2
0
mirror of https://github.com/edk2-porting/linux-next.git synced 2024-12-25 05:34:00 +08:00
linux-next/fs/binfmt_elf.c

2249 lines
58 KiB
C
Raw Normal View History

/*
* linux/fs/binfmt_elf.c
*
* These are the functions used to load ELF format executables as used
* on SVr4 machines. Information on the format may be found in the book
* "UNIX SYSTEM V RELEASE 4 Programmers Guide: Ansi C and Programming Support
* Tools".
*
* Copyright 1993, 1994: Eric Youngdale (ericy@cais.com).
*/
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/fs.h>
#include <linux/mm.h>
#include <linux/mman.h>
#include <linux/errno.h>
#include <linux/signal.h>
#include <linux/binfmts.h>
#include <linux/string.h>
#include <linux/file.h>
#include <linux/slab.h>
#include <linux/personality.h>
#include <linux/elfcore.h>
#include <linux/init.h>
#include <linux/highuid.h>
#include <linux/compiler.h>
#include <linux/highmem.h>
#include <linux/pagemap.h>
#include <linux/vmalloc.h>
#include <linux/security.h>
#include <linux/random.h>
#include <linux/elf.h>
#include <linux/utsname.h>
#include <linux/coredump.h>
#include <linux/sched.h>
#include <asm/uaccess.h>
#include <asm/param.h>
#include <asm/page.h>
#ifndef user_long_t
#define user_long_t long
#endif
#ifndef user_siginfo_t
#define user_siginfo_t siginfo_t
#endif
static int load_elf_binary(struct linux_binprm *bprm);
static int load_elf_library(struct file *);
x86: PIE executable randomization, checkpatch fixes #39: FILE: arch/ia64/ia32/binfmt_elf32.c:229: +elf32_map (struct file *filep, unsigned long addr, struct elf_phdr *eppnt, int prot, int type, unsigned long unused) WARNING: no space between function name and open parenthesis '(' #39: FILE: arch/ia64/ia32/binfmt_elf32.c:229: +elf32_map (struct file *filep, unsigned long addr, struct elf_phdr *eppnt, int prot, int type, unsigned long unused) WARNING: line over 80 characters #67: FILE: arch/x86/kernel/sys_x86_64.c:80: + new_begin = randomize_range(*begin, *begin + 0x02000000, 0); ERROR: use tabs not spaces #110: FILE: arch/x86/kernel/sys_x86_64.c:185: + ^I mm->cached_hole_size = 0;$ ERROR: use tabs not spaces #111: FILE: arch/x86/kernel/sys_x86_64.c:186: + ^I^Imm->free_area_cache = mm->mmap_base;$ ERROR: use tabs not spaces #112: FILE: arch/x86/kernel/sys_x86_64.c:187: + ^I}$ ERROR: use tabs not spaces #141: FILE: arch/x86/kernel/sys_x86_64.c:216: + ^I^I/* remember the largest hole we saw so far */$ ERROR: use tabs not spaces #142: FILE: arch/x86/kernel/sys_x86_64.c:217: + ^I^Iif (addr + mm->cached_hole_size < vma->vm_start)$ ERROR: use tabs not spaces #143: FILE: arch/x86/kernel/sys_x86_64.c:218: + ^I^I mm->cached_hole_size = vma->vm_start - addr;$ ERROR: use tabs not spaces #157: FILE: arch/x86/kernel/sys_x86_64.c:232: + ^Imm->free_area_cache = TASK_UNMAPPED_BASE;$ ERROR: need a space before the open parenthesis '(' #291: FILE: arch/x86/mm/mmap_64.c:101: + } else if(mmap_is_legacy()) { WARNING: braces {} are not necessary for single statement blocks #302: FILE: arch/x86/mm/mmap_64.c:112: + if (current->flags & PF_RANDOMIZE) { + mm->mmap_base += ((long)rnd) << PAGE_SHIFT; + } WARNING: line over 80 characters #314: FILE: fs/binfmt_elf.c:48: +static unsigned long elf_map (struct file *, unsigned long, struct elf_phdr *, int, int, unsigned long); WARNING: no space between function name and open parenthesis '(' #314: FILE: fs/binfmt_elf.c:48: +static unsigned long elf_map (struct file *, unsigned long, struct elf_phdr *, int, int, unsigned long); WARNING: line over 80 characters #429: FILE: fs/binfmt_elf.c:438: + eppnt, elf_prot, elf_type, total_size); ERROR: need space after that ',' (ctx:VxV) #480: FILE: fs/binfmt_elf.c:939: + elf_prot, elf_flags,0); ^ total: 9 errors, 7 warnings, 461 lines checked Your patch has style problems, please review. If any of these errors are false positives report them to the maintainer, see CHECKPATCH in MAINTAINERS. Please run checkpatch prior to sending patches Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Arjan van de Ven <arjan@infradead.org> Cc: Jakub Jelinek <jakub@redhat.com> Cc: Jiri Kosina <jkosina@suse.cz> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Roland McGrath <roland@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 20:31:07 +08:00
static unsigned long elf_map(struct file *, unsigned long, struct elf_phdr *,
int, int, unsigned long);
/*
* If we don't support core dumping, then supply a NULL so we
* don't even try.
*/
#ifdef CONFIG_ELF_CORE
static int elf_core_dump(struct coredump_params *cprm);
#else
#define elf_core_dump NULL
#endif
#if ELF_EXEC_PAGESIZE > PAGE_SIZE
#define ELF_MIN_ALIGN ELF_EXEC_PAGESIZE
#else
#define ELF_MIN_ALIGN PAGE_SIZE
#endif
#ifndef ELF_CORE_EFLAGS
#define ELF_CORE_EFLAGS 0
#endif
#define ELF_PAGESTART(_v) ((_v) & ~(unsigned long)(ELF_MIN_ALIGN-1))
#define ELF_PAGEOFFSET(_v) ((_v) & (ELF_MIN_ALIGN-1))
#define ELF_PAGEALIGN(_v) (((_v) + ELF_MIN_ALIGN - 1) & ~(ELF_MIN_ALIGN - 1))
static struct linux_binfmt elf_format = {
.module = THIS_MODULE,
.load_binary = load_elf_binary,
.load_shlib = load_elf_library,
.core_dump = elf_core_dump,
.min_coredump = ELF_EXEC_PAGESIZE,
};
#define BAD_ADDR(x) ((unsigned long)(x) >= TASK_SIZE)
static int set_brk(unsigned long start, unsigned long end)
{
start = ELF_PAGEALIGN(start);
end = ELF_PAGEALIGN(end);
if (end > start) {
unsigned long addr;
addr = vm_brk(start, end - start);
if (BAD_ADDR(addr))
return addr;
}
current->mm->start_brk = current->mm->brk = end;
return 0;
}
/* We need to explicitly zero any fractional pages
after the data section (i.e. bss). This would
contain the junk from the file that should not
be in memory
*/
static int padzero(unsigned long elf_bss)
{
unsigned long nbyte;
nbyte = ELF_PAGEOFFSET(elf_bss);
if (nbyte) {
nbyte = ELF_MIN_ALIGN - nbyte;
if (clear_user((void __user *) elf_bss, nbyte))
return -EFAULT;
}
return 0;
}
/* Let's use some macros to make this stack manipulation a little clearer */
#ifdef CONFIG_STACK_GROWSUP
#define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) + (items))
#define STACK_ROUND(sp, items) \
((15 + (unsigned long) ((sp) + (items))) &~ 15UL)
#define STACK_ALLOC(sp, len) ({ \
elf_addr_t __user *old_sp = (elf_addr_t __user *)sp; sp += len; \
old_sp; })
#else
#define STACK_ADD(sp, items) ((elf_addr_t __user *)(sp) - (items))
#define STACK_ROUND(sp, items) \
(((unsigned long) (sp - items)) &~ 15UL)
#define STACK_ALLOC(sp, len) ({ sp -= len ; sp; })
#endif
#ifndef ELF_BASE_PLATFORM
/*
* AT_BASE_PLATFORM indicates the "real" hardware/microarchitecture.
* If the arch defines ELF_BASE_PLATFORM (in asm/elf.h), the value
* will be copied to the user stack in the same manner as AT_PLATFORM.
*/
#define ELF_BASE_PLATFORM NULL
#endif
static int
create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
unsigned long load_addr, unsigned long interp_load_addr)
{
unsigned long p = bprm->p;
int argc = bprm->argc;
int envc = bprm->envc;
elf_addr_t __user *argv;
elf_addr_t __user *envp;
elf_addr_t __user *sp;
elf_addr_t __user *u_platform;
elf_addr_t __user *u_base_platform;
ELF: implement AT_RANDOM for glibc PRNG seeding While discussing[1] the need for glibc to have access to random bytes during program load, it seems that an earlier attempt to implement AT_RANDOM got stalled. This implements a random 16 byte string, available to every ELF program via a new auxv AT_RANDOM vector. [1] http://sourceware.org/ml/libc-alpha/2008-10/msg00006.html Ulrich said: glibc needs right after startup a bit of random data for internal protections (stack canary etc). What is now in upstream glibc is that we always unconditionally open /dev/urandom, read some data, and use it. For every process startup. That's slow. ... The solution is to provide a limited amount of random data to the starting process in the aux vector. I suggested 16 bytes and this is what the patch implements. If we need only 16 bytes or less we use the data directly. If we need more we'll use the 16 bytes to see a PRNG. This avoids the costly /dev/urandom use and it allows the kernel to use the most adequate source of random data for this purpose. It might not be the same pool as that for /dev/urandom. Concerns were expressed about the depletion of the randomness pool. But this patch doesn't make the situation worse, it doesn't deplete entropy more than happens now. Signed-off-by: Kees Cook <kees.cook@canonical.com> Cc: Jakub Jelinek <jakub@redhat.com> Cc: Andi Kleen <andi@firstfloor.org> Cc: Ulrich Drepper <drepper@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 10:08:52 +08:00
elf_addr_t __user *u_rand_bytes;
const char *k_platform = ELF_PLATFORM;
const char *k_base_platform = ELF_BASE_PLATFORM;
ELF: implement AT_RANDOM for glibc PRNG seeding While discussing[1] the need for glibc to have access to random bytes during program load, it seems that an earlier attempt to implement AT_RANDOM got stalled. This implements a random 16 byte string, available to every ELF program via a new auxv AT_RANDOM vector. [1] http://sourceware.org/ml/libc-alpha/2008-10/msg00006.html Ulrich said: glibc needs right after startup a bit of random data for internal protections (stack canary etc). What is now in upstream glibc is that we always unconditionally open /dev/urandom, read some data, and use it. For every process startup. That's slow. ... The solution is to provide a limited amount of random data to the starting process in the aux vector. I suggested 16 bytes and this is what the patch implements. If we need only 16 bytes or less we use the data directly. If we need more we'll use the 16 bytes to see a PRNG. This avoids the costly /dev/urandom use and it allows the kernel to use the most adequate source of random data for this purpose. It might not be the same pool as that for /dev/urandom. Concerns were expressed about the depletion of the randomness pool. But this patch doesn't make the situation worse, it doesn't deplete entropy more than happens now. Signed-off-by: Kees Cook <kees.cook@canonical.com> Cc: Jakub Jelinek <jakub@redhat.com> Cc: Andi Kleen <andi@firstfloor.org> Cc: Ulrich Drepper <drepper@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 10:08:52 +08:00
unsigned char k_rand_bytes[16];
int items;
elf_addr_t *elf_info;
int ei_index = 0;
const struct cred *cred = current_cred();
struct vm_area_struct *vma;
/*
* In some cases (e.g. Hyper-Threading), we want to avoid L1
* evictions by the processes running on the same package. One
* thing we can do is to shuffle the initial stack for them.
*/
p = arch_align_stack(p);
/*
* If this architecture has a platform capability string, copy it
* to userspace. In some cases (Sparc), this info is impossible
* for userspace to get any other way, in others (i386) it is
* merely difficult.
*/
u_platform = NULL;
if (k_platform) {
size_t len = strlen(k_platform) + 1;
u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
if (__copy_to_user(u_platform, k_platform, len))
return -EFAULT;
}
/*
* If this architecture has a "base" platform capability
* string, copy it to userspace.
*/
u_base_platform = NULL;
if (k_base_platform) {
size_t len = strlen(k_base_platform) + 1;
u_base_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
if (__copy_to_user(u_base_platform, k_base_platform, len))
return -EFAULT;
}
ELF: implement AT_RANDOM for glibc PRNG seeding While discussing[1] the need for glibc to have access to random bytes during program load, it seems that an earlier attempt to implement AT_RANDOM got stalled. This implements a random 16 byte string, available to every ELF program via a new auxv AT_RANDOM vector. [1] http://sourceware.org/ml/libc-alpha/2008-10/msg00006.html Ulrich said: glibc needs right after startup a bit of random data for internal protections (stack canary etc). What is now in upstream glibc is that we always unconditionally open /dev/urandom, read some data, and use it. For every process startup. That's slow. ... The solution is to provide a limited amount of random data to the starting process in the aux vector. I suggested 16 bytes and this is what the patch implements. If we need only 16 bytes or less we use the data directly. If we need more we'll use the 16 bytes to see a PRNG. This avoids the costly /dev/urandom use and it allows the kernel to use the most adequate source of random data for this purpose. It might not be the same pool as that for /dev/urandom. Concerns were expressed about the depletion of the randomness pool. But this patch doesn't make the situation worse, it doesn't deplete entropy more than happens now. Signed-off-by: Kees Cook <kees.cook@canonical.com> Cc: Jakub Jelinek <jakub@redhat.com> Cc: Andi Kleen <andi@firstfloor.org> Cc: Ulrich Drepper <drepper@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 10:08:52 +08:00
/*
* Generate 16 random bytes for userspace PRNG seeding.
*/
get_random_bytes(k_rand_bytes, sizeof(k_rand_bytes));
u_rand_bytes = (elf_addr_t __user *)
STACK_ALLOC(p, sizeof(k_rand_bytes));
if (__copy_to_user(u_rand_bytes, k_rand_bytes, sizeof(k_rand_bytes)))
return -EFAULT;
/* Create the ELF interpreter info */
elf_info = (elf_addr_t *)current->mm->saved_auxv;
/* update AT_VECTOR_SIZE_BASE if the number of NEW_AUX_ENT() changes */
#define NEW_AUX_ENT(id, val) \
do { \
elf_info[ei_index++] = id; \
elf_info[ei_index++] = val; \
} while (0)
#ifdef ARCH_DLINFO
/*
* ARCH_DLINFO must come first so PPC can do its special alignment of
* AUXV.
* update AT_VECTOR_SIZE_ARCH if the number of NEW_AUX_ENT() in
* ARCH_DLINFO changes
*/
ARCH_DLINFO;
#endif
NEW_AUX_ENT(AT_HWCAP, ELF_HWCAP);
NEW_AUX_ENT(AT_PAGESZ, ELF_EXEC_PAGESIZE);
NEW_AUX_ENT(AT_CLKTCK, CLOCKS_PER_SEC);
NEW_AUX_ENT(AT_PHDR, load_addr + exec->e_phoff);
NEW_AUX_ENT(AT_PHENT, sizeof(struct elf_phdr));
NEW_AUX_ENT(AT_PHNUM, exec->e_phnum);
NEW_AUX_ENT(AT_BASE, interp_load_addr);
NEW_AUX_ENT(AT_FLAGS, 0);
NEW_AUX_ENT(AT_ENTRY, exec->e_entry);
NEW_AUX_ENT(AT_UID, from_kuid_munged(cred->user_ns, cred->uid));
NEW_AUX_ENT(AT_EUID, from_kuid_munged(cred->user_ns, cred->euid));
NEW_AUX_ENT(AT_GID, from_kgid_munged(cred->user_ns, cred->gid));
NEW_AUX_ENT(AT_EGID, from_kgid_munged(cred->user_ns, cred->egid));
NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
ELF: implement AT_RANDOM for glibc PRNG seeding While discussing[1] the need for glibc to have access to random bytes during program load, it seems that an earlier attempt to implement AT_RANDOM got stalled. This implements a random 16 byte string, available to every ELF program via a new auxv AT_RANDOM vector. [1] http://sourceware.org/ml/libc-alpha/2008-10/msg00006.html Ulrich said: glibc needs right after startup a bit of random data for internal protections (stack canary etc). What is now in upstream glibc is that we always unconditionally open /dev/urandom, read some data, and use it. For every process startup. That's slow. ... The solution is to provide a limited amount of random data to the starting process in the aux vector. I suggested 16 bytes and this is what the patch implements. If we need only 16 bytes or less we use the data directly. If we need more we'll use the 16 bytes to see a PRNG. This avoids the costly /dev/urandom use and it allows the kernel to use the most adequate source of random data for this purpose. It might not be the same pool as that for /dev/urandom. Concerns were expressed about the depletion of the randomness pool. But this patch doesn't make the situation worse, it doesn't deplete entropy more than happens now. Signed-off-by: Kees Cook <kees.cook@canonical.com> Cc: Jakub Jelinek <jakub@redhat.com> Cc: Andi Kleen <andi@firstfloor.org> Cc: Ulrich Drepper <drepper@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-01-08 10:08:52 +08:00
NEW_AUX_ENT(AT_RANDOM, (elf_addr_t)(unsigned long)u_rand_bytes);
#ifdef ELF_HWCAP2
NEW_AUX_ENT(AT_HWCAP2, ELF_HWCAP2);
#endif
execve filename: document and export via auxiliary vector The Linux kernel puts the filename argument of execve() into the new address space. Many developers are surprised to learn this. Those who know and could use it, object "But it's not documented." Those who want to use it dislike the expression (char *)(1+ strlen(env[-1+ n_env]) + env[-1+ n_env]) because it requires locating the last original environment variable, and assumes that the filename follows the characters. This patch documents the insertion of the filename, and makes it easier to find by adding a new tag AT_EXECFN in the ElfXX_auxv_t; see <elf.h>. In many cases readlink("/proc/self/exe",) gives the same answer. But if all the original pages get unmapped, then the kernel erases the symlink for /proc/self/exe. This can happen when a program decompressor does a good job of cleaning up after uncompressing directly to memory, so that the address space of the target program looks the same as if compression had never happened. One example is http://upx.sourceforge.net . One notable use of the underlying concept (what path containED the executable) is glibc expanding $ORIGIN in DT_RUNPATH. In practice for the near term, it may be a good idea for user-mode code to use both /proc/self/exe and AT_EXECFN as fall-back methods for each other. /proc/self/exe can fail due to unmapping, AT_EXECFN can fail because it won't be present on non-new systems. The auxvec or {AT_EXECFN}.d_val also can get overwritten, although in nearly all cases this would be the result of a bug. The runtime cost is one NEW_AUX_ENT using two words of stack space. The underlying value is maintained already as bprm->exec; setup_arg_pages() in fs/exec.c slides it for stack_shift, etc. Signed-off-by: John Reiser <jreiser@BitWagon.com> Cc: Roland McGrath <roland@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Cc: Ulrich Drepper <drepper@redhat.com> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-22 05:21:32 +08:00
NEW_AUX_ENT(AT_EXECFN, bprm->exec);
if (k_platform) {
NEW_AUX_ENT(AT_PLATFORM,
(elf_addr_t)(unsigned long)u_platform);
}
if (k_base_platform) {
NEW_AUX_ENT(AT_BASE_PLATFORM,
(elf_addr_t)(unsigned long)u_base_platform);
}
if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) {
NEW_AUX_ENT(AT_EXECFD, bprm->interp_data);
}
#undef NEW_AUX_ENT
/* AT_NULL is zero; clear the rest too */
memset(&elf_info[ei_index], 0,
sizeof current->mm->saved_auxv - ei_index * sizeof elf_info[0]);
/* And advance past the AT_NULL entry. */
ei_index += 2;
sp = STACK_ADD(p, ei_index);
items = (argc + 1) + (envc + 1) + 1;
bprm->p = STACK_ROUND(sp, items);
/* Point sp at the lowest address on the stack */
#ifdef CONFIG_STACK_GROWSUP
sp = (elf_addr_t __user *)bprm->p - items - ei_index;
bprm->exec = (unsigned long)sp; /* XXX: PARISC HACK */
#else
sp = (elf_addr_t __user *)bprm->p;
#endif
/*
* Grow the stack manually; some architectures have a limit on how
* far ahead a user-space access may be in order to grow the stack.
