19454b2d8SWarner Losh /*- 28a36da99SPedro F. Giffuni * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 38a36da99SPedro F. Giffuni * 426f9a767SRodney W. Grimes * Copyright (c) 1993, David Greenman 526f9a767SRodney W. Grimes * All rights reserved. 6df8bae1dSRodney W. Grimes * 7df8bae1dSRodney W. Grimes * Redistribution and use in source and binary forms, with or without 8df8bae1dSRodney W. Grimes * modification, are permitted provided that the following conditions 9df8bae1dSRodney W. Grimes * are met: 10df8bae1dSRodney W. Grimes * 1. Redistributions of source code must retain the above copyright 11df8bae1dSRodney W. Grimes * notice, this list of conditions and the following disclaimer. 12df8bae1dSRodney W. Grimes * 2. Redistributions in binary form must reproduce the above copyright 13df8bae1dSRodney W. Grimes * notice, this list of conditions and the following disclaimer in the 14df8bae1dSRodney W. Grimes * documentation and/or other materials provided with the distribution. 15df8bae1dSRodney W. Grimes * 1626f9a767SRodney W. Grimes * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17df8bae1dSRodney W. Grimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18df8bae1dSRodney W. Grimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 1926f9a767SRodney W. Grimes * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20df8bae1dSRodney W. Grimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21df8bae1dSRodney W. Grimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22df8bae1dSRodney W. Grimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23df8bae1dSRodney W. Grimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24df8bae1dSRodney W. Grimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25df8bae1dSRodney W. Grimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26df8bae1dSRodney W. Grimes * SUCH DAMAGE. 27df8bae1dSRodney W. Grimes */ 28df8bae1dSRodney W. Grimes 29677b542eSDavid E. O'Brien #include <sys/cdefs.h> 30677b542eSDavid E. O'Brien __FBSDID("$FreeBSD$"); 31677b542eSDavid E. O'Brien 3212bc222eSJonathan Anderson #include "opt_capsicum.h" 334da0d332SPeter Wemm #include "opt_hwpmc_hooks.h" 341e9b3d91SPeter Wemm #include "opt_ktrace.h" 35f8a47341SAlan Cox #include "opt_vm.h" 361e9b3d91SPeter Wemm 37df8bae1dSRodney W. Grimes #include <sys/param.h> 3826f9a767SRodney W. Grimes #include <sys/systm.h> 3926f9a767SRodney W. Grimes #include <sys/acct.h> 40eeeaa7baSMark Johnston #include <sys/capsicum.h> 41eeeaa7baSMark Johnston #include <sys/eventhandler.h> 4226f9a767SRodney W. Grimes #include <sys/exec.h> 43eeeaa7baSMark Johnston #include <sys/fcntl.h> 44eeeaa7baSMark Johnston #include <sys/filedesc.h> 4526f9a767SRodney W. Grimes #include <sys/imgact.h> 46e1743d02SSøren Schmidt #include <sys/imgact_elf.h> 47eeeaa7baSMark Johnston #include <sys/kernel.h> 48eeeaa7baSMark Johnston #include <sys/lock.h> 49333ea485SGuido van Rooij #include <sys/malloc.h> 50ec492b13SMark Johnston #include <sys/mman.h> 51eeeaa7baSMark Johnston #include <sys/mount.h> 52eeeaa7baSMark Johnston #include <sys/mutex.h> 53eeeaa7baSMark Johnston #include <sys/namei.h> 54eeeaa7baSMark Johnston #include <sys/pioctl.h> 55acd3428bSRobert Watson #include <sys/priv.h> 56a794e791SBruce Evans #include <sys/proc.h> 578d570f64SJohn Baldwin #include <sys/ptrace.h> 586004362eSDavid Schultz #include <sys/resourcevar.h> 5989f6b863SAttilio Rao #include <sys/rwlock.h> 6044ad5475SJohn Baldwin #include <sys/sched.h> 615d217f17SJohn Birrell #include <sys/sdt.h> 6259c8bc40SAlan Cox #include <sys/sf_buf.h> 63797f2d22SPoul-Henning Kamp #include <sys/shm.h> 64eeeaa7baSMark Johnston #include <sys/signalvar.h> 65ec492b13SMark Johnston #include <sys/smp.h> 66dfa7fd1dSEdward Tomasz Napierala #include <sys/stat.h> 67eeeaa7baSMark Johnston #include <sys/syscallsubr.h> 68eeeaa7baSMark Johnston #include <sys/sysctl.h> 69eeeaa7baSMark Johnston #include <sys/sysent.h> 70eeeaa7baSMark Johnston #include <sys/sysproto.h> 71eeeaa7baSMark Johnston #include <sys/vnode.h> 72eeeaa7baSMark Johnston #include <sys/wait.h> 73c781aea8SPeter Wemm #ifdef KTRACE 74c781aea8SPeter Wemm #include <sys/ktrace.h> 75c781aea8SPeter Wemm #endif 7626f9a767SRodney W. Grimes 7726f9a767SRodney W. Grimes #include <vm/vm.h> 78efeaf95aSDavid Greenman #include <vm/vm_param.h> 79efeaf95aSDavid Greenman #include <vm/pmap.h> 801616db3cSJohn Dyson #include <vm/vm_page.h> 81efeaf95aSDavid Greenman #include <vm/vm_map.h> 8226f9a767SRodney W. Grimes #include <vm/vm_kern.h> 83efeaf95aSDavid Greenman #include <vm/vm_extern.h> 841ebd0c59SDavid Greenman #include <vm/vm_object.h> 851616db3cSJohn Dyson #include <vm/vm_pager.h> 8626f9a767SRodney W. Grimes 87ebccf1e3SJoseph Koshy #ifdef HWPMC_HOOKS 88ebccf1e3SJoseph Koshy #include <sys/pmckern.h> 89ebccf1e3SJoseph Koshy #endif 90ebccf1e3SJoseph Koshy 9126f9a767SRodney W. Grimes #include <machine/reg.h> 9226f9a767SRodney W. Grimes 93ae1078d6SWayne Salamon #include <security/audit/audit.h> 94aed55708SRobert Watson #include <security/mac/mac_framework.h> 95ae1078d6SWayne Salamon 965d217f17SJohn Birrell #ifdef KDTRACE_HOOKS 975d217f17SJohn Birrell #include <sys/dtrace_bsd.h> 985d217f17SJohn Birrell dtrace_execexit_func_t dtrace_fasttrap_exec; 995d217f17SJohn Birrell #endif 1005d217f17SJohn Birrell 1015d217f17SJohn Birrell SDT_PROVIDER_DECLARE(proc); 10236160958SMark Johnston SDT_PROBE_DEFINE1(proc, , , exec, "char *"); 10336160958SMark Johnston SDT_PROBE_DEFINE1(proc, , , exec__failure, "int"); 10436160958SMark Johnston SDT_PROBE_DEFINE1(proc, , , exec__success, "char *"); 1055d217f17SJohn Birrell 106b9df5231SPoul-Henning Kamp MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 107b9df5231SPoul-Henning Kamp 108bcb60d52SConrad Meyer int coredump_pack_fileinfo = 1; 109bcb60d52SConrad Meyer SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN, 110bcb60d52SConrad Meyer &coredump_pack_fileinfo, 0, 111bcb60d52SConrad Meyer "Enable file path packing in 'procstat -f' coredump notes"); 112bcb60d52SConrad Meyer 113e6b95927SConrad Meyer int coredump_pack_vmmapinfo = 1; 114e6b95927SConrad Meyer SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN, 115e6b95927SConrad Meyer &coredump_pack_vmmapinfo, 0, 116e6b95927SConrad Meyer "Enable file path packing in 'procstat -v' coredump notes"); 117e6b95927SConrad Meyer 11805ba50f5SJake Burkholder static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS); 11905ba50f5SJake Burkholder static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS); 1205dadd17bSJake Burkholder static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS); 121610ecfe0SMaxim Sobolev static int do_execve(struct thread *td, struct image_args *args, 122610ecfe0SMaxim Sobolev struct mac *mac_p); 12305ba50f5SJake Burkholder 1249701cd40SJohn Baldwin /* XXX This should be vm_size_t. */ 12553dc58f2SMateusz Guzik SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD| 1262fe6aeffSMariusz Zaborski CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", ""); 127486bddb0SDoug Rabson 1289701cd40SJohn Baldwin /* XXX This should be vm_size_t. */ 129ff66f6a4SRobert Watson SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD| 13053dc58f2SMateusz Guzik CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", ""); 13199ac3bc8SPeter Wemm 13253dc58f2SMateusz Guzik SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE, 1335dadd17bSJake Burkholder NULL, 0, sysctl_kern_stackprot, "I", ""); 1345dadd17bSJake Burkholder 135b9df5231SPoul-Henning Kamp u_long ps_arg_cache_limit = PAGE_SIZE / 16; 136c3699b5fSPeter Wemm SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 1379701cd40SJohn Baldwin &ps_arg_cache_limit, 0, ""); 138b9df5231SPoul-Henning Kamp 139eda6009cSKonstantin Belousov static int disallow_high_osrel; 140eda6009cSKonstantin Belousov SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW, 141eda6009cSKonstantin Belousov &disallow_high_osrel, 0, 142eda6009cSKonstantin Belousov "Disallow execution of binaries built for higher version of the world"); 143eda6009cSKonstantin Belousov 144878adb85SBjoern A. Zeeb static int map_at_zero = 0; 145af3b2549SHans Petter Selasky SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0, 146878adb85SBjoern A. Zeeb "Permit processes to map an object at virtual address 0."); 147878adb85SBjoern A. Zeeb 1482ca45184SMatt Joras EVENTHANDLER_LIST_DECLARE(process_exec); 1492ca45184SMatt Joras 15005ba50f5SJake Burkholder static int 15105ba50f5SJake Burkholder sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) 15205ba50f5SJake Burkholder { 15305ba50f5SJake Burkholder struct proc *p; 154df7c361eSPeter Wemm int error; 15505ba50f5SJake Burkholder 15605ba50f5SJake Burkholder p = curproc; 157a7bc3102SPeter Wemm #ifdef SCTL_MASK32 158a7bc3102SPeter Wemm if (req->flags & SCTL_MASK32) { 159df7c361eSPeter Wemm unsigned int val; 160df7c361eSPeter Wemm val = (unsigned int)p->p_sysent->sv_psstrings; 161df7c361eSPeter Wemm error = SYSCTL_OUT(req, &val, sizeof(val)); 162df7c361eSPeter Wemm } else 163df7c361eSPeter Wemm #endif 164df7c361eSPeter Wemm error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings, 165df7c361eSPeter Wemm sizeof(p->p_sysent->sv_psstrings)); 166df7c361eSPeter Wemm return error; 16705ba50f5SJake Burkholder } 16805ba50f5SJake Burkholder 16905ba50f5SJake Burkholder static int 17005ba50f5SJake Burkholder sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) 17105ba50f5SJake Burkholder { 17205ba50f5SJake Burkholder struct proc *p; 173df7c361eSPeter Wemm int error; 17405ba50f5SJake Burkholder 17505ba50f5SJake Burkholder p = curproc; 176a7bc3102SPeter Wemm #ifdef SCTL_MASK32 177a7bc3102SPeter Wemm if (req->flags & SCTL_MASK32) { 178df7c361eSPeter Wemm unsigned int val; 179df7c361eSPeter Wemm val = (unsigned int)p->p_sysent->sv_usrstack; 180df7c361eSPeter Wemm error = SYSCTL_OUT(req, &val, sizeof(val)); 181df7c361eSPeter Wemm } else 182df7c361eSPeter Wemm #endif 183df7c361eSPeter Wemm error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack, 184df7c361eSPeter Wemm sizeof(p->p_sysent->sv_usrstack)); 185df7c361eSPeter Wemm return error; 18605ba50f5SJake Burkholder } 18705ba50f5SJake Burkholder 1885dadd17bSJake Burkholder static int 1895dadd17bSJake Burkholder sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) 1905dadd17bSJake Burkholder { 1915dadd17bSJake Burkholder struct proc *p; 1925dadd17bSJake Burkholder 1935dadd17bSJake Burkholder p = curproc; 1945dadd17bSJake Burkholder return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, 1955dadd17bSJake Burkholder sizeof(p->p_sysent->sv_stackprot))); 1965dadd17bSJake Burkholder } 1975dadd17bSJake Burkholder 19899ac3bc8SPeter Wemm /* 199aa855a59SPeter Wemm * Each of the items is a pointer to a `const struct execsw', hence the 200aa855a59SPeter Wemm * double pointer here. 201df8bae1dSRodney W. Grimes */ 202aa855a59SPeter Wemm static const struct execsw **execsw; 20326f9a767SRodney W. Grimes 204ca46e90eSBruce Evans #ifndef _SYS_SYSPROTO_H_ 205ca46e90eSBruce Evans struct execve_args { 206ca46e90eSBruce Evans char *fname; 207ca46e90eSBruce Evans char **argv; 208ca46e90eSBruce Evans char **envv; 209ca46e90eSBruce Evans }; 210ca46e90eSBruce Evans #endif 211ca46e90eSBruce Evans 212ca46e90eSBruce Evans int 2137b445033SKonstantin Belousov sys_execve(struct thread *td, struct execve_args *uap) 214ca46e90eSBruce Evans { 215610ecfe0SMaxim Sobolev struct image_args args; 2167b445033SKonstantin Belousov struct vmspace *oldvmspace; 2177b445033SKonstantin Belousov int error; 218ca46e90eSBruce Evans 2197b445033SKonstantin Belousov error = pre_execve(td, &oldvmspace); 2207b445033SKonstantin Belousov if (error != 0) 2217b445033SKonstantin Belousov return (error); 222610ecfe0SMaxim