sys/kern/kern_ktrace.c

   1 /*
   2  * Copyright (c) 1989, 1993
   3  *      The Regents of the University of California.  All rights reserved.
   4  *
   5  * Redistribution and use in source and binary forms, with or without
   6  * modification, are permitted provided that the following conditions
   7  * are met:
   8  * 1. Redistributions of source code must retain the above copyright
   9  *    notice, this list of conditions and the following disclaimer.
  10  * 2. Redistributions in binary form must reproduce the above copyright
  11  *    notice, this list of conditions and the following disclaimer in the
  12  *    documentation and/or other materials provided with the distribution.
  13  * 3. All advertising materials mentioning features or use of this software
  14  *    must display the following acknowledgement:
  15  *      This product includes software developed by the University of
  16  *      California, Berkeley and its contributors.
  17  * 4. Neither the name of the University nor the names of its contributors
  18  *    may be used to endorse or promote products derived from this software
  19  *    without specific prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  31  * SUCH DAMAGE.
  32  *
  33  *      @(#)kern_ktrace.c       8.2 (Berkeley) 9/23/93
  34  * $FreeBSD: src/sys/kern/kern_ktrace.c,v 1.35.2.6 2002/07/05 22:36:38 darrenr Exp $
  35  * $DragonFly: src/sys/kern/kern_ktrace.c,v 1.29 2007/05/07 15:43:30 dillon Exp $
  36  */
  37
  38 #include "opt_ktrace.h"
  39
  40 #include <sys/param.h>
  41 #include <sys/systm.h>
  42 #include <sys/sysproto.h>
  43 #include <sys/kernel.h>
  44 #include <sys/proc.h>
  45 #include <sys/fcntl.h>
  46 #include <sys/lock.h>
  47 #include <sys/nlookup.h>
  48 #include <sys/vnode.h>
  49 #include <sys/ktrace.h>
  50 #include <sys/malloc.h>
  51 #include <sys/syslog.h>
  52 #include <sys/sysent.h>
  53
  54 #include <vm/vm_zone.h>
  55 static MALLOC_DEFINE(M_KTRACE, "KTRACE", "KTRACE");
  56
  57 #ifdef KTRACE
  58 static struct ktr_header *ktrgetheader (int type);
  59 static void ktrwrite (struct proc *, struct ktr_header *, struct uio *);
  60 static int ktrcanset (struct proc *,struct proc *);
  61 static int ktrsetchildren (struct proc *,struct proc *,int,int, ktrace_node_t);
  62 static int ktrops (struct proc *,struct proc *,int,int, ktrace_node_t);
  63
  64 static struct ktr_header *
  65 ktrgetheader(int type)
  66 {
  67         struct ktr_header *kth;
  68         struct proc *p = curproc;       /* XXX */
  69         struct lwp *lp = curthread->td_lwp;
  70
  71         MALLOC(kth, struct ktr_header *, sizeof (struct ktr_header),
  72                 M_KTRACE, M_WAITOK);
  73         kth->ktr_type = type;
  74         /* XXX threaded flag is a hack at the moment */
  75         kth->ktr_flags = (p->p_nthreads > 1) ? KTRH_THREADED : 0;
  76         microtime(&kth->ktr_time);
  77         kth->ktr_pid = p->p_pid;
  78         kth->ktr_tid = lp->lwp_tid;
  79         bcopy(p->p_comm, kth->ktr_comm, MAXCOMLEN + 1);
  80         return (kth);
  81 }
  82
  83 void
  84 ktrsyscall(struct proc *p, int code, int narg, register_t args[])
  85 {
  86         struct  ktr_header *kth;
  87         struct  ktr_syscall *ktp;
  88         int len;
  89         register_t *argp;
  90         int i;
  91
  92         len = offsetof(struct ktr_syscall, ktr_args) +
  93               (narg * sizeof(register_t));
  94
  95         /*
  96          * Setting the active bit prevents a ktrace recursion from the
  97          * ktracing op itself.
