]> git.saurik.com Git - apple/xnu.git/blobdiff - bsd/kern/kern_ktrace.c
xnu-3789.60.24.tar.gz
[apple/xnu.git] / bsd / kern / kern_ktrace.c
index c88bbf81ae57fccb8666e7600854c98e35297075..af4573ef80e8f5169c38d64d281dc5bb303e79b1 100644 (file)
 /*
- * Copyright (c) 2000 Apple Computer, Inc. All rights reserved.
+ * Copyright (c) 2015 Apple Inc. All rights reserved.
  *
- * @APPLE_LICENSE_HEADER_START@
- * 
- * The contents of this file constitute Original Code as defined in and
- * are subject to the Apple Public Source License Version 1.1 (the
- * "License").  You may not use this file except in compliance with the
- * License.  Please obtain a copy of the License at
- * http://www.apple.com/publicsource and read it before using this file.
- * 
- * This Original Code and all software distributed under the License are
- * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
+ *
+ * This file contains Original Code and/or Modifications of Original Code
+ * as defined in and that are subject to the Apple Public Source License
+ * Version 2.0 (the 'License'). You may not use this file except in
+ * compliance with the License. The rights granted to you under the License
+ * may not be used to create, or enable the creation or redistribution of,
+ * unlawful or unlicensed copies of an Apple operating system, or to
+ * circumvent, violate, or enable the circumvention or violation of, any
+ * terms of an Apple operating system software license agreement.
+ *
+ * Please obtain a copy of the License at
+ * http://www.opensource.apple.com/apsl/ and read it before using this file.
+ *
+ * The Original Code and all software distributed under the License are
+ * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT.  Please see the
- * License for the specific language governing rights and limitations
- * under the License.
- * 
- * @APPLE_LICENSE_HEADER_END@
+ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
+ * Please see the License for the specific language governing rights and
+ * limitations under the License.
+ *
+ * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
-/* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */
+
 /*
- * Copyright (c) 1989, 1993
- *     The Regents of the University of California.  All rights reserved.
+ * This file manages the ownership of ktrace and its subsystems, like kdebug
+ * and kperf, as well as the overall state of the system, whether it is in
+ * foreground or background mode.
  *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- * 1. Redistributions of source code must retain the above copyright
- *    notice, this list of conditions and the following disclaimer.
- * 2. Redistributions in binary form must reproduce the above copyright
- *    notice, this list of conditions and the following disclaimer in the
- *    documentation and/or other materials provided with the distribution.
- * 3. All advertising materials mentioning features or use of this software
- *    must display the following acknowledgement:
- *     This product includes software developed by the University of
- *     California, Berkeley and its contributors.
- * 4. Neither the name of the University nor the names of its contributors
- *    may be used to endorse or promote products derived from this software
- *    without specific prior written permission.
+ * When unconfigured or in background mode, any root process can take ownership
+ * of ktrace and configure it, changing the state to foreground and, in the case
+ * of a transition out of background, resetting the background configuration.
  *
- * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
- * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
- * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
- * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
- * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
- * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
- * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
- * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
- * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
- * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
- * SUCH DAMAGE.
+ * When in foreground mode, if the owning process is still running, only it may
+ * configure ktrace.  If it exits, ktrace keeps running but any root process can
+ * change the configuration.  When ktrace is reset, the state changes back to
+ * unconfigured and a notification is sent on the ktrace_background host special
+ * port.
  *
- *     @(#)kern_ktrace.c       8.2 (Berkeley) 9/23/93
+ * If a process has set itself as the background tool, using the init_background
+ * sysctl, it can configure ktrace only when ktrace is off or already in
+ * background mode.  The first attempt to configure ktrace by the background pid
+ * when it is off results in the transition to background mode.
  */
 
+#include <sys/ktrace.h>
 
-#include <sys/param.h>
-#include <sys/systm.h>
+#include <mach/host_priv.h>
+#include <mach/mach_types.h>
+#include <mach/ktrace_background.h>
+
+#include <sys/kauth.h>
+#include <sys/priv.h>
 #include <sys/proc.h>
-#include <sys/file.h>
-#include <sys/namei.h>
-#include <sys/vnode.h>
-#include <sys/ktrace.h>
-#include <sys/malloc.h>
-#include <sys/syslog.h>
+char *proc_name_address(void *p);
+#include <sys/sysctl.h>
+#include <sys/vm.h>
+
+#include <kern/locks.h>
+#include <kern/assert.h>
+
+#include <sys/kdebug.h>
+#include <kperf/kperf.h>
+
+#include <kern/host.h>
 
