target: Fix bug in handling of FILEIO + block_device resize ops
[linux-flexiantxendom0-3.2.10.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         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 #include <linux/coredump.h>
134
135 #define DUMP_WRITE(addr, nr)                         \
136         if (!dump_write(file, (void *)(addr), (nr))) \
137                 goto end_coredump;
138
139 #define DUMP_SEEK(offset)               \
140         if (!dump_seek(file, offset))   \
141                 goto end_coredump;
142
143 #define START_DATA()    (u.u_tsize << PAGE_SHIFT)
144 #define START_STACK(u)  (u.start_stack)
145
146 /*
147  * Routine writes a core dump image in the current directory.
148  * Currently only a stub-function.
149  *
150  * Note that setuid/setgid files won't make a core-dump if the uid/gid
151  * changed due to the set[u|g]id. It's enforced by the "current->mm->dumpable"
152  * field, which also makes sure the core-dumps won't be recursive if the
153  * dumping of the process results in another error..
154  */
155
156 static int aout_core_dump(long signr, struct pt_regs *regs, struct file *file,
157                           unsigned long limit)
158 {
159         mm_segment_t fs;
160         int has_dumped = 0;
161         unsigned long dump_start, dump_size;
162         struct user32 dump;
163
164         fs = get_fs();
165         set_fs(KERNEL_DS);
166         has_dumped = 1;
167         current->flags |= PF_DUMPCORE;
168         strncpy(dump.u_comm, current->comm, sizeof(current->comm));
169         dump.u_ar0 = offsetof(struct user32, regs);
170         dump.signal = signr;
171         dump_thread32(regs, &dump);
172
173         /*
174          * If the size of the dump file exceeds the rlimit, then see
175          * what would happen if we wrote the stack, but not the data
176          * area.
177          */
178         if ((dump.u_dsize + dump.u_ssize + 1) * PAGE_SIZE > limit)
179                 dump.u_dsize = 0;
180
181         /* Make sure we have enough room to write the stack and data areas. */
182         if ((dump.u_ssize + 1) * PAGE_SIZE > limit)
183                 dump.u_ssize = 0;
184
185         /* make sure we actually have a data and stack area to dump */
186         set_fs(USER_DS);
187         if (!access_ok(VERIFY_READ, (void *) (unsigned long)START_DATA(dump),
188                        dump.u_dsize << PAGE_SHIFT))
189                 dump.u_dsize = 0;
190         if (!access_ok(VERIFY_READ, (void *) (unsigned long)START_STACK(dump),
191                        dump.u_ssize << PAGE_SHIFT))
192                 dump.u_ssize = 0;
193
194         set_fs(KERNEL_DS);
195         /* struct user */
196         DUMP_WRITE(&dump, sizeof(dump));
197         /* Now dump all of the user data.  Include malloced stuff as well */
198         DUMP_SEEK(PAGE_SIZE);
199         /* now we start writing out the user space info */
200         set_fs(USER_DS);
201         /* Dump the data area */
202         if (dump.u_dsize != 0) {
203                 dump_start = START_DATA(dump);
204                 dump_size = dump.u_dsize << PAGE_SHIFT;
205                 DUMP_WRITE(dump_start, dump_size);
206         }
207         /* Now prepare to dump the stack area */
208         if (dump.u_ssize != 0) {
209                 dump_start = START_STACK(dump);
210                 dump_size = dump.u_ssize << PAGE_SHIFT;
211                 DUMP_WRITE(dump_start, dump_size);
212         }
213 end_coredump:
214         set_fs(fs);
215         return has_dumped;
216 }
217 #endif
218
219 /*
220  * create_aout_tables() parses the env- and arg-strings in new user
221  * memory and creates the pointer tables from them, and puts their
222  * addresses on the "stack", returning the new stack pointer value.
223  */
224 static u32 __user *create_aout_tables(char __user *p, struct linux_binprm *bprm)
225 {
226         u32 __user *argv, *envp, *sp;
227         int argc = bprm->argc, envc = bprm->envc;
228
229         sp = (u32 __user *) ((-(unsigned long)sizeof(u32)) & (unsigned long) p);
230         sp -= envc+1;
231         envp = sp;
232         sp -= argc+1;
233         argv = sp;
234         put_user((unsigned long) envp, --sp);
235         put_user((unsigned long) argv, --sp);
236         put_user(argc, --sp);
237         current->mm->arg_start = (unsigned long) p;
238         while (argc-- > 0) {
239                 char c;
240
241                 put_user((u32)(unsigned long)p, argv++);
242                 do {
243                         get_user(c, p++);
244                 } while (c);
245         }
246         put_user(0, argv);
247         current->mm->arg_end = current->mm->env_start = (unsigned long) p;
248         while (envc-- > 0) {
249                 char c;
250
251                 put_user((u32)(unsigned long)p, envp++);
252                 do {
253                         get_user(c, p++);
254                 } while (c);
255         }
256         put_user(0, envp);
257         current->mm->env_end = (unsigned long) p;
258         return sp;
259 }
260
261 /*
262  * These are the functions used to load a.out style executables and shared
263  * libraries.  There is no binary dependent code anywhere else.
