coredump: unify dump_seek() implementations for each binfmt_*.c
[linux-2.6.git] / fs / binfmt_aout.c
1 /*
2  *  linux/fs/binfmt_aout.c
3  *
4  *  Copyright (C) 1991, 1992, 1996  Linus Torvalds
5  */
6
7 #include <linux/module.h>
8
9 #include <linux/time.h>
10 #include <linux/kernel.h>
11 #include <linux/mm.h>
12 #include <linux/mman.h>
13 #include <linux/a.out.h>
14 #include <linux/errno.h>
15 #include <linux/signal.h>
16 #include <linux/string.h>
17 #include <linux/fs.h>
18 #include <linux/file.h>
19 #include <linux/stat.h>
20 #include <linux/fcntl.h>
21 #include <linux/ptrace.h>
22 #include <linux/user.h>
23 #include <linux/slab.h>
24 #include <linux/binfmts.h>
25 #include <linux/personality.h>
26 #include <linux/init.h>
27
28 #include <asm/system.h>
29 #include <asm/uaccess.h>
30 #include <asm/cacheflush.h>
31 #include <asm/a.out-core.h>
32
33 static int load_aout_binary(struct linux_binprm *, struct pt_regs * regs);
34 static int load_aout_library(struct file*);
35 static int aout_core_dump(struct coredump_params *cprm);
36
37 static struct linux_binfmt aout_format = {
38         .module         = THIS_MODULE,
39         .load_binary    = load_aout_binary,
40         .load_shlib     = load_aout_library,
41         .core_dump      = aout_core_dump,
42         .min_coredump   = PAGE_SIZE
43 };
44
45 #define BAD_ADDR(x)     ((unsigned long)(x) >= TASK_SIZE)
46
47 static int set_brk(unsigned long start, unsigned long end)
48 {
49         start = PAGE_ALIGN(start);
50         end = PAGE_ALIGN(end);
51         if (end > start) {
52                 unsigned long addr;
53                 down_write(&current->mm->mmap_sem);
54                 addr = do_brk(start, end - start);
55                 up_write(&current->mm->mmap_sem);
56                 if (BAD_ADDR(addr))
57                         return addr;
58         }
59         return 0;
60 }
61
62 /*
63  * These are the only things you should do on a core-file: use only these
64  * macros to write out all the necessary info.
65  */
66
67 static int dump_write(struct file *file, const void *addr, int nr)
68 {
69         return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
70 }
71
72 static int dump_seek(struct file *file, loff_t off)
73 {
74         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
75                 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
76                         return 0;
77         } else {
78                 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
79                 if (!buf)
80                         return 0;
81                 while (off > 0) {
82                         unsigned long n = off;
83                         if (n > PAGE_SIZE)
84                                 n = PAGE_SIZE;
85                         if (!dump_write(file, buf, n))
86                                 return 0;
87                         off -= n;
88                 }
89                 free_page((unsigned long)buf);
90         }
91         return 1;
92 }
93
94 #define DUMP_WRITE(addr, nr)    \
95         if (!dump_write(file, (void *)(addr), (nr))) \
96                 goto end_coredump;
97
98 /*
99  * Routine writes a core dump image in the current directory.
100  * Currently only a stub-function.
101  *
102  * Note that setuid/setgid files won't make a core-dump if the uid/gid
103  * changed due to the set[u|g]id. It's enforced by the "current->mm->dumpable"
104  * field, which also makes sure the core-dumps won't be recursive if the
105  * dumping of the process results in another error..
