blob: e3ff2b9e602fd8fc8d35b24ae5a6dad841bc05f1 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#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>
Jesper Juhlf4e5cc22006-06-23 02:05:35 -070039#include <linux/elf.h>
Alexey Dobriyan7e80d0d2007-05-08 00:28:59 -070040#include <linux/utsname.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <asm/uaccess.h>
42#include <asm/param.h>
43#include <asm/page.h>
44
Jesper Juhlf4e5cc22006-06-23 02:05:35 -070045static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs);
46static int load_elf_library(struct file *);
Andrew Mortonbb1ad822008-01-30 13:31:07 +010047static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
48 int, int, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Linus Torvalds1da177e2005-04-16 15:20:36 -070050/*
51 * If we don't support core dumping, then supply a NULL so we
52 * don't even try.
53 */
Matt Mackall708e9a72006-01-08 01:05:25 -080054#if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
Neil Horman7dc0b222007-10-16 23:26:34 -070055static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, unsigned long limit);
Linus Torvalds1da177e2005-04-16 15:20:36 -070056#else
57#define elf_core_dump NULL
58#endif
59
60#if ELF_EXEC_PAGESIZE > PAGE_SIZE
Jesper Juhlf4e5cc22006-06-23 02:05:35 -070061#define ELF_MIN_ALIGN ELF_EXEC_PAGESIZE
Linus Torvalds1da177e2005-04-16 15:20:36 -070062#else
Jesper Juhlf4e5cc22006-06-23 02:05:35 -070063#define ELF_MIN_ALIGN PAGE_SIZE
Linus Torvalds1da177e2005-04-16 15:20:36 -070064#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
74static 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,
Andi Kleen9fbbd4d2007-02-13 13:26:26 +010079 .min_coredump = ELF_EXEC_PAGESIZE,
80 .hasvdso = 1
Linus Torvalds1da177e2005-04-16 15:20:36 -070081};
82
Andrew Mortond4e3cc32007-07-21 04:37:32 -070083#define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
85static 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101/* 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
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700104 be in memory
105 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106static 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
Ohad Ben-Cohen09c6dd32008-02-03 18:05:15 +0200119/* Let's use some macros to make this stack manipulation a little clearer */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120#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)
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700124#define STACK_ALLOC(sp, len) ({ \
125 elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
126 old_sp; })
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127#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
Nathan Lynch483fad12008-07-22 04:48:46 +1000134#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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143static int
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700144create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
Andi Kleend20894a2008-02-08 04:21:54 -0800145 unsigned long load_addr, unsigned long interp_load_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146{
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;
Nathan Lynch483fad12008-07-22 04:48:46 +1000154 elf_addr_t __user *u_base_platform;
Kees Cookf06295b2009-01-07 18:08:52 -0800155 elf_addr_t __user *u_rand_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 const char *k_platform = ELF_PLATFORM;
Nathan Lynch483fad12008-07-22 04:48:46 +1000157 const char *k_base_platform = ELF_BASE_PLATFORM;
Kees Cookf06295b2009-01-07 18:08:52 -0800158 unsigned char k_rand_bytes[16];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 int items;
160 elf_addr_t *elf_info;
161 int ei_index = 0;
David Howells86a264a2008-11-14 10:39:18 +1100162 const struct cred *cred = current_cred();
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700163 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
165 /*
Franck Bui-Huud68c9d62007-10-16 23:30:24 -0700166 * In some cases (e.g. Hyper-Threading), we want to avoid L1
167 * evictions by the processes running on the same package. One
168 * thing we can do is to shuffle the initial stack for them.
169 */
170
171 p = arch_align_stack(p);
172
173 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 * If this architecture has a platform capability string, copy it
175 * to userspace. In some cases (Sparc), this info is impossible
176 * for userspace to get any other way, in others (i386) it is
177 * merely difficult.
178 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 u_platform = NULL;
180 if (k_platform) {
181 size_t len = strlen(k_platform) + 1;
182
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
184 if (__copy_to_user(u_platform, k_platform, len))
185 return -EFAULT;
186 }
187
Nathan Lynch483fad12008-07-22 04:48:46 +1000188 /*
189 * If this architecture has a "base" platform capability
190 * string, copy it to userspace.
191 */
192 u_base_platform = NULL;
193 if (k_base_platform) {
194 size_t len = strlen(k_base_platform) + 1;
195
196 u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
197 if (__copy_to_user(u_base_platform, k_base_platform, len))
198 return -EFAULT;
199 }
200
Kees Cookf06295b2009-01-07 18:08:52 -0800201 /*
202 * Generate 16 random bytes for userspace PRNG seeding.
203 */
204 get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
205 u_rand_bytes = (elf_addr_t __user *)
206 STACK_ALLOC(p, sizeof(k_rand_bytes));
207 if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
208 return -EFAULT;
209
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 /* Create the ELF interpreter info */
Jesper Juhl785d5572006-06-23 02:05:35 -0700211 elf_info = (elf_addr_t *)current->mm->saved_auxv;
Olaf Hering4f9a58d2007-10-16 23:30:12 -0700212 /* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213#define NEW_AUX_ENT(id, val) \
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700214 do { \
Jesper Juhl785d5572006-06-23 02:05:35 -0700215 elf_info[ei_index++] = id; \
216 elf_info[ei_index++] = val; \
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700217 } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218
219#ifdef ARCH_DLINFO
220 /*
221 * ARCH_DLINFO must come first so PPC can do its special alignment of
222 * AUXV.
Olaf Hering4f9a58d2007-10-16 23:30:12 -0700223 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
224 * ARCH_DLINFO changes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 */
226 ARCH_DLINFO;
227#endif
228 NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
229 NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
230 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
231 NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700232 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
234 NEW_AUX_ENT(AT_BASE, interp_load_addr);
235 NEW_AUX_ENT(AT_FLAGS, 0);
236 NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
David Howells86a264a2008-11-14 10:39:18 +1100237 NEW_AUX_ENT(AT_UID, cred->uid);
238 NEW_AUX_ENT(AT_EUID, cred->euid);
239 NEW_AUX_ENT(AT_GID, cred->gid);
240 NEW_AUX_ENT(AT_EGID, cred->egid);
Jesper Juhl785d5572006-06-23 02:05:35 -0700241 NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
Kees Cookf06295b2009-01-07 18:08:52 -0800242 NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
John Reiser65191082008-07-21 14:21:32 -0700243 NEW_AUX_ENT(AT_EXECFN, bprm->exec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 if (k_platform) {
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700245 NEW_AUX_ENT(AT_PLATFORM,
Jesper Juhl785d5572006-06-23 02:05:35 -0700246 (elf_addr_t)(unsigned long)u_platform);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 }
Nathan Lynch483fad12008-07-22 04:48:46 +1000248 if (k_base_platform) {
249 NEW_AUX_ENT(AT_BASE_PLATFORM,
250 (elf_addr_t)(unsigned long)u_base_platform);
251 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
Jesper Juhl785d5572006-06-23 02:05:35 -0700253 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 }
255#undef NEW_AUX_ENT
256 /* AT_NULL is zero; clear the rest too */
257 memset(&elf_info[ei_index], 0,
258 sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
259
260 /* And advance past the AT_NULL entry. */
261 ei_index += 2;
262
263 sp = STACK_ADD(p, ei_index);
264
Andi Kleend20894a2008-02-08 04:21:54 -0800265 items = (argc + 1) + (envc + 1) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 bprm->p = STACK_ROUND(sp, items);
267
268 /* Point sp at the lowest address on the stack */
269#ifdef CONFIG_STACK_GROWSUP
270 sp = (elf_addr_t __user *)bprm->p - items - ei_index;
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700271 bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272#else
273 sp = (elf_addr_t __user *)bprm->p;
274#endif
275
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700276
277 /*
278 * Grow the stack manually; some architectures have a limit on how
279 * far ahead a user-space access may be in order to grow the stack.
280 */
281 vma = find_extend_vma(current->mm, bprm->p);
282 if (!vma)
283 return -EFAULT;
284
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 /* Now, let's put argc (and argv, envp if appropriate) on the stack */
286 if (__put_user(argc, sp++))
287 return -EFAULT;
Andi Kleend20894a2008-02-08 04:21:54 -0800288 argv = sp;
289 envp = argv + argc + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290
291 /* Populate argv and envp */
Greg Kroah-Hartmana84a5052005-05-11 00:10:44 -0700292 p = current->mm->arg_end = current->mm->arg_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 while (argc-- > 0) {
294 size_t len;
Heiko Carstens841d5fb2006-12-06 20:36:35 -0800295 if (__put_user((elf_addr_t)p, argv++))
296 return -EFAULT;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700297 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
298 if (!len || len > MAX_ARG_STRLEN)
WANG Cong23c49712008-05-08 21:52:33 +0800299 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 p += len;
301 }
302 if (__put_user(0, argv))
303 return -EFAULT;
304 current->mm->arg_end = current->mm->env_start = p;
305 while (envc-- > 0) {
306 size_t len;
Heiko Carstens841d5fb2006-12-06 20:36:35 -0800307 if (__put_user((elf_addr_t)p, envp++))
308 return -EFAULT;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700309 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
310 if (!len || len > MAX_ARG_STRLEN)
WANG Cong23c49712008-05-08 21:52:33 +0800311 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 p += len;
313 }
314 if (__put_user(0, envp))
315 return -EFAULT;
316 current->mm->env_end = p;
317
318 /* Put the elf_info on the stack in the right place. */
319 sp = (elf_addr_t __user *)envp + 1;
320 if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
321 return -EFAULT;
322 return 0;
323}
324
325#ifndef elf_map
326
327static unsigned long elf_map(struct file *filep, unsigned long addr,
Jiri Kosinacc503c12008-01-30 13:31:07 +0100328 struct elf_phdr *eppnt, int prot, int type,
329 unsigned long total_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330{
331 unsigned long map_addr;
Jiri Kosinacc503c12008-01-30 13:31:07 +0100332 unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
333 unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
334 addr = ELF_PAGESTART(addr);
335 size = ELF_PAGEALIGN(size);
Jan Kratochvil60bfba72007-07-15 23:40:06 -0700336
Andrew Mortond4e3cc32007-07-21 04:37:32 -0700337 /* mmap() will return -EINVAL if given a zero size, but a
338 * segment with zero filesize is perfectly valid */
Jiri Kosinacc503c12008-01-30 13:31:07 +0100339 if (!size)
340 return addr;
341
342 down_write(&current->mm->mmap_sem);
343 /*
344 * total_size is the size of the ELF (interpreter) image.
