blob: f34078d702d3bdbe1ee69bc1122b63d3facf5009 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/mm.h>
16#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/errno.h>
18#include <linux/signal.h>
19#include <linux/binfmts.h>
20#include <linux/string.h>
21#include <linux/file.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/personality.h>
24#include <linux/elfcore.h>
25#include <linux/init.h>
26#include <linux/highuid.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/compiler.h>
28#include <linux/highmem.h>
29#include <linux/pagemap.h>
30#include <linux/security.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/random.h>
Jesper Juhlf4e5cc22006-06-23 02:05:35 -070032#include <linux/elf.h>
Alexey Dobriyan7e80d0d2007-05-08 00:28:59 -070033#include <linux/utsname.h>
Daisuke HATAYAMA088e7af2010-03-05 13:44:06 -080034#include <linux/coredump.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/uaccess.h>
36#include <asm/param.h>
37#include <asm/page.h>
38
Jesper Juhlf4e5cc22006-06-23 02:05:35 -070039static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs);
40static int load_elf_library(struct file *);
Andrew Mortonbb1ad822008-01-30 13:31:07 +010041static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
42 int, int, unsigned long);
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Linus Torvalds1da177e2005-04-16 15:20:36 -070044/*
45 * If we don't support core dumping, then supply a NULL so we
46 * don't even try.
47 */
Christoph Hellwig698ba7b2009-12-15 16:47:37 -080048#ifdef CONFIG_ELF_CORE
Masami Hiramatsuf6151df2009-12-17 15:27:16 -080049static int elf_core_dump(struct coredump_params *cprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#else
51#define elf_core_dump NULL
52#endif
53
54#if ELF_EXEC_PAGESIZE > PAGE_SIZE
Jesper Juhlf4e5cc22006-06-23 02:05:35 -070055#define ELF_MIN_ALIGN ELF_EXEC_PAGESIZE
Linus Torvalds1da177e2005-04-16 15:20:36 -070056#else
Jesper Juhlf4e5cc22006-06-23 02:05:35 -070057#define ELF_MIN_ALIGN PAGE_SIZE
Linus Torvalds1da177e2005-04-16 15:20:36 -070058#endif
59
60#ifndef ELF_CORE_EFLAGS
61#define ELF_CORE_EFLAGS 0
62#endif
63
64#define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
65#define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
66#define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
67
68static struct linux_binfmt elf_format = {
Mikael Petterssonf670d0e2011-01-12 17:00:02 -080069 .module = THIS_MODULE,
70 .load_binary = load_elf_binary,
71 .load_shlib = load_elf_library,
72 .core_dump = elf_core_dump,
73 .min_coredump = ELF_EXEC_PAGESIZE,
Linus Torvalds1da177e2005-04-16 15:20:36 -070074};
75
Andrew Mortond4e3cc32007-07-21 04:37:32 -070076#define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
78static int set_brk(unsigned long start, unsigned long end)
79{
80 start = ELF_PAGEALIGN(start);
81 end = ELF_PAGEALIGN(end);
82 if (end > start) {
83 unsigned long addr;
84 down_write(&current->mm->mmap_sem);
85 addr = do_brk(start, end - start);
86 up_write(&current->mm->mmap_sem);
87 if (BAD_ADDR(addr))
88 return addr;
89 }
90 current->mm->start_brk = current->mm->brk = end;
91 return 0;
92}
93
Linus Torvalds1da177e2005-04-16 15:20:36 -070094/* We need to explicitly zero any fractional pages
95 after the data section (i.e. bss). This would
96 contain the junk from the file that should not
Jesper Juhlf4e5cc22006-06-23 02:05:35 -070097 be in memory
98 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070099static int padzero(unsigned long elf_bss)
100{
101 unsigned long nbyte;
102
103 nbyte = ELF_PAGEOFFSET(elf_bss);
104 if (nbyte) {
105 nbyte = ELF_MIN_ALIGN - nbyte;
106 if (clear_user((void __user *) elf_bss, nbyte))
107 return -EFAULT;
108 }
109 return 0;
110}
111
Ohad Ben-Cohen09c6dd32008-02-03 18:05:15 +0200112/* Let's use some macros to make this stack manipulation a little clearer */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113#ifdef CONFIG_STACK_GROWSUP
114#define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
115#define STACK_ROUND(sp, items) \
116 ((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700117#define STACK_ALLOC(sp, len) ({ \
118 elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
119 old_sp; })
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120#else
121#define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
122#define STACK_ROUND(sp, items) \
123 (((unsigned long) (sp - items)) &~ 15UL)
124#define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
125#endif
126
Nathan Lynch483fad12008-07-22 04:48:46 +1000127#ifndef ELF_BASE_PLATFORM
128/*
129 * AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
130 * If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
131 * will be copied to the user stack in the same manner as AT_PLATFORM.
132 */
133#define ELF_BASE_PLATFORM NULL
134#endif
135
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136static int
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700137create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
Andi Kleend20894a2008-02-08 04:21:54 -0800138 unsigned long load_addr, unsigned long interp_load_addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139{
140 unsigned long p = bprm->p;
141 int argc = bprm->argc;
142 int envc = bprm->envc;
143 elf_addr_t __user *argv;
144 elf_addr_t __user *envp;
145 elf_addr_t __user *sp;
146 elf_addr_t __user *u_platform;
Nathan Lynch483fad12008-07-22 04:48:46 +1000147 elf_addr_t __user *u_base_platform;
Kees Cookf06295b2009-01-07 18:08:52 -0800148 elf_addr_t __user *u_rand_bytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 const char *k_platform = ELF_PLATFORM;
Nathan Lynch483fad12008-07-22 04:48:46 +1000150 const char *k_base_platform = ELF_BASE_PLATFORM;
Kees Cookf06295b2009-01-07 18:08:52 -0800151 unsigned char k_rand_bytes[16];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 int items;
153 elf_addr_t *elf_info;
154 int ei_index = 0;
David Howells86a264a2008-11-14 10:39:18 +1100155 const struct cred *cred = current_cred();
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700156 struct vm_area_struct *vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
158 /*
Franck Bui-Huud68c9d62007-10-16 23:30:24 -0700159 * In some cases (e.g. Hyper-Threading), we want to avoid L1
160 * evictions by the processes running on the same package. One
161 * thing we can do is to shuffle the initial stack for them.
162 */
163
164 p = arch_align_stack(p);
165
166 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 * If this architecture has a platform capability string, copy it
168 * to userspace. In some cases (Sparc), this info is impossible
169 * for userspace to get any other way, in others (i386) it is
170 * merely difficult.
171 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 u_platform = NULL;
173 if (k_platform) {
174 size_t len = strlen(k_platform) + 1;
175
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
177 if (__copy_to_user(u_platform, k_platform, len))
178 return -EFAULT;
179 }
180
Nathan Lynch483fad12008-07-22 04:48:46 +1000181 /*
182 * If this architecture has a "base" platform capability
183 * string, copy it to userspace.
184 */
185 u_base_platform = NULL;
186 if (k_base_platform) {
187 size_t len = strlen(k_base_platform) + 1;
188
189 u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
190 if (__copy_to_user(u_base_platform, k_base_platform, len))
191 return -EFAULT;
192 }
193
Kees Cookf06295b2009-01-07 18:08:52 -0800194 /*
195 * Generate 16 random bytes for userspace PRNG seeding.
196 */
197 get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
198 u_rand_bytes = (elf_addr_t __user *)
199 STACK_ALLOC(p, sizeof(k_rand_bytes));
200 if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
201 return -EFAULT;
202
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 /* Create the ELF interpreter info */
Jesper Juhl785d5572006-06-23 02:05:35 -0700204 elf_info = (elf_addr_t *)current->mm->saved_auxv;
Olaf Hering4f9a58d2007-10-16 23:30:12 -0700205 /* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206#define NEW_AUX_ENT(id, val) \
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700207 do { \
Jesper Juhl785d5572006-06-23 02:05:35 -0700208 elf_info[ei_index++] = id; \
209 elf_info[ei_index++] = val; \
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700210 } while (0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211
212#ifdef ARCH_DLINFO
213 /*
214 * ARCH_DLINFO must come first so PPC can do its special alignment of
215 * AUXV.
Olaf Hering4f9a58d2007-10-16 23:30:12 -0700216 * update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
217 * ARCH_DLINFO changes
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 */
219 ARCH_DLINFO;
220#endif
221 NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
222 NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
223 NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
224 NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700225 NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
227 NEW_AUX_ENT(AT_BASE, interp_load_addr);
228 NEW_AUX_ENT(AT_FLAGS, 0);
229 NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
David Howells86a264a2008-11-14 10:39:18 +1100230 NEW_AUX_ENT(AT_UID, cred->uid);
231 NEW_AUX_ENT(AT_EUID, cred->euid);
232 NEW_AUX_ENT(AT_GID, cred->gid);
233 NEW_AUX_ENT(AT_EGID, cred->egid);
Jesper Juhl785d5572006-06-23 02:05:35 -0700234 NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
Kees Cookf06295b2009-01-07 18:08:52 -0800235 NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
John Reiser65191082008-07-21 14:21:32 -0700236 NEW_AUX_ENT(AT_EXECFN, bprm->exec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 if (k_platform) {
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700238 NEW_AUX_ENT(AT_PLATFORM,
Jesper Juhl785d5572006-06-23 02:05:35 -0700239 (elf_addr_t)(unsigned long)u_platform);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 }
Nathan Lynch483fad12008-07-22 04:48:46 +1000241 if (k_base_platform) {
242 NEW_AUX_ENT(AT_BASE_PLATFORM,
243 (elf_addr_t)(unsigned long)u_base_platform);
244 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
Jesper Juhl785d5572006-06-23 02:05:35 -0700246 NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 }
248#undef NEW_AUX_ENT
249 /* AT_NULL is zero; clear the rest too */
250 memset(&elf_info[ei_index], 0,
251 sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
252
253 /* And advance past the AT_NULL entry. */
254 ei_index += 2;
255
256 sp = STACK_ADD(p, ei_index);
257
Andi Kleend20894a2008-02-08 04:21:54 -0800258 items = (argc + 1) + (envc + 1) + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 bprm->p = STACK_ROUND(sp, items);
260
261 /* Point sp at the lowest address on the stack */
262#ifdef CONFIG_STACK_GROWSUP
263 sp = (elf_addr_t __user *)bprm->p - items - ei_index;
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700264 bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265#else
266 sp = (elf_addr_t __user *)bprm->p;
267#endif
268
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700269
270 /*
271 * Grow the stack manually; some architectures have a limit on how
272 * far ahead a user-space access may be in order to grow the stack.
