29
30#include "opt_hwpmc_hooks.h"
31#include "opt_kdtrace.h"
32#include "opt_ktrace.h"
33#include "opt_vm.h"
34
35#include <sys/param.h>
36#include <sys/systm.h>
37#include <sys/eventhandler.h>
38#include <sys/lock.h>
39#include <sys/mutex.h>
40#include <sys/sysproto.h>
41#include <sys/signalvar.h>
42#include <sys/kernel.h>
43#include <sys/mount.h>
44#include <sys/filedesc.h>
45#include <sys/fcntl.h>
46#include <sys/acct.h>
47#include <sys/exec.h>
48#include <sys/imgact.h>
49#include <sys/imgact_elf.h>
50#include <sys/wait.h>
51#include <sys/malloc.h>
52#include <sys/priv.h>
53#include <sys/proc.h>
54#include <sys/pioctl.h>
55#include <sys/namei.h>
56#include <sys/resourcevar.h>
57#include <sys/sdt.h>
58#include <sys/sf_buf.h>
59#include <sys/syscallsubr.h>
60#include <sys/sysent.h>
61#include <sys/shm.h>
62#include <sys/sysctl.h>
63#include <sys/vnode.h>
64#include <sys/stat.h>
65#ifdef KTRACE
66#include <sys/ktrace.h>
67#endif
68
69#include <vm/vm.h>
70#include <vm/vm_param.h>
71#include <vm/pmap.h>
72#include <vm/vm_page.h>
73#include <vm/vm_map.h>
74#include <vm/vm_kern.h>
75#include <vm/vm_extern.h>
76#include <vm/vm_object.h>
77#include <vm/vm_pager.h>
78
79#ifdef HWPMC_HOOKS
80#include <sys/pmckern.h>
81#endif
82
83#include <machine/reg.h>
84
85#include <security/audit/audit.h>
86#include <security/mac/mac_framework.h>
87
88#ifdef KDTRACE_HOOKS
89#include <sys/dtrace_bsd.h>
90dtrace_execexit_func_t dtrace_fasttrap_exec;
91#endif
92
93SDT_PROVIDER_DECLARE(proc);
94SDT_PROBE_DEFINE(proc, kernel, , exec);
95SDT_PROBE_ARGTYPE(proc, kernel, , exec, 0, "char *");
96SDT_PROBE_DEFINE(proc, kernel, , exec_failure);
97SDT_PROBE_ARGTYPE(proc, kernel, , exec_failure, 0, "int");
98SDT_PROBE_DEFINE(proc, kernel, , exec_success);
99SDT_PROBE_ARGTYPE(proc, kernel, , exec_success, 0, "char *");
100
101MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
102
103static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
104static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
105static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
106static int do_execve(struct thread *td, struct image_args *args,
107 struct mac *mac_p);
108static void exec_free_args(struct image_args *);
109
110/* XXX This should be vm_size_t. */
111SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
112 NULL, 0, sysctl_kern_ps_strings, "LU", "");
113
114/* XXX This should be vm_size_t. */
115SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
116 NULL, 0, sysctl_kern_usrstack, "LU", "");
117
118SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
119 NULL, 0, sysctl_kern_stackprot, "I", "");
120
121u_long ps_arg_cache_limit = PAGE_SIZE / 16;
122SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
123 &ps_arg_cache_limit, 0, "");
124
125static int
126sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
127{
128 struct proc *p;
129 int error;
130
131 p = curproc;
132#ifdef SCTL_MASK32
133 if (req->flags & SCTL_MASK32) {
134 unsigned int val;
135 val = (unsigned int)p->p_sysent->sv_psstrings;
136 error = SYSCTL_OUT(req, &val, sizeof(val));
137 } else
138#endif
139 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
140 sizeof(p->p_sysent->sv_psstrings));
141 return error;
142}
143
144static int
145sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
146{
147 struct proc *p;
148 int error;
149
150 p = curproc;
151#ifdef SCTL_MASK32
152 if (req->flags & SCTL_MASK32) {
153 unsigned int val;
154 val = (unsigned int)p->p_sysent->sv_usrstack;
155 error = SYSCTL_OUT(req, &val, sizeof(val));
156 } else
157#endif
158 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
159 sizeof(p->p_sysent->sv_usrstack));
160 return error;
161}
162
163static int
164sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
165{
166 struct proc *p;
167
168 p = curproc;
169 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
170 sizeof(p->p_sysent->sv_stackprot)));
171}
172
173/*
174 * Each of the items is a pointer to a `const struct execsw', hence the
175 * double pointer here.
176 */
177static const struct execsw **execsw;
178
179#ifndef _SYS_SYSPROTO_H_
180struct execve_args {
181 char *fname;
182 char **argv;
183 char **envv;
184};
185#endif
186
187int
188execve(td, uap)
189 struct thread *td;
190 struct execve_args /* {
191 char *fname;
192 char **argv;
193 char **envv;
194 } */ *uap;
195{
196 int error;
197 struct image_args args;
198
199 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
200 uap->argv, uap->envv);
201 if (error == 0)
202 error = kern_execve(td, &args, NULL);
203 return (error);
204}
205
206#ifndef _SYS_SYSPROTO_H_
207struct fexecve_args {
208 int fd;
209 char **argv;
210 char **envv;
211}
212#endif
213int
214fexecve(struct thread *td, struct fexecve_args *uap)
215{
216 int error;
217 struct image_args args;
218
219 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
220 uap->argv, uap->envv);
221 if (error == 0) {
222 args.fd = uap->fd;
223 error = kern_execve(td, &args, NULL);
224 }
225 return (error);
226}
227
228#ifndef _SYS_SYSPROTO_H_
229struct __mac_execve_args {
230 char *fname;
231 char **argv;
232 char **envv;
233 struct mac *mac_p;
234};
235#endif
236
237int
238__mac_execve(td, uap)
239 struct thread *td;
240 struct __mac_execve_args /* {
241 char *fname;
242 char **argv;
243 char **envv;
244 struct mac *mac_p;
245 } */ *uap;
246{
247#ifdef MAC
248 int error;
249 struct image_args args;
250
251 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
252 uap->argv, uap->envv);
253 if (error == 0)
254 error = kern_execve(td, &args, uap->mac_p);
255 return (error);
256#else
257 return (ENOSYS);
258#endif
259}
260
261/*
262 * XXX: kern_execve has the astonishing property of not always returning to
263 * the caller. If sufficiently bad things happen during the call to
264 * do_execve(), it can end up calling exit1(); as a result, callers must
265 * avoid doing anything which they might need to undo (e.g., allocating
266 * memory).
267 */
268int
269kern_execve(td, args, mac_p)
270 struct thread *td;
271 struct image_args *args;
272 struct mac *mac_p;
273{
274 struct proc *p = td->td_proc;
275 int error;
276
277 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
278 args->begin_envv - args->begin_argv);
279 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
280 args->endp - args->begin_envv);
281 if (p->p_flag & P_HADTHREADS) {
282 PROC_LOCK(p);
283 if (thread_single(SINGLE_BOUNDARY)) {
284 PROC_UNLOCK(p);
285 exec_free_args(args);
286 return (ERESTART); /* Try again later. */
287 }
288 PROC_UNLOCK(p);
289 }
290
291 error = do_execve(td, args, mac_p);
292
293 if (p->p_flag & P_HADTHREADS) {
294 PROC_LOCK(p);
295 /*
296 * If success, we upgrade to SINGLE_EXIT state to
297 * force other threads to suicide.
298 */
299 if (error == 0)
300 thread_single(SINGLE_EXIT);
301 else
302 thread_single_end();
303 PROC_UNLOCK(p);
304 }
305
306 return (error);
307}
308
309/*
310 * In-kernel implementation of execve(). All arguments are assumed to be
311 * userspace pointers from the passed thread.
312 */
313static int
314do_execve(td, args, mac_p)
315 struct thread *td;
316 struct image_args *args;
317 struct mac *mac_p;
318{
319 struct proc *p = td->td_proc;
320 struct nameidata nd;
321 struct ucred *newcred = NULL, *oldcred;
322 struct uidinfo *euip;
323 register_t *stack_base;
324 int error, len = 0, i;
325 struct image_params image_params, *imgp;
326 struct vattr attr;
327 int (*img_first)(struct image_params *);
328 struct pargs *oldargs = NULL, *newargs = NULL;
329 struct sigacts *oldsigacts, *newsigacts;
330#ifdef KTRACE
331 struct vnode *tracevp = NULL;
332 struct ucred *tracecred = NULL;
333#endif
334 struct vnode *textvp = NULL, *binvp = NULL;
335 int credential_changing;
336 int vfslocked;
337 int textset;
338#ifdef MAC
339 struct label *interpvplabel = NULL;
340 int will_transition;
341#endif
342#ifdef HWPMC_HOOKS
343 struct pmckern_procexec pe;
344#endif
345 static const char fexecv_proc_title[] = "(fexecv)";
346
347 vfslocked = 0;
348 imgp = &image_params;
349
350 /*
351 * Lock the process and set the P_INEXEC flag to indicate that
352 * it should be left alone until we're done here. This is
353 * necessary to avoid race conditions - e.g. in ptrace() -
354 * that might allow a local user to illicitly obtain elevated
355 * privileges.
356 */
357 PROC_LOCK(p);
358 KASSERT((p->p_flag & P_INEXEC) == 0,
359 ("%s(): process already has P_INEXEC flag", __func__));
360 p->p_flag |= P_INEXEC;
361 PROC_UNLOCK(p);
362
363 /*
364 * Initialize part of the common data
365 */
366 imgp->proc = p;
367 imgp->execlabel = NULL;
368 imgp->attr = &attr;
369 imgp->entry_addr = 0;
370 imgp->vmspace_destroyed = 0;
371 imgp->interpreted = 0;
372 imgp->opened = 0;
373 imgp->interpreter_name = args->buf + PATH_MAX + ARG_MAX;
374 imgp->auxargs = NULL;
375 imgp->vp = NULL;
376 imgp->object = NULL;
377 imgp->firstpage = NULL;
378 imgp->ps_strings = 0;
379 imgp->auxarg_size = 0;
380 imgp->args = args;
381 imgp->execpath = imgp->freepath = NULL;
382 imgp->execpathp = 0;
383
384#ifdef MAC
385 error = mac_execve_enter(imgp, mac_p);
386 if (error)
387 goto exec_fail;
388#endif
389
390 imgp->image_header = NULL;
391
392 /*
393 * Translate the file name. namei() returns a vnode pointer
394 * in ni_vp amoung other things.
