include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit...
[safe/jmp/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/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 /* probably 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 static int dump_write(struct file *file, const void *addr, int nr)
135 {
136         return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
137 }
138
139 #define DUMP_WRITE(addr, nr)                         \
140         if (!dump_write(file, (void *)(addr), (nr))) \
141                 goto end_coredump;
142
143 #define DUMP_SEEK(offset)                                               \
144         if (file->f_op->llseek) {                                       \
145                 if (file->f_op->llseek(file, (offset), 0) != (offset))  \
146                         goto end_coredump;                              \
147         } else                                                          \
148                 file->f_pos = (offset)
149
150 #define START_DATA()    (u.u_tsize << PAGE_SHIFT)
151 #define START_STACK(u)  (u.start_stack)
152
153 /*
154  * Routine writes a core dump image in the current directory.
155  * Currently only a stub-function.
156  *
157  * Note that setuid/setgid files won't make a core-dump if the uid/gid
158  * changed due to the set[u|g]id. It's enforced by the "current->mm->dumpable"
159  * field, which also makes sure the core-dumps won't be recursive if the
160  * dumping of the process results in another error..
161  */
162
163 static int aout_core_dump(long signr, struct pt_regs *regs, struct file *file,
164                           unsigned long limit)
165 {
166         mm_segment_t fs;
167         int has_dumped = 0;
168         unsigned long dump_start, dump_size;
169         struct user32 dump;
170
171         fs = get_fs();
172         set_fs(KERNEL_DS);
173         has_dumped = 1;
174         current->flags |= PF_DUMPCORE;
175         strncpy(dump.u_comm, current->comm, sizeof(current->comm));
176         dump.u_ar0 = offsetof(struct user32, regs);
177         dump.signal = signr;
178         dump_thread32(regs, &dump);
179
180         /*
181          * If the size of the dump file exceeds the rlimit, then see
182          * what would happen if we wrote the stack, but not the data
183          * area.
184          */
185         if ((dump.u_dsize + dump.u_ssize + 1) * PAGE_SIZE > limit)
186                 dump.u_dsize = 0;
187
188         /* Make sure we have enough room to write the stack and data areas. */
189         if ((dump.u_ssize + 1) * PAGE_SIZE > limit)
190                 dump.u_ssize = 0;
191
192         /* make sure we actually have a data and stack area to dump */
193         set_fs(USER_DS);
194         if (!access_ok(VERIFY_READ, (void *) (unsigned long)START_DATA(dump),
195                        dump.u_dsize << PAGE_SHIFT))
196                 dump.u_dsize = 0;
197         if (!access_ok(VERIFY_READ, (void *) (unsigned long)START_STACK(dump),
198                        dump.u_ssize << PAGE_SHIFT))
199                 dump.u_ssize = 0;
200
201         set_fs(KERNEL_DS);
202         /* struct user */
203         DUMP_WRITE(&dump, sizeof(dump));
204         /* Now dump all of the user data.  Include malloced stuff as well */
205         DUMP_SEEK(PAGE_SIZE);
206         /* now we start writing out the user space info */
207         set_fs(USER_DS);
208         /* Dump the data area */
209         if (dump.u_dsize != 0) {
210                 dump_start = START_DATA(dump);
211                 dump_size = dump.u_dsize << PAGE_SHIFT;
212                 DUMP_WRITE(dump_start, dump_size);
213         }
214         /* Now prepare to dump the stack area */
215         if (dump.u_ssize != 0) {
216                 dump_start = START_STACK(dump);
217                 dump_size = dump.u_ssize << PAGE_SHIFT;
218                 DUMP_WRITE(dump_start, dump_size);
219         }
220         /*
221          * Finally dump the task struct.  Not be used by gdb, but
222          * could be useful
223          */
224         set_fs(KERNEL_DS);
225         DUMP_WRITE(current, sizeof(*current));
226 end_coredump:
227         set_fs(fs);
228         return has_dumped;
229 }
230 #endif
231
232 /*
233  * create_aout_tables() parses the env- and arg-strings in new user
234  * memory and creates the pointer tables from them, and puts their
235  * addresses on the "stack", returning the new stack pointer value.
