blob: 303983fabfd63391f3961a58be352303b7af0064 [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
Jiri Kosina4471a672011-04-14 15:22:09 -0700944 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);
Jiri Kosina4471a672011-04-14 15:22:09 -0700947#ifdef CONFIG_COMPAT_BRK
948 current->brk_randomized = 1;
949#endif
950 }
Jiri Kosinac1d171a2008-01-30 13:30:40 +0100951#endif
952
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 if (current->personality & MMAP_PAGE_ZERO) {
954 /* Why this, you ask??? Well SVr4 maps page 0 as read-only,
955 and some applications "depend" upon this behavior.
956 Since we do not have the power to recompile these, we
Jesper Juhlf4e5cc22006-06-23 02:05:35 -0700957 emulate the SVr4 behavior. Sigh. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 down_write(&current->mm->mmap_sem);
959 error = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
960 MAP_FIXED | MAP_PRIVATE, 0);
961 up_write(&current->mm->mmap_sem);
962 }
963
964#ifdef ELF_PLAT_INIT
965 /*
966 * The ABI may specify that certain registers be set up in special
967 * ways (on i386 %edx is the address of a DT_FINI function, for
968 * example. In addition, it may also specify (eg, PowerPC64 ELF)
969 * that the e_entry field is the address of the function descriptor
970 * for the startup routine, rather than the address of the startup
971 * routine itself. This macro performs whatever initialization to
972 * the regs structure is required as well as any relocations to the
973 * function descriptor entries when executing dynamically links apps.
974 */
975 ELF_PLAT_INIT(regs, reloc_func_desc);
976#endif
977
978 start_thread(regs, elf_entry, bprm->p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 retval = 0;
980out:
981 kfree(loc);
982out_ret:
983 return retval;
984
985 /* error cleanup */
986out_free_dentry:
987 allow_write_access(interpreter);
988 if (interpreter)
989 fput(interpreter);
990out_free_interp:
Jesper Juhlf99d49a2005-11-07 01:01:34 -0800991 kfree(elf_interpreter);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992out_free_ph:
993 kfree(elf_phdata);
994 goto out;
995}
996
997/* This is really simpleminded and specialized - we are loading an
998 a.out library that is given an ELF header. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999static int load_elf_library(struct file *file)
1000{
1001 struct elf_phdr *elf_phdata;
1002 struct elf_phdr *eppnt;
1003 unsigned long elf_bss, bss, len;
1004 int retval, error, i, j;
1005 struct elfhdr elf_ex;
1006
1007 error = -ENOEXEC;
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001008 retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 if (retval != sizeof(elf_ex))
1010 goto out;
1011
1012 if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
1013 goto out;
1014
1015 /* First of all, some simple consistency checks */
1016 if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001017 !elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 goto out;
1019
1020 /* Now read in all of the header information */
1021
1022 j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
1023 /* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
1024
1025 error = -ENOMEM;
1026 elf_phdata = kmalloc(j, GFP_KERNEL);
1027 if (!elf_phdata)
1028 goto out;
1029
1030 eppnt = elf_phdata;
1031 error = -ENOEXEC;
1032 retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
1033 if (retval != j)
1034 goto out_free_ph;
1035
1036 for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
1037 if ((eppnt + i)->p_type == PT_LOAD)
1038 j++;
1039 if (j != 1)
1040 goto out_free_ph;
1041
1042 while (eppnt->p_type != PT_LOAD)
1043 eppnt++;
1044
1045 /* Now use mmap to map the library into memory. */
1046 down_write(&current->mm->mmap_sem);
1047 error = do_mmap(file,
1048 ELF_PAGESTART(eppnt->p_vaddr),
1049 (eppnt->p_filesz +
1050 ELF_PAGEOFFSET(eppnt->p_vaddr)),
1051 PROT_READ | PROT_WRITE | PROT_EXEC,
1052 MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
1053 (eppnt->p_offset -
1054 ELF_PAGEOFFSET(eppnt->p_vaddr)));
1055 up_write(&current->mm->mmap_sem);
1056 if (error != ELF_PAGESTART(eppnt->p_vaddr))
1057 goto out_free_ph;
1058
1059 elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
1060 if (padzero(elf_bss)) {
1061 error = -EFAULT;
1062 goto out_free_ph;
1063 }
1064
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001065 len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
1066 ELF_MIN_ALIGN - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 bss = eppnt->p_memsz + eppnt->p_vaddr;
1068 if (bss > len) {
1069 down_write(&current->mm->mmap_sem);
1070 do_brk(len, bss - len);
1071 up_write(&current->mm->mmap_sem);
1072 }
1073 error = 0;
1074
1075out_free_ph:
1076 kfree(elf_phdata);
1077out:
1078 return error;
1079}
1080
Christoph Hellwig698ba7b2009-12-15 16:47:37 -08001081#ifdef CONFIG_ELF_CORE
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082/*
1083 * ELF core dumper
1084 *
1085 * Modelled on fs/exec.c:aout_core_dump()
1086 * Jeremy Fitzhardinge <jeremy@sw.oz.au>
1087 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088
1089/*
Roland McGrath82df3972007-10-16 23:27:02 -07001090 * Decide what to dump of a segment, part, all or none.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 */
Roland McGrath82df3972007-10-16 23:27:02 -07001092static unsigned long vma_dump_size(struct vm_area_struct *vma,
1093 unsigned long mm_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094{
KOSAKI Motohiroe575f112008-10-18 20:27:08 -07001095#define FILTER(type) (mm_flags & (1UL << MMF_DUMP_##type))
1096
Roland McGrathe5b97dd2007-01-26 00:56:48 -08001097 /* The vma can be set up to tell us the answer directly. */
1098 if (vma->vm_flags & VM_ALWAYSDUMP)
Roland McGrath82df3972007-10-16 23:27:02 -07001099 goto whole;
Roland McGrathe5b97dd2007-01-26 00:56:48 -08001100
KOSAKI Motohiroe575f112008-10-18 20:27:08 -07001101 /* Hugetlb memory check */
1102 if (vma->vm_flags & VM_HUGETLB) {
1103 if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
1104 goto whole;
1105 if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
1106 goto whole;
1107 }
1108
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 /* Do not dump I/O mapped devices or special mappings */
1110 if (vma->vm_flags & (VM_IO | VM_RESERVED))
1111 return 0;
1112
Kawai, Hidehiroa1b59e82007-07-19 01:48:29 -07001113 /* By default, dump shared memory if mapped from an anonymous file. */
1114 if (vma->vm_flags & VM_SHARED) {
Roland McGrath82df3972007-10-16 23:27:02 -07001115 if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0 ?
