Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / 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 #include <linux/perf_event.h>
29
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
38 static int load_aout_binary(struct linux_binprm *);
39 static int load_aout_library(struct file *);
40
41 #ifdef CONFIG_COREDUMP
42 static int aout_core_dump(struct coredump_params *);
43
44 static unsigned long get_dr(int n)
45 {
46         struct perf_event *bp = current->thread.ptrace_bps[n];
47         return bp ? bp->hw.info.address : 0;
48 }
49
50 /*
51  * fill in the user structure for a core dump..
52  */
53 static void dump_thread32(struct pt_regs *regs, struct user32 *dump)
54 {
55         u32 fs, gs;
56         memset(dump, 0, sizeof(*dump));
57
58 /* changed the size calculations - should hopefully work better. lbt */
59         dump->magic = CMAGIC;
60         dump->start_code = 0;
61         dump->start_stack = regs->sp & ~(PAGE_SIZE - 1);
62         dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
63         dump->u_dsize = ((unsigned long)
64                          (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
65         dump->u_dsize -= dump->u_tsize;
66         dump->u_debugreg[0] = get_dr(0);
67         dump->u_debugreg[1] = get_dr(1);
68         dump->u_debugreg[2] = get_dr(2);
69         dump->u_debugreg[3] = get_dr(3);
70         dump->u_debugreg[6] = current->thread.debugreg6;
71         dump->u_debugreg[7] = current->thread.ptrace_dr7;
72
73         if (dump->start_stack < 0xc0000000) {
74                 unsigned long tmp;
75
76                 tmp = (unsigned long) (0xc0000000 - dump->start_stack);
77                 dump->u_ssize = tmp >> PAGE_SHIFT;
78         }
79
80         dump->regs.ebx = regs->bx;
81         dump->regs.ecx = regs->cx;
82         dump->regs.edx = regs->dx;
83         dump->regs.esi = regs->si;
84         dump->regs.edi = regs->di;
85         dump->regs.ebp = regs->bp;
86         dump->regs.eax = regs->ax;
87         dump->regs.ds = current->thread.ds;
88         dump->regs.es = current->thread.es;
89         savesegment(fs, fs);
90         dump->regs.fs = fs;
91         savesegment(gs, gs);
92         dump->regs.gs = gs;
93         dump->regs.orig_eax = regs->orig_ax;
94         dump->regs.eip = regs->ip;
95         dump->regs.cs = regs->cs;
96         dump->regs.eflags = regs->flags;
97         dump->regs.esp = regs->sp;
98         dump->regs.ss = regs->ss;
99
100 #if 1 /* FIXME */
101         dump->u_fpvalid = 0;
102 #else
103         dump->u_fpvalid = dump_fpu(regs, &dump->i387);
104 #endif
105 }
106
107 #endif
108
109 static struct linux_binfmt aout_format = {
110         .module         = THIS_MODULE,
111         .load_binary    = load_aout_binary,
112         .load_shlib     = load_aout_library,
113 #ifdef CONFIG_COREDUMP
114         .core_dump      = aout_core_dump,
115 #endif
116         .min_coredump   = PAGE_SIZE
117 };
118
119 static void set_brk(unsigned long start, unsigned long end)
120 {
121         start = PAGE_ALIGN(start);
122         end = PAGE_ALIGN(end);
123         if (end <= start)
124                 return;
125         vm_brk(start, end - start);
126 }
127
128 #ifdef CONFIG_COREDUMP
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 START_DATA(u)   (u.u_tsize << PAGE_SHIFT)
137 #define START_STACK(u)  (u.start_stack)
138
139 /*
140  * Routine writes a core dump image in the current directory.
141  * Currently only a stub-function.
142  *
143  * Note that setuid/setgid files won't make a core-dump if the uid/gid
144  * changed due to the set[u|g]id. It's enforced by the "current->mm->dumpable"
145  * field, which also makes sure the core-dumps won't be recursive if the
146  * dumping of the process results in another error..