345 * The _first_ mmap needs to know the full size, otherwise
346 * randomization might put this image into an overlapping
347 * position with the ELF binary image. (since size < total_size)
348 * So we first map the 'big' image - and unmap the remainder at
349 * the end. (which unmap is needed for ELF images with holes.)
350 */
351 if (total_size) {
352 total_size = ELF_PAGEALIGN(total_size);
353 map_addr = do_mmap(filep, addr, total_size, prot, type, off);
354 if (!BAD_ADDR(map_addr))
355 do_munmap(current->mm, map_addr+size, total_size-size);
356 } else
357 map_addr = do_mmap(filep, addr, size, prot, type, off);
358
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 up_write(&current->mm->mmap_sem);
360 return(map_addr);
361}
362
363#endif /* !elf_map */
364
Jiri Kosinacc503c12008-01-30 13:31:07 +0100365static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
366{
367 int i, first_idx = -1, last_idx = -1;
368
369 for (i = 0; i < nr; i++) {
370 if (cmds[i].p_type == PT_LOAD) {
371 last_idx = i;
372 if (first_idx == -1)
373 first_idx = i;
374 }
375 }
376 if (first_idx == -1)
377 return 0;
378
379 return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
380 ELF_PAGESTART(cmds[first_idx].p_vaddr);
381}
382
383
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384/* This is much more generalized than the library routine read function,
385 so we keep this separate. Technically the library read function
386 is only provided so that we can read a.out libraries that have
387 an ELF header */
388
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700389static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
Jiri Kosinacc503c12008-01-30 13:31:07 +0100390 struct file *interpreter, unsigned long *interp_map_addr,
391 unsigned long no_base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392{
393 struct elf_phdr *elf_phdata;
394 struct elf_phdr *eppnt;
395 unsigned long load_addr = 0;
396 int load_addr_set = 0;
397 unsigned long last_bss = 0, elf_bss = 0;
398 unsigned long error = ~0UL;
Jiri Kosinacc503c12008-01-30 13:31:07 +0100399 unsigned long total_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 int retval, i, size;
401
402 /* First of all, some simple consistency checks */
403 if (interp_elf_ex->e_type != ET_EXEC &&
404 interp_elf_ex->e_type != ET_DYN)
405 goto out;
406 if (!elf_check_arch(interp_elf_ex))
407 goto out;
408 if (!interpreter->f_op || !interpreter->f_op->mmap)
409 goto out;
410
411 /*
412 * If the size of this structure has changed, then punt, since
413 * we will be doing the wrong thing.
414 */
415 if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
416 goto out;
417 if (interp_elf_ex->e_phnum < 1 ||
418 interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
419 goto out;
420
421 /* Now read in all of the header information */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
423 if (size > ELF_MIN_ALIGN)
424 goto out;
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700425 elf_phdata = kmalloc(size, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 if (!elf_phdata)
427 goto out;
428
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700429 retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
430 (char *)elf_phdata,size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 error = -EIO;
432 if (retval != size) {
433 if (retval < 0)
434 error = retval;
435 goto out_close;
436 }
437
Jiri Kosinacc503c12008-01-30 13:31:07 +0100438 total_size = total_mapping_size(elf_phdata, interp_elf_ex->e_phnum);
439 if (!total_size) {
440 error = -EINVAL;
441 goto out_close;
442 }
443
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 eppnt = elf_phdata;
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700445 for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
446 if (eppnt->p_type == PT_LOAD) {
447 int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
448 int elf_prot = 0;
449 unsigned long vaddr = 0;
450 unsigned long k, map_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700452 if (eppnt->p_flags & PF_R)
453 elf_prot = PROT_READ;
454 if (eppnt->p_flags & PF_W)
455 elf_prot |= PROT_WRITE;
456 if (eppnt->p_flags & PF_X)
457 elf_prot |= PROT_EXEC;
458 vaddr = eppnt->p_vaddr;
459 if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
460 elf_type |= MAP_FIXED;
Jiri Kosinacc503c12008-01-30 13:31:07 +0100461 else if (no_base && interp_elf_ex->e_type == ET_DYN)
462 load_addr = -vaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700464 map_addr = elf_map(interpreter, load_addr + vaddr,
Andrew Mortonbb1ad822008-01-30 13:31:07 +0100465 eppnt, elf_prot, elf_type, total_size);
Jiri Kosinacc503c12008-01-30 13:31:07 +0100466 total_size = 0;
467 if (!*interp_map_addr)
468 *interp_map_addr = map_addr;
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700469 error = map_addr;
470 if (BAD_ADDR(map_addr))
471 goto out_close;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700473 if (!load_addr_set &&
474 interp_elf_ex->e_type == ET_DYN) {
475 load_addr = map_addr - ELF_PAGESTART(vaddr);
476 load_addr_set = 1;
477 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700479 /*
480 * Check to see if the section's size will overflow the
481 * allowed task size. Note that p_filesz must always be
482 * <= p_memsize so it's only necessary to check p_memsz.
483 */
484 k = load_addr + eppnt->p_vaddr;
Chuck Ebbertce510592006-07-03 00:24:14 -0700485 if (BAD_ADDR(k) ||
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700486 eppnt->p_filesz > eppnt->p_memsz ||
487 eppnt->p_memsz > TASK_SIZE ||
488 TASK_SIZE - eppnt->p_memsz < k) {
489 error = -ENOMEM;
490 goto out_close;
491 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700493 /*
494 * Find the end of the file mapping for this phdr, and
495 * keep track of the largest address we see for this.
496 */
497 k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
498 if (k > elf_bss)
499 elf_bss = k;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700501 /*
502 * Do the same thing for the memory mapping - between
503 * elf_bss and last_bss is the bss section.
504 */
505 k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
506 if (k > last_bss)
507 last_bss = k;
508 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 }
510
511 /*
512 * Now fill out the bss section. First pad the last page up
513 * to the page boundary, and then perform a mmap to make sure
514 * that there are zero-mapped pages up to and including the
515 * last bss page.
516 */
517 if (padzero(elf_bss)) {
518 error = -EFAULT;
519 goto out_close;
520 }
521
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700522 /* What we have mapped so far */
523 elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
525 /* Map the last of the bss segment */
526 if (last_bss > elf_bss) {
527 down_write(&current->mm->mmap_sem);
528 error = do_brk(elf_bss, last_bss - elf_bss);
529 up_write(&current->mm->mmap_sem);
530 if (BAD_ADDR(error))
531 goto out_close;
532 }
533
Jiri Kosinacc503c12008-01-30 13:31:07 +0100534 error = load_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535
536out_close:
537 kfree(elf_phdata);
538out:
539 return error;
540}
541
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542/*
543 * These are the functions used to load ELF style executables and shared
544 * libraries. There is no binary dependent code anywhere else.
