Merge branch 'core/rcu' into core/rcu-for-linus
[linux-2.6] / 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/slab.h>
25 #include <linux/binfmts.h>
26 #include <linux/personality.h>
27 #include <linux/init.h>
28 #include <linux/jiffies.h>
29
30 #include <asm/system.h>
31 #include <asm/uaccess.h>
32 #include <asm/pgalloc.h>
33 #include <asm/cacheflush.h>
34 #include <asm/user32.h>
35 #include <asm/ia32.h>
36
37 #undef WARN_OLD
38 #undef CORE_DUMP /* probably broken */
39
40 static int load_aout_binary(struct linux_binprm *, struct pt_regs *regs);
41 static int load_aout_library(struct file *);
42
43 #ifdef CORE_DUMP
44 static int aout_core_dump(long signr, struct pt_regs *regs, struct file *file,
45                           unsigned long limit);
46
47 /*
48  * fill in the user structure for a core dump..
49  */
50 static void dump_thread32(struct pt_regs *regs, struct user32 *dump)
51 {
52         u32 fs, gs;
53
54 /* changed the size calculations - should hopefully work better. lbt */
55         dump->magic = CMAGIC;
56         dump->start_code = 0;
57         dump->start_stack = regs->sp & ~(PAGE_SIZE - 1);
58         dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
59         dump->u_dsize = ((unsigned long)
60                          (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
61         dump->u_dsize -= dump->u_tsize;
62         dump->u_ssize = 0;
63         dump->u_debugreg[0] = current->thread.debugreg0;
64         dump->u_debugreg[1] = current->thread.debugreg1;
65         dump->u_debugreg[2] = current->thread.debugreg2;
66         dump->u_debugreg[3] = current->thread.debugreg3;
67         dump->u_debugreg[4] = 0;
68         dump->u_debugreg[5] = 0;
69         dump->u_debugreg[6] = current->thread.debugreg6;
70         dump->u_debugreg[7] = current->thread.debugreg7;
71
72         if (dump->start_stack < 0xc0000000) {
73                 unsigned long tmp;
74
75                 tmp = (unsigned long) (0xc0000000 - dump->start_stack);
76                 dump->u_ssize = tmp >> PAGE_SHIFT;
77         }
78
79         dump->regs.bx = regs->bx;
80         dump->regs.cx = regs->cx;
81         dump->regs.dx = regs->dx;
82         dump->regs.si = regs->si;
83         dump->regs.di = regs->di;
84         dump->regs.bp = regs->bp;
85         dump->regs.ax = regs->ax;
86         dump->regs.ds = current->thread.ds;
87         dump->regs.es = current->thread.es;
88         asm("movl %%fs,%0" : "=r" (fs)); dump->regs.fs = fs;
89         asm("movl %%gs,%0" : "=r" (gs)); dump->regs.gs = gs;
90         dump->regs.orig_ax = regs->orig_ax;
91         dump->regs.ip = regs->ip;
92         dump->regs.cs = regs->cs;
93         dump->regs.flags = regs->flags;
94         dump->regs.sp = regs->sp;
95         dump->regs.ss = regs->ss;
96
97 #if 1 /* FIXME */
98         dump->u_fpvalid = 0;
99 #else
100         dump->u_fpvalid = dump_fpu(regs, &dump->i387);
101 #endif
102 }
103
104 #endif
105
106 static struct linux_binfmt aout_format = {
107         .module         = THIS_MODULE,
108         .load_binary    = load_aout_binary,
109         .load_shlib     = load_aout_library,
110 #ifdef CORE_DUMP
111         .core_dump      = aout_core_dump,
112 #endif
113         .min_coredump   = PAGE_SIZE
114 };
115
116 static void set_brk(unsigned long start, unsigned long end)
117 {
118         start = PAGE_ALIGN(start);
119         end = PAGE_ALIGN(end);
120         if (end <= start)
121                 return;
122         down_write(&current->mm->mmap_sem);
123         do_brk(start, end - start);
124         up_write(&current->mm->mmap_sem);
125 }
126
127 #ifdef CORE_DUMP
128 /*
129  * These are the only things you should do on a core-file: use only these
130  * macros to write out all the necessary info.
