vserver 2.0 rc7
[linux-2.6.git] / fs / binfmt_aout.c
1 /*
2  *  linux/fs/binfmt_aout.c
3  *
4  *  Copyright (C) 1991, 1992, 1996  Linus Torvalds
5  */
6
7 #include <linux/module.h>
8
9 #include <linux/time.h>
10 #include <linux/kernel.h>
11 #include <linux/mm.h>
12 #include <linux/mman.h>
13 #include <linux/a.out.h>
14 #include <linux/errno.h>
15 #include <linux/signal.h>
16 #include <linux/string.h>
17 #include <linux/fs.h>
18 #include <linux/file.h>
19 #include <linux/stat.h>
20 #include <linux/fcntl.h>
21 #include <linux/ptrace.h>
22 #include <linux/user.h>
23 #include <linux/slab.h>
24 #include <linux/binfmts.h>
25 #include <linux/personality.h>
26 #include <linux/init.h>
27 #include <linux/vs_memory.h>
28
29 #include <asm/system.h>
30 #include <asm/uaccess.h>
31 #include <asm/cacheflush.h>
32
33 static int load_aout_binary(struct linux_binprm *, struct pt_regs * regs);
34 static int load_aout_library(struct file*);
35 static int aout_core_dump(long signr, struct pt_regs * regs, struct file *file);
36
37 extern void dump_thread(struct pt_regs *, struct user *);
38
39 static struct linux_binfmt aout_format = {
40         .module         = THIS_MODULE,
41         .load_binary    = load_aout_binary,
42         .load_shlib     = load_aout_library,
43         .core_dump      = aout_core_dump,
44         .min_coredump   = PAGE_SIZE
45 };
46
47 #define BAD_ADDR(x)     ((unsigned long)(x) >= TASK_SIZE)
48
49 static int set_brk(unsigned long start, unsigned long end)
50 {
51         start = PAGE_ALIGN(start);
52         end = PAGE_ALIGN(end);
53         if (end > start) {
54                 unsigned long addr;
55                 down_write(&current->mm->mmap_sem);
56                 addr = do_brk(start, end - start);
57                 up_write(&current->mm->mmap_sem);
58                 if (BAD_ADDR(addr))
59                         return addr;
60         }
61         return 0;
62 }
63
64 /*
65  * These are the only things you should do on a core-file: use only these
66  * macros to write out all the necessary info.
67  */
68
69 static int dump_write(struct file *file, const void *addr, int nr)
70 {
71         return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
72 }
73
74 #define DUMP_WRITE(addr, nr)    \
75         if (!dump_write(file, (void *)(addr), (nr))) \
76                 goto end_coredump;
77
78 #define DUMP_SEEK(offset) \
79 if (file->f_op->llseek) { \
80         if (file->f_op->llseek(file,(offset),0) != (offset)) \
81                 goto end_coredump; \
82 } else file->f_pos = (offset)
83
84 /*
85  * Routine writes a core dump image in the current directory.
86  * Currently only a stub-function.
87  *
88  * Note that setuid/setgid files won't make a core-dump if the uid/gid
89  * changed due to the set[u|g]id. It's enforced by the "current->mm->dumpable"
90  * field, which also makes sure the core-dumps won't be recursive if the
91  * dumping of the process results in another error..