147  */
148
149 static int aout_core_dump(struct coredump_params *cprm)
150 {
151         mm_segment_t fs;
152         int has_dumped = 0;
153         unsigned long dump_start, dump_size;
154         struct user32 dump;
155
156         fs = get_fs();
157         set_fs(KERNEL_DS);
158         has_dumped = 1;
159         strncpy(dump.u_comm, current->comm, sizeof(current->comm));
160         dump.u_ar0 = offsetof(struct user32, regs);
161         dump.signal = cprm->siginfo->si_signo;
162         dump_thread32(cprm->regs, &dump);
163
164         /*
165          * If the size of the dump file exceeds the rlimit, then see
166          * what would happen if we wrote the stack, but not the data
167          * area.
168          */
169         if ((dump.u_dsize + dump.u_ssize + 1) * PAGE_SIZE > cprm->limit)
170                 dump.u_dsize = 0;
171
172         /* Make sure we have enough room to write the stack and data areas. */
173         if ((dump.u_ssize + 1) * PAGE_SIZE > cprm->limit)
174                 dump.u_ssize = 0;
175
176         /* make sure we actually have a data and stack area to dump */
177         set_fs(USER_DS);
178         if (!access_ok(VERIFY_READ, (void *) (unsigned long)START_DATA(dump),
179                        dump.u_dsize << PAGE_SHIFT))
180                 dump.u_dsize = 0;
181         if (!access_ok(VERIFY_READ, (void *) (unsigned long)START_STACK(dump),
182                        dump.u_ssize << PAGE_SHIFT))
183                 dump.u_ssize = 0;
184
185         set_fs(KERNEL_DS);
186         /* struct user */
187         if (!dump_emit(cprm, &dump, sizeof(dump)))
188                 goto end_coredump;
189         /* Now dump all of the user data.  Include malloced stuff as well */
190         if (!dump_skip(cprm, PAGE_SIZE - sizeof(dump)))
191                 goto end_coredump;
192         /* now we start writing out the user space info */
193         set_fs(USER_DS);
194         /* Dump the data area */
195         if (dump.u_dsize != 0) {
196                 dump_start = START_DATA(dump);
197                 dump_size = dump.u_dsize << PAGE_SHIFT;
198                 if (!dump_emit(cprm, (void *)dump_start, dump_size))
199                         goto end_coredump;
200         }
201         /* Now prepare to dump the stack area */
202         if (dump.u_ssize != 0) {
203                 dump_start = START_STACK(dump);
204                 dump_size = dump.u_ssize << PAGE_SHIFT;
205                 if (!dump_emit(cprm, (void *)dump_start, dump_size))
206                         goto end_coredump;
207         }
208 end_coredump:
209         set_fs(fs);
210         return has_dumped;
211 }
212 #endif
213
214 /*
215  * create_aout_tables() parses the env- and arg-strings in new user
216  * memory and creates the pointer tables from them, and puts their
217  * addresses on the "stack", returning the new stack pointer value.
218  */
219 static u32 __user *create_aout_tables(char __user *p, struct linux_binprm *bprm)
220 {
221         u32 __user *argv, *envp, *sp;
222         int argc = bprm->argc, envc = bprm->envc;
223
224         sp = (u32 __user *) ((-(unsigned long)sizeof(u32)) & (unsigned long) p);
225         sp -= envc+1;
226         envp = sp;
227         sp -= argc+1;
228         argv = sp;
229         put_user((unsigned long) envp, --sp);
230         put_user((unsigned long) argv, --sp);
231         put_user(argc, --sp);
232         current->mm->arg_start = (unsigned long) p;
233         while (argc-- > 0) {
234                 char c;
235
236                 put_user((u32)(unsigned long)p, argv++);
237                 do {
238                         get_user(c, p++);
239                 } while (c);
240         }
241         put_user(0, argv);
242         current->mm->arg_end = current->mm->env_start = (unsigned long) p;
243         while (envc-- > 0) {
244                 char c;
245
246                 put_user((u32)(unsigned long)p, envp++);
247                 do {
248                         get_user(c, p++);
249                 } while (c);
250         }
251         put_user(0, envp);
252         current->mm->env_end = (unsigned long) p;
253         return sp;
254 }
255
256 /*
257  * These are the functions used to load a.out style executables and shared
258  * libraries.  There is no binary dependent code anywhere else.
