b6817ee9033fcf7befff4926be2e8a9acf7cfbc1
[~shefty/rdma-dev.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/uaccess.h>
30 #include <asm/pgalloc.h>
31 #include <asm/cacheflush.h>
32 #include <asm/user32.h>
33 #include <asm/ia32.h>
34
35 #undef WARN_OLD
36 #undef CORE_DUMP /* definitely broken */
37
38 static int load_aout_binary(struct linux_binprm *, struct pt_regs *regs);
39 static int load_aout_library(struct file *);
40
41 #ifdef CORE_DUMP
42 static int aout_core_dump(long signr, struct pt_regs *regs, struct file *file,
43                           unsigned long limit);
44
45 /*
46  * fill in the user structure for a core dump..
47  */
48 static void dump_thread32(struct pt_regs *regs, struct user32 *dump)
49 {
50         u32 fs, gs;
51
52 /* changed the size calculations - should hopefully work better. lbt */
53         dump->magic = CMAGIC;
54         dump->start_code = 0;
55         dump->start_stack = regs->sp & ~(PAGE_SIZE - 1);
56         dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
57         dump->u_dsize = ((unsigned long)
58                          (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
59         dump->u_dsize -= dump->u_tsize;
60         dump->u_ssize = 0;
61         dump->u_debugreg[0] = current->thread.debugreg0;
62         dump->u_debugreg[1] = current->thread.debugreg1;
63         dump->u_debugreg[2] = current->thread.debugreg2;
64         dump->u_debugreg[3] = current->thread.debugreg3;
65         dump->u_debugreg[4] = 0;
66         dump->u_debugreg[5] = 0;
67         dump->u_debugreg[6] = current->thread.debugreg6;
68         dump->u_debugreg[7] = current->thread.debugreg7;
69
70         if (dump->start_stack < 0xc0000000) {
71                 unsigned long tmp;
72
73                 tmp = (unsigned long) (0xc0000000 - dump->start_stack);
74                 dump->u_ssize = tmp >> PAGE_SHIFT;
75         }
76
77         dump->regs.bx = regs->bx;
78         dump->regs.cx = regs->cx;
79         dump->regs.dx = regs->dx;
80         dump->regs.si = regs->si;
81         dump->regs.di = regs->di;
82         dump->regs.bp = regs->bp;
83         dump->regs.ax = regs->ax;
84         dump->regs.ds = current->thread.ds;
85         dump->regs.es = current->thread.es;
86         savesegment(fs, fs);
87         dump->regs.fs = fs;
88         savesegment(gs, gs);
89         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         vm_brk(start, end - start);
123 }
124
125 #ifdef CORE_DUMP
126 /*
127  * These are the only things you should do on a core-file: use only these
128  * macros to write out all the necessary info.
129  */
130
131 #include <linux/coredump.h>
132
133 #define DUMP_WRITE(addr, nr)                         \
134         if (!dump_write(file, (void *)(addr), (nr))) \
135                 goto end_coredump;
136
137 #define DUMP_SEEK(offset)               \
138         if (!dump_seek(file, offset))   \
139                 goto end_coredump;
140
141 #define START_DATA()    (u.u_tsize << PAGE_SHIFT)
142 #define START_STACK(u)  (u.start_stack)
143
144 /*
145  * Routine writes a core dump image in the current directory.
146  * Currently only a stub-function.
147  *
148  * Note that setuid/setgid files won't make a core-dump if the uid/gid
149  * changed due to the set[u|g]id. It's enforced by the "current->mm->dumpable"
150  * field, which also makes sure the core-dumps won't be recursive if the
151  * dumping of the process results in another error..
152  */
153
154 static int aout_core_dump(long signr, struct pt_regs *regs, struct file *file,
155                           unsigned long limit)
156 {
157         mm_segment_t fs;
158         int has_dumped = 0;
159         unsigned long dump_start, dump_size;
160         struct user32 dump;
161
162         fs = get_fs();
163         set_fs(KERNEL_DS);
164         has_dumped = 1;
165         current->flags |= PF_DUMPCORE;
166         strncpy(dump.u_comm, current->comm, sizeof(current->comm));
167         dump.u_ar0 = offsetof(struct user32, regs);
168         dump.signal = signr;
169         dump_thread32(regs, &dump);
170
171         /*
172          * If the size of the dump file exceeds the rlimit, then see
173          * what would happen if we wrote the stack, but not the data
174          * area.