545 */
546
547#define INTERPRETER_NONE 0
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548#define INTERPRETER_ELF 2
549
Andi Kleen913bd902006-03-25 16:29:09 +0100550#ifndef STACK_RND_MASK
James Bottomleyd1cabd62007-03-16 13:38:35 -0800551#define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
Andi Kleen913bd902006-03-25 16:29:09 +0100552#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553
554static unsigned long randomize_stack_top(unsigned long stack_top)
555{
556 unsigned int random_variable = 0;
557
Andi Kleenc16b63e02006-09-26 10:52:28 +0200558 if ((current->flags & PF_RANDOMIZE) &&
559 !(current->personality & ADDR_NO_RANDOMIZE)) {
Andi Kleen913bd902006-03-25 16:29:09 +0100560 random_variable = get_random_int() & STACK_RND_MASK;
561 random_variable <<= PAGE_SHIFT;
562 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563#ifdef CONFIG_STACK_GROWSUP
Andi Kleen913bd902006-03-25 16:29:09 +0100564 return PAGE_ALIGN(stack_top) + random_variable;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565#else
Andi Kleen913bd902006-03-25 16:29:09 +0100566 return PAGE_ALIGN(stack_top) - random_variable;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567#endif
568}
569
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700570static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571{
572 struct file *interpreter = NULL; /* to shut gcc up */
573 unsigned long load_addr = 0, load_bias = 0;
574 int load_addr_set = 0;
575 char * elf_interpreter = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 unsigned long error;
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700577 struct elf_phdr *elf_ppnt, *elf_phdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 unsigned long elf_bss, elf_brk;
579 int elf_exec_fileno;
580 int retval, i;
581 unsigned int size;
Jiri Kosinacc503c12008-01-30 13:31:07 +0100582 unsigned long elf_entry;
583 unsigned long interp_load_addr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 unsigned long start_code, end_code, start_data, end_data;
585 unsigned long reloc_func_desc = 0;
David Rientjes8de61e62006-12-06 20:40:16 -0800586 int executable_stack = EXSTACK_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 unsigned long def_flags = 0;
588 struct {
589 struct elfhdr elf_ex;
590 struct elfhdr interp_elf_ex;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 } *loc;
592
593 loc = kmalloc(sizeof(*loc), GFP_KERNEL);
594 if (!loc) {
595 retval = -ENOMEM;
596 goto out_ret;
597 }
598
599 /* Get the exec-header */
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700600 loc->elf_ex = *((struct elfhdr *)bprm->buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601
602 retval = -ENOEXEC;
603 /* First of all, some simple consistency checks */
604 if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
605 goto out;
606
607 if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
608 goto out;
609 if (!elf_check_arch(&loc->elf_ex))
610 goto out;
611 if (!bprm->file->f_op||!bprm->file->f_op->mmap)
612 goto out;
613
614 /* Now read in all of the header information */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
616 goto out;
617 if (loc->elf_ex.e_phnum < 1 ||
618 loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
619 goto out;
620 size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
621 retval = -ENOMEM;
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700622 elf_phdata = kmalloc(size, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 if (!elf_phdata)
624 goto out;
625
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700626 retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
627 (char *)elf_phdata, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 if (retval != size) {
629 if (retval >= 0)
630 retval = -EIO;
631 goto out_free_ph;
632 }
633
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 retval = get_unused_fd();
635 if (retval < 0)
Al Virofd8328b2008-04-22 05:11:59 -0400636 goto out_free_ph;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 get_file(bprm->file);
638 fd_install(elf_exec_fileno = retval, bprm->file);
639
640 elf_ppnt = elf_phdata;
641 elf_bss = 0;
642 elf_brk = 0;
643
644 start_code = ~0UL;
645 end_code = 0;
646 start_data = 0;
647 end_data = 0;
648
649 for (i = 0; i < loc->elf_ex.e_phnum; i++) {
650 if (elf_ppnt->p_type == PT_INTERP) {
651 /* This is the program interpreter used for
652 * shared libraries - for now assume that this
653 * is an a.out format binary
654 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 retval = -ENOEXEC;
656 if (elf_ppnt->p_filesz > PATH_MAX ||
657 elf_ppnt->p_filesz < 2)
658 goto out_free_file;
659
660 retval = -ENOMEM;
Jesper Juhl792db3a2006-01-09 20:54:45 -0800661 elf_interpreter = kmalloc(elf_ppnt->p_filesz,
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700662 GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 if (!elf_interpreter)
664 goto out_free_file;
665
666 retval = kernel_read(bprm->file, elf_ppnt->p_offset,
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700667 elf_interpreter,
668 elf_ppnt->p_filesz);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 if (retval != elf_ppnt->p_filesz) {
670 if (retval >= 0)
671 retval = -EIO;
672 goto out_free_interp;
673 }
674 /* make sure path is NULL terminated */
675 retval = -ENOEXEC;
676 if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
677 goto out_free_interp;
678
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 /*
680 * The early SET_PERSONALITY here is so that the lookup
681 * for the interpreter happens in the namespace of the
682 * to-be-execed image. SET_PERSONALITY can select an
683 * alternate root.
684 *
685 * However, SET_PERSONALITY is NOT allowed to switch
686 * this task into the new images's memory mapping
687 * policy - that is, TASK_SIZE must still evaluate to
688 * that which is appropriate to the execing application.
689 * This is because exit_mmap() needs to have TASK_SIZE
690 * evaluate to the size of the old image.
691 *
692 * So if (say) a 64-bit application is execing a 32-bit
693 * application it is the architecture's responsibility
694 * to defer changing the value of TASK_SIZE until the
695 * switch really is going to happen - do this in
696 * flush_thread(). - akpm
697 */
Martin Schwidefsky0b592682008-10-16 15:39:57 +0200698 SET_PERSONALITY(loc->elf_ex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699
700 interpreter = open_exec(elf_interpreter);
701 retval = PTR_ERR(interpreter);
702 if (IS_ERR(interpreter))
703 goto out_free_interp;
Alexey Dobriyan1fb84492007-01-26 00:57:16 -0800704
705 /*
706 * If the binary is not readable then enforce
707 * mm->dumpable = 0 regardless of the interpreter's
708 * permissions.
709 */
710 if (file_permission(interpreter, MAY_READ) < 0)
711 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
712
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700713 retval = kernel_read(interpreter, 0, bprm->buf,
714 BINPRM_BUF_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 if (retval != BINPRM_BUF_SIZE) {
716 if (retval >= 0)
717 retval = -EIO;
718 goto out_free_dentry;
719 }
720
721 /* Get the exec headers */
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700722 loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 break;
724 }
725 elf_ppnt++;
726 }
727
728 elf_ppnt = elf_phdata;
729 for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
730 if (elf_ppnt->p_type == PT_GNU_STACK) {
731 if (elf_ppnt->p_flags & PF_X)
732 executable_stack = EXSTACK_ENABLE_X;
733 else
734 executable_stack = EXSTACK_DISABLE_X;
735 break;
736 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737
738 /* Some simple consistency checks for the interpreter */
739 if (elf_interpreter) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 retval = -ELIBBAD;
Andi Kleend20894a2008-02-08 04:21:54 -0800741 /* Not an ELF interpreter */
742 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 goto out_free_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 /* Verify the interpreter has a valid arch */
Andi Kleend20894a2008-02-08 04:21:54 -0800745 if (!elf_check_arch(&loc->interp_elf_ex))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 goto out_free_dentry;
747 } else {
748 /* Executables without an interpreter also need a personality */
Martin Schwidefsky0b592682008-10-16 15:39:57 +0200749 SET_PERSONALITY(loc->elf_ex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 }
751
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 /* Flush all traces of the currently running executable */
753 retval = flush_old_exec(bprm);
754 if (retval)
755 goto out_free_dentry;
756
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 /* OK, This is the point of no return */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 current->flags &= ~PF_FORKNOEXEC;
759 current->mm->def_flags = def_flags;
760
761 /* Do this immediately, since STACK_TOP as used in setup_arg_pages
762 may depend on the personality. */
Martin Schwidefsky0b592682008-10-16 15:39:57 +0200763 SET_PERSONALITY(loc->elf_ex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 if (elf_read_implies_exec(loc->elf_ex, executable_stack))
765 current->personality |= READ_IMPLIES_EXEC;
766
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700767 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 current->flags |= PF_RANDOMIZE;
769 arch_pick_mmap_layout(current->mm);
770
771 /* Do this so that we can load the interpreter, if need be. We will
772 change some of these later */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 current->mm->free_area_cache = current->mm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -0700774 current->mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
776 executable_stack);
777 if (retval < 0) {
778 send_sig(SIGKILL, current, 0);
779 goto out_free_dentry;
780 }
781
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 current->mm->start_stack = bprm->p;
783
784 /* Now we do a little grungy work by mmaping the ELF image into
Jiri Kosinacc503c12008-01-30 13:31:07 +0100785 the correct location in memory. */
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700786 for(i = 0, elf_ppnt = elf_phdata;
787 i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 int elf_prot = 0, elf_flags;
789 unsigned long k, vaddr;
790
791 if (elf_ppnt->p_type != PT_LOAD)
792 continue;
793
794 if (unlikely (elf_brk > elf_bss)) {
795 unsigned long nbyte;
796
797 /* There was a PT_LOAD segment with p_memsz > p_filesz
798 before this one. Map anonymous pages, if needed,
799 and clear the area. */
800 retval = set_brk (elf_bss + load_bias,
801 elf_brk + load_bias);
802 if (retval) {
803 send_sig(SIGKILL, current, 0);
804 goto out_free_dentry;
805 }
806 nbyte = ELF_PAGEOFFSET(elf_bss);
807 if (nbyte) {
808 nbyte = ELF_MIN_ALIGN - nbyte;
809 if (nbyte > elf_brk - elf_bss)
810 nbyte = elf_brk - elf_bss;
811 if (clear_user((void __user *)elf_bss +
812 load_bias, nbyte)) {
813 /*
814 * This bss-zeroing can fail if the ELF
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700815 * file specifies odd protections. So
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 * we don't check the return value
817 */
818 }
819 }
820 }
821
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700822 if (elf_ppnt->p_flags & PF_R)
823 elf_prot |= PROT_READ;
824 if (elf_ppnt->p_flags & PF_W)
825 elf_prot |= PROT_WRITE;
826 if (elf_ppnt->p_flags & PF_X)
827 elf_prot |= PROT_EXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700829 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830
831 vaddr = elf_ppnt->p_vaddr;
832 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
833 elf_flags |= MAP_FIXED;
834 } else if (loc->elf_ex.e_type == ET_DYN) {
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700835 /* Try and get dynamic programs out of the way of the
836 * default mmap base, as well as whatever program they
837 * might try to exec. This is because the brk will
838 * follow the loader, and is not movable. */
Jiri Kosinacc503c12008-01-30 13:31:07 +0100839#ifdef CONFIG_X86
840 load_bias = 0;
841#else
Linus Torvalds90cb28e2007-01-06 13:28:21 -0800842 load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
Jiri Kosinacc503c12008-01-30 13:31:07 +0100843#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 }
845
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700846 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
Andrew Mortonbb1ad822008-01-30 13:31:07 +0100847 elf_prot, elf_flags, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 if (BAD_ADDR(error)) {
849 send_sig(SIGKILL, current, 0);
Alexey Kuznetsovb140f2512007-05-08 00:31:57 -0700850 retval = IS_ERR((void *)error) ?