-#include <sys/mount.h>
+kern_return_t ktrace_background_available_notify_user(void);
 
-#if KTRACE
+lck_mtx_t *ktrace_lock;
 
-struct ktr_header *
-ktrgetheader(type)
-       int type;
+/*
+ * The overall state of ktrace, whether it is unconfigured, in foreground mode,
+ * or in background mode.  The state determines which processes can configure
+ * ktrace.
+ */
+static enum ktrace_state ktrace_state = KTRACE_STATE_OFF;
+
+/* The true owner of ktrace, checked by ktrace_access_check(). */
+static uint64_t ktrace_owning_unique_id = 0;
+static pid_t ktrace_owning_pid = 0;
+
+/*
+ * The background pid of ktrace, automatically made the owner when
+ * transitioning to background mode.
+ */
+static uint64_t ktrace_bg_unique_id = 0;
+static pid_t ktrace_bg_pid = 0;
+
+/* The name of the last process to configure ktrace. */
+static char ktrace_last_owner_execname[MAXCOMLEN + 1] = { 0 };
+
+/*
+ * Which subsystems of ktrace (currently kdebug and kperf) are active.
+ */
+static uint32_t ktrace_active_mask = 0;
+
+/*
+ * At boot or when a daemon has been newly loaded, it's necessary to bootstrap
+ * user space background tools by sending a background available notification
+ * when the init_background sysctl is made.
+ *
+ * Background tools must be RunAtLoad daemons.
+ */
+static boolean_t should_notify_on_init = TRUE;
+
+/* Set the owning process of ktrace. */
+static void ktrace_set_owning_proc(proc_t p);
+
+/* Reset ktrace ownership back to unowned. */
+static void ktrace_release_ownership(void);
+
+/* Make the background tool the owner of ktrace. */
+static void ktrace_promote_background(void);
+
+/*
+ * If user space sets a pid manually (through kperf "blessing"), ktrace should
+ * not treat resets as releasing ownership.  At that point, ownership is only
+ * released when the owner is set to an invalid pid.
+ *
+ * This is managed by the user space-oriented function ktrace_set_owning_pid
+ * and ktrace_unset_owning_pid.
+ */
+boolean_t ktrace_keep_ownership_on_reset = FALSE;
+
+/* Allow user space to unset the owning pid and potentially reset ktrace. */
+static void ktrace_set_invalid_owning_pid(void);
+
+/*
+ * This flag allows any root process to set a new ktrace owner.  It is
+ * currently used by Instruments.
+ */
+int ktrace_root_set_owner_allowed = 0;
+
+static void
+ktrace_reset_internal(uint32_t reset_mask)
 {
-       register struct ktr_header *kth;
-       struct proc *p = current_proc();        /* XXX */
-
-       MALLOC(kth, struct ktr_header *, sizeof (struct ktr_header), 
-               M_TEMP, M_WAITOK);
-       kth->ktr_type = type;
-       microtime(&kth->ktr_time);
-       kth->ktr_pid = p->p_pid;
-       bcopy(p->p_comm, kth->ktr_comm, MAXCOMLEN);
-       return (kth);
+       if (!ktrace_keep_ownership_on_reset) {
+               ktrace_active_mask &= ~reset_mask;
+       }
+
+       if (reset_mask & KTRACE_KPERF) {
+               kperf_reset();
+       }
+       if (reset_mask & KTRACE_KDEBUG) {
+               kdebug_reset();
+       }
+
+       if (ktrace_active_mask == 0) {
+               if (ktrace_state == KTRACE_STATE_FG) {
+                       /* transition from foreground to background */
+                       ktrace_promote_background();
+               } else if (ktrace_state == KTRACE_STATE_BG) {
+                       /* background tool is resetting ktrace */
+                       should_notify_on_init = TRUE;
+                       ktrace_release_ownership();
+                       ktrace_state = KTRACE_STATE_OFF;
+               }
+       }
 }
 
