dt_proc.c revision 238979
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> 80178565Sjb#if defined(sun) 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 92211554Srpaulo#if !defined(sun) 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{ 146211554Srpaulo#if defined(sun) 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 155211554Srpaulo#if !defined(sun) 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)) { 162211554Srpaulo#if defined(sun) 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", 173211554Srpaulo#if defined(sun) 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, 345211554Srpaulo#if defined(sun) 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{ 360211554Srpaulo#if defined(sun) 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) { 369211554Srpaulo#if defined(sun) 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); 375211554Srpaulo#if defined(sun) 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) { 381211554Srpaulo#if defined(sun) 382178479Sjb dt_proc_rdwatch(dpr, RD_PREINIT, "RD_PREINIT"); 383211554Srpaulo#endif 384178479Sjb dt_proc_rdwatch(dpr, RD_POSTINIT, "RD_POSTINIT"); 385211554Srpaulo#if defined(sun) 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{ 429211554Srpauloprintf("%s:%s(%d): DOODAD\n",__FUNCTION__,__FILE__,__LINE__); 430211554Srpaulo#ifdef DOODAD 431178479Sjb struct ps_prochandle *P = dpr->dpr_proc; 432178479Sjb const lwpstatus_t *psp = &Pstatus(P)->pr_lwp; 433178479Sjb 434178479Sjb int krflag = psp->pr_flags & (PR_KLC | PR_RLC); 435178479Sjb timestruc_t tstamp = psp->pr_tstamp; 436178479Sjb lwpid_t lwpid = psp->pr_lwpid; 437178479Sjb 438178479Sjb const long wstop = PCWSTOP; 439178479Sjb int pfd = Pctlfd(P); 440178479Sjb 441210775Srpaulo assert(DT_MUTEX_HELD(&dpr->dpr_lock)); 442178479Sjb assert(psp->pr_flags & PR_STOPPED); 443178479Sjb assert(Pstate(P) == PS_STOP); 444178479Sjb 445178479Sjb /* 446178479Sjb * While we are waiting for the victim to run, clear PR_KLC and PR_RLC 447178479Sjb * so that if the libdtrace client is killed, the victim stays stopped. 448178479Sjb * dt_proc_destroy() will also observe this and perform PRELEASE_HANG. 449178479Sjb */ 450178479Sjb (void) Punsetflags(P, krflag); 451178479Sjb Psync(P); 452178479Sjb 453178479Sjb (void) pthread_mutex_unlock(&dpr->dpr_lock); 454178479Sjb 455178479Sjb while (!dpr->dpr_quit) { 456178479Sjb if (write(pfd, &wstop, sizeof (wstop)) == -1 && errno == EINTR) 457178479Sjb continue; /* check dpr_quit and continue waiting */ 458178479Sjb 459178479Sjb (void) pthread_mutex_lock(&dpr->dpr_lock); 460178479Sjb (void) Pstopstatus(P, PCNULL, 0); 461178479Sjb psp = &Pstatus(P)->pr_lwp; 462178479Sjb 463178479Sjb /* 464178479Sjb * If we've reached a new state, found a new representative, or 465178479Sjb * the stop timestamp has changed, restore PR_KLC/PR_RLC to its 466178479Sjb * original setting and then return with dpr_lock held. 467178479Sjb */ 468178479Sjb if (Pstate(P) != PS_STOP || psp->pr_lwpid != lwpid || 469178479Sjb bcmp(&psp->pr_tstamp, &tstamp, sizeof (tstamp)) != 0) { 470178479Sjb (void) Psetflags(P, krflag); 471178479Sjb Psync(P); 472178479Sjb return; 473178479Sjb } 474178479Sjb 475178479Sjb (void) pthread_mutex_unlock(&dpr->dpr_lock); 476178479Sjb (void) poll(NULL, 0, MILLISEC / 2); 477178479Sjb } 478178479Sjb 479178479Sjb (void) pthread_mutex_lock(&dpr->dpr_lock); 480211554Srpaulo#endif 481178479Sjb} 482178479Sjb 483178479Sjbtypedef struct dt_proc_control_data { 484178479Sjb dtrace_hdl_t *dpcd_hdl; /* DTrace handle */ 485178479Sjb dt_proc_t *dpcd_proc; /* proccess to control */ 486178479Sjb} dt_proc_control_data_t; 487178479Sjb 488178479Sjb/* 489178479Sjb * Main loop for all victim process control threads. We initialize all the 490178479Sjb * appropriate /proc control mechanisms, and then enter a loop waiting for 491178479Sjb * the process to stop on an event or die. We process any events by calling 492178479Sjb * appropriate subroutines, and exit when the victim dies or we lose control. 