Skip to contentSkip to conversation

BPF Standard Streams

PATCH Superseded Kumar Kartikeya Dwivedi 4 participants 23 messages Jun 24, 2025 – Jun 25, 2025

Subsystems (8)
x86 Kselftest BPF BPF selftests +4 more subsystems

v3 of 5 Superseded Latest is v5 →

lore
Share & tools
lore

Kumar Kartikeya Dwivedi

author maintainer
#0 lore raw

v3 cover BPF Standard Streams

This set introduces a standard output interface with two streams, namely stdout and stderr, for BPF programs. The idea is that these streams will be written to by BPF programs and the kernel, and serve as standard interfaces for informing user space of any BPF runtime violations. Users can also utilize them for printing normal messages for debugging usage, as is the case with bpf_printk() and trace pipe interface.

BPF programs and the kernel can use these streams to output messages. User space can dump these messages using bpftool.

The stream interface itself is implemented using a lockless list, so that we can queue messages from any context. Every printk statement into the stream leads to memory allocation. Allocation itself relies on try_alloc_pages() to construct a bespoke bump allocator to carve out elements. If this fails, we finally give up and drop the message.

See commit logs for more details.

Two scenarios are covered:
- Deadlocks and timeouts in rqspinlock.
- Timeouts for may_goto.

In each we provide the stack trace and source information for the offending BPF programs. Both the C source line and the file and line numbers are printed. The output format is as follows:

ERROR: AA or ABBA deadlock detected for bpf_res_spin_lock
Attempted lock = 0xff11000108f3a5e0
Total held locks = 1
Held lock[ 0] = 0xff11000108f3a5e0

CPU: 48 UID: 0 PID: 786 Comm: test_progs
Call trace:
bpf_stream_stage_dump_stack+0xb0/0xd0
bpf_prog_report_rqspinlock_violation+0x10b/0x130
bpf_res_spin_lock+0x8c/0xa0
bpf_prog_3699ea119d1f6ed8_foo+0xe5/0x140
  if (!bpf_res_spin_lock(&v2->lock)) @ stream_bpftool.c:62
bpf_prog_9b324ec4a1b2a5c0_stream_bpftool_dump_prog_stream+0x7e/0x2d0
  foo(stream); @ stream_bpftool.c:93
bpf_prog_test_run_syscall+0x102/0x240
__sys_bpf+0xd68/0x2bf0
__x64_sys_bpf+0x1e/0x30
do_syscall_64+0x68/0x140
entry_SYSCALL_64_after_hwframe+0x76/0x7e

ERROR: Timeout detected for may_goto instruction

CPU: 48 UID: 0 PID: 786 Comm: test_progs
Call trace:
bpf_stream_stage_dump_stack+0xb0/0xd0
bpf_prog_report_may_goto_violation+0x6a/0x90
bpf_check_timed_may_goto+0x4d/0xa0
arch_bpf_timed_may_goto+0x21/0x40
bpf_prog_3699ea119d1f6ed8_foo+0x12f/0x140
  while (can_loop) @ stream_bpftool.c:71
bpf_prog_9b324ec4a1b2a5c0_stream_bpftool_dump_prog_stream+0x7e/0x2d0
  foo(stream); @ stream_bpftool.c:93
bpf_prog_test_run_syscall+0x102/0x240
__sys_bpf+0xd68/0x2bf0
__x64_sys_bpf+0x1e/0x30
do_syscall_64+0x68/0x140
entry_SYSCALL_64_after_hwframe+0x76/0x7e

Changelog:
----------
v2 -> v3
v2: https://lore.kernel.org/bpf/[email protected]

* Fix bug when handling single element stream stage. (Eduard)
* Move to mutex for protection of stream read and copy_to_user(). (Alexei)
* Split bprintf refactor into its own patch. (Alexei)
* Move kfunc definition to common_btf_ids to avoid initcall proliferation. (Alexei)
* Return line number by reference in bpf_prog_get_file_line. (Alexei)
* Remove NULL checks for BTF name pointer. (Alexei)
* Add WARN_ON_ONCE(!rcu_read_lock_held()) in bpf_prog_ksym_find. (Eduard)
* Remove hardcoded stream stage from macros. (Alexei, Eduard)
* Move refactoring hunks to their own patch. (Alexei)
* Add empty opts parameter for future extensibility to libbpf API. (Andrii, Eduard)
* Add BPF_STREAM_{STDOUT,STDERR} to UAPI. (Andrii)
* Add code to match on backtrace output. (Eduard)
* Fix misc nits.
* Add acks.

v1 -> v2
v1: https://lore.kernel.org/bpf/[email protected]

* Drop arena page fault prints, will be done as follow up. (Alexei)
* Defer Andrii's request to reuse code and Alan's suggestion of error counts to follow up.
* Drop bpf_dynptr_from_mem_slice patch.
* Drop some acks due to heavy reworking.
* Fix KASAN splat in bpf_prog_get_file_line. (Eduard)
* Collapse bpf_prog_ksym_find and is_bpf_text_address into single call. (Eduard)
* Add missing RCU read lock in bpf_prog_ksym_find.
* Fix incorrect error handling in dump_stack_cb.
* Simplify libbpf macro. (Eduard, Andrii)
* Introduce bpf_prog_stream_read() libbpf API. (Eduard, Alexei, Andrii)
* Drop BPF prog from the bpftool, use libbpf API.
* Rework selftests.

RFC v1 -> v1
RFC v1: https://lore.kernel.org/bpf/[email protected]

* Rebase on bpf-next/master.
* Change output in dump_stack to also print source line. (Alexei)
* Simplify API to single pop() operation. (Eduard, Alexei)
* Add kdoc for bpf_dynptr_from_mem_slice.
* Fix -EINVAL returned from prog_dump_stream. (Eduard)
* Split dump_stack() patch into multiple commits.
* Add macro wrapping stream staging API.
* Change bpftool command from dump to tracelog. (Quentin)
* Add bpftool documentation and bash completion. (Quentin)
* Change license of bpftool to Dual BSD/GPL.
* Simplify memory allocator. (Alexei)
* No overflow into second page.
* Remove bpf_mem_alloc() fallback.
* Symlink bpftool BPF program and exercise as selftest. (Eduard)
* Verify output after dumping from ringbuf. (Eduard)
* More failure cases to check API invariants.
* Remove patches for dynptr lifetime fixes (split into separate set).
* Limit maximum error messages, and add stream capacity. (Eduard)

Kumar Kartikeya Dwivedi (12):
bpf: Refactor bprintf buffer support
bpf: Introduce BPF standard streams
bpf: Add function to extract program source info
bpf: Ensure RCU lock is held around bpf_prog_ksym_find
bpf: Add function to find program from stack trace
bpf: Add dump_stack() analogue to print to BPF stderr
bpf: Report may_goto timeout to BPF stderr
bpf: Report rqspinlock deadlocks/timeout to BPF stderr
libbpf: Add bpf_stream_printk() macro
libbpf: Introduce bpf_prog_stream_read() API
bpftool: Add support for dumping streams selftests/bpf: Add tests for prog streams

 arch/x86/net/bpf_jit_comp.c                   |   1 -
 include/linux/bpf.h                           |  80 ++-
 include/uapi/linux/bpf.h                      |  24 +
 kernel/bpf/Makefile                           |   2 +-
 kernel/bpf/core.c                             | 105 +++-
 kernel/bpf/helpers.c                          |  29 +-
 kernel/bpf/rqspinlock.c                       |  23 +
 kernel/bpf/stream.c                           | 529 ++++++++++++++++++
 kernel/bpf/syscall.c                          |  27 +-
 .../bpftool/Documentation/bpftool-prog.rst    |   7 +
 tools/bpf/bpftool/bash-completion/bpftool     |  16 +-
 tools/bpf/bpftool/prog.c                      |  49 +-
 tools/include/uapi/linux/bpf.h                |  24 +
 tools/lib/bpf/bpf.c                           |  20 +
 tools/lib/bpf/bpf.h                           |  21 +
 tools/lib/bpf/bpf_helpers.h                   |  16 +
 tools/lib/bpf/libbpf.map                      |   1 +
 .../testing/selftests/bpf/prog_tests/stream.c | 140 +++++
 tools/testing/selftests/bpf/progs/stream.c    |  75 +++
 .../testing/selftests/bpf/progs/stream_fail.c |  17 +
 20 files changed, 1183 insertions(+), 23 deletions(-)
 create mode 100644 kernel/bpf/stream.c
 create mode 100644 tools/testing/selftests/bpf/prog_tests/stream.c
 create mode 100644 tools/testing/selftests/bpf/progs/stream.c
 create mode 100644 tools/testing/selftests/bpf/progs/stream_fail.c

base-commit: 3ce7cdde66e65a400b2d1b2a7f72c499e1db26b6

Signature · 1 line
2.47.1
author maintainer
#1 lore raw

PATCH v3 1/12 bpf: Refactor bprintf buffer support

Refactor code to be able to get and put bprintf buffers and use bpf_printf_prepare independently. This will be used in the next patch to implement BPF streams support, particularly as a staging buffer for strings that need to be formatted and then allocated and pushed into a stream.

---
 include/linux/bpf.h  | 15 ++++++++++++++-
 kernel/bpf/helpers.c | 26 +++++++++++---------------
 2 files changed, 25 insertions(+), 16 deletions(-)
include/linux/bpf.h +14 -1
index 5dd556e89cce..4fff0cee8622 100644
@@ -3550,6 +3550,16 @@ bool btf_id_set_contains(const struct btf_id_set *set, u32 id);
 #define MAX_BPRINTF_VARARGS		12
 #define MAX_BPRINTF_BUF			1024
 
+/* Per-cpu temp buffers used by printf-like helpers to store the bprintf binary
+ * arguments representation.
+ */
+#define MAX_BPRINTF_BIN_ARGS	512
+
+struct bpf_bprintf_buffers {
+	char bin_args[MAX_BPRINTF_BIN_ARGS];
+	char buf[MAX_BPRINTF_BUF];
+};
+
 struct bpf_bprintf_data {
 	u32 *bin_args;
 	char *buf;
@@ -3557,9 +3567,12 @@ struct bpf_bprintf_data {
 	bool get_buf;
 };
 
-int bpf_bprintf_prepare(char *fmt, u32 fmt_size, const u64 *raw_args,
+int bpf_bprintf_prepare(const char *fmt, u32 fmt_size, const u64 *raw_args,
 			u32 num_args, struct bpf_bprintf_data *data);
 void bpf_bprintf_cleanup(struct bpf_bprintf_data *data);
+int bpf_try_get_buffers(struct bpf_bprintf_buffers **bufs);
+void bpf_put_buffers(void);
+
 
 #ifdef CONFIG_BPF_LSM
 void bpf_cgroup_atype_get(u32 attach_btf_id, int cgroup_atype);
kernel/bpf/helpers.c +11 -15
index b71e428ad936..67d48f9fb173 100644
@@ -763,22 +763,13 @@ static int bpf_trace_copy_string(char *buf, void *unsafe_ptr, char fmt_ptype,
 	return -EINVAL;
 }
 
-/* Per-cpu temp buffers used by printf-like helpers to store the bprintf binary
- * arguments representation.
- */
-#define MAX_BPRINTF_BIN_ARGS	512
-
 /* Support executing three nested bprintf helper calls on a given CPU */
 #define MAX_BPRINTF_NEST_LEVEL	3
-struct bpf_bprintf_buffers {
-	char bin_args[MAX_BPRINTF_BIN_ARGS];
-	char buf[MAX_BPRINTF_BUF];
-};
 
 static DEFINE_PER_CPU(struct bpf_bprintf_buffers[MAX_BPRINTF_NEST_LEVEL], bpf_bprintf_bufs);
 static DEFINE_PER_CPU(int, bpf_bprintf_nest_level);
 
-static int try_get_buffers(struct bpf_bprintf_buffers **bufs)
+int bpf_try_get_buffers(struct bpf_bprintf_buffers **bufs)
 {
 	int nest_level;
 
@@ -794,16 +785,21 @@ static int try_get_buffers(struct bpf_bprintf_buffers **bufs)
 	return 0;
 }
 
-void bpf_bprintf_cleanup(struct bpf_bprintf_data *data)
+void bpf_put_buffers(void)
 {
-	if (!data->bin_args && !data->buf)
-		return;
 	if (WARN_ON_ONCE(this_cpu_read(bpf_bprintf_nest_level) == 0))
 		return;
 	this_cpu_dec(bpf_bprintf_nest_level);
 	preempt_enable();
 }
 
+void bpf_bprintf_cleanup(struct bpf_bprintf_data *data)
+{
+	if (!data->bin_args && !data->buf)
+		return;
+	bpf_put_buffers();
+}
+
 /*
  * bpf_bprintf_prepare - Generic pass on format strings for bprintf-like helpers
  *
@@ -818,7 +814,7 @@ void bpf_bprintf_cleanup(struct bpf_bprintf_data *data)
  * In argument preparation mode, if 0 is returned, safe temporary buffers are
  * allocated and bpf_bprintf_cleanup should be called to free them after use.
  */
-int bpf_bprintf_prepare(char *fmt, u32 fmt_size, const u64 *raw_args,
+int bpf_bprintf_prepare(const char *fmt, u32 fmt_size, const u64 *raw_args,
 			u32 num_args, struct bpf_bprintf_data *data)
 {
 	bool get_buffers = (data->get_bin_args && num_args) || data->get_buf;
@@ -834,7 +830,7 @@ int bpf_bprintf_prepare(char *fmt, u32 fmt_size, const u64 *raw_args,
 		return -EINVAL;
 	fmt_size = fmt_end - fmt;
 
-	if (get_buffers && try_get_buffers(&buffers))
+	if (get_buffers && bpf_try_get_buffers(&buffers))
 		return -EBUSY;
 
 	if (data->get_bin_args) {
Signature · 1 line
2.47.1
author maintainer
#2 lore raw

PATCH v3 2/12 bpf: Introduce BPF standard streams

In reply to Kumar Kartikeya Dwivedi

Add support for a stream API to the kernel and expose related kfuncs to BPF programs. Two streams are exposed, BPF_STDOUT and BPF_STDERR. These can be used for printing messages that can be consumed from user space, thus it's similar in spirit to existing trace_pipe interface.