273 */
274 vma = find_extend_vma(current->mm, bprm->p);
275 if (!vma)
276 return -EFAULT;
277
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 /* Now, let's put argc (and argv, envp if appropriate) on the stack */
279 if (__put_user(argc, sp++))
280 return -EFAULT;
Andi Kleend20894a2008-02-08 04:21:54 -0800281 argv = sp;
282 envp = argv + argc + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283
284 /* Populate argv and envp */
Greg Kroah-Hartmana84a5052005-05-11 00:10:44 -0700285 p = current->mm->arg_end = current->mm->arg_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 while (argc-- > 0) {
287 size_t len;
Heiko Carstens841d5fb2006-12-06 20:36:35 -0800288 if (__put_user((elf_addr_t)p, argv++))
289 return -EFAULT;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700290 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
291 if (!len || len > MAX_ARG_STRLEN)
WANG Cong23c49712008-05-08 21:52:33 +0800292 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293 p += len;
294 }
295 if (__put_user(0, argv))
296 return -EFAULT;
297 current->mm->arg_end = current->mm->env_start = p;
298 while (envc-- > 0) {
299 size_t len;
Heiko Carstens841d5fb2006-12-06 20:36:35 -0800300 if (__put_user((elf_addr_t)p, envp++))
301 return -EFAULT;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700302 len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
303 if (!len || len > MAX_ARG_STRLEN)
WANG Cong23c49712008-05-08 21:52:33 +0800304 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 p += len;
306 }
307 if (__put_user(0, envp))
308 return -EFAULT;
309 current->mm->env_end = p;
310
311 /* Put the elf_info on the stack in the right place. */
312 sp = (elf_addr_t __user *)envp + 1;
313 if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
314 return -EFAULT;
315 return 0;
316}
317
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318static unsigned long elf_map(struct file *filep, unsigned long addr,
Jiri Kosinacc503c12008-01-30 13:31:07 +0100319 struct elf_phdr *eppnt, int prot, int type,
320 unsigned long total_size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321{
322 unsigned long map_addr;
Jiri Kosinacc503c12008-01-30 13:31:07 +0100323 unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
324 unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
325 addr = ELF_PAGESTART(addr);
326 size = ELF_PAGEALIGN(size);
Jan Kratochvil60bfba72007-07-15 23:40:06 -0700327
Andrew Mortond4e3cc32007-07-21 04:37:32 -0700328 /* mmap() will return -EINVAL if given a zero size, but a
329 * segment with zero filesize is perfectly valid */
Jiri Kosinacc503c12008-01-30 13:31:07 +0100330 if (!size)
331 return addr;
332
333 down_write(&current->mm->mmap_sem);
334 /*
335 * total_size is the size of the ELF (interpreter) image.
336 * The _first_ mmap needs to know the full size, otherwise
337 * randomization might put this image into an overlapping
338 * position with the ELF binary image. (since size < total_size)
339 * So we first map the 'big' image - and unmap the remainder at
340 * the end. (which unmap is needed for ELF images with holes.)
341 */
342 if (total_size) {
343 total_size = ELF_PAGEALIGN(total_size);
344 map_addr = do_mmap(filep, addr, total_size, prot, type, off);
345 if (!BAD_ADDR(map_addr))
346 do_munmap(current->mm, map_addr+size, total_size-size);
347 } else
348 map_addr = do_mmap(filep, addr, size, prot, type, off);
349
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350 up_write(&current->mm->mmap_sem);
351 return(map_addr);
352}
353
Jiri Kosinacc503c12008-01-30 13:31:07 +0100354static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
355{
356 int i, first_idx = -1, last_idx = -1;
357
358 for (i = 0; i < nr; i++) {
359 if (cmds[i].p_type == PT_LOAD) {
360 last_idx = i;
361 if (first_idx == -1)
362 first_idx = i;
363 }
364 }
365 if (first_idx == -1)
366 return 0;
367
368 return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
369 ELF_PAGESTART(cmds[first_idx].p_vaddr);
370}
371
372
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373/* This is much more generalized than the library routine read function,
374 so we keep this separate. Technically the library read function
375 is only provided so that we can read a.out libraries that have
376 an ELF header */
377
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700378static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
Jiri Kosinacc503c12008-01-30 13:31:07 +0100379 struct file *interpreter, unsigned long *interp_map_addr,
380 unsigned long no_base)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381{
382 struct elf_phdr *elf_phdata;
383 struct elf_phdr *eppnt;
384 unsigned long load_addr = 0;
385 int load_addr_set = 0;
386 unsigned long last_bss = 0, elf_bss = 0;
387 unsigned long error = ~0UL;
Jiri Kosinacc503c12008-01-30 13:31:07 +0100388 unsigned long total_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 int retval, i, size;
390
391 /* First of all, some simple consistency checks */
392 if (interp_elf_ex->e_type != ET_EXEC &&
393 interp_elf_ex->e_type != ET_DYN)
394 goto out;
395 if (!elf_check_arch(interp_elf_ex))
396 goto out;
397 if (!interpreter->f_op || !interpreter->f_op->mmap)
398 goto out;
399
400 /*
401 * If the size of this structure has changed, then punt, since
402 * we will be doing the wrong thing.
403 */
404 if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
405 goto out;
406 if (interp_elf_ex->e_phnum < 1 ||
407 interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
408 goto out;
409
410 /* Now read in all of the header information */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
412 if (size > ELF_MIN_ALIGN)
413 goto out;
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700414 elf_phdata = kmalloc(size, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 if (!elf_phdata)
416 goto out;
417
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700418 retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
Mikael Petterssonf670d0e2011-01-12 17:00:02 -0800419 (char *)elf_phdata, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 error = -EIO;
421 if (retval != size) {
422 if (retval < 0)
423 error = retval;
424 goto out_close;
425 }
426
Jiri Kosinacc503c12008-01-30 13:31:07 +0100427 total_size = total_mapping_size(elf_phdata, interp_elf_ex->e_phnum);
428 if (!total_size) {
429 error = -EINVAL;
430 goto out_close;
431 }
432
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 eppnt = elf_phdata;
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700434 for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
435 if (eppnt->p_type == PT_LOAD) {
436 int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
437 int elf_prot = 0;
438 unsigned long vaddr = 0;
439 unsigned long k, map_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700441 if (eppnt->p_flags & PF_R)
442 elf_prot = PROT_READ;
443 if (eppnt->p_flags & PF_W)
444 elf_prot |= PROT_WRITE;
445 if (eppnt->p_flags & PF_X)
446 elf_prot |= PROT_EXEC;
447 vaddr = eppnt->p_vaddr;
448 if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
449 elf_type |= MAP_FIXED;
Jiri Kosinacc503c12008-01-30 13:31:07 +0100450 else if (no_base && interp_elf_ex->e_type == ET_DYN)
451 load_addr = -vaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700453 map_addr = elf_map(interpreter, load_addr + vaddr,
Andrew Mortonbb1ad822008-01-30 13:31:07 +0100454 eppnt, elf_prot, elf_type, total_size);
Jiri Kosinacc503c12008-01-30 13:31:07 +0100455 total_size = 0;
456 if (!*interp_map_addr)
457 *interp_map_addr = map_addr;
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700458 error = map_addr;
459 if (BAD_ADDR(map_addr))
460 goto out_close;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700462 if (!load_addr_set &&
463 interp_elf_ex->e_type == ET_DYN) {
464 load_addr = map_addr - ELF_PAGESTART(vaddr);
465 load_addr_set = 1;
466 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700468 /*
469 * Check to see if the section's size will overflow the
470 * allowed task size. Note that p_filesz must always be
471 * <= p_memsize so it's only necessary to check p_memsz.
472 */
473 k = load_addr + eppnt->p_vaddr;
Chuck Ebbertce510592006-07-03 00:24:14 -0700474 if (BAD_ADDR(k) ||
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700475 eppnt->p_filesz > eppnt->p_memsz ||
476 eppnt->p_memsz > TASK_SIZE ||
477 TASK_SIZE - eppnt->p_memsz < k) {
478 error = -ENOMEM;
479 goto out_close;
480 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700482 /*
483 * Find the end of the file mapping for this phdr, and
484 * keep track of the largest address we see for this.
485 */
486 k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
487 if (k > elf_bss)
488 elf_bss = k;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700490 /*
491 * Do the same thing for the memory mapping - between
492 * elf_bss and last_bss is the bss section.
493 */
494 k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
495 if (k > last_bss)
496 last_bss = k;
497 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 }
499
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 if (last_bss > elf_bss) {
Roland McGrath752015d2009-09-08 19:49:40 -0700501 /*
502 * Now fill out the bss section. First pad the last page up
503 * to the page boundary, and then perform a mmap to make sure
504 * that there are zero-mapped pages up to and including the
505 * last bss page.
506 */
507 if (padzero(elf_bss)) {
508 error = -EFAULT;
509 goto out_close;
510 }
511
512 /* What we have mapped so far */
513 elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
514
515 /* Map the last of the bss segment */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 down_write(&current->mm->mmap_sem);
517 error = do_brk(elf_bss, last_bss - elf_bss);
518 up_write(&current->mm->mmap_sem);
519 if (BAD_ADDR(error))
520 goto out_close;
521 }
522
Jiri Kosinacc503c12008-01-30 13:31:07 +0100523 error = load_addr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
525out_close:
526 kfree(elf_phdata);
527out:
528 return error;
529}
530
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531/*
532 * These are the functions used to load ELF style executables and shared
533 * libraries. There is no binary dependent code anywhere else.
