Skip to contentSkip to conversation

bpf: Add bpf_copy_from_user_str kfunc

PATCH Superseded Jordan Rome 3 participants 6 messages Aug 15, 2024 – Aug 19, 2024

Subsystems (3)

v5 of 10 Superseded Latest is v10 →

lore
Share & tools
lore
#0 lore raw

PATCH v5 1/2 bpf: Add bpf_copy_from_user_str kfunc

This adds a kfunc wrapper around strncpy_from_user, which can be called from sleepable BPF programs.

This matches the non-sleepable 'bpf_probe_read_user_str' helper except it includes an additional 'flags' param, which allows consumers to clear the entire destination buffer on success.

---
 include/uapi/linux/bpf.h       |  8 +++++++
 kernel/bpf/helpers.c           | 41 ++++++++++++++++++++++++++++++++++
 tools/include/uapi/linux/bpf.h |  8 +++++++
 3 files changed, 57 insertions(+)
include/uapi/linux/bpf.h +8 -0
index e05b39e39c3f..e207175981be 100644
@@ -7513,4 +7513,12 @@ struct bpf_iter_num {
 	__u64 __opaque[1];
 } __attribute__((aligned(8)));
 
+/*
+ * Flags to control bpf_copy_from_user_str() behaviour.
+ *     - BPF_ZERO_BUFFER: Memset 0 the tail of the destination buffer on success
+ */
+enum {
+	BPF_ZERO_BUFFER = (1ULL << 0)
+};
+
 #endif /* _UAPI__LINUX_BPF_H__ */
kernel/bpf/helpers.c +41 -0

53 lines not on this page · raw message

tools/include/uapi/linux/bpf.h +8 -0
index e05b39e39c3f..15c2c3431e0f 100644
@@ -7513,4 +7513,12 @@ struct bpf_iter_num {
 	__u64 __opaque[1];
 } __attribute__((aligned(8)));
 
+/*
+ * Flags to control bpf_copy_from_user_str() behaviour.
+ *     - BPF_ZERO_BUFFER: Memset 0 the entire destination buffer on success
+ */
+enum {
+	BPF_ZERO_BUFFER = (1ULL << 0)
+};
+
 #endif /* _UAPI__LINUX_BPF_H__ */

--
2.43.5

#1 lore raw

PATCH v5 2/2 bpf: Add tests for bpf_copy_from_user_str kfunc

This adds tests for both the happy path and
the error path.

---
 .../selftests/bpf/prog_tests/attach_probe.c   |  8 ++--
 .../selftests/bpf/prog_tests/read_vsyscall.c  |  1 +
 .../selftests/bpf/progs/read_vsyscall.c       |  9 ++++-
 .../selftests/bpf/progs/test_attach_probe.c   | 38 +++++++++++++++++--
 4 files changed, 49 insertions(+), 7 deletions(-)
tools/testing/selftests/bpf/prog_tests/attach_probe.c +5 -3
index 7175af39134f..329c7862b52d 100644
@@ -283,9 +283,11 @@ static void test_uprobe_sleepable(struct test_attach_probe *skel)
 	trigger_func3();
 
 	ASSERT_EQ(skel->bss->uprobe_byname3_sleepable_res, 9, "check_uprobe_byname3_sleepable_res");
-	ASSERT_EQ(skel->bss->uprobe_byname3_res, 10, "check_uprobe_byname3_res");
-	ASSERT_EQ(skel->bss->uretprobe_byname3_sleepable_res, 11, "check_uretprobe_byname3_sleepable_res");
-	ASSERT_EQ(skel->bss->uretprobe_byname3_res, 12, "check_uretprobe_byname3_res");
+	ASSERT_EQ(skel->bss->uprobe_byname3_str_sleepable_res, 10, "check_uprobe_byname3_str_sleepable_res");
+	ASSERT_EQ(skel->bss->uprobe_byname3_res, 11, "check_uprobe_byname3_res");
+	ASSERT_EQ(skel->bss->uretprobe_byname3_sleepable_res, 12, "check_uretprobe_byname3_sleepable_res");
+	ASSERT_EQ(skel->bss->uretprobe_byname3_str_sleepable_res, 13, "check_uretprobe_byname3_str_sleepable_res");
+	ASSERT_EQ(skel->bss->uretprobe_byname3_res, 14, "check_uretprobe_byname3_res");
 }
 
