diff mbox series

[v3,1/2] capability: Introduce capability API

Message ID 20190823094621.21747-1-rpalethorpe@suse.com
State Superseded
Headers show
Series [v3,1/2] capability: Introduce capability API | expand

Commit Message

Richard Palethorpe Aug. 23, 2019, 9:46 a.m. UTC
Allow users to easily ensure particular capabilities are either present or not
present during testing without requiring libcap.

Signed-off-by: Richard Palethorpe <rpalethorpe@suse.com>
Reviewed-by: Yang Xu <xuyang2018.jy@cn.fujitsu.com>
Reviewed-by: Jan Stancek <jstancek@redhat.com>
Reviewed-by: Li Wang <liwang@redhat.com>
Reviewed-by: Cyril Hrubis <chrubis@suse.cz>
---

V3:
* Rebase on master after guard buffers patch
* close socket in test
* Fix checking if cap is permitted, but not effective
* checkpatch fixes

 doc/test-writing-guidelines.txt |  79 ++++++++++++++++++++++
 include/lapi/capability.h       |  27 ++++++++
 include/tst_capability.h        |  48 ++++++++++++++
 include/tst_test.h              |   6 ++
 lib/tst_capability.c            | 112 ++++++++++++++++++++++++++++++++
 lib/tst_test.c                  |   3 +
 6 files changed, 275 insertions(+)
 create mode 100644 include/lapi/capability.h
 create mode 100644 include/tst_capability.h
 create mode 100644 lib/tst_capability.c

Comments

Li Wang Aug. 28, 2019, 10:43 a.m. UTC | #1
> The capability will be dropped in between 'setup' and 'run'.

I'm not sure to put this cap function behind 'setup' is a better choice.

Although it provides more capability in different test phase and makes
test flexible, that also involves more complexity for LTP users,
sometimes test needs to spawn children in the 'setup' and do more
testing in next 'run' phase, which obviously makes us have to consider
more in this case writing.

Anyway, the patchset looks good. We can do some improvements in real
situations then.

    Acked-by: Li Wang <liwang@redhat.com>
Richard Palethorpe Aug. 28, 2019, 11:58 a.m. UTC | #2
Hello Li,

Li Wang <liwang@redhat.com> writes:

>> The capability will be dropped in between 'setup' and 'run'.
>
> I'm not sure to put this cap function behind 'setup' is a better
> choice.
>
> Although it provides more capability in different test phase and makes
> test flexible, that also involves more complexity for LTP users,
> sometimes test needs to spawn children in the 'setup' and do more
> testing in next 'run' phase, which obviously makes us have to consider
> more in this case writing.

Children will need to drop and check for privs themselves anyway unless
one uses ambient privileges (which I guess could still be overriden by
the environment).

Maybe it would make sense to check for privileges before setup. However
I can't think of a situation where one would want to drop them before
setup. Meanwhile it seems likely that setup requires privs, but the test
should not have them.

>
> Anyway, the patchset looks good. We can do some improvements in real
> situations then.
>
>     Acked-by: Li Wang <liwang@redhat.com>


--
Thank you,
Richard.
Petr Vorel Aug. 29, 2019, 9:08 p.m. UTC | #3
Hi Richie,

> Hello Li,

> Li Wang <liwang@redhat.com> writes:

> >> The capability will be dropped in between 'setup' and 'run'.

> > I'm not sure to put this cap function behind 'setup' is a better
> > choice.

> > Although it provides more capability in different test phase and makes
> > test flexible, that also involves more complexity for LTP users,
> > sometimes test needs to spawn children in the 'setup' and do more
> > testing in next 'run' phase, which obviously makes us have to consider
> > more in this case writing.

> Children will need to drop and check for privs themselves anyway unless
> one uses ambient privileges (which I guess could still be overriden by
> the environment).

> Maybe it would make sense to check for privileges before setup. However
> I can't think of a situation where one would want to drop them before
> setup. Meanwhile it seems likely that setup requires privs, but the test
> should not have them.