92  */
93
94 static int aout_core_dump(long signr, struct pt_regs * regs, struct file *file)
95 {
96         mm_segment_t fs;
97         int has_dumped = 0;
98         unsigned long dump_start, dump_size;
99         struct user dump;
100 #if defined(__alpha__)
101 #       define START_DATA(u)    (u.start_data)
102 #elif defined(__arm__)
103 #       define START_DATA(u)    ((u.u_tsize << PAGE_SHIFT) + u.start_code)
104 #elif defined(__sparc__)
105 #       define START_DATA(u)    (u.u_tsize)
106 #elif defined(__i386__) || defined(__mc68000__) || defined(__arch_um__)
107 #       define START_DATA(u)    (u.u_tsize << PAGE_SHIFT)
108 #endif
109 #ifdef __sparc__
110 #       define START_STACK(u)   ((regs->u_regs[UREG_FP]) & ~(PAGE_SIZE - 1))
111 #else
112 #       define START_STACK(u)   (u.start_stack)
113 #endif
114
115         fs = get_fs();
116         set_fs(KERNEL_DS);
117         has_dumped = 1;
118         current->flags |= PF_DUMPCORE;
119         strncpy(dump.u_comm, current->comm, sizeof(dump.u_comm));
120 #ifndef __sparc__
121         dump.u_ar0 = (void *)(((unsigned long)(&dump.regs)) - ((unsigned long)(&dump)));
122 #endif
123         dump.signal = signr;
124         dump_thread(regs, &dump);
125
126 /* If the size of the dump file exceeds the rlimit, then see what would happen
127    if we wrote the stack, but not the data area.  */
128 #ifdef __sparc__
129         if ((dump.u_dsize+dump.u_ssize) >
130             current->signal->rlim[RLIMIT_CORE].rlim_cur)
131                 dump.u_dsize = 0;
132 #else
133         if ((dump.u_dsize+dump.u_ssize+1) * PAGE_SIZE >
134             current->signal->rlim[RLIMIT_CORE].rlim_cur)
135                 dump.u_dsize = 0;
136 #endif
137
138 /* Make sure we have enough room to write the stack and data areas. */
139 #ifdef __sparc__
140         if ((dump.u_ssize) >
141             current->signal->rlim[RLIMIT_CORE].rlim_cur)
142                 dump.u_ssize = 0;
143 #else
144         if ((dump.u_ssize+1) * PAGE_SIZE >
145             current->signal->rlim[RLIMIT_CORE].rlim_cur)
146                 dump.u_ssize = 0;
147 #endif
148
149 /* make sure we actually have a data and stack area to dump */
150         set_fs(USER_DS);
151 #ifdef __sparc__
152         if (!access_ok(VERIFY_READ, (void __user *)START_DATA(dump), dump.u_dsize))
153                 dump.u_dsize = 0;
154         if (!access_ok(VERIFY_READ, (void __user *)START_STACK(dump), dump.u_ssize))
155                 dump.u_ssize = 0;
156 #else
157         if (!access_ok(VERIFY_READ, (void __user *)START_DATA(dump), dump.u_dsize << PAGE_SHIFT))
158                 dump.u_dsize = 0;
159         if (!access_ok(VERIFY_READ, (void __user *)START_STACK(dump), dump.u_ssize << PAGE_SHIFT))
160                 dump.u_ssize = 0;
161 #endif
162
163         set_fs(KERNEL_DS);
164 /* struct user */
165         DUMP_WRITE(&dump,sizeof(dump));
166 /* Now dump all of the user data.  Include malloced stuff as well */
167 #ifndef __sparc__
168         DUMP_SEEK(PAGE_SIZE);
169 #endif
170 /* now we start writing out the user space info */
171         set_fs(USER_DS);
172 /* Dump the data area */
173         if (dump.u_dsize != 0) {
174                 dump_start = START_DATA(dump);
175 #ifdef __sparc__
176                 dump_size = dump.u_dsize;
177 #else
178                 dump_size = dump.u_dsize << PAGE_SHIFT;
179 #endif
180                 DUMP_WRITE(dump_start,dump_size);
181         }
182 /* Now prepare to dump the stack area */
183         if (dump.u_ssize != 0) {
184                 dump_start = START_STACK(dump);
185 #ifdef __sparc__
186                 dump_size = dump.u_ssize;
187 #else
188                 dump_size = dump.u_ssize << PAGE_SHIFT;
189 #endif
190                 DUMP_WRITE(dump_start,dump_size);
191         }
192 /* Finally dump the task struct.  Not be used by gdb, but could be useful */
193         set_fs(KERNEL_DS);
194         DUMP_WRITE(current,sizeof(*current));
195 end_coredump:
196         set_fs(fs);
197         return has_dumped;
198 }
199
200 /*
201  * create_aout_tables() parses the env- and arg-strings in new user
202  * memory and creates the pointer tables from them, and puts their
203  * addresses on the "stack", returning the new stack pointer value.
