xref: /minix3/minix/servers/vfs/exec.c (revision 0b98e8aad89f2bd4ba80b523d73cf29e9dd82ce1)
1 /* This file handles the EXEC system call.  It performs the work as follows:
2  *    - see if the permissions allow the file to be executed
3  *    - read the header and extract the sizes
4  *    - fetch the initial args and environment from the user space
5  *    - allocate the memory for the new process
6  *    - copy the initial stack from PM to the process
7  *    - read in the text and data segments and copy to the process
8  *    - take care of setuid and setgid bits
9  *    - fix up 'mproc' table
10  *    - tell kernel about EXEC
11  *    - save offset to initial argc (for ps)
12  *
13  * The entry points into this file are:
14  *   pm_exec:	 perform the EXEC system call
15  */
16 
17 #include "fs.h"
18 #include <sys/stat.h>
19 #include <sys/mman.h>
20 #include <minix/callnr.h>
21 #include <minix/endpoint.h>
22 #include <minix/com.h>
23 #include <minix/u64.h>
24 #include <signal.h>
25 #include <stdlib.h>
26 #include <string.h>
27 #include <sys/dirent.h>
28 #include <sys/exec.h>
29 #include <sys/param.h>
30 #include "path.h"
31 #include "vnode.h"
32 #include "file.h"
33 #include <minix/vfsif.h>
34 #include <machine/vmparam.h>
35 #include <assert.h>
36 #include <fcntl.h>
37 
38 #define _KERNEL	/* for ELF_AUX_ENTRIES */
39 #include <libexec.h>
40 
41 /* fields only used by elf and in VFS */
42 struct vfs_exec_info {
43     struct exec_info args;		/* libexec exec args */
44     struct vnode *vp;			/* Exec file's vnode */
45     struct vmnt *vmp;			/* Exec file's vmnt */
46     struct stat sb;			/* Exec file's stat structure */
47     int userflags;			/* exec() flags from userland */
48     int is_dyn;				/* Dynamically linked executable */
49     int elf_main_fd;			/* Dyn: FD of main program execuatble */
50     char execname[PATH_MAX];		/* Full executable invocation */
51     int vmfd;
52     int vmfd_used;
53 };
54 
55 static int patch_stack(struct vnode *vp, char stack[ARG_MAX],
56 	size_t *stk_bytes, char path[PATH_MAX], vir_bytes *vsp);
57 static int is_script(struct vfs_exec_info *execi);
58 static int insert_arg(char stack[ARG_MAX], size_t *stk_bytes, char *arg,
59 	vir_bytes *vsp, char replace);
60 static void clo_exec(struct fproc *rfp);
61 static int stack_prepare_elf(struct vfs_exec_info *execi,
62 	char *curstack, size_t *frame_len, vir_bytes *vsp);
63 static int map_header(struct vfs_exec_info *execi);
64 static int read_seg(struct exec_info *execi, off_t off, vir_bytes seg_addr, size_t seg_bytes);
65 
66 #define PTRSIZE	sizeof(char *) /* Size of pointers in argv[] and envp[]. */
67 
68 /* Array of loaders for different object file formats */
69 typedef int (*exechook_t)(struct vfs_exec_info *execpackage);
70 typedef int (*stackhook_t)(struct vfs_exec_info *execi, char *curstack,
71 	size_t *frame_len, vir_bytes *vsp);
72 struct exec_loaders {
73 	libexec_exec_loadfunc_t load_object;	 /* load executable into memory */
74 	stackhook_t setup_stack; /* prepare stack before argc and argv push */
75 };
76 
77 static const struct exec_loaders exec_loaders[] = {
78 	{ libexec_load_elf,  stack_prepare_elf },
79 	{ NULL, NULL }
80 };
81 
82 #define lock_exec() lock_proc(fproc_addr(VM_PROC_NR))
83 #define unlock_exec() unlock_proc(fproc_addr(VM_PROC_NR))
84 
85 extern struct minix_kerninfo *_minix_kerninfo;
86 
87 /*===========================================================================*
88  *				get_read_vp				     *
89  *===========================================================================*/
90 static int get_read_vp(struct vfs_exec_info *execi,
91   char *fullpath, int copyprogname, int sugid, struct lookup *resolve, struct fproc *fp)
92 {
93 /* Make the executable that we want to exec() into the binary pointed
94  * to by 'fullpath.' This function fills in necessary details in the execi
95  * structure, such as opened vnode. It unlocks and releases the vnode if
96  * it was already there. This makes it easy to change the executable
97  * during the exec(), which is often necessary, by calling this function
98  * more than once. This is specifically necessary when we discover the
99  * executable is actually a script or a dynamically linked executable.