Sobolev error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 223610ecfe0SMaxim Sobolev uap->argv, uap->envv); 224610ecfe0SMaxim Sobolev if (error == 0) 225610ecfe0SMaxim Sobolev error = kern_execve(td, &args, NULL); 2267b445033SKonstantin Belousov post_execve(td, error, oldvmspace); 227610ecfe0SMaxim Sobolev return (error); 228ca46e90eSBruce Evans } 229ca46e90eSBruce Evans 230ca46e90eSBruce Evans #ifndef _SYS_SYSPROTO_H_ 231632dbc19SKonstantin Belousov struct fexecve_args { 232632dbc19SKonstantin Belousov int fd; 233632dbc19SKonstantin Belousov char **argv; 234632dbc19SKonstantin Belousov char **envv; 235632dbc19SKonstantin Belousov } 236632dbc19SKonstantin Belousov #endif 237632dbc19SKonstantin Belousov int 2388451d0ddSKip Macy sys_fexecve(struct thread *td, struct fexecve_args *uap) 239632dbc19SKonstantin Belousov { 240632dbc19SKonstantin Belousov struct image_args args; 2417b445033SKonstantin Belousov struct vmspace *oldvmspace; 2427b445033SKonstantin Belousov int error; 243632dbc19SKonstantin Belousov 2447b445033SKonstantin Belousov error = pre_execve(td, &oldvmspace); 2457b445033SKonstantin Belousov if (error != 0) 2467b445033SKonstantin Belousov return (error); 247632dbc19SKonstantin Belousov error = exec_copyin_args(&args, NULL, UIO_SYSSPACE, 248632dbc19SKonstantin Belousov uap->argv, uap->envv); 249632dbc19SKonstantin Belousov if (error == 0) { 250632dbc19SKonstantin Belousov args.fd = uap->fd; 251632dbc19SKonstantin Belousov error = kern_execve(td, &args, NULL); 252632dbc19SKonstantin Belousov } 2537b445033SKonstantin Belousov post_execve(td, error, oldvmspace); 254632dbc19SKonstantin Belousov return (error); 255632dbc19SKonstantin Belousov } 256632dbc19SKonstantin Belousov 257632dbc19SKonstantin Belousov #ifndef _SYS_SYSPROTO_H_ 258ca46e90eSBruce Evans struct __mac_execve_args { 259ca46e90eSBruce Evans char *fname; 260ca46e90eSBruce Evans char **argv; 261ca46e90eSBruce Evans char **envv; 262ca46e90eSBruce Evans struct mac *mac_p; 263ca46e90eSBruce Evans }; 264ca46e90eSBruce Evans #endif 265ca46e90eSBruce Evans 266ca46e90eSBruce Evans int 2677b445033SKonstantin Belousov sys___mac_execve(struct thread *td, struct __mac_execve_args *uap) 268ca46e90eSBruce Evans { 269ca46e90eSBruce Evans #ifdef MAC 270610ecfe0SMaxim Sobolev struct image_args args; 2717b445033SKonstantin Belousov struct vmspace *oldvmspace; 2727b445033SKonstantin Belousov int error; 273610ecfe0SMaxim Sobolev 2747b445033SKonstantin Belousov error = pre_execve(td, &oldvmspace); 2757b445033SKonstantin Belousov if (error != 0) 2767b445033SKonstantin Belousov return (error); 277610ecfe0SMaxim Sobolev error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 278610ecfe0SMaxim Sobolev uap->argv, uap->envv); 279610ecfe0SMaxim Sobolev if (error == 0) 280610ecfe0SMaxim Sobolev error = kern_execve(td, &args, uap->mac_p); 2817b445033SKonstantin Belousov post_execve(td, error, oldvmspace); 282610ecfe0SMaxim Sobolev return (error); 283ca46e90eSBruce Evans #else 284ca46e90eSBruce Evans return (ENOSYS); 285ca46e90eSBruce Evans #endif 286ca46e90eSBruce Evans } 287ca46e90eSBruce Evans 28884e0b075SDavid Xu int 2897b445033SKonstantin Belousov pre_execve(struct thread *td, struct vmspace **oldvmspace) 290906ac69dSDavid Xu { 2917b445033SKonstantin Belousov struct proc *p; 292906ac69dSDavid Xu int error; 293906ac69dSDavid Xu 2947b445033SKonstantin Belousov KASSERT(td == curthread, ("non-current thread %p", td)); 2957b445033SKonstantin Belousov error = 0; 2967b445033SKonstantin Belousov p = td->td_proc; 2977b445033SKonstantin Belousov if ((p->p_flag & P_HADTHREADS) != 0) { 298906ac69dSDavid Xu PROC_LOCK(p); 2997b445033SKonstantin Belousov if (thread_single(p, SINGLE_BOUNDARY) != 0) 3007b445033SKonstantin Belousov error = ERESTART; 301906ac69dSDavid Xu PROC_UNLOCK(p); 302906ac69dSDavid Xu } 3037b445033SKonstantin Belousov KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0, 3047b445033SKonstantin Belousov ("nested execve")); 3057b445033SKonstantin Belousov *oldvmspace = p->p_vmspace; 3067b445033SKonstantin Belousov return (error); 3077b445033SKonstantin Belousov } 308906ac69dSDavid Xu 3097b445033SKonstantin Belousov void 3107b445033SKonstantin Belousov post_execve(struct thread *td, int error, struct vmspace *oldvmspace) 3117b445033SKonstantin Belousov { 3127b445033SKonstantin Belousov struct proc *p; 313906ac69dSDavid Xu 3147b445033SKonstantin Belousov KASSERT(td == curthread, ("non-current thread %p", td)); 3157b445033SKonstantin Belousov p = td->td_proc; 3167b445033SKonstantin Belousov if ((p->p_flag & P_HADTHREADS) != 0) { 317906ac69dSDavid Xu PROC_LOCK(p); 318906ac69dSDavid Xu /* 319906ac69dSDavid Xu * If success, we upgrade to SINGLE_EXIT state to 320906ac69dSDavid Xu * force other threads to suicide. 321906ac69dSDavid Xu */ 322814629ddSEd Schouten if (error == EJUSTRETURN) 3236ddcc233SKonstantin Belousov thread_single(p, SINGLE_EXIT); 324906ac69dSDavid Xu else 3256ddcc233SKonstantin Belousov thread_single_end(p, SINGLE_BOUNDARY); 326906ac69dSDavid Xu PROC_UNLOCK(p); 327906ac69dSDavid Xu } 3287032434eSKonstantin Belousov if ((td->td_pflags & TDP_EXECVMSPC) != 0) { 3297b445033SKonstantin Belousov KASSERT(p->p_vmspace != oldvmspace, 3307032434eSKonstantin Belousov ("oldvmspace still used")); 3317032434eSKonstantin Belousov vmspace_free(oldvmspace); 3327032434eSKonstantin Belousov td->td_pflags &= ~TDP_EXECVMSPC; 3337032434eSKonstantin Belousov } 3347b445033SKonstantin Belousov } 335906ac69dSDavid Xu 3367b445033SKonstantin Belousov /* 3377b445033SKonstantin Belousov * XXX: kern_execve has the astonishing property of not always returning to 3387b445033SKonstantin Belousov * the caller. If sufficiently bad things happen during the call to 3397b445033SKonstantin Belousov * do_execve(), it can end up calling exit1(); as a result, callers must 3407b445033SKonstantin Belousov * avoid doing anything which they might need to undo (e.g., allocating 3417b445033SKonstantin Belousov * memory). 3427b445033SKonstantin Belousov */ 3437b445033SKonstantin Belousov int 3447b445033SKonstantin Belousov kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p) 3457b445033SKonstantin Belousov { 3467b445033SKonstantin Belousov 3477b445033SKonstantin Belousov AUDIT_ARG_ARGV(args->begin_argv, args->argc, 348f373437aSBrooks Davis exec_args_get_begin_envv(args) - args->begin_argv); 349f373437aSBrooks Davis AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc, 350f373437aSBrooks Davis args->endp - exec_args_get_begin_envv(args)); 3517b445033SKonstantin Belousov return (do_execve(td, args, mac_p)); 352906ac69dSDavid Xu } 353906ac69dSDavid Xu 35426f9a767SRodney W. Grimes /* 355450ffb44SRobert Watson * In-kernel implementation of execve(). All arguments are assumed to be 356450ffb44SRobert Watson * userspace pointers from the passed thread. 35726f9a767SRodney W. Grimes */ 358450ffb44SRobert Watson static int 3593e85b721SEd Maste do_execve(struct thread *td, struct image_args *args, struct mac *mac_p) 360df8bae1dSRodney W. Grimes { 361b40ce416SJulian Elischer struct proc *p = td->td_proc; 36269c9eff8SEd Schouten struct nameidata nd; 363881010f0SBryan Drewery struct ucred *oldcred; 364350d5181SMateusz Guzik struct uidinfo *euip = NULL; 365654f6be1SBruce Evans register_t *stack_base; 3665ca4819dSJohn Baldwin int error, i; 367c52007c2SDavid Greenman struct image_params image_params, *imgp; 3689f5c1d19SDiomidis Spinellis struct vattr attr; 3694d77a549SAlfred Perlstein int (*img_first)(struct image_params *); 37069be5db9SAlfred Perlstein struct pargs *oldargs = NULL, *newargs = NULL; 371cda68844SMateusz Guzik struct sigacts *oldsigacts = NULL, *newsigacts = NULL; 3726c84de02SJohn Baldwin #ifdef KTRACE 3736c84de02SJohn Baldwin struct vnode *tracevp = NULL; 374a5881ea5SJohn Baldwin struct ucred *tracecred = NULL; 3756c84de02SJohn Baldwin #endif 37661617058SMateusz Guzik struct vnode *oldtextvp = NULL, *newtextvp; 377d06c0d4dSRobert Watson int credential_changing; 378ccafe7ebSRobert Watson #ifdef MAC 3796356dba0SRobert Watson struct label *interpvplabel = NULL; 380eca8a663SRobert Watson int will_transition; 381ccafe7ebSRobert Watson #endif 38215139246SJoseph Koshy #ifdef HWPMC_HOOKS 38315139246SJoseph Koshy struct pmckern_procexec pe; 38415139246SJoseph Koshy #endif 385632dbc19SKonstantin Belousov static const char fexecv_proc_title[] = "(fexecv)"; 38626f9a767SRodney W. Grimes 387c52007c2SDavid Greenman imgp = &image_params; 388df8bae1dSRodney W. Grimes 3899ca45e81SDag-Erling Smørgrav /* 3909ca45e81SDag-Erling Smørgrav * Lock the process and set the P_INEXEC flag to indicate that 3919ca45e81SDag-Erling Smørgrav * it should be left alone until we're done here. This is 3929ca45e81SDag-Erling Smørgrav * necessary to avoid race conditions - e.g. in ptrace() - 3939ca45e81SDag-Erling Smørgrav * that might allow a local user to illicitly obtain elevated 3949ca45e81SDag-Erling Smørgrav * privileges. 3959ca45e81SDag-Erling Smørgrav */ 3969ca45e81SDag-Erling Smørgrav PROC_LOCK(p); 3979ca45e81SDag-Erling Smørgrav KASSERT((p->p_flag & P_INEXEC) == 0, 39891f91617SDavid E. O'Brien ("%s(): process already has P_INEXEC flag", __func__)); 3999ca45e81SDag-Erling Smørgrav p->p_flag |= P_INEXEC; 4009ca45e81SDag-Erling Smørgrav PROC_UNLOCK(p); 4019ca45e81SDag-Erling Smørgrav 402df8bae1dSRodney W. Grimes /* 403c52007c2SDavid Greenman * Initialize part of the common data 404df8bae1dSRodney W. Grimes */ 4058e572983SMateusz Guzik bzero(imgp, sizeof(*imgp)); 406c52007c2SDavid Greenman imgp->proc = p; 407c52007c2SDavid Greenman imgp->attr = &attr; 408610ecfe0SMaxim Sobolev imgp->args = args; 409881010f0SBryan Drewery oldcred = p->p_ucred; 41026f9a767SRodney W. Grimes 411670cb89bSRobert Watson #ifdef MAC 412eca8a663SRobert Watson error = mac_execve_enter(imgp, mac_p); 413532c491bSJeff Roberson if (error) 414670cb89bSRobert Watson goto exec_fail; 415670cb89bSRobert Watson #endif 416670cb89bSRobert Watson 41726f9a767SRodney W. Grimes /* 41826f9a767SRodney W. Grimes * Translate the file name. namei() returns a vnode pointer 419e3043798SPedro F. Giffuni * in ni_vp among other things. 420ae1078d6SWayne Salamon * 421ae1078d6SWayne Salamon * XXXAUDIT: It would be desirable to also audit the name of the 422ae1078d6SWayne Salamon * interpreter if this is an interpreted binary. 42326f9a767SRodney W. Grimes */ 424632dbc19SKonstantin Belousov if (args->fname != NULL) { 425*78022527SKonstantin Belousov NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW | 426*78022527SKonstantin Belousov SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td); 427632dbc19SKonstantin Belousov } 42826f9a767SRodney W. Grimes 42936160958SMark Johnston SDT_PROBE1(proc, , , exec, args->fname); 430e506f34bSCraig Rodrigues 43126f9a767SRodney W. Grimes interpret: 432632dbc19SKonstantin Belousov if (args->fname != NULL) { 43312bc222eSJonathan Anderson #ifdef CAPABILITY_MODE 43412bc222eSJonathan Anderson /* 43512bc222eSJonathan Anderson * While capability mode can't reach this point via direct 43612bc222eSJonathan Anderson * path arguments to execve(), we also don't allow 43712bc222eSJonathan Anderson * interpreters to be used in capability mode (for now). 43812bc222eSJonathan Anderson * Catch indirect lookups and return a permissions error. 43912bc222eSJonathan Anderson */ 44012bc222eSJonathan Anderson if (IN_CAPABILITY_MODE(td)) { 44112bc222eSJonathan Anderson error = ECAPMODE; 44212bc222eSJonathan Anderson goto exec_fail; 44312bc222eSJonathan Anderson } 44412bc222eSJonathan Anderson #endif 44569c9eff8SEd Schouten error = namei(&nd); 446610ecfe0SMaxim Sobolev if (error) 44726f9a767SRodney W. Grimes goto exec_fail; 44826f9a767SRodney W. Grimes 44961617058SMateusz Guzik newtextvp = nd.ni_vp; 45061617058SMateusz Guzik imgp->vp = newtextvp; 451632dbc19SKonstantin Belousov } else { 45214961ba7SRobert Watson AUDIT_ARG_FD(args->fd); 453a9d2f8d8SRobert Watson /* 454c8e781f6SPawel Jakub Dawidek * Descriptors opened only with O_EXEC or O_RDONLY are allowed. 