Message ID | 20211104070016.2463668-3-songliubraving@fb.com (mailing list archive) |
---|---|
State | Superseded |
Delegated to: | BPF |
Headers | show |
Series | introduce bpf_find_vma | expand |
> On Nov 4, 2021, at 12:00 AM, Song Liu <songliubraving@fb.com> wrote: > > Add tests for bpf_find_vma in perf_event program and kprobe program. The > perf_event program is triggered from NMI context, so the second call of > bpf_find_vma() will return -EBUSY (irq_work busy). The kprobe program, > on the other hand, does not have this constraint. > > Also add test for illegal writes to task or vma from the callback > function. The verifier should reject both cases. > > Signed-off-by: Song Liu <songliubraving@fb.com> > --- > .../selftests/bpf/prog_tests/find_vma.c | 115 ++++++++++++++++++ > tools/testing/selftests/bpf/progs/find_vma.c | 70 +++++++++++ > .../selftests/bpf/progs/find_vma_fail1.c | 30 +++++ > .../selftests/bpf/progs/find_vma_fail2.c | 30 +++++ > 4 files changed, 245 insertions(+) > create mode 100644 tools/testing/selftests/bpf/prog_tests/find_vma.c > create mode 100644 tools/testing/selftests/bpf/progs/find_vma.c > create mode 100644 tools/testing/selftests/bpf/progs/find_vma_fail1.c > create mode 100644 tools/testing/selftests/bpf/progs/find_vma_fail2.c > > diff --git a/tools/testing/selftests/bpf/prog_tests/find_vma.c b/tools/testing/selftests/bpf/prog_tests/find_vma.c > new file mode 100644 > index 0000000000000..3955a92d4c152 > --- /dev/null > +++ b/tools/testing/selftests/bpf/prog_tests/find_vma.c > @@ -0,0 +1,115 @@ > +// SPDX-License-Identifier: GPL-2.0 > +/* Copyright (c) 2021 Facebook */ > +#include <test_progs.h> > +#include <sys/types.h> > +#include <unistd.h> > +#include "find_vma.skel.h" > +#include "find_vma_fail1.skel.h" > +#include "find_vma_fail2.skel.h" > + > +static void test_and_reset_skel(struct find_vma *skel, int expected_find_zero_ret) > +{ > + ASSERT_EQ(skel->bss->found_vm_exec, 1, "found_vm_exec"); > + ASSERT_EQ(skel->data->find_addr_ret, 0, "find_addr_ret"); > + ASSERT_EQ(skel->data->find_zero_ret, expected_find_zero_ret, "find_zero_ret"); > + ASSERT_OK_PTR(strstr(skel->bss->d_iname, "test_progs"), "find_test_progs"); > + > + skel->bss->found_vm_exec = 0; > + skel->data->find_addr_ret = -1; > + skel->data->find_zero_ret = -1; > + skel->bss->d_iname[0] = 0; > +} > + > +static int open_pe(void) > +{ > + struct perf_event_attr attr = {0}; > + int pfd; > + > + /* create perf event */ > + attr.size = sizeof(attr); > + attr.type = PERF_TYPE_HARDWARE; > + attr.config = PERF_COUNT_HW_CPU_CYCLES; > + attr.freq = 1; > + attr.sample_freq = 4000; > + pfd = syscall(__NR_perf_event_open, &attr, 0, -1, -1, PERF_FLAG_FD_CLOEXEC); > + > + return pfd >= 0 ? pfd : -errno; > +} > + > +static void test_find_vma_pe(struct find_vma *skel) > +{ > + struct bpf_link *link = NULL; > + volatile int j = 0; > + int pfd = -1, i; > + > + pfd = open_pe(); > + if (pfd < 0) { > + if (pfd == -ENOENT || pfd == -EOPNOTSUPP) { > + printf("%s:SKIP:no PERF_COUNT_HW_CPU_CYCLES\n", __func__); > + test__skip(); Ah, I missed a goto cleanup here, so it breaks vmtest. I can fix this in v3, or we can fix it when applying the patch. > + } > + if (!ASSERT_GE(pfd, 0, "perf_event_open")) > + goto cleanup; > + } > + > + link = bpf_program__attach_perf_event(skel->progs.handle_pe, pfd); > + if (!ASSERT_OK_PTR(link, "attach_perf_event")) > + goto cleanup; > + > + for (i = 0; i < 1000000; ++i) > + ++j; > + > + test_and_reset_skel(skel, -EBUSY /* in nmi, irq_work is busy */); > +cleanup: > + bpf_link__destroy(link); > + close(pfd); > + /* caller will clean up skel */ > +} > + > +static void test_find_vma_kprobe(struct find_vma *skel) > +{ > + int err; > + > + err = find_vma__attach(skel); > + if (!ASSERT_OK(err, "get_branch_snapshot__attach")) > + return; /* caller will cleanup skel */ > + > + getpgid(skel->bss->target_pid); > + test_and_reset_skel(skel, -ENOENT /* could not find vma for ptr 0 */); > +} > + > +static void test_illegal_write_vma(void) > +{ > + struct find_vma_fail1 *skel; > + > + skel = find_vma_fail1__open_and_load(); > + ASSERT_ERR_PTR(skel, "find_vma_fail1__open_and_load"); > +} > + > +static void test_illegal_write_task(void) > +{ > + struct find_vma_fail2 *skel; > + > + skel = find_vma_fail2__open_and_load(); > + ASSERT_ERR_PTR(skel, "find_vma_fail2__open_and_load"); > +} > + > +void serial_test_find_vma(void) > +{ > + struct find_vma *skel; > + > + skel = find_vma__open_and_load(); > + if (!ASSERT_OK_PTR(skel, "find_vma__open_and_load")) > + return; > + > + skel->bss->target_pid = getpid(); > + skel->bss->addr = (__u64)test_find_vma_pe; > + > + test_find_vma_pe(skel); > + usleep(100000); /* allow the irq_work to finish */ > + test_find_vma_kprobe(skel); > + > + find_vma__destroy(skel); > + test_illegal_write_vma(); > + test_illegal_write_task(); > +} > diff --git a/tools/testing/selftests/bpf/progs/find_vma.c b/tools/testing/selftests/bpf/progs/find_vma.c > new file mode 100644 > index 0000000000000..2776718a54e29 > --- /dev/null > +++ b/tools/testing/selftests/bpf/progs/find_vma.c > @@ -0,0 +1,70 @@ > +// SPDX-License-Identifier: GPL-2.0 > +/* Copyright (c) 2021 Facebook */ > +#include "vmlinux.h" > +#include <bpf/bpf_helpers.h> > +#include <bpf/bpf_tracing.h> > + > +char _license[] SEC("license") = "GPL"; > + > +struct callback_ctx { > + int dummy; > +}; > + > +#define VM_EXEC 0x00000004 > +#define DNAME_INLINE_LEN 32 > + > +pid_t target_pid = 0; > +char d_iname[DNAME_INLINE_LEN] = {0}; > +__u32 found_vm_exec = 0; > +__u64 addr = 0; > +int find_zero_ret = -1; > +int find_addr_ret = -1; > + > +static __u64 > +check_vma(struct task_struct *task, struct vm_area_struct *vma, > + struct callback_ctx *data) > +{ > + if (vma->vm_file) > + bpf_probe_read_kernel_str(d_iname, DNAME_INLINE_LEN - 1, > + vma->vm_file->f_path.dentry->d_iname); > + > + /* check for VM_EXEC */ > + if (vma->vm_flags & VM_EXEC) > + found_vm_exec = 1; > + > + return 0; > +} > + > +SEC("kprobe/__x64_sys_getpgid") > +int handle_getpid(void) > +{ > + struct task_struct *task = bpf_get_current_task_btf(); > + struct callback_ctx data = {0}; > + > + if (task->pid != target_pid) > + return 0; > + > + find_addr_ret = bpf_find_vma(task, addr, check_vma, &data, 0); > + > + /* this should return -ENOENT */ > + find_zero_ret = bpf_find_vma(task, 0, check_vma, &data, 0); > + return 0; > +} > + > +SEC("perf_event") > +int handle_pe(void) > +{ > + struct task_struct *task = bpf_get_current_task_btf(); > + struct callback_ctx data = {0}; > + > + if (task->pid != target_pid) > + return 0; > + > + find_addr_ret = bpf_find_vma(task, addr, check_vma, &data, 0); > + > + /* In NMI, this should return -EBUSY, as the previous call is using > + * the irq_work. > + */ > + find_zero_ret = bpf_find_vma(task, 0, check_vma, &data, 0); > + return 0; > +} > diff --git a/tools/testing/selftests/bpf/progs/find_vma_fail1.c b/tools/testing/selftests/bpf/progs/find_vma_fail1.c > new file mode 100644 > index 0000000000000..d17bdcdf76f07 > --- /dev/null > +++ b/tools/testing/selftests/bpf/progs/find_vma_fail1.c > @@ -0,0 +1,30 @@ > +// SPDX-License-Identifier: GPL-2.0 > +/* Copyright (c) 2021 Facebook */ > +#include "vmlinux.h" > +#include <bpf/bpf_helpers.h> > + > +char _license[] SEC("license") = "GPL"; > + > +struct callback_ctx { > + int dummy; > +}; > + > +static __u64 > +write_vma(struct task_struct *task, struct vm_area_struct *vma, > + struct callback_ctx *data) > +{ > + /* writing to vma, which is illegal */ > + vma->vm_flags |= 0x55; > + > + return 0; > +} > + > +SEC("kprobe/__x64_sys_getpgid") > +int handle_getpid(void) > +{ > + struct task_struct *task = bpf_get_current_task_btf(); > + struct callback_ctx data = {0}; > + > + bpf_find_vma(task, 0, write_vma, &data, 0); > + return 0; > +} > diff --git a/tools/testing/selftests/bpf/progs/find_vma_fail2.c b/tools/testing/selftests/bpf/progs/find_vma_fail2.c > new file mode 100644 > index 0000000000000..079c4594c095d > --- /dev/null > +++ b/tools/testing/selftests/bpf/progs/find_vma_fail2.c > @@ -0,0 +1,30 @@ > +// SPDX-License-Identifier: GPL-2.0 > +/* Copyright (c) 2021 Facebook */ > +#include "vmlinux.h" > +#include <bpf/bpf_helpers.h> > + > +char _license[] SEC("license") = "GPL"; > + > +struct callback_ctx { > + int dummy; > +}; > + > +static __u64 > +write_task(struct task_struct *task, struct vm_area_struct *vma, > + struct callback_ctx *data) > +{ > + /* writing to task, which is illegal */ > + task->mm = NULL; > + > + return 0; > +} > + > +SEC("kprobe/__x64_sys_getpgid") > +int handle_getpid(void) > +{ > + struct task_struct *task = bpf_get_current_task_btf(); > + struct callback_ctx data = {0}; > + > + bpf_find_vma(task, 0, write_task, &data, 0); > + return 0; > +} > -- > 2.30.2 >
On 11/4/21 12:00 AM, Song Liu wrote: > Add tests for bpf_find_vma in perf_event program and kprobe program. The > perf_event program is triggered from NMI context, so the second call of > bpf_find_vma() will return -EBUSY (irq_work busy). The kprobe program, > on the other hand, does not have this constraint. > > Also add test for illegal writes to task or vma from the callback > function. The verifier should reject both cases. > > Signed-off-by: Song Liu <songliubraving@fb.com> > --- > .../selftests/bpf/prog_tests/find_vma.c | 115 ++++++++++++++++++ > tools/testing/selftests/bpf/progs/find_vma.c | 70 +++++++++++ > .../selftests/bpf/progs/find_vma_fail1.c | 30 +++++ > .../selftests/bpf/progs/find_vma_fail2.c | 30 +++++ > 4 files changed, 245 insertions(+) > create mode 100644 tools/testing/selftests/bpf/prog_tests/find_vma.c > create mode 100644 tools/testing/selftests/bpf/progs/find_vma.c > create mode 100644 tools/testing/selftests/bpf/progs/find_vma_fail1.c > create mode 100644 tools/testing/selftests/bpf/progs/find_vma_fail2.c > > diff --git a/tools/testing/selftests/bpf/prog_tests/find_vma.c b/tools/testing/selftests/bpf/prog_tests/find_vma.c > new file mode 100644 > index 0000000000000..3955a92d4c152 > --- /dev/null > +++ b/tools/testing/selftests/bpf/prog_tests/find_vma.c > @@ -0,0 +1,115 @@ > +// SPDX-License-Identifier: GPL-2.0 > +/* Copyright (c) 2021 