395 *
396 * XXXAUDIT: It would be desirable to also audit the name of the
397 * interpreter if this is an interpreted binary.
398 */
399 if (args->fname != NULL) {
400 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
401 | MPSAFE | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
402 }
403
404 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
405
406interpret:
407 if (args->fname != NULL) {
408 error = namei(&nd);
409 if (error)
410 goto exec_fail;
411
412 vfslocked = NDHASGIANT(&nd);
413 binvp = nd.ni_vp;
414 imgp->vp = binvp;
415 } else {
416 AUDIT_ARG_FD(args->fd);
417 error = fgetvp(td, args->fd, &binvp);
418 if (error)
419 goto exec_fail;
420 vfslocked = VFS_LOCK_GIANT(binvp->v_mount);
421 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
| 29
30#include "opt_hwpmc_hooks.h"
31#include "opt_kdtrace.h"
32#include "opt_ktrace.h"
33#include "opt_vm.h"
34
35#include <sys/param.h>
36#include <sys/systm.h>
37#include <sys/eventhandler.h>
38#include <sys/lock.h>
39#include <sys/mutex.h>
40#include <sys/sysproto.h>
41#include <sys/signalvar.h>
42#include <sys/kernel.h>
43#include <sys/mount.h>
44#include <sys/filedesc.h>
45#include <sys/fcntl.h>
46#include <sys/acct.h>
47#include <sys/exec.h>
48#include <sys/imgact.h>
49#include <sys/imgact_elf.h>
50#include <sys/wait.h>
51#include <sys/malloc.h>
52#include <sys/priv.h>
53#include <sys/proc.h>
54#include <sys/pioctl.h>
55#include <sys/namei.h>
56#include <sys/resourcevar.h>
57#include <sys/sdt.h>
58#include <sys/sf_buf.h>
59#include <sys/syscallsubr.h>
60#include <sys/sysent.h>
61#include <sys/shm.h>
62#include <sys/sysctl.h>
63#include <sys/vnode.h>
64#include <sys/stat.h>
65#ifdef KTRACE
66#include <sys/ktrace.h>
67#endif
68
69#include <vm/vm.h>
70#include <vm/vm_param.h>
71#include <vm/pmap.h>
72#include <vm/vm_page.h>
73#include <vm/vm_map.h>
74#include <vm/vm_kern.h>
75#include <vm/vm_extern.h>
76#include <vm/vm_object.h>
77#include <vm/vm_pager.h>
78
79#ifdef HWPMC_HOOKS
80#include <sys/pmckern.h>
81#endif
82
83#include <machine/reg.h>
84
85#include <security/audit/audit.h>
86#include <security/mac/mac_framework.h>
87
88#ifdef KDTRACE_HOOKS
89#include <sys/dtrace_bsd.h>
90dtrace_execexit_func_t dtrace_fasttrap_exec;
91#endif
92
93SDT_PROVIDER_DECLARE(proc);
94SDT_PROBE_DEFINE(proc, kernel, , exec);
95SDT_PROBE_ARGTYPE(proc, kernel, , exec, 0, "char *");
96SDT_PROBE_DEFINE(proc, kernel, , exec_failure);
97SDT_PROBE_ARGTYPE(proc, kernel, , exec_failure, 0, "int");
98SDT_PROBE_DEFINE(proc, kernel, , exec_success);
99SDT_PROBE_ARGTYPE(proc, kernel, , exec_success, 0, "char *");
100
101MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
102
103static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
104static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
105static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
106static int do_execve(struct thread *td, struct image_args *args,
107 struct mac *mac_p);
108static void exec_free_args(struct image_args *);
109
110/* XXX This should be vm_size_t. */
111SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
112 NULL, 0, sysctl_kern_ps_strings, "LU", "");
113
114/* XXX This should be vm_size_t. */
115SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
116 NULL, 0, sysctl_kern_usrstack, "LU", "");
117
118SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
119 NULL, 0, sysctl_kern_stackprot, "I", "");
120
121u_long ps_arg_cache_limit = PAGE_SIZE / 16;
122SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
123 &ps_arg_cache_limit, 0, "");
124
125static int
126sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
127{
128 struct proc *p;
129 int error;
130
131 p = curproc;
132#ifdef SCTL_MASK32
133 if (req->flags & SCTL_MASK32) {
134 unsigned int val;
135 val = (unsigned int)p->p_sysent->sv_psstrings;
136 error = SYSCTL_OUT(req, &val, sizeof(val));
137 } else
138#endif
139 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
140 sizeof(p->p_sysent->sv_psstrings));
141 return error;
142}
143
144static int
145sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
146{
147 struct proc *p;
148 int error;
149
150 p = curproc;
151#ifdef SCTL_MASK32
152 if (req->flags & SCTL_MASK32) {
153 unsigned int val;
154 val = (unsigned int)p->p_sysent->sv_usrstack;
155 error = SYSCTL_OUT(req, &val, sizeof(val));
156 } else
157#endif
158 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
159 sizeof(p->p_sysent->sv_usrstack));
160 return error;
161}
162
163static int
164sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
165{
166 struct proc *p;
167
168 p = curproc;
169 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
170 sizeof(p->p_sysent->sv_stackprot)));
171}
172
173/*
174 * Each of the items is a pointer to a `const struct execsw', hence the
175 * double pointer here.
176 */
177static const struct execsw **execsw;
178
179#ifndef _SYS_SYSPROTO_H_
180struct execve_args {
181 char *fname;
182 char **argv;
183 char **envv;
184};
185#endif
186
187int
188execve(td, uap)
189 struct thread *td;
190 struct execve_args /* {
191 char *fname;
192 char **argv;
193 char **envv;
194 } */ *uap;
195{
196 int error;
197 struct image_args args;
198
199 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
200 uap->argv, uap->envv);
201 if (error == 0)
202 error = kern_execve(td, &args, NULL);
203 return (error);
204}
205
206#ifndef _SYS_SYSPROTO_H_
207struct fexecve_args {
208 int fd;
209 char **argv;
210 char **envv;
211}
212#endif
213int
214fexecve(struct thread *td, struct fexecve_args *uap)
215{
216 int error;
217 struct image_args args;
218
219 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
220 uap->argv, uap->envv);
221 if (error == 0) {
222 args.fd = uap->fd;
223 error = kern_execve(td, &args, NULL);
224 }
225 return (error);
226}
227
228#ifndef _SYS_SYSPROTO_H_
229struct __mac_execve_args {
230 char *fname;
231 char **argv;
232 char **envv;
233 struct mac *mac_p;
234};
235#endif
236
237int
238__mac_execve(td, uap)
239 struct thread *td;
240 struct __mac_execve_args /* {
241 char *fname;
242 char **argv;
243 char **envv;
244 struct mac *mac_p;
245 } */ *uap;
246{
247#ifdef MAC
248 int error;
249 struct image_args args;
250
251 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
252 uap->argv, uap->envv);
253 if (error == 0)
254 error = kern_execve(td, &args, uap->mac_p);
255 return (error);
256#else
257 return (ENOSYS);
258#endif
259}
260
261/*
262 * XXX: kern_execve has the astonishing property of not always returning to
263 * the caller. If sufficiently bad things happen during the call to
264 * do_execve(), it can end up calling exit1(); as a result, callers must
265 * avoid doing anything which they might need to undo (e.g., allocating
266 * memory).
267 */
268int
269kern_execve(td, args, mac_p)
270 struct thread *td;
271 struct image_args *args;
272 struct mac *mac_p;
273{
274 struct proc *p = td->td_proc;
275 int error;
276
277 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
278 args->begin_envv - args->begin_argv);
279 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
280 args->endp - args->begin_envv);
281 if (p->p_flag & P_HADTHREADS) {
282 PROC_LOCK(p);
283 if (thread_single(SINGLE_BOUNDARY)) {
284 PROC_UNLOCK(p);
285 exec_free_args(args);
286 return (ERESTART); /* Try again later. */
287 }
288 PROC_UNLOCK(p);
289 }
290
291 error = do_execve(td, args, mac_p);
292
293 if (p->p_flag & P_HADTHREADS) {
294 PROC_LOCK(p);
295 /*
296 * If success, we upgrade to SINGLE_EXIT state to
297 * force other threads to suicide.
298 */
299 if (error == 0)
300 thread_single(SINGLE_EXIT);
301 else
302 thread_single_end();
303 PROC_UNLOCK(p);
304 }
305
306 return (error);
307}
308
309/*
310 * In-kernel implementation of execve(). All arguments are assumed to be
311 * userspace pointers from the passed thread.
312 */
313static int
314do_execve(td, args, mac_p)
315 struct thread *td;
316 struct image_args *args;
317 struct mac *mac_p;
318{
319 struct proc *p = td->td_proc;
320 struct nameidata nd;
321 struct ucred *newcred = NULL, *oldcred;
322 struct uidinfo *euip;
323 register_t *stack_base;
324 int error, len = 0, i;
325 struct image_params image_params, *imgp;
326 struct vattr attr;
327 int (*img_first)(struct image_params *);
328 struct pargs *oldargs = NULL, *newargs = NULL;
329 struct sigacts *oldsigacts, *newsigacts;
330#ifdef KTRACE
331 struct vnode *tracevp = NULL;
332 struct ucred *tracecred = NULL;
333#endif
334 struct vnode *textvp = NULL, *binvp = NULL;
335 int credential_changing;
336 int vfslocked;
337 int textset;
338#ifdef MAC
339 struct label *interpvplabel = NULL;
340 int will_transition;
341#endif
342#ifdef HWPMC_HOOKS
343 struct pmckern_procexec pe;
344#endif
345 static const char fexecv_proc_title[] = "(fexecv)";
346
347 vfslocked = 0;
348 imgp = &image_params;
349
350 /*
351 * Lock the process and set the P_INEXEC flag to indicate that
352 * it should be left alone until we're done here. This is
353 * necessary to avoid race conditions - e.g. in ptrace() -
354 * that might allow a local user to illicitly obtain elevated
355 * privileges.
