Merge branch 'release' of git://git.kernel.org/pub/scm/linux/kernel/git/aegl/linux-2.6
[linux-2.6] / fs / binfmt_elf.c
1 /*
2  * linux/fs/binfmt_elf.c
3  *
4  * These are the functions used to load ELF format executables as used
5  * on SVr4 machines.  Information on the format may be found in the book
6  * "UNIX SYSTEM V RELEASE 4 Programmers Guide: Ansi C and Programming Support
7  * Tools".
8  *
9  * Copyright 1993, 1994: Eric Youngdale (ericy@cais.com).
10  */
11
12 #include <linux/module.h>
13 #include <linux/kernel.h>
14 #include <linux/fs.h>
15 #include <linux/stat.h>
16 #include <linux/time.h>
17 #include <linux/mm.h>
18 #include <linux/mman.h>
19 #include <linux/errno.h>
20 #include <linux/signal.h>
21 #include <linux/binfmts.h>
22 #include <linux/string.h>
23 #include <linux/file.h>
24 #include <linux/fcntl.h>
25 #include <linux/ptrace.h>
26 #include <linux/slab.h>
27 #include <linux/shm.h>
28 #include <linux/personality.h>
29 #include <linux/elfcore.h>
30 #include <linux/init.h>
31 #include <linux/highuid.h>
32 #include <linux/smp.h>
33 #include <linux/compiler.h>
34 #include <linux/highmem.h>
35 #include <linux/pagemap.h>
36 #include <linux/security.h>
37 #include <linux/syscalls.h>
38 #include <linux/random.h>
39 #include <linux/elf.h>
40 #include <linux/utsname.h>
41 #include <asm/uaccess.h>
42 #include <asm/param.h>
43 #include <asm/page.h>
44
45 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs);
46 static int load_elf_library(struct file *);
47 static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
48                                 int, int, unsigned long);
49
50 /*
51  * If we don't support core dumping, then supply a NULL so we
52  * don't even try.
53  */
54 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
55 static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, unsigned long limit);
56 #else
57 #define elf_core_dump   NULL
58 #endif
59
60 #if ELF_EXEC_PAGESIZE > PAGE_SIZE
61 #define ELF_MIN_ALIGN   ELF_EXEC_PAGESIZE
62 #else
63 #define ELF_MIN_ALIGN   PAGE_SIZE
64 #endif
65
66 #ifndef ELF_CORE_EFLAGS
67 #define ELF_CORE_EFLAGS 0
68 #endif
69
70 #define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
71 #define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
72 #define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
73
74 static struct linux_binfmt elf_format = {
75                 .module         = THIS_MODULE,
76                 .load_binary    = load_elf_binary,
77                 .load_shlib     = load_elf_library,
78                 .core_dump      = elf_core_dump,
79                 .min_coredump   = ELF_EXEC_PAGESIZE,
80                 .hasvdso        = 1
81 };
82
83 #define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
84
85 static int set_brk(unsigned long start, unsigned long end)
86 {
87         start = ELF_PAGEALIGN(start);
88         end = ELF_PAGEALIGN(end);
89         if (end > start) {
90                 unsigned long addr;
91                 down_write(&current->mm->mmap_sem);
92                 addr = do_brk(start, end - start);
93                 up_write(&current->mm->mmap_sem);
94                 if (BAD_ADDR(addr))
95                         return addr;
96         }
97         current->mm->start_brk = current->mm->brk = end;
98         return 0;
99 }
100
101 /* We need to explicitly zero any fractional pages
102    after the data section (i.e. bss).  This would
103    contain the junk from the file that should not
104    be in memory
105  */
106 static int padzero(unsigned long elf_bss)
107 {
108         unsigned long nbyte;
109
110         nbyte = ELF_PAGEOFFSET(elf_bss);
111         if (nbyte) {
112                 nbyte = ELF_MIN_ALIGN - nbyte;
113                 if (clear_user((void __user *) elf_bss, nbyte))
114                         return -EFAULT;
115         }
116         return 0;
117 }
118
119 /* Let's use some macros to make this stack manipulation a little clearer */
120 #ifdef CONFIG_STACK_GROWSUP
121 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
122 #define STACK_ROUND(sp, items) \
123         ((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
124 #define STACK_ALLOC(sp, len) ({ \
125         elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
126         old_sp; })
127 #else
128 #define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
129 #define STACK_ROUND(sp, items) \
130         (((unsigned long) (sp - items)) &~ 15UL)
131 #define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
132 #endif
133
134 #ifndef ELF_BASE_PLATFORM
135 /*
136  * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
137  * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
138  * will be copied to the user stack in the same manner as AT_PLATFORM.
139  */
140 #define ELF_BASE_PLATFORM NULL
141 #endif
142
143 static int
144 create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
145                 unsigned long load_addr, unsigned long interp_load_addr)
146 {
147         unsigned long p = bprm->p;
148         int argc = bprm->argc;
149         int envc = bprm->envc;
150         elf_addr_t __user *argv;
151         elf_addr_t __user *envp;
152         elf_addr_t __user *sp;
153         elf_addr_t __user *u_platform;
154         elf_addr_t __user *u_base_platform;
155         const char *k_platform = ELF_PLATFORM;
156         const char *k_base_platform = ELF_BASE_PLATFORM;
157         int items;
158         elf_addr_t *elf_info;
159         int ei_index = 0;
160         struct task_struct *tsk = current;
161         struct vm_area_struct *vma;
162
163         /*
164          * In some cases (e.g. Hyper-Threading), we want to avoid L1
165          * evictions by the processes running on the same package. One
166          * thing we can do is to shuffle the initial stack for them.
167          */
168
169         p = arch_align_stack(p);
170
171         /*
172          * If this architecture has a platform capability string, copy it
173          * to userspace.  In some cases (Sparc), this info is impossible
174          * for userspace to get any other way, in others (i386) it is
175          * merely difficult.
176          */
177         u_platform = NULL;
178         if (k_platform) {
179                 size_t len = strlen(k_platform) + 1;
180
181                 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
182                 if (__copy_to_user(u_platform, k_platform, len))
183                         return -EFAULT;
184         }
185
186         /*
187          * If this architecture has a "base" platform capability
188          * string, copy it to userspace.
189          */
190         u_base_platform = NULL;
191         if (k_base_platform) {
192                 size_t len = strlen(k_base_platform) + 1;
193
194                 u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
195                 if (__copy_to_user(u_base_platform, k_base_platform, len))
196                         return -EFAULT;
197         }
198
199         /* Create the ELF interpreter info */
200         elf_info = (elf_addr_t *)current->mm->saved_auxv;
201         /* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
202 #define NEW_AUX_ENT(id, val) \
203         do { \
204                 elf_info[ei_index++] = id; \
205                 elf_info[ei_index++] = val; \
206         } while (0)
207
208 #ifdef ARCH_DLINFO
209         /* 
210          * ARCH_DLINFO must come first so PPC can do its special alignment of
211          * AUXV.
212          * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
213          * ARCH_DLINFO changes
214          */
215         ARCH_DLINFO;
216 #endif
217         NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
218         NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
219         NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
220         NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
221         NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
222         NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
223         NEW_AUX_ENT(AT_BASE, interp_load_addr);
224         NEW_AUX_ENT(AT_FLAGS, 0);
225         NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
226         NEW_AUX_ENT(AT_UID, tsk->uid);
227         NEW_AUX_ENT(AT_EUID, tsk->euid);
228         NEW_AUX_ENT(AT_GID, tsk->gid);
229         NEW_AUX_ENT(AT_EGID, tsk->egid);
230         NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
231         NEW_AUX_ENT(AT_EXECFN, bprm->exec);
232         if (k_platform) {
233                 NEW_AUX_ENT(AT_PLATFORM,
234                             (elf_addr_t)(unsigned long)u_platform);
235         }
236         if (k_base_platform) {
237                 NEW_AUX_ENT(AT_BASE_PLATFORM,
238                             (elf_addr_t)(unsigned long)u_base_platform);
239         }
240         if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
241                 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
242         }
243 #undef NEW_AUX_ENT
244         /* AT_NULL is zero; clear the rest too */
245         memset(&elf_info[ei_index], 0,
246                sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
247
248         /* And advance past the AT_NULL entry.  */
249         ei_index += 2;
250
251         sp = STACK_ADD(p, ei_index);
252
253         items = (argc + 1) + (envc + 1) + 1;
254         bprm->p = STACK_ROUND(sp, items);
255
256         /* Point sp at the lowest address on the stack */
257 #ifdef CONFIG_STACK_GROWSUP
258         sp = (elf_addr_t __user *)bprm->p - items - ei_index;
259         bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
260 #else
261         sp = (elf_addr_t __user *)bprm->p;
262 #endif
263
264
265         /*
266          * Grow the stack manually; some architectures have a limit on how
267          * far ahead a user-space access may be in order to grow the stack.
