xref: /minix3/minix/servers/vfs/exec.c (revision b5e2faaaaf60a8b9a02f8d72f64caa56a87eb312)
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 /*===========================================================================*
86  *				get_read_vp				     *
87  *===========================================================================*/
88 static int get_read_vp(struct vfs_exec_info *execi,
89   char *fullpath, int copyprogname, int sugid, struct lookup *resolve, struct fproc *fp)
90 {
91 /* Make the executable that we want to exec() into the binary pointed
92  * to by 'fullpath.' This function fills in necessary details in the execi
93  * structure, such as opened vnode. It unlocks and releases the vnode if
94  * it was already there. This makes it easy to change the executable
95  * during the exec(), which is often necessary, by calling this function
96  * more than once. This is specifically necessary when we discover the
97  * executable is actually a script or a dynamically linked executable.
98  */
99 	int r;
100 
101 	/* Caller wants to switch vp to the file in 'fullpath.'
102 	 * unlock and put it first if there is any there.
103 	 */
104 	if(execi->vp) {
105 		unlock_vnode(execi->vp);
106 		put_vnode(execi->vp);
107 		execi->vp = NULL;
108 	}
109 
110 	/* Remember/overwrite the executable name if requested. */
111 	if(copyprogname) {
112 		char *cp = strrchr(fullpath, '/');
113 		if(cp) cp++;
114 		else cp = fullpath;
115 		strlcpy(execi->args.progname, cp, sizeof(execi->args.progname));
116 		execi->args.progname[sizeof(execi->args.progname)-1] = '\0';
117 	}
118 
119 	/* Open executable */
120 	if ((execi->vp = eat_path(resolve, fp)) == NULL)
121 		return err_code;
122 
123 	unlock_vmnt(execi->vmp);
124 
125 	if (!S_ISREG(execi->vp->v_mode))
126 		return ENOEXEC;
127 	else if ((r = forbidden(fp, execi->vp, X_BIT)) != OK)
128 		return r;
129 	else
130 		r = req_stat(execi->vp->v_fs_e, execi->vp->v_inode_nr,
131 			VFS_PROC_NR, (vir_bytes) &(execi->sb));
132 
133 	if (r != OK) return r;
134 
135 	/* If caller wants us to, honour suid/guid mode bits. */
136         if (sugid) {
137 		/* Deal with setuid/setgid executables */
138 		if (execi->vp->v_mode & I_SET_UID_BIT) {
139 			execi->args.new_uid = execi->vp->v_uid;
140 			execi->args.allow_setuid = 1;
141 		}
142 		if (execi->vp->v_mode & I_SET_GID_BIT) {
143 			execi->args.new_gid = execi->vp->v_gid;
144 			execi->args.allow_setuid = 1;
145 		}
146         }
147 
148 	/* Read in first chunk of file. */
149 	if((r=map_header(execi)) != OK)
150 		return r;
151 
152 	return OK;
153 }
154 
155 #define FAILCHECK(expr) if((r=(expr)) != OK) { goto pm_execfinal; } while(0)
156 #define Get_read_vp(e,f,p,s,rs,fp) do { \
157 	r=get_read_vp(&e,f,p,s,rs,fp); if(r != OK) { FAILCHECK(r); }	\
158 	} while(0)
159 
160 static int vfs_memmap(struct exec_info *execi,
161         vir_bytes vaddr, vir_bytes len, vir_bytes foffset, u16_t clearend,
162 	int protflags)
163 {
164 	struct vfs_exec_info *vi = (struct vfs_exec_info *) execi->opaque;
165 	struct vnode *vp = ((struct vfs_exec_info *) execi->opaque)->vp;
166 	int r;
167 	u16_t flags = 0;
168 
169 	if(protflags & PROT_WRITE)
170 		flags |= MVM_WRITABLE;
171 
172 	r = minix_vfs_mmap(execi->proc_e, foffset, len,
173 	        vp->v_dev, vp->v_inode_nr, vi->vmfd, vaddr, clearend, flags);
174 	if(r == OK) {
175 		vi->vmfd_used = 1;
176 	}
177 
178 	return r;
179 }
180 
181 /*===========================================================================*
182  *				pm_exec					     *
183  *===========================================================================*/
184 int pm_exec(vir_bytes path, size_t path_len, vir_bytes frame, size_t frame_len,
185 	vir_bytes *pc, vir_bytes *newsp, vir_bytes *UNUSED(ps_str))
186 {
187 /* Perform the execve(name, argv, envp) call.  The user library builds a
188  * complete stack image, including pointers, args, environ, etc.  The stack
189  * is copied to a buffer inside VFS, and then to the new core image.