356 */
357 PROC_LOCK(p);
358 KASSERT((p->p_flag & P_INEXEC) == 0,
359 ("%s(): process already has P_INEXEC flag", __func__));
360 p->p_flag |= P_INEXEC;
361 PROC_UNLOCK(p);
362
363 /*
364 * Initialize part of the common data
365 */
366 imgp->proc = p;
367 imgp->execlabel = NULL;
368 imgp->attr = &attr;
369 imgp->entry_addr = 0;
370 imgp->vmspace_destroyed = 0;
371 imgp->interpreted = 0;
372 imgp->opened = 0;
373 imgp->interpreter_name = args->buf + PATH_MAX + ARG_MAX;
374 imgp->auxargs = NULL;
375 imgp->vp = NULL;
376 imgp->object = NULL;
377 imgp->firstpage = NULL;
378 imgp->ps_strings = 0;
379 imgp->auxarg_size = 0;
380 imgp->args = args;
381 imgp->execpath = imgp->freepath = NULL;
382 imgp->execpathp = 0;
383
384#ifdef MAC
385 error = mac_execve_enter(imgp, mac_p);
386 if (error)
387 goto exec_fail;
388#endif
389
390 imgp->image_header = NULL;
391
392 /*
393 * Translate the file name. namei() returns a vnode pointer
394 * in ni_vp amoung other things.
395 *
396 * XXXAUDIT: It would be desirable to also audit the name of the
397 * interpreter if this is an interpreted binary.
398 */
399 if (args->fname != NULL) {
400 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
401 | MPSAFE | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
402 }
403
404 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
405
406interpret:
407 if (args->fname != NULL) {
408 error = namei(&nd);
409 if (error)
410 goto exec_fail;
411
412 vfslocked = NDHASGIANT(&nd);
413 binvp = nd.ni_vp;
414 imgp->vp = binvp;
415 } else {
416 AUDIT_ARG_FD(args->fd);
417 error = fgetvp(td, args->fd, &binvp);
418 if (error)
419 goto exec_fail;
420 vfslocked = VFS_LOCK_GIANT(binvp->v_mount);
421 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
|
423 imgp->vp = binvp;
424 }
425
426 /*
427 * Check file permissions (also 'opens' file)
428 */
429 error = exec_check_permissions(imgp);
430 if (error)
431 goto exec_fail_dealloc;
432
433 imgp->object = imgp->vp->v_object;
434 if (imgp->object != NULL)
435 vm_object_reference(imgp->object);
436
437 /*
438 * Set VV_TEXT now so no one can write to the executable while we're
439 * activating it.
440 *
441 * Remember if this was set before and unset it in case this is not
442 * actually an executable image.
443 */
444 textset = imgp->vp->v_vflag & VV_TEXT;
445 imgp->vp->v_vflag |= VV_TEXT;
446
447 error = exec_map_first_page(imgp);
448 if (error)
449 goto exec_fail_dealloc;
450
451 imgp->proc->p_osrel = 0;
452 /*
453 * If the current process has a special image activator it
454 * wants to try first, call it. For example, emulating shell
455 * scripts differently.
456 */
457 error = -1;
458 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
459 error = img_first(imgp);
460
461 /*
462 * Loop through the list of image activators, calling each one.
463 * An activator returns -1 if there is no match, 0 on success,
464 * and an error otherwise.
465 */
466 for (i = 0; error == -1 && execsw[i]; ++i) {
467 if (execsw[i]->ex_imgact == NULL ||
468 execsw[i]->ex_imgact == img_first) {
469 continue;
470 }
471 error = (*execsw[i]->ex_imgact)(imgp);
472 }
473
474 if (error) {
475 if (error == -1) {
476 if (textset == 0)
477 imgp->vp->v_vflag &= ~VV_TEXT;
478 error = ENOEXEC;
479 }
480 goto exec_fail_dealloc;
481 }
482
483 /*
484 * Special interpreter operation, cleanup and loop up to try to
485 * activate the interpreter.
486 */
487 if (imgp->interpreted) {
488 exec_unmap_first_page(imgp);
489 /*
490 * VV_TEXT needs to be unset for scripts. There is a short
491 * period before we determine that something is a script where
492 * VV_TEXT will be set. The vnode lock is held over this
493 * entire period so nothing should illegitimately be blocked.
494 */
495 imgp->vp->v_vflag &= ~VV_TEXT;
496 /* free name buffer and old vnode */
497 if (args->fname != NULL)
498 NDFREE(&nd, NDF_ONLY_PNBUF);
499#ifdef MAC
500 mac_execve_interpreter_enter(binvp, &interpvplabel);
501#endif
502 if (imgp->opened) {
503 VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
504 imgp->opened = 0;
505 }
506 vput(binvp);
507 vm_object_deallocate(imgp->object);
508 imgp->object = NULL;
509 VFS_UNLOCK_GIANT(vfslocked);
510 vfslocked = 0;
511 /* set new name to that of the interpreter */
512 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
513 UIO_SYSSPACE, imgp->interpreter_name, td);
514 args->fname = imgp->interpreter_name;
515 goto interpret;
516 }
517
518 /*
519 * NB: We unlock the vnode here because it is believed that none
520 * of the sv_copyout_strings/sv_fixup operations require the vnode.
521 */
522 VOP_UNLOCK(imgp->vp, 0);
523
524 /*
525 * Do the best to calculate the full path to the image file.
526 */
527 if (imgp->auxargs != NULL &&
528 ((args->fname != NULL && args->fname[0] == '/') ||
529 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
530 imgp->execpath = args->fname;
531
532 /*
533 * Copy out strings (args and env) and initialize stack base
534 */
535 if (p->p_sysent->sv_copyout_strings)
536 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
537 else
538 stack_base = exec_copyout_strings(imgp);
539
540 /*
541 * If custom stack fixup routine present for this process
542 * let it do the stack setup.
543 * Else stuff argument count as first item on stack
544 */
545 if (p->p_sysent->sv_fixup != NULL)
546 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
547 else
548 suword(--stack_base, imgp->args->argc);
549
550 /*
551 * For security and other reasons, the file descriptor table cannot
552 * be shared after an exec.
553 */
554 fdunshare(p, td);
555
556 /*
557 * Malloc things before we need locks.
558 */
559 newcred = crget();
560 euip = uifind(attr.va_uid);
561 i = imgp->args->begin_envv - imgp->args->begin_argv;
562 /* Cache arguments if they fit inside our allowance */
563 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
564 newargs = pargs_alloc(i);
565 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
566 }
567
568 /* close files on exec */
569 fdcloseexec(td);
570 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
571
572 /* Get a reference to the vnode prior to locking the proc */
573 VREF(binvp);
574
575 /*
576 * For security and other reasons, signal handlers cannot
577 * be shared after an exec. The new process gets a copy of the old
578 * handlers. In execsigs(), the new process will have its signals
579 * reset.
580 */
581 PROC_LOCK(p);
582 oldcred = crcopysafe(p, newcred);
583 if (sigacts_shared(p->p_sigacts)) {
584 oldsigacts = p->p_sigacts;
585 PROC_UNLOCK(p);
586 newsigacts = sigacts_alloc();
587 sigacts_copy(newsigacts, oldsigacts);
588 PROC_LOCK(p);
589 p->p_sigacts = newsigacts;
590 } else
591 oldsigacts = NULL;
592
593 /* Stop profiling */
594 stopprofclock(p);
595
596 /* reset caught signals */
597 execsigs(p);
598
599 /* name this process - nameiexec(p, ndp) */
600 if (args->fname) {
601 len = min(nd.ni_cnd.cn_namelen,MAXCOMLEN);
602 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, len);
603 } else {
604 if (vn_commname(binvp, p->p_comm, MAXCOMLEN + 1) == 0)
605 len = MAXCOMLEN;
606 else {
607 len = sizeof(fexecv_proc_title);
608 bcopy(fexecv_proc_title, p->p_comm, len);
609 }
610 }
611 p->p_comm[len] = 0;
612 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
613
614 /*
615 * mark as execed, wakeup the process that vforked (if any) and tell
616 * it that it now has its own resources back
617 */
618 p->p_flag |= P_EXEC;
619 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
620 p->p_flag &= ~P_PPWAIT;
621 cv_broadcast(&p->p_pwait);
622 }
623
624 /*
625 * Implement image setuid/setgid.
626 *
627 * Don't honor setuid/setgid if the filesystem prohibits it or if
628 * the process is being traced.
629 *
630 * XXXMAC: For the time being, use NOSUID to also prohibit
631 * transitions on the file system.
632 */
633 credential_changing = 0;
634 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
635 attr.va_uid;
636 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
637 attr.va_gid;
638#ifdef MAC
639 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
640 interpvplabel, imgp);
641 credential_changing |= will_transition;
642#endif
643
644 if (credential_changing &&
645 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
646 (p->p_flag & P_TRACED) == 0) {
647 /*
648 * Turn off syscall tracing for set-id programs, except for
649 * root. Record any set-id flags first to make sure that
650 * we do not regain any tracing during a possible block.
651 */
652 setsugid(p);
653
654#ifdef KTRACE
655 if (p->p_tracevp != NULL &&
656 priv_check_cred(oldcred, PRIV_DEBUG_DIFFCRED, 0)) {
657 mtx_lock(&ktrace_mtx);
658 p->p_traceflag = 0;
659 tracevp = p->p_tracevp;
660 p->p_tracevp = NULL;
661 tracecred = p->p_tracecred;
662 p->p_tracecred = NULL;
663 mtx_unlock(&ktrace_mtx);
664 }
665#endif
666 /*
667 * Close any file descriptors 0..2 that reference procfs,
668 * then make sure file descriptors 0..2 are in use.
669 *
670 * setugidsafety() may call closef() and then pfind()
671 * which may grab the process lock.
672 * fdcheckstd() may call falloc() which may block to
673 * allocate memory, so temporarily drop the process lock.
