On Wed, 2021-08-18 at 14:59 -0400, Paul Moore wrote:
> On Sat, Aug 14, 2021 at 2:34 PM Michael Weiß
> <michael.weiss@xxxxxxxxxxxxxxxxxxx> wrote:
> > To be able to send auditing events to user space, we introduce
> > a generic dm-audit module. It provides helper functions to emit
> > audit events through the kernel audit subsystem. We claim the
> > AUDIT_DM type=1336 out of the audit event messages range in the
> > corresponding userspace api in 'include/uapi/linux/audit.h' for
> > those events.
> >
> > Following commits to device mapper targets actually will make
> > use of this to emit those events in relevant cases.
> >
> > Signed-off-by: Michael Weiß <michael.weiss@xxxxxxxxxxxxxxxxxxx>
>
> Hi Michael,
>
> You went into more detail in your patchset cover letter, e.g. example
> audit records, which I think would be helpful here in the commit
> description so we have it as part of the git log. I don't want to
> discourage you from writing cover letters, but don't forget that the
> cover letters can be lost to the ether after a couple of years whereas
> the git log has a much longer lifetime (we hope!) and a tighter
> binding to the related code.
Hi Paul,
at first thank you for your comments.
I see your point and I will respect that in providing the next version of
this patch-set.
>
> > ---
> > drivers/md/Kconfig | 10 +++++++
> > drivers/md/Makefile | 4 +++
> > drivers/md/dm-audit.c | 59 ++++++++++++++++++++++++++++++++++++++
> > drivers/md/dm-audit.h | 33 +++++++++++++++++++++
> > include/uapi/linux/audit.h | 1 +
> > 5 files changed, 107 insertions(+)
> > create mode 100644 drivers/md/dm-audit.c
> > create mode 100644 drivers/md/dm-audit.h
> >
> > diff --git a/drivers/md/Kconfig b/drivers/md/Kconfig
> > index 0602e82a9516..48adbec12148 100644
> > --- a/drivers/md/Kconfig
> > +++ b/drivers/md/Kconfig
> > @@ -608,6 +608,7 @@ config DM_INTEGRITY
> > select CRYPTO
> > select CRYPTO_SKCIPHER
> > select ASYNC_XOR
> > + select DM_AUDIT if AUDIT
> > help
> > This device-mapper target emulates a block device that has
> > additional per-sector tags that can be used for storing
> > @@ -640,4 +641,13 @@ config DM_ZONED
> >
> > If unsure, say N.
> >
> > +config DM_AUDIT
> > + bool "DM audit events"
> > + depends on AUDIT
> > + help
> > + Generate audit events for device-mapper.
> > +
> > + Enables audit logging of several security relevant events in the
> > + particular device-mapper targets, especially the integrity target.
> > +
> > endif # MD
> > diff --git a/drivers/md/Makefile b/drivers/md/Makefile
> > index a74aaf8b1445..4cd47623c742 100644
> > --- a/drivers/md/Makefile
> > +++ b/drivers/md/Makefile
> > @@ -103,3 +103,7 @@ endif
> > ifeq ($(CONFIG_DM_VERITY_VERIFY_ROOTHASH_SIG),y)
> > dm-verity-objs += dm-verity-verify-sig.o
> > endif
> > +
> > +ifeq ($(CONFIG_DM_AUDIT),y)
> > +dm-mod-objs += dm-audit.o
> > +endif
> > diff --git a/drivers/md/dm-audit.c b/drivers/md/dm-audit.c
> > new file mode 100644
> > index 000000000000..c7e5824821bb
> > --- /dev/null
> > +++ b/drivers/md/dm-audit.c
> > @@ -0,0 +1,59 @@
> > +// SPDX-License-Identifier: GPL-2.0
> > +/*
> > + * Creating audit records for mapped devices.
> > + *
> > + * Copyright (C) 2021 Fraunhofer AISEC. All rights reserved.
