1
linux/fs/binfmt_elf.c

1794 lines
48 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/stat.h>
#include <linux/time.h>
#include <linux/mm.h>
#include <linux/mman.h>
#include <linux/a.out.h>
#include <linux/errno.h>
#include <linux/signal.h>
#include <linux/binfmts.h>
#include <linux/string.h>
#include <linux/file.h>
#include <linux/fcntl.h>
#include <linux/ptrace.h>
#include <linux/slab.h>
#include <linux/shm.h>
#include <linux/personality.h>
#include <linux/elfcore.h>
#include <linux/init.h>
#include <linux/highuid.h>
#include <linux/smp.h>
#include <linux/compiler.h>
#include <linux/highmem.h>
#include <linux/pagemap.h>
#include <linux/security.h>
#include <linux/syscalls.h>
#include <linux/random.h>
#include <linux/elf.h>
#include <linux/utsname.h>
#include <asm/uaccess.h>
#include <asm/param.h>
#include <asm/page.h>
static int load_elf_binary(struct linux_binprm *bprm, struct pt_regs *regs);
static int load_elf_library(struct file *);
static unsigned long elf_map (struct file *, unsigned long, struct elf_phdr *, int, int);
/*
* If we don't support core dumping, then supply a NULL so we
* don't even try.
*/
#if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file);
#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,
.hasvdso = 1
};
#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;
down_write(&current->mm->mmap_sem);
addr = do_brk(start, end - start);
up_write(&current->mm->mmap_sem);
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 litle 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
static int
create_elf_tables(struct linux_binprm *bprm, struct elfhdr *exec,
int interp_aout, 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;
const char *k_platform = ELF_PLATFORM;
int items;
elf_addr_t *elf_info;
int ei_index = 0;
struct task_struct *tsk = current;
struct vm_area_struct *vma;
/*
* 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;
/*
* 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);
u_platform = (elf_addr_t __user *)STACK_ALLOC(p, len);
if (__copy_to_user(u_platform, k_platform, len))
return -EFAULT;
}
/* Create the ELF interpreter info */
elf_info = (elf_addr_t *)current->mm->saved_auxv;
#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.
*/
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, tsk->uid);
NEW_AUX_ENT(AT_EUID, tsk->euid);
NEW_AUX_ENT(AT_GID, tsk->gid);
NEW_AUX_ENT(AT_EGID, tsk->egid);
NEW_AUX_ENT(AT_SECURE, security_bprm_secureexec(bprm));
if (k_platform) {
NEW_AUX_ENT(AT_PLATFORM,
(elf_addr_t)(unsigned long)u_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);
if (interp_aout) {
items += 3; /* a.out interpreters require argv & envp too */
} else {
items += 1; /* ELF interpreters only put argc on the stack */
}
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;
if (interp_aout) {
argv = sp + 2;
envp = argv + argc + 1;
if (__put_user((elf_addr_t)(unsigned long)argv, sp++) ||
__put_user((elf_addr_t)(unsigned long)envp, sp++))
return -EFAULT;
} else {
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 0;
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 0;
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 map_addr;
unsigned long pageoffset = ELF_PAGEOFFSET(eppnt->p_vaddr);
down_write(&current->mm->mmap_sem);
/* mmap() will return -EINVAL if given a zero size, but a
* segment with zero filesize is perfectly valid */
if (eppnt->p_filesz + pageoffset)
map_addr = do_mmap(filep, ELF_PAGESTART(addr),
eppnt->p_filesz + pageoffset, prot, type,
eppnt->p_offset - pageoffset);
else
map_addr = ELF_PAGESTART(addr);
up_write(&current->mm->mmap_sem);
return(map_addr);
}
#endif /* !elf_map */
/* 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_load_addr)
{
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;
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;
}
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;
map_addr = elf_map(interpreter, load_addr + vaddr,
eppnt, elf_prot, elf_type);
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 00:24:14 -07:00