268          */
269         vma = find_extend_vma(current->mm, bprm->p);
270         if (!vma)
271                 return -EFAULT;
272
273         /* Now, let's put argc (and argv, envp if appropriate) on the stack */
274         if (__put_user(argc, sp++))
275                 return -EFAULT;
276         argv = sp;
277         envp = argv + argc + 1;
278
279         /* Populate argv and envp */
280         p = current->mm->arg_end = current->mm->arg_start;
281         while (argc-- > 0) {
282                 size_t len;
283                 if (__put_user((elf_addr_t)p, argv++))
284                         return -EFAULT;
285                 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
286                 if (!len || len > MAX_ARG_STRLEN)
287                         return -EINVAL;
288                 p += len;
289         }
290         if (__put_user(0, argv))
291                 return -EFAULT;
292         current->mm->arg_end = current->mm->env_start = p;
293         while (envc-- > 0) {
294                 size_t len;
295                 if (__put_user((elf_addr_t)p, envp++))
296                         return -EFAULT;
297                 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
298                 if (!len || len > MAX_ARG_STRLEN)
299                         return -EINVAL;
300                 p += len;
301         }
302         if (__put_user(0, envp))
303                 return -EFAULT;
304         current->mm->env_end = p;
305
306         /* Put the elf_info on the stack in the right place.  */
307         sp = (elf_addr_t __user *)envp + 1;
308         if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
309                 return -EFAULT;
310         return 0;
311 }
312
313 #ifndef elf_map
314
315 static unsigned long elf_map(struct file *filep, unsigned long addr,
316                 struct elf_phdr *eppnt, int prot, int type,
317                 unsigned long total_size)
318 {
319         unsigned long map_addr;
320         unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
321         unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
322         addr = ELF_PAGESTART(addr);
323         size = ELF_PAGEALIGN(size);
324
325         /* mmap() will return -EINVAL if given a zero size, but a
326          * segment with zero filesize is perfectly valid */
327         if (!size)
328                 return addr;
329
330         down_write(&current->mm->mmap_sem);
331         /*
332         * total_size is the size of the ELF (interpreter) image.
333         * The _first_ mmap needs to know the full size, otherwise
334         * randomization might put this image into an overlapping
335         * position with the ELF binary image. (since size < total_size)
336         * So we first map the 'big' image - and unmap the remainder at
337         * the end. (which unmap is needed for ELF images with holes.)
338         */
339         if (total_size) {
340                 total_size = ELF_PAGEALIGN(total_size);
341                 map_addr = do_mmap(filep, addr, total_size, prot, type, off);
342                 if (!BAD_ADDR(map_addr))
343                         do_munmap(current->mm, map_addr+size, total_size-size);
344         } else
345                 map_addr = do_mmap(filep, addr, size, prot, type, off);
346
347         up_write(&current->mm->mmap_sem);
348         return(map_addr);
349 }
350
351 #endif /* !elf_map */
352
353 static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
354 {
355         int i, first_idx = -1, last_idx = -1;
356
357         for (i = 0; i < nr; i++) {
358                 if (cmds[i].p_type == PT_LOAD) {
359                         last_idx = i;
360                         if (first_idx == -1)
361                                 first_idx = i;
362                 }
363         }
364         if (first_idx == -1)
365                 return 0;
366
367         return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
368                                 ELF_PAGESTART(cmds[first_idx].p_vaddr);
369 }
370
371
372 /* This is much more generalized than the library routine read function,
373    so we keep this separate.  Technically the library read function
374    is only provided so that we can read a.out libraries that have
375    an ELF header */
376
377 static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
378                 struct file *interpreter, unsigned long *interp_map_addr,
379                 unsigned long no_base)
380 {
381         struct elf_phdr *elf_phdata;
382         struct elf_phdr *eppnt;
383         unsigned long load_addr = 0;
384         int load_addr_set = 0;
385         unsigned long last_bss = 0, elf_bss = 0;
386         unsigned long error = ~0UL;
387         unsigned long total_size;
388         int retval, i, size;
389
390         /* First of all, some simple consistency checks */
391         if (interp_elf_ex->e_type != ET_EXEC &&
392             interp_elf_ex->e_type != ET_DYN)
393                 goto out;
394         if (!elf_check_arch(interp_elf_ex))
395                 goto out;
396         if (!interpreter->f_op || !interpreter->f_op->mmap)
397                 goto out;
398
399         /*
400          * If the size of this structure has changed, then punt, since
401          * we will be doing the wrong thing.
402          */
403         if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
404                 goto out;
405         if (interp_elf_ex->e_phnum < 1 ||
406                 interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
407                 goto out;
408
409         /* Now read in all of the header information */
410         size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
411         if (size > ELF_MIN_ALIGN)
412                 goto out;
413         elf_phdata = kmalloc(size, GFP_KERNEL);
414         if (!elf_phdata)
415                 goto out;
416
417         retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
418                              (char *)elf_phdata,size);
419         error = -EIO;
420         if (retval != size) {
421                 if (retval < 0)
422                         error = retval; 
423                 goto out_close;
424         }
425
426         total_size = total_mapping_size(elf_phdata, interp_elf_ex->e_phnum);
427         if (!total_size) {
428                 error = -EINVAL;
429                 goto out_close;
430         }
431
432         eppnt = elf_phdata;
433         for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
434                 if (eppnt->p_type == PT_LOAD) {
435                         int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
436                         int elf_prot = 0;
437                         unsigned long vaddr = 0;
438                         unsigned long k, map_addr;
439
440                         if (eppnt->p_flags & PF_R)
441                                 elf_prot = PROT_READ;
442                         if (eppnt->p_flags & PF_W)
443                                 elf_prot |= PROT_WRITE;
444                         if (eppnt->p_flags & PF_X)
445                                 elf_prot |= PROT_EXEC;
446                         vaddr = eppnt->p_vaddr;
447                         if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
448                                 elf_type |= MAP_FIXED;
449                         else if (no_base && interp_elf_ex->e_type == ET_DYN)
450                                 load_addr = -vaddr;
451
452                         map_addr = elf_map(interpreter, load_addr + vaddr,
453                                         eppnt, elf_prot, elf_type, total_size);
454                         total_size = 0;
455                         if (!*interp_map_addr)
456                                 *interp_map_addr = map_addr;
457                         error = map_addr;
458                         if (BAD_ADDR(map_addr))
459                                 goto out_close;
460
461                         if (!load_addr_set &&
462                             interp_elf_ex->e_type == ET_DYN) {
463                                 load_addr = map_addr - ELF_PAGESTART(vaddr);
464                                 load_addr_set = 1;
465                         }
466
467                         /*
468                          * Check to see if the section's size will overflow the
469                          * allowed task size. Note that p_filesz must always be
470                          * <= p_memsize so it's only necessary to check p_memsz.
471                          */
472                         k = load_addr + eppnt->p_vaddr;
473                         if (BAD_ADDR(k) ||
474                             eppnt->p_filesz > eppnt->p_memsz ||
475                             eppnt->p_memsz > TASK_SIZE ||
476                             TASK_SIZE - eppnt->p_memsz < k) {
477                                 error = -ENOMEM;
478                                 goto out_close;
479                         }
480
481                         /*
482                          * Find the end of the file mapping for this phdr, and
483                          * keep track of the largest address we see for this.
484                          */
485                         k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
486                         if (k > elf_bss)
487                                 elf_bss = k;
488
489                         /*
490                          * Do the same thing for the memory mapping - between
491                          * elf_bss and last_bss is the bss section.
492                          */
493                         k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
494                         if (k > last_bss)
495                                 last_bss = k;
496                 }
497         }
498
499         /*
500          * Now fill out the bss section.  First pad the last page up
501          * to the page boundary, and then perform a mmap to make sure
502          * that there are zero-mapped pages up to and including the 
503          * last bss page.
504          */
505         if (padzero(elf_bss)) {
506                 error = -EFAULT;
507                 goto out_close;
508         }
509
510         /* What we have mapped so far */
511         elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
512
513         /* Map the last of the bss segment */
514         if (last_bss > elf_bss) {
515                 down_write(&current->mm->mmap_sem);
516                 error = do_brk(elf_bss, last_bss - elf_bss);
517                 up_write(&current->mm->mmap_sem);
518                 if (BAD_ADDR(error))
519                         goto out_close;
520         }
521
522         error = load_addr;
523
524 out_close:
525         kfree(elf_phdata);
526 out:
527         return error;
528 }
529
530 /*
531  * These are the functions used to load ELF style executables and shared
532  * libraries.  There is no binary dependent code anywhere else.
