4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
23 * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
24 * Use is subject to license terms.
28 * DTrace Process Control
30 * This file provides a set of routines that permit libdtrace and its clients
31 * to create and grab process handles using libproc, and to share these handles
32 * between library mechanisms that need libproc access, such as ustack(), and
33 * client mechanisms that need libproc access, such as dtrace(1M) -c and -p.
34 * The library provides several mechanisms in the libproc control layer:
36 * Reference Counting: The library code and client code can independently grab
37 * the same process handles without interfering with one another. Only when
38 * the reference count drops to zero and the handle is not being cached (see
39 * below for more information on caching) will Prelease() be called on it.
41 * Handle Caching: If a handle is grabbed PGRAB_RDONLY (e.g. by ustack()) and
42 * the reference count drops to zero, the handle is not immediately released.
43 * Instead, libproc handles are maintained on dph_lrulist in order from most-
44 * recently accessed to least-recently accessed. Idle handles are maintained
45 * until a pre-defined LRU cache limit is exceeded, permitting repeated calls
46 * to ustack() to avoid the overhead of releasing and re-grabbing processes.
48 * Process Control: For processes that are grabbed for control (~PGRAB_RDONLY)
49 * or created by dt_proc_create(), a control thread is created to provide
50 * callbacks on process exit and symbol table caching on dlopen()s.
52 * MT-Safety: Libproc is not MT-Safe, so dt_proc_lock() and dt_proc_unlock()
53 * are provided to synchronize access to the libproc handle between libdtrace
54 * code and client code and the control thread's use of the ps_prochandle.
56 * NOTE: MT-Safety is NOT provided for libdtrace itself, or for use of the
57 * dtrace_proc_grab/dtrace_proc_create mechanisms. Like all exported libdtrace
58 * calls, these are assumed to be MT-Unsafe. MT-Safety is ONLY provided for
59 * synchronization between libdtrace control threads and the client thread.
61 * The ps_prochandles themselves are maintained along with a dt_proc_t struct
62 * in a hash table indexed by PID. This provides basic locking and reference
63 * counting. The dt_proc_t is also maintained in LRU order on dph_lrulist.
64 * The dph_lrucnt and dph_lrulim count the number of cacheable processes and
65 * the current limit on the number of actively cached entries.
67 * The control thread for a process establishes breakpoints at the rtld_db
68 * locations of interest, updates mappings and symbol tables at these points,
69 * and handles exec and fork (by always following the parent). The control
70 * thread automatically exits when the process dies or control is lost.
72 * A simple notification mechanism is provided for libdtrace clients using
73 * dtrace_handle_proc() for notification of PS_UNDEAD or PS_LOST events. If
74 * such an event occurs, the dt_proc_t itself is enqueued on a notification
75 * list and the control thread broadcasts to dph_cv. dtrace_sleep() will wake
76 * up using this condition and will then call the client handler as necessary.
93 #include <sys/syscall.h>
94 #include <libproc_compat.h>
95 #define SYS_forksys SYS_fork
98 #define IS_SYS_EXEC(w) (w == SYS_execve)
99 #define IS_SYS_FORK(w) (w == SYS_vfork || w == SYS_forksys)
102 dt_proc_bpcreate(dt_proc_t *dpr, uintptr_t addr, dt_bkpt_f *func, void *data)
104 struct ps_prochandle *P = dpr->dpr_proc;
107 assert(DT_MUTEX_HELD(&dpr->dpr_lock));
109 if ((dbp = dt_zalloc(dpr->dpr_hdl, sizeof (dt_bkpt_t))) != NULL) {
110 dbp->dbp_func = func;
111 dbp->dbp_data = data;
112 dbp->dbp_addr = addr;
114 if (Psetbkpt(P, dbp->dbp_addr, &dbp->dbp_instr) == 0)
115 dbp->dbp_active = B_TRUE;
117 dt_list_append(&dpr->dpr_bps, dbp);
124 dt_proc_bpdestroy(dt_proc_t *dpr, int delbkpts)
126 int state = Pstate(dpr->dpr_proc);
127 dt_bkpt_t *dbp, *nbp;
129 assert(DT_MUTEX_HELD(&dpr->dpr_lock));
131 for (dbp = dt_list_next(&dpr->dpr_bps); dbp != NULL; dbp = nbp) {
132 if (delbkpts && dbp->dbp_active &&
133 state != PS_LOST && state != PS_UNDEAD) {
134 (void) Pdelbkpt(dpr->dpr_proc,
135 dbp->dbp_addr, dbp->dbp_instr);
137 nbp = dt_list_next(dbp);
138 dt_list_delete(&dpr->dpr_bps, dbp);
139 dt_free(dpr->dpr_hdl, dbp);
144 dt_proc_bpmatch(dtrace_hdl_t *dtp, dt_proc_t *dpr)