+1

Nice work.
Reviewed-by: Petr Vorel <pvorel@suse.cz>

There is a warning, but I guess that's just gcc being paranoid:
test_guarded_buf.c:93:1: warning: missing initializer for field ‘caps’ of ‘struct tst_test’ [-Wmissing-field-initializers]
   93 | };
      | ^
In file included from test_guarded_buf.c:12:
../../include/tst_test.h:214:18: note: ‘caps’ declared here
  214 |  struct tst_cap *caps;
      |                  ^~~~


Kind regards,
Petr
Cyril Hrubis Aug. 30, 2019, 2:48 p.m. UTC | #4
Hi!
> There is a warning, but I guess that's just gcc being paranoid:
> test_guarded_buf.c:93:1: warning: missing initializer for field ???caps??? of ???struct tst_test??? [-Wmissing-field-initializers]
>    93 | };
>       | ^
> In file included from test_guarded_buf.c:12:
> ../../include/tst_test.h:214:18: note: ???caps??? declared here
>   214 |  struct tst_cap *caps;
>       |                  ^~~~

That sounds like a gcc bug since for designated initializers omitted
fields are implicitly initialized to zero, there is no point in
producing warnings like this one. I guess that we trigger that by
inlining a structure that is inialized by mixing plain and designated
initalizers in the middle of the structure.
diff mbox series

Patch

diff --git a/doc/test-writing-guidelines.txt b/doc/test-writing-guidelines.txt
index 1e933c49e..d21b89bd4 100644
--- a/doc/test-writing-guidelines.txt
+++ b/doc/test-writing-guidelines.txt
@@ -1769,6 +1769,85 @@  setting up the size or struct iovec, which is allocated recursively including
 the individual buffers as described by an '-1' terminated array of buffer
 sizes.
 