533  */
534
535 #define INTERPRETER_NONE 0
536 #define INTERPRETER_ELF 2
537
538 #ifndef STACK_RND_MASK
539 #define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12))     /* 8MB of VA */
540 #endif
541
542 static unsigned long randomize_stack_top(unsigned long stack_top)
543 {
544         unsigned int random_variable = 0;
545
546         if ((current->flags & PF_RANDOMIZE) &&
547                 !(current->personality & ADDR_NO_RANDOMIZE)) {
548                 random_variable = get_random_int() & STACK_RND_MASK;
549                 random_variable <<= PAGE_SHIFT;
550         }
551 #ifdef CONFIG_STACK_GROWSUP
552         return PAGE_ALIGN(stack_top) + random_variable;
553 #else
554         return PAGE_ALIGN(stack_top) - random_variable;
555 #endif
556 }
557
558 static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
559 {
560         struct file *interpreter = NULL; /* to shut gcc up */
561         unsigned long load_addr = 0, load_bias = 0;
562         int load_addr_set = 0;
563         char * elf_interpreter = NULL;
564         unsigned long error;
565         struct elf_phdr *elf_ppnt, *elf_phdata;
566         unsigned long elf_bss, elf_brk;
567         int elf_exec_fileno;
568         int retval, i;
569         unsigned int size;
570         unsigned long elf_entry;
571         unsigned long interp_load_addr = 0;
572         unsigned long start_code, end_code, start_data, end_data;
573         unsigned long reloc_func_desc = 0;
574         int executable_stack = EXSTACK_DEFAULT;
575         unsigned long def_flags = 0;
576         struct {
577                 struct elfhdr elf_ex;
578                 struct elfhdr interp_elf_ex;
579         } *loc;
580
581         loc = kmalloc(sizeof(*loc), GFP_KERNEL);
582         if (!loc) {
583                 retval = -ENOMEM;
584                 goto out_ret;
585         }
586         
587         /* Get the exec-header */
588         loc->elf_ex = *((struct elfhdr *)bprm->buf);
589
590         retval = -ENOEXEC;
591         /* First of all, some simple consistency checks */
592         if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
593                 goto out;
594
595         if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
596                 goto out;
597         if (!elf_check_arch(&loc->elf_ex))
598                 goto out;
599         if (!bprm->file->f_op||!bprm->file->f_op->mmap)
600                 goto out;
601
602         /* Now read in all of the header information */
603         if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
604                 goto out;
605         if (loc->elf_ex.e_phnum < 1 ||
606                 loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
607                 goto out;
608         size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
609         retval = -ENOMEM;
610         elf_phdata = kmalloc(size, GFP_KERNEL);
611         if (!elf_phdata)
612                 goto out;
613
614         retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
615                              (char *)elf_phdata, size);
616         if (retval != size) {
617                 if (retval >= 0)
618                         retval = -EIO;
619                 goto out_free_ph;
620         }
621
622         retval = get_unused_fd();
623         if (retval < 0)
624                 goto out_free_ph;
625         get_file(bprm->file);
626         fd_install(elf_exec_fileno = retval, bprm->file);
627
628         elf_ppnt = elf_phdata;
629         elf_bss = 0;
630         elf_brk = 0;
631
632         start_code = ~0UL;
633         end_code = 0;
634         start_data = 0;
635         end_data = 0;
636
637         for (i = 0; i < loc->elf_ex.e_phnum; i++) {
638                 if (elf_ppnt->p_type == PT_INTERP) {
639                         /* This is the program interpreter used for
640                          * shared libraries - for now assume that this
641                          * is an a.out format binary
642                          */
643                         retval = -ENOEXEC;
644                         if (elf_ppnt->p_filesz > PATH_MAX || 
645                             elf_ppnt->p_filesz < 2)
646                                 goto out_free_file;
647
648                         retval = -ENOMEM;
649                         elf_interpreter = kmalloc(elf_ppnt->p_filesz,
650                                                   GFP_KERNEL);
651                         if (!elf_interpreter)
652                                 goto out_free_file;
653
654                         retval = kernel_read(bprm->file, elf_ppnt->p_offset,
655                                              elf_interpreter,
656                                              elf_ppnt->p_filesz);
657                         if (retval != elf_ppnt->p_filesz) {
658                                 if (retval >= 0)
659                                         retval = -EIO;
660                                 goto out_free_interp;
661                         }
662                         /* make sure path is NULL terminated */
663                         retval = -ENOEXEC;
664                         if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
665                                 goto out_free_interp;
666
667                         /*
668                          * The early SET_PERSONALITY here is so that the lookup
669                          * for the interpreter happens in the namespace of the 
670                          * to-be-execed image.  SET_PERSONALITY can select an
671                          * alternate root.
672                          *
673                          * However, SET_PERSONALITY is NOT allowed to switch
674                          * this task into the new images's memory mapping
675                          * policy - that is, TASK_SIZE must still evaluate to
676                          * that which is appropriate to the execing application.
677                          * This is because exit_mmap() needs to have TASK_SIZE
678                          * evaluate to the size of the old image.
679                          *
680                          * So if (say) a 64-bit application is execing a 32-bit
681                          * application it is the architecture's responsibility
682                          * to defer changing the value of TASK_SIZE until the
683                          * switch really is going to happen - do this in
684                          * flush_thread().      - akpm
685                          */
686                         SET_PERSONALITY(loc->elf_ex, 0);
687
688                         interpreter = open_exec(elf_interpreter);
689                         retval = PTR_ERR(interpreter);
690                         if (IS_ERR(interpreter))
691                                 goto out_free_interp;
692
693                         /*
694                          * If the binary is not readable then enforce
695                          * mm->dumpable = 0 regardless of the interpreter's
696                          * permissions.
697                          */
698                         if (file_permission(interpreter, MAY_READ) < 0)
699                                 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
700
701                         retval = kernel_read(interpreter, 0, bprm->buf,
702                                              BINPRM_BUF_SIZE);
703                         if (retval != BINPRM_BUF_SIZE) {
704                                 if (retval >= 0)
705                                         retval = -EIO;
706                                 goto out_free_dentry;
707                         }
708
709                         /* Get the exec headers */
710                         loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
711                         break;
712                 }
713                 elf_ppnt++;
714         }
715
716         elf_ppnt = elf_phdata;
717         for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
718                 if (elf_ppnt->p_type == PT_GNU_STACK) {
719                         if (elf_ppnt->p_flags & PF_X)
720                                 executable_stack = EXSTACK_ENABLE_X;
721                         else
722                                 executable_stack = EXSTACK_DISABLE_X;
723                         break;
724                 }
725
726         /* Some simple consistency checks for the interpreter */
727         if (elf_interpreter) {
728                 retval = -ELIBBAD;
729                 /* Not an ELF interpreter */
730                 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
731                         goto out_free_dentry;
732                 /* Verify the interpreter has a valid arch */
733                 if (!elf_check_arch(&loc->interp_elf_ex))
734                         goto out_free_dentry;
735         } else {
736                 /* Executables without an interpreter also need a personality  */
737                 SET_PERSONALITY(loc->elf_ex, 0);
738         }
739
740         /* Flush all traces of the currently running executable */
741         retval = flush_old_exec(bprm);
742         if (retval)
743                 goto out_free_dentry;
744
745         /* OK, This is the point of no return */
746         current->flags &= ~PF_FORKNOEXEC;
747         current->mm->def_flags = def_flags;
748
749         /* Do this immediately, since STACK_TOP as used in setup_arg_pages
750            may depend on the personality.  */
751         SET_PERSONALITY(loc->elf_ex, 0);
752         if (elf_read_implies_exec(loc->elf_ex, executable_stack))
753                 current->personality |= READ_IMPLIES_EXEC;
754
755         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
756                 current->flags |= PF_RANDOMIZE;
757         arch_pick_mmap_layout(current->mm);
758
759         /* Do this so that we can load the interpreter, if need be.  We will
760            change some of these later */
761         current->mm->free_area_cache = current->mm->mmap_base;
762         current->mm->cached_hole_size = 0;
763         retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
764                                  executable_stack);
765         if (retval < 0) {
766                 send_sig(SIGKILL, current, 0);
767                 goto out_free_dentry;
768         }
769         
770         current->mm->start_stack = bprm->p;
771
772         /* Now we do a little grungy work by mmaping the ELF image into
773            the correct location in memory. */
774         for(i = 0, elf_ppnt = elf_phdata;
775             i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
776                 int elf_prot = 0, elf_flags;
777                 unsigned long k, vaddr;
778
779                 if (elf_ppnt->p_type != PT_LOAD)
780                         continue;
781
782                 if (unlikely (elf_brk > elf_bss)) {
783                         unsigned long nbyte;
784                     
785                         /* There was a PT_LOAD segment with p_memsz > p_filesz
786                            before this one. Map anonymous pages, if needed,
787                            and clear the area.  */
788                         retval = set_brk (elf_bss + load_bias,
789                                           elf_brk + load_bias);
790                         if (retval) {
791                                 send_sig(SIGKILL, current, 0);
792                                 goto out_free_dentry;
793                         }
794                         nbyte = ELF_PAGEOFFSET(elf_bss);
795                         if (nbyte) {
796                                 nbyte = ELF_MIN_ALIGN - nbyte;
797                                 if (nbyte > elf_brk - elf_bss)
798                                         nbyte = elf_brk - elf_bss;
799                                 if (clear_user((void __user *)elf_bss +
800                                                         load_bias, nbyte)) {
801                                         /*
802                                          * This bss-zeroing can fail if the ELF
803                                          * file specifies odd protections. So
804                                          * we don't check the return value
805                                          */
806                                 }
807                         }
808                 }
809
810                 if (elf_ppnt->p_flags & PF_R)
811                         elf_prot |= PROT_READ;
812                 if (elf_ppnt->p_flags & PF_W)
813                         elf_prot |= PROT_WRITE;
814                 if (elf_ppnt->p_flags & PF_X)
815                         elf_prot |= PROT_EXEC;
816
817                 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
818
819                 vaddr = elf_ppnt->p_vaddr;
820                 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
821                         elf_flags |= MAP_FIXED;
822                 } else if (loc->elf_ex.e_type == ET_DYN) {
823                         /* Try and get dynamic programs out of the way of the
824                          * default mmap base, as well as whatever program they
825                          * might try to exec.  This is because the brk will
826                          * follow the loader, and is not movable.  */
827 #ifdef CONFIG_X86
828                         load_bias = 0;
829 #else
830                         load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
831 #endif
832                 }
833
834                 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
835                                 elf_prot, elf_flags, 0);
836                 if (BAD_ADDR(error)) {
837                         send_sig(SIGKILL, current, 0);
838                         retval = IS_ERR((void *)error) ?