Facebook */ > +#include <test_progs.h> > +#include <sys/types.h> > +#include <unistd.h> > +#include "find_vma.skel.h" > +#include "find_vma_fail1.skel.h" > +#include "find_vma_fail2.skel.h" > + > +static void test_and_reset_skel(struct find_vma *skel, int expected_find_zero_ret) > +{ > + ASSERT_EQ(skel->bss->found_vm_exec, 1, "found_vm_exec"); > + ASSERT_EQ(skel->data->find_addr_ret, 0, "find_addr_ret"); > + ASSERT_EQ(skel->data->find_zero_ret, expected_find_zero_ret, "find_zero_ret"); > + ASSERT_OK_PTR(strstr(skel->bss->d_iname, "test_progs"), "find_test_progs"); > + > + skel->bss->found_vm_exec = 0; > + skel->data->find_addr_ret = -1; > + skel->data->find_zero_ret = -1; > + skel->bss->d_iname[0] = 0; > +} > + > +static int open_pe(void) > +{ > + struct perf_event_attr attr = {0}; > + int pfd; > + > + /* create perf event */ > + attr.size = sizeof(attr); > + attr.type = PERF_TYPE_HARDWARE; > + attr.config = PERF_COUNT_HW_CPU_CYCLES; > + attr.freq = 1; > + attr.sample_freq = 4000; > + pfd = syscall(__NR_perf_event_open, &attr, 0, -1, -1, PERF_FLAG_FD_CLOEXEC); > + > + return pfd >= 0 ? pfd : -errno; > +} > + > +static void test_find_vma_pe(struct find_vma *skel) > +{ > + struct bpf_link *link = NULL; > + volatile int j = 0; > + int pfd = -1, i; > + > + pfd = open_pe(); > + if (pfd < 0) { > + if (pfd == -ENOENT || pfd == -EOPNOTSUPP) { > + printf("%s:SKIP:no PERF_COUNT_HW_CPU_CYCLES\n", __func__); > + test__skip(); > + } > + if (!ASSERT_GE(pfd, 0, "perf_event_open")) > + goto cleanup; > + } > + > + link = bpf_program__attach_perf_event(skel->progs.handle_pe, pfd); > + if (!ASSERT_OK_PTR(link, "attach_perf_event")) > + goto cleanup; > + > + for (i = 0; i < 1000000; ++i) > + ++j; Does this really work? Compiler could do j += 1000000; > + > + test_and_reset_skel(skel, -EBUSY /* in nmi, irq_work is busy */); > +cleanup: > + bpf_link__destroy(link); > + close(pfd); > + /* caller will clean up skel */ Above comment is not needed. It should be clear from the code. > +} > + > +static void test_find_vma_kprobe(struct find_vma *skel) > +{ > + int err; > + > + err = find_vma__attach(skel); > + if (!ASSERT_OK(err, "get_branch_snapshot__attach")) > + return; /* caller will cleanup skel */ > + > + getpgid(skel->bss->target_pid); > + test_and_reset_skel(skel, -ENOENT /* could not find vma for ptr 0 */); > +} > + > +static void test_illegal_write_vma(void) > +{ > + struct find_vma_fail1 *skel; > + > + skel = find_vma_fail1__open_and_load(); > + ASSERT_ERR_PTR(skel, "find_vma_fail1__open_and_load"); > +} > + > +static void test_illegal_write_task(void) > +{ > + struct find_vma_fail2 *skel; > + > + skel = find_vma_fail2__open_and_load(); > + ASSERT_ERR_PTR(skel, "find_vma_fail2__open_and_load"); > +} > + > +void serial_test_find_vma(void) > +{ > + struct find_vma *skel; > + > + skel = find_vma__open_and_load(); > + if (!ASSERT_OK_PTR(skel, "find_vma__open_and_load")) > + return; > + > + skel->bss->target_pid = getpid(); > + skel->bss->addr = (__u64)test_find_vma_pe; > + > + test_find_vma_pe(skel); > + usleep(100000); /* allow the irq_work to finish */ > + test_find_vma_kprobe(skel); > + > + find_vma__destroy(skel); > + test_illegal_write_vma(); > + test_illegal_write_task(); > +} > diff --git a/tools/testing/selftests/bpf/progs/find_vma.c b/tools/testing/selftests/bpf/progs/find_vma.c > new file mode 100644 > index 0000000000000..2776718a54e29 > --- /dev/null > +++ b/tools/testing/selftests/bpf/progs/find_vma.c > @@ -0,0 +1,70 @@ > +// SPDX-License-Identifier: GPL-2.0 > +/* Copyright (c) 2021 Facebook */ > +#include "vmlinux.h" > +#include <bpf/bpf_helpers.h> > +#include <bpf/bpf_tracing.h> > + > +char _license[] SEC("license") = "GPL"; > + > +struct callback_ctx { > + int dummy; > +}; > + > +#define VM_EXEC 0x00000004 > +#define DNAME_INLINE_LEN 32 > + > +pid_t target_pid = 0; > +char d_iname[DNAME_INLINE_LEN] = {0}; > +__u32 found_vm_exec = 0; > +__u64 addr = 0; > +int find_zero_ret = -1; > +int find_addr_ret = -1; > + > +static __u64 Let us 'long' instead of '__u64' to match uapi bpf.h. > +check_vma(struct task_struct *task, struct vm_area_struct *vma, > + struct callback_ctx *data) > +{ > + if (vma->vm_file) > + bpf_probe_read_kernel_str(d_iname, DNAME_INLINE_LEN - 1, > + vma->vm_file->f_path.dentry->d_iname); > + > + /* check for VM_EXEC */ > + if (vma->vm_flags & VM_EXEC) > + found_vm_exec = 1; > + > + return 0; > +} > + > +SEC("kprobe/__x64_sys_getpgid") The test will fail for non x86_64 architecture. I had some tweaks in test_probe_user.c. Please take a look. We can refactor to make tweaks in test_probe_user.c reusable by other files. > +int handle_getpid(void) > +{ > + struct task_struct *task = bpf_get_current_task_btf(); > + struct callback_ctx data = {0}; > + > + if (task->pid != target_pid) > + return 0; > + > + find_addr_ret = bpf_find_vma(task, addr, check_vma, &data, 0); > + > + /* this should return -ENOENT */ > + find_zero_ret = bpf_find_vma(task, 0, check_vma, &data, 0); > + return 0; > +} > + > +SEC("perf_event") > +int handle_pe(void) > +{ > + struct task_struct *task = bpf_get_current_task_btf(); > + struct callback_ctx data = {0}; > + > + if (task->pid != target_pid) > + return 0; This is tricky. How do we guarantee task->pid == target_pid hit? This probably mostly okay in serial running mode. But it may become more challenging if test_progs is running in parallel mode? > + > + find_addr_ret = bpf_find_vma(task, addr, check_vma, &data, 0); > + > + /* In NMI, this should return -EBUSY, as the previous call is using > + * the irq_work. > + */ > + find_zero_ret = bpf_find_vma(task, 0, check_vma, &data, 0); > + return 0; > +} > diff --git a/tools/testing/selftests/bpf/progs/find_vma_fail1.c b/tools/testing/selftests/bpf/progs/find_vma_fail1.c > new file mode 100644 > index 0000000000000..d17bdcdf76f07 > --- /dev/null > +++ b/tools/testing/selftests/bpf/progs/find_vma_fail1.c > @@ -0,0 +1,30 @@ > +// SPDX-License-Identifier: GPL-2.0 > +/* Copyright (c) 2021 Facebook */ > +#include "vmlinux.h" > +#include <bpf/bpf_helpers.h> > + > +char _license[] SEC("license") = "GPL"; > + > +struct callback_ctx { > + int dummy; > +}; > + > +static __u64 __u64 => long > +write_vma(struct task_struct *task, struct vm_area_struct *vma, > + struct callback_ctx *data) > +{ > + /* writing to vma, which is illegal */ > + vma->vm_flags |= 0x55; > + > + return 0; > +} > + > +SEC("kprobe/__x64_sys_getpgid") > +int handle_getpid(void) > +{ > + struct task_struct *task = bpf_get_current_task_btf(); > + struct callback_ctx data = {0}; > + > + bpf_find_vma(task, 0, write_vma, &data, 0); > + return 0; > +} > diff --git a/tools/testing/selftests/bpf/progs/find_vma_fail2.c b/tools/testing/selftests/bpf/progs/find_vma_fail2.c > new file mode 100644 > index 0000000000000..079c4594c095d > --- /dev/null > +++ b/tools/testing/selftests/bpf/progs/find_vma_fail2.c > @@ -0,0 +1,30 @@ > +// SPDX-License-Identifier: GPL-2.0 > +/* Copyright (c) 2021 Facebook */ > +#include "vmlinux.h" > +#include <bpf/bpf_helpers.h> > + > +char _license[] SEC("license") = "GPL"; > + > +struct callback_ctx { > + int dummy; > +}; > + > +static __u64 __u64 => long > +write_task(struct task_struct *task, struct vm_area_struct *vma, > + struct callback_ctx *data) > +{ > + /* writing to task, which is illegal */ > + task->mm = NULL; > + > + return 0; > +} > + > +SEC("kprobe/__x64_sys_getpgid") > +int handle_getpid(void) > +{ > + struct task_struct *task = bpf_get_current_task_btf(); > + struct callback_ctx data = {0}; > + > + bpf_find_vma(task, 0, write_task, &data, 0); > + return 0; > +} >
> On Nov 4, 2021, at 10:07 AM, Yonghong Song <yhs@fb.com> wrote: > > > > On 11/4/21 12:00 AM, Song Liu wrote: >> Add tests for bpf_find_vma in perf_event program and kprobe program. The >> perf_event program is triggered from NMI context, so the second call of >> bpf_find_vma() will return -EBUSY (irq_work busy). The kprobe program, >> on the other hand, does not have this constraint. >> Also add test for illegal writes to task or vma from the callback >> function. The verifier should reject both cases. >> Signed-off-by: Song Liu <songliubraving@fb.com> [...] >> +static void test_find_vma_pe(struct find_vma *skel) >> +{ >> + struct bpf_link *link = NULL; >> + volatile int j = 0; >> + int pfd = -1, i; >> + >> + pfd = open_pe(); >> + if (pfd < 0) { >> + if (pfd == -ENOENT || pfd == -EOPNOTSUPP) { >> + printf("%s:SKIP:no PERF_COUNT_HW_CPU_CYCLES\n", __func__); >> + test__skip(); >> + } >> + if (!ASSERT_GE(pfd, 0, "perf_event_open")) >> + goto cleanup; >> + } >> + >> + link = bpf_program__attach_perf_event(skel->progs.handle_pe, pfd); >> + if (!ASSERT_OK_PTR(link, "attach_perf_event")) >> + goto cleanup; >> + >> + for (i = 0; i < 1000000; ++i) >> + ++j; > > Does this really work? Compiler could do > j += 1000000; I think compiler won't do it with volatile j? > >> + >> + test_and_reset_skel(skel, -EBUSY /* in nmi, irq_work is busy */); >> +cleanup: >> + bpf_link__destroy(link); >> + close(pfd); >> + /* caller will clean up skel */ > > Above comment is not needed. It should be clear from the code. > >> +} >> + >> +static void test_find_vma_kprobe(struct find_vma *skel) >> +{ >> + int err; >> + >> + err = find_vma__attach(skel); >> + if (!ASSERT_OK(err, "get_branch_snapshot__attach")) >> + return; /* caller will cleanup skel */ >> + >> + getpgid(skel->bss->target_pid); >> + test_and_reset_skel(skel, -ENOENT /* could not find vma for ptr 0 */); >> +} >> + >> +static void test_illegal_write_vma(void) >> +{ >> + struct find_vma_fail1 *skel; >> + >> + skel = find_vma_fail1__open_and_load(); >> + ASSERT_ERR_PTR(skel, "find_vma_fail1__open_and_load"); >> +} >> + >> +static void test_illegal_write_task(void) >> +{ >> + struct find_vma_fail2 *skel; >> + >> + skel = find_vma_fail2__open_and_load(); >> + ASSERT_ERR_PTR(skel, "find_vma_fail2__open_and_load"); >> +} >> + >> +void serial_test_find_vma(void) >> +{ >> + struct find_vma *skel; >> + >> + skel = find_vma__open_and_load(); >> + if (!ASSERT_OK_PTR(skel, "find_vma__open_and_load")) >> + return; >> + >> + skel->bss->target_pid = getpid(); >> + skel->bss->addr = (__u64)test_find_vma_pe; >> + >> + test_find_vma_pe(skel); >> + usleep(100000); /* allow the irq_work to finish */ >> + test_find_vma_kprobe(skel); >> + >> + find_vma__destroy(skel); >> + test_illegal_write_vma(); >> + test_illegal_write_task(); >> +} >> diff --git a/tools/testing/selftests/bpf/progs/find_vma.c b/tools/testing/selftests/bpf/progs/find_vma.c >> new file mode 100644 >> index 0000000000000..2776718a54e29 >> --- /dev/null >> +++ b/tools/testing/selftests/bpf/progs/find_vma.c >> @@ -0,0 +1,70 @@ >> +// SPDX-License-Identifier: GPL-2.0 >> +/* Copyright (c) 2021 Facebook */ >> +#include "vmlinux.h" >> +#include <bpf/bpf_helpers.h> >> +#include <bpf/bpf_tracing.h> >> + >> +char _license[] SEC("license") = "GPL"; >> + >> +struct callback_ctx { >> + int dummy; >> +}; >> + >> +#define VM_EXEC 0x00000004 >> +#define DNAME_INLINE_LEN 32 >> + >> +pid_t target_pid = 0; >> +char d_iname[DNAME_INLINE_LEN] = {0}; >> +__u32 found_vm_exec = 0; >> +__u64 addr = 0; >> +int find_zero_ret = -1; >> +int find_addr_ret = -1; >> + >> +static __u64 > > Let us 'long' instead of '__u64' to match uapi bpf.h. > >> +check_vma(struct task_struct *task, struct vm_area_struct *vma, >> + struct callback_ctx *data) >> +{ >> + if (vma->vm_file) >> + bpf_probe_read_kernel_str(d_iname, DNAME_INLINE_LEN - 1, >> + vma->vm_file->f_path.dentry->d_iname); >> + >> + /* check for VM_EXEC */ >> + if (vma->vm_flags & VM_EXEC) >> + found_vm_exec = 1; >> + >> + return 0; >> +} >> + >> +SEC("kprobe/__x64_sys_getpgid") > > The test will fail for non x86_64 architecture. > I had some tweaks in test_probe_user.c. Please take a look. > We can refactor to make tweaks in test_probe_user.c reusable > by other files. Good point. I will look into this. > >> +int handle_getpid(void) >> +{ >> + struct task_struct *task = bpf_get_current_task_btf(); >> + struct callback_ctx data = {0}; >> + >> + if (task->pid != target_pid) >> + return 0; >> + >> + find_addr_ret = bpf_find_vma(task, addr, check_vma, &data, 0); >> + >> + /* this should return -ENOENT */ >> + find_zero_ret = bpf_find_vma(task, 0, check_vma, &data, 0); >> + return 0; >> +} >> + >> +SEC("perf_event") >> +int handle_pe(void) >> +{ >> + struct task_struct *task = bpf_get_current_task_btf(); >> + struct callback_ctx data = {0}; >> + >> + if (task->pid != target_pid) >> + return 0; > > This is tricky. How do we guarantee task->pid == target_pid hit? > This probably mostly okay in serial running mode. But it may > become more challenging if test_progs is running in parallel mode? This is on a per task perf_event, so it shouldn't hit other tasks. > >> + >> + find_addr_ret = bpf_find_vma(task, addr, check_vma, &data, 0); >> + >> + /* In NMI, this should return -EBUSY, as the previous call is using >> + * the irq_work. >> + */ >> + find_zero_ret = bpf_find_vma(task, 0, check_vma, &data, 0); >> + return 0; >> +} >> diff --git a/tools/testing/selftests/bpf/progs/find_vma_fail1.c b/tools/testing/selftests/bpf/progs/find_vma_fail1.c >> new file mode 100644 >> index 0000000000000..d17bdcdf76f07 >> --- /dev/null >> +++ b/tools/testing/selftests/bpf/progs/find_vma_fail1.c >> @@ -0,0 +1,30 @@ >> +// SPDX-License-Identifier: GPL-2.0 >> +/* Copyright (c) 2021 Facebook */ >> +#include "vmlinux.h" >> +#include <bpf/bpf_helpers.h> >> + >> +char _license[] SEC("license") = "GPL"; >> + >> +struct callback_ctx { >> + int dummy; >> +}; >> + >> +static __u64 > > __u64 => long > >> +write_vma(struct task_struct *task, struct vm_area_struct *vma, >> + struct callback_ctx *data) >> +{ >> + /* writing to vma, which is illegal */ >> + vma->vm_flags |= 0x55; >> + >> + return 0; >> +} >> + >> +SEC("kprobe/__x64_sys_getpgid") >> +int handle_getpid(void) >> +{ >> + struct task_struct *task = bpf_get_current_task_btf(); >> + struct callback_ctx data = {0}; >> + >> + bpf_find_vma(task, 0, write_vma, &data, 0); >> + return 0; >> +} >> diff --git a/tools/testing/selftests/bpf/progs/find_vma_fail2.c b/tools/testing/selftests/bpf/progs/find_vma_fail2.c >> new file mode 100644 >> index 0000000000000..079c4594c095d >> --- /dev/null >> +++ b/tools/testing/selftests/bpf/progs/find_vma_fail2.c >> @@ -0,0 +1,30 @@ >> +// SPDX-License-Identifier: GPL-2.0 >> +/* Copyright (c) 2021 Facebook */ >> +#include "vmlinux.h" >> +#include <bpf/bpf_helpers.h> >> + >> +char _license[] SEC("license") = "GPL"; >> + >> +struct callback_ctx { >> + int dummy; >> +}; >> + >> +static __u64 > > __u64 => long > >> +write_task(struct task_struct *task, struct vm_area_struct *vma, >> + struct callback_ctx *data) >> +{ >> + /* writing to task, which is illegal */ >> + task->mm = NULL; >> + >> + return 0; >> +} >> + >> +SEC("kprobe/__x64_sys_getpgid") >> +int handle_getpid(void) >> +{ >> + struct task_struct *task = bpf_get_current_task_btf(); >> + struct callback_ctx data = {0}; >> + >> + bpf_find_vma(task, 0, write_task, &data, 0); >> + return 0; >> +}
