kern_ktrace.c revision 147576
1139804Simp/*- 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 * 4. Neither the name of the University nor the names of its contributors 141541Srgrimes * may be used to endorse or promote products derived from this software 151541Srgrimes * without specific prior written permission. 161541Srgrimes * 171541Srgrimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 181541Srgrimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 191541Srgrimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 201541Srgrimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 211541Srgrimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 221541Srgrimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 231541Srgrimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 241541Srgrimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 251541Srgrimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 261541Srgrimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 271541Srgrimes * SUCH DAMAGE. 281541Srgrimes * 291541Srgrimes * @(#)kern_ktrace.c 8.2 (Berkeley) 9/23/93 301541Srgrimes */ 311541Srgrimes 32116182Sobrien#include <sys/cdefs.h> 33116182Sobrien__FBSDID("$FreeBSD: head/sys/kern/kern_ktrace.c 147576 2005-06-24 12:05:24Z pjd $"); 34116182Sobrien 3513203Swollman#include "opt_ktrace.h" 36101123Srwatson#include "opt_mac.h" 371541Srgrimes 381541Srgrimes#include <sys/param.h> 392112Swollman#include <sys/systm.h> 4097993Sjhb#include <sys/fcntl.h> 4197993Sjhb#include <sys/kernel.h> 4297993Sjhb#include <sys/kthread.h> 4376166Smarkm#include <sys/lock.h> 4476166Smarkm#include <sys/mutex.h> 45101123Srwatson#include <sys/mac.h> 4697993Sjhb#include <sys/malloc.h> 4797993Sjhb#include <sys/namei.h> 481541Srgrimes#include <sys/proc.h> 4997993Sjhb#include <sys/unistd.h> 501541Srgrimes#include <sys/vnode.h> 511541Srgrimes#include <sys/ktrace.h> 5274927Sjhb#include <sys/sx.h> 5397993Sjhb#include <sys/sysctl.h> 541541Srgrimes#include <sys/syslog.h> 5597993Sjhb#include <sys/sysproto.h> 561541Srgrimes 5730354Sphkstatic MALLOC_DEFINE(M_KTRACE, "KTRACE", "KTRACE"); 5830309Sphk 5913203Swollman#ifdef KTRACE 6012577Sbde 6197993Sjhb#ifndef KTRACE_REQUEST_POOL 6297993Sjhb#define KTRACE_REQUEST_POOL 100 6397993Sjhb#endif 6412819Sphk 6597993Sjhbstruct ktr_request { 6697993Sjhb struct ktr_header ktr_header; 6797993Sjhb struct ucred *ktr_cred; 6897993Sjhb struct vnode *ktr_vp; 6997993Sjhb union { 7097993Sjhb struct ktr_syscall ktr_syscall; 7197993Sjhb struct ktr_sysret ktr_sysret; 7297993Sjhb struct ktr_genio ktr_genio; 7397993Sjhb struct ktr_psig ktr_psig; 7497993Sjhb struct ktr_csw ktr_csw; 7597993Sjhb } ktr_data; 7697993Sjhb STAILQ_ENTRY(ktr_request) ktr_list; 7797993Sjhb}; 7897993Sjhb 7997993Sjhbstatic int data_lengths[] = { 8097993Sjhb 0, /* none */ 8197993Sjhb offsetof(struct ktr_syscall, ktr_args), /* KTR_SYSCALL */ 8297993Sjhb sizeof(struct ktr_sysret), /* KTR_SYSRET */ 8397993Sjhb 0, /* KTR_NAMEI */ 8497993Sjhb sizeof(struct ktr_genio), /* KTR_GENIO */ 8597993Sjhb sizeof(struct ktr_psig), /* KTR_PSIG */ 8697993Sjhb sizeof(struct ktr_csw), /* KTR_CSW */ 8797993Sjhb 0 /* KTR_USER */ 8897993Sjhb}; 8997993Sjhb 9097993Sjhbstatic STAILQ_HEAD(, ktr_request) ktr_todo; 9197993Sjhbstatic STAILQ_HEAD(, ktr_request) ktr_free; 9297993Sjhb 93141633Sphkstatic SYSCTL_NODE(_kern, OID_AUTO, ktrace, CTLFLAG_RD, 0, "KTRACE options"); 94103234Sjhb 95118607Sjhbstatic u_int ktr_requestpool = KTRACE_REQUEST_POOL; 96103234SjhbTUNABLE_INT("kern.ktrace.request_pool", &ktr_requestpool); 9797993Sjhb 98118607Sjhbstatic u_int ktr_geniosize = PAGE_SIZE; 99103234SjhbTUNABLE_INT("kern.ktrace.genio_size", &ktr_geniosize); 100103234SjhbSYSCTL_UINT(_kern_ktrace, OID_AUTO, genio_size, CTLFLAG_RW, &ktr_geniosize, 101103234Sjhb 0, "Maximum size of genio event payload"); 102103234Sjhb 10397993Sjhbstatic int print_message = 1; 10497993Sjhbstruct mtx ktrace_mtx; 105126295Sjhbstatic struct cv ktrace_cv; 10697993Sjhb 10797993Sjhbstatic void ktrace_init(void *dummy); 10897993Sjhbstatic int sysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS); 