455a9d2f8d8SRobert Watson */ 456cbd92ce6SMatt Macy error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp); 457632dbc19SKonstantin Belousov if (error) 458632dbc19SKonstantin Belousov goto exec_fail; 459*78022527SKonstantin Belousov vn_lock(newtextvp, LK_SHARED | LK_RETRY); 46061617058SMateusz Guzik AUDIT_ARG_VNODE1(newtextvp); 46161617058SMateusz Guzik imgp->vp = newtextvp; 462632dbc19SKonstantin Belousov } 46326f9a767SRodney W. Grimes 46426f9a767SRodney W. Grimes /* 465*78022527SKonstantin Belousov * Check file permissions. Also 'opens' file and sets its vnode to 466*78022527SKonstantin Belousov * text mode. 4679022e4adSDavid Greenman */ 468c52007c2SDavid Greenman error = exec_check_permissions(imgp); 469619eb6e5SJeff Roberson if (error) 47026f9a767SRodney W. Grimes goto exec_fail_dealloc; 471619eb6e5SJeff Roberson 4728516dd18SPoul-Henning Kamp imgp->object = imgp->vp->v_object; 4738516dd18SPoul-Henning Kamp if (imgp->object != NULL) 4740b2ed1aeSJeff Roberson vm_object_reference(imgp->object); 47526f9a767SRodney W. Grimes 4761616db3cSJohn Dyson error = exec_map_first_page(imgp); 4776d5a0a8cSDavid Greenman if (error) 47826f9a767SRodney W. Grimes goto exec_fail_dealloc; 47926f9a767SRodney W. Grimes 480f231de47SKonstantin Belousov imgp->proc->p_osrel = 0; 481f5cf7589SKonstantin Belousov imgp->proc->p_fctl0 = 0; 482881010f0SBryan Drewery 483881010f0SBryan Drewery /* 484881010f0SBryan Drewery * Implement image setuid/setgid. 485881010f0SBryan Drewery * 486881010f0SBryan Drewery * Determine new credentials before attempting image activators 487881010f0SBryan Drewery * so that it can be used by process_exec handlers to determine 488881010f0SBryan Drewery * credential/setid changes. 489881010f0SBryan Drewery * 490881010f0SBryan Drewery * Don't honor setuid/setgid if the filesystem prohibits it or if 491881010f0SBryan Drewery * the process is being traced. 492881010f0SBryan Drewery * 493881010f0SBryan Drewery * We disable setuid/setgid/etc in capability mode on the basis 494881010f0SBryan Drewery * that most setugid applications are not written with that 495881010f0SBryan Drewery * environment in mind, and will therefore almost certainly operate 496881010f0SBryan Drewery * incorrectly. In principle there's no reason that setugid 497881010f0SBryan Drewery * applications might not be useful in capability mode, so we may want 498881010f0SBryan Drewery * to reconsider this conservative design choice in the future. 499881010f0SBryan Drewery * 500881010f0SBryan Drewery * XXXMAC: For the time being, use NOSUID to also prohibit 501881010f0SBryan Drewery * transitions on the file system. 502881010f0SBryan Drewery */ 503881010f0SBryan Drewery credential_changing = 0; 504881010f0SBryan Drewery credential_changing |= (attr.va_mode & S_ISUID) && 505881010f0SBryan Drewery oldcred->cr_uid != attr.va_uid; 506881010f0SBryan Drewery credential_changing |= (attr.va_mode & S_ISGID) && 507881010f0SBryan Drewery oldcred->cr_gid != attr.va_gid; 508881010f0SBryan Drewery #ifdef MAC 509881010f0SBryan Drewery will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 510881010f0SBryan Drewery interpvplabel, imgp); 511881010f0SBryan Drewery credential_changing |= will_transition; 512881010f0SBryan Drewery #endif 513881010f0SBryan Drewery 5141302eea7SKonstantin Belousov /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */ 515b9408863SKonstantin Belousov if (credential_changing) 516b9408863SKonstantin Belousov imgp->proc->p_pdeathsig = 0; 517b9408863SKonstantin Belousov 518881010f0SBryan Drewery if (credential_changing && 519881010f0SBryan Drewery #ifdef CAPABILITY_MODE 520881010f0SBryan Drewery ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 521881010f0SBryan Drewery #endif 522881010f0SBryan Drewery (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 523881010f0SBryan Drewery (p->p_flag & P_TRACED) == 0) { 524881010f0SBryan Drewery imgp->credential_setid = true; 525881010f0SBryan Drewery VOP_UNLOCK(imgp->vp, 0); 526881010f0SBryan Drewery imgp->newcred = crdup(oldcred); 527881010f0SBryan Drewery if (attr.va_mode & S_ISUID) { 528881010f0SBryan Drewery euip = uifind(attr.va_uid); 529881010f0SBryan Drewery change_euid(imgp->newcred, euip); 530881010f0SBryan Drewery } 531881010f0SBryan Drewery vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 532881010f0SBryan Drewery if (attr.va_mode & S_ISGID) 533881010f0SBryan Drewery change_egid(imgp->newcred, attr.va_gid); 534881010f0SBryan Drewery /* 535881010f0SBryan Drewery * Implement correct POSIX saved-id behavior. 536881010f0SBryan Drewery * 537881010f0SBryan Drewery * XXXMAC: Note that the current logic will save the 538881010f0SBryan Drewery * uid and gid if a MAC domain transition occurs, even 539881010f0SBryan Drewery * though maybe it shouldn't. 540881010f0SBryan Drewery */ 541881010f0SBryan Drewery change_svuid(imgp->newcred, imgp->newcred->cr_uid); 542881010f0SBryan Drewery change_svgid(imgp->newcred, imgp->newcred->cr_gid); 543881010f0SBryan Drewery } else { 544881010f0SBryan Drewery /* 545881010f0SBryan Drewery * Implement correct POSIX saved-id behavior. 546881010f0SBryan Drewery * 547881010f0SBryan Drewery * XXX: It's not clear that the existing behavior is 548881010f0SBryan Drewery * POSIX-compliant. A number of sources indicate that the 549881010f0SBryan Drewery * saved uid/gid should only be updated if the new ruid is 550881010f0SBryan Drewery * not equal to the old ruid, or the new euid is not equal 551881010f0SBryan Drewery * to the old euid and the new euid is not equal to the old 552881010f0SBryan Drewery * ruid. The FreeBSD code always updates the saved uid/gid. 553881010f0SBryan Drewery * Also, this code uses the new (replaced) euid and egid as 554881010f0SBryan Drewery * the source, which may or may not be the right ones to use. 555881010f0SBryan Drewery */ 556881010f0SBryan Drewery if (oldcred->cr_svuid != oldcred->cr_uid || 557881010f0SBryan Drewery oldcred->cr_svgid != oldcred->cr_gid) { 558881010f0SBryan Drewery VOP_UNLOCK(imgp->vp, 0); 559881010f0SBryan Drewery imgp->newcred = crdup(oldcred); 560881010f0SBryan Drewery vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 561881010f0SBryan Drewery change_svuid(imgp->newcred, imgp->newcred->cr_uid); 562881010f0SBryan Drewery change_svgid(imgp->newcred, imgp->newcred->cr_gid); 563881010f0SBryan Drewery } 564881010f0SBryan Drewery } 565881010f0SBryan Drewery /* The new credentials are installed into the process later. */ 566881010f0SBryan Drewery 56726f9a767SRodney W. Grimes /* 5681afd78b3SBryan Drewery * Do the best to calculate the full path to the image file. 5691afd78b3SBryan Drewery */ 5701afd78b3SBryan Drewery if (args->fname != NULL && args->fname[0] == '/') 5711afd78b3SBryan Drewery imgp->execpath = args->fname; 5721afd78b3SBryan Drewery else { 5731afd78b3SBryan Drewery VOP_UNLOCK(imgp->vp, 0); 5741afd78b3SBryan Drewery if (vn_fullpath(td, imgp->vp, &imgp->execpath, 5751afd78b3SBryan Drewery &imgp->freepath) != 0) 5761afd78b3SBryan Drewery imgp->execpath = args->fname; 5771afd78b3SBryan Drewery vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 5781afd78b3SBryan Drewery } 5791afd78b3SBryan Drewery 5801afd78b3SBryan Drewery /* 581d323ddf3SMatthew Dillon * If the current process has a special image activator it 582d323ddf3SMatthew Dillon * wants to try first, call it. For example, emulating shell 583d323ddf3SMatthew Dillon * scripts differently. 58426f9a767SRodney W. Grimes */ 585d323ddf3SMatthew Dillon error = -1; 586d323ddf3SMatthew Dillon if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 587d323ddf3SMatthew Dillon error = img_first(imgp); 588d323ddf3SMatthew Dillon 589d323ddf3SMatthew Dillon /* 590d323ddf3SMatthew Dillon * Loop through the list of image activators, calling each one. 591d323ddf3SMatthew Dillon * An activator returns -1 if there is no match, 0 on success, 592d323ddf3SMatthew Dillon * and an error otherwise. 593d323ddf3SMatthew Dillon */ 594d323ddf3SMatthew Dillon for (i = 0; error == -1 && execsw[i]; ++i) { 595d323ddf3SMatthew Dillon if (execsw[i]->ex_imgact == NULL || 596d323ddf3SMatthew Dillon execsw[i]->ex_imgact == img_first) { 597d323ddf3SMatthew Dillon continue; 598d323ddf3SMatthew Dillon } 599c52007c2SDavid Greenman error = (*execsw[i]->ex_imgact)(imgp); 600d323ddf3SMatthew Dillon } 601d323ddf3SMatthew Dillon 602d323ddf3SMatthew Dillon if (error) { 603*78022527SKonstantin Belousov if (error == -1) 604d323ddf3SMatthew Dillon error = ENOEXEC; 60526f9a767SRodney W. Grimes goto exec_fail_dealloc; 606d323ddf3SMatthew Dillon } 607d323ddf3SMatthew Dillon 608d323ddf3SMatthew Dillon /* 609d323ddf3SMatthew Dillon * Special interpreter operation, cleanup and loop up to try to 610d323ddf3SMatthew Dillon * activate the interpreter. 611d323ddf3SMatthew Dillon */ 612c52007c2SDavid Greenman if (imgp->interpreted) { 6131616db3cSJohn Dyson exec_unmap_first_page(imgp); 614619eb6e5SJeff Roberson /* 615*78022527SKonstantin Belousov * The text reference needs to be removed for scripts. 616*78022527SKonstantin Belousov * There is a short period before we determine that 617*78022527SKonstantin Belousov * something is a script where text reference is active. 618*78022527SKonstantin Belousov * The vnode lock is held over this entire period 619*78022527SKonstantin Belousov * so nothing should illegitimately be blocked. 620619eb6e5SJeff Roberson */ 621*78022527SKonstantin Belousov VOP_UNSET_TEXT_CHECKED(imgp->vp); 622762e6b85SEivind Eklund /* free name buffer and old vnode */ 623632dbc19SKonstantin Belousov if (args->fname != NULL) 62469c9eff8SEd Schouten NDFREE(&nd, NDF_ONLY_PNBUF); 625670cb89bSRobert Watson #ifdef MAC 62661617058SMateusz Guzik mac_execve_interpreter_enter(newtextvp, &interpvplabel); 627670cb89bSRobert Watson #endif 6289a75ea23SKonstantin Belousov if (imgp->opened) { 62961617058SMateusz Guzik VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td); 6309a75ea23SKonstantin Belousov imgp->opened = 0; 6319a75ea23SKonstantin Belousov } 63261617058SMateusz Guzik vput(newtextvp); 6330b2ed1aeSJeff Roberson vm_object_deallocate(imgp->object); 6340b2ed1aeSJeff Roberson imgp->object = NULL; 635881010f0SBryan Drewery imgp->credential_setid = false; 636881010f0SBryan Drewery if (imgp->newcred != NULL) { 637881010f0SBryan Drewery crfree(imgp->newcred); 638881010f0SBryan Drewery imgp->newcred = NULL; 639881010f0SBryan Drewery } 6401afd78b3SBryan Drewery imgp->execpath = NULL; 6411afd78b3SBryan Drewery free(imgp->freepath, M_TEMP); 6421afd78b3SBryan Drewery imgp->freepath = NULL; 64326f9a767SRodney W. Grimes /* set new name to that of the interpreter */ 6445050aa86SKonstantin Belousov NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME, 645b40ce416SJulian Elischer UIO_SYSSPACE, imgp->interpreter_name, td); 646632dbc19SKonstantin Belousov args->fname = imgp->interpreter_name; 64726f9a767SRodney W. Grimes goto interpret; 64826f9a767SRodney W. Grimes } 64926f9a767SRodney W. Grimes 65026f9a767SRodney W. Grimes /* 651ded7d39cSChristian S.J. Peron * NB: We unlock the vnode here because it is believed that none 652ded7d39cSChristian S.J. Peron * of the sv_copyout_strings/sv_fixup operations require the vnode. 