147 const lwpstatus_t *psp = &Pstatus(dpr->dpr_proc)->pr_lwp;
153 assert(DT_MUTEX_HELD(&dpr->dpr_lock));
156 proc_regget(dpr->dpr_proc, REG_PC, &pc);
157 proc_bkptregadj(&pc);
160 for (dbp = dt_list_next(&dpr->dpr_bps);
161 dbp != NULL; dbp = dt_list_next(dbp)) {
163 if (psp->pr_reg[R_PC] == dbp->dbp_addr)
166 if (pc == dbp->dbp_addr)
172 dt_dprintf("pid %d: spurious breakpoint wakeup for %lx\n",
174 (int)dpr->dpr_pid, (ulong_t)psp->pr_reg[R_PC]);
176 (int)dpr->dpr_pid, pc);
181 dt_dprintf("pid %d: hit breakpoint at %lx (%lu)\n",
182 (int)dpr->dpr_pid, (ulong_t)dbp->dbp_addr, ++dbp->dbp_hits);
184 dbp->dbp_func(dtp, dpr, dbp->dbp_data);
185 (void) Pxecbkpt(dpr->dpr_proc, dbp->dbp_instr);
189 dt_proc_bpenable(dt_proc_t *dpr)
193 assert(DT_MUTEX_HELD(&dpr->dpr_lock));
195 for (dbp = dt_list_next(&dpr->dpr_bps);
196 dbp != NULL; dbp = dt_list_next(dbp)) {
197 if (!dbp->dbp_active && Psetbkpt(dpr->dpr_proc,
198 dbp->dbp_addr, &dbp->dbp_instr) == 0)
199 dbp->dbp_active = B_TRUE;
202 dt_dprintf("breakpoints enabled\n");
206 dt_proc_bpdisable(dt_proc_t *dpr)
210 assert(DT_MUTEX_HELD(&dpr->dpr_lock));
212 for (dbp = dt_list_next(&dpr->dpr_bps);
213 dbp != NULL; dbp = dt_list_next(dbp)) {
214 if (dbp->dbp_active && Pdelbkpt(dpr->dpr_proc,
215 dbp->dbp_addr, dbp->dbp_instr) == 0)
216 dbp->dbp_active = B_FALSE;
219 dt_dprintf("breakpoints disabled\n");
223 dt_proc_notify(dtrace_hdl_t *dtp, dt_proc_hash_t *dph, dt_proc_t *dpr,
226 dt_proc_notify_t *dprn = dt_alloc(dtp, sizeof (dt_proc_notify_t));
229 dt_dprintf("failed to allocate notification for %d %s\n",
230 (int)dpr->dpr_pid, msg);
232 dprn->dprn_dpr = dpr;
234 dprn->dprn_errmsg[0] = '\0';
236 (void) strlcpy(dprn->dprn_errmsg, msg,
237 sizeof (dprn->dprn_errmsg));
239 (void) pthread_mutex_lock(&dph->dph_lock);
241 dprn->dprn_next = dph->dph_notify;
242 dph->dph_notify = dprn;
244 (void) pthread_cond_broadcast(&dph->dph_cv);
245 (void) pthread_mutex_unlock(&dph->dph_lock);
250 * Check to see if the control thread was requested to stop when the victim
251 * process reached a particular event (why) rather than continuing the victim.
252 * If 'why' is set in the stop mask, we wait on dpr_cv for dt_proc_continue().
253 * If 'why' is not set, this function returns immediately and does nothing.
256 dt_proc_stop(dt_proc_t *dpr, uint8_t why)
258 assert(DT_MUTEX_HELD(&dpr->dpr_lock));
259 assert(why != DT_PROC_STOP_IDLE);
261 if (dpr->dpr_stop & why) {
262 dpr->dpr_stop |= DT_PROC_STOP_IDLE;
263 dpr->dpr_stop &= ~why;
265 (void) pthread_cond_broadcast(&dpr->dpr_cv);
268 * We disable breakpoints while stopped to preserve the
269 * integrity of the program text for both our own disassembly
270 * and that of the kernel.
272 dt_proc_bpdisable(dpr);
274 while (dpr->dpr_stop & DT_PROC_STOP_IDLE)
275 (void) pthread_cond_wait(&dpr->dpr_cv, &dpr->dpr_lock);
277 dt_proc_bpenable(dpr);
283 dt_proc_bpmain(dtrace_hdl_t *dtp, dt_proc_t *dpr, const char *fname)
285 dt_dprintf("pid %d: breakpoint at %s()\n", (int)dpr->dpr_pid, fname);
286 dt_proc_stop(dpr, DT_PROC_STOP_MAIN);
290 dt_proc_rdevent(dtrace_hdl_t *dtp, dt_proc_t *dpr, const char *evname)
295 if ((err = rd_event_getmsg(dpr->dpr_rtld, &rdm)) != RD_OK) {
296 dt_dprintf("pid %d: failed to get %s event message: %s\n",
297 (int)dpr->dpr_pid, evname, rd_errstr(err));
301 dt_dprintf("pid %d: rtld event %s type=%d state %d\n",
302 (int)dpr->dpr_pid, evname, rdm.type, rdm.u.state);
306 if (rdm.u.state != RD_CONSISTENT)
309 Pupdate_syms(dpr->dpr_proc);
310 if (dt_pid_create_probes_module(dtp, dpr) != 0)
311 dt_proc_notify(dtp, dtp->dt_procs, dpr,
316 Pupdate_syms(dpr->dpr_proc);
317 dt_proc_stop(dpr, DT_PROC_STOP_PREINIT);
320 Pupdate_syms(dpr->dpr_proc);
321 dt_proc_stop(dpr, DT_PROC_STOP_POSTINIT);
327 dt_proc_rdwatch(dt_proc_t *dpr, rd_event_e event, const char *evname)
332 if ((err = rd_event_addr(dpr->dpr_rtld, event, &rdn)) != RD_OK) {
333 dt_dprintf("pid %d: failed to get event address for %s: %s\n",
334 (int)dpr->dpr_pid, evname, rd_errstr(err));
338 if (rdn.type != RD_NOTIFY_BPT) {
339 dt_dprintf("pid %d: event %s has unexpected type %d\n",
340 (int)dpr->dpr_pid, evname, rdn.type);
344 (void) dt_proc_bpcreate(dpr, rdn.u.bptaddr,
346 (dt_bkpt_f *)dt_proc_rdevent, (void *)evname);
349 (dt_bkpt_f *)dt_proc_rdevent, __DECONST(void *, evname));
354 * Common code for enabling events associated with the run-time linker after
355 * attaching to a process or after a victim process completes an exec(2).