 void test_attach_probe(void)
tools/testing/selftests/bpf/prog_tests/read_vsyscall.c +1 -0
index 3405923fe4e6..c7b9ba8b1d06 100644
@@ -23,6 +23,7 @@ struct read_ret_desc {
 	{ .name = "probe_read_user_str", .ret = -EFAULT },
 	{ .name = "copy_from_user", .ret = -EFAULT },
 	{ .name = "copy_from_user_task", .ret = -EFAULT },
+	{ .name = "copy_from_user_str", .ret = -EFAULT },
 };
 
 void test_read_vsyscall(void)
tools/testing/selftests/bpf/progs/read_vsyscall.c +8 -1
index 986f96687ae1..39ebef430059 100644
@@ -1,5 +1,6 @@
 // SPDX-License-Identifier: GPL-2.0
 /* Copyright (C) 2024. Huawei Technologies Co., Ltd */
+#include "vmlinux.h"
 #include <linux/types.h>
 #include <bpf/bpf_helpers.h>
 
@@ -7,10 +8,15 @@
 
 int target_pid = 0;
 void *user_ptr = 0;
-int read_ret[8];
+int read_ret[9];
 
 char _license[] SEC("license") = "GPL";
 
+/*
+ * This is the only kfunc, the others are helpers
+ */
+int bpf_copy_from_user_str(void *dst, u32, const void *, u64) __weak __ksym;
+
 SEC("fentry/" SYS_PREFIX "sys_nanosleep")
 int do_probe_read(void *ctx)
 {
@@ -40,6 +46,7 @@ int do_copy_from_user(void *ctx)
 	read_ret[6] = bpf_copy_from_user(buf, sizeof(buf), user_ptr);
 	read_ret[7] = bpf_copy_from_user_task(buf, sizeof(buf), user_ptr,
 					      bpf_get_current_task_btf(), 0);
+	read_ret[8] = bpf_copy_from_user_str((char *)buf, sizeof(buf), user_ptr, 0);
 
 	return 0;
 }
tools/testing/selftests/bpf/progs/test_attach_probe.c +35 -3
index 68466a6ad18c..705830d44101 100644
@@ -14,11 +14,15 @@ int uretprobe_byname_res = 0;
 int uprobe_byname2_res = 0;
 int uretprobe_byname2_res = 0;
 int uprobe_byname3_sleepable_res = 0;
+int uprobe_byname3_str_sleepable_res = 0;
 int uprobe_byname3_res = 0;
 int uretprobe_byname3_sleepable_res = 0;
+int uretprobe_byname3_str_sleepable_res = 0;
 int uretprobe_byname3_res = 0;
 void *user_ptr = 0;
 
+int bpf_copy_from_user_str(void *dst, u32, const void *, u64) __weak __ksym;
+
 SEC("ksyscall/nanosleep")
 int BPF_KSYSCALL(handle_kprobe_auto, struct __kernel_timespec *req, struct __kernel_timespec *rem)
 {
@@ -87,11 +91,37 @@ static __always_inline bool verify_sleepable_user_copy(void)
 	return bpf_strncmp(data, sizeof(data), "test_data") == 0;
 }
 
