kern_ktrace.c revision 112199
11541Srgrimes/* 21541Srgrimes * Copyright (c) 1989, 1993 31541Srgrimes * The Regents of the University of California. All rights reserved. 41541Srgrimes * 51541Srgrimes * Redistribution and use in source and binary forms, with or without 61541Srgrimes * modification, are permitted provided that the following conditions 71541Srgrimes * are met: 81541Srgrimes * 1. Redistributions of source code must retain the above copyright 91541Srgrimes * notice, this list of conditions and the following disclaimer. 101541Srgrimes * 2. Redistributions in binary form must reproduce the above copyright 111541Srgrimes * notice, this list of conditions and the following disclaimer in the 121541Srgrimes * documentation and/or other materials provided with the distribution. 131541Srgrimes * 3. All advertising materials mentioning features or use of this software 141541Srgrimes * must display the following acknowledgement: 151541Srgrimes * This product includes software developed by the University of 161541Srgrimes * California, Berkeley and its contributors. 171541Srgrimes * 4. Neither the name of the University nor the names of its contributors 181541Srgrimes * may be used to endorse or promote products derived from this software 191541Srgrimes * without specific prior written permission. 201541Srgrimes * 211541Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 221541Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 231541Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 241541Srgrimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 251541Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 261541Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 271541Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 281541Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 291541Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 301541Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 311541Srgrimes * SUCH DAMAGE. 321541Srgrimes * 331541Srgrimes * @(#)kern_ktrace.c 8.2 (Berkeley) 9/23/93 3450477Speter * $FreeBSD: head/sys/kern/kern_ktrace.c 112199 2003-03-13 18:31:15Z jhb $ 351541Srgrimes */ 361541Srgrimes 3713203Swollman#include "opt_ktrace.h" 38101123Srwatson#include "opt_mac.h" 391541Srgrimes 401541Srgrimes#include <sys/param.h> 412112Swollman#include <sys/systm.h> 4297993Sjhb#include <sys/fcntl.h> 4397993Sjhb#include <sys/jail.h> 4497993Sjhb#include <sys/kernel.h> 4597993Sjhb#include <sys/kthread.h> 4676166Smarkm#include <sys/lock.h> 4776166Smarkm#include <sys/mutex.h> 48101123Srwatson#include <sys/mac.h> 4997993Sjhb#include <sys/malloc.h> 5097993Sjhb#include <sys/namei.h> 511541Srgrimes#include <sys/proc.h> 5297993Sjhb#include <sys/unistd.h> 531541Srgrimes#include <sys/vnode.h> 541541Srgrimes#include <sys/ktrace.h> 5597993Sjhb#include <sys/sema.h> 5674927Sjhb#include <sys/sx.h> 5797993Sjhb#include <sys/sysctl.h> 581541Srgrimes#include <sys/syslog.h> 5997993Sjhb#include <sys/sysproto.h> 601541Srgrimes 6130354Sphkstatic MALLOC_DEFINE(M_KTRACE, "KTRACE", "KTRACE"); 6230309Sphk 6313203Swollman#ifdef KTRACE 6412577Sbde 6597993Sjhb#ifndef KTRACE_REQUEST_POOL 6697993Sjhb#define KTRACE_REQUEST_POOL 100 6797993Sjhb#endif 6812819Sphk 6997993Sjhbstruct ktr_request { 7097993Sjhb struct ktr_header ktr_header; 7197993Sjhb struct ucred *ktr_cred; 7297993Sjhb struct vnode *ktr_vp; 7397993Sjhb union { 7497993Sjhb struct ktr_syscall ktr_syscall; 7597993Sjhb struct ktr_sysret ktr_sysret; 7697993Sjhb struct ktr_genio ktr_genio; 7797993Sjhb struct ktr_psig ktr_psig; 7897993Sjhb struct ktr_csw ktr_csw; 7997993Sjhb } ktr_data; 8097993Sjhb STAILQ_ENTRY(ktr_request) ktr_list; 8197993Sjhb}; 8297993Sjhb 8397993Sjhbstatic int data_lengths[] = { 8497993Sjhb 0, /* none */ 8597993Sjhb offsetof(struct ktr_syscall, ktr_args), /* KTR_SYSCALL */ 8697993Sjhb sizeof(struct ktr_sysret), /* KTR_SYSRET */ 8797993Sjhb 0, /* KTR_NAMEI */ 8897993Sjhb sizeof(struct ktr_genio), /* KTR_GENIO */ 8997993Sjhb sizeof(struct ktr_psig), /* KTR_PSIG */ 9097993Sjhb