839                                 PTR_ERR((void*)error) : -EINVAL;
840                         goto out_free_dentry;
841                 }
842
843                 if (!load_addr_set) {
844                         load_addr_set = 1;
845                         load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
846                         if (loc->elf_ex.e_type == ET_DYN) {
847                                 load_bias += error -
848                                              ELF_PAGESTART(load_bias + vaddr);
849                                 load_addr += load_bias;
850                                 reloc_func_desc = load_bias;
851                         }
852                 }
853                 k = elf_ppnt->p_vaddr;
854                 if (k < start_code)
855                         start_code = k;
856                 if (start_data < k)
857                         start_data = k;
858
859                 /*
860                  * Check to see if the section's size will overflow the
861                  * allowed task size. Note that p_filesz must always be
862                  * <= p_memsz so it is only necessary to check p_memsz.
863                  */
864                 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
865                     elf_ppnt->p_memsz > TASK_SIZE ||
866                     TASK_SIZE - elf_ppnt->p_memsz < k) {
867                         /* set_brk can never work. Avoid overflows. */
868                         send_sig(SIGKILL, current, 0);
869                         retval = -EINVAL;
870                         goto out_free_dentry;
871                 }
872
873                 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
874
875                 if (k > elf_bss)
876                         elf_bss = k;
877                 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
878                         end_code = k;
879                 if (end_data < k)
880                         end_data = k;
881                 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
882                 if (k > elf_brk)
883                         elf_brk = k;
884         }
885
886         loc->elf_ex.e_entry += load_bias;
887         elf_bss += load_bias;
888         elf_brk += load_bias;
889         start_code += load_bias;
890         end_code += load_bias;
891         start_data += load_bias;
892         end_data += load_bias;
893
894         /* Calling set_brk effectively mmaps the pages that we need
895          * for the bss and break sections.  We must do this before
896          * mapping in the interpreter, to make sure it doesn't wind
897          * up getting placed where the bss needs to go.
898          */
899         retval = set_brk(elf_bss, elf_brk);
900         if (retval) {
901                 send_sig(SIGKILL, current, 0);
902                 goto out_free_dentry;
903         }
904         if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
905                 send_sig(SIGSEGV, current, 0);
906                 retval = -EFAULT; /* Nobody gets to see this, but.. */
907                 goto out_free_dentry;
908         }
909
910         if (elf_interpreter) {
911                 unsigned long uninitialized_var(interp_map_addr);
912
913                 elf_entry = load_elf_interp(&loc->interp_elf_ex,
914                                             interpreter,
915                                             &interp_map_addr,
916                                             load_bias);
917                 if (!IS_ERR((void *)elf_entry)) {
918                         /*
919                          * load_elf_interp() returns relocation
920                          * adjustment
921                          */
922                         interp_load_addr = elf_entry;
923                         elf_entry += loc->interp_elf_ex.e_entry;
924                 }
925                 if (BAD_ADDR(elf_entry)) {
926                         force_sig(SIGSEGV, current);
927                         retval = IS_ERR((void *)elf_entry) ?
928                                         (int)elf_entry : -EINVAL;
929                         goto out_free_dentry;
930                 }
931                 reloc_func_desc = interp_load_addr;
932
933                 allow_write_access(interpreter);
934                 fput(interpreter);
935                 kfree(elf_interpreter);
936         } else {
937                 elf_entry = loc->elf_ex.e_entry;
938                 if (BAD_ADDR(elf_entry)) {
939                         force_sig(SIGSEGV, current);
940                         retval = -EINVAL;
941                         goto out_free_dentry;
942                 }
943         }
944
945         kfree(elf_phdata);
946
947         sys_close(elf_exec_fileno);
948
949         set_binfmt(&elf_format);
950
951 #ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
952         retval = arch_setup_additional_pages(bprm, executable_stack);
953         if (retval < 0) {
954                 send_sig(SIGKILL, current, 0);
955                 goto out;
956         }
957 #endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
958
959         compute_creds(bprm);
960         current->flags &= ~PF_FORKNOEXEC;
961         retval = create_elf_tables(bprm, &loc->elf_ex,
962                           load_addr, interp_load_addr);
963         if (retval < 0) {
964                 send_sig(SIGKILL, current, 0);
965                 goto out;
966         }
967         /* N.B. passed_fileno might not be initialized? */
968         current->mm->end_code = end_code;
969         current->mm->start_code = start_code;
970         current->mm->start_data = start_data;
971         current->mm->end_data = end_data;
972         current->mm->start_stack = bprm->p;
973
974 #ifdef arch_randomize_brk
975         if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1))
976                 current->mm->brk = current->mm->start_brk =
977                         arch_randomize_brk(current->mm);
978 #endif
979
980         if (current->personality & MMAP_PAGE_ZERO) {
981                 /* Why this, you ask???  Well SVr4 maps page 0 as read-only,
982                    and some applications "depend" upon this behavior.
983                    Since we do not have the power to recompile these, we
984                    emulate the SVr4 behavior. Sigh. */
985                 down_write(&current->mm->mmap_sem);
986                 error = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
987                                 MAP_FIXED | MAP_PRIVATE, 0);
988                 up_write(&current->mm->mmap_sem);
989         }
990
991 #ifdef ELF_PLAT_INIT
992         /*
993          * The ABI may specify that certain registers be set up in special
994          * ways (on i386 %edx is the address of a DT_FINI function, for
995          * example.  In addition, it may also specify (eg, PowerPC64 ELF)
996          * that the e_entry field is the address of the function descriptor
997          * for the startup routine, rather than the address of the startup
998          * routine itself.  This macro performs whatever initialization to
999          * the regs structure is required as well as any relocations to the
1000          * function descriptor entries when executing dynamically links apps.