+static __always_inline bool verify_sleepable_user_copy_str(void)
+{
+	int ret;
+	char data_long[20];
+	char data_short[4];
+
+	ret = bpf_copy_from_user_str(data_short, sizeof(data_short), user_ptr, 0);
+
+	if (bpf_strncmp(data_short, 4, "tes\0") != 0 || ret != 4)
+		return false;
+
+	ret = bpf_copy_from_user_str(data_long, sizeof(data_long), user_ptr, BPF_ZERO_BUFFER);
+
+	if (bpf_strncmp(data_long, 10, "test_data\0") != 0 || ret != 10 || data_long[19] != '\0')
+		return false;
+
+	ret = bpf_copy_from_user_str(data_long, sizeof(data_long), user_ptr, 0);
+
+	if (bpf_strncmp(data_long, 10, "test_data\0") != 0 || ret != 10)
+		return false;
+
+	return true;
+}
+
 SEC("uprobe.s//proc/self/exe:trigger_func3")
 int handle_uprobe_byname3_sleepable(struct pt_regs *ctx)
 {
 	if (verify_sleepable_user_copy())
 		uprobe_byname3_sleepable_res = 9;
+	if (verify_sleepable_user_copy_str())
+		uprobe_byname3_str_sleepable_res = 10;
 	return 0;
 }
 
@@ -102,7 +132,7 @@ int handle_uprobe_byname3_sleepable(struct pt_regs *ctx)
 SEC("uprobe//proc/self/exe:trigger_func3")
 int handle_uprobe_byname3(struct pt_regs *ctx)
 {
-	uprobe_byname3_res = 10;
+	uprobe_byname3_res = 11;
 	return 0;
 }
 
@@ -110,14 +140,16 @@ SEC("uretprobe.s//proc/self/exe:trigger_func3")
 int handle_uretprobe_byname3_sleepable(struct pt_regs *ctx)
 {
 	if (verify_sleepable_user_copy())
-		uretprobe_byname3_sleepable_res = 11;
+		uretprobe_byname3_sleepable_res = 12;
+	if (verify_sleepable_user_copy_str())
+		uretprobe_byname3_str_sleepable_res = 13;
 	return 0;
 }
 
 SEC("uretprobe//proc/self/exe:trigger_func3")
 int handle_uretprobe_byname3(struct pt_regs *ctx)
 {
-	uretprobe_byname3_res = 12;
+	uretprobe_byname3_res = 14;
 	return 0;
 }
 

--
2.43.5

#2 lore raw

Re: [bpf-next v5 2/2] bpf: Add tests for bpf_copy_from_user_str kfunc

Quoted 4 lines · Jordan Rome

This adds tests for both the happy path and
the error path.

Quoted 8 lines
---
 .../selftests/bpf/prog_tests/attach_probe.c   |  8 ++--
 .../selftests/bpf/prog_tests/read_vsyscall.c  |  1 +
 .../selftests/bpf/progs/read_vsyscall.c       |  9 ++++-
 .../selftests/bpf/progs/test_attach_probe.c   | 38 +++++++++++++++++--
 4 files changed, 49 insertions(+), 7 deletions(-)

As I mentioned in the first patch, it would be better to have a bit more extensive testing. All those rare conditions:

- dst_size is zero
- dst_size is one
- string is empty
- string exactly fits
- string is truncated
- plus various error conditions.

Quoted 7 lines
tools/testing/selftests/bpf/prog_tests/attach_probe.c +0 -0
index 7175af39134f..329c7862b52d 100644
@@ -283,9 +283,11 @@ static void test_uprobe_sleepable(struct test_attach_probe *skel)
        trigger_func3();
 

[...]

#3 lore raw
In reply to Jordan Rome
Quoted 21 lines · Jordan Rome

This adds a kfunc wrapper around strncpy_from_user, which can be called from sleepable BPF programs.

This matches the non-sleepable 'bpf_probe_read_user_str' helper except it includes an additional 'flags' param, which allows consumers to clear the entire destination buffer on success.