1116 FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
1117 goto whole;
1118 return 0;
Kawai, Hidehiroa1b59e82007-07-19 01:48:29 -07001119 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120
Roland McGrath82df3972007-10-16 23:27:02 -07001121 /* Dump segments that have been written to. */
1122 if (vma->anon_vma && FILTER(ANON_PRIVATE))
1123 goto whole;
1124 if (vma->vm_file == NULL)
1125 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
Roland McGrath82df3972007-10-16 23:27:02 -07001127 if (FILTER(MAPPED_PRIVATE))
1128 goto whole;
1129
1130 /*
1131 * If this looks like the beginning of a DSO or executable mapping,
1132 * check for an ELF header. If we find one, dump the first page to
1133 * aid in determining what was mapped here.
1134 */
Roland McGrath92dc07b2009-02-06 17:34:07 -08001135 if (FILTER(ELF_HEADERS) &&
1136 vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
Roland McGrath82df3972007-10-16 23:27:02 -07001137 u32 __user *header = (u32 __user *) vma->vm_start;
1138 u32 word;
Roland McGrath92dc07b2009-02-06 17:34:07 -08001139 mm_segment_t fs = get_fs();
Roland McGrath82df3972007-10-16 23:27:02 -07001140 /*
1141 * Doing it this way gets the constant folded by GCC.
1142 */
1143 union {
1144 u32 cmp;
1145 char elfmag[SELFMAG];
1146 } magic;
1147 BUILD_BUG_ON(SELFMAG != sizeof word);
1148 magic.elfmag[EI_MAG0] = ELFMAG0;
1149 magic.elfmag[EI_MAG1] = ELFMAG1;
1150 magic.elfmag[EI_MAG2] = ELFMAG2;
1151 magic.elfmag[EI_MAG3] = ELFMAG3;
Roland McGrath92dc07b2009-02-06 17:34:07 -08001152 /*
1153 * Switch to the user "segment" for get_user(),
1154 * then put back what elf_core_dump() had in place.
1155 */
1156 set_fs(USER_DS);
1157 if (unlikely(get_user(word, header)))
1158 word = 0;
1159 set_fs(fs);
1160 if (word == magic.cmp)
Roland McGrath82df3972007-10-16 23:27:02 -07001161 return PAGE_SIZE;
1162 }
1163
1164#undef FILTER
1165
1166 return 0;
1167
1168whole:
1169 return vma->vm_end - vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170}
1171
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172/* An ELF note in memory */
1173struct memelfnote
1174{
1175 const char *name;
1176 int type;
1177 unsigned int datasz;
1178 void *data;
1179};
1180
1181static int notesize(struct memelfnote *en)
1182{
1183 int sz;
1184
1185 sz = sizeof(struct elf_note);
1186 sz += roundup(strlen(en->name) + 1, 4);
1187 sz += roundup(en->datasz, 4);
1188
1189 return sz;
1190}
1191
Andi Kleend025c9d2006-09-30 23:29:28 -07001192#define DUMP_WRITE(addr, nr, foffset) \
1193 do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194
Andi Kleend025c9d2006-09-30 23:29:28 -07001195static int alignfile(struct file *file, loff_t *foffset)
1196{
Petr Vandroveca7a0d862006-10-13 18:42:07 +02001197 static const char buf[4] = { 0, };
Andi Kleend025c9d2006-09-30 23:29:28 -07001198 DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
1199 return 1;
1200}
1201
1202static int writenote(struct memelfnote *men, struct file *file,
1203 loff_t *foffset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204{
1205 struct elf_note en;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206 en.n_namesz = strlen(men->name) + 1;
1207 en.n_descsz = men->datasz;
1208 en.n_type = men->type;
1209
Andi Kleend025c9d2006-09-30 23:29:28 -07001210 DUMP_WRITE(&en, sizeof(en), foffset);
1211 DUMP_WRITE(men->name, en.n_namesz, foffset);
1212 if (!alignfile(file, foffset))
1213 return 0;
1214 DUMP_WRITE(men->data, men->datasz, foffset);
1215 if (!alignfile(file, foffset))
1216 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217
1218 return 1;
1219}
1220#undef DUMP_WRITE
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221
Roland McGrath3aba4812008-01-30 13:31:44 +01001222static void fill_elf_header(struct elfhdr *elf, int segs,
1223 u16 machine, u32 flags, u8 osabi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224{
Cyrill Gorcunov6970c8e2008-04-29 01:01:18 -07001225 memset(elf, 0, sizeof(*elf));
1226
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 memcpy(elf->e_ident, ELFMAG, SELFMAG);
1228 elf->e_ident[EI_CLASS] = ELF_CLASS;
1229 elf->e_ident[EI_DATA] = ELF_DATA;
1230 elf->e_ident[EI_VERSION] = EV_CURRENT;
1231 elf->e_ident[EI_OSABI] = ELF_OSABI;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232
1233 elf->e_type = ET_CORE;
Roland McGrath3aba4812008-01-30 13:31:44 +01001234 elf->e_machine = machine;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 elf->e_version = EV_CURRENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 elf->e_phoff = sizeof(struct elfhdr);
Roland McGrath3aba4812008-01-30 13:31:44 +01001237 elf->e_flags = flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 elf->e_ehsize = sizeof(struct elfhdr);
1239 elf->e_phentsize = sizeof(struct elf_phdr);
1240 elf->e_phnum = segs;
Cyrill Gorcunov6970c8e2008-04-29 01:01:18 -07001241
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 return;
1243}
1244
Andrew Morton8d6b5eee2006-09-25 23:32:04 -07001245static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246{
1247 phdr->p_type = PT_NOTE;
1248 phdr->p_offset = offset;
1249 phdr->p_vaddr = 0;
1250 phdr->p_paddr = 0;
