Merge git://git.kernel.org/pub/scm/linux/kernel/git/steve/gfs2-2.6-nmw
[pandora-kernel.git] / arch / x86 / ia32 / ia32_aout.c
1 /*
2  *  a.out loader for x86-64
3  *
4  *  Copyright (C) 1991, 1992, 1996  Linus Torvalds
5  *  Hacked together by Andi Kleen
6  */
7
8 #include <linux/module.h>
9
10 #include <linux/time.h>
11 #include <linux/kernel.h>
12 #include <linux/mm.h>
13 #include <linux/mman.h>
14 #include <linux/a.out.h>
15 #include <linux/errno.h>
16 #include <linux/signal.h>
17 #include <linux/string.h>
18 #include <linux/fs.h>
19 #include <linux/file.h>
20 #include <linux/stat.h>
21 #include <linux/fcntl.h>
22 #include <linux/ptrace.h>
23 #include <linux/user.h>
24 #include <linux/binfmts.h>
25 #include <linux/personality.h>
26 #include <linux/init.h>
27 #include <linux/jiffies.h>
28
29 #include <asm/system.h>
30 #include <asm/uaccess.h>
31 #include <asm/pgalloc.h>
32 #include <asm/cacheflush.h>
33 #include <asm/user32.h>
34 #include <asm/ia32.h>
35
36 #undef WARN_OLD
37 #undef CORE_DUMP /* definitely broken */
38
39 static int load_aout_binary(struct linux_binprm *, struct pt_regs *regs);
40 static int load_aout_library(struct file *);
41
42 #ifdef CORE_DUMP
43 static int aout_core_dump(long signr, struct pt_regs *regs, struct file *file,
44                           unsigned long limit);
45
46 /*
47  * fill in the user structure for a core dump..
48  */
49 static void dump_thread32(struct pt_regs *regs, struct user32 *dump)
50 {
51         u32 fs, gs;
52
53 /* changed the size calculations - should hopefully work better. lbt */
54         dump->magic = CMAGIC;
55         dump->start_code = 0;
56         dump->start_stack = regs->sp & ~(PAGE_SIZE - 1);
57         dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
58         dump->u_dsize = ((unsigned long)
59                          (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
60         dump->u_dsize -= dump->u_tsize;
61         dump->u_ssize = 0;
62         dump->u_debugreg[0] = current->thread.debugreg0;
63         dump->u_debugreg[1] = current->thread.debugreg1;
64         dump->u_debugreg[2] = current->thread.debugreg2;
65         dump->u_debugreg[3] = current->thread.debugreg3;
66         dump->u_debugreg[4] = 0;
67         dump->u_debugreg[5] = 0;
68         dump->u_debugreg[6] = current->thread.debugreg6;
69         dump->u_debugreg[7] = current->thread.debugreg7;
70
71         if (dump->start_stack < 0xc0000000) {
72                 unsigned long tmp;
73
74                 tmp = (unsigned long) (0xc0000000 - dump->start_stack);
75                 dump->u_ssize = tmp >> PAGE_SHIFT;
76         }
77
78         dump->regs.bx = regs->bx;
79         dump->regs.cx = regs->cx;
80         dump->regs.dx = regs->dx;
81         dump->regs.si = regs->si;
82         dump->regs.di = regs->di;
83         dump->regs.bp = regs->bp;
84         dump->regs.ax = regs->ax;
85         dump->regs.ds = current->thread.ds;
86         dump->regs.es = current->thread.es;
87         savesegment(fs, fs);
88         dump->regs.fs = fs;
89         savesegment(gs, gs);
90         dump->regs.gs = gs;
91         dump->regs.orig_ax = regs->orig_ax;
92         dump->regs.ip = regs->ip;
93         dump->regs.cs = regs->cs;
94         dump->regs.flags = regs->flags;
95         dump->regs.sp = regs->sp;
96         dump->regs.ss = regs->ss;
97
98 #if 1 /* FIXME */
99         dump->u_fpvalid = 0;
100 #else
101         dump->u_fpvalid = dump_fpu(regs, &dump->i387);
102 #endif
103 }
104
105 #endif
106
107 static struct linux_binfmt aout_format = {
108         .module         = THIS_MODULE,
109         .load_binary    = load_aout_binary,
110         .load_shlib     = load_aout_library,
111 #ifdef CORE_DUMP
112         .core_dump      = aout_core_dump,
113 #endif
114         .min_coredump   = PAGE_SIZE
115 };
116
117 static void set_brk(unsigned long start, unsigned long end)
118 {
119         start = PAGE_ALIGN(start);
120         end = PAGE_ALIGN(end);
121         if (end <= start)
122                 return;
123         down_write(&current->mm->mmap_sem);
124         do_brk(start, end - start);
125         up_write(&current->mm->mmap_sem);
126 }
127
128 #ifdef CORE_DUMP
129 /*
130  * These are the only things you should do on a core-file: use only these
131  * macros to write out all the necessary info.