  98          */
  99         p->p_traceflag |= KTRFAC_ACTIVE;
 100         kth = ktrgetheader(KTR_SYSCALL);
 101         MALLOC(ktp, struct ktr_syscall *, len, M_KTRACE, M_WAITOK);
 102         ktp->ktr_code = code;
 103         ktp->ktr_narg = narg;
 104         argp = &ktp->ktr_args[0];
 105         for (i = 0; i < narg; i++)
 106                 *argp++ = args[i];
 107         kth->ktr_buf = (caddr_t)ktp;
 108         kth->ktr_len = len;
 109         ktrwrite(p, kth, NULL);
 110         FREE(ktp, M_KTRACE);
 111         FREE(kth, M_KTRACE);
 112         p->p_traceflag &= ~KTRFAC_ACTIVE;
 113 }
 114
 115 void
 116 ktrsysret(struct proc *p, int code, int error, register_t retval)
 117 {
 118         struct ktr_header *kth;
 119         struct ktr_sysret ktp;
 120
 121         p->p_traceflag |= KTRFAC_ACTIVE;
 122         kth = ktrgetheader(KTR_SYSRET);
 123         ktp.ktr_code = code;
 124         ktp.ktr_error = error;
 125         ktp.ktr_retval = retval;                /* what about val2 ? */
 126
 127         kth->ktr_buf = (caddr_t)&ktp;
 128         kth->ktr_len = sizeof(struct ktr_sysret);
 129
 130         ktrwrite(p, kth, NULL);
 131         FREE(kth, M_KTRACE);
 132         p->p_traceflag &= ~KTRFAC_ACTIVE;
 133 }
 134
 135 void
 136 ktrnamei(struct proc *p, char *path)
 137 {
 138         struct ktr_header *kth;
 139
 140         p->p_traceflag |= KTRFAC_ACTIVE;
 141         kth = ktrgetheader(KTR_NAMEI);
 142         kth->ktr_len = strlen(path);
 143         kth->ktr_buf = path;
 144
 145         ktrwrite(p, kth, NULL);
 146         FREE(kth, M_KTRACE);
 147         p->p_traceflag &= ~KTRFAC_ACTIVE;
 148 }
 149
 150 void
 151 ktrgenio(struct proc *p, int fd, enum uio_rw rw, struct uio *uio, int error)
 152 {
 153         struct ktr_header *kth;
 154         struct ktr_genio ktg;
 155
 156         if (error)
 157                 return;
 158         p->p_traceflag |= KTRFAC_ACTIVE;
 159         kth = ktrgetheader(KTR_GENIO);
 160         ktg.ktr_fd = fd;
 161         ktg.ktr_rw = rw;
 162         kth->ktr_buf = (caddr_t)&ktg;
 163         kth->ktr_len = sizeof(struct ktr_genio);
 164         uio->uio_offset = 0;
 165         uio->uio_rw = UIO_WRITE;
 166
 167         ktrwrite(p, kth, uio);
 168         FREE(kth, M_KTRACE);
 169         p->p_traceflag &= ~KTRFAC_ACTIVE;
 170 }
 171
 172 void
 173 ktrpsig(struct proc *p, int sig, sig_t action, sigset_t *mask, int code)
 174 {
 175         struct ktr_header *kth;
 176         struct ktr_psig kp;
 177
 178         p->p_traceflag |= KTRFAC_ACTIVE;
 179         kth = ktrgetheader(KTR_PSIG);
 180         kp.signo = (char)sig;
 181         kp.action = action;
 182         kp.mask = *mask;
 183         kp.code = code;
 184         kth->ktr_buf = (caddr_t)&kp;
 185         kth->ktr_len = sizeof (struct ktr_psig);
 186
 187         ktrwrite(p, kth, NULL);
 188         FREE(kth, M_KTRACE);
 189         p->p_traceflag &= ~KTRFAC_ACTIVE;
 190 }
 191
 192 void
 193 ktrcsw(struct proc *p, int out, int user)
 194 {
 195         struct ktr_header *kth;
 196         struct  ktr_csw kc;
 197
 198         p->p_traceflag |= KTRFAC_ACTIVE;
 199         kth = ktrgetheader(KTR_CSW);
 200         kc.out = out;
 201         kc.user = user;
 202         kth->ktr_buf = (caddr_t)&kc;
 203         kth->ktr_len = sizeof (struct ktr_csw);
 204
 205         ktrwrite(p, kth, NULL);
 206         FREE(kth, M_KTRACE);
 207         p->p_traceflag &= ~KTRFAC_ACTIVE;
 208 }
 209 #endif
 210
 211 /* Interface and common routines */
 212
 213 #ifdef KTRACE
 214 /*
 215  * ktrace system call
 216  */
 217 struct ktrace_clear_info {
 218         ktrace_node_t tracenode;
 219         int rootclear;
 220         int error;
 221 };
 222
 223 static int ktrace_clear_callback(struct proc *p, void *data);
 224
 225 #endif
 226
 227 int
 228 sys_ktrace(struct ktrace_args *uap)
 229 {
 230 #ifdef KTRACE
 231         struct ktrace_clear_info info;
 232         struct thread *td = curthread;
 233         struct proc *curp = td->td_proc;
 234         struct proc *p;
 235         struct pgrp *pg;
 236         int facs = uap->facs & ~KTRFAC_ROOT;
 237         int ops = KTROP(uap->ops);
 238         int descend = uap->ops & KTRFLAG_DESCEND;
 239         int ret = 0;
 240         int error = 0;
 241         struct nlookupdata nd;
 242         ktrace_node_t tracenode = NULL;
 243
 244         curp->p_traceflag |= KTRFAC_ACTIVE;
 245         if (ops != KTROP_CLEAR) {
 246                 /*
 247                  * an operation which requires a file argument.
