[PATCH] mm: update comments to pte lock
[linux-2.6.git] / arch / x86_64 / 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
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 extern int ia32_setup_arg_pages(struct linux_binprm *bprm,
40                                 unsigned long stack_top, int exec_stack);
41
42 static int load_aout_binary(struct linux_binprm *, struct pt_regs * regs);
43 static int load_aout_library(struct file*);
44
45 #ifdef CORE_DUMP
46 static int aout_core_dump(long signr, struct pt_regs * regs, struct file *file);
47
48 /*
49  * fill in the user structure for a core dump..
50  */
51 static void dump_thread32(struct pt_regs * regs, struct user32 * dump)
52 {
53         u32 fs,gs;
54
55 /* changed the size calculations - should hopefully work better. lbt */
56         dump->magic = CMAGIC;
57         dump->start_code = 0;
58         dump->start_stack = regs->rsp & ~(PAGE_SIZE - 1);
59         dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
60         dump->u_dsize = ((unsigned long) (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                 dump->u_ssize = ((unsigned long) (0xc0000000 - dump->start_stack)) >> PAGE_SHIFT;
74
75         dump->regs.ebx = regs->rbx;
76         dump->regs.ecx = regs->rcx;
77         dump->regs.edx = regs->rdx;
78         dump->regs.esi = regs->rsi;
79         dump->regs.edi = regs->rdi;
80         dump->regs.ebp = regs->rbp;
81         dump->regs.eax = regs->rax;
82         dump->regs.ds = current->thread.ds;
83         dump->regs.es = current->thread.es;
84         asm("movl %%fs,%0" : "=r" (fs)); dump->regs.fs = fs;
85         asm("movl %%gs,%0" : "=r" (gs)); dump->regs.gs = gs; 
86         dump->regs.orig_eax = regs->orig_rax;
87         dump->regs.eip = regs->rip;
88         dump->regs.cs = regs->cs;
89         dump->regs.eflags = regs->eflags;
90         dump->regs.esp = regs->rsp;
91         dump->regs.ss = regs->ss;
92
93 #if 1 /* FIXME */
94         dump->u_fpvalid = 0;
95 #else
96         dump->u_fpvalid = dump_fpu (regs, &dump->i387);
97 #endif
98 }
99
100 #endif
101
102 static struct linux_binfmt aout_format = {
103         .module         = THIS_MODULE,
104         .load_binary    = load_aout_binary,
105         .load_shlib     = load_aout_library,
106 #ifdef CORE_DUMP
107         .core_dump      = aout_core_dump,
108 #endif
109         .min_coredump   = PAGE_SIZE
110 };
111
112 static void set_brk(unsigned long start, unsigned long end)
113 {
114         start = PAGE_ALIGN(start);
115         end = PAGE_ALIGN(end);
116         if (end <= start)
117                 return;
118         down_write(&current->mm->mmap_sem);
119         do_brk(start, end - start);
120         up_write(&current->mm->mmap_sem);
121 }
122
123 #ifdef CORE_DUMP
124 /*
125  * These are the only things you should do on a core-file: use only these
126  * macros to write out all the necessary info.