534 */
535
536#define INTERPRETER_NONE 0
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537#define INTERPRETER_ELF 2
538
Andi Kleen913bd902006-03-25 16:29:09 +0100539#ifndef STACK_RND_MASK
James Bottomleyd1cabd62007-03-16 13:38:35 -0800540#define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
Andi Kleen913bd902006-03-25 16:29:09 +0100541#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542
543static unsigned long randomize_stack_top(unsigned long stack_top)
544{
545 unsigned int random_variable = 0;
546
Andi Kleenc16b63e02006-09-26 10:52:28 +0200547 if ((current->flags & PF_RANDOMIZE) &&
548 !(current->personality & ADDR_NO_RANDOMIZE)) {
Andi Kleen913bd902006-03-25 16:29:09 +0100549 random_variable = get_random_int() & STACK_RND_MASK;
550 random_variable <<= PAGE_SHIFT;
551 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552#ifdef CONFIG_STACK_GROWSUP
Andi Kleen913bd902006-03-25 16:29:09 +0100553 return PAGE_ALIGN(stack_top) + random_variable;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554#else
Andi Kleen913bd902006-03-25 16:29:09 +0100555 return PAGE_ALIGN(stack_top) - random_variable;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556#endif
557}
558
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700559static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560{
561 struct file *interpreter = NULL; /* to shut gcc up */
562 unsigned long load_addr = 0, load_bias = 0;
563 int load_addr_set = 0;
564 char * elf_interpreter = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 unsigned long error;
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700566 struct elf_phdr *elf_ppnt, *elf_phdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 unsigned long elf_bss, elf_brk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 int retval, i;
569 unsigned int size;
Jiri Kosinacc503c12008-01-30 13:31:07 +0100570 unsigned long elf_entry;
571 unsigned long interp_load_addr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 unsigned long start_code, end_code, start_data, end_data;
David Daney1a530a62011-03-22 16:34:48 -0700573 unsigned long reloc_func_desc __maybe_unused = 0;
David Rientjes8de61e62006-12-06 20:40:16 -0800574 int executable_stack = EXSTACK_DEFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 unsigned long def_flags = 0;
576 struct {
577 struct elfhdr elf_ex;
578 struct elfhdr interp_elf_ex;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 } *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 */
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700588 loc->elf_ex = *((struct elfhdr *)bprm->buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
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;
Mikael Petterssonf670d0e2011-01-12 17:00:02 -0800599 if (!bprm->file->f_op || !bprm->file->f_op->mmap)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 goto out;
601
602 /* Now read in all of the header information */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 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;
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700610 elf_phdata = kmalloc(size, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 if (!elf_phdata)
612 goto out;
613
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700614 retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
615 (char *)elf_phdata, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 if (retval != size) {
617 if (retval >= 0)
618 retval = -EIO;
619 goto out_free_ph;
620 }
621
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 elf_ppnt = elf_phdata;
623 elf_bss = 0;
624 elf_brk = 0;
625
626 start_code = ~0UL;
627 end_code = 0;
628 start_data = 0;
629 end_data = 0;
630
631 for (i = 0; i < loc->elf_ex.e_phnum; i++) {
632 if (elf_ppnt->p_type == PT_INTERP) {
633 /* This is the program interpreter used for
634 * shared libraries - for now assume that this
635 * is an a.out format binary
636 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 retval = -ENOEXEC;
638 if (elf_ppnt->p_filesz > PATH_MAX ||
639 elf_ppnt->p_filesz < 2)
Al Viroe7b9b552009-03-29 16:31:16 -0400640 goto out_free_ph;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641
642 retval = -ENOMEM;
Jesper Juhl792db3a2006-01-09 20:54:45 -0800643 elf_interpreter = kmalloc(elf_ppnt->p_filesz,
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700644 GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 if (!elf_interpreter)
Al Viroe7b9b552009-03-29 16:31:16 -0400646 goto out_free_ph;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647
648 retval = kernel_read(bprm->file, elf_ppnt->p_offset,
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700649 elf_interpreter,
650 elf_ppnt->p_filesz);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 if (retval != elf_ppnt->p_filesz) {
652 if (retval >= 0)
653 retval = -EIO;
654 goto out_free_interp;
655 }
656 /* make sure path is NULL terminated */
657 retval = -ENOEXEC;
658 if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
659 goto out_free_interp;
660
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 interpreter = open_exec(elf_interpreter);
662 retval = PTR_ERR(interpreter);
663 if (IS_ERR(interpreter))
664 goto out_free_interp;
Alexey Dobriyan1fb84492007-01-26 00:57:16 -0800665
666 /*
667 * If the binary is not readable then enforce
668 * mm->dumpable = 0 regardless of the interpreter's
669 * permissions.
670 */
671 if (file_permission(interpreter, MAY_READ) < 0)
672 bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
673
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700674 retval = kernel_read(interpreter, 0, bprm->buf,
675 BINPRM_BUF_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 if (retval != BINPRM_BUF_SIZE) {
677 if (retval >= 0)
678 retval = -EIO;
679 goto out_free_dentry;
680 }
681
682 /* Get the exec headers */
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700683 loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 break;
685 }
686 elf_ppnt++;
687 }
688
689 elf_ppnt = elf_phdata;
690 for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
691 if (elf_ppnt->p_type == PT_GNU_STACK) {
692 if (elf_ppnt->p_flags & PF_X)
693 executable_stack = EXSTACK_ENABLE_X;
694 else
695 executable_stack = EXSTACK_DISABLE_X;
696 break;
697 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
699 /* Some simple consistency checks for the interpreter */
700 if (elf_interpreter) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 retval = -ELIBBAD;
Andi Kleend20894a2008-02-08 04:21:54 -0800702 /* Not an ELF interpreter */
703 if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 goto out_free_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705 /* Verify the interpreter has a valid arch */
Andi Kleend20894a2008-02-08 04:21:54 -0800706 if (!elf_check_arch(&loc->interp_elf_ex))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 goto out_free_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 }
709
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 /* Flush all traces of the currently running executable */
711 retval = flush_old_exec(bprm);
712 if (retval)
713 goto out_free_dentry;
714
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 /* OK, This is the point of no return */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 current->flags &= ~PF_FORKNOEXEC;
717 current->mm->def_flags = def_flags;
718
719 /* Do this immediately, since STACK_TOP as used in setup_arg_pages
720 may depend on the personality. */
Martin Schwidefsky0b592682008-10-16 15:39:57 +0200721 SET_PERSONALITY(loc->elf_ex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 if (elf_read_implies_exec(loc->elf_ex, executable_stack))
723 current->personality |= READ_IMPLIES_EXEC;
724
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700725 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 current->flags |= PF_RANDOMIZE;
Linus Torvalds221af7f2010-01-28 22:14:42 -0800727
728 setup_new_exec(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729
730 /* Do this so that we can load the interpreter, if need be. We will
731 change some of these later */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 current->mm->free_area_cache = current->mm->mmap_base;
Wolfgang Wander1363c3c2005-06-21 17:14:49 -0700733 current->mm->cached_hole_size = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
735 executable_stack);
736 if (retval < 0) {
737 send_sig(SIGKILL, current, 0);
738 goto out_free_dentry;
739 }
740
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741 current->mm->start_stack = bprm->p;
742
André Goddard Rosaaf901ca2009-11-14 13:09:05 -0200743 /* Now we do a little grungy work by mmapping the ELF image into
Jiri Kosinacc503c12008-01-30 13:31:07 +0100744 the correct location in memory. */
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700745 for(i = 0, elf_ppnt = elf_phdata;
746 i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 int elf_prot = 0, elf_flags;
748 unsigned long k, vaddr;
749
750 if (elf_ppnt->p_type != PT_LOAD)
751 continue;
752
753 if (unlikely (elf_brk > elf_bss)) {
754 unsigned long nbyte;
755
756 /* There was a PT_LOAD segment with p_memsz > p_filesz
757 before this one. Map anonymous pages, if needed,
758 and clear the area. */
Mikael Petterssonf670d0e2011-01-12 17:00:02 -0800759 retval = set_brk(elf_bss + load_bias,
760 elf_brk + load_bias);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 if (retval) {
762 send_sig(SIGKILL, current, 0);
763 goto out_free_dentry;
764 }
765 nbyte = ELF_PAGEOFFSET(elf_bss);
766 if (nbyte) {
767 nbyte = ELF_MIN_ALIGN - nbyte;
768 if (nbyte > elf_brk - elf_bss)
769 nbyte = elf_brk - elf_bss;
770 if (clear_user((void __user *)elf_bss +
771 load_bias, nbyte)) {
772 /*
773 * This bss-zeroing can fail if the ELF
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700774 * file specifies odd protections. So
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 * we don't check the return value
776 */
777 }
778 }
779 }
780
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700781 if (elf_ppnt->p_flags & PF_R)
782 elf_prot |= PROT_READ;
783 if (elf_ppnt->p_flags & PF_W)
784 elf_prot |= PROT_WRITE;
785 if (elf_ppnt->p_flags & PF_X)
786 elf_prot |= PROT_EXEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700788 elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
790 vaddr = elf_ppnt->p_vaddr;
791 if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
792 elf_flags |= MAP_FIXED;
793 } else if (loc->elf_ex.e_type == ET_DYN) {
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700794 /* Try and get dynamic programs out of the way of the
795 * default mmap base, as well as whatever program they
796 * might try to exec. This is because the brk will
797 * follow the loader, and is not movable. */
Nicolas Pitree4eab082010-08-20 21:14:46 +0100798#if defined(CONFIG_X86) || defined(CONFIG_ARM)
Jiri Kosinacc503c12008-01-30 13:31:07 +0100799 load_bias = 0;
800#else
Linus Torvalds90cb28e2007-01-06 13:28:21 -0800801 load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
Jiri Kosinacc503c12008-01-30 13:31:07 +0100802#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803 }
804
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700805 error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
Andrew Mortonbb1ad822008-01-30 13:31:07 +0100806 elf_prot, elf_flags, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 if (BAD_ADDR(error)) {
808 send_sig(SIGKILL, current, 0);
Alexey Kuznetsovb140f2512007-05-08 00:31:57 -0700809 retval = IS_ERR((void *)error) ?
