DomainKey-Signature: a=rsa-sha1; c=nofws;
        d=gmail.com; s=gamma;
        h=mime-version:in-reply-to:references:date:message-id:subject:from:to
         :cc:content-type:content-transfer-encoding;
        b=AyyTVeFCB08O1ho3wB1NZMYOZw2x/pEDDUmiI1PEYUUzucIKeIA7nY2jD6H63eK7Kp
         ZpBAAcRAc0M3sd6vRFLBzys+ETqLj0v2PpIdj42s6wyYS+DmKp0A/cQbxURTMaxDAdn5
         73oC6Lt2o02HcIX9jrdHf/BhXRnZqDj2xo5RI=
MIME-Version: 1.0
In-Reply-To: <20091215.114149.189724375.d.hatayama@jp.fujitsu.com>
References: <20091215.114149.189724375.d.hatayama@jp.fujitsu.com>
Date: Tue, 15 Dec 2009 15:56:38 +0800
Message-ID: <2375c9f90912142356s6bd6708fi1cb04e6ff0c72c4a@mail.gmail.com>
Subject: Re: [RFC, PATCH 4/4] elf_core_dump(): Add extended numbering support
From: =?UTF-8?Q?Am=C3=A9rico_Wang?= <xiyou.wangcong@gmail.com>
To: Daisuke HATAYAMA <d.hatayama@jp.fujitsu.com>
Cc: linux-kernel@vger.kernel.org, akpm@linux-foundation.org, jdike@addtoit.com,
       tony.luck@intel.com, mhiramat@redhat.com
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8BIT
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 12062
Lines: 313

On Tue, Dec 15, 2009 at 10:41 AM, Daisuke HATAYAMA
<d.hatayama@jp.fujitsu.com> wrote:
> The current ELF dumper implementation can produce broken corefiles
> if program headers exceed 65535. This number is determined by the
> number of vmas which the process have. In particular, some extreme
> programs may use more than 65535 vmas. (If you google max_map_count,
> you can find some users facing this problem.) This kind of program
> never be able to generate correct coredumps.
>
> This patch implements ``extended numbering'' that uses sh_info
> field of the first section header instead of e_phnum field in order
> to represent upto 4294967295 vmas.
>
> This is supported by AMD64-ABI(http://www.x86-64.org/documentation.html)
> and Solaris(http://docs.sun.com/app/docs/doc/817-1984/). Of course,
> we are preparing patches for gdb and binutils.
>
> Signed-off-by: Daisuke HATAYAMA <d.hatayama@jp.fujitsu.com>


Hi,

Can you reorder your patches please?
Your patch 0/4 depends on 1/4, I am afraid. :-/

Thanks!