653ded7d39cSChristian S.J. Peron */ 654ded7d39cSChristian S.J. Peron VOP_UNLOCK(imgp->vp, 0); 6553ff06357SKonstantin Belousov 656eda6009cSKonstantin Belousov if (disallow_high_osrel && 657eda6009cSKonstantin Belousov P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) { 658eda6009cSKonstantin Belousov error = ENOEXEC; 659eda6009cSKonstantin Belousov uprintf("Osrel %d for image %s too high\n", p->p_osrel, 660eda6009cSKonstantin Belousov imgp->execpath != NULL ? imgp->execpath : "<unresolved>"); 661eda6009cSKonstantin Belousov vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 662eda6009cSKonstantin Belousov goto exec_fail_dealloc; 663eda6009cSKonstantin Belousov } 664eda6009cSKonstantin Belousov 66539f5ebb7SEd Schouten /* ABI enforces the use of Capsicum. Switch into capabilities mode. */ 66639f5ebb7SEd Schouten if (SV_PROC_FLAG(p, SV_CAPSICUM)) 66739f5ebb7SEd Schouten sys_cap_enter(td, NULL); 66839f5ebb7SEd Schouten 669ded7d39cSChristian S.J. Peron /* 67091ff2d48SEdward Tomasz Napierala * Copy out strings (args and env) and initialize stack base. 67126f9a767SRodney W. Grimes */ 6723ebc1248SPeter Wemm stack_base = (*p->p_sysent->sv_copyout_strings)(imgp); 67326f9a767SRodney W. Grimes 67426f9a767SRodney W. Grimes /* 67591ff2d48SEdward Tomasz Napierala * Stack setup. 67626f9a767SRodney W. Grimes */ 6775f77b8a8SBrooks Davis error = (*p->p_sysent->sv_fixup)(&stack_base, imgp); 6786f26dd50SKonstantin Belousov if (error != 0) { 6796f26dd50SKonstantin Belousov vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 6805f77b8a8SBrooks Davis goto exec_fail_dealloc; 6816f26dd50SKonstantin Belousov } 68226f9a767SRodney W. Grimes 683457f7e23SEd Schouten if (args->fdp != NULL) { 684457f7e23SEd Schouten /* Install a brand new file descriptor table. */ 685457f7e23SEd Schouten fdinstall_remapped(td, args->fdp); 686457f7e23SEd Schouten args->fdp = NULL; 687457f7e23SEd Schouten } else { 688a78e8d2aSDavid Greenman /* 689457f7e23SEd Schouten * Keep on using the existing file descriptor table. For 690457f7e23SEd Schouten * security and other reasons, the file descriptor table 691457f7e23SEd Schouten * cannot be shared after an exec. 692a78e8d2aSDavid Greenman */ 693b9d32c36SMateusz Guzik fdunshare(td); 694b0bc0cadSMateusz Guzik /* close files on exec */ 695b0bc0cadSMateusz Guzik fdcloseexec(td); 696457f7e23SEd Schouten } 697a78e8d2aSDavid Greenman 698333ea485SGuido van Rooij /* 6999b3b1c5fSJohn Baldwin * Malloc things before we need locks. 7009b3b1c5fSJohn Baldwin */ 701f373437aSBrooks Davis i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv; 7025997cae9SDon Lewis /* Cache arguments if they fit inside our allowance */ 7035997cae9SDon Lewis if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 7049b3b1c5fSJohn Baldwin newargs = pargs_alloc(i); 7055997cae9SDon Lewis bcopy(imgp->args->begin_argv, newargs->ar_args, i); 7065997cae9SDon Lewis } 7079b3b1c5fSJohn Baldwin 7089b3b1c5fSJohn Baldwin /* 709333ea485SGuido van Rooij * For security and other reasons, signal handlers cannot 7108688bb93SJohn Baldwin * be shared after an exec. The new process gets a copy of the old 711b2c3fa70SDima Dorfman * handlers. In execsigs(), the new process will have its signals 712333ea485SGuido van Rooij * reset. 713333ea485SGuido van Rooij */ 71490af4afaSJohn Baldwin if (sigacts_shared(p->p_sigacts)) { 71590af4afaSJohn Baldwin oldsigacts = p->p_sigacts; 71690af4afaSJohn Baldwin newsigacts = sigacts_alloc(); 71790af4afaSJohn Baldwin sigacts_copy(newsigacts, oldsigacts); 718a6bad85eSMateusz Guzik } 719333ea485SGuido van Rooij 720cda68844SMateusz Guzik vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 721cda68844SMateusz Guzik 722d00c8ea4SMateusz Guzik PROC_LOCK(p); 723d00c8ea4SMateusz Guzik if (oldsigacts) 724d00c8ea4SMateusz Guzik p->p_sigacts = newsigacts; 725fdf4e8b3SWarner Losh /* Stop profiling */ 726fdf4e8b3SWarner Losh stopprofclock(p); 727fdf4e8b3SWarner Losh 72826f9a767SRodney W. Grimes /* reset caught signals */ 72926f9a767SRodney W. Grimes execsigs(p); 73026f9a767SRodney W. Grimes 73126f9a767SRodney W. Grimes /* name this process - nameiexec(p, ndp) */ 7325ca4819dSJohn Baldwin bzero(p->p_comm, sizeof(p->p_comm)); 7335ca4819dSJohn Baldwin if (args->fname) 7345ca4819dSJohn Baldwin bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, 7355ca4819dSJohn Baldwin min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); 73661617058SMateusz Guzik else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0) 7375ca4819dSJohn Baldwin bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); 738ca081fdbSJulian Elischer bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 73944ad5475SJohn Baldwin #ifdef KTR 74044ad5475SJohn Baldwin sched_clear_tdname(td); 74144ad5475SJohn Baldwin #endif 74226f9a767SRodney W. Grimes 74326f9a767SRodney W. Grimes /* 74424b34f09SSujal Patel * mark as execed, wakeup the process that vforked (if any) and tell 745dc733423SDag-Erling Smørgrav * it that it now has its own resources back 74626f9a767SRodney W. Grimes */ 74726f9a767SRodney W. Grimes p->p_flag |= P_EXEC; 748677258f7SKonstantin Belousov if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0) 749677258f7SKonstantin Belousov p->p_flag2 &= ~P2_NOTRACE; 750965d0860SMateusz Guzik if (p->p_flag & P_PPWAIT) { 751888d4d4fSKonstantin Belousov p->p_flag &= ~(P_PPWAIT | P_PPTRACE); 752aeb32571SKonstantin Belousov cv_broadcast(&p->p_pwait); 75377d68094SKonstantin Belousov /* STOPs are no longer ignored, arrange for AST */ 75477d68094SKonstantin Belousov signotify(td); 75526f9a767SRodney W. Grimes } 75626f9a767SRodney W. Grimes 75726f9a767SRodney W. Grimes /* 758881010f0SBryan Drewery * Implement image setuid/setgid installation. 759c52007c2SDavid Greenman */ 760881010f0SBryan Drewery if (imgp->credential_setid) { 761c52007c2SDavid Greenman /* 762c52007c2SDavid Greenman * Turn off syscall tracing for set-id programs, except for 763b85db196SPeter Wemm * root. Record any set-id flags first to make sure that 764b85db196SPeter Wemm * we do not regain any tracing during a possible block. 76526f9a767SRodney W. Grimes */ 766b85db196SPeter Wemm setsugid(p); 767acd3428bSRobert Watson 7686c84de02SJohn Baldwin #ifdef KTRACE 769c331c970SAndriy Gapon if (p->p_tracecred != NULL && 770cc426dd3SMateusz Guzik priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED)) 771d680caabSJohn Baldwin ktrprocexec(p, &tracecred, &tracevp); 7726c84de02SJohn Baldwin #endif 77328b325aaSDon Lewis /* 774c1e2d386SNate Lawson * Close any file descriptors 0..2 that reference procfs, 775c1e2d386SNate Lawson * then make sure file descriptors 0..2 are in use. 77628b325aaSDon Lewis * 7778a5177ccSMateusz Guzik * Both fdsetugidsafety() and fdcheckstd() may call functions 7788a5177ccSMateusz Guzik * taking sleepable locks, so temporarily drop our locks. 77928b325aaSDon Lewis */ 78028b325aaSDon Lewis PROC_UNLOCK(p); 78122db15c0SAttilio Rao VOP_UNLOCK(imgp->vp, 0); 78211888da8SMateusz Guzik fdsetugidsafety(td); 783e983a376SJacques Vidrine error = fdcheckstd(td); 7845e2554b7SMateusz Guzik vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 785cda68844SMateusz Guzik if (error != 0) 786cda68844SMateusz Guzik goto exec_fail_dealloc; 787e1b1aa3bSJohn Baldwin PROC_LOCK(p); 788ccafe7ebSRobert Watson #ifdef MAC 789670cb89bSRobert Watson if (will_transition) { 790881010f0SBryan Drewery mac_vnode_execve_transition(oldcred, imgp->newcred, 791881010f0SBryan Drewery imgp->vp, interpvplabel, imgp); 792670cb89bSRobert Watson } 793ccafe7ebSRobert Watson #endif 794c52007c2SDavid Greenman } else { 795b1fc0ec1SRobert Watson if (oldcred->cr_uid == oldcred->cr_ruid && 796b1fc0ec1SRobert Watson oldcred->cr_gid == oldcred->cr_rgid) 797c52007c2SDavid Greenman p->p_flag &= ~P_SUGID; 798881010f0SBryan Drewery } 79926f9a767SRodney W. Grimes /* 800881010f0SBryan Drewery * Set the new credentials. 80126f9a767SRodney W. Grimes */ 8023fc292d5SKonstantin Belousov if (imgp->newcred != NULL) { 803881010f0SBryan Drewery proc_set_cred(p, imgp->newcred); 8043fc292d5SKonstantin Belousov crfree(oldcred); 8053fc292d5SKonstantin Belousov oldcred = NULL; 8063fc292d5SKonstantin Belousov } 80726f9a767SRodney W. Grimes 80826f9a767SRodney W. Grimes /* 809853be5ffSMateusz Guzik * Store the vp for use in procfs. This vnode was referenced by namei 810853be5ffSMateusz Guzik * or fgetvp_exec. 8112a531c80SDavid Greenman */ 81261617058SMateusz Guzik oldtextvp = p->p_textvp; 81361617058SMateusz Guzik p->p_textvp = newtextvp; 8142a531c80SDavid Greenman 8155d217f17SJohn Birrell #ifdef KDTRACE_HOOKS 8165d217f17SJohn Birrell /* 8175d217f17SJohn Birrell * Tell the DTrace fasttrap provider about the exec if it 8185d217f17SJohn Birrell * has declared an interest. 8195d217f17SJohn Birrell */ 8205d217f17SJohn Birrell if (dtrace_fasttrap_exec) 8215d217f17SJohn Birrell dtrace_fasttrap_exec(p); 8225d217f17SJohn Birrell #endif 8235d217f17SJohn Birrell 8242a531c80SDavid Greenman /* 8259ca45e81SDag-Erling Smørgrav * Notify others that we exec'd, and clear the P_INEXEC flag 8269ca45e81SDag-Erling Smørgrav * as we're now a bona fide freshly-execed process. 827cb679c38SJonathan Lemon */ 8289e590ff0SKonstantin Belousov KNOTE_LOCKED(p->p_klist, NOTE_EXEC); 8299ca45e81SDag-Erling Smørgrav p->p_flag &= ~P_INEXEC; 830cb679c38SJonathan Lemon 83126f9a767SRodney W. Grimes /* clear "fork but no exec" flag, as we _are_ execing */ 83226f9a767SRodney W. Grimes p->p_acflag &= ~AFORK; 83326f9a767SRodney W. Grimes 8345997cae9SDon Lewis /* 83534ea500bSDon Lewis * Free any previous argument cache and replace it with 8365997cae9SDon Lewis * the new argument cache, if any. 8375997cae9SDon Lewis */ 8389b3b1c5fSJohn Baldwin oldargs = p->p_args; 839e1b1aa3bSJohn Baldwin p->p_args = newargs; 840e1b1aa3bSJohn Baldwin newargs = NULL; 841ebccf1e3SJoseph Koshy 84280a2397aSMateusz Guzik PROC_UNLOCK(p); 84380a2397aSMateusz Guzik 844ebccf1e3SJoseph Koshy #ifdef HWPMC_HOOKS 845ebccf1e3SJoseph Koshy /* 846f263522aSJoseph Koshy * Check if system-wide sampling is in effect or if the 847f263522aSJoseph Koshy * current process is using PMCs. If so, do exec() time 848ebccf1e3SJoseph Koshy * processing. This processing needs to happen AFTER the 849ebccf1e3SJoseph Koshy * P_INEXEC flag is cleared. 