810 PTR_ERR((void*)error) : -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811 goto out_free_dentry;
812 }
813
814 if (!load_addr_set) {
815 load_addr_set = 1;
816 load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
817 if (loc->elf_ex.e_type == ET_DYN) {
818 load_bias += error -
819 ELF_PAGESTART(load_bias + vaddr);
820 load_addr += load_bias;
821 reloc_func_desc = load_bias;
822 }
823 }
824 k = elf_ppnt->p_vaddr;
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700825 if (k < start_code)
826 start_code = k;
827 if (start_data < k)
828 start_data = k;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829
830 /*
831 * Check to see if the section's size will overflow the
832 * allowed task size. Note that p_filesz must always be
833 * <= p_memsz so it is only necessary to check p_memsz.
834 */
Chuck Ebbertce510592006-07-03 00:24:14 -0700835 if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 elf_ppnt->p_memsz > TASK_SIZE ||
837 TASK_SIZE - elf_ppnt->p_memsz < k) {
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700838 /* set_brk can never work. Avoid overflows. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839 send_sig(SIGKILL, current, 0);
Alexey Kuznetsovb140f2512007-05-08 00:31:57 -0700840 retval = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 goto out_free_dentry;
842 }
843
844 k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
845
846 if (k > elf_bss)
847 elf_bss = k;
848 if ((elf_ppnt->p_flags & PF_X) && end_code < k)
849 end_code = k;
850 if (end_data < k)
851 end_data = k;
852 k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
853 if (k > elf_brk)
854 elf_brk = k;
855 }
856
857 loc->elf_ex.e_entry += load_bias;
858 elf_bss += load_bias;
859 elf_brk += load_bias;
860 start_code += load_bias;
861 end_code += load_bias;
862 start_data += load_bias;
863 end_data += load_bias;
864
865 /* Calling set_brk effectively mmaps the pages that we need
866 * for the bss and break sections. We must do this before
867 * mapping in the interpreter, to make sure it doesn't wind
868 * up getting placed where the bss needs to go.
869 */
870 retval = set_brk(elf_bss, elf_brk);
871 if (retval) {
872 send_sig(SIGKILL, current, 0);
873 goto out_free_dentry;
874 }
akpm@osdl.org6de50512005-10-11 08:29:08 -0700875 if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 send_sig(SIGSEGV, current, 0);
877 retval = -EFAULT; /* Nobody gets to see this, but.. */
878 goto out_free_dentry;
879 }
880
881 if (elf_interpreter) {
Andi Kleend20894a2008-02-08 04:21:54 -0800882 unsigned long uninitialized_var(interp_map_addr);
Jiri Kosinacc503c12008-01-30 13:31:07 +0100883
Andi Kleend20894a2008-02-08 04:21:54 -0800884 elf_entry = load_elf_interp(&loc->interp_elf_ex,
885 interpreter,
886 &interp_map_addr,
887 load_bias);
888 if (!IS_ERR((void *)elf_entry)) {
889 /*
890 * load_elf_interp() returns relocation
891 * adjustment
892 */
893 interp_load_addr = elf_entry;
894 elf_entry += loc->interp_elf_ex.e_entry;
Jiri Kosinacc503c12008-01-30 13:31:07 +0100895 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 if (BAD_ADDR(elf_entry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 force_sig(SIGSEGV, current);
Chuck Ebbertce510592006-07-03 00:24:14 -0700898 retval = IS_ERR((void *)elf_entry) ?
899 (int)elf_entry : -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 goto out_free_dentry;
901 }
902 reloc_func_desc = interp_load_addr;
903
904 allow_write_access(interpreter);
905 fput(interpreter);
906 kfree(elf_interpreter);
907 } else {
908 elf_entry = loc->elf_ex.e_entry;
Suresh Siddha5342fba2006-02-26 04:18:28 +0100909 if (BAD_ADDR(elf_entry)) {
Chuck Ebbertce510592006-07-03 00:24:14 -0700910 force_sig(SIGSEGV, current);
911 retval = -EINVAL;
Suresh Siddha5342fba2006-02-26 04:18:28 +0100912 goto out_free_dentry;
913 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 }
915
916 kfree(elf_phdata);
917
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918 set_binfmt(&elf_format);
919
Benjamin Herrenschmidt547ee842005-04-16 15:24:35 -0700920#ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
Martin Schwidefskyfc5243d2008-12-25 13:38:35 +0100921 retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
Benjamin Herrenschmidt547ee842005-04-16 15:24:35 -0700922 if (retval < 0) {
923 send_sig(SIGKILL, current, 0);
Roland McGrath18c8baf2005-04-28 15:17:19 -0700924 goto out;
Benjamin Herrenschmidt547ee842005-04-16 15:24:35 -0700925 }
926#endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
927
David Howellsa6f76f22008-11-14 10:39:24 +1100928 install_exec_creds(bprm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 current->flags &= ~PF_FORKNOEXEC;
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700930 retval = create_elf_tables(bprm, &loc->elf_ex,
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700931 load_addr, interp_load_addr);
Ollie Wildb6a2fea2007-07-19 01:48:16 -0700932 if (retval < 0) {
933 send_sig(SIGKILL, current, 0);
934 goto out;
935 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 /* N.B. passed_fileno might not be initialized? */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 current->mm->end_code = end_code;
938 current->mm->start_code = start_code;
939 current->mm->start_data = start_data;
940 current->mm->end_data = end_data;
941 current->mm->start_stack = bprm->p;
942
Jiri Kosinac1d171a2008-01-30 13:30:40 +0100943#ifdef arch_randomize_brk
Ingo Molnar32a93232008-02-06 22:39:44 +0100944 if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1))
Jiri Kosinac1d171a2008-01-30 13:30:40 +0100945 current->mm->brk = current->mm->start_brk =
946 arch_randomize_brk(current->mm);
947#endif
948
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 if (current->personality & MMAP_PAGE_ZERO) {
950 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
951 and some applications "depend" upon this behavior.
952 Since we do not have the power to recompile these, we
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700953 emulate the SVr4 behavior. Sigh. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 down_write(&current->mm->mmap_sem);
955 error = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
956 MAP_FIXED | MAP_PRIVATE, 0);
957 up_write(&current->mm->mmap_sem);
958 }
959
960#ifdef ELF_PLAT_INIT
961 /*
962 * The ABI may specify that certain registers be set up in special
963 * ways (on i386 %edx is the address of a DT_FINI function, for
964 * example. In addition, it may also specify (eg, PowerPC64 ELF)
965 * that the e_entry field is the address of the function descriptor
966 * for the startup routine, rather than the address of the startup
967 * routine itself. This macro performs whatever initialization to
968 * the regs structure is required as well as any relocations to the
969 * function descriptor entries when executing dynamically links apps.
970 */
971 ELF_PLAT_INIT(regs, reloc_func_desc);
972#endif
973
974 start_thread(regs, elf_entry, bprm->p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 retval = 0;
976out:
977 kfree(loc);
978out_ret:
979 return retval;
980
981 /* error cleanup */
982out_free_dentry:
983 allow_write_access(interpreter);
984 if (interpreter)
985 fput(interpreter);
986out_free_interp:
Jesper Juhlf99d49a2005-11-07 01:01:34 -0800987 kfree(elf_interpreter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988out_free_ph:
989 kfree(elf_phdata);
990 goto out;
991}
992
993/* This is really simpleminded and specialized - we are loading an
994 a.out library that is given an ELF header. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995static int load_elf_library(struct file *file)
996{
997 struct elf_phdr *elf_phdata;
998 struct elf_phdr *eppnt;
999 unsigned long elf_bss, bss, len;
1000 int retval, error, i, j;
1001 struct elfhdr elf_ex;
1002
1003 error = -ENOEXEC;
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001004 retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 if (retval != sizeof(elf_ex))
1006 goto out;
1007
1008 if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1009 goto out;
1010
1011 /* First of all, some simple consistency checks */
1012 if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001013 !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 goto out;
1015
1016 /* Now read in all of the header information */
1017
1018 j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1019 /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1020
1021 error = -ENOMEM;
1022 elf_phdata = kmalloc(j, GFP_KERNEL);
1023 if (!elf_phdata)
1024 goto out;
1025
1026 eppnt = elf_phdata;
1027 error = -ENOEXEC;
1028 retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1029 if (retval != j)
1030 goto out_free_ph;
1031
1032 for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1033 if ((eppnt + i)->p_type == PT_LOAD)
1034 j++;
1035 if (j != 1)
1036 goto out_free_ph;
1037
1038 while (eppnt->p_type != PT_LOAD)
1039 eppnt++;
1040
1041 /* Now use mmap to map the library into memory. */
1042 down_write(&current->mm->mmap_sem);
1043 error = do_mmap(file,
1044 ELF_PAGESTART(eppnt->p_vaddr),
1045 (eppnt->p_filesz +
1046 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1047 PROT_READ | PROT_WRITE | PROT_EXEC,
1048 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1049 (eppnt->p_offset -
1050 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1051 up_write(&current->mm->mmap_sem);
1052 if (error != ELF_PAGESTART(eppnt->p_vaddr))
1053 goto out_free_ph;
1054
1055 elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1056 if (padzero(elf_bss)) {
1057 error = -EFAULT;
1058 goto out_free_ph;
1059 }
1060
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001061 len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1062 ELF_MIN_ALIGN - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 bss = eppnt->p_memsz + eppnt->p_vaddr;
1064 if (bss > len) {
1065 down_write(&current->mm->mmap_sem);
1066 do_brk(len, bss - len);
1067 up_write(&current->mm->mmap_sem);
1068 }
1069 error = 0;
1070
1071out_free_ph:
1072 kfree(elf_phdata);
1073out:
1074 return error;
1075}
1076
Christoph Hellwig698ba7b2009-12-15 16:47:37 -08001077#ifdef CONFIG_ELF_CORE
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078/*
1079 * ELF core dumper
1080 *
1081 * Modelled on fs/exec.c:aout_core_dump()
1082 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1083 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084
1085/*
Roland McGrath82df3972007-10-16 23:27:02 -07001086 * Decide what to dump of a segment, part, all or none.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 */
Roland McGrath82df3972007-10-16 23:27:02 -07001088static unsigned long vma_dump_size(struct vm_area_struct *vma,
1089 unsigned long mm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090{
KOSAKI Motohiroe575f112008-10-18 20:27:08 -07001091#define FILTER(type) (mm_flags & (1UL << MMF_DUMP_##type))
1092
Roland McGrathe5b97dd2007-01-26 00:56:48 -08001093 /* The vma can be set up to tell us the answer directly. */
1094 if (vma->vm_flags & VM_ALWAYSDUMP)
Roland McGrath82df3972007-10-16 23:27:02 -07001095 goto whole;
Roland McGrathe5b97dd2007-01-26 00:56:48 -08001096
KOSAKI Motohiroe575f112008-10-18 20:27:08 -07001097 /* Hugetlb memory check */
1098 if (vma->vm_flags & VM_HUGETLB) {
1099 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1100 goto whole;
1101 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1102 goto whole;
1103 }
1104
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 /* Do not dump I/O mapped devices or special mappings */
1106 if (vma->vm_flags & (VM_IO | VM_RESERVED))
1107 return 0;
1108
Kawai, Hidehiroa1b59e82007-07-19 01:48:29 -07001109 /* By default, dump shared memory if mapped from an anonymous file. */
1110 if (vma->vm_flags & VM_SHARED) {
Roland McGrath82df3972007-10-16 23:27:02 -07001111 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0 ?