+2.2.32 Adding and removing capabilities
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+Some tests may require the presence or absence of particular
+capabilities. Using the API provided by 'tst_capability.h' the test author can
+try to ensure that some capabilities are either present or absent during the
+test.
+
+For example; below we try to create a raw socket, which requires
+CAP_NET_ADMIN. During setup we should be able to do it, then during run it
+should be impossible. The LTP capability library should drop CAP_NET_RAW
+(assuming we have it) after setup completes.
+
+[source,c]
+--------------------------------------------------------------------------------
+#include "tst_test.h"
+#include "tst_capability.h"
+#include "tst_safe_net.h"
+
+#include "lapi/socket.h"
+
+static void run(void)
+{
+	TEST(socket(AF_INET, SOCK_RAW, 1));
+	if (TST_RET > -1) {
+		tst_res(TFAIL, "Created raw socket");
+	} else if (TST_ERR != EPERM) {
+		tst_res(TBROK | TTERRNO,
+			"Failed to create socket for wrong reason");
+	} else {
+		tst_res(TPASS | TTERRNO, "Didn't create raw socket");
+	}
+}
+
+static void setup(void)
+{
+	TEST(socket(AF_INET, SOCK_RAW, 1));
+	if (TST_RET < 0)
+		tst_brk(TCONF | TTERRNO, "We don't have CAP_NET_RAW to begin with");
+
+	SAFE_CLOSE(TST_RET);
+}
+
+static struct tst_test test = {
+	.setup = setup,
+	.test_all = run,
+	.caps = (struct tst_cap []) {
+		TST_CAP(TST_CAP_DROP, CAP_NET_RAW),
+		{}
+	},
+};
+--------------------------------------------------------------------------------
+
+Look at the test struct at the bottom. We have filled in the 'caps' field with
+a NULL terminated array containing a single 'tst_cap'. This indicates to the
+library that we should drop CAP_NET_RAW if we have it. The capability will be
+dropped in between 'setup' and 'run'.
+
+[source,c]
+--------------------------------------------------------------------------------
+static struct tst_test test = {
+	.test_all = run,
+	.caps = (struct tst_cap []) {
+		TST_CAP(TST_CAP_REQ, CAP_NET_RAW),
+		TST_CAP(TST_CAP_DROP, CAP_SYS_ADMIN),
+		{}
+	},
+};
+--------------------------------------------------------------------------------
+
+Here we request 'CAP_NET_RAW', but drop 'CAP_SYS_ADMIN'. If the capability is
+in the permitted set, but not the effective set, the library will try to
+permit it. If it is not in the permitted set, then it will fail with 'TCONF'.
+
+This API does not require 'libcap' to be installed. However it has limited
+features relative to 'libcap'. It only tries to add or remove capabilities
+from the effective set. This means that tests which need to spawn child
+processes with various capabilities will probably need 'libcap'.
+
 2.3 Writing a testcase in shell
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 
diff --git a/include/lapi/capability.h b/include/lapi/capability.h
new file mode 100644
index 000000000..02d7a9fda
--- /dev/null
+++ b/include/lapi/capability.h
@@ -0,0 +1,27 @@ 
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Copyright (c) 2019 Richard Palethorpe <rpalethorpe@suse.com>
+ */
+
+#ifndef LAPI_CAPABILITY_H
+#define LAPI_CAPABILITY_H
+
+#include "config.h"
+
+#ifdef HAVE_SYS_CAPABILITY_H
+# include <sys/capability.h>
+#endif
+
+#ifndef CAP_SYS_ADMIN
+# define CAP_SYS_ADMIN        21
+#endif
+
+#ifndef CAP_TO_INDEX
+# define CAP_TO_INDEX(x)     ((x) >> 5)
+#endif
+
+#ifndef CAP_TO_MASK
+# define CAP_TO_MASK(x)      (1 << ((x) & 31))
+#endif
+
+#endif
diff --git a/include/tst_capability.h b/include/tst_capability.h
new file mode 100644
index 000000000..6b5a140a3
--- /dev/null
+++ b/include/tst_capability.h
@@ -0,0 +1,48 @@ 
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/*
+ * Copyright (c) 2019 Richard Palethorpe <rpalethorpe@suse.com>
+ */
+/**
+ * @file tst_capability.h
+ *
+ * Limited capability operations without libcap.
+ */
+
+#ifndef TST_CAPABILITY_H
+#define TST_CAPABILITY_H
+
+#include <stdint.h>
+
+#include "lapi/capability.h"
+
+#define TST_CAP_DROP 1
+#define TST_CAP_REQ  (1 << 1)
+
+#define TST_CAP(action, capability) {action, capability, #capability}
+
+struct tst_cap_user_header {
+	uint32_t version;
+	int pid;
+};
+
+struct tst_cap_user_data {
+	uint32_t effective;
+	uint32_t permitted;
+	uint32_t inheritable;
+};
+
+struct tst_cap {
+	uint32_t action;
+	uint32_t id;
+	char *name;
+};
+
+int tst_capget(struct tst_cap_user_header *hdr,
+	       struct tst_cap_user_data *data);
+int tst_capset(struct tst_cap_user_header *hdr,
+	       const struct tst_cap_user_data *data);
+
+void tst_cap_action(struct tst_cap *cap);
+void tst_cap_setup(struct tst_cap *cap);
+
+#endif
diff --git a/include/tst_test.h b/include/tst_test.h
index cdeaf6ad0..84acf2c59 100644
--- a/include/tst_test.h
+++ b/include/tst_test.h
@@ -36,6 +36,7 @@ 
 #include "tst_sys_conf.h"
 #include "tst_coredump.h"
 #include "tst_buffers.h"
+#include "tst_capability.h"
 
 /*
  * Reports testcase result.
@@ -206,6 +207,11 @@  struct tst_test {
 	 * NULL-terminated array to be allocated buffers.
 	 */
 	struct tst_buffers *bufs;
+
+	/*
+	 * NULL-terminated array of capability settings
+	 */
+	struct tst_cap *caps;
 };
 