---
 include/uapi/linux/bpf.h       |  8 +++++++
 kernel/bpf/helpers.c           | 41 ++++++++++++++++++++++++++++++++++
 tools/include/uapi/linux/bpf.h |  8 +++++++
 3 files changed, 57 insertions(+)
include/uapi/linux/bpf.h +0 -0
index e05b39e39c3f..e207175981be 100644

include/uapi/linux/bpf.h struct bpf_iter_num { · line 7521

Quoted 9 lines
include/uapi/linux/bpf.h +6 -0
@@ -7513,4 +7513,12 @@ struct bpf_iter_num {
        __u64 __opaque[1];
 } __attribute__((aligned(8)));
 
+/*
+ * Flags to control bpf_copy_from_user_str() behaviour.
+ *     - BPF_ZERO_BUFFER: Memset 0 the tail of the destination buffer on success
+ */
+enum {
+       BPF_ZERO_BUFFER = (1ULL << 0)

We call all flags BPF_F_<something>, so let's stay consistent.

And just for a bit of bikeshedding, "zero buffer" isn't immediately clear and it would be nice to have a clearer verb in there. I don't have a perfect name, but something like BPF_F_PAD_ZEROS or something with "pad" maybe?

Also, should we keep behavior a bit more consistent and say that on failure this flag will also ensure that buffer is cleared?

Quoted 20 lines
include/uapi/linux/bpf.h +2 -0
+};
+
 #endif /* _UAPI__LINUX_BPF_H__ */
kernel/bpf/helpers.c +9 -0
index d02ae323996b..fe4348679d38 100644
@@ -2939,6 +2939,46 @@ __bpf_kfunc void bpf_iter_bits_destroy(struct bpf_iter_bits *it)
        bpf_mem_free(&bpf_global_ma, kit->bits);
 }
 
+/**
+ * bpf_copy_from_user_str() - Copy a string from an unsafe user address
+ * @dst:             Destination address, in kernel space.  This buffer must be at
+ *                   least @dst__szk bytes long.
+ * @dst__szk:        Maximum number of bytes to copy, including the trailing NUL.
+ * @unsafe_ptr__ign: Source address, in user space.
+ * @flags:           The only supported flag is BPF_ZERO_BUFFER
+ *
+ * Copies a NUL-terminated string from userspace to BPF space. If user string is

kernel/bpf/helpers.c __bpf_kfunc void bpf_iter_bits_destroy(struct bpf_iter_bits *it) · line 2960

Quoted 10 lines
kernel/bpf/helpers.c +10 -0
+ * too long this will still ensure zero termination in the dst buffer unless
+ * buffer size is 0.
+ *
+ * If BPF_ZERO_BUFFER flag is set, memset the tail of @dst to 0 on success.
+ */
+__bpf_kfunc int bpf_copy_from_user_str(void *dst, u32 dst__szk, const void __user *unsafe_ptr__ign, u64 flags)
+{
+       int ret;
+       int count;
+

validate that flags doesn't have any unknown flags

if (unlikely(flags & ~BPF_F_ZERO_BUFFER))

    return -EINVAL;

kernel/bpf/helpers.c __bpf_kfunc void bpf_iter_bits_destroy(struct bpf_iter_bits *it) · line 2968

Quoted 8 lines
kernel/bpf/helpers.c +8 -0
+       if (unlikely(!dst__szk))
+               return 0;
+
+       count = dst__szk - 1;
+       if (unlikely(!count)) {
+               ((char *)dst)[0] = '\0';
+               return 1;
+       }

Do we need to special-case this unlikely scenario? Especially that it's unlikely, why write code for it and pay a tiny price for an extra check?

kernel/bpf/helpers.c __bpf_kfunc void bpf_iter_bits_destroy(struct bpf_iter_bits *it) · line 2976