493178479Sjb * 494178479Sjb * The control thread synchronizes the use of dpr_proc with other libdtrace 495178479Sjb * threads using dpr_lock. We hold the lock for all of our operations except 496178479Sjb * waiting while the process is running: this is accomplished by writing a 497178479Sjb * PCWSTOP directive directly to the underlying /proc/<pid>/ctl file. If the 498178479Sjb * libdtrace client wishes to exit or abort our wait, SIGCANCEL can be used. 499178479Sjb */ 500178479Sjbstatic void * 501178479Sjbdt_proc_control(void *arg) 502178479Sjb{ 503178479Sjb dt_proc_control_data_t *datap = arg; 504178479Sjb dtrace_hdl_t *dtp = datap->dpcd_hdl; 505178479Sjb dt_proc_t *dpr = datap->dpcd_proc; 506178479Sjb dt_proc_hash_t *dph = dpr->dpr_hdl->dt_procs; 507178479Sjb struct ps_prochandle *P = dpr->dpr_proc; 508178565Sjb int pid = dpr->dpr_pid; 509178479Sjb 510178565Sjb#if defined(sun) 511178479Sjb int pfd = Pctlfd(P); 512178479Sjb 513178479Sjb const long wstop = PCWSTOP; 514178565Sjb#endif 515178479Sjb int notify = B_FALSE; 516178479Sjb 517178479Sjb /* 518178479Sjb * We disable the POSIX thread cancellation mechanism so that the 519178479Sjb * client program using libdtrace can't accidentally cancel our thread. 520178479Sjb * dt_proc_destroy() uses SIGCANCEL explicitly to simply poke us out 521178479Sjb * of PCWSTOP with EINTR, at which point we will see dpr_quit and exit. 522178479Sjb */ 523178479Sjb (void) pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, NULL); 524178479Sjb 525178479Sjb /* 526178479Sjb * Set up the corresponding process for tracing by libdtrace. We want 527178479Sjb * to be able to catch breakpoints and efficiently single-step over 528178479Sjb * them, and we need to enable librtld_db to watch libdl activity. 529178479Sjb */ 530178479Sjb (void) pthread_mutex_lock(&dpr->dpr_lock); 531178479Sjb 532178565Sjb#if defined(sun) 533178479Sjb (void) Punsetflags(P, PR_ASYNC); /* require synchronous mode */ 534178479Sjb (void) Psetflags(P, PR_BPTADJ); /* always adjust eip on x86 */ 535178479Sjb (void) Punsetflags(P, PR_FORK); /* do not inherit on fork */ 536178479Sjb 537178479Sjb (void) Pfault(P, FLTBPT, B_TRUE); /* always trace breakpoints */ 538178479Sjb (void) Pfault(P, FLTTRACE, B_TRUE); /* always trace single-step */ 539178479Sjb 540178479Sjb /* 541178479Sjb * We must trace exit from exec() system calls so that if the exec is 542178479Sjb * successful, we can reset our breakpoints and re-initialize libproc. 543178479Sjb */ 544178479Sjb (void) Psysexit(P, SYS_execve, B_TRUE); 545178479Sjb 546178479Sjb /* 547178479Sjb * We must trace entry and exit for fork() system calls in order to 548178479Sjb * disable our breakpoints temporarily during the fork. We do not set 549178479Sjb * the PR_FORK flag, so if fork succeeds the child begins executing and 550178479Sjb * does not inherit any other tracing behaviors or a control thread. 551178479Sjb */ 552178479Sjb (void) Psysentry(P, SYS_vfork, B_TRUE); 553178479Sjb (void) Psysexit(P, SYS_vfork, B_TRUE); 554178479Sjb (void) Psysentry(P, SYS_forksys, B_TRUE); 555178479Sjb (void) Psysexit(P, SYS_forksys, B_TRUE); 556178479Sjb 557178479Sjb Psync(P); /* enable all /proc changes */ 558211554Srpaulo#endif 559178479Sjb dt_proc_attach(dpr, B_FALSE); /* enable rtld breakpoints */ 560178479Sjb 561178479Sjb /* 562178479Sjb * If PR_KLC is set, we created the process; otherwise we grabbed it. 563178479Sjb * Check for an appropriate stop request and wait for dt_proc_continue. 564178479Sjb */ 565211554Srpaulo#if defined(sun) 566178479Sjb if (Pstatus(P)->pr_flags & PR_KLC) 567211554Srpaulo#else 568211554Srpaulo if (proc_getflags(P) & PR_KLC) 569211554Srpaulo#endif 570178479Sjb dt_proc_stop(dpr, DT_PROC_STOP_CREATE); 571178479Sjb else 572178479Sjb dt_proc_stop(dpr, DT_PROC_STOP_GRAB); 573178479Sjb 574178479Sjb if (Psetrun(P, 0, 0) == -1) { 575178479Sjb dt_dprintf("pid %d: failed to set running: %s\n", 576178479Sjb (int)dpr->dpr_pid, strerror(errno)); 577178479Sjb } 578178479Sjb 579178479Sjb (void) pthread_mutex_unlock(&dpr->dpr_lock); 580178479Sjb 581178479Sjb /* 582178479Sjb * Wait for the process corresponding to this control thread to stop, 583178479Sjb * process the event, and then set it running again. We want to sleep 584178479Sjb * with dpr_lock *unheld* so that other parts of libdtrace can use the 585178479Sjb * ps_prochandle in the meantime (e.g. ustack()). To do this, we write 586178479Sjb * a PCWSTOP directive directly to the underlying /proc/<pid>/ctl file. 587178479Sjb * Once the process stops, we wake up, grab dpr_lock, and then call 588178479Sjb * Pwait() (which will return immediately) and do our processing. 