175          */
176         if ((dump.u_dsize + dump.u_ssize + 1) * PAGE_SIZE > limit)
177                 dump.u_dsize = 0;
178
179         /* Make sure we have enough room to write the stack and data areas. */
180         if ((dump.u_ssize + 1) * PAGE_SIZE > limit)
181                 dump.u_ssize = 0;
182
183         /* make sure we actually have a data and stack area to dump */
184         set_fs(USER_DS);
185         if (!access_ok(VERIFY_READ, (void *) (unsigned long)START_DATA(dump),
186                        dump.u_dsize << PAGE_SHIFT))
187                 dump.u_dsize = 0;
188         if (!access_ok(VERIFY_READ, (void *) (unsigned long)START_STACK(dump),
189                        dump.u_ssize << PAGE_SHIFT))
190                 dump.u_ssize = 0;
191
192         set_fs(KERNEL_DS);
193         /* struct user */
194         DUMP_WRITE(&dump, sizeof(dump));
195         /* Now dump all of the user data.  Include malloced stuff as well */
196         DUMP_SEEK(PAGE_SIZE);
197         /* now we start writing out the user space info */
198         set_fs(USER_DS);
199         /* Dump the data area */
200         if (dump.u_dsize != 0) {
201                 dump_start = START_DATA(dump);
202                 dump_size = dump.u_dsize << PAGE_SHIFT;
203                 DUMP_WRITE(dump_start, dump_size);
204         }
205         /* Now prepare to dump the stack area */
206         if (dump.u_ssize != 0) {
207                 dump_start = START_STACK(dump);
208                 dump_size = dump.u_ssize << PAGE_SHIFT;
209                 DUMP_WRITE(dump_start, dump_size);
210         }
211 end_coredump:
212         set_fs(fs);
213         return has_dumped;
214 }
215 #endif
216
217 /*
218  * create_aout_tables() parses the env- and arg-strings in new user
219  * memory and creates the pointer tables from them, and puts their
220  * addresses on the "stack", returning the new stack pointer value.
221  */
222 static u32 __user *create_aout_tables(char __user *p, struct linux_binprm *bprm)
223 {
224         u32 __user *argv, *envp, *sp;
225         int argc = bprm->argc, envc = bprm->envc;
226
227         sp = (u32 __user *) ((-(unsigned long)sizeof(u32)) & (unsigned long) p);
228         sp -= envc+1;
229         envp = sp;
230         sp -= argc+1;
231         argv = sp;
232         put_user((unsigned long) envp, --sp);
233         put_user((unsigned long) argv, --sp);
234         put_user(argc, --sp);
235         current->mm->arg_start = (unsigned long) p;
236         while (argc-- > 0) {
237                 char c;
238
239                 put_user((u32)(unsigned long)p, argv++);
240                 do {
241                         get_user(c, p++);
242                 } while (c);
243         }
244         put_user(0, argv);
245         current->mm->arg_end = current->mm->env_start = (unsigned long) p;
246         while (envc-- > 0) {
247                 char c;
248
249                 put_user((u32)(unsigned long)p, envp++);
250                 do {
251                         get_user(c, p++);
252                 } while (c);
253         }
254         put_user(0, envp);
255         current->mm->env_end = (unsigned long) p;
256         return sp;
257 }
258
259 /*
260  * These are the functions used to load a.out style executables and shared
261  * libraries.  There is no binary dependent code anywhere else.