1001          */
1002         ELF_PLAT_INIT(regs, reloc_func_desc);
1003 #endif
1004
1005         start_thread(regs, elf_entry, bprm->p);
1006         if (unlikely(current->ptrace & PT_PTRACED)) {
1007                 if (current->ptrace & PT_TRACE_EXEC)
1008                         ptrace_notify ((PTRACE_EVENT_EXEC << 8) | SIGTRAP);
1009                 else
1010                         send_sig(SIGTRAP, current, 0);
1011         }
1012         retval = 0;
1013 out:
1014         kfree(loc);
1015 out_ret:
1016         return retval;
1017
1018         /* error cleanup */
1019 out_free_dentry:
1020         allow_write_access(interpreter);
1021         if (interpreter)
1022                 fput(interpreter);
1023 out_free_interp:
1024         kfree(elf_interpreter);
1025 out_free_file:
1026         sys_close(elf_exec_fileno);
1027 out_free_ph:
1028         kfree(elf_phdata);
1029         goto out;
1030 }
1031
1032 /* This is really simpleminded and specialized - we are loading an
1033    a.out library that is given an ELF header. */
1034 static int load_elf_library(struct file *file)
1035 {
1036         struct elf_phdr *elf_phdata;
1037         struct elf_phdr *eppnt;
1038         unsigned long elf_bss, bss, len;
1039         int retval, error, i, j;
1040         struct elfhdr elf_ex;
1041
1042         error = -ENOEXEC;
1043         retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
1044         if (retval != sizeof(elf_ex))
1045                 goto out;
1046
1047         if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1048                 goto out;
1049
1050         /* First of all, some simple consistency checks */
1051         if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
1052             !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
1053                 goto out;
1054
1055         /* Now read in all of the header information */
1056
1057         j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1058         /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1059
1060         error = -ENOMEM;
1061         elf_phdata = kmalloc(j, GFP_KERNEL);
1062         if (!elf_phdata)
1063                 goto out;
1064
1065         eppnt = elf_phdata;
1066         error = -ENOEXEC;
1067         retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1068         if (retval != j)
1069                 goto out_free_ph;
1070
1071         for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1072                 if ((eppnt + i)->p_type == PT_LOAD)
1073                         j++;
1074         if (j != 1)
1075                 goto out_free_ph;
1076
1077         while (eppnt->p_type != PT_LOAD)
1078                 eppnt++;
1079
1080         /* Now use mmap to map the library into memory. */
1081         down_write(&current->mm->mmap_sem);
1082         error = do_mmap(file,
1083                         ELF_PAGESTART(eppnt->p_vaddr),
1084                         (eppnt->p_filesz +
1085                          ELF_PAGEOFFSET(eppnt->p_vaddr)),
1086                         PROT_READ | PROT_WRITE | PROT_EXEC,
1087                         MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1088                         (eppnt->p_offset -
1089                          ELF_PAGEOFFSET(eppnt->p_vaddr)));
1090         up_write(&current->mm->mmap_sem);
1091         if (error != ELF_PAGESTART(eppnt->p_vaddr))
1092                 goto out_free_ph;
1093
1094         elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1095         if (padzero(elf_bss)) {
1096                 error = -EFAULT;
1097                 goto out_free_ph;
1098         }
1099
1100         len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1101                             ELF_MIN_ALIGN - 1);
1102         bss = eppnt->p_memsz + eppnt->p_vaddr;
1103         if (bss > len) {
1104                 down_write(&current->mm->mmap_sem);
1105                 do_brk(len, bss - len);
1106                 up_write(&current->mm->mmap_sem);
1107         }
1108         error = 0;
1109
1110 out_free_ph:
1111         kfree(elf_phdata);
1112 out:
1113         return error;
1114 }
1115
1116 /*
1117  * Note that some platforms still use traditional core dumps and not
1118  * the ELF core dump.  Each platform can select it as appropriate.
1119  */
1120 #if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
1121
1122 /*
1123  * ELF core dumper
1124  *
1125  * Modelled on fs/exec.c:aout_core_dump()
1126  * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1127  */
1128 /*
1129  * These are the only things you should do on a core-file: use only these
1130  * functions to write out all the necessary info.
1131  */
1132 static int dump_write(struct file *file, const void *addr, int nr)
1133 {
1134         return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1135 }
1136
1137 static int dump_seek(struct file *file, loff_t off)
1138 {
1139         if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
1140                 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
1141                         return 0;
1142         } else {
1143                 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
1144                 if (!buf)
1145                         return 0;
1146                 while (off > 0) {
1147                         unsigned long n = off;
1148                         if (n > PAGE_SIZE)
1149                                 n = PAGE_SIZE;
1150                         if (!dump_write(file, buf, n))
1151                                 return 0;
1152                         off -= n;
1153                 }
1154                 free_page((unsigned long)buf);
1155         }
1156         return 1;
1157 }
1158
1159 /*
1160  * Decide what to dump of a segment, part, all or none.
1161  */
1162 static unsigned long vma_dump_size(struct vm_area_struct *vma,
1163                                    unsigned long mm_flags)
1164 {
1165         /* The vma can be set up to tell us the answer directly.  */
1166         if (vma->vm_flags & VM_ALWAYSDUMP)
1167                 goto whole;
1168
1169         /* Do not dump I/O mapped devices or special mappings */
1170         if (vma->vm_flags & (VM_IO | VM_RESERVED))
1171                 return 0;
1172
1173 #define FILTER(type)    (mm_flags & (1UL << MMF_DUMP_##type))
1174
1175         /* By default, dump shared memory if mapped from an anonymous file. */
1176         if (vma->vm_flags & VM_SHARED) {
1177                 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0 ?
1178                     FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1179                         goto whole;
1180                 return 0;
1181         }
1182
1183         /* Dump segments that have been written to.  */
1184         if (vma->anon_vma && FILTER(ANON_PRIVATE))
1185                 goto whole;
1186         if (vma->vm_file == NULL)
1187                 return 0;
1188
1189         if (FILTER(MAPPED_PRIVATE))
1190                 goto whole;
1191
1192         /*
1193          * If this looks like the beginning of a DSO or executable mapping,
1194          * check for an ELF header.  If we find one, dump the first page to
1195          * aid in determining what was mapped here.
1196          */
1197         if (FILTER(ELF_HEADERS) && vma->vm_file != NULL && vma->vm_pgoff == 0) {
1198                 u32 __user *header = (u32 __user *) vma->vm_start;
1199                 u32 word;
1200                 /*
1201                  * Doing it this way gets the constant folded by GCC.
1202                  */
1203                 union {
1204                         u32 cmp;
1205                         char elfmag[SELFMAG];
1206                 } magic;
1207                 BUILD_BUG_ON(SELFMAG != sizeof word);
1208                 magic.elfmag[EI_MAG0] = ELFMAG0;
1209                 magic.elfmag[EI_MAG1] = ELFMAG1;
1210                 magic.elfmag[EI_MAG2] = ELFMAG2;
1211                 magic.elfmag[EI_MAG3] = ELFMAG3;
1212                 if (get_user(word, header) == 0 && word == magic.cmp)
1213                         return PAGE_SIZE;
1214         }
1215
1216 #undef  FILTER
1217
1218         return 0;
1219
1220 whole:
1221         return vma->vm_end - vma->vm_start;
1222 }
1223
1224 /* An ELF note in memory */
1225 struct memelfnote
1226 {
1227         const char *name;
1228         int type;
1229         unsigned int datasz;
1230         void *data;
1231 };
1232
1233 static int notesize(struct memelfnote *en)
1234 {
1235         int sz;
1236
1237         sz = sizeof(struct elf_note);
1238         sz += roundup(strlen(en->name) + 1, 4);
1239         sz += roundup(en->datasz, 4);
1240
1241         return sz;
1242 }
1243
1244 #define DUMP_WRITE(addr, nr, foffset)   \
1245         do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
1246
1247 static int alignfile(struct file *file, loff_t *foffset)
1248 {
1249         static const char buf[4] = { 0, };
1250         DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1251         return 1;
1252 }
1253
1254 static int writenote(struct memelfnote *men, struct file *file,
1255                         loff_t *foffset)
1256 {
1257         struct elf_note en;
1258         en.n_namesz = strlen(men->name) + 1;
1259         en.n_descsz = men->datasz;
1260         en.n_type = men->type;
1261
1262         DUMP_WRITE(&en, sizeof(en), foffset);
1263         DUMP_WRITE(men->name, en.n_namesz, foffset);
1264         if (!alignfile(file, foffset))
1265                 return 0;
1266         DUMP_WRITE(men->data, men->datasz, foffset);
1267         if (!alignfile(file, foffset))
1268                 return 0;
1269
1270         return 1;
1271 }
1272 #undef DUMP_WRITE
1273
1274 #define DUMP_WRITE(addr, nr)    \
1275         if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
1276                 goto end_coredump;
1277 #define DUMP_SEEK(off)  \
1278         if (!dump_seek(file, (off))) \
1279                 goto end_coredump;
1280
1281 static void fill_elf_header(struct elfhdr *elf, int segs,
1282                             u16 machine, u32 flags, u8 osabi)
1283 {
1284         memset(elf, 0, sizeof(*elf));
1285
1286         memcpy(elf->e_ident, ELFMAG, SELFMAG);
1287         elf->e_ident[EI_CLASS] = ELF_CLASS;
1288         elf->e_ident[EI_DATA] = ELF_DATA;
1289         elf->e_ident[EI_VERSION] = EV_CURRENT;
1290         elf->e_ident[EI_OSABI] = ELF_OSABI;
1291
1292         elf->e_type = ET_CORE;
1293         elf->e_machine = machine;
1294         elf->e_version = EV_CURRENT;
1295         elf->e_phoff = sizeof(struct elfhdr);
1296         elf->e_flags = flags;
1297         elf->e_ehsize = sizeof(struct elfhdr);
1298         elf->e_phentsize = sizeof(struct elf_phdr);
1299         elf->e_phnum = segs;
1300
1301         return;
1302 }
1303
1304 static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
1305 {
1306         phdr->p_type = PT_NOTE;
1307         phdr->p_offset = offset;
1308         phdr->p_vaddr = 0;
1309         phdr->p_paddr = 0;
1310         phdr->p_filesz = sz;
1311         phdr->p_memsz = 0;
1312         phdr->p_flags = 0;
1313         phdr->p_align = 0;
1314         return;
1315 }
1316
1317 static void fill_note(struct memelfnote *note, const char *name, int type, 
1318                 unsigned int sz, void *data)
1319 {
1320         note->name = name;
1321         note->type = type;
1322         note->datasz = sz;
1323         note->data = data;
1324         return;
1325 }
1326
1327 /*
1328  * fill up all the fields in prstatus from the given task struct, except
1329  * registers which need to be filled up separately.