190  *
191  * ps_str is not currently used, but may be if the ps_strings structure has to
192  * be moved to another location.
193  */
194   int r;
195   vir_bytes vsp;
196   static char mbuf[ARG_MAX];	/* buffer for stack and zeroes */
197   struct vfs_exec_info execi;
198   int i;
199   static char fullpath[PATH_MAX],
200   	elf_interpreter[PATH_MAX],
201 	firstexec[PATH_MAX],
202 	finalexec[PATH_MAX];
203   struct lookup resolve;
204   struct fproc *vmfp = fproc_addr(VM_PROC_NR);
205   stackhook_t makestack = NULL;
206   struct filp *newfilp = NULL;
207 
208   lock_exec();
209 
210   /* unset execi values are 0. */
211   memset(&execi, 0, sizeof(execi));
212   execi.vmfd = -1;
213 
214   /* passed from exec() libc code */
215   execi.userflags = 0;
216   execi.args.stack_high = kinfo.user_sp;
217   execi.args.stack_size = DEFAULT_STACK_LIMIT;
218 
219   fp->text_size = 0;
220   fp->data_size = 0;
221 
222   lookup_init(&resolve, fullpath, PATH_NOFLAGS, &execi.vmp, &execi.vp);
223 
224   resolve.l_vmnt_lock = VMNT_READ;
225   resolve.l_vnode_lock = VNODE_READ;
226 
227   /* Fetch the stack from the user before destroying the old core image. */
228   if (frame_len > ARG_MAX)
229 	FAILCHECK(ENOMEM); /* stack too big */
230 
231   r = sys_datacopy_wrapper(fp->fp_endpoint, (vir_bytes) frame, SELF, (vir_bytes) mbuf,
232 		   (size_t) frame_len);
233   if (r != OK) { /* can't fetch stack (e.g. bad virtual addr) */
234         printf("VFS: pm_exec: sys_datacopy failed\n");
235 	FAILCHECK(r);
236   }
237 
238   /* Compute the current virtual stack pointer, has to be done before calling
239    * patch_stack, which needs it, and will adapt as required. */
240   vsp = execi.args.stack_high - frame_len;
241 
242   /* The default is to keep the original user and group IDs */
243   execi.args.new_uid = fp->fp_effuid;
244   execi.args.new_gid = fp->fp_effgid;
245 
246   /* Get the exec file name. */
247   FAILCHECK(fetch_name(path, path_len, fullpath));
248   strlcpy(finalexec, fullpath, PATH_MAX);
249   strlcpy(firstexec, fullpath, PATH_MAX);
250 
251   /* Get_read_vp will return an opened vn in execi.
252    * if necessary it releases the existing vp so we can
253    * switch after we find out what's inside the file.
254    * It reads the start of the file.
255    */
256   Get_read_vp(execi, fullpath, 1, 1, &resolve, fp);
257 
258   /* If this is a script (i.e. has a #!/interpreter line),
259    * retrieve the name of the interpreter and open that
260    * executable instead.
261    */
262   if(is_script(&execi)) {
263   	/* patch_stack will add interpreter name and
264 	 * args to stack and retrieve the new binary
265 	 * name into fullpath.
266 	 */
267 	FAILCHECK(fetch_name(path, path_len, fullpath));
268 	FAILCHECK(patch_stack(execi.vp, mbuf, &frame_len, fullpath, &vsp));
269 
270 	strlcpy(finalexec, fullpath, PATH_MAX);
271   	strlcpy(firstexec, fullpath, PATH_MAX);
272 	Get_read_vp(execi, fullpath, 1, 0, &resolve, fp);
273   }
274 
275   /* If this is a dynamically linked executable, retrieve
276    * the name of that interpreter in elf_interpreter and open that
277    * executable instead. But open the current executable in an
278    * fd for the current process.