358 dt_proc_attach(dt_proc_t *dpr, int exec)
361 const pstatus_t *psp = Pstatus(dpr->dpr_proc);
366 assert(DT_MUTEX_HELD(&dpr->dpr_lock));
370 if (psp->pr_lwp.pr_errno != 0)
371 return; /* exec failed: nothing needs to be done */
374 dt_proc_bpdestroy(dpr, B_FALSE);
376 Preset_maps(dpr->dpr_proc);
379 if ((dpr->dpr_rtld = Prd_agent(dpr->dpr_proc)) != NULL &&
380 (err = rd_event_enable(dpr->dpr_rtld, B_TRUE)) == RD_OK) {
382 dt_proc_rdwatch(dpr, RD_PREINIT, "RD_PREINIT");
384 dt_proc_rdwatch(dpr, RD_POSTINIT, "RD_POSTINIT");
386 dt_proc_rdwatch(dpr, RD_DLACTIVITY, "RD_DLACTIVITY");
389 dt_dprintf("pid %d: failed to enable rtld events: %s\n",
390 (int)dpr->dpr_pid, dpr->dpr_rtld ? rd_errstr(err) :
391 "rtld_db agent initialization failed");
394 Pupdate_maps(dpr->dpr_proc);
396 if (Pxlookup_by_name(dpr->dpr_proc, LM_ID_BASE,
397 "a.out", "main", &sym, NULL) == 0) {
398 (void) dt_proc_bpcreate(dpr, (uintptr_t)sym.st_value,
399 (dt_bkpt_f *)dt_proc_bpmain, "a.out`main");
401 dt_dprintf("pid %d: failed to find a.out`main: %s\n",
402 (int)dpr->dpr_pid, strerror(errno));
407 * Wait for a stopped process to be set running again by some other debugger.
408 * This is typically not required by /proc-based debuggers, since the usual
409 * model is that one debugger controls one victim. But DTrace, as usual, has
410 * its own needs: the stop() action assumes that prun(1) or some other tool
411 * will be applied to resume the victim process. This could be solved by
412 * adding a PCWRUN directive to /proc, but that seems like overkill unless
413 * other debuggers end up needing this functionality, so we implement a cheap
414 * equivalent to PCWRUN using the set of existing kernel mechanisms.
416 * Our intent is really not just to wait for the victim to run, but rather to
417 * wait for it to run and then stop again for a reason other than the current
418 * PR_REQUESTED stop. Since PCWSTOP/Pstopstatus() can be applied repeatedly
419 * to a stopped process and will return the same result without affecting the
420 * victim, we can just perform these operations repeatedly until Pstate()
421 * changes, the representative LWP ID changes, or the stop timestamp advances.
422 * dt_proc_control() will then rediscover the new state and continue as usual.
423 * When the process is still stopped in the same exact state, we sleep for a
424 * brief interval before waiting again so as not to spin consuming CPU cycles.
427 dt_proc_waitrun(dt_proc_t *dpr)
429 printf("%s:%s(%d): DOODAD\n",__FUNCTION__,__FILE__,__LINE__);
431 struct ps_prochandle *P = dpr->dpr_proc;
432 const lwpstatus_t *psp = &Pstatus(P)->pr_lwp;
434 int krflag = psp->pr_flags & (PR_KLC | PR_RLC);
435 timestruc_t tstamp = psp->pr_tstamp;
436 lwpid_t lwpid = psp->pr_lwpid;
438 const long wstop = PCWSTOP;
441 assert(DT_MUTEX_HELD(&dpr->dpr_lock));
442 assert(psp->pr_flags & PR_STOPPED);
443 assert(Pstate(P) == PS_STOP);
446 * While we are waiting for the victim to run, clear PR_KLC and PR_RLC
447 * so that if the libdtrace client is killed, the victim stays stopped.
448 * dt_proc_destroy() will also observe this and perform PRELEASE_HANG.
450 (void) Punsetflags(P, krflag);
453 (void) pthread_mutex_unlock(&dpr->dpr_lock);
455 while (!dpr->dpr_quit) {
456 if (write(pfd, &wstop, sizeof (wstop)) == -1 && errno == EINTR)
457 continue; /* check dpr_quit and continue waiting */
459 (void) pthread_mutex_lock(&dpr->dpr_lock);
460 (void) Pstopstatus(P, PCNULL, 0);
461 psp = &Pstatus(P)->pr_lwp;
464 * If we've reached a new state, found a new representative, or
465 * the stop timestamp has changed, restore PR_KLC/PR_RLC to its
466 * original setting and then return with dpr_lock held.
