Skip to contentSkip to conversation

tracing/user_events: Update user_events ABI from

PATCH Superseded Beau Belgrave 2 participants 14 messages Apr 25, 2022 – Jul 27, 2022

Subsystems (4)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

v2 cover tracing/user_events: Update user_events ABI from

This series covers the changes that were brought up once user_events went into 5.18. The largest change is moving away from byte index to a bit index, as first suggested by Mathieu Desnoyers.

The other changes are either fixes that have accumulated or found by Mathieu.

NOTE: The sample and self-tests do not build unless you manually install user_events.h into usr/include/linux.

Link:
https://lore.kernel.org/all/[email protected]/

Psuedo code example of typical usage with the new ABI:
struct user_reg reg;

int page_fd = open("user_events_status", O_RDWR);
char *page_data = mmap(NULL, PAGE_SIZE, PROT_READ, MAP_SHARED, page_fd, 0);
close(page_fd);

int data_fd = open("user_events_data", O_RDWR);

reg.size = sizeof(reg);
reg.name_args = (__u64)"test";

ioctl(data_fd, DIAG_IOCSREG, &reg);
int status_id = reg.status_bit / 8;
int status_mask = 1 << (reg.status_bit % 8);
int write_id = reg.write_index;

struct iovec io[2];
io[0].iov_base = &write_id;
io[0].iov_len = sizeof(write_id);
io[1].iov_base = payload;
io[1].iov_len = sizeof(payload);

if (page_data[status_id] & status_mask)

	writev(data_fd, io, 2);

V2 Updates:
Changed from status_index and status_mask on user_reg to just status_bit. Updated and added byte-wise and long-wise indexing pattern into docs. Updated tests to use byte-wise index pattern.
Updated sample to show long-wise index pattern.

Beau Belgrave (7):
tracing/user_events: Fix repeated word in comments tracing/user_events: Use NULL for strstr checks tracing/user_events: Use WRITE instead of READ for io vector import tracing/user_events: Ensure user provided strings are safely formatted tracing/user_events: Use refcount instead of atomic for ref tracking tracing/user_events: Use bits vs bytes for enabled status page data tracing/user_events: Update ABI documentation to align to bits vs
bytes

 Documentation/trace/user_events.rst           |  86 ++++---
 include/linux/user_events.h                   |  15 +-
 kernel/trace/trace_events_user.c              | 212 ++++++++++++------
 samples/user_events/example.c                 |  25 ++-
 .../selftests/user_events/ftrace_test.c       |  47 +++-
 .../testing/selftests/user_events/perf_test.c |  11 +-
 6 files changed, 264 insertions(+), 132 deletions(-)

base-commit: 5cfff569cab8bf544bab62c911c5d6efd5af5e05

Signature · 1 line
2.25.1
#1 lore raw

PATCH v2 1/7 tracing/user_events: Fix repeated word in comments

Trivial fix to remove accidental double wording of have in comments.

---
 kernel/trace/trace_events_user.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/trace/trace_events_user.c +1 -1
index 706e1686b5eb..a6621c52ce45 100644
@@ -567,7 +567,7 @@ static int user_event_set_call_visible(struct user_event *user, bool visible)
 	 * to allow user_event files to be less locked down. The extreme case
 	 * being "other" has read/write access to user_events_data/status.
 	 *
-	 * When not locked down, processes may not have have permissions to
+	 * When not locked down, processes may not have permissions to
 	 * add/remove calls themselves to tracefs. We need to temporarily
 	 * switch to root file permission to allow for this scenario.
 	 */
Signature · 1 line
2.25.1
#2 lore raw

PATCH v2 2/7 tracing/user_events: Use NULL for strstr checks

In reply to Beau Belgrave

Trivial fix to ensure strstr checks use NULL instead of 0.

