selftests/bpf: add test for arena vma split and fork rejection
Subsystems (3)
PATCH 1/1 selftests/bpf: add test for arena vma split and fork rejection
Verify that an arena map's mmap()ed VMA cannot be split with a partial munmap() or inherited across fork().
Before commit 4fddde2a732d ("bpf: Fix use-after-free in arena_vm_close on fork") the arena VMA had no .may_split callback and was not marked VM_DONTCOPY, so a splitting munmap() and a fork() both succeeded. The fork() case left the child with a VMA whose vml->vma still pointed at the parent, dangling after the parent unmapped and leading to a use-after-free if the child freed arena pages.
The test creates an arena map, mmap()s it, and asserts that a splitting munmap() fails with -EINVAL and that the child of a fork() does not inherit the arena VMA.
--- .../selftests/bpf/prog_tests/arena_fork.c | 58 +++++++++++++++++++ 1 file changed, 58 insertions(+) create mode 100644 tools/testing/selftests/bpf/prog_tests/arena_fork.c
tools/testing/selftests/bpf/prog_tests/arena_fork.c new file +58 -0
@@ -0,0 +1,58 @@ +// SPDX-License-Identifier: GPL-2.0 +#include <test_progs.h> +#include <errno.h> +#include <sys/mman.h> +#include <sys/wait.h> + +/* Regression test for commit 4fddde2a732d ("bpf: Fix use-after-free in +* arena_vm_close on fork"): an arena VMA must reject a splitting munmap() +* (.may_split) and must not be inherited across fork() (VM_DONTCOPY). On +* an unfixed kernel both operations succeed. +*/ + +#define NR_PAGES 3 + +void test_arena_fork (void ) +{ + LIBBPF_OPTS(bpf_map_create_opts, opts, .map_flags = BPF_F_MMAPABLE); +long ps =sysconf (_SC_PAGESIZE); +size_t sz = (size_t )NR_PAGES * ps; +int fd, ret, status, err; +void *area; +pid_t pid; + + fd =bpf_map_create (BPF_MAP_TYPE_ARENA ,"arena_fork" ,0 ,0 ,NR_PAGES , &opts); +if (!ASSERT_OK_FD (fd,"arena map create" )) +return ; + + area =mmap (NULL , sz,PROT_READ |PROT_WRITE ,MAP_SHARED , fd,0 ); +if (!ASSERT_NEQ (area,MAP_FAILED ,"mmap arena" )) + goto close_fd; + +/* A split of the arena VMA must be rejected (.may_split). */ + ret =munmap ((char *)area + ps, ps); + err = errno; +ASSERT_ERR (ret,"split munmap rejected" ); +ASSERT_EQ (err,EINVAL ,"split munmap errno" ); + +/* The child of a fork() must not inherit the arena VMA (VM_DONTCOPY); +* mincore() returns ENOMEM for the unmapped range. +*/ + pid =fork (); +if (ASSERT_GE (pid,0 ,"fork" )) { +if (pid ==0 ) { +unsigned char vec; + +_exit (mincore (area, ps, &vec) <0 && errno ==ENOMEM ?0 :1 ); + } +while ((ret =waitpid (pid, &status,0 )) <0 && errno ==EINTR ) + ; +if (ASSERT_EQ (ret, pid,"waitpid" )) +ASSERT_TRUE (WIFEXITED (status) &&WEXITSTATUS (status) ==0 , +"child must not inherit arena vma" ); + } + +munmap (area, sz); +close_fd: +close (fd); +}
base-commit: 174914ea551314c52a61713b9c4bde9e42d48073
Signature · 1 line
2.43.0
[email protected]
AI reviewThank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Low] Multi-line comments do not follow the BPF subsystem style guide. --
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Low] Multi-line comments do not follow the BPF subsystem style guide.
--
commit 68912cf41a7e5ee9f6445ed2bc4cecbbb14bad14
Author: Ruslan Valiyev <[email protected]>
selftests/bpf: add test for arena vma split and fork rejection
This commit adds a selftest to verify that an arena map's mmap()ed VMA cannot be split with a partial munmap() or inherited across fork().