468 if (Pstate(P) != PS_STOP || psp->pr_lwpid != lwpid ||
469 bcmp(&psp->pr_tstamp, &tstamp, sizeof (tstamp)) != 0) {
470 (void) Psetflags(P, krflag);
475 (void) pthread_mutex_unlock(&dpr->dpr_lock);
476 (void) poll(NULL, 0, MILLISEC / 2);
479 (void) pthread_mutex_lock(&dpr->dpr_lock);
483 typedef struct dt_proc_control_data {
484 dtrace_hdl_t *dpcd_hdl; /* DTrace handle */
485 dt_proc_t *dpcd_proc; /* proccess to control */
486 } dt_proc_control_data_t;
489 * Main loop for all victim process control threads. We initialize all the
490 * appropriate /proc control mechanisms, and then enter a loop waiting for
491 * the process to stop on an event or die. We process any events by calling
492 * appropriate subroutines, and exit when the victim dies or we lose control.
494 * The control thread synchronizes the use of dpr_proc with other libdtrace
495 * threads using dpr_lock. We hold the lock for all of our operations except
496 * waiting while the process is running: this is accomplished by writing a
497 * PCWSTOP directive directly to the underlying /proc/<pid>/ctl file. If the
498 * libdtrace client wishes to exit or abort our wait, SIGCANCEL can be used.
501 dt_proc_control(void *arg)
503 dt_proc_control_data_t *datap = arg;
504 dtrace_hdl_t *dtp = datap->dpcd_hdl;
505 dt_proc_t *dpr = datap->dpcd_proc;
506 dt_proc_hash_t *dph = dpr->dpr_hdl->dt_procs;
507 struct ps_prochandle *P = dpr->dpr_proc;
508 int pid = dpr->dpr_pid;
513 const long wstop = PCWSTOP;
515 int notify = B_FALSE;
518 * We disable the POSIX thread cancellation mechanism so that the
519 * client program using libdtrace can't accidentally cancel our thread.
520 * dt_proc_destroy() uses SIGCANCEL explicitly to simply poke us out
521 * of PCWSTOP with EINTR, at which point we will see dpr_quit and exit.
523 (void) pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, NULL);
526 * Set up the corresponding process for tracing by libdtrace. We want
527 * to be able to catch breakpoints and efficiently single-step over
528 * them, and we need to enable librtld_db to watch libdl activity.
530 (void) pthread_mutex_lock(&dpr->dpr_lock);
533 (void) Punsetflags(P, PR_ASYNC); /* require synchronous mode */
534 (void) Psetflags(P, PR_BPTADJ); /* always adjust eip on x86 */
535 (void) Punsetflags(P, PR_FORK); /* do not inherit on fork */
537 (void) Pfault(P, FLTBPT, B_TRUE); /* always trace breakpoints */
538 (void) Pfault(P, FLTTRACE, B_TRUE); /* always trace single-step */
541 * We must trace exit from exec() system calls so that if the exec is
542 * successful, we can reset our breakpoints and re-initialize libproc.
544 (void) Psysexit(P, SYS_execve, B_TRUE);
547 * We must trace entry and exit for fork() system calls in order to
548 * disable our breakpoints temporarily during the fork. We do not set
549 * the PR_FORK flag, so if fork succeeds the child begins executing and
550 * does not inherit any other tracing behaviors or a control thread.
552 (void) Psysentry(P, SYS_vfork, B_TRUE);
553 (void) Psysexit(P, SYS_vfork, B_TRUE);
554 (void) Psysentry(P, SYS_forksys, B_TRUE);
555 (void) Psysexit(P, SYS_forksys, B_TRUE);
557 Psync(P); /* enable all /proc changes */
559 dt_proc_attach(dpr, B_FALSE); /* enable rtld breakpoints */
562 * If PR_KLC is set, we created the process; otherwise we grabbed it.
563 * Check for an appropriate stop request and wait for dt_proc_continue.
566 if (Pstatus(P)->pr_flags & PR_KLC)
568 if (proc_getflags(P) & PR_KLC)
570 dt_proc_stop(dpr, DT_PROC_STOP_CREATE);
572 dt_proc_stop(dpr, DT_PROC_STOP_GRAB);
574 if (Psetrun(P, 0, 0) == -1) {
575 dt_dprintf("pid %d: failed to set running: %s\n",
576 (int)dpr->dpr_pid, strerror(errno));
579 (void) pthread_mutex_unlock(&dpr->dpr_lock);
582 * Wait for the process corresponding to this control thread to stop,
583 * process the event, and then set it running again. We want to sleep
584 * with dpr_lock *unheld* so that other parts of libdtrace can use the
585 * ps_prochandle in the meantime (e.g. ustack()). To do this, we write
586 * a PCWSTOP directive directly to the underlying /proc/<pid>/ctl file.
587 * Once the process stops, we wake up, grab dpr_lock, and then call
588 * Pwait() (which will return immediately) and do our processing.
590 while (!dpr->dpr_quit) {
591 const lwpstatus_t *psp;
594 if (write(pfd, &wstop, sizeof (wstop)) == -1 && errno == EINTR)
595 continue; /* check dpr_quit and continue waiting */
597 /* Wait for the process to report status. */
600 continue; /* check dpr_quit and continue waiting */
603 (void) pthread_mutex_lock(&dpr->dpr_lock);
607 if (Pstopstatus(P, PCNULL, 0) == -1 && errno == EINTR) {
608 (void) pthread_mutex_unlock(&dpr->dpr_lock);
609 continue; /* check dpr_quit and continue waiting */
616 psp = &Pstatus(P)->pr_lwp;
618 psp = proc_getlwpstatus(P);
621 dt_dprintf("pid %d: proc stopped showing %d/%d\n",
622 pid, psp->pr_why, psp->pr_what);
625 * If the process stops showing PR_REQUESTED, then the
626 * DTrace stop() action was applied to it or another
627 * debugging utility (e.g. pstop(1)) asked it to stop.