 void
-ktrsyscall(vp, code, argsize, args)
-       struct vnode *vp;
-       register_t code;
-       size_t argsize;
-       register_t args[];
+ktrace_reset(uint32_t reset_mask)
 {
-       struct  ktr_header *kth;
-       struct  ktr_syscall *ktp;
-       register len = sizeof(struct ktr_syscall) + argsize;
-       struct proc *p = current_proc();        /* XXX */
-       register_t *argp;
-       int i;
-
-       p->p_traceflag |= KTRFAC_ACTIVE;
-       kth = ktrgetheader(KTR_SYSCALL);
-       MALLOC(ktp, struct ktr_syscall *, len, M_TEMP, M_WAITOK);
-       ktp->ktr_code = code;
-       ktp->ktr_argsize = argsize;
-       argp = (register_t *)((char *)ktp + sizeof(struct ktr_syscall));
-       for (i = 0; i < (argsize / sizeof *argp); i++)
-               *argp++ = args[i];
-       kth->ktr_buf = (caddr_t)ktp;
-       kth->ktr_len = len;
-       ktrwrite(vp, kth);
-       FREE(ktp, M_TEMP);
-       FREE(kth, M_TEMP);
-       p->p_traceflag &= ~KTRFAC_ACTIVE;
+       lck_mtx_assert(ktrace_lock, LCK_MTX_ASSERT_OWNED);
+
+       if (ktrace_active_mask == 0) {
+               if (!ktrace_keep_ownership_on_reset) {
+                       assert(ktrace_state == KTRACE_STATE_OFF);
+               }
+               return;
+       }
+
+       ktrace_reset_internal(reset_mask);
 }
 
-void
-ktrsysret(vp, code, error, retval)
-       struct vnode *vp;
-       register_t code;
-       int error;
-       register_t retval;
+static void
+ktrace_promote_background(void)
+{
+       lck_mtx_assert(ktrace_lock, LCK_MTX_ASSERT_OWNED);
+       assert(ktrace_state != KTRACE_STATE_BG);
+
+       /*
+        * Remember to send a background available notification on the next init
+        * if the notification failed (meaning no task holds the receive right
+        * for the host special port).
+        */
+       if (ktrace_background_available_notify_user() == KERN_FAILURE) {
+               should_notify_on_init = TRUE;
+       } else {
+               should_notify_on_init = FALSE;
+       }
+
+       ktrace_release_ownership();
+       ktrace_state = KTRACE_STATE_OFF;
+}
+
+bool
+ktrace_background_active(void)
 {
-       struct ktr_header *kth;
-       struct ktr_sysret ktp;
-       struct proc *p = current_proc();        /* XXX */
-
-       p->p_traceflag |= KTRFAC_ACTIVE;
-       kth = ktrgetheader(KTR_SYSRET);
-       ktp.ktr_code = code;
-       ktp.ktr_error = error;
-       ktp.ktr_retval = retval;                /* what about val2 ? */
-
-       kth->ktr_buf = (caddr_t)&ktp;
-       kth->ktr_len = sizeof(struct ktr_sysret);
-
-       ktrwrite(vp, kth);
-       FREE(kth, M_TEMP);
-       p->p_traceflag &= ~KTRFAC_ACTIVE;
+       lck_mtx_assert(ktrace_lock, LCK_MTX_ASSERT_OWNED);
+       return (ktrace_state == KTRACE_STATE_BG);
 }
 
-void
-ktrnamei(vp, path)
-       struct vnode *vp;
-       char *path;
+int
+ktrace_read_check(void)
 {
-       struct ktr_header *kth;
-       struct proc *p = current_proc();        /* XXX */
+       lck_mtx_assert(ktrace_lock, LCK_MTX_ASSERT_OWNED);
 
-       p->p_traceflag |= KTRFAC_ACTIVE;
-       kth = ktrgetheader(KTR_NAMEI);
-       kth->ktr_len = strlen(path);
-       kth->ktr_buf = path;
+       if (proc_uniqueid(current_proc()) == ktrace_owning_unique_id)
+       {
+               return 0;
+       }
 