 248                  */
 249                 error = nlookup_init(&nd, uap->fname,
 250                                         UIO_USERSPACE, NLC_LOCKVP);
 251                 if (error == 0)
 252                         error = vn_open(&nd, NULL, FREAD|FWRITE|O_NOFOLLOW, 0);
 253                 if (error == 0 && nd.nl_open_vp->v_type != VREG)
 254                         error = EACCES;
 255                 if (error) {
 256                         curp->p_traceflag &= ~KTRFAC_ACTIVE;
 257                         nlookup_done(&nd);
 258                         return (error);
 259                 }
 260                 MALLOC(tracenode, ktrace_node_t, sizeof (struct ktrace_node),
 261                        M_KTRACE, M_WAITOK | M_ZERO);
 262                 tracenode->kn_vp = nd.nl_open_vp;
 263                 tracenode->kn_refs = 1;
 264                 nd.nl_open_vp = NULL;
 265                 nlookup_done(&nd);
 266                 vn_unlock(tracenode->kn_vp);
 267         }
 268         /*
 269          * Clear all uses of the tracefile.  Not the most efficient operation
 270          * in the world.
 271          */
 272         if (ops == KTROP_CLEARFILE) {
 273                 info.tracenode = tracenode;
 274                 info.error = 0;
 275                 info.rootclear = 0;
 276                 allproc_scan(ktrace_clear_callback, &info);
 277                 error = info.error;
 278                 goto done;
 279         }
 280         /*
 281          * need something to (un)trace (XXX - why is this here?)
 282          */
 283         if (!facs) {
 284                 error = EINVAL;
 285                 goto done;
 286         }
 287         /*
 288          * do it
 289          */
 290         if (uap->pid < 0) {
 291                 /*
 292                  * by process group
 293                  */
 294                 pg = pgfind(-uap->pid);
 295                 if (pg == NULL) {
 296                         error = ESRCH;
 297                         goto done;
 298                 }
 299                 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
 300                         if (descend)
 301                                 ret |= ktrsetchildren(curp, p, ops, facs, tracenode);
 302                         else
 303                                 ret |= ktrops(curp, p, ops, facs, tracenode);
 304                 }
 305         } else {
 306                 /*
 307                  * by pid
 308                  */
 309                 p = pfind(uap->pid);
 310                 if (p == NULL) {
 311                         error = ESRCH;
 312                         goto done;
 313                 }
 314                 if (descend)
 315                         ret |= ktrsetchildren(curp, p, ops, facs, tracenode);
 316                 else
 317                         ret |= ktrops(curp, p, ops, facs, tracenode);
 318         }
 319         if (!ret)
 320                 error = EPERM;
 321 done:
 322         if (tracenode)
 323                 ktrdestroy(&tracenode);
 324         curp->p_traceflag &= ~KTRFAC_ACTIVE;
 325         return (error);
 326 #else
 327         return ENOSYS;
 328 #endif
 329 }
 330
 331 #ifdef KTRACE
 332
 333 /*
 334  * NOTE: NOT MPSAFE (yet)
 335  */
 336 static int
 337 ktrace_clear_callback(struct proc *p, void *data)
 338 {
 339         struct ktrace_clear_info *info = data;
 340
 341         if (info->rootclear) {
 342                 if (p->p_tracenode == info->tracenode) {
 343                         ktrdestroy(&p->p_tracenode);
 344                         p->p_traceflag = 0;
 345                 }
 346         } else {
 347                 if (p->p_tracenode->kn_vp == info->tracenode->kn_vp) {
 348                         if (ktrcanset(curproc, p)) {
 349                                 ktrdestroy(&p->p_tracenode);
 350                                 p->p_traceflag = 0;