628 * In either case, the user's intention is for the
629 * process to remain stopped until another external
630 * mechanism (e.g. prun(1)) is applied. So instead of
631 * setting the process running ourself, we wait for
632 * someone else to do so. Once that happens, we return
633 * to our normal loop waiting for an event of interest.
635 if (psp->pr_why == PR_REQUESTED) {
636 dt_proc_waitrun(dpr);
637 (void) pthread_mutex_unlock(&dpr->dpr_lock);
642 * If the process stops showing one of the events that
643 * we are tracing, perform the appropriate response.
644 * Note that we ignore PR_SUSPENDED, PR_CHECKPOINT, and
645 * PR_JOBCONTROL by design: if one of these conditions
646 * occurs, we will fall through to Psetrun() but the
647 * process will remain stopped in the kernel by the
648 * corresponding mechanism (e.g. job control stop).
650 if (psp->pr_why == PR_FAULTED && psp->pr_what == FLTBPT)
651 dt_proc_bpmatch(dtp, dpr);
652 else if (psp->pr_why == PR_SYSENTRY &&
653 IS_SYS_FORK(psp->pr_what))
654 dt_proc_bpdisable(dpr);
655 else if (psp->pr_why == PR_SYSEXIT &&
656 IS_SYS_FORK(psp->pr_what))
657 dt_proc_bpenable(dpr);
658 else if (psp->pr_why == PR_SYSEXIT &&
659 IS_SYS_EXEC(psp->pr_what))
660 dt_proc_attach(dpr, B_TRUE);
669 dt_dprintf("pid %d: proc lost: %s\n",
670 pid, strerror(errno));
672 dpr->dpr_quit = B_TRUE;
677 dt_dprintf("pid %d: proc died\n", pid);
678 dpr->dpr_quit = B_TRUE;
683 if (Pstate(P) != PS_UNDEAD && Psetrun(P, 0, 0) == -1) {
684 dt_dprintf("pid %d: failed to set running: %s\n",
685 (int)dpr->dpr_pid, strerror(errno));
688 (void) pthread_mutex_unlock(&dpr->dpr_lock);
692 * If the control thread detected PS_UNDEAD or PS_LOST, then enqueue
693 * the dt_proc_t structure on the dt_proc_hash_t notification list.
696 dt_proc_notify(dtp, dph, dpr, NULL);
699 * Destroy and remove any remaining breakpoints, set dpr_done and clear
700 * dpr_tid to indicate the control thread has exited, and notify any
701 * waiting thread in dt_proc_destroy() that we have succesfully exited.
703 (void) pthread_mutex_lock(&dpr->dpr_lock);
705 dt_proc_bpdestroy(dpr, B_TRUE);
706 dpr->dpr_done = B_TRUE;
709 (void) pthread_cond_broadcast(&dpr->dpr_cv);
710 (void) pthread_mutex_unlock(&dpr->dpr_lock);
716 static struct ps_prochandle *
717 dt_proc_error(dtrace_hdl_t *dtp, dt_proc_t *dpr, const char *format, ...)
721 va_start(ap, format);
722 dt_set_errmsg(dtp, NULL, NULL, NULL, 0, format, ap);
725 if (dpr->dpr_proc != NULL)
726 Prelease(dpr->dpr_proc, 0);
729 (void) dt_set_errno(dtp, EDT_COMPILER);
734 dt_proc_lookup(dtrace_hdl_t *dtp, struct ps_prochandle *P, int remove)
736 dt_proc_hash_t *dph = dtp->dt_procs;
738 pid_t pid = Pstatus(P)->pr_pid;
740 pid_t pid = proc_getpid(P);
742 dt_proc_t *dpr, **dpp = &dph->dph_hash[pid & (dph->dph_hashlen - 1)];
744 for (dpr = *dpp; dpr != NULL; dpr = dpr->dpr_hash) {
745 if (dpr->dpr_pid == pid)
748 dpp = &dpr->dpr_hash;
752 assert(dpr->dpr_proc == P);
755 *dpp = dpr->dpr_hash; /* remove from pid hash chain */
761 dt_proc_destroy(dtrace_hdl_t *dtp, struct ps_prochandle *P)
763 dt_proc_t *dpr = dt_proc_lookup(dtp, P, B_FALSE);
764 dt_proc_hash_t *dph = dtp->dt_procs;
765 dt_proc_notify_t *npr, **npp;
771 * If neither PR_KLC nor PR_RLC is set, then the process is stopped by
772 * an external debugger and we were waiting in dt_proc_waitrun().
773 * Leave the process in this condition using PRELEASE_HANG.