264  */
265 static int load_aout_binary(struct linux_binprm *bprm, struct pt_regs *regs)
266 {
267         unsigned long error, fd_offset, rlim;
268         struct exec ex;
269         int retval;
270
271         ex = *((struct exec *) bprm->buf);              /* exec-header */
272         if ((N_MAGIC(ex) != ZMAGIC && N_MAGIC(ex) != OMAGIC &&
273              N_MAGIC(ex) != QMAGIC && N_MAGIC(ex) != NMAGIC) ||
274             N_TRSIZE(ex) || N_DRSIZE(ex) ||
275             i_size_read(bprm->file->f_path.dentry->d_inode) <
276             ex.a_text+ex.a_data+N_SYMSIZE(ex)+N_TXTOFF(ex)) {
277                 return -ENOEXEC;
278         }
279
280         fd_offset = N_TXTOFF(ex);
281
282         /* Check initial limits. This avoids letting people circumvent
283          * size limits imposed on them by creating programs with large
284          * arrays in the data or bss.
285          */
286         rlim = rlimit(RLIMIT_DATA);
287         if (rlim >= RLIM_INFINITY)
288                 rlim = ~0;
289         if (ex.a_data + ex.a_bss > rlim)
290                 return -ENOMEM;
291
292         /* Flush all traces of the currently running executable */
293         retval = flush_old_exec(bprm);
294         if (retval)
295                 return retval;
296
297         /* OK, This is the point of no return */
298         set_personality(PER_LINUX);
299         set_thread_flag(TIF_IA32);
300         current->mm->context.ia32_compat = 1;
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         retval = setup_arg_pages(bprm, IA32_STACK_TOP, EXSTACK_DEFAULT);
318         if (retval < 0) {
319                 /* Someone check-me: is this error path enough? */
320                 send_sig(SIGKILL, current, 0);
321                 return retval;
322         }
323
324         install_exec_creds(bprm);
325
326         if (N_MAGIC(ex) == OMAGIC) {
327                 unsigned long text_addr, map_size;
328                 loff_t pos;
329
330                 text_addr = N_TXTADDR(ex);
331
332                 pos = 32;
333                 map_size = ex.a_text+ex.a_data;
334
335                 down_write(&current->mm->mmap_sem);
336                 error = do_brk(text_addr & PAGE_MASK, map_size);
337                 up_write(&current->mm->mmap_sem);
338
339                 if (error != (text_addr & PAGE_MASK)) {
340                         send_sig(SIGKILL, current, 0);
341                         return error;
342                 }
343
344                 error = bprm->file->f_op->read(bprm->file,
345                          (char __user *)text_addr,
346                           ex.a_text+ex.a_data, &pos);
347                 if ((signed long)error < 0) {
348                         send_sig(SIGKILL, current, 0);
349                         return error;
350                 }
351
352                 flush_icache_range(text_addr, text_addr+ex.a_text+ex.a_data);
353         } else {
354 #ifdef WARN_OLD
355                 static unsigned long error_time, error_time2;
356                 if ((ex.a_text & 0xfff || ex.a_data & 0xfff) &&
357                     (N_MAGIC(ex) != NMAGIC) &&
358                                 time_after(jiffies, error_time2 + 5*HZ)) {
359                         printk(KERN_NOTICE "executable not page aligned\n");
360                         error_time2 = jiffies;
361                 }
362
363                 if ((fd_offset & ~PAGE_MASK) != 0 &&
364                             time_after(jiffies, error_time + 5*HZ)) {
365                         printk(KERN_WARNING
366                                "fd_offset is not page aligned. Please convert "
367                                "program: %s\n",
368                                bprm->file->f_path.dentry->d_name.name);
369                         error_time = jiffies;
370                 }
371 #endif
372
373                 if (!bprm->file->f_op->mmap || (fd_offset & ~PAGE_MASK) != 0) {
374                         loff_t pos = fd_offset;
375
376                         down_write(&current->mm->mmap_sem);
377                         do_brk(N_TXTADDR(ex), ex.a_text+ex.a_data);
378                         up_write(&current->mm->mmap_sem);
379                         bprm->file->f_op->read(bprm->file,
380                                         (char __user *)N_TXTADDR(ex),
381                                         ex.a_text+ex.a_data, &pos);
382                         flush_icache_range((unsigned long) N_TXTADDR(ex),
383                                            (unsigned long) N_TXTADDR(ex) +
384                                            ex.a_text+ex.a_data);
385                         goto beyond_if;
386                 }
387
388                 down_write(&current->mm->mmap_sem);
389                 error = do_mmap(bprm->file, N_TXTADDR(ex), ex.a_text,
390                                 PROT_READ | PROT_EXEC,
391                                 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE |
392                                 MAP_EXECUTABLE | MAP_32BIT,
393                                 fd_offset);
394                 up_write(&current->mm->mmap_sem);