850ebccf1e3SJoseph Koshy */ 851f263522aSJoseph Koshy if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 8521ef6ea9bSKonstantin Belousov VOP_UNLOCK(imgp->vp, 0); 85315139246SJoseph Koshy pe.pm_credentialschanged = credential_changing; 85415139246SJoseph Koshy pe.pm_entryaddr = imgp->entry_addr; 85515139246SJoseph Koshy 85615139246SJoseph Koshy PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 8572974cc36SKonstantin Belousov vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 85880a2397aSMateusz Guzik } 859ebccf1e3SJoseph Koshy #endif 860e1b1aa3bSJohn Baldwin 861e913ca22SDoug Rabson /* Set values passed into the program in registers. */ 86291ff2d48SEdward Tomasz Napierala (*p->p_sysent->sv_setregs)(td, imgp, (u_long)(uintptr_t)stack_base); 863e913ca22SDoug Rabson 8640359a12eSAttilio Rao vfs_mark_atime(imgp->vp, td->td_ucred); 8656341095eSKen Smith 86636160958SMark Johnston SDT_PROBE1(proc, , , exec__success, args->fname); 8675d217f17SJohn Birrell 8681616db3cSJohn Dyson exec_fail_dealloc: 869d6c847f3SBruce Evans if (imgp->firstpage != NULL) 8701616db3cSJohn Dyson exec_unmap_first_page(imgp); 87126f9a767SRodney W. Grimes 872d6c847f3SBruce Evans if (imgp->vp != NULL) { 873632dbc19SKonstantin Belousov if (args->fname) 87469c9eff8SEd Schouten NDFREE(&nd, NDF_ONLY_PNBUF); 8759a75ea23SKonstantin Belousov if (imgp->opened) 8769a75ea23SKonstantin Belousov VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 877*78022527SKonstantin Belousov if (imgp->textset) 878*78022527SKonstantin Belousov VOP_UNSET_TEXT_CHECKED(imgp->vp); 879853be5ffSMateusz Guzik if (error != 0) 880619eb6e5SJeff Roberson vput(imgp->vp); 881853be5ffSMateusz Guzik else 882853be5ffSMateusz Guzik VOP_UNLOCK(imgp->vp, 0); 883a3cf6ebaSDavid Greenman } 88426f9a767SRodney W. Grimes 885d6c847f3SBruce Evans if (imgp->object != NULL) 8860b2ed1aeSJeff Roberson vm_object_deallocate(imgp->object); 8870b2ed1aeSJeff Roberson 8883ff06357SKonstantin Belousov free(imgp->freepath, M_TEMP); 8893ff06357SKonstantin Belousov 890d1989db5SRobert Drehmel if (error == 0) { 891e6b645aeSMateusz Guzik if (p->p_ptevents & PTRACE_EXEC) { 892afe1a688SKonstantin Belousov PROC_LOCK(p); 8938d570f64SJohn Baldwin if (p->p_ptevents & PTRACE_EXEC) 894afe1a688SKonstantin Belousov td->td_dbgflags |= TDB_EXEC; 895afe1a688SKonstantin Belousov PROC_UNLOCK(p); 896e6b645aeSMateusz Guzik } 897afe1a688SKonstantin Belousov 898d1989db5SRobert Drehmel /* 899d1989db5SRobert Drehmel * Stop the process here if its stop event mask has 900d1989db5SRobert Drehmel * the S_EXEC bit set. 901d1989db5SRobert Drehmel */ 902d1989db5SRobert Drehmel STOPEVENT(p, S_EXEC, 0); 903cda68844SMateusz Guzik } else { 90426f9a767SRodney W. Grimes exec_fail: 9059ca45e81SDag-Erling Smørgrav /* we're done here, clear P_INEXEC */ 9069ca45e81SDag-Erling Smørgrav PROC_LOCK(p); 9079ca45e81SDag-Erling Smørgrav p->p_flag &= ~P_INEXEC; 9089ca45e81SDag-Erling Smørgrav PROC_UNLOCK(p); 9099ca45e81SDag-Erling Smørgrav 91036160958SMark Johnston SDT_PROBE1(proc, , , exec__failure, error); 911cda68844SMateusz Guzik } 9125d217f17SJohn Birrell 9133fc292d5SKonstantin Belousov if (imgp->newcred != NULL && oldcred != NULL) 9143fc292d5SKonstantin Belousov crfree(imgp->newcred); 9153fc292d5SKonstantin Belousov 916670cb89bSRobert Watson #ifdef MAC 917670cb89bSRobert Watson mac_execve_exit(imgp); 9186356dba0SRobert Watson mac_execve_interpreter_exit(interpvplabel); 919670cb89bSRobert Watson #endif 9208917b8d2SJohn Baldwin exec_free_args(args); 9218917b8d2SJohn Baldwin 922cda68844SMateusz Guzik /* 923cda68844SMateusz Guzik * Handle deferred decrement of ref counts. 924cda68844SMateusz Guzik */ 925cda68844SMateusz Guzik if (oldtextvp != NULL) 926cda68844SMateusz Guzik vrele(oldtextvp); 927cda68844SMateusz Guzik #ifdef KTRACE 928cda68844SMateusz Guzik if (tracevp != NULL) 929cda68844SMateusz Guzik vrele(tracevp); 930cda68844SMateusz Guzik if (tracecred != NULL) 931cda68844SMateusz Guzik crfree(tracecred); 932cda68844SMateusz Guzik #endif 933cda68844SMateusz Guzik pargs_drop(oldargs); 934cda68844SMateusz Guzik pargs_drop(newargs); 935cda68844SMateusz Guzik if (oldsigacts != NULL) 936cda68844SMateusz Guzik sigacts_free(oldsigacts); 937cda68844SMateusz Guzik if (euip != NULL) 938cda68844SMateusz Guzik uifree(euip); 939cda68844SMateusz Guzik 9408917b8d2SJohn Baldwin if (error && imgp->vmspace_destroyed) { 9418917b8d2SJohn Baldwin /* sorry, no more process anymore. exit gracefully */ 942b4490c6eSKonstantin Belousov exit1(td, 0, SIGABRT); 9438917b8d2SJohn Baldwin /* NOT REACHED */ 9448917b8d2SJohn Baldwin } 9457705d4b2SDmitry Chagin 9467705d4b2SDmitry Chagin #ifdef KTRACE 9477705d4b2SDmitry Chagin if (error == 0) 9487705d4b2SDmitry Chagin ktrprocctor(p); 9497705d4b2SDmitry Chagin #endif 9507705d4b2SDmitry Chagin 951814629ddSEd Schouten /* 952814629ddSEd Schouten * We don't want cpu_set_syscall_retval() to overwrite any of 953814629ddSEd Schouten * the register values put in place by exec_setregs(). 954814629ddSEd Schouten * Implementations of cpu_set_syscall_retval() will leave 955814629ddSEd Schouten * registers unmodified when returning EJUSTRETURN. 956814629ddSEd Schouten */ 957814629ddSEd Schouten return (error == 0 ? EJUSTRETURN : error); 95826f9a767SRodney W. Grimes } 95926f9a767SRodney W. Grimes 9601616db3cSJohn Dyson int 961b8d908b7SEd Maste exec_map_first_page(struct image_params *imgp) 9621616db3cSJohn Dyson { 963b0cd2017SGleb Smirnoff int rv, i, after, initial_pagein; 96495461b45SJohn Dyson vm_page_t ma[VM_INITIAL_PAGEIN]; 9651616db3cSJohn Dyson vm_object_t object; 9661616db3cSJohn Dyson 967d6c847f3SBruce Evans if (imgp->firstpage != NULL) 9681616db3cSJohn Dyson exec_unmap_first_page(imgp); 9691616db3cSJohn Dyson 9708516dd18SPoul-Henning Kamp object = imgp->vp->v_object; 9714d7d2993SJeff Roberson if (object == NULL) 9724d7d2993SJeff Roberson return (EACCES); 97389f6b863SAttilio Rao VM_OBJECT_WLOCK(object); 974f8a47341SAlan Cox #if VM_NRESERVLEVEL > 0 9753d653db0SAlan Cox vm_object_color(object, 0); 976f8a47341SAlan Cox #endif 977ce3ee09bSAlan Cox ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY); 9780a2e596aSAlan Cox if (ma[0]->valid != VM_PAGE_BITS_ALL) { 979ce3ee09bSAlan Cox vm_page_xbusy(ma[0]); 980b0cd2017SGleb Smirnoff if (!vm_pager_has_page(object, 0, NULL, &after)) { 981b0cd2017SGleb Smirnoff vm_page_lock(ma[0]); 982b0cd2017SGleb Smirnoff vm_page_free(ma[0]); 983b0cd2017SGleb Smirnoff vm_page_unlock(ma[0]); 984b0cd2017SGleb Smirnoff VM_OBJECT_WUNLOCK(object); 985b0cd2017SGleb Smirnoff return (EIO); 986b0cd2017SGleb Smirnoff } 987b0cd2017SGleb Smirnoff initial_pagein = min(after, VM_INITIAL_PAGEIN); 988b0cd2017SGleb Smirnoff KASSERT(initial_pagein <= object->size, 989b0cd2017SGleb Smirnoff ("%s: initial_pagein %d object->size %ju", 990b0cd2017SGleb Smirnoff __func__, initial_pagein, (uintmax_t )object->size)); 99195461b45SJohn Dyson for (i = 1; i < initial_pagein; i++) { 99241890423SAlan Cox if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) { 9936ec2fca5SAlan Cox if (ma[i]->valid) 9946ec2fca5SAlan Cox break; 995b2387aa6SAndriy Gapon if (!vm_page_tryxbusy(ma[i])) 99606fa71cdSAlan Cox break; 99795461b45SJohn Dyson } else { 998ca0387efSJake Burkholder ma[i] = vm_page_alloc(object, i, 9997667839aSAlan Cox VM_ALLOC_NORMAL); 100095461b45SJohn Dyson if (ma[i] == NULL) 100195461b45SJohn Dyson break; 100295461b45SJohn Dyson } 100395461b45SJohn Dyson } 100495461b45SJohn Dyson initial_pagein = i; 1005b0cd2017SGleb Smirnoff rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL); 1006093c7f39SGleb Smirnoff if (rv != VM_PAGER_OK) { 1007b0cd2017SGleb Smirnoff for (i = 0; i < initial_pagein; i++) { 1008b0cd2017SGleb Smirnoff vm_page_lock(ma[i]); 1009b0cd2017SGleb Smirnoff vm_page_free(ma[i]); 1010b0cd2017SGleb Smirnoff vm_page_unlock(ma[i]); 1011b0cd2017SGleb Smirnoff } 101289f6b863SAttilio Rao VM_OBJECT_WUNLOCK(object); 1013a7cddfedSJake Burkholder return (EIO); 10141616db3cSJohn Dyson } 1015ce3ee09bSAlan Cox vm_page_xunbusy(ma[0]); 1016b0cd2017SGleb Smirnoff for (i = 1; i < initial_pagein; i++) 1017b0cd2017SGleb Smirnoff vm_page_readahead_finish(ma[i]); 10181616db3cSJohn Dyson } 10192965a453SKip Macy vm_page_lock(ma[0]); 1020be996836SAttilio Rao vm_page_hold(ma[0]); 102170978c93SKonstantin Belousov vm_page_activate(ma[0]); 10222965a453SKip Macy vm_page_unlock(ma[0]); 102389f6b863SAttilio Rao VM_OBJECT_WUNLOCK(object); 10241616db3cSJohn Dyson 102559c8bc40SAlan Cox imgp->firstpage = sf_buf_alloc(ma[0], 0); 102659c8bc40SAlan Cox imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 10271616db3cSJohn Dyson 1028a7cddfedSJake Burkholder return (0); 10291616db3cSJohn Dyson } 10301616db3cSJohn Dyson 10311616db3cSJohn Dyson void 10323e85b721SEd Maste exec_unmap_first_page(struct image_params *imgp) 10331616db3cSJohn Dyson { 103459c8bc40SAlan Cox vm_page_t m; 103523955314SAlfred Perlstein 1036d6c847f3SBruce Evans if (imgp->firstpage != NULL) { 103759c8bc40SAlan Cox m = sf_buf_page(imgp->firstpage); 103859c8bc40SAlan Cox sf_buf_free(imgp->firstpage); 10391616db3cSJohn Dyson imgp->firstpage = NULL; 10402965a453SKip Macy vm_page_lock(m); 1041be996836SAttilio Rao vm_page_unhold(m); 10422965a453SKip Macy vm_page_unlock(m); 10431616db3cSJohn Dyson } 10441616db3cSJohn Dyson } 10451616db3cSJohn Dyson 104626f9a767SRodney W. Grimes /* 1047467b3975SKonstantin Belousov * Destroy old address space, and allocate a new stack. 1048467b3975SKonstantin Belousov * The new stack is only sgrowsiz large because it is grown 1049467b3975SKonstantin Belousov * automatically on a page fault. 105026f9a767SRodney W. Grimes */ 105126f9a767SRodney W. Grimes int 10523e85b721SEd Maste exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv) 105326f9a767SRodney W. Grimes { 105426f9a767SRodney W. Grimes int error; 10555e20c11fSAlan Cox struct proc *p = imgp->proc; 10565e20c11fSAlan Cox struct vmspace *vmspace = p->p_vmspace; 10576297a3d8SKonstantin Belousov vm_object_t obj; 1058316b3843SKonstantin Belousov struct rlimit rlim_stack; 1059878adb85SBjoern A. Zeeb vm_offset_t sv_minuser, stack_addr; 106005ba50f5SJake Burkholder vm_map_t map; 106159d8f3ffSJohn Baldwin u_long ssiz; 106226f9a767SRodney W. Grimes 1063c52007c2SDavid Greenman imgp->vmspace_destroyed = 1; 1064993182e5SAlexander Leidinger imgp->sysent = sv; 106526f9a767SRodney W. Grimes 1066bd37fd72SKris Kennaway /* May be called with Giant held */ 10672ca45184SMatt Joras EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp); 10686f5dafeaSAlan Cox 10696f5dafeaSAlan Cox /* 1070af9ec885SJohn Dyson * Blow away entire process VM, if address space not shared, 1071af9ec885SJohn Dyson * otherwise, create a new VM space so that other threads are 1072af9ec885SJohn Dyson * not disrupted 1073af9ec885SJohn Dyson */ 107405ba50f5SJake Burkholder map = &vmspace->vm_map; 1075878adb85SBjoern A. Zeeb if (map_at_zero) 1076878adb85SBjoern A. Zeeb sv_minuser = sv->sv_minuser; 1077878adb85SBjoern A. Zeeb else 1078878adb85SBjoern A. Zeeb sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1079878adb85SBjoern A. Zeeb if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser && 10806f1fe330SKonstantin Belousov vm_map_max(map) == sv->sv_maxuser && 10816f1fe330SKonstantin Belousov cpu_exec_vmspace_reuse(p, map)) { 10823db161e0SMatthew Dillon shmexit(vmspace); 1083b9eee07eSPeter Wemm pmap_remove_pages(vmspace_pmap(vmspace)); 108405ba50f5SJake Burkholder vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1085fa50a355SKonstantin Belousov /* 1086fa50a355SKonstantin Belousov * An exec terminates mlockall(MCL_FUTURE), ASLR state 1087fa50a355SKonstantin Belousov * must be re-evaluated. 1088fa50a355SKonstantin Belousov */ 10898056df6eSAlan Cox vm_map_lock(map); 1090fa50a355SKonstantin Belousov vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR | 1091fa50a355SKonstantin Belousov MAP_ASLR_IGNSTART); 10928056df6eSAlan Cox vm_map_unlock(map); 1093af9ec885SJohn Dyson } else { 1094878adb85SBjoern A. Zeeb error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 109589b57fcfSKonstantin Belousov if (error) 109689b57fcfSKonstantin Belousov return (error); 10975e20c11fSAlan Cox vmspace = p->p_vmspace; 109805ba50f5SJake Burkholder map = &vmspace->vm_map; 1099af9ec885SJohn Dyson } 1100fa50a355SKonstantin Belousov map->flags |= imgp->map_flags; 110126f9a767SRodney W. Grimes 11026297a3d8SKonstantin Belousov /* Map a shared page */ 11036297a3d8SKonstantin Belousov obj = sv->sv_shared_page_obj; 11046297a3d8SKonstantin Belousov if (obj != NULL) { 11056297a3d8SKonstantin Belousov vm_object_reference(obj); 11066297a3d8SKonstantin Belousov error = vm_map_fixed(map, obj, 0, 11076297a3d8SKonstantin Belousov sv->sv_shared_page_base, sv->sv_shared_page_len, 11081e65d73cSKonstantin Belousov VM_PROT_READ | VM_PROT_EXECUTE, 11091e65d73cSKonstantin Belousov VM_PROT_READ | VM_PROT_EXECUTE, 11101e65d73cSKonstantin Belousov MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1111467b3975SKonstantin Belousov if (error != KERN_SUCCESS) { 11126297a3d8SKonstantin Belousov vm_object_deallocate(obj); 1113467b3975SKonstantin Belousov return (vm_mmap_to_errno(error)); 11146297a3d8SKonstantin Belousov } 11156297a3d8SKonstantin Belousov } 11166297a3d8SKonstantin Belousov 111726f9a767SRodney W. Grimes /* Allocate a new stack */ 1118316b3843SKonstantin Belousov if (imgp->stack_sz != 0) { 1119fe8a824cSKonstantin Belousov ssiz = trunc_page(imgp->stack_sz); 1120316b3843SKonstantin Belousov PROC_LOCK(p); 1121f6f6d240SMateusz Guzik lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack); 1122316b3843SKonstantin Belousov PROC_UNLOCK(p); 1123316b3843SKonstantin Belousov if (ssiz > rlim_stack.rlim_max) 1124316b3843SKonstantin Belousov ssiz = rlim_stack.rlim_max; 1125316b3843SKonstantin Belousov if (ssiz > rlim_stack.rlim_cur) { 1126316b3843SKonstantin Belousov rlim_stack.rlim_cur = ssiz; 1127316b3843SKonstantin Belousov kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack); 1128316b3843SKonstantin Belousov } 1129316b3843SKonstantin Belousov } else if (sv->sv_maxssiz != NULL) { 113059d8f3ffSJohn Baldwin ssiz = *sv->sv_maxssiz; 1131316b3843SKonstantin Belousov } else { 113259d8f3ffSJohn Baldwin ssiz = maxssiz; 1133316b3843SKonstantin Belousov } 113459d8f3ffSJohn Baldwin stack_addr = sv->sv_usrstack - ssiz; 113559d8f3ffSJohn Baldwin error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 11366297a3d8SKonstantin Belousov obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1137467b3975SKonstantin Belousov sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1138467b3975SKonstantin Belousov if (error != KERN_SUCCESS) 1139467b3975SKonstantin Belousov return (vm_mmap_to_errno(error)); 11402267af78SJulian Elischer 114178960940SAlan Cox /* 114278960940SAlan Cox * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they 114378960940SAlan Cox * are still used to enforce the stack rlimit on the process stack. 