1251 phdr->p_filesz = sz;
1252 phdr->p_memsz = 0;
1253 phdr->p_flags = 0;
1254 phdr->p_align = 0;
1255 return;
1256}
1257
1258static void fill_note(struct memelfnote *note, const char *name, int type,
1259 unsigned int sz, void *data)
1260{
1261 note->name = name;
1262 note->type = type;
1263 note->datasz = sz;
1264 note->data = data;
1265 return;
1266}
1267
1268/*
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001269 * fill up all the fields in prstatus from the given task struct, except
1270 * registers which need to be filled up separately.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 */
1272static void fill_prstatus(struct elf_prstatus *prstatus,
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001273 struct task_struct *p, long signr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274{
1275 prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
1276 prstatus->pr_sigpend = p->pending.signal.sig[0];
1277 prstatus->pr_sighold = p->blocked.sig[0];
Oleg Nesterov3b34fc52009-06-17 16:27:38 -07001278 rcu_read_lock();
1279 prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1280 rcu_read_unlock();
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001281 prstatus->pr_pid = task_pid_vnr(p);
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001282 prstatus->pr_pgrp = task_pgrp_vnr(p);
1283 prstatus->pr_sid = task_session_vnr(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 if (thread_group_leader(p)) {
Frank Mayharf06febc2008-09-12 09:54:39 -07001285 struct task_cputime cputime;
1286
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 /*
Frank Mayharf06febc2008-09-12 09:54:39 -07001288 * This is the record for the group leader. It shows the
1289 * group-wide total, not its individual thread total.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 */
Frank Mayharf06febc2008-09-12 09:54:39 -07001291 thread_group_cputime(p, &cputime);
1292 cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
1293 cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 } else {
1295 cputime_to_timeval(p->utime, &prstatus->pr_utime);
1296 cputime_to_timeval(p->stime, &prstatus->pr_stime);
1297 }
1298 cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
1299 cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
1300}
1301
1302static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
1303 struct mm_struct *mm)
1304{
David Howellsc69e8d92008-11-14 10:39:19 +11001305 const struct cred *cred;
Greg Kroah-Hartmana84a5052005-05-11 00:10:44 -07001306 unsigned int i, len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307
1308 /* first copy the parameters from user space */
1309 memset(psinfo, 0, sizeof(struct elf_prpsinfo));
1310
1311 len = mm->arg_end - mm->arg_start;
1312 if (len >= ELF_PRARGSZ)
1313 len = ELF_PRARGSZ-1;
1314 if (copy_from_user(&psinfo->pr_psargs,
1315 (const char __user *)mm->arg_start, len))
1316 return -EFAULT;
1317 for(i = 0; i < len; i++)
1318 if (psinfo->pr_psargs[i] == 0)
1319 psinfo->pr_psargs[i] = ' ';
1320 psinfo->pr_psargs[len] = 0;
1321
Oleg Nesterov3b34fc52009-06-17 16:27:38 -07001322 rcu_read_lock();
1323 psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
1324 rcu_read_unlock();
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001325 psinfo->pr_pid = task_pid_vnr(p);
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001326 psinfo->pr_pgrp = task_pgrp_vnr(p);
1327 psinfo->pr_sid = task_session_vnr(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328
1329 i = p->state ? ffz(~p->state) + 1 : 0;
1330 psinfo->pr_state = i;
Carsten Otte55148542006-03-25 03:08:22 -08001331 psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 psinfo->pr_zomb = psinfo->pr_sname == 'Z';
1333 psinfo->pr_nice = task_nice(p);
1334 psinfo->pr_flag = p->flags;
David Howellsc69e8d92008-11-14 10:39:19 +11001335 rcu_read_lock();
1336 cred = __task_cred(p);
1337 SET_UID(psinfo->pr_uid, cred->uid);
1338 SET_GID(psinfo->pr_gid, cred->gid);
1339 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
1341
1342 return 0;
1343}
1344
Roland McGrath3aba4812008-01-30 13:31:44 +01001345static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
1346{
1347 elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
1348 int i = 0;
1349 do
1350 i += 2;
1351 while (auxv[i - 2] != AT_NULL);
1352 fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
1353}
1354
Roland McGrath4206d3a2008-01-30 13:31:45 +01001355#ifdef CORE_DUMP_USE_REGSET
1356#include <linux/regset.h>
1357
1358struct elf_thread_core_info {
1359 struct elf_thread_core_info *next;
1360 struct task_struct *task;
1361 struct elf_prstatus prstatus;
1362 struct memelfnote notes[0];
1363};
1364
1365struct elf_note_info {
1366 struct elf_thread_core_info *thread;
1367 struct memelfnote psinfo;
1368 struct memelfnote auxv;
1369 size_t size;
1370 int thread_notes;
1371};
1372
Roland McGrathd31472b2008-03-04 14:28:30 -08001373/*
1374 * When a regset has a writeback hook, we call it on each thread before
1375 * dumping user memory. On register window machines, this makes sure the
1376 * user memory backing the register data is up to date before we read it.