279    */
280   if(elf_has_interpreter(execi.args.hdr, execi.args.hdr_len,
281 	elf_interpreter, sizeof(elf_interpreter))) {
282 	/* Switch the executable vnode to the interpreter */
283 	execi.is_dyn = 1;
284 
285 	/* The interpreter (loader) needs an fd to the main program,
286 	 * which is currently in finalexec
287 	 */
288 	if ((r = execi.elf_main_fd =
289 	    common_open(finalexec, O_RDONLY, 0, TRUE /*for_exec*/)) < 0) {
290 		printf("VFS: exec: dynamic: open main exec failed %s (%d)\n",
291 			fullpath, r);
292 		FAILCHECK(r);
293 	}
294 
295 	/* ld.so is linked at 0, but it can relocate itself; we
296 	 * want it higher to trap NULL pointer dereferences.
297 	 * Let's put it below the stack, and reserve 10MB for ld.so.
298 	 */
299 	execi.args.load_offset =
300 		 execi.args.stack_high - execi.args.stack_size - 0xa00000;
301 
302 	/* Remember it */
303 	strlcpy(execi.execname, finalexec, PATH_MAX);
304 
305 	/* The executable we need to execute first (loader)
306 	 * is in elf_interpreter, and has to be in fullpath to
307 	 * be looked up
308 	 */
309 	strlcpy(fullpath, elf_interpreter, PATH_MAX);
310 	strlcpy(firstexec, elf_interpreter, PATH_MAX);
311 	Get_read_vp(execi, fullpath, 0, 0, &resolve, fp);
312   }
313 
314   /* We also want an FD for VM to mmap() the process in if possible. */
315   {
316 	struct vnode *vp = execi.vp;
317 	assert(vp);
318 	if ((vp->v_vmnt->m_fs_flags & RES_HASPEEK) &&
319 			major(vp->v_dev) != MEMORY_MAJOR) {
320 		int newfd = -1;
321 		if(get_fd(vmfp, 0, R_BIT, &newfd, &newfilp) == OK) {
322 			assert(newfd >= 0 && newfd < OPEN_MAX);
323 			assert(!vmfp->fp_filp[newfd]);
324 			newfilp->filp_count = 1;
325 			newfilp->filp_vno = vp;
326 			newfilp->filp_flags = O_RDONLY;
327 			vmfp->fp_filp[newfd] = newfilp;
328 			/* dup_vnode(vp); */
329 			execi.vmfd = newfd;
330 			execi.args.memmap = vfs_memmap;
331 		}
332 	}
333   }
334 
335   /* callback functions and data */
336   execi.args.copymem = read_seg;
337   execi.args.clearproc = libexec_clearproc_vm_procctl;
338   execi.args.clearmem = libexec_clear_sys_memset;
339   execi.args.allocmem_prealloc_cleared = libexec_alloc_mmap_prealloc_cleared;
340   execi.args.allocmem_prealloc_junk = libexec_alloc_mmap_prealloc_junk;
341   execi.args.allocmem_ondemand = libexec_alloc_mmap_ondemand;
342   execi.args.opaque = &execi;
343 
344   execi.args.proc_e = fp->fp_endpoint;
345   execi.args.frame_len = frame_len;
346   execi.args.filesize = execi.vp->v_size;
347 
348   for (i = 0; exec_loaders[i].load_object != NULL; i++) {
349       r = (*exec_loaders[i].load_object)(&execi.args);
350       /* Loaded successfully, so no need to try other loaders */
351       if (r == OK) { makestack = exec_loaders[i].setup_stack; break; }
352   }
353 
354   FAILCHECK(r);
355 
356   /* Inform PM */
357   FAILCHECK(libexec_pm_newexec(fp->fp_endpoint, &execi.args));
358 
359   /* Save off PC */
360   *pc = execi.args.pc;
361 
362   /* call a stack-setup function if this executable type wants it */
363   if(makestack) FAILCHECK(makestack(&execi, mbuf, &frame_len, &vsp));
364 
365   /* Copy the stack from VFS to new core image. */