131  */
132
133 static int dump_write(struct file *file, const void *addr, int nr)
134 {
135         return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
136 }
137
138 #define DUMP_WRITE(addr, nr)                         \
139         if (!dump_write(file, (void *)(addr), (nr))) \
140                 goto end_coredump;
141
142 #define DUMP_SEEK(offset)                                               \
143         if (file->f_op->llseek) {                                       \
144                 if (file->f_op->llseek(file, (offset), 0) != (offset))  \
145                         goto end_coredump;                              \
146         } else                                                          \
147                 file->f_pos = (offset)
148
149 #define START_DATA()    (u.u_tsize << PAGE_SHIFT)
150 #define START_STACK(u)  (u.start_stack)
151
152 /*
153  * Routine writes a core dump image in the current directory.
154  * Currently only a stub-function.
155  *
156  * Note that setuid/setgid files won't make a core-dump if the uid/gid
157  * changed due to the set[u|g]id. It's enforced by the "current->mm->dumpable"
158  * field, which also makes sure the core-dumps won't be recursive if the
159  * dumping of the process results in another error..
160  */
161
162 static int aout_core_dump(long signr, struct pt_regs *regs, struct file *file,
163                           unsigned long limit)
164 {
165         mm_segment_t fs;
166         int has_dumped = 0;
167         unsigned long dump_start, dump_size;
168         struct user32 dump;
169
170         fs = get_fs();
171         set_fs(KERNEL_DS);
172         has_dumped = 1;
173         current->flags |= PF_DUMPCORE;
174         strncpy(dump.u_comm, current->comm, sizeof(current->comm));
175         dump.u_ar0 = offsetof(struct user32, regs);
176         dump.signal = signr;
177         dump_thread32(regs, &dump);
178
179         /*
180          * If the size of the dump file exceeds the rlimit, then see
181          * what would happen if we wrote the stack, but not the data
182          * area.
183          */
184         if ((dump.u_dsize + dump.u_ssize + 1) * PAGE_SIZE > limit)
185                 dump.u_dsize = 0;
186
187         /* Make sure we have enough room to write the stack and data areas. */
188         if ((dump.u_ssize + 1) * PAGE_SIZE > limit)
189                 dump.u_ssize = 0;
190
191         /* make sure we actually have a data and stack area to dump */
192         set_fs(USER_DS);
193         if (!access_ok(VERIFY_READ, (void *) (unsigned long)START_DATA(dump),
194                        dump.u_dsize << PAGE_SHIFT))
195                 dump.u_dsize = 0;
196         if (!access_ok(VERIFY_READ, (void *) (unsigned long)START_STACK(dump),
197                        dump.u_ssize << PAGE_SHIFT))
198                 dump.u_ssize = 0;
199
200         set_fs(KERNEL_DS);
201         /* struct user */
202         DUMP_WRITE(&dump, sizeof(dump));
203         /* Now dump all of the user data.  Include malloced stuff as well */
204         DUMP_SEEK(PAGE_SIZE);
205         /* now we start writing out the user space info */
206         set_fs(USER_DS);
207         /* Dump the data area */
208         if (dump.u_dsize != 0) {
209                 dump_start = START_DATA(dump);
210                 dump_size = dump.u_dsize << PAGE_SHIFT;
211                 DUMP_WRITE(dump_start, dump_size);
212         }
213         /* Now prepare to dump the stack area */
214         if (dump.u_ssize != 0) {
215                 dump_start = START_STACK(dump);
216                 dump_size = dump.u_ssize << PAGE_SHIFT;
217                 DUMP_WRITE(dump_start, dump_size);
218         }
219         /*
220          * Finally dump the task struct.  Not be used by gdb, but
221          * could be useful
222          */
223         set_fs(KERNEL_DS);
224         DUMP_WRITE(current, sizeof(*current));
225 end_coredump:
226         set_fs(fs);
227         return has_dumped;
228 }
229 #endif
230
231 /*
232  * create_aout_tables() parses the env- and arg-strings in new user
233  * memory and creates the pointer tables from them, and puts their
234  * addresses on the "stack", returning the new stack pointer value.
