Re: [PATCH v2] efi: fix a race and a buffer overflow while reading efivars via sysfs

From: Vladis Dronov
Date: Wed Mar 04 2020 - 12:18:26 EST


Hello, Ard, Joye, all,

----- Original Message -----
> From: "Ard Biesheuvel" <ardb@xxxxxxxxxx>
> To: "Vladis Dronov" <vdronov@xxxxxxxxxx>
> Cc: "linux-efi" <linux-efi@xxxxxxxxxxxxxxx>, "joeyli" <jlee@xxxxxxxx>, "Linux Kernel Mailing List"
> <linux-kernel@xxxxxxxxxxxxxxx>
> Sent: Wednesday, March 4, 2020 4:57:16 PM
> Subject: Re: [PATCH v2] efi: fix a race and a buffer overflow while reading efivars via sysfs
>
> On Wed, 4 Mar 2020 at 16:50, Vladis Dronov <vdronov@xxxxxxxxxx> wrote:
> >
> > There is a race and a buffer overflow corrupting a kernel memory while
> > reading an efi variable with a size more than 1024 bytes via the older
> > sysfs method. This happens because accessing struct efi_variable in
> > efivar_{attr,size,data}_read() and friends is not protected from
> > a concurrent access leading to a kernel memory corruption and, at best,
> > to a crash. The race scenario is the following:
> >
> > CPU0: CPU1:
> > efivar_attr_read()
> > var->DataSize = 1024;
> > efivar_entry_get(... &var->DataSize)
> > down_interruptible(&efivars_lock)
> > efivar_attr_read() // same efi var
> > var->DataSize = 1024;
> > efivar_entry_get(... &var->DataSize)
> > down_interruptible(&efivars_lock)
> > virt_efi_get_variable()
> > // returns EFI_BUFFER_TOO_SMALL but
> > // var->DataSize is set to a real
> > // var size more than 1024 bytes
> > up(&efivars_lock)
> > virt_efi_get_variable()
> > // called with var->DataSize set
> > // to a real var size, returns
> > // successfully and overwrites
> > // a 1024-bytes kernel buffer
> > up(&efivars_lock)
> >
> > This can be reproduced by concurrent reading of an efi variable which size
> > is more than 1024 bytes:
> >
> > ts# for cpu in $(seq 0 $(nproc --ignore=1)); do ( taskset -c $cpu \
> > cat /sys/firmware/efi/vars/KEKDefault*/size & ) ; done
> >
> > Fix this by using a local variable for a var's data buffer size so it
> > does not get overwritten. Also add a sanity check to efivar_store_raw().
> >
> > Reported-by: Bob Sanders <bob.sanders@xxxxxxx> and the LTP testsuite
> > Signed-off-by: Vladis Dronov <vdronov@xxxxxxxxxx>

AFAIU, you can modify suggested patches, could you please, add a link here
so further reader has a reference (I forgot to do it myself):

Link: https://lore.kernel.org/linux-efi/20200303085528.27658-1-vdronov@xxxxxxxxxx/T/#u

> > ---
> > drivers/firmware/efi/efi-pstore.c | 2 +-
> > drivers/firmware/efi/efivars.c | 32 ++++++++++++++++++++++---------
> > drivers/firmware/efi/vars.c | 2 +-
> > 3 files changed, 25 insertions(+), 11 deletions(-)
> >
> > diff --git a/drivers/firmware/efi/efi-pstore.c
> > b/drivers/firmware/efi/efi-pstore.c
> > index 9ea13e8d12ec..e4767a7ce973 100644
> > --- a/drivers/firmware/efi/efi-pstore.c
> > +++ b/drivers/firmware/efi/efi-pstore.c
> > @@ -161,7 +161,7 @@ static int efi_pstore_scan_sysfs_exit(struct
> > efivar_entry *pos,
> > *
> > * @record: pstore record to pass to callback
> > *
> > - * You MUST call efivar_enter_iter_begin() before this function, and
> > + * You MUST call efivar_entry_iter_begin() before this function, and
> > * efivar_entry_iter_end() afterwards.
> > *
> > */
>
> This hunk can be dropped now, I guess

I surely do not have much experience in writing upstream patches. But I saw people
doing small fixes like this one, say, commit 589b7289 ("While we are here, the previous
line has some trailing whitespace; clean that up as well"). This is a small mistype
and I just wanted to fix it and did not wanted to allocate a whole commit for that.
I believe a bigger commit is an acceptable place to fix mistypes.

AFAIU, a maintainer can modify suggested patches, so please, feel free to drop this
hunk, if you feel this is a right thing.

> > diff --git a/drivers/firmware/efi/efivars.c
> > b/drivers/firmware/efi/efivars.c
> > index 7576450c8254..16a617f9c5cf 100644
> > --- a/drivers/firmware/efi/efivars.c
> > +++ b/drivers/firmware/efi/efivars.c
> > @@ -83,13 +83,16 @@ static ssize_t
> > efivar_attr_read(struct efivar_entry *entry, char *buf)
> > {
> > struct efi_variable *var = &entry->var;
> > + unsigned long size = sizeof(var->Data);
> > char *str = buf;
> > + int ret;
> >
> > if (!entry || !buf)
> > return -EINVAL;
> >
> > - var->DataSize = 1024;
> > - if (efivar_entry_get(entry, &var->Attributes, &var->DataSize,
> > var->Data))
> > + ret = efivar_entry_get(entry, &var->Attributes, &size, var->Data);
> > + var->DataSize = size;
>
> For my understanding, could you explain why we do the assignment here?
> Does var->DataSize matter in this case? Can it deviate from 1024?