1377 */
1378static void do_thread_regset_writeback(struct task_struct *task,
1379 const struct user_regset *regset)
1380{
1381 if (regset->writeback)
1382 regset->writeback(task, regset, 1);
1383}
1384
Roland McGrath4206d3a2008-01-30 13:31:45 +01001385static int fill_thread_core_info(struct elf_thread_core_info *t,
1386 const struct user_regset_view *view,
1387 long signr, size_t *total)
1388{
1389 unsigned int i;
1390
1391 /*
1392 * NT_PRSTATUS is the one special case, because the regset data
1393 * goes into the pr_reg field inside the note contents, rather
1394 * than being the whole note contents. We fill the reset in here.
1395 * We assume that regset 0 is NT_PRSTATUS.
1396 */
1397 fill_prstatus(&t->prstatus, t->task, signr);
1398 (void) view->regsets[0].get(t->task, &view->regsets[0],
1399 0, sizeof(t->prstatus.pr_reg),
1400 &t->prstatus.pr_reg, NULL);
1401
1402 fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
1403 sizeof(t->prstatus), &t->prstatus);
1404 *total += notesize(&t->notes[0]);
1405
Roland McGrathd31472b2008-03-04 14:28:30 -08001406 do_thread_regset_writeback(t->task, &view->regsets[0]);
1407
Roland McGrath4206d3a2008-01-30 13:31:45 +01001408 /*
1409 * Each other regset might generate a note too. For each regset
1410 * that has no core_note_type or is inactive, we leave t->notes[i]
1411 * all zero and we'll know to skip writing it later.
1412 */
1413 for (i = 1; i < view->n; ++i) {
1414 const struct user_regset *regset = &view->regsets[i];
Roland McGrathd31472b2008-03-04 14:28:30 -08001415 do_thread_regset_writeback(t->task, regset);
Roland McGrath4206d3a2008-01-30 13:31:45 +01001416 if (regset->core_note_type &&
1417 (!regset->active || regset->active(t->task, regset))) {
1418 int ret;
1419 size_t size = regset->n * regset->size;
1420 void *data = kmalloc(size, GFP_KERNEL);
1421 if (unlikely(!data))
1422 return 0;
1423 ret = regset->get(t->task, regset,
1424 0, size, data, NULL);
1425 if (unlikely(ret))
1426 kfree(data);
1427 else {
1428 if (regset->core_note_type != NT_PRFPREG)
1429 fill_note(&t->notes[i], "LINUX",
1430 regset->core_note_type,
1431 size, data);
1432 else {
1433 t->prstatus.pr_fpvalid = 1;
1434 fill_note(&t->notes[i], "CORE",
1435 NT_PRFPREG, size, data);
1436 }
1437 *total += notesize(&t->notes[i]);
1438 }
1439 }
1440 }
1441
1442 return 1;
1443}
1444
1445static int fill_note_info(struct elfhdr *elf, int phdrs,
1446 struct elf_note_info *info,
1447 long signr, struct pt_regs *regs)
1448{
1449 struct task_struct *dump_task = current;
1450 const struct user_regset_view *view = task_user_regset_view(dump_task);
1451 struct elf_thread_core_info *t;
1452 struct elf_prpsinfo *psinfo;
Oleg Nesterov83914442008-07-25 01:47:45 -07001453 struct core_thread *ct;
Roland McGrath4206d3a2008-01-30 13:31:45 +01001454 unsigned int i;
1455
1456 info->size = 0;
1457 info->thread = NULL;
1458
1459 psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
Roland McGrath4206d3a2008-01-30 13:31:45 +01001460 if (psinfo == NULL)
1461 return 0;
1462
Amerigo Wange2dbe122009-07-01 01:06:26 -04001463 fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
1464
Roland McGrath4206d3a2008-01-30 13:31:45 +01001465 /*
1466 * Figure out how many notes we're going to need for each thread.
1467 */
1468 info->thread_notes = 0;
1469 for (i = 0; i < view->n; ++i)
1470 if (view->regsets[i].core_note_type != 0)
1471 ++info->thread_notes;
1472
1473 /*
1474 * Sanity check. We rely on regset 0 being in NT_PRSTATUS,
1475 * since it is our one special case.
1476 */
1477 if (unlikely(info->thread_notes == 0) ||
1478 unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
1479 WARN_ON(1);
1480 return 0;
1481 }
1482
1483 /*
1484 * Initialize the ELF file header.
1485 */
1486 fill_elf_header(elf, phdrs,
1487 view->e_machine, view->e_flags, view->ei_osabi);
1488
1489 /*
1490 * Allocate a structure for each thread.
1491 */
Oleg Nesterov83914442008-07-25 01:47:45 -07001492 for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
1493 t = kzalloc(offsetof(struct elf_thread_core_info,
1494 notes[info->thread_notes]),
1495 GFP_KERNEL);
1496 if (unlikely(!t))
1497 return 0;
Oleg Nesterov24d52882008-07-25 01:47:40 -07001498
Oleg Nesterov83914442008-07-25 01:47:45 -07001499 t->task = ct->task;
1500 if (ct->task == dump_task || !info->thread) {
1501 t->next = info->thread;
1502 info->thread = t;
1503 } else {
1504 /*
1505 * Make sure to keep the original task at
1506 * the head of the list.
1507 */
1508 t->next = info->thread->next;
1509 info->thread->next = t;
Roland McGrath4206d3a2008-01-30 13:31:45 +01001510 }
Oleg Nesterov83914442008-07-25 01:47:45 -07001511 }
Roland McGrath4206d3a2008-01-30 13:31:45 +01001512
1513 /*
1514 * Now fill in each thread's information.
1515 */
1516 for (t = info->thread; t != NULL; t = t->next)
1517 if (!fill_thread_core_info(t, view, signr, &info->size))
1518 return 0;
1519
1520 /*
1521 * Fill in the two process-wide notes.