100  */
101 	int r;
102 
103 	/* Caller wants to switch vp to the file in 'fullpath.'
104 	 * unlock and put it first if there is any there.
105 	 */
106 	if(execi->vp) {
107 		unlock_vnode(execi->vp);
108 		put_vnode(execi->vp);
109 		execi->vp = NULL;
110 	}
111 
112 	/* Remember/overwrite the executable name if requested. */
113 	if(copyprogname) {
114 		char *cp = strrchr(fullpath, '/');
115 		if(cp) cp++;
116 		else cp = fullpath;
117 		strlcpy(execi->args.progname, cp, sizeof(execi->args.progname));
118 		execi->args.progname[sizeof(execi->args.progname)-1] = '\0';
119 	}
120 
121 	/* Open executable */
122 	if ((execi->vp = eat_path(resolve, fp)) == NULL)
123 		return err_code;
124 
125 	unlock_vmnt(execi->vmp);
126 
127 	if (!S_ISREG(execi->vp->v_mode))
128 		return ENOEXEC;
129 	else if ((r = forbidden(fp, execi->vp, X_BIT)) != OK)
130 		return r;
131 	else
132 		r = req_stat(execi->vp->v_fs_e, execi->vp->v_inode_nr,
133 			VFS_PROC_NR, (vir_bytes) &(execi->sb));
134 
135 	if (r != OK) return r;
136 
137 	/* If caller wants us to, honour suid/guid mode bits. */
138         if (sugid) {
139 		/* Deal with setuid/setgid executables */
140 		if (execi->vp->v_mode & I_SET_UID_BIT) {
141 			execi->args.new_uid = execi->vp->v_uid;
142 			execi->args.allow_setuid = 1;
143 		}
144 		if (execi->vp->v_mode & I_SET_GID_BIT) {
145 			execi->args.new_gid = execi->vp->v_gid;
146 			execi->args.allow_setuid = 1;
147 		}
148         }
149 
150 	/* Read in first chunk of file. */
151 	if((r=map_header(execi)) != OK)
152 		return r;
153 
154 	return OK;
155 }
156 
157 #define FAILCHECK(expr) if((r=(expr)) != OK) { goto pm_execfinal; } while(0)
158 #define Get_read_vp(e,f,p,s,rs,fp) do { \
159 	r=get_read_vp(&e,f,p,s,rs,fp); if(r != OK) { FAILCHECK(r); }	\
160 	} while(0)
161 
162 static int vfs_memmap(struct exec_info *execi,
163         vir_bytes vaddr, vir_bytes len, vir_bytes foffset, u16_t clearend,
164 	int protflags)
165 {
166 	struct vfs_exec_info *vi = (struct vfs_exec_info *) execi->opaque;
167 	struct vnode *vp = ((struct vfs_exec_info *) execi->opaque)->vp;
168 	int r;
169 	u16_t flags = 0;
170 
171 	if(protflags & PROT_WRITE)
172 		flags |= MVM_WRITABLE;
173 
174 	r = minix_vfs_mmap(execi->proc_e, foffset, len,
175 	        vp->v_dev, vp->v_inode_nr, vi->vmfd, vaddr, clearend, flags);
176 	if(r == OK) {
177 		vi->vmfd_used = 1;
178 	}
179 
180 	return r;
181 }
182 
183 /*===========================================================================*
184  *				pm_exec					     *
185  *===========================================================================*/
186 int pm_exec(vir_bytes path, size_t path_len, vir_bytes frame, size_t frame_len,
187 	vir_bytes *pc, vir_bytes *newsp, vir_bytes *UNUSED(ps_str))
188 {
189 /* Perform the execve(name, argv, envp) call.  The user library builds a
190  * complete stack image, including pointers, args, environ, etc.  The stack
191  * is copied to a buffer inside VFS, and then to the new core image.