Quoted 7 lines
tools/testing/selftests/bpf/prog_tests/arena_fork.c new file +1
-0@@ -0,0 +1,58 @@ +// SPDX-License-Identifier: GPL-2.0
tools/testing/selftests/bpf/prog_tests/arena_fork.c line 11
Quoted 10 lines
tools/testing/selftests/bpf/prog_tests/arena_fork.c new file +10
-0+#include <test_progs.h> +#include <errno.h> +#include <sys/mman.h> +#include <sys/wait.h> + +/* Regression test for commit 4fddde2a732d ("bpf: Fix use-after-free in +* arena_vm_close on fork"): an arena VMA must reject a splitting munmap() +* (.may_split) and must not be inherited across fork() (VM_DONTCOPY). On +* an unfixed kernel both operations succeed. +*/
[Severity: Low]
This isn't a bug, but does this multi-line comment format follow the BPF subsystem style? The subsystem guidelines state that multi-line comments should have the opening /* on its own line.
[ ... ]
tools/testing/selftests/bpf/prog_tests/arena_fork.c line 40
Quoted 9 lines
tools/testing/selftests/bpf/prog_tests/arena_fork.c new file +9
-0+/* A split of the arena VMA must be rejected (.may_split). */ + ret =munmap ((char *)area + ps, ps); + err = errno; +ASSERT_ERR (ret,"split munmap rejected" ); +ASSERT_EQ (err,EINVAL ,"split munmap errno" ); + +/* The child of a fork() must not inherit the arena VMA (VM_DONTCOPY); +* mincore() returns ENOMEM for the unmapped range. +*/
[Severity: Low]
This isn't a bug, but does this comment also need the opening /* on its own line to match the preferred BPF multi-line comment style?
[ ... ]
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
Emil Tsalapatis
reviewerQuoted 27 lines · Ruslan Valiyev
Verify that an arena map's mmap()ed VMA cannot be split with a partial
munmap()or inherited acrossfork().Before commit
4fddde2a732d("bpf: Fix use-after-free inarena_vm_closeon fork") the arena VMA had no .may_split callback and was not markedVM_DONTCOPY, so a splittingmunmap()and afork()both succeeded. Thefork()case left the child with a VMA whosevml->vmastill pointed at the parent, dangling after the parent unmapped and leading to a use-after-free if the child freed arena pages.The test creates an arena map, mmap()s it, and asserts that a splitting
munmap()fails with-EINVALand that the child of afork()does not inherit the arena VMA.Signed-off-by Ruslan Valiyev <[email protected]>--- .../selftests/bpf/prog_tests/arena_fork.c | 58 +++++++++++++++++++ 1 file changed, 58 insertions(+) create mode 100644 tools/testing/selftests/bpf/prog_tests/arena_fork.ctools/testing/selftests/bpf/prog_tests/arena_fork.c new file +1
-0@@ -0,0 +1,58 @@ +// SPDX-License-Identifier: GPL-2.0
tools/testing/selftests/bpf/prog_tests/arena_fork.c line 11
Quoted 10 lines
tools/testing/selftests/bpf/prog_tests/arena_fork.c new file +10
-0+#include <test_progs.h> +#include <errno.h> +#include <sys/mman.h> +#include <sys/wait.h> + +/* Regression test for commit 4fddde2a732d ("bpf: Fix use-after-free in +* arena_vm_close on fork"): an arena VMA must reject a splitting munmap() +* (.may_split) and must not be inherited across fork() (VM_DONTCOPY). On +* an unfixed kernel both operations succeed. +*/
Remove this comment, we don't frame tests as regression tests for specific commits. The details about .may_split and VM_DONTCOPY also don't matter. Something like
/* Make sure arena mappings cannot be split or inherited. */
The name is also misleading, this is two tests in one. It's not just the fork test, you are also checking may_split. I think it's worth splitting this into two separate tests, one for partial unmaps and one for forks.