*/
vma = find_extend_vma(current->mm, bprm->p);
if (!vma)
return -EFAULT;
/* Now, let's put argc (and argv, envp if appropriate) on the stack */
if (__put_user(argc, sp++))
return -EFAULT;
argv = sp;
envp = argv + argc + 1;
/* Populate argv and envp */
p = current->mm->arg_end = current->mm->arg_start;
while (argc-- > 0) {
size_t len;
if (__put_user((elf_addr_t)p, argv++))
return -EFAULT;
len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
if (!len || len > MAX_ARG_STRLEN)
return -EINVAL;
p += len;
}
if (__put_user(0, argv))
return -EFAULT;
current->mm->arg_end = current->mm->env_start = p;
while (envc-- > 0) {
size_t len;
if (__put_user((elf_addr_t)p, envp++))
return -EFAULT;
len = strnlen_user((void __user *)p, MAX_ARG_STRLEN);
if (!len || len > MAX_ARG_STRLEN)
return -EINVAL;
p += len;
}
if (__put_user(0, envp))
return -EFAULT;
current->mm->env_end = p;
/* Put the elf_info on the stack in the right place. */
sp = (elf_addr_t __user *)envp + 1;
if (copy_to_user(sp, elf_info, ei_index * sizeof(elf_addr_t)))
return -EFAULT;
return 0;
}
#ifndef elf_map
static unsigned long elf_map(struct file *filep, unsigned long addr,
struct elf_phdr *eppnt, int prot, int type,
unsigned long total_size)
{
unsigned long map_addr;
unsigned long size = eppnt->p_filesz + ELF_PAGEOFFSET(eppnt->p_vaddr);
unsigned long off = eppnt->p_offset - ELF_PAGEOFFSET(eppnt->p_vaddr);
addr = ELF_PAGESTART(addr);
size = ELF_PAGEALIGN(size);
/* mmap() will return -EINVAL if given a zero size, but a
* segment with zero filesize is perfectly valid */
if (!size)
return addr;
/*
* total_size is the size of the ELF (interpreter) image.
* The _first_ mmap needs to know the full size, otherwise
* randomization might put this image into an overlapping
* position with the ELF binary image. (since size < total_size)
* So we first map the 'big' image - and unmap the remainder at
* the end. (which unmap is needed for ELF images with holes.)
*/
if (total_size) {
total_size = ELF_PAGEALIGN(total_size);
map_addr = vm_mmap(filep, addr, total_size, prot, type, off);
if (!BAD_ADDR(map_addr))
vm_munmap(map_addr+size, total_size-size);
} else
map_addr = vm_mmap(filep, addr, size, prot, type, off);
return(map_addr);
}
#endif /* !elf_map */
static unsigned long total_mapping_size(struct elf_phdr *cmds, int nr)
{
int i, first_idx = -1, last_idx = -1;
for (i = 0; i < nr; i++) {
if (cmds[i].p_type == PT_LOAD) {
last_idx = i;
if (first_idx == -1)
first_idx = i;
}
}
if (first_idx == -1)
return 0;
return cmds[last_idx].p_vaddr + cmds[last_idx].p_memsz -
ELF_PAGESTART(cmds[first_idx].p_vaddr);
}
/* This is much more generalized than the library routine read function,
so we keep this separate. Technically the library read function
is only provided so that we can read a.out libraries that have
an ELF header */
static unsigned long load_elf_interp(struct elfhdr *interp_elf_ex,
struct file *interpreter, unsigned long *interp_map_addr,
unsigned long no_base)
{
struct elf_phdr *elf_phdata;
struct elf_phdr *eppnt;
unsigned long load_addr = 0;
int load_addr_set = 0;
unsigned long last_bss = 0, elf_bss = 0;
unsigned long error = ~0UL;
unsigned long total_size;
int retval, i, size;
/* First of all, some simple consistency checks */
if (interp_elf_ex->e_type != ET_EXEC &&
interp_elf_ex->e_type != ET_DYN)
goto out;
if (!elf_check_arch(interp_elf_ex))
goto out;
if (!interpreter->f_op || !interpreter->f_op->mmap)
goto out;
/*
* If the size of this structure has changed, then punt, since
* we will be doing the wrong thing.
*/
if (interp_elf_ex->e_phentsize != sizeof(struct elf_phdr))
goto out;
if (interp_elf_ex->e_phnum < 1 ||
interp_elf_ex->e_phnum > 65536U / sizeof(struct elf_phdr))
goto out;
/* Now read in all of the header information */
size = sizeof(struct elf_phdr) * interp_elf_ex->e_phnum;
if (size > ELF_MIN_ALIGN)
goto out;
elf_phdata = kmalloc(size, GFP_KERNEL);
if (!elf_phdata)
goto out;
retval = kernel_read(interpreter, interp_elf_ex->e_phoff,
(char *)elf_phdata, size);
error = -EIO;
if (retval != size) {
if (retval < 0)
error = retval;
goto out_close;
}
total_size = total_mapping_size(elf_phdata, interp_elf_ex->e_phnum);
if (!total_size) {
error = -EINVAL;
goto out_close;
}
eppnt = elf_phdata;
for (i = 0; i < interp_elf_ex->e_phnum; i++, eppnt++) {
if (eppnt->p_type == PT_LOAD) {
int elf_type = MAP_PRIVATE | MAP_DENYWRITE;
int elf_prot = 0;
unsigned long vaddr = 0;
unsigned long k, map_addr;
if (eppnt->p_flags & PF_R)
elf_prot = PROT_READ;
if (eppnt->p_flags & PF_W)
elf_prot |= PROT_WRITE;
if (eppnt->p_flags & PF_X)
elf_prot |= PROT_EXEC;
vaddr = eppnt->p_vaddr;
if (interp_elf_ex->e_type == ET_EXEC || load_addr_set)
elf_type |= MAP_FIXED;
else if (no_base && interp_elf_ex->e_type == ET_DYN)
load_addr = -vaddr;
map_addr = elf_map(interpreter, load_addr + vaddr,
x86: PIE executable randomization, checkpatch fixes #39: FILE: arch/ia64/ia32/binfmt_elf32.c:229: +elf32_map (struct file *filep, unsigned long addr, struct elf_phdr *eppnt, int prot, int type, unsigned long unused) WARNING: no space between function name and open parenthesis '(' #39: FILE: arch/ia64/ia32/binfmt_elf32.c:229: +elf32_map (struct file *filep, unsigned long addr, struct elf_phdr *eppnt, int prot, int type, unsigned long unused) WARNING: line over 80 characters #67: FILE: arch/x86/kernel/sys_x86_64.c:80: + new_begin = randomize_range(*begin, *begin + 0x02000000, 0); ERROR: use tabs not spaces #110: FILE: arch/x86/kernel/sys_x86_64.c:185: + ^I mm->cached_hole_size = 0;$ ERROR: use tabs not spaces #111: FILE: arch/x86/kernel/sys_x86_64.c:186: + ^I^Imm->free_area_cache = mm->mmap_base;$ ERROR: use tabs not spaces #112: FILE: arch/x86/kernel/sys_x86_64.c:187: + ^I}$ ERROR: use tabs not spaces #141: FILE: arch/x86/kernel/sys_x86_64.c:216: + ^I^I/* remember the largest hole we saw so far */$ ERROR: use tabs not spaces #142: FILE: arch/x86/kernel/sys_x86_64.c:217: + ^I^Iif (addr + mm->cached_hole_size < vma->vm_start)$ ERROR: use tabs not spaces #143: FILE: arch/x86/kernel/sys_x86_64.c:218: + ^I^I mm->cached_hole_size = vma->vm_start - addr;$ ERROR: use tabs not spaces #157: FILE: arch/x86/kernel/sys_x86_64.c:232: + ^Imm->free_area_cache = TASK_UNMAPPED_BASE;$ ERROR: need a space before the open parenthesis '(' #291: FILE: arch/x86/mm/mmap_64.c:101: + } else if(mmap_is_legacy()) { WARNING: braces {} are not necessary for single statement blocks #302: FILE: arch/x86/mm/mmap_64.c:112: + if (current->flags & PF_RANDOMIZE) { + mm->mmap_base += ((long)rnd) << PAGE_SHIFT; + } WARNING: line over 80 characters #314: FILE: fs/binfmt_elf.c:48: +static unsigned long elf_map (struct file *, unsigned long, struct elf_phdr *, int, int, unsigned long); WARNING: no space between function name and open parenthesis '(' #314: FILE: fs/binfmt_elf.c:48: +static unsigned long elf_map (struct file *, unsigned long, struct elf_phdr *, int, int, unsigned long); WARNING: line over 80 characters #429: FILE: fs/binfmt_elf.c:438: + eppnt, elf_prot, elf_type, total_size); ERROR: need space after that ',' (ctx:VxV) #480: FILE: fs/binfmt_elf.c:939: + elf_prot, elf_flags,0); ^ total: 9 errors, 7 warnings, 461 lines checked Your patch has style problems, please review. If any of these errors are false positives report them to the maintainer, see CHECKPATCH in MAINTAINERS. Please run checkpatch prior to sending patches Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Arjan van de Ven <arjan@infradead.org> Cc: Jakub Jelinek <jakub@redhat.com> Cc: Jiri Kosina <jkosina@suse.cz> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Roland McGrath <roland@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 20:31:07 +08:00
eppnt, elf_prot, elf_type, total_size);
total_size = 0;
if (!*interp_map_addr)
*interp_map_addr = map_addr;
error = map_addr;
if (BAD_ADDR(map_addr))
goto out_close;
if (!load_addr_set &&
interp_elf_ex->e_type == ET_DYN) {
load_addr = map_addr - ELF_PAGESTART(vaddr);
load_addr_set = 1;
}
/*
* Check to see if the section's size will overflow the
* allowed task size. Note that p_filesz must always be
* <= p_memsize so it's only necessary to check p_memsz.
*/
k = load_addr + eppnt->p_vaddr;
[PATCH] binfmt_elf: fix checks for bad address Fix check for bad address; use macro instead of open-coding two checks. Taken from RHEL4 kernel update. From: Ernie Petrides <petrides@redhat.com> For background, the BAD_ADDR() macro should return TRUE if the address is TASK_SIZE, because that's the lowest address that is *not* valid for user-space mappings. The macro was correct in binfmt_aout.c but was wrong for the "equal to" case in binfmt_elf.c. There were two in-line validations of user-space addresses in binfmt_elf.c, which have been appropriately converted to use the corrected BAD_ADDR() macro in the patch you posted yesterday. Note that the size checks against TASK_SIZE are okay as coded. The additional changes that I propose are below. These are in the error paths for bad ELF entry addresses once load_elf_binary() has already committed to exec'ing the new image (following the tearing down of the task's original address space). The 1st hunk deals with the interp-side of the outer "if". There were two problems here. The printk() should be removed because this path can be triggered at will by a bogus interpreter image created and used by a malicious user. Further, the error code should not be ENOEXEC, because that causes the loop in search_binary_handler() to continue trying other exec handlers (twice, in fact). But it's too late for this to work correctly, because the user address space has already been torn down, and an exec() failure cannot be returned to the user code because the code no longer exists. The only recovery is to force a SIGSEGV, but it's best to terminate the search loop immediately. I somewhat arbitrarily chose EINVAL as a fallback error code, but any error returned by load_elf_interp() will override that (but this value will never be seen by user-space). The 2nd hunk deals with the non-interp-side of the outer "if". There were two problems here as well. The SIGSEGV needs to be forced, because a prior sigaction() syscall might have set the associated disposition to SIG_IGN. And the ENOEXEC should be changed to EINVAL as described above. Signed-off-by: Chuck Ebbert <76306.1226@compuserve.com> Signed-off-by: Ernie Petrides <petrides@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-07-03 15:24:14 +08:00
if (BAD_ADDR(k) ||
eppnt->p_filesz > eppnt->p_memsz ||
eppnt->p_memsz > TASK_SIZE ||
TASK_SIZE - eppnt->p_memsz < k) {
error = -ENOMEM;
goto out_close;
}
/*
* Find the end of the file mapping for this phdr, and
* keep track of the largest address we see for this.
*/
k = load_addr + eppnt->p_vaddr + eppnt->p_filesz;
if (k > elf_bss)
elf_bss = k;
/*
* Do the same thing for the memory mapping - between
* elf_bss and last_bss is the bss section.
*/
k = load_addr + eppnt->p_memsz + eppnt->p_vaddr;
if (k > last_bss)
last_bss = k;
}
}
if (last_bss > elf_bss) {
/*
* Now fill out the bss section. First pad the last page up
* to the page boundary, and then perform a mmap to make sure
* that there are zero-mapped pages up to and including the
* last bss page.
*/
if (padzero(elf_bss)) {
error = -EFAULT;
goto out_close;
}
/* What we have mapped so far */
elf_bss = ELF_PAGESTART(elf_bss + ELF_MIN_ALIGN - 1);
/* Map the last of the bss segment */
error = vm_brk(elf_bss, last_bss - elf_bss);
if (BAD_ADDR(error))
goto out_close;
}
error = load_addr;
out_close:
kfree(elf_phdata);
out:
return error;
}
/*
* These are the functions used to load ELF style executables and shared
* libraries. There is no binary dependent code anywhere else.
*/
#define INTERPRETER_NONE 0
#define INTERPRETER_ELF 2
#ifndef STACK_RND_MASK
#define STACK_RND_MASK (0x7ff >> (PAGE_SHIFT - 12)) /* 8MB of VA */
#endif
static unsigned long randomize_stack_top(unsigned long stack_top)
{
unsigned int random_variable = 0;
if ((current->flags & PF_RANDOMIZE) &&
!(current->personality & ADDR_NO_RANDOMIZE)) {
random_variable = get_random_int() & STACK_RND_MASK;
random_variable <<= PAGE_SHIFT;
}
#ifdef CONFIG_STACK_GROWSUP
return PAGE_ALIGN(stack_top) + random_variable;
#else
return PAGE_ALIGN(stack_top) - random_variable;
#endif
}
static int load_elf_binary(struct linux_binprm *bprm)
{
struct file *interpreter = NULL; /* to shut gcc up */
unsigned long load_addr = 0, load_bias = 0;
int load_addr_set = 0;
char * elf_interpreter = NULL;
unsigned long error;
struct elf_phdr *elf_ppnt, *elf_phdata;
unsigned long elf_bss, elf_brk;
int retval, i;
unsigned int size;
unsigned long elf_entry;
unsigned long interp_load_addr = 0;
unsigned long start_code, end_code, start_data, end_data;
unsigned long reloc_func_desc __maybe_unused = 0;
int executable_stack = EXSTACK_DEFAULT;
unsigned long def_flags = 0;
struct pt_regs *regs = current_pt_regs();
struct {
struct elfhdr elf_ex;
struct elfhdr interp_elf_ex;
} *loc;
loc = kmalloc(sizeof(*loc), GFP_KERNEL);
if (!loc) {
retval = -ENOMEM;
goto out_ret;
}
/* Get the exec-header */
loc->elf_ex = *((struct elfhdr *)bprm->buf);
retval = -ENOEXEC;
/* First of all, some simple consistency checks */
if (memcmp(loc->elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
goto out;
if (loc->elf_ex.e_type != ET_EXEC && loc->elf_ex.e_type != ET_DYN)
goto out;
if (!elf_check_arch(&loc->elf_ex))
goto out;
if (!bprm->file->f_op || !bprm->file->f_op->mmap)
goto out;
/* Now read in all of the header information */
if (loc->elf_ex.e_phentsize != sizeof(struct elf_phdr))
goto out;
if (loc->elf_ex.e_phnum < 1 ||
loc->elf_ex.e_phnum > 65536U / sizeof(struct elf_phdr))
goto out;
size = loc->elf_ex.e_phnum * sizeof(struct elf_phdr);
retval = -ENOMEM;
elf_phdata = kmalloc(size, GFP_KERNEL);
if (!elf_phdata)
goto out;
retval = kernel_read(bprm->file, loc->elf_ex.e_phoff,
(char *)elf_phdata, size);
if (retval != size) {
if (retval >= 0)
retval = -EIO;
goto out_free_ph;
}
elf_ppnt = elf_phdata;
elf_bss = 0;
elf_brk = 0;
start_code = ~0UL;
end_code = 0;
start_data = 0;
end_data = 0;
for (i = 0; i < loc->elf_ex.e_phnum; i++) {
if (elf_ppnt->p_type == PT_INTERP) {
/* This is the program interpreter used for
* shared libraries - for now assume that this
* is an a.out format binary
*/
retval = -ENOEXEC;
if (elf_ppnt->p_filesz > PATH_MAX ||
elf_ppnt->p_filesz < 2)
goto out_free_ph;
retval = -ENOMEM;
elf_interpreter = kmalloc(elf_ppnt->p_filesz,
GFP_KERNEL);
if (!elf_interpreter)
goto out_free_ph;
retval = kernel_read(bprm->file, elf_ppnt->p_offset,
elf_interpreter,
elf_ppnt->p_filesz);
if (retval != elf_ppnt->p_filesz) {
if (retval >= 0)
retval = -EIO;
goto out_free_interp;
}
/* make sure path is NULL terminated */
retval = -ENOEXEC;
if (elf_interpreter[elf_ppnt->p_filesz - 1] != '\0')
goto out_free_interp;
interpreter = open_exec(elf_interpreter);
retval = PTR_ERR(interpreter);
if (IS_ERR(interpreter))
goto out_free_interp;
/*
* If the binary is not readable then enforce
* mm->dumpable = 0 regardless of the interpreter's
* permissions.