674 */
675 PROC_UNLOCK(p);
676 setugidsafety(td);
677 VOP_UNLOCK(imgp->vp, 0);
678 error = fdcheckstd(td);
679 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
680 if (error != 0)
681 goto done1;
682 PROC_LOCK(p);
683 /*
684 * Set the new credentials.
685 */
686 if (attr.va_mode & S_ISUID)
687 change_euid(newcred, euip);
688 if (attr.va_mode & S_ISGID)
689 change_egid(newcred, attr.va_gid);
690#ifdef MAC
691 if (will_transition) {
692 mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
693 interpvplabel, imgp);
694 }
695#endif
696 /*
697 * Implement correct POSIX saved-id behavior.
698 *
699 * XXXMAC: Note that the current logic will save the
700 * uid and gid if a MAC domain transition occurs, even
701 * though maybe it shouldn't.
702 */
703 change_svuid(newcred, newcred->cr_uid);
704 change_svgid(newcred, newcred->cr_gid);
705 p->p_ucred = newcred;
706 newcred = NULL;
707 } else {
708 if (oldcred->cr_uid == oldcred->cr_ruid &&
709 oldcred->cr_gid == oldcred->cr_rgid)
710 p->p_flag &= ~P_SUGID;
711 /*
712 * Implement correct POSIX saved-id behavior.
713 *
714 * XXX: It's not clear that the existing behavior is
715 * POSIX-compliant. A number of sources indicate that the
716 * saved uid/gid should only be updated if the new ruid is
717 * not equal to the old ruid, or the new euid is not equal
718 * to the old euid and the new euid is not equal to the old
719 * ruid. The FreeBSD code always updates the saved uid/gid.
720 * Also, this code uses the new (replaced) euid and egid as
721 * the source, which may or may not be the right ones to use.
722 */
723 if (oldcred->cr_svuid != oldcred->cr_uid ||
724 oldcred->cr_svgid != oldcred->cr_gid) {
725 change_svuid(newcred, newcred->cr_uid);
726 change_svgid(newcred, newcred->cr_gid);
727 p->p_ucred = newcred;
728 newcred = NULL;
729 }
730 }
731
732 /*
733 * Store the vp for use in procfs. This vnode was referenced prior
734 * to locking the proc lock.
735 */
736 textvp = p->p_textvp;
737 p->p_textvp = binvp;
738
739#ifdef KDTRACE_HOOKS
740 /*
741 * Tell the DTrace fasttrap provider about the exec if it
742 * has declared an interest.
743 */
744 if (dtrace_fasttrap_exec)
745 dtrace_fasttrap_exec(p);
746#endif
747
748 /*
749 * Notify others that we exec'd, and clear the P_INEXEC flag
750 * as we're now a bona fide freshly-execed process.
751 */
752 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
753 p->p_flag &= ~P_INEXEC;
754
755 /*
756 * If tracing the process, trap to debugger so breakpoints
757 * can be set before the program executes.
758 * Use tdsignal to deliver signal to current thread, use
759 * psignal may cause the signal to be delivered to wrong thread
760 * because that thread will exit, remember we are going to enter
761 * single thread mode.
762 */
763 if (p->p_flag & P_TRACED)
764 tdsignal(p, td, SIGTRAP, NULL);
765
766 /* clear "fork but no exec" flag, as we _are_ execing */
767 p->p_acflag &= ~AFORK;
768
769 /*
770 * Free any previous argument cache and replace it with
771 * the new argument cache, if any.
772 */
773 oldargs = p->p_args;
774 p->p_args = newargs;
775 newargs = NULL;
776
777#ifdef HWPMC_HOOKS
778 /*
779 * Check if system-wide sampling is in effect or if the
780 * current process is using PMCs. If so, do exec() time
781 * processing. This processing needs to happen AFTER the
782 * P_INEXEC flag is cleared.
783 *
784 * The proc lock needs to be released before taking the PMC
785 * SX.
786 */
787 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
788 PROC_UNLOCK(p);
789 pe.pm_credentialschanged = credential_changing;
790 pe.pm_entryaddr = imgp->entry_addr;
791
792 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
793 } else
794 PROC_UNLOCK(p);
795#else /* !HWPMC_HOOKS */
796 PROC_UNLOCK(p);
797#endif
798
799 /* Set values passed into the program in registers. */
800 if (p->p_sysent->sv_setregs)
801 (*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
802 (u_long)(uintptr_t)stack_base, imgp->ps_strings);
803 else
804 exec_setregs(td, imgp->entry_addr,
805 (u_long)(uintptr_t)stack_base, imgp->ps_strings);
806
807 vfs_mark_atime(imgp->vp, td->td_ucred);
808
809 SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0);
810
811done1:
812 /*
813 * Free any resources malloc'd earlier that we didn't use.
814 */
815 uifree(euip);
816 if (newcred == NULL)
817 crfree(oldcred);
818 else
819 crfree(newcred);
820 VOP_UNLOCK(imgp->vp, 0);
821
822 /*
823 * Handle deferred decrement of ref counts.
824 */
825 if (textvp != NULL) {
826 int tvfslocked;
827
828 tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
829 vrele(textvp);
830 VFS_UNLOCK_GIANT(tvfslocked);
831 }
832 if (binvp && error != 0)
833 vrele(binvp);
834#ifdef KTRACE
835 if (tracevp != NULL) {
836 int tvfslocked;
837
838 tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
839 vrele(tracevp);
840 VFS_UNLOCK_GIANT(tvfslocked);
841 }
842 if (tracecred != NULL)
843 crfree(tracecred);
844#endif
845 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
846 pargs_drop(oldargs);
847 pargs_drop(newargs);
848 if (oldsigacts != NULL)
849 sigacts_free(oldsigacts);
850
851exec_fail_dealloc:
852
853 /*
854 * free various allocated resources
855 */
856 if (imgp->firstpage != NULL)
857 exec_unmap_first_page(imgp);
858
859 if (imgp->vp != NULL) {
860 if (args->fname)
861 NDFREE(&nd, NDF_ONLY_PNBUF);
862 if (imgp->opened)
863 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
864 vput(imgp->vp);
865 }
866
867 if (imgp->object != NULL)
868 vm_object_deallocate(imgp->object);
869
870 free(imgp->freepath, M_TEMP);
871
872 if (error == 0) {
873 /*
874 * Stop the process here if its stop event mask has
875 * the S_EXEC bit set.
876 */
877 STOPEVENT(p, S_EXEC, 0);
878 goto done2;
879 }
880
881exec_fail:
882 /* we're done here, clear P_INEXEC */
883 PROC_LOCK(p);
884 p->p_flag &= ~P_INEXEC;
885 PROC_UNLOCK(p);
886
887 SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0);
888
889done2:
890#ifdef MAC
891 mac_execve_exit(imgp);
892 mac_execve_interpreter_exit(interpvplabel);
893#endif
894 VFS_UNLOCK_GIANT(vfslocked);
895 exec_free_args(args);
896
897 if (error && imgp->vmspace_destroyed) {
898 /* sorry, no more process anymore. exit gracefully */
899 exit1(td, W_EXITCODE(0, SIGABRT));
900 /* NOT REACHED */
901 }
902 return (error);
903}
904
905int
906exec_map_first_page(imgp)
907 struct image_params *imgp;
908{
909 int rv, i;
910 int initial_pagein;
911 vm_page_t ma[VM_INITIAL_PAGEIN];
912 vm_object_t object;
913
914 if (imgp->firstpage != NULL)
915 exec_unmap_first_page(imgp);
916
917 object = imgp->vp->v_object;
918 if (object == NULL)
919 return (EACCES);
920 VM_OBJECT_LOCK(object);
921#if VM_NRESERVLEVEL > 0
922 if ((object->flags & OBJ_COLORED) == 0) {
923 object->flags |= OBJ_COLORED;
924 object->pg_color = 0;
925 }
926#endif
927 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
928 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
929 initial_pagein = VM_INITIAL_PAGEIN;
930 if (initial_pagein > object->size)
931 initial_pagein = object->size;
932 for (i = 1; i < initial_pagein; i++) {
933 if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
934 if (ma[i]->valid)
935 break;
936 if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy)
937 break;
938 vm_page_busy(ma[i]);
939 } else {
940 ma[i] = vm_page_alloc(object, i,
941 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
942 if (ma[i] == NULL)
943 break;
944 }
945 }
946 initial_pagein = i;
947 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
948 ma[0] = vm_page_lookup(object, 0);
949 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
950 if (ma[0]) {
951 vm_page_lock_queues();
952 vm_page_free(ma[0]);
953 vm_page_unlock_queues();
954 }
955 VM_OBJECT_UNLOCK(object);
956 return (EIO);
957 }
958 }
959 vm_page_lock_queues();
960 vm_page_hold(ma[0]);
961 vm_page_unlock_queues();
962 vm_page_wakeup(ma[0]);
963 VM_OBJECT_UNLOCK(object);
964
965 imgp->firstpage = sf_buf_alloc(ma[0], 0);
966 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
967
968 return (0);
969}
970
971void
972exec_unmap_first_page(imgp)
973 struct image_params *imgp;
974{
975 vm_page_t m;
976
977 if (imgp->firstpage != NULL) {
978 m = sf_buf_page(imgp->firstpage);
979 sf_buf_free(imgp->firstpage);
980 imgp->firstpage = NULL;
981 vm_page_lock_queues();
982 vm_page_unhold(m);
983 vm_page_unlock_queues();
984 }
985}
986
987/*
988 * Destroy old address space, and allocate a new stack
989 * The new stack is only SGROWSIZ large because it is grown
990 * automatically in trap.c.