132  */
133
134 #include <linux/coredump.h>
135
136 #define DUMP_WRITE(addr, nr)                         \
137         if (!dump_write(file, (void *)(addr), (nr))) \
138                 goto end_coredump;
139
140 #define DUMP_SEEK(offset)               \
141         if (!dump_seek(file, offset))   \
142                 goto end_coredump;
143
144 #define START_DATA()    (u.u_tsize << PAGE_SHIFT)
145 #define START_STACK(u)  (u.start_stack)
146
147 /*
148  * Routine writes a core dump image in the current directory.
149  * Currently only a stub-function.
150  *
151  * Note that setuid/setgid files won't make a core-dump if the uid/gid
152  * changed due to the set[u|g]id. It's enforced by the "current->mm->dumpable"
153  * field, which also makes sure the core-dumps won't be recursive if the
154  * dumping of the process results in another error..
155  */
156
157 static int aout_core_dump(long signr, struct pt_regs *regs, struct file *file,
158                           unsigned long limit)
159 {
160         mm_segment_t fs;
161         int has_dumped = 0;
162         unsigned long dump_start, dump_size;
163         struct user32 dump;
164
165         fs = get_fs();
166         set_fs(KERNEL_DS);
167         has_dumped = 1;
168         current->flags |= PF_DUMPCORE;
169         strncpy(dump.u_comm, current->comm, sizeof(current->comm));
170         dump.u_ar0 = offsetof(struct user32, regs);
171         dump.signal = signr;
172         dump_thread32(regs, &dump);
173
174         /*
175          * If the size of the dump file exceeds the rlimit, then see
176          * what would happen if we wrote the stack, but not the data
177          * area.
178          */
179         if ((dump.u_dsize + dump.u_ssize + 1) * PAGE_SIZE > limit)
180                 dump.u_dsize = 0;
181
182         /* Make sure we have enough room to write the stack and data areas. */
183         if ((dump.u_ssize + 1) * PAGE_SIZE > limit)
184                 dump.u_ssize = 0;
185
186         /* make sure we actually have a data and stack area to dump */
187         set_fs(USER_DS);
188         if (!access_ok(VERIFY_READ, (void *) (unsigned long)START_DATA(dump),
189                        dump.u_dsize << PAGE_SHIFT))
190                 dump.u_dsize = 0;
191         if (!access_ok(VERIFY_READ, (void *) (unsigned long)START_STACK(dump),
192                        dump.u_ssize << PAGE_SHIFT))
193                 dump.u_ssize = 0;
194
195         set_fs(KERNEL_DS);
196         /* struct user */
197         DUMP_WRITE(&dump, sizeof(dump));
198         /* Now dump all of the user data.  Include malloced stuff as well */
199         DUMP_SEEK(PAGE_SIZE);
200         /* now we start writing out the user space info */
201         set_fs(USER_DS);
202         /* Dump the data area */
203         if (dump.u_dsize != 0) {
204                 dump_start = START_DATA(dump);
205                 dump_size = dump.u_dsize << PAGE_SHIFT;
206                 DUMP_WRITE(dump_start, dump_size);
207         }
208         /* Now prepare to dump the stack area */
209         if (dump.u_ssize != 0) {
210                 dump_start = START_STACK(dump);
211                 dump_size = dump.u_ssize << PAGE_SHIFT;
212                 DUMP_WRITE(dump_start, dump_size);
213         }
214 end_coredump:
215         set_fs(fs);
216         return has_dumped;
217 }
218 #endif
219
220 /*
221  * create_aout_tables() parses the env- and arg-strings in new user
222  * memory and creates the pointer tables from them, and puts their
223  * addresses on the "stack", returning the new stack pointer value.