776 if (!(Pstatus(dpr->dpr_proc)->pr_flags & (PR_KLC | PR_RLC))) {
778 if (!(proc_getflags(dpr->dpr_proc) & (PR_KLC | PR_RLC))) {
780 dt_dprintf("abandoning pid %d\n", (int)dpr->dpr_pid);
781 rflag = PRELEASE_HANG;
783 } else if (Pstatus(dpr->dpr_proc)->pr_flags & PR_KLC) {
785 } else if (proc_getflags(dpr->dpr_proc) & PR_KLC) {
787 dt_dprintf("killing pid %d\n", (int)dpr->dpr_pid);
788 rflag = PRELEASE_KILL; /* apply kill-on-last-close */
790 dt_dprintf("releasing pid %d\n", (int)dpr->dpr_pid);
791 rflag = 0; /* apply run-on-last-close */
796 * Set the dpr_quit flag to tell the daemon thread to exit. We
797 * send it a SIGCANCEL to poke it out of PCWSTOP or any other
798 * long-term /proc system call. Our daemon threads have POSIX
799 * cancellation disabled, so EINTR will be the only effect. We
800 * then wait for dpr_done to indicate the thread has exited.
802 * We can't use pthread_kill() to send SIGCANCEL because the
803 * interface forbids it and we can't use pthread_cancel()
804 * because with cancellation disabled it won't actually
805 * send SIGCANCEL to the target thread, so we use _lwp_kill()
806 * to do the job. This is all built on evil knowledge of
807 * the details of the cancellation mechanism in libc.
809 (void) pthread_mutex_lock(&dpr->dpr_lock);
810 dpr->dpr_quit = B_TRUE;
812 (void) _lwp_kill(dpr->dpr_tid, SIGCANCEL);
814 pthread_kill(dpr->dpr_tid, SIGTHR);
818 * If the process is currently idling in dt_proc_stop(), re-
819 * enable breakpoints and poke it into running again.
821 if (dpr->dpr_stop & DT_PROC_STOP_IDLE) {
822 dt_proc_bpenable(dpr);
823 dpr->dpr_stop &= ~DT_PROC_STOP_IDLE;
824 (void) pthread_cond_broadcast(&dpr->dpr_cv);
827 while (!dpr->dpr_done)
828 (void) pthread_cond_wait(&dpr->dpr_cv, &dpr->dpr_lock);
830 (void) pthread_mutex_unlock(&dpr->dpr_lock);
834 * Before we free the process structure, remove this dt_proc_t from the
835 * lookup hash, and then walk the dt_proc_hash_t's notification list
836 * and remove this dt_proc_t if it is enqueued.
838 (void) pthread_mutex_lock(&dph->dph_lock);
839 (void) dt_proc_lookup(dtp, P, B_TRUE);
840 npp = &dph->dph_notify;
842 while ((npr = *npp) != NULL) {
843 if (npr->dprn_dpr == dpr) {
844 *npp = npr->dprn_next;
847 npp = &npr->dprn_next;
851 (void) pthread_mutex_unlock(&dph->dph_lock);
854 * Remove the dt_proc_list from the LRU list, release the underlying
855 * libproc handle, and free our dt_proc_t data structure.
857 if (dpr->dpr_cacheable) {
858 assert(dph->dph_lrucnt != 0);
862 dt_list_delete(&dph->dph_lrulist, dpr);
863 Prelease(dpr->dpr_proc, rflag);
868 dt_proc_create_thread(dtrace_hdl_t *dtp, dt_proc_t *dpr, uint_t stop)
870 dt_proc_control_data_t data;
875 (void) pthread_mutex_lock(&dpr->dpr_lock);
876 dpr->dpr_stop |= stop; /* set bit for initial rendezvous */
878 (void) pthread_attr_init(&a);
879 (void) pthread_attr_setdetachstate(&a, PTHREAD_CREATE_DETACHED);
881 (void) sigfillset(&nset);
882 (void) sigdelset(&nset, SIGABRT); /* unblocked for assert() */
884 (void) sigdelset(&nset, SIGCANCEL); /* see dt_proc_destroy() */
886 (void) sigdelset(&nset, SIGUSR1); /* see dt_proc_destroy() */
890 data.dpcd_proc = dpr;
892 (void) pthread_sigmask(SIG_SETMASK, &nset, &oset);
893 err = pthread_create(&dpr->dpr_tid, &a, dt_proc_control, &data);
894 (void) pthread_sigmask(SIG_SETMASK, &oset, NULL);
897 * If the control thread was created, then wait on dpr_cv for either
898 * dpr_done to be set (the victim died or the control thread failed)
899 * or DT_PROC_STOP_IDLE to be set, indicating that the victim is now
900 * stopped by /proc and the control thread is at the rendezvous event.
901 * On success, we return with the process and control thread stopped:
902 * the caller can then apply dt_proc_continue() to resume both.
905 while (!dpr->dpr_done && !(dpr->dpr_stop & DT_PROC_STOP_IDLE))
906 (void) pthread_cond_wait(&dpr->dpr_cv, &dpr->dpr_lock);
909 * If dpr_done is set, the control thread aborted before it
910 * reached the rendezvous event. This is either due to PS_LOST
911 * or PS_UNDEAD (i.e. the process died). We try to provide a
912 * small amount of useful information to help figure it out.