366   FAILCHECK(sys_datacopy_wrapper(SELF, (vir_bytes) mbuf, fp->fp_endpoint,
367 	(vir_bytes) vsp, (phys_bytes)frame_len));
368 
369   /* Return new stack pointer to caller */
370   *newsp = vsp;
371 
372   clo_exec(fp);
373 
374   if (execi.args.allow_setuid) {
375 	/* If after loading the image we're still allowed to run with
376 	 * setuid or setgid, change credentials now */
377 	fp->fp_effuid = execi.args.new_uid;
378 	fp->fp_effgid = execi.args.new_gid;
379   }
380 
381   /* Remember the new name of the process */
382   strlcpy(fp->fp_name, execi.args.progname, PROC_NAME_LEN);
383   fp->text_size = execi.args.text_size;
384   fp->data_size = execi.args.data_size;
385 
386 pm_execfinal:
387   if(newfilp) unlock_filp(newfilp);
388   else if (execi.vp != NULL) {
389 	unlock_vnode(execi.vp);
390 	put_vnode(execi.vp);
391   }
392 
393   if(execi.vmfd >= 0 && !execi.vmfd_used) {
394   	if(OK != close_fd(vmfp, execi.vmfd)) {
395 		printf("VFS: unexpected close fail of vm fd\n");
396 	}
397   }
398 
399   unlock_exec();
400 
401   return(r);
402 }
403 
404 /* This is a copy-paste of the same macro in minix/lib/libc/sys/stack_utils.c.
405  * Keep it synchronized. */
406 #define STACK_MIN_SZ \
407 ( \
408        sizeof(int) + sizeof(void *) * 2 + \
409        sizeof(AuxInfo) * PMEF_AUXVECTORS + PMEF_EXECNAMELEN1 + \
410        sizeof(struct ps_strings) \
411 )
412 
413 static int stack_prepare_elf(struct vfs_exec_info *execi, char *frame, size_t *frame_size,
414 	vir_bytes *vsp)
415 {
416 	AuxInfo *aux_vec, *aux_vec_end;
417 	vir_bytes vap; /* Address in proc space of the first AuxVec. */
418 	Elf_Ehdr const * const elf_header = (Elf_Ehdr *) execi->args.hdr;
419 	struct ps_strings const * const psp = (struct ps_strings *)
420 		(frame + (*frame_size - sizeof(struct ps_strings)));
421 
422 	size_t const execname_len = strlen(execi->execname);
423 
424 	if (!execi->is_dyn)
425 		return OK;
426 
427 	if (execi->args.hdr_len < sizeof(*elf_header)) {
428 		printf("VFS: malformed ELF headers for exec\n");
429 		return ENOEXEC;
430 	}
431 
432 	if (*frame_size < STACK_MIN_SZ) {
433 		printf("VFS: malformed stack for exec(), smaller than minimum"
434 			" possible size.\n");
435 		return ENOEXEC;
436 	}
437 
438 	/* Find first Aux vector in the stack frame. */
439 	vap = (vir_bytes)(psp->ps_envstr + (psp->ps_nenvstr + 1));
440 	aux_vec = (AuxInfo *) (frame + (vap - *vsp));
441 	aux_vec_end = aux_vec + PMEF_AUXVECTORS;
442 
443 	if (((char *)aux_vec < frame) ||
444 		((char *)aux_vec > (frame + *frame_size))) {
445 		printf("VFS: malformed stack for exec(), first AuxVector is"
446 		       " not on the stack.\n");
447 		return ENOEXEC;
448 	}
449 
450 	if (((char *)aux_vec_end < frame) ||
451 		((char *)aux_vec_end > (frame + *frame_size))) {
452 		printf("VFS: malformed stack for exec(), last AuxVector is"
453 		       " not on the stack.\n");
454 		return ENOEXEC;
455 	}
456 
457 	/* Userland provides a fully filled stack frame, with argc, argv, envp
458 	 * and then all the argv and envp strings; consistent with ELF ABI,
459 	 * except for a list of Aux vectors that should be between envp points
460 	 * and the start of the strings.