236  */
237 static u32 __user *create_aout_tables(char __user *p, struct linux_binprm *bprm)
238 {
239         u32 __user *argv, *envp, *sp;
240         int argc = bprm->argc, envc = bprm->envc;
241
242         sp = (u32 __user *) ((-(unsigned long)sizeof(u32)) & (unsigned long) p);
243         sp -= envc+1;
244         envp = sp;
245         sp -= argc+1;
246         argv = sp;
247         put_user((unsigned long) envp, --sp);
248         put_user((unsigned long) argv, --sp);
249         put_user(argc, --sp);
250         current->mm->arg_start = (unsigned long) p;
251         while (argc-- > 0) {
252                 char c;
253
254                 put_user((u32)(unsigned long)p, argv++);
255                 do {
256                         get_user(c, p++);
257                 } while (c);
258         }
259         put_user(0, argv);
260         current->mm->arg_end = current->mm->env_start = (unsigned long) p;
261         while (envc-- > 0) {
262                 char c;
263
264                 put_user((u32)(unsigned long)p, envp++);
265                 do {
266                         get_user(c, p++);
267                 } while (c);
268         }
269         put_user(0, envp);
270         current->mm->env_end = (unsigned long) p;
271         return sp;
272 }
273
274 /*
275  * These are the functions used to load a.out style executables and shared
276  * libraries.  There is no binary dependent code anywhere else.
277  */
278 static int load_aout_binary(struct linux_binprm *bprm, struct pt_regs *regs)
279 {
280         unsigned long error, fd_offset, rlim;
281         struct exec ex;
282         int retval;
283
284         ex = *((struct exec *) bprm->buf);              /* exec-header */
285         if ((N_MAGIC(ex) != ZMAGIC && N_MAGIC(ex) != OMAGIC &&
286              N_MAGIC(ex) != QMAGIC && N_MAGIC(ex) != NMAGIC) ||
287             N_TRSIZE(ex) || N_DRSIZE(ex) ||
288             i_size_read(bprm->file->f_path.dentry->d_inode) <
289             ex.a_text+ex.a_data+N_SYMSIZE(ex)+N_TXTOFF(ex)) {
290                 return -ENOEXEC;
291         }
292
293         fd_offset = N_TXTOFF(ex);
294
295         /* Check initial limits. This avoids letting people circumvent
296          * size limits imposed on them by creating programs with large
297          * arrays in the data or bss.
298          */
299         rlim = rlimit(RLIMIT_DATA);
300         if (rlim >= RLIM_INFINITY)
301                 rlim = ~0;
302         if (ex.a_data + ex.a_bss > rlim)
303                 return -ENOMEM;
304
305         /* Flush all traces of the currently running executable */
306         retval = flush_old_exec(bprm);
307         if (retval)
308                 return retval;
309
310         /* OK, This is the point of no return */
311         set_personality(PER_LINUX);
312         set_thread_flag(TIF_IA32);
313
314         setup_new_exec(bprm);
315
316         regs->cs = __USER32_CS;
317         regs->r8 = regs->r9 = regs->r10 = regs->r11 = regs->r12 =
318                 regs->r13 = regs->r14 = regs->r15 = 0;
319
320         current->mm->end_code = ex.a_text +
321                 (current->mm->start_code = N_TXTADDR(ex));
322         current->mm->end_data = ex.a_data +
323                 (current->mm->start_data = N_DATADDR(ex));
324         current->mm->brk = ex.a_bss +
325                 (current->mm->start_brk = N_BSSADDR(ex));
326         current->mm->free_area_cache = TASK_UNMAPPED_BASE;
327         current->mm->cached_hole_size = 0;
328
329         install_exec_creds(bprm);
330         current->flags &= ~PF_FORKNOEXEC;
331
332         if (N_MAGIC(ex) == OMAGIC) {
333                 unsigned long text_addr, map_size;
334                 loff_t pos;
335
336                 text_addr = N_TXTADDR(ex);
337
338                 pos = 32;
339                 map_size = ex.a_text+ex.a_data;
340
341                 down_write(&current->mm->mmap_sem);
342                 error = do_brk(text_addr & PAGE_MASK, map_size);
343                 up_write(&current->mm->mmap_sem);
344
345                 if (error != (text_addr & PAGE_MASK)) {
346                         send_sig(SIGKILL, current, 0);