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;
}
}
/*
* 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 */
if (last_bss > elf_bss) {
down_write(&current->mm->mmap_sem);
error = do_brk(elf_bss, last_bss - elf_bss);
up_write(&current->mm->mmap_sem);
if (BAD_ADDR(error))
goto out_close;
}
*interp_load_addr = load_addr;
error = ((unsigned long)interp_elf_ex->e_entry) + load_addr;
out_close:
kfree(elf_phdata);
out:
return error;
}
static unsigned long load_aout_interp(struct exec *interp_ex,
struct file *interpreter)
{
unsigned long text_data, elf_entry = ~0UL;
char __user * addr;
loff_t offset;
current->mm->end_code = interp_ex->a_text;
text_data = interp_ex->a_text + interp_ex->a_data;
current->mm->end_data = text_data;
current->mm->brk = interp_ex->a_bss + text_data;
switch (N_MAGIC(*interp_ex)) {
case OMAGIC:
offset = 32;
addr = (char __user *)0;
break;
case ZMAGIC:
case QMAGIC:
offset = N_TXTOFF(*interp_ex);
addr = (char __user *)N_TXTADDR(*interp_ex);
break;
default:
goto out;
}
down_write(&current->mm->mmap_sem);
do_brk(0, text_data);
up_write(&current->mm->mmap_sem);
if (!interpreter->f_op || !interpreter->f_op->read)
goto out;
if (interpreter->f_op->read(interpreter, addr, text_data, &offset) < 0)
goto out;
flush_icache_range((unsigned long)addr,
(unsigned long)addr + text_data);
down_write(&current->mm->mmap_sem);
do_brk(ELF_PAGESTART(text_data + ELF_MIN_ALIGN - 1),
interp_ex->a_bss);
up_write(&current->mm->mmap_sem);
elf_entry = interp_ex->a_entry;
out:
return elf_entry;
}
/*
* 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_AOUT 1
#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 pt_regs *regs)
{
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 int interpreter_type = INTERPRETER_NONE;
unsigned char ibcs2_interpreter = 0;
unsigned long error;
struct elf_phdr *elf_ppnt, *elf_phdata;
unsigned long elf_bss, elf_brk;
int elf_exec_fileno;
int retval, i;
unsigned int size;
unsigned long elf_entry, interp_load_addr = 0;
unsigned long start_code, end_code, start_data, end_data;
unsigned long reloc_func_desc = 0;
char passed_fileno[6];
struct files_struct *files;
int executable_stack = EXSTACK_DEFAULT;
unsigned long def_flags = 0;
struct {
struct elfhdr elf_ex;
struct elfhdr interp_elf_ex;
struct exec interp_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;
}
files = current->files; /* Refcounted so ok */
retval = unshare_files();
if (retval < 0)
goto out_free_ph;
if (files == current->files) {
put_files_struct(files);
files = NULL;
}
/* exec will make our files private anyway, but for the a.out
loader stuff we need to do it earlier */
retval = get_unused_fd();
if (retval < 0)
goto out_free_fh;
get_file(bprm->file);
fd_install(elf_exec_fileno = retval, bprm->file);
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_file;
retval = -ENOMEM;
elf_interpreter = kmalloc(elf_ppnt->p_filesz,
GFP_KERNEL);
if (!elf_interpreter)
goto out_free_file;
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;
/* If the program interpreter is one of these two,
* then assume an iBCS2 image. Otherwise assume
* a native linux image.
*/
if (strcmp(elf_interpreter,"/usr/lib/libc.so.1") == 0 ||
strcmp(elf_interpreter,"/usr/lib/ld.so.1") == 0)
ibcs2_interpreter = 1;
/*
* The early SET_PERSONALITY here is so that the lookup
* for the interpreter happens in the namespace of the
* to-be-execed image. SET_PERSONALITY can select an
* alternate root.
*
* However, SET_PERSONALITY is NOT allowed to switch
* this task into the new images's memory mapping
* policy - that is, TASK_SIZE must still evaluate to
* that which is appropriate to the execing application.
* This is because exit_mmap() needs to have TASK_SIZE
* evaluate to the size of the old image.
*
* So if (say) a 64-bit application is execing a 32-bit
* application it is the architecture's responsibility
* to defer changing the value of TASK_SIZE until the
* switch really is going to happen - do this in
* flush_thread(). - akpm
*/
SET_PERSONALITY(loc->elf_ex, ibcs2_interpreter);
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.
