kern_ktrace.c revision 225617
1139804Simp/*- 21541Srgrimes * Copyright (c) 1989, 1993 3152376Srwatson * The Regents of the University of California. 4152376Srwatson * Copyright (c) 2005 Robert N. M. Watson 5152376Srwatson * All rights reserved. 61541Srgrimes * 71541Srgrimes * Redistribution and use in source and binary forms, with or without 81541Srgrimes * modification, are permitted provided that the following conditions 91541Srgrimes * are met: 101541Srgrimes * 1. Redistributions of source code must retain the above copyright 111541Srgrimes * notice, this list of conditions and the following disclaimer. 121541Srgrimes * 2. Redistributions in binary form must reproduce the above copyright 131541Srgrimes * notice, this list of conditions and the following disclaimer in the 141541Srgrimes * documentation and/or other materials provided with the distribution. 151541Srgrimes * 4. Neither the name of the University nor the names of its contributors 161541Srgrimes * may be used to endorse or promote products derived from this software 171541Srgrimes * without specific prior written permission. 181541Srgrimes * 191541Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 201541Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 211541Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 221541Srgrimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 231541Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 241541Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 251541Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 261541Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 271541Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 281541Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 291541Srgrimes * SUCH DAMAGE. 301541Srgrimes * 311541Srgrimes * @(#)kern_ktrace.c 8.2 (Berkeley) 9/23/93 321541Srgrimes */ 331541Srgrimes 34116182Sobrien#include <sys/cdefs.h> 35116182Sobrien__FBSDID("$FreeBSD: head/sys/kern/kern_ktrace.c 225617 2011-09-16 13:58:51Z kmacy $"); 36116182Sobrien 3713203Swollman#include "opt_ktrace.h" 381541Srgrimes 391541Srgrimes#include <sys/param.h> 402112Swollman#include <sys/systm.h> 4197993Sjhb#include <sys/fcntl.h> 4297993Sjhb#include <sys/kernel.h> 4397993Sjhb#include <sys/kthread.h> 4476166Smarkm#include <sys/lock.h> 4576166Smarkm#include <sys/mutex.h> 4697993Sjhb#include <sys/malloc.h> 47155031Sjeff#include <sys/mount.h> 4897993Sjhb#include <sys/namei.h> 49164033Srwatson#include <sys/priv.h> 501541Srgrimes#include <sys/proc.h> 5197993Sjhb#include <sys/unistd.h> 521541Srgrimes#include <sys/vnode.h> 53176471Sdes#include <sys/socket.h> 54176471Sdes#include <sys/stat.h> 551541Srgrimes#include <sys/ktrace.h> 5674927Sjhb#include <sys/sx.h> 5797993Sjhb#include <sys/sysctl.h> 58219042Sdchagin#include <sys/sysent.h> 591541Srgrimes#include <sys/syslog.h> 6097993Sjhb#include <sys/sysproto.h> 611541Srgrimes 62163606Srwatson#include <security/mac/mac_framework.h> 63163606Srwatson 64152376Srwatson/* 65152376Srwatson * The ktrace facility allows the tracing of certain key events in user space 66152376Srwatson * processes, such as system calls, signal delivery, context switches, and 67152376Srwatson * user generated events using utrace(2). It works by streaming event 68152376Srwatson * records and data to a vnode associated with the process using the 69152376Srwatson * ktrace(2) system call. In general, records can be written directly from 70152376Srwatson * the context that generates the event. One important exception to this is 71152376Srwatson * during a context switch, where sleeping is not permitted. To handle this 72152376Srwatson * case, trace events are generated using in-kernel ktr_request records, and 73152376Srwatson * then delivered to disk at a convenient moment -- either immediately, the 74152376Srwatson * next traceable event, at system call return, or at process exit. 75152376Srwatson * 76152376Srwatson * When dealing with multiple threads or processes writing to the same event 77152376Srwatson * log, ordering guarantees are weak: specifically, if an event has multiple 78152376Srwatson * records (i.e., system call enter and return), they may be interlaced with 79152376Srwatson * records from another event. Process and thread ID information is provided 80152376Srwatson * in the record, and user applications can de-interlace events if required. 81152376Srwatson */ 82152376Srwatson 8330354Sphkstatic MALLOC_DEFINE(M_KTRACE, "KTRACE", "KTRACE"); 8430309Sphk 8513203Swollman#ifdef KTRACE 8612577Sbde 87219028SnetchildFEATURE(ktrace, "Kernel support for system-call tracing"); 88219028Snetchild 8997993Sjhb#ifndef KTRACE_REQUEST_POOL 9097993Sjhb#define KTRACE_REQUEST_POOL 100 9197993Sjhb#endif 9212819Sphk 9397993Sjhbstruct ktr_request { 9497993Sjhb struct ktr_header ktr_header; 95151927Srwatson void *ktr_buffer; 9697993Sjhb union { 97219042Sdchagin struct ktr_proc_ctor ktr_proc_ctor; 9897993Sjhb struct ktr_syscall ktr_syscall; 9997993Sjhb struct ktr_sysret ktr_sysret; 10097993Sjhb struct ktr_genio ktr_genio; 10197993Sjhb struct ktr_psig ktr_psig; 10297993Sjhb struct ktr_csw ktr_csw; 10397993Sjhb } ktr_data; 10497993Sjhb STAILQ_ENTRY(ktr_request) ktr_list; 10597993Sjhb}; 10697993Sjhb 10797993Sjhbstatic int data_lengths[] = { 10897993Sjhb 0, /* none */ 10997993Sjhb offsetof(struct ktr_syscall, ktr_args), /* KTR_SYSCALL */ 11097993Sjhb sizeof(struct ktr_sysret), /* KTR_SYSRET */ 11197993Sjhb 0, /* KTR_NAMEI */ 11297993Sjhb sizeof(struct ktr_genio), /* KTR_GENIO */ 11397993Sjhb sizeof(struct ktr_psig), /* KTR_PSIG */ 114219312Sdchagin sizeof(struct ktr_csw), /* KTR_CSW */ 115176471Sdes 0, /* KTR_USER */ 116176471Sdes 0, /* KTR_STRUCT */ 117189707Sjhb 0, /* KTR_SYSCTL */ 118219042Sdchagin sizeof(struct ktr_proc_ctor), /* KTR_PROCCTOR */ 119219042Sdchagin 0, /* KTR_PROCDTOR */ 12097993Sjhb}; 12197993Sjhb 12297993Sjhbstatic STAILQ_HEAD(, ktr_request) ktr_free; 12397993Sjhb 124141633Sphkstatic SYSCTL_NODE(_kern, OID_AUTO, ktrace, CTLFLAG_RD, 0, "KTRACE options"); 125103234Sjhb 126118607Sjhbstatic u_int ktr_requestpool = KTRACE_REQUEST_POOL; 127103234SjhbTUNABLE_INT("kern.ktrace.request_pool", &ktr_requestpool); 12897993Sjhb 129118607Sjhbstatic u_int ktr_geniosize = PAGE_SIZE; 130103234SjhbTUNABLE_INT("kern.ktrace.genio_size", &ktr_geniosize); 131103234SjhbSYSCTL_UINT(_kern_ktrace, OID_AUTO, genio_size, CTLFLAG_RW, &ktr_geniosize, 132103234Sjhb 0, "Maximum size of genio event payload"); 133103234Sjhb 13497993Sjhbstatic int print_message = 1; 135214158Sjhbstatic struct mtx ktrace_mtx; 136152376Srwatsonstatic struct sx ktrace_sx; 13797993Sjhb 13897993Sjhbstatic void ktrace_init(void *dummy); 13997993Sjhbstatic int sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS); 140219041Sdchaginstatic u_int ktrace_resize_pool(u_int oldsize, u_int newsize); 141219311Sdchaginstatic struct ktr_request *ktr_getrequest_entered(struct thread *td, int type); 14297993Sjhbstatic struct ktr_request *ktr_getrequest(int type); 143152376Srwatsonstatic void ktr_submitrequest(struct thread *td, struct ktr_request *req); 144214158Sjhbstatic void ktr_freeproc(struct proc *p, struct ucred **uc, 145214158Sjhb struct vnode **vp); 14697993Sjhbstatic void ktr_freerequest(struct ktr_request *req); 147214158Sjhbstatic void ktr_freerequest_locked(struct ktr_request *req); 148152376Srwatsonstatic void ktr_writerequest(struct thread *td, struct ktr_request *req); 14997993Sjhbstatic int ktrcanset(struct thread *,struct proc *); 15097993Sjhbstatic int ktrsetchildren(struct thread *,struct proc *,int,int,struct vnode *); 15197993Sjhbstatic int ktrops(struct thread *,struct proc *,int,int,struct vnode *); 152219311Sdchaginstatic void ktrprocctor_entered(struct thread *, struct proc *); 15397993Sjhb 154152376Srwatson/* 155152376Srwatson * ktrace itself generates events, such as context switches, which we do not 156152376Srwatson * wish to trace. Maintain a flag, TDP_INKTRACE, on each thread to determine 157152376Srwatson * whether or not it is in a region where tracing of events should be 158152376Srwatson * suppressed. 159152376Srwatson */ 16097993Sjhbstatic void 161152376Srwatsonktrace_enter(struct thread *td) 162152376Srwatson{ 163152376Srwatson 164152376Srwatson KASSERT(!(td->td_pflags & TDP_INKTRACE), ("ktrace_enter: flag set")); 165152376Srwatson td->td_pflags |= TDP_INKTRACE; 166152376Srwatson} 167152376Srwatson 168152376Srwatsonstatic void 169152376Srwatsonktrace_exit(struct thread *td) 170152376Srwatson{ 171152376Srwatson 172152376Srwatson KASSERT(td->td_pflags & TDP_INKTRACE, ("ktrace_exit: flag not set")); 173152376Srwatson td->td_pflags &= ~TDP_INKTRACE; 174152376Srwatson} 175152376Srwatson 176152376Srwatsonstatic void 177152376Srwatsonktrace_assert(struct thread *td) 178152376Srwatson{ 179152376Srwatson 180152376Srwatson KASSERT(td->td_pflags & TDP_INKTRACE, ("ktrace_assert: flag not set")); 181152376Srwatson} 182152376Srwatson 183152376Srwatsonstatic void 18497993Sjhbktrace_init(void *dummy) 1851541Srgrimes{ 18697993Sjhb struct ktr_request *req; 18797993Sjhb int i; 1881541Srgrimes 18997993Sjhb mtx_init(&ktrace_mtx, "ktrace", NULL, MTX_DEF | MTX_QUIET); 190152376Srwatson sx_init(&ktrace_sx, "ktrace_sx"); 19197993Sjhb STAILQ_INIT(&ktr_free); 19297993Sjhb for (i = 0; i < ktr_requestpool; i++) { 193111119Simp req = malloc(sizeof(struct ktr_request), M_KTRACE, M_WAITOK); 19497993Sjhb STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list); 19597993Sjhb } 1961541Srgrimes} 19797993SjhbSYSINIT(ktrace_init, SI_SUB_KTRACE, SI_ORDER_ANY, ktrace_init, NULL); 1981541Srgrimes 19997993Sjhbstatic int 20097993Sjhbsysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS) 20197993Sjhb{ 20297993Sjhb struct thread *td; 203118607Sjhb u_int newsize, oldsize, wantsize; 20497993Sjhb int error; 20597993Sjhb 20697993Sjhb /* Handle easy read-only case first to avoid warnings from GCC. */ 20797993Sjhb if (!req->newptr) { 20897993Sjhb oldsize = ktr_requestpool; 209118607Sjhb return (SYSCTL_OUT(req, &oldsize, sizeof(u_int))); 21097993Sjhb } 21197993Sjhb 212118607Sjhb error = SYSCTL_IN(req, &wantsize, sizeof(u_int)); 21397993Sjhb if (error) 21497993Sjhb return (error); 21597993Sjhb td = curthread; 216152376Srwatson ktrace_enter(td); 21797993Sjhb oldsize = ktr_requestpool; 218219041Sdchagin newsize = ktrace_resize_pool(oldsize, wantsize); 219152376Srwatson ktrace_exit(td); 220118607Sjhb error = SYSCTL_OUT(req, &oldsize, sizeof(u_int)); 22197993Sjhb if (error) 22297993Sjhb return (error); 223122478Sjkoshy if (wantsize > oldsize && newsize < wantsize) 22497993Sjhb return (ENOSPC); 22597993Sjhb return (0); 22697993Sjhb} 227103234SjhbSYSCTL_PROC(_kern_ktrace, OID_AUTO, request_pool, CTLTYPE_UINT|CTLFLAG_RW, 228211102Sgavin &ktr_requestpool, 0, sysctl_kern_ktrace_request_pool, "IU", 229211102Sgavin "Pool buffer size for ktrace(1)"); 23097993Sjhb 231118607Sjhbstatic u_int 232219041Sdchaginktrace_resize_pool(u_int oldsize, u_int newsize) 23397993Sjhb{ 234219041Sdchagin STAILQ_HEAD(, ktr_request) ktr_new; 23597993Sjhb struct ktr_request *req; 236122478Sjkoshy int bound; 23797993Sjhb 23897993Sjhb print_message = 1; 239219041Sdchagin bound = newsize - oldsize; 240122478Sjkoshy if (bound == 0) 241122478Sjkoshy return (ktr_requestpool); 242219041Sdchagin if (bound < 0) { 243219041Sdchagin mtx_lock(&ktrace_mtx); 24497993Sjhb /* Shrink pool down to newsize if possible. */ 245122478Sjkoshy while (bound++ < 0) { 24697993Sjhb req = STAILQ_FIRST(&ktr_free); 24797993Sjhb if (req == NULL) 248219041Sdchagin break; 24997993Sjhb STAILQ_REMOVE_HEAD(&ktr_free, ktr_list); 25097993Sjhb ktr_requestpool--; 25197993Sjhb