235  */
236 static u32 __user *create_aout_tables(char __user *p, struct linux_binprm *bprm)
237 {
238         u32 __user *argv, *envp, *sp;
239         int argc = bprm->argc, envc = bprm->envc;
240
241         sp = (u32 __user *) ((-(unsigned long)sizeof(u32)) & (unsigned long) p);
242         sp -= envc+1;
243         envp = sp;
244         sp -= argc+1;
245         argv = sp;
246         put_user((unsigned long) envp, --sp);
247         put_user((unsigned long) argv, --sp);
248         put_user(argc, --sp);
249         current->mm->arg_start = (unsigned long) p;
250         while (argc-- > 0) {
251                 char c;
252
253                 put_user((u32)(unsigned long)p, argv++);
254                 do {
255                         get_user(c, p++);
256                 } while (c);
257         }
258         put_user(0, argv);
259         current->mm->arg_end = current->mm->env_start = (unsigned long) p;
260         while (envc-- > 0) {
261                 char c;
262
263                 put_user((u32)(unsigned long)p, envp++);
264                 do {
265                         get_user(c, p++);
266                 } while (c);
267         }
268         put_user(0, envp);
269         current->mm->env_end = (unsigned long) p;
270         return sp;
271 }
272
273 /*
274  * These are the functions used to load a.out style executables and shared
275  * libraries.  There is no binary dependent code anywhere else.
276  */
277 static int load_aout_binary(struct linux_binprm *bprm, struct pt_regs *regs)
278 {
279         unsigned long error, fd_offset, rlim;
280         struct exec ex;
281         int retval;
282
283         ex = *((struct exec *) bprm->buf);              /* exec-header */
284         if ((N_MAGIC(ex) != ZMAGIC && N_MAGIC(ex) != OMAGIC &&
285              N_MAGIC(ex) != QMAGIC && N_MAGIC(ex) != NMAGIC) ||
286             N_TRSIZE(ex) || N_DRSIZE(ex) ||
287             i_size_read(bprm->file->f_path.dentry->d_inode) <
288             ex.a_text+ex.a_data+N_SYMSIZE(ex)+N_TXTOFF(ex)) {
289                 return -ENOEXEC;
290         }
291
292         fd_offset = N_TXTOFF(ex);
293
294         /* Check initial limits. This avoids letting people circumvent
295          * size limits imposed on them by creating programs with large
296          * arrays in the data or bss.
297          */
298         rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
299         if (rlim >= RLIM_INFINITY)
300                 rlim = ~0;
301         if (ex.a_data + ex.a_bss > rlim)
302                 return -ENOMEM;
303
304         /* Flush all traces of the currently running executable */
305         retval = flush_old_exec(bprm);
306         if (retval)
307                 return retval;
308
309         regs->cs = __USER32_CS;
310         regs->r8 = regs->r9 = regs->r10 = regs->r11 = regs->r12 =
311                 regs->r13 = regs->r14 = regs->r15 = 0;
312
313         /* OK, This is the point of no return */
314         set_personality(PER_LINUX);
315         set_thread_flag(TIF_IA32);
316         clear_thread_flag(TIF_ABI_PENDING);
317
318         current->mm->end_code = ex.a_text +
319                 (current->mm->start_code = N_TXTADDR(ex));
320         current->mm->end_data = ex.a_data +
321                 (current->mm->start_data = N_DATADDR(ex));
322         current->mm->brk = ex.a_bss +
323                 (current->mm->start_brk = N_BSSADDR(ex));
324         current->mm->free_area_cache = TASK_UNMAPPED_BASE;
325         current->mm->cached_hole_size = 0;
326
327         current->mm->mmap = NULL;
328         compute_creds(bprm);
329         current->flags &= ~PF_FORKNOEXEC;
330
331         if (N_MAGIC(ex) == OMAGIC) {
332                 unsigned long text_addr, map_size;
333                 loff_t pos;
334
335                 text_addr = N_TXTADDR(ex);
336
337                 pos = 32;
338                 map_size = ex.a_text+ex.a_data;
339
340                 down_write(&current->mm->mmap_sem);
341                 error = do_brk(text_addr & PAGE_MASK, map_size);
342                 up_write(&current->mm->mmap_sem);
343
344                 if (error != (text_addr & PAGE_MASK)) {
345                         send_sig(SIGKILL, current, 0);
346                         return error;