224  */
225 static u32 __user *create_aout_tables(char __user *p, struct linux_binprm *bprm)
226 {
227         u32 __user *argv, *envp, *sp;
228         int argc = bprm->argc, envc = bprm->envc;
229
230         sp = (u32 __user *) ((-(unsigned long)sizeof(u32)) & (unsigned long) p);
231         sp -= envc+1;
232         envp = sp;
233         sp -= argc+1;
234         argv = sp;
235         put_user((unsigned long) envp, --sp);
236         put_user((unsigned long) argv, --sp);
237         put_user(argc, --sp);
238         current->mm->arg_start = (unsigned long) p;
239         while (argc-- > 0) {
240                 char c;
241
242                 put_user((u32)(unsigned long)p, argv++);
243                 do {
244                         get_user(c, p++);
245                 } while (c);
246         }
247         put_user(0, argv);
248         current->mm->arg_end = current->mm->env_start = (unsigned long) p;
249         while (envc-- > 0) {
250                 char c;
251
252                 put_user((u32)(unsigned long)p, envp++);
253                 do {
254                         get_user(c, p++);
255                 } while (c);
256         }
257         put_user(0, envp);
258         current->mm->env_end = (unsigned long) p;
259         return sp;
260 }
261
262 /*
263  * These are the functions used to load a.out style executables and shared
264  * libraries.  There is no binary dependent code anywhere else.
265  */
266 static int load_aout_binary(struct linux_binprm *bprm, struct pt_regs *regs)
267 {
268         unsigned long error, fd_offset, rlim;
269         struct exec ex;
270         int retval;
271
272         ex = *((struct exec *) bprm->buf);              /* exec-header */
273         if ((N_MAGIC(ex) != ZMAGIC && N_MAGIC(ex) != OMAGIC &&
274              N_MAGIC(ex) != QMAGIC && N_MAGIC(ex) != NMAGIC) ||
275             N_TRSIZE(ex) || N_DRSIZE(ex) ||
276             i_size_read(bprm->file->f_path.dentry->d_inode) <
277             ex.a_text+ex.a_data+N_SYMSIZE(ex)+N_TXTOFF(ex)) {
278                 return -ENOEXEC;
279         }
280
281         fd_offset = N_TXTOFF(ex);
282
283         /* Check initial limits. This avoids letting people circumvent
284          * size limits imposed on them by creating programs with large
285          * arrays in the data or bss.
286          */
287         rlim = rlimit(RLIMIT_DATA);
288         if (rlim >= RLIM_INFINITY)
289                 rlim = ~0;
290         if (ex.a_data + ex.a_bss > rlim)
291                 return -ENOMEM;
292
293         /* Flush all traces of the currently running executable */
294         retval = flush_old_exec(bprm);
295         if (retval)
296                 return retval;
297
298         /* OK, This is the point of no return */
299         set_personality(PER_LINUX);
300         set_thread_flag(TIF_IA32);
301
302         setup_new_exec(bprm);
303
304         regs->cs = __USER32_CS;
305         regs->r8 = regs->r9 = regs->r10 = regs->r11 = regs->r12 =
306                 regs->r13 = regs->r14 = regs->r15 = 0;
307
308         current->mm->end_code = ex.a_text +
309                 (current->mm->start_code = N_TXTADDR(ex));
310         current->mm->end_data = ex.a_data +
311                 (current->mm->start_data = N_DATADDR(ex));
312         current->mm->brk = ex.a_bss +
313                 (current->mm->start_brk = N_BSSADDR(ex));
314         current->mm->free_area_cache = TASK_UNMAPPED_BASE;
315         current->mm->cached_hole_size = 0;
316
317         install_exec_creds(bprm);
318         current->flags &= ~PF_FORKNOEXEC;
319
320         if (N_MAGIC(ex) == OMAGIC) {
321                 unsigned long text_addr, map_size;
322                 loff_t pos;
323
324                 text_addr = N_TXTADDR(ex);
325
326                 pos = 32;
327                 map_size = ex.a_text+ex.a_data;
328
329                 down_write(&current->mm->mmap_sem);
330                 error = do_brk(text_addr & PAGE_MASK, map_size);
331                 up_write(&current->mm->mmap_sem);
332
333                 if (error != (text_addr & PAGE_MASK)) {
334                         send_sig(SIGKILL, current, 0);
