Change the way ELF coredumps are handled. Instead of unconditionally

skipping read-only pages, which can result in valuable non-text-related
data not getting dumped, the ELF loader and the dynamic loader now mark
read-only text pages NOCORE and the coredump code only checks (primarily) for
complete inaccessibility of the page or NOCORE being set.

Certain applications which map large amounts of read-only data will
produce much larger cores.  A new sysctl has been added,
debug.elf_legacy_coredump, which will revert to the old behavior.

This commit represents collaborative work by all parties involved.
The PR contains a program demonstrating the problem.

PR:		kern/45994
Submitted by:	"Peter Edwards" <pmedwards@eircom.net>, Archie Cobbs <archie@dellroad.org>
Reviewed by:	jdp, dillon
MFC after:	7 days
This commit is contained in:
Matthew Dillon 2002-12-16 19:24:43 +00:00
parent 47770b6fd5
commit fa7dd9c5bc
Notes: svn2git 2020-12-20 02:59:44 +00:00
svn path=/head/; revision=107948
3 changed files with 55 additions and 17 deletions

View File

@ -38,7 +38,8 @@
#include "debug.h"
#include "rtld.h"
static int protflags(int); /* Elf flags -> mmap protection */
static int convert_prot(int); /* Elf flags -> mmap protection */
static int convert_flags(int); /* Elf flags -> mmap flags */
/*
* Map a shared object into memory. The "fd" argument is a file descriptor,
@ -75,6 +76,7 @@ map_object(int fd, const char *path, const struct stat *sb)
Elf_Addr data_vlimit;
caddr_t data_addr;
int data_prot;
int data_flags;
Elf_Addr clear_vaddr;
caddr_t clear_addr;
caddr_t clear_page;
@ -189,8 +191,8 @@ map_object(int fd, const char *path, const struct stat *sb)
mapsize = base_vlimit - base_vaddr;
base_addr = u.hdr.e_type == ET_EXEC ? (caddr_t) base_vaddr : NULL;
mapbase = mmap(base_addr, mapsize, protflags(segs[0]->p_flags),
MAP_PRIVATE, fd, base_offset);
mapbase = mmap(base_addr, mapsize, convert_prot(segs[0]->p_flags),
convert_flags(segs[0]->p_flags), fd, base_offset);
if (mapbase == (caddr_t) -1) {
_rtld_error("%s: mmap of entire address space failed: %s",
path, strerror(errno));
@ -209,10 +211,11 @@ map_object(int fd, const char *path, const struct stat *sb)
data_vaddr = trunc_page(segs[i]->p_vaddr);
data_vlimit = round_page(segs[i]->p_vaddr + segs[i]->p_filesz);
data_addr = mapbase + (data_vaddr - base_vaddr);
data_prot = protflags(segs[i]->p_flags);
data_prot = convert_prot(segs[i]->p_flags);
data_flags = convert_flags(segs[i]->p_flags) | MAP_FIXED;
/* Do not call mmap on the first segment - this is redundant */
if (i && mmap(data_addr, data_vlimit - data_vaddr, data_prot,
MAP_PRIVATE|MAP_FIXED, fd, data_offset) == (caddr_t) -1) {
data_flags, fd, data_offset) == (caddr_t) -1) {
_rtld_error("%s: mmap of data failed: %s", path, strerror(errno));
return NULL;
}
@ -315,7 +318,7 @@ obj_new(void)
* flags for MMAP.
*/
static int
protflags(int elfflags)
convert_prot(int elfflags)
{
int prot = 0;
if (elfflags & PF_R)
@ -326,3 +329,17 @@ protflags(int elfflags)
prot |= PROT_EXEC;
return prot;
}
static int
convert_flags(int elfflags)
{
int flags = MAP_PRIVATE; /* All mappings are private */
/*
* Readonly mappings are marked "MAP_NOCORE", because they can be
* reconstructed by a debugger.
*/
if (!(elfflags & PF_W))
flags |= MAP_NOCORE;
return flags;
}

View File

@ -87,6 +87,9 @@ SYSCTL_INT(_debug, OID_AUTO, elf32_trace, CTLFLAG_RW, &elf_trace, 0, "");
#else
SYSCTL_INT(_debug, OID_AUTO, elf64_trace, CTLFLAG_RW, &elf_trace, 0, "");
#endif
static int elf_legacy_coredump = 0;
SYSCTL_INT(_debug, OID_AUTO, elf_legacy_coredump, CTLFLAG_RW,
&elf_legacy_coredump, 0, "");
static Elf_Brandinfo *elf_brand_list[MAX_BRANDS];
extern int fallback_elf_brand;
@ -349,7 +352,7 @@ __elfN(load_section)(struct proc *p, struct vmspace *vmspace,
{
size_t map_len;
vm_offset_t map_addr;
int error, rv;
int error, rv, cow;
size_t copy_len;
vm_offset_t file_addr;
vm_offset_t data_buf = 0;
@ -392,6 +395,11 @@ __elfN(load_section)(struct proc *p, struct vmspace *vmspace,
if (map_len != 0) {
vm_object_reference(object);
/* cow flags: don't dump readonly sections in core */
cow = MAP_COPY_ON_WRITE | MAP_PREFAULT |
(prot & VM_PROT_WRITE ? 0 : MAP_DISABLE_COREDUMP);
rv = __elfN(map_insert)(&vmspace->vm_map,
object,
file_addr, /* file offset */
@ -399,7 +407,7 @@ __elfN(load_section)(struct proc *p, struct vmspace *vmspace,
map_addr + map_len,/* virtual end */
prot,
VM_PROT_ALL,
MAP_COPY_ON_WRITE | MAP_PREFAULT);
cow);
if (rv != KERN_SUCCESS) {
vm_object_deallocate(object);
return (EINVAL);
@ -1042,17 +1050,29 @@ each_writable_segment(p, func, closure)
entry = entry->next) {
vm_object_t obj;
if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) ||
(entry->protection & (VM_PROT_READ|VM_PROT_WRITE)) !=
(VM_PROT_READ|VM_PROT_WRITE))
continue;
/*
* Don't dump inaccessible mappings, deal with legacy
* coredump mode.
*
* Note that read-only segments related to the elf binary
* are marked MAP_ENTRY_NOCOREDUMP now so we no longer
* need to arbitrarily ignore such segments.
*/
if (elf_legacy_coredump) {
if ((entry->protection & VM_PROT_RW) != VM_PROT_RW)
continue;
} else {
if ((entry->protection & VM_PROT_ALL) == 0)
continue;
}
/*
** Dont include memory segment in the coredump if
** MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
** madvise(2).
*/
if (entry->eflags & MAP_ENTRY_NOCOREDUMP)
* Dont include memory segment in the coredump if
* MAP_NOCORE is set in mmap(2) or MADV_NOCORE in
* madvise(2). Do not dump submaps (i.e. parts of the
* kernel map).
*/
if (entry->eflags & (MAP_ENTRY_NOCOREDUMP|MAP_ENTRY_IS_SUB_MAP))
continue;
if ((obj = entry->object.vm_object) == NULL)

View File

@ -81,6 +81,7 @@ typedef u_char vm_prot_t; /* protection codes */
#define VM_PROT_OVERRIDE_WRITE ((vm_prot_t) 0x08) /* copy-on-write */
#define VM_PROT_ALL (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE)
#define VM_PROT_RW (VM_PROT_READ|VM_PROT_WRITE)
#define VM_PROT_DEFAULT VM_PROT_ALL
union vm_map_object;