On 11/4/21 10:17 AM, Song Liu wrote: > > >> On Nov 4, 2021, at 10:07 AM, Yonghong Song <yhs@fb.com> wrote: >> >> >> >> On 11/4/21 12:00 AM, Song Liu wrote: >>> Add tests for bpf_find_vma in perf_event program and kprobe program. The >>> perf_event program is triggered from NMI context, so the second call of >>> bpf_find_vma() will return -EBUSY (irq_work busy). The kprobe program, >>> on the other hand, does not have this constraint. >>> Also add test for illegal writes to task or vma from the callback >>> function. The verifier should reject both cases. >>> Signed-off-by: Song Liu <songliubraving@fb.com> > > [...] > >>> +static void test_find_vma_pe(struct find_vma *skel) >>> +{ >>> + struct bpf_link *link = NULL; >>> + volatile int j = 0; >>> + int pfd = -1, i; >>> + >>> + pfd = open_pe(); >>> + if (pfd < 0) { >>> + if (pfd == -ENOENT || pfd == -EOPNOTSUPP) { >>> + printf("%s:SKIP:no PERF_COUNT_HW_CPU_CYCLES\n", __func__); >>> + test__skip(); >>> + } >>> + if (!ASSERT_GE(pfd, 0, "perf_event_open")) >>> + goto cleanup; >>> + } >>> + >>> + link = bpf_program__attach_perf_event(skel->progs.handle_pe, pfd); >>> + if (!ASSERT_OK_PTR(link, "attach_perf_event")) >>> + goto cleanup; >>> + >>> + for (i = 0; i < 1000000; ++i) >>> + ++j; >> >> Does this really work? Compiler could do >> j += 1000000; > > I think compiler won't do it with volatile j? Ya. volatile j should be fine. > >> >>> + >>> + test_and_reset_skel(skel, -EBUSY /* in nmi, irq_work is busy */); >>> +cleanup: >>> + bpf_link__destroy(link); >>> + close(pfd); >>> + /* caller will clean up skel */ >> >> Above comment is not needed. It should be clear from the code. >> [...] >> >>> +int handle_getpid(void) >>> +{ >>> + struct task_struct *task = bpf_get_current_task_btf(); >>> + struct callback_ctx data = {0}; >>> + >>> + if (task->pid != target_pid) >>> + return 0; >>> + >>> + find_addr_ret = bpf_find_vma(task, addr, check_vma, &data, 0); >>> + >>> + /* this should return -ENOENT */ >>> + find_zero_ret = bpf_find_vma(task, 0, check_vma, &data, 0); >>> + return 0; >>> +} >>> + >>> +SEC("perf_event") >>> +int handle_pe(void) >>> +{ >>> + struct task_struct *task = bpf_get_current_task_btf(); >>> + struct callback_ctx data = {0}; >>> + >>> + if (task->pid != target_pid) >>> + return 0; >> >> This is tricky. How do we guarantee task->pid == target_pid hit? >> This probably mostly okay in serial running mode. But it may >> become more challenging if test_progs is running in parallel mode? > > This is on a per task perf_event, so it shouldn't hit other tasks. I see. we have the following parameters for perf_event open. pid == 0 and cpu == -1 This measures the calling process/thread on any CPU. So yes, we are fine then. > >> >>> + >>> + find_addr_ret = bpf_find_vma(task, addr, check_vma, &data, 0); >>> + >>> + /* In NMI, this should return -EBUSY, as the previous call is using >>> + * the irq_work. >>> + */ >>> + find_zero_ret = bpf_find_vma(task, 0, check_vma, &data, 0); >>> + return 0; >>> +} [...]
> On Nov 4, 2021, at 10:17 AM, Song Liu <songliubraving@fb.com> wrote: > > > >> On Nov 4, 2021, at 10:07 AM, Yonghong Song <yhs@fb.com> wrote: >> >> >> >> On 11/4/21 12:00 AM, Song Liu wrote: >>> Add tests for bpf_find_vma in perf_event program and kprobe program. The >>> perf_event program is triggered from NMI context, so the second call of >>> bpf_find_vma() will return -EBUSY (irq_work busy). The kprobe program, >>> on the other hand, does not have this constraint. >>> Also add test for illegal writes to task or vma from the callback >>> function. The verifier should reject both cases. >>> Signed-off-by: Song Liu <songliubraving@fb.com> > > [...] > >>> +static void test_find_vma_pe(struct find_vma *skel) >>> +{ >>> + struct bpf_link *link = NULL; >>> + volatile int j = 0; >>> + int pfd = -1, i; >>> + >>> + pfd = open_pe(); >>> + if (pfd < 0) { >>> + if (pfd == -ENOENT || pfd == -EOPNOTSUPP) { >>> + printf("%s:SKIP:no PERF_COUNT_HW_CPU_CYCLES\n", __func__); >>> + test__skip(); >>> + } >>> + if (!ASSERT_GE(pfd, 0, "perf_event_open")) >>> + goto cleanup; >>> + } >>> + >>> + link = bpf_program__attach_perf_event(skel->progs.handle_pe, pfd); >>> + if (!ASSERT_OK_PTR(link, "attach_perf_event")) >>> + goto cleanup; >>> + >>> + for (i = 0; i < 1000000; ++i) >>> + ++j; >> >> Does this really work? Compiler could do >> j += 1000000; > > I think compiler won't do it with volatile j? > >> >>> + >>> + test_and_reset_skel(skel, -EBUSY /* in nmi, irq_work is busy */); >>> +cleanup: >>> + bpf_link__destroy(link); >>> + close(pfd); >>> + /* caller will clean up skel */ >> >> Above comment is not needed. It should be clear from the code. >> >>> +} >>> + >>> +static void test_find_vma_kprobe(struct find_vma *skel) >>> +{ >>> + int err; >>> + >>> + err = find_vma__attach(skel); >>> + if (!ASSERT_OK(err, "get_branch_snapshot__attach")) >>> + return; /* caller will cleanup skel */ >>> + >>> + getpgid(skel->bss->target_pid); >>> + test_and_reset_skel(skel, -ENOENT /* could not find vma for ptr 0 */); >>> +} >>> + >>> +static void test_illegal_write_vma(void) >>> +{ >>> + struct find_vma_fail1 *skel; >>> + >>> + skel = find_vma_fail1__open_and_load(); >>> + ASSERT_ERR_PTR(skel, "find_vma_fail1__open_and_load"); >>> +} >>> + >>> +static