1330  */
1331 static void fill_prstatus(struct elf_prstatus *prstatus,
1332                 struct task_struct *p, long signr)
1333 {
1334         prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1335         prstatus->pr_sigpend = p->pending.signal.sig[0];
1336         prstatus->pr_sighold = p->blocked.sig[0];
1337         prstatus->pr_pid = task_pid_vnr(p);
1338         prstatus->pr_ppid = task_pid_vnr(p->real_parent);
1339         prstatus->pr_pgrp = task_pgrp_vnr(p);
1340         prstatus->pr_sid = task_session_vnr(p);
1341         if (thread_group_leader(p)) {
1342                 /*
1343                  * This is the record for the group leader.  Add in the
1344                  * cumulative times of previous dead threads.  This total
1345                  * won't include the time of each live thread whose state
1346                  * is included in the core dump.  The final total reported
1347                  * to our parent process when it calls wait4 will include
1348                  * those sums as well as the little bit more time it takes
1349                  * this and each other thread to finish dying after the
1350                  * core dump synchronization phase.
1351                  */
1352                 cputime_to_timeval(cputime_add(p->utime, p->signal->utime),
1353                                    &prstatus->pr_utime);
1354                 cputime_to_timeval(cputime_add(p->stime, p->signal->stime),
1355                                    &prstatus->pr_stime);
1356         } else {
1357                 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1358                 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1359         }
1360         cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1361         cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1362 }
1363
1364 static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1365                        struct mm_struct *mm)
1366 {
1367         unsigned int i, len;
1368         
1369         /* first copy the parameters from user space */
1370         memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1371
1372         len = mm->arg_end - mm->arg_start;
1373         if (len >= ELF_PRARGSZ)
1374                 len = ELF_PRARGSZ-1;
1375         if (copy_from_user(&psinfo->pr_psargs,
1376                            (const char __user *)mm->arg_start, len))
1377                 return -EFAULT;
1378         for(i = 0; i < len; i++)
1379                 if (psinfo->pr_psargs[i] == 0)
1380                         psinfo->pr_psargs[i] = ' ';
1381         psinfo->pr_psargs[len] = 0;
1382
1383         psinfo->pr_pid = task_pid_vnr(p);
1384         psinfo->pr_ppid = task_pid_vnr(p->real_parent);
1385         psinfo->pr_pgrp = task_pgrp_vnr(p);
1386         psinfo->pr_sid = task_session_vnr(p);
1387
1388         i = p->state ? ffz(~p->state) + 1 : 0;
1389         psinfo->pr_state = i;
1390         psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
1391         psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1392         psinfo->pr_nice = task_nice(p);
1393         psinfo->pr_flag = p->flags;
1394         SET_UID(psinfo->pr_uid, p->uid);
1395         SET_GID(psinfo->pr_gid, p->gid);
1396         strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1397         
1398         return 0;
1399 }
1400
1401 static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1402 {
1403         elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1404         int i = 0;
1405         do
1406                 i += 2;
1407         while (auxv[i - 2] != AT_NULL);
1408         fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1409 }
1410
1411 #ifdef CORE_DUMP_USE_REGSET
1412 #include <linux/regset.h>
1413
1414 struct elf_thread_core_info {
1415         struct elf_thread_core_info *next;
1416         struct task_struct *task;
1417         struct elf_prstatus prstatus;
1418         struct memelfnote notes[0];
1419 };
1420
1421 struct elf_note_info {
1422         struct elf_thread_core_info *thread;
1423         struct memelfnote psinfo;
1424         struct memelfnote auxv;
1425         size_t size;
1426         int thread_notes;
1427 };
1428
1429 /*
1430  * When a regset has a writeback hook, we call it on each thread before
1431  * dumping user memory.  On register window machines, this makes sure the
1432  * user memory backing the register data is up to date before we read it.
1433  */
1434 static void do_thread_regset_writeback(struct task_struct *task,
1435                                        const struct user_regset *regset)
1436 {
1437         if (regset->writeback)
1438                 regset->writeback(task, regset, 1);
1439 }
1440
1441 static int fill_thread_core_info(struct elf_thread_core_info *t,
1442                                  const struct user_regset_view *view,
1443                                  long signr, size_t *total)
1444 {
1445         unsigned int i;
1446
1447         /*
1448          * NT_PRSTATUS is the one special case, because the regset data
1449          * goes into the pr_reg field inside the note contents, rather
1450          * than being the whole note contents.  We fill the reset in here.
1451          * We assume that regset 0 is NT_PRSTATUS.
1452          */
1453         fill_prstatus(&t->prstatus, t->task, signr);
1454         (void) view->regsets[0].get(t->task, &view->regsets[0],
1455                                     0, sizeof(t->prstatus.pr_reg),
1456                                     &t->prstatus.pr_reg, NULL);
1457
1458         fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1459                   sizeof(t->prstatus), &t->prstatus);
1460         *total += notesize(&t->notes[0]);
1461
1462         do_thread_regset_writeback(t->task, &view->regsets[0]);
1463
1464         /*
1465          * Each other regset might generate a note too.  For each regset
1466          * that has no core_note_type or is inactive, we leave t->notes[i]
1467          * all zero and we'll know to skip writing it later.
1468          */
1469         for (i = 1; i < view->n; ++i) {
1470                 const struct user_regset *regset = &view->regsets[i];
1471                 do_thread_regset_writeback(t->task, regset);
1472                 if (regset->core_note_type &&
1473                     (!regset->active || regset->active(t->task, regset))) {
1474                         int ret;
1475                         size_t size = regset->n * regset->size;
1476                         void *data = kmalloc(size, GFP_KERNEL);
1477                         if (unlikely(!data))
1478                                 return 0;
1479                         ret = regset->get(t->task, regset,
1480                                           0, size, data, NULL);
1481                         if (unlikely(ret))
1482                                 kfree(data);
1483                         else {
1484                                 if (regset->core_note_type != NT_PRFPREG)
1485                                         fill_note(&t->notes[i], "LINUX",
1486                                                   regset->core_note_type,
1487                                                   size, data);
1488                                 else {
1489                                         t->prstatus.pr_fpvalid = 1;
1490                                         fill_note(&t->notes[i], "CORE",
1491                                                   NT_PRFPREG, size, data);
1492                                 }
1493                                 *total += notesize(&t->notes[i]);
1494                         }
1495                 }
1496         }
1497
1498         return 1;
1499 }
1500
1501 static int fill_note_info(struct elfhdr *elf, int phdrs,
1502                           struct elf_note_info *info,
1503                           long signr, struct pt_regs *regs)
1504 {
1505         struct task_struct *dump_task = current;
1506         const struct user_regset_view *view = task_user_regset_view(dump_task);
1507         struct elf_thread_core_info *t;
1508         struct elf_prpsinfo *psinfo;
1509         struct core_thread *ct;
1510         unsigned int i;
1511
1512         info->size = 0;
1513         info->thread = NULL;
1514
1515         psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1516         fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1517
1518         if (psinfo == NULL)
1519                 return 0;
1520
1521         /*
1522          * Figure out how many notes we're going to need for each thread.
1523          */
1524         info->thread_notes = 0;
1525         for (i = 0; i < view->n; ++i)
1526                 if (view->regsets[i].core_note_type != 0)
1527                         ++info->thread_notes;
1528
1529         /*
1530          * Sanity check.  We rely on regset 0 being in NT_PRSTATUS,
1531          * since it is our one special case.
1532          */
1533         if (unlikely(info->thread_notes == 0) ||
1534             unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1535                 WARN_ON(1);
1536                 return 0;
1537         }
1538
1539         /*
1540          * Initialize the ELF file header.
1541          */
1542         fill_elf_header(elf, phdrs,
1543                         view->e_machine, view->e_flags, view->ei_osabi);
1544
1545         /*
1546          * Allocate a structure for each thread.