192  *
193  * ps_str is not currently used, but may be if the ps_strings structure has to
194  * be moved to another location.
195  */
196   int r;
197   vir_bytes vsp;
198   static char mbuf[ARG_MAX];	/* buffer for stack and zeroes */
199   struct vfs_exec_info execi;
200   int i;
201   static char fullpath[PATH_MAX],
202   	elf_interpreter[PATH_MAX],
203 	firstexec[PATH_MAX],
204 	finalexec[PATH_MAX];
205   struct lookup resolve;
206   struct fproc *vmfp = fproc_addr(VM_PROC_NR);
207   stackhook_t makestack = NULL;
208   struct filp *newfilp = NULL;
209 
210   lock_exec();
211 
212   /* unset execi values are 0. */
213   memset(&execi, 0, sizeof(execi));
214   execi.vmfd = -1;
215 
216   /* passed from exec() libc code */
217   execi.userflags = 0;
218   execi.args.stack_high = _minix_kerninfo->kinfo->user_sp;
219   execi.args.stack_size = DEFAULT_STACK_LIMIT;
220 
221   fp->text_size = 0;
222   fp->data_size = 0;
223 
224   lookup_init(&resolve, fullpath, PATH_NOFLAGS, &execi.vmp, &execi.vp);
225 
226   resolve.l_vmnt_lock = VMNT_READ;
227   resolve.l_vnode_lock = VNODE_READ;
228 
229   /* Fetch the stack from the user before destroying the old core image. */
230   if (frame_len > ARG_MAX)
231 	FAILCHECK(ENOMEM); /* stack too big */
232 
233   r = sys_datacopy_wrapper(fp->fp_endpoint, (vir_bytes) frame, SELF, (vir_bytes) mbuf,
234 		   (size_t) frame_len);
235   if (r != OK) { /* can't fetch stack (e.g. bad virtual addr) */
236         printf("VFS: pm_exec: sys_datacopy failed\n");
237 	FAILCHECK(r);
238   }
239 
240   /* Compute the current virtual stack pointer, has to be done before calling
241    * patch_stack, which needs it, and will adapt as required. */
242   vsp = execi.args.stack_high - frame_len;
243 
244   /* The default is to keep the original user and group IDs */
245   execi.args.new_uid = fp->fp_effuid;
246   execi.args.new_gid = fp->fp_effgid;
247 
248   /* Get the exec file name. */
249   FAILCHECK(fetch_name(path, path_len, fullpath));
250   strlcpy(finalexec, fullpath, PATH_MAX);
251   strlcpy(firstexec, fullpath, PATH_MAX);
252 
253   /* Get_read_vp will return an opened vn in execi.
254    * if necessary it releases the existing vp so we can
255    * switch after we find out what's inside the file.
256    * It reads the start of the file.
257    */
258   Get_read_vp(execi, fullpath, 1, 1, &resolve, fp);
259 
260   /* If this is a script (i.e. has a #!/interpreter line),
261    * retrieve the name of the interpreter and open that
262    * executable instead.
263    */
264   if(is_script(&execi)) {
265   	/* patch_stack will add interpreter name and
266 	 * args to stack and retrieve the new binary
267 	 * name into fullpath.
268 	 */
269 	FAILCHECK(fetch_name(path, path_len, fullpath));
270 	FAILCHECK(patch_stack(execi.vp, mbuf, &frame_len, fullpath, &vsp));
271 
272 	strlcpy(finalexec, fullpath, PATH_MAX);
273   	strlcpy(firstexec, fullpath, PATH_MAX);
274 	Get_read_vp(execi, fullpath, 1, 0, &resolve, fp);
275   }
276 
277   /* If this is a dynamically linked executable, retrieve
278    * the name of that interpreter in elf_interpreter and open that
279    * executable instead. But open the current executable in an
280    * fd for the current process.