> > + *
> > + * Authors: Michael Weiß <michael.weiss@xxxxxxxxxxxxxxxxxxx>
> > + */
> > +
> > +#include <linux/audit.h>
> > +#include <linux/module.h>
> > +#include <linux/device-mapper.h>
> > +#include <linux/bio.h>
> > +#include <linux/blkdev.h>
> > +
> > +#include "dm-audit.h"
> > +#include "dm-core.h"
> > +
> > +void dm_audit_log_bio(const char *dm_msg_prefix, const char *op,
> > + struct bio *bio, sector_t sector, int result)
> > +{
> > + struct audit_buffer *ab;
> > +
> > + if (audit_enabled == AUDIT_OFF)
> > + return;
> > +
> > + ab = audit_log_start(audit_context(), GFP_KERNEL, AUDIT_DM);
> > + if (unlikely(!ab))
> > + return;
> > +
> > + audit_log_format(ab, "module=%s dev=%d:%d op=%s sector=%llu res=%d",
> > + dm_msg_prefix, MAJOR(bio->bi_bdev->bd_dev),
> > + MINOR(bio->bi_bdev->bd_dev), op, sector, result);
> > + audit_log_end(ab);
> > +}
> > +EXPORT_SYMBOL_GPL(dm_audit_log_bio);
> > +
> > +void dm_audit_log_target(const char *dm_msg_prefix, const char *op,
> > + struct dm_target *ti, int result)
> > +{
> > + struct audit_buffer *ab;
> > + struct mapped_device *md = dm_table_get_md(ti->table);
> > +
> > + if (audit_enabled == AUDIT_OFF)
> > + return;
> > +
> > + ab = audit_log_start(audit_context(), GFP_KERNEL, AUDIT_DM);
> > + if (unlikely(!ab))
> > + return;
> > +
> > + audit_log_format(ab, "module=%s dev=%s op=%s",
> > + dm_msg_prefix, dm_device_name(md), op);
> > +
> > + if (!result && !strcmp("ctr", op))
> > + audit_log_format(ab, " error_msg='%s'", ti->error);
> > + audit_log_format(ab, " res=%d", result);
> > + audit_log_end(ab);
> > +}
>
> Generally speaking we try to keep a consistent format and ordering
> within a given audit record type. However you appear to have at least
> three different formats for the AUDIT_DM record in this patch:
>
> "... module=%s dev=%d:%d op=%s sector=%llu res=%d"
> "... module=%s dev=%s op=%s error_msg='%s' res=%d"
> "... module=%s dev=%s op=%s res=%d"
>
> The first thing that jumps out is that some fields, e.g. "sector", are
> not always present in the record; we typically handle this by using a
> "?" for the field value in those cases where you would otherwise drop
> the field from the record, for example the following record:
>
> "... module=%s dev=%s op=%s res=%d"
>
> ... would be rewritten like this:
>
> "... module=%s dev=%s op=%s sector=? res=%d"
Well, I didn't know that.
For target creation and destruction a sector is not relevant.
So would it also be an option for you if we just define two different
type of messages like this in audit.h?
#define AUDIT_DM_CTRL 1336 /* Device Mapper target control */
#define AUDIT_DM_EVENT 1337 /* Device Mapper events */
>
> The second thing that I noticed is that the "dev" field changes from a
> "major:minor" number representation to an arbitrary string value, e.g.
> "dev=%s". This generally isn't something we do with audit records,
> please stick to a single representation for a given audit
> record-type/field combination.
dm_device_name(md) already does provide a major:minor in string
representation, that is why I used it
directly with dev=%s
and in the bio case build it up manually out of major and minor
of the bdev.
I see two options here to be more clear on this in the code.
First, just provide a
comment or second use the major minor directly from
dm_disk(md)->major, dm_disk(md)->first_minor.
I'm not sure what's better here.
Thanks,
Michael