*/
would_dump(bprm, interpreter);
retval = kernel_read(interpreter, 0, bprm->buf,
BINPRM_BUF_SIZE);
if (retval != BINPRM_BUF_SIZE) {
if (retval >= 0)
retval = -EIO;
goto out_free_dentry;
}
/* Get the exec headers */
loc->interp_elf_ex = *((struct elfhdr *)bprm->buf);
break;
}
elf_ppnt++;
}
elf_ppnt = elf_phdata;
for (i = 0; i < loc->elf_ex.e_phnum; i++, elf_ppnt++)
if (elf_ppnt->p_type == PT_GNU_STACK) {
if (elf_ppnt->p_flags & PF_X)
executable_stack = EXSTACK_ENABLE_X;
else
executable_stack = EXSTACK_DISABLE_X;
break;
}
/* Some simple consistency checks for the interpreter */
if (elf_interpreter) {
retval = -ELIBBAD;
/* Not an ELF interpreter */
if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
goto out_free_dentry;
/* Verify the interpreter has a valid arch */
if (!elf_check_arch(&loc->interp_elf_ex))
goto out_free_dentry;
}
/* Flush all traces of the currently running executable */
retval = flush_old_exec(bprm);
if (retval)
goto out_free_dentry;
/* OK, This is the point of no return */
current->mm->def_flags = def_flags;
/* Do this immediately, since STACK_TOP as used in setup_arg_pages
may depend on the personality. */
SET_PERSONALITY(loc->elf_ex);
if (elf_read_implies_exec(loc->elf_ex, executable_stack))
current->personality |= READ_IMPLIES_EXEC;
if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
current->flags |= PF_RANDOMIZE;
setup_new_exec(bprm);
/* Do this so that we can load the interpreter, if need be. We will
change some of these later */
current->mm->free_area_cache = current->mm->mmap_base;
[PATCH] Avoiding mmap fragmentation Ingo recently introduced a great speedup for allocating new mmaps using the free_area_cache pointer which boosts the specweb SSL benchmark by 4-5% and causes huge performance increases in thread creation. The downside of this patch is that it does lead to fragmentation in the mmap-ed areas (visible via /proc/self/maps), such that some applications that work fine under 2.4 kernels quickly run out of memory on any 2.6 kernel. The problem is twofold: 1) the free_area_cache is used to continue a search for memory where the last search ended. Before the change new areas were always searched from the base address on. So now new small areas are cluttering holes of all sizes throughout the whole mmap-able region whereas before small holes tended to close holes near the base leaving holes far from the base large and available for larger requests. 2) the free_area_cache also is set to the location of the last munmap-ed area so in scenarios where we allocate e.g. five regions of 1K each, then free regions 4 2 3 in this order the next request for 1K will be placed in the position of the old region 3, whereas before we appended it to the still active region 1, placing it at the location of the old region 2. Before we had 1 free region of 2K, now we only get two free regions of 1K -> fragmentation. The patch addresses thes issues by introducing yet another cache descriptor cached_hole_size that contains the largest known hole size below the current free_area_cache. If a new request comes in the size is compared against the cached_hole_size and if the request can be filled with a hole below free_area_cache the search is started from the base instead. The results look promising: Whereas 2.6.12-rc4 fragments quickly and my (earlier posted) leakme.c test program terminates after 50000+ iterations with 96 distinct and fragmented maps in /proc/self/maps it performs nicely (as expected) with thread creation, Ingo's test_str02 with 20000 threads requires 0.7s system time. Taking out Ingo's patch (un-patch available per request) by basically deleting all mentions of free_area_cache from the kernel and starting the search for new memory always at the respective bases we observe: leakme terminates successfully with 11 distinctive hardly fragmented areas in /proc/self/maps but thread creating is gringdingly slow: 30+s(!) system time for Ingo's test_str02 with 20000 threads. Now - drumroll ;-) the appended patch works fine with leakme: it ends with only 7 distinct areas in /proc/self/maps and also thread creation seems sufficiently fast with 0.71s for 20000 threads. Signed-off-by: Wolfgang Wander <wwc@rentec.com> Credit-to: "Richard Purdie" <rpurdie@rpsys.net> Signed-off-by: Ken Chen <kenneth.w.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> (partly) Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2005-06-22 08:14:49 +08:00
current->mm->cached_hole_size = 0;
retval = setup_arg_pages(bprm, randomize_stack_top(STACK_TOP),
executable_stack);
if (retval < 0) {
send_sig(SIGKILL, current, 0);
goto out_free_dentry;
}
current->mm->start_stack = bprm->p;
/* Now we do a little grungy work by mmapping the ELF image into
the correct location in memory. */
for(i = 0, elf_ppnt = elf_phdata;
i < loc->elf_ex.e_phnum; i++, elf_ppnt++) {
int elf_prot = 0, elf_flags;
unsigned long k, vaddr;
if (elf_ppnt->p_type != PT_LOAD)
continue;
if (unlikely (elf_brk > elf_bss)) {
unsigned long nbyte;
/* There was a PT_LOAD segment with p_memsz > p_filesz
before this one. Map anonymous pages, if needed,
and clear the area. */
retval = set_brk(elf_bss + load_bias,
elf_brk + load_bias);
if (retval) {
send_sig(SIGKILL, current, 0);
goto out_free_dentry;
}
nbyte = ELF_PAGEOFFSET(elf_bss);
if (nbyte) {
nbyte = ELF_MIN_ALIGN - nbyte;
if (nbyte > elf_brk - elf_bss)
nbyte = elf_brk - elf_bss;
if (clear_user((void __user *)elf_bss +
load_bias, nbyte)) {
/*
* This bss-zeroing can fail if the ELF
* file specifies odd protections. So
* we don't check the return value
*/
}
}
}
if (elf_ppnt->p_flags & PF_R)
elf_prot |= PROT_READ;
if (elf_ppnt->p_flags & PF_W)
elf_prot |= PROT_WRITE;
if (elf_ppnt->p_flags & PF_X)
elf_prot |= PROT_EXEC;
elf_flags = MAP_PRIVATE | MAP_DENYWRITE | MAP_EXECUTABLE;
vaddr = elf_ppnt->p_vaddr;
if (loc->elf_ex.e_type == ET_EXEC || load_addr_set) {
elf_flags |= MAP_FIXED;
} else if (loc->elf_ex.e_type == ET_DYN) {
/* Try and get dynamic programs out of the way of the
* default mmap base, as well as whatever program they
* might try to exec. This is because the brk will
* follow the loader, and is not movable. */
#ifdef CONFIG_ARCH_BINFMT_ELF_RANDOMIZE_PIE
/* Memory randomization might have been switched off
* in runtime via sysctl or explicit setting of
* personality flags.
* If that is the case, retain the original non-zero
* load_bias value in order to establish proper
* non-randomized mappings.
*/
if (current->flags & PF_RANDOMIZE)
load_bias = 0;
else
load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
#else
load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
#endif
}
error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
x86: PIE executable randomization, checkpatch fixes #39: FILE: arch/ia64/ia32/binfmt_elf32.c:229: +elf32_map (struct file *filep, unsigned long addr, struct elf_phdr *eppnt, int prot, int type, unsigned long unused) WARNING: no space between function name and open parenthesis '(' #39: FILE: arch/ia64/ia32/binfmt_elf32.c:229: +elf32_map (struct file *filep, unsigned long addr, struct elf_phdr *eppnt, int prot, int type, unsigned long unused) WARNING: line over 80 characters #67: FILE: arch/x86/kernel/sys_x86_64.c:80: + new_begin = randomize_range(*begin, *begin + 0x02000000, 0); ERROR: use tabs not spaces #110: FILE: arch/x86/kernel/sys_x86_64.c:185: + ^I mm->cached_hole_size = 0;$ ERROR: use tabs not spaces #111: FILE: arch/x86/kernel/sys_x86_64.c:186: + ^I^Imm->free_area_cache = mm->mmap_base;$ ERROR: use tabs not spaces #112: FILE: arch/x86/kernel/sys_x86_64.c:187: + ^I}$ ERROR: use tabs not spaces #141: FILE: arch/x86/kernel/sys_x86_64.c:216: + ^I^I/* remember the largest hole we saw so far */$ ERROR: use tabs not spaces #142: FILE: arch/x86/kernel/sys_x86_64.c:217: + ^I^Iif (addr + mm->cached_hole_size < vma->vm_start)$ ERROR: use tabs not spaces #143: FILE: arch/x86/kernel/sys_x86_64.c:218: + ^I^I mm->cached_hole_size = vma->vm_start - addr;$ ERROR: use tabs not spaces #157: FILE: arch/x86/kernel/sys_x86_64.c:232: + ^Imm->free_area_cache = TASK_UNMAPPED_BASE;$ ERROR: need a space before the open parenthesis '(' #291: FILE: arch/x86/mm/mmap_64.c:101: + } else if(mmap_is_legacy()) { WARNING: braces {} are not necessary for single statement blocks #302: FILE: arch/x86/mm/mmap_64.c:112: + if (current->flags & PF_RANDOMIZE) { + mm->mmap_base += ((long)rnd) << PAGE_SHIFT; + } WARNING: line over 80 characters #314: FILE: fs/binfmt_elf.c:48: +static unsigned long elf_map (struct file *, unsigned long, struct elf_phdr *, int, int, unsigned long); WARNING: no space between function name and open parenthesis '(' #314: FILE: fs/binfmt_elf.c:48: +static unsigned long elf_map (struct file *, unsigned long, struct elf_phdr *, int, int, unsigned long); WARNING: line over 80 characters #429: FILE: fs/binfmt_elf.c:438: + eppnt, elf_prot, elf_type, total_size); ERROR: need space after that ',' (ctx:VxV) #480: FILE: fs/binfmt_elf.c:939: + elf_prot, elf_flags,0); ^ total: 9 errors, 7 warnings, 461 lines checked Your patch has style problems, please review. If any of these errors are false positives report them to the maintainer, see CHECKPATCH in MAINTAINERS. Please run checkpatch prior to sending patches Cc: "Luck, Tony" <tony.luck@intel.com> Cc: Arjan van de Ven <arjan@infradead.org> Cc: Jakub Jelinek <jakub@redhat.com> Cc: Jiri Kosina <jkosina@suse.cz> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Roland McGrath <roland@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 20:31:07 +08:00
elf_prot, elf_flags, 0);
if (BAD_ADDR(error)) {
send_sig(SIGKILL, current, 0);
retval = IS_ERR((void *)error) ?
PTR_ERR((void*)error) : -EINVAL;
goto out_free_dentry;
}
if (!load_addr_set) {
load_addr_set = 1;
load_addr = (elf_ppnt->p_vaddr - elf_ppnt->p_offset);
if (loc->elf_ex.e_type == ET_DYN) {
load_bias += error -
ELF_PAGESTART(load_bias + vaddr);
load_addr += load_bias;
reloc_func_desc = load_bias;
}
}
k = elf_ppnt->p_vaddr;
if (k < start_code)
start_code = k;
if (start_data < k)
start_data = k;
/*
* Check to see if the section's size will overflow the
* allowed task size. Note that p_filesz must always be
* <= p_memsz so it is only necessary to check p_memsz.
*/
[PATCH] binfmt_elf: fix checks for bad address Fix check for bad address; use macro instead of open-coding two checks. Taken from RHEL4 kernel update. From: Ernie Petrides <petrides@redhat.com> For background, the BAD_ADDR() macro should return TRUE if the address is TASK_SIZE, because that's the lowest address that is *not* valid for user-space mappings. The macro was correct in binfmt_aout.c but was wrong for the "equal to" case in binfmt_elf.c. There were two in-line validations of user-space addresses in binfmt_elf.c, which have been appropriately converted to use the corrected BAD_ADDR() macro in the patch you posted yesterday. Note that the size checks against TASK_SIZE are okay as coded. The additional changes that I propose are below. These are in the error paths for bad ELF entry addresses once load_elf_binary() has already committed to exec'ing the new image (following the tearing down of the task's original address space). The 1st hunk deals with the interp-side of the outer "if". There were two problems here. The printk() should be removed because this path can be triggered at will by a bogus interpreter image created and used by a malicious user. Further, the error code should not be ENOEXEC, because that causes the loop in search_binary_handler() to continue trying other exec handlers (twice, in fact). But it's too late for this to work correctly, because the user address space has already been torn down, and an exec() failure cannot be returned to the user code because the code no longer exists. The only recovery is to force a SIGSEGV, but it's best to terminate the search loop immediately. I somewhat arbitrarily chose EINVAL as a fallback error code, but any error returned by load_elf_interp() will override that (but this value will never be seen by user-space). The 2nd hunk deals with the non-interp-side of the outer "if". There were two problems here as well. The SIGSEGV needs to be forced, because a prior sigaction() syscall might have set the associated disposition to SIG_IGN. And the ENOEXEC should be changed to EINVAL as described above. Signed-off-by: Chuck Ebbert <76306.1226@compuserve.com> Signed-off-by: Ernie Petrides <petrides@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-07-03 15:24:14 +08:00
if (BAD_ADDR(k) || elf_ppnt->p_filesz > elf_ppnt->p_memsz ||
elf_ppnt->p_memsz > TASK_SIZE ||
TASK_SIZE - elf_ppnt->p_memsz < k) {
/* set_brk can never work. Avoid overflows. */
send_sig(SIGKILL, current, 0);
retval = -EINVAL;
goto out_free_dentry;
}
k = elf_ppnt->p_vaddr + elf_ppnt->p_filesz;
if (k > elf_bss)
elf_bss = k;
if ((elf_ppnt->p_flags & PF_X) && end_code < k)
end_code = k;
if (end_data < k)
end_data = k;
k = elf_ppnt->p_vaddr + elf_ppnt->p_memsz;
if (k > elf_brk)
elf_brk = k;
}
loc->elf_ex.e_entry += load_bias;
elf_bss += load_bias;
elf_brk += load_bias;
start_code += load_bias;
end_code += load_bias;
start_data += load_bias;
end_data += load_bias;
/* Calling set_brk effectively mmaps the pages that we need
* for the bss and break sections. We must do this before
* mapping in the interpreter, to make sure it doesn't wind
* up getting placed where the bss needs to go.
*/
retval = set_brk(elf_bss, elf_brk);
if (retval) {
send_sig(SIGKILL, current, 0);
goto out_free_dentry;
}
if (likely(elf_bss != elf_brk) && unlikely(padzero(elf_bss))) {
send_sig(SIGSEGV, current, 0);
retval = -EFAULT; /* Nobody gets to see this, but.. */
goto out_free_dentry;
}
if (elf_interpreter) {
unsigned long interp_map_addr = 0;
elf_entry = load_elf_interp(&loc->interp_elf_ex,
interpreter,
&interp_map_addr,
load_bias);
if (!IS_ERR((void *)elf_entry)) {
/*
* load_elf_interp() returns relocation
* adjustment
*/
interp_load_addr = elf_entry;
elf_entry += loc->interp_elf_ex.e_entry;
}
if (BAD_ADDR(elf_entry)) {
force_sig(SIGSEGV, current);
[PATCH] binfmt_elf: fix checks for bad address Fix check for bad address; use macro instead of open-coding two checks. Taken from RHEL4 kernel update. From: Ernie Petrides <petrides@redhat.com> For background, the BAD_ADDR() macro should return TRUE if the address is TASK_SIZE, because that's the lowest address that is *not* valid for user-space mappings. The macro was correct in binfmt_aout.c but was wrong for the "equal to" case in binfmt_elf.c. There were two in-line validations of user-space addresses in binfmt_elf.c, which have been appropriately converted to use the corrected BAD_ADDR() macro in the patch you posted yesterday. Note that the size checks against TASK_SIZE are okay as coded. The additional changes that I propose are below. These are in the error paths for bad ELF entry addresses once load_elf_binary() has already committed to exec'ing the new image (following the tearing down of the task's original address space). The 1st hunk deals with the interp-side of the outer "if". There were two problems here. The printk() should be removed because this path can be triggered at will by a bogus interpreter image created and used by a malicious user. Further, the error code should not be ENOEXEC, because that causes the loop in search_binary_handler() to continue trying other exec handlers (twice, in fact). But it's too late for this to work correctly, because the user address space has already been torn down, and an exec() failure cannot be returned to the user code because the code no longer exists. The only recovery is to force a SIGSEGV, but it's best to terminate the search loop immediately. I somewhat arbitrarily chose EINVAL as a fallback error code, but any error returned by load_elf_interp() will override that (but this value will never be seen by user-space). The 2nd hunk deals with the non-interp-side of the outer "if". There were two problems here as well. The SIGSEGV needs to be forced, because a prior sigaction() syscall might have set the associated disposition to SIG_IGN. And the ENOEXEC should be changed to EINVAL as described above. Signed-off-by: Chuck Ebbert <76306.1226@compuserve.com> Signed-off-by: Ernie Petrides <petrides@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-07-03 15:24:14 +08:00
retval = IS_ERR((void *)elf_entry) ?