461 	 *
462 	 * It would take some very unpleasant hackery to insert the aux vectors
463 	 * before the strings, and correct all the pointers, so the exec code
464 	 * in libc makes space for us.
465 	 */
466 
467 #define AUXINFO(a, type, value) \
468 	do { \
469 		if (a < aux_vec_end) { \
470 			a->a_type = type; \
471 			a->a_v = value; \
472 			a++; \
473 		} else { \
474 			printf("VFS: No more room for ELF AuxVec type %d, skipping it for %s\n", type, execi->execname); \
475 			(aux_vec_end - 1)->a_type = AT_NULL; \
476 			(aux_vec_end - 1)->a_v = 0; \
477 		} \
478 	} while(0)
479 
480 	AUXINFO(aux_vec, AT_BASE, execi->args.load_base);
481 	AUXINFO(aux_vec, AT_ENTRY, execi->args.pc);
482 	AUXINFO(aux_vec, AT_EXECFD, execi->elf_main_fd);
483 #if 0
484 	AUXINFO(aux_vec, AT_PHDR, XXX ); /* should be &phdr[0] */
485 	AUXINFO(aux_vec, AT_PHENT, elf_header->e_phentsize);
486 	AUXINFO(aux_vec, AT_PHNUM, elf_header->e_phnum);
487 
488 	AUXINFO(aux_vec, AT_RUID, XXX);
489 	AUXINFO(aux_vec, AT_RGID, XXX);
490 #endif
491 	AUXINFO(aux_vec, AT_EUID, execi->args.new_uid);
492 	AUXINFO(aux_vec, AT_EGID, execi->args.new_gid);
493 	AUXINFO(aux_vec, AT_PAGESZ, PAGE_SIZE);
494 
495 	if(execname_len < PMEF_EXECNAMELEN1) {
496 		char *spacestart;
497 		vir_bytes userp;
498 
499 		/* Empty space starts after aux_vec table; we can put the name
500 		 * here. */
501 		spacestart = (char *) aux_vec + 2 * sizeof(AuxInfo);
502 		strlcpy(spacestart, execi->execname, PMEF_EXECNAMELEN1);
503 		memset(spacestart + execname_len, '\0',
504 			PMEF_EXECNAMELEN1 - execname_len);
505 
506 		/* What will the address of the string for the user be */
507 		userp = *vsp + (spacestart - frame);
508 
509 		/* Move back to where the AT_NULL is */
510 		AUXINFO(aux_vec, AT_SUN_EXECNAME, userp);
511 	}
512 
513 	/* Always terminate with AT_NULL */
514 	AUXINFO(aux_vec, AT_NULL, 0);
515 
516 	return OK;
517 }
518 
519 /*===========================================================================*
520  *				is_script				     *
521  *===========================================================================*/
522 static int is_script(struct vfs_exec_info *execi)
523 {
524 /* Is Interpreted script? */
525   assert(execi->args.hdr != NULL);
526 
527   return(execi->args.hdr[0] == '#' && execi->args.hdr[1] == '!'
528   	&& execi->args.hdr_len >= 2);
529 }
530 
531 /*===========================================================================*
532  *				patch_stack				     *
533  *===========================================================================*/
534 static int patch_stack(vp, stack, stk_bytes, path, vsp)
535 struct vnode *vp;		/* pointer for open script file */
536 char stack[ARG_MAX];		/* pointer to stack image within VFS */
537 size_t *stk_bytes;		/* size of initial stack */
538 char path[PATH_MAX];		/* path to script file */
539 vir_bytes *vsp;
540 {
541 /* Patch the argument vector to include the path name of the script to be
542  * interpreted, and all strings on the #! line.  Returns the path name of
543  * the interpreter.