1112 FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1113 goto whole;
1114 return 0;
Kawai, Hidehiroa1b59e82007-07-19 01:48:29 -07001115 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116
Roland McGrath82df3972007-10-16 23:27:02 -07001117 /* Dump segments that have been written to. */
1118 if (vma->anon_vma && FILTER(ANON_PRIVATE))
1119 goto whole;
1120 if (vma->vm_file == NULL)
1121 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122
Roland McGrath82df3972007-10-16 23:27:02 -07001123 if (FILTER(MAPPED_PRIVATE))
1124 goto whole;
1125
1126 /*
1127 * If this looks like the beginning of a DSO or executable mapping,
1128 * check for an ELF header. If we find one, dump the first page to
1129 * aid in determining what was mapped here.
1130 */
Roland McGrath92dc07b2009-02-06 17:34:07 -08001131 if (FILTER(ELF_HEADERS) &&
1132 vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
Roland McGrath82df3972007-10-16 23:27:02 -07001133 u32 __user *header = (u32 __user *) vma->vm_start;
1134 u32 word;
Roland McGrath92dc07b2009-02-06 17:34:07 -08001135 mm_segment_t fs = get_fs();
Roland McGrath82df3972007-10-16 23:27:02 -07001136 /*
1137 * Doing it this way gets the constant folded by GCC.
1138 */
1139 union {
1140 u32 cmp;
1141 char elfmag[SELFMAG];
1142 } magic;
1143 BUILD_BUG_ON(SELFMAG != sizeof word);
1144 magic.elfmag[EI_MAG0] = ELFMAG0;
1145 magic.elfmag[EI_MAG1] = ELFMAG1;
1146 magic.elfmag[EI_MAG2] = ELFMAG2;
1147 magic.elfmag[EI_MAG3] = ELFMAG3;
Roland McGrath92dc07b2009-02-06 17:34:07 -08001148 /*
1149 * Switch to the user "segment" for get_user(),
1150 * then put back what elf_core_dump() had in place.
1151 */
1152 set_fs(USER_DS);
1153 if (unlikely(get_user(word, header)))
1154 word = 0;
1155 set_fs(fs);
1156 if (word == magic.cmp)
Roland McGrath82df3972007-10-16 23:27:02 -07001157 return PAGE_SIZE;
1158 }
1159
1160#undef FILTER
1161
1162 return 0;
1163
1164whole:
1165 return vma->vm_end - vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166}
1167
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168/* An ELF note in memory */
1169struct memelfnote
1170{
1171 const char *name;
1172 int type;
1173 unsigned int datasz;
1174 void *data;
1175};
1176
1177static int notesize(struct memelfnote *en)
1178{
1179 int sz;
1180
1181 sz = sizeof(struct elf_note);
1182 sz += roundup(strlen(en->name) + 1, 4);
1183 sz += roundup(en->datasz, 4);
1184
1185 return sz;
1186}
1187
Andi Kleend025c9d2006-09-30 23:29:28 -07001188#define DUMP_WRITE(addr, nr, foffset) \
1189 do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190
Andi Kleend025c9d2006-09-30 23:29:28 -07001191static int alignfile(struct file *file, loff_t *foffset)
1192{
Petr Vandroveca7a0d862006-10-13 18:42:07 +02001193 static const char buf[4] = { 0, };
Andi Kleend025c9d2006-09-30 23:29:28 -07001194 DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1195 return 1;
1196}
1197
1198static int writenote(struct memelfnote *men, struct file *file,
1199 loff_t *foffset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200{
1201 struct elf_note en;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 en.n_namesz = strlen(men->name) + 1;
1203 en.n_descsz = men->datasz;
1204 en.n_type = men->type;
1205
Andi Kleend025c9d2006-09-30 23:29:28 -07001206 DUMP_WRITE(&en, sizeof(en), foffset);
1207 DUMP_WRITE(men->name, en.n_namesz, foffset);
1208 if (!alignfile(file, foffset))
1209 return 0;
1210 DUMP_WRITE(men->data, men->datasz, foffset);
1211 if (!alignfile(file, foffset))
1212 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213
1214 return 1;
1215}
1216#undef DUMP_WRITE
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217
Roland McGrath3aba4812008-01-30 13:31:44 +01001218static void fill_elf_header(struct elfhdr *elf, int segs,
1219 u16 machine, u32 flags, u8 osabi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220{
Cyrill Gorcunov6970c8e2008-04-29 01:01:18 -07001221 memset(elf, 0, sizeof(*elf));
1222
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1224 elf->e_ident[EI_CLASS] = ELF_CLASS;
1225 elf->e_ident[EI_DATA] = ELF_DATA;
1226 elf->e_ident[EI_VERSION] = EV_CURRENT;
1227 elf->e_ident[EI_OSABI] = ELF_OSABI;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228
1229 elf->e_type = ET_CORE;
Roland McGrath3aba4812008-01-30 13:31:44 +01001230 elf->e_machine = machine;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 elf->e_version = EV_CURRENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 elf->e_phoff = sizeof(struct elfhdr);
Roland McGrath3aba4812008-01-30 13:31:44 +01001233 elf->e_flags = flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 elf->e_ehsize = sizeof(struct elfhdr);
1235 elf->e_phentsize = sizeof(struct elf_phdr);
1236 elf->e_phnum = segs;
Cyrill Gorcunov6970c8e2008-04-29 01:01:18 -07001237
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 return;
1239}
1240
Andrew Morton8d6b5eee2006-09-25 23:32:04 -07001241static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242{
1243 phdr->p_type = PT_NOTE;
1244 phdr->p_offset = offset;
1245 phdr->p_vaddr = 0;
1246 phdr->p_paddr = 0;
1247 phdr->p_filesz = sz;
1248 phdr->p_memsz = 0;
1249 phdr->p_flags = 0;
1250 phdr->p_align = 0;
1251 return;
1252}
1253
1254static void fill_note(struct memelfnote *note, const char *name, int type,
1255 unsigned int sz, void *data)
1256{
1257 note->name = name;
1258 note->type = type;
1259 note->datasz = sz;
1260 note->data = data;
1261 return;
1262}
1263
1264/*
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001265 * fill up all the fields in prstatus from the given task struct, except
1266 * registers which need to be filled up separately.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267 */
1268static void fill_prstatus(struct elf_prstatus *prstatus,
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001269 struct task_struct *p, long signr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270{
1271 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1272 prstatus->pr_sigpend = p->pending.signal.sig[0];
1273 prstatus->pr_sighold = p->blocked.sig[0];
Oleg Nesterov3b34fc52009-06-17 16:27:38 -07001274 rcu_read_lock();
1275 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1276 rcu_read_unlock();
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001277 prstatus->pr_pid = task_pid_vnr(p);
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001278 prstatus->pr_pgrp = task_pgrp_vnr(p);
1279 prstatus->pr_sid = task_session_vnr(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280 if (thread_group_leader(p)) {
Frank Mayharf06febc2008-09-12 09:54:39 -07001281 struct task_cputime cputime;
1282
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 /*
Frank Mayharf06febc2008-09-12 09:54:39 -07001284 * This is the record for the group leader. It shows the
1285 * group-wide total, not its individual thread total.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 */
Frank Mayharf06febc2008-09-12 09:54:39 -07001287 thread_group_cputime(p, &cputime);
1288 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1289 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 } else {
1291 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1292 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1293 }
1294 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1295 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1296}
1297
1298static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1299 struct mm_struct *mm)
1300{
David Howellsc69e8d92008-11-14 10:39:19 +11001301 const struct cred *cred;
Greg Kroah-Hartmana84a5052005-05-11 00:10:44 -07001302 unsigned int i, len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303
1304 /* first copy the parameters from user space */
1305 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1306
1307 len = mm->arg_end - mm->arg_start;
1308 if (len >= ELF_PRARGSZ)
1309 len = ELF_PRARGSZ-1;
1310 if (copy_from_user(&psinfo->pr_psargs,
1311 (const char __user *)mm->arg_start, len))
1312 return -EFAULT;
1313 for(i = 0; i < len; i++)
1314 if (psinfo->pr_psargs[i] == 0)
1315 psinfo->pr_psargs[i] = ' ';
1316 psinfo->pr_psargs[len] = 0;
1317
Oleg Nesterov3b34fc52009-06-17 16:27:38 -07001318 rcu_read_lock();
1319 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1320 rcu_read_unlock();
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001321 psinfo->pr_pid = task_pid_vnr(p);
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001322 psinfo->pr_pgrp = task_pgrp_vnr(p);
1323 psinfo->pr_sid = task_session_vnr(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
1325 i = p->state ? ffz(~p->state) + 1 : 0;
1326 psinfo->pr_state = i;
Carsten Otte55148542006-03-25 03:08:22 -08001327 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1329 psinfo->pr_nice = task_nice(p);
1330 psinfo->pr_flag = p->flags;
David Howellsc69e8d92008-11-14 10:39:19 +11001331 rcu_read_lock();
1332 cred = __task_cred(p);
1333 SET_UID(psinfo->pr_uid, cred->uid);
1334 SET_GID(psinfo->pr_gid, cred->gid);
1335 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1337
1338 return 0;
1339}
1340
Roland McGrath3aba4812008-01-30 13:31:44 +01001341static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1342{
1343 elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1344 int i = 0;
1345 do
1346 i += 2;
1347 while (auxv[i - 2] != AT_NULL);
1348 fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1349}
1350
Roland McGrath4206d3a2008-01-30 13:31:45 +01001351#ifdef CORE_DUMP_USE_REGSET
1352#include <linux/regset.h>
1353
1354struct elf_thread_core_info {
1355 struct elf_thread_core_info *next;
1356 struct task_struct *task;
1357 struct elf_prstatus prstatus;
1358 struct memelfnote notes[0];
1359};
1360
1361struct elf_note_info {
1362 struct elf_thread_core_info *thread;
1363 struct memelfnote psinfo;
1364 struct memelfnote auxv;
1365 size_t size;
1366 int thread_notes;
1367};
1368
Roland McGrathd31472b2008-03-04 14:28:30 -08001369/*
1370 * When a regset has a writeback hook, we call it on each thread before
1371 * dumping user memory. On register window machines, this makes sure the
1372 * user memory backing the register data is up to date before we read it.