(int)elf_entry : -EINVAL;
goto out_free_dentry;
}
reloc_func_desc = interp_load_addr;
allow_write_access(interpreter);
fput(interpreter);
kfree(elf_interpreter);
} else {
elf_entry = loc->elf_ex.e_entry;
if (BAD_ADDR(elf_entry)) {
[PATCH] binfmt_elf: fix checks for bad address Fix check for bad address; use macro instead of open-coding two checks. Taken from RHEL4 kernel update. From: Ernie Petrides <petrides@redhat.com> For background, the BAD_ADDR() macro should return TRUE if the address is TASK_SIZE, because that's the lowest address that is *not* valid for user-space mappings. The macro was correct in binfmt_aout.c but was wrong for the "equal to" case in binfmt_elf.c. There were two in-line validations of user-space addresses in binfmt_elf.c, which have been appropriately converted to use the corrected BAD_ADDR() macro in the patch you posted yesterday. Note that the size checks against TASK_SIZE are okay as coded. The additional changes that I propose are below. These are in the error paths for bad ELF entry addresses once load_elf_binary() has already committed to exec'ing the new image (following the tearing down of the task's original address space). The 1st hunk deals with the interp-side of the outer "if". There were two problems here. The printk() should be removed because this path can be triggered at will by a bogus interpreter image created and used by a malicious user. Further, the error code should not be ENOEXEC, because that causes the loop in search_binary_handler() to continue trying other exec handlers (twice, in fact). But it's too late for this to work correctly, because the user address space has already been torn down, and an exec() failure cannot be returned to the user code because the code no longer exists. The only recovery is to force a SIGSEGV, but it's best to terminate the search loop immediately. I somewhat arbitrarily chose EINVAL as a fallback error code, but any error returned by load_elf_interp() will override that (but this value will never be seen by user-space). The 2nd hunk deals with the non-interp-side of the outer "if". There were two problems here as well. The SIGSEGV needs to be forced, because a prior sigaction() syscall might have set the associated disposition to SIG_IGN. And the ENOEXEC should be changed to EINVAL as described above. Signed-off-by: Chuck Ebbert <76306.1226@compuserve.com> Signed-off-by: Ernie Petrides <petrides@redhat.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-07-03 15:24:14 +08:00
force_sig(SIGSEGV, current);
retval = -EINVAL;
goto out_free_dentry;
}
}
kfree(elf_phdata);
set_binfmt(&elf_format);
#ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
retval = arch_setup_additional_pages(bprm, !!elf_interpreter);
if (retval < 0) {
send_sig(SIGKILL, current, 0);
goto out;
}
#endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
CRED: Make execve() take advantage of copy-on-write credentials Make execve() take advantage of copy-on-write credentials, allowing it to set up the credentials in advance, and then commit the whole lot after the point of no return. This patch and the preceding patches have been tested with the LTP SELinux testsuite. This patch makes several logical sets of alteration: (1) execve(). The credential bits from struct linux_binprm are, for the most part, replaced with a single credentials pointer (bprm->cred). This means that all the creds can be calculated in advance and then applied at the point of no return with no possibility of failure. I would like to replace bprm->cap_effective with: cap_isclear(bprm->cap_effective) but this seems impossible due to special behaviour for processes of pid 1 (they always retain their parent's capability masks where normally they'd be changed - see cap_bprm_set_creds()). The following sequence of events now happens: (a) At the start of do_execve, the current task's cred_exec_mutex is locked to prevent PTRACE_ATTACH from obsoleting the calculation of creds that we make. (a) prepare_exec_creds() is then called to make a copy of the current task's credentials and prepare it. This copy is then assigned to bprm->cred. This renders security_bprm_alloc() and security_bprm_free() unnecessary, and so they've been removed. (b) The determination of unsafe execution is now performed immediately after (a) rather than later on in the code. The result is stored in bprm->unsafe for future reference. (c) prepare_binprm() is called, possibly multiple times. (i) This applies the result of set[ug]id binaries to the new creds attached to bprm->cred. Personality bit clearance is recorded, but now deferred on the basis that the exec procedure may yet fail. (ii) This then calls the new security_bprm_set_creds(). This should calculate the new LSM and capability credentials into *bprm->cred. This folds together security_bprm_set() and parts of security_bprm_apply_creds() (these two have been removed). Anything that might fail must be done at this point. (iii) bprm->cred_prepared is set to 1. bprm->cred_prepared is 0 on the first pass of the security calculations, and 1 on all subsequent passes. This allows SELinux in (ii) to base its calculations only on the initial script and not on the interpreter. (d) flush_old_exec() is called to commit the task to execution. This performs the following steps with regard to credentials: (i) Clear pdeath_signal and set dumpable on certain circumstances that may not be covered by commit_creds(). (ii) Clear any bits in current->personality that were deferred from (c.i). (e) install_exec_creds() [compute_creds() as was] is called to install the new credentials. This performs the following steps with regard to credentials: (i) Calls security_bprm_committing_creds() to apply any security requirements, such as flushing unauthorised files in SELinux, that must be done before the credentials are changed. This is made up of bits of security_bprm_apply_creds() and security_bprm_post_apply_creds(), both of which have been removed. This function is not allowed to fail; anything that might fail must have been done in (c.ii). (ii) Calls commit_creds() to apply the new credentials in a single assignment (more or less). Possibly pdeath_signal and dumpable should be part of struct creds. (iii) Unlocks the task's cred_replace_mutex, thus allowing PTRACE_ATTACH to take place. (iv) Clears The bprm->cred pointer as the credentials it was holding are now immutable. (v) Calls security_bprm_committed_creds() to apply any security alterations that must be done after the creds have been changed. SELinux uses this to flush signals and signal handlers. (f) If an error occurs before (d.i), bprm_free() will call abort_creds() to destroy the proposed new credentials and will then unlock cred_replace_mutex. No changes to the credentials will have been made. (2) LSM interface. A number of functions have been changed, added or removed: (*) security_bprm_alloc(), ->bprm_alloc_security() (*) security_bprm_free(), ->bprm_free_security() Removed in favour of preparing new credentials and modifying those. (*) security_bprm_apply_creds(), ->bprm_apply_creds() (*) security_bprm_post_apply_creds(), ->bprm_post_apply_creds() Removed; split between security_bprm_set_creds(), security_bprm_committing_creds() and security_bprm_committed_creds(). (*) security_bprm_set(), ->bprm_set_security() Removed; folded into security_bprm_set_creds(). (*) security_bprm_set_creds(), ->bprm_set_creds() New. The new credentials in bprm->creds should be checked and set up as appropriate. bprm->cred_prepared is 0 on the first call, 1 on the second and subsequent calls. (*) security_bprm_committing_creds(), ->bprm_committing_creds() (*) security_bprm_committed_creds(), ->bprm_committed_creds() New. Apply the security effects of the new credentials. This includes closing unauthorised files in SELinux. This function may not fail. When the former is called, the creds haven't yet been applied to the process; when the latter is called, they have. The former may access bprm->cred, the latter may not. (3) SELinux. SELinux has a number of changes, in addition to those to support the LSM interface changes mentioned above: (a) The bprm_security_struct struct has been removed in favour of using the credentials-under-construction approach. (c) flush_unauthorized_files() now takes a cred pointer and passes it on to inode_has_perm(), file_has_perm() and dentry_open(). Signed-off-by: David Howells <dhowells@redhat.com> Acked-by: James Morris <jmorris@namei.org> Acked-by: Serge Hallyn <serue@us.ibm.com> Signed-off-by: James Morris <jmorris@namei.org>
2008-11-14 07:39:24 +08:00
install_exec_creds(bprm);
retval = create_elf_tables(bprm, &loc->elf_ex,
load_addr, interp_load_addr);
if (retval < 0) {
send_sig(SIGKILL, current, 0);
goto out;
}
/* N.B. passed_fileno might not be initialized? */
current->mm->end_code = end_code;
current->mm->start_code = start_code;
current->mm->start_data = start_data;
current->mm->end_data = end_data;
current->mm->start_stack = bprm->p;
x86: randomize brk Randomize the location of the heap (brk) for i386 and x86_64. The range is randomized in the range starting at current brk location up to 0x02000000 offset for both architectures. This, together with pie-executable-randomization.patch and pie-executable-randomization-fix.patch, should make the address space randomization on i386 and x86_64 complete. Arjan says: This is known to break older versions of some emacs variants, whose dumper code assumed that the last variable declared in the program is equal to the start of the dynamically allocated memory region. (The dumper is the code where emacs effectively dumps core at the end of it's compilation stage; this coredump is then loaded as the main program during normal use) iirc this was 5 years or so; we found this way back when I was at RH and we first did the security stuff there (including this brk randomization). It wasn't all variants of emacs, and it got fixed as a result (I vaguely remember that emacs already had code to deal with it for other archs/oses, just ifdeffed wrongly). It's a rare and wrong assumption as a general thing, just on x86 it mostly happened to be true (but to be honest, it'll break too if gcc does something fancy or if the linker does a non-standard order). Still its something we should at least document. Note 2: afaik it only broke the emacs *build*. I'm not 100% sure about that (it IS 5 years ago) though. [ akpm@linux-foundation.org: deuglification ] Signed-off-by: Jiri Kosina <jkosina@suse.cz> Cc: Arjan van de Ven <arjan@infradead.org> Cc: Roland McGrath <roland@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 20:30:40 +08:00
#ifdef arch_randomize_brk
if ((current->flags & PF_RANDOMIZE) && (randomize_va_space > 1)) {
x86: randomize brk Randomize the location of the heap (brk) for i386 and x86_64. The range is randomized in the range starting at current brk location up to 0x02000000 offset for both architectures. This, together with pie-executable-randomization.patch and pie-executable-randomization-fix.patch, should make the address space randomization on i386 and x86_64 complete. Arjan says: This is known to break older versions of some emacs variants, whose dumper code assumed that the last variable declared in the program is equal to the start of the dynamically allocated memory region. (The dumper is the code where emacs effectively dumps core at the end of it's compilation stage; this coredump is then loaded as the main program during normal use) iirc this was 5 years or so; we found this way back when I was at RH and we first did the security stuff there (including this brk randomization). It wasn't all variants of emacs, and it got fixed as a result (I vaguely remember that emacs already had code to deal with it for other archs/oses, just ifdeffed wrongly). It's a rare and wrong assumption as a general thing, just on x86 it mostly happened to be true (but to be honest, it'll break too if gcc does something fancy or if the linker does a non-standard order). Still its something we should at least document. Note 2: afaik it only broke the emacs *build*. I'm not 100% sure about that (it IS 5 years ago) though. [ akpm@linux-foundation.org: deuglification ] Signed-off-by: Jiri Kosina <jkosina@suse.cz> Cc: Arjan van de Ven <arjan@infradead.org> Cc: Roland McGrath <roland@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 20:30:40 +08:00
current->mm->brk = current->mm->start_brk =
arch_randomize_brk(current->mm);
#ifdef CONFIG_COMPAT_BRK
current->brk_randomized = 1;
#endif
}
x86: randomize brk Randomize the location of the heap (brk) for i386 and x86_64. The range is randomized in the range starting at current brk location up to 0x02000000 offset for both architectures. This, together with pie-executable-randomization.patch and pie-executable-randomization-fix.patch, should make the address space randomization on i386 and x86_64 complete. Arjan says: This is known to break older versions of some emacs variants, whose dumper code assumed that the last variable declared in the program is equal to the start of the dynamically allocated memory region. (The dumper is the code where emacs effectively dumps core at the end of it's compilation stage; this coredump is then loaded as the main program during normal use) iirc this was 5 years or so; we found this way back when I was at RH and we first did the security stuff there (including this brk randomization). It wasn't all variants of emacs, and it got fixed as a result (I vaguely remember that emacs already had code to deal with it for other archs/oses, just ifdeffed wrongly). It's a rare and wrong assumption as a general thing, just on x86 it mostly happened to be true (but to be honest, it'll break too if gcc does something fancy or if the linker does a non-standard order). Still its something we should at least document. Note 2: afaik it only broke the emacs *build*. I'm not 100% sure about that (it IS 5 years ago) though. [ akpm@linux-foundation.org: deuglification ] Signed-off-by: Jiri Kosina <jkosina@suse.cz> Cc: Arjan van de Ven <arjan@infradead.org> Cc: Roland McGrath <roland@redhat.com> Cc: Jakub Jelinek <jakub@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2008-01-30 20:30:40 +08:00
#endif
if (current->personality & MMAP_PAGE_ZERO) {
/* Why this, you ask??? Well SVr4 maps page 0 as read-only,
and some applications "depend" upon this behavior.
Since we do not have the power to recompile these, we
emulate the SVr4 behavior. Sigh. */
error = vm_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
MAP_FIXED | MAP_PRIVATE, 0);
}
#ifdef ELF_PLAT_INIT
/*
* The ABI may specify that certain registers be set up in special
* ways (on i386 %edx is the address of a DT_FINI function, for
* example. In addition, it may also specify (eg, PowerPC64 ELF)
* that the e_entry field is the address of the function descriptor
* for the startup routine, rather than the address of the startup
* routine itself. This macro performs whatever initialization to
* the regs structure is required as well as any relocations to the
* function descriptor entries when executing dynamically links apps.
*/
ELF_PLAT_INIT(regs, reloc_func_desc);
#endif
start_thread(regs, elf_entry, bprm->p);
retval = 0;
out:
kfree(loc);
out_ret:
return retval;
/* error cleanup */
out_free_dentry:
allow_write_access(interpreter);
if (interpreter)
fput(interpreter);
out_free_interp:
kfree(elf_interpreter);
out_free_ph:
kfree(elf_phdata);
goto out;
}
/* This is really simpleminded and specialized - we are loading an
a.out library that is given an ELF header. */
static int load_elf_library(struct file *file)
{
struct elf_phdr *elf_phdata;
struct elf_phdr *eppnt;
unsigned long elf_bss, bss, len;
int retval, error, i, j;
struct elfhdr elf_ex;
error = -ENOEXEC;
retval = kernel_read(file, 0, (char *)&elf_ex, sizeof(elf_ex));
if (retval != sizeof(elf_ex))
goto out;
if (memcmp(elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
goto out;
/* First of all, some simple consistency checks */
if (elf_ex.e_type != ET_EXEC || elf_ex.e_phnum > 2 ||
!elf_check_arch(&elf_ex) || !file->f_op || !file->f_op->mmap)
goto out;
/* Now read in all of the header information */
j = sizeof(struct elf_phdr) * elf_ex.e_phnum;
/* j < ELF_MIN_ALIGN because elf_ex.e_phnum <= 2 */
error = -ENOMEM;
elf_phdata = kmalloc(j, GFP_KERNEL);
if (!elf_phdata)
goto out;
eppnt = elf_phdata;
error = -ENOEXEC;
retval = kernel_read(file, elf_ex.e_phoff, (char *)eppnt, j);
if (retval != j)
goto out_free_ph;
for (j = 0, i = 0; i<elf_ex.e_phnum; i++)
if ((eppnt + i)->p_type == PT_LOAD)
j++;
if (j != 1)
goto out_free_ph;
while (eppnt->p_type != PT_LOAD)
eppnt++;
/* Now use mmap to map the library into memory. */
error = vm_mmap(file,
ELF_PAGESTART(eppnt->p_vaddr),
(eppnt->p_filesz +
ELF_PAGEOFFSET(eppnt->p_vaddr)),
PROT_READ | PROT_WRITE | PROT_EXEC,
MAP_FIXED | MAP_PRIVATE | MAP_DENYWRITE,
(eppnt->p_offset -
ELF_PAGEOFFSET(eppnt->p_vaddr)));
if (error != ELF_PAGESTART(eppnt->p_vaddr))
goto out_free_ph;
elf_bss = eppnt->p_vaddr + eppnt->p_filesz;
if (padzero(elf_bss)) {
error = -EFAULT;
goto out_free_ph;
}
len = ELF_PAGESTART(eppnt->p_filesz + eppnt->p_vaddr +
ELF_MIN_ALIGN - 1);
bss = eppnt->p_memsz + eppnt->p_vaddr;
if (bss > len)
vm_brk(len, bss - len);
error = 0;
out_free_ph:
kfree(elf_phdata);
out:
return error;
}
#ifdef CONFIG_ELF_CORE
/*
* ELF core dumper
*
* Modelled on fs/exec.c:aout_core_dump()
* Jeremy Fitzhardinge <jeremy@sw.oz.au>
*/
coredump: remove VM_ALWAYSDUMP flag The motivation for this patchset was that I was looking at a way for a qemu-kvm process, to exclude the guest memory from its core dump, which can be quite large. There are already a number of filter flags in /proc/<pid>/coredump_filter, however, these allow one to specify 'types' of kernel memory, not specific address ranges (which is needed in this case). Since there are no more vma flags available, the first patch eliminates the need for the 'VM_ALWAYSDUMP' flag. The flag is used internally by the kernel to mark vdso and vsyscall pages. However, it is simple enough to check if a vma covers a vdso or vsyscall page without the need for this flag. The second patch then replaces the 'VM_ALWAYSDUMP' flag with a new 'VM_NODUMP' flag, which can be set by userspace using new madvise flags: 'MADV_DONTDUMP', and unset via 'MADV_DODUMP'. The core dump filters continue to work the same as before unless 'MADV_DONTDUMP' is set on the region. The qemu code which implements this features is at: http://people.redhat.com/~jbaron/qemu-dump/qemu-dump.patch In my testing the qemu core dump shrunk from 383MB -> 13MB with this patch. I also believe that the 'MADV_DONTDUMP' flag might be useful for security sensitive apps, which might want to select which areas are dumped. This patch: The VM_ALWAYSDUMP flag is currently used by the coredump code to indicate that a vma is part of a vsyscall or vdso section. However, we can determine if a vma is in one these sections by checking it against the gate_vma and checking for a non-NULL return value from arch_vma_name(). Thus, freeing a valuable vma bit. Signed-off-by: Jason Baron <jbaron@redhat.com> Acked-by: Roland McGrath <roland@hack.frob.com> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Avi Kivity <avi@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-03-24 06:02:51 +08:00
/*
* The purpose of always_dump_vma() is to make sure that special kernel mappings
* that are useful for post-mortem analysis are included in every core dump.
* In that way we ensure that the core dump is fully interpretable later
* without matching up the same kernel and hardware config to see what PC values
* meant. These special mappings include - vDSO, vsyscall, and other
* architecture specific mappings
*/
static bool always_dump_vma(struct vm_area_struct *vma)
{
/* Any vsyscall mappings? */
if (vma == get_gate_vma(vma->vm_mm))
return true;
/*
* arch_vma_name() returns non-NULL for special architecture mappings,
* such as vDSO sections.
*/
if (arch_vma_name(vma))
return true;
return false;
}
/*
* Decide what to dump of a segment, part, all or none.