109118607Sjhbstatic u_int ktrace_resize_pool(u_int newsize); 11097993Sjhbstatic struct ktr_request *ktr_getrequest(int type); 11197993Sjhbstatic void ktr_submitrequest(struct ktr_request *req); 11297993Sjhbstatic void ktr_freerequest(struct ktr_request *req); 11397993Sjhbstatic void ktr_loop(void *dummy); 11497993Sjhbstatic void ktr_writerequest(struct ktr_request *req); 11597993Sjhbstatic int ktrcanset(struct thread *,struct proc *); 11697993Sjhbstatic int ktrsetchildren(struct thread *,struct proc *,int,int,struct vnode *); 11797993Sjhbstatic int ktrops(struct thread *,struct proc *,int,int,struct vnode *); 11897993Sjhb 11997993Sjhbstatic void 12097993Sjhbktrace_init(void *dummy) 1211541Srgrimes{ 12297993Sjhb struct ktr_request *req; 12397993Sjhb int i; 1241541Srgrimes 12597993Sjhb mtx_init(&ktrace_mtx, "ktrace", NULL, MTX_DEF | MTX_QUIET); 126126295Sjhb cv_init(&ktrace_cv, "ktrace"); 12797993Sjhb STAILQ_INIT(&ktr_todo); 12897993Sjhb STAILQ_INIT(&ktr_free); 12997993Sjhb for (i = 0; i < ktr_requestpool; i++) { 130111119Simp req = malloc(sizeof(struct ktr_request), M_KTRACE, M_WAITOK); 13197993Sjhb STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list); 13297993Sjhb } 133104354Sscottl kthread_create(ktr_loop, NULL, NULL, RFHIGHPID, 0, "ktrace"); 1341541Srgrimes} 13597993SjhbSYSINIT(ktrace_init, SI_SUB_KTRACE, SI_ORDER_ANY, ktrace_init, NULL); 1361541Srgrimes 13797993Sjhbstatic int 13897993Sjhbsysctl_kern_ktrace_request_pool(SYSCTL_HANDLER_ARGS) 13997993Sjhb{ 14097993Sjhb struct thread *td; 141118607Sjhb u_int newsize, oldsize, wantsize; 14297993Sjhb int error; 14397993Sjhb 14497993Sjhb /* Handle easy read-only case first to avoid warnings from GCC. */ 14597993Sjhb if (!req->newptr) { 14697993Sjhb mtx_lock(&ktrace_mtx); 14797993Sjhb oldsize = ktr_requestpool; 14897993Sjhb mtx_unlock(&ktrace_mtx); 149118607Sjhb return (SYSCTL_OUT(req, &oldsize, sizeof(u_int))); 15097993Sjhb } 15197993Sjhb 152118607Sjhb error = SYSCTL_IN(req, &wantsize, sizeof(u_int)); 15397993Sjhb if (error) 15497993Sjhb return (error); 15597993Sjhb td = curthread; 156116101Sjhb td->td_pflags |= TDP_INKTRACE; 15797993Sjhb mtx_lock(&ktrace_mtx); 15897993Sjhb oldsize = ktr_requestpool; 15997993Sjhb newsize = ktrace_resize_pool(wantsize); 16097993Sjhb mtx_unlock(&ktrace_mtx); 161116101Sjhb td->td_pflags &= ~TDP_INKTRACE; 162118607Sjhb error = SYSCTL_OUT(req, &oldsize, sizeof(u_int)); 16397993Sjhb if (error) 16497993Sjhb return (error); 165122478Sjkoshy if (wantsize > oldsize && newsize < wantsize) 16697993Sjhb return (ENOSPC); 16797993Sjhb return (0); 16897993Sjhb} 169103234SjhbSYSCTL_PROC(_kern_ktrace, OID_AUTO, request_pool, CTLTYPE_UINT|CTLFLAG_RW, 17097993Sjhb &ktr_requestpool, 0, sysctl_kern_ktrace_request_pool, "IU", ""); 17197993Sjhb 172118607Sjhbstatic u_int 173118607Sjhbktrace_resize_pool(u_int newsize) 17497993Sjhb{ 17597993Sjhb struct ktr_request *req; 176122478Sjkoshy int bound; 17797993Sjhb 17897993Sjhb mtx_assert(&ktrace_mtx, MA_OWNED); 17997993Sjhb print_message = 1; 180122478Sjkoshy bound = newsize - ktr_requestpool; 181122478Sjkoshy if (bound == 0) 182122478Sjkoshy return (ktr_requestpool); 183122478Sjkoshy if (bound < 0) 18497993Sjhb /* Shrink pool down to newsize if possible. */ 185122478Sjkoshy while (bound++ < 0) { 18697993Sjhb req = STAILQ_FIRST(&ktr_free); 18797993Sjhb if (req == NULL) 18897993Sjhb return (ktr_requestpool); 18997993Sjhb STAILQ_REMOVE_HEAD(&ktr_free, ktr_list); 19097993Sjhb ktr_requestpool--; 19197993Sjhb mtx_unlock(&ktrace_mtx); 19297993Sjhb free(req, M_KTRACE); 19397993Sjhb mtx_lock(&ktrace_mtx); 19497993Sjhb } 19597993Sjhb else 19697993Sjhb /* Grow pool up to newsize. */ 197122478Sjkoshy while (bound-- > 0) { 19897993Sjhb mtx_unlock(&ktrace_mtx); 19997993Sjhb req = malloc(sizeof(struct ktr_request), M_KTRACE, 200111119Simp M_WAITOK); 20197993Sjhb mtx_lock(&ktrace_mtx); 20297993Sjhb STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list); 20397993Sjhb ktr_requestpool++; 20497993Sjhb } 20597993Sjhb return (ktr_requestpool); 20697993Sjhb} 20797993Sjhb 20897993Sjhbstatic struct ktr_request * 