*/
if (file_permission(interpreter, MAY_READ) < 0)
bprm->interp_flags |= BINPRM_FLAGS_ENFORCE_NONDUMP;
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_ex = *((struct exec *)bprm->buf);
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) {
interpreter_type = INTERPRETER_ELF | INTERPRETER_AOUT;
/* Now figure out which format our binary is */
if ((N_MAGIC(loc->interp_ex) != OMAGIC) &&
(N_MAGIC(loc->interp_ex) != ZMAGIC) &&
(N_MAGIC(loc->interp_ex) != QMAGIC))
interpreter_type = INTERPRETER_ELF;
if (memcmp(loc->interp_elf_ex.e_ident, ELFMAG, SELFMAG) != 0)
interpreter_type &= ~INTERPRETER_ELF;
retval = -ELIBBAD;
if (!interpreter_type)
goto out_free_dentry;
/* Make sure only one type was selected */
if ((interpreter_type & INTERPRETER_ELF) &&
interpreter_type != INTERPRETER_ELF) {
// FIXME - ratelimit this before re-enabling
// printk(KERN_WARNING "ELF: Ambiguous type, using ELF\n");
interpreter_type = INTERPRETER_ELF;
}
/* Verify the interpreter has a valid arch */
if ((interpreter_type == INTERPRETER_ELF) &&
!elf_check_arch(&loc->interp_elf_ex))
goto out_free_dentry;
} else {
/* Executables without an interpreter also need a personality */
SET_PERSONALITY(loc->elf_ex, ibcs2_interpreter);
}
/* OK, we are done with that, now set up the arg stuff,
and then start this sucker up */
if ((!bprm->sh_bang) && (interpreter_type == INTERPRETER_AOUT)) {
char *passed_p = passed_fileno;
sprintf(passed_fileno, "%d", elf_exec_fileno);
if (elf_interpreter) {
retval = copy_strings_kernel(1, &passed_p, bprm);
if (retval)
goto out_free_dentry;
bprm->argc++;
}
}
/* Flush all traces of the currently running executable */
retval = flush_old_exec(bprm);
if (retval)
goto out_free_dentry;
/* Discard our unneeded old files struct */
if (files) {
put_files_struct(files);
files = NULL;
}
/* OK, This is the point of no return */
current->flags &= ~PF_FORKNOEXEC;
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, ibcs2_interpreter);
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;
arch_pick_mmap_layout(current->mm);
/* 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>
2005-06-21 17:14:49 -07:00
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 mmaping the ELF image into
the correct location in memory. At this point, we assume that
the image should be loaded at fixed address, not at a variable
address. */
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. */
load_bias = ELF_PAGESTART(ELF_ET_DYN_BASE - vaddr);
}
error = elf_map(bprm->file, load_bias + vaddr, elf_ppnt,
elf_prot, elf_flags);
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 00:24:14 -07:00
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) {
if (interpreter_type == INTERPRETER_AOUT)
elf_entry = load_aout_interp(&loc->interp_ex,
interpreter);
else
elf_entry = load_elf_interp(&loc->interp_elf_ex,
interpreter,
&interp_load_addr);
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 00:24:14 -07:00
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 00:24:14 -07:00
force_sig(SIGSEGV, current);
retval = -EINVAL;
goto out_free_dentry;
}
}
kfree(elf_phdata);
if (interpreter_type != INTERPRETER_AOUT)
sys_close(elf_exec_fileno);
set_binfmt(&elf_format);
#ifdef ARCH_HAS_SETUP_ADDITIONAL_PAGES
retval = arch_setup_additional_pages(bprm, executable_stack);
if (retval < 0) {
send_sig(SIGKILL, current, 0);
goto out;
}
#endif /* ARCH_HAS_SETUP_ADDITIONAL_PAGES */
compute_creds(bprm);
current->flags &= ~PF_FORKNOEXEC;
retval = create_elf_tables(bprm, &loc->elf_ex,
(interpreter_type == INTERPRETER_AOUT),
load_addr, interp_load_addr);
if (retval < 0) {
send_sig(SIGKILL, current, 0);
goto out;
}
/* N.B. passed_fileno might not be initialized? */
if (interpreter_type == INTERPRETER_AOUT)
current->mm->arg_start += strlen(passed_fileno) + 1;
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;
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. */
down_write(&current->mm->mmap_sem);
error = do_mmap(NULL, 0, PAGE_SIZE, PROT_READ | PROT_EXEC,
MAP_FIXED | MAP_PRIVATE, 0);
up_write(&current->mm->mmap_sem);
}
#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);
if (unlikely(current->ptrace & PT_PTRACED)) {
if (current->ptrace & PT_TRACE_EXEC)
ptrace_notify ((PTRACE_EVENT_EXEC << 8) | SIGTRAP);
else
send_sig(SIGTRAP, current, 0);
}
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_file:
sys_close(elf_exec_fileno);
out_free_fh:
if (files)
reset_files_struct(current, files);
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. */
down_write(&current->mm->mmap_sem);
error = do_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)));
up_write(&current->mm->mmap_sem);
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) {
down_write(&current->mm->mmap_sem);
do_brk(len, bss - len);
up_write(&current->mm->mmap_sem);
}
error = 0;
out_free_ph:
kfree(elf_phdata);
out:
return error;
}
/*
* Note that some platforms still use traditional core dumps and not
* the ELF core dump. Each platform can select it as appropriate.