347                         return error;
348                 }
349
350                 error = bprm->file->f_op->read(bprm->file,
351                          (char __user *)text_addr,
352                           ex.a_text+ex.a_data, &pos);
353                 if ((signed long)error < 0) {
354                         send_sig(SIGKILL, current, 0);
355                         return error;
356                 }
357
358                 flush_icache_range(text_addr, text_addr+ex.a_text+ex.a_data);
359         } else {
360 #ifdef WARN_OLD
361                 static unsigned long error_time, error_time2;
362                 if ((ex.a_text & 0xfff || ex.a_data & 0xfff) &&
363                     (N_MAGIC(ex) != NMAGIC) &&
364                                 time_after(jiffies, error_time2 + 5*HZ)) {
365                         printk(KERN_NOTICE "executable not page aligned\n");
366                         error_time2 = jiffies;
367                 }
368
369                 if ((fd_offset & ~PAGE_MASK) != 0 &&
370                             time_after(jiffies, error_time + 5*HZ)) {
371                         printk(KERN_WARNING
372                                "fd_offset is not page aligned. Please convert "
373                                "program: %s\n",
374                                bprm->file->f_path.dentry->d_name.name);
375                         error_time = jiffies;
376                 }
377 #endif
378
379                 if (!bprm->file->f_op->mmap || (fd_offset & ~PAGE_MASK) != 0) {
380                         loff_t pos = fd_offset;
381
382                         down_write(&current->mm->mmap_sem);
383                         do_brk(N_TXTADDR(ex), ex.a_text+ex.a_data);
384                         up_write(&current->mm->mmap_sem);
385                         bprm->file->f_op->read(bprm->file,
386                                         (char __user *)N_TXTADDR(ex),
387                                         ex.a_text+ex.a_data, &pos);
388                         flush_icache_range((unsigned long) N_TXTADDR(ex),
389                                            (unsigned long) N_TXTADDR(ex) +
390                                            ex.a_text+ex.a_data);
391                         goto beyond_if;
392                 }
393
394                 down_write(&current->mm->mmap_sem);
395                 error = do_mmap(bprm->file, N_TXTADDR(ex), ex.a_text,
396                                 PROT_READ | PROT_EXEC,
397                                 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE |
398                                 MAP_EXECUTABLE | MAP_32BIT,
399                                 fd_offset);
400                 up_write(&current->mm->mmap_sem);
401
402                 if (error != N_TXTADDR(ex)) {
403                         send_sig(SIGKILL, current, 0);
404                         return error;
405                 }
406
407                 down_write(&current->mm->mmap_sem);
408                 error = do_mmap(bprm->file, N_DATADDR(ex), ex.a_data,
409                                 PROT_READ | PROT_WRITE | PROT_EXEC,
410                                 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE |
411                                 MAP_EXECUTABLE | MAP_32BIT,
412                                 fd_offset + ex.a_text);
413                 up_write(&current->mm->mmap_sem);
414                 if (error != N_DATADDR(ex)) {
415                         send_sig(SIGKILL, current, 0);
416                         return error;
417                 }
418         }
419 beyond_if:
420         set_binfmt(&aout_format);
421
422         set_brk(current->mm->start_brk, current->mm->brk);
423
424         retval = setup_arg_pages(bprm, IA32_STACK_TOP, EXSTACK_DEFAULT);
425         if (retval < 0) {
426                 /* Someone check-me: is this error path enough? */
427                 send_sig(SIGKILL, current, 0);
428                 return retval;
429         }
430
431         current->mm->start_stack =
432                 (unsigned long)create_aout_tables((char __user *)bprm->p, bprm);
433         /* start thread */