free(req, M_KTRACE); 25297993Sjhb } 253219041Sdchagin } else { 25497993Sjhb /* Grow pool up to newsize. */ 255219041Sdchagin STAILQ_INIT(&ktr_new); 256122478Sjkoshy while (bound-- > 0) { 25797993Sjhb req = malloc(sizeof(struct ktr_request), M_KTRACE, 258111119Simp M_WAITOK); 259219041Sdchagin STAILQ_INSERT_HEAD(&ktr_new, req, ktr_list); 26097993Sjhb } 261219041Sdchagin mtx_lock(&ktrace_mtx); 262219041Sdchagin STAILQ_CONCAT(&ktr_free, &ktr_new); 263219041Sdchagin ktr_requestpool += (newsize - oldsize); 264219041Sdchagin } 265219041Sdchagin mtx_unlock(&ktrace_mtx); 26697993Sjhb return (ktr_requestpool); 26797993Sjhb} 26897993Sjhb 269198411Sjhb/* ktr_getrequest() assumes that ktr_comm[] is the same size as td_name[]. */ 270198411SjhbCTASSERT(sizeof(((struct ktr_header *)NULL)->ktr_comm) == 271198411Sjhb (sizeof((struct thread *)NULL)->td_name)); 272198411Sjhb 27397993Sjhbstatic struct ktr_request * 274219311Sdchaginktr_getrequest_entered(struct thread *td, int type) 27597993Sjhb{ 27697993Sjhb struct ktr_request *req; 27797993Sjhb struct proc *p = td->td_proc; 27897993Sjhb int pm; 27997993Sjhb 280152430Srwatson mtx_lock(&ktrace_mtx); 28197993Sjhb if (!KTRCHECK(td, type)) { 282152430Srwatson mtx_unlock(&ktrace_mtx); 28397993Sjhb return (NULL); 28497993Sjhb } 28597993Sjhb req = STAILQ_FIRST(&ktr_free); 28697993Sjhb if (req != NULL) { 28797993Sjhb STAILQ_REMOVE_HEAD(&ktr_free, ktr_list); 28897993Sjhb req->ktr_header.ktr_type = type; 289112199Sjhb if (p->p_traceflag & KTRFAC_DROP) { 290112199Sjhb req->ktr_header.ktr_type |= KTR_DROP; 291112199Sjhb p->p_traceflag &= ~KTRFAC_DROP; 292112199Sjhb } 293152430Srwatson mtx_unlock(&ktrace_mtx); 29497993Sjhb microtime(&req->ktr_header.ktr_time); 29597993Sjhb req->ktr_header.ktr_pid = p->p_pid; 296151929Srwatson req->ktr_header.ktr_tid = td->td_tid; 297198411Sjhb bcopy(td->td_name, req->ktr_header.ktr_comm, 298198411Sjhb sizeof(req->ktr_header.ktr_comm)); 299151927Srwatson req->ktr_buffer = NULL; 30097993Sjhb req->ktr_header.ktr_len = 0; 30197993Sjhb } else { 302112199Sjhb p->p_traceflag |= KTRFAC_DROP; 30397993Sjhb pm = print_message; 30497993Sjhb print_message = 0; 30597993Sjhb mtx_unlock(&ktrace_mtx); 30697993Sjhb if (pm) 30797993Sjhb printf("Out of ktrace request objects.\n"); 30897993Sjhb } 30997993Sjhb return (req); 31097993Sjhb} 31197993Sjhb 312219042Sdchaginstatic struct ktr_request * 313219042Sdchaginktr_getrequest(int type) 314219042Sdchagin{ 315219042Sdchagin struct thread *td = curthread; 316219042Sdchagin struct ktr_request *req; 317219042Sdchagin 318219042Sdchagin ktrace_enter(td); 319219311Sdchagin req = ktr_getrequest_entered(td, type); 320219042Sdchagin if (req == NULL) 321219042Sdchagin ktrace_exit(td); 322219042Sdchagin 323219042Sdchagin return (req); 324219042Sdchagin} 325219042Sdchagin 326152376Srwatson/* 327152376Srwatson * Some trace generation environments don't permit direct access to VFS, 328152376Srwatson * such as during a context switch where sleeping is not allowed. Under these 329152376Srwatson * circumstances, queue a request to the thread to be written asynchronously 330152376Srwatson * later. 331152376Srwatson */ 33297993Sjhbstatic void 333152376Srwatsonktr_enqueuerequest(struct thread *td, struct ktr_request *req) 33497993Sjhb{ 33597993Sjhb 33697993Sjhb mtx_lock(&ktrace_mtx); 337152376Srwatson STAILQ_INSERT_TAIL(&td->td_proc->p_ktr, req, ktr_list); 338118599Sjhb mtx_unlock(&ktrace_mtx); 33997993Sjhb} 34097993Sjhb 341152376Srwatson/* 342152376Srwatson * Drain any pending ktrace records from the per-thread queue to disk. This 343152376Srwatson * is used both internally before committing other records, and also on 344152376Srwatson * system call return. We drain all the ones we can find at the time when 345152376Srwatson * drain is requested, but don't keep draining after that as those events 346189707Sjhb * may be approximately "after" the current event. 347152376Srwatson */ 34897993Sjhbstatic void 349152376Srwatsonktr_drain(struct thread *td) 350152376Srwatson{ 351152376Srwatson struct ktr_request *queued_req; 352152376Srwatson STAILQ_HEAD(, ktr_request) local_queue; 353152376Srwatson 354152376Srwatson ktrace_assert(td); 355152376Srwatson sx_assert(&ktrace_sx, SX_XLOCKED); 356152376Srwatson 357211512Sjhb STAILQ_INIT(&local_queue); 358152376Srwatson 359152376Srwatson if (!STAILQ_EMPTY(&td->td_proc->p_ktr)) { 360152376Srwatson mtx_lock(&ktrace_mtx); 361152376Srwatson STAILQ_CONCAT(&local_queue, &td->td_proc->p_ktr); 362152376Srwatson mtx_unlock(&ktrace_mtx); 363152376Srwatson 364152376Srwatson while ((queued_req = STAILQ_FIRST(&local_queue))) { 365152376Srwatson STAILQ_REMOVE_HEAD(&local_queue, ktr_list); 366152376Srwatson ktr_writerequest(td, queued_req); 367152376Srwatson ktr_freerequest(queued_req); 368152376Srwatson } 369152376Srwatson } 370152376Srwatson} 371152376Srwatson 372152376Srwatson/* 373152376Srwatson * Submit a trace record for immediate commit to disk -- to be used only 374152376Srwatson * where entering VFS is OK. First drain any pending records that may have 375152376Srwatson * been cached in the thread. 376152376Srwatson */ 377152376Srwatsonstatic void 378219311Sdchaginktr_submitrequest(struct thread *td, struct ktr_request *req) 379152376Srwatson{ 380152376Srwatson 381152376Srwatson ktrace_assert(td); 382152376Srwatson 383152376Srwatson sx_xlock(&ktrace_sx); 384152376Srwatson ktr_drain(td); 385152376Srwatson ktr_writerequest(td, req); 386152376Srwatson ktr_freerequest(req); 387152376Srwatson sx_xunlock(&ktrace_sx); 388152376Srwatson ktrace_exit(td); 389152376Srwatson} 390152376Srwatson 391152376Srwatsonstatic void 39297993Sjhbktr_freerequest(struct ktr_request *req) 39397993Sjhb{ 39497993Sjhb 395214158Sjhb mtx_lock(&ktrace_mtx); 396214158Sjhb ktr_freerequest_locked(req); 397214158Sjhb mtx_unlock(&ktrace_mtx); 398214158Sjhb} 399214158Sjhb 400214158Sjhbstatic void 401214158Sjhbktr_freerequest_locked(struct ktr_request *req) 402214158Sjhb{ 403214158Sjhb 404214158Sjhb mtx_assert(&ktrace_mtx, MA_OWNED); 405151927Srwatson if (req->ktr_buffer != NULL) 406151927Srwatson free(req->ktr_buffer, M_KTRACE); 40797993Sjhb STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list); 40897993Sjhb} 40997993Sjhb 410214158Sjhb/* 411214158Sjhb * Disable tracing for a process and release all associated resources. 