347                 }
348
349                 error = bprm->file->f_op->read(bprm->file,
350                          (char __user *)text_addr,
351                           ex.a_text+ex.a_data, &pos);
352                 if ((signed long)error < 0) {
353                         send_sig(SIGKILL, current, 0);
354                         return error;
355                 }
356
357                 flush_icache_range(text_addr, text_addr+ex.a_text+ex.a_data);
358         } else {
359 #ifdef WARN_OLD
360                 static unsigned long error_time, error_time2;
361                 if ((ex.a_text & 0xfff || ex.a_data & 0xfff) &&
362                     (N_MAGIC(ex) != NMAGIC) &&
363                                 time_after(jiffies, error_time2 + 5*HZ)) {
364                         printk(KERN_NOTICE "executable not page aligned\n");
365                         error_time2 = jiffies;
366                 }
367
368                 if ((fd_offset & ~PAGE_MASK) != 0 &&
369                             time_after(jiffies, error_time + 5*HZ)) {
370                         printk(KERN_WARNING
371                                "fd_offset is not page aligned. Please convert "
372                                "program: %s\n",
373                                bprm->file->f_path.dentry->d_name.name);
374                         error_time = jiffies;
375                 }
376 #endif
377
378                 if (!bprm->file->f_op->mmap || (fd_offset & ~PAGE_MASK) != 0) {
379                         loff_t pos = fd_offset;
380
381                         down_write(&current->mm->mmap_sem);
382                         do_brk(N_TXTADDR(ex), ex.a_text+ex.a_data);
383                         up_write(&current->mm->mmap_sem);
384                         bprm->file->f_op->read(bprm->file,
385                                         (char __user *)N_TXTADDR(ex),
386                                         ex.a_text+ex.a_data, &pos);
387                         flush_icache_range((unsigned long) N_TXTADDR(ex),
388                                            (unsigned long) N_TXTADDR(ex) +
389                                            ex.a_text+ex.a_data);
390                         goto beyond_if;
391                 }
392
393                 down_write(&current->mm->mmap_sem);
394                 error = do_mmap(bprm->file, N_TXTADDR(ex), ex.a_text,
395                                 PROT_READ | PROT_EXEC,
396                                 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE |
397                                 MAP_EXECUTABLE | MAP_32BIT,
398                                 fd_offset);
399                 up_write(&current->mm->mmap_sem);
400
401                 if (error != N_TXTADDR(ex)) {
402                         send_sig(SIGKILL, current, 0);
403                         return error;
404                 }
405
406                 down_write(&current->mm->mmap_sem);
407                 error = do_mmap(bprm->file, N_DATADDR(ex), ex.a_data,
408                                 PROT_READ | PROT_WRITE | PROT_EXEC,
409                                 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE |
410                                 MAP_EXECUTABLE | MAP_32BIT,
411                                 fd_offset + ex.a_text);
412                 up_write(&current->mm->mmap_sem);
413                 if (error != N_DATADDR(ex)) {
414                         send_sig(SIGKILL, current, 0);
415                         return error;
416                 }
417         }
418 beyond_if:
419         set_binfmt(&aout_format);
420
421         set_brk(current->mm->start_brk, current->mm->brk);
422
423         retval = setup_arg_pages(bprm, IA32_STACK_TOP, EXSTACK_DEFAULT);
424         if (retval < 0) {
425                 /* Someone check-me: is this error path enough? */
426                 send_sig(SIGKILL, current, 0);
427                 return retval;
428         }
429
430         current->mm->start_stack =
431                 (unsigned long)create_aout_tables((char __user *)bprm->p, bprm);
432         /* start thread */
433         asm volatile("movl %0,%%fs" :: "r" (0)); \
434         asm volatile("movl %0,%%es; movl %0,%%ds": :"r" (__USER32_DS));
435         load_gs_index(0);
436         (regs)->ip = ex.a_entry;