589178479Sjb */ 590178479Sjb while (!dpr->dpr_quit) { 591178479Sjb const lwpstatus_t *psp; 592178479Sjb 593211554Srpaulo#if defined(sun) 594178479Sjb if (write(pfd, &wstop, sizeof (wstop)) == -1 && errno == EINTR) 595178479Sjb continue; /* check dpr_quit and continue waiting */ 596178565Sjb#else 597178565Sjb /* Wait for the process to report status. */ 598210695Srpaulo proc_wstatus(P); 599211554Srpaulo if (errno == EINTR) 600211554Srpaulo continue; /* check dpr_quit and continue waiting */ 601178565Sjb#endif 602178479Sjb 603178479Sjb (void) pthread_mutex_lock(&dpr->dpr_lock); 604178565Sjb 605178565Sjb#if defined(sun) 606178479Sjbpwait_locked: 607178479Sjb if (Pstopstatus(P, PCNULL, 0) == -1 && errno == EINTR) { 608178479Sjb (void) pthread_mutex_unlock(&dpr->dpr_lock); 609178479Sjb continue; /* check dpr_quit and continue waiting */ 610178479Sjb } 611178565Sjb#endif 612178479Sjb 613211554Srpaulo switch (Pstate(P)) { 614211554Srpaulo case PS_STOP: 615178565Sjb#if defined(sun) 616211554Srpaulo psp = &Pstatus(P)->pr_lwp; 617178565Sjb#else 618211554Srpaulo psp = proc_getlwpstatus(P); 619178565Sjb#endif 620178479Sjb 621178479Sjb dt_dprintf("pid %d: proc stopped showing %d/%d\n", 622178479Sjb pid, psp->pr_why, psp->pr_what); 623178479Sjb 624178479Sjb /* 625178479Sjb * If the process stops showing PR_REQUESTED, then the 626178479Sjb * DTrace stop() action was applied to it or another 627178479Sjb * debugging utility (e.g. pstop(1)) asked it to stop. 628178479Sjb * In either case, the user's intention is for the 629178479Sjb * process to remain stopped until another external 630178479Sjb * mechanism (e.g. prun(1)) is applied. So instead of 631178479Sjb * setting the process running ourself, we wait for 632178479Sjb * someone else to do so. Once that happens, we return 633178479Sjb * to our normal loop waiting for an event of interest. 634178479Sjb */ 635178479Sjb if (psp->pr_why == PR_REQUESTED) { 636178479Sjb dt_proc_waitrun(dpr); 637178479Sjb (void) pthread_mutex_unlock(&dpr->dpr_lock); 638178479Sjb continue; 639178479Sjb } 640178479Sjb 641178479Sjb /* 642178479Sjb * If the process stops showing one of the events that 643178479Sjb * we are tracing, perform the appropriate response. 644178479Sjb * Note that we ignore PR_SUSPENDED, PR_CHECKPOINT, and 645178479Sjb * PR_JOBCONTROL by design: if one of these conditions 646178479Sjb * occurs, we will fall through to Psetrun() but the 647178479Sjb * process will remain stopped in the kernel by the 648178479Sjb * corresponding mechanism (e.g. job control stop). 649178479Sjb */ 650178479Sjb if (psp->pr_why == PR_FAULTED && psp->pr_what == FLTBPT) 651178479Sjb dt_proc_bpmatch(dtp, dpr); 652178479Sjb else if (psp->pr_why == PR_SYSENTRY && 653178479Sjb IS_SYS_FORK(psp->pr_what)) 654178479Sjb dt_proc_bpdisable(dpr); 655178479Sjb else if (psp->pr_why == PR_SYSEXIT && 656178479Sjb IS_SYS_FORK(psp->pr_what)) 657178479Sjb dt_proc_bpenable(dpr); 658178479Sjb else if (psp->pr_why == PR_SYSEXIT && 659178479Sjb IS_SYS_EXEC(psp->pr_what)) 660178479Sjb dt_proc_attach(dpr, B_TRUE); 661178479Sjb break; 662178479Sjb 663178479Sjb case PS_LOST: 664178565Sjb#if defined(sun) 665178479Sjb if (Preopen(P) == 0) 666178479Sjb goto pwait_locked; 667178565Sjb#endif 668178479Sjb 669178479Sjb dt_dprintf("pid %d: proc lost: %s\n", 670178479Sjb pid, strerror(errno)); 671178479Sjb 672178479Sjb dpr->dpr_quit = B_TRUE; 673178479Sjb notify = B_TRUE; 674178479Sjb break; 675178479Sjb 676178479Sjb case PS_UNDEAD: 677178479Sjb dt_dprintf("pid %d: proc died\n", pid); 678178479Sjb dpr->dpr_quit = B_TRUE; 679178479Sjb notify = B_TRUE; 680178479Sjb break; 681178479Sjb } 682178479Sjb 683178479Sjb if (Pstate(P) != PS_UNDEAD && Psetrun(P, 0, 0) == -1) { 684178479Sjb dt_dprintf("pid %d: failed to set running: %s\n", 685178479Sjb (int)dpr->dpr_pid, strerror(errno)); 686178479Sjb } 687178479Sjb 688178479Sjb (void) pthread_mutex_unlock(&dpr->dpr_lock); 689178479Sjb } 690178479Sjb 691178479Sjb /* 692178479Sjb * If the control thread detected PS_UNDEAD or PS_LOST, then enqueue 693178479Sjb * the dt_proc_t structure on the dt_proc_hash_t notification list. 694178479Sjb */ 695178479Sjb if (notify) 696178479Sjb dt_proc_notify(dtp, dph, dpr, NULL); 697178479Sjb 698178479Sjb /* 699178479Sjb * Destroy and remove any remaining breakpoints, set dpr_done and clear 700178479Sjb * dpr_tid to indicate the control thread has exited, and notify any 701178479Sjb * waiting thread in dt_proc_destroy() that we have succesfully exited. 