1547          */
1548         for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1549                 t = kzalloc(offsetof(struct elf_thread_core_info,
1550                                      notes[info->thread_notes]),
1551                             GFP_KERNEL);
1552                 if (unlikely(!t))
1553                         return 0;
1554
1555                 t->task = ct->task;
1556                 if (ct->task == dump_task || !info->thread) {
1557                         t->next = info->thread;
1558                         info->thread = t;
1559                 } else {
1560                         /*
1561                          * Make sure to keep the original task at
1562                          * the head of the list.
1563                          */
1564                         t->next = info->thread->next;
1565                         info->thread->next = t;
1566                 }
1567         }
1568
1569         /*
1570          * Now fill in each thread's information.
1571          */
1572         for (t = info->thread; t != NULL; t = t->next)
1573                 if (!fill_thread_core_info(t, view, signr, &info->size))
1574                         return 0;
1575
1576         /*
1577          * Fill in the two process-wide notes.
1578          */
1579         fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1580         info->size += notesize(&info->psinfo);
1581
1582         fill_auxv_note(&info->auxv, current->mm);
1583         info->size += notesize(&info->auxv);
1584
1585         return 1;
1586 }
1587
1588 static size_t get_note_info_size(struct elf_note_info *info)
1589 {
1590         return info->size;
1591 }
1592
1593 /*
1594  * Write all the notes for each thread.  When writing the first thread, the
1595  * process-wide notes are interleaved after the first thread-specific note.
1596  */
1597 static int write_note_info(struct elf_note_info *info,
1598                            struct file *file, loff_t *foffset)
1599 {
1600         bool first = 1;
1601         struct elf_thread_core_info *t = info->thread;
1602
1603         do {
1604                 int i;
1605
1606                 if (!writenote(&t->notes[0], file, foffset))
1607                         return 0;
1608
1609                 if (first && !writenote(&info->psinfo, file, foffset))
1610                         return 0;
1611                 if (first && !writenote(&info->auxv, file, foffset))
1612                         return 0;
1613
1614                 for (i = 1; i < info->thread_notes; ++i)
1615                         if (t->notes[i].data &&
1616                             !writenote(&t->notes[i], file, foffset))
1617                                 return 0;
1618
1619                 first = 0;
1620                 t = t->next;
1621         } while (t);
1622
1623         return 1;
1624 }
1625
1626 static void free_note_info(struct elf_note_info *info)
1627 {
1628         struct elf_thread_core_info *threads = info->thread;
1629         while (threads) {
1630                 unsigned int i;
1631                 struct elf_thread_core_info *t = threads;
1632                 threads = t->next;
1633                 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1634                 for (i = 1; i < info->thread_notes; ++i)
1635                         kfree(t->notes[i].data);
1636                 kfree(t);
1637         }
1638         kfree(info->psinfo.data);
1639 }
1640
1641 #else
1642
1643 /* Here is the structure in which status of each thread is captured. */
1644 struct elf_thread_status
1645 {
1646         struct list_head list;
1647         struct elf_prstatus prstatus;   /* NT_PRSTATUS */
1648         elf_fpregset_t fpu;             /* NT_PRFPREG */
1649         struct task_struct *thread;
1650 #ifdef ELF_CORE_COPY_XFPREGS
1651         elf_fpxregset_t xfpu;           /* ELF_CORE_XFPREG_TYPE */
1652 #endif
1653         struct memelfnote notes[3];
1654         int num_notes;
1655 };
1656
1657 /*
1658  * In order to add the specific thread information for the elf file format,
1659  * we need to keep a linked list of every threads pr_status and then create
1660  * a single section for them in the final core file.
1661  */
1662 static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1663 {
1664         int sz = 0;
1665         struct task_struct *p = t->thread;
1666         t->num_notes = 0;
1667
1668         fill_prstatus(&t->prstatus, p, signr);
1669         elf_core_copy_task_regs(p, &t->prstatus.pr_reg);        
1670         
1671         fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1672                   &(t->prstatus));
1673         t->num_notes++;
1674         sz += notesize(&t->notes[0]);
1675
1676         if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1677                                                                 &t->fpu))) {
1678                 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1679                           &(t->fpu));
1680                 t->num_notes++;
1681                 sz += notesize(&t->notes[1]);
1682         }
1683
1684 #ifdef ELF_CORE_COPY_XFPREGS
1685         if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
1686                 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1687                           sizeof(t->xfpu), &t->xfpu);
1688                 t->num_notes++;
1689                 sz += notesize(&t->notes[2]);
1690         }
1691 #endif  
1692         return sz;
1693 }
1694
1695 struct elf_note_info {
1696         struct memelfnote *notes;
1697         struct elf_prstatus *prstatus;  /* NT_PRSTATUS */
1698         struct elf_prpsinfo *psinfo;    /* NT_PRPSINFO */
1699         struct list_head thread_list;
1700         elf_fpregset_t *fpu;
1701 #ifdef ELF_CORE_COPY_XFPREGS
1702         elf_fpxregset_t *xfpu;
1703 #endif
1704         int thread_status_size;
1705         int numnote;
1706 };
1707
1708 static int fill_note_info(struct elfhdr *elf, int phdrs,
1709                           struct elf_note_info *info,
1710                           long signr, struct pt_regs *regs)
1711 {
1712 #define NUM_NOTES       6
1713         struct list_head *t;
1714
1715         info->notes = NULL;
1716         info->prstatus = NULL;
1717         info->psinfo = NULL;
1718         info->fpu = NULL;
1719 #ifdef ELF_CORE_COPY_XFPREGS
1720         info->xfpu = NULL;
1721 #endif
1722         INIT_LIST_HEAD(&info->thread_list);
1723
1724         info->notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote),
1725                               GFP_KERNEL);
1726         if (!info->notes)
1727                 return 0;
1728         info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1729         if (!info->psinfo)
1730                 return 0;
1731         info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1732         if (!info->prstatus)
1733                 return 0;
1734         info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1735         if (!info->fpu)
1736                 return 0;
1737 #ifdef ELF_CORE_COPY_XFPREGS
1738         info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1739         if (!info->xfpu)
1740                 return 0;
1741 #endif
1742
1743         info->thread_status_size = 0;
1744         if (signr) {
1745                 struct core_thread *ct;
1746                 struct elf_thread_status *ets;
1747
1748                 for (ct = current->mm->core_state->dumper.next;
1749                                                 ct; ct = ct->next) {
1750                         ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1751                         if (!ets)
1752                                 return 0;
1753
1754                         ets->thread = ct->task;
1755                         list_add(&ets->list, &info->thread_list);
1756                 }
1757
1758                 list_for_each(t, &info->thread_list) {
1759                         int sz;
1760
1761                         ets = list_entry(t, struct elf_thread_status, list);
1762                         sz = elf_dump_thread_status(signr, ets);
1763                         info->thread_status_size += sz;
1764                 }
1765         }
1766         /* now collect the dump for the current */
1767         memset(info->prstatus, 0, sizeof(*info->prstatus));
1768         fill_prstatus(info->prstatus, current, signr);
1769         elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1770
1771         /* Set up header */
1772         fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS, ELF_OSABI);
1773
1774         /*
1775          * Set up the notes in similar form to SVR4 core dumps made
1776          * with info from their /proc.