Quoted 8 lines
kernel/bpf/helpers.c +8 -0
+
+       ret = strncpy_from_user(dst, unsafe_ptr__ign, count);
+       if (ret >= 0) {
+               if (flags & BPF_ZERO_BUFFER)
+                       memset((char *)dst + ret, 0, dst__szk - ret);
+               else
+                       ((char *)dst)[ret] = '\0';
+               ret++;

so if string is truncated, ret == count, no? And dst[ret] will go beyond the buffer?

we need more tests to validate all those various conditions

I'd also rewrite this a bit, so it's more linear:

ret = strncpy(...);
if (ret < 0)

    return ret;

((char *)dst)[count - 1] = '\0';

if (flags & BPF_F_ZERO_BUF)

      memset(...);

return ret < count ? ret + 1 : count;

or something along those lines

pw-bot: cr

Quoted 36 lines
kernel/bpf/helpers.c +6 -0
+       }
+
+       return ret;
+}
+
 __bpf_kfunc_end_defs();
 
 BTF_KFUNCS_START(generic_btf_ids)
@@ -3024,6 +3064,7 @@ BTF_ID_FLAGS(func, bpf_preempt_enable)
 BTF_ID_FLAGS(func, bpf_iter_bits_new, KF_ITER_NEW)
 BTF_ID_FLAGS(func, bpf_iter_bits_next, KF_ITER_NEXT | KF_RET_NULL)
 BTF_ID_FLAGS(func, bpf_iter_bits_destroy, KF_ITER_DESTROY)
+BTF_ID_FLAGS(func, bpf_copy_from_user_str, KF_SLEEPABLE)
 BTF_KFUNCS_END(common_btf_ids)
 
 static const struct btf_kfunc_id_set common_kfunc_set = {
tools/include/uapi/linux/bpf.h +8 -1
index e05b39e39c3f..15c2c3431e0f 100644
@@ -7513,4 +7513,12 @@ struct bpf_iter_num {
        __u64 __opaque[1];
 } __attribute__((aligned(8)));
 
+/*
+ * Flags to control bpf_copy_from_user_str() behaviour.
+ *     - BPF_ZERO_BUFFER: Memset 0 the entire destination buffer on success
+ */
+enum {
+       BPF_ZERO_BUFFER = (1ULL << 0)
+};
+
 #endif /* _UAPI__LINUX_BPF_H__ */
--

2.43.5

#4 lore raw
Quoted 100 lines · Andrii Nakryiko
Quoted 21 lines · Jordan Rome

This adds a kfunc wrapper around strncpy_from_user, which can be called from sleepable BPF programs.

This matches the non-sleepable 'bpf_probe_read_user_str' helper except it includes an additional 'flags' param, which allows consumers to clear the entire destination buffer on success.

---
  include/uapi/linux/bpf.h       |  8 +++++++
  kernel/bpf/helpers.c           | 41 ++++++++++++++++++++++++++++++++++
  tools/include/uapi/linux/bpf.h |  8 +++++++
  3 files changed, 57 insertions(+)
include/uapi/linux/bpf.h +0 -0
index e05b39e39c3f..e207175981be 100644

include/uapi/linux/bpf.h struct bpf_iter_num { · line 7521

Quoted 9 lines
include/uapi/linux/bpf.h +6 -0
@@ -7513,4 +7513,12 @@ struct bpf_iter_num {
         __u64 __opaque[1];
  } __attribute__((aligned(8)));
 
+/*
+ * Flags to control bpf_copy_from_user_str() behaviour.
+ *     - BPF_ZERO_BUFFER: Memset 0 the tail of the destination buffer on success
+ */
+enum {
+       BPF_ZERO_BUFFER = (1ULL << 0)

We call all flags BPF_F_<something>, so let's stay consistent.

And just for a bit of bikeshedding, "zero buffer" isn't immediately clear and it would be nice to have a clearer verb in there. I don't have a perfect name, but something like BPF_F_PAD_ZEROS or something with "pad" maybe?

Also, should we keep behavior a bit more consistent and say that on failure this flag will also ensure that buffer is cleared?