412214158Sjhb * The caller is responsible for releasing a reference on the returned 413214158Sjhb * vnode and credentials. 414214158Sjhb */ 415214158Sjhbstatic void 416214158Sjhbktr_freeproc(struct proc *p, struct ucred **uc, struct vnode **vp) 417214158Sjhb{ 418214158Sjhb struct ktr_request *req; 419214158Sjhb 420214158Sjhb PROC_LOCK_ASSERT(p, MA_OWNED); 421214158Sjhb mtx_assert(&ktrace_mtx, MA_OWNED); 422214158Sjhb *uc = p->p_tracecred; 423214158Sjhb p->p_tracecred = NULL; 424214158Sjhb if (vp != NULL) 425214158Sjhb *vp = p->p_tracevp; 426214158Sjhb p->p_tracevp = NULL; 427214158Sjhb p->p_traceflag = 0; 428214158Sjhb while ((req = STAILQ_FIRST(&p->p_ktr)) != NULL) { 429214158Sjhb STAILQ_REMOVE_HEAD(&p->p_ktr, ktr_list); 430214158Sjhb ktr_freerequest_locked(req); 431214158Sjhb } 432214158Sjhb} 433214158Sjhb 4341549Srgrimesvoid 43597993Sjhbktrsyscall(code, narg, args) 43647955Sdt int code, narg; 43747955Sdt register_t args[]; 4381541Srgrimes{ 43997993Sjhb struct ktr_request *req; 44097993Sjhb struct ktr_syscall *ktp; 44197993Sjhb size_t buflen; 442103233Sjhb char *buf = NULL; 4431541Srgrimes 444103233Sjhb buflen = sizeof(register_t) * narg; 445103233Sjhb if (buflen > 0) { 446111119Simp buf = malloc(buflen, M_KTRACE, M_WAITOK); 447103233Sjhb bcopy(args, buf, buflen); 448103233Sjhb } 44997993Sjhb req = ktr_getrequest(KTR_SYSCALL); 450104230Sphk if (req == NULL) { 451104230Sphk if (buf != NULL) 452104230Sphk free(buf, M_KTRACE); 45397993Sjhb return; 454104230Sphk } 45597993Sjhb ktp = &req->ktr_data.ktr_syscall; 4561541Srgrimes ktp->ktr_code = code; 4571541Srgrimes ktp->ktr_narg = narg; 45897993Sjhb if (buflen > 0) { 45997993Sjhb req->ktr_header.ktr_len = buflen; 460151927Srwatson req->ktr_buffer = buf; 46197993Sjhb } 462152376Srwatson ktr_submitrequest(curthread, req); 4631541Srgrimes} 4641541Srgrimes 4651549Srgrimesvoid 46697993Sjhbktrsysret(code, error, retval) 46747955Sdt int code, error; 46847955Sdt register_t retval; 4691541Srgrimes{ 47097993Sjhb struct ktr_request *req; 47197993Sjhb struct ktr_sysret *ktp; 4721541Srgrimes 47397993Sjhb req = ktr_getrequest(KTR_SYSRET); 47497993Sjhb if (req == NULL) 47597993Sjhb return; 47697993Sjhb ktp = &req->ktr_data.ktr_sysret; 47797993Sjhb ktp->ktr_code = code; 47897993Sjhb ktp->ktr_error = error; 47997993Sjhb ktp->ktr_retval = retval; /* what about val2 ? */ 480152376Srwatson ktr_submitrequest(curthread, req); 4811541Srgrimes} 4821541Srgrimes 483152376Srwatson/* 484214158Sjhb * When a setuid process execs, disable tracing. 485214158Sjhb * 486214158Sjhb * XXX: We toss any pending asynchronous records. 487152376Srwatson */ 4881549Srgrimesvoid 489214158Sjhbktrprocexec(struct proc *p, struct ucred **uc, struct vnode **vp) 490214158Sjhb{ 491214158Sjhb 492214158Sjhb PROC_LOCK_ASSERT(p, MA_OWNED); 493214158Sjhb mtx_lock(&ktrace_mtx); 494214158Sjhb ktr_freeproc(p, uc, vp); 495214158Sjhb mtx_unlock(&ktrace_mtx); 496214158Sjhb} 497214158Sjhb 498214158Sjhb/* 499214158Sjhb * When a process exits, drain per-process asynchronous trace records 500214158Sjhb * and disable tracing. 501214158Sjhb */ 502214158Sjhbvoid 503152376Srwatsonktrprocexit(struct thread *td) 504152376Srwatson{ 505219042Sdchagin struct ktr_request *req; 506214158Sjhb struct proc *p; 507214158Sjhb struct ucred *cred; 508214158Sjhb struct vnode *vp; 509214158Sjhb int vfslocked; 510152376Srwatson 511214158Sjhb p = td->td_proc; 512214158Sjhb if (p->p_traceflag == 0) 513214158Sjhb return; 514214158Sjhb 515152376Srwatson ktrace_enter(td); 516219311Sdchagin req = ktr_getrequest_entered(td, KTR_PROCDTOR); 517219311Sdchagin if (req != NULL) 518219311Sdchagin ktr_enqueuerequest(td, req); 519152376Srwatson sx_xlock(&ktrace_sx); 520152376Srwatson ktr_drain(td); 521152376Srwatson sx_xunlock(&ktrace_sx); 522214158Sjhb PROC_LOCK(p); 523214158Sjhb mtx_lock(&ktrace_mtx); 524214158Sjhb ktr_freeproc(p, &cred, &vp); 525214158Sjhb mtx_unlock(&ktrace_mtx); 526214158Sjhb PROC_UNLOCK(p); 527214158Sjhb if (vp != NULL) { 528214158Sjhb vfslocked = VFS_LOCK_GIANT(vp->v_mount); 529214158Sjhb vrele(vp); 530214158Sjhb VFS_UNLOCK_GIANT(vfslocked); 531214158Sjhb } 532214158Sjhb if (cred != NULL) 533214158Sjhb crfree(cred); 534152376Srwatson ktrace_exit(td); 535152376Srwatson} 536152376Srwatson 537219042Sdchaginstatic void 538219311Sdchaginktrprocctor_entered(struct thread *td, struct proc *p) 539219042Sdchagin{ 540219042Sdchagin struct ktr_proc_ctor *ktp; 541219042Sdchagin struct ktr_request *req; 542219312Sdchagin struct thread *td2; 543219042Sdchagin 544219042Sdchagin ktrace_assert(td); 545219042Sdchagin td2 = FIRST_THREAD_IN_PROC(p); 546219311Sdchagin req = ktr_getrequest_entered(td2, KTR_PROCCTOR); 547219042Sdchagin if (req == NULL) 548219042Sdchagin return; 549219042Sdchagin ktp = &req->ktr_data.ktr_proc_ctor; 550219042Sdchagin ktp->sv_flags = p->p_sysent->sv_flags; 551219311Sdchagin ktr_enqueuerequest(td2, req); 552219042Sdchagin} 553219042Sdchagin 554219042Sdchaginvoid 555219042Sdchaginktrprocctor(struct proc *p) 556219042Sdchagin{ 557219042Sdchagin struct thread *td = curthread; 558219042Sdchagin 559219042Sdchagin if ((p->p_traceflag & KTRFAC_MASK) == 0) 560219042Sdchagin return; 561219042Sdchagin 562219042Sdchagin ktrace_enter(td); 563219311Sdchagin ktrprocctor_entered(td, p); 564219042Sdchagin ktrace_exit(td); 565219042Sdchagin} 566219042Sdchagin 567152376Srwatson/* 568214158Sjhb * When a process forks, enable tracing in the new process if needed. 