sizeof(struct ktr_csw), /* KTR_CSW */ 9197993Sjhb 0 /* KTR_USER */ 9297993Sjhb}; 9397993Sjhb 9497993Sjhbstatic STAILQ_HEAD(, ktr_request) ktr_todo; 9597993Sjhbstatic STAILQ_HEAD(, ktr_request) ktr_free; 9697993Sjhb 97103234SjhbSYSCTL_NODE(_kern, OID_AUTO, ktrace, CTLFLAG_RD, 0, "KTRACE options"); 98103234Sjhb 9997993Sjhbstatic uint ktr_requestpool = KTRACE_REQUEST_POOL; 100103234SjhbTUNABLE_INT("kern.ktrace.request_pool", &ktr_requestpool); 10197993Sjhb 102103234Sjhbstatic uint ktr_geniosize = PAGE_SIZE; 103103234SjhbTUNABLE_INT("kern.ktrace.genio_size", &ktr_geniosize); 104103234SjhbSYSCTL_UINT(_kern_ktrace, OID_AUTO, genio_size, CTLFLAG_RW, &ktr_geniosize, 105103234Sjhb 0, "Maximum size of genio event payload"); 106103234Sjhb 10797993Sjhbstatic int print_message = 1; 10897993Sjhbstruct mtx ktrace_mtx; 10997993Sjhbstatic struct sema ktrace_sema; 11097993Sjhb 11197993Sjhbstatic void ktrace_init(void *dummy); 11297993Sjhbstatic int sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS); 11397993Sjhbstatic uint ktrace_resize_pool(uint newsize); 11497993Sjhbstatic struct ktr_request *ktr_getrequest(int type); 11597993Sjhbstatic void ktr_submitrequest(struct ktr_request *req); 11697993Sjhbstatic void ktr_freerequest(struct ktr_request *req); 11797993Sjhbstatic void ktr_loop(void *dummy); 11897993Sjhbstatic void ktr_writerequest(struct ktr_request *req); 11997993Sjhbstatic int ktrcanset(struct thread *,struct proc *); 12097993Sjhbstatic int ktrsetchildren(struct thread *,struct proc *,int,int,struct vnode *); 12197993Sjhbstatic int ktrops(struct thread *,struct proc *,int,int,struct vnode *); 12297993Sjhb 12397993Sjhbstatic void 12497993Sjhbktrace_init(void *dummy) 1251541Srgrimes{ 12697993Sjhb struct ktr_request *req; 12797993Sjhb int i; 1281541Srgrimes 12997993Sjhb mtx_init(&ktrace_mtx, "ktrace", NULL, MTX_DEF | MTX_QUIET); 13097993Sjhb sema_init(&ktrace_sema, 0, "ktrace"); 13197993Sjhb STAILQ_INIT(&ktr_todo); 13297993Sjhb STAILQ_INIT(&ktr_free); 13397993Sjhb for (i = 0; i < ktr_requestpool; i++) { 134111119Simp req = malloc(sizeof(struct ktr_request), M_KTRACE, M_WAITOK); 13597993Sjhb STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list); 13697993Sjhb } 137104354Sscottl kthread_create(ktr_loop, NULL, NULL, RFHIGHPID, 0, "ktrace"); 1381541Srgrimes} 13997993SjhbSYSINIT(ktrace_init, SI_SUB_KTRACE, SI_ORDER_ANY, ktrace_init, NULL); 1401541Srgrimes 14197993Sjhbstatic int 14297993Sjhbsysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS) 14397993Sjhb{ 14497993Sjhb struct thread *td; 14597993Sjhb uint newsize, oldsize, wantsize; 14697993Sjhb int error; 14797993Sjhb 14897993Sjhb /* Handle easy read-only case first to avoid warnings from GCC. */ 14997993Sjhb if (!req->newptr) { 15097993Sjhb mtx_lock(&ktrace_mtx); 15197993Sjhb oldsize = ktr_requestpool; 15297993Sjhb mtx_unlock(&ktrace_mtx); 15397993Sjhb return (SYSCTL_OUT(req, &oldsize, sizeof(uint))); 15497993Sjhb } 15597993Sjhb 15697993Sjhb error = SYSCTL_IN(req, &wantsize, sizeof(uint)); 15797993Sjhb if (error) 15897993Sjhb return (error); 15997993Sjhb td = curthread; 16097993Sjhb td->td_inktrace = 1; 16197993Sjhb mtx_lock(&ktrace_mtx); 16297993Sjhb oldsize = ktr_requestpool; 16397993Sjhb newsize = ktrace_resize_pool(wantsize); 16497993Sjhb mtx_unlock(&ktrace_mtx); 16597993Sjhb td->td_inktrace = 0; 16697993Sjhb error = SYSCTL_OUT(req, &oldsize, sizeof(uint)); 16797993Sjhb if (error) 16897993Sjhb return (error); 16997993Sjhb if (newsize != wantsize) 17097993Sjhb return (ENOSPC); 17197993Sjhb return (0); 17297993Sjhb} 173103234SjhbSYSCTL_PROC(_kern_ktrace, OID_AUTO, request_pool, CTLTYPE_UINT|CTLFLAG_RW, 17497993Sjhb &ktr_requestpool, 0, sysctl_kern_ktrace_request_pool, "IU", ""); 17597993Sjhb 17697993Sjhbstatic uint 17797993Sjhbktrace_resize_pool(uint newsize) 17897993Sjhb{ 17997993Sjhb struct ktr_request *req; 18097993Sjhb 18197993Sjhb mtx_assert(&ktrace_mtx, MA_OWNED); 18297993Sjhb print_message = 1; 18397993Sjhb if (newsize == ktr_requestpool) 18497993Sjhb return (newsize); 