544  */
545   enum { INSERT=FALSE, REPLACE=TRUE };
546   int n, r;
547   off_t pos, new_pos;
548   char *sp, *interp = NULL;
549   size_t cum_io;
550   char buf[PAGE_SIZE];
551 
552   /* Make 'path' the new argv[0]. */
553   if (!insert_arg(stack, stk_bytes, path, vsp, REPLACE)) return(ENOMEM);
554 
555   pos = 0;	/* Read from the start of the file */
556 
557   /* Issue request */
558   r = req_readwrite(vp->v_fs_e, vp->v_inode_nr, pos, READING, VFS_PROC_NR,
559 			(vir_bytes) buf, sizeof(buf), &new_pos, &cum_io);
560 
561   if (r != OK) return(r);
562 
563   n = vp->v_size;
564   if (n > sizeof(buf))
565 	n = sizeof(buf);
566   if (n < 2) return ENOEXEC;
567 
568   sp = &(buf[2]);				/* just behind the #! */
569   n -= 2;
570   if (n > PATH_MAX) n = PATH_MAX;
571 
572   /* Use the 'path' variable for temporary storage */
573   memcpy(path, sp, n);
574 
575   if ((sp = memchr(path, '\n', n)) == NULL) /* must be a proper line */
576 	return(ENOEXEC);
577 
578   /* Move sp backwards through script[], prepending each string to stack. */
579   for (;;) {
580 	/* skip spaces behind argument. */
581 	while (sp > path && (*--sp == ' ' || *sp == '\t')) {}
582 	if (sp == path) break;
583 
584 	sp[1] = 0;
585 	/* Move to the start of the argument. */
586 	while (sp > path && sp[-1] != ' ' && sp[-1] != '\t') --sp;
587 
588 	interp = sp;
589 	if (!insert_arg(stack, stk_bytes, sp, vsp, INSERT)) {
590 		printf("VFS: patch_stack: insert_arg failed\n");
591 		return(ENOMEM);
592 	}
593   }
594 
595   if(!interp)
596   	return ENOEXEC;
597 
598   if (interp != path)
599 	memmove(path, interp, strlen(interp)+1);
600 
601   return(OK);
602 }
603 
604 /*===========================================================================*
605  *				insert_arg				     *
606  *===========================================================================*/
607 static int insert_arg(char stack[ARG_MAX], size_t *stk_bytes, char *arg,
608 	vir_bytes *vsp, char replace)
609 {
610 	/* Patch the stack so that arg will become argv[0]. Be careful, the
611 	 * stack may be filled with garbage, although it normally looks like
612 	 * this:
613 	 *	nargs argv[0] ... argv[nargs-1] NULL envp[0] ... NULL
614 	 * followed by the strings "pointed" to by the argv[i] and the envp[i].
615 	 * The * pointers are in the new process address space.
616 	 *
617 	 * Return true iff the operation succeeded.
618 	 */
619 	struct ps_strings *psp;
620 	int offset;
621 	size_t old_bytes = *stk_bytes;
622 
623 	int const arg_len = strlen(arg) + 1;
624 
625 	/* Offset to argv[0][0] in the stack frame. */
626 	int const a0 = (int)(((char **)stack)[1] - *vsp);
627 
628 	/* Check that argv[0] points within the stack frame. */
629 	if ((a0 < 0) || (a0 >= old_bytes)) {
630 		printf("vfs:: argv[0][] not within stack range!! %i\n", a0);
631 		return FALSE;
632 	}
633 
634 	if (!replace) {
635 		/* Prepending arg adds one pointer, one string and a zero byte. */
636 		offset = arg_len + PTRSIZE;
637 	} else {
638 		/* replacing argv[0] with arg adds the difference in length of
639 		 * the two strings. Make sure we don't go beyond the stack size
640 		 * when computing the length of the current argv[0]. */
641 		offset = arg_len - strnlen(stack + a0, ARG_MAX - a0 - 1);
642 	}
643 
644 	/* As ps_strings follows the strings, ensure the offset is word aligned. */
645 	offset = offset + (PTRSIZE - ((PTRSIZE + offset) % PTRSIZE));
646 
647 	/* The stack will grow (or shrink) by offset bytes. */
648 	if ((*stk_bytes += offset) > ARG_MAX) {
649 		printf("vfs:: offset too big!! %zu (max %d)\n", *stk_bytes,
650 			ARG_MAX);
651 		return FALSE;
652 	}