*/
#if defined(USE_ELF_CORE_DUMP) && defined(CONFIG_ELF_CORE)
/*
* ELF core dumper
*
* Modelled on fs/exec.c:aout_core_dump()
* Jeremy Fitzhardinge <jeremy@sw.oz.au>
*/
/*
* These are the only things you should do on a core-file: use only these
* functions to write out all the necessary info.
*/
static int dump_write(struct file *file, const void *addr, int nr)
{
return file->f_op->write(file, addr, nr, &file->f_pos) == nr;
}
static int dump_seek(struct file *file, loff_t off)
{
[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>
2006-09-30 23:29:28 -07:00
if (file->f_op->llseek && file->f_op->llseek != no_llseek) {
if (file->f_op->llseek(file, off, SEEK_CUR) < 0)
return 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>
2006-09-30 23:29:28 -07:00
} else {
char *buf = (char *)get_zeroed_page(GFP_KERNEL);
if (!buf)
return 0;
while (off > 0) {
unsigned long n = off;
if (n > PAGE_SIZE)
n = PAGE_SIZE;
if (!dump_write(file, buf, n))
return 0;
off -= n;
}
free_page((unsigned long)buf);
}
return 1;
}
/*
* Decide whether a segment is worth dumping; default is yes to be
* sure (missing info is worse than too much; etc).
* Personally I'd include everything, and use the coredump limit...
*
* I think we should skip something. But I am not sure how. H.J.
*/
static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
{
/* The vma can be set up to tell us the answer directly. */
if (vma->vm_flags & VM_ALWAYSDUMP)
return 1;
/* Do not dump I/O mapped devices or special mappings */
if (vma->vm_flags & (VM_IO | VM_RESERVED))
return 0;
/* By default, dump shared memory if mapped from an anonymous file. */
if (vma->vm_flags & VM_SHARED) {
if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0)
return test_bit(MMF_DUMP_ANON_SHARED, &mm_flags);
else
return test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags);
}
/* By default, if it hasn't been written to, don't write it out. */
if (!vma->anon_vma)
return test_bit(MMF_DUMP_MAPPED_PRIVATE, &mm_flags);
return test_bit(MMF_DUMP_ANON_PRIVATE, &mm_flags);
}
/* 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>
2006-09-30 23:29:28 -07:00
#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>
2006-09-30 23:29:28 -07:00
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>
2006-09-30 23:29:28 -07:00
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>
2006-09-30 23:29:28 -07:00
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>
2006-09-30 23:29:28 -07:00
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
#define DUMP_WRITE(addr, nr) \
if ((size += (nr)) > limit || !dump_write(file, (addr), (nr))) \
goto end_coredump;
#define DUMP_SEEK(off) \
if (!dump_seek(file, (off))) \
goto end_coredump;
static void fill_elf_header(struct elfhdr *elf, int segs)
{
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;
memset(elf->e_ident+EI_PAD, 0, EI_NIDENT-EI_PAD);
elf->e_type = ET_CORE;
elf->e_machine = ELF_ARCH;
elf->e_version = EV_CURRENT;
elf->e_entry = 0;
elf->e_phoff = sizeof(struct elfhdr);
elf->e_shoff = 0;
elf->e_flags = ELF_CORE_EFLAGS;
elf->e_ehsize = sizeof(struct elfhdr);
elf->e_phentsize = sizeof(struct elf_phdr);
elf->e_phnum = segs;
elf->e_shentsize = 0;
elf->e_shnum = 0;
elf->e_shstrndx = 0;
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];
prstatus->pr_pid = p->pid;
prstatus->pr_ppid = p->parent->pid;
prstatus->pr_pgrp = process_group(p);
prstatus->pr_sid = process_session(p);
if (thread_group_leader(p)) {
/*
* This is the record for the group leader. Add in the
* cumulative times of previous dead threads. This total
* won't include the time of each live thread whose state
* is included in the core dump. The final total reported
* to our parent process when it calls wait4 will include
* those sums as well as the little bit more time it takes
* this and each other thread to finish dying after the
* core dump synchronization phase.