> ---
>  arch/ia64/kernel/elfcore.c |   16 ++++++++
>  arch/um/sys-i386/elfcore.c |   18 +++++++++
>  fs/binfmt_elf.c            |   88 +++++++++++++++++++++++++++++++++++++++-----
>  include/linux/elf.h        |   26 ++++++++++++-
>  4 files changed, 137 insertions(+), 11 deletions(-)
>
> diff --git a/arch/ia64/kernel/elfcore.c b/arch/ia64/kernel/elfcore.c
> index 9c0dd8b..a15d8d4 100644
> --- a/arch/ia64/kernel/elfcore.c
> +++ b/arch/ia64/kernel/elfcore.c
> @@ -73,3 +73,19 @@ int elf_core_write_extra_data(struct file *file, size_t *size,
>        }
>        return 1;
>  }
> +
> +size_t elf_core_extra_data_size(void)
> +{
> +       const struct elf_phdr *const gate_phdrs =
> +               (const struct elf_phdr *) (GATE_ADDR + GATE_EHDR->e_phoff);
> +       int i;
> +       size_t size = 0;
> +
> +       for (i = 0; i < GATE_EHDR->e_phnum; ++i) {
> +               if (gate_phdrs[i].p_type == PT_LOAD) {
> +                       size += PAGE_ALIGN(gate_phdrs[i].p_memsz);
> +                       break;
> +               }
> +       }
> +       return size;
> +}
> diff --git a/arch/um/sys-i386/elfcore.c b/arch/um/sys-i386/elfcore.c
> index 4e320f0..4e34e47 100644
> --- a/arch/um/sys-i386/elfcore.c
> +++ b/arch/um/sys-i386/elfcore.c
> @@ -76,3 +76,21 @@ int elf_core_write_extra_data(struct file *file, size_t *size,
>        }
>        return 1;
>  }
> +
> +size_t elf_core_extra_data_size(void)
> +{
> +       if ( vsyscall_ehdr ) {
> +               const struct elfhdr *const ehdrp =
> +                       (struct elfhdr *)vsyscall_ehdr;
> +               const struct elf_phdr *const phdrp =
> +                       (const struct elf_phdr *) (vsyscall_ehdr + ehdrp->e_phoff);
> +               int i;
> +
> +               for (i = 0; i < ehdrp->e_phnum; ++i) {
> +                       if (phdrp[i].p_type == PT_LOAD) {
> +                               return (size_t) phdrp[i].p_filesz;
> +                       }
> +               }
> +       }
> +       return 0;
> +}
> diff --git a/fs/binfmt_elf.c b/fs/binfmt_elf.c
> index cded1ba..ad2ad5f 100644
> --- a/fs/binfmt_elf.c
> +++ b/fs/binfmt_elf.c
> @@ -1895,6 +1895,38 @@ static struct vm_area_struct *next_vma(struct vm_area_struct *this_vma,
>        return gate_vma;
>  }
>
> +static void fill_extnum_info(struct elfhdr *elf, struct elf_shdr *shdr4extnum,
> +                            elf_addr_t e_shoff, int segs)
> +{
> +       elf->e_shoff = e_shoff;
> +       elf->e_shentsize = sizeof(*shdr4extnum);
> +       elf->e_shnum = 1;
> +       elf->e_shstrndx = SHN_UNDEF;
> +
> +       shdr4extnum->sh_name = 0;
> +       shdr4extnum->sh_addr = 0;
> +       shdr4extnum->sh_offset = 0;
> +       shdr4extnum->sh_type = SHT_NULL;
> +       shdr4extnum->sh_flags = 0;
> +       shdr4extnum->sh_size = elf->e_shnum;
> +       shdr4extnum->sh_link = elf->e_shstrndx;
> +       shdr4extnum->sh_info = segs;
> +       shdr4extnum->sh_addralign = 0;
> +       shdr4extnum->sh_entsize = 0;
> +}
> +
> +static size_t elf_core_vma_data_size(struct vm_area_struct *gate_vma,
> +                                    unsigned long mm_flags)
> +{
> +       struct vm_area_struct *vma;
> +       size_t size = 0;
> +
> +       for (vma = first_vma(current, gate_vma); vma != NULL;
> +            vma = next_vma(vma, gate_vma))
> +               size += vma_dump_size(vma, mm_flags);
> +       return size;
> +}
> +
>  /*
>  * It's been implemented that some architectures write out some extra
>  * data into segments. On the other hand, other architechtures use
> @@ -1917,6 +1949,11 @@ int __weak elf_core_write_extra_data(struct file *file, size_t *size,
>        return 1;
>  }
>
> +size_t __weak elf_core_extra_data_size(void)
> +{
> +       return 0;
> +}
> +
>  /*
>  * Actual dumper
>  *
> @@ -1936,6 +1973,9 @@ static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, un
>        unsigned long mm_flags;
>        struct elf_note_info info;
>        struct elf_phdr *phdr4note = NULL;
> +       struct elf_shdr *shdr4extnum = NULL;
> +       Elf_Half e_phnum = 0;
> +       elf_addr_t e_shoff;
>
>        /*
>         * We no longer stop all VM operations.
> @@ -1964,12 +2004,19 @@ static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, un
>        if (gate_vma != NULL)
>                segs++;
>
> +       /* for notes section */
> +       segs++;
> +
> +       /* If segs > PN_XNUM(0xffff), then e_phnum overflows. To avoid
> +        * this, kernel supports extended numbering. Have a look at
> +        * include/linux/elf.h for further information. */