*/
static unsigned long vma_dump_size(struct vm_area_struct *vma,
unsigned long mm_flags)
{
coredump_filter: add hugepage dumping Presently hugepage's vma has a VM_RESERVED flag in order not to be swapped. But a VM_RESERVED vma isn't core dumped because this flag is often used for some kernel vmas (e.g. vmalloc, sound related). Thus hugepages are never dumped and it can't be debugged easily. Many developers want hugepages to be included into core-dump. However, We can't read generic VM_RESERVED area because this area is often IO mapping area. then these area reading may change device state. it is definitly undesiable side-effect. So adding a hugepage specific bit to the coredump filter is better. It will be able to hugepage core dumping and doesn't cause any side-effect to any i/o devices. In additional, libhugetlb use hugetlb private mapping pages as anonymous page. Then, hugepage private mapping pages should be core dumped by default. Then, /proc/[pid]/core_dump_filter has two new bits. - bit 5 mean hugetlb private mapping pages are dumped or not. (default: yes) - bit 6 mean hugetlb shared mapping pages are dumped or not. (default: no) I tested by following method. % ulimit -c unlimited % ./crash_hugepage 50 % ./crash_hugepage 50 -p % ls -lh % gdb ./crash_hugepage core % % echo 0x43 > /proc/self/coredump_filter % ./crash_hugepage 50 % ./crash_hugepage 50 -p % ls -lh % gdb ./crash_hugepage core #include <stdlib.h> #include <stdio.h> #include <unistd.h> #include <sys/mman.h> #include <string.h> #include "hugetlbfs.h" int main(int argc, char** argv){ char* p; int ch; int mmap_flags = MAP_SHARED; int fd; int nr_pages; while((ch = getopt(argc, argv, "p")) != -1) { switch (ch) { case 'p': mmap_flags &= ~MAP_SHARED; mmap_flags |= MAP_PRIVATE; break; default: /* nothing*/ break; } } argc -= optind; argv += optind; if (argc == 0){ printf("need # of pages\n"); exit(1); } nr_pages = atoi(argv[0]); if (nr_pages < 2) { printf("nr_pages must >2\n"); exit(1); } fd = hugetlbfs_unlinked_fd(); p = mmap(NULL, nr_pages * gethugepagesize(), PROT_READ|PROT_WRITE, mmap_flags, fd, 0); sleep(2); *(p + gethugepagesize()) = 1; /* COW */ sleep(2); /* crash! */ *(int*)0 = 1; return 0; } Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Reviewed-by: Kawai Hidehiro <hidehiro.kawai.ez@hitachi.com> Cc: Hugh Dickins <hugh@veritas.com> Cc: William Irwin <wli@holomorphy.com> Cc: Adam Litke <agl@us.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 11:27:08 +08:00
#define FILTER(type) (mm_flags & (1UL << MMF_DUMP_##type))
coredump: remove VM_ALWAYSDUMP flag The motivation for this patchset was that I was looking at a way for a qemu-kvm process, to exclude the guest memory from its core dump, which can be quite large. There are already a number of filter flags in /proc/<pid>/coredump_filter, however, these allow one to specify 'types' of kernel memory, not specific address ranges (which is needed in this case). Since there are no more vma flags available, the first patch eliminates the need for the 'VM_ALWAYSDUMP' flag. The flag is used internally by the kernel to mark vdso and vsyscall pages. However, it is simple enough to check if a vma covers a vdso or vsyscall page without the need for this flag. The second patch then replaces the 'VM_ALWAYSDUMP' flag with a new 'VM_NODUMP' flag, which can be set by userspace using new madvise flags: 'MADV_DONTDUMP', and unset via 'MADV_DODUMP'. The core dump filters continue to work the same as before unless 'MADV_DONTDUMP' is set on the region. The qemu code which implements this features is at: http://people.redhat.com/~jbaron/qemu-dump/qemu-dump.patch In my testing the qemu core dump shrunk from 383MB -> 13MB with this patch. I also believe that the 'MADV_DONTDUMP' flag might be useful for security sensitive apps, which might want to select which areas are dumped. This patch: The VM_ALWAYSDUMP flag is currently used by the coredump code to indicate that a vma is part of a vsyscall or vdso section. However, we can determine if a vma is in one these sections by checking it against the gate_vma and checking for a non-NULL return value from arch_vma_name(). Thus, freeing a valuable vma bit. Signed-off-by: Jason Baron <jbaron@redhat.com> Acked-by: Roland McGrath <roland@hack.frob.com> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Avi Kivity <avi@redhat.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-03-24 06:02:51 +08:00
/* always dump the vdso and vsyscall sections */
if (always_dump_vma(vma))
goto whole;
if (vma->vm_flags & VM_DONTDUMP)
return 0;
coredump_filter: add hugepage dumping Presently hugepage's vma has a VM_RESERVED flag in order not to be swapped. But a VM_RESERVED vma isn't core dumped because this flag is often used for some kernel vmas (e.g. vmalloc, sound related). Thus hugepages are never dumped and it can't be debugged easily. Many developers want hugepages to be included into core-dump. However, We can't read generic VM_RESERVED area because this area is often IO mapping area. then these area reading may change device state. it is definitly undesiable side-effect. So adding a hugepage specific bit to the coredump filter is better. It will be able to hugepage core dumping and doesn't cause any side-effect to any i/o devices. In additional, libhugetlb use hugetlb private mapping pages as anonymous page. Then, hugepage private mapping pages should be core dumped by default. Then, /proc/[pid]/core_dump_filter has two new bits. - bit 5 mean hugetlb private mapping pages are dumped or not. (default: yes) - bit 6 mean hugetlb shared mapping pages are dumped or not. (default: no) I tested by following method. % ulimit -c unlimited % ./crash_hugepage 50 % ./crash_hugepage 50 -p % ls -lh % gdb ./crash_hugepage core % % echo 0x43 > /proc/self/coredump_filter % ./crash_hugepage 50 % ./crash_hugepage 50 -p % ls -lh % gdb ./crash_hugepage core #include <stdlib.h> #include <stdio.h> #include <unistd.h> #include <sys/mman.h> #include <string.h> #include "hugetlbfs.h" int main(int argc, char** argv){ char* p; int ch; int mmap_flags = MAP_SHARED; int fd; int nr_pages; while((ch = getopt(argc, argv, "p")) != -1) { switch (ch) { case 'p': mmap_flags &= ~MAP_SHARED; mmap_flags |= MAP_PRIVATE; break; default: /* nothing*/ break; } } argc -= optind; argv += optind; if (argc == 0){ printf("need # of pages\n"); exit(1); } nr_pages = atoi(argv[0]); if (nr_pages < 2) { printf("nr_pages must >2\n"); exit(1); } fd = hugetlbfs_unlinked_fd(); p = mmap(NULL, nr_pages * gethugepagesize(), PROT_READ|PROT_WRITE, mmap_flags, fd, 0); sleep(2); *(p + gethugepagesize()) = 1; /* COW */ sleep(2); /* crash! */ *(int*)0 = 1; return 0; } Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Reviewed-by: Kawai Hidehiro <hidehiro.kawai.ez@hitachi.com> Cc: Hugh Dickins <hugh@veritas.com> Cc: William Irwin <wli@holomorphy.com> Cc: Adam Litke <agl@us.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 11:27:08 +08:00
/* Hugetlb memory check */
if (vma->vm_flags & VM_HUGETLB) {
if ((vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_SHARED))
goto whole;
if (!(vma->vm_flags & VM_SHARED) && FILTER(HUGETLB_PRIVATE))
goto whole;
return 0;
coredump_filter: add hugepage dumping Presently hugepage's vma has a VM_RESERVED flag in order not to be swapped. But a VM_RESERVED vma isn't core dumped because this flag is often used for some kernel vmas (e.g. vmalloc, sound related). Thus hugepages are never dumped and it can't be debugged easily. Many developers want hugepages to be included into core-dump. However, We can't read generic VM_RESERVED area because this area is often IO mapping area. then these area reading may change device state. it is definitly undesiable side-effect. So adding a hugepage specific bit to the coredump filter is better. It will be able to hugepage core dumping and doesn't cause any side-effect to any i/o devices. In additional, libhugetlb use hugetlb private mapping pages as anonymous page. Then, hugepage private mapping pages should be core dumped by default. Then, /proc/[pid]/core_dump_filter has two new bits. - bit 5 mean hugetlb private mapping pages are dumped or not. (default: yes) - bit 6 mean hugetlb shared mapping pages are dumped or not. (default: no) I tested by following method. % ulimit -c unlimited % ./crash_hugepage 50 % ./crash_hugepage 50 -p % ls -lh % gdb ./crash_hugepage core % % echo 0x43 > /proc/self/coredump_filter % ./crash_hugepage 50 % ./crash_hugepage 50 -p % ls -lh % gdb ./crash_hugepage core #include <stdlib.h> #include <stdio.h> #include <unistd.h> #include <sys/mman.h> #include <string.h> #include "hugetlbfs.h" int main(int argc, char** argv){ char* p; int ch; int mmap_flags = MAP_SHARED; int fd; int nr_pages; while((ch = getopt(argc, argv, "p")) != -1) { switch (ch) { case 'p': mmap_flags &= ~MAP_SHARED; mmap_flags |= MAP_PRIVATE; break; default: /* nothing*/ break; } } argc -= optind; argv += optind; if (argc == 0){ printf("need # of pages\n"); exit(1); } nr_pages = atoi(argv[0]); if (nr_pages < 2) { printf("nr_pages must >2\n"); exit(1); } fd = hugetlbfs_unlinked_fd(); p = mmap(NULL, nr_pages * gethugepagesize(), PROT_READ|PROT_WRITE, mmap_flags, fd, 0); sleep(2); *(p + gethugepagesize()) = 1; /* COW */ sleep(2); /* crash! */ *(int*)0 = 1; return 0; } Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Reviewed-by: Kawai Hidehiro <hidehiro.kawai.ez@hitachi.com> Cc: Hugh Dickins <hugh@veritas.com> Cc: William Irwin <wli@holomorphy.com> Cc: Adam Litke <agl@us.ibm.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-19 11:27:08 +08:00
}
/* Do not dump I/O mapped devices or special mappings */
mm: kill vma flag VM_RESERVED and mm->reserved_vm counter A long time ago, in v2.4, VM_RESERVED kept swapout process off VMA, currently it lost original meaning but still has some effects: | effect | alternative flags -+------------------------+--------------------------------------------- 1| account as reserved_vm | VM_IO 2| skip in core dump | VM_IO, VM_DONTDUMP 3| do not merge or expand | VM_IO, VM_DONTEXPAND, VM_HUGETLB, VM_PFNMAP 4| do not mlock | VM_IO, VM_DONTEXPAND, VM_HUGETLB, VM_PFNMAP This patch removes reserved_vm counter from mm_struct. Seems like nobody cares about it, it does not exported into userspace directly, it only reduces total_vm showed in proc. Thus VM_RESERVED can be replaced with VM_IO or pair VM_DONTEXPAND | VM_DONTDUMP. remap_pfn_range() and io_remap_pfn_range() set VM_IO|VM_DONTEXPAND|VM_DONTDUMP. remap_vmalloc_range() set VM_DONTEXPAND | VM_DONTDUMP. [akpm@linux-foundation.org: drivers/vfio/pci/vfio_pci.c fixup] Signed-off-by: Konstantin Khlebnikov <khlebnikov@openvz.org> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Carsten Otte <cotte@de.ibm.com> Cc: Chris Metcalf <cmetcalf@tilera.com> Cc: Cyrill Gorcunov <gorcunov@openvz.org> Cc: Eric Paris <eparis@redhat.com> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Hugh Dickins <hughd@google.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: James Morris <james.l.morris@oracle.com> Cc: Jason Baron <jbaron@redhat.com> Cc: Kentaro Takeda <takedakn@nttdata.co.jp> Cc: Matt Helsley <matthltc@us.ibm.com> Cc: Nick Piggin <npiggin@kernel.dk> Cc: Oleg Nesterov <oleg@redhat.com> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Robert Richter <robert.richter@amd.com> Cc: Suresh Siddha <suresh.b.siddha@intel.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Venkatesh Pallipadi <venki@google.com> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-09 07:29:02 +08:00
if (vma->vm_flags & VM_IO)
return 0;
/* By default, dump shared memory if mapped from an anonymous file. */
if (vma->vm_flags & VM_SHARED) {
if (file_inode(vma->vm_file)->i_nlink == 0 ?
FILTER(ANON_SHARED) : FILTER(MAPPED_SHARED))
goto whole;
return 0;
}
/* Dump segments that have been written to. */
if (vma->anon_vma && FILTER(ANON_PRIVATE))
goto whole;
if (vma->vm_file == NULL)
return 0;
if (FILTER(MAPPED_PRIVATE))
goto whole;
/*
* If this looks like the beginning of a DSO or executable mapping,
* check for an ELF header. If we find one, dump the first page to
* aid in determining what was mapped here.
*/
if (FILTER(ELF_HEADERS) &&
vma->vm_pgoff == 0 && (vma->vm_flags & VM_READ)) {
u32 __user *header = (u32 __user *) vma->vm_start;
u32 word;
mm_segment_t fs = get_fs();
/*
* Doing it this way gets the constant folded by GCC.
*/
union {
u32 cmp;
char elfmag[SELFMAG];
} magic;
BUILD_BUG_ON(SELFMAG != sizeof word);
magic.elfmag[EI_MAG0] = ELFMAG0;
magic.elfmag[EI_MAG1] = ELFMAG1;
magic.elfmag[EI_MAG2] = ELFMAG2;
magic.elfmag[EI_MAG3] = ELFMAG3;
/*
* Switch to the user "segment" for get_user(),
* then put back what elf_core_dump() had in place.
*/
set_fs(USER_DS);
if (unlikely(get_user(word, header)))
word = 0;
set_fs(fs);
if (word == magic.cmp)
return PAGE_SIZE;
}
#undef FILTER
return 0;
whole:
return vma->vm_end - vma->vm_start;
}
/* An ELF note in memory */
struct memelfnote
{
const char *name;
int type;
unsigned int datasz;
void *data;
};
static int notesize(struct memelfnote *en)
{
int sz;
sz = sizeof(struct elf_note);
sz += roundup(strlen(en->name) + 1, 4);
sz += roundup(en->datasz, 4);
return sz;
}
[PATCH] Support piping into commands in /proc/sys/kernel/core_pattern Using the infrastructure created in previous patches implement support to pipe core dumps into programs. This is done by overloading the existing core_pattern sysctl with a new syntax: |program When the first character of the pattern is a '|' the kernel will instead threat the rest of the pattern as a command to run. The core dump will be written to the standard input of that program instead of to a file. This is useful for having automatic core dump analysis without filling up disks. The program can do some simple analysis and save only a summary of the core dump. The core dump proces will run with the privileges and in the name space of the process that caused the core dump. I also increased the core pattern size to 128 bytes so that longer command lines fit. Most of the changes comes from allowing core dumps without seeks. They are fairly straight forward though. One small incompatibility is that if someone had a core pattern previously that started with '|' they will get suddenly new behaviour. I think that's unlikely to be a real problem though. Additional background: > Very nice, do you happen to have a program that can accept this kind of > input for crash dumps? I'm guessing that the embedded people will > really want this functionality. I had a cheesy demo/prototype. Basically it wrote the dump to a file again, ran gdb on it to get a backtrace and wrote the summary to a shared directory. Then there was a simple CGI script to generate a "top 10" crashes HTML listing. Unfortunately this still had the disadvantage to needing full disk space for a dump except for deleting it afterwards (in fact it was worse because over the pipe holes didn't work so if you have a holey address map it would require more space). Fortunately gdb seems to be happy to handle /proc/pid/fd/xxx input pipes as cores (at least it worked with zsh's =(cat core) syntax), so it would be likely possible to do it without temporary space with a simple wrapper that calls it in the right way. I ran out of time before doing that though. The demo prototype scripts weren't very good. If there is really interest I can dig them out (they are currently on a laptop disk on the desk with the laptop itself being in service), but I would recommend to rewrite them for any serious application of this and fix the disk space problem. Also to be really useful it should probably find a way to automatically fetch the debuginfos (I cheated and just installed them in advance). If nobody else does it I can probably do the rewrite myself again at some point. My hope at some point was that desktops would support it in their builtin crash reporters, but at least the KDE people I talked too seemed to be happy with their user space only solution. Alan sayeth: I don't believe that piping as such as neccessarily the right model, but the ability to intercept and processes core dumps from user space is asked for by many enterprise users as well. They want to know about, capture, analyse and process core dumps, often centrally and in automated form. [akpm@osdl.org: loff_t != unsigned long] Signed-off-by: Andi Kleen <ak@suse.de> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-10-01 14:29:28 +08:00
#define DUMP_WRITE(addr, nr, foffset) \
do { if (!dump_write(file, (addr), (nr))) return 0; *foffset += (nr); } while(0)
[PATCH] Support piping into commands in /proc/sys/kernel/core_pattern Using the infrastructure created in previous patches implement support to pipe core dumps into programs. This is done by overloading the existing core_pattern sysctl with a new syntax: |program When the first character of the pattern is a '|' the kernel will instead threat the rest of the pattern as a command to run. The core dump will be written to the standard input of that program instead of to a file. This is useful for having automatic core dump analysis without filling up disks. The program can do some simple analysis and save only a summary of the core dump. The core dump proces will run with the privileges and in the name space of the process that caused the core dump. I also increased the core pattern size to 128 bytes so that longer command lines fit. Most of the changes comes from allowing core dumps without seeks. They are fairly straight forward though. One small incompatibility is that if someone had a core pattern previously that started with '|' they will get suddenly new behaviour. I think that's unlikely to be a real problem though. Additional background: > Very nice, do you happen to have a program that can accept this kind of > input for crash dumps? I'm guessing that the embedded people will > really want this functionality. I had a cheesy demo/prototype. Basically it wrote the dump to a file again, ran gdb on it to get a backtrace and wrote the summary to a shared directory. Then there was a simple CGI script to generate a "top 10" crashes HTML listing. Unfortunately this still had the disadvantage to needing full disk space for a dump except for deleting it afterwards (in fact it was worse because over the pipe holes didn't work so if you have a holey address map it would require more space). Fortunately gdb seems to be happy to handle /proc/pid/fd/xxx input pipes as cores (at least it worked with zsh's =(cat core) syntax), so it would be likely possible to do it without temporary space with a simple wrapper that calls it in the right way. I ran out of time before doing that though. The demo prototype scripts weren't very good. If there is really interest I can dig them out (they are currently on a laptop disk on the desk with the laptop itself being in service), but I would recommend to rewrite them for any serious application of this and fix the disk space problem. Also to be really useful it should probably find a way to automatically fetch the debuginfos (I cheated and just installed them in advance). If nobody else does it I can probably do the rewrite myself again at some point. My hope at some point was that desktops would support it in their builtin crash reporters, but at least the KDE people I talked too seemed to be happy with their user space only solution. Alan sayeth: I don't believe that piping as such as neccessarily the right model, but the ability to intercept and processes core dumps from user space is asked for by many enterprise users as well. They want to know about, capture, analyse and process core dumps, often centrally and in automated form. [akpm@osdl.org: loff_t != unsigned long] Signed-off-by: Andi Kleen <ak@suse.de> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-10-01 14:29:28 +08:00
static int alignfile(struct file *file, loff_t *foffset)
{
static const char buf[4] = { 0, };
[PATCH] Support piping into commands in /proc/sys/kernel/core_pattern Using the infrastructure created in previous patches implement support to pipe core dumps into programs. This is done by overloading the existing core_pattern sysctl with a new syntax: |program When the first character of the pattern is a '|' the kernel will instead threat the rest of the pattern as a command to run. The core dump will be written to the standard input of that program instead of to a file. This is useful for having automatic core dump analysis without filling up disks. The program can do some simple analysis and save only a summary of the core dump. The core dump proces will run with the privileges and in the name space of the process that caused the core dump. I also increased the core pattern size to 128 bytes so that longer command lines fit. Most of the changes comes from allowing core dumps without seeks. They are fairly straight forward though. One small incompatibility is that if someone had a core pattern previously that started with '|' they will get suddenly new behaviour. I think that's unlikely to be a real problem though. Additional background: > Very nice, do you happen to have a program that can accept this kind of > input for crash dumps? I'm guessing that the embedded people will > really want this functionality. I had a cheesy demo/prototype. Basically it wrote the dump to a file again, ran gdb on it to get a backtrace and wrote the summary to a shared directory. Then there was a simple CGI script to generate a "top 10" crashes HTML listing. Unfortunately this still had the disadvantage to needing full disk space for a dump except for deleting it afterwards (in fact it was worse because over the pipe holes didn't work so if you have a holey address map it would require more space). Fortunately gdb seems to be happy to handle /proc/pid/fd/xxx input pipes as cores (at least it worked with zsh's =(cat core) syntax), so it would be likely possible to do it without temporary space with a simple wrapper that calls it in the right way. I ran out of time before doing that though. The demo prototype scripts weren't very good. If there is really interest I can dig them out (they are currently on a laptop disk on the desk with the laptop itself being in service), but I would recommend to rewrite them for any serious application of this and fix the disk space problem. Also to be really useful it should probably find a way to automatically fetch the debuginfos (I cheated and just installed them in advance). If nobody else does it I can probably do the rewrite myself again at some point. My hope at some point was that desktops would support it in their builtin crash reporters, but at least the KDE people I talked too seemed to be happy with their user space only solution. Alan sayeth: I don't believe that piping as such as neccessarily the right model, but the ability to intercept and processes core dumps from user space is asked for by many enterprise users as well. They want to know about, capture, analyse and process core dumps, often centrally and in automated form. [akpm@osdl.org: loff_t != unsigned long] Signed-off-by: Andi Kleen <ak@suse.de> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-10-01 14:29:28 +08:00
DUMP_WRITE(buf, roundup(*foffset, 4) - *foffset, foffset);
return 1;
}
[PATCH] Support piping into commands in /proc/sys/kernel/core_pattern Using the infrastructure created in previous patches implement support to pipe core dumps into programs. This is done by overloading the existing core_pattern sysctl with a new syntax: |program When the first character of the pattern is a '|' the kernel will instead threat the rest of the pattern as a command to run. The core dump will be written to the standard input of that program instead of to a file. This is useful for having automatic core dump analysis without filling up disks. The program can do some simple analysis and save only a summary of the core dump. The core dump proces will run with the privileges and in the name space of the process that caused the core dump. I also increased the core pattern size to 128 bytes so that longer command lines fit. Most of the changes comes from allowing core dumps without seeks. They are fairly straight forward though. One small incompatibility is that if someone had a core pattern previously that started with '|' they will get suddenly new behaviour. I think that's unlikely to be a real problem though. Additional background: > Very nice, do you happen to have a program that can accept this kind of > input for crash dumps? I'm guessing that the embedded people will > really want this functionality. I had a cheesy demo/prototype. Basically it wrote the dump to a file again, ran gdb on it to get a backtrace and wrote the summary to a shared directory. Then there was a simple CGI script to generate a "top 10" crashes HTML listing. Unfortunately this still had the disadvantage to needing full disk space for a dump except for deleting it afterwards (in fact it was worse because over the pipe holes didn't work so if you have a holey address map it would require more space). Fortunately gdb seems to be happy to handle /proc/pid/fd/xxx input pipes as cores (at least it worked with zsh's =(cat core) syntax), so it would be likely possible to do it without temporary space with a simple wrapper that calls it in the right way. I ran out of time before doing that though. The demo prototype scripts weren't very good. If there is really interest I can dig them out (they are currently on a laptop disk on the desk with the laptop itself being in service), but I would recommend to rewrite them for any serious application of this and fix the disk space problem. Also to be really useful it should probably find a way to automatically fetch the debuginfos (I cheated and just installed them in advance). If nobody else does it I can probably do the rewrite myself again at some point. My hope at some point was that desktops would support it in their builtin crash reporters, but at least the KDE people I talked too seemed to be happy with their user space only solution. Alan sayeth: I don't believe that piping as such as neccessarily the right model, but the ability to intercept and processes core dumps from user space is asked for by many enterprise users as well. They want to know about, capture, analyse and process core dumps, often centrally and in automated form. [akpm@osdl.org: loff_t != unsigned long] Signed-off-by: Andi Kleen <ak@suse.de> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-10-01 14:29:28 +08:00
static int writenote(struct memelfnote *men, struct file *file,
loff_t *foffset)
{
struct elf_note en;
en.n_namesz = strlen(men->name) + 1;
en.n_descsz = men->datasz;
en.n_type = men->type;
[PATCH] Support piping into commands in /proc/sys/kernel/core_pattern Using the infrastructure created in previous patches implement support to pipe core dumps into programs. This is done by overloading the existing core_pattern sysctl with a new syntax: |program When the first character of the pattern is a '|' the kernel will instead threat the rest of the pattern as a command to run. The core dump will be written to the standard input of that program instead of to a file. This is useful for having automatic core dump analysis without filling up disks. The program can do some simple analysis and save only a summary of the core dump. The core dump proces will run with the privileges and in the name space of the process that caused the core dump. I also increased the core pattern size to 128 bytes so that longer command lines fit. Most of the changes comes from allowing core dumps without seeks. They are fairly straight forward though. One small incompatibility is that if someone had a core pattern previously that started with '|' they will get suddenly new behaviour. I think that's unlikely to be a real problem though. Additional background: > Very nice, do you happen to have a program that can accept this kind of > input for crash dumps? I'm guessing that the embedded people will > really want this functionality. I had a cheesy demo/prototype. Basically it wrote the dump to a file again, ran gdb on it to get a backtrace and wrote the summary to a shared directory. Then there was a simple CGI script to generate a "top 10" crashes HTML listing. Unfortunately this still had the disadvantage to needing full disk space for a dump except for deleting it afterwards (in fact it was worse because over the pipe holes didn't work so if you have a holey address map it would require more space). Fortunately gdb seems to be happy to handle /proc/pid/fd/xxx input pipes as cores (at least it worked with zsh's =(cat core) syntax), so it would be likely possible to do it without temporary space with a simple wrapper that calls it in the right way. I ran out of time before doing that though. The demo prototype scripts weren't very good. If there is really interest I can dig them out (they are currently on a laptop disk on the desk with the laptop itself being in service), but I would recommend to rewrite them for any serious application of this and fix the disk space problem. Also to be really useful it should probably find a way to automatically fetch the debuginfos (I cheated and just installed them in advance). If nobody else does it I can probably do the rewrite myself again at some point. My hope at some point was that desktops would support it in their builtin crash reporters, but at least the KDE people I talked too seemed to be happy with their user space only solution. Alan sayeth: I don't believe that piping as such as neccessarily the right model, but the ability to intercept and processes core dumps from user space is asked for by many enterprise users as well. They want to know about, capture, analyse and process core dumps, often centrally and in automated form. [akpm@osdl.org: loff_t != unsigned long] Signed-off-by: Andi Kleen <ak@suse.de> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-10-01 14:29:28 +08:00
DUMP_WRITE(&en, sizeof(en), foffset);
DUMP_WRITE(men->name, en.n_namesz, foffset);
if (!alignfile(file, foffset))
return 0;
DUMP_WRITE(men->data, men->datasz, foffset);
if (!alignfile(file, foffset))
return 0;
return 1;
}
#undef DUMP_WRITE
static void fill_elf_header(struct elfhdr *elf, int segs,
u16 machine, u32 flags)
{
memset(elf, 0, sizeof(*elf));
memcpy(elf->e_ident, ELFMAG, SELFMAG);
elf->e_ident[EI_CLASS] = ELF_CLASS;
elf->e_ident[EI_DATA] = ELF_DATA;
elf->e_ident[EI_VERSION] = EV_CURRENT;
elf->e_ident[EI_OSABI] = ELF_OSABI;
elf->e_type = ET_CORE;
elf->e_machine = machine;
elf->e_version = EV_CURRENT;
elf->e_phoff = sizeof(struct elfhdr);
elf->e_flags = flags;
elf->e_ehsize = sizeof(struct elfhdr);
elf->e_phentsize = sizeof(struct elf_phdr);
elf->e_phnum = segs;
return;
}
static void fill_elf_note_phdr(struct elf_phdr *phdr, int sz, loff_t offset)
{
phdr->p_type = PT_NOTE;
phdr->p_offset = offset;
phdr->p_vaddr = 0;
phdr->p_paddr = 0;
phdr->p_filesz = sz;
phdr->p_memsz = 0;
phdr->p_flags = 0;
phdr->p_align = 0;
return;
}
static void fill_note(struct memelfnote *note, const char *name, int type,
unsigned int sz, void *data)
{
note->name = name;
note->type = type;
note->datasz = sz;
note->data = data;
return;
}
/*
* fill up all the fields in prstatus from the given task struct, except
* registers which need to be filled up separately.