262  */
263 static int load_aout_binary(struct linux_binprm *bprm, struct pt_regs *regs)
264 {
265         unsigned long error, fd_offset, rlim;
266         struct exec ex;
267         int retval;
268
269         ex = *((struct exec *) bprm->buf);              /* exec-header */
270         if ((N_MAGIC(ex) != ZMAGIC && N_MAGIC(ex) != OMAGIC &&
271              N_MAGIC(ex) != QMAGIC && N_MAGIC(ex) != NMAGIC) ||
272             N_TRSIZE(ex) || N_DRSIZE(ex) ||
273             i_size_read(bprm->file->f_path.dentry->d_inode) <
274             ex.a_text+ex.a_data+N_SYMSIZE(ex)+N_TXTOFF(ex)) {
275                 return -ENOEXEC;
276         }
277
278         fd_offset = N_TXTOFF(ex);
279
280         /* Check initial limits. This avoids letting people circumvent
281          * size limits imposed on them by creating programs with large
282          * arrays in the data or bss.
283          */
284         rlim = rlimit(RLIMIT_DATA);
285         if (rlim >= RLIM_INFINITY)
286                 rlim = ~0;
287         if (ex.a_data + ex.a_bss > rlim)
288                 return -ENOMEM;
289
290         /* Flush all traces of the currently running executable */
291         retval = flush_old_exec(bprm);
292         if (retval)
293                 return retval;
294
295         /* OK, This is the point of no return */
296         set_personality(PER_LINUX);
297         set_thread_flag(TIF_IA32);
298         current->mm->context.ia32_compat = 1;
299
300         setup_new_exec(bprm);
301
302         regs->cs = __USER32_CS;
303         regs->r8 = regs->r9 = regs->r10 = regs->r11 = regs->r12 =
304                 regs->r13 = regs->r14 = regs->r15 = 0;
305
306         current->mm->end_code = ex.a_text +
307                 (current->mm->start_code = N_TXTADDR(ex));
308         current->mm->end_data = ex.a_data +
309                 (current->mm->start_data = N_DATADDR(ex));
310         current->mm->brk = ex.a_bss +
311                 (current->mm->start_brk = N_BSSADDR(ex));
312         current->mm->free_area_cache = TASK_UNMAPPED_BASE;
313         current->mm->cached_hole_size = 0;
314
315         retval = setup_arg_pages(bprm, IA32_STACK_TOP, EXSTACK_DEFAULT);
316         if (retval < 0) {
317                 /* Someone check-me: is this error path enough? */
318                 send_sig(SIGKILL, current, 0);
319                 return retval;
320         }
321
322         install_exec_creds(bprm);
323
324         if (N_MAGIC(ex) == OMAGIC) {
325                 unsigned long text_addr, map_size;
326                 loff_t pos;
327
328                 text_addr = N_TXTADDR(ex);
329
330                 pos = 32;
331                 map_size = ex.a_text+ex.a_data;
332
333                 error = vm_brk(text_addr & PAGE_MASK, map_size);
334
335                 if (error != (text_addr & PAGE_MASK)) {
336                         send_sig(SIGKILL, current, 0);
337                         return error;
338                 }
339
340                 error = bprm->file->f_op->read(bprm->file,
341                          (char __user *)text_addr,
342                           ex.a_text+ex.a_data, &pos);
343                 if ((signed long)error < 0) {
344                         send_sig(SIGKILL, current, 0);
345                         return error;
346                 }
347
348                 flush_icache_range(text_addr, text_addr+ex.a_text+ex.a_data);
349         } else {
350 #ifdef WARN_OLD
351                 static unsigned long error_time, error_time2;
352                 if ((ex.a_text & 0xfff || ex.a_data & 0xfff) &&
353                     (N_MAGIC(ex) != NMAGIC) &&
354                                 time_after(jiffies, error_time2 + 5*HZ)) {
355                         printk(KERN_NOTICE "executable not page aligned\n");
356                         error_time2 = jiffies;
357                 }
358
359                 if ((fd_offset & ~PAGE_MASK) != 0 &&
360                             time_after(jiffies, error_time + 5*HZ)) {
361                         printk(KERN_WARNING
362                                "fd_offset is not page aligned. Please convert "
363                                "program: %s\n",
364                                bprm->file->f_path.dentry->d_name.name);
365                         error_time = jiffies;
366                 }
367 #endif
368