18597993Sjhb if (newsize < ktr_requestpool) 18697993Sjhb /* Shrink pool down to newsize if possible. */ 18797993Sjhb while (ktr_requestpool > newsize) { 18897993Sjhb req = STAILQ_FIRST(&ktr_free); 18997993Sjhb if (req == NULL) 19097993Sjhb return (ktr_requestpool); 19197993Sjhb STAILQ_REMOVE_HEAD(&ktr_free, ktr_list); 19297993Sjhb ktr_requestpool--; 19397993Sjhb mtx_unlock(&ktrace_mtx); 19497993Sjhb free(req, M_KTRACE); 19597993Sjhb mtx_lock(&ktrace_mtx); 19697993Sjhb } 19797993Sjhb else 19897993Sjhb /* Grow pool up to newsize. */ 19997993Sjhb while (ktr_requestpool < newsize) { 20097993Sjhb mtx_unlock(&ktrace_mtx); 20197993Sjhb req = malloc(sizeof(struct ktr_request), M_KTRACE, 202111119Simp M_WAITOK); 20397993Sjhb mtx_lock(&ktrace_mtx); 20497993Sjhb STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list); 20597993Sjhb ktr_requestpool++; 20697993Sjhb } 20797993Sjhb return (ktr_requestpool); 20897993Sjhb} 20997993Sjhb 21097993Sjhbstatic struct ktr_request * 21197993Sjhbktr_getrequest(int type) 21297993Sjhb{ 21397993Sjhb struct ktr_request *req; 21497993Sjhb struct thread *td = curthread; 21597993Sjhb struct proc *p = td->td_proc; 21697993Sjhb int pm; 21797993Sjhb 21897993Sjhb td->td_inktrace = 1; 21997993Sjhb mtx_lock(&ktrace_mtx); 22097993Sjhb if (!KTRCHECK(td, type)) { 22197993Sjhb mtx_unlock(&ktrace_mtx); 22297993Sjhb td->td_inktrace = 0; 22397993Sjhb return (NULL); 22497993Sjhb } 22597993Sjhb req = STAILQ_FIRST(&ktr_free); 22697993Sjhb if (req != NULL) { 22797993Sjhb STAILQ_REMOVE_HEAD(&ktr_free, ktr_list); 22897993Sjhb req->ktr_header.ktr_type = type; 229112199Sjhb if (p->p_traceflag & KTRFAC_DROP) { 230112199Sjhb req->ktr_header.ktr_type |= KTR_DROP; 231112199Sjhb p->p_traceflag &= ~KTRFAC_DROP; 232112199Sjhb } 233112198Sjhb KASSERT(p->p_tracevp != NULL, ("ktrace: no trace vnode")); 234112198Sjhb KASSERT(p->p_tracecred != NULL, ("ktrace: no trace cred")); 235112198Sjhb req->ktr_vp = p->p_tracevp; 236112198Sjhb VREF(p->p_tracevp); 237112198Sjhb req->ktr_cred = crhold(p->p_tracecred); 23897993Sjhb mtx_unlock(&ktrace_mtx); 23997993Sjhb microtime(&req->ktr_header.ktr_time); 24097993Sjhb req->ktr_header.ktr_pid = p->p_pid; 24197993Sjhb bcopy(p->p_comm, req->ktr_header.ktr_comm, MAXCOMLEN + 1); 24297993Sjhb req->ktr_header.ktr_buffer = NULL; 24397993Sjhb req->ktr_header.ktr_len = 0; 24497993Sjhb } else { 245112199Sjhb p->p_traceflag |= KTRFAC_DROP; 24697993Sjhb pm = print_message; 24797993Sjhb print_message = 0; 24897993Sjhb mtx_unlock(&ktrace_mtx); 24997993Sjhb if (pm) 25097993Sjhb printf("Out of ktrace request objects.\n"); 25197993Sjhb td->td_inktrace = 0; 25297993Sjhb } 25397993Sjhb return (req); 25497993Sjhb} 25597993Sjhb 25697993Sjhbstatic void 25797993Sjhbktr_submitrequest(struct ktr_request *req) 25897993Sjhb{ 25997993Sjhb 26097993Sjhb mtx_lock(&ktrace_mtx); 26197993Sjhb STAILQ_INSERT_TAIL(&ktr_todo, req, ktr_list); 26297993Sjhb sema_post(&ktrace_sema); 26397993Sjhb mtx_unlock(&ktrace_mtx); 26497993Sjhb curthread->td_inktrace = 0; 26597993Sjhb} 26697993Sjhb 26797993Sjhbstatic void 26897993Sjhbktr_freerequest(struct ktr_request *req) 26997993Sjhb{ 27097993Sjhb 27197993Sjhb crfree(req->ktr_cred); 272101153Sjhb if (req->ktr_vp != NULL) { 273101153Sjhb mtx_lock(&Giant); 274101153Sjhb vrele(req->ktr_vp); 275101153Sjhb mtx_unlock(&Giant); 276101153Sjhb } 277103235Sjhb if (req->ktr_header.ktr_buffer != NULL) 278103235Sjhb free(req->ktr_header.ktr_buffer, M_KTRACE); 27997993Sjhb mtx_lock(&ktrace_mtx); 28097993Sjhb STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list); 28197993Sjhb mtx_unlock(&ktrace_mtx); 28297993Sjhb} 28397993Sjhb 28497993Sjhbstatic void 28597993Sjhbktr_loop(void *dummy) 28697993Sjhb{ 28797993Sjhb struct ktr_request *req; 28897993Sjhb struct thread *td; 28997993Sjhb struct ucred *cred; 29097993Sjhb 29197993Sjhb /* Only cache these values once. */ 29297993Sjhb td = curthread; 29397993Sjhb cred = td->td_ucred; 29497993Sjhb for (;;) { 29597993Sjhb sema_wait(&ktrace_sema); 29697993Sjhb mtx_lock(&ktrace_mtx); 