916 const psinfo_t *prp = Ppsinfo(dpr->dpr_proc);
917 int stat = prp ? prp->pr_wstat : 0;
918 int pid = dpr->dpr_pid;
920 int stat = proc_getwstat(dpr->dpr_proc);
921 int pid = proc_getpid(dpr->dpr_proc);
923 if (proc_state(dpr->dpr_proc) == PS_LOST) {
924 (void) dt_proc_error(dpr->dpr_hdl, dpr,
925 "failed to control pid %d: process exec'd "
926 "set-id or unobservable program\n", pid);
927 } else if (WIFSIGNALED(stat)) {
928 (void) dt_proc_error(dpr->dpr_hdl, dpr,
929 "failed to control pid %d: process died "
930 "from signal %d\n", pid, WTERMSIG(stat));
932 (void) dt_proc_error(dpr->dpr_hdl, dpr,
933 "failed to control pid %d: process exited "
934 "with status %d\n", pid, WEXITSTATUS(stat));
937 err = ESRCH; /* cause grab() or create() to fail */
940 (void) dt_proc_error(dpr->dpr_hdl, dpr,
941 "failed to create control thread for process-id %d: %s\n",
942 (int)dpr->dpr_pid, strerror(err));
946 (void) pthread_mutex_unlock(&dpr->dpr_lock);
947 (void) pthread_attr_destroy(&a);
952 struct ps_prochandle *
953 dt_proc_create(dtrace_hdl_t *dtp, const char *file, char *const *argv,
954 proc_child_func *pcf, void *child_arg)
956 dt_proc_hash_t *dph = dtp->dt_procs;
960 if ((dpr = dt_zalloc(dtp, sizeof (dt_proc_t))) == NULL)
961 return (NULL); /* errno is set for us */
963 (void) pthread_mutex_init(&dpr->dpr_lock, NULL);
964 (void) pthread_cond_init(&dpr->dpr_cv, NULL);
967 if ((dpr->dpr_proc = Pcreate(file, argv, &err, NULL, 0)) == NULL) {
969 if ((err = proc_create(file, argv, pcf, child_arg,
970 &dpr->dpr_proc)) != 0) {
972 return (dt_proc_error(dtp, dpr,
973 "failed to execute %s: %s\n", file, Pcreate_error(err)));
978 dpr->dpr_pid = Pstatus(dpr->dpr_proc)->pr_pid;
980 dpr->dpr_pid = proc_getpid(dpr->dpr_proc);
983 (void) Punsetflags(dpr->dpr_proc, PR_RLC);
984 (void) Psetflags(dpr->dpr_proc, PR_KLC);
986 if (dt_proc_create_thread(dtp, dpr, dtp->dt_prcmode) != 0)
987 return (NULL); /* dt_proc_error() has been called for us */
989 dpr->dpr_hash = dph->dph_hash[dpr->dpr_pid & (dph->dph_hashlen - 1)];
990 dph->dph_hash[dpr->dpr_pid & (dph->dph_hashlen - 1)] = dpr;
991 dt_list_prepend(&dph->dph_lrulist, dpr);
993 dt_dprintf("created pid %d\n", (int)dpr->dpr_pid);
996 return (dpr->dpr_proc);
999 struct ps_prochandle *
1000 dt_proc_grab(dtrace_hdl_t *dtp, pid_t pid, int flags, int nomonitor)
1002 dt_proc_hash_t *dph = dtp->dt_procs;
1003 uint_t h = pid & (dph->dph_hashlen - 1);
1004 dt_proc_t *dpr, *opr;
1008 * Search the hash table for the pid. If it is already grabbed or
1009 * created, move the handle to the front of the lrulist, increment
1010 * the reference count, and return the existing ps_prochandle.
1012 for (dpr = dph->dph_hash[h]; dpr != NULL; dpr = dpr->dpr_hash) {
1013 if (dpr->dpr_pid == pid && !dpr->dpr_stale) {
1015 * If the cached handle was opened read-only and
1016 * this request is for a writeable handle, mark
1017 * the cached handle as stale and open a new handle.
1018 * Since it's stale, unmark it as cacheable.
1020 if (dpr->dpr_rdonly && !(flags & PGRAB_RDONLY)) {
1021 dt_dprintf("upgrading pid %d\n", (int)pid);
1022 dpr->dpr_stale = B_TRUE;
1023 dpr->dpr_cacheable = B_FALSE;
1028 dt_dprintf("grabbed pid %d (cached)\n", (int)pid);
1029 dt_list_delete(&dph->dph_lrulist, dpr);
1030 dt_list_prepend(&dph->dph_lrulist, dpr);
1032 return (dpr->dpr_proc);
1036 if ((dpr = dt_zalloc(dtp, sizeof (dt_proc_t))) == NULL)
1037 return (NULL); /* errno is set for us */
1039 (void) pthread_mutex_init(&dpr->dpr_lock, NULL);
1040 (void) pthread_cond_init(&dpr->dpr_cv, NULL);
1043 if ((dpr->dpr_proc = Pgrab(pid, flags, &err)) == NULL) {
1045 if ((err = proc_attach(pid, flags, &dpr->dpr_proc)) != 0) {
1047 return (dt_proc_error(dtp, dpr,
1048 "failed to grab pid %d: %s\n", (int)pid, Pgrab_error(err)));
1054 (void) Punsetflags(dpr->dpr_proc, PR_KLC);
1055 (void) Psetflags(dpr->dpr_proc, PR_RLC);
1058 * If we are attempting to grab the process without a monitor
1059 * thread, then mark the process cacheable only if it's being
1060 * grabbed read-only. If we're currently caching more process
1061 * handles than dph_lrulim permits, attempt to find the
1062 * least-recently-used handle that is currently unreferenced and
1063 * release it from the cache. Otherwise we are grabbing the process
1064 * for control: create a control thread for this process and store
1065 * its ID in dpr->dpr_tid.