851 PTR_ERR((void*)error) : -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 goto out_free_dentry;
853 }
854
855 if (!load_addr_set) {
856 load_addr_set = 1;
857 load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
858 if (loc->elf_ex.e_type == ET_DYN) {
859 load_bias += error -
860 ELF_PAGESTART(load_bias + vaddr);
861 load_addr += load_bias;
862 reloc_func_desc = load_bias;
863 }
864 }
865 k = elf_ppnt->p_vaddr;
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700866 if (k < start_code)
867 start_code = k;
868 if (start_data < k)
869 start_data = k;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870
871 /*
872 * Check to see if the section's size will overflow the
873 * allowed task size. Note that p_filesz must always be
874 * <= p_memsz so it is only necessary to check p_memsz.
875 */
Chuck Ebbertce510592006-07-03 00:24:14 -0700876 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 elf_ppnt->p_memsz > TASK_SIZE ||
878 TASK_SIZE - elf_ppnt->p_memsz < k) {
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700879 /* set_brk can never work. Avoid overflows. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 send_sig(SIGKILL, current, 0);
Alexey Kuznetsovb140f2512007-05-08 00:31:57 -0700881 retval = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 goto out_free_dentry;
883 }
884
885 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
886
887 if (k > elf_bss)
888 elf_bss = k;
889 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
890 end_code = k;
891 if (end_data < k)
892 end_data = k;
893 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
894 if (k > elf_brk)
895 elf_brk = k;
896 }
897
898 loc->elf_ex.e_entry += load_bias;
899 elf_bss += load_bias;
900 elf_brk += load_bias;
901 start_code += load_bias;
902 end_code += load_bias;
903 start_data += load_bias;
904 end_data += load_bias;
905
906 /* Calling set_brk effectively mmaps the pages that we need
907 * for the bss and break sections. We must do this before
908 * mapping in the interpreter, to make sure it doesn't wind
909 * up getting placed where the bss needs to go.
910 */
911 retval = set_brk(elf_bss, elf_brk);
912 if (retval) {
913 send_sig(SIGKILL, current, 0);
914 goto out_free_dentry;
915 }
akpm@osdl.org6de50512005-10-11 08:29:08 -0700916 if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 send_sig(SIGSEGV, current, 0);
918 retval = -EFAULT; /* Nobody gets to see this, but.. */
919 goto out_free_dentry;
920 }
921
922 if (elf_interpreter) {
Andi Kleend20894a2008-02-08 04:21:54 -0800923 unsigned long uninitialized_var(interp_map_addr);
Jiri Kosinacc503c12008-01-30 13:31:07 +0100924
Andi Kleend20894a2008-02-08 04:21:54 -0800925 elf_entry = load_elf_interp(&loc->interp_elf_ex,
926 interpreter,
927 &interp_map_addr,
928 load_bias);
929 if (!IS_ERR((void *)elf_entry)) {
930 /*
931 * load_elf_interp() returns relocation
932 * adjustment
933 */
934 interp_load_addr = elf_entry;
935 elf_entry += loc->interp_elf_ex.e_entry;
Jiri Kosinacc503c12008-01-30 13:31:07 +0100936 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 if (BAD_ADDR(elf_entry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 force_sig(SIGSEGV, current);
Chuck Ebbertce510592006-07-03 00:24:14 -0700939 retval = IS_ERR((void *)elf_entry) ?
940 (int)elf_entry : -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941 goto out_free_dentry;
942 }
943 reloc_func_desc = interp_load_addr;
944
945 allow_write_access(interpreter);
946 fput(interpreter);
947 kfree(elf_interpreter);
948 } else {
949 elf_entry = loc->elf_ex.e_entry;
Suresh Siddha5342fba2006-02-26 04:18:28 +0100950 if (BAD_ADDR(elf_entry)) {
Chuck Ebbertce510592006-07-03 00:24:14 -0700951 force_sig(SIGSEGV, current);
952 retval = -EINVAL;
Suresh Siddha5342fba2006-02-26 04:18:28 +0100953 goto out_free_dentry;
954 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 }
956
957 kfree(elf_phdata);
958
Andi Kleend20894a2008-02-08 04:21:54 -0800959 sys_close(elf_exec_fileno);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960
961 set_binfmt(&elf_format);
962
Benjamin Herrenschmidt547ee842005-04-16 15:24:35 -0700963#ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
Martin Schwidefskyfc5243d2008-12-25 13:38:35 +0100964 retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
Benjamin Herrenschmidt547ee842005-04-16 15:24:35 -0700965 if (retval < 0) {
966 send_sig(SIGKILL, current, 0);
Roland McGrath18c8baf2005-04-28 15:17:19 -0700967 goto out;
Benjamin Herrenschmidt547ee842005-04-16 15:24:35 -0700968 }
969#endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
970
David Howellsa6f76f22008-11-14 10:39:24 +1100971 install_exec_creds(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 current->flags &= ~PF_FORKNOEXEC;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700973 retval = create_elf_tables(bprm, &loc->elf_ex,
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700974 load_addr, interp_load_addr);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700975 if (retval < 0) {
976 send_sig(SIGKILL, current, 0);
977 goto out;
978 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 /* N.B. passed_fileno might not be initialized? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 current->mm->end_code = end_code;
981 current->mm->start_code = start_code;
982 current->mm->start_data = start_data;
983 current->mm->end_data = end_data;
984 current->mm->start_stack = bprm->p;
985
Jiri Kosinac1d171a2008-01-30 13:30:40 +0100986#ifdef arch_randomize_brk
Ingo Molnar32a93232008-02-06 22:39:44 +0100987 if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1))
Jiri Kosinac1d171a2008-01-30 13:30:40 +0100988 current->mm->brk = current->mm->start_brk =
989 arch_randomize_brk(current->mm);
990#endif
991
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 if (current->personality & MMAP_PAGE_ZERO) {
993 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
994 and some applications "depend" upon this behavior.
995 Since we do not have the power to recompile these, we
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700996 emulate the SVr4 behavior. Sigh. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 down_write(&current->mm->mmap_sem);
998 error = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
999 MAP_FIXED | MAP_PRIVATE, 0);
1000 up_write(&current->mm->mmap_sem);
1001 }
1002
1003#ifdef ELF_PLAT_INIT
1004 /*
1005 * The ABI may specify that certain registers be set up in special
1006 * ways (on i386 %edx is the address of a DT_FINI function, for
1007 * example. In addition, it may also specify (eg, PowerPC64 ELF)
1008 * that the e_entry field is the address of the function descriptor
1009 * for the startup routine, rather than the address of the startup
1010 * routine itself. This macro performs whatever initialization to
1011 * the regs structure is required as well as any relocations to the
1012 * function descriptor entries when executing dynamically links apps.
1013 */
1014 ELF_PLAT_INIT(regs, reloc_func_desc);
1015#endif
1016
1017 start_thread(regs, elf_entry, bprm->p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 retval = 0;
1019out:
1020 kfree(loc);
1021out_ret:
1022 return retval;
1023
1024 /* error cleanup */
1025out_free_dentry:
1026 allow_write_access(interpreter);
1027 if (interpreter)
1028 fput(interpreter);
1029out_free_interp:
Jesper Juhlf99d49a2005-11-07 01:01:34 -08001030 kfree(elf_interpreter);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031out_free_file:
1032 sys_close(elf_exec_fileno);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033out_free_ph:
1034 kfree(elf_phdata);
1035 goto out;
1036}
1037
1038/* This is really simpleminded and specialized - we are loading an
1039 a.out library that is given an ELF header. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040static int load_elf_library(struct file *file)
1041{
1042 struct elf_phdr *elf_phdata;
1043 struct elf_phdr *eppnt;
1044 unsigned long elf_bss, bss, len;
1045 int retval, error, i, j;
1046 struct elfhdr elf_ex;
1047
1048 error = -ENOEXEC;
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001049 retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 if (retval != sizeof(elf_ex))
1051 goto out;
1052
1053 if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1054 goto out;
1055
1056 /* First of all, some simple consistency checks */
1057 if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001058 !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 goto out;
1060
1061 /* Now read in all of the header information */
1062
1063 j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1064 /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1065
1066 error = -ENOMEM;
1067 elf_phdata = kmalloc(j, GFP_KERNEL);
1068 if (!elf_phdata)
1069 goto out;
1070
1071 eppnt = elf_phdata;
1072 error = -ENOEXEC;
1073 retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1074 if (retval != j)
1075 goto out_free_ph;
1076
1077 for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1078 if ((eppnt + i)->p_type == PT_LOAD)
1079 j++;
1080 if (j != 1)
1081 goto out_free_ph;
1082
1083 while (eppnt->p_type != PT_LOAD)
1084 eppnt++;
1085
1086 /* Now use mmap to map the library into memory. */
1087 down_write(&current->mm->mmap_sem);
1088 error = do_mmap(file,
1089 ELF_PAGESTART(eppnt->p_vaddr),
1090 (eppnt->p_filesz +
1091 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1092 PROT_READ | PROT_WRITE | PROT_EXEC,
1093 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1094 (eppnt->p_offset -
1095 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1096 up_write(&current->mm->mmap_sem);
1097 if (error != ELF_PAGESTART(eppnt->p_vaddr))
1098 goto out_free_ph;
1099
1100 elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1101 if (padzero(elf_bss)) {
1102 error = -EFAULT;
1103 goto out_free_ph;
1104 }
1105
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001106 len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1107 ELF_MIN_ALIGN - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 bss = eppnt->p_memsz + eppnt->p_vaddr;
1109 if (bss > len) {
1110 down_write(&current->mm->mmap_sem);
1111 do_brk(len, bss - len);
1112 up_write(&current->mm->mmap_sem);
1113 }
1114 error = 0;
1115
1116out_free_ph:
1117 kfree(elf_phdata);
1118out:
1119 return error;
1120}
1121
1122/*
1123 * Note that some platforms still use traditional core dumps and not
1124 * the ELF core dump. Each platform can select it as appropriate.