702178479Sjb */ 703178479Sjb (void) pthread_mutex_lock(&dpr->dpr_lock); 704178479Sjb 705178479Sjb dt_proc_bpdestroy(dpr, B_TRUE); 706178479Sjb dpr->dpr_done = B_TRUE; 707178479Sjb dpr->dpr_tid = 0; 708178479Sjb 709178479Sjb (void) pthread_cond_broadcast(&dpr->dpr_cv); 710178479Sjb (void) pthread_mutex_unlock(&dpr->dpr_lock); 711178479Sjb 712178479Sjb return (NULL); 713178479Sjb} 714178479Sjb 715178479Sjb/*PRINTFLIKE3*/ 716178479Sjbstatic struct ps_prochandle * 717178479Sjbdt_proc_error(dtrace_hdl_t *dtp, dt_proc_t *dpr, const char *format, ...) 718178479Sjb{ 719178479Sjb va_list ap; 720178479Sjb 721178479Sjb va_start(ap, format); 722178479Sjb dt_set_errmsg(dtp, NULL, NULL, NULL, 0, format, ap); 723178479Sjb va_end(ap); 724178479Sjb 725178479Sjb if (dpr->dpr_proc != NULL) 726178479Sjb Prelease(dpr->dpr_proc, 0); 727178479Sjb 728178479Sjb dt_free(dtp, dpr); 729178479Sjb (void) dt_set_errno(dtp, EDT_COMPILER); 730178479Sjb return (NULL); 731178479Sjb} 732178479Sjb 733178479Sjbdt_proc_t * 734178479Sjbdt_proc_lookup(dtrace_hdl_t *dtp, struct ps_prochandle *P, int remove) 735178479Sjb{ 736178479Sjb dt_proc_hash_t *dph = dtp->dt_procs; 737178565Sjb#if defined(sun) 738178479Sjb pid_t pid = Pstatus(P)->pr_pid; 739178565Sjb#else 740178565Sjb pid_t pid = proc_getpid(P); 741178565Sjb#endif 742178479Sjb dt_proc_t *dpr, **dpp = &dph->dph_hash[pid & (dph->dph_hashlen - 1)]; 743178479Sjb 744178479Sjb for (dpr = *dpp; dpr != NULL; dpr = dpr->dpr_hash) { 745178479Sjb if (dpr->dpr_pid == pid) 746178479Sjb break; 747178479Sjb else 748178479Sjb dpp = &dpr->dpr_hash; 749178479Sjb } 750178479Sjb 751178479Sjb assert(dpr != NULL); 752178479Sjb assert(dpr->dpr_proc == P); 753178479Sjb 754178479Sjb if (remove) 755178479Sjb *dpp = dpr->dpr_hash; /* remove from pid hash chain */ 756178479Sjb 757178479Sjb return (dpr); 758178479Sjb} 759178479Sjb 760178479Sjbstatic void 761178479Sjbdt_proc_destroy(dtrace_hdl_t *dtp, struct ps_prochandle *P) 762178479Sjb{ 763178479Sjb dt_proc_t *dpr = dt_proc_lookup(dtp, P, B_FALSE); 764178479Sjb dt_proc_hash_t *dph = dtp->dt_procs; 765178479Sjb dt_proc_notify_t *npr, **npp; 766178479Sjb int rflag; 767178479Sjb 768178479Sjb assert(dpr != NULL); 769178479Sjb 770178479Sjb /* 771178479Sjb * If neither PR_KLC nor PR_RLC is set, then the process is stopped by 772178479Sjb * an external debugger and we were waiting in dt_proc_waitrun(). 773178479Sjb * Leave the process in this condition using PRELEASE_HANG. 774178479Sjb */ 775178565Sjb#if defined(sun) 776178479Sjb if (!(Pstatus(dpr->dpr_proc)->pr_flags & (PR_KLC | PR_RLC))) { 777178565Sjb#else 778178565Sjb if (!(proc_getflags(dpr->dpr_proc) & (PR_KLC | PR_RLC))) { 779178565Sjb#endif 780178479Sjb dt_dprintf("abandoning pid %d\n", (int)dpr->dpr_pid); 781210767Srpaulo rflag = PRELEASE_HANG; 782178565Sjb#if defined(sun) 783210767Srpaulo } else if (Pstatus(dpr->dpr_proc)->pr_flags & PR_KLC) { 784178565Sjb#else 785210767Srpaulo } else if (proc_getflags(dpr->dpr_proc) & PR_KLC) { 786178565Sjb#endif 787210767Srpaulo dt_dprintf("killing pid %d\n", (int)dpr->dpr_pid); 788210767Srpaulo rflag = PRELEASE_KILL; /* apply kill-on-last-close */ 789178479Sjb } else { 790178479Sjb dt_dprintf("releasing pid %d\n", (int)dpr->dpr_pid); 791210767Srpaulo rflag = 0; /* apply run-on-last-close */ 792178479Sjb } 793178479Sjb 794178479Sjb if (dpr->dpr_tid) { 795178479Sjb /* 796178479Sjb * Set the dpr_quit flag to tell the daemon thread to exit. We 797178479Sjb * send it a SIGCANCEL to poke it out of PCWSTOP or any other 798178479Sjb * long-term /proc system call. Our daemon threads have POSIX 799178479Sjb * cancellation disabled, so EINTR will be the only effect. We 800178479Sjb * then wait for dpr_done to indicate the thread has exited. 