Quoted 20 lines
include/uapi/linux/bpf.h +2 -0
+};
+
  #endif /* _UAPI__LINUX_BPF_H__ */
kernel/bpf/helpers.c +9 -0
index d02ae323996b..fe4348679d38 100644
@@ -2939,6 +2939,46 @@ __bpf_kfunc void bpf_iter_bits_destroy(struct bpf_iter_bits *it)
         bpf_mem_free(&bpf_global_ma, kit->bits);
  }
 
+/**
+ * bpf_copy_from_user_str() - Copy a string from an unsafe user address
+ * @dst:             Destination address, in kernel space.  This buffer must be at
+ *                   least @dst__szk bytes long.
+ * @dst__szk:        Maximum number of bytes to copy, including the trailing NUL.
+ * @unsafe_ptr__ign: Source address, in user space.
+ * @flags:           The only supported flag is BPF_ZERO_BUFFER
+ *
+ * Copies a NUL-terminated string from userspace to BPF space. If user string is

kernel/bpf/helpers.c __bpf_kfunc void bpf_iter_bits_destroy(struct bpf_iter_bits *it) · line 2960

Quoted 10 lines
kernel/bpf/helpers.c +10 -0
+ * too long this will still ensure zero termination in the dst buffer unless
+ * buffer size is 0.
+ *
+ * If BPF_ZERO_BUFFER flag is set, memset the tail of @dst to 0 on success.
+ */
+__bpf_kfunc int bpf_copy_from_user_str(void *dst, u32 dst__szk, const void __user *unsafe_ptr__ign, u64 flags)
+{
+       int ret;
+       int count;
+

validate that flags doesn't have any unknown flags

if (unlikely(flags & ~BPF_F_ZERO_BUFFER))

     return -EINVAL;

kernel/bpf/helpers.c __bpf_kfunc void bpf_iter_bits_destroy(struct bpf_iter_bits *it) · line 2968

Quoted 8 lines
kernel/bpf/helpers.c +8 -0
+       if (unlikely(!dst__szk))
+               return 0;
+
+       count = dst__szk - 1;
+       if (unlikely(!count)) {
+               ((char *)dst)[0] = '\0';
+               return 1;
+       }

Do we need to special-case this unlikely scenario? Especially that it's unlikely, why write code for it and pay a tiny price for an extra check?

kernel/bpf/helpers.c __bpf_kfunc void bpf_iter_bits_destroy(struct bpf_iter_bits *it) · line 2976

Quoted 8 lines
kernel/bpf/helpers.c +8 -0
+
+       ret = strncpy_from_user(dst, unsafe_ptr__ign, count);
+       if (ret >= 0) {
+               if (flags & BPF_ZERO_BUFFER)
+                       memset((char *)dst + ret, 0, dst__szk - ret);
+               else
+                       ((char *)dst)[ret] = '\0';
+               ret++;
Quoted 2 lines

so if string is truncated, ret == count, no? And dst[ret] will go beyond the buffer?

Since count = dst__szk - 1, it is not going beyond the buffer.

Quoted 61 lines

we need more tests to validate all those various conditions

I'd also rewrite this a bit, so it's more linear:

ret = strncpy(...);
if (ret < 0)

     return ret;

((char *)dst)[count - 1] = '\0';

if (flags & BPF_F_ZERO_BUF)

       memset(...);

return ret < count ? ret + 1 : count;

or something along those lines

pw-bot: cr

Quoted 36 lines
kernel/bpf/helpers.c +6 -0
+       }
+
+       return ret;
+}
+
  __bpf_kfunc_end_defs();
 