1125 */
Matt Mackall708e9a72006-01-08 01:05:25 -08001126#if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127
1128/*
1129 * ELF core dumper
1130 *
1131 * Modelled on fs/exec.c:aout_core_dump()
1132 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1133 */
1134/*
1135 * These are the only things you should do on a core-file: use only these
1136 * functions to write out all the necessary info.
1137 */
1138static int dump_write(struct file *file, const void *addr, int nr)
1139{
1140 return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
1141}
1142
Daniel Jacobowitz5db92852005-06-15 22:26:34 -07001143static int dump_seek(struct file *file, loff_t off)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144{
Andi Kleend025c9d2006-09-30 23:29:28 -07001145 if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
Petr Vandrovec7f14daa2006-10-13 04:13:16 +02001146 if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 return 0;
Andi Kleend025c9d2006-09-30 23:29:28 -07001148 } else {
1149 char *buf = (char *)get_zeroed_page(GFP_KERNEL);
1150 if (!buf)
1151 return 0;
1152 while (off > 0) {
1153 unsigned long n = off;
1154 if (n > PAGE_SIZE)
1155 n = PAGE_SIZE;
1156 if (!dump_write(file, buf, n))
1157 return 0;
1158 off -= n;
1159 }
1160 free_page((unsigned long)buf);
1161 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 return 1;
1163}
1164
1165/*
Roland McGrath82df3972007-10-16 23:27:02 -07001166 * Decide what to dump of a segment, part, all or none.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 */
Roland McGrath82df3972007-10-16 23:27:02 -07001168static unsigned long vma_dump_size(struct vm_area_struct *vma,
1169 unsigned long mm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170{
KOSAKI Motohiroe575f112008-10-18 20:27:08 -07001171#define FILTER(type) (mm_flags & (1UL << MMF_DUMP_##type))
1172
Roland McGrathe5b97dd2007-01-26 00:56:48 -08001173 /* The vma can be set up to tell us the answer directly. */
1174 if (vma->vm_flags & VM_ALWAYSDUMP)
Roland McGrath82df3972007-10-16 23:27:02 -07001175 goto whole;
Roland McGrathe5b97dd2007-01-26 00:56:48 -08001176
KOSAKI Motohiroe575f112008-10-18 20:27:08 -07001177 /* Hugetlb memory check */
1178 if (vma->vm_flags & VM_HUGETLB) {
1179 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1180 goto whole;
1181 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1182 goto whole;
1183 }
1184
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 /* Do not dump I/O mapped devices or special mappings */
1186 if (vma->vm_flags & (VM_IO | VM_RESERVED))
1187 return 0;
1188
Kawai, Hidehiroa1b59e82007-07-19 01:48:29 -07001189 /* By default, dump shared memory if mapped from an anonymous file. */
1190 if (vma->vm_flags & VM_SHARED) {
Roland McGrath82df3972007-10-16 23:27:02 -07001191 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0 ?
1192 FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1193 goto whole;
1194 return 0;
Kawai, Hidehiroa1b59e82007-07-19 01:48:29 -07001195 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196
Roland McGrath82df3972007-10-16 23:27:02 -07001197 /* Dump segments that have been written to. */
1198 if (vma->anon_vma && FILTER(ANON_PRIVATE))
1199 goto whole;
1200 if (vma->vm_file == NULL)
1201 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202
Roland McGrath82df3972007-10-16 23:27:02 -07001203 if (FILTER(MAPPED_PRIVATE))
1204 goto whole;
1205
1206 /*
1207 * If this looks like the beginning of a DSO or executable mapping,
1208 * check for an ELF header. If we find one, dump the first page to
1209 * aid in determining what was mapped here.
1210 */
1211 if (FILTER(ELF_HEADERS) && vma->vm_file != NULL && vma->vm_pgoff == 0) {
1212 u32 __user *header = (u32 __user *) vma->vm_start;
1213 u32 word;
1214 /*
1215 * Doing it this way gets the constant folded by GCC.
1216 */
1217 union {
1218 u32 cmp;
1219 char elfmag[SELFMAG];
1220 } magic;
1221 BUILD_BUG_ON(SELFMAG != sizeof word);
1222 magic.elfmag[EI_MAG0] = ELFMAG0;
1223 magic.elfmag[EI_MAG1] = ELFMAG1;
1224 magic.elfmag[EI_MAG2] = ELFMAG2;
1225 magic.elfmag[EI_MAG3] = ELFMAG3;
1226 if (get_user(word, header) == 0 && word == magic.cmp)
1227 return PAGE_SIZE;
1228 }
1229
1230#undef FILTER
1231
1232 return 0;
1233
1234whole:
1235 return vma->vm_end - vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236}
1237
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238/* An ELF note in memory */
1239struct memelfnote
1240{
1241 const char *name;
1242 int type;
1243 unsigned int datasz;
1244 void *data;
1245};
1246
1247static int notesize(struct memelfnote *en)
1248{
1249 int sz;
1250
1251 sz = sizeof(struct elf_note);
1252 sz += roundup(strlen(en->name) + 1, 4);
1253 sz += roundup(en->datasz, 4);
1254
1255 return sz;
1256}
1257
Andi Kleend025c9d2006-09-30 23:29:28 -07001258#define DUMP_WRITE(addr, nr, foffset) \
1259 do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260
Andi Kleend025c9d2006-09-30 23:29:28 -07001261static int alignfile(struct file *file, loff_t *foffset)
1262{
Petr Vandroveca7a0d862006-10-13 18:42:07 +02001263 static const char buf[4] = { 0, };
Andi Kleend025c9d2006-09-30 23:29:28 -07001264 DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1265 return 1;
1266}
1267
1268static int writenote(struct memelfnote *men, struct file *file,
1269 loff_t *foffset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270{
1271 struct elf_note en;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 en.n_namesz = strlen(men->name) + 1;
1273 en.n_descsz = men->datasz;
1274 en.n_type = men->type;
1275
Andi Kleend025c9d2006-09-30 23:29:28 -07001276 DUMP_WRITE(&en, sizeof(en), foffset);
1277 DUMP_WRITE(men->name, en.n_namesz, foffset);
1278 if (!alignfile(file, foffset))
1279 return 0;
1280 DUMP_WRITE(men->data, men->datasz, foffset);
1281 if (!alignfile(file, foffset))
1282 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283
1284 return 1;
1285}
1286#undef DUMP_WRITE
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287
1288#define DUMP_WRITE(addr, nr) \
1289 if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
1290 goto end_coredump;
1291#define DUMP_SEEK(off) \
1292 if (!dump_seek(file, (off))) \
1293 goto end_coredump;
1294
Roland McGrath3aba4812008-01-30 13:31:44 +01001295static void fill_elf_header(struct elfhdr *elf, int segs,
1296 u16 machine, u32 flags, u8 osabi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297{
Cyrill Gorcunov6970c8e2008-04-29 01:01:18 -07001298 memset(elf, 0, sizeof(*elf));
1299
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1301 elf->e_ident[EI_CLASS] = ELF_CLASS;
1302 elf->e_ident[EI_DATA] = ELF_DATA;
1303 elf->e_ident[EI_VERSION] = EV_CURRENT;
1304 elf->e_ident[EI_OSABI] = ELF_OSABI;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305
1306 elf->e_type = ET_CORE;
Roland McGrath3aba4812008-01-30 13:31:44 +01001307 elf->e_machine = machine;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 elf->e_version = EV_CURRENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 elf->e_phoff = sizeof(struct elfhdr);
Roland McGrath3aba4812008-01-30 13:31:44 +01001310 elf->e_flags = flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 elf->e_ehsize = sizeof(struct elfhdr);
1312 elf->e_phentsize = sizeof(struct elf_phdr);
1313 elf->e_phnum = segs;
Cyrill Gorcunov6970c8e2008-04-29 01:01:18 -07001314
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315 return;
1316}
1317
Andrew Morton8d6b5eee2006-09-25 23:32:04 -07001318static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319{
1320 phdr->p_type = PT_NOTE;
1321 phdr->p_offset = offset;
1322 phdr->p_vaddr = 0;
1323 phdr->p_paddr = 0;
1324 phdr->p_filesz = sz;
1325 phdr->p_memsz = 0;
1326 phdr->p_flags = 0;
1327 phdr->p_align = 0;
1328 return;
1329}
1330
1331static void fill_note(struct memelfnote *note, const char *name, int type,
1332 unsigned int sz, void *data)
1333{
1334 note->name = name;
1335 note->type = type;
1336 note->datasz = sz;
1337 note->data = data;
1338 return;
1339}
1340
1341/*
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001342 * fill up all the fields in prstatus from the given task struct, except
1343 * registers which need to be filled up separately.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 */
1345static void fill_prstatus(struct elf_prstatus *prstatus,
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001346 struct task_struct *p, long signr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347{