259  */
260 static int load_aout_binary(struct linux_binprm *bprm)
261 {
262         unsigned long error, fd_offset, rlim;
263         struct pt_regs *regs = current_pt_regs();
264         struct exec ex;
265         int retval;
266
267         ex = *((struct exec *) bprm->buf);              /* exec-header */
268         if ((N_MAGIC(ex) != ZMAGIC && N_MAGIC(ex) != OMAGIC &&
269              N_MAGIC(ex) != QMAGIC && N_MAGIC(ex) != NMAGIC) ||
270             N_TRSIZE(ex) || N_DRSIZE(ex) ||
271             i_size_read(file_inode(bprm->file)) <
272             ex.a_text+ex.a_data+N_SYMSIZE(ex)+N_TXTOFF(ex)) {
273                 return -ENOEXEC;
274         }
275
276         fd_offset = N_TXTOFF(ex);
277
278         /* Check initial limits. This avoids letting people circumvent
279          * size limits imposed on them by creating programs with large
280          * arrays in the data or bss.
281          */
282         rlim = rlimit(RLIMIT_DATA);
283         if (rlim >= RLIM_INFINITY)
284                 rlim = ~0;
285         if (ex.a_data + ex.a_bss > rlim)
286                 return -ENOMEM;
287
288         /* Flush all traces of the currently running executable */
289         retval = flush_old_exec(bprm);
290         if (retval)
291                 return retval;
292
293         /* OK, This is the point of no return */
294         set_personality(PER_LINUX);
295         set_personality_ia32(false);
296
297         setup_new_exec(bprm);
298
299         regs->cs = __USER32_CS;
300         regs->r8 = regs->r9 = regs->r10 = regs->r11 = regs->r12 =
301                 regs->r13 = regs->r14 = regs->r15 = 0;
302
303         current->mm->end_code = ex.a_text +
304                 (current->mm->start_code = N_TXTADDR(ex));
305         current->mm->end_data = ex.a_data +
306                 (current->mm->start_data = N_DATADDR(ex));
307         current->mm->brk = ex.a_bss +
308                 (current->mm->start_brk = N_BSSADDR(ex));
309
310         retval = setup_arg_pages(bprm, IA32_STACK_TOP, EXSTACK_DEFAULT);
311         if (retval < 0)
312                 return retval;
313
314         install_exec_creds(bprm);
315
316         if (N_MAGIC(ex) == OMAGIC) {
317                 unsigned long text_addr, map_size;
318
319                 text_addr = N_TXTADDR(ex);
320                 map_size = ex.a_text+ex.a_data;
321
322                 error = vm_brk(text_addr & PAGE_MASK, map_size);
323
324                 if (error != (text_addr & PAGE_MASK))
325                         return error;
326
327                 error = read_code(bprm->file, text_addr, 32,
328                                   ex.a_text + ex.a_data);
329                 if ((signed long)error < 0)
330                         return error;
331         } else {
332 #ifdef WARN_OLD
333                 static unsigned long error_time, error_time2;
334                 if ((ex.a_text & 0xfff || ex.a_data & 0xfff) &&
335                     (N_MAGIC(ex) != NMAGIC) &&
336                                 time_after(jiffies, error_time2 + 5*HZ)) {
337                         printk(KERN_NOTICE "executable not page aligned\n");
338                         error_time2 = jiffies;
339                 }
340
341                 if ((fd_offset & ~PAGE_MASK) != 0 &&
342                             time_after(jiffies, error_time + 5*HZ)) {
343                         printk(KERN_WARNING
344                                "fd_offset is not page aligned. Please convert "
345                                "program: %pD\n",
346                                bprm->file);
347                         error_time = jiffies;
348                 }
349 #endif
350
351                 if (!bprm->file->f_op->mmap || (fd_offset & ~PAGE_MASK) != 0) {
352                         vm_brk(N_TXTADDR(ex), ex.a_text+ex.a_data);
353                         read_code(bprm->file, N_TXTADDR(ex), fd_offset,
354                                         ex.a_text+ex.a_data);
355                         goto beyond_if;
356                 }
357
358                 error = vm_mmap(bprm->file, N_TXTADDR(ex), ex.a_text,
359                                 PROT_READ | PROT_EXEC,
360                                 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE |
361                                 MAP_EXECUTABLE | MAP_32BIT,
362                                 fd_offset);