1373 */
1374static void do_thread_regset_writeback(struct task_struct *task,
1375 const struct user_regset *regset)
1376{
1377 if (regset->writeback)
1378 regset->writeback(task, regset, 1);
1379}
1380
Roland McGrath4206d3a2008-01-30 13:31:45 +01001381static int fill_thread_core_info(struct elf_thread_core_info *t,
1382 const struct user_regset_view *view,
1383 long signr, size_t *total)
1384{
1385 unsigned int i;
1386
1387 /*
1388 * NT_PRSTATUS is the one special case, because the regset data
1389 * goes into the pr_reg field inside the note contents, rather
1390 * than being the whole note contents. We fill the reset in here.
1391 * We assume that regset 0 is NT_PRSTATUS.
1392 */
1393 fill_prstatus(&t->prstatus, t->task, signr);
1394 (void) view->regsets[0].get(t->task, &view->regsets[0],
1395 0, sizeof(t->prstatus.pr_reg),
1396 &t->prstatus.pr_reg, NULL);
1397
1398 fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1399 sizeof(t->prstatus), &t->prstatus);
1400 *total += notesize(&t->notes[0]);
1401
Roland McGrathd31472b2008-03-04 14:28:30 -08001402 do_thread_regset_writeback(t->task, &view->regsets[0]);
1403
Roland McGrath4206d3a2008-01-30 13:31:45 +01001404 /*
1405 * Each other regset might generate a note too. For each regset
1406 * that has no core_note_type or is inactive, we leave t->notes[i]
1407 * all zero and we'll know to skip writing it later.
1408 */
1409 for (i = 1; i < view->n; ++i) {
1410 const struct user_regset *regset = &view->regsets[i];
Roland McGrathd31472b2008-03-04 14:28:30 -08001411 do_thread_regset_writeback(t->task, regset);
Roland McGrath4206d3a2008-01-30 13:31:45 +01001412 if (regset->core_note_type &&
1413 (!regset->active || regset->active(t->task, regset))) {
1414 int ret;
1415 size_t size = regset->n * regset->size;
1416 void *data = kmalloc(size, GFP_KERNEL);
1417 if (unlikely(!data))
1418 return 0;
1419 ret = regset->get(t->task, regset,
1420 0, size, data, NULL);
1421 if (unlikely(ret))
1422 kfree(data);
1423 else {
1424 if (regset->core_note_type != NT_PRFPREG)
1425 fill_note(&t->notes[i], "LINUX",
1426 regset->core_note_type,
1427 size, data);
1428 else {
1429 t->prstatus.pr_fpvalid = 1;
1430 fill_note(&t->notes[i], "CORE",
1431 NT_PRFPREG, size, data);
1432 }
1433 *total += notesize(&t->notes[i]);
1434 }
1435 }
1436 }
1437
1438 return 1;
1439}
1440
1441static int fill_note_info(struct elfhdr *elf, int phdrs,
1442 struct elf_note_info *info,
1443 long signr, struct pt_regs *regs)
1444{
1445 struct task_struct *dump_task = current;
1446 const struct user_regset_view *view = task_user_regset_view(dump_task);
1447 struct elf_thread_core_info *t;
1448 struct elf_prpsinfo *psinfo;
Oleg Nesterov83914442008-07-25 01:47:45 -07001449 struct core_thread *ct;
Roland McGrath4206d3a2008-01-30 13:31:45 +01001450 unsigned int i;
1451
1452 info->size = 0;
1453 info->thread = NULL;
1454
1455 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
Roland McGrath4206d3a2008-01-30 13:31:45 +01001456 if (psinfo == NULL)
1457 return 0;
1458
Amerigo Wange2dbe122009-07-01 01:06:26 -04001459 fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1460
Roland McGrath4206d3a2008-01-30 13:31:45 +01001461 /*
1462 * Figure out how many notes we're going to need for each thread.
1463 */
1464 info->thread_notes = 0;
1465 for (i = 0; i < view->n; ++i)
1466 if (view->regsets[i].core_note_type != 0)
1467 ++info->thread_notes;
1468
1469 /*
1470 * Sanity check. We rely on regset 0 being in NT_PRSTATUS,
1471 * since it is our one special case.
1472 */
1473 if (unlikely(info->thread_notes == 0) ||
1474 unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1475 WARN_ON(1);
1476 return 0;
1477 }
1478
1479 /*
1480 * Initialize the ELF file header.
1481 */
1482 fill_elf_header(elf, phdrs,
1483 view->e_machine, view->e_flags, view->ei_osabi);
1484
1485 /*
1486 * Allocate a structure for each thread.
1487 */
Oleg Nesterov83914442008-07-25 01:47:45 -07001488 for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1489 t = kzalloc(offsetof(struct elf_thread_core_info,
1490 notes[info->thread_notes]),
1491 GFP_KERNEL);
1492 if (unlikely(!t))
1493 return 0;
Oleg Nesterov24d52882008-07-25 01:47:40 -07001494
Oleg Nesterov83914442008-07-25 01:47:45 -07001495 t->task = ct->task;
1496 if (ct->task == dump_task || !info->thread) {
1497 t->next = info->thread;
1498 info->thread = t;
1499 } else {
1500 /*
1501 * Make sure to keep the original task at
1502 * the head of the list.
1503 */
1504 t->next = info->thread->next;
1505 info->thread->next = t;
Roland McGrath4206d3a2008-01-30 13:31:45 +01001506 }
Oleg Nesterov83914442008-07-25 01:47:45 -07001507 }
Roland McGrath4206d3a2008-01-30 13:31:45 +01001508
1509 /*
1510 * Now fill in each thread's information.
1511 */
1512 for (t = info->thread; t != NULL; t = t->next)
1513 if (!fill_thread_core_info(t, view, signr, &info->size))
1514 return 0;
1515
1516 /*
1517 * Fill in the two process-wide notes.
1518 */
1519 fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1520 info->size += notesize(&info->psinfo);
1521
1522 fill_auxv_note(&info->auxv, current->mm);
1523 info->size += notesize(&info->auxv);
1524
1525 return 1;
1526}
1527
1528static size_t get_note_info_size(struct elf_note_info *info)
1529{
1530 return info->size;
1531}
1532
1533/*
1534 * Write all the notes for each thread. When writing the first thread, the
1535 * process-wide notes are interleaved after the first thread-specific note.