127  */
128
129 static int dump_write(struct file *file, const void *addr, int nr)
130 {
131         return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
132 }
133
134 #define DUMP_WRITE(addr, nr)    \
135         if (!dump_write(file, (void *)(addr), (nr))) \
136                 goto end_coredump;
137
138 #define DUMP_SEEK(offset) \
139 if (file->f_op->llseek) { \
140         if (file->f_op->llseek(file,(offset),0) != (offset)) \
141                 goto end_coredump; \
142 } else file->f_pos = (offset)
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 {
156         mm_segment_t fs;
157         int has_dumped = 0;
158         unsigned long dump_start, dump_size;
159         struct user32 dump;
160 #       define START_DATA(u)    (u.u_tsize << PAGE_SHIFT)
161 #       define START_STACK(u)   (u.start_stack)
162
163         fs = get_fs();
164         set_fs(KERNEL_DS);
165         has_dumped = 1;
166         current->flags |= PF_DUMPCORE;
167         strncpy(dump.u_comm, current->comm, sizeof(current->comm));
168         dump.u_ar0 = (u32)(((unsigned long)(&dump.regs)) - ((unsigned long)(&dump)));
169         dump.signal = signr;
170         dump_thread32(regs, &dump);
171
172 /* If the size of the dump file exceeds the rlimit, then see what would happen
173    if we wrote the stack, but not the data area.  */
174         if ((dump.u_dsize+dump.u_ssize+1) * PAGE_SIZE >
175             current->signal->rlim[RLIMIT_CORE].rlim_cur)
176                 dump.u_dsize = 0;
177
178 /* Make sure we have enough room to write the stack and data areas. */
179         if ((dump.u_ssize+1) * PAGE_SIZE >
180             current->signal->rlim[RLIMIT_CORE].rlim_cur)
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), dump.u_dsize << PAGE_SHIFT))
186                 dump.u_dsize = 0;
187         if (!access_ok(VERIFY_READ, (void *) (unsigned long)START_STACK(dump), dump.u_ssize << PAGE_SHIFT))
188                 dump.u_ssize = 0;
189
190         set_fs(KERNEL_DS);
191 /* struct user */
192         DUMP_WRITE(&dump,sizeof(dump));
193 /* Now dump all of the user data.  Include malloced stuff as well */
194         DUMP_SEEK(PAGE_SIZE);
195 /* now we start writing out the user space info */
196         set_fs(USER_DS);
197 /* Dump the data area */
198         if (dump.u_dsize != 0) {
199                 dump_start = START_DATA(dump);
200                 dump_size = dump.u_dsize << PAGE_SHIFT;
201                 DUMP_WRITE(dump_start,dump_size);
202         }
203 /* Now prepare to dump the stack area */
204         if (dump.u_ssize != 0) {
205                 dump_start = START_STACK(dump);
206                 dump_size = dump.u_ssize << PAGE_SHIFT;
207                 DUMP_WRITE(dump_start,dump_size);
208         }
209 /* Finally dump the task struct.  Not be used by gdb, but could be useful */
210         set_fs(KERNEL_DS);
211         DUMP_WRITE(current,sizeof(*current));
212 end_coredump:
213         set_fs(fs);
214         return has_dumped;
215 }
216 #endif
217
218 /*
219  * create_aout_tables() parses the env- and arg-strings in new user
220  * memory and creates the pointer tables from them, and puts their
221  * addresses on the "stack", returning the new stack pointer value.
222  */
223 static u32 __user *create_aout_tables(char __user *p, struct linux_binprm *bprm)
224 {
225         u32 __user *argv;
226         u32 __user *envp;
227         u32 __user *sp;
228         int argc = bprm->argc;
229         int envc = bprm->envc;
230
231         sp = (u32 __user *) ((-(unsigned long)sizeof(u32)) & (unsigned long) p);
232         sp -= envc+1;
233         envp = sp;
234         sp -= argc+1;
235         argv = sp;
236         put_user((unsigned long) envp,--sp);
237         put_user((unsigned long) argv,--sp);
238         put_user(argc,--sp);
239         current->mm->arg_start = (unsigned long) p;
240         while (argc-->0) {
241                 char c;
242                 put_user((u32)(unsigned long)p,argv++);
243                 do {
244                         get_user(c,p++);
245                 } while (c);
246         }
247         put_user(NULL,argv);
248         current->mm->arg_end = current->mm->env_start = (unsigned long) p;
249         while (envc-->0) {
250                 char c;
251                 put_user((u32)(unsigned long)p,envp++);
252                 do {
253                         get_user(c,p++);
254                 } while (c);
255         }
256         put_user(NULL,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
266 static int load_aout_binary(struct linux_binprm * bprm, struct pt_regs * regs)
267 {
268         struct exec ex;
269         unsigned long error;
270         unsigned long fd_offset;
271         unsigned long rlim;
272         int retval;
273
274         ex = *((struct exec *) bprm->buf);              /* exec-header */
275         if ((N_MAGIC(ex) != ZMAGIC && N_MAGIC(ex) != OMAGIC &&
276              N_MAGIC(ex) != QMAGIC && N_MAGIC(ex) != NMAGIC) ||
277             N_TRSIZE(ex) || N_DRSIZE(ex) ||
278             i_size_read(bprm->file->f_dentry->d_inode) < ex.a_text+ex.a_data+N_SYMSIZE(ex)+N_TXTOFF(ex)) {
279                 return -ENOEXEC;
280         }
281
282         fd_offset = N_TXTOFF(ex);
283
284         /* Check initial limits. This avoids letting people circumvent
285          * size limits imposed on them by creating programs with large
286          * arrays in the data or bss.
