Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20221013210648.137452-1-gpiccoli@igalia.com> <20221013210648.137452-4-gpiccoli@igalia.com>
In-Reply-To: <20221013210648.137452-4-gpiccoli@igalia.com>
From:   Ard Biesheuvel <ardb@kernel.org>
Date:   Fri, 14 Oct 2022 16:46:43 +0200
Message-ID: <CAMj1kXG7syjMsOL+AcUMfT0_nhGde6qc_6MexpdDtxFQpS2=7A@mail.gmail.com>
Subject: Re: [PATCH V2 3/3] efi: pstore: Add module parameter for setting the
 record size
To:     "Guilherme G. Piccoli" <gpiccoli@igalia.com>
Cc:     linux-kernel@vger.kernel.org, linux-hardening@vger.kernel.org,
        linux-fsdevel@vger.kernel.org, linux-efi@vger.kernel.org,
        kernel-dev@igalia.com, kernel@gpiccoli.net, keescook@chromium.org,
        anton@enomsg.org, ccross@android.com, tony.luck@intel.com
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Thu, 13 Oct 2022 at 23:11, Guilherme G. Piccoli <gpiccoli@igalia.com> wrote:
>
> By default, the efi-pstore backend hardcode the UEFI variable size
> as 1024 bytes. The historical reasons for that were discussed by
> Ard in threads [0][1]:
>
> "there is some cargo cult from prehistoric EFI times going
> on here, it seems. Or maybe just misinterpretation of the maximum
> size for the variable *name* vs the variable itself.".
>
> "OVMF has
> OvmfPkg/OvmfPkgX64.dsc:
> gEfiMdeModulePkgTokenSpaceGuid.PcdMaxVariableSize|0x2000
> OvmfPkg/OvmfPkgX64.dsc:
> gEfiMdeModulePkgTokenSpaceGuid.PcdMaxVariableSize|0x8400
>
> where the first one is without secure boot and the second with secure
> boot. Interestingly, the default is
>
> gEfiMdeModulePkgTokenSpaceGuid.PcdMaxVariableSize|0x400
>
> so this is probably where this 1k number comes from."
>
> With that, and since there is not such a limit in the UEFI spec, we
> have the confidence to hereby add a module parameter to enable advanced
> users to change the UEFI record size for efi-pstore data collection,
> this way allowing a much easier reading of the collected log, which is
> not scattered anymore among many small files.
>
> Through empirical analysis we observed that extreme low values (like 8
> bytes) could eventually cause writing issues, so given that and the OVMF
> default discussed, we limited the minimum value to 1024 bytes, which also
> is still the default.
>
> [0] https://lore.kernel.org/lkml/CAMj1kXF4UyRMh2Y_KakeNBHvkHhTtavASTAxXinDO1rhPe_wYg@mail.gmail.com/
> [1] https://lore.kernel.org/lkml/CAMj1kXFy-2KddGu+dgebAdU9v2sindxVoiHLWuVhqYw+R=kqng@mail.gmail.com/
>
> Cc: Ard Biesheuvel <ardb@kernel.org>
> Signed-off-by: Guilherme G. Piccoli <gpiccoli@igalia.com>
> ---
>
>
> V2:
> - Fixed a memory corruption bug in the code (that wasn't causing
> trouble before due to the fixed sized of record_size), thanks
> Ard for spotting this!
>
> - Added Ard's archeology in the commit message plus a comment
> with the reasoning behind the minimum value.
>
>
>  drivers/firmware/efi/efi-pstore.c | 23 +++++++++++++++++------
>  1 file changed, 17 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/firmware/efi/efi-pstore.c b/drivers/firmware/efi/efi-pstore.c
> index 97a9e84840a0..827e32427ddb 100644
> --- a/drivers/firmware/efi/efi-pstore.c
> +++ b/drivers/firmware/efi/efi-pstore.c
> @@ -10,7 +10,9 @@ MODULE_IMPORT_NS(EFIVAR);
>
>  #define DUMP_NAME_LEN 66
>
> -#define EFIVARS_DATA_SIZE_MAX 1024
> +static unsigned int record_size = 1024;
> +module_param(record_size, uint, 0444);
> +MODULE_PARM_DESC(record_size, "size of each pstore UEFI var (in bytes, min/default=1024)");
>
>  static bool efivars_pstore_disable =
>         IS_ENABLED(CONFIG_EFI_VARS_PSTORE_DEFAULT_DISABLE);
> @@ -30,7 +32,7 @@ static int efi_pstore_open(struct pstore_info *psi)
>         if (err)
>                 return err;
>
> -       psi->data = kzalloc(EFIVARS_DATA_SIZE_MAX, GFP_KERNEL);
> +       psi->data = kzalloc(record_size, GFP_KERNEL);
>         if (!psi->data)
>                 return -ENOMEM;
>
> @@ -52,7 +54,7 @@ static inline u64 generic_id(u64 timestamp, unsigned int part, int count)
>  static int efi_pstore_read_func(struct pstore_record *record,
>                                 efi_char16_t *varname)
>  {
> -       unsigned long wlen, size = EFIVARS_DATA_SIZE_MAX;
> +       unsigned long wlen, size = record_size;
>         char name[DUMP_NAME_LEN], data_type;
>         efi_status_t status;
>         int cnt;
> @@ -133,7 +135,7 @@ static ssize_t efi_pstore_read(struct pstore_record *record)
>         efi_status_t status;
>
>         for (;;) {
> -               varname_size = EFIVARS_DATA_SIZE_MAX;
> +               varname_size = record_size;
>

I don't think we need this - this is the size of the variable name not
the variable itself.

>                 /*
>                  * If this is the first read() call in the pstore enumeration,
> @@ -224,11 +226,20 @@ static __init int efivars_pstore_init(void)
>         if (efivars_pstore_disable)
>                 return 0;
>
> -       efi_pstore_info.buf = kmalloc(4096, GFP_KERNEL);
> +       /*
> +        * Notice that 1024 is the minimum here to prevent issues with
> +        * decompression algorithms that were spotted during tests;
> +        * even in the case of not using compression, smaller values would
> +        * just pollute more the pstore FS with many small collected files.
> +        */
> +       if (record_size < 1024)
> +               record_size = 1024;
> +
> +       efi_pstore_info.buf = kmalloc(record_size, GFP_KERNEL);
>         if (!efi_pstore_info.buf)
>                 return -ENOMEM;
>
> -       efi_pstore_info.bufsize = 1024;
> +       efi_pstore_info.bufsize = record_size;
>
>         if (pstore_register(&efi_pstore_info)) {
>                 kfree(efi_pstore_info.buf);
> --
> 2.38.0
>