Yes, the other code expects var->DataSize to be set to a real size of a var
after efivar_entry_get() call. For example, efivar_show_raw():

compat->DataSize = var->DataSize;

and efivar_data_read():

memcpy(buf, var->Data, var->DataSize);
return var->DataSize;

Yes, we can change the code to use size here, but this will make struct efi_variable
*var inconsistent (name, guid, data, attr set properly, but not size). It feels so
incorrect to leave this struct inconsistent. I'm not sure that code which calls
efivar_{attr,size,data}_read()/efivar_show_raw() is not using this struct's ->DataSize
field later.

> > + if (ret)
> > return -EIO;
> >
> > if (var->Attributes & EFI_VARIABLE_NON_VOLATILE)
> > @@ -116,13 +119,16 @@ static ssize_t
> > efivar_size_read(struct efivar_entry *entry, char *buf)
> > {
> > struct efi_variable *var = &entry->var;
> > + unsigned long size = sizeof(var->Data);
> > char *str = buf;
> > + int ret;
> >
> > if (!entry || !buf)
> > return -EINVAL;
> >
> > - var->DataSize = 1024;
> > - if (efivar_entry_get(entry, &var->Attributes, &var->DataSize,
> > var->Data))
> > + ret = efivar_entry_get(entry, &var->Attributes, &size, var->Data);
> > + var->DataSize = size;
> > + if (ret)
> > return -EIO;
> >
> > str += sprintf(str, "0x%lx\n", var->DataSize);
> > @@ -133,12 +139,15 @@ static ssize_t
> > efivar_data_read(struct efivar_entry *entry, char *buf)
> > {
> > struct efi_variable *var = &entry->var;
> > + unsigned long size = sizeof(var->Data);
> > + int ret;
> >
> > if (!entry || !buf)
> > return -EINVAL;
> >
> > - var->DataSize = 1024;
> > - if (efivar_entry_get(entry, &var->Attributes, &var->DataSize,
> > var->Data))
> > + ret = efivar_entry_get(entry, &var->Attributes, &size, var->Data);
> > + var->DataSize = size;
> > + if (ret)
> > return -EIO;
> >
> > memcpy(buf, var->Data, var->DataSize);
> > @@ -199,6 +208,9 @@ efivar_store_raw(struct efivar_entry *entry, const char
> > *buf, size_t count)
> > u8 *data;
> > int err;
> >
> > + if (!entry || !buf)
> > + return -EINVAL;
> > +
>
> So what are we sanity checking here? When might this occur? Does it
> need to be in the same patch?

efivar_{attr,size,data}_read()/efivar_show_raw() has this check, I believe
it is reasonable to add it here too. In case entry or buf happen to be NULL
it will lead to a NULL-deref later:

compat = (struct compat_efi_variable *)buf;
memcpy(compat->VariableName, var->VariableName, EFI_VAR_NAME_LEN);

I see this as more-or-less related and too small for a whole separate commit.
Please, feel free to drop this hunk, if you believe this is not correct. Would
you like me to send a separate patch adding the check above in this case?

> > if (in_compat_syscall()) {
> > struct compat_efi_variable *compat;
> >
> > @@ -250,14 +262,16 @@ efivar_show_raw(struct efivar_entry *entry, char
> > *buf)
> > {
> > struct efi_variable *var = &entry->var;
> > struct compat_efi_variable *compat;
> > + unsigned long datasize = sizeof(var->Data);
> > size_t size;
> > + int ret;
> >
> > if (!entry || !buf)
> > return 0;
> >
> > - var->DataSize = 1024;
> > - if (efivar_entry_get(entry, &entry->var.Attributes,
> > - &entry->var.DataSize, entry->var.Data))
> > + ret = efivar_entry_get(entry, &var->Attributes, &datasize,
> > var->Data);
> > + var->DataSize = size;
> > + if (ret)
> > return -EIO;
> >
> > if (in_compat_syscall()) {
> > diff --git a/drivers/firmware/efi/vars.c b/drivers/firmware/efi/vars.c
> > index 436d1776bc7b..5f2a4d162795 100644
> > --- a/drivers/firmware/efi/vars.c
> > +++ b/drivers/firmware/efi/vars.c
> > @@ -1071,7 +1071,7 @@ EXPORT_SYMBOL_GPL(efivar_entry_iter_end);
> > * entry on the list. It is safe for @func to remove entries in the
> > * list via efivar_entry_delete().
> > *
> > - * You MUST call efivar_enter_iter_begin() before this function, and
> > + * You MUST call efivar_entry_iter_begin() before this function, and
> > * efivar_entry_iter_end() afterwards.
> > *
> > * It is possible to begin iteration from an arbitrary entry within
>
> We can drop this.
>
> > --
> > 2.20.1

Best regards,
Vladis Dronov | Red Hat, Inc. | The Core Kernel | Senior Software Engineer