void test_illegal_write_task(void) >>> +{ >>> + struct find_vma_fail2 *skel; >>> + >>> + skel = find_vma_fail2__open_and_load(); >>> + ASSERT_ERR_PTR(skel, "find_vma_fail2__open_and_load"); >>> +} >>> + >>> +void serial_test_find_vma(void) >>> +{ >>> + struct find_vma *skel; >>> + >>> + skel = find_vma__open_and_load(); >>> + if (!ASSERT_OK_PTR(skel, "find_vma__open_and_load")) >>> + return; >>> + >>> + skel->bss->target_pid = getpid(); >>> + skel->bss->addr = (__u64)test_find_vma_pe; >>> + >>> + test_find_vma_pe(skel); >>> + usleep(100000); /* allow the irq_work to finish */ >>> + test_find_vma_kprobe(skel); >>> + >>> + find_vma__destroy(skel); >>> + test_illegal_write_vma(); >>> + test_illegal_write_task(); >>> +} >>> diff --git a/tools/testing/selftests/bpf/progs/find_vma.c b/tools/testing/selftests/bpf/progs/find_vma.c >>> new file mode 100644 >>> index 0000000000000..2776718a54e29 >>> --- /dev/null >>> +++ b/tools/testing/selftests/bpf/progs/find_vma.c >>> @@ -0,0 +1,70 @@ >>> +// SPDX-License-Identifier: GPL-2.0 >>> +/* Copyright (c) 2021 Facebook */ >>> +#include "vmlinux.h" >>> +#include <bpf/bpf_helpers.h> >>> +#include <bpf/bpf_tracing.h> >>> + >>> +char _license[] SEC("license") = "GPL"; >>> + >>> +struct callback_ctx { >>> + int dummy; >>> +}; >>> + >>> +#define VM_EXEC 0x00000004 >>> +#define DNAME_INLINE_LEN 32 >>> + >>> +pid_t target_pid = 0; >>> +char d_iname[DNAME_INLINE_LEN] = {0}; >>> +__u32 found_vm_exec = 0; >>> +__u64 addr = 0; >>> +int find_zero_ret = -1; >>> +int find_addr_ret = -1; >>> + >>> +static __u64 >> >> Let us 'long' instead of '__u64' to match uapi bpf.h. >> >>> +check_vma(struct task_struct *task, struct vm_area_struct *vma, >>> + struct callback_ctx *data) >>> +{ >>> + if (vma->vm_file) >>> + bpf_probe_read_kernel_str(d_iname, DNAME_INLINE_LEN - 1, >>> + vma->vm_file->f_path.dentry->d_iname); >>> + >>> + /* check for VM_EXEC */ >>> + if (vma->vm_flags & VM_EXEC) >>> + found_vm_exec = 1; >>> + >>> + return 0; >>> +} >>> + >>> +SEC("kprobe/__x64_sys_getpgid") >> >> The test will fail for non x86_64 architecture. >> I had some tweaks in test_probe_user.c. Please take a look. >> We can refactor to make tweaks in test_probe_user.c reusable >> by other files. > > Good point. I will look into this. Actually, we can just use SEC("raw_tp/sys_enter") here. Thanks, Song
diff --git a/tools/testing/selftests/bpf/prog_tests/find_vma.c b/tools/testing/selftests/bpf/prog_tests/find_vma.c new file mode 100644 index 0000000000000..3955a92d4c152 --- /dev/null +++ b/tools/testing/selftests/bpf/prog_tests/find_vma.c @@ -0,0 +1,115 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright (c) 2021 Facebook */ +#include <test_progs.h> +#include <sys/types.h> +#include <unistd.h> +#include "find_vma.skel.h" +#include "find_vma_fail1.skel.h" +#include "find_vma_fail2.skel.h" + +static void test_and_reset_skel(struct find_vma *skel, int expected_find_zero_ret) +{ + ASSERT_EQ(skel->bss->found_vm_exec, 1, "found_vm_exec"); + ASSERT_EQ(skel->data->find_addr_ret, 0, "find_addr_ret"); + ASSERT_EQ(skel->data->find_zero_ret, expected_find_zero_ret, "find_zero_ret"); + ASSERT_OK_PTR(strstr(skel->bss->d_iname, "test_progs"), "find_test_progs"); + + skel->bss->found_vm_exec = 0; + skel->data->find_addr_ret = -1; + skel->data->find_zero_ret = -1; + skel->bss->d_iname[0] = 0; +} + +static int open_pe(void) +{ + struct perf_event_attr attr = {0}; + int pfd; + + /* create perf event */ + attr.size = sizeof(attr); + attr.type = PERF_TYPE_HARDWARE; + attr.config = PERF_COUNT_HW_CPU_CYCLES; + attr.freq = 1; + attr.sample_freq = 4000; + pfd = syscall(__NR_perf_event_open, &attr, 0, -1, -1, PERF_FLAG_FD_CLOEXEC); + + return pfd >= 0 ? pfd : -errno; +} + +static void test_find_vma_pe(struct find_vma *skel) +{ + struct bpf_link *link = NULL; + volatile int j = 0; + int pfd = -1, i; + + pfd = open_pe(); + if (pfd < 0) { + if (pfd == -ENOENT || pfd == -EOPNOTSUPP) { + printf("%s:SKIP:no PERF_COUNT_HW_CPU_CYCLES\n", __func__); + test__skip(); + } + if (!ASSERT_GE(pfd, 0, "perf_event_open")) + goto cleanup; + } + + link = bpf_program__attach_perf_event(skel->progs.handle_pe, pfd); + if (!ASSERT_OK_PTR(link, "attach_perf_event")) + goto cleanup; + + for (i = 0; i < 1000000; ++i) + ++j; + + test_and_reset_skel(skel, -EBUSY /* in nmi, irq_work is busy */); +cleanup: + bpf_link__destroy(link); + close(pfd); + /* caller will clean up skel */ +} + +static void test_find_vma_kprobe(struct find_vma *skel) +{ + int err; + + err = find_vma__attach(skel); + if (!ASSERT_OK(err, "get_branch_snapshot__attach")) + return; /* caller will cleanup skel */ + + getpgid(skel->bss->target_pid); + test_and_reset_skel(skel, -ENOENT /* could not find vma for ptr 0 */); +} + +static void test_illegal_write_vma(void) +{ + struct find_vma_fail1 *skel; + + skel = find_vma_fail1__open_and_load(); + ASSERT_ERR_PTR(skel, "find_vma_fail1__open_and_load"); +} + +static void test_illegal_write_task(void) +{ + struct find_vma_fail2 *skel; + + skel = find_vma_fail2__open_and_load(); + ASSERT_ERR_PTR(skel, "find_vma_fail2__open_and_load"); +} + +void