1536 */
1537static int write_note_info(struct elf_note_info *info,
1538 struct file *file, loff_t *foffset)
1539{
1540 bool first = 1;
1541 struct elf_thread_core_info *t = info->thread;
1542
1543 do {
1544 int i;
1545
1546 if (!writenote(&t->notes[0], file, foffset))
1547 return 0;
1548
1549 if (first && !writenote(&info->psinfo, file, foffset))
1550 return 0;
1551 if (first && !writenote(&info->auxv, file, foffset))
1552 return 0;
1553
1554 for (i = 1; i < info->thread_notes; ++i)
1555 if (t->notes[i].data &&
1556 !writenote(&t->notes[i], file, foffset))
1557 return 0;
1558
1559 first = 0;
1560 t = t->next;
1561 } while (t);
1562
1563 return 1;
1564}
1565
1566static void free_note_info(struct elf_note_info *info)
1567{
1568 struct elf_thread_core_info *threads = info->thread;
1569 while (threads) {
1570 unsigned int i;
1571 struct elf_thread_core_info *t = threads;
1572 threads = t->next;
1573 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1574 for (i = 1; i < info->thread_notes; ++i)
1575 kfree(t->notes[i].data);
1576 kfree(t);
1577 }
1578 kfree(info->psinfo.data);
1579}
1580
1581#else
1582
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583/* Here is the structure in which status of each thread is captured. */
1584struct elf_thread_status
1585{
1586 struct list_head list;
1587 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1588 elf_fpregset_t fpu; /* NT_PRFPREG */
1589 struct task_struct *thread;
1590#ifdef ELF_CORE_COPY_XFPREGS
Mark Nelson5b20cd82007-10-16 23:25:39 -07001591 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592#endif
1593 struct memelfnote notes[3];
1594 int num_notes;
1595};
1596
1597/*
1598 * In order to add the specific thread information for the elf file format,
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001599 * we need to keep a linked list of every threads pr_status and then create
1600 * a single section for them in the final core file.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 */
1602static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1603{
1604 int sz = 0;
1605 struct task_struct *p = t->thread;
1606 t->num_notes = 0;
1607
1608 fill_prstatus(&t->prstatus, p, signr);
1609 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1610
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001611 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1612 &(t->prstatus));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 t->num_notes++;
1614 sz += notesize(&t->notes[0]);
1615
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001616 if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1617 &t->fpu))) {
1618 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1619 &(t->fpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 t->num_notes++;
1621 sz += notesize(&t->notes[1]);
1622 }
1623
1624#ifdef ELF_CORE_COPY_XFPREGS
1625 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
Mark Nelson5b20cd82007-10-16 23:25:39 -07001626 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1627 sizeof(t->xfpu), &t->xfpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 t->num_notes++;
1629 sz += notesize(&t->notes[2]);
1630 }
1631#endif
1632 return sz;
1633}
1634
Roland McGrath3aba4812008-01-30 13:31:44 +01001635struct elf_note_info {
1636 struct memelfnote *notes;
1637 struct elf_prstatus *prstatus; /* NT_PRSTATUS */
1638 struct elf_prpsinfo *psinfo; /* NT_PRPSINFO */
1639 struct list_head thread_list;
1640 elf_fpregset_t *fpu;
1641#ifdef ELF_CORE_COPY_XFPREGS
1642 elf_fpxregset_t *xfpu;
1643#endif
1644 int thread_status_size;
1645 int numnote;
1646};
1647
Amerigo Wang0cf062d2009-09-23 15:57:05 -07001648static int elf_note_info_init(struct elf_note_info *info)
Roland McGrath3aba4812008-01-30 13:31:44 +01001649{
Amerigo Wang0cf062d2009-09-23 15:57:05 -07001650 memset(info, 0, sizeof(*info));
Roland McGrath3aba4812008-01-30 13:31:44 +01001651 INIT_LIST_HEAD(&info->thread_list);
1652
Amerigo Wang0cf062d2009-09-23 15:57:05 -07001653 /* Allocate space for six ELF notes */
1654 info->notes = kmalloc(6 * sizeof(struct memelfnote), GFP_KERNEL);
Roland McGrath3aba4812008-01-30 13:31:44 +01001655 if (!info->notes)
1656 return 0;
1657 info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1658 if (!info->psinfo)
Amerigo Wang0cf062d2009-09-23 15:57:05 -07001659 goto notes_free;
Roland McGrath3aba4812008-01-30 13:31:44 +01001660 info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1661 if (!info->prstatus)
Amerigo Wang0cf062d2009-09-23 15:57:05 -07001662 goto psinfo_free;
Roland McGrath3aba4812008-01-30 13:31:44 +01001663 info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1664 if (!info->fpu)
Amerigo Wang0cf062d2009-09-23 15:57:05 -07001665 goto prstatus_free;
Roland McGrath3aba4812008-01-30 13:31:44 +01001666#ifdef ELF_CORE_COPY_XFPREGS
1667 info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1668 if (!info->xfpu)
Amerigo Wang0cf062d2009-09-23 15:57:05 -07001669 goto fpu_free;
Roland McGrath3aba4812008-01-30 13:31:44 +01001670#endif
Amerigo Wang0cf062d2009-09-23 15:57:05 -07001671 return 1;
1672#ifdef ELF_CORE_COPY_XFPREGS
1673 fpu_free:
1674 kfree(info->fpu);
1675#endif
1676 prstatus_free:
1677 kfree(info->prstatus);
1678 psinfo_free:
1679 kfree(info->psinfo);
1680 notes_free:
1681 kfree(info->notes);
1682 return 0;
1683}
Roland McGrath3aba4812008-01-30 13:31:44 +01001684
Amerigo Wang0cf062d2009-09-23 15:57:05 -07001685static int fill_note_info(struct elfhdr *elf, int phdrs,
1686 struct elf_note_info *info,
1687 long signr, struct pt_regs *regs)
1688{
1689 struct list_head *t;
1690
1691 if (!elf_note_info_init(info))
1692 return 0;
1693
Roland McGrath3aba4812008-01-30 13:31:44 +01001694 if (signr) {
Oleg Nesterov83914442008-07-25 01:47:45 -07001695 struct core_thread *ct;
WANG Cong4220b7f2008-04-29 01:01:18 -07001696 struct elf_thread_status *ets;
Oleg Nesterov24d52882008-07-25 01:47:40 -07001697
Oleg Nesterov83914442008-07-25 01:47:45 -07001698 for (ct = current->mm->core_state->dumper.next;
1699 ct; ct = ct->next) {
1700 ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1701 if (!ets)
1702 return 0;
1703
1704 ets->thread = ct->task;
1705 list_add(&ets->list, &info->thread_list);
1706 }
1707
Roland McGrath3aba4812008-01-30 13:31:44 +01001708 list_for_each(t, &info->thread_list) {
Roland McGrath3aba4812008-01-30 13:31:44 +01001709 int sz;
1710
WANG Cong4220b7f2008-04-29 01:01:18 -07001711 ets = list_entry(t, struct elf_thread_status, list);
1712 sz = elf_dump_thread_status(signr, ets);
Roland McGrath3aba4812008-01-30 13:31:44 +01001713 info->thread_status_size += sz;
1714 }
1715 }
1716 /* now collect the dump for the current */
1717 memset(info->prstatus, 0, sizeof(*info->prstatus));
1718 fill_prstatus(info->prstatus, current, signr);
1719 elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1720
1721 /* Set up header */
1722 fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS, ELF_OSABI);
1723
1724 /*
1725 * Set up the notes in similar form to SVR4 core dumps made
1726 * with info from their /proc.
1727 */
1728
1729 fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1730 sizeof(*info->prstatus), info->prstatus);
1731 fill_psinfo(info->psinfo, current->group_leader, current->mm);
1732 fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1733 sizeof(*info->psinfo), info->psinfo);
1734
1735 info->numnote = 2;
1736
1737 fill_auxv_note(&info->notes[info->numnote++], current->mm);
1738
1739 /* Try to dump the FPU. */
1740 info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1741 info->fpu);
1742 if (info->prstatus->pr_fpvalid)
1743 fill_note(info->notes + info->numnote++,
1744 "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1745#ifdef ELF_CORE_COPY_XFPREGS
1746 if (elf_core_copy_task_xfpregs(current, info->xfpu))
1747 fill_note(info->notes + info->numnote++,
1748 "LINUX", ELF_CORE_XFPREG_TYPE,
1749 sizeof(*info->xfpu), info->xfpu);
1750#endif
1751
1752 return 1;
Roland McGrath3aba4812008-01-30 13:31:44 +01001753}
1754
1755static size_t get_note_info_size(struct elf_note_info *info)
1756{
1757 int sz = 0;
1758 int i;
1759
1760 for (i = 0; i < info->numnote; i++)
1761 sz += notesize(info->notes + i);
1762
1763 sz += info->thread_status_size;
1764
1765 return sz;
1766}
1767
1768static int write_note_info(struct elf_note_info *info,
1769 struct file *file, loff_t *foffset)
1770{
1771 int i;
1772 struct list_head *t;
1773
1774 for (i = 0; i < info->numnote; i++)
1775 if (!writenote(info->notes + i, file, foffset))
1776 return 0;
1777
1778 /* write out the thread status notes section */
1779 list_for_each(t, &info->thread_list) {
1780 struct elf_thread_status *tmp =
1781 list_entry(t, struct elf_thread_status, list);
1782
1783 for (i = 0; i < tmp->num_notes; i++)
1784 if (!writenote(&tmp->notes[i], file, foffset))
1785 return 0;
1786 }
1787
1788 return 1;
1789}
1790
1791static void free_note_info(struct elf_note_info *info)
1792{
1793 while (!list_empty(&info->thread_list)) {
1794 struct list_head *tmp = info->thread_list.next;
1795 list_del(tmp);
1796 kfree(list_entry(tmp, struct elf_thread_status, list));
1797 }
1798
1799 kfree(info->prstatus);
1800 kfree(info->psinfo);
1801 kfree(info->notes);
1802 kfree(info->fpu);
1803#ifdef ELF_CORE_COPY_XFPREGS
1804 kfree(info->xfpu);
1805#endif
1806}
1807
Roland McGrath4206d3a2008-01-30 13:31:45 +01001808#endif
1809
Roland McGrathf47aef52007-01-26 00:56:49 -08001810static struct vm_area_struct *first_vma(struct task_struct *tsk,
1811 struct vm_area_struct *gate_vma)
1812{
1813 struct vm_area_struct *ret = tsk->mm->mmap;
1814
1815 if (ret)
1816 return ret;
1817 return gate_vma;
1818}
1819/*
1820 * Helper function for iterating across a vma list. It ensures that the caller
1821 * will visit `gate_vma' prior to terminating the search.
1822 */
1823static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1824 struct vm_area_struct *gate_vma)
1825{
1826 struct vm_area_struct *ret;
1827
1828 ret = this_vma->vm_next;
1829 if (ret)
1830 return ret;
1831 if (this_vma == gate_vma)
1832 return NULL;
1833 return gate_vma;
1834}
1835
Daisuke HATAYAMA8d9032b2010-03-05 13:44:10 -08001836static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1837 elf_addr_t e_shoff, int segs)
1838{
1839 elf->e_shoff = e_shoff;
1840 elf->e_shentsize = sizeof(*shdr4extnum);
1841 elf->e_shnum = 1;
1842 elf->e_shstrndx = SHN_UNDEF;
1843
1844 memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1845
1846 shdr4extnum->sh_type = SHT_NULL;
1847 shdr4extnum->sh_size = elf->e_shnum;
1848 shdr4extnum->sh_link = elf->e_shstrndx;
1849 shdr4extnum->sh_info = segs;
1850}
1851
1852static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
1853 unsigned long mm_flags)
1854{
1855 struct vm_area_struct *vma;
1856 size_t size = 0;
1857
1858 for (vma = first_vma(current, gate_vma); vma != NULL;
1859 vma = next_vma(vma, gate_vma))
1860 size += vma_dump_size(vma, mm_flags);
1861 return size;
1862}
1863
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864/*
1865 * Actual dumper
1866 *
1867 * This is a two-pass process; first we find the offsets of the bits,
1868 * and then they are actually written out. If we run out of core limit
1869 * we just truncate.