The kernel will use the BPF_STDERR stream to notify the program of any errors encountered at runtime. BPF programs themselves may use both streams for writing debug messages. BPF library-like code may use BPF_STDERR to print warnings or errors on misuse at runtime.

The implementation of a stream is as follows. Everytime a message is emitted from the kernel (directly, or through a BPF program), a record is allocated by bump allocating from per-cpu region backed by a page obtained using try_alloc_pages. This ensures that we can allocate memory from any context. The eventual plan is to discard this scheme in favor of Alexei's kmalloc_nolock() [0].

This record is then locklessly inserted into a list (llist_add()) so that the printing side doesn't require holding any locks, and works in any context. Each stream has a maximum capacity of 4MB of text, and each printed message is accounted against this limit.

Messages from a program are emitted using the bpf_stream_vprintk kfunc, which takes a stream_id argument in addition to working otherwise similar to bpf_trace_vprintk.

The bprintf buffer helpers are extracted out to be reused for printing the string into them before copying it into the stream, so that we can (with the defined max limit) format a string and know its true length before performing allocations of the stream element.

For consuming elements from a stream, we expose a bpf(2) syscall command named BPF_PROG_STREAM_READ_BY_FD, which allows reading data from the stream of a given prog_fd into a user space buffer. The main logic is implemented in bpf_stream_read(). The log messages are queued in bpf_stream::log by the bpf_stream_vprintk kfunc, and then pulled and ordered correctly in the stream backlog.

For this purpose, we hold a lock around bpf_stream_backlog_peek(), as llist_del_first() (if we maintained a second lockless list for the backlog) wouldn't be safe from multiple threads anyway. Then, if we fail to find something in the backlog log, we splice out everything from the lockless log, and place it in the backlog log, and then return the head of the backlog. Once the full length of the element is consumed, we will pop it and free it.

The lockless list bpf_stream::log is a LIFO stack. Elements obtained using a llist_del_all() operation are in LIFO order, thus would break the chronological ordering if printed directly. Hence, this batch of messages is first reversed. Then, it is stashed into a separate list in the stream, i.e. the backlog_log. The head of this list is the actual message that should always be returned to the caller. All of this is done in bpf_stream_backlog_fill().

From the kernel side, the writing into the stream will be a bit more involved than the typical printk. First, the kernel typically may print a collection of messages into the stream, and parallel writers into the stream may suffer from interleaving of messages. To ensure each group of messages is visible atomically, we can lift the advantage of using a lockless list for pushing in messages.

To enable this, we add a bpf_stream_stage() macro, and require kernel users to use bpf_stream_printk statements for the passed expression to write into the stream. Underneath the macro, we have a message staging API, where a bpf_stream_stage object on the stack accumulates the messages being printed into a local llist_head, and then a commit operation splices the whole batch into the stream's lockless log list.

This is especially pertinent for rqspinlock deadlock messages printed to program streams. After this change, we see each deadlock invocation as a non-interleaving contiguous message without any confusion on the reader's part, improving their user experience in debugging the fault.

While programs cannot benefit from this staged stream writing API, they could just as well hold an rqspinlock around their print statements to serialize messages, hence this is kept kernel-internal for now.

Overall, this infrastructure provides NMI-safe any context printing of messages to two dedicated streams.

Later patches will add support for printing splats in case of BPF arena page faults, rqspinlock deadlocks, and cond_break timeouts, and integration of this facility into bpftool for dumping messages to user space.

Make sure that we don't end up spamming too many errors if the program keeps failing repeatedly and filling up the stream, hence emit at most 512 error messages from the kernel for a given stream.

[0]: https://lore.kernel.org/bpf/[email protected]

---
 include/linux/bpf.h            |  59 ++++
 include/uapi/linux/bpf.h       |  24 ++
 kernel/bpf/Makefile            |   2 +-
 kernel/bpf/core.c              |   5 +
 kernel/bpf/helpers.c           |   1 +
 kernel/bpf/stream.c            | 485 +++++++++++++++++++++++++++++++++
 kernel/bpf/syscall.c           |  27 +-
 tools/include/uapi/linux/bpf.h |  24 ++
 8 files changed, 625 insertions(+), 2 deletions(-)
 create mode 100644 kernel/bpf/stream.c
include/linux/bpf.h +59 -0
index 4fff0cee8622..cdd726cfe622 100644
@@ -1538,6 +1538,36 @@ struct btf_mod_pair {
 
 struct bpf_kfunc_desc_tab;
 
+enum bpf_stream_id {
+	BPF_STDOUT = 1,
+	BPF_STDERR = 2,
+};
+
+struct bpf_stream_elem {
+	struct llist_node node;
+	int total_len;
+	int consumed_len;
+	char str[];
+};
+
+enum {
+	BPF_STREAM_MAX_CAPACITY = (4 * 1024U * 1024U),
+};
+
+struct bpf_stream {
+	atomic_t capacity;
+	struct llist_head log;	/* list of in-flight stream elements in LIFO order */
+
+	struct mutex lock;  /* lock protecting backlog_{head,tail} */
+	struct llist_node *backlog_head; /* list of in-flight stream elements in FIFO order */
+	struct llist_node *backlog_tail; /* tail of the list above */
+};
+
+struct bpf_stream_stage {
+	struct llist_head log;
+	int len;
+};
+
 struct bpf_prog_aux {
 	atomic64_t refcnt;
 	u32 used_map_cnt;
@@ -1646,6 +1676,8 @@ struct bpf_prog_aux {
 		struct work_struct work;
 		struct rcu_head	rcu;
 	};
+	struct bpf_stream stream[2];
+	atomic_t stream_error_cnt;
 };
 
 struct bpf_prog {
@@ -2408,6 +2440,8 @@ int  generic_map_delete_batch(struct bpf_map *map,
 struct bpf_map *bpf_map_get_curr_or_next(u32 *id);
 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id);
 
+
+struct page *__bpf_alloc_page(int nid);
 int bpf_map_alloc_pages(const struct bpf_map *map, int nid,
 			unsigned long nr_pages, struct page **page_array);
 #ifdef CONFIG_MEMCG
@@ -3573,6 +3607,31 @@ void bpf_bprintf_cleanup(struct bpf_bprintf_data *data);
 int bpf_try_get_buffers(struct bpf_bprintf_buffers **bufs);
 void bpf_put_buffers(void);
 
+#define BPF_PROG_STREAM_ERROR_CNT 512
+
+void bpf_prog_stream_init(struct bpf_prog *prog);
+void bpf_prog_stream_free(struct bpf_prog *prog);
+int bpf_prog_stream_read(struct bpf_prog *prog, enum bpf_stream_id stream_id, void __user *buf, int len);
+void bpf_stream_stage_init(struct bpf_stream_stage *ss);
+void bpf_stream_stage_free(struct bpf_stream_stage *ss);
+__printf(2, 3)
+int bpf_stream_stage_printk(struct bpf_stream_stage *ss, const char *fmt, ...);
+int bpf_stream_stage_commit(struct bpf_stream_stage *ss, struct bpf_prog *prog,
+			    enum bpf_stream_id stream_id);
+
+bool bpf_prog_stream_error_limit(struct bpf_prog *prog);
+
+#define bpf_stream_printk(ss, ...) bpf_stream_stage_printk(&ss, __VA_ARGS__)
+
+#define bpf_stream_stage(ss, prog, stream_id, expr)                      \
+	({                                                               \
+		if (!bpf_prog_stream_error_limit(prog)) {                \
+			bpf_stream_stage_init(&ss);			 \
+			(expr);                                          \
+			bpf_stream_stage_commit(&ss, prog, stream_id);	 \
+			bpf_stream_stage_free(&ss);			 \
+		}                                                        \
+	})
 
 #ifdef CONFIG_BPF_LSM
 void bpf_cgroup_atype_get(u32 attach_btf_id, int cgroup_atype);
include/uapi/linux/bpf.h +24 -0
index 39e7818cca80..f2fce6a94523 100644
@@ -906,6 +906,17 @@ union bpf_iter_link_info {
  *		A new file descriptor (a nonnegative integer), or -1 if an
  *		error occurred (in which case, *errno* is set appropriately).
  *
+ * BPF_PROG_STREAM_READ_BY_FD
+ *	Description
+ *		Read data of a program's BPF stream. The program is identified
+ *		by *prog_fd*, and the stream is identified by the *stream_id*.
+ *		The data is copied to a buffer pointed to by *stream_buf*, and
+ *		filled less than or equal to *stream_buf_len* bytes.
+ *
+ *	Return
+ *		Number of bytes read from the stream on success, or -1 if an
+ *		error occurred (in which case, *errno* is set appropriately).
+ *
  * NOTES
  *	eBPF objects (maps and programs) can be shared between processes.
  *
@@ -961,6 +972,7 @@ enum bpf_cmd {
 	BPF_LINK_DETACH,
 	BPF_PROG_BIND_MAP,
 	BPF_TOKEN_CREATE,
+	BPF_PROG_STREAM_READ_BY_FD,
 	__MAX_BPF_CMD,
 };
 
@@ -1463,6 +1475,11 @@ struct bpf_stack_build_id {
 
 #define BPF_OBJ_NAME_LEN 16U
 
+enum {
+	BPF_STREAM_STDOUT = 1,
+	BPF_STREAM_STDERR = 2,
+};
+
 union bpf_attr {
 	struct { /* anonymous struct used by BPF_MAP_CREATE command */
 		__u32	map_type;	/* one of enum bpf_map_type */
@@ -1849,6 +1866,13 @@ union bpf_attr {
 		__u32		bpffs_fd;
 	} token_create;
 
+	struct {
+		__aligned_u64	stream_buf;
+		__u32		stream_buf_len;
+		__u32		stream_id;
+		__u32		prog_fd;
+	} prog_stream_read;
+
 } __attribute__((aligned(8)));
 
 /* The description below is an attempt at providing documentation to eBPF
kernel/bpf/Makefile +1 -1
index 3a335c50e6e3..269c04a24664 100644
@@ -14,7 +14,7 @@ obj-$(CONFIG_BPF_SYSCALL) += bpf_local_storage.o bpf_task_storage.o
 obj-${CONFIG_BPF_LSM}	  += bpf_inode_storage.o
 obj-$(CONFIG_BPF_SYSCALL) += disasm.o mprog.o
 obj-$(CONFIG_BPF_JIT) += trampoline.o
-obj-$(CONFIG_BPF_SYSCALL) += btf.o memalloc.o rqspinlock.o
+obj-$(CONFIG_BPF_SYSCALL) += btf.o memalloc.o rqspinlock.o stream.o
 ifeq ($(CONFIG_MMU)$(CONFIG_64BIT),yy)
 obj-$(CONFIG_BPF_SYSCALL) += arena.o range_tree.o
 endif
kernel/bpf/core.c +5 -0
index e536a34a32c8..f0def24573ae 100644
@@ -134,6 +134,10 @@ struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flag
 	mutex_init(&fp->aux->ext_mutex);
 	mutex_init(&fp->aux->dst_mutex);
 
+#ifdef CONFIG_BPF_SYSCALL
+	bpf_prog_stream_init(fp);
+#endif
+
 	return fp;
 }
 
@@ -2862,6 +2866,7 @@ static void bpf_prog_free_deferred(struct work_struct *work)
 	aux = container_of(work, struct bpf_prog_aux, work);
 #ifdef CONFIG_BPF_SYSCALL
 	bpf_free_kfunc_btf_tab(aux->kfunc_btf_tab);
+	bpf_prog_stream_free(aux->prog);
 #endif
 #ifdef CONFIG_CGROUP_BPF
 	if (aux->cgroup_atype != CGROUP_BPF_ATTACH_TYPE_INVALID)
kernel/bpf/helpers.c +1 -0
index 67d48f9fb173..8fef7b3cbd80 100644
@@ -3393,6 +3393,7 @@ BTF_ID_FLAGS(func, bpf_iter_dmabuf_next, KF_ITER_NEXT | KF_RET_NULL | KF_SLEEPAB
 BTF_ID_FLAGS(func, bpf_iter_dmabuf_destroy, KF_ITER_DESTROY | KF_SLEEPABLE)
 #endif
 BTF_ID_FLAGS(func, __bpf_trap)
+BTF_ID_FLAGS(func, bpf_stream_vprintk, KF_TRUSTED_ARGS)
 BTF_KFUNCS_END(common_btf_ids)
 
 static const struct btf_kfunc_id_set common_kfunc_set = {
kernel/bpf/stream.c new file +485 -0

485 lines not on this page · raw message

kernel/bpf/syscall.c +26 -1
index 56500381c28a..ac1010b9d11b 100644
@@ -576,7 +576,7 @@ static bool can_alloc_pages(void)
 		!IS_ENABLED(CONFIG_PREEMPT_RT);
 }
 
-static struct page *__bpf_alloc_page(int nid)
+struct page *__bpf_alloc_page(int nid)
 {
 	if (!can_alloc_pages())
 		return alloc_pages_nolock(nid, 0);
@@ -5936,6 +5936,28 @@ static int token_create(union bpf_attr *attr)
 	return bpf_token_create(attr);
 }
 