801178479Sjb * 802178479Sjb * We can't use pthread_kill() to send SIGCANCEL because the 803178479Sjb * interface forbids it and we can't use pthread_cancel() 804178479Sjb * because with cancellation disabled it won't actually 805178479Sjb * send SIGCANCEL to the target thread, so we use _lwp_kill() 806178479Sjb * to do the job. This is all built on evil knowledge of 807178479Sjb * the details of the cancellation mechanism in libc. 808178479Sjb */ 809178479Sjb (void) pthread_mutex_lock(&dpr->dpr_lock); 810178479Sjb dpr->dpr_quit = B_TRUE; 811178565Sjb#if defined(sun) 812178479Sjb (void) _lwp_kill(dpr->dpr_tid, SIGCANCEL); 813178565Sjb#else 814234234Sgnn pthread_kill(dpr->dpr_tid, SIGTHR); 815178565Sjb#endif 816178479Sjb 817178479Sjb /* 818178479Sjb * If the process is currently idling in dt_proc_stop(), re- 819178479Sjb * enable breakpoints and poke it into running again. 820178479Sjb */ 821178479Sjb if (dpr->dpr_stop & DT_PROC_STOP_IDLE) { 822178479Sjb dt_proc_bpenable(dpr); 823178479Sjb dpr->dpr_stop &= ~DT_PROC_STOP_IDLE; 824178479Sjb (void) pthread_cond_broadcast(&dpr->dpr_cv); 825178479Sjb } 826178479Sjb 827178479Sjb while (!dpr->dpr_done) 828178479Sjb (void) pthread_cond_wait(&dpr->dpr_cv, &dpr->dpr_lock); 829178479Sjb 830178479Sjb (void) pthread_mutex_unlock(&dpr->dpr_lock); 831178479Sjb } 832178479Sjb 833178479Sjb /* 834178479Sjb * Before we free the process structure, remove this dt_proc_t from the 835178479Sjb * lookup hash, and then walk the dt_proc_hash_t's notification list 836178479Sjb * and remove this dt_proc_t if it is enqueued. 837178479Sjb */ 838178479Sjb (void) pthread_mutex_lock(&dph->dph_lock); 839178479Sjb (void) dt_proc_lookup(dtp, P, B_TRUE); 840178479Sjb npp = &dph->dph_notify; 841178479Sjb 842178479Sjb while ((npr = *npp) != NULL) { 843178479Sjb if (npr->dprn_dpr == dpr) { 844178479Sjb *npp = npr->dprn_next; 845178479Sjb dt_free(dtp, npr); 846178479Sjb } else { 847178479Sjb npp = &npr->dprn_next; 848178479Sjb } 849178479Sjb } 850178479Sjb 851178479Sjb (void) pthread_mutex_unlock(&dph->dph_lock); 852178479Sjb 853178479Sjb /* 854178479Sjb * Remove the dt_proc_list from the LRU list, release the underlying 855178479Sjb * libproc handle, and free our dt_proc_t data structure. 856178479Sjb */ 857178479Sjb if (dpr->dpr_cacheable) { 858178479Sjb assert(dph->dph_lrucnt != 0); 859178479Sjb dph->dph_lrucnt--; 860178479Sjb } 861178479Sjb 862178479Sjb dt_list_delete(&dph->dph_lrulist, dpr); 863178479Sjb Prelease(dpr->dpr_proc, rflag); 864178479Sjb dt_free(dtp, dpr); 865178479Sjb} 866178479Sjb 867178479Sjbstatic int 868178479Sjbdt_proc_create_thread(dtrace_hdl_t *dtp, dt_proc_t *dpr, uint_t stop) 869178479Sjb{ 870178479Sjb dt_proc_control_data_t data; 871178479Sjb sigset_t nset, oset; 872178479Sjb pthread_attr_t a; 873178479Sjb int err; 874178479Sjb 875178479Sjb (void) pthread_mutex_lock(&dpr->dpr_lock); 876178479Sjb dpr->dpr_stop |= stop; /* set bit for initial rendezvous */ 877178479Sjb 878178479Sjb (void) pthread_attr_init(&a); 879178479Sjb (void) pthread_attr_setdetachstate(&a, PTHREAD_CREATE_DETACHED); 880178479Sjb 881178479Sjb (void) sigfillset(&nset); 882178479Sjb (void) sigdelset(&nset, SIGABRT); /* unblocked for assert() */ 883178565Sjb#if defined(sun) 884178479Sjb (void) sigdelset(&nset, SIGCANCEL); /* see dt_proc_destroy() */ 885178565Sjb#else 886178565Sjb (void) sigdelset(&nset, SIGUSR1); /* see dt_proc_destroy() */ 887178565Sjb#endif 888178479Sjb 889178479Sjb data.dpcd_hdl = dtp; 890178479Sjb data.dpcd_proc = dpr; 891178479Sjb 892178479Sjb (void) pthread_sigmask(SIG_SETMASK, &nset, &oset); 893178479Sjb err = pthread_create(&dpr->dpr_tid, &a, dt_proc_control, &data); 894178479Sjb (void) pthread_sigmask(SIG_SETMASK, &oset, NULL); 895178479Sjb 896178479Sjb /* 897178479Sjb * If the control thread was created, then wait on dpr_cv for either 898178479Sjb * dpr_done to be set (the victim died or the control thread failed) 899178479Sjb * or DT_PROC_STOP_IDLE to be set, indicating that the victim is now 900178479Sjb * stopped by /proc and the control thread is at the rendezvous event. 