1348 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1349 prstatus->pr_sigpend = p->pending.signal.sig[0];
1350 prstatus->pr_sighold = p->blocked.sig[0];
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001351 prstatus->pr_pid = task_pid_vnr(p);
Roland McGrath45626bb2008-01-07 14:22:44 -08001352 prstatus->pr_ppid = task_pid_vnr(p->real_parent);
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001353 prstatus->pr_pgrp = task_pgrp_vnr(p);
1354 prstatus->pr_sid = task_session_vnr(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355 if (thread_group_leader(p)) {
Frank Mayharf06febc2008-09-12 09:54:39 -07001356 struct task_cputime cputime;
1357
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 /*
Frank Mayharf06febc2008-09-12 09:54:39 -07001359 * This is the record for the group leader. It shows the
1360 * group-wide total, not its individual thread total.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 */
Frank Mayharf06febc2008-09-12 09:54:39 -07001362 thread_group_cputime(p, &cputime);
1363 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1364 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 } else {
1366 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1367 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1368 }
1369 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1370 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1371}
1372
1373static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1374 struct mm_struct *mm)
1375{
David Howellsc69e8d92008-11-14 10:39:19 +11001376 const struct cred *cred;
Greg Kroah-Hartmana84a5052005-05-11 00:10:44 -07001377 unsigned int i, len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378
1379 /* first copy the parameters from user space */
1380 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1381
1382 len = mm->arg_end - mm->arg_start;
1383 if (len >= ELF_PRARGSZ)
1384 len = ELF_PRARGSZ-1;
1385 if (copy_from_user(&psinfo->pr_psargs,
1386 (const char __user *)mm->arg_start, len))
1387 return -EFAULT;
1388 for(i = 0; i < len; i++)
1389 if (psinfo->pr_psargs[i] == 0)
1390 psinfo->pr_psargs[i] = ' ';
1391 psinfo->pr_psargs[len] = 0;
1392
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001393 psinfo->pr_pid = task_pid_vnr(p);
Roland McGrath45626bb2008-01-07 14:22:44 -08001394 psinfo->pr_ppid = task_pid_vnr(p->real_parent);
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001395 psinfo->pr_pgrp = task_pgrp_vnr(p);
1396 psinfo->pr_sid = task_session_vnr(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397
1398 i = p->state ? ffz(~p->state) + 1 : 0;
1399 psinfo->pr_state = i;
Carsten Otte55148542006-03-25 03:08:22 -08001400 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1402 psinfo->pr_nice = task_nice(p);
1403 psinfo->pr_flag = p->flags;
David Howellsc69e8d92008-11-14 10:39:19 +11001404 rcu_read_lock();
1405 cred = __task_cred(p);
1406 SET_UID(psinfo->pr_uid, cred->uid);
1407 SET_GID(psinfo->pr_gid, cred->gid);
1408 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1410
1411 return 0;
1412}
1413
Roland McGrath3aba4812008-01-30 13:31:44 +01001414static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1415{
1416 elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1417 int i = 0;
1418 do
1419 i += 2;
1420 while (auxv[i - 2] != AT_NULL);
1421 fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1422}
1423
Roland McGrath4206d3a2008-01-30 13:31:45 +01001424#ifdef CORE_DUMP_USE_REGSET
1425#include <linux/regset.h>
1426
1427struct elf_thread_core_info {
1428 struct elf_thread_core_info *next;
1429 struct task_struct *task;
1430 struct elf_prstatus prstatus;
1431 struct memelfnote notes[0];
1432};
1433
1434struct elf_note_info {
1435 struct elf_thread_core_info *thread;
1436 struct memelfnote psinfo;
1437 struct memelfnote auxv;
1438 size_t size;
1439 int thread_notes;
1440};
1441
Roland McGrathd31472b2008-03-04 14:28:30 -08001442/*
1443 * When a regset has a writeback hook, we call it on each thread before
1444 * dumping user memory. On register window machines, this makes sure the
1445 * user memory backing the register data is up to date before we read it.
1446 */
1447static void do_thread_regset_writeback(struct task_struct *task,
1448 const struct user_regset *regset)
1449{
1450 if (regset->writeback)
1451 regset->writeback(task, regset, 1);
1452}
1453
Roland McGrath4206d3a2008-01-30 13:31:45 +01001454static int fill_thread_core_info(struct elf_thread_core_info *t,
1455 const struct user_regset_view *view,
1456 long signr, size_t *total)
1457{
1458 unsigned int i;
1459
1460 /*
1461 * NT_PRSTATUS is the one special case, because the regset data
1462 * goes into the pr_reg field inside the note contents, rather
1463 * than being the whole note contents. We fill the reset in here.
1464 * We assume that regset 0 is NT_PRSTATUS.
1465 */
1466 fill_prstatus(&t->prstatus, t->task, signr);
1467 (void) view->regsets[0].get(t->task, &view->regsets[0],
1468 0, sizeof(t->prstatus.pr_reg),
1469 &t->prstatus.pr_reg, NULL);
1470
1471 fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1472 sizeof(t->prstatus), &t->prstatus);
1473 *total += notesize(&t->notes[0]);
1474
Roland McGrathd31472b2008-03-04 14:28:30 -08001475 do_thread_regset_writeback(t->task, &view->regsets[0]);
1476
Roland McGrath4206d3a2008-01-30 13:31:45 +01001477 /*
1478 * Each other regset might generate a note too. For each regset
1479 * that has no core_note_type or is inactive, we leave t->notes[i]
1480 * all zero and we'll know to skip writing it later.
1481 */
1482 for (i = 1; i < view->n; ++i) {
1483 const struct user_regset *regset = &view->regsets[i];
Roland McGrathd31472b2008-03-04 14:28:30 -08001484 do_thread_regset_writeback(t->task, regset);
Roland McGrath4206d3a2008-01-30 13:31:45 +01001485 if (regset->core_note_type &&
1486 (!regset->active || regset->active(t->task, regset))) {
1487 int ret;
1488 size_t size = regset->n * regset->size;
1489 void *data = kmalloc(size, GFP_KERNEL);
1490 if (unlikely(!data))
1491 return 0;
1492 ret = regset->get(t->task, regset,
1493 0, size, data, NULL);
1494 if (unlikely(ret))
1495 kfree(data);
1496 else {
1497 if (regset->core_note_type != NT_PRFPREG)
1498 fill_note(&t->notes[i], "LINUX",
1499 regset->core_note_type,
1500 size, data);
1501 else {
1502 t->prstatus.pr_fpvalid = 1;
1503 fill_note(&t->notes[i], "CORE",
1504 NT_PRFPREG, size, data);
1505 }
1506 *total += notesize(&t->notes[i]);
1507 }
1508 }
1509 }
1510
1511 return 1;
1512}
1513
1514static int fill_note_info(struct elfhdr *elf, int phdrs,
1515 struct elf_note_info *info,
1516 long signr, struct pt_regs *regs)
1517{
1518 struct task_struct *dump_task = current;
1519 const struct user_regset_view *view = task_user_regset_view(dump_task);
1520 struct elf_thread_core_info *t;
1521 struct elf_prpsinfo *psinfo;
Oleg Nesterov83914442008-07-25 01:47:45 -07001522 struct core_thread *ct;
Roland McGrath4206d3a2008-01-30 13:31:45 +01001523 unsigned int i;
1524
1525 info->size = 0;
1526 info->thread = NULL;
1527
1528 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
1529 fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1530
1531 if (psinfo == NULL)
1532 return 0;
1533
1534 /*
1535 * Figure out how many notes we're going to need for each thread.
1536 */
1537 info->thread_notes = 0;
1538 for (i = 0; i < view->n; ++i)
1539 if (view->regsets[i].core_note_type != 0)
1540 ++info->thread_notes;
1541
1542 /*
1543 * Sanity check. We rely on regset 0 being in NT_PRSTATUS,
1544 * since it is our one special case.
1545 */
1546 if (unlikely(info->thread_notes == 0) ||
1547 unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1548 WARN_ON(1);
1549 return 0;
1550 }
1551
1552 /*
1553 * Initialize the ELF file header.
1554 */
1555 fill_elf_header(elf, phdrs,
1556 view->e_machine, view->e_flags, view->ei_osabi);
1557
1558 /*
1559 * Allocate a structure for each thread.
1560 */
Oleg Nesterov83914442008-07-25 01:47:45 -07001561 for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1562 t = kzalloc(offsetof(struct elf_thread_core_info,
1563 notes[info->thread_notes]),
1564 GFP_KERNEL);
1565 if (unlikely(!t))
1566 return 0;
Oleg Nesterov24d52882008-07-25 01:47:40 -07001567
Oleg Nesterov83914442008-07-25 01:47:45 -07001568 t->task = ct->task;
1569 if (ct->task == dump_task || !info->thread) {
1570 t->next = info->thread;
1571 info->thread = t;
1572 } else {
1573 /*
1574 * Make sure to keep the original task at
1575 * the head of the list.