> +       e_phnum = segs > PN_XNUM ? PN_XNUM : segs;
> +
>        /*
>         * Collect all the non-memory information about the process for the
>         * notes.  This also sets up the file header.
>         */
> -       if (!fill_note_info(elf, segs + 1, /* including notes section */
> -                           &info, signr, regs))
> +       if (!fill_note_info(elf, e_phnum, &info, signr, regs))
>                goto cleanup;
>
>        has_dumped = 1;
> @@ -1979,7 +2026,7 @@ static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, un
>        set_fs(KERNEL_DS);
>
>        offset += sizeof(*elf);                         /* Elf header */
> -       offset += (segs + 1) * sizeof(struct elf_phdr); /* Program headers */
> +       offset += segs * sizeof(struct elf_phdr); /* Program headers */
>        foffset = offset;
>
>        /* Write notes phdr entry */
> @@ -1998,6 +2045,26 @@ static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, un
>
>        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;
> +
> +       offset += elf_core_vma_data_size(gate_vma, mm_flags);
> +       offset += elf_core_extra_data_size();
> +       e_shoff = offset;
> +
> +       if (e_phnum == PN_XNUM) {
> +               shdr4extnum = kmalloc(sizeof(*shdr4extnum), GFP_KERNEL);
> +               if (!shdr4extnum)
> +                       goto end_coredump;
> +               fill_extnum_info(elf, shdr4extnum, e_shoff, segs);
> +       }
> +
> +       offset = dataoff;
> +
>        size += sizeof(*elf);
>        if (size > limit || !dump_write(file, elf, sizeof(*elf)))
>                goto end_coredump;
> @@ -2006,13 +2073,6 @@ static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, un
>        if (size > limit || !dump_write(file, phdr4note, sizeof(*phdr4note)))
>                goto end_coredump;
>
> -       /*
> -        * 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)) {
> @@ -2079,11 +2139,19 @@ static int elf_core_dump(long signr, struct pt_regs *regs, struct file *file, un
>        if (!elf_core_write_extra_data(file, &size, limit))
>                goto end_coredump;
>
> +       if (e_phnum == PN_XNUM) {
> +               size += sizeof(*shdr4extnum);
> +               if (size > limit
> +                   || !dump_write(file, shdr4extnum, sizeof(*shdr4extnum)))
> +                       goto end_coredump;
> +       }
> +
>  end_coredump:
>        set_fs(fs);
>
>  cleanup:
>        free_note_info(&info);
> +       kfree(shdr4extnum);
>        kfree(phdr4note);
>        kfree(elf);
>  out:
> diff --git a/include/linux/elf.h b/include/linux/elf.h
> index d103127..027fdfe 100644
> --- a/include/linux/elf.h
> +++ b/include/linux/elf.h
> @@ -50,6 +50,28 @@ typedef __s64        Elf64_Sxword;
>
>  #define PT_GNU_STACK   (PT_LOOS + 0x474e551)
>
> +/*
> + * Extended Numbering
> + *
> + * If the real number of program header table entries is larger than
> + * or equal to PN_XNUM(0xffff), it is set to sh_info field of the
> + * section header at index 0, and PN_XNUM is set to e_phnum
> + * field. Otherwise, the section header at index 0 is zero
> + * initialized, if it exists.
> + *
> + * Specifications are available in:
> + *
> + * - Sun microsystems: Linker and Libraries.
> + *   Part No: 817-1984-17, September 2008.
> + *   URL: http://docs.sun.com/app/docs/doc/817-1984
> + *
> + * - System V ABI AMD64 Architecture Processor Supplement
> + *   Draft Version 0.99.,
> + *   May 11, 2009.
> + *   URL: http://www.x86-64.org/
> + */
> +#define PN_XNUM 0xffff
> +
>  /* These constants define the different elf file types */
>  #define ET_NONE   0
>  #define ET_REL    1
> @@ -286,7 +308,7 @@ typedef struct elf64_phdr {
>  #define SHN_COMMON     0xfff2
>  #define SHN_HIRESERVE  0xffff
>
> -typedef struct {
> +typedef struct elf32_shdr {
>   Elf32_Word   sh_name;
>   Elf32_Word   sh_type;
>   Elf32_Word   sh_flags;
> @@ -384,6 +406,7 @@ typedef struct elf64_note {
>  extern Elf32_Dyn _DYNAMIC [];
>  #define elfhdr         elf32_hdr
>  #define elf_phdr       elf32_phdr
> +#define elf_shdr       elf32_shdr
>  #define elf_note       elf32_note
>  #define elf_addr_t     Elf32_Off
>  #define Elf_Half       Elf32_Half
> @@ -393,6 +416,7 @@ extern Elf32_Dyn _DYNAMIC [];
>  extern Elf64_Dyn _DYNAMIC [];
>  #define elfhdr         elf64_hdr
>  #define elf_phdr       elf64_phdr
> +#define elf_shdr       elf64_shdr
>  #define elf_note       elf64_note
>  #define elf_addr_t     Elf64_Off
>  #define Elf_Half       Elf64_Half
> --
> 1.6.5.1
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/