---
 kernel/trace/trace_events_user.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
kernel/trace/trace_events_user.c +3 -3
index a6621c52ce45..075d694d20e3 100644
@@ -277,7 +277,7 @@ static int user_event_add_field(struct user_event *user, const char *type,
 	goto add_field;
 
 add_validator:
-	if (strstr(type, "char") != 0)
+	if (strstr(type, "char") != NULL)
 		validator_flags |= VALIDATOR_ENSURE_NULL;
 
 	validator = kmalloc(sizeof(*validator), GFP_KERNEL);
@@ -458,7 +458,7 @@ static const char *user_field_format(const char *type)
 		return "%d";
 	if (strcmp(type, "unsigned char") == 0)
 		return "%u";
-	if (strstr(type, "char[") != 0)
+	if (strstr(type, "char[") != NULL)
 		return "%s";
 
 	/* Unknown, likely struct, allowed treat as 64-bit */
@@ -479,7 +479,7 @@ static bool user_field_is_dyn_string(const char *type, const char **str_func)
 
 	return false;
 check:
-	return strstr(type, "char") != 0;
+	return strstr(type, "char") != NULL;
 }
 
 #define LEN_OR_ZERO (len ? len - pos : 0)
Signature · 1 line
2.25.1
#3 lore raw

PATCH v2 3/7 tracing/user_events: Use WRITE instead of READ for io vector import

In reply to Beau Belgrave

import_single_range expects the direction/rw to be where it came from, not the protection/limit. Since the import is in a write path use WRITE.

---
 kernel/trace/trace_events_user.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
kernel/trace/trace_events_user.c +2 -1
index 075d694d20e3..15edbf6b1e2e 100644
@@ -1245,7 +1245,8 @@ static ssize_t user_events_write(struct file *file, const char __user *ubuf,
 	if (unlikely(*ppos != 0))
 		return -EFAULT;
 
-	if (unlikely(import_single_range(READ, (char *)ubuf, count, &iov, &i)))
+	if (unlikely(import_single_range(WRITE, (char __user *)ubuf,
+					 count, &iov, &i)))
 		return -EFAULT;
 
 	return user_events_write_core(file, &i);
Signature · 1 line
2.25.1
#4 lore raw

PATCH v2 4/7 tracing/user_events: Ensure user provided strings are safely formatted

In reply to Beau Belgrave

User processes can provide bad strings that may cause issues or leak kernel details back out. Don't trust the content of these strings when formatting strings for matching.

This also moves to a consistent dynamic length string creation model.

---
 kernel/trace/trace_events_user.c | 91 +++++++++++++++++++++-----------
 1 file changed, 59 insertions(+), 32 deletions(-)
kernel/trace/trace_events_user.c +59 -32
index 15edbf6b1e2e..f9bb7d37d76f 100644
@@ -45,7 +45,6 @@
 #define MAX_EVENT_DESC 512
 #define EVENT_NAME(user_event) ((user_event)->tracepoint.name)
 #define MAX_FIELD_ARRAY_SIZE 1024
-#define MAX_FIELD_ARG_NAME 256
 
 static char *register_page_data;
 
@@ -483,6 +482,48 @@ static bool user_field_is_dyn_string(const char *type, const char **str_func)
 }
 