-       ktrwrite(vp, kth);
-       FREE(kth, M_TEMP);
-       p->p_traceflag &= ~KTRFAC_ACTIVE;
+       return kauth_cred_issuser(kauth_cred_get()) ? 0 : EPERM;
 }
 
-void
-ktrgenio(vp, fd, rw, iov, len, error)
-       struct vnode *vp;
-       int fd;
-       enum uio_rw rw;
-       register struct iovec *iov;
-       int len, error;
+/* If an owning process has exited, reset the ownership. */
+static void
+ktrace_ownership_maintenance(void)
 {
-       struct ktr_header *kth;
-       register struct ktr_genio *ktp;
-       register caddr_t cp;
-       register int resid = len, cnt;
-       struct proc *p = current_proc();        /* XXX */
-       
-       if (error)
+       lck_mtx_assert(ktrace_lock, LCK_MTX_ASSERT_OWNED);
+
+       /* do nothing if ktrace is not owned */
+       if (ktrace_owning_unique_id == 0) {
                return;
-       p->p_traceflag |= KTRFAC_ACTIVE;
-       kth = ktrgetheader(KTR_GENIO);
-       MALLOC(ktp, struct ktr_genio *, sizeof(struct ktr_genio) + len,
-               M_TEMP, M_WAITOK);
-       ktp->ktr_fd = fd;
-       ktp->ktr_rw = rw;
-       cp = (caddr_t)((char *)ktp + sizeof (struct ktr_genio));
-       while (resid > 0) {
-               if ((cnt = iov->iov_len) > resid)
-                       cnt = resid;
-               if (copyin(iov->iov_base, cp, (unsigned)cnt))
-                       goto done;
-               cp += cnt;
-               resid -= cnt;
-               iov++;
        }
-       kth->ktr_buf = (caddr_t)ktp;
-       kth->ktr_len = sizeof (struct ktr_genio) + len;
-
-       ktrwrite(vp, kth);
-done:
-       FREE(kth, M_TEMP);
-       FREE(ktp, M_TEMP);
-       p->p_traceflag &= ~KTRFAC_ACTIVE;
+
+       /* reset ownership if process cannot be found */
+
+       proc_t owning_proc = proc_find(ktrace_owning_pid);
+
+       if (owning_proc != NULL) {
+               /* make sure the pid was not recycled */
+               if (proc_uniqueid(owning_proc) != ktrace_owning_unique_id) {
+                       ktrace_release_ownership();
+               }
+
+               proc_rele(owning_proc);
+       } else {
+               ktrace_release_ownership();
+       }
 }
 
-void
-ktrpsig(vp, sig, action, mask, code)
-       struct vnode *vp;
-       int sig;
-       sig_t action;
-       int mask, code;
+int
+ktrace_configure(uint32_t config_mask)
 {
-       struct ktr_header *kth;
-       struct ktr_psig kp;
-       struct proc *p = current_proc();        /* XXX */
-
-       p->p_traceflag |= KTRFAC_ACTIVE;
-       kth = ktrgetheader(KTR_PSIG);
-       kp.signo = (char)sig;
-       kp.action = action;
-       kp.mask = mask;
-       kp.code = code;
-       kth->ktr_buf = (caddr_t)&kp;
-       kth->ktr_len = sizeof (struct ktr_psig);
-
-       ktrwrite(vp, kth);
-       FREE(kth, M_TEMP);
-       p->p_traceflag &= ~KTRFAC_ACTIVE;
+       lck_mtx_assert(ktrace_lock, LCK_MTX_ASSERT_OWNED);
+       assert(config_mask != 0);
+
+       proc_t p = current_proc();
+
+       /* if process clearly owns ktrace, allow */
+       if (proc_uniqueid(p) == ktrace_owning_unique_id) {
+               ktrace_active_mask |= config_mask;
+               return 0;
+       }
+
+       /* background configure while foreground is active is not allowed */
+       if (proc_uniqueid(p) == ktrace_bg_unique_id &&
+           ktrace_state == KTRACE_STATE_FG)
+       {
+               return EBUSY;
+       }
+
+       ktrace_ownership_maintenance();
+
+       /* allow process to gain control when unowned or background */
+       if (ktrace_owning_unique_id == 0 || ktrace_state == KTRACE_STATE_BG) {
+               if (!kauth_cred_issuser(kauth_cred_get())) {
+                       return EPERM;
+               }
+
+               ktrace_set_owning_proc(p);
+               ktrace_active_mask |= config_mask;
+               return 0;
+       }
+
+       /* owned by an existing, different process */
+       return EBUSY;
 }
 