653 
654 	/* Reposition the strings by offset bytes */
655 	memmove(stack + a0 + offset, stack + a0, old_bytes - a0);
656 
657 	/* Put arg in the new space, leaving padding in front of it. */
658 	strlcpy(stack + a0 + offset - arg_len, arg, arg_len);
659 
660 	if (!replace) {
661 		/* Make space for a new argv[0]. */
662 		memmove(stack + 2 * PTRSIZE,
663 			stack + 1 * PTRSIZE, a0 - 2 * PTRSIZE);
664 
665 		((char **) stack)[0]++;	/* nargs++; */
666 	}
667 
668 	/* set argv[0] correctly */
669 	((char **) stack)[1] = (char *) a0 - arg_len + *vsp;
670 
671 	/* Update stack pointer in the process address space. */
672 	*vsp -= offset;
673 
674 	/* Update argv and envp in ps_strings */
675 	psp = (struct ps_strings *) (stack + *stk_bytes - sizeof(struct ps_strings));
676 	psp->ps_argvstr -= (offset / PTRSIZE);
677 	if (!replace) {
678 		psp->ps_nargvstr++;
679 	}
680 	psp->ps_envstr = psp->ps_argvstr + psp->ps_nargvstr + 1;
681 
682 	return TRUE;
683 }
684 
685 /*===========================================================================*
686  *				read_seg				     *
687  *===========================================================================*/
688 static int read_seg(struct exec_info *execi, off_t off, vir_bytes seg_addr, size_t seg_bytes)
689 {
690 /*
691  * The byte count on read is usually smaller than the segment count, because
692  * a segment is padded out to a click multiple, and the data segment is only
693  * partially initialized.
694  */
695   int r;
696   off_t new_pos;
697   size_t cum_io;
698   struct vnode *vp = ((struct vfs_exec_info *) execi->opaque)->vp;
699 
700   /* Make sure that the file is big enough */
701   if (off + seg_bytes > LONG_MAX) return(EIO);
702   if ((unsigned long) vp->v_size < off+seg_bytes) return(EIO);
703 
704   if ((r = req_readwrite(vp->v_fs_e, vp->v_inode_nr, off, READING,
705 		 execi->proc_e, (vir_bytes) seg_addr, seg_bytes,
706 		 &new_pos, &cum_io)) != OK) {
707     printf("VFS: read_seg: req_readwrite failed (data)\n");
708     return(r);
709   }
710 
711   if (r == OK && cum_io != seg_bytes)
712 	printf("VFS: read_seg segment has not been read properly\n");
713 
714 	return(r);
715 }
716 
717 
718 /*===========================================================================*
719  *				clo_exec				     *
720  *===========================================================================*/
721 static void clo_exec(struct fproc *rfp)
722 {
723 /* Files can be marked with the FD_CLOEXEC bit (in fp->fp_cloexec).
724  */
725   int i;
726 
727   /* Check the file desriptors one by one for presence of FD_CLOEXEC. */
728   for (i = 0; i < OPEN_MAX; i++)
729 	if ( FD_ISSET(i, &rfp->fp_cloexec_set))
730 		(void) close_fd(rfp, i);
731 }
732 
733 /*===========================================================================*
734  *				map_header				     *
735  *===========================================================================*/
736 static int map_header(struct vfs_exec_info *execi)
737 {
738   int r;
739   size_t cum_io;
740   off_t pos, new_pos;
741   static char hdr[PAGE_SIZE]; /* Assume that header is not larger than a page */
742 
743   pos = 0;	/* Read from the start of the file */
744 
745   /* How much is sensible to read */
746   execi->args.hdr_len = MIN(execi->vp->v_size, sizeof(hdr));
747   execi->args.hdr = hdr;
748 
749   r = req_readwrite(execi->vp->v_fs_e, execi->vp->v_inode_nr,
750   	pos, READING, VFS_PROC_NR, (vir_bytes) hdr,
751 	execi->args.hdr_len, &new_pos, &cum_io);
752   if (r != OK) {
753 	printf("VFS: exec: map_header: req_readwrite failed\n");
754 	return(r);
755   }
756 
757   return(OK);
758 }
759