 351                         } else {
 352                                 info->error = EPERM;
 353                         }
 354                 }
 355         }
 356         return(0);
 357 }
 358
 359 #endif
 360
 361 /*
 362  * utrace system call
 363  */
 364 /* ARGSUSED */
 365 int
 366 sys_utrace(struct utrace_args *uap)
 367 {
 368 #ifdef KTRACE
 369         struct ktr_header *kth;
 370         struct thread *td = curthread;  /* XXX */
 371         struct proc *p = td->td_proc;
 372         caddr_t cp;
 373
 374         if (!KTRPOINT(td, KTR_USER))
 375                 return (0);
 376         if (uap->len > KTR_USER_MAXLEN)
 377                 return (EINVAL);
 378         p->p_traceflag |= KTRFAC_ACTIVE;
 379         kth = ktrgetheader(KTR_USER);
 380         MALLOC(cp, caddr_t, uap->len, M_KTRACE, M_WAITOK);
 381         if (!copyin(uap->addr, cp, uap->len)) {
 382                 kth->ktr_buf = cp;
 383                 kth->ktr_len = uap->len;
 384                 ktrwrite(p, kth, NULL);
 385         }
 386         FREE(kth, M_KTRACE);
 387         FREE(cp, M_KTRACE);
 388         p->p_traceflag &= ~KTRFAC_ACTIVE;
 389
 390         return (0);
 391 #else
 392         return (ENOSYS);
 393 #endif
 394 }
 395
 396 void
 397 ktrdestroy(struct ktrace_node **tracenodep)
 398 {
 399         ktrace_node_t tracenode;
 400
 401         if ((tracenode = *tracenodep) != NULL) {
 402                 *tracenodep = NULL;
 403                 KKASSERT(tracenode->kn_refs > 0);
 404                 /* XXX not MP safe yet */
 405                 --tracenode->kn_refs;
 406                 if (tracenode->kn_refs == 0) {
 407                         vn_close(tracenode->kn_vp, FREAD|FWRITE);
 408                         tracenode->kn_vp = NULL;
 409                         FREE(tracenode, M_KTRACE);
 410                 }
 411         }
 412 }
 413
 414 /*
 415  * This allows a process to inherit a ref on a tracenode and is also used
 416  * as a temporary ref to prevent a tracenode from being destroyed out from
 417  * under an active operation.
 418  */
 419 ktrace_node_t
 420 ktrinherit(ktrace_node_t tracenode)
 421 {
 422         if (tracenode) {
 423                 KKASSERT(tracenode->kn_refs > 0);
 424                 ++tracenode->kn_refs;
 425         }
 426         return(tracenode);
 427 }
 428
 429 #ifdef KTRACE
 430 static int
 431 ktrops(struct proc *curp, struct proc *p, int ops, int facs,
 432        ktrace_node_t tracenode)
 433 {
 434         ktrace_node_t oldnode;
 435
 436         if (!ktrcanset(curp, p))
 437                 return (0);
 438         if (ops == KTROP_SET) {
 439                 if ((oldnode = p->p_tracenode) != tracenode) {
 440                         p->p_tracenode = ktrinherit(tracenode);
 441                         ktrdestroy(&oldnode);
 442                 }
 443                 p->p_traceflag |= facs;
 444                 if (curp->p_ucred->cr_uid == 0)
 445                         p->p_traceflag |= KTRFAC_ROOT;
 446         } else {
 447                 /* KTROP_CLEAR */
 448                 if (((p->p_traceflag &= ~facs) & KTRFAC_MASK) == 0) {
 449                         /* no more tracing */
 450                         p->p_traceflag = 0;
 451                         ktrdestroy(&p->p_tracenode);
 452                 }
 453         }
 454
 455         return (1);
 456 }
 457
 458 static int
 459 ktrsetchildren(struct proc *curp, struct proc *top, int ops, int facs,
 460                ktrace_node_t tracenode)
 461 {
 462         struct proc *p;
 463         int ret = 0;
 464
 465         p = top;
 466         for (;;) {
 467                 ret |= ktrops(curp, p, ops, facs, tracenode);
 468                 /*
 469                  * If this process has children, descend to them next,
 470                  * otherwise do any siblings, and if done with this level,
 471                  * follow back up the tree (but not past top).