29797993Sjhb req = STAILQ_FIRST(&ktr_todo); 29897993Sjhb STAILQ_REMOVE_HEAD(&ktr_todo, ktr_list); 29997993Sjhb KASSERT(req != NULL, ("got a NULL request")); 300103236Sjhb mtx_unlock(&ktrace_mtx); 301103236Sjhb /* 302103236Sjhb * It is not enough just to pass the cached cred 303103236Sjhb * to the VOP's in ktr_writerequest(). Some VFS 304103236Sjhb * operations use curthread->td_ucred, so we need 305103236Sjhb * to modify our thread's credentials as well. 306103236Sjhb * Evil. 307103236Sjhb */ 308103236Sjhb td->td_ucred = req->ktr_cred; 309103236Sjhb ktr_writerequest(req); 310103236Sjhb td->td_ucred = cred; 31197993Sjhb ktr_freerequest(req); 31297993Sjhb } 31397993Sjhb} 31497993Sjhb 31582585Sdillon/* 31682585Sdillon * MPSAFE 31782585Sdillon */ 3181549Srgrimesvoid 31997993Sjhbktrsyscall(code, narg, args) 32047955Sdt int code, narg; 32147955Sdt register_t args[]; 3221541Srgrimes{ 32397993Sjhb struct ktr_request *req; 32497993Sjhb struct ktr_syscall *ktp; 32597993Sjhb size_t buflen; 326103233Sjhb char *buf = NULL; 3271541Srgrimes 328103233Sjhb buflen = sizeof(register_t) * narg; 329103233Sjhb if (buflen > 0) { 330111119Simp buf = malloc(buflen, M_KTRACE, M_WAITOK); 331103233Sjhb bcopy(args, buf, buflen); 332103233Sjhb } 33397993Sjhb req = ktr_getrequest(KTR_SYSCALL); 334104230Sphk if (req == NULL) { 335104230Sphk if (buf != NULL) 336104230Sphk free(buf, M_KTRACE); 33797993Sjhb return; 338104230Sphk } 33997993Sjhb ktp = &req->ktr_data.ktr_syscall; 3401541Srgrimes ktp->ktr_code = code; 3411541Srgrimes ktp->ktr_narg = narg; 34297993Sjhb if (buflen > 0) { 34397993Sjhb req->ktr_header.ktr_len = buflen; 344103233Sjhb req->ktr_header.ktr_buffer = buf; 34597993Sjhb } 34697993Sjhb ktr_submitrequest(req); 3471541Srgrimes} 3481541Srgrimes 34982585Sdillon/* 35082585Sdillon * MPSAFE 35182585Sdillon */ 3521549Srgrimesvoid 35397993Sjhbktrsysret(code, error, retval) 35447955Sdt int code, error; 35547955Sdt register_t retval; 3561541Srgrimes{ 35797993Sjhb struct ktr_request *req; 35897993Sjhb struct ktr_sysret *ktp; 3591541Srgrimes 36097993Sjhb req = ktr_getrequest(KTR_SYSRET); 36197993Sjhb if (req == NULL) 36297993Sjhb return; 36397993Sjhb ktp = &req->ktr_data.ktr_sysret; 36497993Sjhb ktp->ktr_code = code; 36597993Sjhb ktp->ktr_error = error; 36697993Sjhb ktp->ktr_retval = retval; /* what about val2 ? */ 36797993Sjhb ktr_submitrequest(req); 3681541Srgrimes} 3691541Srgrimes 3701549Srgrimesvoid 37197993Sjhbktrnamei(path) 3721541Srgrimes char *path; 3731541Srgrimes{ 37497993Sjhb struct ktr_request *req; 37597993Sjhb int namelen; 376103233Sjhb char *buf = NULL; 3771541Srgrimes 378103233Sjhb namelen = strlen(path); 379103233Sjhb if (namelen > 0) { 380111119Simp buf = malloc(namelen, M_KTRACE, M_WAITOK); 381103233Sjhb bcopy(path, buf, namelen); 382103233Sjhb } 38397993Sjhb req = ktr_getrequest(KTR_NAMEI); 384104230Sphk if (req == NULL) { 385104230Sphk if (buf != NULL) 386104230Sphk free(buf, M_KTRACE); 38797993Sjhb return; 388104230Sphk } 38997993Sjhb if (namelen > 0) { 39097993Sjhb req->ktr_header.ktr_len = namelen; 391103233Sjhb req->ktr_header.ktr_buffer = buf; 39297993Sjhb } 39397993Sjhb ktr_submitrequest(req); 3941541Srgrimes} 3951541Srgrimes 39697993Sjhb/* 39797993Sjhb * Since the uio may not stay valid, we can not hand off this request to 39897993Sjhb * the thread and need to process it synchronously. However, we wish to 39997993Sjhb * keep the relative order of records in a trace file correct, so we 40097993Sjhb * do put this request on the queue (if it isn't empty) and then block. 