20997993Sjhbktr_getrequest(int type) 21097993Sjhb{ 21197993Sjhb struct ktr_request *req; 21297993Sjhb struct thread *td = curthread; 21397993Sjhb struct proc *p = td->td_proc; 21497993Sjhb int pm; 21597993Sjhb 216116101Sjhb td->td_pflags |= TDP_INKTRACE; 21797993Sjhb mtx_lock(&ktrace_mtx); 21897993Sjhb if (!KTRCHECK(td, type)) { 21997993Sjhb mtx_unlock(&ktrace_mtx); 220116101Sjhb td->td_pflags &= ~TDP_INKTRACE; 22197993Sjhb return (NULL); 22297993Sjhb } 22397993Sjhb req = STAILQ_FIRST(&ktr_free); 22497993Sjhb if (req != NULL) { 22597993Sjhb STAILQ_REMOVE_HEAD(&ktr_free, ktr_list); 22697993Sjhb req->ktr_header.ktr_type = type; 227112199Sjhb if (p->p_traceflag & KTRFAC_DROP) { 228112199Sjhb req->ktr_header.ktr_type |= KTR_DROP; 229112199Sjhb p->p_traceflag &= ~KTRFAC_DROP; 230112199Sjhb } 231112198Sjhb KASSERT(p->p_tracevp != NULL, ("ktrace: no trace vnode")); 232112198Sjhb KASSERT(p->p_tracecred != NULL, ("ktrace: no trace cred")); 233112198Sjhb req->ktr_vp = p->p_tracevp; 234112198Sjhb VREF(p->p_tracevp); 235112198Sjhb req->ktr_cred = crhold(p->p_tracecred); 23697993Sjhb mtx_unlock(&ktrace_mtx); 23797993Sjhb microtime(&req->ktr_header.ktr_time); 23897993Sjhb req->ktr_header.ktr_pid = p->p_pid; 23997993Sjhb bcopy(p->p_comm, req->ktr_header.ktr_comm, MAXCOMLEN + 1); 24097993Sjhb req->ktr_header.ktr_buffer = NULL; 24197993Sjhb req->ktr_header.ktr_len = 0; 24297993Sjhb } else { 243112199Sjhb p->p_traceflag |= KTRFAC_DROP; 24497993Sjhb pm = print_message; 24597993Sjhb print_message = 0; 24697993Sjhb mtx_unlock(&ktrace_mtx); 24797993Sjhb if (pm) 24897993Sjhb printf("Out of ktrace request objects.\n"); 249116101Sjhb td->td_pflags &= ~TDP_INKTRACE; 25097993Sjhb } 25197993Sjhb return (req); 25297993Sjhb} 25397993Sjhb 25497993Sjhbstatic void 25597993Sjhbktr_submitrequest(struct ktr_request *req) 25697993Sjhb{ 25797993Sjhb 25897993Sjhb mtx_lock(&ktrace_mtx); 25997993Sjhb STAILQ_INSERT_TAIL(&ktr_todo, req, ktr_list); 260126295Sjhb cv_signal(&ktrace_cv); 261118599Sjhb mtx_unlock(&ktrace_mtx); 262116101Sjhb curthread->td_pflags &= ~TDP_INKTRACE; 26397993Sjhb} 26497993Sjhb 26597993Sjhbstatic void 26697993Sjhbktr_freerequest(struct ktr_request *req) 26797993Sjhb{ 26897993Sjhb 26997993Sjhb crfree(req->ktr_cred); 270101153Sjhb if (req->ktr_vp != NULL) { 271101153Sjhb mtx_lock(&Giant); 272101153Sjhb vrele(req->ktr_vp); 273101153Sjhb mtx_unlock(&Giant); 274101153Sjhb } 275103235Sjhb if (req->ktr_header.ktr_buffer != NULL) 276103235Sjhb free(req->ktr_header.ktr_buffer, M_KTRACE); 27797993Sjhb mtx_lock(&ktrace_mtx); 27897993Sjhb STAILQ_INSERT_HEAD(&ktr_free, req, ktr_list); 27997993Sjhb mtx_unlock(&ktrace_mtx); 28097993Sjhb} 28197993Sjhb 28297993Sjhbstatic void 28397993Sjhbktr_loop(void *dummy) 28497993Sjhb{ 28597993Sjhb struct ktr_request *req; 28697993Sjhb struct thread *td; 28797993Sjhb struct ucred *cred; 28897993Sjhb 28997993Sjhb /* Only cache these values once. */ 29097993Sjhb td = curthread; 29197993Sjhb cred = td->td_ucred; 29297993Sjhb for (;;) { 29397993Sjhb mtx_lock(&ktrace_mtx); 294126295Sjhb while (STAILQ_EMPTY(&ktr_todo)) 295126295Sjhb cv_wait(&ktrace_cv, &ktrace_mtx); 29697993Sjhb req = STAILQ_FIRST(&ktr_todo); 29797993Sjhb STAILQ_REMOVE_HEAD(&ktr_todo, ktr_list); 29897993Sjhb KASSERT(req != NULL, ("got a NULL request")); 299103236Sjhb mtx_unlock(&ktrace_mtx); 300103236Sjhb /* 301103236Sjhb * It is not enough just to pass the cached cred 302103236Sjhb * to the VOP's in ktr_writerequest(). Some VFS 303103236Sjhb * operations use curthread->td_ucred, so we need 304103236Sjhb * to modify our thread's credentials as well. 305103236Sjhb * Evil. 306103236Sjhb */ 307103236Sjhb td->td_ucred = req->ktr_cred; 308103236Sjhb ktr_writerequest(req); 309103236Sjhb td->td_ucred = cred; 31097993Sjhb ktr_freerequest(req); 31197993Sjhb } 31297993Sjhb} 31397993Sjhb 31482585Sdillon/* 31582585Sdillon * MPSAFE 31682585Sdillon */ 3171549Srgrimesvoid 31897993Sjhbktrsyscall(code, narg, args) 31947955Sdt int code, narg; 32047955Sdt register_t args[]; 3211541Srgrimes{ 32297993Sjhb struct ktr_request *req; 32397993Sjhb struct ktr_syscall *ktp; 32497993Sjhb size_t buflen; 325103233Sjhb char *buf = NULL; 3261541Srgrimes 327103233Sjhb buflen = sizeof(register_t) * narg; 328103233Sjhb if (buflen > 0) { 329111119Simp buf = malloc(buflen, M_KTRACE, M_WAITOK); 330103233Sjhb bcopy(args, buf, buflen); 331103233Sjhb } 33297993Sjhb req = ktr_getrequest(KTR_SYSCALL); 333104230Sphk if (req == NULL) { 334104230Sphk if (buf != NULL) 335104230Sphk free(buf, M_KTRACE); 33697993Sjhb return; 337104230Sphk } 33897993Sjhb ktp = &req->ktr_data.ktr_syscall; 3391541Srgrimes ktp->ktr_code = code; 3401541Srgrimes ktp->ktr_narg = narg; 34197993Sjhb if (buflen > 0) { 34297993Sjhb req->ktr_header.ktr_len = buflen; 343103233Sjhb req->ktr_header.ktr_buffer = buf; 34497993Sjhb } 34597993Sjhb ktr_submitrequest(req); 3461541Srgrimes} 3471541Srgrimes 34882585Sdillon/* 34982585Sdillon * MPSAFE 35082585Sdillon */ 3511549Srgrimesvoid 35297993Sjhbktrsysret(code, error, retval) 35347955Sdt int code, error; 35447955Sdt register_t retval; 3551541Srgrimes{ 35697993Sjhb struct ktr_request *req; 35797993Sjhb struct ktr_sysret *ktp; 3581541Srgrimes 35997993Sjhb req = ktr_getrequest(KTR_SYSRET); 36097993Sjhb if (req == NULL) 36197993Sjhb return; 36297993Sjhb ktp = &req->ktr_data.ktr_sysret; 36397993Sjhb ktp->ktr_code = code; 36497993Sjhb ktp->ktr_error = error; 36597993Sjhb ktp->ktr_retval = retval; /* what about val2 ? */ 36697993Sjhb ktr_submitrequest(req); 3671541Srgrimes} 3681541Srgrimes 3691549Srgrimesvoid 37097993Sjhbktrnamei(path) 3711541Srgrimes char *path; 3721541Srgrimes{ 37397993Sjhb struct ktr_request *req; 37497993Sjhb int namelen; 375103233Sjhb char *buf = NULL; 3761541Srgrimes 377103233Sjhb namelen = strlen(path); 378103233Sjhb if (namelen > 0) { 379111119Simp buf = malloc(namelen, M_KTRACE, M_WAITOK); 380103233Sjhb bcopy(path, buf, namelen); 381103233Sjhb } 38297993Sjhb req = ktr_getrequest(KTR_NAMEI); 383104230Sphk if (req == NULL) { 384104230Sphk if (buf != NULL) 385104230Sphk free(buf, M_KTRACE); 38697993Sjhb return; 387104230Sphk } 38897993Sjhb if (namelen > 0) { 38997993Sjhb req->ktr_header.ktr_len = namelen; 390103233Sjhb req->ktr_header.ktr_buffer = buf; 39197993Sjhb } 39297993Sjhb ktr_submitrequest(req); 3931541Srgrimes} 3941541Srgrimes 39597993Sjhb/* 39697993Sjhb * Since the uio may not stay valid, we can not hand off this request to 39797993Sjhb * the thread and need to process it synchronously. However, we wish to 39897993Sjhb * keep the relative order of records in a trace file correct, so we 39997993Sjhb * do put this request on the queue (if it isn't empty) and then block. 40097993Sjhb * The ktrace thread waks us back up when it is time for this event to 40197993Sjhb * be posted and blocks until we have completed writing out the event 40297993Sjhb * and woken it back up. 40397993Sjhb */ 4041549Srgrimesvoid 40597993Sjhbktrgenio(fd, rw, uio, error) 4061541Srgrimes int fd; 4071541Srgrimes enum uio_rw rw; 40862378Sgreen struct uio *uio; 40962378Sgreen int error; 4101541Srgrimes{ 41197993Sjhb struct ktr_request *req; 41297993Sjhb struct ktr_genio *ktg; 413103235Sjhb int datalen; 414103235Sjhb char *buf; 4158876Srgrimes 416131897Sphk if (error) { 417131897Sphk free(uio, M_IOV); 4181541Srgrimes return; 419131897Sphk } 420103235Sjhb uio->uio_offset = 0; 421103235Sjhb uio->uio_rw = UIO_WRITE; 422103235Sjhb datalen = imin(uio->uio_resid, ktr_geniosize); 423111119Simp buf = malloc(datalen, M_KTRACE, M_WAITOK); 424131897Sphk error = uiomove(buf, datalen, uio); 425131897Sphk free(uio, M_IOV); 426131897Sphk if (error) { 427103235Sjhb free(buf, M_KTRACE); 428103235Sjhb return; 429103235Sjhb } 43097993Sjhb req = ktr_getrequest(KTR_GENIO); 431103235Sjhb if (req == NULL) { 432103235Sjhb free(buf, M_KTRACE); 43397993Sjhb return; 434103235Sjhb } 43597993Sjhb ktg = &req->ktr_data.ktr_genio; 43697993Sjhb ktg->ktr_fd = fd; 43797993Sjhb ktg->ktr_rw = rw; 438103235Sjhb req->ktr_header.ktr_len = datalen; 439103235Sjhb req->ktr_header.ktr_buffer = buf; 44097993Sjhb ktr_submitrequest(req); 4411541Srgrimes} 4421541Srgrimes 4431549Srgrimesvoid 44497993Sjhbktrpsig(sig, action, mask, code) 