281    */
282   if(elf_has_interpreter(execi.args.hdr, execi.args.hdr_len,
283 	elf_interpreter, sizeof(elf_interpreter))) {
284 	/* Switch the executable vnode to the interpreter */
285 	execi.is_dyn = 1;
286 
287 	/* The interpreter (loader) needs an fd to the main program,
288 	 * which is currently in finalexec
289 	 */
290 	if ((r = execi.elf_main_fd =
291 	    common_open(finalexec, O_RDONLY, 0, TRUE /*for_exec*/)) < 0) {
292 		printf("VFS: exec: dynamic: open main exec failed %s (%d)\n",
293 			fullpath, r);
294 		FAILCHECK(r);
295 	}
296 
297 	/* ld.so is linked at 0, but it can relocate itself; we
298 	 * want it higher to trap NULL pointer dereferences.
299 	 * Let's put it below the stack, and reserve 10MB for ld.so.
300 	 */
301 	execi.args.load_offset =
302 		 execi.args.stack_high - execi.args.stack_size - 0xa00000;
303 
304 	/* Remember it */
305 	strlcpy(execi.execname, finalexec, PATH_MAX);
306 
307 	/* The executable we need to execute first (loader)
308 	 * is in elf_interpreter, and has to be in fullpath to
309 	 * be looked up
310 	 */
311 	strlcpy(fullpath, elf_interpreter, PATH_MAX);
312 	strlcpy(firstexec, elf_interpreter, PATH_MAX);
313 	Get_read_vp(execi, fullpath, 0, 0, &resolve, fp);
314   }
315 
316   /* We also want an FD for VM to mmap() the process in if possible. */
317   {
318 	struct vnode *vp = execi.vp;
319 	assert(vp);
320 	if ((vp->v_vmnt->m_fs_flags & RES_HASPEEK) &&
321 			major(vp->v_dev) != MEMORY_MAJOR) {
322 		int newfd = -1;
323 		if(get_fd(vmfp, 0, R_BIT, &newfd, &newfilp) == OK) {
324 			assert(newfd >= 0 && newfd < OPEN_MAX);
325 			assert(!vmfp->fp_filp[newfd]);
326 			newfilp->filp_count = 1;
327 			newfilp->filp_vno = vp;
328 			newfilp->filp_flags = O_RDONLY;
329 			vmfp->fp_filp[newfd] = newfilp;
330 			/* dup_vnode(vp); */
331 			execi.vmfd = newfd;
332 			execi.args.memmap = vfs_memmap;
333 		}
334 	}
335   }
336 
337   /* callback functions and data */
338   execi.args.copymem = read_seg;
339   execi.args.clearproc = libexec_clearproc_vm_procctl;
340   execi.args.clearmem = libexec_clear_sys_memset;
341   execi.args.allocmem_prealloc_cleared = libexec_alloc_mmap_prealloc_cleared;
342   execi.args.allocmem_prealloc_junk = libexec_alloc_mmap_prealloc_junk;
343   execi.args.allocmem_ondemand = libexec_alloc_mmap_ondemand;
344   execi.args.opaque = &execi;
345 
346   execi.args.proc_e = fp->fp_endpoint;
347   execi.args.frame_len = frame_len;
348   execi.args.filesize = execi.vp->v_size;
349 
350   for (i = 0; exec_loaders[i].load_object != NULL; i++) {
351       r = (*exec_loaders[i].load_object)(&execi.args);
352       /* Loaded successfully, so no need to try other loaders */
353       if (r == OK) { makestack = exec_loaders[i].setup_stack; break; }
354   }
355 
356   FAILCHECK(r);
357 
358   /* Inform PM */
359   FAILCHECK(libexec_pm_newexec(fp->fp_endpoint, &execi.args));
360 
361   /* Save off PC */
362   *pc = execi.args.pc;
363 
364   /* call a stack-setup function if this executable type wants it */
365   if(makestack) FAILCHECK(makestack(&execi, mbuf, &frame_len, &vsp));
366 
367   /* Copy the stack from VFS to new core image. */