 /*
diff --git a/lib/tst_capability.c b/lib/tst_capability.c
new file mode 100644
index 000000000..21e8fd5ff
--- /dev/null
+++ b/lib/tst_capability.c
@@ -0,0 +1,112 @@ 
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright (c) 2019 Richard Palethorpe <rpalethorpe@suse.com>
+ */
+
+#include <string.h>
+
+#define TST_NO_DEFAULT_MAIN
+#include "tst_test.h"
+#include "tst_capability.h"
+
+#include "lapi/syscalls.h"
+
+/**
+ * Get the capabilities as decided by hdr.
+ *
+ * Note that the memory pointed to by data should be large enough to store two
+ * structs.
+ */
+int tst_capget(struct tst_cap_user_header *hdr,
+	       struct tst_cap_user_data *data)
+{
+	return tst_syscall(__NR_capget, hdr, data);
+}
+
+/**
+ * Set the capabilities as decided by hdr and data
+ *
+ * Note that the memory pointed to by data should be large enough to store two
+ * structs.
+ */
+int tst_capset(struct tst_cap_user_header *hdr,
+	       const struct tst_cap_user_data *data)
+{
+	return tst_syscall(__NR_capset, hdr, data);
+}
+
+static void do_cap_drop(uint32_t *set, uint32_t mask, const struct tst_cap *cap)
+{
+	if (*set & mask) {
+		tst_res(TINFO, "Dropping %s(%d)", cap->name, cap->id);
+		*set &= ~mask;
+	}
+}
+
+static void do_cap_req(uint32_t *permitted, uint32_t *effective, uint32_t mask,
+		       const struct tst_cap *cap)
+{
+	if (!(*permitted & mask))
+		tst_brk(TCONF, "Need %s(%d)", cap->name, cap->id);
+
+	if (!(*effective & mask)) {
+		tst_res(TINFO, "Permitting %s(%d)", cap->name, cap->id);
+		*effective |= mask;
+	}
+}
+
+/**
+ * Add, check or remove capabilities
+ *
+ * Takes a NULL terminated array of structs which describe whether some
+ * capabilities are needed or not.
+ *
+ * It will attempt to drop or add capabilities to the effective set. It will
+ * try to detect if this is needed and whether it can or can't be done. If it
+ * clearly can not add a privilege to the effective set then it will return
+ * TCONF. However it may fail for some other reason and return TBROK.
+ *
+ * This only tries to change the effective set. Some tests may need to change
+ * the inheritable and ambient sets, so that child processes retain some
+ * capability.
+ */
+void tst_cap_action(struct tst_cap *cap)
+{
+	struct tst_cap_user_header hdr = {
+		.version = 0x20080522,
+		.pid = tst_syscall(__NR_gettid),
+	};
+	struct tst_cap_user_data cur[2] = { {0} };
+	struct tst_cap_user_data new[2] = { {0} };
+	uint32_t act = cap->action;
+	uint32_t *pE = &new[CAP_TO_INDEX(cap->id)].effective;
+	uint32_t *pP = &new[CAP_TO_INDEX(cap->id)].permitted;
+	uint32_t mask = CAP_TO_MASK(cap->id);
+
+	if (tst_capget(&hdr, cur))
+		tst_brk(TBROK | TTERRNO, "tst_capget()");
+
+	memcpy(new, cur, sizeof(new));
+
+	switch (act) {
+	case TST_CAP_DROP:
+		do_cap_drop(pE, mask, cap);
+		break;
+	case TST_CAP_REQ:
+		do_cap_req(pP, pE, mask, cap);
+		break;
+	default:
+		tst_brk(TBROK, "Unrecognised action %d", cap->action);
+	}
+
+	if (memcmp(cur, new, sizeof(new)) && tst_capset(&hdr, new))
+		tst_brk(TBROK | TERRNO, "tst_capset(%s)", cap->name);
+}
+
+void tst_cap_setup(struct tst_cap *caps)
+{
+	struct tst_cap *cap;
+
+	for (cap = caps; cap->action; cap++)
+		tst_cap_action(cap);
+}
diff --git a/lib/tst_test.c b/lib/tst_test.c
index 39f261472..b76b0f0b5 100644
--- a/lib/tst_test.c
+++ b/lib/tst_test.c
@@ -896,6 +896,9 @@  static void do_test_setup(void)
 
 	if (main_pid != getpid())
 		tst_brk(TBROK, "Runaway child in setup()!");
+
+	if (tst_test->caps)
+		tst_cap_setup(tst_test->caps);
 }
 
 static void do_cleanup(void)