+#define BPF_PROG_STREAM_READ_BY_FD_LAST_FIELD prog_stream_read.prog_fd
+
+static int prog_stream_read(union bpf_attr *attr)
+{
+	char __user *buf = u64_to_user_ptr(attr->prog_stream_read.stream_buf);
+	u32 len = attr->prog_stream_read.stream_buf_len;
+	struct bpf_prog *prog;
+	int ret;
+
+	if (CHECK_ATTR(BPF_PROG_STREAM_READ_BY_FD))
+		return -EINVAL;
+
+	prog = bpf_prog_get(attr->prog_stream_read.prog_fd);
+	if (IS_ERR(prog))
+		return PTR_ERR(prog);
+
+	ret = bpf_prog_stream_read(prog, attr->prog_stream_read.stream_id, buf, len);
+	bpf_prog_put(prog);
+
+	return ret;
+}
+
 static int __sys_bpf(enum bpf_cmd cmd, bpfptr_t uattr, unsigned int size)
 {
 	union bpf_attr attr;
@@ -6072,6 +6094,9 @@ static int __sys_bpf(enum bpf_cmd cmd, bpfptr_t uattr, unsigned int size)
 	case BPF_TOKEN_CREATE:
 		err = token_create(&attr);
 		break;
+	case BPF_PROG_STREAM_READ_BY_FD:
+		err = prog_stream_read(&attr);
+		break;
 	default:
 		err = -EINVAL;
 		break;
tools/include/uapi/linux/bpf.h +24 -0
index 39e7818cca80..f2fce6a94523 100644
@@ -906,6 +906,17 @@ union bpf_iter_link_info {
  *		A new file descriptor (a nonnegative integer), or -1 if an
  *		error occurred (in which case, *errno* is set appropriately).
  *
+ * BPF_PROG_STREAM_READ_BY_FD
+ *	Description
+ *		Read data of a program's BPF stream. The program is identified
+ *		by *prog_fd*, and the stream is identified by the *stream_id*.
+ *		The data is copied to a buffer pointed to by *stream_buf*, and
+ *		filled less than or equal to *stream_buf_len* bytes.
+ *
+ *	Return
+ *		Number of bytes read from the stream on success, or -1 if an
+ *		error occurred (in which case, *errno* is set appropriately).
+ *
  * NOTES
  *	eBPF objects (maps and programs) can be shared between processes.
  *
@@ -961,6 +972,7 @@ enum bpf_cmd {
 	BPF_LINK_DETACH,
 	BPF_PROG_BIND_MAP,
 	BPF_TOKEN_CREATE,
+	BPF_PROG_STREAM_READ_BY_FD,
 	__MAX_BPF_CMD,
 };
 
@@ -1463,6 +1475,11 @@ struct bpf_stack_build_id {
 
 #define BPF_OBJ_NAME_LEN 16U
 
+enum {
+	BPF_STREAM_STDOUT = 1,
+	BPF_STREAM_STDERR = 2,
+};
+
 union bpf_attr {
 	struct { /* anonymous struct used by BPF_MAP_CREATE command */
 		__u32	map_type;	/* one of enum bpf_map_type */
@@ -1849,6 +1866,13 @@ union bpf_attr {
 		__u32		bpffs_fd;
 	} token_create;
 
+	struct {
+		__aligned_u64	stream_buf;
+		__u32		stream_buf_len;
+		__u32		stream_id;
+		__u32		prog_fd;
+	} prog_stream_read;
+
 } __attribute__((aligned(8)));
 
 /* The description below is an attempt at providing documentation to eBPF
Signature · 1 line
2.47.1
#3 lore raw

Re: [PATCH bpf-next v3 02/12] bpf: Introduce BPF standard streams

Quoted 9 lines · Kumar Kartikeya Dwivedi

Add support for a stream API to the kernel and expose related kfuncs to BPF programs. Two streams are exposed, BPF_STDOUT and BPF_STDERR. These can be used for printing messages that can be consumed from user space, thus it's similar in spirit to existing trace_pipe interface.

The kernel will use the BPF_STDERR stream to notify the program of any errors encountered at runtime. BPF programs themselves may use both streams for writing debug messages. BPF library-like code may use BPF_STDERR to print warnings or errors on misuse at runtime.

just curious, IIUC we can't mix the output of the streams when we dump them, right? I wonder it'd be handy to be able to get combined output and see messages from bpf programs sorted out with messages from kernel

thanks,
jirka

Quoted 894 lines

The implementation of a stream is as follows. Everytime a message is emitted from the kernel (directly, or through a BPF program), a record is allocated by bump allocating from per-cpu region backed by a page obtained using try_alloc_pages. This ensures that we can allocate memory from any context. The eventual plan is to discard this scheme in favor of Alexei's kmalloc_nolock() [0].

This record is then locklessly inserted into a list (llist_add()) so that the printing side doesn't require holding any locks, and works in any context. Each stream has a maximum capacity of 4MB of text, and each printed message is accounted against this limit.

Messages from a program are emitted using the bpf_stream_vprintk kfunc, which takes a stream_id argument in addition to working otherwise similar to bpf_trace_vprintk.

The bprintf buffer helpers are extracted out to be reused for printing the string into them before copying it into the stream, so that we can (with the defined max limit) format a string and know its true length before performing allocations of the stream element.

For consuming elements from a stream, we expose a bpf(2) syscall command named BPF_PROG_STREAM_READ_BY_FD, which allows reading data from the stream of a given prog_fd into a user space buffer. The main logic is implemented in bpf_stream_read(). The log messages are queued in bpf_stream::log by the bpf_stream_vprintk kfunc, and then pulled and ordered correctly in the stream backlog.

For this purpose, we hold a lock around bpf_stream_backlog_peek(), as llist_del_first() (if we maintained a second lockless list for the backlog) wouldn't be safe from multiple threads anyway. Then, if we fail to find something in the backlog log, we splice out everything from the lockless log, and place it in the backlog log, and then return the head of the backlog. Once the full length of the element is consumed, we will pop it and free it.

The lockless list bpf_stream::log is a LIFO stack. Elements obtained using a llist_del_all() operation are in LIFO order, thus would break the chronological ordering if printed directly. Hence, this batch of messages is first reversed. Then, it is stashed into a separate list in the stream, i.e. the backlog_log. The head of this list is the actual message that should always be returned to the caller. All of this is done in bpf_stream_backlog_fill().

From the kernel side, the writing into the stream will be a bit more involved than the typical printk. First, the kernel typically may print a collection of messages into the stream, and parallel writers into the stream may suffer from interleaving of messages. To ensure each group of messages is visible atomically, we can lift the advantage of using a lockless list for pushing in messages.

To enable this, we add a bpf_stream_stage() macro, and require kernel users to use bpf_stream_printk statements for the passed expression to write into the stream. Underneath the macro, we have a message staging API, where a bpf_stream_stage object on the stack accumulates the messages being printed into a local llist_head, and then a commit operation splices the whole batch into the stream's lockless log list.

This is especially pertinent for rqspinlock deadlock messages printed to program streams. After this change, we see each deadlock invocation as a non-interleaving contiguous message without any confusion on the reader's part, improving their user experience in debugging the fault.

While programs cannot benefit from this staged stream writing API, they could just as well hold an rqspinlock around their print statements to serialize messages, hence this is kept kernel-internal for now.

Overall, this infrastructure provides NMI-safe any context printing of messages to two dedicated streams.

Later patches will add support for printing splats in case of BPF arena page faults, rqspinlock deadlocks, and cond_break timeouts, and integration of this facility into bpftool for dumping messages to user space.

Make sure that we don't end up spamming too many errors if the program keeps failing repeatedly and filling up the stream, hence emit at most 512 error messages from the kernel for a given stream.

[0]: https://lore.kernel.org/bpf/[email protected]

---
 include/linux/bpf.h            |  59 ++++
 include/uapi/linux/bpf.h       |  24 ++
 kernel/bpf/Makefile            |   2 +-
 kernel/bpf/core.c              |   5 +
 kernel/bpf/helpers.c           |   1 +
 kernel/bpf/stream.c            | 485 +++++++++++++++++++++++++++++++++
 kernel/bpf/syscall.c           |  27 +-
 tools/include/uapi/linux/bpf.h |  24 ++
 8 files changed, 625 insertions(+), 2 deletions(-)
 create mode 100644 kernel/bpf/stream.c
include/linux/bpf.h +59 -0
index 4fff0cee8622..cdd726cfe622 100644
@@ -1538,6 +1538,36 @@ struct btf_mod_pair {
 
 struct bpf_kfunc_desc_tab;
 
+enum bpf_stream_id {
+	BPF_STDOUT = 1,
+	BPF_STDERR = 2,
+};
+
+struct bpf_stream_elem {
+	struct llist_node node;
+	int total_len;
+	int consumed_len;
+	char str[];
+};
+
+enum {
+	BPF_STREAM_MAX_CAPACITY = (4 * 1024U * 1024U),
+};
+
+struct bpf_stream {
+	atomic_t capacity;
+	struct llist_head log;	/* list of in-flight stream elements in LIFO order */
+
+	struct mutex lock;  /* lock protecting backlog_{head,tail} */
+	struct llist_node *backlog_head; /* list of in-flight stream elements in FIFO order */
+	struct llist_node *backlog_tail; /* tail of the list above */
+};
+
+struct bpf_stream_stage {
+	struct llist_head log;
+	int len;
+};
+
 struct bpf_prog_aux {
 	atomic64_t refcnt;
 	u32 used_map_cnt;
@@ -1646,6 +1676,8 @@ struct bpf_prog_aux {
 		struct work_struct work;
 		struct rcu_head	rcu;
 	};
+	struct bpf_stream stream[2];
+	atomic_t stream_error_cnt;
 };
 
 struct bpf_prog {
@@ -2408,6 +2440,8 @@ int  generic_map_delete_batch(struct bpf_map *map,
 struct bpf_map *bpf_map_get_curr_or_next(u32 *id);
 struct bpf_prog *bpf_prog_get_curr_or_next(u32 *id);
 
+
+struct page *__bpf_alloc_page(int nid);
 int bpf_map_alloc_pages(const struct bpf_map *map, int nid,
 			unsigned long nr_pages, struct page **page_array);
 #ifdef CONFIG_MEMCG
@@ -3573,6 +3607,31 @@ void bpf_bprintf_cleanup(struct bpf_bprintf_data *data);
 int bpf_try_get_buffers(struct bpf_bprintf_buffers **bufs);
 void bpf_put_buffers(void);
 
+#define BPF_PROG_STREAM_ERROR_CNT 512
+
+void bpf_prog_stream_init(struct bpf_prog *prog);
+void bpf_prog_stream_free(struct bpf_prog *prog);
+int bpf_prog_stream_read(struct bpf_prog *prog, enum bpf_stream_id stream_id, void __user *buf, int len);
+void bpf_stream_stage_init(struct bpf_stream_stage *ss);
+void bpf_stream_stage_free(struct bpf_stream_stage *ss);
+__printf(2, 3)
+int bpf_stream_stage_printk(struct bpf_stream_stage *ss, const char *fmt, ...);
+int bpf_stream_stage_commit(struct bpf_stream_stage *ss, struct bpf_prog *prog,
+			    enum bpf_stream_id stream_id);
+
+bool bpf_prog_stream_error_limit(struct bpf_prog *prog);
+
+#define bpf_stream_printk(ss, ...) bpf_stream_stage_printk(&ss, __VA_ARGS__)
+
+#define bpf_stream_stage(ss, prog, stream_id, expr)                      \
+	({                                                               \
+		if (!bpf_prog_stream_error_limit(prog)) {                \
+			bpf_stream_stage_init(&ss);			 \
+			(expr);                                          \
+			bpf_stream_stage_commit(&ss, prog, stream_id);	 \
+			bpf_stream_stage_free(&ss);			 \
+		}                                                        \
+	})
 
 #ifdef CONFIG_BPF_LSM
 void bpf_cgroup_atype_get(u32 attach_btf_id, int cgroup_atype);
include/uapi/linux/bpf.h +24 -0
index 39e7818cca80..f2fce6a94523 100644
@@ -906,6 +906,17 @@ union bpf_iter_link_info {
  *		A new file descriptor (a nonnegative integer), or -1 if an
  *		error occurred (in which case, *errno* is set appropriately).
  *
+ * BPF_PROG_STREAM_READ_BY_FD
+ *	Description
+ *		Read data of a program's BPF stream. The program is identified
+ *		by *prog_fd*, and the stream is identified by the *stream_id*.
+ *		The data is copied to a buffer pointed to by *stream_buf*, and
+ *		filled less than or equal to *stream_buf_len* bytes.
+ *
+ *	Return
+ *		Number of bytes read from the stream on success, or -1 if an
+ *		error occurred (in which case, *errno* is set appropriately).
+ *
  * NOTES
  *	eBPF objects (maps and programs) can be shared between processes.
  *
@@ -961,6 +972,7 @@ enum bpf_cmd {
 	BPF_LINK_DETACH,
 	BPF_PROG_BIND_MAP,
 	BPF_TOKEN_CREATE,
+	BPF_PROG_STREAM_READ_BY_FD,
 	__MAX_BPF_CMD,
 };
 
@@ -1463,6 +1475,11 @@ struct bpf_stack_build_id {
 
 #define BPF_OBJ_NAME_LEN 16U
 
+enum {
+	BPF_STREAM_STDOUT = 1,
+	BPF_STREAM_STDERR = 2,
+};
+
 union bpf_attr {
 	struct { /* anonymous struct used by BPF_MAP_CREATE command */
 		__u32	map_type;	/* one of enum bpf_map_type */
@@ -1849,6 +1866,13 @@ union bpf_attr {
 		__u32		bpffs_fd;
 	} token_create;
 
+	struct {
+		__aligned_u64	stream_buf;
+		__u32		stream_buf_len;
+		__u32		stream_id;
+		__u32		prog_fd;
+	} prog_stream_read;
+
 } __attribute__((aligned(8)));
 
 /* The description below is an attempt at providing documentation to eBPF
kernel/bpf/Makefile +1 -1
index 3a335c50e6e3..269c04a24664 100644
@@ -14,7 +14,7 @@ obj-$(CONFIG_BPF_SYSCALL) += bpf_local_storage.o bpf_task_storage.o
 obj-${CONFIG_BPF_LSM}	  += bpf_inode_storage.o
 obj-$(CONFIG_BPF_SYSCALL) += disasm.o mprog.o
 obj-$(CONFIG_BPF_JIT) += trampoline.o
-obj-$(CONFIG_BPF_SYSCALL) += btf.o memalloc.o rqspinlock.o
+obj-$(CONFIG_BPF_SYSCALL) += btf.o memalloc.o rqspinlock.o stream.o
 ifeq ($(CONFIG_MMU)$(CONFIG_64BIT),yy)
 obj-$(CONFIG_BPF_SYSCALL) += arena.o range_tree.o
 endif
kernel/bpf/core.c +5 -0
index e536a34a32c8..f0def24573ae 100644
@@ -134,6 +134,10 @@ struct bpf_prog *bpf_prog_alloc_no_stats(unsigned int size, gfp_t gfp_extra_flag
 	mutex_init(&fp->aux->ext_mutex);
 	mutex_init(&fp->aux->dst_mutex);
 