368   FAILCHECK(sys_datacopy_wrapper(SELF, (vir_bytes) mbuf, fp->fp_endpoint,
369 	(vir_bytes) vsp, (phys_bytes)frame_len));
370 
371   /* Return new stack pointer to caller */
372   *newsp = vsp;
373 
374   clo_exec(fp);
375 
376   if (execi.args.allow_setuid) {
377 	/* If after loading the image we're still allowed to run with
378 	 * setuid or setgid, change credentials now */
379 	fp->fp_effuid = execi.args.new_uid;
380 	fp->fp_effgid = execi.args.new_gid;
381   }
382 
383   /* Remember the new name of the process */
384   strlcpy(fp->fp_name, execi.args.progname, PROC_NAME_LEN);
385   fp->text_size = execi.args.text_size;
386   fp->data_size = execi.args.data_size;
387 
388 pm_execfinal:
389   if(newfilp) unlock_filp(newfilp);
390   else if (execi.vp != NULL) {
391 	unlock_vnode(execi.vp);
392 	put_vnode(execi.vp);
393   }
394 
395   if(execi.vmfd >= 0 && !execi.vmfd_used) {
396   	if(OK != close_fd(vmfp, execi.vmfd)) {
397 		printf("VFS: unexpected close fail of vm fd\n");
398 	}
399   }
400 
401   unlock_exec();
402 
403   return(r);
404 }
405 
406 /* This is a copy-paste of the same macro in minix/lib/libc/sys/stack_utils.c.
407  * Keep it synchronized. */
408 #define STACK_MIN_SZ \
409 ( \
410        sizeof(int) + sizeof(void *) * 2 + \
411        sizeof(AuxInfo) * PMEF_AUXVECTORS + PMEF_EXECNAMELEN1 + \
412        sizeof(struct ps_strings) \
413 )
414 
415 static int stack_prepare_elf(struct vfs_exec_info *execi, char *frame, size_t *frame_size,
416 	vir_bytes *vsp)
417 {
418 	AuxInfo *aux_vec, *aux_vec_end;
419 	vir_bytes vap; /* Address in proc space of the first AuxVec. */
420 	Elf_Ehdr const * const elf_header = (Elf_Ehdr *) execi->args.hdr;
421 	struct ps_strings const * const psp = (struct ps_strings *)
422 		(frame + (*frame_size - sizeof(struct ps_strings)));
423 
424 	size_t const execname_len = strlen(execi->execname);
425 
426 	if (!execi->is_dyn)
427 		return OK;
428 
429 	if (execi->args.hdr_len < sizeof(*elf_header)) {
430 		printf("VFS: malformed ELF headers for exec\n");
431 		return ENOEXEC;
432 	}
433 
434 	if (*frame_size < STACK_MIN_SZ) {
435 		printf("VFS: malformed stack for exec(), smaller than minimum"
436 			" possible size.\n");
437 		return ENOEXEC;
438 	}
439 
440 	/* Find first Aux vector in the stack frame. */
441 	vap = (vir_bytes)(psp->ps_envstr + (psp->ps_nenvstr + 1));
442 	aux_vec = (AuxInfo *) (frame + (vap - *vsp));
443 	aux_vec_end = aux_vec + PMEF_AUXVECTORS;
444 
445 	if (((char *)aux_vec < frame) ||
446 		((char *)aux_vec > (frame + *frame_size))) {
447 		printf("VFS: malformed stack for exec(), first AuxVector is"
448 		       " not on the stack.\n");
449 		return ENOEXEC;
450 	}
451 
452 	if (((char *)aux_vec_end < frame) ||
453 		((char *)aux_vec_end > (frame + *frame_size))) {
454 		printf("VFS: malformed stack for exec(), last AuxVector is"
455 		       " not on the stack.\n");
456 		return ENOEXEC;
457 	}
458 
459 	/* Userland provides a fully filled stack frame, with argc, argv, envp
460 	 * and then all the argv and envp strings; consistent with ELF ABI,
461 	 * except for a list of Aux vectors that should be between envp points
462 	 * and the start of the strings.