991 */
992int
993exec_new_vmspace(imgp, sv)
994 struct image_params *imgp;
995 struct sysentvec *sv;
996{
997 int error;
998 struct proc *p = imgp->proc;
999 struct vmspace *vmspace = p->p_vmspace;
1000 vm_offset_t stack_addr;
1001 vm_map_t map;
1002 u_long ssiz;
1003
1004 imgp->vmspace_destroyed = 1;
1005 imgp->sysent = sv;
1006
1007 /* May be called with Giant held */
1008 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1009
1010 /*
1011 * Blow away entire process VM, if address space not shared,
1012 * otherwise, create a new VM space so that other threads are
1013 * not disrupted
1014 */
1015 map = &vmspace->vm_map;
1016 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
1017 vm_map_max(map) == sv->sv_maxuser) {
1018 shmexit(vmspace);
1019 pmap_remove_pages(vmspace_pmap(vmspace));
1020 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1021 } else {
1022 error = vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
1023 if (error)
1024 return (error);
1025 vmspace = p->p_vmspace;
1026 map = &vmspace->vm_map;
1027 }
1028
1029 /* Allocate a new stack */
1030 if (sv->sv_maxssiz != NULL)
1031 ssiz = *sv->sv_maxssiz;
1032 else
1033 ssiz = maxssiz;
1034 stack_addr = sv->sv_usrstack - ssiz;
1035 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1036 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1037 if (error)
1038 return (error);
1039
1040#ifdef __ia64__
1041 /* Allocate a new register stack */
1042 stack_addr = IA64_BACKINGSTORE;
1043 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1044 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1045 if (error)
1046 return (error);
1047#endif
1048
1049 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1050 * VM_STACK case, but they are still used to monitor the size of the
1051 * process stack so we can check the stack rlimit.
1052 */
1053 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1054 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1055
1056 return (0);
1057}
1058
1059/*
1060 * Copy out argument and environment strings from the old process address
1061 * space into the temporary string buffer.
1062 */
1063int
1064exec_copyin_args(struct image_args *args, char *fname,
1065 enum uio_seg segflg, char **argv, char **envv)
1066{
1067 char *argp, *envp;
1068 int error;
1069 size_t length;
1070
1071 bzero(args, sizeof(*args));
1072 if (argv == NULL)
1073 return (EFAULT);
1074 /*
1075 * Allocate temporary demand zeroed space for argument and
1076 * environment strings:
1077 *
1078 * o ARG_MAX for argument and environment;
1079 * o MAXSHELLCMDLEN for the name of interpreters.
1080 */
1081 args->buf = (char *) kmem_alloc_wait(exec_map,
1082 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1083 if (args->buf == NULL)
1084 return (ENOMEM);
1085 args->begin_argv = args->buf;
1086 args->endp = args->begin_argv;
1087 args->stringspace = ARG_MAX;
1088 /*
1089 * Copy the file name.
1090 */
1091 if (fname != NULL) {
1092 args->fname = args->buf + ARG_MAX;
1093 error = (segflg == UIO_SYSSPACE) ?
1094 copystr(fname, args->fname, PATH_MAX, &length) :
1095 copyinstr(fname, args->fname, PATH_MAX, &length);
1096 if (error != 0)
1097 goto err_exit;
1098 } else
1099 args->fname = NULL;
1100
1101 /*
1102 * extract arguments first
1103 */
1104 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1105 if (argp == (caddr_t) -1) {
1106 error = EFAULT;
1107 goto err_exit;
1108 }
1109 if ((error = copyinstr(argp, args->endp,
1110 args->stringspace, &length))) {
1111 if (error == ENAMETOOLONG)
1112 error = E2BIG;
1113 goto err_exit;
1114 }
1115 args->stringspace -= length;
1116 args->endp += length;
1117 args->argc++;
1118 }
1119
1120 args->begin_envv = args->endp;
1121
1122 /*
1123 * extract environment strings
1124 */
1125 if (envv) {
1126 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1127 if (envp == (caddr_t)-1) {
1128 error = EFAULT;
1129 goto err_exit;
1130 }
1131 if ((error = copyinstr(envp, args->endp,
1132 args->stringspace, &length))) {
1133 if (error == ENAMETOOLONG)
1134 error = E2BIG;
1135 goto err_exit;
1136 }
1137 args->stringspace -= length;
1138 args->endp += length;
1139 args->envc++;
1140 }
1141 }
1142
1143 return (0);
1144
1145err_exit:
1146 exec_free_args(args);
1147 return (error);
1148}
1149
1150static void
1151exec_free_args(struct image_args *args)
1152{
1153
1154 if (args->buf) {
1155 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1156 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1157 args->buf = NULL;
1158 }
1159}
1160
1161/*
1162 * Copy strings out to the new process address space, constructing new arg
1163 * and env vector tables. Return a pointer to the base so that it can be used
1164 * as the initial stack pointer.
1165 */
1166register_t *
1167exec_copyout_strings(imgp)
1168 struct image_params *imgp;
1169{
1170 int argc, envc;
1171 char **vectp;
1172 char *stringp, *destp;
1173 register_t *stack_base;
1174 struct ps_strings *arginfo;
1175 struct proc *p;
1176 size_t execpath_len;
1177 int szsigcode;
1178
1179 /*
1180 * Calculate string base and vector table pointers.
1181 * Also deal with signal trampoline code for this exec type.
1182 */
1183 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1184 execpath_len = strlen(imgp->execpath) + 1;
1185 else
1186 execpath_len = 0;
1187 p = imgp->proc;
1188 szsigcode = 0;
1189 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1190 if (p->p_sysent->sv_szsigcode != NULL)
1191 szsigcode = *(p->p_sysent->sv_szsigcode);
1192 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1193 roundup(execpath_len, sizeof(char *)) -
1194 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1195
1196 /*
1197 * install sigcode
1198 */
1199 if (szsigcode)
1200 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1201 szsigcode), szsigcode);
1202
1203 /*
1204 * Copy the image path for the rtld.
1205 */
1206 if (execpath_len != 0) {
1207 imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
1208 copyout(imgp->execpath, (void *)imgp->execpathp,
1209 execpath_len);
1210 }
1211
1212 /*
1213 * If we have a valid auxargs ptr, prepare some room
1214 * on the stack.
1215 */
1216 if (imgp->auxargs) {
1217 /*
1218 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1219 * lower compatibility.
1220 */
1221 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1222 (AT_COUNT * 2);
1223 /*
1224 * The '+ 2' is for the null pointers at the end of each of
1225 * the arg and env vector sets,and imgp->auxarg_size is room
1226 * for argument of Runtime loader.
1227 */
1228 vectp = (char **)(destp - (imgp->args->argc +
1229 imgp->args->envc + 2 + imgp->auxarg_size + execpath_len) *
1230 sizeof(char *));
1231 } else {
1232 /*
1233 * The '+ 2' is for the null pointers at the end of each of
1234 * the arg and env vector sets
1235 */
1236 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1237 sizeof(char *));
1238 }
1239
1240 /*
1241 * vectp also becomes our initial stack base
1242 */
1243 stack_base = (register_t *)vectp;
1244
1245 stringp = imgp->args->begin_argv;
1246 argc = imgp->args->argc;
1247 envc = imgp->args->envc;
1248
1249 /*
1250 * Copy out strings - arguments and environment.
1251 */
1252 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1253
1254 /*
1255 * Fill in "ps_strings" struct for ps, w, etc.
1256 */
1257 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1258 suword(&arginfo->ps_nargvstr, argc);
1259
1260 /*
1261 * Fill in argument portion of vector table.
1262 */
1263 for (; argc > 0; --argc) {
1264 suword(vectp++, (long)(intptr_t)destp);
1265 while (*stringp++ != 0)
1266 destp++;
1267 destp++;
1268 }
1269
1270 /* a null vector table pointer separates the argp's from the envp's */
1271 suword(vectp++, 0);
1272
1273 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1274 suword(&arginfo->ps_nenvstr, envc);
1275
1276 /*
1277 * Fill in environment portion of vector table.
1278 */
1279 for (; envc > 0; --envc) {
1280 suword(vectp++, (long)(intptr_t)destp);
1281 while (*stringp++ != 0)
1282 destp++;
1283 destp++;
1284 }
1285
1286 /* end of vector table is a null pointer */
1287 suword(vectp, 0);
1288
1289 return (stack_base);
1290}
1291
1292/*
1293 * Check permissions of file to execute.
1294 * Called with imgp->vp locked.
1295 * Return 0 for success or error code on failure.
1296 */
1297int
1298exec_check_permissions(imgp)
1299 struct image_params *imgp;
1300{
1301 struct vnode *vp = imgp->vp;
1302 struct vattr *attr = imgp->attr;
1303 struct thread *td;
1304 int error;
1305
1306 td = curthread;
1307
1308 /* Get file attributes */
1309 error = VOP_GETATTR(vp, attr, td->td_ucred);
1310 if (error)
1311 return (error);
1312
1313#ifdef MAC
1314 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1315 if (error)
1316 return (error);
1317#endif
1318
1319 /*
1320 * 1) Check if file execution is disabled for the filesystem that this
1321 * file resides on.
1322 * 2) Insure that at least one execute bit is on - otherwise root
1323 * will always succeed, and we don't want to happen unless the
1324 * file really is executable.
1325 * 3) Insure that the file is a regular file.
1326 */
1327 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1328 ((attr->va_mode & 0111) == 0) ||
1329 (attr->va_type != VREG))
1330 return (EACCES);
1331
1332 /*
1333 * Zero length files can't be exec'd
1334 */
1335 if (attr->va_size == 0)
1336 return (ENOEXEC);
1337
1338 /*
1339 * Check for execute permission to file based on current credentials.
1340 */
1341 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1342 if (error)
1343 return (error);
1344
1345 /*
1346 * Check number of open-for-writes on the file and deny execution
1347 * if there are any.
1348 */
1349 if (vp->v_writecount)
1350 return (ETXTBSY);
1351
1352 /*
1353 * Call filesystem specific open routine (which does nothing in the
1354 * general case).