335                         return error;
336                 }
337
338                 error = bprm->file->f_op->read(bprm->file,
339                          (char __user *)text_addr,
340                           ex.a_text+ex.a_data, &pos);
341                 if ((signed long)error < 0) {
342                         send_sig(SIGKILL, current, 0);
343                         return error;
344                 }
345
346                 flush_icache_range(text_addr, text_addr+ex.a_text+ex.a_data);
347         } else {
348 #ifdef WARN_OLD
349                 static unsigned long error_time, error_time2;
350                 if ((ex.a_text & 0xfff || ex.a_data & 0xfff) &&
351                     (N_MAGIC(ex) != NMAGIC) &&
352                                 time_after(jiffies, error_time2 + 5*HZ)) {
353                         printk(KERN_NOTICE "executable not page aligned\n");
354                         error_time2 = jiffies;
355                 }
356
357                 if ((fd_offset & ~PAGE_MASK) != 0 &&
358                             time_after(jiffies, error_time + 5*HZ)) {
359                         printk(KERN_WARNING
360                                "fd_offset is not page aligned. Please convert "
361                                "program: %s\n",
362                                bprm->file->f_path.dentry->d_name.name);
363                         error_time = jiffies;
364                 }
365 #endif
366
367                 if (!bprm->file->f_op->mmap || (fd_offset & ~PAGE_MASK) != 0) {
368                         loff_t pos = fd_offset;
369
370                         down_write(&current->mm->mmap_sem);
371                         do_brk(N_TXTADDR(ex), ex.a_text+ex.a_data);
372                         up_write(&current->mm->mmap_sem);
373                         bprm->file->f_op->read(bprm->file,
374                                         (char __user *)N_TXTADDR(ex),
375                                         ex.a_text+ex.a_data, &pos);
376                         flush_icache_range((unsigned long) N_TXTADDR(ex),
377                                            (unsigned long) N_TXTADDR(ex) +
378                                            ex.a_text+ex.a_data);
379                         goto beyond_if;
380                 }
381
382                 down_write(&current->mm->mmap_sem);
383                 error = do_mmap(bprm->file, N_TXTADDR(ex), ex.a_text,
384                                 PROT_READ | PROT_EXEC,
385                                 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE |
386                                 MAP_EXECUTABLE | MAP_32BIT,
387                                 fd_offset);
388                 up_write(&current->mm->mmap_sem);
389
390                 if (error != N_TXTADDR(ex)) {
391                         send_sig(SIGKILL, current, 0);
392                         return error;
393                 }
394
395                 down_write(&current->mm->mmap_sem);
396                 error = do_mmap(bprm->file, N_DATADDR(ex), ex.a_data,
397                                 PROT_READ | PROT_WRITE | PROT_EXEC,
398                                 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE |
399                                 MAP_EXECUTABLE | MAP_32BIT,
400                                 fd_offset + ex.a_text);
401                 up_write(&current->mm->mmap_sem);
402                 if (error != N_DATADDR(ex)) {
403                         send_sig(SIGKILL, current, 0);
404                         return error;
405                 }
406         }
407 beyond_if:
408         set_binfmt(&aout_format);
409
410         set_brk(current->mm->start_brk, current->mm->brk);
411
412         retval = setup_arg_pages(bprm, IA32_STACK_TOP, EXSTACK_DEFAULT);
413         if (retval < 0) {
414                 /* Someone check-me: is this error path enough? */
415                 send_sig(SIGKILL, current, 0);
416                 return retval;
417         }
418
419         current->mm->start_stack =
420                 (unsigned long)create_aout_tables((char __user *)bprm->p, bprm);
421         /* start thread */
