On Thu, 26 Jul 2007 02:37:43 -0700 (PDT) Roland McGrath <[EMAIL PROTECTED]>
wrote:
>
> This adds two sysctl items under fs.binfmt_elf for controlling how memory
> segments are dumped in ELF core files. The core_dump_whole_segments option
> can be set to have private file mappings dumped in their entirety. Some
> people prefer the certainty to the saving disk space and i/o at dump time,
> and a global default is easier to set than every individual mm's MMF_DUMP_*.
It's a bit disappointing to route around the newly-added
/proc/pid/coredump_filter. It's inherited across fork so where's the
problem? Just set init's mask appropriately at boot if that's what you
want.
Plus making it system-wide is fairly sucky compared to making it settable
per-process-tree.
And you had to add considerably more code this way.
> The more interesting new feature is the core_dump_elf_headers option.
> This dumps the first page (only) of a read-only private file mapping if it
> appears to be a mapping of an ELF file. Including these pages in the core
> dump may give sufficient identifying information to associate the original
> DSO and executable file images and their debugging information with a core
> file in a generic way just from its contents (e.g. when those binaries were
> built with ld --build-id). I expect this to become the default behavior
> eventually. Existing versions of gdb can be confused by the core dumps it
> creates, so it won't enabled by default for some time to come. Soon many
> people will have systems with a gdb that handle these dumps, so they can
> use the sysctl option.
>
> On biarch machines, there are two sets of options.
> The native dumper uses fs.binfmt_elf.core_dump_*.
> The compat dumper uses fs.binfmt_elf.32bit.core_dump_*.
Documentation/filesystems/proc.txt...
> -static int maydump(struct vm_area_struct *vma, unsigned long mm_flags)
> +static unsigned long vma_dump_size(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;
> + goto whole;
>
> /* Do not dump I/O mapped devices or special mappings */
> if (vma->vm_flags & (VM_IO | VM_RESERVED))
> @@ -1211,17 +1213,51 @@ static int maydump(struct vm_area_struct *vma,
> unsigned long mm_flags)
>
> /* 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);
> + if (vma->vm_file->f_path.dentry->d_inode->i_nlink == 0 ?
hm, I'm now wondering whether testing i_nlink was the best (or official)
way of determining whether a MAP_SHARED vma is anonymous or file-backed.
Hugh will know.
> + test_bit(MMF_DUMP_ANON_SHARED, &mm_flags) :
> + test_bit(MMF_DUMP_MAPPED_SHARED, &mm_flags))
It was peculiar of us to use bitops against a local variable - normally
we'd use open-coded &'s here. Maybe it's OK - test_bit() is pretty cheap
(on x86, at least), but one wonders whether gcc could do better if it knew
what was going on.
> ...
>
> +
> +static ctl_table elf_core_dump_sysctls[] = {
> + {
> + .ctl_name = CTL_UNNUMBERED,
> + .procname = "core_dump_whole_segments",
> + .data = &dump_whole_segments,
> + .maxlen = sizeof(int),
> + .mode = 0644,
> + .proc_handler = &proc_dointvec,
> + },
> + {
> + .ctl_name = CTL_UNNUMBERED,
> + .procname = "core_dump_elf_headers",
> + .data = &dump_elf_headers,
> + .maxlen = sizeof(int),
> + .mode = 0644,
> + .proc_handler = &proc_dointvec,
> + },
> + { .ctl_name = 0 }
> +};
> +
> +#if BITS_PER_LONG > 32 && ELF_CLASS == ELFCLASS32
> +static ctl_table binfmt_elf_sysctl_32bit_dir[] = {
> + {
> + .ctl_name = CTL_UNNUMBERED,
> + .procname = "32bit",
> + .mode = 0555,
> + .child = elf_core_dump_sysctls,
> + },
> + { .ctl_name = 0 }
> +};
> +#define elf_core_dump_sysctls binfmt_elf_sysctl_32bit_dir
> +#endif
The above makes elf_core_dump_sysctls[] inaccessible from here on. It will
warn, surely?
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