Quoted 25 lines
tools/testing/selftests/bpf/prog_tests/arena_fork.c new file +25
-0+ +#define NR_PAGES 3 + +void test_arena_fork (void ) +{ + LIBBPF_OPTS(bpf_map_create_opts, opts, .map_flags = BPF_F_MMAPABLE); +long ps =sysconf (_SC_PAGESIZE); +size_t sz = (size_t )NR_PAGES * ps; +int fd, ret, status, err; +void *area; +pid_t pid; + + fd =bpf_map_create (BPF_MAP_TYPE_ARENA ,"arena_fork" ,0 ,0 ,NR_PAGES , &opts); +if (!ASSERT_OK_FD (fd,"arena map create" )) +return ; + + area =mmap (NULL , sz,PROT_READ |PROT_WRITE ,MAP_SHARED , fd,0 ); +if (!ASSERT_NEQ (area,MAP_FAILED ,"mmap arena" )) + goto close_fd; + +/* A split of the arena VMA must be rejected (.may_split). */ + ret =munmap ((char *)area + ps, ps); + err = errno; +ASSERT_ERR (ret,"split munmap rejected" ); +ASSERT_EQ (err,EINVAL ,"split munmap errno" );
tools/testing/selftests/bpf/prog_tests/arena_fork.c line 46
Quoted 10 lines
tools/testing/selftests/bpf/prog_tests/arena_fork.c new file +10
-0+ +/* The child of a fork() must not inherit the arena VMA (VM_DONTCOPY); +* mincore() returns ENOMEM for the unmapped range. +*/ + pid =fork (); +if (ASSERT_GE (pid,0 ,"fork" )) { +if (pid ==0 ) { +unsigned char vec; + +_exit (mincore (area, ps, &vec) <0 && errno ==ENOMEM ?0 :1 );
This expression is needlessly dense, unpack it.
Quoted 14 lines
tools/testing/selftests/bpf/prog_tests/arena_fork.c new file +12
-0+ } +while ((ret =waitpid (pid, &status,0 )) <0 && errno ==EINTR ) + ; +if (ASSERT_EQ (ret, pid,"waitpid" )) +ASSERT_TRUE (WIFEXITED (status) &&WEXITSTATUS (status) ==0 , +"child must not inherit arena vma" ); + } + +munmap (area, sz); +close_fd: +close (fd); +}base-commit:
174914ea551314c52a61713b9c4bde9e42d48073
PATCH v2 1/1 selftests/bpf: add arena split and fork tests
Add two tests for an arena map's mmap()ed region: arena_split checks that a partial munmap() that would split the mapping fails with -EINVAL, and arena_fork checks that a forked child does not inherit the mapping (mincore() reports the range as unmapped).
---
v2:
- Drop the regression-test framing and per-callback details; use short one-line comments.
- Split into two separate tests, arena_split (partial unmap) and arena_fork.
- Unpack the mincore() child check into explicit statements.