  BTF_KFUNCS_START(generic_btf_ids)
@@ -3024,6 +3064,7 @@ BTF_ID_FLAGS(func, bpf_preempt_enable)
  BTF_ID_FLAGS(func, bpf_iter_bits_new, KF_ITER_NEW)
  BTF_ID_FLAGS(func, bpf_iter_bits_next, KF_ITER_NEXT | KF_RET_NULL)
  BTF_ID_FLAGS(func, bpf_iter_bits_destroy, KF_ITER_DESTROY)
+BTF_ID_FLAGS(func, bpf_copy_from_user_str, KF_SLEEPABLE)
  BTF_KFUNCS_END(common_btf_ids)
 
  static const struct btf_kfunc_id_set common_kfunc_set = {
tools/include/uapi/linux/bpf.h +8 -1
index e05b39e39c3f..15c2c3431e0f 100644
@@ -7513,4 +7513,12 @@ struct bpf_iter_num {
         __u64 __opaque[1];
  } __attribute__((aligned(8)));
 
+/*
+ * Flags to control bpf_copy_from_user_str() behaviour.
+ *     - BPF_ZERO_BUFFER: Memset 0 the entire destination buffer on success
+ */
+enum {
+       BPF_ZERO_BUFFER = (1ULL << 0)
+};
+
  #endif /* _UAPI__LINUX_BPF_H__ */
--

2.43.5

#5 lore raw
Quoted 104 lines · Kui-Feng Lee
Quoted 100 lines · Andrii Nakryiko
Quoted 21 lines · Jordan Rome

This adds a kfunc wrapper around strncpy_from_user, which can be called from sleepable BPF programs.

This matches the non-sleepable 'bpf_probe_read_user_str' helper except it includes an additional 'flags' param, which allows consumers to clear the entire destination buffer on success.

---
  include/uapi/linux/bpf.h       |  8 +++++++
  kernel/bpf/helpers.c           | 41 ++++++++++++++++++++++++++++++++++
  tools/include/uapi/linux/bpf.h |  8 +++++++
  3 files changed, 57 insertions(+)
include/uapi/linux/bpf.h +0 -0
index e05b39e39c3f..e207175981be 100644

include/uapi/linux/bpf.h struct bpf_iter_num { · line 7521

Quoted 9 lines
include/uapi/linux/bpf.h +6 -0
@@ -7513,4 +7513,12 @@ struct bpf_iter_num {
         __u64 __opaque[1];
  } __attribute__((aligned(8)));
 
+/*
+ * Flags to control bpf_copy_from_user_str() behaviour.
+ *     - BPF_ZERO_BUFFER: Memset 0 the tail of the destination buffer on success
+ */
+enum {
+       BPF_ZERO_BUFFER = (1ULL << 0)

We call all flags BPF_F_<something>, so let's stay consistent.

And just for a bit of bikeshedding, "zero buffer" isn't immediately clear and it would be nice to have a clearer verb in there. I don't have a perfect name, but something like BPF_F_PAD_ZEROS or something with "pad" maybe?

Also, should we keep behavior a bit more consistent and say that on failure this flag will also ensure that buffer is cleared?

Quoted 20 lines
include/uapi/linux/bpf.h +2 -0
+};
+
  #endif /* _UAPI__LINUX_BPF_H__ */
kernel/bpf/helpers.c +9 -0
index d02ae323996b..fe4348679d38 100644
@@ -2939,6 +2939,46 @@ __bpf_kfunc void bpf_iter_bits_destroy(struct bpf_iter_bits *it)
         bpf_mem_free(&bpf_global_ma, kit->bits);
  }
 
+/**
+ * bpf_copy_from_user_str() - Copy a string from an unsafe user address
+ * @dst:             Destination address, in kernel space.  This buffer must be at
+ *                   least @dst__szk bytes long.
+ * @dst__szk:        Maximum number of bytes to copy, including the trailing NUL.
+ * @unsafe_ptr__ign: Source address, in user space.
+ * @flags:           The only supported flag is BPF_ZERO_BUFFER
+ *
+ * Copies a NUL-terminated string from userspace to BPF space. If user string is

kernel/bpf/helpers.c __bpf_kfunc void bpf_iter_bits_destroy(struct bpf_iter_bits *it) · line 2960