44551941Smarcel int sig; 4461541Srgrimes sig_t action; 44751791Smarcel sigset_t *mask; 44851941Smarcel int code; 4491541Srgrimes{ 45097993Sjhb struct ktr_request *req; 45197993Sjhb struct ktr_psig *kp; 4521541Srgrimes 45397993Sjhb req = ktr_getrequest(KTR_PSIG); 45497993Sjhb if (req == NULL) 45597993Sjhb return; 45697993Sjhb kp = &req->ktr_data.ktr_psig; 45797993Sjhb kp->signo = (char)sig; 45897993Sjhb kp->action = action; 45997993Sjhb kp->mask = *mask; 46097993Sjhb kp->code = code; 46197993Sjhb ktr_submitrequest(req); 4621541Srgrimes} 4631541Srgrimes 4641549Srgrimesvoid 46597993Sjhbktrcsw(out, user) 4661541Srgrimes int out, user; 4671541Srgrimes{ 46897993Sjhb struct ktr_request *req; 46997993Sjhb struct ktr_csw *kc; 4701541Srgrimes 47197993Sjhb req = ktr_getrequest(KTR_CSW); 47297993Sjhb if (req == NULL) 47397993Sjhb return; 47497993Sjhb kc = &req->ktr_data.ktr_csw; 47597993Sjhb kc->out = out; 47697993Sjhb kc->user = user; 47797993Sjhb ktr_submitrequest(req); 4781541Srgrimes} 479114026Sjhb#endif /* KTRACE */ 4801541Srgrimes 4811541Srgrimes/* Interface and common routines */ 4821541Srgrimes 4831541Srgrimes/* 4841541Srgrimes * ktrace system call 485114026Sjhb * 486114026Sjhb * MPSAFE 4871541Srgrimes */ 48812221Sbde#ifndef _SYS_SYSPROTO_H_ 4891541Srgrimesstruct ktrace_args { 4901541Srgrimes char *fname; 4911541Srgrimes int ops; 4921541Srgrimes int facs; 4931541Srgrimes int pid; 4941541Srgrimes}; 49512221Sbde#endif 4961541Srgrimes/* ARGSUSED */ 4971549Srgrimesint 49883366Sjulianktrace(td, uap) 49983366Sjulian struct thread *td; 5001541Srgrimes register struct ktrace_args *uap; 5011541Srgrimes{ 50213203Swollman#ifdef KTRACE 5031541Srgrimes register struct vnode *vp = NULL; 5041541Srgrimes register struct proc *p; 5051541Srgrimes struct pgrp *pg; 5061541Srgrimes int facs = uap->facs & ~KTRFAC_ROOT; 5071541Srgrimes int ops = KTROP(uap->ops); 5081541Srgrimes int descend = uap->ops & KTRFLAG_DESCEND; 509147576Spjd int nfound, ret = 0; 51062550Smckusick int flags, error = 0; 5111541Srgrimes struct nameidata nd; 512112198Sjhb struct ucred *cred; 5131541Srgrimes 514114026Sjhb /* 515114026Sjhb * Need something to (un)trace. 516114026Sjhb */ 517114026Sjhb if (ops != KTROP_CLEARFILE && facs == 0) 518114026Sjhb return (EINVAL); 519114026Sjhb 520116101Sjhb td->td_pflags |= TDP_INKTRACE; 5211541Srgrimes if (ops != KTROP_CLEAR) { 5221541Srgrimes /* 5231541Srgrimes * an operation which requires a file argument. 5241541Srgrimes */ 52583366Sjulian NDINIT(&nd, LOOKUP, NOFOLLOW, UIO_USERSPACE, uap->fname, td); 52662550Smckusick flags = FREAD | FWRITE | O_NOFOLLOW; 527114026Sjhb mtx_lock(&Giant); 528118094Sphk error = vn_open(&nd, &flags, 0, -1); 5293308Sphk if (error) { 530114026Sjhb mtx_unlock(&Giant); 531116101Sjhb td->td_pflags &= ~TDP_INKTRACE; 5321541Srgrimes return (error); 5331541Srgrimes } 53454655Seivind NDFREE(&nd, NDF_ONLY_PNBUF); 5351541Srgrimes vp = nd.ni_vp; 53683366Sjulian VOP_UNLOCK(vp, 0, td); 5371541Srgrimes if (vp->v_type != VREG) { 53891406Sjhb (void) vn_close(vp, FREAD|FWRITE, td->td_ucred, td); 539114026Sjhb mtx_unlock(&Giant); 540116101Sjhb td->td_pflags &= ~TDP_INKTRACE; 5411541Srgrimes return (EACCES); 5421541Srgrimes } 543114026Sjhb mtx_unlock(&Giant); 5441541Srgrimes } 5451541Srgrimes /* 54685397Sdillon * Clear all uses of the tracefile. 5471541Srgrimes */ 5481541Srgrimes if (ops == KTROP_CLEARFILE) { 54974927Sjhb sx_slock(&allproc_lock); 55053212Sphk LIST_FOREACH(p, &allproc, p_list) { 55194618Sjhb PROC_LOCK(p); 552112198Sjhb if (p->p_tracevp == vp) { 55397993Sjhb if (ktrcanset(td, p)) { 55497993Sjhb mtx_lock(&ktrace_mtx); 555112198Sjhb cred = p->p_tracecred; 556112198Sjhb p->p_tracecred = NULL; 557112198Sjhb p->p_tracevp = NULL; 5581541Srgrimes p->p_traceflag = 0; 55997993Sjhb mtx_unlock(&ktrace_mtx); 56094618Sjhb PROC_UNLOCK(p); 561114026Sjhb mtx_lock(&Giant); 5621541Srgrimes (void) vn_close(vp, FREAD|FWRITE, 563112198Sjhb cred, td); 564114026Sjhb mtx_unlock(&Giant); 565112198Sjhb crfree(cred); 56685397Sdillon } else { 56794618Sjhb PROC_UNLOCK(p); 5681541Srgrimes error = EPERM; 56985397Sdillon } 57094618Sjhb } else 57194618Sjhb PROC_UNLOCK(p); 5721541Srgrimes } 57374927Sjhb sx_sunlock(&allproc_lock); 5741541Srgrimes goto done; 5751541Srgrimes } 5761541Srgrimes /* 5771541Srgrimes * do it 5781541Srgrimes */ 579114026Sjhb sx_slock(&proctree_lock); 5801541Srgrimes if (uap->pid < 0) { 5811541Srgrimes /* 5821541Srgrimes * by process group 5831541Srgrimes */ 5841541Srgrimes pg = pgfind(-uap->pid); 5851541Srgrimes if (pg == NULL) { 58694861Sjhb sx_sunlock(&proctree_lock); 5871541Srgrimes error = ESRCH; 5881541Srgrimes goto done; 5891541Srgrimes } 59091140Stanimura /* 59191140Stanimura * ktrops() may call vrele(). Lock pg_members 59294861Sjhb * by the proctree_lock rather than pg_mtx. 59391140Stanimura */ 59491140Stanimura PGRP_UNLOCK(pg); 595147576Spjd nfound = 0; 596147576Spjd LIST_FOREACH(p, &pg->pg_members, p_pglist) { 597147576Spjd PROC_LOCK(p); 598147576Spjd if (p_cansee(td, p) != 0) { 599147576Spjd PROC_UNLOCK(p); 600147576Spjd continue; 601147576Spjd } 602147576Spjd PROC_UNLOCK(p); 603147576Spjd nfound++; 6041541Srgrimes if (descend) 60594618Sjhb ret |= ktrsetchildren(td, p, ops, facs, vp); 6068876Srgrimes else 60794618Sjhb ret |= ktrops(td, p, ops, facs, vp); 608147576Spjd } 609147576Spjd if (nfound == 0) { 610147576Spjd sx_sunlock(&proctree_lock); 611147576Spjd error = ESRCH; 612147576Spjd goto done; 613147576Spjd } 6141541Srgrimes } else { 6151541Srgrimes /* 6161541Srgrimes * by pid 6171541Srgrimes */ 6181541Srgrimes p = pfind(uap->pid); 6191541Srgrimes if (p == NULL) { 620114026Sjhb sx_sunlock(&proctree_lock); 6211541Srgrimes error = ESRCH; 6221541Srgrimes goto done; 6231541Srgrimes } 624147183Spjd error = p_cansee(td, p); 625114026Sjhb /* 626114026Sjhb * The slock of the proctree lock will keep this process 627114026Sjhb * from going away, so unlocking the proc here is ok. 628114026Sjhb */ 62975893Sjhb PROC_UNLOCK(p); 630147520Spjd if (error) { 631147520Spjd sx_sunlock(&proctree_lock); 632147183Spjd goto done; 633147520Spjd } 6341541Srgrimes if (descend) 63594618Sjhb ret |= ktrsetchildren(td, p, ops, facs, vp); 6361541Srgrimes else 63794618Sjhb ret |= ktrops(td, p, ops, facs, vp); 6381541Srgrimes } 639114026Sjhb sx_sunlock(&proctree_lock); 6401541Srgrimes if (!ret) 6411541Srgrimes error = EPERM; 6421541Srgrimesdone: 643114026Sjhb if (vp != NULL) { 644114026Sjhb mtx_lock(&Giant); 64591406Sjhb (void) vn_close(vp, FWRITE, td->td_ucred, td); 646114026Sjhb mtx_unlock(&Giant); 647114026Sjhb } 648116101Sjhb td->td_pflags &= ~TDP_INKTRACE; 6491541Srgrimes return (error); 650114026Sjhb#else /* !KTRACE */ 651114026Sjhb return (ENOSYS); 652114026Sjhb#endif /* KTRACE */ 6531541Srgrimes} 6541541Srgrimes 65518398Sphk/* 65618398Sphk * utrace system call 657114026Sjhb * 658114026Sjhb * MPSAFE 65918398Sphk */ 66018398Sphk/* ARGSUSED */ 66118398Sphkint 66283366Sjulianutrace(td, uap) 66383366Sjulian struct thread *td; 66418398Sphk register struct utrace_args *uap; 66518398Sphk{ 66683366Sjulian 66713203Swollman#ifdef KTRACE 66897993Sjhb struct ktr_request *req; 66999009Salfred void *cp; 670103237Sjhb int error; 67118398Sphk 672103237Sjhb if (!KTRPOINT(td, KTR_USER)) 673103237Sjhb return (0); 67470792Salfred if (uap->len > KTR_USER_MAXLEN) 67570707Salfred return (EINVAL); 676111119Simp cp = malloc(uap->len, M_KTRACE, M_WAITOK); 677103237Sjhb error = copyin(uap->addr, cp, uap->len); 678104230Sphk if (error) { 679104230Sphk free(cp, M_KTRACE); 680103237Sjhb return (error); 681104230Sphk } 68297993Sjhb req = ktr_getrequest(KTR_USER); 683104230Sphk if (req == NULL) { 684104230Sphk free(cp, M_KTRACE); 685122457Sjkoshy return (ENOMEM); 686104230Sphk } 687103237Sjhb req->ktr_header.ktr_buffer = cp; 688103237Sjhb req->ktr_header.ktr_len = uap->len; 689103237Sjhb ktr_submitrequest(req); 69018398Sphk return (0); 691114026Sjhb#else /* !KTRACE */ 69218398Sphk return (ENOSYS); 693114026Sjhb#endif /* KTRACE */ 69418398Sphk} 69518398Sphk 69618398Sphk#ifdef KTRACE 69712819Sphkstatic int 69894618Sjhbktrops(td, p, ops, facs, vp) 69994618Sjhb struct thread *td; 70094618Sjhb struct proc *p; 7011541Srgrimes int ops, facs; 