204  */
205 static unsigned long __user *create_aout_tables(char __user *p, struct linux_binprm * bprm)
206 {
207         char __user * __user *argv;
208         char __user * __user *envp;
209         unsigned long __user *sp;
210         int argc = bprm->argc;
211         int envc = bprm->envc;
212
213         sp = (void __user *)((-(unsigned long)sizeof(char *)) & (unsigned long) p);
214 #ifdef __sparc__
215         /* This imposes the proper stack alignment for a new process. */
216         sp = (void __user *) (((unsigned long) sp) & ~7);
217         if ((envc+argc+3)&1) --sp;
218 #endif
219 #ifdef __alpha__
220 /* whee.. test-programs are so much fun. */
221         put_user(0, --sp);
222         put_user(0, --sp);
223         if (bprm->loader) {
224                 put_user(0, --sp);
225                 put_user(0x3eb, --sp);
226                 put_user(bprm->loader, --sp);
227                 put_user(0x3ea, --sp);
228         }
229         put_user(bprm->exec, --sp);
230         put_user(0x3e9, --sp);
231 #endif
232         sp -= envc+1;
233         envp = (char __user * __user *) sp;
234         sp -= argc+1;
235         argv = (char __user * __user *) sp;
236 #if defined(__i386__) || defined(__mc68000__) || defined(__arm__) || defined(__arch_um__)
237         put_user((unsigned long) envp,--sp);
238         put_user((unsigned long) argv,--sp);
239 #endif
240         put_user(argc,--sp);
241         current->mm->arg_start = (unsigned long) p;
242         while (argc-->0) {
243                 char c;
244                 put_user(p,argv++);
245                 do {
246                         get_user(c,p++);
247                 } while (c);
248         }
249         put_user(NULL,argv);
250         current->mm->arg_end = current->mm->env_start = (unsigned long) p;
251         while (envc-->0) {
252                 char c;
253                 put_user(p,envp++);
254                 do {
255                         get_user(c,p++);
256                 } while (c);
257         }
258         put_user(NULL,envp);
259         current->mm->env_end = (unsigned long) p;
260         return sp;
261 }
262
263 /*
264  * These are the functions used to load a.out style executables and shared
265  * libraries.  There is no binary dependent code anywhere else.
266  */
267
268 static int load_aout_binary(struct linux_binprm * bprm, struct pt_regs * regs)
269 {
270         struct exec ex;
271         unsigned long error;
272         unsigned long fd_offset;
273         unsigned long rlim;
274         int retval;
275
276         ex = *((struct exec *) bprm->buf);              /* exec-header */
277         if ((N_MAGIC(ex) != ZMAGIC && N_MAGIC(ex) != OMAGIC &&
278              N_MAGIC(ex) != QMAGIC && N_MAGIC(ex) != NMAGIC) ||
279             N_TRSIZE(ex) || N_DRSIZE(ex) ||
280             i_size_read(bprm->file->f_dentry->d_inode) < ex.a_text+ex.a_data+N_SYMSIZE(ex)+N_TXTOFF(ex)) {
281                 return -ENOEXEC;
282         }
283
284         fd_offset = N_TXTOFF(ex);
285
286         /* Check initial limits. This avoids letting people circumvent
287          * size limits imposed on them by creating programs with large
288          * arrays in the data or bss.
289          */
290         rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
291         if (rlim >= RLIM_INFINITY)
292                 rlim = ~0;
293         if (ex.a_data + ex.a_bss > rlim)
294                 return -ENOMEM;
295
296         /* Flush all traces of the currently running executable */
297         retval = flush_old_exec(bprm);
298         if (retval)
299                 return retval;
300
301         /* OK, This is the point of no return */
302 #if defined(__alpha__)
303         SET_AOUT_PERSONALITY(bprm, ex);
304 #elif defined(__sparc__)
305         set_personality(PER_SUNOS);
306 #if !defined(__sparc_v9__)
307         memcpy(&current->thread.core_exec, &ex, sizeof(struct exec));
308 #endif
309 #else
310         set_personality(PER_LINUX);
311 #endif
312
313         current->mm->end_code = ex.a_text +
314                 (current->mm->start_code = N_TXTADDR(ex));
315         current->mm->end_data = ex.a_data +
316                 (current->mm->start_data = N_DATADDR(ex));
317         current->mm->brk = ex.a_bss +
318                 (current->mm->start_brk = N_BSSADDR(ex));
319         current->mm->free_area_cache = current->mm->mmap_base;
320
321         set_mm_counter(current->mm, rss, 0);
322         current->mm->mmap = NULL;
323         compute_creds(bprm);
324         current->flags &= ~PF_FORKNOEXEC;
325 #ifdef __sparc__
326         if (N_MAGIC(ex) == NMAGIC) {
327                 loff_t pos = fd_offset;
328                 /* Fuck me plenty... */