.../selftests/bpf/prog_tests/arena_fork.c | 53 +++++++++++++++++++ .../selftests/bpf/prog_tests/arena_split.c | 34 ++++++++++++ 2 files changed, 87 insertions(+) create mode 100644 tools/testing/selftests/bpf/prog_tests/arena_fork.c create mode 100644 tools/testing/selftests/bpf/prog_tests/arena_split.c
tools/testing/selftests/bpf/prog_tests/arena_fork.c new file +53 -0
@@ -0,0 +1,53 @@ +// SPDX-License-Identifier: GPL-2.0 +#include <test_progs.h> +#include <errno.h> +#include <sys/mman.h> +#include <sys/wait.h> + +/* Make sure an arena mapping is not inherited across fork(). */ + +#define NR_PAGES 3 + +void test_arena_fork (void ) +{ + LIBBPF_OPTS(bpf_map_create_opts, opts, .map_flags = BPF_F_MMAPABLE); +long ps =sysconf (_SC_PAGESIZE); +size_t sz = (size_t )NR_PAGES * ps; +void *area; +int fd, ret, status; +pid_t pid; + + fd =bpf_map_create (BPF_MAP_TYPE_ARENA ,"arena_fork" ,0 ,0 ,NR_PAGES , &opts); +if (!ASSERT_OK_FD (fd,"arena map create" )) +return ; + + area =mmap (NULL , sz,PROT_READ |PROT_WRITE ,MAP_SHARED , fd,0 ); +if (!ASSERT_NEQ (area,MAP_FAILED ,"mmap arena" )) + goto close_fd; + + pid =fork (); +if (pid ==0 ) { +unsigned char vec; +int rc; + +/* If the mapping was not inherited the range is unmapped in +* the child, so mincore() fails with ENOMEM. A success means +* the child wrongly inherited the mapping. +*/ + rc =mincore (area, ps, &vec); +if (rc ==0 ) +_exit (1 ); +_exit (errno ==ENOMEM ?0 :2 ); + } +if (ASSERT_GE (pid,0 ,"fork" )) { +while ((ret =waitpid (pid, &status,0 )) <0 && errno ==EINTR ) + ; +if (ASSERT_EQ (ret, pid,"waitpid" )) +ASSERT_TRUE (WIFEXITED (status) &&WEXITSTATUS (status) ==0 , +"child must not inherit arena mapping" ); + } + +munmap (area, sz); +close_fd: +close (fd); +}
tools/testing/selftests/bpf/prog_tests/arena_split.c new file +34 -0
@@ -0,0 +1,34 @@ +// SPDX-License-Identifier: GPL-2.0 +#include <test_progs.h> +#include <errno.h> +#include <sys/mman.h> + +/* Make sure an arena mapping cannot be split by a partial munmap(). */ + +#define NR_PAGES 3 + +void test_arena_split (void ) +{ + LIBBPF_OPTS(bpf_map_create_opts, opts, .map_flags = BPF_F_MMAPABLE); +long ps =sysconf (_SC_PAGESIZE); +size_t sz = (size_t )NR_PAGES * ps; +void *area; +int fd, ret, err; + + fd =bpf_map_create (BPF_MAP_TYPE_ARENA ,"arena_split" ,0 ,0 ,NR_PAGES , &opts); +if (!ASSERT_OK_FD (fd,"arena map create" )) +return ; + + area =mmap (NULL , sz,PROT_READ |PROT_WRITE ,MAP_SHARED , fd,0 ); +if (!ASSERT_NEQ (area,MAP_FAILED ,"mmap arena" )) + goto close_fd; + + ret =munmap ((char *)area + ps, ps); + err = errno; +if (ASSERT_ERR (ret,"split munmap" )) +ASSERT_EQ (err,EINVAL ,"split munmap errno" ); + +munmap (area, sz); +close_fd: +close (fd); +}
base-commit: 174914ea551314c52a61713b9c4bde9e42d48073
Signature · 1 line
2.43.0
[email protected]
AI reviewRe: [PATCH v2 bpf-next] selftests/bpf: add arena split and fork tests
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Low] Multi-line comment does not follow the required BPF subsystem comment style. --
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Low] Multi-line comment does not follow the required BPF subsystem comment style. --
commit f3309fb333fa0cbe670a595203f862bdefb2a073
Author: Ruslan Valiyev <[email protected]>
selftests/bpf: add arena split and fork tests
This commit adds two tests for an arena map's mmap()ed region. The arena_split test checks that a partial munmap() that would split the mapping fails with EINVAL, while arena_fork verifies that a forked child does not inherit the mapping.
Quoted 5 lines
tools/testing/selftests/bpf/prog_tests/arena_fork.c new file +0
-0
[ ... ]
tools/testing/selftests/bpf/prog_tests/arena_fork.c line 36
Quoted 9 lines
tools/testing/selftests/bpf/prog_tests/arena_fork.c new file +9
-0+ pid =fork (); +if (pid ==0 ) { +unsigned char vec; +int rc; + +/* If the mapping was not inherited the range is unmapped in +* the child, so mincore() fails with ENOMEM. A success means +* the child wrongly inherited the mapping. +*/
[Severity: Low]
This isn't a bug, but does this multi-line comment in test_arena_fork() follow the preferred BPF subsystem comment style?