40197993Sjhb * The ktrace thread waks us back up when it is time for this event to 40297993Sjhb * be posted and blocks until we have completed writing out the event 40397993Sjhb * and woken it back up. 40497993Sjhb */ 4051549Srgrimesvoid 40697993Sjhbktrgenio(fd, rw, uio, error) 4071541Srgrimes int fd; 4081541Srgrimes enum uio_rw rw; 40962378Sgreen struct uio *uio; 41062378Sgreen int error; 4111541Srgrimes{ 41297993Sjhb struct ktr_request *req; 41397993Sjhb struct ktr_genio *ktg; 414103235Sjhb int datalen; 415103235Sjhb char *buf; 4168876Srgrimes 4171541Srgrimes if (error) 4181541Srgrimes return; 419103235Sjhb uio->uio_offset = 0; 420103235Sjhb uio->uio_rw = UIO_WRITE; 421103235Sjhb datalen = imin(uio->uio_resid, ktr_geniosize); 422111119Simp buf = malloc(datalen, M_KTRACE, M_WAITOK); 423103235Sjhb if (uiomove(buf, datalen, uio)) { 424103235Sjhb free(buf, M_KTRACE); 425103235Sjhb return; 426103235Sjhb } 42797993Sjhb req = ktr_getrequest(KTR_GENIO); 428103235Sjhb if (req == NULL) { 429103235Sjhb free(buf, M_KTRACE); 43097993Sjhb return; 431103235Sjhb } 43297993Sjhb ktg = &req->ktr_data.ktr_genio; 43397993Sjhb ktg->ktr_fd = fd; 43497993Sjhb ktg->ktr_rw = rw; 435103235Sjhb req->ktr_header.ktr_len = datalen; 436103235Sjhb req->ktr_header.ktr_buffer = buf; 43797993Sjhb ktr_submitrequest(req); 4381541Srgrimes} 4391541Srgrimes 4401549Srgrimesvoid 44197993Sjhbktrpsig(sig, action, mask, code) 44251941Smarcel int sig; 4431541Srgrimes sig_t action; 44451791Smarcel sigset_t *mask; 44551941Smarcel int code; 4461541Srgrimes{ 44797993Sjhb struct ktr_request *req; 44897993Sjhb struct ktr_psig *kp; 4491541Srgrimes 45097993Sjhb req = ktr_getrequest(KTR_PSIG); 45197993Sjhb if (req == NULL) 45297993Sjhb return; 45397993Sjhb kp = &req->ktr_data.ktr_psig; 45497993Sjhb kp->signo = (char)sig; 45597993Sjhb kp->action = action; 45697993Sjhb kp->mask = *mask; 45797993Sjhb kp->code = code; 45897993Sjhb ktr_submitrequest(req); 4591541Srgrimes} 4601541Srgrimes 4611549Srgrimesvoid 46297993Sjhbktrcsw(out, user) 4631541Srgrimes int out, user; 4641541Srgrimes{ 46597993Sjhb struct ktr_request *req; 46697993Sjhb struct ktr_csw *kc; 4671541Srgrimes 46897993Sjhb req = ktr_getrequest(KTR_CSW); 46997993Sjhb if (req == NULL) 47097993Sjhb return; 47197993Sjhb kc = &req->ktr_data.ktr_csw; 47297993Sjhb kc->out = out; 47397993Sjhb kc->user = user; 47497993Sjhb ktr_submitrequest(req); 4751541Srgrimes} 47613203Swollman#endif 4771541Srgrimes 4781541Srgrimes/* Interface and common routines */ 4791541Srgrimes 4801541Srgrimes/* 4811541Srgrimes * ktrace system call 4821541Srgrimes */ 48312221Sbde#ifndef _SYS_SYSPROTO_H_ 4841541Srgrimesstruct ktrace_args { 4851541Srgrimes char *fname; 4861541Srgrimes int ops; 4871541Srgrimes int facs; 4881541Srgrimes int pid; 4891541Srgrimes}; 49012221Sbde#endif 4911541Srgrimes/* ARGSUSED */ 4921549Srgrimesint 49383366Sjulianktrace(td, uap) 49483366Sjulian struct thread *td; 4951541Srgrimes register struct ktrace_args *uap; 4961541Srgrimes{ 49713203Swollman#ifdef KTRACE 4981541Srgrimes register struct vnode *vp = NULL; 4991541Srgrimes register struct proc *p; 5001541Srgrimes struct pgrp *pg; 5011541Srgrimes int facs = uap->facs & ~KTRFAC_ROOT; 5021541Srgrimes int ops = KTROP(uap->ops); 5031541Srgrimes int descend = uap->ops & KTRFLAG_DESCEND; 5041541Srgrimes int ret = 0; 50562550Smckusick int flags, error = 0; 5061541Srgrimes struct nameidata nd; 507112198Sjhb struct ucred *cred; 5081541Srgrimes 50997993Sjhb td->td_inktrace = 1; 5101541Srgrimes if (ops != KTROP_CLEAR) { 5111541Srgrimes /* 5121541Srgrimes * an operation which requires a file argument. 5131541Srgrimes */ 51483366Sjulian NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, uap->fname, td); 51562550Smckusick flags = FREAD | FWRITE | O_NOFOLLOW; 51662550Smckusick error = vn_open(&nd, &flags, 0); 5173308Sphk if (error) { 51897993Sjhb td->td_inktrace = 0; 5191541Srgrimes return (error); 5201541Srgrimes } 52154655Seivind NDFREE(&nd, NDF_ONLY_PNBUF); 5221541Srgrimes vp = nd.ni_vp; 52383366Sjulian VOP_UNLOCK(vp, 0, td); 5241541Srgrimes if (vp->v_type != VREG) { 52591406Sjhb (void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td); 52697993Sjhb td->td_inktrace = 0; 5271541Srgrimes return (EACCES); 5281541Srgrimes } 5291541Srgrimes } 5301541Srgrimes /* 53185397Sdillon * Clear all uses of the tracefile. 