437         (regs)->sp = current->mm->start_stack;
438         (regs)->flags = 0x200;
439         (regs)->cs = __USER32_CS;
440         (regs)->ss = __USER32_DS;
441         regs->r8 = regs->r9 = regs->r10 = regs->r11 =
442         regs->r12 = regs->r13 = regs->r14 = regs->r15 = 0;
443         set_fs(USER_DS);
444         if (unlikely(current->ptrace & PT_PTRACED)) {
445                 if (current->ptrace & PT_TRACE_EXEC)
446                         ptrace_notify((PTRACE_EVENT_EXEC << 8) | SIGTRAP);
447                 else
448                         send_sig(SIGTRAP, current, 0);
449         }
450         return 0;
451 }
452
453 static int load_aout_library(struct file *file)
454 {
455         struct inode *inode;
456         unsigned long bss, start_addr, len, error;
457         int retval;
458         struct exec ex;
459
460         inode = file->f_path.dentry->d_inode;
461
462         retval = -ENOEXEC;
463         error = kernel_read(file, 0, (char *) &ex, sizeof(ex));
464         if (error != sizeof(ex))
465                 goto out;
466
467         /* We come in here for the regular a.out style of shared libraries */
468         if ((N_MAGIC(ex) != ZMAGIC && N_MAGIC(ex) != QMAGIC) || N_TRSIZE(ex) ||
469             N_DRSIZE(ex) || ((ex.a_entry & 0xfff) && N_MAGIC(ex) == ZMAGIC) ||
470             i_size_read(inode) <
471             ex.a_text+ex.a_data+N_SYMSIZE(ex)+N_TXTOFF(ex)) {
472                 goto out;
473         }
474
475         if (N_FLAGS(ex))
476                 goto out;
477
478         /* For  QMAGIC, the starting address is 0x20 into the page.  We mask
479            this off to get the starting address for the page */
480
481         start_addr =  ex.a_entry & 0xfffff000;
482
483         if ((N_TXTOFF(ex) & ~PAGE_MASK) != 0) {
484                 loff_t pos = N_TXTOFF(ex);
485
486 #ifdef WARN_OLD
487                 static unsigned long error_time;
488                 if (time_after(jiffies, error_time + 5*HZ)) {
489                         printk(KERN_WARNING
490                                "N_TXTOFF is not page aligned. Please convert "
491                                "library: %s\n",
492                                file->f_path.dentry->d_name.name);
493                         error_time = jiffies;
494                 }
495 #endif
496                 down_write(&current->mm->mmap_sem);
497                 do_brk(start_addr, ex.a_text + ex.a_data + ex.a_bss);
498                 up_write(&current->mm->mmap_sem);
499
500                 file->f_op->read(file, (char __user *)start_addr,
501                         ex.a_text + ex.a_data, &pos);
502                 flush_icache_range((unsigned long) start_addr,
503                                    (unsigned long) start_addr + ex.a_text +
504                                    ex.a_data);
505
506                 retval = 0;
507                 goto out;
508         }
509         /* Now use mmap to map the library into memory. */
510         down_write(&current->mm->mmap_sem);
511         error = do_mmap(file, start_addr, ex.a_text + ex.a_data,
512                         PROT_READ | PROT_WRITE | PROT_EXEC,
513                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE | MAP_32BIT,
514                         N_TXTOFF(ex));
515         up_write(&current->mm->mmap_sem);
516         retval = error;
517         if (error != start_addr)
518                 goto out;
519
520         len = PAGE_ALIGN(ex.a_text + ex.a_data);
521         bss = ex.a_text + ex.a_data + ex.a_bss;
522         if (bss > len) {
523                 down_write(&current->mm->mmap_sem);
524                 error = do_brk(start_addr + len, bss - len);
525                 up_write(&current->mm->mmap_sem);
526                 retval = error;
527                 if (error != start_addr + len)
528                         goto out;
529         }
530         retval = 0;
531 out:
532         return retval;
533 }
534
535 static int __init init_aout_binfmt(void)
536 {
537         return register_binfmt(&aout_format);
538 }
539
540 static void __exit exit_aout_binfmt(void)
541 {
542         unregister_binfmt(&aout_format);
543 }
544
545 module_init(init_aout_binfmt);
546 module_exit(exit_aout_binfmt);
547 MODULE_LICENSE("GPL");