjb 752178479Sjb assert(dpr != NULL); 753178479Sjb assert(dpr->dpr_proc == P); 754178479Sjb 755178479Sjb if (remove) 756178479Sjb *dpp = dpr->dpr_hash; /* remove from pid hash chain */ 757178479Sjb 758178479Sjb return (dpr); 759178479Sjb} 760178479Sjb 761178479Sjbstatic void 762178479Sjbdt_proc_destroy(dtrace_hdl_t *dtp, struct ps_prochandle *P) 763178479Sjb{ 764178479Sjb dt_proc_t *dpr = dt_proc_lookup(dtp, P, B_FALSE); 765178479Sjb dt_proc_hash_t *dph = dtp->dt_procs; 766178479Sjb dt_proc_notify_t *npr, **npp; 767178479Sjb int rflag; 768178479Sjb 769178479Sjb assert(dpr != NULL); 770178479Sjb 771178479Sjb /* 772178479Sjb * If neither PR_KLC nor PR_RLC is set, then the process is stopped by 773178479Sjb * an external debugger and we were waiting in dt_proc_waitrun(). 774178479Sjb * Leave the process in this condition using PRELEASE_HANG. 775178479Sjb */ 776277300Ssmh#ifdef illumos 777178479Sjb if (!(Pstatus(dpr->dpr_proc)->pr_flags & (PR_KLC | PR_RLC))) { 778178565Sjb#else 779178565Sjb if (!(proc_getflags(dpr->dpr_proc) & (PR_KLC | PR_RLC))) { 780178565Sjb#endif 781178479Sjb dt_dprintf("abandoning pid %d\n", (int)dpr->dpr_pid); 782210767Srpaulo rflag = PRELEASE_HANG; 783277300Ssmh#ifdef illumos 784210767Srpaulo } else if (Pstatus(dpr->dpr_proc)->pr_flags & PR_KLC) { 785178565Sjb#else 786210767Srpaulo } else if (proc_getflags(dpr->dpr_proc) & PR_KLC) { 787178565Sjb#endif 788210767Srpaulo dt_dprintf("killing pid %d\n", (int)dpr->dpr_pid); 789210767Srpaulo rflag = PRELEASE_KILL; /* apply kill-on-last-close */ 790178479Sjb } else { 791178479Sjb dt_dprintf("releasing pid %d\n", (int)dpr->dpr_pid); 792210767Srpaulo rflag = 0; /* apply run-on-last-close */ 793178479Sjb } 794178479Sjb 795178479Sjb if (dpr->dpr_tid) { 796178479Sjb /* 797178479Sjb * Set the dpr_quit flag to tell the daemon thread to exit. We 798178479Sjb * send it a SIGCANCEL to poke it out of PCWSTOP or any other 799178479Sjb * long-term /proc system call. Our daemon threads have POSIX 800178479Sjb * cancellation disabled, so EINTR will be the only effect. We 801178479Sjb * then wait for dpr_done to indicate the thread has exited. 802178479Sjb * 803178479Sjb * We can't use pthread_kill() to send SIGCANCEL because the 804178479Sjb * interface forbids it and we can't use pthread_cancel() 805178479Sjb * because with cancellation disabled it won't actually 806178479Sjb * send SIGCANCEL to the target thread, so we use _lwp_kill() 807178479Sjb * to do the job. This is all built on evil knowledge of 808178479Sjb * the details of the cancellation mechanism in libc. 809178479Sjb */ 810178479Sjb (void) pthread_mutex_lock(&dpr->dpr_lock); 811178479Sjb dpr->dpr_quit = B_TRUE; 812277300Ssmh#ifdef illumos 813178479Sjb (void) _lwp_kill(dpr->dpr_tid, SIGCANCEL); 814178565Sjb#else 815234234Sgnn pthread_kill(dpr->dpr_tid, SIGTHR); 816178565Sjb#endif 817178479Sjb 818178479Sjb /* 819178479Sjb * If the process is currently idling in dt_proc_stop(), re- 820178479Sjb * enable breakpoints and poke it into running again. 