422         loadsegment(fs, 0);
423         loadsegment(ds, __USER32_DS);
424         loadsegment(es, __USER32_DS);
425         load_gs_index(0);
426         (regs)->ip = ex.a_entry;
427         (regs)->sp = current->mm->start_stack;
428         (regs)->flags = 0x200;
429         (regs)->cs = __USER32_CS;
430         (regs)->ss = __USER32_DS;
431         regs->r8 = regs->r9 = regs->r10 = regs->r11 =
432         regs->r12 = regs->r13 = regs->r14 = regs->r15 = 0;
433         set_fs(USER_DS);
434         return 0;
435 }
436
437 static int load_aout_library(struct file *file)
438 {
439         struct inode *inode;
440         unsigned long bss, start_addr, len, error;
441         int retval;
442         struct exec ex;
443
444         inode = file->f_path.dentry->d_inode;
445
446         retval = -ENOEXEC;
447         error = kernel_read(file, 0, (char *) &ex, sizeof(ex));
448         if (error != sizeof(ex))
449                 goto out;
450
451         /* We come in here for the regular a.out style of shared libraries */
452         if ((N_MAGIC(ex) != ZMAGIC && N_MAGIC(ex) != QMAGIC) || N_TRSIZE(ex) ||
453             N_DRSIZE(ex) || ((ex.a_entry & 0xfff) && N_MAGIC(ex) == ZMAGIC) ||
454             i_size_read(inode) <
455             ex.a_text+ex.a_data+N_SYMSIZE(ex)+N_TXTOFF(ex)) {
456                 goto out;
457         }
458
459         if (N_FLAGS(ex))
460                 goto out;
461
462         /* For  QMAGIC, the starting address is 0x20 into the page.  We mask
463            this off to get the starting address for the page */
464
465         start_addr =  ex.a_entry & 0xfffff000;
466
467         if ((N_TXTOFF(ex) & ~PAGE_MASK) != 0) {
468                 loff_t pos = N_TXTOFF(ex);
469
470 #ifdef WARN_OLD
471                 static unsigned long error_time;
472                 if (time_after(jiffies, error_time + 5*HZ)) {
473                         printk(KERN_WARNING
474                                "N_TXTOFF is not page aligned. Please convert "
475                                "library: %s\n",
476                                file->f_path.dentry->d_name.name);
477                         error_time = jiffies;
478                 }
479 #endif
480                 down_write(&current->mm->mmap_sem);
481                 do_brk(start_addr, ex.a_text + ex.a_data + ex.a_bss);
482                 up_write(&current->mm->mmap_sem);
483
484                 file->f_op->read(file, (char __user *)start_addr,
485                         ex.a_text + ex.a_data, &pos);
486                 flush_icache_range((unsigned long) start_addr,
487                                    (unsigned long) start_addr + ex.a_text +
488                                    ex.a_data);
489
490                 retval = 0;
491                 goto out;
492         }
493         /* Now use mmap to map the library into memory. */
494         down_write(&current->mm->mmap_sem);
495         error = do_mmap(file, start_addr, ex.a_text + ex.a_data,
496                         PROT_READ | PROT_WRITE | PROT_EXEC,
497                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE | MAP_32BIT,
498                         N_TXTOFF(ex));
499         up_write(&current->mm->mmap_sem);
500         retval = error;
501         if (error != start_addr)
502                 goto out;
503
504         len = PAGE_ALIGN(ex.a_text + ex.a_data);
505         bss = ex.a_text + ex.a_data + ex.a_bss;
506         if (bss > len) {
507                 down_write(&current->mm->mmap_sem);
508                 error = do_brk(start_addr + len, bss - len);
509                 up_write(&current->mm->mmap_sem);
510                 retval = error;
511                 if (error != start_addr + len)
512                         goto out;
513         }
514         retval = 0;
515 out:
516         return retval;
517 }
518
519 static int __init init_aout_binfmt(void)
520 {
521         return register_binfmt(&aout_format);
522 }
523
524 static void __exit exit_aout_binfmt(void)
525 {
526         unregister_binfmt(&aout_format);
527 }
528
529 module_init(init_aout_binfmt);
530 module_exit(exit_aout_binfmt);
531 MODULE_LICENSE("GPL");