1777          */
1778
1779         fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1780                   sizeof(*info->prstatus), info->prstatus);
1781         fill_psinfo(info->psinfo, current->group_leader, current->mm);
1782         fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1783                   sizeof(*info->psinfo), info->psinfo);
1784
1785         info->numnote = 2;
1786
1787         fill_auxv_note(&info->notes[info->numnote++], current->mm);
1788
1789         /* Try to dump the FPU. */
1790         info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1791                                                                info->fpu);
1792         if (info->prstatus->pr_fpvalid)
1793                 fill_note(info->notes + info->numnote++,
1794                           "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1795 #ifdef ELF_CORE_COPY_XFPREGS
1796         if (elf_core_copy_task_xfpregs(current, info->xfpu))
1797                 fill_note(info->notes + info->numnote++,
1798                           "LINUX", ELF_CORE_XFPREG_TYPE,
1799                           sizeof(*info->xfpu), info->xfpu);
1800 #endif
1801
1802         return 1;
1803
1804 #undef NUM_NOTES
1805 }
1806
1807 static size_t get_note_info_size(struct elf_note_info *info)
1808 {
1809         int sz = 0;
1810         int i;
1811
1812         for (i = 0; i < info->numnote; i++)
1813                 sz += notesize(info->notes + i);
1814
1815         sz += info->thread_status_size;
1816
1817         return sz;
1818 }
1819
1820 static int write_note_info(struct elf_note_info *info,
1821                            struct file *file, loff_t *foffset)
1822 {
1823         int i;
1824         struct list_head *t;
1825
1826         for (i = 0; i < info->numnote; i++)
1827                 if (!writenote(info->notes + i, file, foffset))
1828                         return 0;
1829
1830         /* write out the thread status notes section */
1831         list_for_each(t, &info->thread_list) {
1832                 struct elf_thread_status *tmp =
1833                                 list_entry(t, struct elf_thread_status, list);
1834
1835                 for (i = 0; i < tmp->num_notes; i++)
1836                         if (!writenote(&tmp->notes[i], file, foffset))
1837                                 return 0;
1838         }
1839
1840         return 1;
1841 }
1842
1843 static void free_note_info(struct elf_note_info *info)
1844 {
1845         while (!list_empty(&info->thread_list)) {
1846                 struct list_head *tmp = info->thread_list.next;
1847                 list_del(tmp);
1848                 kfree(list_entry(tmp, struct elf_thread_status, list));
1849         }
1850
1851         kfree(info->prstatus);
1852         kfree(info->psinfo);
1853         kfree(info->notes);
1854         kfree(info->fpu);
1855 #ifdef ELF_CORE_COPY_XFPREGS
1856         kfree(info->xfpu);
1857 #endif
1858 }
1859
1860 #endif
1861
1862 static struct vm_area_struct *first_vma(struct task_struct *tsk,
1863                                         struct vm_area_struct *gate_vma)
1864 {
1865         struct vm_area_struct *ret = tsk->mm->mmap;
1866
1867         if (ret)
1868                 return ret;
1869         return gate_vma;
1870 }
1871 /*
1872  * Helper function for iterating across a vma list.  It ensures that the caller
1873  * will visit `gate_vma' prior to terminating the search.
1874  */
1875 static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1876                                         struct vm_area_struct *gate_vma)
1877 {
1878         struct vm_area_struct *ret;
1879
1880         ret = this_vma->vm_next;
1881         if (ret)
1882                 return ret;
1883         if (this_vma == gate_vma)
1884                 return NULL;
1885         return gate_vma;
1886 }
1887
1888 /*
1889  * Actual dumper
1890  *
1891  * This is a two-pass process; first we find the offsets of the bits,
1892  * and then they are actually written out.  If we run out of core limit
1893  * we just truncate.
1894  */
1895 static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, unsigned long limit)
1896 {
1897         int has_dumped = 0;
1898         mm_segment_t fs;
1899         int segs;
1900         size_t size = 0;
1901         struct vm_area_struct *vma, *gate_vma;
1902         struct elfhdr *elf = NULL;
1903         loff_t offset = 0, dataoff, foffset;
1904         unsigned long mm_flags;
1905         struct elf_note_info info;
1906
1907         /*
1908          * We no longer stop all VM operations.
1909          * 
1910          * This is because those proceses that could possibly change map_count
1911          * or the mmap / vma pages are now blocked in do_exit on current
1912          * finishing this core dump.
1913          *
1914          * Only ptrace can touch these memory addresses, but it doesn't change
1915          * the map_count or the pages allocated. So no possibility of crashing
1916          * exists while dumping the mm->vm_next areas to the core file.
1917          */
1918   
1919         /* alloc memory for large data structures: too large to be on stack */
1920         elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1921         if (!elf)
1922                 goto out;
1923         
1924         segs = current->mm->map_count;
1925 #ifdef ELF_CORE_EXTRA_PHDRS
1926         segs += ELF_CORE_EXTRA_PHDRS;
1927 #endif
1928
1929         gate_vma = get_gate_vma(current);
1930         if (gate_vma != NULL)
1931                 segs++;
1932
1933         /*
1934          * Collect all the non-memory information about the process for the
1935          * notes.  This also sets up the file header.
1936          */
1937         if (!fill_note_info(elf, segs + 1, /* including notes section */
1938                             &info, signr, regs))
1939                 goto cleanup;
1940
1941         has_dumped = 1;
1942         current->flags |= PF_DUMPCORE;
1943   
1944         fs = get_fs();
1945         set_fs(KERNEL_DS);
1946
1947         DUMP_WRITE(elf, sizeof(*elf));
1948         offset += sizeof(*elf);                         /* Elf header */
1949         offset += (segs + 1) * sizeof(struct elf_phdr); /* Program headers */
1950         foffset = offset;
1951
1952         /* Write notes phdr entry */
1953         {
1954                 struct elf_phdr phdr;
1955                 size_t sz = get_note_info_size(&info);
1956
1957                 sz += elf_coredump_extra_notes_size();
1958
1959                 fill_elf_note_phdr(&phdr, sz, offset);
1960                 offset += sz;
1961                 DUMP_WRITE(&phdr, sizeof(phdr));
1962         }
1963
1964         dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1965
1966         /*
1967          * We must use the same mm->flags while dumping core to avoid
1968          * inconsistency between the program headers and bodies, otherwise an
1969          * unusable core file can be generated.
1970          */
1971         mm_flags = current->mm->flags;
1972
1973         /* Write program headers for segments dump */
1974         for (vma = first_vma(current, gate_vma); vma != NULL;
1975                         vma = next_vma(vma, gate_vma)) {
1976                 struct elf_phdr phdr;
1977
1978                 phdr.p_type = PT_LOAD;
1979                 phdr.p_offset = offset;
1980                 phdr.p_vaddr = vma->vm_start;
1981                 phdr.p_paddr = 0;
1982                 phdr.p_filesz = vma_dump_size(vma, mm_flags);
1983                 phdr.p_memsz = vma->vm_end - vma->vm_start;
1984                 offset += phdr.p_filesz;
1985                 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
1986                 if (vma->vm_flags & VM_WRITE)
1987                         phdr.p_flags |= PF_W;
1988                 if (vma->vm_flags & VM_EXEC)
1989                         phdr.p_flags |= PF_X;
1990                 phdr.p_align = ELF_EXEC_PAGESIZE;
1991
1992                 DUMP_WRITE(&phdr, sizeof(phdr));
1993         }
1994
1995 #ifdef ELF_CORE_WRITE_EXTRA_PHDRS
1996         ELF_CORE_WRITE_EXTRA_PHDRS;
1997 #endif
1998
1999         /* write out the notes section */
2000         if (!write_note_info(&info, file, &foffset))
2001                 goto end_coredump;
2002
2003         if (elf_coredump_extra_notes_write(file, &foffset))
2004                 goto end_coredump;
2005
2006         /* Align to page */
2007         DUMP_SEEK(dataoff - foffset);
2008
2009         for (vma = first_vma(current, gate_vma); vma != NULL;
2010                         vma = next_vma(vma, gate_vma)) {
2011                 unsigned long addr;
2012                 unsigned long end;
2013
2014                 end = vma->vm_start + vma_dump_size(vma, mm_flags);
2015
2016                 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
2017                         struct page *page;
2018                         struct vm_area_struct *tmp_vma;
2019
2020                         if (get_user_pages(current, current->mm, addr, 1, 0, 1,
2021                                                 &page, &tmp_vma) <= 0) {
2022                                 DUMP_SEEK(PAGE_SIZE);
2023                         } else {
2024                                 if (page == ZERO_PAGE(0)) {
2025                                         if (!dump_seek(file, PAGE_SIZE)) {
2026                                                 page_cache_release(page);
2027                                                 goto end_coredump;
2028                                         }
2029                                 } else {
2030                                         void *kaddr;
2031                                         flush_cache_page(tmp_vma, addr,
2032                                                          page_to_pfn(page));
2033                                         kaddr = kmap(page);
2034                                         if ((size += PAGE_SIZE) > limit ||
2035                                             !dump_write(file, kaddr,
2036                                             PAGE_SIZE)) {
2037                                                 kunmap(page);
2038                                                 page_cache_release(page);
2039                                                 goto end_coredump;
2040                                         }
2041                                         kunmap(page);
2042                                 }
2043                                 page_cache_release(page);
2044                         }
2045                 }
2046         }
2047
2048 #ifdef ELF_CORE_WRITE_EXTRA_DATA
2049         ELF_CORE_WRITE_EXTRA_DATA;
2050 #endif
2051
2052 end_coredump:
2053         set_fs(fs);
2054
2055 cleanup:
2056         free_note_info(&info);
2057         kfree(elf);
2058 out:
2059         return has_dumped;
2060 }
2061
2062 #endif          /* USE_ELF_CORE_DUMP */
2063
2064 static int __init init_elf_binfmt(void)
2065 {
2066         return register_binfmt(&elf_format);
2067 }
2068
2069 static void __exit exit_elf_binfmt(void)
2070 {
2071         /* Remove the COFF and ELF loaders. */
2072         unregister_binfmt(&elf_format);
2073 }
2074
2075 core_initcall(init_elf_binfmt);
2076 module_exit(exit_elf_binfmt);
2077 MODULE_LICENSE("GPL");