The BPF guidelines require the opening /* to be on its own line for multi-line comments, like this:
/*
* If the mapping was not inherited the range is unmapped in
* the child...
*/
Quoted 5 lines
tools/testing/selftests/bpf/prog_tests/arena_fork.c new file +5
-0+ rc =mincore (area, ps, &vec); +if (rc ==0 ) +_exit (1 ); +_exit (errno ==ENOMEM ?0 :2 ); + }
[ ... ]
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
[email protected]
AI reviewRe: [PATCH v2 bpf-next] selftests/bpf: add arena split and fork tests
In reply to Ruslan ValiyevThis isn't a bug, but does this multi-line comment match the preferred BPF subsystem comment style? The guidelines ask for the opening /* to be on its own line, for example:
Quoted 6 lines
tools/testing/selftests/bpf/prog_tests/arena_fork.c new file +0
-0@@ -0,0 +1,53 @@
[ ... ]
tools/testing/selftests/bpf/prog_tests/arena_fork.c line 36
Quoted 9 lines
tools/testing/selftests/bpf/prog_tests/arena_fork.c new file +9
-0+ pid =fork (); +if (pid ==0 ) { +unsigned char vec; +int rc; + +/* If the mapping was not inherited the range is unmapped in +* the child, so mincore() fails with ENOMEM. A success means +* the child wrongly inherited the mapping. +*/
This isn't a bug, but does this multi-line comment match the preferred BPF subsystem comment style? The guidelines ask for the opening /* to be on its own line, for example:
/*
* If the mapping was not inherited the range is unmapped in
* the child, so mincore() fails with ENOMEM. A success means
* the child wrongly inherited the mapping.
*/
This was raised on v2 and does not appear to be addressed yet.
---
AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md
CI run summary: https://github.com/kernel-patches/bpf/actions/runs/27013893868
Re: [PATCH v2 bpf-next] selftests/bpf: add arena split and fork tests
In reply to Ruslan ValiyevQuoted 7 lines · Ruslan Valiyev
Add two tests for an arena map's mmap()ed region:
arena_splitchecks that a partialmunmap()that would split the mapping fails with-EINVAL, andarena_forkchecks that a forked child does not inherit the mapping (mincore()reports the range as unmapped).Signed-off-by Ruslan Valiyev <[email protected]>---
Two tests are quite small and share common setup. Does it make sense to put them into the same file, extract common function and turn them into 2 subtests. You can check other BPF selftests to see how to implement subtests.
Quoted 114 lines
v2:
- Drop the regression-test framing and per-callback details; use short one-line comments.
- Split into two separate tests,arena_split(partial unmap) andarena_fork.