+#ifdef CONFIG_BPF_SYSCALL
+	bpf_prog_stream_init(fp);
+#endif
+
 	return fp;
 }
 
@@ -2862,6 +2866,7 @@ static void bpf_prog_free_deferred(struct work_struct *work)
 	aux = container_of(work, struct bpf_prog_aux, work);
 #ifdef CONFIG_BPF_SYSCALL
 	bpf_free_kfunc_btf_tab(aux->kfunc_btf_tab);
+	bpf_prog_stream_free(aux->prog);
 #endif
 #ifdef CONFIG_CGROUP_BPF
 	if (aux->cgroup_atype != CGROUP_BPF_ATTACH_TYPE_INVALID)
kernel/bpf/helpers.c +1 -0
index 67d48f9fb173..8fef7b3cbd80 100644
@@ -3393,6 +3393,7 @@ BTF_ID_FLAGS(func, bpf_iter_dmabuf_next, KF_ITER_NEXT | KF_RET_NULL | KF_SLEEPAB
 BTF_ID_FLAGS(func, bpf_iter_dmabuf_destroy, KF_ITER_DESTROY | KF_SLEEPABLE)
 #endif
 BTF_ID_FLAGS(func, __bpf_trap)
+BTF_ID_FLAGS(func, bpf_stream_vprintk, KF_TRUSTED_ARGS)
 BTF_KFUNCS_END(common_btf_ids)
 
 static const struct btf_kfunc_id_set common_kfunc_set = {
kernel/bpf/stream.c new file +485 -0
new file mode 100644
index 000000000000..75ceb6379368
@@ -0,0 +1,485 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/* Copyright (c) 2025 Meta Platforms, Inc. and affiliates. */
+
+#include <linux/bpf.h>
+#include <linux/bpf_mem_alloc.h>
+#include <linux/percpu.h>
+#include <linux/refcount.h>
+#include <linux/gfp.h>
+#include <linux/memory.h>
+#include <linux/local_lock.h>
+#include <linux/mutex.h>
+
+/*
+ * Simple per-CPU NMI-safe bump allocation mechanism, backed by the NMI-safe
+ * try_alloc_pages()/free_pages_nolock() primitives. We allocate a page and
+ * stash it in a local per-CPU variable, and bump allocate from the page
+ * whenever items need to be printed to a stream. Each page holds a global
+ * atomic refcount in its first 4 bytes, and then records of variable length
+ * that describe the printed messages. Once the global refcount has dropped to
+ * zero, it is a signal to free the page back to the kernel's page allocator,
+ * given all the individual records in it have been consumed.
+ *
+ * It is possible the same page is used to serve allocations across different
+ * programs, which may be consumed at different times individually, hence
+ * maintaining a reference count per-page is critical for correct lifetime
+ * tracking.
+ *
+ * The bpf_stream_page code will be replaced to use kmalloc_nolock() once it
+ * lands.
+ */
+struct bpf_stream_page {
+	refcount_t ref;
+	u32 consumed;
+	char buf[];
+};
+
+/* Available room to add data to a refcounted page. */
+#define BPF_STREAM_PAGE_SZ (PAGE_SIZE - offsetofend(struct bpf_stream_page, consumed))
+
+static DEFINE_PER_CPU(local_trylock_t, stream_local_lock) = INIT_LOCAL_TRYLOCK(stream_local_lock);
+static DEFINE_PER_CPU(struct bpf_stream_page *, stream_pcpu_page);
+
+static bool bpf_stream_page_local_lock(unsigned long *flags)
+{
+	return local_trylock_irqsave(&stream_local_lock, *flags);
+}
+
+static void bpf_stream_page_local_unlock(unsigned long *flags)
+{
+	local_unlock_irqrestore(&stream_local_lock, *flags);
+}
+
+static void bpf_stream_page_free(struct bpf_stream_page *stream_page)
+{
+	struct page *p;
+
+	if (!stream_page)
+		return;
+	p = virt_to_page(stream_page);
+	free_pages_nolock(p, 0);
+}
+
+static void bpf_stream_page_get(struct bpf_stream_page *stream_page)
+{
+	refcount_inc(&stream_page->ref);
+}
+
+static void bpf_stream_page_put(struct bpf_stream_page *stream_page)
+{
+	if (refcount_dec_and_test(&stream_page->ref))
+		bpf_stream_page_free(stream_page);
+}
+
+static void bpf_stream_page_init(struct bpf_stream_page *stream_page)
+{
+	refcount_set(&stream_page->ref, 1);
+	stream_page->consumed = 0;
+}
+
+static struct bpf_stream_page *bpf_stream_page_replace(void)
+{
+	struct bpf_stream_page *stream_page, *old_stream_page;
+	struct page *page;
+
+	page = __bpf_alloc_page(NUMA_NO_NODE);
+	if (!page)
+		return NULL;
+	stream_page = page_address(page);
+	bpf_stream_page_init(stream_page);
+
+	old_stream_page = this_cpu_read(stream_pcpu_page);
+	if (old_stream_page)
+		bpf_stream_page_put(old_stream_page);
+	this_cpu_write(stream_pcpu_page, stream_page);
+	return stream_page;
+}
+
+static int bpf_stream_page_check_room(struct bpf_stream_page *stream_page, int len)
+{
+	int min = offsetof(struct bpf_stream_elem, str[0]);
+	int consumed = stream_page->consumed;
+	int total = BPF_STREAM_PAGE_SZ;
+	int rem = max(0, total - consumed - min);
+
+	/* Let's give room of at least 8 bytes. */
+	WARN_ON_ONCE(rem % 8 != 0);
+	rem = rem < 8 ? 0 : rem;
+	return min(len, rem);
+}
+
+static void bpf_stream_elem_init(struct bpf_stream_elem *elem, int len)
+{
+	init_llist_node(&elem->node);
+	elem->total_len = len;
+	elem->consumed_len = 0;
+}
+
+static struct bpf_stream_page *bpf_stream_page_from_elem(struct bpf_stream_elem *elem)
+{
+	unsigned long addr = (unsigned long)elem;
+
+	return (struct bpf_stream_page *)PAGE_ALIGN_DOWN(addr);
+}
+
+static struct bpf_stream_elem *bpf_stream_page_push_elem(struct bpf_stream_page *stream_page, int len)
+{
+	u32 consumed = stream_page->consumed;
+
+	stream_page->consumed += round_up(offsetof(struct bpf_stream_elem, str[len]), 8);
+	return (struct bpf_stream_elem *)&stream_page->buf[consumed];
+}
+
+static noinline struct bpf_stream_elem *bpf_stream_page_reserve_elem(int len)
+{
+	struct bpf_stream_elem *elem = NULL;
+	struct bpf_stream_page *page;
+	int room = 0;
+
+	page = this_cpu_read(stream_pcpu_page);
+	if (!page)
+		page = bpf_stream_page_replace();
+	if (!page)
+		return NULL;
+
+	room = bpf_stream_page_check_room(page, len);
+	if (room != len)
+		page = bpf_stream_page_replace();
+	if (!page)
+		return NULL;
+	bpf_stream_page_get(page);
+	room = bpf_stream_page_check_room(page, len);
+	WARN_ON_ONCE(room != len);
+
+	elem = bpf_stream_page_push_elem(page, room);
+	bpf_stream_elem_init(elem, room);
+	return elem;
+}
+
+static struct bpf_stream_elem *bpf_stream_elem_alloc(int len)
+{
+	const int max_len = ARRAY_SIZE((struct bpf_bprintf_buffers){}.buf);
+	struct bpf_stream_elem *elem;
+	unsigned long flags;
+
+	BUILD_BUG_ON(max_len > BPF_STREAM_PAGE_SZ);
+	/*
+	 * Length denotes the amount of data to be written as part of stream element,
+	 * thus includes '\0' byte. We're capped by how much bpf_bprintf_buffers can
+	 * accomodate, therefore deny allocations that won't fit into them.
+	 */
+	if (len < 0 || len > max_len)
+		return NULL;
+
+	if (!bpf_stream_page_local_lock(&flags))
+		return NULL;
+	elem = bpf_stream_page_reserve_elem(len);
+	bpf_stream_page_local_unlock(&flags);
+	return elem;
+}
+
+static int __bpf_stream_push_str(struct llist_head *log, const char *str, int len)
+{
+	struct bpf_stream_elem *elem = NULL;
+
+	/*
+	 * Allocate a bpf_prog_stream_elem and push it to the bpf_prog_stream
+	 * log, elements will be popped at once and reversed to print the log.
+	 */
+	elem = bpf_stream_elem_alloc(len);
+	if (!elem)
+		return -ENOMEM;
+
+	memcpy(elem->str, str, len);
+	llist_add(&elem->node, log);
+
+	return 0;
+}
+
+static int bpf_stream_consume_capacity(struct bpf_stream *stream, int len)
+{
+	if (atomic_read(&stream->capacity) >= BPF_STREAM_MAX_CAPACITY)
+		return -ENOSPC;
+	if (atomic_add_return(len, &stream->capacity) >= BPF_STREAM_MAX_CAPACITY) {
+		atomic_sub(len, &stream->capacity);
+		return -ENOSPC;
+	}
+	return 0;
+}
+
+static void bpf_stream_release_capacity(struct bpf_stream *stream, struct bpf_stream_elem *elem)
+{
+	int len = elem->total_len;
+
+	atomic_sub(len, &stream->capacity);
+}
+
+static int bpf_stream_push_str(struct bpf_stream *stream, const char *str, int len)
+{
+	int ret = bpf_stream_consume_capacity(stream, len);
+
+	return ret ?: __bpf_stream_push_str(&stream->log, str, len);
+}
+
+static struct bpf_stream *bpf_stream_get(enum bpf_stream_id stream_id, struct bpf_prog_aux *aux)
+{
+	if (stream_id != BPF_STDOUT && stream_id != BPF_STDERR)
+		return NULL;
+	return &aux->stream[stream_id - 1];
+}
+
+static void bpf_stream_free_elem(struct bpf_stream_elem *elem)
+{
+	struct bpf_stream_page *p;
+
+	p = bpf_stream_page_from_elem(elem);
+	bpf_stream_page_put(p);
+}
+
+static void bpf_stream_free_list(struct llist_node *list)
+{
+	struct bpf_stream_elem *elem, *tmp;
+
+	llist_for_each_entry_safe(elem, tmp, list, node)
+		bpf_stream_free_elem(elem);
+}
+
+static struct llist_node *bpf_stream_backlog_peek(struct bpf_stream *stream)
+{
+	return stream->backlog_head;
+}
+
+static struct llist_node *bpf_stream_backlog_pop(struct bpf_stream *stream)
+{
+	struct llist_node *node;
+
+	node = stream->backlog_head;
+	if (stream->backlog_head == stream->backlog_tail)
+		stream->backlog_head = stream->backlog_tail = NULL;
+	else
+		stream->backlog_head = node->next;
+	return node;
+}
+
+static void bpf_stream_backlog_fill(struct bpf_stream *stream)
+{
+	struct llist_node *head, *tail;
+
+	if (llist_empty(&stream->log))
+		return;
+	tail = llist_del_all(&stream->log);
+	if (!tail)
+		return;
+	head = llist_reverse_order(tail);
+
+	if (!stream->backlog_head) {
+		stream->backlog_head = head;
+		stream->backlog_tail = tail;
+	} else {
+		stream->backlog_tail->next = head;
+		stream->backlog_tail = tail;
+	}
+
+	return;
+}
+
+static bool bpf_stream_consume_elem(struct bpf_stream_elem *elem, int *len)
+{
+	int rem = elem->total_len - elem->consumed_len;
+	int used = min(rem, *len);
+
+	elem->consumed_len += used;
+	*len -= used;
+
+	return elem->consumed_len == elem->total_len;
+}
+
+static int bpf_stream_read(struct bpf_stream *stream, void __user *buf, int len)
+{
+	int rem_len = len, cons_len, ret = 0;
+	struct bpf_stream_elem *elem = NULL;
+	struct llist_node *node;
+
+	mutex_lock(&stream->lock);
+
+	while (rem_len) {
+		int pos = len - rem_len;
+		bool cont;
+
+		node = bpf_stream_backlog_peek(stream);
+		if (!node) {
+			bpf_stream_backlog_fill(stream);
+			node = bpf_stream_backlog_peek(stream);
+		}
+		if (!node)
+			break;
+		elem = container_of(node, typeof(*elem), node);
+
+		cons_len = elem->consumed_len;
+		cont = bpf_stream_consume_elem(elem, &rem_len) == false;
+
+		ret = copy_to_user(buf + pos, elem->str + cons_len,
+				   elem->consumed_len - cons_len);
+		/* Restore in case of error. */
+		if (ret) {
+			ret = -EFAULT;
+			elem->consumed_len = cons_len;
+			break;
+		}
+
+		if (cont)
+			continue;
+		bpf_stream_backlog_pop(stream);
+		bpf_stream_release_capacity(stream, elem);
+		bpf_stream_free_elem(elem);
+	}
+
+	mutex_unlock(&stream->lock);
+	return ret ? ret : len - rem_len;
+}
+
+int bpf_prog_stream_read(struct bpf_prog *prog, enum bpf_stream_id stream_id, void __user *buf, int len)
+{
+	struct bpf_stream *stream;
+
+	stream = bpf_stream_get(stream_id, prog->aux);
+	if (!stream)
+		return -ENOENT;
+	return bpf_stream_read(stream, buf, len);
+}
+
+__bpf_kfunc_start_defs();
+
+/*
+ * Avoid using enum bpf_stream_id so that kfunc users don't have to pull in the
+ * enum in headers.
+ */
+__bpf_kfunc int bpf_stream_vprintk(int stream_id, const char *fmt__str, const void *args, u32 len__sz, void *aux__prog)
+{
+	struct bpf_bprintf_data data = {
+		.get_bin_args	= true,
+		.get_buf	= true,
+	};
+	struct bpf_prog_aux *aux = aux__prog;
+	u32 fmt_size = strlen(fmt__str) + 1;
+	struct bpf_stream *stream;
+	u32 data_len = len__sz;
+	int ret, num_args;
+
+	stream = bpf_stream_get(stream_id, aux);
+	if (!stream)
+		return -ENOENT;
+
+	if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
+	    (data_len && !args))
+		return -EINVAL;
+	num_args = data_len / 8;
+
+	ret = bpf_bprintf_prepare(fmt__str, fmt_size, args, num_args, &data);
+	if (ret < 0)
+		return ret;
+
+	ret = bstr_printf(data.buf, MAX_BPRINTF_BUF, fmt__str, data.bin_args);
+	/* If the string was truncated, we only wrote until the size of buffer. */
+	ret = min_t(u32, ret + 1, MAX_BPRINTF_BUF);
+	ret = bpf_stream_push_str(stream, data.buf, ret);
+	bpf_bprintf_cleanup(&data);
+
+	return ret;
+}
+
+__bpf_kfunc_end_defs();
+
+/* Added kfunc to common_btf_ids */
+
+void bpf_prog_stream_init(struct bpf_prog *prog)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(prog->aux->stream); i++) {
+		atomic_set(&prog->aux->stream[i].capacity, 0);
+		init_llist_head(&prog->aux->stream[i].log);
+		mutex_init(&prog->aux->stream[i].lock);
+		prog->aux->stream[i].backlog_head = NULL;
+		prog->aux->stream[i].backlog_tail = NULL;
+	}
+}
+
+void bpf_prog_stream_free(struct bpf_prog *prog)
+{
+	struct llist_node *list;
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(prog->aux->stream); i++) {
+		list = llist_del_all(&prog->aux->stream[i].log);
+		bpf_stream_free_list(list);
+		bpf_stream_free_list(prog->aux->stream[i].backlog_head);
+	}
+}
+
+void bpf_stream_stage_init(struct bpf_stream_stage *ss)
+{
+	init_llist_head(&ss->log);
+	ss->len = 0;
+}
+
+void bpf_stream_stage_free(struct bpf_stream_stage *ss)
+{
+	struct llist_node *node;
+
+	node = llist_del_all(&ss->log);
+	bpf_stream_free_list(node);
+}
+
+int bpf_stream_stage_printk(struct bpf_stream_stage *ss, const char *fmt, ...)
+{
+	struct bpf_bprintf_buffers *buf;
+	va_list args;
+	int ret;
+
+	if (bpf_try_get_buffers(&buf))
+		return -EBUSY;
+
+	va_start(args, fmt);
+	ret = vsnprintf(buf->buf, ARRAY_SIZE(buf->buf), fmt, args);
+	va_end(args);
+	/* If the string was truncated, we only wrote until the size of buffer. */
+	ret = min_t(u32, ret + 1, ARRAY_SIZE(buf->buf));
+	ss->len += ret;
+	ret = __bpf_stream_push_str(&ss->log, buf->buf, ret);
+	bpf_put_buffers();
+	return ret;
+}
+
+int bpf_stream_stage_commit(struct bpf_stream_stage *ss, struct bpf_prog *prog,
+			    enum bpf_stream_id stream_id)
+{
+	struct llist_node *list, *head, *tail;
+	struct bpf_stream *stream;
+	int ret;
+
+	stream = bpf_stream_get(stream_id, prog->aux);
+	if (!stream)
+		return -EINVAL;
+
+	ret = bpf_stream_consume_capacity(stream, ss->len);
+	if (ret)
+		return ret;
+
+	list = llist_del_all(&ss->log);
+	head = tail = list;
+
+	if (!list)
+		return 0;
+	while (llist_next(list)) {
+		tail = llist_next(list);
+		list = tail;
+	}
+	llist_add_batch(head, tail, &stream->log);
+	return 0;
+}
+
+bool bpf_prog_stream_error_limit(struct bpf_prog *prog)
+{
+	return atomic_fetch_add(1, &prog->aux->stream_error_cnt) >= BPF_PROG_STREAM_ERROR_CNT;
+}
kernel/bpf/syscall.c +26 -1
index 56500381c28a..ac1010b9d11b 100644
@@ -576,7 +576,7 @@ static bool can_alloc_pages(void)
 		!IS_ENABLED(CONFIG_PREEMPT_RT);
 }
 