329                 /* <AOL></AOL> */
330                 down_write(&current->mm->mmap_sem);     
331                 error = do_brk(N_TXTADDR(ex), ex.a_text);
332                 up_write(&current->mm->mmap_sem);
333                 bprm->file->f_op->read(bprm->file, (char *) N_TXTADDR(ex),
334                           ex.a_text, &pos);
335                 down_write(&current->mm->mmap_sem);
336                 error = do_brk(N_DATADDR(ex), ex.a_data);
337                 up_write(&current->mm->mmap_sem);
338                 bprm->file->f_op->read(bprm->file, (char *) N_DATADDR(ex),
339                           ex.a_data, &pos);
340                 goto beyond_if;
341         }
342 #endif
343
344         if (N_MAGIC(ex) == OMAGIC) {
345                 unsigned long text_addr, map_size;
346                 loff_t pos;
347
348                 text_addr = N_TXTADDR(ex);
349
350 #if defined(__alpha__) || defined(__sparc__)
351                 pos = fd_offset;
352                 map_size = ex.a_text+ex.a_data + PAGE_SIZE - 1;
353 #else
354                 pos = 32;
355                 map_size = ex.a_text+ex.a_data;
356 #endif
357                 down_write(&current->mm->mmap_sem);
358                 error = do_brk(text_addr & PAGE_MASK, map_size);
359                 up_write(&current->mm->mmap_sem);
360                 if (error != (text_addr & PAGE_MASK)) {
361                         send_sig(SIGKILL, current, 0);
362                         return error;
363                 }
364
365                 error = bprm->file->f_op->read(bprm->file,
366                           (char __user *)text_addr,
367                           ex.a_text+ex.a_data, &pos);
368                 if ((signed long)error < 0) {
369                         send_sig(SIGKILL, current, 0);
370                         return error;
371                 }
372                          
373                 flush_icache_range(text_addr, text_addr+ex.a_text+ex.a_data);
374         } else {
375                 static unsigned long error_time, error_time2;
376                 if ((ex.a_text & 0xfff || ex.a_data & 0xfff) &&
377                     (N_MAGIC(ex) != NMAGIC) && (jiffies-error_time2) > 5*HZ)
378                 {
379                         printk(KERN_NOTICE "executable not page aligned\n");
380                         error_time2 = jiffies;
381                 }
382
383                 if ((fd_offset & ~PAGE_MASK) != 0 &&
384                     (jiffies-error_time) > 5*HZ)
385                 {
386                         printk(KERN_WARNING 
387                                "fd_offset is not page aligned. Please convert program: %s\n",
388                                bprm->file->f_dentry->d_name.name);
389                         error_time = jiffies;
390                 }
391
392                 if (!bprm->file->f_op->mmap||((fd_offset & ~PAGE_MASK) != 0)) {
393                         loff_t pos = fd_offset;
394                         down_write(&current->mm->mmap_sem);
395                         do_brk(N_TXTADDR(ex), ex.a_text+ex.a_data);
396                         up_write(&current->mm->mmap_sem);
397                         bprm->file->f_op->read(bprm->file,
398                                         (char __user *)N_TXTADDR(ex),
399                                         ex.a_text+ex.a_data, &pos);
400                         flush_icache_range((unsigned long) N_TXTADDR(ex),
401                                            (unsigned long) N_TXTADDR(ex) +
402                                            ex.a_text+ex.a_data);
403                         goto beyond_if;
404                 }
405
406                 down_write(&current->mm->mmap_sem);
407                 error = do_mmap(bprm->file, N_TXTADDR(ex), ex.a_text,
408                         PROT_READ | PROT_EXEC,
409                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE,
410                         fd_offset);
411                 up_write(&current->mm->mmap_sem);
412
413                 if (error != N_TXTADDR(ex)) {
414                         send_sig(SIGKILL, current, 0);
415                         return error;
416                 }
417
418                 down_write(&current->mm->mmap_sem);
419                 error = do_mmap(bprm->file, N_DATADDR(ex), ex.a_data,
420                                 PROT_READ | PROT_WRITE | PROT_EXEC,
421                                 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE,
422                                 fd_offset + ex.a_text);
423                 up_write(&current->mm->mmap_sem);
424                 if (error != N_DATADDR(ex)) {
425                         send_sig(SIGKILL, current, 0);