 #define LEN_OR_ZERO (len ? len - pos : 0)
+static int user_dyn_field_set_string(int argc, const char **argv, int *iout,
+				     char *buf, int len, bool *colon)
+{
+	int pos = 0, i = *iout;
+
+	*colon = false;
+
+	for (; i < argc; ++i) {
+		if (i != *iout)
+			pos += snprintf(buf + pos, LEN_OR_ZERO, " ");
+
+		pos += snprintf(buf + pos, LEN_OR_ZERO, "%s", argv[i]);
+
+		if (strchr(argv[i], ';')) {
+			++i;
+			*colon = true;
+			break;
+		}
+	}
+
+	/* Actual set, advance i */
+	if (len != 0)
+		*iout = i;
+
+	return pos + 1;
+}
+
+static int user_field_set_string(struct ftrace_event_field *field,
+				 char *buf, int len, bool colon)
+{
+	int pos = 0;
+
+	pos += snprintf(buf + pos, LEN_OR_ZERO, "%s", field->type);
+	pos += snprintf(buf + pos, LEN_OR_ZERO, " ");
+	pos += snprintf(buf + pos, LEN_OR_ZERO, "%s", field->name);
+
+	if (colon)
+		pos += snprintf(buf + pos, LEN_OR_ZERO, ";");
+
+	return pos + 1;
+}
+
 static int user_event_set_print_fmt(struct user_event *user, char *buf, int len)
 {
 	struct ftrace_event_field *field, *next;
@@ -926,49 +967,35 @@ static int user_event_free(struct dyn_event *ev)
 static bool user_field_match(struct ftrace_event_field *field, int argc,
 			     const char **argv, int *iout)
 {
-	char *field_name, *arg_name;
-	int len, pos, i = *iout;
+	char *field_name = NULL, *dyn_field_name = NULL;
 	bool colon = false, match = false;
+	int dyn_len, len;
 
-	if (i >= argc)
+	if (*iout >= argc)
 		return false;
 
-	len = MAX_FIELD_ARG_NAME;
-	field_name = kmalloc(len, GFP_KERNEL);
-	arg_name = kmalloc(len, GFP_KERNEL);
+	dyn_len = user_dyn_field_set_string(argc, argv, iout, dyn_field_name,
+					    0, &colon);
 
-	if (!arg_name || !field_name)
-		goto out;
-
-	pos = 0;
-
-	for (; i < argc; ++i) {
-		if (i != *iout)
-			pos += snprintf(arg_name + pos, len - pos, " ");
+	len = user_field_set_string(field, field_name, 0, colon);
 
-		pos += snprintf(arg_name + pos, len - pos, argv[i]);
-
-		if (strchr(argv[i], ';')) {
-			++i;
-			colon = true;
-			break;
-		}
-	}
+	if (dyn_len != len)
+		return false;
 
-	pos = 0;
+	dyn_field_name = kmalloc(dyn_len, GFP_KERNEL);
+	field_name = kmalloc(len, GFP_KERNEL);
 
-	pos += snprintf(field_name + pos, len - pos, field->type);
-	pos += snprintf(field_name + pos, len - pos, " ");
-	pos += snprintf(field_name + pos, len - pos, field->name);
+	if (!dyn_field_name || !field_name)
+		goto out;
 
-	if (colon)
-		pos += snprintf(field_name + pos, len - pos, ";");
+	user_dyn_field_set_string(argc, argv, iout, dyn_field_name,
+				  dyn_len, &colon);
 
-	*iout = i;
+	user_field_set_string(field, field_name, len, colon);
 
-	match = strcmp(arg_name, field_name) == 0;
+	match = strcmp(dyn_field_name, field_name) == 0;
 out:
-	kfree(arg_name);
+	kfree(dyn_field_name);
 	kfree(field_name);
 
 	return match;
Signature · 1 line
2.25.1
#5 lore raw

PATCH v2 5/7 tracing/user_events: Use refcount instead of atomic for ref tracking

In reply to Beau Belgrave

User processes could open up enough event references to cause rollovers. These could cause use after free scenarios, which we do not want. Switching to refcount APIs prevent this, but will leak memory once saturated.

Once saturated, user processes can still use the events. This prevents a bad user process from stopping existing telemetry from being emitted.