- Unpack themincore()child check into explicit statements..../selftests/bpf/prog_tests/arena_fork.c | 53 +++++++++++++++++++ .../selftests/bpf/prog_tests/arena_split.c | 34 ++++++++++++ 2 files changed, 87 insertions(+) create mode 100644 tools/testing/selftests/bpf/prog_tests/arena_fork.c create mode 100644 tools/testing/selftests/bpf/prog_tests/arena_split.ctools/testing/selftests/bpf/prog_tests/arena_fork.c new file +53
-0@@ -0,0 +1,53 @@ +// SPDX-License-Identifier: GPL-2.0 +#include <test_progs.h> +#include <errno.h> +#include <sys/mman.h> +#include <sys/wait.h> + +/* Make sure an arena mapping is not inherited across fork(). */ + +#define NR_PAGES 3 + +void test_arena_fork (void ) +{ + LIBBPF_OPTS(bpf_map_create_opts, opts, .map_flags = BPF_F_MMAPABLE); +long ps =sysconf (_SC_PAGESIZE); +size_t sz = (size_t )NR_PAGES * ps; +void *area; +int fd, ret, status; +pid_t pid; + + fd =bpf_map_create (BPF_MAP_TYPE_ARENA ,"arena_fork" ,0 ,0 ,NR_PAGES , &opts); +if (!ASSERT_OK_FD (fd,"arena map create" )) +return ; + + area =mmap (NULL , sz,PROT_READ |PROT_WRITE ,MAP_SHARED , fd,0 ); +if (!ASSERT_NEQ (area,MAP_FAILED ,"mmap arena" )) + goto close_fd; + + pid =fork (); +if (pid ==0 ) { +unsigned char vec; +int rc; + +/* If the mapping was not inherited the range is unmapped in +* the child, so mincore() fails with ENOMEM. A success means +* the child wrongly inherited the mapping. +*/ + rc =mincore (area, ps, &vec); +if (rc ==0 ) +_exit (1 ); +_exit (errno ==ENOMEM ?0 :2 ); + } +if (ASSERT_GE (pid,0 ,"fork" )) { +while ((ret =waitpid (pid, &status,0 )) <0 && errno ==EINTR ) + ; +if (ASSERT_EQ (ret, pid,"waitpid" )) +ASSERT_TRUE (WIFEXITED (status) &&WEXITSTATUS (status) ==0 , +"child must not inherit arena mapping" ); + } + +munmap (area, sz); +close_fd: +close (fd); +}tools/testing/selftests/bpf/prog_tests/arena_split.c new file +34
-0@@ -0,0 +1,34 @@ +// SPDX-License-Identifier: GPL-2.0 +#include <test_progs.h> +#include <errno.h> +#include <sys/mman.h> + +/* Make sure an arena mapping cannot be split by a partial munmap(). */ + +#define NR_PAGES 3 + +void test_arena_split (void ) +{ + LIBBPF_OPTS(bpf_map_create_opts, opts, .map_flags = BPF_F_MMAPABLE); +long ps =sysconf (_SC_PAGESIZE); +size_t sz = (size_t )NR_PAGES * ps; +void *area; +int fd, ret, err; + + fd =bpf_map_create (BPF_MAP_TYPE_ARENA ,"arena_split" ,0 ,0 ,NR_PAGES , &opts); +if (!ASSERT_OK_FD (fd,"arena map create" )) +return ; + + area =mmap (NULL , sz,PROT_READ |PROT_WRITE ,MAP_SHARED , fd,0 ); +if (!ASSERT_NEQ (area,MAP_FAILED ,"mmap arena" )) + goto close_fd; + + ret =munmap ((char *)area + ps, ps); + err = errno; +if (ASSERT_ERR (ret,"split munmap" )) +ASSERT_EQ (err,EINVAL ,"split munmap errno" ); + +munmap (area, sz); +close_fd: +close (fd); +}base-commit:
174914ea551314c52a61713b9c4bde9e42d48073
Re: [PATCH v2 bpf-next] selftests/bpf: add arena split and fork tests
Quoted 4 lines · Mykyta Yatsenko
Two tests are quite small and share common setup. Does it make sense to put them into the same file, extract common function and turn them into 2 subtests. You can check other BPF selftests to see how to implement subtests.
Sounds good, will do in v3: one file with a shared setup helper and two subtests, "split" and "fork".
Thanks,
Ruslan
PATCH v3 1/1 selftests/bpf: add tests for arena vma split and fork
In reply to Mykyta YatsenkoAdd a test that an arena map's mmap()ed region cannot be split by a partial munmap() (the "split" subtest, which expects -EINVAL) and is not inherited across fork() (the "fork" subtest, where mincore() reports the child's range as unmapped).
---
v3:
- Consolidate into one file with a shared setup helper and two subtests.
- Put the opening /* of the multi-line comment on its own line. v2:
- Drop the regression-test framing and per-callback details.
- Split the original single test; unpack the mincore() check.