11442267af78SJulian Elischer */ 1145cbc89bfbSPaul Saab vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 11466ef12002SKonstantin Belousov vmspace->vm_maxsaddr = (char *)stack_addr; 114726f9a767SRodney W. Grimes 114826f9a767SRodney W. Grimes return (0); 114926f9a767SRodney W. Grimes } 115026f9a767SRodney W. Grimes 115126f9a767SRodney W. Grimes /* 1152873fbcd7SRobert Watson * Copy out argument and environment strings from the old process address 1153873fbcd7SRobert Watson * space into the temporary string buffer. 115426f9a767SRodney W. Grimes */ 115526f9a767SRodney W. Grimes int 115612e69f96SBrooks Davis exec_copyin_args(struct image_args *args, const char *fname, 1157610ecfe0SMaxim Sobolev enum uio_seg segflg, char **argv, char **envv) 115826f9a767SRodney W. Grimes { 1159f373437aSBrooks Davis u_long arg, env; 1160ecbb00a2SDoug Rabson int error; 116126f9a767SRodney W. Grimes 1162610ecfe0SMaxim Sobolev bzero(args, sizeof(*args)); 1163610ecfe0SMaxim Sobolev if (argv == NULL) 1164610ecfe0SMaxim Sobolev return (EFAULT); 11652af6e14dSAlan Cox 116626f9a767SRodney W. Grimes /* 11672af6e14dSAlan Cox * Allocate demand-paged memory for the file name, argument, and 11682af6e14dSAlan Cox * environment strings. 116926f9a767SRodney W. Grimes */ 11702af6e14dSAlan Cox error = exec_alloc_args(args); 11712af6e14dSAlan Cox if (error != 0) 11722af6e14dSAlan Cox return (error); 11739e4e5114SAlan Cox 1174610ecfe0SMaxim Sobolev /* 1175610ecfe0SMaxim Sobolev * Copy the file name. 1176610ecfe0SMaxim Sobolev */ 1177f373437aSBrooks Davis error = exec_args_add_fname(args, fname, segflg); 1178c30af532SMaxim Sobolev if (error != 0) 117968ff3c24SStephan Uphoff goto err_exit; 118026f9a767SRodney W. Grimes 1181610ecfe0SMaxim Sobolev /* 1182610ecfe0SMaxim Sobolev * extract arguments first 1183610ecfe0SMaxim Sobolev */ 11840a2c94b8SKonstantin Belousov for (;;) { 1185f373437aSBrooks Davis error = fueword(argv++, &arg); 11860a2c94b8SKonstantin Belousov if (error == -1) { 118768ff3c24SStephan Uphoff error = EFAULT; 118868ff3c24SStephan Uphoff goto err_exit; 118968ff3c24SStephan Uphoff } 1190f373437aSBrooks Davis if (arg == 0) 11910a2c94b8SKonstantin Belousov break; 1192f373437aSBrooks Davis error = exec_args_add_arg(args, (char *)(uintptr_t)arg, 1193f373437aSBrooks Davis UIO_USERSPACE); 1194f373437aSBrooks Davis if (error != 0) 119568ff3c24SStephan Uphoff goto err_exit; 1196610ecfe0SMaxim Sobolev } 1197b9df5231SPoul-Henning Kamp 119826f9a767SRodney W. Grimes /* 119926f9a767SRodney W. Grimes * extract environment strings 120026f9a767SRodney W. Grimes */ 12010fd1a014SDavid Greenman if (envv) { 12020a2c94b8SKonstantin Belousov for (;;) { 1203f373437aSBrooks Davis error = fueword(envv++, &env); 12040a2c94b8SKonstantin Belousov if (error == -1) { 120568ff3c24SStephan Uphoff error = EFAULT; 120668ff3c24SStephan Uphoff goto err_exit; 120768ff3c24SStephan Uphoff } 1208f373437aSBrooks Davis if (env == 0) 12090a2c94b8SKonstantin Belousov break; 1210f373437aSBrooks Davis error = exec_args_add_env(args, 1211f373437aSBrooks Davis (char *)(uintptr_t)env, UIO_USERSPACE); 1212f373437aSBrooks Davis if (error != 0) 121368ff3c24SStephan Uphoff goto err_exit; 12140fd1a014SDavid Greenman } 12150fd1a014SDavid Greenman } 121626f9a767SRodney W. Grimes 121726f9a767SRodney W. Grimes return (0); 121868ff3c24SStephan Uphoff 121968ff3c24SStephan Uphoff err_exit: 122068ff3c24SStephan Uphoff exec_free_args(args); 122168ff3c24SStephan Uphoff return (error); 122226f9a767SRodney W. Grimes } 122326f9a767SRodney W. Grimes 1224457f7e23SEd Schouten int 1225457f7e23SEd Schouten exec_copyin_data_fds(struct thread *td, struct image_args *args, 1226457f7e23SEd Schouten const void *data, size_t datalen, const int *fds, size_t fdslen) 1227457f7e23SEd Schouten { 1228457f7e23SEd Schouten struct filedesc *ofdp; 1229457f7e23SEd Schouten const char *p; 1230457f7e23SEd Schouten int *kfds; 1231457f7e23SEd Schouten int error; 1232457f7e23SEd Schouten 1233457f7e23SEd Schouten memset(args, '\0', sizeof(*args)); 1234457f7e23SEd Schouten ofdp = td->td_proc->p_fd; 1235457f7e23SEd Schouten if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1) 1236457f7e23SEd Schouten return (E2BIG); 1237457f7e23SEd Schouten error = exec_alloc_args(args); 1238457f7e23SEd Schouten if (error != 0) 1239457f7e23SEd Schouten return (error); 1240457f7e23SEd Schouten 1241457f7e23SEd Schouten args->begin_argv = args->buf; 1242457f7e23SEd Schouten args->stringspace = ARG_MAX; 1243457f7e23SEd Schouten 1244457f7e23SEd Schouten if (datalen > 0) { 1245457f7e23SEd Schouten /* 1246457f7e23SEd Schouten * Argument buffer has been provided. Copy it into the 1247457f7e23SEd Schouten * kernel as a single string and add a terminating null 1248457f7e23SEd Schouten * byte. 1249457f7e23SEd Schouten */ 1250457f7e23SEd Schouten error = copyin(data, args->begin_argv, datalen); 1251457f7e23SEd Schouten if (error != 0) 1252457f7e23SEd Schouten goto err_exit; 1253457f7e23SEd Schouten args->begin_argv[datalen] = '\0'; 1254457f7e23SEd Schouten args->endp = args->begin_argv + datalen + 1; 1255457f7e23SEd Schouten args->stringspace -= datalen + 1; 1256457f7e23SEd Schouten 1257457f7e23SEd Schouten /* 1258457f7e23SEd Schouten * Traditional argument counting. Count the number of 1259457f7e23SEd Schouten * null bytes. 1260457f7e23SEd Schouten */ 1261457f7e23SEd Schouten for (p = args->begin_argv; p < args->endp; ++p) 1262457f7e23SEd Schouten if (*p == '\0') 1263457f7e23SEd Schouten ++args->argc; 1264457f7e23SEd Schouten } else { 1265457f7e23SEd Schouten /* No argument buffer provided. */ 1266457f7e23SEd Schouten args->endp = args->begin_argv; 1267457f7e23SEd Schouten } 1268457f7e23SEd Schouten 1269457f7e23SEd Schouten /* Create new file descriptor table. */ 1270457f7e23SEd Schouten kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK); 1271457f7e23SEd Schouten error = copyin(fds, kfds, fdslen * sizeof(int)); 1272457f7e23SEd Schouten if (error != 0) { 1273457f7e23SEd Schouten free(kfds, M_TEMP); 1274457f7e23SEd Schouten goto err_exit; 1275457f7e23SEd Schouten } 1276457f7e23SEd Schouten error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp); 1277457f7e23SEd Schouten free(kfds, M_TEMP); 1278457f7e23SEd Schouten if (error != 0) 1279457f7e23SEd Schouten goto err_exit; 1280457f7e23SEd Schouten 1281457f7e23SEd Schouten return (0); 1282457f7e23SEd Schouten err_exit: 1283457f7e23SEd Schouten exec_free_args(args); 1284457f7e23SEd Schouten return (error); 1285457f7e23SEd Schouten } 1286457f7e23SEd Schouten 1287ec492b13SMark Johnston struct exec_args_kva { 1288ec492b13SMark Johnston vm_offset_t addr; 1289c6a4ba5aSMark Johnston u_int gen; 1290ec492b13SMark Johnston SLIST_ENTRY(exec_args_kva) next; 1291ec492b13SMark Johnston }; 1292ec492b13SMark Johnston 12932bf95012SAndrew Turner DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva); 1294ec492b13SMark Johnston 1295ec492b13SMark Johnston static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist; 1296ec492b13SMark Johnston static struct mtx exec_args_kva_mtx; 1297c6a4ba5aSMark Johnston static u_int exec_args_gen; 1298ec492b13SMark Johnston 1299ec492b13SMark Johnston static void 1300ec492b13SMark Johnston exec_prealloc_args_kva(void *arg __unused) 1301ec492b13SMark Johnston { 1302ec492b13SMark Johnston struct exec_args_kva *argkva; 1303ec492b13SMark Johnston u_int i; 1304ec492b13SMark Johnston 1305ec492b13SMark Johnston SLIST_INIT(&exec_args_kva_freelist); 1306ec492b13SMark Johnston mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF); 1307ec492b13SMark Johnston for (i = 0; i < exec_map_entries; i++) { 1308ec492b13SMark Johnston argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK); 1309ec492b13SMark Johnston argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size); 1310c6a4ba5aSMark Johnston argkva->gen = exec_args_gen; 1311ec492b13SMark Johnston SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1312ec492b13SMark Johnston } 1313ec492b13SMark Johnston } 1314ec492b13SMark Johnston SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL); 1315ec492b13SMark Johnston 1316ec492b13SMark Johnston static vm_offset_t 1317ec492b13SMark Johnston exec_alloc_args_kva(void **cookie) 1318ec492b13SMark Johnston { 1319ec492b13SMark Johnston struct exec_args_kva *argkva; 1320ec492b13SMark Johnston 1321ec492b13SMark Johnston argkva = (void *)atomic_readandclear_ptr( 1322ec492b13SMark Johnston (uintptr_t *)DPCPU_PTR(exec_args_kva)); 1323ec492b13SMark Johnston if (argkva == NULL) { 1324ec492b13SMark Johnston mtx_lock(&exec_args_kva_mtx); 1325ec492b13SMark Johnston while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL) 1326ec492b13SMark Johnston (void)mtx_sleep(&exec_args_kva_freelist, 1327ec492b13SMark Johnston &exec_args_kva_mtx, 0, "execkva", 0); 1328ec492b13SMark Johnston SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next); 1329ec492b13SMark Johnston mtx_unlock(&exec_args_kva_mtx); 1330ec492b13SMark Johnston } 1331ec492b13SMark Johnston *(struct exec_args_kva **)cookie = argkva; 1332ec492b13SMark Johnston return (argkva->addr); 1333ec492b13SMark Johnston } 1334ec492b13SMark Johnston 1335ec492b13SMark Johnston static void 1336c6a4ba5aSMark Johnston exec_release_args_kva(struct exec_args_kva *argkva, u_int gen) 1337ec492b13SMark Johnston { 1338ec492b13SMark Johnston vm_offset_t base; 1339ec492b13SMark Johnston 1340ec492b13SMark Johnston base = argkva->addr; 1341c6a4ba5aSMark Johnston if (argkva->gen != gen) { 13423e7cb27cSAlan Cox (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size, 1343c6a4ba5aSMark Johnston MADV_FREE); 1344c6a4ba5aSMark Johnston argkva->gen = gen; 1345c6a4ba5aSMark Johnston } 1346ec492b13SMark Johnston if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva), 1347ec492b13SMark Johnston (uintptr_t)NULL, (uintptr_t)argkva)) { 1348ec492b13SMark Johnston mtx_lock(&exec_args_kva_mtx); 1349ec492b13SMark Johnston SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1350ec492b13SMark Johnston wakeup_one(&exec_args_kva_freelist); 1351ec492b13SMark Johnston mtx_unlock(&exec_args_kva_mtx); 1352ec492b13SMark Johnston } 1353ec492b13SMark Johnston } 1354ec492b13SMark Johnston 1355c6a4ba5aSMark Johnston static void 1356c6a4ba5aSMark Johnston exec_free_args_kva(void *cookie) 1357c6a4ba5aSMark Johnston { 1358c6a4ba5aSMark Johnston 1359c6a4ba5aSMark Johnston exec_release_args_kva(cookie, exec_args_gen); 1360c6a4ba5aSMark Johnston } 1361c6a4ba5aSMark Johnston 1362c6a4ba5aSMark Johnston static void 1363c6a4ba5aSMark Johnston exec_args_kva_lowmem(void *arg __unused) 1364c6a4ba5aSMark Johnston { 1365c6a4ba5aSMark Johnston SLIST_HEAD(, exec_args_kva) head; 1366c6a4ba5aSMark Johnston struct exec_args_kva *argkva; 1367c6a4ba5aSMark Johnston u_int gen; 1368c6a4ba5aSMark Johnston int i; 1369c6a4ba5aSMark Johnston 1370c6a4ba5aSMark Johnston gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1; 1371c6a4ba5aSMark Johnston 1372c6a4ba5aSMark Johnston /* 1373c6a4ba5aSMark Johnston * Force an madvise of each KVA range. Any currently allocated ranges 1374c6a4ba5aSMark Johnston * will have MADV_FREE applied once they are freed. 