1067 if (nomonitor || (flags & PGRAB_RDONLY)) {
1068 if (dph->dph_lrucnt >= dph->dph_lrulim) {
1069 for (opr = dt_list_prev(&dph->dph_lrulist);
1070 opr != NULL; opr = dt_list_prev(opr)) {
1071 if (opr->dpr_cacheable && opr->dpr_refs == 0) {
1072 dt_proc_destroy(dtp, opr->dpr_proc);
1078 if (flags & PGRAB_RDONLY) {
1079 dpr->dpr_cacheable = B_TRUE;
1080 dpr->dpr_rdonly = B_TRUE;
1084 } else if (dt_proc_create_thread(dtp, dpr, DT_PROC_STOP_GRAB) != 0)
1085 return (NULL); /* dt_proc_error() has been called for us */
1087 dpr->dpr_hash = dph->dph_hash[h];
1088 dph->dph_hash[h] = dpr;
1089 dt_list_prepend(&dph->dph_lrulist, dpr);
1091 dt_dprintf("grabbed pid %d\n", (int)pid);
1094 return (dpr->dpr_proc);
1098 dt_proc_release(dtrace_hdl_t *dtp, struct ps_prochandle *P)
1100 dt_proc_t *dpr = dt_proc_lookup(dtp, P, B_FALSE);
1101 dt_proc_hash_t *dph = dtp->dt_procs;
1103 assert(dpr != NULL);
1104 assert(dpr->dpr_refs != 0);
1106 if (--dpr->dpr_refs == 0 &&
1107 (!dpr->dpr_cacheable || dph->dph_lrucnt > dph->dph_lrulim))
1108 dt_proc_destroy(dtp, P);
1112 dt_proc_continue(dtrace_hdl_t *dtp, struct ps_prochandle *P)
1114 dt_proc_t *dpr = dt_proc_lookup(dtp, P, B_FALSE);
1116 (void) pthread_mutex_lock(&dpr->dpr_lock);
1118 if (dpr->dpr_stop & DT_PROC_STOP_IDLE) {
1119 dpr->dpr_stop &= ~DT_PROC_STOP_IDLE;
1120 (void) pthread_cond_broadcast(&dpr->dpr_cv);
1123 (void) pthread_mutex_unlock(&dpr->dpr_lock);
1127 dt_proc_lock(dtrace_hdl_t *dtp, struct ps_prochandle *P)
1129 dt_proc_t *dpr = dt_proc_lookup(dtp, P, B_FALSE);
1130 int err = pthread_mutex_lock(&dpr->dpr_lock);
1131 assert(err == 0); /* check for recursion */
1135 dt_proc_unlock(dtrace_hdl_t *dtp, struct ps_prochandle *P)
1137 dt_proc_t *dpr = dt_proc_lookup(dtp, P, B_FALSE);
1138 int err = pthread_mutex_unlock(&dpr->dpr_lock);
1139 assert(err == 0); /* check for unheld lock */
1143 dt_proc_hash_create(dtrace_hdl_t *dtp)
1145 if ((dtp->dt_procs = dt_zalloc(dtp, sizeof (dt_proc_hash_t) +
1146 sizeof (dt_proc_t *) * _dtrace_pidbuckets - 1)) != NULL) {
1148 (void) pthread_mutex_init(&dtp->dt_procs->dph_lock, NULL);
1149 (void) pthread_cond_init(&dtp->dt_procs->dph_cv, NULL);
1151 dtp->dt_procs->dph_hashlen = _dtrace_pidbuckets;
1152 dtp->dt_procs->dph_lrulim = _dtrace_pidlrulim;
1157 dt_proc_hash_destroy(dtrace_hdl_t *dtp)
1159 dt_proc_hash_t *dph = dtp->dt_procs;
1162 while ((dpr = dt_list_next(&dph->dph_lrulist)) != NULL)
1163 dt_proc_destroy(dtp, dpr->dpr_proc);
1165 dtp->dt_procs = NULL;
1169 struct ps_prochandle *
1170 dtrace_proc_create(dtrace_hdl_t *dtp, const char *file, char *const *argv,
1171 proc_child_func *pcf, void *child_arg)
1173 dt_ident_t *idp = dt_idhash_lookup(dtp->dt_macros, "target");
1174 struct ps_prochandle *P = dt_proc_create(dtp, file, argv, pcf, child_arg);
1176 if (P != NULL && idp != NULL && idp->di_id == 0) {
1178 idp->di_id = Pstatus(P)->pr_pid; /* $target = created pid */
1180 idp->di_id = proc_getpid(P); /* $target = created pid */
1187 struct ps_prochandle *
1188 dtrace_proc_grab(dtrace_hdl_t *dtp, pid_t pid, int flags)
1190 dt_ident_t *idp = dt_idhash_lookup(dtp->dt_macros, "target");
1191 struct ps_prochandle *P = dt_proc_grab(dtp, pid, flags, 0);
1193 if (P != NULL && idp != NULL && idp->di_id == 0)
1194 idp->di_id = pid; /* $target = grabbed pid */
1200 dtrace_proc_release(dtrace_hdl_t *dtp, struct ps_prochandle *P)
1202 dt_proc_release(dtp, P);
1206 dtrace_proc_continue(dtrace_hdl_t *dtp, struct ps_prochandle *P)
1208 dt_proc_continue(dtp, P);