1522 */
1523 fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
1524 info->size += notesize(&info->psinfo);
1525
1526 fill_auxv_note(&info->auxv, current->mm);
1527 info->size += notesize(&info->auxv);
1528
1529 return 1;
1530}
1531
1532static size_t get_note_info_size(struct elf_note_info *info)
1533{
1534 return info->size;
1535}
1536
1537/*
1538 * Write all the notes for each thread. When writing the first thread, the
1539 * process-wide notes are interleaved after the first thread-specific note.
1540 */
1541static int write_note_info(struct elf_note_info *info,
1542 struct file *file, loff_t *foffset)
1543{
1544 bool first = 1;
1545 struct elf_thread_core_info *t = info->thread;
1546
1547 do {
1548 int i;
1549
1550 if (!writenote(&t->notes[0], file, foffset))
1551 return 0;
1552
1553 if (first && !writenote(&info->psinfo, file, foffset))
1554 return 0;
1555 if (first && !writenote(&info->auxv, file, foffset))
1556 return 0;
1557
1558 for (i = 1; i < info->thread_notes; ++i)
1559 if (t->notes[i].data &&
1560 !writenote(&t->notes[i], file, foffset))
1561 return 0;
1562
1563 first = 0;
1564 t = t->next;
1565 } while (t);
1566
1567 return 1;
1568}
1569
1570static void free_note_info(struct elf_note_info *info)
1571{
1572 struct elf_thread_core_info *threads = info->thread;
1573 while (threads) {
1574 unsigned int i;
1575 struct elf_thread_core_info *t = threads;
1576 threads = t->next;
1577 WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
1578 for (i = 1; i < info->thread_notes; ++i)
1579 kfree(t->notes[i].data);
1580 kfree(t);
1581 }
1582 kfree(info->psinfo.data);
1583}
1584
1585#else
1586
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587/* Here is the structure in which status of each thread is captured. */
1588struct elf_thread_status
1589{
1590 struct list_head list;
1591 struct elf_prstatus prstatus; /* NT_PRSTATUS */
1592 elf_fpregset_t fpu; /* NT_PRFPREG */
1593 struct task_struct *thread;
1594#ifdef ELF_CORE_COPY_XFPREGS
Mark Nelson5b20cd82007-10-16 23:25:39 -07001595 elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596#endif
1597 struct memelfnote notes[3];
1598 int num_notes;
1599};
1600
1601/*
1602 * In order to add the specific thread information for the elf file format,
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001603 * we need to keep a linked list of every threads pr_status and then create
1604 * a single section for them in the final core file.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 */
1606static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
1607{
1608 int sz = 0;
1609 struct task_struct *p = t->thread;
1610 t->num_notes = 0;
1611
1612 fill_prstatus(&t->prstatus, p, signr);
1613 elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
1614
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001615 fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
1616 &(t->prstatus));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 t->num_notes++;
1618 sz += notesize(&t->notes[0]);
1619
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001620 if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
1621 &t->fpu))) {
1622 fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
1623 &(t->fpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001624 t->num_notes++;
1625 sz += notesize(&t->notes[1]);
1626 }
1627
1628#ifdef ELF_CORE_COPY_XFPREGS
1629 if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
Mark Nelson5b20cd82007-10-16 23:25:39 -07001630 fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
1631 sizeof(t->xfpu), &t->xfpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 t->num_notes++;
1633 sz += notesize(&t->notes[2]);
1634 }
1635#endif
1636 return sz;
1637}
1638
Roland McGrath3aba4812008-01-30 13:31:44 +01001639struct elf_note_info {
1640 struct memelfnote *notes;
1641 struct elf_prstatus *prstatus; /* NT_PRSTATUS */
1642 struct elf_prpsinfo *psinfo; /* NT_PRPSINFO */
1643 struct list_head thread_list;
1644 elf_fpregset_t *fpu;
1645#ifdef ELF_CORE_COPY_XFPREGS
1646 elf_fpxregset_t *xfpu;
1647#endif
1648 int thread_status_size;
1649 int numnote;
1650};
1651
Amerigo Wang0cf062d2009-09-23 15:57:05 -07001652static int elf_note_info_init(struct elf_note_info *info)
Roland McGrath3aba4812008-01-30 13:31:44 +01001653{
Amerigo Wang0cf062d2009-09-23 15:57:05 -07001654 memset(info, 0, sizeof(*info));
Roland McGrath3aba4812008-01-30 13:31:44 +01001655 INIT_LIST_HEAD(&info->thread_list);
1656
Amerigo Wang0cf062d2009-09-23 15:57:05 -07001657 /* Allocate space for six ELF notes */
1658 info->notes = kmalloc(6 * sizeof(struct memelfnote), GFP_KERNEL);
Roland McGrath3aba4812008-01-30 13:31:44 +01001659 if (!info->notes)
1660 return 0;
1661 info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
1662 if (!info->psinfo)
Amerigo Wang0cf062d2009-09-23 15:57:05 -07001663 goto notes_free;
Roland McGrath3aba4812008-01-30 13:31:44 +01001664 info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
1665 if (!info->prstatus)
Amerigo Wang0cf062d2009-09-23 15:57:05 -07001666 goto psinfo_free;
Roland McGrath3aba4812008-01-30 13:31:44 +01001667 info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