.../selftests/bpf/prog_tests/arena_vma.c | 94 +++++++++++++++++++ 1 file changed, 94 insertions(+) create mode 100644 tools/testing/selftests/bpf/prog_tests/arena_vma.c
tools/testing/selftests/bpf/prog_tests/arena_vma.c new file +94 -0
@@ -0,0 +1,94 @@ +// SPDX-License-Identifier: GPL-2.0 +#include <test_progs.h> +#include <errno.h> +#include <sys/mman.h> +#include <sys/wait.h> + +/* Make sure arena mappings cannot be split or inherited. */ + +#define NR_PAGES 3 + +static int arena_mmap (int *fd,void **area,size_t *sz) +{ + LIBBPF_OPTS(bpf_map_create_opts, opts, .map_flags = BPF_F_MMAPABLE); + + *sz = (size_t )NR_PAGES *sysconf (_SC_PAGESIZE); + + *fd =bpf_map_create (BPF_MAP_TYPE_ARENA ,"arena_vma" ,0 ,0 ,NR_PAGES , &opts); +if (!ASSERT_OK_FD (*fd,"arena map create" )) +return -1 ; + + *area =mmap (NULL , *sz,PROT_READ |PROT_WRITE ,MAP_SHARED , *fd,0 ); +if (!ASSERT_NEQ (*area,MAP_FAILED ,"mmap arena" )) { +close (*fd); +return -1 ; + } +return 0 ; +} + +/* A partial munmap that would split the arena mapping must be rejected. */ +static void split_test (void ) +{ +long ps =sysconf (_SC_PAGESIZE); +void *area; +size_t sz; +int fd, ret, err; + +if (arena_mmap (&fd, &area, &sz)) +return ; + + ret =munmap ((char *)area + ps, ps); + err = errno; +if (ASSERT_ERR (ret,"split munmap" )) +ASSERT_EQ (err,EINVAL ,"split munmap errno" ); + +munmap (area, sz); +close (fd); +} + +/* A forked child must not inherit the arena mapping. */ +static void fork_test (void ) +{ +long ps =sysconf (_SC_PAGESIZE); +void *area; +size_t sz; +int fd, ret, status; +pid_t pid; + +if (arena_mmap (&fd, &area, &sz)) +return ; + + pid =fork (); +if (pid ==0 ) { +unsigned char vec; +int rc; + +/* +* If the mapping was not inherited the range is unmapped in +* the child, so mincore() fails with ENOMEM. A success means +* the child wrongly inherited the mapping. +*/ + rc =mincore (area, ps, &vec); +if (rc ==0 ) +_exit (1 ); +_exit (errno ==ENOMEM ?0 :2 ); + } +if (ASSERT_GE (pid,0 ,"fork" )) { +while ((ret =waitpid (pid, &status,0 )) <0 && errno ==EINTR ) + ; +if (ASSERT_EQ (ret, pid,"waitpid" )) +ASSERT_TRUE (WIFEXITED (status) &&WEXITSTATUS (status) ==0 , +"child must not inherit arena mapping" ); + } + +munmap (area, sz); +close (fd); +} + +void test_arena_vma (void ) +{ +if (test__start_subtest ("split" )) +split_test (); +if (test__start_subtest ("fork" )) +fork_test (); +}
base-commit: 174914ea551314c52a61713b9c4bde9e42d48073
Signature · 1 line
2.43.0
Re: [PATCH v3 bpf-next] selftests/bpf: add tests for arena vma split and fork
Quoted 6 lines · Ruslan Valiyev
Add a test that an arena map's mmap()ed region cannot be split by a partial
munmap()(the "split" subtest, which expects-EINVAL) and is not inherited acrossfork()(the "fork" subtest, wheremincore()reports the child's range as unmapped).Signed-off-by Ruslan Valiyev <[email protected]>
Quoted 115 lines
---
v3:
- Consolidate into one file with a shared setup helper and two subtests.
- Put the opening /* of the multi-line comment on its own line. v2:
- Drop the regression-test framing and per-callback details.