---
 kernel/trace/trace_events_user.c | 53 +++++++++++++++-----------------
 1 file changed, 24 insertions(+), 29 deletions(-)
kernel/trace/trace_events_user.c +24 -29
index f9bb7d37d76f..2bcae7abfa81 100644
@@ -14,6 +14,7 @@
 #include <linux/uio.h>
 #include <linux/ioctl.h>
 #include <linux/jhash.h>
+#include <linux/refcount.h>
 #include <linux/trace_events.h>
 #include <linux/tracefs.h>
 #include <linux/types.h>
@@ -57,7 +58,7 @@ static DECLARE_BITMAP(page_bitmap, MAX_EVENTS);
  * within a file a user_event might be created if it does not
  * already exist. These are globally used and their lifetime
  * is tied to the refcnt member. These cannot go away until the
- * refcnt reaches zero.
+ * refcnt reaches one.
  */
 struct user_event {
 	struct tracepoint tracepoint;
@@ -67,7 +68,7 @@ struct user_event {
 	struct hlist_node node;
 	struct list_head fields;
 	struct list_head validators;
-	atomic_t refcnt;
+	refcount_t refcnt;
 	int index;
 	int flags;
 	int min_size;
@@ -105,6 +106,12 @@ static u32 user_event_key(char *name)
 	return jhash(name, strlen(name), 0);
 }
 
+static __always_inline __must_check
+bool user_event_last_ref(struct user_event *user)
+{
+	return refcount_read(&user->refcnt) == 1;
+}
+
 static __always_inline __must_check
 size_t copy_nofault(void *addr, size_t bytes, struct iov_iter *i)
 {
@@ -662,7 +669,7 @@ static struct user_event *find_user_event(char *name, u32 *outkey)
 
 	hash_for_each_possible(register_table, user, node, key)
 		if (!strcmp(EVENT_NAME(user), name)) {
-			atomic_inc(&user->refcnt);
+			refcount_inc(&user->refcnt);
 			return user;
 		}
 
@@ -876,12 +883,12 @@ static int user_event_reg(struct trace_event_call *call,
 
 	return ret;
 inc:
-	atomic_inc(&user->refcnt);
+	refcount_inc(&user->refcnt);
 	update_reg_page_for(user);
 	return 0;
 dec:
 	update_reg_page_for(user);
-	atomic_dec(&user->refcnt);
+	refcount_dec(&user->refcnt);
 	return 0;
 }
 
@@ -907,7 +914,7 @@ static int user_event_create(const char *raw_command)
 	ret = user_event_parse_cmd(name, &user);
 
 	if (!ret)
-		atomic_dec(&user->refcnt);
+		refcount_dec(&user->refcnt);
 
 	mutex_unlock(&reg_mutex);
 
@@ -951,14 +958,14 @@ static bool user_event_is_busy(struct dyn_event *ev)
 {
 	struct user_event *user = container_of(ev, struct user_event, devent);
 
-	return atomic_read(&user->refcnt) != 0;
+	return !user_event_last_ref(user);
 }
 
 static int user_event_free(struct dyn_event *ev)
 {
 	struct user_event *user = container_of(ev, struct user_event, devent);
 
-	if (atomic_read(&user->refcnt) != 0)
+	if (!user_event_last_ref(user))
 		return -EBUSY;
 
 	return destroy_user_event(user);
@@ -1137,8 +1144,8 @@ static int user_event_parse(char *name, char *args, char *flags,
 
 	user->index = index;
 
-	/* Ensure we track ref */
-	atomic_inc(&user->refcnt);
+	/* Ensure we track self ref and caller ref (2) */
+	refcount_set(&user->refcnt, 2);
 
 	dyn_event_init(&user->devent, &user_event_dops);
 	dyn_event_add(&user->devent, &user->call);
@@ -1164,29 +1171,17 @@ static int user_event_parse(char *name, char *args, char *flags,
 static int delete_user_event(char *name)
 {
 	u32 key;
-	int ret;
 	struct user_event *user = find_user_event(name, &key);
 
 	if (!user)
 		return -ENOENT;
 
-	/* Ensure we are the last ref */
-	if (atomic_read(&user->refcnt) != 1) {
-		ret = -EBUSY;
-		goto put_ref;
-	}
-
-	ret = destroy_user_event(user);
+	refcount_dec(&user->refcnt);
 
-	if (ret)
-		goto put_ref;
-
-	return ret;
-put_ref:
-	/* No longer have this ref */
-	atomic_dec(&user->refcnt);
+	if (!user_event_last_ref(user))
+		return -EBUSY;
 