Quoted 10 lines
kernel/bpf/helpers.c +10 -0
+ * too long this will still ensure zero termination in the dst buffer unless
+ * buffer size is 0.
+ *
+ * If BPF_ZERO_BUFFER flag is set, memset the tail of @dst to 0 on success.
+ */
+__bpf_kfunc int bpf_copy_from_user_str(void *dst, u32 dst__szk, const void __user *unsafe_ptr__ign, u64 flags)
+{
+       int ret;
+       int count;
+

validate that flags doesn't have any unknown flags

if (unlikely(flags & ~BPF_F_ZERO_BUFFER))

     return -EINVAL;

kernel/bpf/helpers.c __bpf_kfunc void bpf_iter_bits_destroy(struct bpf_iter_bits *it) · line 2968

Quoted 8 lines
kernel/bpf/helpers.c +8 -0
+       if (unlikely(!dst__szk))
+               return 0;
+
+       count = dst__szk - 1;
+       if (unlikely(!count)) {
+               ((char *)dst)[0] = '\0';
+               return 1;
+       }

Do we need to special-case this unlikely scenario? Especially that it's unlikely, why write code for it and pay a tiny price for an extra check?

Quoted 8 lines
kernel/bpf/helpers.c +8 -0
+
+       ret = strncpy_from_user(dst, unsafe_ptr__ign, count);
+       if (ret >= 0) {
+               if (flags & BPF_ZERO_BUFFER)
+                       memset((char *)dst + ret, 0, dst__szk - ret);
+               else
+                       ((char *)dst)[ret] = '\0';
+               ret++;
Quoted 5 lines
Quoted 2 lines

so if string is truncated, ret == count, no? And dst[ret] will go beyond the buffer?

Since count = dst__szk - 1, it is not going beyond the buffer.

ah, I forgot that count is adjusted size already, ok

Quoted 61 lines
Quoted 61 lines

we need more tests to validate all those various conditions

I'd also rewrite this a bit, so it's more linear:

ret = strncpy(...);
if (ret < 0)

     return ret;

((char *)dst)[count - 1] = '\0';

if (flags & BPF_F_ZERO_BUF)

       memset(...);

return ret < count ? ret + 1 : count;

or something along those lines

pw-bot: cr

Quoted 36 lines
kernel/bpf/helpers.c +6 -0
+       }
+
+       return ret;
+}
+
  __bpf_kfunc_end_defs();
 
  BTF_KFUNCS_START(generic_btf_ids)
@@ -3024,6 +3064,7 @@ BTF_ID_FLAGS(func, bpf_preempt_enable)
  BTF_ID_FLAGS(func, bpf_iter_bits_new, KF_ITER_NEW)
  BTF_ID_FLAGS(func, bpf_iter_bits_next, KF_ITER_NEXT | KF_RET_NULL)
  BTF_ID_FLAGS(func, bpf_iter_bits_destroy, KF_ITER_DESTROY)
+BTF_ID_FLAGS(func, bpf_copy_from_user_str, KF_SLEEPABLE)
  BTF_KFUNCS_END(common_btf_ids)
 
  static const struct btf_kfunc_id_set common_kfunc_set = {
tools/include/uapi/linux/bpf.h +8 -1
index e05b39e39c3f..15c2c3431e0f 100644
@@ -7513,4 +7513,12 @@ struct bpf_iter_num {
         __u64 __opaque[1];
  } __attribute__((aligned(8)));
 
+/*
+ * Flags to control bpf_copy_from_user_str() behaviour.
+ *     - BPF_ZERO_BUFFER: Memset 0 the entire destination buffer on success
+ */
+enum {
+       BPF_ZERO_BUFFER = (1ULL << 0)
+};
+
  #endif /* _UAPI__LINUX_BPF_H__ */
--

2.43.5