106  */
107
108 static int aout_core_dump(struct coredump_params *cprm)
109 {
110         struct file *file = cprm->file;
111         mm_segment_t fs;
112         int has_dumped = 0;
113         unsigned long dump_start, dump_size;
114         struct user dump;
115 #ifdef __alpha__
116 #       define START_DATA(u)    (u.start_data)
117 #else
118 #       define START_DATA(u)    ((u.u_tsize << PAGE_SHIFT) + u.start_code)
119 #endif
120 #       define START_STACK(u)   (u.start_stack)
121
122         fs = get_fs();
123         set_fs(KERNEL_DS);
124         has_dumped = 1;
125         current->flags |= PF_DUMPCORE;
126         strncpy(dump.u_comm, current->comm, sizeof(dump.u_comm));
127         dump.u_ar0 = offsetof(struct user, regs);
128         dump.signal = cprm->signr;
129         aout_dump_thread(cprm->regs, &dump);
130
131 /* If the size of the dump file exceeds the rlimit, then see what would happen
132    if we wrote the stack, but not the data area.  */
133         if ((dump.u_dsize + dump.u_ssize+1) * PAGE_SIZE > cprm->limit)
134                 dump.u_dsize = 0;
135
136 /* Make sure we have enough room to write the stack and data areas. */
137         if ((dump.u_ssize + 1) * PAGE_SIZE > cprm->limit)
138                 dump.u_ssize = 0;
139
140 /* make sure we actually have a data and stack area to dump */
141         set_fs(USER_DS);
142         if (!access_ok(VERIFY_READ, (void __user *)START_DATA(dump), dump.u_dsize << PAGE_SHIFT))
143                 dump.u_dsize = 0;
144         if (!access_ok(VERIFY_READ, (void __user *)START_STACK(dump), dump.u_ssize << PAGE_SHIFT))
145                 dump.u_ssize = 0;
146
147         set_fs(KERNEL_DS);
148 /* struct user */
149         DUMP_WRITE(&dump,sizeof(dump));
150 /* Now dump all of the user data.  Include malloced stuff as well */
151         if (!dump_seek(cprm->file, PAGE_SIZE - sizeof(dump)))
152                 goto end_coredump;
153 /* now we start writing out the user space info */
154         set_fs(USER_DS);
155 /* Dump the data area */
156         if (dump.u_dsize != 0) {
157                 dump_start = START_DATA(dump);
158                 dump_size = dump.u_dsize << PAGE_SHIFT;
159                 DUMP_WRITE(dump_start,dump_size);
160         }
161 /* Now prepare to dump the stack area */
162         if (dump.u_ssize != 0) {
163                 dump_start = START_STACK(dump);
164                 dump_size = dump.u_ssize << PAGE_SHIFT;
165                 DUMP_WRITE(dump_start,dump_size);
166         }
167 /* Finally dump the task struct.  Not be used by gdb, but could be useful */
168         set_fs(KERNEL_DS);
169         DUMP_WRITE(current,sizeof(*current));
170 end_coredump:
171         set_fs(fs);
172         return has_dumped;
173 }
174
175 /*
176  * create_aout_tables() parses the env- and arg-strings in new user
177  * memory and creates the pointer tables from them, and puts their
178  * addresses on the "stack", returning the new stack pointer value.
179  */
180 static unsigned long __user *create_aout_tables(char __user *p, struct linux_binprm * bprm)
181 {
182         char __user * __user *argv;
183         char __user * __user *envp;
184         unsigned long __user *sp;
185         int argc = bprm->argc;
186         int envc = bprm->envc;
187
188         sp = (void __user *)((-(unsigned long)sizeof(char *)) & (unsigned long) p);
189 #ifdef __alpha__
190 /* whee.. test-programs are so much fun. */
191         put_user(0, --sp);
192         put_user(0, --sp);
193         if (bprm->loader) {
194                 put_user(0, --sp);
195                 put_user(1003, --sp);
196                 put_user(bprm->loader, --sp);
197                 put_user(1002, --sp);
198         }
199         put_user(bprm->exec, --sp);
200         put_user(1001, --sp);
201 #endif
202         sp -= envc+1;
203         envp = (char __user * __user *) sp;
204         sp -= argc+1;
205         argv = (char __user * __user *) sp;
206 #ifndef __alpha__
207         put_user((unsigned long) envp,--sp);
208         put_user((unsigned long) argv,--sp);
209 #endif
210         put_user(argc,--sp);
211         current->mm->arg_start = (unsigned long) p;
212         while (argc-->0) {
213                 char c;
214                 put_user(p,argv++);
215                 do {
216                         get_user(c,p++);
217                 } while (c);
218         }
219         put_user(NULL,argv);
220         current->mm->arg_end = current->mm->env_start = (unsigned long) p;
221         while (envc-->0) {
222                 char c;
223                 put_user(p,envp++);
224                 do {
225                         get_user(c,p++);
226                 } while (c);
227         }
228         put_user(NULL,envp);
229         current->mm->env_end = (unsigned long) p;
230         return sp;
231 }
232
233 /*
234  * These are the functions used to load a.out style executables and shared
235  * libraries.  There is no binary dependent code anywhere else.
236  */
237
238 static int load_aout_binary(struct linux_binprm * bprm, struct pt_regs * regs)
239 {
240         struct exec ex;
241         unsigned long error;
242         unsigned long fd_offset;
243         unsigned long rlim;
244         int retval;
245
246         ex = *((struct exec *) bprm->buf);              /* exec-header */
247         if ((N_MAGIC(ex) != ZMAGIC && N_MAGIC(ex) != OMAGIC &&
248              N_MAGIC(ex) != QMAGIC && N_MAGIC(ex) != NMAGIC) ||
249             N_TRSIZE(ex) || N_DRSIZE(ex) ||
250             i_size_read(bprm->file->f_path.dentry->d_inode) < ex.a_text+ex.a_data+N_SYMSIZE(ex)+N_TXTOFF(ex)) {
251                 return -ENOEXEC;
252         }
253
254         /*
255          * Requires a mmap handler. This prevents people from using a.out
256          * as part of an exploit attack against /proc-related vulnerabilities.