463 	 *
464 	 * It would take some very unpleasant hackery to insert the aux vectors
465 	 * before the strings, and correct all the pointers, so the exec code
466 	 * in libc makes space for us.
467 	 */
468 
469 #define AUXINFO(a, type, value) \
470 	do { \
471 		if (a < aux_vec_end) { \
472 			a->a_type = type; \
473 			a->a_v = value; \
474 			a++; \
475 		} else { \
476 			printf("VFS: No more room for ELF AuxVec type %d, skipping it for %s\n", type, execi->execname); \
477 			(aux_vec_end - 1)->a_type = AT_NULL; \
478 			(aux_vec_end - 1)->a_v = 0; \
479 		} \
480 	} while(0)
481 
482 	AUXINFO(aux_vec, AT_BASE, execi->args.load_base);
483 	AUXINFO(aux_vec, AT_ENTRY, execi->args.pc);
484 	AUXINFO(aux_vec, AT_EXECFD, execi->elf_main_fd);
485 #if 0
486 	AUXINFO(aux_vec, AT_PHDR, XXX ); /* should be &phdr[0] */
487 	AUXINFO(aux_vec, AT_PHENT, elf_header->e_phentsize);
488 	AUXINFO(aux_vec, AT_PHNUM, elf_header->e_phnum);
489 
490 	AUXINFO(aux_vec, AT_RUID, XXX);
491 	AUXINFO(aux_vec, AT_RGID, XXX);
492 #endif
493 	AUXINFO(aux_vec, AT_EUID, execi->args.new_uid);
494 	AUXINFO(aux_vec, AT_EGID, execi->args.new_gid);
495 	AUXINFO(aux_vec, AT_PAGESZ, PAGE_SIZE);
496 
497 	if(execname_len < PMEF_EXECNAMELEN1) {
498 		char *spacestart;
499 		vir_bytes userp;
500 
501 		/* Empty space starts after aux_vec table; we can put the name
502 		 * here. */
503 		spacestart = (char *) aux_vec + 2 * sizeof(AuxInfo);
504 		strlcpy(spacestart, execi->execname, PMEF_EXECNAMELEN1);
505 		memset(spacestart + execname_len, '\0',
506 			PMEF_EXECNAMELEN1 - execname_len);
507 
508 		/* What will the address of the string for the user be */
509 		userp = *vsp + (spacestart - frame);
510 
511 		/* Move back to where the AT_NULL is */
512 		AUXINFO(aux_vec, AT_SUN_EXECNAME, userp);
513 	}
514 
515 	/* Always terminate with AT_NULL */
516 	AUXINFO(aux_vec, AT_NULL, 0);
517 
518 	return OK;
519 }
520 
521 /*===========================================================================*
522  *				is_script				     *
523  *===========================================================================*/
524 static int is_script(struct vfs_exec_info *execi)
525 {
526 /* Is Interpreted script? */
527   assert(execi->args.hdr != NULL);
528 
529   return(execi->args.hdr[0] == '#' && execi->args.hdr[1] == '!'
530   	&& execi->args.hdr_len >= 2);
531 }
532 
533 /*===========================================================================*
534  *				patch_stack				     *
535  *===========================================================================*/
536 static int patch_stack(vp, stack, stk_bytes, path, vsp)
537 struct vnode *vp;		/* pointer for open script file */
538 char stack[ARG_MAX];		/* pointer to stack image within VFS */
539 size_t *stk_bytes;		/* size of initial stack */
540 char path[PATH_MAX];		/* path to script file */
541 vir_bytes *vsp;
542 {
543 /* Patch the argument vector to include the path name of the script to be
544  * interpreted, and all strings on the #! line.  Returns the path name of
545  * the interpreter.