369                 if (!bprm->file->f_op->mmap || (fd_offset & ~PAGE_MASK) != 0) {
370                         loff_t pos = fd_offset;
371
372                         vm_brk(N_TXTADDR(ex), ex.a_text+ex.a_data);
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         current->mm->start_stack =
413                 (unsigned long)create_aout_tables((char __user *)bprm->p, bprm);
414         /* start thread */
415         loadsegment(fs, 0);
416         loadsegment(ds, __USER32_DS);
417         loadsegment(es, __USER32_DS);
418         load_gs_index(0);
419         (regs)->ip = ex.a_entry;
420         (regs)->sp = current->mm->start_stack;
421         (regs)->flags = 0x200;
422         (regs)->cs = __USER32_CS;
423         (regs)->ss = __USER32_DS;
424         regs->r8 = regs->r9 = regs->r10 = regs->r11 =
425         regs->r12 = regs->r13 = regs->r14 = regs->r15 = 0;
426         set_fs(USER_DS);
427         return 0;
428 }
429
430 static int load_aout_library(struct file *file)
431 {
432         struct inode *inode;
433         unsigned long bss, start_addr, len, error;
434         int retval;
435         struct exec ex;
436
437         inode = file->f_path.dentry->d_inode;
438
439         retval = -ENOEXEC;
440         error = kernel_read(file, 0, (char *) &ex, sizeof(ex));
441         if (error != sizeof(ex))
442                 goto out;
443
444         /* We come in here for the regular a.out style of shared libraries */
445         if ((N_MAGIC(ex) != ZMAGIC && N_MAGIC(ex) != QMAGIC) || N_TRSIZE(ex) ||
446             N_DRSIZE(ex) || ((ex.a_entry & 0xfff) && N_MAGIC(ex) == ZMAGIC) ||
447             i_size_read(inode) <
448             ex.a_text+ex.a_data+N_SYMSIZE(ex)+N_TXTOFF(ex)) {
449                 goto out;
450         }
451
452         if (N_FLAGS(ex))
453                 goto out;
454
455         /* For  QMAGIC, the starting address is 0x20 into the page.  We mask
456            this off to get the starting address for the page */
457
458         start_addr =  ex.a_entry & 0xfffff000;
459
460         if ((N_TXTOFF(ex) & ~PAGE_MASK) != 0) {
461                 loff_t pos = N_TXTOFF(ex);
462
463 #ifdef WARN_OLD
464                 static unsigned long error_time;
465                 if (time_after(jiffies, error_time + 5*HZ)) {
466                         printk(KERN_WARNING
467                                "N_TXTOFF is not page aligned. Please convert "
468                                "library: %s\n",
469                                file->f_path.dentry->d_name.name);
470                         error_time = jiffies;
471                 }
472 #endif
473                 vm_brk(start_addr, ex.a_text + ex.a_data + ex.a_bss);
474
475                 file->f_op->read(file, (char __user *)start_addr,
476                         ex.a_text + ex.a_data, &pos);
477                 flush_icache_range((unsigned long) start_addr,
478                                    (unsigned long) start_addr + ex.a_text +
479                                    ex.a_data);
480
481                 retval = 0;
482                 goto out;
483         }
484         /* Now use mmap to map the library into memory. */
485         down_write(&current->mm->mmap_sem);
486         error = do_mmap(file, start_addr, ex.a_text + ex.a_data,
487                         PROT_READ | PROT_WRITE | PROT_EXEC,
488                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE | MAP_32BIT,
489                         N_TXTOFF(ex));
490         up_write(&current->mm->mmap_sem);
491         retval = error;
492         if (error != start_addr)
493                 goto out;
494
495         len = PAGE_ALIGN(ex.a_text + ex.a_data);
496         bss = ex.a_text + ex.a_data + ex.a_bss;
497         if (bss > len) {
498                 error = vm_brk(start_addr + len, bss - len);
499                 retval = error;
500                 if (error != start_addr + len)
501                         goto out;
502         }
503         retval = 0;
504 out:
505         return retval;
506 }
507
508 static int __init init_aout_binfmt(void)
509 {
510         register_binfmt(&aout_format);
511         return 0;
512 }
513
514 static void __exit exit_aout_binfmt(void)
515 {
516         unregister_binfmt(&aout_format);
517 }
518
519 module_init(init_aout_binfmt);
520 module_exit(exit_aout_binfmt);
521 MODULE_LICENSE("GPL");