-static struct page *__bpf_alloc_page(int nid)
+struct page *__bpf_alloc_page(int nid)
 {
 	if (!can_alloc_pages())
 		return alloc_pages_nolock(nid, 0);
@@ -5936,6 +5936,28 @@ static int token_create(union bpf_attr *attr)
 	return bpf_token_create(attr);
 }
 
+#define BPF_PROG_STREAM_READ_BY_FD_LAST_FIELD prog_stream_read.prog_fd
+
+static int prog_stream_read(union bpf_attr *attr)
+{
+	char __user *buf = u64_to_user_ptr(attr->prog_stream_read.stream_buf);
+	u32 len = attr->prog_stream_read.stream_buf_len;
+	struct bpf_prog *prog;
+	int ret;
+
+	if (CHECK_ATTR(BPF_PROG_STREAM_READ_BY_FD))
+		return -EINVAL;
+
+	prog = bpf_prog_get(attr->prog_stream_read.prog_fd);
+	if (IS_ERR(prog))
+		return PTR_ERR(prog);
+
+	ret = bpf_prog_stream_read(prog, attr->prog_stream_read.stream_id, buf, len);
+	bpf_prog_put(prog);
+
+	return ret;
+}
+
 static int __sys_bpf(enum bpf_cmd cmd, bpfptr_t uattr, unsigned int size)
 {
 	union bpf_attr attr;
@@ -6072,6 +6094,9 @@ static int __sys_bpf(enum bpf_cmd cmd, bpfptr_t uattr, unsigned int size)
 	case BPF_TOKEN_CREATE:
 		err = token_create(&attr);
 		break;
+	case BPF_PROG_STREAM_READ_BY_FD:
+		err = prog_stream_read(&attr);
+		break;
 	default:
 		err = -EINVAL;
 		break;
tools/include/uapi/linux/bpf.h +24 -0
index 39e7818cca80..f2fce6a94523 100644
@@ -906,6 +906,17 @@ union bpf_iter_link_info {
  *		A new file descriptor (a nonnegative integer), or -1 if an
  *		error occurred (in which case, *errno* is set appropriately).
  *
+ * BPF_PROG_STREAM_READ_BY_FD
+ *	Description
+ *		Read data of a program's BPF stream. The program is identified
+ *		by *prog_fd*, and the stream is identified by the *stream_id*.
+ *		The data is copied to a buffer pointed to by *stream_buf*, and
+ *		filled less than or equal to *stream_buf_len* bytes.
+ *
+ *	Return
+ *		Number of bytes read from the stream on success, or -1 if an
+ *		error occurred (in which case, *errno* is set appropriately).
+ *
  * NOTES
  *	eBPF objects (maps and programs) can be shared between processes.
  *
@@ -961,6 +972,7 @@ enum bpf_cmd {
 	BPF_LINK_DETACH,
 	BPF_PROG_BIND_MAP,
 	BPF_TOKEN_CREATE,
+	BPF_PROG_STREAM_READ_BY_FD,
 	__MAX_BPF_CMD,
 };
 
@@ -1463,6 +1475,11 @@ struct bpf_stack_build_id {
 
 #define BPF_OBJ_NAME_LEN 16U
 
+enum {
+	BPF_STREAM_STDOUT = 1,
+	BPF_STREAM_STDERR = 2,
+};
+
 union bpf_attr {
 	struct { /* anonymous struct used by BPF_MAP_CREATE command */
 		__u32	map_type;	/* one of enum bpf_map_type */
@@ -1849,6 +1866,13 @@ union bpf_attr {
 		__u32		bpffs_fd;
 	} token_create;
 
+	struct {
+		__aligned_u64	stream_buf;
+		__u32		stream_buf_len;
+		__u32		stream_id;
+		__u32		prog_fd;
+	} prog_stream_read;
+
 } __attribute__((aligned(8)));
 
 /* The description below is an attempt at providing documentation to eBPF
Signature · 1 line
2.47.1
author maintainer
#4 lore raw

Re: [PATCH bpf-next v3 02/12] bpf: Introduce BPF standard streams

Quoted 7 lines · Jiri Olsa
Quoted 5 lines · Kumar Kartikeya Dwivedi

Add support for a stream API to the kernel and expose related kfuncs to BPF programs. Two streams are exposed, BPF_STDOUT and BPF_STDERR. These can be used for printing messages that can be consumed from user space, thus it's similar in spirit to existing trace_pipe interface.

Quoted 9 lines
Quoted 4 lines

The kernel will use the BPF_STDERR stream to notify the program of any errors encountered at runtime. BPF programs themselves may use both streams for writing debug messages. BPF library-like code may use BPF_STDERR to print warnings or errors on misuse at runtime.

just curious, IIUC we can't mix the output of the streams when we dump them, right? I wonder it'd be handy to be able to get combined output and see messages from bpf programs sorted out with messages from kernel

Yeah, this is a good point.
Right now, no, in the sense that sequentiality is definitely broken across the two streams.
We can force print a timestamp for every message and do the sorting from bpftool side, or it can just be piped to sort after dumping both stdout and stderr.
Output will look like trace_pipe with some fixed format before the actual message.
WDYT? Others are also welcome to chime in.

Quoted 5 lines

thanks,
jirka

Quoted 1 line

[...]

#5 lore raw

Re: [PATCH bpf-next v3 02/12] bpf: Introduce BPF standard streams

Quoted 18 lines · Kumar Kartikeya Dwivedi
Quoted 7 lines · Jiri Olsa
Quoted 5 lines · Kumar Kartikeya Dwivedi

Add support for a stream API to the kernel and expose related kfuncs to BPF programs. Two streams are exposed, BPF_STDOUT and BPF_STDERR. These can be used for printing messages that can be consumed from user space, thus it's similar in spirit to existing trace_pipe interface.

Quoted 9 lines
Quoted 4 lines

The kernel will use the BPF_STDERR stream to notify the program of any errors encountered at runtime. BPF programs themselves may use both streams for writing debug messages. BPF library-like code may use BPF_STDERR to print warnings or errors on misuse at runtime.

just curious, IIUC we can't mix the output of the streams when we dump them, right? I wonder it'd be handy to be able to get combined output and see messages from bpf programs sorted out with messages from kernel

Quoted 9 lines

Yeah, this is a good point.
Right now, no, in the sense that sequentiality is definitely broken across the two streams.
We can force print a timestamp for every message and do the sorting from bpftool side, or it can just be piped to sort after dumping both stdout and stderr.
Output will look like trace_pipe with some fixed format before the actual message.
WDYT? Others are also welcome to chime in.

yes, keeping the kernel simple (just adding timestamp) and sorting it in bpftool seems good to me

jirka

#6 lore raw

Re: [PATCH bpf-next v3 02/12] bpf: Introduce BPF standard streams

Quoted 30 lines · Jiri Olsa
Quoted 18 lines · Kumar Kartikeya Dwivedi
Quoted 7 lines · Jiri Olsa
Quoted 5 lines · Kumar Kartikeya Dwivedi

Add support for a stream API to the kernel and expose related kfuncs to BPF programs. Two streams are exposed, BPF_STDOUT and BPF_STDERR. These can be used for printing messages that can be consumed from user space, thus it's similar in spirit to existing trace_pipe interface.

Quoted 9 lines
Quoted 4 lines

The kernel will use the BPF_STDERR stream to notify the program of any errors encountered at runtime. BPF programs themselves may use both streams for writing debug messages. BPF library-like code may use BPF_STDERR to print warnings or errors on misuse at runtime.

just curious, IIUC we can't mix the output of the streams when we dump them, right? I wonder it'd be handy to be able to get combined output and see messages from bpf programs sorted out with messages from kernel

Quoted 9 lines

Yeah, this is a good point.
Right now, no, in the sense that sequentiality is definitely broken across the two streams.
We can force print a timestamp for every message and do the sorting from bpftool side, or it can just be piped to sort after dumping both stdout and stderr.
Output will look like trace_pipe with some fixed format before the actual message.
WDYT? Others are also welcome to chime in.

Quoted 2 lines

yes, keeping the kernel simple (just adding timestamp) and sorting it in bpftool seems good to me

I don't see the point in all that complication. If bpf prog wants to separate its printk vs kernel printks it should use a different stream.
Right now there is only stdout and stderr. Eventually we will allow for more.

#7 lore raw

Re: [PATCH bpf-next v3 02/12] bpf: Introduce BPF standard streams

In reply to Kumar Kartikeya Dwivedi

kernel/bpf/stream.c line 87

Quoted 10 lines · Kumar Kartikeya Dwivedi
kernel/bpf/stream.c new file +9 -0
+
+static struct bpf_stream_page *bpf_stream_page_replace(void)
+{
+       struct bpf_stream_page *stream_page, *old_stream_page;
+       struct page *page;
+
+       page = __bpf_alloc_page(NUMA_NO_NODE);
+       if (!page)
+               return NULL;

__bpf_alloc_page() is using GFP_ACCOUNT in both nolock and normal cases, but active_memcg is random at this point, so the page accounting is incorrect.
I think we need to remember objcg in prog_aux similar to what we do with maps, and then store that in bpf_stream and stream_stage objects in corresponding init() functions.
Then do set_active_memcg() here the way we do in bpf_mem_alloc and in map_*alloc()s.

Or use alloc_page_nolock() directly here without GFP_ACCOUNT.

Also I think it's strange to limit kernel messages to 4M, since the kernel messages are essential debug info. While it makes sense to limit kfunc's spam.
I suspect the idea here is to avoid OOM if the kernel is spammy due to malicious bpf prog that forces the kernel to warn so much? But dmesg doesn't do it.
And 4M * 2 * number_of_progs can be many gigabytes.

Maybe let's drop stream->capacity and rely on memcg to limit the spam for both kernel and kfunc?
Accounting a page at a time seems sufficient.

kernel/bpf/stream.c line 385

Quoted 4 lines
kernel/bpf/stream.c new file +4 -0
+       ret = bstr_printf(data.buf, MAX_BPRINTF_BUF, fmt__str, data.bin_args);
+       /* If the string was truncated, we only wrote until the size of buffer. */
+       ret = min_t(u32, ret + 1, MAX_BPRINTF_BUF);
+       ret = bpf_stream_push_str(stream, data.buf, ret);

We discussed it offline, so mentioning here for the list. Let's not emit \0 into the stream. Looks unnecessary.

author maintainer
#8 lore raw

PATCH v3 3/12 bpf: Add function to extract program source info

In reply to Kumar Kartikeya Dwivedi

Prepare a function for use in future patches that can extract the file info, line info, and the source line number for a given BPF program provided it's program counter.