1576 */
1577 t->next = info->thread->next;
1578 info->thread->next = t;
Roland McGrath4206d3a2008-01-30 13:31:45 +01001579 }
Oleg Nesterov83914442008-07-25 01:47:45 -07001580 }
Roland McGrath4206d3a2008-01-30 13:31:45 +01001581
1582 /*
1583 * Now fill in each thread's information.
1584 */
1585 for (t = info->thread; t != NULL; t = t->next)
1586 if (!fill_thread_core_info(t, view, signr, &info->size))
1587 return 0;
1588
1589 /*
1590 * Fill in the two process-wide notes.
1591 */
1592 fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1593 info->size += notesize(&info->psinfo);
1594
1595 fill_auxv_note(&info->auxv, current->mm);
1596 info->size += notesize(&info->auxv);
1597
1598 return 1;
1599}
1600
1601static size_t get_note_info_size(struct elf_note_info *info)
1602{
1603 return info->size;
1604}
1605
1606/*
1607 * Write all the notes for each thread. When writing the first thread, the
1608 * process-wide notes are interleaved after the first thread-specific note.
1609 */
1610static int write_note_info(struct elf_note_info *info,
1611 struct file *file, loff_t *foffset)
1612{
1613 bool first = 1;
1614 struct elf_thread_core_info *t = info->thread;
1615
1616 do {
1617 int i;
1618
1619 if (!writenote(&t->notes[0], file, foffset))
1620 return 0;
1621
1622 if (first && !writenote(&info->psinfo, file, foffset))
1623 return 0;
1624 if (first && !writenote(&info->auxv, file, foffset))
1625 return 0;
1626
1627 for (i = 1; i < info->thread_notes; ++i)
1628 if (t->notes[i].data &&
1629 !writenote(&t->notes[i], file, foffset))
1630 return 0;
1631
1632 first = 0;
1633 t = t->next;
1634 } while (t);
1635
1636 return 1;
1637}
1638
1639static void free_note_info(struct elf_note_info *info)
1640{
1641 struct elf_thread_core_info *threads = info->thread;
1642 while (threads) {
1643 unsigned int i;
1644 struct elf_thread_core_info *t = threads;
1645 threads = t->next;
1646 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1647 for (i = 1; i < info->thread_notes; ++i)
1648 kfree(t->notes[i].data);
1649 kfree(t);
1650 }
1651 kfree(info->psinfo.data);
1652}
1653
1654#else
1655
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656/* Here is the structure in which status of each thread is captured. */
1657struct elf_thread_status
1658{
1659 struct list_head list;
1660 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1661 elf_fpregset_t fpu; /* NT_PRFPREG */
1662 struct task_struct *thread;
1663#ifdef ELF_CORE_COPY_XFPREGS
Mark Nelson5b20cd82007-10-16 23:25:39 -07001664 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665#endif
1666 struct memelfnote notes[3];
1667 int num_notes;
1668};
1669
1670/*
1671 * In order to add the specific thread information for the elf file format,
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001672 * we need to keep a linked list of every threads pr_status and then create
1673 * a single section for them in the final core file.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 */
1675static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1676{
1677 int sz = 0;
1678 struct task_struct *p = t->thread;
1679 t->num_notes = 0;
1680
1681 fill_prstatus(&t->prstatus, p, signr);
1682 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1683
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001684 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1685 &(t->prstatus));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 t->num_notes++;
1687 sz += notesize(&t->notes[0]);
1688
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001689 if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1690 &t->fpu))) {
1691 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1692 &(t->fpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 t->num_notes++;
1694 sz += notesize(&t->notes[1]);
1695 }
1696
1697#ifdef ELF_CORE_COPY_XFPREGS
1698 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
Mark Nelson5b20cd82007-10-16 23:25:39 -07001699 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1700 sizeof(t->xfpu), &t->xfpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 t->num_notes++;
1702 sz += notesize(&t->notes[2]);
1703 }
1704#endif
1705 return sz;
1706}
1707
Roland McGrath3aba4812008-01-30 13:31:44 +01001708struct elf_note_info {
1709 struct memelfnote *notes;
1710 struct elf_prstatus *prstatus; /* NT_PRSTATUS */
1711 struct elf_prpsinfo *psinfo; /* NT_PRPSINFO */
1712 struct list_head thread_list;
1713 elf_fpregset_t *fpu;
1714#ifdef ELF_CORE_COPY_XFPREGS
1715 elf_fpxregset_t *xfpu;
1716#endif
1717 int thread_status_size;
1718 int numnote;
1719};
1720
1721static int fill_note_info(struct elfhdr *elf, int phdrs,
1722 struct elf_note_info *info,
1723 long signr, struct pt_regs *regs)
1724{
1725#define NUM_NOTES 6
1726 struct list_head *t;
Roland McGrath3aba4812008-01-30 13:31:44 +01001727
1728 info->notes = NULL;
1729 info->prstatus = NULL;
1730 info->psinfo = NULL;
1731 info->fpu = NULL;
1732#ifdef ELF_CORE_COPY_XFPREGS
1733 info->xfpu = NULL;
1734#endif
1735 INIT_LIST_HEAD(&info->thread_list);
1736
1737 info->notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote),
1738 GFP_KERNEL);
1739 if (!info->notes)
1740 return 0;
1741 info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1742 if (!info->psinfo)
1743 return 0;
1744 info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1745 if (!info->prstatus)
1746 return 0;
1747 info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1748 if (!info->fpu)
1749 return 0;
1750#ifdef ELF_CORE_COPY_XFPREGS
1751 info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1752 if (!info->xfpu)
1753 return 0;
1754#endif
1755
1756 info->thread_status_size = 0;
1757 if (signr) {
Oleg Nesterov83914442008-07-25 01:47:45 -07001758 struct core_thread *ct;
WANG Cong4220b7f2008-04-29 01:01:18 -07001759 struct elf_thread_status *ets;
Oleg Nesterov24d52882008-07-25 01:47:40 -07001760
Oleg Nesterov83914442008-07-25 01:47:45 -07001761 for (ct = current->mm->core_state->dumper.next;
1762 ct; ct = ct->next) {
1763 ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1764 if (!ets)
1765 return 0;
1766
1767 ets->thread = ct->task;
1768 list_add(&ets->list, &info->thread_list);
1769 }
1770
Roland McGrath3aba4812008-01-30 13:31:44 +01001771 list_for_each(t, &info->thread_list) {
Roland McGrath3aba4812008-01-30 13:31:44 +01001772 int sz;
1773
WANG Cong4220b7f2008-04-29 01:01:18 -07001774 ets = list_entry(t, struct elf_thread_status, list);
1775 sz = elf_dump_thread_status(signr, ets);
Roland McGrath3aba4812008-01-30 13:31:44 +01001776 info->thread_status_size += sz;
1777 }
1778 }
1779 /* now collect the dump for the current */
1780 memset(info->prstatus, 0, sizeof(*info->prstatus));
1781 fill_prstatus(info->prstatus, current, signr);
1782 elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1783
1784 /* Set up header */
1785 fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS, ELF_OSABI);
1786
1787 /*
1788 * Set up the notes in similar form to SVR4 core dumps made
1789 * with info from their /proc.
1790 */
1791
1792 fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1793 sizeof(*info->prstatus), info->prstatus);
1794 fill_psinfo(info->psinfo, current->group_leader, current->mm);
1795 fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1796 sizeof(*info->psinfo), info->psinfo);
1797
1798 info->numnote = 2;
1799
1800 fill_auxv_note(&info->notes[info->numnote++], current->mm);
1801
1802 /* Try to dump the FPU. */
1803 info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1804 info->fpu);
1805 if (info->prstatus->pr_fpvalid)
1806 fill_note(info->notes + info->numnote++,
1807 "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1808#ifdef ELF_CORE_COPY_XFPREGS
1809 if (elf_core_copy_task_xfpregs(current, info->xfpu))
1810 fill_note(info->notes + info->numnote++,
1811 "LINUX", ELF_CORE_XFPREG_TYPE,
1812 sizeof(*info->xfpu), info->xfpu);
1813#endif
1814
1815 return 1;
1816
1817#undef NUM_NOTES
1818}
1819
1820static size_t get_note_info_size(struct elf_note_info *info)
1821{
1822 int sz = 0;
1823 int i;
1824
1825 for (i = 0; i < info->numnote; i++)
1826 sz += notesize(info->notes + i);
1827
1828 sz += info->thread_status_size;
1829
1830 return sz;
1831}
1832
1833static int write_note_info(struct elf_note_info *info,
1834 struct file *file, loff_t *foffset)
1835{
1836 int i;
1837 struct list_head *t;
1838
1839 for (i = 0; i < info->numnote; i++)
1840 if (!writenote(info->notes + i, file, foffset))
1841 return 0;
1842
1843 /* write out the thread status notes section */
1844 list_for_each(t, &info->thread_list) {
1845 struct elf_thread_status *tmp =
1846 list_entry(t, struct elf_thread_status, list);
1847
1848 for (i = 0; i < tmp->num_notes; i++)
1849 if (!writenote(&tmp->notes[i], file, foffset))
1850 return 0;
1851 }
1852
1853 return 1;
1854}
1855
1856static void free_note_info(struct elf_note_info *info)
1857{
1858 while (!list_empty(&info->thread_list)) {
1859 struct list_head *tmp = info->thread_list.next;
1860 list_del(tmp);
1861 kfree(list_entry(tmp, struct elf_thread_status, list));
1862 }
1863
1864 kfree(info->prstatus);
1865 kfree(info->psinfo);
1866 kfree(info->notes);
1867 kfree(info->fpu);
1868#ifdef ELF_CORE_COPY_XFPREGS
1869 kfree(info->xfpu);
1870#endif
1871}
1872
Roland McGrath4206d3a2008-01-30 13:31:45 +01001873#endif
1874
Roland McGrathf47aef52007-01-26 00:56:49 -08001875static struct vm_area_struct *first_vma(struct task_struct *tsk,
1876 struct vm_area_struct *gate_vma)
1877{
1878 struct vm_area_struct *ret = tsk->mm->mmap;
1879
1880 if (ret)
1881 return ret;
1882 return gate_vma;
1883}
1884/*
1885 * Helper function for iterating across a vma list. It ensures that the caller
1886 * will visit `gate_vma' prior to terminating the search.