7021541Srgrimes struct vnode *vp; 7031541Srgrimes{ 70497993Sjhb struct vnode *tracevp = NULL; 705112198Sjhb struct ucred *tracecred = NULL; 7061541Srgrimes 70794618Sjhb PROC_LOCK(p); 70894618Sjhb if (!ktrcanset(td, p)) { 70994618Sjhb PROC_UNLOCK(p); 7101541Srgrimes return (0); 71194618Sjhb } 71297993Sjhb mtx_lock(&ktrace_mtx); 7131541Srgrimes if (ops == KTROP_SET) { 714112198Sjhb if (p->p_tracevp != vp) { 7151541Srgrimes /* 71694618Sjhb * if trace file already in use, relinquish below 7171541Srgrimes */ 718112198Sjhb tracevp = p->p_tracevp; 71997993Sjhb VREF(vp); 720112198Sjhb p->p_tracevp = vp; 7211541Srgrimes } 722112198Sjhb if (p->p_tracecred != td->td_ucred) { 723112198Sjhb tracecred = p->p_tracecred; 724112198Sjhb p->p_tracecred = crhold(td->td_ucred); 725112198Sjhb } 7261541Srgrimes p->p_traceflag |= facs; 72794618Sjhb if (td->td_ucred->cr_uid == 0) 7281541Srgrimes p->p_traceflag |= KTRFAC_ROOT; 7298876Srgrimes } else { 7301541Srgrimes /* KTROP_CLEAR */ 7311541Srgrimes if (((p->p_traceflag &= ~facs) & KTRFAC_MASK) == 0) { 7321541Srgrimes /* no more tracing */ 7331541Srgrimes p->p_traceflag = 0; 734112198Sjhb tracevp = p->p_tracevp; 735112198Sjhb p->p_tracevp = NULL; 736112198Sjhb tracecred = p->p_tracecred; 737112198Sjhb p->p_tracecred = NULL; 7381541Srgrimes } 7391541Srgrimes } 74097993Sjhb mtx_unlock(&ktrace_mtx); 74194618Sjhb PROC_UNLOCK(p); 742114026Sjhb if (tracevp != NULL) { 743114026Sjhb mtx_lock(&Giant); 74497993Sjhb vrele(tracevp); 745114026Sjhb mtx_unlock(&Giant); 746114026Sjhb } 747112198Sjhb if (tracecred != NULL) 748112198Sjhb crfree(tracecred); 7491541Srgrimes 7501541Srgrimes return (1); 7511541Srgrimes} 7521541Srgrimes 75312819Sphkstatic int 75494618Sjhbktrsetchildren(td, top, ops, facs, vp) 75594618Sjhb struct thread *td; 75694618Sjhb struct proc *top; 7571541Srgrimes int ops, facs; 7581541Srgrimes struct vnode *vp; 7591541Srgrimes{ 7601541Srgrimes register struct proc *p; 7611541Srgrimes register int ret = 0; 7621541Srgrimes 7631541Srgrimes p = top; 764114026Sjhb sx_assert(&proctree_lock, SX_LOCKED); 7651541Srgrimes for (;;) { 76694618Sjhb ret |= ktrops(td, p, ops, facs, vp); 7671541Srgrimes /* 7681541Srgrimes * If this process has children, descend to them next, 7691541Srgrimes * otherwise do any siblings, and if done with this level, 7701541Srgrimes * follow back up the tree (but not past top). 7711541Srgrimes */ 77253212Sphk if (!LIST_EMPTY(&p->p_children)) 77353212Sphk p = LIST_FIRST(&p->p_children); 7741541Srgrimes else for (;;) { 775114026Sjhb if (p == top) 7761541Srgrimes return (ret); 77753212Sphk if (LIST_NEXT(p, p_sibling)) { 77853212Sphk p = LIST_NEXT(p, p_sibling); 7791541Srgrimes break; 7801541Srgrimes } 78114529Shsu p = p->p_pptr; 7821541Srgrimes } 7831541Srgrimes } 7841541Srgrimes /*NOTREACHED*/ 7851541Srgrimes} 7861541Srgrimes 78712819Sphkstatic void 78897993Sjhbktr_writerequest(struct ktr_request *req) 78997993Sjhb{ 79097993Sjhb struct ktr_header *kth; 7911541Srgrimes struct vnode *vp; 79297993Sjhb struct proc *p; 79397993Sjhb struct thread *td; 79497993Sjhb struct ucred *cred; 7951541Srgrimes struct uio auio; 79697993Sjhb struct iovec aiov[3]; 79762976Smckusick struct mount *mp; 79897993Sjhb int datalen, buflen, vrele_count; 7991541Srgrimes int error; 8001541Srgrimes 80197993Sjhb vp = req->ktr_vp; 80297993Sjhb /* 80397993Sjhb * If vp is NULL, the vp has been cleared out from under this 80497993Sjhb * request, so just drop it. 80597993Sjhb */ 8061541Srgrimes if (vp == NULL) 8071541Srgrimes return; 80897993Sjhb kth = &req->ktr_header; 809118607Sjhb datalen = data_lengths[(u_short)kth->ktr_type & ~KTR_DROP]; 81097993Sjhb buflen = kth->ktr_len; 81197993Sjhb cred = req->ktr_cred; 81297993Sjhb td = curthread; 8131541Srgrimes auio.uio_iov = &aiov[0]; 8141541Srgrimes auio.uio_offset = 0; 8151541Srgrimes auio.uio_segflg = UIO_SYSSPACE; 8161541Srgrimes auio.uio_rw = UIO_WRITE; 8171541Srgrimes aiov[0].iov_base = (caddr_t)kth; 8181541Srgrimes aiov[0].iov_len = sizeof(struct ktr_header); 8191541Srgrimes auio.uio_resid = sizeof(struct ktr_header); 8201541Srgrimes auio.uio_iovcnt = 1; 82197993Sjhb auio.uio_td = td; 82297993Sjhb if (datalen != 0) { 82397993Sjhb aiov[1].iov_base = (caddr_t)&req->ktr_data; 82497993Sjhb aiov[1].iov_len = datalen; 82597993Sjhb auio.uio_resid += datalen; 8261541Srgrimes auio.uio_iovcnt++; 82797993Sjhb kth->ktr_len += datalen; 8281541Srgrimes } 82997993Sjhb if (buflen != 0) { 83097993Sjhb KASSERT(kth->ktr_buffer != NULL, ("ktrace: nothing to write")); 83197993Sjhb aiov[auio.uio_iovcnt].iov_base = kth->ktr_buffer; 83297993Sjhb aiov[auio.uio_iovcnt].iov_len = buflen; 83397993Sjhb auio.uio_resid += buflen; 83497993Sjhb auio.uio_iovcnt++; 835103235Sjhb } 83697993Sjhb mtx_lock(&Giant); 83762976Smckusick vn_start_write(vp, &mp, V_WAIT); 83883366Sjulian vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td); 83997993Sjhb (void)VOP_LEASE(vp, td, cred, LEASE_WRITE); 840101123Srwatson#ifdef MAC 841102129Srwatson error = mac_check_vnode_write(cred, NOCRED, vp); 842101123Srwatson if (error == 0) 843101123Srwatson#endif 844101123Srwatson error = VOP_WRITE(vp, &auio, IO_UNIT | IO_APPEND, cred); 84583366Sjulian VOP_UNLOCK(vp, 0, td); 84662976Smckusick vn_finished_write(mp); 84797993Sjhb mtx_unlock(&Giant); 8481541Srgrimes if (!error) 8491541Srgrimes return; 8501541Srgrimes /* 85197993Sjhb * If error encountered, give up tracing on this vnode. We defer 85297993Sjhb * all the vrele()'s on the vnode until after we are finished walking 85397993Sjhb * the various lists to avoid needlessly holding locks. 8541541Srgrimes */ 8551541Srgrimes log(LOG_NOTICE, "ktrace write failed, errno %d, tracing stopped\n", 8561541Srgrimes error); 85797993Sjhb vrele_count = 0; 85897993Sjhb /* 85997993Sjhb * First, clear this vnode from being used by any processes in the 86097993Sjhb * system. 86197993Sjhb * XXX - If one process gets an EPERM writing to the vnode, should 86297993Sjhb * we really do this? Other processes might have suitable 86397993Sjhb * credentials for the operation. 86497993Sjhb */ 865112198Sjhb cred = NULL; 86674927Sjhb sx_slock(&allproc_lock); 86753212Sphk LIST_FOREACH(p, &allproc, p_list) { 86897993Sjhb PROC_LOCK(p); 869112198Sjhb if (p->p_tracevp == vp) { 87097993Sjhb mtx_lock(&ktrace_mtx); 871112198Sjhb p->p_tracevp = NULL; 8721541Srgrimes p->p_traceflag = 0; 873112198Sjhb cred = p->p_tracecred; 874112198Sjhb p->p_tracecred = NULL; 87597993Sjhb mtx_unlock(&ktrace_mtx); 87697993Sjhb vrele_count++; 8771541Srgrimes } 87897993Sjhb PROC_UNLOCK(p); 879112198Sjhb if (cred != NULL) { 880112198Sjhb crfree(cred); 881112198Sjhb cred = NULL; 882112198Sjhb } 8831541Srgrimes } 88474927Sjhb sx_sunlock(&allproc_lock); 88597993Sjhb /* 88697993Sjhb * Second, clear this vnode from any pending requests. 88797993Sjhb */ 88897993Sjhb mtx_lock(&ktrace_mtx); 88997993Sjhb STAILQ_FOREACH(req, &ktr_todo, ktr_list) { 89097993Sjhb if (req->ktr_vp == vp) { 89197993Sjhb req->ktr_vp = NULL; 89297993Sjhb vrele_count++; 89397993Sjhb } 89497993Sjhb } 89597993Sjhb mtx_unlock(&ktrace_mtx); 89697993Sjhb mtx_lock(&Giant); 89797993Sjhb while (vrele_count-- > 0) 89897993Sjhb vrele(vp); 89997993Sjhb mtx_unlock(&Giant); 9001541Srgrimes} 9011541Srgrimes 9021541Srgrimes/* 9031541Srgrimes * Return true if caller has permission to set the ktracing state 9041541Srgrimes * of target. Essentially, the target can't possess any 9051541Srgrimes * more permissions than the caller. KTRFAC_ROOT signifies that 9068876Srgrimes * root previously set the tracing status on the target process, and 9071541Srgrimes * so, only root may further change it. 9081541Srgrimes */ 90912819Sphkstatic int 91094618Sjhbktrcanset(td, targetp) 91194618Sjhb struct thread *td; 91294618Sjhb struct proc *targetp; 9131541Srgrimes{ 9141541Srgrimes 91594618Sjhb PROC_LOCK_ASSERT(targetp, MA_OWNED); 91679335Srwatson if (targetp->p_traceflag & KTRFAC_ROOT && 917132653Scperciva suser_cred(td->td_ucred, SUSER_ALLOWJAIL)) 91846155Sphk return (0); 9191541Srgrimes 92096886Sjhb if (p_candebug(td, targetp) != 0) 92179335Srwatson return (0); 92279335Srwatson 92379335Srwatson return (1); 9241541Srgrimes} 9251541Srgrimes 92613203Swollman#endif /* KTRACE */ 927