 void
-ktrcsw(vp, out, user)
-       struct vnode *vp;
-       int out, user;
+ktrace_disable(enum ktrace_state state_to_match)
 {
-       struct ktr_header *kth;
-       struct  ktr_csw kc;
-       struct proc *p = current_proc();        /* XXX */
-
-       p->p_traceflag |= KTRFAC_ACTIVE;
-       kth = ktrgetheader(KTR_CSW);
-       kc.out = out;
-       kc.user = user;
-       kth->ktr_buf = (caddr_t)&kc;
-       kth->ktr_len = sizeof (struct ktr_csw);
-
-       ktrwrite(vp, kth);
-       FREE(kth, M_TEMP);
-       p->p_traceflag &= ~KTRFAC_ACTIVE;
+       if (ktrace_state == state_to_match) {
+               kernel_debug_disable();
+               kperf_sampling_disable();
+       }
 }
 
-/* Interface and common routines */
-
-/*
- * ktrace system call
- */
-struct ktrace_args {
-       char *  fname;
-       int             ops;
-       int             facs;
-       int             pid;
-};
-/* ARGSUSED */
 int
-ktrace(curp, uap, retval)
-       struct proc *curp;
-       register struct ktrace_args *uap;
-       register_t *retval;
+ktrace_get_owning_pid(void)
 {
-       register struct vnode *vp = NULL;
-       register struct proc *p;
-       struct pgrp *pg;
-       int facs = SCARG(uap, facs) & ~KTRFAC_ROOT;
-       int ops = KTROP(SCARG(uap, ops));
-       int descend = SCARG(uap, ops) & KTRFLAG_DESCEND;
-       int ret = 0;
-       int error = 0;
-       struct nameidata nd;
-
-       curp->p_traceflag |= KTRFAC_ACTIVE;
-       if (ops != KTROP_CLEAR) {
-               /*
-                * an operation which requires a file argument.
-                */
-               NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE, SCARG(uap, fname),
-                   curp);
-               if (error = vn_open(&nd, FREAD|FWRITE, 0)) {
-                       curp->p_traceflag &= ~KTRFAC_ACTIVE;
-                       return (error);
+       lck_mtx_assert(ktrace_lock, LCK_MTX_ASSERT_OWNED);
+
+       ktrace_ownership_maintenance();
+       return ktrace_owning_pid;
+}
+
+void
+ktrace_kernel_configure(uint32_t config_mask)
+{
+       lck_mtx_assert(ktrace_lock, LCK_MTX_ASSERT_OWNED);
+
+       if (ktrace_state != KTRACE_STATE_OFF) {
+               if (ktrace_active_mask & KTRACE_KPERF) {
+                       kperf_reset();
                }
-               vp = nd.ni_vp;
-               VOP_UNLOCK(vp, 0, p);
-               if (vp->v_type != VREG) {
-                       (void) vn_close(vp, FREAD|FWRITE, curp->p_ucred, curp);
-                       curp->p_traceflag &= ~KTRFAC_ACTIVE;
-                       return (EACCES);
+               if (ktrace_active_mask & KTRACE_KDEBUG) {
+                       kdebug_reset();
                }
        }
+
+       ktrace_active_mask = config_mask;
+       ktrace_state = KTRACE_STATE_FG;
+
+       ktrace_release_ownership();
+       strlcpy(ktrace_last_owner_execname, "kernel_task",
+               sizeof(ktrace_last_owner_execname));
+}
+
+static errno_t
+ktrace_init_background(void)
+{
+       int err = 0;
+
+       lck_mtx_assert(ktrace_lock, LCK_MTX_ASSERT_OWNED);
+
+       if ((err = priv_check_cred(kauth_cred_get(), PRIV_KTRACE_BACKGROUND, 0))) {
+               return err;
+       }
+
        /*
-        * Clear all uses of the tracefile
+        * When a background tool first checks in, send a notification if ktrace
+        * is available.
         */
-       if (ops == KTROP_CLEARFILE) {
-               for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
-                       if (p->p_tracep == vp) {
-                               if (ktrcanset(curp, p)) {
-                                       p->p_tracep = NULL;
-                                       p->p_traceflag = 0;
-                                       (void) vn_close(vp, FREAD|FWRITE,
-                                               p->p_ucred, p);
-                               } else
-                                       error = EPERM;
+       if (should_notify_on_init) {
+               if (ktrace_state == KTRACE_STATE_OFF) {
+                       /*
+                        * This notification can only fail if a process does not
+                        * hold the receive right for the host special port.
+                        * Return an error and don't make the current process
+                        * the background tool.
+                        */
+                       if (ktrace_background_available_notify_user() == KERN_FAILURE) {
+                               return EINVAL;
                        }
                }
-               goto done;
+               should_notify_on_init = FALSE;
        }
-       /*
-        * need something to (un)trace (XXX - why is this here?)
-        */
-       if (!facs) {
-               error = EINVAL;
-               goto done;
+
+       proc_t p = current_proc();
+
+       ktrace_bg_unique_id = proc_uniqueid(p);
+       ktrace_bg_pid = proc_pid(p);
+
+       if (ktrace_state == KTRACE_STATE_BG) {
+               ktrace_set_owning_proc(p);
        }
-       /* 
-        * do it
-        */
-       if (SCARG(uap, pid) < 0) {
-               /*
-                * by process group
-                */
-               pg = pgfind(-SCARG(uap, pid));
-               if (pg == NULL) {
-                       error = ESRCH;
-                       goto done;
-               }
-               for (p = pg->pg_members.lh_first; p != 0; p = p->p_pglist.le_next)
-                       if (descend)
-                               ret |= ktrsetchildren(curp, p, ops, facs, vp);
-                       else 
-                               ret |= ktrops(curp, p, ops, facs, vp);
-                                       
-       } else {
-               /*
-                * by pid
-                */
-               p = pfind(SCARG(uap, pid));
-               if (p == NULL) {
-                       error = ESRCH;
-                       goto done;
-               }
-               if (descend)
-                       ret |= ktrsetchildren(curp, p, ops, facs, vp);
-               else
-                       ret |= ktrops(curp, p, ops, facs, vp);
+
+       return 0;
+}
+
+void
+ktrace_set_invalid_owning_pid(void)
+{
+       if (ktrace_keep_ownership_on_reset) {
+               ktrace_keep_ownership_on_reset = FALSE;
+               ktrace_reset_internal(ktrace_active_mask);
        }
-       if (!ret)
-               error = EPERM;
-done:
-       if (vp != NULL)
-               (void) vn_close(vp, FWRITE, curp->p_ucred, curp);
-       curp->p_traceflag &= ~KTRFAC_ACTIVE;
-       return (error);
 }
 