257          */
258         if (!bprm->file->f_op || !bprm->file->f_op->mmap)
259                 return -ENOEXEC;
260
261         fd_offset = N_TXTOFF(ex);
262
263         /* Check initial limits. This avoids letting people circumvent
264          * size limits imposed on them by creating programs with large
265          * arrays in the data or bss.
266          */
267         rlim = rlimit(RLIMIT_DATA);
268         if (rlim >= RLIM_INFINITY)
269                 rlim = ~0;
270         if (ex.a_data + ex.a_bss > rlim)
271                 return -ENOMEM;
272
273         /* Flush all traces of the currently running executable */
274         retval = flush_old_exec(bprm);
275         if (retval)
276                 return retval;
277
278         /* OK, This is the point of no return */
279 #ifdef __alpha__
280         SET_AOUT_PERSONALITY(bprm, ex);
281 #else
282         set_personality(PER_LINUX);
283 #endif
284         setup_new_exec(bprm);
285
286         current->mm->end_code = ex.a_text +
287                 (current->mm->start_code = N_TXTADDR(ex));
288         current->mm->end_data = ex.a_data +
289                 (current->mm->start_data = N_DATADDR(ex));
290         current->mm->brk = ex.a_bss +
291                 (current->mm->start_brk = N_BSSADDR(ex));
292         current->mm->free_area_cache = current->mm->mmap_base;
293         current->mm->cached_hole_size = 0;
294
295         install_exec_creds(bprm);
296         current->flags &= ~PF_FORKNOEXEC;
297
298         if (N_MAGIC(ex) == OMAGIC) {
299                 unsigned long text_addr, map_size;
300                 loff_t pos;
301
302                 text_addr = N_TXTADDR(ex);
303
304 #ifdef __alpha__
305                 pos = fd_offset;
306                 map_size = ex.a_text+ex.a_data + PAGE_SIZE - 1;
307 #else
308                 pos = 32;
309                 map_size = ex.a_text+ex.a_data;
310 #endif
311                 down_write(&current->mm->mmap_sem);
312                 error = do_brk(text_addr & PAGE_MASK, map_size);
313                 up_write(&current->mm->mmap_sem);
314                 if (error != (text_addr & PAGE_MASK)) {
315                         send_sig(SIGKILL, current, 0);
316                         return error;
317                 }
318
319                 error = bprm->file->f_op->read(bprm->file,
320                           (char __user *)text_addr,
321                           ex.a_text+ex.a_data, &pos);
322                 if ((signed long)error < 0) {
323                         send_sig(SIGKILL, current, 0);
324                         return error;
325                 }
326                          
327                 flush_icache_range(text_addr, text_addr+ex.a_text+ex.a_data);
328         } else {
329                 if ((ex.a_text & 0xfff || ex.a_data & 0xfff) &&
330                     (N_MAGIC(ex) != NMAGIC) && printk_ratelimit())
331                 {
332                         printk(KERN_NOTICE "executable not page aligned\n");
333                 }
334
335                 if ((fd_offset & ~PAGE_MASK) != 0 && printk_ratelimit())
336                 {
337                         printk(KERN_WARNING 
338                                "fd_offset is not page aligned. Please convert program: %s\n",
339                                bprm->file->f_path.dentry->d_name.name);
340                 }
341
342                 if (!bprm->file->f_op->mmap||((fd_offset & ~PAGE_MASK) != 0)) {
343                         loff_t pos = fd_offset;
344                         down_write(&current->mm->mmap_sem);
345                         do_brk(N_TXTADDR(ex), ex.a_text+ex.a_data);
346                         up_write(&current->mm->mmap_sem);
347                         bprm->file->f_op->read(bprm->file,
348                                         (char __user *)N_TXTADDR(ex),
349                                         ex.a_text+ex.a_data, &pos);
350                         flush_icache_range((unsigned long) N_TXTADDR(ex),
351                                            (unsigned long) N_TXTADDR(ex) +
352                                            ex.a_text+ex.a_data);
353                         goto beyond_if;
354                 }
355
356                 down_write(&current->mm->mmap_sem);
357                 error = do_mmap(bprm->file, N_TXTADDR(ex), ex.a_text,
358                         PROT_READ | PROT_EXEC,
359                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE,
360                         fd_offset);
361                 up_write(&current->mm->mmap_sem);
362
363                 if (error != N_TXTADDR(ex)) {
364                         send_sig(SIGKILL, current, 0);
365                         return error;
366                 }
367
368                 down_write(&current->mm->mmap_sem);
369                 error = do_mmap(bprm->file, N_DATADDR(ex), ex.a_data,
370                                 PROT_READ | PROT_WRITE | PROT_EXEC,