901178479Sjb * On success, we return with the process and control thread stopped: 902178479Sjb * the caller can then apply dt_proc_continue() to resume both. 903178479Sjb */ 904178479Sjb if (err == 0) { 905178479Sjb while (!dpr->dpr_done && !(dpr->dpr_stop & DT_PROC_STOP_IDLE)) 906178479Sjb (void) pthread_cond_wait(&dpr->dpr_cv, &dpr->dpr_lock); 907178479Sjb 908178479Sjb /* 909178479Sjb * If dpr_done is set, the control thread aborted before it 910178479Sjb * reached the rendezvous event. This is either due to PS_LOST 911178479Sjb * or PS_UNDEAD (i.e. the process died). We try to provide a 912178479Sjb * small amount of useful information to help figure it out. 913178479Sjb */ 914178479Sjb if (dpr->dpr_done) { 915178565Sjb#if defined(sun) 916178479Sjb const psinfo_t *prp = Ppsinfo(dpr->dpr_proc); 917178479Sjb int stat = prp ? prp->pr_wstat : 0; 918178479Sjb int pid = dpr->dpr_pid; 919178565Sjb#else 920211554Srpaulo int stat = proc_getwstat(dpr->dpr_proc); 921211554Srpaulo int pid = proc_getpid(dpr->dpr_proc); 922211554Srpaulo#endif 923178565Sjb if (proc_state(dpr->dpr_proc) == PS_LOST) { 924178479Sjb (void) dt_proc_error(dpr->dpr_hdl, dpr, 925178479Sjb "failed to control pid %d: process exec'd " 926178479Sjb "set-id or unobservable program\n", pid); 927178479Sjb } else if (WIFSIGNALED(stat)) { 928178479Sjb (void) dt_proc_error(dpr->dpr_hdl, dpr, 929178479Sjb "failed to control pid %d: process died " 930178479Sjb "from signal %d\n", pid, WTERMSIG(stat)); 931178479Sjb } else { 932178479Sjb (void) dt_proc_error(dpr->dpr_hdl, dpr, 933178479Sjb "failed to control pid %d: process exited " 934178479Sjb "with status %d\n", pid, WEXITSTATUS(stat)); 935178479Sjb } 936178479Sjb 937178479Sjb err = ESRCH; /* cause grab() or create() to fail */ 938178479Sjb } 939178479Sjb } else { 940178479Sjb (void) dt_proc_error(dpr->dpr_hdl, dpr, 941178479Sjb "failed to create control thread for process-id %d: %s\n", 942178479Sjb (int)dpr->dpr_pid, strerror(err)); 943178479Sjb } 944178479Sjb 945238979Sgnn if (err == 0) 946238979Sgnn (void) pthread_mutex_unlock(&dpr->dpr_lock); 947178479Sjb (void) pthread_attr_destroy(&a); 948178479Sjb 949178479Sjb return (err); 950178479Sjb} 951178479Sjb 952178479Sjbstruct ps_prochandle * 953184696Srodrigcdt_proc_create(dtrace_hdl_t *dtp, const char *file, char *const *argv, 954184696Srodrigc proc_child_func *pcf, void *child_arg) 955178479Sjb{ 956178479Sjb dt_proc_hash_t *dph = dtp->dt_procs; 957178479Sjb dt_proc_t *dpr; 958178479Sjb int err; 959178479Sjb 960178479Sjb if ((dpr = dt_zalloc(dtp, sizeof (dt_proc_t))) == NULL) 961178479Sjb return (NULL); /* errno is set for us */ 962178479Sjb 963178479Sjb (void) pthread_mutex_init(&dpr->dpr_lock, NULL); 964178479Sjb (void) pthread_cond_init(&dpr->dpr_cv, NULL); 965178479Sjb 966178565Sjb#if defined(sun) 967178479Sjb if ((dpr->dpr_proc = Pcreate(file, argv, &err, NULL, 0)) == NULL) { 968211554Srpaulo#else 969211554Srpaulo if ((err = proc_create(file, argv, pcf, child_arg, 970211554Srpaulo &dpr->dpr_proc)) != 0) { 971211554Srpaulo#endif 972178479Sjb return (dt_proc_error(dtp, dpr, 973178479Sjb "failed to execute %s: %s\n", file, Pcreate_error(err))); 974178479Sjb } 975178479Sjb 976178479Sjb dpr->dpr_hdl = dtp; 977211554Srpaulo#if defined(sun) 978178479Sjb dpr->dpr_pid = Pstatus(dpr->dpr_proc)->pr_pid; 979178565Sjb#else 980178565Sjb dpr->dpr_pid = proc_getpid(dpr->dpr_proc); 981178565Sjb#endif 982178479Sjb 983211554Srpaulo (void) Punsetflags(dpr->dpr_proc, PR_RLC); 984211554Srpaulo (void) Psetflags(dpr->dpr_proc, PR_KLC); 985211554Srpaulo 986178479Sjb if (dt_proc_create_thread(dtp, dpr, dtp->dt_prcmode) != 0) 987178479Sjb return (NULL); /* dt_proc_error() has been called for us */ 988178479Sjb 989178479Sjb dpr->dpr_hash = dph->dph_hash[dpr->dpr_pid & (dph->dph_hashlen - 1)]; 990178479Sjb dph->dph_hash[dpr->dpr_pid & (dph->dph_hashlen - 1)] = dpr; 991178479Sjb dt_list_prepend(&dph->dph_lrulist, dpr); 992178479Sjb 993178479Sjb dt_dprintf("created pid %d\n", (int)dpr->dpr_pid); 994178479Sjb dpr->dpr_refs++; 995178479Sjb 996178479Sjb return (dpr->dpr_proc); 997178479Sjb} 998178479Sjb 999178479Sjbstruct ps_prochandle * 1000178479Sjbdt_proc_grab(dtrace_hdl_t *dtp, pid_t pid, int flags, int nomonitor) 1001178479Sjb{ 1002178479Sjb dt_proc_hash_t *dph = dtp->dt_procs; 1003178479Sjb uint_t h = pid & (dph->dph_hashlen - 1); 1004178479Sjb dt_proc_t *dpr, *opr; 1005178479Sjb int err; 1006178479Sjb 1007178479Sjb /* 1008178479Sjb * Search the hash table for the pid. If it is already grabbed or 1009178479Sjb * created, move the handle to the front of the lrulist, increment 1010178479Sjb * the reference count, and return the existing ps_prochandle. 