546  */
547   enum { INSERT=FALSE, REPLACE=TRUE };
548   int n, r;
549   off_t pos, new_pos;
550   char *sp, *interp = NULL;
551   size_t cum_io;
552   char buf[PAGE_SIZE];
553 
554   /* Make 'path' the new argv[0]. */
555   if (!insert_arg(stack, stk_bytes, path, vsp, REPLACE)) return(ENOMEM);
556 
557   pos = 0;	/* Read from the start of the file */
558 
559   /* Issue request */
560   r = req_readwrite(vp->v_fs_e, vp->v_inode_nr, pos, READING, VFS_PROC_NR,
561 			(vir_bytes) buf, sizeof(buf), &new_pos, &cum_io);
562 
563   if (r != OK) return(r);
564 
565   n = vp->v_size;
566   if (n > sizeof(buf))
567 	n = sizeof(buf);
568   if (n < 2) return ENOEXEC;
569 
570   sp = &(buf[2]);				/* just behind the #! */
571   n -= 2;
572   if (n > PATH_MAX) n = PATH_MAX;
573 
574   /* Use the 'path' variable for temporary storage */
575   memcpy(path, sp, n);
576 
577   if ((sp = memchr(path, '\n', n)) == NULL) /* must be a proper line */
578 	return(ENOEXEC);
579 
580   /* Move sp backwards through script[], prepending each string to stack. */
581   for (;;) {
582 	/* skip spaces behind argument. */
583 	while (sp > path && (*--sp == ' ' || *sp == '\t')) {}
584 	if (sp == path) break;
585 
586 	sp[1] = 0;
587 	/* Move to the start of the argument. */
588 	while (sp > path && sp[-1] != ' ' && sp[-1] != '\t') --sp;
589 
590 	interp = sp;
591 	if (!insert_arg(stack, stk_bytes, sp, vsp, INSERT)) {
592 		printf("VFS: patch_stack: insert_arg failed\n");
593 		return(ENOMEM);
594 	}
595   }
596 
597   if(!interp)
598   	return ENOEXEC;
599 
600   if (interp != path)
601 	memmove(path, interp, strlen(interp)+1);
602 
603   return(OK);
604 }
605 
606 /*===========================================================================*
607  *				insert_arg				     *
608  *===========================================================================*/
609 static int insert_arg(char stack[ARG_MAX], size_t *stk_bytes, char *arg,
610 	vir_bytes *vsp, char replace)
611 {
612 	/* Patch the stack so that arg will become argv[0]. Be careful, the
613 	 * stack may be filled with garbage, although it normally looks like
614 	 * this:
615 	 *	nargs argv[0] ... argv[nargs-1] NULL envp[0] ... NULL
616 	 * followed by the strings "pointed" to by the argv[i] and the envp[i].
617 	 * The * pointers are in the new process address space.
618 	 *
619 	 * Return true iff the operation succeeded.
620 	 */
621 	struct ps_strings *psp;
622 	int offset;
623 	size_t old_bytes = *stk_bytes;
624 
625 	int const arg_len = strlen(arg) + 1;
626 
627 	/* Offset to argv[0][0] in the stack frame. */
628 	int const a0 = (int)(((char **)stack)[1] - *vsp);
629 
630 	/* Check that argv[0] points within the stack frame. */
631 	if ((a0 < 0) || (a0 >= old_bytes)) {
632 		printf("vfs:: argv[0][] not within stack range!! %i\n", a0);
633 		return FALSE;
634 	}
635 
636 	if (!replace) {
637 		/* Prepending arg adds one pointer, one string and a zero byte. */
638 		offset = arg_len + PTRSIZE;
639 	} else {
640 		/* replacing argv[0] with arg adds the difference in length of
641 		 * the two strings. Make sure we don't go beyond the stack size
642 		 * when computing the length of the current argv[0]. */
643 		offset = arg_len - strnlen(stack + a0, ARG_MAX - a0 - 1);
644 	}
645 
646 	/* As ps_strings follows the strings, ensure the offset is word aligned. */
647 	offset = offset + (PTRSIZE - ((PTRSIZE + offset) % PTRSIZE));
648 
649 	/* The stack will grow (or shrink) by offset bytes. */
650 	if ((*stk_bytes += offset) > ARG_MAX) {
651 		printf("vfs:: offset too big!! %zu (max %d)\n", *stk_bytes,
652 			ARG_MAX);
653 		return FALSE;
654 	}