- Split the original single test; unpack themincore()check..../selftests/bpf/prog_tests/arena_vma.c | 94 +++++++++++++++++++ 1 file changed, 94 insertions(+) create mode 100644 tools/testing/selftests/bpf/prog_tests/arena_vma.ctools/testing/selftests/bpf/prog_tests/arena_vma.c new file +94
-0@@ -0,0 +1,94 @@ +// SPDX-License-Identifier: GPL-2.0 +#include <test_progs.h> +#include <errno.h> +#include <sys/mman.h> +#include <sys/wait.h> + +/* Make sure arena mappings cannot be split or inherited. */ + +#define NR_PAGES 3 + +static int arena_mmap (int *fd,void **area,size_t *sz) +{ + LIBBPF_OPTS(bpf_map_create_opts, opts, .map_flags = BPF_F_MMAPABLE); + + *sz = (size_t )NR_PAGES *sysconf (_SC_PAGESIZE); + + *fd =bpf_map_create (BPF_MAP_TYPE_ARENA ,"arena_vma" ,0 ,0 ,NR_PAGES , &opts); +if (!ASSERT_OK_FD (*fd,"arena map create" )) +return -1 ; + + *area =mmap (NULL , *sz,PROT_READ |PROT_WRITE ,MAP_SHARED , *fd,0 ); +if (!ASSERT_NEQ (*area,MAP_FAILED ,"mmap arena" )) { +close (*fd); +return -1 ; + } +return 0 ; +} + +/* A partial munmap that would split the arena mapping must be rejected. */ +static void split_test (void ) +{ +long ps =sysconf (_SC_PAGESIZE); +void *area; +size_t sz; +int fd, ret, err; + +if (arena_mmap (&fd, &area, &sz)) +return ; + + ret =munmap ((char *)area + ps, ps); + err = errno; +if (ASSERT_ERR (ret,"split munmap" )) +ASSERT_EQ (err,EINVAL ,"split munmap errno" ); + +munmap (area, sz); +close (fd); +} + +/* A forked child must not inherit the arena mapping. */ +static void fork_test (void ) +{ +long ps =sysconf (_SC_PAGESIZE); +void *area; +size_t sz; +int fd, ret, status; +pid_t pid; + +if (arena_mmap (&fd, &area, &sz)) +return ; + + pid =fork (); +if (pid ==0 ) { +unsigned char vec; +int rc; + +/* +* If the mapping was not inherited the range is unmapped in +* the child, so mincore() fails with ENOMEM. A success means +* the child wrongly inherited the mapping. +*/ + rc =mincore (area, ps, &vec); +if (rc ==0 ) +_exit (1 ); +_exit (errno ==ENOMEM ?0 :2 ); + } +if (ASSERT_GE (pid,0 ,"fork" )) { +while ((ret =waitpid (pid, &status,0 )) <0 && errno ==EINTR ) + ; +if (ASSERT_EQ (ret, pid,"waitpid" )) +ASSERT_TRUE (WIFEXITED (status) &&WEXITSTATUS (status) ==0 , +"child must not inherit arena mapping" ); + } + +munmap (area, sz); +close (fd); +} + +void test_arena_vma (void ) +{ +if (test__start_subtest ("split" )) +split_test (); +if (test__start_subtest ("fork" )) +fork_test (); +}base-commit:
174914ea551314c52a61713b9c4bde9e42d48073
Re: [PATCH v3 bpf-next] selftests/bpf: add tests for arena vma split and fork
In reply to Ruslan ValiyevQuoted 6 lines · Ruslan Valiyev
Add a test that an arena map's mmap()ed region cannot be split by a partial
munmap()(the "split" subtest, which expects-EINVAL) and is not inherited acrossfork()(the "fork" subtest, wheremincore()reports the child's range as unmapped).
Quoted 6 lines
Signed-off-by Ruslan Valiyev <[email protected]>---
v3:
- Consolidate into one file with a shared setup helper and two subtests.
- Put the opening /* of the multi-line comment on its own line.
Sorry, not every branch in the kernel needs a selftest. I don't think we need to spend CI cycles for every future patch to test for this unlikely regression.
pw-bot: cr