569214158Sjhb */ 570214158Sjhbvoid 571214158Sjhbktrprocfork(struct proc *p1, struct proc *p2) 572214158Sjhb{ 573214158Sjhb 574219042Sdchagin PROC_LOCK(p1); 575214158Sjhb mtx_lock(&ktrace_mtx); 576214158Sjhb KASSERT(p2->p_tracevp == NULL, ("new process has a ktrace vnode")); 577214158Sjhb if (p1->p_traceflag & KTRFAC_INHERIT) { 578214158Sjhb p2->p_traceflag = p1->p_traceflag; 579214158Sjhb if ((p2->p_tracevp = p1->p_tracevp) != NULL) { 580214158Sjhb VREF(p2->p_tracevp); 581214158Sjhb KASSERT(p1->p_tracecred != NULL, 582214158Sjhb ("ktrace vnode with no cred")); 583214158Sjhb p2->p_tracecred = crhold(p1->p_tracecred); 584214158Sjhb } 585214158Sjhb } 586214158Sjhb mtx_unlock(&ktrace_mtx); 587219042Sdchagin PROC_UNLOCK(p1); 588219042Sdchagin 589219042Sdchagin ktrprocctor(p2); 590214158Sjhb} 591214158Sjhb 592214158Sjhb/* 593152376Srwatson * When a thread returns, drain any asynchronous records generated by the 594152376Srwatson * system call. 595152376Srwatson */ 596152376Srwatsonvoid 597152376Srwatsonktruserret(struct thread *td) 598152376Srwatson{ 599152376Srwatson 600152376Srwatson ktrace_enter(td); 601152376Srwatson sx_xlock(&ktrace_sx); 602152376Srwatson ktr_drain(td); 603152376Srwatson sx_xunlock(&ktrace_sx); 604152376Srwatson ktrace_exit(td); 605152376Srwatson} 606152376Srwatson 607152376Srwatsonvoid 60897993Sjhbktrnamei(path) 6091541Srgrimes char *path; 6101541Srgrimes{ 61197993Sjhb struct ktr_request *req; 61297993Sjhb int namelen; 613103233Sjhb char *buf = NULL; 6141541Srgrimes 615103233Sjhb namelen = strlen(path); 616103233Sjhb if (namelen > 0) { 617111119Simp buf = malloc(namelen, M_KTRACE, M_WAITOK); 618103233Sjhb bcopy(path, buf, namelen); 619103233Sjhb } 62097993Sjhb req = ktr_getrequest(KTR_NAMEI); 621104230Sphk if (req == NULL) { 622104230Sphk if (buf != NULL) 623104230Sphk free(buf, M_KTRACE); 62497993Sjhb return; 625104230Sphk } 62697993Sjhb if (namelen > 0) { 62797993Sjhb req->ktr_header.ktr_len = namelen; 628151927Srwatson req->ktr_buffer = buf; 62997993Sjhb } 630152376Srwatson ktr_submitrequest(curthread, req); 6311541Srgrimes} 6321541Srgrimes 6331549Srgrimesvoid 634189707Sjhbktrsysctl(name, namelen) 635189707Sjhb int *name; 636189707Sjhb u_int namelen; 637189707Sjhb{ 638189707Sjhb struct ktr_request *req; 639189707Sjhb u_int mib[CTL_MAXNAME + 2]; 640189707Sjhb char *mibname; 641189707Sjhb size_t mibnamelen; 642189707Sjhb int error; 643189707Sjhb 644189707Sjhb /* Lookup name of mib. */ 645189707Sjhb KASSERT(namelen <= CTL_MAXNAME, ("sysctl MIB too long")); 646189707Sjhb mib[0] = 0; 647189707Sjhb mib[1] = 1; 648189707Sjhb bcopy(name, mib + 2, namelen * sizeof(*name)); 649189707Sjhb mibnamelen = 128; 650189707Sjhb mibname = malloc(mibnamelen, M_KTRACE, M_WAITOK); 651189707Sjhb error = kernel_sysctl(curthread, mib, namelen + 2, mibname, &mibnamelen, 652189707Sjhb NULL, 0, &mibnamelen, 0); 653189707Sjhb if (error) { 654189707Sjhb free(mibname, M_KTRACE); 655189707Sjhb return; 656189707Sjhb } 657189707Sjhb req = ktr_getrequest(KTR_SYSCTL); 658189707Sjhb if (req == NULL) { 659189707Sjhb free(mibname, M_KTRACE); 660189707Sjhb return; 661189707Sjhb } 662189707Sjhb req->ktr_header.ktr_len = mibnamelen; 663189707Sjhb req->ktr_buffer = mibname; 664189707Sjhb ktr_submitrequest(curthread, req); 665189707Sjhb} 666189707Sjhb 667189707Sjhbvoid 66897993Sjhbktrgenio(fd, rw, uio, error) 6691541Srgrimes int fd; 6701541Srgrimes enum uio_rw rw; 67162378Sgreen struct uio *uio; 67262378Sgreen int error; 6731541Srgrimes{ 67497993Sjhb struct ktr_request *req; 67597993Sjhb struct ktr_genio *ktg; 676103235Sjhb int datalen; 677103235Sjhb char *buf; 6788876Srgrimes 679131897Sphk if (error) { 680131897Sphk free(uio, M_IOV); 6811541Srgrimes return; 682131897Sphk } 683103235Sjhb uio->uio_offset = 0; 684103235Sjhb uio->uio_rw = UIO_WRITE; 685103235Sjhb datalen = imin(uio->uio_resid, ktr_geniosize); 686111119Simp buf = malloc(datalen, M_KTRACE, M_WAITOK); 687131897Sphk error = uiomove(buf, datalen, uio); 688131897Sphk free(uio, M_IOV); 689131897Sphk if (error) { 690103235Sjhb free(buf, M_KTRACE); 691103235Sjhb return; 692103235Sjhb } 69397993Sjhb req = ktr_getrequest(KTR_GENIO); 694103235Sjhb if (req == NULL) { 695103235Sjhb free(buf, M_KTRACE); 69697993Sjhb return; 697103235Sjhb } 69897993Sjhb ktg = &req->ktr_data.ktr_genio; 69997993Sjhb ktg->ktr_fd = fd; 70097993Sjhb ktg->ktr_rw = rw; 701103235Sjhb req->ktr_header.ktr_len = datalen; 702151927Srwatson req->ktr_buffer = buf; 703152376Srwatson ktr_submitrequest(curthread, req); 7041541Srgrimes} 7051541Srgrimes 7061549Srgrimesvoid 70797993Sjhbktrpsig(sig, action, mask, code) 70851941Smarcel int sig; 7091541Srgrimes sig_t action; 71051791Smarcel sigset_t *mask; 71151941Smarcel int code; 7121541Srgrimes{ 713219311Sdchagin struct thread *td = curthread; 71497993Sjhb struct ktr_request *req; 71597993Sjhb struct ktr_psig *kp; 7161541Srgrimes 71797993Sjhb req = ktr_getrequest(KTR_PSIG); 71897993Sjhb if (req == NULL) 71997993Sjhb return; 72097993Sjhb kp = &req->ktr_data.ktr_psig; 72197993Sjhb kp->signo = (char)sig; 72297993Sjhb kp->action = action; 72397993Sjhb kp->mask = *mask; 72497993Sjhb kp->code = code; 725219311Sdchagin ktr_enqueuerequest(td, req); 726219311Sdchagin ktrace_exit(td); 7271541Srgrimes} 7281541Srgrimes 7291549Srgrimesvoid 73097993Sjhbktrcsw(out, user) 7311541Srgrimes int out, user; 7321541Srgrimes{ 733219311Sdchagin struct thread *td = curthread; 73497993Sjhb struct ktr_request *req; 73597993Sjhb struct ktr_csw *kc; 7361541Srgrimes 73797993Sjhb req = ktr_getrequest(KTR_CSW); 73897993Sjhb if (req == NULL) 73997993Sjhb return; 74097993Sjhb kc = &req->ktr_data.ktr_csw; 74197993Sjhb kc->out = out; 74297993Sjhb kc->user = user; 