*/
cputime_to_timeval(cputime_add(p->utime, p->signal->utime),
&prstatus->pr_utime);
cputime_to_timeval(cputime_add(p->stime, p->signal->stime),
&prstatus->pr_stime);
} else {
cputime_to_timeval(p->utime, &prstatus->pr_utime);
cputime_to_timeval(p->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)
{
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;
psinfo->pr_pid = p->pid;
psinfo->pr_ppid = p->parent->pid;
psinfo->pr_pgrp = process_group(p);
psinfo->pr_sid = process_session(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;
SET_UID(psinfo->pr_uid, p->uid);
SET_GID(psinfo->pr_gid, p->gid);
strncpy(psinfo->pr_fname, p->comm, sizeof(psinfo->pr_fname));
return 0;
}
/* 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; /* NT_PRXFPREG */
#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", NT_PRXFPREG, sizeof(t->xfpu),
&t->xfpu);
t->num_notes++;
sz += notesize(&t->notes[2]);
}
#endif
return sz;
}
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;
}
/*
* 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(long signr, struct pt_regs *regs, struct file *file)
{
#define NUM_NOTES 6
int has_dumped = 0;
mm_segment_t fs;
int segs;
size_t size = 0;
int i;
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>
2006-09-30 23:29:28 -07:00
loff_t offset = 0, dataoff, foffset;
unsigned long limit = current->signal->rlim[RLIMIT_CORE].rlim_cur;
int numnote;
struct memelfnote *notes = NULL;
struct elf_prstatus *prstatus = NULL; /* NT_PRSTATUS */
struct elf_prpsinfo *psinfo = NULL; /* NT_PRPSINFO */
struct task_struct *g, *p;
LIST_HEAD(thread_list);
struct list_head *t;
elf_fpregset_t *fpu = NULL;
#ifdef ELF_CORE_COPY_XFPREGS
elf_fpxregset_t *xfpu = NULL;
#endif
int thread_status_size = 0;
elf_addr_t *auxv;
unsigned long mm_flags;
/*
* 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 cleanup;
prstatus = kmalloc(sizeof(*prstatus), GFP_KERNEL);
if (!prstatus)
goto cleanup;
psinfo = kmalloc(sizeof(*psinfo), GFP_KERNEL);
if (!psinfo)
goto cleanup;
notes = kmalloc(NUM_NOTES * sizeof(struct memelfnote), GFP_KERNEL);
if (!notes)
goto cleanup;
fpu = kmalloc(sizeof(*fpu), GFP_KERNEL);
if (!fpu)
goto cleanup;
#ifdef ELF_CORE_COPY_XFPREGS
xfpu = kmalloc(sizeof(*xfpu), GFP_KERNEL);
if (!xfpu)
goto cleanup;
#endif
if (signr) {
struct elf_thread_status *tmp;
rcu_read_lock();
do_each_thread(g,p)
if (current->mm == p->mm && current != p) {
tmp = kzalloc(sizeof(*tmp), GFP_ATOMIC);
if (!tmp) {
rcu_read_unlock();
goto cleanup;
}
tmp->thread = p;
list_add(&tmp->list, &thread_list);
}
while_each_thread(g,p);
rcu_read_unlock();
list_for_each(t, &thread_list) {
struct elf_thread_status *tmp;
int sz;
tmp = list_entry(t, struct elf_thread_status, list);
sz = elf_dump_thread_status(signr, tmp);
thread_status_size += sz;
}
}
/* now collect the dump for the current */
memset(prstatus, 0, sizeof(*prstatus));
fill_prstatus(prstatus, current, signr);
elf_core_copy_regs(&prstatus->pr_reg, regs);
segs = current->mm->map_count;
#ifdef ELF_CORE_EXTRA_PHDRS
segs += ELF_CORE_EXTRA_PHDRS;
#endif
gate_vma = get_gate_vma(current);
if (gate_vma != NULL)
segs++;
/* Set up header */
fill_elf_header(elf, segs + 1); /* including notes section */
has_dumped = 1;
current->flags |= PF_DUMPCORE;
/*
* Set up the notes in similar form to SVR4 core dumps made
* with info from their /proc.