363
364                 if (error != N_TXTADDR(ex))
365                         return error;
366
367                 error = vm_mmap(bprm->file, N_DATADDR(ex), ex.a_data,
368                                 PROT_READ | PROT_WRITE | PROT_EXEC,
369                                 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE |
370                                 MAP_EXECUTABLE | MAP_32BIT,
371                                 fd_offset + ex.a_text);
372                 if (error != N_DATADDR(ex))
373                         return error;
374         }
375 beyond_if:
376         set_binfmt(&aout_format);
377
378         set_brk(current->mm->start_brk, current->mm->brk);
379
380         current->mm->start_stack =
381                 (unsigned long)create_aout_tables((char __user *)bprm->p, bprm);
382         /* start thread */
383         loadsegment(fs, 0);
384         loadsegment(ds, __USER32_DS);
385         loadsegment(es, __USER32_DS);
386         load_gs_index(0);
387         (regs)->ip = ex.a_entry;
388         (regs)->sp = current->mm->start_stack;
389         (regs)->flags = 0x200;
390         (regs)->cs = __USER32_CS;
391         (regs)->ss = __USER32_DS;
392         regs->r8 = regs->r9 = regs->r10 = regs->r11 =
393         regs->r12 = regs->r13 = regs->r14 = regs->r15 = 0;
394         set_fs(USER_DS);
395         return 0;
396 }
397
398 static int load_aout_library(struct file *file)
399 {
400         unsigned long bss, start_addr, len, error;
401         int retval;
402         struct exec ex;
403
404
405         retval = -ENOEXEC;
406         error = kernel_read(file, 0, (char *) &ex, sizeof(ex));
407         if (error != sizeof(ex))
408                 goto out;
409
410         /* We come in here for the regular a.out style of shared libraries */
411         if ((N_MAGIC(ex) != ZMAGIC && N_MAGIC(ex) != QMAGIC) || N_TRSIZE(ex) ||
412             N_DRSIZE(ex) || ((ex.a_entry & 0xfff) && N_MAGIC(ex) == ZMAGIC) ||
413             i_size_read(file_inode(file)) <
414             ex.a_text+ex.a_data+N_SYMSIZE(ex)+N_TXTOFF(ex)) {
415                 goto out;
416         }
417
418         if (N_FLAGS(ex))
419                 goto out;
420
421         /* For  QMAGIC, the starting address is 0x20 into the page.  We mask
422            this off to get the starting address for the page */
423
424         start_addr =  ex.a_entry & 0xfffff000;
425
426         if ((N_TXTOFF(ex) & ~PAGE_MASK) != 0) {
427 #ifdef WARN_OLD
428                 static unsigned long error_time;
429                 if (time_after(jiffies, error_time + 5*HZ)) {
430                         printk(KERN_WARNING
431                                "N_TXTOFF is not page aligned. Please convert "
432                                "library: %pD\n",
433                                file);
434                         error_time = jiffies;
435                 }
436 #endif
437                 vm_brk(start_addr, ex.a_text + ex.a_data + ex.a_bss);
438
439                 read_code(file, start_addr, N_TXTOFF(ex),
440                           ex.a_text + ex.a_data);
441                 retval = 0;
442                 goto out;
443         }
444         /* Now use mmap to map the library into memory. */
445         error = vm_mmap(file, start_addr, ex.a_text + ex.a_data,
446                         PROT_READ | PROT_WRITE | PROT_EXEC,
447                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE | MAP_32BIT,
448                         N_TXTOFF(ex));
449         retval = error;
450         if (error != start_addr)
451                 goto out;
452
453         len = PAGE_ALIGN(ex.a_text + ex.a_data);
454         bss = ex.a_text + ex.a_data + ex.a_bss;
455         if (bss > len) {
456                 error = vm_brk(start_addr + len, bss - len);
457                 retval = error;
458                 if (error != start_addr + len)
459                         goto out;
460         }
461         retval = 0;
462 out:
463         return retval;
464 }
465
466 static int __init init_aout_binfmt(void)
467 {
468         register_binfmt(&aout_format);
469         return 0;
470 }
471
472 static void __exit exit_aout_binfmt(void)
473 {
474         unregister_binfmt(&aout_format);
475 }
476
477 module_init(init_aout_binfmt);
478 module_exit(exit_aout_binfmt);
479 MODULE_LICENSE("GPL");