426                         return error;
427                 }
428         }
429 beyond_if:
430         set_binfmt(&aout_format);
431
432         retval = set_brk(current->mm->start_brk, current->mm->brk);
433         if (retval < 0) {
434                 send_sig(SIGKILL, current, 0);
435                 return retval;
436         }
437
438         retval = setup_arg_pages(bprm, STACK_TOP, EXSTACK_DEFAULT);
439         if (retval < 0) { 
440                 /* Someone check-me: is this error path enough? */ 
441                 send_sig(SIGKILL, current, 0); 
442                 return retval;
443         }
444
445         current->mm->start_stack =
446                 (unsigned long) create_aout_tables((char __user *) bprm->p, bprm);
447 #ifdef __alpha__
448         regs->gp = ex.a_gpvalue;
449 #endif
450         start_thread(regs, ex.a_entry, current->mm->start_stack);
451         if (unlikely(current->ptrace & PT_PTRACED)) {
452                 if (current->ptrace & PT_TRACE_EXEC)
453                         ptrace_notify ((PTRACE_EVENT_EXEC << 8) | SIGTRAP);
454                 else
455                         send_sig(SIGTRAP, current, 0);
456         }
457         return 0;
458 }
459
460 static int load_aout_library(struct file *file)
461 {
462         struct inode * inode;
463         unsigned long bss, start_addr, len;
464         unsigned long error;
465         int retval;
466         struct exec ex;
467
468         inode = file->f_dentry->d_inode;
469
470         retval = -ENOEXEC;
471         error = kernel_read(file, 0, (char *) &ex, sizeof(ex));
472         if (error != sizeof(ex))
473                 goto out;
474
475         /* We come in here for the regular a.out style of shared libraries */
476         if ((N_MAGIC(ex) != ZMAGIC && N_MAGIC(ex) != QMAGIC) || N_TRSIZE(ex) ||
477             N_DRSIZE(ex) || ((ex.a_entry & 0xfff) && N_MAGIC(ex) == ZMAGIC) ||
478             i_size_read(inode) < ex.a_text+ex.a_data+N_SYMSIZE(ex)+N_TXTOFF(ex)) {
479                 goto out;
480         }
481
482         if (N_FLAGS(ex))
483                 goto out;
484
485         /* For  QMAGIC, the starting address is 0x20 into the page.  We mask
486            this off to get the starting address for the page */
487
488         start_addr =  ex.a_entry & 0xfffff000;
489
490         if ((N_TXTOFF(ex) & ~PAGE_MASK) != 0) {
491                 static unsigned long error_time;
492                 loff_t pos = N_TXTOFF(ex);
493
494                 if ((jiffies-error_time) > 5*HZ)
495                 {
496                         printk(KERN_WARNING 
497                                "N_TXTOFF is not page aligned. Please convert library: %s\n",
498                                file->f_dentry->d_name.name);
499                         error_time = jiffies;
500                 }
501                 down_write(&current->mm->mmap_sem);
502                 do_brk(start_addr, ex.a_text + ex.a_data + ex.a_bss);
503                 up_write(&current->mm->mmap_sem);
504                 
505                 file->f_op->read(file, (char __user *)start_addr,
506                         ex.a_text + ex.a_data, &pos);
507                 flush_icache_range((unsigned long) start_addr,
508                                    (unsigned long) start_addr + ex.a_text + ex.a_data);
509
510                 retval = 0;
511                 goto out;
512         }
513         /* Now use mmap to map the library into memory. */
514         down_write(&current->mm->mmap_sem);
515         error = do_mmap(file, start_addr, ex.a_text + ex.a_data,
516                         PROT_READ | PROT_WRITE | PROT_EXEC,
517                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
518                         N_TXTOFF(ex));
519         up_write(&current->mm->mmap_sem);
520         retval = error;
521         if (error != start_addr)
522                 goto out;
523
524         len = PAGE_ALIGN(ex.a_text + ex.a_data);
525         bss = ex.a_text + ex.a_data + ex.a_bss;
526         if (bss > len) {
527                 down_write(&current->mm->mmap_sem);
528                 error = do_brk(start_addr + len, bss - len);
529                 up_write(&current->mm->mmap_sem);
530                 retval = error;
531                 if (error != start_addr + len)
532                         goto out;
533         }
534         retval = 0;
535 out:
536         return retval;
537 }
538
539 static int __init init_aout_binfmt(void)
540 {
541         return register_binfmt(&aout_format);
542 }
543
544 static void __exit exit_aout_binfmt(void)
545 {
546         unregister_binfmt(&aout_format);
547 }
548
549 core_initcall(init_aout_binfmt);
550 module_exit(exit_aout_binfmt);
551 MODULE_LICENSE("GPL");