1887 */
1888static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1889 struct vm_area_struct *gate_vma)
1890{
1891 struct vm_area_struct *ret;
1892
1893 ret = this_vma->vm_next;
1894 if (ret)
1895 return ret;
1896 if (this_vma == gate_vma)
1897 return NULL;
1898 return gate_vma;
1899}
1900
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901/*
1902 * Actual dumper
1903 *
1904 * This is a two-pass process; first we find the offsets of the bits,
1905 * and then they are actually written out. If we run out of core limit
1906 * we just truncate.
1907 */
Neil Horman7dc0b222007-10-16 23:26:34 -07001908static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, unsigned long limit)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 int has_dumped = 0;
1911 mm_segment_t fs;
1912 int segs;
1913 size_t size = 0;
Roland McGrathf47aef52007-01-26 00:56:49 -08001914 struct vm_area_struct *vma, *gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915 struct elfhdr *elf = NULL;
Andi Kleend025c9d2006-09-30 23:29:28 -07001916 loff_t offset = 0, dataoff, foffset;
Kawai, Hidehiroa1b59e82007-07-19 01:48:29 -07001917 unsigned long mm_flags;
Roland McGrath3aba4812008-01-30 13:31:44 +01001918 struct elf_note_info info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919
1920 /*
1921 * We no longer stop all VM operations.
1922 *
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001923 * This is because those proceses that could possibly change map_count
1924 * or the mmap / vma pages are now blocked in do_exit on current
1925 * finishing this core dump.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 *
1927 * Only ptrace can touch these memory addresses, but it doesn't change
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001928 * the map_count or the pages allocated. So no possibility of crashing
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 * exists while dumping the mm->vm_next areas to the core file.
1930 */
1931
1932 /* alloc memory for large data structures: too large to be on stack */
1933 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1934 if (!elf)
WANG Cong5f719552008-05-06 12:45:35 +08001935 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936
1937 segs = current->mm->map_count;
1938#ifdef ELF_CORE_EXTRA_PHDRS
1939 segs += ELF_CORE_EXTRA_PHDRS;
1940#endif
1941
Roland McGrathf47aef52007-01-26 00:56:49 -08001942 gate_vma = get_gate_vma(current);
1943 if (gate_vma != NULL)
1944 segs++;
1945
Roland McGrath3aba4812008-01-30 13:31:44 +01001946 /*
1947 * Collect all the non-memory information about the process for the
1948 * notes. This also sets up the file header.
1949 */
1950 if (!fill_note_info(elf, segs + 1, /* including notes section */
1951 &info, signr, regs))
1952 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953
1954 has_dumped = 1;
1955 current->flags |= PF_DUMPCORE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956
1957 fs = get_fs();
1958 set_fs(KERNEL_DS);
1959
1960 DUMP_WRITE(elf, sizeof(*elf));
1961 offset += sizeof(*elf); /* Elf header */
Petr Vandroveca7a0d862006-10-13 18:42:07 +02001962 offset += (segs + 1) * sizeof(struct elf_phdr); /* Program headers */
1963 foffset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964
1965 /* Write notes phdr entry */
1966 {
1967 struct elf_phdr phdr;
Roland McGrath3aba4812008-01-30 13:31:44 +01001968 size_t sz = get_note_info_size(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969
Michael Ellermane5501492007-09-19 14:38:12 +10001970 sz += elf_coredump_extra_notes_size();
Dwayne Grant McConnellbf1ab972006-11-23 00:46:37 +01001971
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 fill_elf_note_phdr(&phdr, sz, offset);
1973 offset += sz;
1974 DUMP_WRITE(&phdr, sizeof(phdr));
1975 }
1976
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1978
Kawai, Hidehiroa1b59e82007-07-19 01:48:29 -07001979 /*
1980 * We must use the same mm->flags while dumping core to avoid
1981 * inconsistency between the program headers and bodies, otherwise an
1982 * unusable core file can be generated.
1983 */
1984 mm_flags = current->mm->flags;
1985
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 /* Write program headers for segments dump */
Roland McGrathf47aef52007-01-26 00:56:49 -08001987 for (vma = first_vma(current, gate_vma); vma != NULL;
1988 vma = next_vma(vma, gate_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 struct elf_phdr phdr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990
1991 phdr.p_type = PT_LOAD;
1992 phdr.p_offset = offset;
1993 phdr.p_vaddr = vma->vm_start;
1994 phdr.p_paddr = 0;
Roland McGrath82df3972007-10-16 23:27:02 -07001995 phdr.p_filesz = vma_dump_size(vma, mm_flags);
1996 phdr.p_memsz = vma->vm_end - vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 offset += phdr.p_filesz;
1998 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001999 if (vma->vm_flags & VM_WRITE)
2000 phdr.p_flags |= PF_W;
2001 if (vma->vm_flags & VM_EXEC)
2002 phdr.p_flags |= PF_X;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 phdr.p_align = ELF_EXEC_PAGESIZE;
2004
2005 DUMP_WRITE(&phdr, sizeof(phdr));
2006 }
2007
2008#ifdef ELF_CORE_WRITE_EXTRA_PHDRS
2009 ELF_CORE_WRITE_EXTRA_PHDRS;
2010#endif
2011
2012 /* write out the notes section */
Roland McGrath3aba4812008-01-30 13:31:44 +01002013 if (!write_note_info(&info, file, &foffset))
2014 goto end_coredump;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015
Michael Ellermane5501492007-09-19 14:38:12 +10002016 if (elf_coredump_extra_notes_write(file, &foffset))
2017 goto end_coredump;
Dwayne Grant McConnellbf1ab972006-11-23 00:46:37 +01002018
Andi Kleend025c9d2006-09-30 23:29:28 -07002019 /* Align to page */
2020 DUMP_SEEK(dataoff - foffset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021
Roland McGrathf47aef52007-01-26 00:56:49 -08002022 for (vma = first_vma(current, gate_vma); vma != NULL;
2023 vma = next_vma(vma, gate_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 unsigned long addr;
Roland McGrath82df3972007-10-16 23:27:02 -07002025 unsigned long end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002026
Roland McGrath82df3972007-10-16 23:27:02 -07002027 end = vma->vm_start + vma_dump_size(vma, mm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002028
Roland McGrath82df3972007-10-16 23:27:02 -07002029 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07002030 struct page *page;
WANG Cong4220b7f2008-04-29 01:01:18 -07002031 struct vm_area_struct *tmp_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032
2033 if (get_user_pages(current, current->mm, addr, 1, 0, 1,
WANG Cong4220b7f2008-04-29 01:01:18 -07002034 &page, &tmp_vma) <= 0) {
Andi Kleend025c9d2006-09-30 23:29:28 -07002035 DUMP_SEEK(PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 } else {
Nick Piggin557ed1f2007-10-16 01:24:40 -07002037 if (page == ZERO_PAGE(0)) {
Brian Pomerantz03221702007-04-01 23:49:41 -07002038 if (!dump_seek(file, PAGE_SIZE)) {
2039 page_cache_release(page);
2040 goto end_coredump;
2041 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 } else {
2043 void *kaddr;
WANG Cong4220b7f2008-04-29 01:01:18 -07002044 flush_cache_page(tmp_vma, addr,
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07002045 page_to_pfn(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 kaddr = kmap(page);
2047 if ((size += PAGE_SIZE) > limit ||
2048 !dump_write(file, kaddr,
2049 PAGE_SIZE)) {
2050 kunmap(page);
2051 page_cache_release(page);
2052 goto end_coredump;
2053 }
2054 kunmap(page);
2055 }
2056 page_cache_release(page);
2057 }
2058 }
2059 }
2060
2061#ifdef ELF_CORE_WRITE_EXTRA_DATA
2062 ELF_CORE_WRITE_EXTRA_DATA;
2063#endif
2064
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065end_coredump:
2066 set_fs(fs);
2067
2068cleanup:
Roland McGrath3aba4812008-01-30 13:31:44 +01002069 free_note_info(&info);
WANG Cong5f719552008-05-06 12:45:35 +08002070 kfree(elf);
2071out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 return has_dumped;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073}
2074
2075#endif /* USE_ELF_CORE_DUMP */
2076
2077static int __init init_elf_binfmt(void)
2078{
2079 return register_binfmt(&elf_format);
2080}
2081
2082static void __exit exit_elf_binfmt(void)
2083{
2084 /* Remove the COFF and ELF loaders. */
2085 unregister_binfmt(&elf_format);
2086}
2087
2088core_initcall(init_elf_binfmt);
2089module_exit(exit_elf_binfmt);
2090MODULE_LICENSE("GPL");