1668 if (!info->fpu)
Amerigo Wang0cf062d2009-09-23 15:57:05 -07001669 goto prstatus_free;
Roland McGrath3aba4812008-01-30 13:31:44 +01001670#ifdef ELF_CORE_COPY_XFPREGS
1671 info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
1672 if (!info->xfpu)
Amerigo Wang0cf062d2009-09-23 15:57:05 -07001673 goto fpu_free;
Roland McGrath3aba4812008-01-30 13:31:44 +01001674#endif
Amerigo Wang0cf062d2009-09-23 15:57:05 -07001675 return 1;
1676#ifdef ELF_CORE_COPY_XFPREGS
1677 fpu_free:
1678 kfree(info->fpu);
1679#endif
1680 prstatus_free:
1681 kfree(info->prstatus);
1682 psinfo_free:
1683 kfree(info->psinfo);
1684 notes_free:
1685 kfree(info->notes);
1686 return 0;
1687}
Roland McGrath3aba4812008-01-30 13:31:44 +01001688
Amerigo Wang0cf062d2009-09-23 15:57:05 -07001689static int fill_note_info(struct elfhdr *elf, int phdrs,
1690 struct elf_note_info *info,
1691 long signr, struct pt_regs *regs)
1692{
1693 struct list_head *t;
1694
1695 if (!elf_note_info_init(info))
1696 return 0;
1697
Roland McGrath3aba4812008-01-30 13:31:44 +01001698 if (signr) {
Oleg Nesterov83914442008-07-25 01:47:45 -07001699 struct core_thread *ct;
WANG Cong4220b7f2008-04-29 01:01:18 -07001700 struct elf_thread_status *ets;
Oleg Nesterov24d52882008-07-25 01:47:40 -07001701
Oleg Nesterov83914442008-07-25 01:47:45 -07001702 for (ct = current->mm->core_state->dumper.next;
1703 ct; ct = ct->next) {
1704 ets = kzalloc(sizeof(*ets), GFP_KERNEL);
1705 if (!ets)
1706 return 0;
1707
1708 ets->thread = ct->task;
1709 list_add(&ets->list, &info->thread_list);
1710 }
1711
Roland McGrath3aba4812008-01-30 13:31:44 +01001712 list_for_each(t, &info->thread_list) {
Roland McGrath3aba4812008-01-30 13:31:44 +01001713 int sz;
1714
WANG Cong4220b7f2008-04-29 01:01:18 -07001715 ets = list_entry(t, struct elf_thread_status, list);
1716 sz = elf_dump_thread_status(signr, ets);
Roland McGrath3aba4812008-01-30 13:31:44 +01001717 info->thread_status_size += sz;
1718 }
1719 }
1720 /* now collect the dump for the current */
1721 memset(info->prstatus, 0, sizeof(*info->prstatus));
1722 fill_prstatus(info->prstatus, current, signr);
1723 elf_core_copy_regs(&info->prstatus->pr_reg, regs);
1724
1725 /* Set up header */
1726 fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS, ELF_OSABI);
1727
1728 /*
1729 * Set up the notes in similar form to SVR4 core dumps made
1730 * with info from their /proc.
1731 */
1732
1733 fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
1734 sizeof(*info->prstatus), info->prstatus);
1735 fill_psinfo(info->psinfo, current->group_leader, current->mm);
1736 fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
1737 sizeof(*info->psinfo), info->psinfo);
1738
1739 info->numnote = 2;
1740
1741 fill_auxv_note(&info->notes[info->numnote++], current->mm);
1742
1743 /* Try to dump the FPU. */
1744 info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
1745 info->fpu);
1746 if (info->prstatus->pr_fpvalid)
1747 fill_note(info->notes + info->numnote++,
1748 "CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
1749#ifdef ELF_CORE_COPY_XFPREGS
1750 if (elf_core_copy_task_xfpregs(current, info->xfpu))
1751 fill_note(info->notes + info->numnote++,
1752 "LINUX", ELF_CORE_XFPREG_TYPE,
1753 sizeof(*info->xfpu), info->xfpu);
1754#endif
1755
1756 return 1;
Roland McGrath3aba4812008-01-30 13:31:44 +01001757}
1758
1759static size_t get_note_info_size(struct elf_note_info *info)
1760{
1761 int sz = 0;
1762 int i;
1763
1764 for (i = 0; i < info->numnote; i++)
1765 sz += notesize(info->notes + i);
1766
1767 sz += info->thread_status_size;
1768
1769 return sz;
1770}
1771
1772static int write_note_info(struct elf_note_info *info,
1773 struct file *file, loff_t *foffset)
1774{
1775 int i;
1776 struct list_head *t;
1777
1778 for (i = 0; i < info->numnote; i++)
1779 if (!writenote(info->notes + i, file, foffset))
1780 return 0;
1781
1782 /* write out the thread status notes section */
1783 list_for_each(t, &info->thread_list) {
1784 struct elf_thread_status *tmp =
1785 list_entry(t, struct elf_thread_status, list);
1786
1787 for (i = 0; i < tmp->num_notes; i++)
1788 if (!writenote(&tmp->notes[i], file, foffset))
1789 return 0;
1790 }
1791
1792 return 1;
1793}
1794
1795static void free_note_info(struct elf_note_info *info)
1796{
1797 while (!list_empty(&info->thread_list)) {
1798 struct list_head *tmp = info->thread_list.next;
1799 list_del(tmp);
1800 kfree(list_entry(tmp, struct elf_thread_status, list));
1801 }
1802
1803 kfree(info->prstatus);
1804 kfree(info->psinfo);
1805 kfree(info->notes);
1806 kfree(info->fpu);
1807#ifdef ELF_CORE_COPY_XFPREGS
1808 kfree(info->xfpu);
1809#endif
1810}
1811
Roland McGrath4206d3a2008-01-30 13:31:45 +01001812#endif
1813
Roland McGrathf47aef52007-01-26 00:56:49 -08001814static struct vm_area_struct *first_vma(struct task_struct *tsk,
1815 struct vm_area_struct *gate_vma)
1816{
1817 struct vm_area_struct *ret = tsk->mm->mmap;
1818
1819 if (ret)
1820 return ret;
1821 return gate_vma;
1822}
1823/*
1824 * Helper function for iterating across a vma list. It ensures that the caller
1825 * will visit `gate_vma' prior to terminating the search.