*/
fill_note(notes + 0, "CORE", NT_PRSTATUS, sizeof(*prstatus), prstatus);
fill_psinfo(psinfo, current->group_leader, current->mm);
fill_note(notes + 1, "CORE", NT_PRPSINFO, sizeof(*psinfo), psinfo);
numnote = 2;
auxv = (elf_addr_t *)current->mm->saved_auxv;
i = 0;
do
i += 2;
while (auxv[i - 2] != AT_NULL);
fill_note(&notes[numnote++], "CORE", NT_AUXV,
i * sizeof(elf_addr_t), auxv);
/* Try to dump the FPU. */
if ((prstatus->pr_fpvalid =
elf_core_copy_task_fpregs(current, regs, fpu)))
fill_note(notes + numnote++,
"CORE", NT_PRFPREG, sizeof(*fpu), fpu);
#ifdef ELF_CORE_COPY_XFPREGS
if (elf_core_copy_task_xfpregs(current, xfpu))
fill_note(notes + numnote++,
"LINUX", NT_PRXFPREG, sizeof(*xfpu), xfpu);
#endif
fs = get_fs();
set_fs(KERNEL_DS);
DUMP_WRITE(elf, sizeof(*elf));
offset += sizeof(*elf); /* Elf header */
offset += (segs + 1) * sizeof(struct elf_phdr); /* Program headers */
foffset = offset;
/* Write notes phdr entry */
{
struct elf_phdr phdr;
int sz = 0;
for (i = 0; i < numnote; i++)
sz += notesize(notes + i);
sz += thread_status_size;
sz += elf_coredump_extra_notes_size();
fill_elf_note_phdr(&phdr, sz, offset);
offset += sz;
DUMP_WRITE(&phdr, sizeof(phdr));
}
dataoff = offset = roundup(offset, ELF_EXEC_PAGESIZE);
/*
* We must use the same mm->flags while dumping core to avoid
* inconsistency between the program headers and bodies, otherwise an
* unusable core file can be generated.
*/
mm_flags = current->mm->flags;
/* 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;
size_t sz;
sz = vma->vm_end - vma->vm_start;
phdr.p_type = PT_LOAD;
phdr.p_offset = offset;
phdr.p_vaddr = vma->vm_start;
phdr.p_paddr = 0;
phdr.p_filesz = maydump(vma, mm_flags) ? sz : 0;
phdr.p_memsz = sz;
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;
DUMP_WRITE(&phdr, sizeof(phdr));
}
#ifdef ELF_CORE_WRITE_EXTRA_PHDRS
ELF_CORE_WRITE_EXTRA_PHDRS;
#endif
/* write out the notes section */
for (i = 0; i < numnote; i++)
[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>
2006-09-30 23:29:28 -07:00
if (!writenote(notes + i, file, &foffset))
goto end_coredump;
if (elf_coredump_extra_notes_write(file, &foffset))
goto end_coredump;
/* write out the thread status notes section */
list_for_each(t, &thread_list) {
struct elf_thread_status *tmp =
list_entry(t, struct elf_thread_status, list);
for (i = 0; i < tmp->num_notes; i++)
[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>
2006-09-30 23:29:28 -07:00
if (!writenote(&tmp->notes[i], 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>
2006-09-30 23:29:28 -07:00
/* Align to page */
DUMP_SEEK(dataoff - foffset);
for (vma = first_vma(current, gate_vma); vma != NULL;
vma = next_vma(vma, gate_vma)) {
unsigned long addr;
if (!maydump(vma, mm_flags))
continue;
for (addr = vma->vm_start;
addr < vma->vm_end;
addr += PAGE_SIZE) {
struct page *page;
struct vm_area_struct *vma;
if (get_user_pages(current, current->mm, addr, 1, 0, 1,
&page, &vma) <= 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>
2006-09-30 23:29:28 -07:00
DUMP_SEEK(PAGE_SIZE);
} else {