 472                  */
 473                 if (!LIST_EMPTY(&p->p_children))
 474                         p = LIST_FIRST(&p->p_children);
 475                 else for (;;) {
 476                         if (p == top)
 477                                 return (ret);
 478                         if (LIST_NEXT(p, p_sibling)) {
 479                                 p = LIST_NEXT(p, p_sibling);
 480                                 break;
 481                         }
 482                         p = p->p_pptr;
 483                 }
 484         }
 485         /*NOTREACHED*/
 486 }
 487
 488 static void
 489 ktrwrite(struct proc *p, struct ktr_header *kth, struct uio *uio)
 490 {
 491         struct ktrace_clear_info info;
 492         struct uio auio;
 493         struct iovec aiov[2];
 494         int error;
 495         ktrace_node_t tracenode;
 496
 497         /*
 498          * We have to ref our tracenode to prevent it from being ripped out
 499          * from under us while we are trying to use it.   p_tracenode can
 500          * go away at any time if another process gets a write error.
 501          *
 502          * XXX not MP safe
 503          */
 504         if (p->p_tracenode == NULL)
 505                 return;
 506         tracenode = ktrinherit(p->p_tracenode);
 507         auio.uio_iov = &aiov[0];
 508         auio.uio_offset = 0;
 509         auio.uio_segflg = UIO_SYSSPACE;
 510         auio.uio_rw = UIO_WRITE;
 511         aiov[0].iov_base = (caddr_t)kth;
 512         aiov[0].iov_len = sizeof(struct ktr_header);
 513         auio.uio_resid = sizeof(struct ktr_header);
 514         auio.uio_iovcnt = 1;
 515         auio.uio_td = curthread;
 516         if (kth->ktr_len > 0) {
 517                 auio.uio_iovcnt++;
 518                 aiov[1].iov_base = kth->ktr_buf;
 519                 aiov[1].iov_len = kth->ktr_len;
 520                 auio.uio_resid += kth->ktr_len;
 521                 if (uio != NULL)
 522                         kth->ktr_len += uio->uio_resid;
 523         }
 524         vn_lock(tracenode->kn_vp, LK_EXCLUSIVE | LK_RETRY);
 525         error = VOP_WRITE(tracenode->kn_vp, &auio,
 526                           IO_UNIT | IO_APPEND, p->p_ucred);
 527         if (error == 0 && uio != NULL) {
 528                 error = VOP_WRITE(tracenode->kn_vp, uio,
 529                                   IO_UNIT | IO_APPEND, p->p_ucred);
 530         }
 531         vn_unlock(tracenode->kn_vp);
 532         if (error) {
 533                 /*
 534                  * If an error occured, give up tracing on all processes
 535                  * using this tracenode.  This is not MP safe but is
 536                  * blocking-safe.
 537                  */
 538                 log(LOG_NOTICE,
 539                     "ktrace write failed, errno %d, tracing stopped\n", error);
 540                 info.tracenode = tracenode;
 541                 info.error = 0;
 542                 info.rootclear = 1;
 543                 allproc_scan(ktrace_clear_callback, &info);
 544         }
 545         ktrdestroy(&tracenode);
 546 }
 547
 548 /*
 549  * Return true if caller has permission to set the ktracing state
 550  * of target.  Essentially, the target can't possess any
 551  * more permissions than the caller.  KTRFAC_ROOT signifies that
 552  * root previously set the tracing status on the target process, and
 553  * so, only root may further change it.
 554  *
 555  * TODO: check groups.  use caller effective gid.
 556  */
 557 static int
 558 ktrcanset(struct proc *callp, struct proc *targetp)
 559 {
 560         struct ucred *caller = callp->p_ucred;
 561         struct ucred *target = targetp->p_ucred;
 562
 563         if (!PRISON_CHECK(caller, target))
 564                 return (0);
 565         if ((caller->cr_uid == target->cr_ruid &&
 566              target->cr_ruid == target->cr_svuid &&
 567              caller->cr_rgid == target->cr_rgid &&      /* XXX */
 568              target->cr_rgid == target->cr_svgid &&
 569              (targetp->p_traceflag & KTRFAC_ROOT) == 0 &&
 570              (targetp->p_flag & P_SUGID) == 0) ||
 571              caller->cr_uid == 0)
 572                 return (1);
 573
 574         return (0);
 575 }
 576
 577 #endif /* KTRACE */