serial_test_find_vma(void) +{ + struct find_vma *skel; + + skel = find_vma__open_and_load(); + if (!ASSERT_OK_PTR(skel, "find_vma__open_and_load")) + return; + + skel->bss->target_pid = getpid(); + skel->bss->addr = (__u64)test_find_vma_pe; + + test_find_vma_pe(skel); + usleep(100000); /* allow the irq_work to finish */ + test_find_vma_kprobe(skel); + + find_vma__destroy(skel); + test_illegal_write_vma(); + test_illegal_write_task(); +} diff --git a/tools/testing/selftests/bpf/progs/find_vma.c b/tools/testing/selftests/bpf/progs/find_vma.c new file mode 100644 index 0000000000000..2776718a54e29 --- /dev/null +++ b/tools/testing/selftests/bpf/progs/find_vma.c @@ -0,0 +1,70 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright (c) 2021 Facebook */ +#include "vmlinux.h" +#include <bpf/bpf_helpers.h> +#include <bpf/bpf_tracing.h> + +char _license[] SEC("license") = "GPL"; + +struct callback_ctx { + int dummy; +}; + +#define VM_EXEC 0x00000004 +#define DNAME_INLINE_LEN 32 + +pid_t target_pid = 0; +char d_iname[DNAME_INLINE_LEN] = {0}; +__u32 found_vm_exec = 0; +__u64 addr = 0; +int find_zero_ret = -1; +int find_addr_ret = -1; + +static __u64 +check_vma(struct task_struct *task, struct vm_area_struct *vma, + struct callback_ctx *data) +{ + if (vma->vm_file) + bpf_probe_read_kernel_str(d_iname, DNAME_INLINE_LEN - 1, + vma->vm_file->f_path.dentry->d_iname); + + /* check for VM_EXEC */ + if (vma->vm_flags & VM_EXEC) + found_vm_exec = 1; + + return 0; +} + +SEC("kprobe/__x64_sys_getpgid") +int handle_getpid(void) +{ + struct task_struct *task = bpf_get_current_task_btf(); + struct callback_ctx data = {0}; + + if (task->pid != target_pid) + return 0; + + find_addr_ret = bpf_find_vma(task, addr, check_vma, &data, 0); + + /* this should return -ENOENT */ + find_zero_ret = bpf_find_vma(task, 0, check_vma, &data, 0); + return 0; +} + +SEC("perf_event") +int handle_pe(void) +{ + struct task_struct *task = bpf_get_current_task_btf(); + struct callback_ctx data = {0}; + + if (task->pid != target_pid) + return 0; + + find_addr_ret = bpf_find_vma(task, addr, check_vma, &data, 0); + + /* In NMI, this should return -EBUSY, as the previous call is using + * the irq_work. + */ + find_zero_ret = bpf_find_vma(task, 0, check_vma, &data, 0); + return 0; +} diff --git a/tools/testing/selftests/bpf/progs/find_vma_fail1.c b/tools/testing/selftests/bpf/progs/find_vma_fail1.c new file mode 100644 index 0000000000000..d17bdcdf76f07 --- /dev/null +++ b/tools/testing/selftests/bpf/progs/find_vma_fail1.c @@ -0,0 +1,30 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright (c) 2021 Facebook */ +#include "vmlinux.h" +#include <bpf/bpf_helpers.h> + +char _license[] SEC("license") = "GPL"; + +struct callback_ctx { + int dummy; +}; + +static __u64 +write_vma(struct task_struct *task, struct vm_area_struct *vma, + struct callback_ctx *data) +{ + /* writing to vma, which is illegal */ + vma->vm_flags |= 0x55; + + return 0; +} + +SEC("kprobe/__x64_sys_getpgid") +int handle_getpid(void) +{ + struct task_struct *task = bpf_get_current_task_btf(); + struct callback_ctx data = {0}; + + bpf_find_vma(task, 0, write_vma, &data, 0); + return 0; +} diff --git a/tools/testing/selftests/bpf/progs/find_vma_fail2.c b/tools/testing/selftests/bpf/progs/find_vma_fail2.c new file mode 100644 index 0000000000000..079c4594c095d --- /dev/null +++ b/tools/testing/selftests/bpf/progs/find_vma_fail2.c @@ -0,0 +1,30 @@ +// SPDX-License-Identifier: GPL-2.0 +/* Copyright (c) 2021 Facebook */ +#include "vmlinux.h" +#include <bpf/bpf_helpers.h> + +char _license[] SEC("license") = "GPL"; + +struct callback_ctx { + int dummy; +}; + +static __u64 +write_task(struct task_struct *task, struct vm_area_struct *vma, + struct callback_ctx *data) +{ + /* writing to task, which is illegal */ + task->mm = NULL; + + return 0; +} + +SEC("kprobe/__x64_sys_getpgid") +int handle_getpid(void) +{ + struct task_struct *task = bpf_get_current_task_btf(); + struct callback_ctx data = {0}; + + bpf_find_vma(task, 0, write_task, &data, 0); + return 0; +}
Add tests for bpf_find_vma in perf_event program and kprobe program. The perf_event program is triggered from NMI context, so the second call of bpf_find_vma() will return -EBUSY (irq_work busy). The kprobe program, on the other hand, does not have this constraint. Also add test for illegal writes to task or vma from the callback function. The verifier should reject both cases. Signed-off-by: Song Liu <songliubraving@fb.com> --- .../selftests/bpf/prog_tests/find_vma.c | 115 ++++++++++++++++++ tools/testing/selftests/bpf/progs/find_vma.c | 70 +++++++++++ .../selftests/bpf/progs/find_vma_fail1.c | 30 +++++ .../selftests/bpf/progs/find_vma_fail2.c | 30 +++++ 4 files changed, 245 insertions(+) create mode 100644 tools/testing/selftests/bpf/prog_tests/find_vma.c create mode 100644 tools/testing/selftests/bpf/progs/find_vma.c create mode 100644 tools/testing/selftests/bpf/progs/find_vma_fail1.c create mode 100644 tools/testing/selftests/bpf/progs/find_vma_fail2.c