Only the basename of the file path is provided, given it can be excessively long in some cases.

This will be used in later patches to print source info to the BPF stream.

---
 include/linux/bpf.h |  3 +++
 kernel/bpf/core.c   | 47 +++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 50 insertions(+)
include/linux/bpf.h +3 -0
index cdd726cfe622..f30697c72ba9 100644
@@ -3667,4 +3667,7 @@ static inline bool bpf_is_subprog(const struct bpf_prog *prog)
 	return prog->aux->func_idx != 0;
 }
 
+int bpf_prog_get_file_line(struct bpf_prog *prog, unsigned long ip, const char **filep,
+			   const char **linep, int *nump);
+
 #endif /* _LINUX_BPF_H */
kernel/bpf/core.c +47 -0

50 lines not on this page · raw message

Signature · 1 line
2.47.1
author maintainer
#9 lore raw

PATCH v3 4/12 bpf: Ensure RCU lock is held around bpf_prog_ksym_find

In reply to Kumar Kartikeya Dwivedi

Add a warning to ensure RCU lock is held around tree lookup, and then fix one of the invocations in bpf_stack_walker. The program has an active stack frame and won't disappear.

Fixes f18b03fabaa9 ("bpf: Implement BPF exceptions")
---
 kernel/bpf/core.c    | 5 ++++-
 kernel/bpf/helpers.c | 2 ++
 2 files changed, 6 insertions(+), 1 deletion(-)
kernel/bpf/core.c +4 -1
index 5c6e9fbb5508..b4203f68cf33 100644
@@ -782,7 +782,10 @@ bool is_bpf_text_address(unsigned long addr)
 
 struct bpf_prog *bpf_prog_ksym_find(unsigned long addr)
 {
-	struct bpf_ksym *ksym = bpf_ksym_find(addr);
+	struct bpf_ksym *ksym;
+
+	WARN_ON_ONCE(!rcu_read_lock_held());
+	ksym = bpf_ksym_find(addr);
 
 	return ksym && ksym->prog ?
 	       container_of(ksym, struct bpf_prog_aux, ksym)->prog :
kernel/bpf/helpers.c +2 -0
index 8fef7b3cbd80..61b69eb08c4a 100644
@@ -2936,7 +2936,9 @@ static bool bpf_stack_walker(void *cookie, u64 ip, u64 sp, u64 bp)
 
 	if (!is_bpf_text_address(ip))
 		return !ctx->cnt;
+	rcu_read_lock();
 	prog = bpf_prog_ksym_find(ip);
+	rcu_read_unlock();
 	ctx->cnt++;
 	if (bpf_is_subprog(prog))
 		return true;
Signature · 1 line
2.47.1
#10 lore raw

Re: [PATCH bpf-next v3 04/12] bpf: Ensure RCU lock is held around bpf_prog_ksym_find

Quoted 33 lines · Kumar Kartikeya Dwivedi

Add a warning to ensure RCU lock is held around tree lookup, and then fix one of the invocations in bpf_stack_walker. The program has an active stack frame and won't disappear.

Fixes f18b03fabaa9 ("bpf: Implement BPF exceptions")
---
 kernel/bpf/core.c    | 5 ++++-
 kernel/bpf/helpers.c | 2 ++
 2 files changed, 6 insertions(+), 1 deletion(-)
kernel/bpf/core.c +4 -1
index 5c6e9fbb5508..b4203f68cf33 100644
@@ -782,7 +782,10 @@ bool is_bpf_text_address(unsigned long addr)
 
 struct bpf_prog *bpf_prog_ksym_find(unsigned long addr)
 {
-       struct bpf_ksym *ksym = bpf_ksym_find(addr);
+       struct bpf_ksym *ksym;
+
+       WARN_ON_ONCE(!rcu_read_lock_held());
+       ksym = bpf_ksym_find(addr);
 
        return ksym && ksym->prog ?
               container_of(ksym, struct bpf_prog_aux, ksym)->prog :
kernel/bpf/helpers.c +0 -0
index 8fef7b3cbd80..61b69eb08c4a 100644

kernel/bpf/helpers.c static bool bpf_stack_walker(void *cookie, u64 ip, u64 sp, u64 bp) · line 2941

Quoted 6 lines
kernel/bpf/helpers.c +2 -0
@@ -2936,7 +2936,9 @@ static bool bpf_stack_walker(void *cookie, u64 ip, u64 sp, u64 bp)
 
        if (!is_bpf_text_address(ip))
                return !ctx->cnt;
+       rcu_read_lock();
        prog = bpf_prog_ksym_find(ip);
+       rcu_read_unlock();

Please add a comment here explaining that rcu lock protects struct latch_tree_root bpf_tree access
and returned prog pointer won't disappear. Otherwise the rcu lock usage looks highly suspicious.

author maintainer
#11 lore raw

PATCH v3 5/12 bpf: Add function to find program from stack trace

In reply to Kumar Kartikeya Dwivedi

In preparation of figuring out the closest program that led to the current point in the kernel, implement a function that scans through the stack trace and finds out the closest BPF program when walking down the stack trace.

Special care needs to be taken to skip over kernel and BPF subprog frames. We basically scan until we find a BPF main prog frame. The assumption is that if a program calls into us transitively, we'll hit it along the way. If not, we end up returning NULL.

Contextually the function will be used in places where we know the program may have called into us.

Due to reliance on arch_bpf_stack_walk(), this function only works on x86 with CONFIG_UNWINDER_ORC, arm64, and s390. Remove the warning from arch_bpf_stack_walk as well since we call it outside bpf_throw() context.

---
 arch/x86/net/bpf_jit_comp.c |  1 -
 include/linux/bpf.h         |  1 +
 kernel/bpf/core.c           | 28 ++++++++++++++++++++++++++++
 3 files changed, 29 insertions(+), 1 deletion(-)
arch/x86/net/bpf_jit_comp.c +0 -1
index 15672cb926fc..40e1b3b9634f 100644
@@ -3845,7 +3845,6 @@ void arch_bpf_stack_walk(bool (*consume_fn)(void *cookie, u64 ip, u64 sp, u64 bp
 	}
 	return;
 #endif
-	WARN(1, "verification of programs using bpf_throw should have failed\n");
 }
 
 void bpf_arch_poke_desc_update(struct bpf_jit_poke_descriptor *poke,
include/linux/bpf.h +1 -0
index f30697c72ba9..cc14ff8e0b88 100644
@@ -3669,5 +3669,6 @@ static inline bool bpf_is_subprog(const struct bpf_prog *prog)
 
 int bpf_prog_get_file_line(struct bpf_prog *prog, unsigned long ip, const char **filep,
 			   const char **linep, int *nump);
+struct bpf_prog *bpf_prog_find_from_stack(void);
 
 #endif /* _LINUX_BPF_H */
kernel/bpf/core.c +28 -0
index b4203f68cf33..3871d817396d 100644
@@ -3262,4 +3262,32 @@ int bpf_prog_get_file_line(struct bpf_prog *prog, unsigned long ip, const char *
 	return 0;
 }
 
+struct walk_stack_ctx {
+	struct bpf_prog *prog;
+};
+
+static bool find_from_stack_cb(void *cookie, u64 ip, u64 sp, u64 bp)
+{
+	struct walk_stack_ctx *ctxp = cookie;
+	struct bpf_prog *prog;
+
+	rcu_read_lock();
+	prog = bpf_prog_ksym_find(ip);
+	rcu_read_unlock();
+	if (!prog)
+		return true;
+	if (bpf_is_subprog(prog))
+		return true;
+	ctxp->prog = prog;
+	return false;
+}
+
+struct bpf_prog *bpf_prog_find_from_stack(void)
+{
+	struct walk_stack_ctx ctx = {};
+
+	arch_bpf_stack_walk(find_from_stack_cb, &ctx);
+	return ctx.prog;
+}
+
 #endif
Signature · 1 line
2.47.1
#12 lore raw

Re: [PATCH bpf-next v3 05/12] bpf: Add function to find program from stack trace

Quoted 60 lines · Kumar Kartikeya Dwivedi

In preparation of figuring out the closest program that led to the current point in the kernel, implement a function that scans through the stack trace and finds out the closest BPF program when walking down the stack trace.

Special care needs to be taken to skip over kernel and BPF subprog frames. We basically scan until we find a BPF main prog frame. The assumption is that if a program calls into us transitively, we'll hit it along the way. If not, we end up returning NULL.

Contextually the function will be used in places where we know the program may have called into us.

Due to reliance on arch_bpf_stack_walk(), this function only works on x86 with CONFIG_UNWINDER_ORC, arm64, and s390. Remove the warning from arch_bpf_stack_walk as well since we call it outside bpf_throw() context.

---
 arch/x86/net/bpf_jit_comp.c |  1 -
 include/linux/bpf.h         |  1 +
 kernel/bpf/core.c           | 28 ++++++++++++++++++++++++++++
 3 files changed, 29 insertions(+), 1 deletion(-)
arch/x86/net/bpf_jit_comp.c +0 -1
index 15672cb926fc..40e1b3b9634f 100644
@@ -3845,7 +3845,6 @@ void arch_bpf_stack_walk(bool (*consume_fn)(void *cookie, u64 ip, u64 sp, u64 bp
        }
        return;
 #endif
-       WARN(1, "verification of programs using bpf_throw should have failed\n");
 }
 
 void bpf_arch_poke_desc_update(struct bpf_jit_poke_descriptor *poke,
include/linux/bpf.h +1 -0
index f30697c72ba9..cc14ff8e0b88 100644
@@ -3669,5 +3669,6 @@ static inline bool bpf_is_subprog(const struct bpf_prog *prog)
 
 int bpf_prog_get_file_line(struct bpf_prog *prog, unsigned long ip, const char **filep,
                           const char **linep, int *nump);
+struct bpf_prog *bpf_prog_find_from_stack(void);
 
 #endif /* _LINUX_BPF_H */
kernel/bpf/core.c +2 -0
index b4203f68cf33..3871d817396d 100644
@@ -3262,4 +3262,32 @@ int bpf_prog_get_file_line(struct bpf_prog *prog, unsigned long ip, const char *
        return 0;
 }
 
+struct walk_stack_ctx {
+       struct bpf_prog *prog;

kernel/bpf/core.c int bpf_prog_get_file_line(struct bpf_prog *prog, unsigned long ip, const char * · line 3276

Quoted 10 lines
kernel/bpf/core.c +10 -0
+};
+
+static bool find_from_stack_cb(void *cookie, u64 ip, u64 sp, u64 bp)
+{
+       struct walk_stack_ctx *ctxp = cookie;
+       struct bpf_prog *prog;
+
+       rcu_read_lock();
+       prog = bpf_prog_ksym_find(ip);
+       rcu_read_unlock();

Same here.
Otherwise it looks like an rcu noob mistake.

Quoted 20 lines
kernel/bpf/core.c +16 -1
+       if (!prog)
+               return true;
+       if (bpf_is_subprog(prog))
+               return true;
+       ctxp->prog = prog;
+       return false;
+}
+
+struct bpf_prog *bpf_prog_find_from_stack(void)
+{
+       struct walk_stack_ctx ctx = {};
+
+       arch_bpf_stack_walk(find_from_stack_cb, &ctx);
+       return ctx.prog;
+}
+
 #endif
--

2.47.1

author maintainer
#13 lore raw

PATCH v3 6/12 bpf: Add dump_stack() analogue to print to BPF stderr

In reply to Kumar Kartikeya Dwivedi

Introduce a kernel function which is the analogue of dump_stack() printing some useful information and the stack trace. This is not exposed to BPF programs yet, but can be made available in the future.

When we have a program counter for a BPF program in the stack trace, also additionally output the filename and line number to make the trace helpful. The rest of the trace can be passed into ./decode_stacktrace.sh to obtain the line numbers for kernel symbols.

---
 include/linux/bpf.h |  2 ++
 kernel/bpf/stream.c | 44 ++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 46 insertions(+)
include/linux/bpf.h +2 -0
index cc14ff8e0b88..bd3cde41795e 100644
@@ -3618,10 +3618,12 @@ __printf(2, 3)
 int bpf_stream_stage_printk(struct bpf_stream_stage *ss, const char *fmt, ...);
 int bpf_stream_stage_commit(struct bpf_stream_stage *ss, struct bpf_prog *prog,
 			    enum bpf_stream_id stream_id);
+int bpf_stream_stage_dump_stack(struct bpf_stream_stage *ss);
 
 bool bpf_prog_stream_error_limit(struct bpf_prog *prog);
 
 #define bpf_stream_printk(ss, ...) bpf_stream_stage_printk(&ss, __VA_ARGS__)
+#define bpf_stream_dump_stack(ss) bpf_stream_stage_dump_stack(&ss)
 
 #define bpf_stream_stage(ss, prog, stream_id, expr)                      \
 	({                                                               \
kernel/bpf/stream.c +44 -0

53 lines not on this page · raw message

Signature · 1 line
2.47.1
#14 lore raw

Re: [PATCH bpf-next v3 06/12] bpf: Add dump_stack() analogue to print to BPF stderr

On Mon, Jun 23, 2025 at 08:12:46PM -0700, Kumar Kartikeya Dwivedi wrote:

SNIP

Quoted 23 lines
kernel/bpf/stream.c +8 -0
index 75ceb6379368..5fb11202ab9c 100644
@@ -2,6 +2,7 @@
 /* Copyright (c) 2025 Meta Platforms, Inc. and affiliates. */
 
 #include <linux/bpf.h>
+#include <linux/filter.h>
 #include <linux/bpf_mem_alloc.h>
 #include <linux/percpu.h>
 #include <linux/refcount.h>
@@ -483,3 +484,46 @@ bool bpf_prog_stream_error_limit(struct bpf_prog *prog)
 {
 	return atomic_fetch_add(1, &prog->aux->stream_error_cnt) >= BPF_PROG_STREAM_ERROR_CNT;
 }
+
+struct dump_stack_ctx {
+	struct bpf_stream_stage *ss;
+	int err;
+};
+
+static bool dump_stack_cb(void *cookie, u64 ip, u64 sp, u64 bp)

kernel/bpf/stream.c bool bpf_prog_stream_error_limit(struct bpf_prog *prog) · line 503