1355 */
1356 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1357 if (error == 0)
1358 imgp->opened = 1;
1359 return (error);
1360}
1361
1362/*
1363 * Exec handler registration
1364 */
1365int
1366exec_register(execsw_arg)
1367 const struct execsw *execsw_arg;
1368{
1369 const struct execsw **es, **xs, **newexecsw;
1370 int count = 2; /* New slot and trailing NULL */
1371
1372 if (execsw)
1373 for (es = execsw; *es; es++)
1374 count++;
1375 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1376 if (newexecsw == NULL)
1377 return (ENOMEM);
1378 xs = newexecsw;
1379 if (execsw)
1380 for (es = execsw; *es; es++)
1381 *xs++ = *es;
1382 *xs++ = execsw_arg;
1383 *xs = NULL;
1384 if (execsw)
1385 free(execsw, M_TEMP);
1386 execsw = newexecsw;
1387 return (0);
1388}
1389
1390int
1391exec_unregister(execsw_arg)
1392 const struct execsw *execsw_arg;
1393{
1394 const struct execsw **es, **xs, **newexecsw;
1395 int count = 1;
1396
1397 if (execsw == NULL)
1398 panic("unregister with no handlers left?\n");
1399
1400 for (es = execsw; *es; es++) {
1401 if (*es == execsw_arg)
1402 break;
1403 }
1404 if (*es == NULL)
1405 return (ENOENT);
1406 for (es = execsw; *es; es++)
1407 if (*es != execsw_arg)
1408 count++;
1409 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1410 if (newexecsw == NULL)
1411 return (ENOMEM);
1412 xs = newexecsw;
1413 for (es = execsw; *es; es++)
1414 if (*es != execsw_arg)
1415 *xs++ = *es;
1416 *xs = NULL;
1417 if (execsw)
1418 free(execsw, M_TEMP);
1419 execsw = newexecsw;
1420 return (0);
1421}
| 423 imgp->vp = binvp;
424 }
425
426 /*
427 * Check file permissions (also 'opens' file)
428 */
429 error = exec_check_permissions(imgp);
430 if (error)
431 goto exec_fail_dealloc;
432
433 imgp->object = imgp->vp->v_object;
434 if (imgp->object != NULL)
435 vm_object_reference(imgp->object);
436
437 /*
438 * Set VV_TEXT now so no one can write to the executable while we're
439 * activating it.
440 *
441 * Remember if this was set before and unset it in case this is not
442 * actually an executable image.
443 */
444 textset = imgp->vp->v_vflag & VV_TEXT;
445 imgp->vp->v_vflag |= VV_TEXT;
446
447 error = exec_map_first_page(imgp);
448 if (error)
449 goto exec_fail_dealloc;
450
451 imgp->proc->p_osrel = 0;
452 /*
453 * If the current process has a special image activator it
454 * wants to try first, call it. For example, emulating shell
455 * scripts differently.
456 */
457 error = -1;
458 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
459 error = img_first(imgp);
460
461 /*
462 * Loop through the list of image activators, calling each one.
463 * An activator returns -1 if there is no match, 0 on success,
464 * and an error otherwise.
465 */
466 for (i = 0; error == -1 && execsw[i]; ++i) {
467 if (execsw[i]->ex_imgact == NULL ||
468 execsw[i]->ex_imgact == img_first) {
469 continue;
470 }
471 error = (*execsw[i]->ex_imgact)(imgp);
472 }
473
474 if (error) {
475 if (error == -1) {
476 if (textset == 0)
477 imgp->vp->v_vflag &= ~VV_TEXT;
478 error = ENOEXEC;
479 }
480 goto exec_fail_dealloc;
481 }
482
483 /*
484 * Special interpreter operation, cleanup and loop up to try to
485 * activate the interpreter.
486 */
487 if (imgp->interpreted) {
488 exec_unmap_first_page(imgp);
489 /*
490 * VV_TEXT needs to be unset for scripts. There is a short
491 * period before we determine that something is a script where
492 * VV_TEXT will be set. The vnode lock is held over this
493 * entire period so nothing should illegitimately be blocked.
494 */
495 imgp->vp->v_vflag &= ~VV_TEXT;
496 /* free name buffer and old vnode */
497 if (args->fname != NULL)
498 NDFREE(&nd, NDF_ONLY_PNBUF);
499#ifdef MAC
500 mac_execve_interpreter_enter(binvp, &interpvplabel);
501#endif
502 if (imgp->opened) {
503 VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
504 imgp->opened = 0;
505 }
506 vput(binvp);
507 vm_object_deallocate(imgp->object);
508 imgp->object = NULL;
509 VFS_UNLOCK_GIANT(vfslocked);
510 vfslocked = 0;
511 /* set new name to that of the interpreter */
512 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
513 UIO_SYSSPACE, imgp->interpreter_name, td);
514 args->fname = imgp->interpreter_name;
515 goto interpret;
516 }
517
518 /*
519 * NB: We unlock the vnode here because it is believed that none
520 * of the sv_copyout_strings/sv_fixup operations require the vnode.
521 */
522 VOP_UNLOCK(imgp->vp, 0);
523
524 /*
525 * Do the best to calculate the full path to the image file.
526 */
527 if (imgp->auxargs != NULL &&
528 ((args->fname != NULL && args->fname[0] == '/') ||
529 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
530 imgp->execpath = args->fname;
531
532 /*
533 * Copy out strings (args and env) and initialize stack base
534 */
535 if (p->p_sysent->sv_copyout_strings)
536 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
537 else
538 stack_base = exec_copyout_strings(imgp);
539
540 /*
541 * If custom stack fixup routine present for this process
542 * let it do the stack setup.
543 * Else stuff argument count as first item on stack
544 */
545 if (p->p_sysent->sv_fixup != NULL)
546 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
547 else
548 suword(--stack_base, imgp->args->argc);
549
550 /*
551 * For security and other reasons, the file descriptor table cannot
552 * be shared after an exec.
553 */
554 fdunshare(p, td);
555
556 /*
557 * Malloc things before we need locks.
558 */
559 newcred = crget();
560 euip = uifind(attr.va_uid);
561 i = imgp->args->begin_envv - imgp->args->begin_argv;
562 /* Cache arguments if they fit inside our allowance */
563 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
564 newargs = pargs_alloc(i);
565 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
566 }
567
568 /* close files on exec */
569 fdcloseexec(td);
570 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
571
572 /* Get a reference to the vnode prior to locking the proc */
573 VREF(binvp);
574
575 /*
576 * For security and other reasons, signal handlers cannot
577 * be shared after an exec. The new process gets a copy of the old
578 * handlers. In execsigs(), the new process will have its signals
579 * reset.
580 */
581 PROC_LOCK(p);
582 oldcred = crcopysafe(p, newcred);
583 if (sigacts_shared(p->p_sigacts)) {
584 oldsigacts = p->p_sigacts;
585 PROC_UNLOCK(p);
586 newsigacts = sigacts_alloc();
587 sigacts_copy(newsigacts, oldsigacts);
588 PROC_LOCK(p);
589 p->p_sigacts = newsigacts;
590 } else
591 oldsigacts = NULL;
592
593 /* Stop profiling */
594 stopprofclock(p);
595
596 /* reset caught signals */
597 execsigs(p);
598
599 /* name this process - nameiexec(p, ndp) */
600 if (args->fname) {
601 len = min(nd.ni_cnd.cn_namelen,MAXCOMLEN);
602 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, len);
603 } else {
604 if (vn_commname(binvp, p->p_comm, MAXCOMLEN + 1) == 0)
605 len = MAXCOMLEN;
606 else {
607 len = sizeof(fexecv_proc_title);
608 bcopy(fexecv_proc_title, p->p_comm, len);
609 }
610 }
611 p->p_comm[len] = 0;
612 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
613
614 /*
615 * mark as execed, wakeup the process that vforked (if any) and tell
616 * it that it now has its own resources back
617 */
618 p->p_flag |= P_EXEC;
619 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
620 p->p_flag &= ~P_PPWAIT;
621 cv_broadcast(&p->p_pwait);
622 }
623
624 /*
625 * Implement image setuid/setgid.
626 *
627 * Don't honor setuid/setgid if the filesystem prohibits it or if
628 * the process is being traced.
629 *
630 * XXXMAC: For the time being, use NOSUID to also prohibit
631 * transitions on the file system.
632 */
633 credential_changing = 0;
634 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
635 attr.va_uid;
636 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
637 attr.va_gid;
638#ifdef MAC
639 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
640 interpvplabel, imgp);
641 credential_changing |= will_transition;
642#endif
643
644 if (credential_changing &&
645 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
646 (p->p_flag & P_TRACED) == 0) {
647 /*
648 * Turn off syscall tracing for set-id programs, except for
649 * root. Record any set-id flags first to make sure that
650 * we do not regain any tracing during a possible block.
651 */
652 setsugid(p);
653
654#ifdef KTRACE
655 if (p->p_tracevp != NULL &&
656 priv_check_cred(oldcred, PRIV_DEBUG_DIFFCRED, 0)) {
657 mtx_lock(&ktrace_mtx);
658 p->p_traceflag = 0;
659 tracevp = p->p_tracevp;
660 p->p_tracevp = NULL;
661 tracecred = p->p_tracecred;
662 p->p_tracecred = NULL;
663 mtx_unlock(&ktrace_mtx);
664 }
665#endif
666 /*
667 * Close any file descriptors 0..2 that reference procfs,
668 * then make sure file descriptors 0..2 are in use.
669 *
670 * setugidsafety() may call closef() and then pfind()
671 * which may grab the process lock.
672 * fdcheckstd() may call falloc() which may block to
673 * allocate memory, so temporarily drop the process lock.
674 */
675 PROC_UNLOCK(p);
676 setugidsafety(td);
677 VOP_UNLOCK(imgp->vp, 0);
678 error = fdcheckstd(td);
679 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
680 if (error != 0)
681 goto done1;
682 PROC_LOCK(p);
683 /*
684 * Set the new credentials.
685 */
686 if (attr.va_mode & S_ISUID)
687 change_euid(newcred, euip);
688 if (attr.va_mode & S_ISGID)
689 change_egid(newcred, attr.va_gid);
690#ifdef MAC
691 if (will_transition) {
692 mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
693 interpvplabel, imgp);
694 }
695#endif
696 /*
697 * Implement correct POSIX saved-id behavior.
698 *
699 * XXXMAC: Note that the current logic will save the
700 * uid and gid if a MAC domain transition occurs, even
701 * though maybe it shouldn't.
702 */
703 change_svuid(newcred, newcred->cr_uid);
704 change_svgid(newcred, newcred->cr_gid);
705 p->p_ucred = newcred;
706 newcred = NULL;
707 } else {
708 if (oldcred->cr_uid == oldcred->cr_ruid &&
709 oldcred->cr_gid == oldcred->cr_rgid)
710 p->p_flag &= ~P_SUGID;
711 /*
712 * Implement correct POSIX saved-id behavior.