 int
-ktrops(curp, p, ops, facs, vp)
-       struct proc *p, *curp;
-       int ops, facs;
-       struct vnode *vp;
+ktrace_set_owning_pid(int pid)
 {
+       lck_mtx_assert(ktrace_lock, LCK_MTX_ASSERT_OWNED);
 
-       if (!ktrcanset(curp, p))
-               return (0);
-       if (ops == KTROP_SET) {
-               if (p->p_tracep != vp) { 
-                       /*
-                        * if trace file already in use, relinquish
-                        */
-                       if (p->p_tracep != NULL)
-                               vrele(p->p_tracep);
-                       VREF(vp);
-                       p->p_tracep = vp;
-               }
-               p->p_traceflag |= facs;
-               if (curp->p_ucred->cr_uid == 0)
-                       p->p_traceflag |= KTRFAC_ROOT;
-       } else {        
-               /* KTROP_CLEAR */
-               if (((p->p_traceflag &= ~facs) & KTRFAC_MASK) == 0) {
-                       /* no more tracing */
-                       p->p_traceflag = 0;
-                       if (p->p_tracep != NULL) {
-                               vrele(p->p_tracep);
-                               p->p_tracep = NULL;
-                       }
-               }
+       /* allow user space to successfully unset owning pid */
+       if (pid == -1) {
+               ktrace_set_invalid_owning_pid();
+               return 0;
+       }
+
+       /* use ktrace_reset or ktrace_release_ownership, not this */
+       if (pid == 0) {
+               ktrace_set_invalid_owning_pid();
+               return EINVAL;
        }
 