821178479Sjb */ 822178479Sjb if (dpr->dpr_stop & DT_PROC_STOP_IDLE) { 823178479Sjb dt_proc_bpenable(dpr); 824178479Sjb dpr->dpr_stop &= ~DT_PROC_STOP_IDLE; 825178479Sjb (void) pthread_cond_broadcast(&dpr->dpr_cv); 826178479Sjb } 827178479Sjb 828178479Sjb while (!dpr->dpr_done) 829178479Sjb (void) pthread_cond_wait(&dpr->dpr_cv, &dpr->dpr_lock); 830178479Sjb 831178479Sjb (void) pthread_mutex_unlock(&dpr->dpr_lock); 832178479Sjb } 833178479Sjb 834178479Sjb /* 835178479Sjb * Before we free the process structure, remove this dt_proc_t from the 836178479Sjb * lookup hash, and then walk the dt_proc_hash_t's notification list 837178479Sjb * and remove this dt_proc_t if it is enqueued. 838178479Sjb */ 839178479Sjb (void) pthread_mutex_lock(&dph->dph_lock); 840178479Sjb (void) dt_proc_lookup(dtp, P, B_TRUE); 841178479Sjb npp = &dph->dph_notify; 842178479Sjb 843178479Sjb while ((npr = *npp) != NULL) { 844178479Sjb if (npr->dprn_dpr == dpr) { 845178479Sjb *npp = npr->dprn_next; 846178479Sjb dt_free(dtp, npr); 847178479Sjb } else { 848178479Sjb npp = &npr->dprn_next; 849178479Sjb } 850178479Sjb } 851178479Sjb 852178479Sjb (void) pthread_mutex_unlock(&dph->dph_lock); 853178479Sjb 854178479Sjb /* 855178479Sjb * Remove the dt_proc_list from the LRU list, release the underlying 856178479Sjb * libproc handle, and free our dt_proc_t data structure. 857178479Sjb */ 858178479Sjb if (dpr->dpr_cacheable) { 859178479Sjb assert(dph->dph_lrucnt != 0); 860178479Sjb dph->dph_lrucnt--; 861178479Sjb } 862178479Sjb 863178479Sjb dt_list_delete(&dph->dph_lrulist, dpr); 864178479Sjb Prelease(dpr->dpr_proc, rflag); 865178479Sjb dt_free(dtp, dpr); 866178479Sjb} 867178479Sjb 868178479Sjbstatic int 869178479Sjbdt_proc_create_thread(dtrace_hdl_t *dtp, dt_proc_t *dpr, uint_t stop) 870178479Sjb{ 871178479Sjb dt_proc_control_data_t data; 872178479Sjb sigset_t nset, oset; 873178479Sjb pthread_attr_t a; 874178479Sjb int err; 875178479Sjb 876178479Sjb (void) pthread_mutex_lock(&dpr->dpr_lock); 877178479Sjb dpr->dpr_stop |= stop; /* set bit for initial rendezvous */ 878178479Sjb 879178479Sjb (void) pthread_attr_init(&a); 880178479Sjb (void) pthread_attr_setdetachstate(&a, PTHREAD_CREATE_DETACHED); 881178479Sjb 882178479Sjb (void) sigfillset(&nset); 883178479Sjb (void) sigdelset(&nset, SIGABRT); /* unblocked for assert() */ 884277300Ssmh#ifdef illumos 885178479Sjb (void) sigdelset(&nset, SIGCANCEL); /* see dt_proc_destroy() */ 886178565Sjb#else 887178565Sjb (void) sigdelset(&nset, SIGUSR1); /* see dt_proc_destroy() */ 888178565Sjb#endif 889178479Sjb 890178479Sjb data.dpcd_hdl = dtp; 891178479Sjb data.dpcd_proc = dpr; 892178479Sjb 893178479Sjb (void) pthread_sigmask(SIG_SETMASK, &nset, &oset); 894178479Sjb err = pthread_create(&dpr->dpr_tid, &a, dt_proc_control, &data); 895178479Sjb (void) pthread_sigmask(SIG_SETMASK, &oset, NULL); 896178479Sjb 897178479Sjb /* 898178479Sjb * If the control thread was created, then wait on dpr_cv for either 899178479Sjb * dpr_done to be set (the victim died or the control thread failed) 900178479Sjb * or DT_PROC_STOP_IDLE to be set, indicating that the victim is now 901178479Sjb * stopped by /proc and the control thread is at the rendezvous event. 