713 *
714 * XXX: It's not clear that the existing behavior is
715 * POSIX-compliant. A number of sources indicate that the
716 * saved uid/gid should only be updated if the new ruid is
717 * not equal to the old ruid, or the new euid is not equal
718 * to the old euid and the new euid is not equal to the old
719 * ruid. The FreeBSD code always updates the saved uid/gid.
720 * Also, this code uses the new (replaced) euid and egid as
721 * the source, which may or may not be the right ones to use.
722 */
723 if (oldcred->cr_svuid != oldcred->cr_uid ||
724 oldcred->cr_svgid != oldcred->cr_gid) {
725 change_svuid(newcred, newcred->cr_uid);
726 change_svgid(newcred, newcred->cr_gid);
727 p->p_ucred = newcred;
728 newcred = NULL;
729 }
730 }
731
732 /*
733 * Store the vp for use in procfs. This vnode was referenced prior
734 * to locking the proc lock.
735 */
736 textvp = p->p_textvp;
737 p->p_textvp = binvp;
738
739#ifdef KDTRACE_HOOKS
740 /*
741 * Tell the DTrace fasttrap provider about the exec if it
742 * has declared an interest.
743 */
744 if (dtrace_fasttrap_exec)
745 dtrace_fasttrap_exec(p);
746#endif
747
748 /*
749 * Notify others that we exec'd, and clear the P_INEXEC flag
750 * as we're now a bona fide freshly-execed process.
751 */
752 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
753 p->p_flag &= ~P_INEXEC;
754
755 /*
756 * If tracing the process, trap to debugger so breakpoints
757 * can be set before the program executes.
758 * Use tdsignal to deliver signal to current thread, use
759 * psignal may cause the signal to be delivered to wrong thread
760 * because that thread will exit, remember we are going to enter
761 * single thread mode.
762 */
763 if (p->p_flag & P_TRACED)
764 tdsignal(p, td, SIGTRAP, NULL);
765
766 /* clear "fork but no exec" flag, as we _are_ execing */
767 p->p_acflag &= ~AFORK;
768
769 /*
770 * Free any previous argument cache and replace it with
771 * the new argument cache, if any.
772 */
773 oldargs = p->p_args;
774 p->p_args = newargs;
775 newargs = NULL;
776
777#ifdef HWPMC_HOOKS
778 /*
779 * Check if system-wide sampling is in effect or if the
780 * current process is using PMCs. If so, do exec() time
781 * processing. This processing needs to happen AFTER the
782 * P_INEXEC flag is cleared.
783 *
784 * The proc lock needs to be released before taking the PMC
785 * SX.
786 */
787 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
788 PROC_UNLOCK(p);
789 pe.pm_credentialschanged = credential_changing;
790 pe.pm_entryaddr = imgp->entry_addr;
791
792 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
793 } else
794 PROC_UNLOCK(p);
795#else /* !HWPMC_HOOKS */
796 PROC_UNLOCK(p);
797#endif
798
799 /* Set values passed into the program in registers. */
800 if (p->p_sysent->sv_setregs)
801 (*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
802 (u_long)(uintptr_t)stack_base, imgp->ps_strings);
803 else
804 exec_setregs(td, imgp->entry_addr,
805 (u_long)(uintptr_t)stack_base, imgp->ps_strings);
806
807 vfs_mark_atime(imgp->vp, td->td_ucred);
808
809 SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0);
810
811done1:
812 /*
813 * Free any resources malloc'd earlier that we didn't use.
814 */
815 uifree(euip);
816 if (newcred == NULL)
817 crfree(oldcred);
818 else
819 crfree(newcred);
820 VOP_UNLOCK(imgp->vp, 0);
821
822 /*
823 * Handle deferred decrement of ref counts.
824 */
825 if (textvp != NULL) {
826 int tvfslocked;
827
828 tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
829 vrele(textvp);
830 VFS_UNLOCK_GIANT(tvfslocked);
831 }
832 if (binvp && error != 0)
833 vrele(binvp);
834#ifdef KTRACE
835 if (tracevp != NULL) {
836 int tvfslocked;
837
838 tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
839 vrele(tracevp);
840 VFS_UNLOCK_GIANT(tvfslocked);
841 }
842 if (tracecred != NULL)
843 crfree(tracecred);
844#endif
845 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
846 pargs_drop(oldargs);
847 pargs_drop(newargs);
848 if (oldsigacts != NULL)
849 sigacts_free(oldsigacts);
850
851exec_fail_dealloc:
852
853 /*
854 * free various allocated resources
855 */
856 if (imgp->firstpage != NULL)
857 exec_unmap_first_page(imgp);
858
859 if (imgp->vp != NULL) {
860 if (args->fname)
861 NDFREE(&nd, NDF_ONLY_PNBUF);
862 if (imgp->opened)
863 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
864 vput(imgp->vp);
865 }
866
867 if (imgp->object != NULL)
868 vm_object_deallocate(imgp->object);
869
870 free(imgp->freepath, M_TEMP);
871
872 if (error == 0) {
873 /*
874 * Stop the process here if its stop event mask has
875 * the S_EXEC bit set.
876 */
877 STOPEVENT(p, S_EXEC, 0);
878 goto done2;
879 }
880
881exec_fail:
882 /* we're done here, clear P_INEXEC */
883 PROC_LOCK(p);
884 p->p_flag &= ~P_INEXEC;
885 PROC_UNLOCK(p);
886
887 SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0);
888
889done2:
890#ifdef MAC
891 mac_execve_exit(imgp);
892 mac_execve_interpreter_exit(interpvplabel);
893#endif
894 VFS_UNLOCK_GIANT(vfslocked);
895 exec_free_args(args);
896
897 if (error && imgp->vmspace_destroyed) {
898 /* sorry, no more process anymore. exit gracefully */
899 exit1(td, W_EXITCODE(0, SIGABRT));
900 /* NOT REACHED */
901 }
902 return (error);
903}
904
905int
906exec_map_first_page(imgp)
907 struct image_params *imgp;
908{
909 int rv, i;
910 int initial_pagein;
911 vm_page_t ma[VM_INITIAL_PAGEIN];
912 vm_object_t object;
913
914 if (imgp->firstpage != NULL)
915 exec_unmap_first_page(imgp);
916
917 object = imgp->vp->v_object;
918 if (object == NULL)
919 return (EACCES);
920 VM_OBJECT_LOCK(object);
921#if VM_NRESERVLEVEL > 0
922 if ((object->flags & OBJ_COLORED) == 0) {
923 object->flags |= OBJ_COLORED;
924 object->pg_color = 0;
925 }
926#endif
927 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
928 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
929 initial_pagein = VM_INITIAL_PAGEIN;
930 if (initial_pagein > object->size)
931 initial_pagein = object->size;
932 for (i = 1; i < initial_pagein; i++) {
933 if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
934 if (ma[i]->valid)
935 break;
936 if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy)
937 break;
938 vm_page_busy(ma[i]);
939 } else {
940 ma[i] = vm_page_alloc(object, i,
941 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
942 if (ma[i] == NULL)
943 break;
944 }
945 }
946 initial_pagein = i;
947 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
948 ma[0] = vm_page_lookup(object, 0);
949 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
950 if (ma[0]) {
951 vm_page_lock_queues();
952 vm_page_free(ma[0]);
953 vm_page_unlock_queues();
954 }
955 VM_OBJECT_UNLOCK(object);
956 return (EIO);
957 }
958 }
959 vm_page_lock_queues();
960 vm_page_hold(ma[0]);
961 vm_page_unlock_queues();
962 vm_page_wakeup(ma[0]);
963 VM_OBJECT_UNLOCK(object);
964
965 imgp->firstpage = sf_buf_alloc(ma[0], 0);
966 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
967
968 return (0);
969}
970
971void
972exec_unmap_first_page(imgp)
973 struct image_params *imgp;
974{
975 vm_page_t m;
976
977 if (imgp->firstpage != NULL) {
978 m = sf_buf_page(imgp->firstpage);
979 sf_buf_free(imgp->firstpage);
980 imgp->firstpage = NULL;
981 vm_page_lock_queues();
982 vm_page_unhold(m);
983 vm_page_unlock_queues();
984 }
985}
986
987/*
988 * Destroy old address space, and allocate a new stack
989 * The new stack is only SGROWSIZ large because it is grown
990 * automatically in trap.c.
991 */
992int
993exec_new_vmspace(imgp, sv)
994 struct image_params *imgp;
995 struct sysentvec *sv;
996{
997 int error;
998 struct proc *p = imgp->proc;
999 struct vmspace *vmspace = p->p_vmspace;
1000 vm_offset_t stack_addr;
1001 vm_map_t map;
1002 u_long ssiz;
1003
1004 imgp->vmspace_destroyed = 1;
1005 imgp->sysent = sv;
1006
1007 /* May be called with Giant held */
1008 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1009
1010 /*
1011 * Blow away entire process VM, if address space not shared,
1012 * otherwise, create a new VM space so that other threads are
1013 * not disrupted
1014 */
1015 map = &vmspace->vm_map;
1016 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
1017 vm_map_max(map) == sv->sv_maxuser) {
1018 shmexit(vmspace);
1019 pmap_remove_pages(vmspace_pmap(vmspace));
1020 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1021 } else {
1022 error = vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
1023 if (error)
1024 return (error);
1025 vmspace = p->p_vmspace;
1026 map = &vmspace->vm_map;
1027 }
1028
1029 /* Allocate a new stack */
1030 if (sv->sv_maxssiz != NULL)
1031 ssiz = *sv->sv_maxssiz;
1032 else
1033 ssiz = maxssiz;
1034 stack_addr = sv->sv_usrstack - ssiz;
1035 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1036 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1037 if (error)
1038 return (error);
1039
1040#ifdef __ia64__
1041 /* Allocate a new register stack */
1042 stack_addr = IA64_BACKINGSTORE;
1043 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1044 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1045 if (error)
1046 return (error);
1047#endif
1048
1049 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1050 * VM_STACK case, but they are still used to monitor the size of the
1051 * process stack so we can check the stack rlimit.