743219311Sdchagin ktr_enqueuerequest(td, req); 744219311Sdchagin ktrace_exit(td); 7451541Srgrimes} 746176471Sdes 747176471Sdesvoid 748210064Sjhbktrstruct(name, data, datalen) 749176471Sdes const char *name; 750176471Sdes void *data; 751176471Sdes size_t datalen; 752176471Sdes{ 753176471Sdes struct ktr_request *req; 754176471Sdes char *buf = NULL; 755176471Sdes size_t buflen; 756176471Sdes 757176471Sdes if (!data) 758176471Sdes datalen = 0; 759210064Sjhb buflen = strlen(name) + 1 + datalen; 760176471Sdes buf = malloc(buflen, M_KTRACE, M_WAITOK); 761210064Sjhb strcpy(buf, name); 762210064Sjhb bcopy(data, buf + strlen(name) + 1, datalen); 763176471Sdes if ((req = ktr_getrequest(KTR_STRUCT)) == NULL) { 764176471Sdes free(buf, M_KTRACE); 765176471Sdes return; 766176471Sdes } 767176471Sdes req->ktr_buffer = buf; 768176471Sdes req->ktr_header.ktr_len = buflen; 769176471Sdes ktr_submitrequest(curthread, req); 770176471Sdes} 771114026Sjhb#endif /* KTRACE */ 7721541Srgrimes 7731541Srgrimes/* Interface and common routines */ 7741541Srgrimes 77512221Sbde#ifndef _SYS_SYSPROTO_H_ 7761541Srgrimesstruct ktrace_args { 7771541Srgrimes char *fname; 7781541Srgrimes int ops; 7791541Srgrimes int facs; 7801541Srgrimes int pid; 7811541Srgrimes}; 78212221Sbde#endif 7831541Srgrimes/* ARGSUSED */ 7841549Srgrimesint 785225617Skmacysys_ktrace(td, uap) 78683366Sjulian struct thread *td; 7871541Srgrimes register struct ktrace_args *uap; 7881541Srgrimes{ 78913203Swollman#ifdef KTRACE 7901541Srgrimes register struct vnode *vp = NULL; 7911541Srgrimes register struct proc *p; 7921541Srgrimes struct pgrp *pg; 7931541Srgrimes int facs = uap->facs & ~KTRFAC_ROOT; 7941541Srgrimes int ops = KTROP(uap->ops); 7951541Srgrimes int descend = uap->ops & KTRFLAG_DESCEND; 796147576Spjd int nfound, ret = 0; 797157233Sjhb int flags, error = 0, vfslocked; 7981541Srgrimes struct nameidata nd; 799112198Sjhb struct ucred *cred; 8001541Srgrimes 801114026Sjhb /* 802114026Sjhb * Need something to (un)trace. 803114026Sjhb */ 804114026Sjhb if (ops != KTROP_CLEARFILE && facs == 0) 805114026Sjhb return (EINVAL); 806114026Sjhb 807152376Srwatson ktrace_enter(td); 8081541Srgrimes if (ops != KTROP_CLEAR) { 8091541Srgrimes /* 8101541Srgrimes * an operation which requires a file argument. 8111541Srgrimes */ 812157233Sjhb NDINIT(&nd, LOOKUP, NOFOLLOW | MPSAFE, UIO_USERSPACE, 813157233Sjhb uap->fname, td); 81462550Smckusick flags = FREAD | FWRITE | O_NOFOLLOW; 815170152Skib error = vn_open(&nd, &flags, 0, NULL); 8163308Sphk if (error) { 817152376Srwatson ktrace_exit(td); 8181541Srgrimes return (error); 8191541Srgrimes } 820157233Sjhb vfslocked = NDHASGIANT(&nd); 82154655Seivind NDFREE(&nd, NDF_ONLY_PNBUF); 8221541Srgrimes vp = nd.ni_vp; 823175294Sattilio VOP_UNLOCK(vp, 0); 8241541Srgrimes if (vp->v_type != VREG) { 82591406Sjhb (void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td); 826157233Sjhb VFS_UNLOCK_GIANT(vfslocked); 827152376Srwatson ktrace_exit(td); 8281541Srgrimes return (EACCES); 8291541Srgrimes } 830157233Sjhb VFS_UNLOCK_GIANT(vfslocked); 8311541Srgrimes } 8321541Srgrimes /* 83385397Sdillon * Clear all uses of the tracefile. 8341541Srgrimes */ 8351541Srgrimes if (ops == KTROP_CLEARFILE) { 836166678Smpp int vrele_count; 837166678Smpp 838166678Smpp vrele_count = 0; 83974927Sjhb sx_slock(&allproc_lock); 840166073Sdelphij FOREACH_PROC_IN_SYSTEM(p) { 84194618Sjhb PROC_LOCK(p); 842112198Sjhb if (p->p_tracevp == vp) { 84397993Sjhb if (ktrcanset(td, p)) { 84497993Sjhb mtx_lock(&ktrace_mtx); 845214158Sjhb ktr_freeproc(p, &cred, NULL); 84697993Sjhb mtx_unlock(&ktrace_mtx); 847166678Smpp vrele_count++; 848112198Sjhb crfree(cred); 849166678Smpp } else 8501541Srgrimes error = EPERM; 851166678Smpp } 852166678Smpp PROC_UNLOCK(p); 8531541Srgrimes } 85474927Sjhb sx_sunlock(&allproc_lock); 855166678Smpp if (vrele_count > 0) { 856166678Smpp vfslocked = VFS_LOCK_GIANT(vp->v_mount); 857166678Smpp while (vrele_count-- > 0) 858166678Smpp vrele(vp); 859166678Smpp VFS_UNLOCK_GIANT(vfslocked); 860166678Smpp } 8611541Srgrimes goto done; 8621541Srgrimes } 8631541Srgrimes /* 8641541Srgrimes * do it 8651541Srgrimes */ 866114026Sjhb sx_slock(&proctree_lock); 8671541Srgrimes if (uap->pid < 0) { 8681541Srgrimes /* 8691541Srgrimes * by process group 8701541Srgrimes */ 8711541Srgrimes pg = pgfind(-uap->pid); 8721541Srgrimes if (pg == NULL) { 87394861Sjhb sx_sunlock(&proctree_lock); 8741541Srgrimes error = ESRCH; 8751541Srgrimes goto done; 8761541Srgrimes } 87791140Stanimura /* 87891140Stanimura * ktrops() may call vrele(). Lock pg_members 87994861Sjhb * by the proctree_lock rather than pg_mtx. 88091140Stanimura */ 88191140Stanimura PGRP_UNLOCK(pg); 882147576Spjd nfound = 0; 883147576Spjd LIST_FOREACH(p, &pg->pg_members, p_pglist) { 884147576Spjd PROC_LOCK(p); 885220390Sjhb if (p->p_state == PRS_NEW || 886220390Sjhb p_cansee(td, p) != 0) { 887147576Spjd PROC_UNLOCK(p); 888147576Spjd continue; 889147576Spjd } 890147576Spjd nfound++; 8911541Srgrimes if (descend) 89294618Sjhb ret |= ktrsetchildren(td, p, ops, facs, vp); 8938876Srgrimes else 89494618Sjhb ret |= ktrops(td, p, ops, facs, vp); 895147576Spjd } 896147576Spjd if (nfound == 0) { 897147576Spjd sx_sunlock(&proctree_lock); 898147576Spjd error = ESRCH; 899147576Spjd goto done; 900147576Spjd } 9011541Srgrimes } else { 9021541Srgrimes /* 9031541Srgrimes * by pid 9041541Srgrimes */ 9051541Srgrimes p = pfind(uap->pid); 906211439Sjhb if (p == NULL) 9071541Srgrimes error = ESRCH; 908211439Sjhb else 909211439Sjhb error = p_cansee(td, p); 910147520Spjd if (error) { 911211439Sjhb if (p != NULL) 912211439Sjhb PROC_UNLOCK(p); 913147520Spjd sx_sunlock(&proctree_lock); 914147183Spjd goto done; 915147520Spjd } 9161541Srgrimes if (descend) 91794618Sjhb ret |= ktrsetchildren(td, p, ops, facs, vp); 9181541Srgrimes else 91994618Sjhb ret |= ktrops(td, p, ops, facs, vp); 9201541Srgrimes } 921114026Sjhb sx_sunlock(&proctree_lock); 9221541Srgrimes if (!ret) 9231541Srgrimes error = EPERM; 9241541Srgrimesdone: 925114026Sjhb if (vp != NULL) { 926157233Sjhb vfslocked = VFS_LOCK_GIANT(vp->v_mount); 92791406Sjhb (void) vn_close(vp, FWRITE, td->td_ucred, td); 928157233Sjhb VFS_UNLOCK_GIANT(vfslocked); 929114026Sjhb } 930152376Srwatson ktrace_exit(td); 9311541Srgrimes return (error); 932114026Sjhb#else /* !KTRACE */ 933114026Sjhb return (ENOSYS); 934114026Sjhb#endif /* KTRACE */ 9351541Srgrimes} 9361541Srgrimes 93718398Sphk/* ARGSUSED */ 93818398Sphkint 939225617Skmacysys_utrace(td, uap) 94083366Sjulian struct thread *td; 94118398Sphk register struct utrace_args *uap; 94218398Sphk{ 94383366Sjulian 94413203Swollman#ifdef KTRACE 94597993Sjhb struct ktr_request *req; 94699009Salfred void *cp; 947103237Sjhb int error; 94818398Sphk 949103237Sjhb if (!KTRPOINT(td, KTR_USER)) 950103237Sjhb return (0); 95170792Salfred if (uap->len > KTR_USER_MAXLEN) 95270707Salfred return (EINVAL); 953111119Simp cp = malloc(uap->len, M_KTRACE, M_WAITOK); 954103237Sjhb error = copyin(uap->addr, cp, uap->len); 955104230Sphk if (error) { 956104230Sphk free(cp, M_KTRACE); 957103237Sjhb return (error); 958104230Sphk } 95997993Sjhb req = ktr_getrequest(KTR_USER); 960104230Sphk if (req == NULL) { 961104230Sphk free(cp, M_KTRACE); 962122457Sjkoshy return (ENOMEM); 963104230Sphk } 964151927Srwatson req->ktr_buffer = cp; 965103237Sjhb req->ktr_header.ktr_len = uap->len; 966152376Srwatson ktr_submitrequest(td, req); 96718398Sphk return (0); 968114026Sjhb#else /* !KTRACE */ 96918398Sphk return (ENOSYS); 970114026Sjhb#endif /* KTRACE */ 97118398Sphk} 97218398Sphk 97318398Sphk#ifdef KTRACE 97412819Sphkstatic int 97594618Sjhbktrops(td, p, ops, facs, vp) 97694618Sjhb struct thread *td; 97794618Sjhb struct proc *p; 9781541Srgrimes int ops, facs; 9791541Srgrimes struct vnode *vp; 9801541Srgrimes{ 98197993Sjhb struct vnode *tracevp = NULL; 982112198Sjhb struct ucred *tracecred = NULL; 9831541Srgrimes 984211439Sjhb PROC_LOCK_ASSERT(p, MA_OWNED); 98594618Sjhb if (!ktrcanset(td, p)) { 98694618Sjhb PROC_UNLOCK(p); 9871541Srgrimes return (0); 98894618Sjhb } 989211439Sjhb if (p->p_flag & P_WEXIT) { 990211439Sjhb /* If the process is exiting, just ignore it. */ 991211439Sjhb PROC_UNLOCK(p); 992211439Sjhb return (1); 993211439Sjhb } 99497993Sjhb mtx_lock(&ktrace_mtx); 9951541Srgrimes if (ops == KTROP_SET) { 996112198Sjhb if (p->p_tracevp != vp) { 9971541Srgrimes /* 99894618Sjhb * if trace file already in use, relinquish below 9991541Srgrimes */ 1000112198Sjhb tracevp = p->p_tracevp; 100197993Sjhb VREF(vp); 1002112198Sjhb p->p_tracevp = vp; 10031541Srgrimes } 1004112198Sjhb if (p->p_tracecred != td->td_ucred) { 1005112198Sjhb tracecred = p->p_tracecred; 1006112198Sjhb p->p_tracecred = crhold(td->td_ucred); 1007112198Sjhb } 10081541Srgrimes p->p_traceflag |= facs; 1009170587Srwatson if (priv_check(td, PRIV_KTRACE) == 0) 10101541Srgrimes p->p_traceflag |= KTRFAC_ROOT; 10118876Srgrimes } else { 10121541Srgrimes /* KTROP_CLEAR */ 1013214158Sjhb if (((p->p_traceflag &= ~facs) & KTRFAC_MASK) == 0) 10141541Srgrimes /* no more tracing */ 1015214158Sjhb ktr_freeproc(p, &tracecred, &tracevp); 10161541Srgrimes } 101797993Sjhb mtx_unlock(&ktrace_mtx); 1018219311Sdchagin if ((p->p_traceflag & KTRFAC_MASK) != 0) 1019219311Sdchagin ktrprocctor_entered(td, p); 102094618Sjhb PROC_UNLOCK(p); 1021114026Sjhb if (tracevp != NULL) { 1022155031Sjeff int vfslocked; 1023155031Sjeff 1024155031Sjeff vfslocked = VFS_LOCK_GIANT(tracevp->v_mount); 102597993Sjhb vrele(tracevp); 1026155031Sjeff VFS_UNLOCK_GIANT(vfslocked); 1027114026Sjhb } 1028112198Sjhb if (tracecred != NULL) 1029112198Sjhb crfree(tracecred); 10301541Srgrimes 10311541Srgrimes return (1); 10321541Srgrimes} 10331541Srgrimes 103412819Sphkstatic int 103594618Sjhbktrsetchildren(td, top, ops, facs, vp) 103694618Sjhb struct thread *td; 103794618Sjhb struct proc *top; 10381541Srgrimes int ops, facs; 10391541Srgrimes struct vnode *vp; 10401541Srgrimes{ 10411541Srgrimes register struct proc *p; 10421541Srgrimes register int ret = 0; 10431541Srgrimes 10441541Srgrimes p = top; 1045211439Sjhb PROC_LOCK_ASSERT(p, MA_OWNED); 1046114026Sjhb sx_assert(&proctree_lock, SX_LOCKED); 10471541Srgrimes for (;;) { 104894618Sjhb ret |= ktrops(td, p, ops, facs, vp); 10491541Srgrimes /* 10501541Srgrimes * If this process has children, descend to them next, 10511541Srgrimes * otherwise do any siblings, and if done with this level, 10521541Srgrimes * follow back up the tree (but not past top). 10531541Srgrimes */ 105453212Sphk if (!LIST_EMPTY(&p->p_children)) 105553212Sphk p = LIST_FIRST(&p->p_children); 10561541Srgrimes else for (;;) { 1057114026Sjhb if (p == top) 10581541Srgrimes return (ret); 105953212Sphk if (LIST_NEXT(p, p_sibling)) { 106053212Sphk p = LIST_NEXT(p, p_sibling); 10611541Srgrimes break; 10621541Srgrimes } 106314529Shsu p = p->p_pptr; 10641541Srgrimes } 1065211439Sjhb PROC_LOCK(p); 10661541Srgrimes } 10671541Srgrimes /*NOTREACHED*/ 10681541Srgrimes} 10691541Srgrimes 107012819Sphkstatic void 1071152376Srwatsonktr_writerequest(struct thread *td, struct ktr_request *req) 107297993Sjhb{ 107397993Sjhb struct ktr_header *kth; 10741541Srgrimes struct vnode *vp; 107597993Sjhb struct proc *p; 107697993Sjhb struct ucred *cred; 10771541Srgrimes struct uio auio; 107897993Sjhb struct iovec aiov[3]; 107962976Smckusick struct mount *mp; 108097993Sjhb int datalen, buflen, vrele_count; 1081157233Sjhb int error, vfslocked; 10821541Srgrimes 108397993Sjhb /* 1084152376Srwatson * We hold the vnode and credential for use in I/O in case ktrace is 1085152376Srwatson * disabled on the process as we write out the request. 