434         loadsegment(fs, 0);
435         loadsegment(ds, __USER32_DS);
436         loadsegment(es, __USER32_DS);
437         load_gs_index(0);
438         (regs)->ip = ex.a_entry;
439         (regs)->sp = current->mm->start_stack;
440         (regs)->flags = 0x200;
441         (regs)->cs = __USER32_CS;
442         (regs)->ss = __USER32_DS;
443         regs->r8 = regs->r9 = regs->r10 = regs->r11 =
444         regs->r12 = regs->r13 = regs->r14 = regs->r15 = 0;
445         set_fs(USER_DS);
446         return 0;
447 }
448
449 static int load_aout_library(struct file *file)
450 {
451         struct inode *inode;
452         unsigned long bss, start_addr, len, error;
453         int retval;
454         struct exec ex;
455
456         inode = file->f_path.dentry->d_inode;
457
458         retval = -ENOEXEC;
459         error = kernel_read(file, 0, (char *) &ex, sizeof(ex));
460         if (error != sizeof(ex))
461                 goto out;
462
463         /* We come in here for the regular a.out style of shared libraries */
464         if ((N_MAGIC(ex) != ZMAGIC && N_MAGIC(ex) != QMAGIC) || N_TRSIZE(ex) ||
465             N_DRSIZE(ex) || ((ex.a_entry & 0xfff) && N_MAGIC(ex) == ZMAGIC) ||
466             i_size_read(inode) <
467             ex.a_text+ex.a_data+N_SYMSIZE(ex)+N_TXTOFF(ex)) {
468                 goto out;
469         }
470
471         if (N_FLAGS(ex))
472                 goto out;
473
474         /* For  QMAGIC, the starting address is 0x20 into the page.  We mask
475            this off to get the starting address for the page */
476
477         start_addr =  ex.a_entry & 0xfffff000;
478
479         if ((N_TXTOFF(ex) & ~PAGE_MASK) != 0) {
480                 loff_t pos = N_TXTOFF(ex);
481
482 #ifdef WARN_OLD
483                 static unsigned long error_time;
484                 if (time_after(jiffies, error_time + 5*HZ)) {
485                         printk(KERN_WARNING
486                                "N_TXTOFF is not page aligned. Please convert "
487                                "library: %s\n",
488                                file->f_path.dentry->d_name.name);
489                         error_time = jiffies;
490                 }
491 #endif
492                 down_write(&current->mm->mmap_sem);
493                 do_brk(start_addr, ex.a_text + ex.a_data + ex.a_bss);
494                 up_write(&current->mm->mmap_sem);
495
496                 file->f_op->read(file, (char __user *)start_addr,
497                         ex.a_text + ex.a_data, &pos);
498                 flush_icache_range((unsigned long) start_addr,
499                                    (unsigned long) start_addr + ex.a_text +
500                                    ex.a_data);
501
502                 retval = 0;
503                 goto out;
504         }
505         /* Now use mmap to map the library into memory. */
506         down_write(&current->mm->mmap_sem);
507         error = do_mmap(file, start_addr, ex.a_text + ex.a_data,
508                         PROT_READ | PROT_WRITE | PROT_EXEC,
509                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE | MAP_32BIT,
510                         N_TXTOFF(ex));
511         up_write(&current->mm->mmap_sem);
512         retval = error;
513         if (error != start_addr)
514                 goto out;
515
516         len = PAGE_ALIGN(ex.a_text + ex.a_data);
517         bss = ex.a_text + ex.a_data + ex.a_bss;
518         if (bss > len) {
519                 down_write(&current->mm->mmap_sem);
520                 error = do_brk(start_addr + len, bss - len);
521                 up_write(&current->mm->mmap_sem);
522                 retval = error;
523                 if (error != start_addr + len)
524                         goto out;
525         }
526         retval = 0;
527 out:
528         return retval;
529 }
530
531 static int __init init_aout_binfmt(void)
532 {
533         return register_binfmt(&aout_format);
534 }
535
536 static void __exit exit_aout_binfmt(void)
537 {
538         unregister_binfmt(&aout_format);
539 }
540
541 module_init(init_aout_binfmt);
542 module_exit(exit_aout_binfmt);
543 MODULE_LICENSE("GPL");