1826 */
1827static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
1828 struct vm_area_struct *gate_vma)
1829{
1830 struct vm_area_struct *ret;
1831
1832 ret = this_vma->vm_next;
1833 if (ret)
1834 return ret;
1835 if (this_vma == gate_vma)
1836 return NULL;
1837 return gate_vma;
1838}
1839
Daisuke HATAYAMA8d9032b2010-03-05 13:44:10 -08001840static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
1841 elf_addr_t e_shoff, int segs)
1842{
1843 elf->e_shoff = e_shoff;
1844 elf->e_shentsize = sizeof(*shdr4extnum);
1845 elf->e_shnum = 1;
1846 elf->e_shstrndx = SHN_UNDEF;
1847
1848 memset(shdr4extnum, 0, sizeof(*shdr4extnum));
1849
1850 shdr4extnum->sh_type = SHT_NULL;
1851 shdr4extnum->sh_size = elf->e_shnum;
1852 shdr4extnum->sh_link = elf->e_shstrndx;
1853 shdr4extnum->sh_info = segs;
1854}
1855
1856static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
1857 unsigned long mm_flags)
1858{
1859 struct vm_area_struct *vma;
1860 size_t size = 0;
1861
1862 for (vma = first_vma(current, gate_vma); vma != NULL;
1863 vma = next_vma(vma, gate_vma))
1864 size += vma_dump_size(vma, mm_flags);
1865 return size;
1866}
1867
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868/*
1869 * Actual dumper
1870 *
1871 * This is a two-pass process; first we find the offsets of the bits,
1872 * and then they are actually written out. If we run out of core limit
1873 * we just truncate.
1874 */
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08001875static int elf_core_dump(struct coredump_params *cprm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 int has_dumped = 0;
1878 mm_segment_t fs;
1879 int segs;
1880 size_t size = 0;
Roland McGrathf47aef52007-01-26 00:56:49 -08001881 struct vm_area_struct *vma, *gate_vma;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 struct elfhdr *elf = NULL;
Andi Kleend025c9d2006-09-30 23:29:28 -07001883 loff_t offset = 0, dataoff, foffset;
Roland McGrath3aba4812008-01-30 13:31:44 +01001884 struct elf_note_info info;
Daisuke HATAYAMA93eb2112010-03-05 13:44:09 -08001885 struct elf_phdr *phdr4note = NULL;
Daisuke HATAYAMA8d9032b2010-03-05 13:44:10 -08001886 struct elf_shdr *shdr4extnum = NULL;
1887 Elf_Half e_phnum;
1888 elf_addr_t e_shoff;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889
1890 /*
1891 * We no longer stop all VM operations.
1892 *
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001893 * This is because those proceses that could possibly change map_count
1894 * or the mmap / vma pages are now blocked in do_exit on current
1895 * finishing this core dump.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 *
1897 * Only ptrace can touch these memory addresses, but it doesn't change
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001898 * the map_count or the pages allocated. So no possibility of crashing
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 * exists while dumping the mm->vm_next areas to the core file.
1900 */
1901
1902 /* alloc memory for large data structures: too large to be on stack */
1903 elf = kmalloc(sizeof(*elf), GFP_KERNEL);
1904 if (!elf)
WANG Cong5f719552008-05-06 12:45:35 +08001905 goto out;
KAMEZAWA Hiroyuki341c87b2009-06-30 11:41:23 -07001906 /*
1907 * The number of segs are recored into ELF header as 16bit value.
1908 * Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
1909 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 segs = current->mm->map_count;
Daisuke HATAYAMA1fcccba2010-03-05 13:44:07 -08001911 segs += elf_core_extra_phdrs();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912
Stephen Wilson31db58b2011-03-13 15:49:15 -04001913 gate_vma = get_gate_vma(current->mm);
Roland McGrathf47aef52007-01-26 00:56:49 -08001914 if (gate_vma != NULL)
1915 segs++;
1916
Daisuke HATAYAMA8d9032b2010-03-05 13:44:10 -08001917 /* for notes section */
1918 segs++;
1919
1920 /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
1921 * this, kernel supports extended numbering. Have a look at
1922 * include/linux/elf.h for further information. */
1923 e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
1924
Roland McGrath3aba4812008-01-30 13:31:44 +01001925 /*
1926 * Collect all the non-memory information about the process for the
1927 * notes. This also sets up the file header.