371                                 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE,
372                                 fd_offset + ex.a_text);
373                 up_write(&current->mm->mmap_sem);
374                 if (error != N_DATADDR(ex)) {
375                         send_sig(SIGKILL, current, 0);
376                         return error;
377                 }
378         }
379 beyond_if:
380         set_binfmt(&aout_format);
381
382         retval = set_brk(current->mm->start_brk, current->mm->brk);
383         if (retval < 0) {
384                 send_sig(SIGKILL, current, 0);
385                 return retval;
386         }
387
388         retval = setup_arg_pages(bprm, STACK_TOP, EXSTACK_DEFAULT);
389         if (retval < 0) { 
390                 /* Someone check-me: is this error path enough? */ 
391                 send_sig(SIGKILL, current, 0); 
392                 return retval;
393         }
394
395         current->mm->start_stack =
396                 (unsigned long) create_aout_tables((char __user *) bprm->p, bprm);
397 #ifdef __alpha__
398         regs->gp = ex.a_gpvalue;
399 #endif
400         start_thread(regs, ex.a_entry, current->mm->start_stack);
401         return 0;
402 }
403
404 static int load_aout_library(struct file *file)
405 {
406         struct inode * inode;
407         unsigned long bss, start_addr, len;
408         unsigned long error;
409         int retval;
410         struct exec ex;
411
412         inode = file->f_path.dentry->d_inode;
413
414         retval = -ENOEXEC;
415         error = kernel_read(file, 0, (char *) &ex, sizeof(ex));
416         if (error != sizeof(ex))
417                 goto out;
418
419         /* We come in here for the regular a.out style of shared libraries */
420         if ((N_MAGIC(ex) != ZMAGIC && N_MAGIC(ex) != QMAGIC) || N_TRSIZE(ex) ||
421             N_DRSIZE(ex) || ((ex.a_entry & 0xfff) && N_MAGIC(ex) == ZMAGIC) ||
422             i_size_read(inode) < ex.a_text+ex.a_data+N_SYMSIZE(ex)+N_TXTOFF(ex)) {
423                 goto out;
424         }
425
426         /*
427          * Requires a mmap handler. This prevents people from using a.out
428          * as part of an exploit attack against /proc-related vulnerabilities.
429          */
430         if (!file->f_op || !file->f_op->mmap)
431                 goto out;
432
433         if (N_FLAGS(ex))
434                 goto out;
435
436         /* For  QMAGIC, the starting address is 0x20 into the page.  We mask
437            this off to get the starting address for the page */
438
439         start_addr =  ex.a_entry & 0xfffff000;
440
441         if ((N_TXTOFF(ex) & ~PAGE_MASK) != 0) {
442                 loff_t pos = N_TXTOFF(ex);
443
444                 if (printk_ratelimit())
445                 {
446                         printk(KERN_WARNING 
447                                "N_TXTOFF is not page aligned. Please convert library: %s\n",
448                                file->f_path.dentry->d_name.name);
449                 }
450                 down_write(&current->mm->mmap_sem);
451                 do_brk(start_addr, ex.a_text + ex.a_data + ex.a_bss);
452                 up_write(&current->mm->mmap_sem);
453                 
454                 file->f_op->read(file, (char __user *)start_addr,
455                         ex.a_text + ex.a_data, &pos);
456                 flush_icache_range((unsigned long) start_addr,
457                                    (unsigned long) start_addr + ex.a_text + ex.a_data);
458
459                 retval = 0;
460                 goto out;
461         }
462         /* Now use mmap to map the library into memory. */
463         down_write(&current->mm->mmap_sem);
464         error = do_mmap(file, start_addr, ex.a_text + ex.a_data,
465                         PROT_READ | PROT_WRITE | PROT_EXEC,
466                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
467                         N_TXTOFF(ex));
468         up_write(&current->mm->mmap_sem);
469         retval = error;
470         if (error != start_addr)
471                 goto out;
472
473         len = PAGE_ALIGN(ex.a_text + ex.a_data);
474         bss = ex.a_text + ex.a_data + ex.a_bss;
475         if (bss > len) {
476                 down_write(&current->mm->mmap_sem);
477                 error = do_brk(start_addr + len, bss - len);
478                 up_write(&current->mm->mmap_sem);
479                 retval = error;
480                 if (error != start_addr + len)
481                         goto out;
482         }
483         retval = 0;
484 out:
485         return retval;
486 }
487
488 static int __init init_aout_binfmt(void)
489 {
490         return register_binfmt(&aout_format);
491 }
492
493 static void __exit exit_aout_binfmt(void)
494 {
495         unregister_binfmt(&aout_format);
496 }
497
498 core_initcall(init_aout_binfmt);
499 module_exit(exit_aout_binfmt);
500 MODULE_LICENSE("GPL");