Quoted 10 lines
kernel/bpf/stream.c +10 -0
+{
+	struct dump_stack_ctx *ctxp = cookie;
+	const char *file = "", *line = "";
+	struct bpf_prog *prog;
+	int num, ret;
+
+	if (is_bpf_text_address(ip)) {
+		rcu_read_lock();
+		prog = bpf_prog_ksym_find(ip);
+		rcu_read_unlock();

do you need to check prog != NULL ?

also is_bpf_text_address calls bpf_ksym_find and bpf_prog_ksym_find calls it again, I think it'd be better just to call bpf_prog_ksym_find from here

jirka

Quoted 30 lines
kernel/bpf/stream.c +26 -0
+		ret = bpf_prog_get_file_line(prog, ip, &file, &line, &num);
+		if (ret < 0)
+			goto end;
+		ctxp->err = bpf_stream_stage_printk(ctxp->ss, "%pS\n  %s @ %s:%d\n",
+						    (void *)ip, line, file, num);
+		return !ctxp->err;
+	}
+end:
+	ctxp->err = bpf_stream_stage_printk(ctxp->ss, "%pS\n", (void *)ip);
+	return !ctxp->err;
+}
+
+int bpf_stream_stage_dump_stack(struct bpf_stream_stage *ss)
+{
+	struct dump_stack_ctx ctx = { .ss = ss };
+	int ret;
+
+	ret = bpf_stream_stage_printk(ss, "CPU: %d UID: %d PID: %d Comm: %s\n",
+				      raw_smp_processor_id(), __kuid_val(current_real_cred()->euid),
+				      current->pid, current->comm);
+	ret = ret ?: bpf_stream_stage_printk(ss, "Call trace:\n");
+	if (!ret)
+		arch_bpf_stack_walk(dump_stack_cb, &ctx);
+	ret = ret ?: ctx.err;
+	return ret ?: bpf_stream_stage_printk(ss, "\n");
+}
Signature · 1 line
2.47.1
author maintainer
#15 lore raw

Re: [PATCH bpf-next v3 06/12] bpf: Add dump_stack() analogue to print to BPF stderr

Quoted 32 lines · Jiri Olsa

On Mon, Jun 23, 2025 at 08:12:46PM -0700, Kumar Kartikeya Dwivedi wrote:

SNIP

Quoted 27 lines
kernel/bpf/stream.c +12 -0
index 75ceb6379368..5fb11202ab9c 100644
@@ -2,6 +2,7 @@
 /* Copyright (c) 2025 Meta Platforms, Inc. and affiliates. */
 
 #include <linux/bpf.h>
+#include <linux/filter.h>
 #include <linux/bpf_mem_alloc.h>
 #include <linux/percpu.h>
 #include <linux/refcount.h>
@@ -483,3 +484,46 @@ bool bpf_prog_stream_error_limit(struct bpf_prog *prog)
 {
      return atomic_fetch_add(1, &prog->aux->stream_error_cnt) >= BPF_PROG_STREAM_ERROR_CNT;
 }
+
+struct dump_stack_ctx {
+     struct bpf_stream_stage *ss;
+     int err;
+};
+
+static bool dump_stack_cb(void *cookie, u64 ip, u64 sp, u64 bp)
+{
+     struct dump_stack_ctx *ctxp = cookie;
+     const char *file = "", *line = "";
+     struct bpf_prog *prog;
Quoted 8 lines

kernel/bpf/stream.c bool bpf_prog_stream_error_limit(struct bpf_prog *prog) · line 503

Quoted 6 lines
kernel/bpf/stream.c +6 -0
+     int num, ret;
+
+     if (is_bpf_text_address(ip)) {
+             rcu_read_lock();
+             prog = bpf_prog_ksym_find(ip);
+             rcu_read_unlock();

do you need to check prog != NULL ?

I think it should be non-NULL, given we're walking IPs of progs with an active stack frame so they're not going away.

Quoted 4 lines

also is_bpf_text_address calls bpf_ksym_find and bpf_prog_ksym_find calls it again, I think it'd be better just to call bpf_prog_ksym_find from here

Good point, I will fix that. Then just check it for NULL once and continue using it.

Thanks!

Quoted 33 lines

jirka

Quoted 30 lines
kernel/bpf/stream.c +26 -1
+             ret = bpf_prog_get_file_line(prog, ip, &file, &line, &num);
+             if (ret < 0)
+                     goto end;
+             ctxp->err = bpf_stream_stage_printk(ctxp->ss, "%pS\n  %s @ %s:%d\n",
+                                                 (void *)ip, line, file, num);
+             return !ctxp->err;
+     }
+end:
+     ctxp->err = bpf_stream_stage_printk(ctxp->ss, "%pS\n", (void *)ip);
+     return !ctxp->err;
+}
+
+int bpf_stream_stage_dump_stack(struct bpf_stream_stage *ss)
+{
+     struct dump_stack_ctx ctx = { .ss = ss };
+     int ret;
+
+     ret = bpf_stream_stage_printk(ss, "CPU: %d UID: %d PID: %d Comm: %s\n",
+                                   raw_smp_processor_id(), __kuid_val(current_real_cred()->euid),
+                                   current->pid, current->comm);
+     ret = ret ?: bpf_stream_stage_printk(ss, "Call trace:\n");
+     if (!ret)
+             arch_bpf_stack_walk(dump_stack_cb, &ctx);
+     ret = ret ?: ctx.err;
+     return ret ?: bpf_stream_stage_printk(ss, "\n");
+}
--

2.47.1

author maintainer
#16 lore raw

PATCH v3 7/12 bpf: Report may_goto timeout to BPF stderr

In reply to Kumar Kartikeya Dwivedi

Begin reporting may_goto timeouts to BPF program's stderr stream.

---
 kernel/bpf/core.c | 20 +++++++++++++++++++-
 1 file changed, 19 insertions(+), 1 deletion(-)
kernel/bpf/core.c +19 -1
index 3871d817396d..7ff2d37625f5 100644
@@ -3168,6 +3168,22 @@ u64 __weak arch_bpf_timed_may_goto(void)
 	return 0;
 }
 
+static noinline void bpf_prog_report_may_goto_violation(void)
+{
+#ifdef CONFIG_BPF_SYSCALL
+	struct bpf_stream_stage ss;
+	struct bpf_prog *prog;
+
+	prog = bpf_prog_find_from_stack();
+	if (!prog)
+		return;
+	bpf_stream_stage(ss, prog, BPF_STDERR, ({
+		bpf_stream_printk(ss, "ERROR: Timeout detected for may_goto instruction\n");
+		bpf_stream_dump_stack(ss);
+	}));
+#endif
+}
+
 u64 bpf_check_timed_may_goto(struct bpf_timed_may_goto *p)
 {
 	u64 time = ktime_get_mono_fast_ns();
@@ -3178,8 +3194,10 @@ u64 bpf_check_timed_may_goto(struct bpf_timed_may_goto *p)
 		return BPF_MAX_TIMED_LOOPS;
 	}
 	/* Check if we've exhausted our time slice, and zero count. */
-	if (time - p->timestamp >= (NSEC_PER_SEC / 4))
+	if (unlikely(time - p->timestamp >= (NSEC_PER_SEC / 4))) {
+		bpf_prog_report_may_goto_violation();
 		return 0;
+	}
 	/* Refresh the count for the stack frame. */
 	return BPF_MAX_TIMED_LOOPS;
 }
Signature · 1 line
2.47.1
author maintainer
#17 lore raw

PATCH v3 8/12 bpf: Report rqspinlock deadlocks/timeout to BPF stderr

In reply to Kumar Kartikeya Dwivedi

Begin reporting rqspinlock deadlocks and timeout to BPF program's stderr.

---
 kernel/bpf/rqspinlock.c | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)
kernel/bpf/rqspinlock.c +23 -0
index 338305c8852c..5ab354d55d82 100644
@@ -666,6 +666,27 @@ EXPORT_SYMBOL_GPL(resilient_queued_spin_lock_slowpath);
 
 __bpf_kfunc_start_defs();
 
+static void bpf_prog_report_rqspinlock_violation(const char *str, void *lock, bool irqsave)
+{
+	struct rqspinlock_held *rqh = this_cpu_ptr(&rqspinlock_held_locks);
+	struct bpf_stream_stage ss;
+	struct bpf_prog *prog;
+
+	prog = bpf_prog_find_from_stack();
+	if (!prog)
+		return;
+	bpf_stream_stage(ss, prog, BPF_STDERR, ({
+		bpf_stream_printk(ss, "ERROR: %s for bpf_res_spin_lock%s\n", str, irqsave ? "_irqsave" : "");
+		bpf_stream_printk(ss, "Attempted lock   = 0x%px\n", lock);
+		bpf_stream_printk(ss, "Total held locks = %d\n", rqh->cnt);
+		for (int i = 0; i < min(RES_NR_HELD, rqh->cnt); i++)
+			bpf_stream_printk(ss, "Held lock[%2d] = 0x%px\n", i, rqh->locks[i]);
+		bpf_stream_dump_stack(ss);
+	}));
+}
+
+#define REPORT_STR(ret) ({ (ret) == -ETIMEDOUT ? "Timeout detected" : "AA or ABBA deadlock detected"; })
+
 __bpf_kfunc int bpf_res_spin_lock(struct bpf_res_spin_lock *lock)
 {
 	int ret;
@@ -676,6 +697,7 @@ __bpf_kfunc int bpf_res_spin_lock(struct bpf_res_spin_lock *lock)
 	preempt_disable();
 	ret = res_spin_lock((rqspinlock_t *)lock);
 	if (unlikely(ret)) {
+		bpf_prog_report_rqspinlock_violation(REPORT_STR(ret), lock, false);
 		preempt_enable();
 		return ret;
 	}
@@ -698,6 +720,7 @@ __bpf_kfunc int bpf_res_spin_lock_irqsave(struct bpf_res_spin_lock *lock, unsign
 	local_irq_save(flags);
 	ret = res_spin_lock((rqspinlock_t *)lock);
 	if (unlikely(ret)) {
+		bpf_prog_report_rqspinlock_violation(REPORT_STR(ret), lock, true);
 		local_irq_restore(flags);
 		preempt_enable();
 		return ret;
Signature · 1 line
2.47.1
author maintainer
#18 lore raw

PATCH v3 9/12 libbpf: Add bpf_stream_printk() macro

In reply to Kumar Kartikeya Dwivedi

Add a convenience macro to print data to the BPF streams. BPF_STDOUT and BPF_STDERR stream IDs in the vmlinux.h can be passed to the macro to print to the respective streams.

---
 tools/lib/bpf/bpf_helpers.h | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
tools/lib/bpf/bpf_helpers.h +16 -0
index a50773d4616e..76b127a9f24d 100644
@@ -314,6 +314,22 @@ enum libbpf_tristate {
 			  ___param, sizeof(___param));		\
 })
 
+extern int bpf_stream_vprintk(int stream_id, const char *fmt__str, const void *args,
+			      __u32 len__sz, void *aux__prog) __weak __ksym;
+
+#define bpf_stream_printk(stream_id, fmt, args...)				\
+({										\
+	static const char ___fmt[] = fmt;					\
+	unsigned long long ___param[___bpf_narg(args)];				\
+										\
+	_Pragma("GCC diagnostic push")						\
+	_Pragma("GCC diagnostic ignored \"-Wint-conversion\"")			\
+	___bpf_fill(___param, args);						\
+	_Pragma("GCC diagnostic pop")						\
+										\
+	bpf_stream_vprintk(stream_id, ___fmt, ___param, sizeof(___param), NULL);\
+})
+
 /* Use __bpf_printk when bpf_printk call has 3 or fewer fmt args
  * Otherwise use __bpf_vprintk
  */
Signature · 1 line
2.47.1
Acked-by on v3 9/12
#19 lore raw

Re: [PATCH bpf-next v3 09/12] libbpf: Add bpf_stream_printk() macro

Quoted 11 lines · Kumar Kartikeya Dwivedi

Add a convenience macro to print data to the BPF streams. BPF_STDOUT and BPF_STDERR stream IDs in the vmlinux.h can be passed to the macro to print to the respective streams.

---
 tools/lib/bpf/bpf_helpers.h | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
Quoted 30 lines
tools/lib/bpf/bpf_helpers.h +16 -1
index a50773d4616e..76b127a9f24d 100644
@@ -314,6 +314,22 @@ enum libbpf_tristate {
                          ___param, sizeof(___param));          \
 })
 
+extern int bpf_stream_vprintk(int stream_id, const char *fmt__str, const void *args,
+                             __u32 len__sz, void *aux__prog) __weak __ksym;
+
+#define bpf_stream_printk(stream_id, fmt, args...)                             \
+({                                                                             \
+       static const char ___fmt[] = fmt;                                       \
+       unsigned long long ___param[___bpf_narg(args)];                         \
+                                                                               \
+       _Pragma("GCC diagnostic push")                                          \
+       _Pragma("GCC diagnostic ignored \"-Wint-conversion\"")                  \
+       ___bpf_fill(___param, args);                                            \
+       _Pragma("GCC diagnostic pop")                                           \
+                                                                               \
+       bpf_stream_vprintk(stream_id, ___fmt, ___param, sizeof(___param), NULL);\
+})
+
 /* Use __bpf_printk when bpf_printk call has 3 or fewer fmt args
  * Otherwise use __bpf_vprintk
  */
--

2.47.1

author maintainer
#20 lore raw

PATCH v3 10/12 libbpf: Introduce bpf_prog_stream_read() API

In reply to Kumar Kartikeya Dwivedi

Introduce a libbpf API so that users can read data from a given BPF stream for a BPF prog fd. For now, only the low-level syscall wrapper is provided, we can add a bpf_program__* accessor as a follow up if needed.