1375c6a4ba5aSMark Johnston */ 1376c6a4ba5aSMark Johnston SLIST_INIT(&head); 1377c6a4ba5aSMark Johnston mtx_lock(&exec_args_kva_mtx); 1378c6a4ba5aSMark Johnston SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva); 1379c6a4ba5aSMark Johnston mtx_unlock(&exec_args_kva_mtx); 1380c6a4ba5aSMark Johnston while ((argkva = SLIST_FIRST(&head)) != NULL) { 1381c6a4ba5aSMark Johnston SLIST_REMOVE_HEAD(&head, next); 1382c6a4ba5aSMark Johnston exec_release_args_kva(argkva, gen); 1383c6a4ba5aSMark Johnston } 1384c6a4ba5aSMark Johnston 1385c6a4ba5aSMark Johnston CPU_FOREACH(i) { 1386c6a4ba5aSMark Johnston argkva = (void *)atomic_readandclear_ptr( 1387c6a4ba5aSMark Johnston (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva)); 1388c6a4ba5aSMark Johnston if (argkva != NULL) 1389c6a4ba5aSMark Johnston exec_release_args_kva(argkva, gen); 1390c6a4ba5aSMark Johnston } 1391c6a4ba5aSMark Johnston } 1392c6a4ba5aSMark Johnston EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL, 1393c6a4ba5aSMark Johnston EVENTHANDLER_PRI_ANY); 1394c6a4ba5aSMark Johnston 13952af6e14dSAlan Cox /* 13962af6e14dSAlan Cox * Allocate temporary demand-paged, zero-filled memory for the file name, 1397ec492b13SMark Johnston * argument, and environment strings. 13982af6e14dSAlan Cox */ 13992af6e14dSAlan Cox int 14002af6e14dSAlan Cox exec_alloc_args(struct image_args *args) 14012af6e14dSAlan Cox { 14022af6e14dSAlan Cox 1403ec492b13SMark Johnston args->buf = (char *)exec_alloc_args_kva(&args->bufkva); 1404ec492b13SMark Johnston return (0); 14052af6e14dSAlan Cox } 14062af6e14dSAlan Cox 140769a8f9e3SAlan Cox void 1408610ecfe0SMaxim Sobolev exec_free_args(struct image_args *args) 1409610ecfe0SMaxim Sobolev { 1410610ecfe0SMaxim Sobolev 141169a8f9e3SAlan Cox if (args->buf != NULL) { 1412ec492b13SMark Johnston exec_free_args_kva(args->bufkva); 1413610ecfe0SMaxim Sobolev args->buf = NULL; 1414610ecfe0SMaxim Sobolev } 1415a14a9498SAlan Cox if (args->fname_buf != NULL) { 1416a14a9498SAlan Cox free(args->fname_buf, M_TEMP); 1417a14a9498SAlan Cox args->fname_buf = NULL; 1418a14a9498SAlan Cox } 1419457f7e23SEd Schouten if (args->fdp != NULL) 1420457f7e23SEd Schouten fdescfree_remapped(args->fdp); 1421610ecfe0SMaxim Sobolev } 1422610ecfe0SMaxim Sobolev 142326f9a767SRodney W. Grimes /* 1424f373437aSBrooks Davis * A set to functions to fill struct image args. 1425f373437aSBrooks Davis * 1426f373437aSBrooks Davis * NOTE: exec_args_add_fname() must be called (possibly with a NULL 1427f373437aSBrooks Davis * fname) before the other functions. All exec_args_add_arg() calls must 1428f373437aSBrooks Davis * be made before any exec_args_add_env() calls. exec_args_adjust_args() 1429f373437aSBrooks Davis * may be called any time after exec_args_add_fname(). 1430f373437aSBrooks Davis * 1431f373437aSBrooks Davis * exec_args_add_fname() - install path to be executed 1432f373437aSBrooks Davis * exec_args_add_arg() - append an argument string 1433f373437aSBrooks Davis * exec_args_add_env() - append an env string 1434f373437aSBrooks Davis * exec_args_adjust_args() - adjust location of the argument list to 1435f373437aSBrooks Davis * allow new arguments to be prepended 1436f373437aSBrooks Davis */ 1437f373437aSBrooks Davis int 1438f373437aSBrooks Davis exec_args_add_fname(struct image_args *args, const char *fname, 1439f373437aSBrooks Davis enum uio_seg segflg) 1440f373437aSBrooks Davis { 1441f373437aSBrooks Davis int error; 1442f373437aSBrooks Davis size_t length; 1443f373437aSBrooks Davis 1444f373437aSBrooks Davis KASSERT(args->fname == NULL, ("fname already appended")); 1445f373437aSBrooks Davis KASSERT(args->endp == NULL, ("already appending to args")); 1446f373437aSBrooks Davis 1447f373437aSBrooks Davis if (fname != NULL) { 1448f373437aSBrooks Davis args->fname = args->buf; 1449f373437aSBrooks Davis error = segflg == UIO_SYSSPACE ? 1450f373437aSBrooks Davis copystr(fname, args->fname, PATH_MAX, &length) : 1451f373437aSBrooks Davis copyinstr(fname, args->fname, PATH_MAX, &length); 1452f373437aSBrooks Davis if (error != 0) 1453f373437aSBrooks Davis return (error == ENAMETOOLONG ? E2BIG : error); 1454f373437aSBrooks Davis } else 1455f373437aSBrooks Davis length = 0; 1456f373437aSBrooks Davis 1457f373437aSBrooks Davis /* Set up for _arg_*()/_env_*() */ 1458f373437aSBrooks Davis args->endp = args->buf + length; 1459f373437aSBrooks Davis /* begin_argv must be set and kept updated */ 1460f373437aSBrooks Davis args->begin_argv = args->endp; 1461f373437aSBrooks Davis KASSERT(exec_map_entry_size - length >= ARG_MAX, 1462f373437aSBrooks Davis ("too little space remaining for arguments %zu < %zu", 1463f373437aSBrooks Davis exec_map_entry_size - length, (size_t)ARG_MAX)); 1464f373437aSBrooks Davis args->stringspace = ARG_MAX; 1465f373437aSBrooks Davis 1466f373437aSBrooks Davis return (0); 1467f373437aSBrooks Davis } 1468f373437aSBrooks Davis 1469f373437aSBrooks Davis static int 1470f373437aSBrooks Davis exec_args_add_str(struct image_args *args, const char *str, 1471f373437aSBrooks Davis enum uio_seg segflg, int *countp) 1472f373437aSBrooks Davis { 1473f373437aSBrooks Davis int error; 1474f373437aSBrooks Davis size_t length; 1475f373437aSBrooks Davis 1476f373437aSBrooks Davis KASSERT(args->endp != NULL, ("endp not initialized")); 1477f373437aSBrooks Davis KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1478f373437aSBrooks Davis 1479f373437aSBrooks Davis error = (segflg == UIO_SYSSPACE) ? 1480f373437aSBrooks Davis copystr(str, args->endp, args->stringspace, &length) : 1481f373437aSBrooks Davis copyinstr(str, args->endp, args->stringspace, &length); 1482f373437aSBrooks Davis if (error != 0) 1483f373437aSBrooks Davis return (error == ENAMETOOLONG ? E2BIG : error); 1484f373437aSBrooks Davis args->stringspace -= length; 1485f373437aSBrooks Davis args->endp += length; 1486f373437aSBrooks Davis (*countp)++; 1487f373437aSBrooks Davis 1488f373437aSBrooks Davis return (0); 1489f373437aSBrooks Davis } 1490f373437aSBrooks Davis 1491f373437aSBrooks Davis int 1492f373437aSBrooks Davis exec_args_add_arg(struct image_args *args, const char *argp, 1493f373437aSBrooks Davis enum uio_seg segflg) 1494f373437aSBrooks Davis { 1495f373437aSBrooks Davis 1496f373437aSBrooks Davis KASSERT(args->envc == 0, ("appending args after env")); 1497f373437aSBrooks Davis 1498f373437aSBrooks Davis return (exec_args_add_str(args, argp, segflg, &args->argc)); 1499f373437aSBrooks Davis } 1500f373437aSBrooks Davis 1501f373437aSBrooks Davis int 1502f373437aSBrooks Davis exec_args_add_env(struct image_args *args, const char *envp, 1503f373437aSBrooks Davis enum uio_seg segflg) 1504f373437aSBrooks Davis { 1505f373437aSBrooks Davis 1506f373437aSBrooks Davis if (args->envc == 0) 1507f373437aSBrooks Davis args->begin_envv = args->endp; 1508f373437aSBrooks Davis 1509f373437aSBrooks Davis return (exec_args_add_str(args, envp, segflg, &args->envc)); 1510f373437aSBrooks Davis } 1511f373437aSBrooks Davis 1512f373437aSBrooks Davis int 1513f373437aSBrooks Davis exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend) 1514f373437aSBrooks Davis { 1515f373437aSBrooks Davis ssize_t offset; 1516f373437aSBrooks Davis 1517f373437aSBrooks Davis KASSERT(args->endp != NULL, ("endp not initialized")); 1518f373437aSBrooks Davis KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1519f373437aSBrooks Davis 1520f373437aSBrooks Davis offset = extend - consume; 1521f373437aSBrooks Davis if (args->stringspace < offset) 1522f373437aSBrooks Davis return (E2BIG); 1523f373437aSBrooks Davis memmove(args->begin_argv + extend, args->begin_argv + consume, 1524f373437aSBrooks Davis args->endp - args->begin_argv + consume); 1525f373437aSBrooks Davis if (args->envc > 0) 1526f373437aSBrooks Davis args->begin_envv += offset; 1527f373437aSBrooks Davis args->endp += offset; 1528f373437aSBrooks Davis args->stringspace -= offset; 1529f373437aSBrooks Davis return (0); 1530f373437aSBrooks Davis } 1531f373437aSBrooks Davis 1532f373437aSBrooks Davis char * 1533f373437aSBrooks Davis exec_args_get_begin_envv(struct image_args *args) 1534f373437aSBrooks Davis { 1535f373437aSBrooks Davis 1536f373437aSBrooks Davis KASSERT(args->endp != NULL, ("endp not initialized")); 1537f373437aSBrooks Davis 1538f373437aSBrooks Davis if (args->envc > 0) 1539f373437aSBrooks Davis return (args->begin_envv); 1540f373437aSBrooks Davis return (args->endp); 1541f373437aSBrooks Davis } 1542f373437aSBrooks Davis 1543f373437aSBrooks Davis /* 1544873fbcd7SRobert Watson * Copy strings out to the new process address space, constructing new arg 1545873fbcd7SRobert Watson * and env vector tables. Return a pointer to the base so that it can be used 1546873fbcd7SRobert Watson * as the initial stack pointer. 154726f9a767SRodney W. Grimes */ 1548654f6be1SBruce Evans register_t * 15493e85b721SEd Maste exec_copyout_strings(struct image_params *imgp) 155026f9a767SRodney W. Grimes { 155126f9a767SRodney W. Grimes int argc, envc; 155226f9a767SRodney W. Grimes char **vectp; 155388b124ceSKonstantin Belousov char *stringp; 155488b124ceSKonstantin Belousov uintptr_t destp; 1555654f6be1SBruce Evans register_t *stack_base; 155693f6448cSDavid Greenman struct ps_strings *arginfo; 1557f3bec5d7SJake Burkholder struct proc *p; 15583ff06357SKonstantin Belousov size_t execpath_len; 1559ee235befSKonstantin Belousov int szsigcode, szps; 1560ee235befSKonstantin Belousov char canary[sizeof(long) * 8]; 156126f9a767SRodney W. Grimes 1562ee235befSKonstantin Belousov szps = sizeof(pagesizes[0]) * MAXPAGESIZES; 156326f9a767SRodney W. Grimes /* 156426f9a767SRodney W. Grimes * Calculate string base and vector table pointers. 