1086152376Srwatson * 1087152376Srwatson * XXXRW: This is not ideal: we could end up performing a write after 1088152376Srwatson * the vnode has been closed. 1089152376Srwatson */ 1090152376Srwatson mtx_lock(&ktrace_mtx); 1091152376Srwatson vp = td->td_proc->p_tracevp; 1092152376Srwatson cred = td->td_proc->p_tracecred; 1093152376Srwatson 1094152376Srwatson /* 109597993Sjhb * If vp is NULL, the vp has been cleared out from under this 1096152376Srwatson * request, so just drop it. Make sure the credential and vnode are 1097152376Srwatson * in sync: we should have both or neither. 109897993Sjhb */ 1099152376Srwatson if (vp == NULL) { 1100152376Srwatson KASSERT(cred == NULL, ("ktr_writerequest: cred != NULL")); 1101185583Sbz mtx_unlock(&ktrace_mtx); 11021541Srgrimes return; 1103152376Srwatson } 1104185583Sbz VREF(vp); 1105152376Srwatson KASSERT(cred != NULL, ("ktr_writerequest: cred == NULL")); 1106185583Sbz crhold(cred); 1107185583Sbz mtx_unlock(&ktrace_mtx); 1108152376Srwatson 110997993Sjhb kth = &req->ktr_header; 1110189707Sjhb KASSERT(((u_short)kth->ktr_type & ~KTR_DROP) < 1111189707Sjhb sizeof(data_lengths) / sizeof(data_lengths[0]), 1112189707Sjhb ("data_lengths array overflow")); 1113118607Sjhb datalen = data_lengths[(u_short)kth->ktr_type & ~KTR_DROP]; 111497993Sjhb buflen = kth->ktr_len; 11151541Srgrimes auio.uio_iov = &aiov[0]; 11161541Srgrimes auio.uio_offset = 0; 11171541Srgrimes auio.uio_segflg = UIO_SYSSPACE; 11181541Srgrimes auio.uio_rw = UIO_WRITE; 11191541Srgrimes aiov[0].iov_base = (caddr_t)kth; 11201541Srgrimes aiov[0].iov_len = sizeof(struct ktr_header); 11211541Srgrimes auio.uio_resid = sizeof(struct ktr_header); 11221541Srgrimes auio.uio_iovcnt = 1; 112397993Sjhb auio.uio_td = td; 112497993Sjhb if (datalen != 0) { 112597993Sjhb aiov[1].iov_base = (caddr_t)&req->ktr_data; 112697993Sjhb aiov[1].iov_len = datalen; 112797993Sjhb auio.uio_resid += datalen; 11281541Srgrimes auio.uio_iovcnt++; 112997993Sjhb kth->ktr_len += datalen; 11301541Srgrimes } 113197993Sjhb if (buflen != 0) { 1132151927Srwatson KASSERT(req->ktr_buffer != NULL, ("ktrace: nothing to write")); 1133151927Srwatson aiov[auio.uio_iovcnt].iov_base = req->ktr_buffer; 113497993Sjhb aiov[auio.uio_iovcnt].iov_len = buflen; 113597993Sjhb auio.uio_resid += buflen; 113697993Sjhb auio.uio_iovcnt++; 1137103235Sjhb } 1138152376Srwatson 1139157233Sjhb vfslocked = VFS_LOCK_GIANT(vp->v_mount); 114062976Smckusick vn_start_write(vp, &mp, V_WAIT); 1141175202Sattilio vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 1142101123Srwatson#ifdef MAC 1143172930Srwatson error = mac_vnode_check_write(cred, NOCRED, vp); 1144101123Srwatson if (error == 0) 1145101123Srwatson#endif 1146101123Srwatson error = VOP_WRITE(vp, &auio, IO_UNIT | IO_APPEND, cred); 1147175294Sattilio VOP_UNLOCK(vp, 0); 114862976Smckusick vn_finished_write(mp); 1149185583Sbz crfree(cred); 1150185583Sbz if (!error) { 1151185583Sbz vrele(vp); 1152185583Sbz VFS_UNLOCK_GIANT(vfslocked); 1153185583Sbz return; 1154185583Sbz } 1155157233Sjhb VFS_UNLOCK_GIANT(vfslocked); 1156185583Sbz 11571541Srgrimes /* 115897993Sjhb * If error encountered, give up tracing on this vnode. We defer 115997993Sjhb * all the vrele()'s on the vnode until after we are finished walking 116097993Sjhb * the various lists to avoid needlessly holding locks. 1161185583Sbz * NB: at this point we still hold the vnode reference that must 1162185583Sbz * not go away as we need the valid vnode to compare with. Thus let 1163185583Sbz * vrele_count start at 1 and the reference will be freed 1164185583Sbz * by the loop at the end after our last use of vp. 11651541Srgrimes */ 11661541Srgrimes log(LOG_NOTICE, "ktrace write failed, errno %d, tracing stopped\n", 11671541Srgrimes error); 1168185583Sbz vrele_count = 1; 116997993Sjhb /* 117097993Sjhb * First, clear this vnode from being used by any processes in the 117197993Sjhb * system. 117297993Sjhb * XXX - If one process gets an EPERM writing to the vnode, should 117397993Sjhb * we really do this? Other processes might have suitable 117497993Sjhb * credentials for the operation. 117597993Sjhb */ 1176112198Sjhb cred = NULL; 117774927Sjhb sx_slock(&allproc_lock); 1178166073Sdelphij FOREACH_PROC_IN_SYSTEM(p) { 117997993Sjhb PROC_LOCK(p); 1180112198Sjhb if (p->p_tracevp == vp) { 118197993Sjhb mtx_lock(&ktrace_mtx); 1182214158Sjhb ktr_freeproc(p, &cred, NULL); 118397993Sjhb mtx_unlock(&ktrace_mtx); 118497993Sjhb vrele_count++; 11851541Srgrimes } 118697993Sjhb PROC_UNLOCK(p); 1187112198Sjhb if (cred != NULL) { 1188112198Sjhb crfree(cred); 1189112198Sjhb cred = NULL; 1190112198Sjhb } 11911541Srgrimes } 119274927Sjhb sx_sunlock(&allproc_lock); 1193152376Srwatson 1194157233Sjhb vfslocked = VFS_LOCK_GIANT(vp->v_mount); 119597993Sjhb while (vrele_count-- > 0) 119697993Sjhb vrele(vp); 1197157233Sjhb VFS_UNLOCK_GIANT(vfslocked); 11981541Srgrimes} 11991541Srgrimes 12001541Srgrimes/* 12011541Srgrimes * Return true if caller has permission to set the ktracing state 12021541Srgrimes * of target. Essentially, the target can't possess any 12031541Srgrimes * more permissions than the caller. KTRFAC_ROOT signifies that 12048876Srgrimes * root previously set the tracing status on the target process, and 12051541Srgrimes * so, only root may further change it. 12061541Srgrimes */ 120712819Sphkstatic int 120894618Sjhbktrcanset(td, targetp) 120994618Sjhb struct thread *td; 121094618Sjhb struct proc *targetp; 12111541Srgrimes{ 12121541Srgrimes 121394618Sjhb PROC_LOCK_ASSERT(targetp, MA_OWNED); 121479335Srwatson if (targetp->p_traceflag & KTRFAC_ROOT && 1215170587Srwatson priv_check(td, PRIV_KTRACE)) 121646155Sphk return (0); 12171541Srgrimes 121896886Sjhb if (p_candebug(td, targetp) != 0) 121979335Srwatson return (0); 122079335Srwatson 122179335Srwatson return (1); 12221541Srgrimes} 12231541Srgrimes 122413203Swollman#endif /* KTRACE */ 1225