---
 tools/lib/bpf/bpf.c      | 20 ++++++++++++++++++++
 tools/lib/bpf/bpf.h      | 21 +++++++++++++++++++++
 tools/lib/bpf/libbpf.map |  1 +
 3 files changed, 42 insertions(+)
tools/lib/bpf/bpf.c +20 -0
index 6eb421ccf91b..ab40dbf9f020 100644
@@ -1375,3 +1375,23 @@ int bpf_token_create(int bpffs_fd, struct bpf_token_create_opts *opts)
 	fd = sys_bpf_fd(BPF_TOKEN_CREATE, &attr, attr_sz);
 	return libbpf_err_errno(fd);
 }
+
+int bpf_prog_stream_read(int prog_fd, __u32 stream_id, void *buf, __u32 buf_len,
+			 struct bpf_prog_stream_read_opts *opts)
+{
+	const size_t attr_sz = offsetofend(union bpf_attr, prog_stream_read);
+	union bpf_attr attr;
+	int err;
+
+	if (!OPTS_VALID(opts, bpf_prog_stream_read_opts))
+		return libbpf_err(-EINVAL);
+
+	memset(&attr, 0, attr_sz);
+	attr.prog_stream_read.stream_buf = ptr_to_u64(buf);
+	attr.prog_stream_read.stream_buf_len = buf_len;
+	attr.prog_stream_read.stream_id = stream_id;
+	attr.prog_stream_read.prog_fd = prog_fd;
+
+	err = sys_bpf(BPF_PROG_STREAM_READ_BY_FD, &attr, attr_sz);
+	return libbpf_err_errno(err);
+}
tools/lib/bpf/bpf.h +21 -0
index 1342564214c8..7252150e7ad3 100644
@@ -709,6 +709,27 @@ struct bpf_token_create_opts {
 LIBBPF_API int bpf_token_create(int bpffs_fd,
 				struct bpf_token_create_opts *opts);
 
+struct bpf_prog_stream_read_opts {
+	size_t sz;
+	size_t :0;
+};
+#define bpf_prog_stream_read_opts__last_field sz
+/**
+ * @brief **bpf_prog_stream_read** reads data from the BPF stream of a given BPF
+ * program.
+ *
+ * @param prog_fd FD for the BPF program whose BPF stream is to be read.
+ * @param stream_id ID of the BPF stream to be read.
+ * @param buf Buffer to read data into from the BPF stream.
+ * @param buf_len Maximum number of bytes to read from the BPF stream.
+ * @param opts optional options, can be NULL
+ *
+ * @return The number of bytes read, on success; negative error code, otherwise
+ * (errno is also set to the error code)
+ */
+LIBBPF_API int bpf_prog_stream_read(int prog_fd, __u32 stream_id, void *buf, __u32 buf_len,
+				    struct bpf_prog_stream_read_opts *opts);
+
 #ifdef __cplusplus
 } /* extern "C" */
 #endif
tools/lib/bpf/libbpf.map +1 -0
index c7fc0bde5648..1bbf77326420 100644
@@ -437,6 +437,7 @@ LIBBPF_1.6.0 {
 		bpf_linker__add_fd;
 		bpf_linker__new_fd;
 		bpf_object__prepare;
+		bpf_prog_stream_read;
 		bpf_program__attach_cgroup_opts;
 		bpf_program__func_info;
 		bpf_program__func_info_cnt;
Signature · 1 line
2.47.1
author maintainer
#21 lore raw

PATCH v3 11/12 bpftool: Add support for dumping streams

In reply to Kumar Kartikeya Dwivedi

Add support for printing the BPF stream contents of a program in bpftool. The new bpftool prog tracelog command is extended to take stdout and stderr arguments, and then the prog specification.

The bpf_prog_stream_read() API added in previous patch is simply reused to grab data and then it is dumped to the respective file. The stdout data is sent to stdout, and stderr is printed to stderr.

---
 .../bpftool/Documentation/bpftool-prog.rst    |  7 +++
 tools/bpf/bpftool/bash-completion/bpftool     | 16 +++++-
 tools/bpf/bpftool/prog.c                      | 49 ++++++++++++++++++-
 3 files changed, 70 insertions(+), 2 deletions(-)
tools/bpf/bpftool/Documentation/bpftool-prog.rst +7 -0
index da3152c16228..f69fd92df8d8 100644
@@ -35,6 +35,7 @@ PROG COMMANDS
 | **bpftool** **prog attach** *PROG* *ATTACH_TYPE* [*MAP*]
 | **bpftool** **prog detach** *PROG* *ATTACH_TYPE* [*MAP*]
 | **bpftool** **prog tracelog**
+| **bpftool** **prog tracelog** [ { **stdout** | **stderr**  } *PROG* ]
 | **bpftool** **prog run** *PROG* **data_in** *FILE* [**data_out** *FILE* [**data_size_out** *L*]] [**ctx_in** *FILE* [**ctx_out** *FILE* [**ctx_size_out** *M*]]] [**repeat** *N*]
 | **bpftool** **prog profile** *PROG* [**duration** *DURATION*] *METRICs*
 | **bpftool** **prog help**
@@ -179,6 +180,12 @@ bpftool prog tracelog
     purposes. For streaming data from BPF programs to user space, one can use
     perf events (see also **bpftool-map**\ (8)).
 
+bpftool prog tracelog { stdout | stderr } *PROG*
+    Dump the BPF stream of the program. BPF programs can write to these streams
+    at runtime with the **bpf_stream_vprintk**\ () kfunc. The kernel may write
+    error messages to the standard error stream. This facility should be used
+    only for debugging purposes.
+
 bpftool prog run *PROG* data_in *FILE* [data_out *FILE* [data_size_out *L*]] [ctx_in *FILE* [ctx_out *FILE* [ctx_size_out *M*]]] [repeat *N*]
     Run BPF program *PROG* in the kernel testing infrastructure for BPF,
     meaning that the program works on the data and context provided by the
tools/bpf/bpftool/bash-completion/bpftool +15 -1
index 27512feb5c70..a759ba24471d 100644
@@ -518,7 +518,21 @@ _bpftool()
                     esac
                     ;;
                 tracelog)
-                    return 0
+                    case $prev in
+                        $command)
+                            COMPREPLY+=( $( compgen -W "stdout stderr" -- \
+                                "$cur" ) )
+                            return 0
+                            ;;
+                        stdout|stderr)
+                            COMPREPLY=( $( compgen -W "$PROG_TYPE" -- \
+                                "$cur" ) )
+                            return 0
+                            ;;
+                        *)
+                            return 0
+                            ;;
+                    esac
                     ;;
                 profile)
                     case $cword in
tools/bpf/bpftool/prog.c +48 -1

67 lines not on this page · raw message

Signature · 1 line
2.47.1
author maintainer
#22 lore raw

PATCH v3 12/12 selftests/bpf: Add tests for prog streams

In reply to Kumar Kartikeya Dwivedi

Add selftests to stress test the various facets of the stream API, memory allocation pattern, and ensuring dumping support is tested and functional.

---
 .../testing/selftests/bpf/prog_tests/stream.c | 140 ++++++++++++++++++
 tools/testing/selftests/bpf/progs/stream.c    |  75 ++++++++++
 .../testing/selftests/bpf/progs/stream_fail.c |  17 +++
 3 files changed, 232 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/stream.c
 create mode 100644 tools/testing/selftests/bpf/progs/stream.c
 create mode 100644 tools/testing/selftests/bpf/progs/stream_fail.c
tools/testing/selftests/bpf/prog_tests/stream.c new file +140 -0
new file mode 100644
index 000000000000..b14ce6e682aa
@@ -0,0 +1,140 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2025 Meta Platforms, Inc. and affiliates. */
+#include <test_progs.h>
+#include <sys/mman.h>
+#include <regex.h>
+
+#include "stream.skel.h"
+#include "stream_fail.skel.h"
+
+void test_stream_failure(void)
+{
+	RUN_TESTS(stream_fail);
+}
+
+void test_stream_success(void)
+{
+	RUN_TESTS(stream);
+	return;
+}
+
+struct {
+	int prog_off;
+	const char *errstr;
+} stream_error_arr[] = {
+	{
+		offsetof(struct stream, progs.stream_cond_break),
+		"ERROR: Timeout detected for may_goto instruction\n\0"
+		"CPU: [0-9]+ UID: 0 PID: [0-9]+ Comm: test_progs\n\0"
+		"Call trace:\n\0"
+		"([a-zA-Z_][a-zA-Z0-9_]*\\+0x[0-9a-fA-F]+/0x[0-9a-fA-F]+\n\0"
+		"|[ \t]+[^\n]+\n\0)*",
+	},
+	{
+		offsetof(struct stream, progs.stream_deadlock),
+		"ERROR: AA or ABBA deadlock detected for bpf_res_spin_lock\n\0"
+		"Attempted lock   = (0x[0-9a-fA-F]+)\n\0"
+		"Total held locks = 1\n\0"
+		"Held lock\\[ 0\\] = \\1\n\0"  // Lock address must match
+		"CPU: [0-9]+ UID: 0 PID: [0-9]+ Comm: test_progs\n\0"
+		"Call trace:\n\0"
+		"([a-zA-Z_][a-zA-Z0-9_]*\\+0x[0-9a-fA-F]+/0x[0-9a-fA-F]+\n\0"
+		"|[ \t]+[^\n]+\n\0)*",
+	},
+};
+
+static int match_regex(const char *pattern, const char *string)
+{
+	int err, rc;
+	regex_t re;
+
+	err = regcomp(&re, pattern, REG_EXTENDED | REG_NEWLINE);
+	if (err)
+		return -1;
+	rc = regexec(&re, string, 0, NULL, 0);
+	regfree(&re);
+	return rc == 0 ? 1 : 0;
+}
+
+void test_stream_errors(void)
+{
+	LIBBPF_OPTS(bpf_test_run_opts, opts);
+	LIBBPF_OPTS(bpf_prog_stream_read_opts, ropts);
+	char buf[1024] = {};
+	struct stream *skel;
+	int ret, prog_fd;
+
+	skel = stream__open_and_load();
+	if (!ASSERT_OK_PTR(skel, "stream__open_and_load"))
+		return;
+
+	for (int i = 0; i < ARRAY_SIZE(stream_error_arr); i++) {
+		struct bpf_program **prog;
+
+		prog = (struct bpf_program **)(((char *)skel) + stream_error_arr[i].prog_off);
+		prog_fd = bpf_program__fd(*prog);
+		ret = bpf_prog_test_run_opts(prog_fd, &opts);
+		ASSERT_OK(ret, "ret");
+		ASSERT_OK(opts.retval, "retval");
+
+#if !defined(__x86_64__)
+		ASSERT_TRUE(1, "Timed may_goto unsupported, skip.");
+		if (i == 0) {
+			ret = bpf_prog_stream_read(prog_fd, 2, buf, sizeof(buf), &ropts);
+			ASSERT_EQ(ret, 0, "stream read");
+			continue;
+		}
+#endif
+
+		ret = bpf_prog_stream_read(prog_fd, BPF_STREAM_STDERR, buf, sizeof(buf), &ropts);
+		ASSERT_GT(ret, 0, "stream read");
+		ASSERT_LE(ret, 1023, "len for buf");
+		buf[ret] = '\0';
+
+		ret = match_regex(stream_error_arr[i].errstr, buf);
+		ASSERT_TRUE(ret == 1, "regex match");
+	}
+
+	stream__destroy(skel);
+}
+
+void test_stream_syscall(void)
+{
+	LIBBPF_OPTS(bpf_test_run_opts, opts);
+	LIBBPF_OPTS(bpf_prog_stream_read_opts, ropts);
+	struct stream *skel;
+	int ret, prog_fd;
+	char buf[64];
+
+	skel = stream__open_and_load();
+	if (!ASSERT_OK_PTR(skel, "stream__open_and_load"))
+		return;
+
+	prog_fd = bpf_program__fd(skel->progs.stream_syscall);
+	ret = bpf_prog_test_run_opts(prog_fd, &opts);
+	ASSERT_OK(ret, "ret");
+	ASSERT_OK(opts.retval, "retval");
+
+	ASSERT_LT(bpf_prog_stream_read(0, BPF_STREAM_STDOUT, buf, sizeof(buf), &ropts), 0, "error");
+	ret = -errno;
+	ASSERT_EQ(ret, -EINVAL, "bad prog_fd");
+
+	ASSERT_LT(bpf_prog_stream_read(prog_fd, 0, buf, sizeof(buf), &ropts), 0, "error");
+	ret = -errno;
+	ASSERT_EQ(ret, -ENOENT, "bad stream id");
+
+	ASSERT_LT(bpf_prog_stream_read(prog_fd, BPF_STREAM_STDOUT, NULL, sizeof(buf), NULL), 0, "error");
+	ret = -errno;
+	ASSERT_EQ(ret, -EFAULT, "bad stream buf");
+
+	ret = bpf_prog_stream_read(prog_fd, BPF_STREAM_STDOUT, buf, 2, NULL);
+	ASSERT_EQ(ret, 2, "bytes");
+	ret = bpf_prog_stream_read(prog_fd, BPF_STREAM_STDOUT, buf, 2, NULL);
+	ASSERT_EQ(ret, 2, "bytes");
+	ret = bpf_prog_stream_read(prog_fd, BPF_STREAM_STDOUT, buf, 1, &ropts);
+	ASSERT_EQ(ret, 0, "no bytes stdout");
+	ret = bpf_prog_stream_read(prog_fd, BPF_STREAM_STDERR, buf, 1, &ropts);
+	ASSERT_EQ(ret, 0, "no bytes stderr");
+
+	stream__destroy(skel);
+}
tools/testing/selftests/bpf/progs/stream.c new file +75 -0

75 lines not on this page · raw message

tools/testing/selftests/bpf/progs/stream_fail.c new file +17 -0
new file mode 100644
index 000000000000..12004d5092b7
@@ -0,0 +1,17 @@
+// SPDX-License-Identifier: GPL-2.0
+/* Copyright (c) 2025 Meta Platforms, Inc. and affiliates. */
+#include <vmlinux.h>
+#include <bpf/bpf_tracing.h>
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_core_read.h>
+#include "bpf_misc.h"
+
+SEC("syscall")
+__failure __msg("Possibly NULL pointer passed")
+int stream_vprintk_null_arg(void *ctx)
+{
+	bpf_stream_vprintk(BPF_STDOUT, "", NULL, 0, NULL);
+	return 0;
+}
+
+char _license[] SEC("license") = "GPL";
Signature · 1 line
2.47.1