*/
static void fill_prstatus(struct elf_prstatus *prstatus,
struct task_struct *p, long signr)
{
prstatus->pr_info.si_signo = prstatus->pr_cursig = signr;
prstatus->pr_sigpend = p->pending.signal.sig[0];
prstatus->pr_sighold = p->blocked.sig[0];
rcu_read_lock();
prstatus->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
rcu_read_unlock();
prstatus->pr_pid = task_pid_vnr(p);
prstatus->pr_pgrp = task_pgrp_vnr(p);
prstatus->pr_sid = task_session_vnr(p);
if (thread_group_leader(p)) {
timers: fix itimer/many thread hang Overview This patch reworks the handling of POSIX CPU timers, including the ITIMER_PROF, ITIMER_VIRT timers and rlimit handling. It was put together with the help of Roland McGrath, the owner and original writer of this code. The problem we ran into, and the reason for this rework, has to do with using a profiling timer in a process with a large number of threads. It appears that the performance of the old implementation of run_posix_cpu_timers() was at least O(n*3) (where "n" is the number of threads in a process) or worse. Everything is fine with an increasing number of threads until the time taken for that routine to run becomes the same as or greater than the tick time, at which point things degrade rather quickly. This patch fixes bug 9906, "Weird hang with NPTL and SIGPROF." Code Changes This rework corrects the implementation of run_posix_cpu_timers() to make it run in constant time for a particular machine. (Performance may vary between one machine and another depending upon whether the kernel is built as single- or multiprocessor and, in the latter case, depending upon the number of running processors.) To do this, at each tick we now update fields in signal_struct as well as task_struct. The run_posix_cpu_timers() function uses those fields to make its decisions. We define a new structure, "task_cputime," to contain user, system and scheduler times and use these in appropriate places: struct task_cputime { cputime_t utime; cputime_t stime; unsigned long long sum_exec_runtime; }; This is included in the structure "thread_group_cputime," which is a new substructure of signal_struct and which varies for uniprocessor versus multiprocessor kernels. For uniprocessor kernels, it uses "task_cputime" as a simple substructure, while for multiprocessor kernels it is a pointer: struct thread_group_cputime { struct task_cputime totals; }; struct thread_group_cputime { struct task_cputime *totals; }; We also add a new task_cputime substructure directly to signal_struct, to cache the earliest expiration of process-wide timers, and task_cputime also replaces the it_*_expires fields of task_struct (used for earliest expiration of thread timers). The "thread_group_cputime" structure contains process-wide timers that are updated via account_user_time() and friends. In the non-SMP case the structure is a simple aggregator; unfortunately in the SMP case that simplicity was not achievable due to cache-line contention between CPUs (in one measured case performance was actually _worse_ on a 16-cpu system than the same test on a 4-cpu system, due to this contention). For SMP, the thread_group_cputime counters are maintained as a per-cpu structure allocated using alloc_percpu(). The timer functions update only the timer field in the structure corresponding to the running CPU, obtained using per_cpu_ptr(). We define a set of inline functions in sched.h that we use to maintain the thread_group_cputime structure and hide the differences between UP and SMP implementations from the rest of the kernel. The thread_group_cputime_init() function initializes the thread_group_cputime structure for the given task. The thread_group_cputime_alloc() is a no-op for UP; for SMP it calls the out-of-line function thread_group_cputime_alloc_smp() to allocate and fill in the per-cpu structures and fields. The thread_group_cputime_free() function, also a no-op for UP, in SMP frees the per-cpu structures. The thread_group_cputime_clone_thread() function (also a UP no-op) for SMP calls thread_group_cputime_alloc() if the per-cpu structures haven't yet been allocated. The thread_group_cputime() function fills the task_cputime structure it is passed with the contents of the thread_group_cputime fields; in UP it's that simple but in SMP it must also safely check that tsk->signal is non-NULL (if it is it just uses the appropriate fields of task_struct) and, if so, sums the per-cpu values for each online CPU. Finally, the three functions account_group_user_time(), account_group_system_time() and account_group_exec_runtime() are used by timer functions to update the respective fields of the thread_group_cputime structure. Non-SMP operation is trivial and will not be mentioned further. The per-cpu structure is always allocated when a task creates its first new thread, via a call to thread_group_cputime_clone_thread() from copy_signal(). It is freed at process exit via a call to thread_group_cputime_free() from cleanup_signal(). All functions that formerly summed utime/stime/sum_sched_runtime values from from all threads in the thread group now use thread_group_cputime() to snapshot the values in the thread_group_cputime structure or the values in the task structure itself if the per-cpu structure hasn't been allocated. Finally, the code in kernel/posix-cpu-timers.c has changed quite a bit. The run_posix_cpu_timers() function has been split into a fast path and a slow path; the former safely checks whether there are any expired thread timers and, if not, just returns, while the slow path does the heavy lifting. With the dedicated thread group fields, timers are no longer "rebalanced" and the process_timer_rebalance() function and related code has gone away. All summing loops are gone and all code that used them now uses the thread_group_cputime() inline. When process-wide timers are set, the new task_cputime structure in signal_struct is used to cache the earliest expiration; this is checked in the fast path. Performance The fix appears not to add significant overhead to existing operations. It generally performs the same as the current code except in two cases, one in which it performs slightly worse (Case 5 below) and one in which it performs very significantly better (Case 2 below). Overall it's a wash except in those two cases. I've since done somewhat more involved testing on a dual-core Opteron system. Case 1: With no itimer running, for a test with 100,000 threads, the fixed kernel took 1428.5 seconds, 513 seconds more than the unfixed system, all of which was spent in the system. There were twice as many voluntary context switches with the fix as without it. Case 2: With an itimer running at .01 second ticks and 4000 threads (the most an unmodified kernel can handle), the fixed kernel ran the test in eight percent of the time (5.8 seconds as opposed to 70 seconds) and had better tick accuracy (.012 seconds per tick as opposed to .023 seconds per tick). Case 3: A 4000-thread test with an initial timer tick of .01 second and an interval of 10,000 seconds (i.e. a timer that ticks only once) had very nearly the same performance in both cases: 6.3 seconds elapsed for the fixed kernel versus 5.5 seconds for the unfixed kernel. With fewer threads (eight in these tests), the Case 1 test ran in essentially the same time on both the modified and unmodified kernels (5.2 seconds versus 5.8 seconds). The Case 2 test ran in about the same time as well, 5.9 seconds versus 5.4 seconds but again with much better tick accuracy, .013 seconds per tick versus .025 seconds per tick for the unmodified kernel. Since the fix affected the rlimit code, I also tested soft and hard CPU limits. Case 4: With a hard CPU limit of 20 seconds and eight threads (and an itimer running), the modified kernel was very slightly favored in that while it killed the process in 19.997 seconds of CPU time (5.002 seconds of wall time), only .003 seconds of that was system time, the rest was user time. The unmodified kernel killed the process in 20.001 seconds of CPU (5.014 seconds of wall time) of which .016 seconds was system time. Really, though, the results were too close to call. The results were essentially the same with no itimer running. Case 5: With a soft limit of 20 seconds and a hard limit of 2000 seconds (where the hard limit would never be reached) and an itimer running, the modified kernel exhibited worse tick accuracy than the unmodified kernel: .050 seconds/tick versus .028 seconds/tick. Otherwise, performance was almost indistinguishable. With no itimer running this test exhibited virtually identical behavior and times in both cases. In times past I did some limited performance testing. those results are below. On a four-cpu Opteron system without this fix, a sixteen-thread test executed in 3569.991 seconds, of which user was 3568.435s and system was 1.556s. On the same system with the fix, user and elapsed time were about the same, but system time dropped to 0.007 seconds. Performance with eight, four and one thread were comparable. Interestingly, the timer ticks with the fix seemed more accurate: The sixteen-thread test with the fix received 149543 ticks for 0.024 seconds per tick, while the same test without the fix received 58720 for 0.061 seconds per tick. Both cases were configured for an interval of 0.01 seconds. Again, the other tests were comparable. Each thread in this test computed the primes up to 25,000,000. I also did a test with a large number of threads, 100,000 threads, which is impossible without the fix. In this case each thread computed the primes only up to 10,000 (to make the runtime manageable). System time dominated, at 1546.968 seconds out of a total 2176.906 seconds (giving a user time of 629.938s). It received 147651 ticks for 0.015 seconds per tick, still quite accurate. There is obviously no comparable test without the fix. Signed-off-by: Frank Mayhar <fmayhar@google.com> Cc: Roland McGrath <roland@redhat.com> Cc: Alexey Dobriyan <adobriyan@gmail.com> Cc: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-09-13 00:54:39 +08:00
struct task_cputime cputime;
/*
timers: fix itimer/many thread hang Overview This patch reworks the handling of POSIX CPU timers, including the ITIMER_PROF, ITIMER_VIRT timers and rlimit handling. It was put together with the help of Roland McGrath, the owner and original writer of this code. The problem we ran into, and the reason for this rework, has to do with using a profiling timer in a process with a large number of threads. It appears that the performance of the old implementation of run_posix_cpu_timers() was at least O(n*3) (where "n" is the number of threads in a process) or worse. Everything is fine with an increasing number of threads until the time taken for that routine to run becomes the same as or greater than the tick time, at which point things degrade rather quickly. This patch fixes bug 9906, "Weird hang with NPTL and SIGPROF." Code Changes This rework corrects the implementation of run_posix_cpu_timers() to make it run in constant time for a particular machine. (Performance may vary between one machine and another depending upon whether the kernel is built as single- or multiprocessor and, in the latter case, depending upon the number of running processors.) To do this, at each tick we now update fields in signal_struct as well as task_struct. The run_posix_cpu_timers() function uses those fields to make its decisions. We define a new structure, "task_cputime," to contain user, system and scheduler times and use these in appropriate places: struct task_cputime { cputime_t utime; cputime_t stime; unsigned long long sum_exec_runtime; }; This is included in the structure "thread_group_cputime," which is a new substructure of signal_struct and which varies for uniprocessor versus multiprocessor kernels. For uniprocessor kernels, it uses "task_cputime" as a simple substructure, while for multiprocessor kernels it is a pointer: struct thread_group_cputime { struct task_cputime totals; }; struct thread_group_cputime { struct task_cputime *totals; }; We also add a new task_cputime substructure directly to signal_struct, to cache the earliest expiration of process-wide timers, and task_cputime also replaces the it_*_expires fields of task_struct (used for earliest expiration of thread timers). The "thread_group_cputime" structure contains process-wide timers that are updated via account_user_time() and friends. In the non-SMP case the structure is a simple aggregator; unfortunately in the SMP case that simplicity was not achievable due to cache-line contention between CPUs (in one measured case performance was actually _worse_ on a 16-cpu system than the same test on a 4-cpu system, due to this contention). For SMP, the thread_group_cputime counters are maintained as a per-cpu structure allocated using alloc_percpu(). The timer functions update only the timer field in the structure corresponding to the running CPU, obtained using per_cpu_ptr(). We define a set of inline functions in sched.h that we use to maintain the thread_group_cputime structure and hide the differences between UP and SMP implementations from the rest of the kernel. The thread_group_cputime_init() function initializes the thread_group_cputime structure for the given task. The thread_group_cputime_alloc() is a no-op for UP; for SMP it calls the out-of-line function thread_group_cputime_alloc_smp() to allocate and fill in the per-cpu structures and fields. The thread_group_cputime_free() function, also a no-op for UP, in SMP frees the per-cpu structures. The thread_group_cputime_clone_thread() function (also a UP no-op) for SMP calls thread_group_cputime_alloc() if the per-cpu structures haven't yet been allocated. The thread_group_cputime() function fills the task_cputime structure it is passed with the contents of the thread_group_cputime fields; in UP it's that simple but in SMP it must also safely check that tsk->signal is non-NULL (if it is it just uses the appropriate fields of task_struct) and, if so, sums the per-cpu values for each online CPU. Finally, the three functions account_group_user_time(), account_group_system_time() and account_group_exec_runtime() are used by timer functions to update the respective fields of the thread_group_cputime structure. Non-SMP operation is trivial and will not be mentioned further. The per-cpu structure is always allocated when a task creates its first new thread, via a call to thread_group_cputime_clone_thread() from copy_signal(). It is freed at process exit via a call to thread_group_cputime_free() from cleanup_signal(). All functions that formerly summed utime/stime/sum_sched_runtime values from from all threads in the thread group now use thread_group_cputime() to snapshot the values in the thread_group_cputime structure or the values in the task structure itself if the per-cpu structure hasn't been allocated. Finally, the code in kernel/posix-cpu-timers.c has changed quite a bit. The run_posix_cpu_timers() function has been split into a fast path and a slow path; the former safely checks whether there are any expired thread timers and, if not, just returns, while the slow path does the heavy lifting. With the dedicated thread group fields, timers are no longer "rebalanced" and the process_timer_rebalance() function and related code has gone away. All summing loops are gone and all code that used them now uses the thread_group_cputime() inline. When process-wide timers are set, the new task_cputime structure in signal_struct is used to cache the earliest expiration; this is checked in the fast path. Performance The fix appears not to add significant overhead to existing operations. It generally performs the same as the current code except in two cases, one in which it performs slightly worse (Case 5 below) and one in which it performs very significantly better (Case 2 below). Overall it's a wash except in those two cases. I've since done somewhat more involved testing on a dual-core Opteron system. Case 1: With no itimer running, for a test with 100,000 threads, the fixed kernel took 1428.5 seconds, 513 seconds more than the unfixed system, all of which was spent in the system. There were twice as many voluntary context switches with the fix as without it. Case 2: With an itimer running at .01 second ticks and 4000 threads (the most an unmodified kernel can handle), the fixed kernel ran the test in eight percent of the time (5.8 seconds as opposed to 70 seconds) and had better tick accuracy (.012 seconds per tick as opposed to .023 seconds per tick). Case 3: A 4000-thread test with an initial timer tick of .01 second and an interval of 10,000 seconds (i.e. a timer that ticks only once) had very nearly the same performance in both cases: 6.3 seconds elapsed for the fixed kernel versus 5.5 seconds for the unfixed kernel. With fewer threads (eight in these tests), the Case 1 test ran in essentially the same time on both the modified and unmodified kernels (5.2 seconds versus 5.8 seconds). The Case 2 test ran in about the same time as well, 5.9 seconds versus 5.4 seconds but again with much better tick accuracy, .013 seconds per tick versus .025 seconds per tick for the unmodified kernel. Since the fix affected the rlimit code, I also tested soft and hard CPU limits. Case 4: With a hard CPU limit of 20 seconds and eight threads (and an itimer running), the modified kernel was very slightly favored in that while it killed the process in 19.997 seconds of CPU time (5.002 seconds of wall time), only .003 seconds of that was system time, the rest was user time. The unmodified kernel killed the process in 20.001 seconds of CPU (5.014 seconds of wall time) of which .016 seconds was system time. Really, though, the results were too close to call. The results were essentially the same with no itimer running. Case 5: With a soft limit of 20 seconds and a hard limit of 2000 seconds (where the hard limit would never be reached) and an itimer running, the modified kernel exhibited worse tick accuracy than the unmodified kernel: .050 seconds/tick versus .028 seconds/tick. Otherwise, performance was almost indistinguishable. With no itimer running this test exhibited virtually identical behavior and times in both cases. In times past I did some limited performance testing. those results are below. On a four-cpu Opteron system without this fix, a sixteen-thread test executed in 3569.991 seconds, of which user was 3568.435s and system was 1.556s. On the same system with the fix, user and elapsed time were about the same, but system time dropped to 0.007 seconds. Performance with eight, four and one thread were comparable. Interestingly, the timer ticks with the fix seemed more accurate: The sixteen-thread test with the fix received 149543 ticks for 0.024 seconds per tick, while the same test without the fix received 58720 for 0.061 seconds per tick. Both cases were configured for an interval of 0.01 seconds. Again, the other tests were comparable. Each thread in this test computed the primes up to 25,000,000. I also did a test with a large number of threads, 100,000 threads, which is impossible without the fix. In this case each thread computed the primes only up to 10,000 (to make the runtime manageable). System time dominated, at 1546.968 seconds out of a total 2176.906 seconds (giving a user time of 629.938s). It received 147651 ticks for 0.015 seconds per tick, still quite accurate. There is obviously no comparable test without the fix. Signed-off-by: Frank Mayhar <fmayhar@google.com> Cc: Roland McGrath <roland@redhat.com> Cc: Alexey Dobriyan <adobriyan@gmail.com> Cc: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-09-13 00:54:39 +08:00
* This is the record for the group leader. It shows the
* group-wide total, not its individual thread total.