1870 */
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001871static int elf_core_dump(struct coredump_params *cprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 int has_dumped = 0;
1874 mm_segment_t fs;
1875 int segs;
1876 size_t size = 0;
Roland McGrathf47aef52007-01-26 00:56:49 -08001877 struct vm_area_struct *vma, *gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 struct elfhdr *elf = NULL;
Andi Kleend025c9d2006-09-30 23:29:28 -07001879 loff_t offset = 0, dataoff, foffset;
Roland McGrath3aba4812008-01-30 13:31:44 +01001880 struct elf_note_info info;
Daisuke HATAYAMA93eb2112010-03-05 13:44:09 -08001881 struct elf_phdr *phdr4note = NULL;
Daisuke HATAYAMA8d9032b2010-03-05 13:44:10 -08001882 struct elf_shdr *shdr4extnum = NULL;
1883 Elf_Half e_phnum;
1884 elf_addr_t e_shoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885
1886 /*
1887 * We no longer stop all VM operations.
1888 *
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001889 * This is because those proceses that could possibly change map_count
1890 * or the mmap / vma pages are now blocked in do_exit on current
1891 * finishing this core dump.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 *
1893 * Only ptrace can touch these memory addresses, but it doesn't change
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001894 * the map_count or the pages allocated. So no possibility of crashing
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895 * exists while dumping the mm->vm_next areas to the core file.
1896 */
1897
1898 /* alloc memory for large data structures: too large to be on stack */
1899 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1900 if (!elf)
WANG Cong5f719552008-05-06 12:45:35 +08001901 goto out;
KAMEZAWA Hiroyuki341c87b2009-06-30 11:41:23 -07001902 /*
1903 * The number of segs are recored into ELF header as 16bit value.
1904 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
1905 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 segs = current->mm->map_count;
Daisuke HATAYAMA1fcccba2010-03-05 13:44:07 -08001907 segs += elf_core_extra_phdrs();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908
Stephen Wilson31db58b2011-03-13 15:49:15 -04001909 gate_vma = get_gate_vma(current->mm);
Roland McGrathf47aef52007-01-26 00:56:49 -08001910 if (gate_vma != NULL)
1911 segs++;
1912
Daisuke HATAYAMA8d9032b2010-03-05 13:44:10 -08001913 /* for notes section */
1914 segs++;
1915
1916 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1917 * this, kernel supports extended numbering. Have a look at
1918 * include/linux/elf.h for further information. */
1919 e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1920
Roland McGrath3aba4812008-01-30 13:31:44 +01001921 /*
1922 * Collect all the non-memory information about the process for the
1923 * notes. This also sets up the file header.
1924 */
Daisuke HATAYAMA8d9032b2010-03-05 13:44:10 -08001925 if (!fill_note_info(elf, e_phnum, &info, cprm->signr, cprm->regs))
Roland McGrath3aba4812008-01-30 13:31:44 +01001926 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927
1928 has_dumped = 1;
1929 current->flags |= PF_DUMPCORE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001930
1931 fs = get_fs();
1932 set_fs(KERNEL_DS);
1933
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 offset += sizeof(*elf); /* Elf header */
Daisuke HATAYAMA8d9032b2010-03-05 13:44:10 -08001935 offset += segs * sizeof(struct elf_phdr); /* Program headers */
Petr Vandroveca7a0d862006-10-13 18:42:07 +02001936 foffset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937
1938 /* Write notes phdr entry */
1939 {
Roland McGrath3aba4812008-01-30 13:31:44 +01001940 size_t sz = get_note_info_size(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941
Michael Ellermane5501492007-09-19 14:38:12 +10001942 sz += elf_coredump_extra_notes_size();
Dwayne Grant McConnellbf1ab972006-11-23 00:46:37 +01001943
Daisuke HATAYAMA93eb2112010-03-05 13:44:09 -08001944 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1945 if (!phdr4note)
Daisuke HATAYAMA088e7af2010-03-05 13:44:06 -08001946 goto end_coredump;
Daisuke HATAYAMA93eb2112010-03-05 13:44:09 -08001947
1948 fill_elf_note_phdr(phdr4note, sz, offset);
1949 offset += sz;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 }
1951
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1953
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001954 offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
Daisuke HATAYAMA8d9032b2010-03-05 13:44:10 -08001955 offset += elf_core_extra_data_size();
1956 e_shoff = offset;
1957
1958 if (e_phnum == PN_XNUM) {
1959 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1960 if (!shdr4extnum)
1961 goto end_coredump;
1962 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1963 }
1964
1965 offset = dataoff;
1966
Daisuke HATAYAMA93eb2112010-03-05 13:44:09 -08001967 size += sizeof(*elf);
1968 if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
1969 goto end_coredump;
1970
1971 size += sizeof(*phdr4note);
1972 if (size > cprm->limit
1973 || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
1974 goto end_coredump;
1975
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976 /* Write program headers for segments dump */
Roland McGrathf47aef52007-01-26 00:56:49 -08001977 for (vma = first_vma(current, gate_vma); vma != NULL;
1978 vma = next_vma(vma, gate_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 struct elf_phdr phdr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980
1981 phdr.p_type = PT_LOAD;
1982 phdr.p_offset = offset;
1983 phdr.p_vaddr = vma->vm_start;
1984 phdr.p_paddr = 0;
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001985 phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
Roland McGrath82df3972007-10-16 23:27:02 -07001986 phdr.p_memsz = vma->vm_end - vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 offset += phdr.p_filesz;
1988 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001989 if (vma->vm_flags & VM_WRITE)
1990 phdr.p_flags |= PF_W;
1991 if (vma->vm_flags & VM_EXEC)
1992 phdr.p_flags |= PF_X;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993 phdr.p_align = ELF_EXEC_PAGESIZE;
1994
Daisuke HATAYAMA088e7af2010-03-05 13:44:06 -08001995 size += sizeof(phdr);
1996 if (size > cprm->limit
1997 || !dump_write(cprm->file, &phdr, sizeof(phdr)))
1998 goto end_coredump;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999 }
2000
Daisuke HATAYAMA1fcccba2010-03-05 13:44:07 -08002001 if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
2002 goto end_coredump;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003
2004 /* write out the notes section */
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002005 if (!write_note_info(&info, cprm->file, &foffset))
Roland McGrath3aba4812008-01-30 13:31:44 +01002006 goto end_coredump;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002008 if (elf_coredump_extra_notes_write(cprm->file, &foffset))
Michael Ellermane5501492007-09-19 14:38:12 +10002009 goto end_coredump;
Dwayne Grant McConnellbf1ab972006-11-23 00:46:37 +01002010
Andi Kleend025c9d2006-09-30 23:29:28 -07002011 /* Align to page */
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002012 if (!dump_seek(cprm->file, dataoff - foffset))
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07002013 goto end_coredump;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014
Roland McGrathf47aef52007-01-26 00:56:49 -08002015 for (vma = first_vma(current, gate_vma); vma != NULL;
2016 vma = next_vma(vma, gate_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 unsigned long addr;
Roland McGrath82df3972007-10-16 23:27:02 -07002018 unsigned long end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019
Masami Hiramatsu30736a42010-03-05 13:44:12 -08002020 end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021
Roland McGrath82df3972007-10-16 23:27:02 -07002022 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07002023 struct page *page;
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07002024 int stop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07002026 page = get_dump_page(addr);
2027 if (page) {
2028 void *kaddr = kmap(page);
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002029 stop = ((size += PAGE_SIZE) > cprm->limit) ||
2030 !dump_write(cprm->file, kaddr,
2031 PAGE_SIZE);
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07002032 kunmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 page_cache_release(page);
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07002034 } else
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002035 stop = !dump_seek(cprm->file, PAGE_SIZE);
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07002036 if (stop)
2037 goto end_coredump;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 }
2039 }
2040
Daisuke HATAYAMA1fcccba2010-03-05 13:44:07 -08002041 if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
2042 goto end_coredump;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043
Daisuke HATAYAMA8d9032b2010-03-05 13:44:10 -08002044 if (e_phnum == PN_XNUM) {
2045 size += sizeof(*shdr4extnum);
2046 if (size > cprm->limit
2047 || !dump_write(cprm->file, shdr4extnum,
2048 sizeof(*shdr4extnum)))
2049 goto end_coredump;
2050 }
2051
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052end_coredump:
2053 set_fs(fs);
2054
2055cleanup:
Roland McGrath3aba4812008-01-30 13:31:44 +01002056 free_note_info(&info);
Daisuke HATAYAMA8d9032b2010-03-05 13:44:10 -08002057 kfree(shdr4extnum);
Daisuke HATAYAMA93eb2112010-03-05 13:44:09 -08002058 kfree(phdr4note);
WANG Cong5f719552008-05-06 12:45:35 +08002059 kfree(elf);
2060out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 return has_dumped;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062}
2063
Christoph Hellwig698ba7b2009-12-15 16:47:37 -08002064#endif /* CONFIG_ELF_CORE */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065
2066static int __init init_elf_binfmt(void)
2067{
2068 return register_binfmt(&elf_format);
2069}
2070
2071static void __exit exit_elf_binfmt(void)
2072{
2073 /* Remove the COFF and ELF loaders. */
2074 unregister_binfmt(&elf_format);
2075}
2076
2077core_initcall(init_elf_binfmt);
2078module_exit(exit_elf_binfmt);
2079MODULE_LICENSE("GPL");