1011178479Sjb */ 1012178479Sjb for (dpr = dph->dph_hash[h]; dpr != NULL; dpr = dpr->dpr_hash) { 1013178479Sjb if (dpr->dpr_pid == pid && !dpr->dpr_stale) { 1014178479Sjb /* 1015178479Sjb * If the cached handle was opened read-only and 1016178479Sjb * this request is for a writeable handle, mark 1017178479Sjb * the cached handle as stale and open a new handle. 1018178479Sjb * Since it's stale, unmark it as cacheable. 1019178479Sjb */ 1020178479Sjb if (dpr->dpr_rdonly && !(flags & PGRAB_RDONLY)) { 1021178479Sjb dt_dprintf("upgrading pid %d\n", (int)pid); 1022178479Sjb dpr->dpr_stale = B_TRUE; 1023178479Sjb dpr->dpr_cacheable = B_FALSE; 1024178479Sjb dph->dph_lrucnt--; 1025178479Sjb break; 1026178479Sjb } 1027178479Sjb 1028178479Sjb dt_dprintf("grabbed pid %d (cached)\n", (int)pid); 1029178479Sjb dt_list_delete(&dph->dph_lrulist, dpr); 1030178479Sjb dt_list_prepend(&dph->dph_lrulist, dpr); 1031178479Sjb dpr->dpr_refs++; 1032178479Sjb return (dpr->dpr_proc); 1033178479Sjb } 1034178479Sjb } 1035178479Sjb 1036178479Sjb if ((dpr = dt_zalloc(dtp, sizeof (dt_proc_t))) == NULL) 1037178479Sjb return (NULL); /* errno is set for us */ 1038178479Sjb 1039178479Sjb (void) pthread_mutex_init(&dpr->dpr_lock, NULL); 1040178479Sjb (void) pthread_cond_init(&dpr->dpr_cv, NULL); 1041178479Sjb 1042178565Sjb#if defined(sun) 1043178479Sjb if ((dpr->dpr_proc = Pgrab(pid, flags, &err)) == NULL) { 1044211554Srpaulo#else 1045211554Srpaulo if ((err = proc_attach(pid, flags, &dpr->dpr_proc)) != 0) { 1046211554Srpaulo#endif 1047178479Sjb return (dt_proc_error(dtp, dpr, 1048178479Sjb "failed to grab pid %d: %s\n", (int)pid, Pgrab_error(err))); 1049178479Sjb } 1050178479Sjb 1051178479Sjb dpr->dpr_hdl = dtp; 1052178479Sjb dpr->dpr_pid = pid; 1053178479Sjb 1054178479Sjb (void) Punsetflags(dpr->dpr_proc, PR_KLC); 1055178479Sjb (void) Psetflags(dpr->dpr_proc, PR_RLC); 1056178479Sjb 1057178479Sjb /* 1058178479Sjb * If we are attempting to grab the process without a monitor 1059178479Sjb * thread, then mark the process cacheable only if it's being 1060178479Sjb * grabbed read-only. If we're currently caching more process 1061178479Sjb * handles than dph_lrulim permits, attempt to find the 1062178479Sjb * least-recently-used handle that is currently unreferenced and 1063178479Sjb * release it from the cache. Otherwise we are grabbing the process 1064178479Sjb * for control: create a control thread for this process and store 1065178479Sjb * its ID in dpr->dpr_tid. 1066178479Sjb */ 1067178479Sjb if (nomonitor || (flags & PGRAB_RDONLY)) { 1068178479Sjb if (dph->dph_lrucnt >= dph->dph_lrulim) { 1069178479Sjb for (opr = dt_list_prev(&dph->dph_lrulist); 1070178479Sjb opr != NULL; opr = dt_list_prev(opr)) { 1071178479Sjb if (opr->dpr_cacheable && opr->dpr_refs == 0) { 1072178479Sjb dt_proc_destroy(dtp, opr->dpr_proc); 1073178479Sjb break; 1074178479Sjb } 1075178479Sjb } 1076178479Sjb } 1077178479Sjb 1078178479Sjb if (flags & PGRAB_RDONLY) { 1079178479Sjb dpr->dpr_cacheable = B_TRUE; 1080178479Sjb dpr->dpr_rdonly = B_TRUE; 1081178479Sjb dph->dph_lrucnt++; 1082178479Sjb } 1083178479Sjb 1084178479Sjb } else if (dt_proc_create_thread(dtp, dpr, DT_PROC_STOP_GRAB) != 0) 1085178479Sjb return (NULL); /* dt_proc_error() has been called for us */ 1086178479Sjb 1087178479Sjb dpr->dpr_hash = dph->dph_hash[h]; 1088178479Sjb dph->dph_hash[h] = dpr; 1089178479Sjb dt_list_prepend(&dph->dph_lrulist, dpr); 1090178479Sjb 1091178479Sjb dt_dprintf("grabbed pid %d\n", (int)pid); 1092178479Sjb dpr->dpr_refs++; 1093178479Sjb 1094178479Sjb return (dpr->dpr_proc); 1095178479Sjb} 1096178479Sjb 1097178479Sjbvoid 1098178479Sjbdt_proc_release(dtrace_hdl_t *dtp, struct