1928 */
Daisuke HATAYAMA8d9032b2010-03-05 13:44:10 -08001929 if (!fill_note_info(elf, e_phnum, &info, cprm->signr, cprm->regs))
Roland McGrath3aba4812008-01-30 13:31:44 +01001930 goto cleanup;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931
1932 has_dumped = 1;
1933 current->flags |= PF_DUMPCORE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934
1935 fs = get_fs();
1936 set_fs(KERNEL_DS);
1937
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 offset += sizeof(*elf); /* Elf header */
Daisuke HATAYAMA8d9032b2010-03-05 13:44:10 -08001939 offset += segs * sizeof(struct elf_phdr); /* Program headers */
Petr Vandroveca7a0d862006-10-13 18:42:07 +02001940 foffset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941
1942 /* Write notes phdr entry */
1943 {
Roland McGrath3aba4812008-01-30 13:31:44 +01001944 size_t sz = get_note_info_size(&info);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945
Michael Ellermane5501492007-09-19 14:38:12 +10001946 sz += elf_coredump_extra_notes_size();
Dwayne Grant McConnellbf1ab972006-11-23 00:46:37 +01001947
Daisuke HATAYAMA93eb2112010-03-05 13:44:09 -08001948 phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
1949 if (!phdr4note)
Daisuke HATAYAMA088e7af2010-03-05 13:44:06 -08001950 goto end_coredump;
Daisuke HATAYAMA93eb2112010-03-05 13:44:09 -08001951
1952 fill_elf_note_phdr(phdr4note, sz, offset);
1953 offset += sz;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 }
1955
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
1957
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001958 offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
Daisuke HATAYAMA8d9032b2010-03-05 13:44:10 -08001959 offset += elf_core_extra_data_size();
1960 e_shoff = offset;
1961
1962 if (e_phnum == PN_XNUM) {
1963 shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
1964 if (!shdr4extnum)
1965 goto end_coredump;
1966 fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
1967 }
1968
1969 offset = dataoff;
1970
Daisuke HATAYAMA93eb2112010-03-05 13:44:09 -08001971 size += sizeof(*elf);
1972 if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
1973 goto end_coredump;
1974
1975 size += sizeof(*phdr4note);
1976 if (size > cprm->limit
1977 || !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
1978 goto end_coredump;
1979
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 /* Write program headers for segments dump */
Roland McGrathf47aef52007-01-26 00:56:49 -08001981 for (vma = first_vma(current, gate_vma); vma != NULL;
1982 vma = next_vma(vma, gate_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 struct elf_phdr phdr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984
1985 phdr.p_type = PT_LOAD;
1986 phdr.p_offset = offset;
1987 phdr.p_vaddr = vma->vm_start;
1988 phdr.p_paddr = 0;
Masami Hiramatsu30736a42010-03-05 13:44:12 -08001989 phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
Roland McGrath82df3972007-10-16 23:27:02 -07001990 phdr.p_memsz = vma->vm_end - vma->vm_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991 offset += phdr.p_filesz;
1992 phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07001993 if (vma->vm_flags & VM_WRITE)
1994 phdr.p_flags |= PF_W;
1995 if (vma->vm_flags & VM_EXEC)
1996 phdr.p_flags |= PF_X;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 phdr.p_align = ELF_EXEC_PAGESIZE;
1998
Daisuke HATAYAMA088e7af2010-03-05 13:44:06 -08001999 size += sizeof(phdr);
2000 if (size > cprm->limit
2001 || !dump_write(cprm->file, &phdr, sizeof(phdr)))
2002 goto end_coredump;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 }
2004
Daisuke HATAYAMA1fcccba2010-03-05 13:44:07 -08002005 if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
2006 goto end_coredump;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007
2008 /* write out the notes section */
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002009 if (!write_note_info(&info, cprm->file, &foffset))
Roland McGrath3aba4812008-01-30 13:31:44 +01002010 goto end_coredump;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002011
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002012 if (elf_coredump_extra_notes_write(cprm->file, &foffset))
Michael Ellermane5501492007-09-19 14:38:12 +10002013 goto end_coredump;
Dwayne Grant McConnellbf1ab972006-11-23 00:46:37 +01002014
Andi Kleend025c9d2006-09-30 23:29:28 -07002015 /* Align to page */
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002016 if (!dump_seek(cprm->file, dataoff - foffset))
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07002017 goto end_coredump;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018
Roland McGrathf47aef52007-01-26 00:56:49 -08002019 for (vma = first_vma(current, gate_vma); vma != NULL;
2020 vma = next_vma(vma, gate_vma)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 unsigned long addr;
Roland McGrath82df3972007-10-16 23:27:02 -07002022 unsigned long end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002023
Masami Hiramatsu30736a42010-03-05 13:44:12 -08002024 end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025
Roland McGrath82df3972007-10-16 23:27:02 -07002026 for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
Jesper Juhlf4e5cc22006-06-23 02:05:35 -07002027 struct page *page;
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07002028 int stop;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07002030 page = get_dump_page(addr);
2031 if (page) {
2032 void *kaddr = kmap(page);
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002033 stop = ((size += PAGE_SIZE) > cprm->limit) ||
2034 !dump_write(cprm->file, kaddr,
2035 PAGE_SIZE);
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07002036 kunmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 page_cache_release(page);
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07002038 } else
Masami Hiramatsuf6151df2009-12-17 15:27:16 -08002039 stop = !dump_seek(cprm->file, PAGE_SIZE);
Hugh Dickinsf3e8fcc2009-09-21 17:03:25 -07002040 if (stop)
2041 goto end_coredump;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042 }
2043 }
2044
Daisuke HATAYAMA1fcccba2010-03-05 13:44:07 -08002045 if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
2046 goto end_coredump;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047
Daisuke HATAYAMA8d9032b2010-03-05 13:44:10 -08002048 if (e_phnum == PN_XNUM) {
2049 size += sizeof(*shdr4extnum);
2050 if (size > cprm->limit
2051 || !dump_write(cprm->file, shdr4extnum,
2052 sizeof(*shdr4extnum)))
2053 goto end_coredump;
2054 }
2055
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056end_coredump:
2057 set_fs(fs);
2058
2059cleanup:
Roland McGrath3aba4812008-01-30 13:31:44 +01002060 free_note_info(&info);
Daisuke HATAYAMA8d9032b2010-03-05 13:44:10 -08002061 kfree(shdr4extnum);
Daisuke HATAYAMA93eb2112010-03-05 13:44:09 -08002062 kfree(phdr4note);
WANG Cong5f719552008-05-06 12:45:35 +08002063 kfree(elf);
2064out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 return has_dumped;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066}
2067
Christoph Hellwig698ba7b2009-12-15 16:47:37 -08002068#endif /* CONFIG_ELF_CORE */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069
2070static int __init init_elf_binfmt(void)
2071{
2072 return register_binfmt(&elf_format);
2073}
2074
2075static void __exit exit_elf_binfmt(void)
2076{
2077 /* Remove the COFF and ELF loaders. */
2078 unregister_binfmt(&elf_format);
2079}
2080
2081core_initcall(init_elf_binfmt);
2082module_exit(exit_elf_binfmt);
2083MODULE_LICENSE("GPL");