287          */
288         rlim = current->signal->rlim[RLIMIT_DATA].rlim_cur;
289         if (rlim >= RLIM_INFINITY)
290                 rlim = ~0;
291         if (ex.a_data + ex.a_bss > rlim)
292                 return -ENOMEM;
293
294         /* Flush all traces of the currently running executable */
295         retval = flush_old_exec(bprm);
296         if (retval)
297                 return retval;
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         /* OK, This is the point of no return */
304         set_personality(PER_LINUX);
305         set_thread_flag(TIF_IA32); 
306         clear_thread_flag(TIF_ABI_PENDING);
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         current->mm->mmap = NULL;
318         compute_creds(bprm);
319         current->flags &= ~PF_FORKNOEXEC;
320
321         if (N_MAGIC(ex) == OMAGIC) {
322                 unsigned long text_addr, map_size;
323                 loff_t pos;
324
325                 text_addr = N_TXTADDR(ex);
326
327                 pos = 32;
328                 map_size = ex.a_text+ex.a_data;
329
330                 down_write(&current->mm->mmap_sem);
331                 error = do_brk(text_addr & PAGE_MASK, map_size);
332                 up_write(&current->mm->mmap_sem);
333
334                 if (error != (text_addr & PAGE_MASK)) {
335                         send_sig(SIGKILL, current, 0);
336                         return error;
337                 }
338
339                 error = bprm->file->f_op->read(bprm->file, (char *)text_addr,
340                           ex.a_text+ex.a_data, &pos);
341                 if ((signed long)error < 0) {
342                         send_sig(SIGKILL, current, 0);
343                         return error;
344                 }
345                          
346                 flush_icache_range(text_addr, text_addr+ex.a_text+ex.a_data);
347         } else {
348 #ifdef WARN_OLD
349                 static unsigned long error_time, error_time2;
350                 if ((ex.a_text & 0xfff || ex.a_data & 0xfff) &&
351                     (N_MAGIC(ex) != NMAGIC) && (jiffies-error_time2) > 5*HZ)
352                 {
353                         printk(KERN_NOTICE "executable not page aligned\n");
354                         error_time2 = jiffies;
355                 }
356
357                 if ((fd_offset & ~PAGE_MASK) != 0 &&
358                     (jiffies-error_time) > 5*HZ)
359                 {
360                         printk(KERN_WARNING 
361                                "fd_offset is not page aligned. Please convert program: %s\n",
362                                bprm->file->f_dentry->d_name.name);
363                         error_time = jiffies;
364                 }
365 #endif
366
367                 if (!bprm->file->f_op->mmap||((fd_offset & ~PAGE_MASK) != 0)) {
368                         loff_t pos = fd_offset;
369                         down_write(&current->mm->mmap_sem);
370                         do_brk(N_TXTADDR(ex), ex.a_text+ex.a_data);
371                         up_write(&current->mm->mmap_sem);
372                         bprm->file->f_op->read(bprm->file,(char *)N_TXTADDR(ex),
373                                         ex.a_text+ex.a_data, &pos);
374                         flush_icache_range((unsigned long) N_TXTADDR(ex),
375                                            (unsigned long) N_TXTADDR(ex) +
376                                            ex.a_text+ex.a_data);
377                         goto beyond_if;
378                 }
379
380                 down_write(&current->mm->mmap_sem);
381                 error = do_mmap(bprm->file, N_TXTADDR(ex), ex.a_text,
382                         PROT_READ | PROT_EXEC,
383                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE | MAP_32BIT,
384                         fd_offset);
385                 up_write(&current->mm->mmap_sem);
386
387                 if (error != N_TXTADDR(ex)) {
388                         send_sig(SIGKILL, current, 0);
389                         return error;
390                 }
391
392                 down_write(&current->mm->mmap_sem);
393                 error = do_mmap(bprm->file, N_DATADDR(ex), ex.a_data,