395
396                 if (error != N_TXTADDR(ex)) {
397                         send_sig(SIGKILL, current, 0);
398                         return error;
399                 }
400
401                 down_write(&current->mm->mmap_sem);
402                 error = do_mmap(bprm->file, N_DATADDR(ex), ex.a_data,
403                                 PROT_READ | PROT_WRITE | PROT_EXEC,
404                                 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE |
405                                 MAP_EXECUTABLE | MAP_32BIT,
406                                 fd_offset + ex.a_text);
407                 up_write(&current->mm->mmap_sem);
408                 if (error != N_DATADDR(ex)) {
409                         send_sig(SIGKILL, current, 0);
410                         return error;
411                 }
412         }
413 beyond_if:
414         set_binfmt(&aout_format);
415
416         set_brk(current->mm->start_brk, current->mm->brk);
417
418         current->mm->start_stack =
419                 (unsigned long)create_aout_tables((char __user *)bprm->p, bprm);
420         /* start thread */
421         loadsegment(fs, 0);
422         loadsegment(ds, __USER32_DS);
423         loadsegment(es, __USER32_DS);
424         load_gs_index(0);
425         (regs)->ip = ex.a_entry;
426         (regs)->sp = current->mm->start_stack;
427         (regs)->flags = 0x200;
428         (regs)->cs = __USER32_CS;
429         (regs)->ss = __USER32_DS;
430         regs->r8 = regs->r9 = regs->r10 = regs->r11 =
431         regs->r12 = regs->r13 = regs->r14 = regs->r15 = 0;
432         set_fs(USER_DS);
433         return 0;
434 }
435
436 static int load_aout_library(struct file *file)
437 {
438         struct inode *inode;
439         unsigned long bss, start_addr, len, error;
440         int retval;
441         struct exec ex;
442
443         inode = file->f_path.dentry->d_inode;
444
445         retval = -ENOEXEC;
446         error = kernel_read(file, 0, (char *) &ex, sizeof(ex));
447         if (error != sizeof(ex))
448                 goto out;
449
450         /* We come in here for the regular a.out style of shared libraries */
451         if ((N_MAGIC(ex) != ZMAGIC && N_MAGIC(ex) != QMAGIC) || N_TRSIZE(ex) ||
452             N_DRSIZE(ex) || ((ex.a_entry & 0xfff) && N_MAGIC(ex) == ZMAGIC) ||
453             i_size_read(inode) <
454             ex.a_text+ex.a_data+N_SYMSIZE(ex)+N_TXTOFF(ex)) {
455                 goto out;
456         }
457
458         if (N_FLAGS(ex))
459                 goto out;
460
461         /* For  QMAGIC, the starting address is 0x20 into the page.  We mask
462            this off to get the starting address for the page */
463
464         start_addr =  ex.a_entry & 0xfffff000;
465
466         if ((N_TXTOFF(ex) & ~PAGE_MASK) != 0) {
467                 loff_t pos = N_TXTOFF(ex);
468
469 #ifdef WARN_OLD
470                 static unsigned long error_time;
471                 if (time_after(jiffies, error_time + 5*HZ)) {
472                         printk(KERN_WARNING
473                                "N_TXTOFF is not page aligned. Please convert "
474                                "library: %s\n",
475                                file->f_path.dentry->d_name.name);
476                         error_time = jiffies;
477                 }
478 #endif
479                 down_write(&current->mm->mmap_sem);
480                 do_brk(start_addr, ex.a_text + ex.a_data + ex.a_bss);
481                 up_write(&current->mm->mmap_sem);
482
483                 file->f_op->read(file, (char __user *)start_addr,
484                         ex.a_text + ex.a_data, &pos);
485                 flush_icache_range((unsigned long) start_addr,
486                                    (unsigned long) start_addr + ex.a_text +
487                                    ex.a_data);
488
489                 retval = 0;
490                 goto out;
491         }
492         /* Now use mmap to map the library into memory. */
493         down_write(&current->mm->mmap_sem);
494         error = do_mmap(file, start_addr, ex.a_text + ex.a_data,
495                         PROT_READ | PROT_WRITE | PROT_EXEC,
496                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE | MAP_32BIT,
497                         N_TXTOFF(ex));
498         up_write(&current->mm->mmap_sem);
499         retval = error;
500         if (error != start_addr)
501                 goto out;
502
503         len = PAGE_ALIGN(ex.a_text + ex.a_data);
504         bss = ex.a_text + ex.a_data + ex.a_bss;
505         if (bss > len) {
506                 down_write(&current->mm->mmap_sem);
507                 error = do_brk(start_addr + len, bss - len);
508                 up_write(&current->mm->mmap_sem);
509                 retval = error;
510                 if (error != start_addr + len)
511                         goto out;
512         }
513         retval = 0;
514 out:
515         return retval;
516 }
517
518 static int __init init_aout_binfmt(void)
519 {
520         register_binfmt(&aout_format);
521         return 0;
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");