remove ZERO_PAGE The commit b5810039a54e5babf428e9a1e89fc1940fabff11 contains the note A last caveat: the ZERO_PAGE is now refcounted and managed with rmap (and thus mapcounted and count towards shared rss). These writes to the struct page could cause excessive cacheline bouncing on big systems. There are a number of ways this could be addressed if it is an issue. And indeed this cacheline bouncing has shown up on large SGI systems. There was a situation where an Altix system was essentially livelocked tearing down ZERO_PAGE pagetables when an HPC app aborted during startup. This situation can be avoided in userspace, but it does highlight the potential scalability problem with refcounting ZERO_PAGE, and corner cases where it can really hurt (we don't want the system to livelock!). There are several broad ways to fix this problem: 1. add back some special casing to avoid refcounting ZERO_PAGE 2. per-node or per-cpu ZERO_PAGES 3. remove the ZERO_PAGE completely I will argue for 3. The others should also fix the problem, but they result in more complex code than does 3, with little or no real benefit that I can see. Why? Inserting a ZERO_PAGE for anonymous read faults appears to be a false optimisation: if an application is performance critical, it would not be doing many read faults of new memory, or at least it could be expected to write to that memory soon afterwards. If cache or memory use is critical, it should not be working with a significant number of ZERO_PAGEs anyway (a more compact representation of zeroes should be used). As a sanity check -- mesuring on my desktop system, there are never many mappings to the ZERO_PAGE (eg. 2 or 3), thus memory usage here should not increase much without it. When running a make -j4 kernel compile on my dual core system, there are about 1,000 mappings to the ZERO_PAGE created per second, but about 1,000 ZERO_PAGE COW faults per second (less than 1 ZERO_PAGE mapping per second is torn down without being COWed). So removing ZERO_PAGE will save 1,000 page faults per second when running kbuild, while keeping it only saves less than 1 page clearing operation per second. 1 page clear is cheaper than a thousand faults, presumably, so there isn't an obvious loss. Neither the logical argument nor these basic tests give a guarantee of no regressions. However, this is a reasonable opportunity to try to remove the ZERO_PAGE from the pagefault path. If it is found to cause regressions, we can reintroduce it and just avoid refcounting it. The /dev/zero ZERO_PAGE usage and TLB tricks also get nuked. I don't see much use to them except on benchmarks. All other users of ZERO_PAGE are converted just to use ZERO_PAGE(0) for simplicity. We can look at replacing them all and maybe ripping out ZERO_PAGE completely when we are more satisfied with this solution. Signed-off-by: Nick Piggin <npiggin@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus "snif" Torvalds <torvalds@linux-foundation.org>
2007-10-16 01:24:40 -07:00
if (page == ZERO_PAGE(0)) {
if (!dump_seek(file, PAGE_SIZE)) {
page_cache_release(page);
goto end_coredump;
}
} else {
void *kaddr;
flush_cache_page(vma, addr,
page_to_pfn(page));
kaddr = kmap(page);
if ((size += PAGE_SIZE) > limit ||
!dump_write(file, kaddr,
PAGE_SIZE)) {
kunmap(page);
page_cache_release(page);
goto end_coredump;
}
kunmap(page);
}
page_cache_release(page);
}
}
}
#ifdef ELF_CORE_WRITE_EXTRA_DATA
ELF_CORE_WRITE_EXTRA_DATA;
#endif
end_coredump:
set_fs(fs);
cleanup:
while (!list_empty(&thread_list)) {
struct list_head *tmp = thread_list.next;
list_del(tmp);
kfree(list_entry(tmp, struct elf_thread_status, list));
}
kfree(elf);
kfree(prstatus);
kfree(psinfo);
kfree(notes);
kfree(fpu);
#ifdef ELF_CORE_COPY_XFPREGS
kfree(xfpu);
#endif
return has_dumped;
#undef NUM_NOTES
}
#endif /* USE_ELF_CORE_DUMP */
static int __init init_elf_binfmt(void)
{
return register_binfmt(&elf_format);
}
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");