394                                 PROT_READ | PROT_WRITE | PROT_EXEC,
395                                 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE | MAP_32BIT,
396                                 fd_offset + ex.a_text);
397                 up_write(&current->mm->mmap_sem);
398                 if (error != N_DATADDR(ex)) {
399                         send_sig(SIGKILL, current, 0);
400                         return error;
401                 }
402         }
403 beyond_if:
404         set_binfmt(&aout_format);
405
406         set_brk(current->mm->start_brk, current->mm->brk);
407
408         retval = ia32_setup_arg_pages(bprm, IA32_STACK_TOP, EXSTACK_DEFAULT);
409         if (retval < 0) { 
410                 /* Someone check-me: is this error path enough? */ 
411                 send_sig(SIGKILL, current, 0); 
412                 return retval;
413         }
414
415         current->mm->start_stack =
416                 (unsigned long)create_aout_tables((char __user *)bprm->p, bprm);
417         /* start thread */
418         asm volatile("movl %0,%%fs" :: "r" (0)); \
419         asm volatile("movl %0,%%es; movl %0,%%ds": :"r" (__USER32_DS));
420         load_gs_index(0); 
421         (regs)->rip = ex.a_entry;
422         (regs)->rsp = current->mm->start_stack;
423         (regs)->eflags = 0x200;
424         (regs)->cs = __USER32_CS;
425         (regs)->ss = __USER32_DS;
426         set_fs(USER_DS);
427         if (unlikely(current->ptrace & PT_PTRACED)) {
428                 if (current->ptrace & PT_TRACE_EXEC)
429                         ptrace_notify ((PTRACE_EVENT_EXEC << 8) | SIGTRAP);
430                 else
431                         send_sig(SIGTRAP, current, 0);
432         }
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;
440         unsigned long error;
441         int retval;
442         struct exec ex;
443
444         inode = file->f_dentry->d_inode;
445
446         retval = -ENOEXEC;
447         error = kernel_read(file, 0, (char *) &ex, sizeof(ex));
448         if (error != sizeof(ex))
449                 goto out;
450
451         /* We come in here for the regular a.out style of shared libraries */
452         if ((N_MAGIC(ex) != ZMAGIC && N_MAGIC(ex) != QMAGIC) || N_TRSIZE(ex) ||
453             N_DRSIZE(ex) || ((ex.a_entry & 0xfff) && N_MAGIC(ex) == ZMAGIC) ||
454             i_size_read(inode) < 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 ((jiffies-error_time) > 5*HZ)
472                 {
473                         printk(KERN_WARNING 
474                                "N_TXTOFF is not page aligned. Please convert library: %s\n",
475                                file->f_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 *)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 + ex.a_data);
487
488                 retval = 0;
489                 goto out;
490         }
491         /* Now use mmap to map the library into memory. */
492         down_write(&current->mm->mmap_sem);
493         error = do_mmap(file, start_addr, ex.a_text + ex.a_data,
494                         PROT_READ | PROT_WRITE | PROT_EXEC,
495                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE | MAP_32BIT,
496                         N_TXTOFF(ex));
497         up_write(&current->mm->mmap_sem);
498         retval = error;
499         if (error != start_addr)
500                 goto out;
501
502         len = PAGE_ALIGN(ex.a_text + ex.a_data);
503         bss = ex.a_text + ex.a_data + ex.a_bss;
504         if (bss > len) {
505                 down_write(&current->mm->mmap_sem);
506                 error = do_brk(start_addr + len, bss - len);
507                 up_write(&current->mm->mmap_sem);
508                 retval = error;
509                 if (error != start_addr + len)
510                         goto out;
511         }
512         retval = 0;
513 out:
514         return retval;
515 }
516
517 static int __init init_aout_binfmt(void)
518 {
519         return register_binfmt(&aout_format);
520 }
521
522 static void __exit exit_aout_binfmt(void)
523 {
524         unregister_binfmt(&aout_format);
525 }
526
527 module_init(init_aout_binfmt);
528 module_exit(exit_aout_binfmt);
529 MODULE_LICENSE("GPL");