*/
timers: fix itimer/many thread hang Overview This patch reworks the handling of POSIX CPU timers, including the ITIMER_PROF, ITIMER_VIRT timers and rlimit handling. It was put together with the help of Roland McGrath, the owner and original writer of this code. The problem we ran into, and the reason for this rework, has to do with using a profiling timer in a process with a large number of threads. It appears that the performance of the old implementation of run_posix_cpu_timers() was at least O(n*3) (where "n" is the number of threads in a process) or worse. Everything is fine with an increasing number of threads until the time taken for that routine to run becomes the same as or greater than the tick time, at which point things degrade rather quickly. This patch fixes bug 9906, "Weird hang with NPTL and SIGPROF." Code Changes This rework corrects the implementation of run_posix_cpu_timers() to make it run in constant time for a particular machine. (Performance may vary between one machine and another depending upon whether the kernel is built as single- or multiprocessor and, in the latter case, depending upon the number of running processors.) To do this, at each tick we now update fields in signal_struct as well as task_struct. The run_posix_cpu_timers() function uses those fields to make its decisions. We define a new structure, "task_cputime," to contain user, system and scheduler times and use these in appropriate places: struct task_cputime { cputime_t utime; cputime_t stime; unsigned long long sum_exec_runtime; }; This is included in the structure "thread_group_cputime," which is a new substructure of signal_struct and which varies for uniprocessor versus multiprocessor kernels. For uniprocessor kernels, it uses "task_cputime" as a simple substructure, while for multiprocessor kernels it is a pointer: struct thread_group_cputime { struct task_cputime totals; }; struct thread_group_cputime { struct task_cputime *totals; }; We also add a new task_cputime substructure directly to signal_struct, to cache the earliest expiration of process-wide timers, and task_cputime also replaces the it_*_expires fields of task_struct (used for earliest expiration of thread timers). The "thread_group_cputime" structure contains process-wide timers that are updated via account_user_time() and friends. In the non-SMP case the structure is a simple aggregator; unfortunately in the SMP case that simplicity was not achievable due to cache-line contention between CPUs (in one measured case performance was actually _worse_ on a 16-cpu system than the same test on a 4-cpu system, due to this contention). For SMP, the thread_group_cputime counters are maintained as a per-cpu structure allocated using alloc_percpu(). The timer functions update only the timer field in the structure corresponding to the running CPU, obtained using per_cpu_ptr(). We define a set of inline functions in sched.h that we use to maintain the thread_group_cputime structure and hide the differences between UP and SMP implementations from the rest of the kernel. The thread_group_cputime_init() function initializes the thread_group_cputime structure for the given task. The thread_group_cputime_alloc() is a no-op for UP; for SMP it calls the out-of-line function thread_group_cputime_alloc_smp() to allocate and fill in the per-cpu structures and fields. The thread_group_cputime_free() function, also a no-op for UP, in SMP frees the per-cpu structures. The thread_group_cputime_clone_thread() function (also a UP no-op) for SMP calls thread_group_cputime_alloc() if the per-cpu structures haven't yet been allocated. The thread_group_cputime() function fills the task_cputime structure it is passed with the contents of the thread_group_cputime fields; in UP it's that simple but in SMP it must also safely check that tsk->signal is non-NULL (if it is it just uses the appropriate fields of task_struct) and, if so, sums the per-cpu values for each online CPU. Finally, the three functions account_group_user_time(), account_group_system_time() and account_group_exec_runtime() are used by timer functions to update the respective fields of the thread_group_cputime structure. Non-SMP operation is trivial and will not be mentioned further. The per-cpu structure is always allocated when a task creates its first new thread, via a call to thread_group_cputime_clone_thread() from copy_signal(). It is freed at process exit via a call to thread_group_cputime_free() from cleanup_signal(). All functions that formerly summed utime/stime/sum_sched_runtime values from from all threads in the thread group now use thread_group_cputime() to snapshot the values in the thread_group_cputime structure or the values in the task structure itself if the per-cpu structure hasn't been allocated. Finally, the code in kernel/posix-cpu-timers.c has changed quite a bit. The run_posix_cpu_timers() function has been split into a fast path and a slow path; the former safely checks whether there are any expired thread timers and, if not, just returns, while the slow path does the heavy lifting. With the dedicated thread group fields, timers are no longer "rebalanced" and the process_timer_rebalance() function and related code has gone away. All summing loops are gone and all code that used them now uses the thread_group_cputime() inline. When process-wide timers are set, the new task_cputime structure in signal_struct is used to cache the earliest expiration; this is checked in the fast path. Performance The fix appears not to add significant overhead to existing operations. It generally performs the same as the current code except in two cases, one in which it performs slightly worse (Case 5 below) and one in which it performs very significantly better (Case 2 below). Overall it's a wash except in those two cases. I've since done somewhat more involved testing on a dual-core Opteron system. Case 1: With no itimer running, for a test with 100,000 threads, the fixed kernel took 1428.5 seconds, 513 seconds more than the unfixed system, all of which was spent in the system. There were twice as many voluntary context switches with the fix as without it. Case 2: With an itimer running at .01 second ticks and 4000 threads (the most an unmodified kernel can handle), the fixed kernel ran the test in eight percent of the time (5.8 seconds as opposed to 70 seconds) and had better tick accuracy (.012 seconds per tick as opposed to .023 seconds per tick). Case 3: A 4000-thread test with an initial timer tick of .01 second and an interval of 10,000 seconds (i.e. a timer that ticks only once) had very nearly the same performance in both cases: 6.3 seconds elapsed for the fixed kernel versus 5.5 seconds for the unfixed kernel. With fewer threads (eight in these tests), the Case 1 test ran in essentially the same time on both the modified and unmodified kernels (5.2 seconds versus 5.8 seconds). The Case 2 test ran in about the same time as well, 5.9 seconds versus 5.4 seconds but again with much better tick accuracy, .013 seconds per tick versus .025 seconds per tick for the unmodified kernel. Since the fix affected the rlimit code, I also tested soft and hard CPU limits. Case 4: With a hard CPU limit of 20 seconds and eight threads (and an itimer running), the modified kernel was very slightly favored in that while it killed the process in 19.997 seconds of CPU time (5.002 seconds of wall time), only .003 seconds of that was system time, the rest was user time. The unmodified kernel killed the process in 20.001 seconds of CPU (5.014 seconds of wall time) of which .016 seconds was system time. Really, though, the results were too close to call. The results were essentially the same with no itimer running. Case 5: With a soft limit of 20 seconds and a hard limit of 2000 seconds (where the hard limit would never be reached) and an itimer running, the modified kernel exhibited worse tick accuracy than the unmodified kernel: .050 seconds/tick versus .028 seconds/tick. Otherwise, performance was almost indistinguishable. With no itimer running this test exhibited virtually identical behavior and times in both cases. In times past I did some limited performance testing. those results are below. On a four-cpu Opteron system without this fix, a sixteen-thread test executed in 3569.991 seconds, of which user was 3568.435s and system was 1.556s. On the same system with the fix, user and elapsed time were about the same, but system time dropped to 0.007 seconds. Performance with eight, four and one thread were comparable. Interestingly, the timer ticks with the fix seemed more accurate: The sixteen-thread test with the fix received 149543 ticks for 0.024 seconds per tick, while the same test without the fix received 58720 for 0.061 seconds per tick. Both cases were configured for an interval of 0.01 seconds. Again, the other tests were comparable. Each thread in this test computed the primes up to 25,000,000. I also did a test with a large number of threads, 100,000 threads, which is impossible without the fix. In this case each thread computed the primes only up to 10,000 (to make the runtime manageable). System time dominated, at 1546.968 seconds out of a total 2176.906 seconds (giving a user time of 629.938s). It received 147651 ticks for 0.015 seconds per tick, still quite accurate. There is obviously no comparable test without the fix. Signed-off-by: Frank Mayhar <fmayhar@google.com> Cc: Roland McGrath <roland@redhat.com> Cc: Alexey Dobriyan <adobriyan@gmail.com> Cc: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-09-13 00:54:39 +08:00
thread_group_cputime(p, &cputime);
cputime_to_timeval(cputime.utime, &prstatus->pr_utime);
cputime_to_timeval(cputime.stime, &prstatus->pr_stime);
} else {
cputime_t utime, stime;
task_cputime(p, &utime, &stime);
cputime_to_timeval(utime, &prstatus->pr_utime);
cputime_to_timeval(stime, &prstatus->pr_stime);
}
cputime_to_timeval(p->signal->cutime, &prstatus->pr_cutime);
cputime_to_timeval(p->signal->cstime, &prstatus->pr_cstime);
}
static int fill_psinfo(struct elf_prpsinfo *psinfo, struct task_struct *p,
struct mm_struct *mm)
{
const struct cred *cred;
unsigned int i, len;
/* first copy the parameters from user space */
memset(psinfo, 0, sizeof(struct elf_prpsinfo));
len = mm->arg_end - mm->arg_start;
if (len >= ELF_PRARGSZ)
len = ELF_PRARGSZ-1;
if (copy_from_user(&psinfo->pr_psargs,
(const char __user *)mm->arg_start, len))
return -EFAULT;
for(i = 0; i < len; i++)
if (psinfo->pr_psargs[i] == 0)
psinfo->pr_psargs[i] = ' ';
psinfo->pr_psargs[len] = 0;
rcu_read_lock();
psinfo->pr_ppid = task_pid_vnr(rcu_dereference(p->real_parent));
rcu_read_unlock();
psinfo->pr_pid = task_pid_vnr(p);
psinfo->pr_pgrp = task_pgrp_vnr(p);
psinfo->pr_sid = task_session_vnr(p);
i = p->state ? ffz(~p->state) + 1 : 0;
psinfo->pr_state = i;
psinfo->pr_sname = (i > 5) ? '.' : "RSDTZW"[i];
psinfo->pr_zomb = psinfo->pr_sname == 'Z';
psinfo->pr_nice = task_nice(p);
psinfo->pr_flag = p->flags;
rcu_read_lock();
cred = __task_cred(p);
SET_UID(psinfo->pr_uid, from_kuid_munged(cred->user_ns, cred->uid));
SET_GID(psinfo->pr_gid, from_kgid_munged(cred->user_ns, cred->gid));
rcu_read_unlock();
strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
return 0;
}
static void fill_auxv_note(struct memelfnote *note, struct mm_struct *mm)
{
elf_addr_t *auxv = (elf_addr_t *) mm->saved_auxv;
int i = 0;
do
i += 2;
while (auxv[i - 2] != AT_NULL);
fill_note(note, "CORE", NT_AUXV, i * sizeof(elf_addr_t), auxv);
}
static void fill_siginfo_note(struct memelfnote *note, user_siginfo_t *csigdata,
siginfo_t *siginfo)
{
mm_segment_t old_fs = get_fs();
set_fs(KERNEL_DS);
copy_siginfo_to_user((user_siginfo_t __user *) csigdata, siginfo);
set_fs(old_fs);
fill_note(note, "CORE", NT_SIGINFO, sizeof(*csigdata), csigdata);
}
#define MAX_FILE_NOTE_SIZE (4*1024*1024)
/*
* Format of NT_FILE note:
*
* long count -- how many files are mapped
* long page_size -- units for file_ofs
* array of [COUNT] elements of
* long start
* long end
* long file_ofs
* followed by COUNT filenames in ASCII: "FILE1" NUL "FILE2" NUL...
*/
static void fill_files_note(struct memelfnote *note)
{
struct vm_area_struct *vma;
unsigned count, size, names_ofs, remaining, n;
user_long_t *data;
user_long_t *start_end_ofs;
char *name_base, *name_curpos;
/* *Estimated* file count and total data size needed */
count = current->mm->map_count;
size = count * 64;
names_ofs = (2 + 3 * count) * sizeof(data[0]);
alloc:
if (size >= MAX_FILE_NOTE_SIZE) /* paranoia check */
goto err;
size = round_up(size, PAGE_SIZE);
data = vmalloc(size);
if (!data)
goto err;
start_end_ofs = data + 2;
name_base = name_curpos = ((char *)data) + names_ofs;
remaining = size - names_ofs;
count = 0;
for (vma = current->mm->mmap; vma != NULL; vma = vma->vm_next) {
struct file *file;
const char *filename;
file = vma->vm_file;
if (!file)
continue;
filename = d_path(&file->f_path, name_curpos, remaining);
if (IS_ERR(filename)) {
if (PTR_ERR(filename) == -ENAMETOOLONG) {
vfree(data);
size = size * 5 / 4;
goto alloc;
}
continue;
}
/* d_path() fills at the end, move name down */
/* n = strlen(filename) + 1: */
n = (name_curpos + remaining) - filename;
remaining = filename - name_curpos;
memmove(name_curpos, filename, n);
name_curpos += n;
*start_end_ofs++ = vma->vm_start;
*start_end_ofs++ = vma->vm_end;
*start_end_ofs++ = vma->vm_pgoff;
count++;
}
/* Now we know exact count of files, can store it */
data[0] = count;
data[1] = PAGE_SIZE;
/*
* Count usually is less than current->mm->map_count,
* we need to move filenames down.
*/
n = current->mm->map_count - count;
if (n != 0) {
unsigned shift_bytes = n * 3 * sizeof(data[0]);
memmove(name_base - shift_bytes, name_base,
name_curpos - name_base);
name_curpos -= shift_bytes;
}
size = name_curpos - (char *)data;
fill_note(note, "CORE", NT_FILE, size, data);
err: ;
}
#ifdef CORE_DUMP_USE_REGSET
#include <linux/regset.h>
struct elf_thread_core_info {
struct elf_thread_core_info *next;
struct task_struct *task;
struct elf_prstatus prstatus;
struct memelfnote notes[0];
};
struct elf_note_info {
struct elf_thread_core_info *thread;
struct memelfnote psinfo;
struct memelfnote signote;
struct memelfnote auxv;
struct memelfnote files;
user_siginfo_t csigdata;
size_t size;
int thread_notes;
};
/*
* When a regset has a writeback hook, we call it on each thread before
* dumping user memory. On register window machines, this makes sure the
* user memory backing the register data is up to date before we read it.
*/
static void do_thread_regset_writeback(struct task_struct *task,
const struct user_regset *regset)
{
if (regset->writeback)
regset->writeback(task, regset, 1);
}
#ifndef PR_REG_SIZE
#define PR_REG_SIZE(S) sizeof(S)
#endif
#ifndef PRSTATUS_SIZE
#define PRSTATUS_SIZE(S) sizeof(S)
#endif
#ifndef PR_REG_PTR
#define PR_REG_PTR(S) (&((S)->pr_reg))
#endif
#ifndef SET_PR_FPVALID
#define SET_PR_FPVALID(S, V) ((S)->pr_fpvalid = (V))
#endif
static int fill_thread_core_info(struct elf_thread_core_info *t,
const struct user_regset_view *view,
long signr, size_t *total)
{
unsigned int i;
/*
* NT_PRSTATUS is the one special case, because the regset data
* goes into the pr_reg field inside the note contents, rather
* than being the whole note contents. We fill the reset in here.
* We assume that regset 0 is NT_PRSTATUS.
*/
fill_prstatus(&t->prstatus, t->task, signr);
(void) view->regsets[0].get(t->task, &view->regsets[0],
0, PR_REG_SIZE(t->prstatus.pr_reg),
PR_REG_PTR(&t->prstatus), NULL);
fill_note(&t->notes[0], "CORE", NT_PRSTATUS,
PRSTATUS_SIZE(t->prstatus), &t->prstatus);
*total += notesize(&t->notes[0]);
do_thread_regset_writeback(t->task, &view->regsets[0]);
/*
* Each other regset might generate a note too. For each regset
* that has no core_note_type or is inactive, we leave t->notes[i]
* all zero and we'll know to skip writing it later.
*/
for (i = 1; i < view->n; ++i) {
const struct user_regset *regset = &view->regsets[i];
do_thread_regset_writeback(t->task, regset);
if (regset->core_note_type && regset->get &&
(!regset->active || regset->active(t->task, regset))) {
int ret;
size_t size = regset->n * regset->size;
void *data = kmalloc(size, GFP_KERNEL);
if (unlikely(!data))
return 0;
ret = regset->get(t->task, regset,
0, size, data, NULL);
if (unlikely(ret))
kfree(data);
else {
if (regset->core_note_type != NT_PRFPREG)
fill_note(&t->notes[i], "LINUX",
regset->core_note_type,
size, data);
else {
SET_PR_FPVALID(&t->prstatus, 1);
fill_note(&t->notes[i], "CORE",
NT_PRFPREG, size, data);
}
*total += notesize(&t->notes[i]);
}
}
}
return 1;
}
static int fill_note_info(struct elfhdr *elf, int phdrs,
struct elf_note_info *info,
siginfo_t *siginfo, struct pt_regs *regs)
{
struct task_struct *dump_task = current;
const struct user_regset_view *view = task_user_regset_view(dump_task);
struct elf_thread_core_info *t;
struct elf_prpsinfo *psinfo;
struct core_thread *ct;
unsigned int i;
info->size = 0;
info->thread = NULL;
psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
if (psinfo == NULL) {
info->psinfo.data = NULL; /* So we don't free this wrongly */
return 0;
}
fill_note(&info->psinfo, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
/*
* Figure out how many notes we're going to need for each thread.
*/
info->thread_notes = 0;
for (i = 0; i < view->n; ++i)
if (view->regsets[i].core_note_type != 0)
++info->thread_notes;
/*
* Sanity check. We rely on regset 0 being in NT_PRSTATUS,
* since it is our one special case.
*/
if (unlikely(info->thread_notes == 0) ||
unlikely(view->regsets[0].core_note_type != NT_PRSTATUS)) {
WARN_ON(1);
return 0;
}
/*
* Initialize the ELF file header.
*/
fill_elf_header(elf, phdrs,
view->e_machine, view->e_flags);
/*
* Allocate a structure for each thread.
*/
for (ct = &dump_task->mm->core_state->dumper; ct; ct = ct->next) {
t = kzalloc(offsetof(struct elf_thread_core_info,
notes[info->thread_notes]),
GFP_KERNEL);
if (unlikely(!t))
return 0;
t->task = ct->task;
if (ct->task == dump_task || !info->thread) {
t->next = info->thread;
info->thread = t;
} else {
/*
* Make sure to keep the original task at
* the head of the list.
*/
t->next = info->thread->next;
info->thread->next = t;
}
}
/*
* Now fill in each thread's information.
*/
for (t = info->thread; t != NULL; t = t->next)
if (!fill_thread_core_info(t, view, siginfo->si_signo, &info->size))
return 0;
/*
* Fill in the two process-wide notes.
*/
fill_psinfo(psinfo, dump_task->group_leader, dump_task->mm);
info->size += notesize(&info->psinfo);
fill_siginfo_note(&info->signote, &info->csigdata, siginfo);
info->size += notesize(&info->signote);
fill_auxv_note(&info->auxv, current->mm);
info->size += notesize(&info->auxv);
fill_files_note(&info->files);
info->size += notesize(&info->files);
return 1;
}
static size_t get_note_info_size(struct elf_note_info *info)
{
return info->size;
}
/*
* Write all the notes for each thread. When writing the first thread, the
* process-wide notes are interleaved after the first thread-specific note.