-       return (1);
+       proc_t p = proc_find(pid);
+       if (!p) {
+               ktrace_set_invalid_owning_pid();
+               return ESRCH;
+       }
+
+       ktrace_keep_ownership_on_reset = TRUE;
+       ktrace_set_owning_proc(p);
+
+       proc_rele(p);
+       return 0;
 }
 
-ktrsetchildren(curp, top, ops, facs, vp)
-       struct proc *curp, *top;
-       int ops, facs;
-       struct vnode *vp;
+static void
+ktrace_set_owning_proc(proc_t p)
 {
-       register struct proc *p;
-       register int ret = 0;
-
-       p = top;
-       for (;;) {
-               ret |= ktrops(curp, p, ops, facs, vp);
-               /*
-                * If this process has children, descend to them next,
-                * otherwise do any siblings, and if done with this level,
-                * follow back up the tree (but not past top).
-                */
-               if (p->p_children.lh_first)
-                       p = p->p_children.lh_first;
-               else for (;;) {
-                       if (p == top)
-                               return (ret);
-                       if (p->p_sibling.le_next) {
-                               p = p->p_sibling.le_next;
-                               break;
+       lck_mtx_assert(ktrace_lock, LCK_MTX_ASSERT_OWNED);
+       assert(p);
+
+       if (ktrace_state != KTRACE_STATE_FG) {
+               if (proc_uniqueid(p) == ktrace_bg_unique_id) {
+                       ktrace_state = KTRACE_STATE_BG;
+               } else {
+                       if (ktrace_state == KTRACE_STATE_BG) {
+                               if (ktrace_active_mask & KTRACE_KPERF) {
+                                       kperf_reset();
+                               }
+                               if (ktrace_active_mask & KTRACE_KDEBUG) {
+                                       kdebug_reset();
+                               }
+
+                               ktrace_active_mask = 0;
                        }
-                       p = p->p_pptr;
+                       ktrace_state = KTRACE_STATE_FG;
+                       should_notify_on_init = FALSE;
                }
        }
-       /*NOTREACHED*/
+
+       ktrace_owning_unique_id = proc_uniqueid(p);
+       ktrace_owning_pid = proc_pid(p);
+       strlcpy(ktrace_last_owner_execname, proc_name_address(p),
+               sizeof(ktrace_last_owner_execname));
 }
 
-ktrwrite(vp, kth)
-       struct vnode *vp;
-       register struct ktr_header *kth;
+static void
+ktrace_release_ownership(void)
 {
-       struct uio auio;
-       struct iovec aiov[2];
-       register struct proc *p = current_proc();       /* XXX */
-       int error;
+       lck_mtx_assert(ktrace_lock, LCK_MTX_ASSERT_OWNED);
 