902178479Sjb * On success, we return with the process and control thread stopped: 903178479Sjb * the caller can then apply dt_proc_continue() to resume both. 904178479Sjb */ 905178479Sjb if (err == 0) { 906178479Sjb while (!dpr->dpr_done && !(dpr->dpr_stop & DT_PROC_STOP_IDLE)) 907178479Sjb (void) pthread_cond_wait(&dpr->dpr_cv, &dpr->dpr_lock); 908178479Sjb 909178479Sjb /* 910178479Sjb * If dpr_done is set, the control thread aborted before it 911178479Sjb * reached the rendezvous event. This is either due to PS_LOST 912178479Sjb * or PS_UNDEAD (i.e. the process died). We try to provide a 913178479Sjb * small amount of useful information to help figure it out. 914178479Sjb */ 915178479Sjb if (dpr->dpr_done) { 916277300Ssmh#ifdef illumos 917178479Sjb const psinfo_t *prp = Ppsinfo(dpr->dpr_proc); 918178479Sjb int stat = prp ? prp->pr_wstat : 0; 919178479Sjb int pid = dpr->dpr_pid; 920178565Sjb#else 921211554Srpaulo int stat = proc_getwstat(dpr->dpr_proc); 922211554Srpaulo int pid = proc_getpid(dpr->dpr_proc); 923211554Srpaulo#endif 924178565Sjb if (proc_state(dpr->dpr_proc) == PS_LOST) { 925178479Sjb (void) dt_proc_error(dpr->dpr_hdl, dpr, 926178479Sjb "failed to control pid %d: process exec'd " 927178479Sjb "set-id or unobservable program\n", pid); 928178479Sjb } else if (WIFSIGNALED(stat)) { 929178479Sjb (void) dt_proc_error(dpr->dpr_hdl, dpr, 930178479Sjb "failed to control pid %d: process died " 931178479Sjb "from signal %d\n", pid, WTERMSIG(stat)); 932178479Sjb } else { 933178479Sjb (void) dt_proc_error(dpr->dpr_hdl, dpr, 934178479Sjb "failed to control pid %d: process exited " 935178479Sjb "with status %d\n", pid, WEXITSTATUS(stat)); 936178479Sjb } 937178479Sjb 938178479Sjb err = ESRCH; /* cause grab() or create() to fail */ 939178479Sjb } 940178479Sjb } else { 941178479Sjb (void) dt_proc_error(dpr->dpr_hdl, dpr, 942178479Sjb "failed to create control thread for process-id %d: %s\n", 943178479Sjb (int)dpr->dpr_pid, strerror(err)); 944178479Sjb } 945178479Sjb 946238979Sgnn if (err == 0) 947238979Sgnn (void) pthread_mutex_unlock(&dpr->dpr_lock); 948178479Sjb (void) pthread_attr_destroy(&a); 949178479Sjb 950178479Sjb return (err); 951178479Sjb} 952178479Sjb 953178479Sjbstruct ps_prochandle * 954184696Srodrigcdt_proc_create(dtrace_hdl_t *dtp, const char *file, char *const *argv, 955184696Srodrigc proc_child_func *pcf, void *child_arg) 956178479Sjb{ 957178479Sjb dt_proc_hash_t *dph = dtp->dt_procs; 958178479Sjb dt_proc_t *dpr; 959178479Sjb int err; 960178479Sjb 961178479Sjb if ((dpr = dt_zalloc(dtp, sizeof (dt_proc_t))) == NULL) 962178479Sjb return (NULL); /* errno is set for us */ 963178479Sjb 964178479Sjb (void) pthread_mutex_init(&dpr->dpr_lock, NULL); 965178479Sjb (void) pthread_cond_init(&dpr->dpr_cv, NULL); 966178479Sjb 967277300Ssmh#ifdef illumos 968249573Spfg if ((dpr->dpr_proc = Pcreate(file, argv, &err, NULL, 0)) == NULL) { 969211554Srpaulo#else 970211554Srpaulo if ((err = proc_create(file, argv, pcf, child_arg, 971211554Srpaulo &dpr->dpr_proc)) != 0) { 972211554Srpaulo#endif 973178479Sjb return (dt_proc_error(dtp, dpr, 974178479Sjb "failed to execute %s: %s\n", file, Pcreate_error(err))); 