655 
656 	/* Reposition the strings by offset bytes */
657 	memmove(stack + a0 + offset, stack + a0, old_bytes - a0);
658 
659 	/* Put arg in the new space, leaving padding in front of it. */
660 	strlcpy(stack + a0 + offset - arg_len, arg, arg_len);
661 
662 	if (!replace) {
663 		/* Make space for a new argv[0]. */
664 		memmove(stack + 2 * PTRSIZE,
665 			stack + 1 * PTRSIZE, a0 - 2 * PTRSIZE);
666 
667 		((char **) stack)[0]++;	/* nargs++; */
668 	}
669 
670 	/* set argv[0] correctly */
671 	((char **) stack)[1] = (char *) a0 - arg_len + *vsp;
672 
673 	/* Update stack pointer in the process address space. */
674 	*vsp -= offset;
675 
676 	/* Update argv and envp in ps_strings */
677 	psp = (struct ps_strings *) (stack + *stk_bytes - sizeof(struct ps_strings));
678 	psp->ps_argvstr -= (offset / PTRSIZE);
679 	if (!replace) {
680 		psp->ps_nargvstr++;
681 	}
682 	psp->ps_envstr = psp->ps_argvstr + psp->ps_nargvstr + 1;
683 
684 	return TRUE;
685 }
686 
687 /*===========================================================================*
688  *				read_seg				     *
689  *===========================================================================*/
690 static int read_seg(struct exec_info *execi, off_t off, vir_bytes seg_addr, size_t seg_bytes)
691 {
692 /*
693  * The byte count on read is usually smaller than the segment count, because
694  * a segment is padded out to a click multiple, and the data segment is only
695  * partially initialized.
696  */
697   int r;
698   off_t new_pos;
699   size_t cum_io;
700   struct vnode *vp = ((struct vfs_exec_info *) execi->opaque)->vp;
701 
702   /* Make sure that the file is big enough */
703   if (off + seg_bytes > LONG_MAX) return(EIO);
704   if ((unsigned long) vp->v_size < off+seg_bytes) return(EIO);
705 
706   if ((r = req_readwrite(vp->v_fs_e, vp->v_inode_nr, off, READING,
707 		 execi->proc_e, (vir_bytes) seg_addr, seg_bytes,
708 		 &new_pos, &cum_io)) != OK) {
709     printf("VFS: read_seg: req_readwrite failed (data)\n");
710     return(r);
711   }
712 
713   if (r == OK && cum_io != seg_bytes)
714 	printf("VFS: read_seg segment has not been read properly\n");
715 
716 	return(r);
717 }
718 
719 
720 /*===========================================================================*
721  *				clo_exec				     *
722  *===========================================================================*/
723 static void clo_exec(struct fproc *rfp)
724 {
725 /* Files can be marked with the FD_CLOEXEC bit (in fp->fp_cloexec).
726  */
727   int i;
728 
729   /* Check the file desriptors one by one for presence of FD_CLOEXEC. */
730   for (i = 0; i < OPEN_MAX; i++)
731 	if ( FD_ISSET(i, &rfp->fp_cloexec_set))
732 		(void) close_fd(rfp, i);
733 }
734 
735 /*===========================================================================*
736  *				map_header				     *
737  *===========================================================================*/
738 static int map_header(struct vfs_exec_info *execi)
739 {
740   int r;
741   size_t cum_io;
742   off_t pos, new_pos;
743   static char hdr[PAGE_SIZE]; /* Assume that header is not larger than a page */
744 
745   pos = 0;	/* Read from the start of the file */
746 
747   /* How much is sensible to read */
748   execi->args.hdr_len = MIN(execi->vp->v_size, sizeof(hdr));
749   execi->args.hdr = hdr;
750 
751   r = req_readwrite(execi->vp->v_fs_e, execi->vp->v_inode_nr,
752   	pos, READING, VFS_PROC_NR, (vir_bytes) hdr,
753 	execi->args.hdr_len, &new_pos, &cum_io);
754   if (r != OK) {
755 	printf("VFS: exec: map_header: req_readwrite failed\n");
756 	return(r);
757   }
758 
759   return(OK);
760 }
761