ps_prochandle *P) 1099178479Sjb{ 1100178479Sjb dt_proc_t *dpr = dt_proc_lookup(dtp, P, B_FALSE); 1101178479Sjb dt_proc_hash_t *dph = dtp->dt_procs; 1102178479Sjb 1103178479Sjb assert(dpr != NULL); 1104178479Sjb assert(dpr->dpr_refs != 0); 1105178479Sjb 1106178479Sjb if (--dpr->dpr_refs == 0 && 1107178479Sjb (!dpr->dpr_cacheable || dph->dph_lrucnt > dph->dph_lrulim)) 1108178479Sjb dt_proc_destroy(dtp, P); 1109178479Sjb} 1110178479Sjb 1111178479Sjbvoid 1112178479Sjbdt_proc_continue(dtrace_hdl_t *dtp, struct ps_prochandle *P) 1113178479Sjb{ 1114178479Sjb dt_proc_t *dpr = dt_proc_lookup(dtp, P, B_FALSE); 1115178479Sjb 1116178479Sjb (void) pthread_mutex_lock(&dpr->dpr_lock); 1117178479Sjb 1118178479Sjb if (dpr->dpr_stop & DT_PROC_STOP_IDLE) { 1119178479Sjb dpr->dpr_stop &= ~DT_PROC_STOP_IDLE; 1120178479Sjb (void) pthread_cond_broadcast(&dpr->dpr_cv); 1121178479Sjb } 1122178479Sjb 1123178479Sjb (void) pthread_mutex_unlock(&dpr->dpr_lock); 1124178479Sjb} 1125178479Sjb 1126178479Sjbvoid 1127178479Sjbdt_proc_lock(dtrace_hdl_t *dtp, struct ps_prochandle *P) 1128178479Sjb{ 1129178479Sjb dt_proc_t *dpr = dt_proc_lookup(dtp, P, B_FALSE); 1130178479Sjb int err = pthread_mutex_lock(&dpr->dpr_lock); 1131178479Sjb assert(err == 0); /* check for recursion */ 1132178479Sjb} 1133178479Sjb 1134178479Sjbvoid 1135178479Sjbdt_proc_unlock(dtrace_hdl_t *dtp, struct ps_prochandle *P) 1136178479Sjb{ 1137178479Sjb dt_proc_t *dpr = dt_proc_lookup(dtp, P, B_FALSE); 1138178479Sjb int err = pthread_mutex_unlock(&dpr->dpr_lock); 1139178479Sjb assert(err == 0); /* check for unheld lock */ 1140178479Sjb} 1141178479Sjb 1142178479Sjbvoid 1143178479Sjbdt_proc_hash_create(dtrace_hdl_t *dtp) 1144178479Sjb{ 1145178479Sjb if ((dtp->dt_procs = dt_zalloc(dtp, sizeof (dt_proc_hash_t) + 1146178479Sjb sizeof (dt_proc_t *) * _dtrace_pidbuckets - 1)) != NULL) { 1147178479Sjb 1148178479Sjb (void) pthread_mutex_init(&dtp->dt_procs->dph_lock, NULL); 1149178479Sjb (void) pthread_cond_init(&dtp->dt_procs->dph_cv, NULL); 1150178479Sjb 1151178479Sjb dtp->dt_procs->dph_hashlen = _dtrace_pidbuckets; 1152178479Sjb dtp->dt_procs->dph_lrulim = _dtrace_pidlrulim; 1153178479Sjb } 1154178479Sjb} 1155178479Sjb 1156178479Sjbvoid 1157178479Sjbdt_proc_hash_destroy(dtrace_hdl_t *dtp) 1158178479Sjb{ 1159178479Sjb dt_proc_hash_t *dph = dtp->dt_procs; 1160178479Sjb dt_proc_t *dpr; 1161178479Sjb 1162178479Sjb while ((dpr = dt_list_next(&dph->dph_lrulist)) != NULL) 1163178479Sjb dt_proc_destroy(dtp, dpr->dpr_proc); 1164178479Sjb 1165178479Sjb dtp->dt_procs = NULL; 1166178479Sjb dt_free(dtp, dph); 1167178479Sjb} 1168178479Sjb 1169178479Sjbstruct ps_prochandle * 1170184696Srodrigcdtrace_proc_create(dtrace_hdl_t *dtp, const char *file, char *const *argv, 1171184696Srodrigc proc_child_func *pcf, void *child_arg) 1172178479Sjb{ 1173178479Sjb dt_ident_t *idp = dt_idhash_lookup(dtp->dt_macros, "target"); 1174184696Srodrigc struct ps_prochandle *P = dt_proc_create(dtp, file, argv, pcf, child_arg); 1175178479Sjb 1176211554Srpaulo if (P != NULL && idp != NULL && idp->di_id == 0) { 1177178565Sjb#if defined(sun) 1178178479Sjb idp->di_id = Pstatus(P)->pr_pid; /* $target = created pid */ 1179178565Sjb#else 1180178565Sjb idp->di_id = proc_getpid(P); /* $target = created pid */ 1181178565Sjb#endif 1182211554Srpaulo } 1183178479Sjb 1184178479Sjb return (P); 1185178479Sjb} 1186178479Sjb 1187178479Sjbstruct ps_prochandle * 1188178479Sjbdtrace_proc_grab(dtrace_hdl_t *dtp, pid_t pid, int flags) 1189178479Sjb{ 1190178479Sjb dt_ident_t *idp = dt_idhash_lookup(dtp->dt_macros, "target"); 1191178479Sjb struct ps_prochandle *P = dt_proc_grab(dtp, pid, flags, 0); 1192178479Sjb 1193178479Sjb if (P != NULL && idp != NULL && idp->di_id == 0) 1194178479Sjb idp->di_id = pid; /* $target = grabbed pid */ 1195178479Sjb 1196178479Sjb return (P); 1197178479Sjb} 1198178479Sjb 1199178479Sjbvoid 1200178479Sjbdtrace_proc_release(dtrace_hdl_t *dtp, struct ps_prochandle *P) 1201178479Sjb{ 1202178479Sjb dt_proc_release(dtp, P); 1203178479Sjb} 1204178479Sjb 1205178479Sjbvoid 1206178479Sjbdtrace_proc_continue(dtrace_hdl_t *dtp, struct ps_prochandle *P) 1207178479Sjb{ 1208178479Sjb dt_proc_continue(dtp, P); 1209178479Sjb} 1210