975178479Sjb } 976178479Sjb 977178479Sjb dpr->dpr_hdl = dtp; 978277300Ssmh#ifdef illumos 979178479Sjb dpr->dpr_pid = Pstatus(dpr->dpr_proc)->pr_pid; 980178565Sjb#else 981178565Sjb dpr->dpr_pid = proc_getpid(dpr->dpr_proc); 982178565Sjb#endif 983178479Sjb 984211554Srpaulo (void) Punsetflags(dpr->dpr_proc, PR_RLC); 985211554Srpaulo (void) Psetflags(dpr->dpr_proc, PR_KLC); 986211554Srpaulo 987178479Sjb if (dt_proc_create_thread(dtp, dpr, dtp->dt_prcmode) != 0) 988178479Sjb return (NULL); /* dt_proc_error() has been called for us */ 989178479Sjb 990178479Sjb dpr->dpr_hash = dph->dph_hash[dpr->dpr_pid & (dph->dph_hashlen - 1)]; 991178479Sjb dph->dph_hash[dpr->dpr_pid & (dph->dph_hashlen - 1)] = dpr; 992178479Sjb dt_list_prepend(&dph->dph_lrulist, dpr); 993178479Sjb 994178479Sjb dt_dprintf("created pid %d\n", (int)dpr->dpr_pid); 995178479Sjb dpr->dpr_refs++; 996178479Sjb 997178479Sjb return (dpr->dpr_proc); 998178479Sjb} 999178479Sjb 1000178479Sjbstruct ps_prochandle * 1001178479Sjbdt_proc_grab(dtrace_hdl_t *dtp, pid_t pid, int flags, int nomonitor) 1002178479Sjb{ 1003178479Sjb dt_proc_hash_t *dph = dtp->dt_procs; 1004178479Sjb uint_t h = pid & (dph->dph_hashlen - 1); 1005178479Sjb dt_proc_t *dpr, *opr; 1006178479Sjb int err; 1007178479Sjb 1008178479Sjb /* 1009178479Sjb * Search the hash table for the pid. If it is already grabbed or 1010178479Sjb * created, move the handle to the front of the lrulist, increment 1011178479Sjb * the reference count, and return the existing ps_prochandle. 1012178479Sjb */ 1013178479Sjb for (dpr = dph->dph_hash[h]; dpr != NULL; dpr = dpr->dpr_hash) { 1014178479Sjb if (dpr->dpr_pid == pid && !dpr->dpr_stale) { 1015178479Sjb /* 1016178479Sjb * If the cached handle was opened read-only and 1017178479Sjb * this request is for a writeable handle, mark 1018178479Sjb * the cached handle as stale and open a new handle. 1019178479Sjb * Since it's stale, unmark it as cacheable. 1020178479Sjb */ 1021178479Sjb if (dpr->dpr_rdonly && !(flags & PGRAB_RDONLY)) { 1022178479Sjb dt_dprintf("upgrading pid %d\n", (int)pid); 1023178479Sjb dpr->dpr_stale = B_TRUE; 1024178479Sjb dpr->dpr_cacheable = B_FALSE; 1025178479Sjb dph->dph_lrucnt--; 1026178479Sjb break; 1027178479Sjb } 1028178479Sjb 1029178479Sjb dt_dprintf("grabbed pid %d (cached)\n", (int)pid); 1030178479Sjb dt_list_delete(&dph->dph_lrulist, dpr); 1031178479Sjb dt_list_prepend(&dph->dph_lrulist, dpr); 1032178479Sjb dpr->dpr_refs++; 1033178479Sjb return (dpr->dpr_proc); 1034178479Sjb } 1035178479Sjb } 1036178479Sjb 1037178479Sjb if ((dpr = dt_zalloc(dtp, sizeof (dt_proc_t))) == NULL) 1038178479Sjb return (NULL); /* errno is set for us */ 1039178479Sjb 1040178479Sjb (void) pthread_mutex_init(&dpr->dpr_lock, NULL); 1041178479Sjb (void) pthread_cond_init(&dpr->dpr_cv, NULL); 1042178479Sjb 1043277300Ssmh#ifdef illumos 1044178479Sjb if ((dpr->dpr_proc = Pgrab(pid, flags, &err)) == NULL) { 1045211554Srpaulo#else 1046211554Srpaulo if ((err = proc_attach(pid, flags, &dpr->dpr_proc)) != 0) { 1047211554Srpaulo#endif 1048178479Sjb return (dt_proc_error(dtp, dpr, 1049178479Sjb "failed to grab pid %d: %s\n", (int)pid, Pgrab_error(err))); 1050178479Sjb } 1051178479Sjb 1052178479Sjb dpr->dpr_hdl = dtp; 1053178479Sjb dpr->dpr_pid = pid; 1054178479Sjb 1055178479Sjb (void) Punsetflags(dpr->dpr_proc, PR_KLC); 1056178479Sjb (void) Psetflags(dpr->dpr_proc, PR_RLC); 1057178479Sjb 1058178479Sjb /* 1059178479Sjb * If we are attempting to grab the process without a monitor 1060178479Sjb * thread, then mark the process cacheable only if it's being 1061178479Sjb * grabbed read-only. If we're currently caching more process 1062178479Sjb * handles than dph_lrulim permits, attempt to find the 1063178479Sjb * least-recently-used handle that is currently unreferenced and 1064178479Sjb * release it from the cache. Otherwise we are grabbing the process 1065178479Sjb * for control: create a control thread for this process and store 1066178479Sjb * its ID in dpr->dpr_tid. 1067178479Sjb */ 1068178479Sjb if (nomonitor || (flags & PGRAB_RDONLY)) { 1069178479Sjb if (dph->dph_lrucnt >= dph->dph_lrulim) { 1070178479Sjb for (opr = dt_list_prev(&dph->dph_lrulist); 1071178479Sjb opr != NULL; opr = dt_list_prev(opr)) { 1072178479Sjb if (opr->dpr_cacheable && opr->dpr_refs == 0) { 1073178479Sjb dt_proc_destroy(dtp, opr->dpr_proc); 1074178479Sjb break; 1075178479Sjb } 1076178479Sjb } 1077178479Sjb } 1078178479Sjb 1079178479Sjb if (flags & PGRAB_RDONLY) { 1080178479Sjb dpr->dpr_cacheable = B_TRUE; 1081178479Sjb dpr->dpr_rdonly = B_TRUE; 1082178479Sjb dph->dph_lrucnt++; 1083178479Sjb } 1084178479Sjb 1085178479Sjb } else if (dt_proc_create_thread(dtp, dpr, DT_PROC_STOP_GRAB) != 0) 1086178479Sjb return (NULL); /* dt_proc_error() has been called for us */ 1087178479Sjb 1088178479Sjb dpr->dpr_hash = dph->dph_hash[h]; 1089178479Sjb dph->dph_hash[h] = dpr; 1090178479Sjb dt_list_prepend(&dph->dph_lrulist, dpr); 1091178479Sjb 1092178479Sjb dt_dprintf("grabbed pid %d\n", (int)pid); 1093178479Sjb dpr->dpr_refs++; 1094178479Sjb 1095178479Sjb return (dpr->dpr_proc); 1096178479Sjb} 1097178479Sjb 1098178479Sjbvoid 1099178479Sjbdt_proc_release(dtrace_hdl_t *dtp, struct ps_prochandle *P) 1100178479Sjb{ 1101178479Sjb dt_proc_t *dpr = dt_proc_lookup(dtp, P, B_FALSE); 1102178479Sjb dt_proc_hash_t *dph = dtp->dt_procs; 1103178479Sjb 1104178479Sjb assert(dpr != NULL); 1105178479Sjb assert(dpr->dpr_refs != 0); 1106178479Sjb 1107178479Sjb if (--dpr->dpr_refs == 0 && 1108178479Sjb (!dpr->dpr_cacheable || dph->dph_lrucnt > dph->dph_lrulim)) 1109178479Sjb dt_proc_destroy(dtp, P); 1110178479Sjb} 1111178479Sjb 1112178479Sjbvoid 1113178479Sjbdt_proc_continue(dtrace_hdl_t *dtp, struct ps_prochandle *P) 1114178479Sjb{ 1115178479Sjb dt_proc_t *dpr = dt_proc_lookup(dtp, P, B_FALSE); 1116178479Sjb 1117178479Sjb (void) pthread_mutex_lock(&dpr->dpr_lock); 1118178479Sjb 1119178479Sjb if (dpr->dpr_stop & DT_PROC_STOP_IDLE) { 1120178479Sjb dpr->dpr_stop &= ~DT_PROC_STOP_IDLE; 1121178479Sjb (void) pthread_cond_broadcast(&dpr->dpr_cv); 1122178479Sjb } 1123178479Sjb 1124178479Sjb (void) pthread_mutex_unlock(&dpr->dpr_lock); 1125178479Sjb} 1126178479Sjb 1127178479Sjbvoid 1128178479Sjbdt_proc_lock(dtrace_hdl_t *dtp, struct ps_prochandle *P) 