1565d66a5066SPeter Wemm * Also deal with signal trampoline code for this exec type. 156626f9a767SRodney W. Grimes */ 15673ff06357SKonstantin Belousov if (imgp->execpath != NULL && imgp->auxargs != NULL) 15683ff06357SKonstantin Belousov execpath_len = strlen(imgp->execpath) + 1; 15693ff06357SKonstantin Belousov else 15703ff06357SKonstantin Belousov execpath_len = 0; 1571f3bec5d7SJake Burkholder p = imgp->proc; 1572f3bec5d7SJake Burkholder szsigcode = 0; 157305ba50f5SJake Burkholder arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 15746297a3d8SKonstantin Belousov if (p->p_sysent->sv_sigcode_base == 0) { 1575f3bec5d7SJake Burkholder if (p->p_sysent->sv_szsigcode != NULL) 1576f3bec5d7SJake Burkholder szsigcode = *(p->p_sysent->sv_szsigcode); 15776297a3d8SKonstantin Belousov } 157888b124ceSKonstantin Belousov destp = (uintptr_t)arginfo; 15791ed012f9SPeter Wemm 158026f9a767SRodney W. Grimes /* 1581d66a5066SPeter Wemm * install sigcode 1582d66a5066SPeter Wemm */ 158388b124ceSKonstantin Belousov if (szsigcode != 0) { 158488b124ceSKonstantin Belousov destp -= szsigcode; 158588b124ceSKonstantin Belousov destp = rounddown2(destp, sizeof(void *)); 158688b124ceSKonstantin Belousov copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode); 158788b124ceSKonstantin Belousov } 1588d66a5066SPeter Wemm 1589d66a5066SPeter Wemm /* 15903ff06357SKonstantin Belousov * Copy the image path for the rtld. 15913ff06357SKonstantin Belousov */ 15923ff06357SKonstantin Belousov if (execpath_len != 0) { 159388b124ceSKonstantin Belousov destp -= execpath_len; 1594d92da759SBrooks Davis destp = rounddown2(destp, sizeof(void *)); 159588b124ceSKonstantin Belousov imgp->execpathp = destp; 159688b124ceSKonstantin Belousov copyout(imgp->execpath, (void *)destp, execpath_len); 15973ff06357SKonstantin Belousov } 15983ff06357SKonstantin Belousov 15993ff06357SKonstantin Belousov /* 1600ee235befSKonstantin Belousov * Prepare the canary for SSP. 1601ee235befSKonstantin Belousov */ 1602ee235befSKonstantin Belousov arc4rand(canary, sizeof(canary), 0); 160388b124ceSKonstantin Belousov destp -= sizeof(canary); 160488b124ceSKonstantin Belousov imgp->canary = destp; 160588b124ceSKonstantin Belousov copyout(canary, (void *)destp, sizeof(canary)); 1606ee235befSKonstantin Belousov imgp->canarylen = sizeof(canary); 1607ee235befSKonstantin Belousov 1608ee235befSKonstantin Belousov /* 1609ee235befSKonstantin Belousov * Prepare the pagesizes array. 1610ee235befSKonstantin Belousov */ 161188b124ceSKonstantin Belousov destp -= szps; 161288b124ceSKonstantin Belousov destp = rounddown2(destp, sizeof(void *)); 161388b124ceSKonstantin Belousov imgp->pagesizes = destp; 161488b124ceSKonstantin Belousov copyout(pagesizes, (void *)destp, szps); 1615ee235befSKonstantin Belousov imgp->pagesizeslen = szps; 1616ee235befSKonstantin Belousov 161788b124ceSKonstantin Belousov destp -= ARG_MAX - imgp->args->stringspace; 161888b124ceSKonstantin Belousov destp = rounddown2(destp, sizeof(void *)); 161988b124ceSKonstantin Belousov 162073c8686eSJohn Baldwin vectp = (char **)destp; 1621b9a22da4STakanori Watanabe if (imgp->auxargs) { 1622e1743d02SSøren Schmidt /* 162373c8686eSJohn Baldwin * Allocate room on the stack for the ELF auxargs 162473c8686eSJohn Baldwin * array. It has up to AT_COUNT entries. 1625b9a22da4STakanori Watanabe */ 162673c8686eSJohn Baldwin vectp -= howmany(AT_COUNT * sizeof(Elf_Auxinfo), 162773c8686eSJohn Baldwin sizeof(*vectp)); 1628610ecfe0SMaxim Sobolev } 162926f9a767SRodney W. Grimes 163026f9a767SRodney W. Grimes /* 163173c8686eSJohn Baldwin * Allocate room for the argv[] and env vectors including the 163273c8686eSJohn Baldwin * terminating NULL pointers. 163373c8686eSJohn Baldwin */ 163473c8686eSJohn Baldwin vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 163573c8686eSJohn Baldwin 163673c8686eSJohn Baldwin /* 163726f9a767SRodney W. Grimes * vectp also becomes our initial stack base 163826f9a767SRodney W. Grimes */ 1639654f6be1SBruce Evans stack_base = (register_t *)vectp; 164026f9a767SRodney W. Grimes 1641610ecfe0SMaxim Sobolev stringp = imgp->args->begin_argv; 1642610ecfe0SMaxim Sobolev argc = imgp->args->argc; 1643610ecfe0SMaxim Sobolev envc = imgp->args->envc; 164426f9a767SRodney W. Grimes 164593f6448cSDavid Greenman /* 16463aed948bSDavid Greenman * Copy out strings - arguments and environment. 164793f6448cSDavid Greenman */ 164888b124ceSKonstantin Belousov copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace); 164993f6448cSDavid Greenman 165093f6448cSDavid Greenman /* 16513aed948bSDavid Greenman * Fill in "ps_strings" struct for ps, w, etc. 16523aed948bSDavid Greenman */ 1653aae0aa45SBruce Evans suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 1654f4e26adeSNathan Whitehorn suword32(&arginfo->ps_nargvstr, argc); 16553aed948bSDavid Greenman 16563aed948bSDavid Greenman /* 16573aed948bSDavid Greenman * Fill in argument portion of vector table. 165893f6448cSDavid Greenman */ 165926f9a767SRodney W. Grimes for (; argc > 0; --argc) { 1660aae0aa45SBruce Evans suword(vectp++, (long)(intptr_t)destp); 16613aed948bSDavid Greenman while (*stringp++ != 0) 16623aed948bSDavid Greenman destp++; 16633aed948bSDavid Greenman destp++; 166426f9a767SRodney W. Grimes } 166526f9a767SRodney W. Grimes 16661a6e52d0SJeroen Ruigrok van der Werven /* a null vector table pointer separates the argp's from the envp's */ 16676ab46d52SBruce Evans suword(vectp++, 0); 166826f9a767SRodney W. Grimes 1669aae0aa45SBruce Evans suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 1670f4e26adeSNathan Whitehorn suword32(&arginfo->ps_nenvstr, envc); 167193f6448cSDavid Greenman 167293f6448cSDavid Greenman /* 16733aed948bSDavid Greenman * Fill in environment portion of vector table. 167493f6448cSDavid Greenman */ 167526f9a767SRodney W. Grimes for (; envc > 0; --envc) { 1676aae0aa45SBruce Evans suword(vectp++, (long)(intptr_t)destp); 16773aed948bSDavid Greenman while (*stringp++ != 0) 16783aed948bSDavid Greenman destp++; 16793aed948bSDavid Greenman destp++; 168026f9a767SRodney W. Grimes } 168126f9a767SRodney W. Grimes 168226f9a767SRodney W. Grimes /* end of vector table is a null pointer */ 16836ab46d52SBruce Evans suword(vectp, 0); 168426f9a767SRodney W. Grimes 168526f9a767SRodney W. Grimes return (stack_base); 168626f9a767SRodney W. Grimes } 168726f9a767SRodney W. Grimes 168826f9a767SRodney W. Grimes /* 168926f9a767SRodney W. Grimes * Check permissions of file to execute. 1690cf64863aSRobert Watson * Called with imgp->vp locked. 169126f9a767SRodney W. Grimes * Return 0 for success or error code on failure. 169226f9a767SRodney W. Grimes */ 1693c8a79999SPeter Wemm int 16943e85b721SEd Maste exec_check_permissions(struct image_params *imgp) 169526f9a767SRodney W. Grimes { 1696c52007c2SDavid Greenman struct vnode *vp = imgp->vp; 1697c52007c2SDavid Greenman struct vattr *attr = imgp->attr; 1698a854ed98SJohn Baldwin struct thread *td; 1699*78022527SKonstantin Belousov int error; 170026f9a767SRodney W. Grimes 17016617724cSJeff Roberson td = curthread; 1702339b79b9SRobert Watson 1703ec35c2afSRobert Watson /* Get file attributes */ 17040359a12eSAttilio Rao error = VOP_GETATTR(vp, attr, td->td_ucred); 1705ec35c2afSRobert Watson if (error) 1706ec35c2afSRobert Watson return (error); 1707ec35c2afSRobert Watson 1708339b79b9SRobert Watson #ifdef MAC 170930d239bcSRobert Watson error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1710339b79b9SRobert Watson if (error) 1711339b79b9SRobert Watson return (error); 1712339b79b9SRobert Watson #endif 1713339b79b9SRobert Watson 171426f9a767SRodney W. Grimes /* 1715de478dd4SJaakko Heinonen * 1) Check if file execution is disabled for the filesystem that 1716de478dd4SJaakko Heinonen * this file resides on. 1717de478dd4SJaakko Heinonen * 2) Ensure that at least one execute bit is on. Otherwise, a 1718de478dd4SJaakko Heinonen * privileged user will always succeed, and we don't want this 1719de478dd4SJaakko Heinonen * to happen unless the file really is executable. 1720de478dd4SJaakko Heinonen * 3) Ensure that the file is a regular file. 172126f9a767SRodney W. Grimes */ 1722c52007c2SDavid Greenman if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1723de478dd4SJaakko Heinonen (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1724a854ed98SJohn Baldwin (attr->va_type != VREG)) 172526f9a767SRodney W. Grimes return (EACCES); 172626f9a767SRodney W. Grimes 172726f9a767SRodney W. Grimes /* 172826f9a767SRodney W. Grimes * Zero length files can't be exec'd 172926f9a767SRodney W. Grimes */ 173026f9a767SRodney W. Grimes if (attr->va_size == 0) 173126f9a767SRodney W. Grimes return (ENOEXEC); 173226f9a767SRodney W. Grimes 173326f9a767SRodney W. Grimes /* 173426f9a767SRodney W. Grimes * Check for execute permission to file based on current credentials. 173526f9a767SRodney W. Grimes */ 1736a854ed98SJohn Baldwin error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 173726f9a767SRodney W. Grimes if (error) 173826f9a767SRodney W. Grimes return (error); 173926f9a767SRodney W. Grimes 17406d5a0a8cSDavid Greenman /* 17416d5a0a8cSDavid Greenman * Check number of open-for-writes on the file and deny execution 17426d5a0a8cSDavid Greenman * if there are any. 1743*78022527SKonstantin Belousov * 1744*78022527SKonstantin Belousov * Add a text reference now so no one can write to the 1745*78022527SKonstantin Belousov * executable while we're activating it. 1746*78022527SKonstantin Belousov * 1747*78022527SKonstantin Belousov * Remember if this was set before and unset it in case this is not 1748*78022527SKonstantin Belousov * actually an executable image. 17496d5a0a8cSDavid Greenman */ 1750*78022527SKonstantin Belousov error = VOP_SET_TEXT(vp); 1751140dedb8SKonstantin Belousov if (error != 0) 1752140dedb8SKonstantin Belousov return (error); 1753*78022527SKonstantin Belousov imgp->textset = true; 17546d5a0a8cSDavid Greenman 17556d5a0a8cSDavid Greenman /* 17566d5a0a8cSDavid Greenman * Call filesystem specific open routine (which does nothing in the 17576d5a0a8cSDavid Greenman * general case). 17586d5a0a8cSDavid Greenman */ 17599e223287SKonstantin Belousov error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 17609a75ea23SKonstantin Belousov if (error == 0) 17619a75ea23SKonstantin Belousov imgp->opened = 1; 176226f9a767SRodney W. Grimes return (error); 1763df8bae1dSRodney W. Grimes } 1764aa855a59SPeter Wemm 1765aa855a59SPeter Wemm /* 1766aa855a59SPeter Wemm * Exec handler registration 1767aa855a59SPeter Wemm */ 1768aa855a59SPeter Wemm int 17693e85b721SEd Maste exec_register(const struct execsw *execsw_arg) 1770aa855a59SPeter Wemm { 1771aa855a59SPeter Wemm const struct execsw **es, **xs, **newexecsw; 1772d821d364SPedro F. Giffuni u_int count = 2; /* New slot and trailing NULL */ 1773aa855a59SPeter Wemm 1774aa855a59SPeter Wemm if (execsw) 1775aa855a59SPeter Wemm for (es = execsw; *es; es++) 1776aa855a59SPeter Wemm count++; 1777a163d034SWarner Losh newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1778aa855a59SPeter Wemm xs = newexecsw; 1779aa855a59SPeter Wemm if (execsw) 1780aa855a59SPeter Wemm for (es = execsw; *es; es++) 1781aa855a59SPeter Wemm *xs++ = *es; 1782aa855a59SPeter Wemm *xs++ = execsw_arg; 1783aa855a59SPeter Wemm *xs = NULL; 1784aa855a59SPeter Wemm if (execsw) 1785aa855a59SPeter Wemm free(execsw, M_TEMP); 1786aa855a59SPeter Wemm execsw = newexecsw; 1787a7cddfedSJake Burkholder return (0); 1788aa855a59SPeter Wemm } 1789aa855a59SPeter Wemm 1790aa855a59SPeter Wemm int 17913e85b721SEd Maste exec_unregister(const struct execsw *execsw_arg) 1792aa855a59SPeter Wemm { 1793aa855a59SPeter Wemm const struct execsw **es, **xs, **newexecsw; 1794aa855a59SPeter Wemm int count = 1; 1795aa855a59SPeter Wemm 1796aa855a59SPeter Wemm if (execsw == NULL) 1797aa855a59SPeter Wemm panic("unregister with no handlers left?\n"); 1798aa855a59SPeter Wemm 1799aa855a59SPeter Wemm for (es = execsw; *es; es++) { 1800aa855a59SPeter Wemm if (*es == execsw_arg) 1801aa855a59SPeter Wemm break; 1802aa855a59SPeter Wemm } 1803aa855a59SPeter Wemm if (*es == NULL) 1804a7cddfedSJake Burkholder return (ENOENT); 1805aa855a59SPeter Wemm for (es = execsw; *es; es++) 1806aa855a59SPeter Wemm if (*es != execsw_arg) 1807aa855a59SPeter Wemm count++; 1808a163d034SWarner Losh newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1809aa855a59SPeter Wemm xs = newexecsw; 1810aa855a59SPeter Wemm for (es = execsw; *es; es++) 1811aa855a59SPeter Wemm if (*es != execsw_arg) 1812aa855a59SPeter Wemm *xs++ = *es; 1813aa855a59SPeter Wemm *xs = NULL; 1814aa855a59SPeter Wemm if (execsw) 1815aa855a59SPeter Wemm free(execsw, M_TEMP); 1816aa855a59SPeter Wemm execsw = newexecsw; 1817a7cddfedSJake Burkholder return (0); 1818aa855a59SPeter Wemm } 1819