*/
static int write_note_info(struct elf_note_info *info,
struct file *file, loff_t *foffset)
{
bool first = 1;
struct elf_thread_core_info *t = info->thread;
do {
int i;
if (!writenote(&t->notes[0], file, foffset))
return 0;
if (first && !writenote(&info->psinfo, file, foffset))
return 0;
if (first && !writenote(&info->signote, file, foffset))
return 0;
if (first && !writenote(&info->auxv, file, foffset))
return 0;
if (first && !writenote(&info->files, file, foffset))
return 0;
for (i = 1; i < info->thread_notes; ++i)
if (t->notes[i].data &&
!writenote(&t->notes[i], file, foffset))
return 0;
first = 0;
t = t->next;
} while (t);
return 1;
}
static void free_note_info(struct elf_note_info *info)
{
struct elf_thread_core_info *threads = info->thread;
while (threads) {
unsigned int i;
struct elf_thread_core_info *t = threads;
threads = t->next;
WARN_ON(t->notes[0].data && t->notes[0].data != &t->prstatus);
for (i = 1; i < info->thread_notes; ++i)
kfree(t->notes[i].data);
kfree(t);
}
kfree(info->psinfo.data);
vfree(info->files.data);
}
#else
/* Here is the structure in which status of each thread is captured. */
struct elf_thread_status
{
struct list_head list;
struct elf_prstatus prstatus; /* NT_PRSTATUS */
elf_fpregset_t fpu; /* NT_PRFPREG */
struct task_struct *thread;
#ifdef ELF_CORE_COPY_XFPREGS
elf_fpxregset_t xfpu; /* ELF_CORE_XFPREG_TYPE */
#endif
struct memelfnote notes[3];
int num_notes;
};
/*
* In order to add the specific thread information for the elf file format,
* we need to keep a linked list of every threads pr_status and then create
* a single section for them in the final core file.
*/
static int elf_dump_thread_status(long signr, struct elf_thread_status *t)
{
int sz = 0;
struct task_struct *p = t->thread;
t->num_notes = 0;
fill_prstatus(&t->prstatus, p, signr);
elf_core_copy_task_regs(p, &t->prstatus.pr_reg);
fill_note(&t->notes[0], "CORE", NT_PRSTATUS, sizeof(t->prstatus),
&(t->prstatus));
t->num_notes++;
sz += notesize(&t->notes[0]);
if ((t->prstatus.pr_fpvalid = elf_core_copy_task_fpregs(p, NULL,
&t->fpu))) {
fill_note(&t->notes[1], "CORE", NT_PRFPREG, sizeof(t->fpu),
&(t->fpu));
t->num_notes++;
sz += notesize(&t->notes[1]);
}
#ifdef ELF_CORE_COPY_XFPREGS
if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
sizeof(t->xfpu), &t->xfpu);
t->num_notes++;
sz += notesize(&t->notes[2]);
}
#endif
return sz;
}
struct elf_note_info {
struct memelfnote *notes;
struct elf_prstatus *prstatus; /* NT_PRSTATUS */
struct elf_prpsinfo *psinfo; /* NT_PRPSINFO */
struct list_head thread_list;
elf_fpregset_t *fpu;
#ifdef ELF_CORE_COPY_XFPREGS
elf_fpxregset_t *xfpu;
#endif
user_siginfo_t csigdata;
int thread_status_size;
int numnote;
};
static int elf_note_info_init(struct elf_note_info *info)
{
memset(info, 0, sizeof(*info));
INIT_LIST_HEAD(&info->thread_list);
/* Allocate space for ELF notes */
info->notes = kmalloc(8 * sizeof(struct memelfnote), GFP_KERNEL);
if (!info->notes)
return 0;
info->psinfo = kmalloc(sizeof(*info->psinfo), GFP_KERNEL);
if (!info->psinfo)
return 0;
info->prstatus = kmalloc(sizeof(*info->prstatus), GFP_KERNEL);
if (!info->prstatus)
return 0;
info->fpu = kmalloc(sizeof(*info->fpu), GFP_KERNEL);
if (!info->fpu)
return 0;
#ifdef ELF_CORE_COPY_XFPREGS
info->xfpu = kmalloc(sizeof(*info->xfpu), GFP_KERNEL);
if (!info->xfpu)
return 0;
#endif
return 1;
}
static int fill_note_info(struct elfhdr *elf, int phdrs,
struct elf_note_info *info,
siginfo_t *siginfo, struct pt_regs *regs)
{
struct list_head *t;
if (!elf_note_info_init(info))
return 0;
if (siginfo->si_signo) {
struct core_thread *ct;
struct elf_thread_status *ets;
for (ct = current->mm->core_state->dumper.next;
ct; ct = ct->next) {
ets = kzalloc(sizeof(*ets), GFP_KERNEL);
if (!ets)
return 0;
ets->thread = ct->task;
list_add(&ets->list, &info->thread_list);
}
list_for_each(t, &info->thread_list) {
int sz;
ets = list_entry(t, struct elf_thread_status, list);
sz = elf_dump_thread_status(siginfo->si_signo, ets);
info->thread_status_size += sz;
}
}
/* now collect the dump for the current */
memset(info->prstatus, 0, sizeof(*info->prstatus));
fill_prstatus(info->prstatus, current, siginfo->si_signo);
elf_core_copy_regs(&info->prstatus->pr_reg, regs);
/* Set up header */
fill_elf_header(elf, phdrs, ELF_ARCH, ELF_CORE_EFLAGS);
/*
* Set up the notes in similar form to SVR4 core dumps made
* with info from their /proc.
*/
fill_note(info->notes + 0, "CORE", NT_PRSTATUS,
sizeof(*info->prstatus), info->prstatus);
fill_psinfo(info->psinfo, current->group_leader, current->mm);
fill_note(info->notes + 1, "CORE", NT_PRPSINFO,
sizeof(*info->psinfo), info->psinfo);
fill_siginfo_note(info->notes + 2, &info->csigdata, siginfo);
fill_auxv_note(info->notes + 3, current->mm);
fill_files_note(info->notes + 4);
info->numnote = 5;
/* Try to dump the FPU. */
info->prstatus->pr_fpvalid = elf_core_copy_task_fpregs(current, regs,
info->fpu);
if (info->prstatus->pr_fpvalid)
fill_note(info->notes + info->numnote++,
"CORE", NT_PRFPREG, sizeof(*info->fpu), info->fpu);
#ifdef ELF_CORE_COPY_XFPREGS
if (elf_core_copy_task_xfpregs(current, info->xfpu))
fill_note(info->notes + info->numnote++,
"LINUX", ELF_CORE_XFPREG_TYPE,
sizeof(*info->xfpu), info->xfpu);
#endif
return 1;
}
static size_t get_note_info_size(struct elf_note_info *info)
{
int sz = 0;
int i;
for (i = 0; i < info->numnote; i++)
sz += notesize(info->notes + i);
sz += info->thread_status_size;
return sz;
}
static int write_note_info(struct elf_note_info *info,
struct file *file, loff_t *foffset)
{
int i;
struct list_head *t;
for (i = 0; i < info->numnote; i++)
if (!writenote(info->notes + i, file, foffset))
return 0;
/* write out the thread status notes section */
list_for_each(t, &info->thread_list) {
struct elf_thread_status *tmp =
list_entry(t, struct elf_thread_status, list);
for (i = 0; i < tmp->num_notes; i++)
if (!writenote(&tmp->notes[i], file, foffset))
return 0;
}
return 1;
}
static void free_note_info(struct elf_note_info *info)
{
while (!list_empty(&info->thread_list)) {
struct list_head *tmp = info->thread_list.next;
list_del(tmp);
kfree(list_entry(tmp, struct elf_thread_status, list));
}
/* Free data allocated by fill_files_note(): */
vfree(info->notes[4].data);
kfree(info->prstatus);
kfree(info->psinfo);
kfree(info->notes);
kfree(info->fpu);
#ifdef ELF_CORE_COPY_XFPREGS
kfree(info->xfpu);
#endif
}
#endif
static struct vm_area_struct *first_vma(struct task_struct *tsk,
struct vm_area_struct *gate_vma)
{
struct vm_area_struct *ret = tsk->mm->mmap;
if (ret)
return ret;
return gate_vma;
}
/*
* Helper function for iterating across a vma list. It ensures that the caller
* will visit `gate_vma' prior to terminating the search.
*/
static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
struct vm_area_struct *gate_vma)
{
struct vm_area_struct *ret;
ret = this_vma->vm_next;
if (ret)
return ret;
if (this_vma == gate_vma)
return NULL;
return gate_vma;
}
static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
elf_addr_t e_shoff, int segs)
{
elf->e_shoff = e_shoff;
elf->e_shentsize = sizeof(*shdr4extnum);
elf->e_shnum = 1;
elf->e_shstrndx = SHN_UNDEF;
memset(shdr4extnum, 0, sizeof(*shdr4extnum));
shdr4extnum->sh_type = SHT_NULL;
shdr4extnum->sh_size = elf->e_shnum;
shdr4extnum->sh_link = elf->e_shstrndx;
shdr4extnum->sh_info = segs;
}
static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
unsigned long mm_flags)
{
struct vm_area_struct *vma;
size_t size = 0;
for (vma = first_vma(current, gate_vma); vma != NULL;
vma = next_vma(vma, gate_vma))
size += vma_dump_size(vma, mm_flags);
return size;
}
/*
* Actual dumper
*
* This is a two-pass process; first we find the offsets of the bits,
* and then they are actually written out. If we run out of core limit
* we just truncate.
*/
static int elf_core_dump(struct coredump_params *cprm)
{
int has_dumped = 0;
mm_segment_t fs;
int segs;
size_t size = 0;
struct vm_area_struct *vma, *gate_vma;
struct elfhdr *elf = NULL;
[PATCH] Support piping into commands in /proc/sys/kernel/core_pattern Using the infrastructure created in previous patches implement support to pipe core dumps into programs. This is done by overloading the existing core_pattern sysctl with a new syntax: |program When the first character of the pattern is a '|' the kernel will instead threat the rest of the pattern as a command to run. The core dump will be written to the standard input of that program instead of to a file. This is useful for having automatic core dump analysis without filling up disks. The program can do some simple analysis and save only a summary of the core dump. The core dump proces will run with the privileges and in the name space of the process that caused the core dump. I also increased the core pattern size to 128 bytes so that longer command lines fit. Most of the changes comes from allowing core dumps without seeks. They are fairly straight forward though. One small incompatibility is that if someone had a core pattern previously that started with '|' they will get suddenly new behaviour. I think that's unlikely to be a real problem though. Additional background: > Very nice, do you happen to have a program that can accept this kind of > input for crash dumps? I'm guessing that the embedded people will > really want this functionality. I had a cheesy demo/prototype. Basically it wrote the dump to a file again, ran gdb on it to get a backtrace and wrote the summary to a shared directory. Then there was a simple CGI script to generate a "top 10" crashes HTML listing. Unfortunately this still had the disadvantage to needing full disk space for a dump except for deleting it afterwards (in fact it was worse because over the pipe holes didn't work so if you have a holey address map it would require more space). Fortunately gdb seems to be happy to handle /proc/pid/fd/xxx input pipes as cores (at least it worked with zsh's =(cat core) syntax), so it would be likely possible to do it without temporary space with a simple wrapper that calls it in the right way. I ran out of time before doing that though. The demo prototype scripts weren't very good. If there is really interest I can dig them out (they are currently on a laptop disk on the desk with the laptop itself being in service), but I would recommend to rewrite them for any serious application of this and fix the disk space problem. Also to be really useful it should probably find a way to automatically fetch the debuginfos (I cheated and just installed them in advance). If nobody else does it I can probably do the rewrite myself again at some point. My hope at some point was that desktops would support it in their builtin crash reporters, but at least the KDE people I talked too seemed to be happy with their user space only solution. Alan sayeth: I don't believe that piping as such as neccessarily the right model, but the ability to intercept and processes core dumps from user space is asked for by many enterprise users as well. They want to know about, capture, analyse and process core dumps, often centrally and in automated form. [akpm@osdl.org: loff_t != unsigned long] Signed-off-by: Andi Kleen <ak@suse.de> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-10-01 14:29:28 +08:00
loff_t offset = 0, dataoff, foffset;
struct elf_note_info info;
struct elf_phdr *phdr4note = NULL;
struct elf_shdr *shdr4extnum = NULL;
Elf_Half e_phnum;
elf_addr_t e_shoff;
/*
* We no longer stop all VM operations.
*
* This is because those proceses that could possibly change map_count
* or the mmap / vma pages are now blocked in do_exit on current
* finishing this core dump.
*
* Only ptrace can touch these memory addresses, but it doesn't change
* the map_count or the pages allocated. So no possibility of crashing
* exists while dumping the mm->vm_next areas to the core file.
*/
/* alloc memory for large data structures: too large to be on stack */
elf = kmalloc(sizeof(*elf), GFP_KERNEL);
if (!elf)
goto out;
/*
* The number of segs are recored into ELF header as 16bit value.
* Please check DEFAULT_MAX_MAP_COUNT definition when you modify here.
*/
segs = current->mm->map_count;
segs += elf_core_extra_phdrs();
gate_vma = get_gate_vma(current->mm);
if (gate_vma != NULL)
segs++;
/* for notes section */
segs++;
/* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
* this, kernel supports extended numbering. Have a look at
* include/linux/elf.h for further information. */
e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
/*
* Collect all the non-memory information about the process for the
* notes. This also sets up the file header.
*/
if (!fill_note_info(elf, e_phnum, &info, cprm->siginfo, cprm->regs))
goto cleanup;
has_dumped = 1;
fs = get_fs();
set_fs(KERNEL_DS);
offset += sizeof(*elf); /* Elf header */
offset += segs * sizeof(struct elf_phdr); /* Program headers */
foffset = offset;
/* Write notes phdr entry */
{
size_t sz = get_note_info_size(&info);
sz += elf_coredump_extra_notes_size();
phdr4note = kmalloc(sizeof(*phdr4note), GFP_KERNEL);
if (!phdr4note)
goto end_coredump;
fill_elf_note_phdr(phdr4note, sz, offset);
offset += sz;
}
dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
offset += elf_core_vma_data_size(gate_vma, cprm->mm_flags);
offset += elf_core_extra_data_size();
e_shoff = offset;
if (e_phnum == PN_XNUM) {
shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
if (!shdr4extnum)
goto end_coredump;
fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
}
offset = dataoff;
size += sizeof(*elf);
if (size > cprm->limit || !dump_write(cprm->file, elf, sizeof(*elf)))
goto end_coredump;
size += sizeof(*phdr4note);
if (size > cprm->limit
|| !dump_write(cprm->file, phdr4note, sizeof(*phdr4note)))
goto end_coredump;
/* Write program headers for segments dump */
for (vma = first_vma(current, gate_vma); vma != NULL;
vma = next_vma(vma, gate_vma)) {
struct elf_phdr phdr;
phdr.p_type = PT_LOAD;
phdr.p_offset = offset;
phdr.p_vaddr = vma->vm_start;
phdr.p_paddr = 0;
phdr.p_filesz = vma_dump_size(vma, cprm->mm_flags);
phdr.p_memsz = vma->vm_end - vma->vm_start;
offset += phdr.p_filesz;
phdr.p_flags = vma->vm_flags & VM_READ ? PF_R : 0;
if (vma->vm_flags & VM_WRITE)
phdr.p_flags |= PF_W;
if (vma->vm_flags & VM_EXEC)
phdr.p_flags |= PF_X;
phdr.p_align = ELF_EXEC_PAGESIZE;
size += sizeof(phdr);
if (size > cprm->limit
|| !dump_write(cprm->file, &phdr, sizeof(phdr)))
goto end_coredump;
}
if (!elf_core_write_extra_phdrs(cprm->file, offset, &size, cprm->limit))
goto end_coredump;
/* write out the notes section */
if (!write_note_info(&info, cprm->file, &foffset))
goto end_coredump;
if (elf_coredump_extra_notes_write(cprm->file, &foffset))
goto end_coredump;
[PATCH] Support piping into commands in /proc/sys/kernel/core_pattern Using the infrastructure created in previous patches implement support to pipe core dumps into programs. This is done by overloading the existing core_pattern sysctl with a new syntax: |program When the first character of the pattern is a '|' the kernel will instead threat the rest of the pattern as a command to run. The core dump will be written to the standard input of that program instead of to a file. This is useful for having automatic core dump analysis without filling up disks. The program can do some simple analysis and save only a summary of the core dump. The core dump proces will run with the privileges and in the name space of the process that caused the core dump. I also increased the core pattern size to 128 bytes so that longer command lines fit. Most of the changes comes from allowing core dumps without seeks. They are fairly straight forward though. One small incompatibility is that if someone had a core pattern previously that started with '|' they will get suddenly new behaviour. I think that's unlikely to be a real problem though. Additional background: > Very nice, do you happen to have a program that can accept this kind of > input for crash dumps? I'm guessing that the embedded people will > really want this functionality. I had a cheesy demo/prototype. Basically it wrote the dump to a file again, ran gdb on it to get a backtrace and wrote the summary to a shared directory. Then there was a simple CGI script to generate a "top 10" crashes HTML listing. Unfortunately this still had the disadvantage to needing full disk space for a dump except for deleting it afterwards (in fact it was worse because over the pipe holes didn't work so if you have a holey address map it would require more space). Fortunately gdb seems to be happy to handle /proc/pid/fd/xxx input pipes as cores (at least it worked with zsh's =(cat core) syntax), so it would be likely possible to do it without temporary space with a simple wrapper that calls it in the right way. I ran out of time before doing that though. The demo prototype scripts weren't very good. If there is really interest I can dig them out (they are currently on a laptop disk on the desk with the laptop itself being in service), but I would recommend to rewrite them for any serious application of this and fix the disk space problem. Also to be really useful it should probably find a way to automatically fetch the debuginfos (I cheated and just installed them in advance). If nobody else does it I can probably do the rewrite myself again at some point. My hope at some point was that desktops would support it in their builtin crash reporters, but at least the KDE people I talked too seemed to be happy with their user space only solution. Alan sayeth: I don't believe that piping as such as neccessarily the right model, but the ability to intercept and processes core dumps from user space is asked for by many enterprise users as well. They want to know about, capture, analyse and process core dumps, often centrally and in automated form. [akpm@osdl.org: loff_t != unsigned long] Signed-off-by: Andi Kleen <ak@suse.de> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-10-01 14:29:28 +08:00
/* Align to page */
if (!dump_seek(cprm->file, dataoff - foffset))
goto end_coredump;
for (vma = first_vma(current, gate_vma); vma != NULL;
vma = next_vma(vma, gate_vma)) {
unsigned long addr;
unsigned long end;
end = vma->vm_start + vma_dump_size(vma, cprm->mm_flags);
for (addr = vma->vm_start; addr < end; addr += PAGE_SIZE) {
struct page *page;
int stop;
page = get_dump_page(addr);
if (page) {
void *kaddr = kmap(page);
stop = ((size += PAGE_SIZE) > cprm->limit) ||
!dump_write(cprm->file, kaddr,
PAGE_SIZE);
kunmap(page);
page_cache_release(page);
} else
stop = !dump_seek(cprm->file, PAGE_SIZE);
if (stop)
goto end_coredump;
}
}
if (!elf_core_write_extra_data(cprm->file, &size, cprm->limit))
goto end_coredump;
if (e_phnum == PN_XNUM) {
size += sizeof(*shdr4extnum);
if (size > cprm->limit
|| !dump_write(cprm->file, shdr4extnum,
sizeof(*shdr4extnum)))
goto end_coredump;
}
end_coredump:
set_fs(fs);
cleanup:
free_note_info(&info);
kfree(shdr4extnum);
kfree(phdr4note);
kfree(elf);
out:
return has_dumped;
}
#endif /* CONFIG_ELF_CORE */
static int __init init_elf_binfmt(void)
{
register_binfmt(&elf_format);
return 0;
}
static void __exit exit_elf_binfmt(void)
{
/* Remove the COFF and ELF loaders. */
unregister_binfmt(&elf_format);
}
core_initcall(init_elf_binfmt);
module_exit(exit_elf_binfmt);
MODULE_LICENSE("GPL");