1129178479Sjb{ 1130178479Sjb dt_proc_t *dpr = dt_proc_lookup(dtp, P, B_FALSE); 1131178479Sjb int err = pthread_mutex_lock(&dpr->dpr_lock); 1132178479Sjb assert(err == 0); /* check for recursion */ 1133178479Sjb} 1134178479Sjb 1135178479Sjbvoid 1136178479Sjbdt_proc_unlock(dtrace_hdl_t *dtp, struct ps_prochandle *P) 1137178479Sjb{ 1138178479Sjb dt_proc_t *dpr = dt_proc_lookup(dtp, P, B_FALSE); 1139178479Sjb int err = pthread_mutex_unlock(&dpr->dpr_lock); 1140178479Sjb assert(err == 0); /* check for unheld lock */ 1141178479Sjb} 1142178479Sjb 1143178479Sjbvoid 1144249573Spfgdt_proc_hash_create(dtrace_hdl_t *dtp) 1145178479Sjb{ 1146178479Sjb if ((dtp->dt_procs = dt_zalloc(dtp, sizeof (dt_proc_hash_t) + 1147249573Spfg sizeof (dt_proc_t *) * _dtrace_pidbuckets - 1)) != NULL) { 1148178479Sjb 1149249573Spfg (void) pthread_mutex_init(&dtp->dt_procs->dph_lock, NULL); 1150249573Spfg (void) pthread_cond_init(&dtp->dt_procs->dph_cv, NULL); 1151178479Sjb 1152249573Spfg dtp->dt_procs->dph_hashlen = _dtrace_pidbuckets; 1153249573Spfg dtp->dt_procs->dph_lrulim = _dtrace_pidlrulim; 1154178479Sjb } 1155178479Sjb} 1156178479Sjb 1157178479Sjbvoid 1158249573Spfgdt_proc_hash_destroy(dtrace_hdl_t *dtp) 1159178479Sjb{ 1160178479Sjb dt_proc_hash_t *dph = dtp->dt_procs; 1161178479Sjb dt_proc_t *dpr; 1162178479Sjb 1163178479Sjb while ((dpr = dt_list_next(&dph->dph_lrulist)) != NULL) 1164178479Sjb dt_proc_destroy(dtp, dpr->dpr_proc); 1165178479Sjb 1166178479Sjb dtp->dt_procs = NULL; 1167178479Sjb dt_free(dtp, dph); 1168178479Sjb} 1169178479Sjb 1170178479Sjbstruct ps_prochandle * 1171184696Srodrigcdtrace_proc_create(dtrace_hdl_t *dtp, const char *file, char *const *argv, 1172184696Srodrigc proc_child_func *pcf, void *child_arg) 1173178479Sjb{ 1174178479Sjb dt_ident_t *idp = dt_idhash_lookup(dtp->dt_macros, "target"); 1175184696Srodrigc struct ps_prochandle *P = dt_proc_create(dtp, file, argv, pcf, child_arg); 1176178479Sjb 1177211554Srpaulo if (P != NULL && idp != NULL && idp->di_id == 0) { 1178277300Ssmh#ifdef illumos 1179178479Sjb idp->di_id = Pstatus(P)->pr_pid; /* $target = created pid */ 1180178565Sjb#else 1181178565Sjb idp->di_id = proc_getpid(P); /* $target = created pid */ 1182178565Sjb#endif 1183211554Srpaulo } 1184178479Sjb 1185178479Sjb return (P); 1186178479Sjb} 1187178479Sjb 1188178479Sjbstruct ps_prochandle * 1189178479Sjbdtrace_proc_grab(dtrace_hdl_t *dtp, pid_t pid, int flags) 1190178479Sjb{ 1191178479Sjb dt_ident_t *idp = dt_idhash_lookup(dtp->dt_macros, "target"); 1192178479Sjb struct ps_prochandle *P = dt_proc_grab(dtp, pid, flags, 0); 1193178479Sjb 1194178479Sjb if (P != NULL && idp != NULL && idp->di_id == 0) 1195178479Sjb idp->di_id = pid; /* $target = grabbed pid */ 1196178479Sjb 1197178479Sjb return (P); 1198178479Sjb} 1199178479Sjb 1200178479Sjbvoid 1201178479Sjbdtrace_proc_release(dtrace_hdl_t *dtp, struct ps_prochandle *P) 1202178479Sjb{ 1203178479Sjb dt_proc_release(dtp, P); 1204178479Sjb} 1205178479Sjb 1206178479Sjbvoid 1207178479Sjbdtrace_proc_continue(dtrace_hdl_t *dtp, struct ps_prochandle *P) 1208178479Sjb{ 1209178479Sjb dt_proc_continue(dtp, P); 1210178479Sjb} 1211