5321541Srgrimes */ 5331541Srgrimes if (ops == KTROP_CLEARFILE) { 53474927Sjhb sx_slock(&allproc_lock); 53553212Sphk LIST_FOREACH(p, &allproc, p_list) { 53694618Sjhb PROC_LOCK(p); 537112198Sjhb if (p->p_tracevp == vp) { 53897993Sjhb if (ktrcanset(td, p)) { 53997993Sjhb mtx_lock(&ktrace_mtx); 540112198Sjhb cred = p->p_tracecred; 541112198Sjhb p->p_tracecred = NULL; 542112198Sjhb p->p_tracevp = NULL; 5431541Srgrimes p->p_traceflag = 0; 54497993Sjhb mtx_unlock(&ktrace_mtx); 54594618Sjhb PROC_UNLOCK(p); 5461541Srgrimes (void) vn_close(vp, FREAD|FWRITE, 547112198Sjhb cred, td); 548112198Sjhb crfree(cred); 54985397Sdillon } else { 55094618Sjhb PROC_UNLOCK(p); 5511541Srgrimes error = EPERM; 55285397Sdillon } 55394618Sjhb } else 55494618Sjhb PROC_UNLOCK(p); 5551541Srgrimes } 55674927Sjhb sx_sunlock(&allproc_lock); 5571541Srgrimes goto done; 5581541Srgrimes } 5591541Srgrimes /* 5601541Srgrimes * need something to (un)trace (XXX - why is this here?) 5611541Srgrimes */ 5621541Srgrimes if (!facs) { 5631541Srgrimes error = EINVAL; 5641541Srgrimes goto done; 5651541Srgrimes } 5668876Srgrimes /* 5671541Srgrimes * do it 5681541Srgrimes */ 5691541Srgrimes if (uap->pid < 0) { 5701541Srgrimes /* 5711541Srgrimes * by process group 5721541Srgrimes */ 57394861Sjhb sx_slock(&proctree_lock); 5741541Srgrimes pg = pgfind(-uap->pid); 5751541Srgrimes if (pg == NULL) { 57694861Sjhb sx_sunlock(&proctree_lock); 5771541Srgrimes error = ESRCH; 5781541Srgrimes goto done; 5791541Srgrimes } 58091140Stanimura /* 58191140Stanimura * ktrops() may call vrele(). Lock pg_members 58294861Sjhb * by the proctree_lock rather than pg_mtx. 58391140Stanimura */ 58491140Stanimura PGRP_UNLOCK(pg); 58553212Sphk LIST_FOREACH(p, &pg->pg_members, p_pglist) 5861541Srgrimes if (descend) 58794618Sjhb ret |= ktrsetchildren(td, p, ops, facs, vp); 5888876Srgrimes else 58994618Sjhb ret |= ktrops(td, p, ops, facs, vp); 59094861Sjhb sx_sunlock(&proctree_lock); 5911541Srgrimes } else { 5921541Srgrimes /* 5931541Srgrimes * by pid 5941541Srgrimes */ 5951541Srgrimes p = pfind(uap->pid); 5961541Srgrimes if (p == NULL) { 5971541Srgrimes error = ESRCH; 5981541Srgrimes goto done; 5991541Srgrimes } 60075893Sjhb PROC_UNLOCK(p); 60194618Sjhb /* XXX: UNLOCK above has a race */ 6021541Srgrimes if (descend) 60394618Sjhb ret |= ktrsetchildren(td, p, ops, facs, vp); 6041541Srgrimes else 60594618Sjhb ret |= ktrops(td, p, ops, facs, vp); 6061541Srgrimes } 6071541Srgrimes if (!ret) 6081541Srgrimes error = EPERM; 6091541Srgrimesdone: 6101541Srgrimes if (vp != NULL) 61191406Sjhb (void) vn_close(vp, FWRITE, td->td_ucred, td); 61297993Sjhb td->td_inktrace = 0; 6131541Srgrimes return (error); 61413203Swollman#else 61513203Swollman return ENOSYS; 61613203Swollman#endif 6171541Srgrimes} 6181541Srgrimes 61918398Sphk/* 62018398Sphk * utrace system call 62118398Sphk */ 62218398Sphk/* ARGSUSED */ 62318398Sphkint 62483366Sjulianutrace(td, uap) 62583366Sjulian struct thread *td; 62618398Sphk register struct utrace_args *uap; 62718398Sphk{ 62883366Sjulian 62913203Swollman#ifdef KTRACE 63097993Sjhb struct ktr_request *req; 63199009Salfred void *cp; 632103237Sjhb int error; 63318398Sphk 634103237Sjhb if (!KTRPOINT(td, KTR_USER)) 635103237Sjhb return (0); 63670792Salfred if (uap->len > KTR_USER_MAXLEN) 63770707Salfred return (EINVAL); 638111119Simp cp = malloc(uap->len, M_KTRACE, M_WAITOK); 639103237Sjhb error = copyin(uap->addr, cp, uap->len); 640104230Sphk if (error) { 641104230Sphk free(cp, M_KTRACE); 642103237Sjhb return (error); 643104230Sphk } 64497993Sjhb req = ktr_getrequest(KTR_USER); 645104230Sphk if (req == NULL) { 646104230Sphk free(cp, M_KTRACE); 64797993Sjhb return (0); 648104230Sphk } 649103237Sjhb req->ktr_header.ktr_buffer = cp; 650103237Sjhb req->ktr_header.ktr_len = uap->len; 651103237Sjhb ktr_submitrequest(req); 65218398Sphk return (0); 65318398Sphk#else 65418398Sphk return (ENOSYS); 65518398Sphk#endif 65618398Sphk} 65718398Sphk 65818398Sphk#ifdef KTRACE 65912819Sphkstatic int 66094618Sjhbktrops(td, p, ops, facs, vp) 66194618Sjhb struct thread *td; 66294618Sjhb struct proc *p; 6631541Srgrimes int ops, facs; 6641541Srgrimes struct vnode *vp; 6651541Srgrimes{ 66697993Sjhb struct vnode *tracevp = NULL; 667112198Sjhb struct ucred *tracecred = NULL; 6681541Srgrimes 66994618Sjhb PROC_LOCK(p); 67094618Sjhb if (!ktrcanset(td, p)) { 67194618Sjhb PROC_UNLOCK(p); 6721541Srgrimes return (0); 67394618Sjhb } 67497993Sjhb mtx_lock(&ktrace_mtx); 6751541Srgrimes if (ops == KTROP_SET) { 676112198Sjhb if (p->p_tracevp != vp) { 6771541Srgrimes /* 67894618Sjhb * if trace file already in use, relinquish below 6791541Srgrimes */ 680112198Sjhb tracevp = p->p_tracevp; 68197993Sjhb VREF(vp); 682112198Sjhb p->p_tracevp = vp; 6831541Srgrimes } 684112198Sjhb if (p->p_tracecred != td->td_ucred) { 685112198Sjhb tracecred = p->p_tracecred; 686112198Sjhb p->p_tracecred = crhold(td->td_ucred); 687112198Sjhb } 6881541Srgrimes p->p_traceflag |= facs; 68994618Sjhb if (td->td_ucred->cr_uid == 0) 6901541Srgrimes p->p_traceflag |= KTRFAC_ROOT; 6918876Srgrimes } else { 6921541Srgrimes /* KTROP_CLEAR */ 6931541Srgrimes if (((p->p_traceflag &= ~facs) & KTRFAC_MASK) == 0) { 6941541Srgrimes /* no more tracing */ 6951541Srgrimes p->p_traceflag = 0; 696112198Sjhb tracevp = p->p_tracevp; 697112198Sjhb p->p_tracevp = NULL; 698112198Sjhb tracecred = p->p_tracecred; 699112198Sjhb p->p_tracecred = NULL; 7001541Srgrimes } 7011541Srgrimes } 70297993Sjhb mtx_unlock(&ktrace_mtx); 70394618Sjhb PROC_UNLOCK(p); 70497993Sjhb if (tracevp != NULL) 70597993Sjhb vrele(tracevp); 706112198Sjhb if (tracecred != NULL) 707112198Sjhb crfree(tracecred); 7081541Srgrimes 7091541Srgrimes return (1); 7101541Srgrimes} 7111541Srgrimes 71212819Sphkstatic int 71394618Sjhbktrsetchildren(td, top, ops, facs, vp) 71494618Sjhb struct thread *td; 71594618Sjhb struct proc *top; 7161541Srgrimes int ops, facs; 7171541Srgrimes struct vnode *vp; 7181541Srgrimes{ 7191541Srgrimes register struct proc *p; 7201541Srgrimes register int ret = 0; 7211541Srgrimes 7221541Srgrimes p = top; 72374927Sjhb sx_slock(&proctree_lock); 7241541Srgrimes for (;;) { 72594618Sjhb ret |= ktrops(td, p, ops, facs, vp); 7261541Srgrimes /* 7271541Srgrimes * If this process has children, descend to them next, 7281541Srgrimes * otherwise do any siblings, and if done with this level, 7291541Srgrimes * follow back up the tree (but not past top). 7301541Srgrimes */ 73153212Sphk if (!LIST_EMPTY(&p->p_children)) 73253212Sphk p = LIST_FIRST(&p->p_children); 7331541Srgrimes else for (;;) { 73470317Sjake if (p == top) { 73574927Sjhb sx_sunlock(&proctree_lock); 7361541Srgrimes return (ret); 73770317Sjake } 73853212Sphk if (LIST_NEXT(p, p_sibling)) { 73953212Sphk p = LIST_NEXT(p, p_sibling); 7401541Srgrimes break; 7411541Srgrimes } 74214529Shsu p = p->p_pptr; 7431541Srgrimes } 7441541Srgrimes } 7451541Srgrimes /*NOTREACHED*/ 7461541Srgrimes} 7471541Srgrimes 74812819Sphkstatic void 74997993Sjhbktr_writerequest(struct ktr_request *req) 75097993Sjhb{ 75197993Sjhb struct ktr_header *kth; 7521541Srgrimes struct vnode *vp; 75397993Sjhb struct proc *p; 75497993Sjhb struct thread *td; 75597993Sjhb struct ucred *cred; 7561541Srgrimes struct uio auio; 75797993Sjhb struct iovec aiov[3]; 75862976Smckusick struct mount *mp; 75997993Sjhb int datalen, buflen, vrele_count; 7601541Srgrimes int error; 7611541Srgrimes 76297993Sjhb vp = req->ktr_vp; 76397993Sjhb /* 76497993Sjhb * If vp is NULL, the vp has been cleared out from under this 76597993Sjhb * request, so just drop it. 76697993Sjhb */ 7671541Srgrimes if (vp == NULL) 7681541Srgrimes return; 76997993Sjhb kth = &req->ktr_header; 770112199Sjhb datalen = data_lengths[(ushort)kth->ktr_type & ~KTR_DROP]; 77197993Sjhb buflen = kth->ktr_len; 77297993Sjhb cred = req->ktr_cred; 77397993Sjhb td = curthread; 7741541Srgrimes auio.uio_iov = &aiov[0]; 7751541Srgrimes auio.uio_offset = 0; 7761541Srgrimes auio.uio_segflg = UIO_SYSSPACE; 7771541Srgrimes auio.uio_rw = UIO_WRITE; 7781541Srgrimes aiov[0].iov_base = (caddr_t)kth; 7791541Srgrimes aiov[0].iov_len = sizeof(struct ktr_header); 7801541Srgrimes auio.uio_resid = sizeof(struct ktr_header); 7811541Srgrimes auio.uio_iovcnt = 1; 78297993Sjhb auio.uio_td = td; 78397993Sjhb