-       if (vp == NULL)
-               return;
-       auio.uio_iov = &aiov[0];
-       auio.uio_offset = 0;
-       auio.uio_segflg = UIO_SYSSPACE;
-       auio.uio_rw = UIO_WRITE;
-       aiov[0].iov_base = (caddr_t)kth;
-       aiov[0].iov_len = sizeof(struct ktr_header);
-       auio.uio_resid = sizeof(struct ktr_header);
-       auio.uio_iovcnt = 1;
-       auio.uio_procp = (struct proc *)0;
-       if (kth->ktr_len > 0) {
-               auio.uio_iovcnt++;
-               aiov[1].iov_base = kth->ktr_buf;
-               aiov[1].iov_len = kth->ktr_len;
-               auio.uio_resid += kth->ktr_len;
+       ktrace_owning_unique_id = 0;
+       ktrace_owning_pid = 0;
+}
+
+#define SYSCTL_INIT_BACKGROUND (1)
+
+static int ktrace_sysctl SYSCTL_HANDLER_ARGS;
+
+SYSCTL_NODE(, OID_AUTO, ktrace, CTLFLAG_RW | CTLFLAG_LOCKED, 0, "ktrace");
+
+SYSCTL_UINT(_ktrace, OID_AUTO, state, CTLFLAG_RD | CTLFLAG_LOCKED,
+            &ktrace_state, 0,
+            "");
+
+SYSCTL_INT(_ktrace, OID_AUTO, owning_pid, CTLFLAG_RD | CTLFLAG_LOCKED,
+           &ktrace_owning_pid, 0,
+           "pid of the process that owns ktrace");
+
+SYSCTL_INT(_ktrace, OID_AUTO, background_pid, CTLFLAG_RD | CTLFLAG_LOCKED,
+           &ktrace_bg_pid, 0,
+           "pid of the background ktrace tool");
+
+SYSCTL_STRING(_ktrace, OID_AUTO, configured_by, CTLFLAG_RD | CTLFLAG_LOCKED,
+              ktrace_last_owner_execname, 0,
+              "execname of process that last configured ktrace");
+
+SYSCTL_PROC(_ktrace, OID_AUTO, init_background, CTLFLAG_RW | CTLFLAG_LOCKED,
+            (void *)SYSCTL_INIT_BACKGROUND, sizeof(int),
+            ktrace_sysctl, "I", "initialize calling process as background");
+
+static int
+ktrace_sysctl SYSCTL_HANDLER_ARGS
+{
+#pragma unused(oidp, arg2)
+       int ret = 0;
+       uintptr_t type = (uintptr_t)arg1;
+
+       lck_mtx_lock(ktrace_lock);
+
+       if (!kauth_cred_issuser(kauth_cred_get())) {
+               ret = EPERM;
+               goto out;
        }
-       vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
-       error = VOP_WRITE(vp, &auio, IO_UNIT|IO_APPEND, p->p_ucred);
-       VOP_UNLOCK(vp, 0, p);
-       if (!error)
-               return;
-       /*
-        * If error encountered, give up tracing on this vnode.
-        */
-       log(LOG_NOTICE, "ktrace write failed, errno %d, tracing stopped\n",
-           error);
-       for (p = allproc.lh_first; p != 0; p = p->p_list.le_next) {
-               if (p->p_tracep == vp) {
-                       p->p_tracep = NULL;
-                       p->p_traceflag = 0;
-                       vrele(vp);
+
+       if (type == SYSCTL_INIT_BACKGROUND) {
+               if (req->newptr != USER_ADDR_NULL) {
+                       ret = ktrace_init_background();
+                       goto out;
+               } else {
+                       ret = EINVAL;
+                       goto out;
                }
+       } else {
+               ret = EINVAL;
+               goto out;
        }
+
+out:
+       lck_mtx_unlock(ktrace_lock);
+       return ret;
 }
 
-/*
- * Return true if caller has permission to set the ktracing state
- * of target.  Essentially, the target can't possess any
- * more permissions than the caller.  KTRFAC_ROOT signifies that
- * root previously set the tracing status on the target process, and 
- * so, only root may further change it.
- *
- * TODO: check groups.  use caller effective gid.
- */
-ktrcanset(callp, targetp)
-       struct proc *callp, *targetp;
+/* This should only be called from the bootstrap thread. */
+void
+ktrace_init(void)
 {
-       register struct pcred *caller = callp->p_cred;
-       register struct pcred *target = targetp->p_cred;
-
-       if ((caller->pc_ucred->cr_uid == target->p_ruid &&
-            target->p_ruid == target->p_svuid &&
-            caller->p_rgid == target->p_rgid &&        /* XXX */
-            target->p_rgid == target->p_svgid &&
-            (targetp->p_traceflag & KTRFAC_ROOT) == 0) ||
-            caller->pc_ucred->cr_uid == 0)
-               return (1);
-
-       return (0);
-}
+       static lck_grp_attr_t *lock_grp_attr = NULL;
+       static lck_grp_t *lock_grp = NULL;
+       static boolean_t initialized = FALSE;
 
-#endif
+       if (initialized) {
+               return;
+       }
+
+       lock_grp_attr = lck_grp_attr_alloc_init();
+       lock_grp = lck_grp_alloc_init("ktrace", lock_grp_attr);
+       lck_grp_attr_free(lock_grp_attr);
+
+       ktrace_lock = lck_mtx_alloc_init(lock_grp, LCK_ATTR_NULL);
+       assert(ktrace_lock);
+       initialized = TRUE;
+}