1052 */
1053 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1054 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1055
1056 return (0);
1057}
1058
1059/*
1060 * Copy out argument and environment strings from the old process address
1061 * space into the temporary string buffer.
1062 */
1063int
1064exec_copyin_args(struct image_args *args, char *fname,
1065 enum uio_seg segflg, char **argv, char **envv)
1066{
1067 char *argp, *envp;
1068 int error;
1069 size_t length;
1070
1071 bzero(args, sizeof(*args));
1072 if (argv == NULL)
1073 return (EFAULT);
1074 /*
1075 * Allocate temporary demand zeroed space for argument and
1076 * environment strings:
1077 *
1078 * o ARG_MAX for argument and environment;
1079 * o MAXSHELLCMDLEN for the name of interpreters.
1080 */
1081 args->buf = (char *) kmem_alloc_wait(exec_map,
1082 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1083 if (args->buf == NULL)
1084 return (ENOMEM);
1085 args->begin_argv = args->buf;
1086 args->endp = args->begin_argv;
1087 args->stringspace = ARG_MAX;
1088 /*
1089 * Copy the file name.
1090 */
1091 if (fname != NULL) {
1092 args->fname = args->buf + ARG_MAX;
1093 error = (segflg == UIO_SYSSPACE) ?
1094 copystr(fname, args->fname, PATH_MAX, &length) :
1095 copyinstr(fname, args->fname, PATH_MAX, &length);
1096 if (error != 0)
1097 goto err_exit;
1098 } else
1099 args->fname = NULL;
1100
1101 /*
1102 * extract arguments first
1103 */
1104 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1105 if (argp == (caddr_t) -1) {
1106 error = EFAULT;
1107 goto err_exit;
1108 }
1109 if ((error = copyinstr(argp, args->endp,
1110 args->stringspace, &length))) {
1111 if (error == ENAMETOOLONG)
1112 error = E2BIG;
1113 goto err_exit;
1114 }
1115 args->stringspace -= length;
1116 args->endp += length;
1117 args->argc++;
1118 }
1119
1120 args->begin_envv = args->endp;
1121
1122 /*
1123 * extract environment strings
1124 */
1125 if (envv) {
1126 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1127 if (envp == (caddr_t)-1) {
1128 error = EFAULT;
1129 goto err_exit;
1130 }
1131 if ((error = copyinstr(envp, args->endp,
1132 args->stringspace, &length))) {
1133 if (error == ENAMETOOLONG)
1134 error = E2BIG;
1135 goto err_exit;
1136 }
1137 args->stringspace -= length;
1138 args->endp += length;
1139 args->envc++;
1140 }
1141 }
1142
1143 return (0);
1144
1145err_exit:
1146 exec_free_args(args);
1147 return (error);
1148}
1149
1150static void
1151exec_free_args(struct image_args *args)
1152{
1153
1154 if (args->buf) {
1155 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1156 PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1157 args->buf = NULL;
1158 }
1159}
1160
1161/*
1162 * Copy strings out to the new process address space, constructing new arg
1163 * and env vector tables. Return a pointer to the base so that it can be used
1164 * as the initial stack pointer.
1165 */
1166register_t *
1167exec_copyout_strings(imgp)
1168 struct image_params *imgp;
1169{
1170 int argc, envc;
1171 char **vectp;
1172 char *stringp, *destp;
1173 register_t *stack_base;
1174 struct ps_strings *arginfo;
1175 struct proc *p;
1176 size_t execpath_len;
1177 int szsigcode;
1178
1179 /*
1180 * Calculate string base and vector table pointers.
1181 * Also deal with signal trampoline code for this exec type.
1182 */
1183 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1184 execpath_len = strlen(imgp->execpath) + 1;
1185 else
1186 execpath_len = 0;
1187 p = imgp->proc;
1188 szsigcode = 0;
1189 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1190 if (p->p_sysent->sv_szsigcode != NULL)
1191 szsigcode = *(p->p_sysent->sv_szsigcode);
1192 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1193 roundup(execpath_len, sizeof(char *)) -
1194 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1195
1196 /*
1197 * install sigcode
1198 */
1199 if (szsigcode)
1200 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1201 szsigcode), szsigcode);
1202
1203 /*
1204 * Copy the image path for the rtld.
1205 */
1206 if (execpath_len != 0) {
1207 imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
1208 copyout(imgp->execpath, (void *)imgp->execpathp,
1209 execpath_len);
1210 }
1211
1212 /*
1213 * If we have a valid auxargs ptr, prepare some room
1214 * on the stack.
1215 */
1216 if (imgp->auxargs) {
1217 /*
1218 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1219 * lower compatibility.
1220 */
1221 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1222 (AT_COUNT * 2);
1223 /*
1224 * The '+ 2' is for the null pointers at the end of each of
1225 * the arg and env vector sets,and imgp->auxarg_size is room
1226 * for argument of Runtime loader.
1227 */
1228 vectp = (char **)(destp - (imgp->args->argc +
1229 imgp->args->envc + 2 + imgp->auxarg_size + execpath_len) *
1230 sizeof(char *));
1231 } else {
1232 /*
1233 * The '+ 2' is for the null pointers at the end of each of
1234 * the arg and env vector sets
1235 */
1236 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1237 sizeof(char *));
1238 }
1239
1240 /*
1241 * vectp also becomes our initial stack base
1242 */
1243 stack_base = (register_t *)vectp;
1244
1245 stringp = imgp->args->begin_argv;
1246 argc = imgp->args->argc;
1247 envc = imgp->args->envc;
1248
1249 /*
1250 * Copy out strings - arguments and environment.
1251 */
1252 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1253
1254 /*
1255 * Fill in "ps_strings" struct for ps, w, etc.
1256 */
1257 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1258 suword(&arginfo->ps_nargvstr, argc);
1259
1260 /*
1261 * Fill in argument portion of vector table.
1262 */
1263 for (; argc > 0; --argc) {
1264 suword(vectp++, (long)(intptr_t)destp);
1265 while (*stringp++ != 0)
1266 destp++;
1267 destp++;
1268 }
1269
1270 /* a null vector table pointer separates the argp's from the envp's */
1271 suword(vectp++, 0);
1272
1273 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1274 suword(&arginfo->ps_nenvstr, envc);
1275
1276 /*
1277 * Fill in environment portion of vector table.
1278 */
1279 for (; envc > 0; --envc) {
1280 suword(vectp++, (long)(intptr_t)destp);
1281 while (*stringp++ != 0)
1282 destp++;
1283 destp++;
1284 }
1285
1286 /* end of vector table is a null pointer */
1287 suword(vectp, 0);
1288
1289 return (stack_base);
1290}
1291
1292/*
1293 * Check permissions of file to execute.
1294 * Called with imgp->vp locked.
1295 * Return 0 for success or error code on failure.
1296 */
1297int
1298exec_check_permissions(imgp)
1299 struct image_params *imgp;
1300{
1301 struct vnode *vp = imgp->vp;
1302 struct vattr *attr = imgp->attr;
1303 struct thread *td;
1304 int error;
1305
1306 td = curthread;
1307
1308 /* Get file attributes */
1309 error = VOP_GETATTR(vp, attr, td->td_ucred);
1310 if (error)
1311 return (error);
1312
1313#ifdef MAC
1314 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1315 if (error)
1316 return (error);
1317#endif
1318
1319 /*
1320 * 1) Check if file execution is disabled for the filesystem that this
1321 * file resides on.
1322 * 2) Insure that at least one execute bit is on - otherwise root
1323 * will always succeed, and we don't want to happen unless the
1324 * file really is executable.
1325 * 3) Insure that the file is a regular file.
1326 */
1327 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1328 ((attr->va_mode & 0111) == 0) ||
1329 (attr->va_type != VREG))
1330 return (EACCES);
1331
1332 /*
1333 * Zero length files can't be exec'd
1334 */
1335 if (attr->va_size == 0)
1336 return (ENOEXEC);
1337
1338 /*
1339 * Check for execute permission to file based on current credentials.
1340 */
1341 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1342 if (error)
1343 return (error);
1344
1345 /*
1346 * Check number of open-for-writes on the file and deny execution
1347 * if there are any.
1348 */
1349 if (vp->v_writecount)
1350 return (ETXTBSY);
1351
1352 /*
1353 * Call filesystem specific open routine (which does nothing in the
1354 * general case).
1355 */
1356 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1357 if (error == 0)
1358 imgp->opened = 1;
1359 return (error);
1360}
1361
1362/*
1363 * Exec handler registration
1364 */
1365int
1366exec_register(execsw_arg)
1367 const struct execsw *execsw_arg;
1368{
1369 const struct execsw **es, **xs, **newexecsw;
1370 int count = 2; /* New slot and trailing NULL */
1371
1372 if (execsw)
1373 for (es = execsw; *es; es++)
1374 count++;
1375 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1376 if (newexecsw == NULL)
1377 return (ENOMEM);
1378 xs = newexecsw;
1379 if (execsw)
1380 for (es = execsw; *es; es++)
1381 *xs++ = *es;
1382 *xs++ = execsw_arg;
1383 *xs = NULL;
1384 if (execsw)
1385 free(execsw, M_TEMP);
1386 execsw = newexecsw;
1387 return (0);
1388}
1389
1390int
1391exec_unregister(execsw_arg)
1392 const struct execsw *execsw_arg;
1393{
1394 const struct execsw **es, **xs, **newexecsw;
1395 int count = 1;
1396
1397 if (execsw == NULL)
1398 panic("unregister with no handlers left?\n");
1399
1400 for (es = execsw; *es; es++) {
1401 if (*es == execsw_arg)
1402 break;
1403 }
1404 if (*es == NULL)
1405 return (ENOENT);
1406 for (es = execsw; *es; es++)
1407 if (*es != execsw_arg)
1408 count++;
1409 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1410 if (newexecsw == NULL)
1411 return (ENOMEM);
1412 xs = newexecsw;
1413 for (es = execsw; *es; es++)
1414 if (*es != execsw_arg)
1415 *xs++ = *es;
1416 *xs = NULL;
1417 if (execsw)
1418 free(execsw, M_TEMP);
1419 execsw = newexecsw;
1420 return (0);
1421}
|