if (datalen != 0) { 78497993Sjhb aiov[1].iov_base = (caddr_t)&req->ktr_data; 78597993Sjhb aiov[1].iov_len = datalen; 78697993Sjhb auio.uio_resid += datalen; 7871541Srgrimes auio.uio_iovcnt++; 78897993Sjhb kth->ktr_len += datalen; 7891541Srgrimes } 79097993Sjhb if (buflen != 0) { 79197993Sjhb KASSERT(kth->ktr_buffer != NULL, ("ktrace: nothing to write")); 79297993Sjhb aiov[auio.uio_iovcnt].iov_base = kth->ktr_buffer; 79397993Sjhb aiov[auio.uio_iovcnt].iov_len = buflen; 79497993Sjhb auio.uio_resid += buflen; 79597993Sjhb auio.uio_iovcnt++; 796103235Sjhb } 79797993Sjhb mtx_lock(&Giant); 79862976Smckusick vn_start_write(vp, &mp, V_WAIT); 79983366Sjulian vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 80097993Sjhb (void)VOP_LEASE(vp, td, cred, LEASE_WRITE); 801101123Srwatson#ifdef MAC 802102129Srwatson error = mac_check_vnode_write(cred, NOCRED, vp); 803101123Srwatson if (error == 0) 804101123Srwatson#endif 805101123Srwatson error = VOP_WRITE(vp, &auio, IO_UNIT | IO_APPEND, cred); 80683366Sjulian VOP_UNLOCK(vp, 0, td); 80762976Smckusick vn_finished_write(mp); 80897993Sjhb mtx_unlock(&Giant); 8091541Srgrimes if (!error) 8101541Srgrimes return; 8111541Srgrimes /* 81297993Sjhb * If error encountered, give up tracing on this vnode. We defer 81397993Sjhb * all the vrele()'s on the vnode until after we are finished walking 81497993Sjhb * the various lists to avoid needlessly holding locks. 8151541Srgrimes */ 8161541Srgrimes log(LOG_NOTICE, "ktrace write failed, errno %d, tracing stopped\n", 8171541Srgrimes error); 81897993Sjhb vrele_count = 0; 81997993Sjhb /* 82097993Sjhb * First, clear this vnode from being used by any processes in the 82197993Sjhb * system. 82297993Sjhb * XXX - If one process gets an EPERM writing to the vnode, should 82397993Sjhb * we really do this? Other processes might have suitable 82497993Sjhb * credentials for the operation. 82597993Sjhb */ 826112198Sjhb cred = NULL; 82774927Sjhb sx_slock(&allproc_lock); 82853212Sphk LIST_FOREACH(p, &allproc, p_list) { 82997993Sjhb PROC_LOCK(p); 830112198Sjhb if (p->p_tracevp == vp) { 83197993Sjhb mtx_lock(&ktrace_mtx); 832112198Sjhb p->p_tracevp = NULL; 8331541Srgrimes p->p_traceflag = 0; 834112198Sjhb cred = p->p_tracecred; 835112198Sjhb p->p_tracecred = NULL; 83697993Sjhb mtx_unlock(&ktrace_mtx); 83797993Sjhb vrele_count++; 8381541Srgrimes } 83997993Sjhb PROC_UNLOCK(p); 840112198Sjhb if (cred != NULL) { 841112198Sjhb crfree(cred); 842112198Sjhb cred = NULL; 843112198Sjhb } 8441541Srgrimes } 84574927Sjhb sx_sunlock(&allproc_lock); 84697993Sjhb /* 84797993Sjhb * Second, clear this vnode from any pending requests. 84897993Sjhb */ 84997993Sjhb mtx_lock(&ktrace_mtx); 85097993Sjhb STAILQ_FOREACH(req, &ktr_todo, ktr_list) { 85197993Sjhb if (req->ktr_vp == vp) { 85297993Sjhb req->ktr_vp = NULL; 85397993Sjhb vrele_count++; 85497993Sjhb } 85597993Sjhb } 85697993Sjhb mtx_unlock(&ktrace_mtx); 85797993Sjhb mtx_lock(&Giant); 85897993Sjhb while (vrele_count-- > 0) 85997993Sjhb vrele(vp); 86097993Sjhb mtx_unlock(&Giant); 8611541Srgrimes} 8621541Srgrimes 8631541Srgrimes/* 8641541Srgrimes * Return true if caller has permission to set the ktracing state 8651541Srgrimes * of target. Essentially, the target can't possess any 8661541Srgrimes * more permissions than the caller. KTRFAC_ROOT signifies that 8678876Srgrimes * root previously set the tracing status on the target process, and 8681541Srgrimes * so, only root may further change it. 8691541Srgrimes */ 87012819Sphkstatic int 87194618Sjhbktrcanset(td, targetp) 87294618Sjhb struct thread *td; 87394618Sjhb struct proc *targetp; 8741541Srgrimes{ 8751541Srgrimes 87694618Sjhb PROC_LOCK_ASSERT(targetp, MA_OWNED); 87779335Srwatson if (targetp->p_traceflag & KTRFAC_ROOT && 87894618Sjhb suser_cred(td->td_ucred, PRISON_ROOT)) 87946155Sphk return (0); 8801541Srgrimes 88196886Sjhb if (p_candebug(td, targetp) != 0) 88279335Srwatson return (0); 88379335Srwatson 88479335Srwatson return (1); 8851541Srgrimes} 8861541Srgrimes 88713203Swollman#endif /* KTRACE */ 888