-	return ret;
+	return destroy_user_event(user);
 }
 
 /*
@@ -1314,7 +1309,7 @@ static int user_events_ref_add(struct file *file, struct user_event *user)
 
 	new_refs->events[i] = user;
 
-	atomic_inc(&user->refcnt);
+	refcount_inc(&user->refcnt);
 
 	rcu_assign_pointer(file->private_data, new_refs);
 
@@ -1374,7 +1369,7 @@ static long user_events_ioctl_reg(struct file *file, unsigned long uarg)
 	ret = user_events_ref_add(file, user);
 
 	/* No longer need parse ref, ref_add either worked or not */
-	atomic_dec(&user->refcnt);
+	refcount_dec(&user->refcnt);
 
 	/* Positive number is index and valid */
 	if (ret < 0)
@@ -1464,7 +1459,7 @@ static int user_events_release(struct inode *node, struct file *file)
 		user = refs->events[i];
 
 		if (user)
-			atomic_dec(&user->refcnt);
+			refcount_dec(&user->refcnt);
 	}
 out:
 	file->private_data = NULL;
Signature · 1 line
2.25.1
#6 lore raw

PATCH v2 6/7 tracing/user_events: Use bits vs bytes for enabled status page data

In reply to Beau Belgrave

User processes may require many events and when they do the cache performance of a byte index status check is less ideal than a bit index. The previous event limit per-page was 4096, the new limit is 32,768.

This change adds a bitwise index to the user_reg struct. Programs check that the bit at status_bit has a bit set within the status page(s).

---
 include/linux/user_events.h                   | 15 +----
 kernel/trace/trace_events_user.c              | 57 ++++++++++++++++---
 samples/user_events/example.c                 | 25 +++++---
 .../selftests/user_events/ftrace_test.c       | 47 ++++++++++++---
 .../testing/selftests/user_events/perf_test.c | 11 +++-
 5 files changed, 117 insertions(+), 38 deletions(-)
include/linux/user_events.h +3 -12
index 736e05603463..592a3fbed98e 100644
@@ -20,15 +20,6 @@
 #define USER_EVENTS_SYSTEM "user_events"
 #define USER_EVENTS_PREFIX "u:"
 
-/* Bits 0-6 are for known probe types, Bit 7 is for unknown probes */
-#define EVENT_BIT_FTRACE 0
-#define EVENT_BIT_PERF 1
-#define EVENT_BIT_OTHER 7
-
-#define EVENT_STATUS_FTRACE (1 << EVENT_BIT_FTRACE)
-#define EVENT_STATUS_PERF (1 << EVENT_BIT_PERF)
-#define EVENT_STATUS_OTHER (1 << EVENT_BIT_OTHER)
-
 /* Create dynamic location entry within a 32-bit value */
 #define DYN_LOC(offset, size) ((size) << 16 | (offset))
 
@@ -45,12 +36,12 @@ struct user_reg {
 	/* Input: Pointer to string with event name, description and flags */
 	__u64 name_args;
 
-	/* Output: Byte index of the event within the status page */
-	__u32 status_index;
+	/* Output: Bitwise index of the event within the status page */
+	__u32 status_bit;
 
 	/* Output: Index of the event to use when writing data */
 	__u32 write_index;
-};
+} __attribute__((__packed__));
 
 #define DIAG_IOC_MAGIC '*'
 
kernel/trace/trace_events_user.c +49 -8

120 lines not on this page · raw message

samples/user_events/example.c +18 -7
index 4f5778e441c0..d06dc24156ec 100644
@@ -12,13 +12,21 @@
 #include <fcntl.h>
 #include <stdio.h>
 #include <unistd.h>
+#include <asm/bitsperlong.h>
+#include <endian.h>
 #include <linux/user_events.h>
 
+#if __BITS_PER_LONG == 64
+#define endian_swap(x) htole64(x)
+#else
+#define endian_swap(x) htole32(x)
+#endif
+
 /* Assumes debugfs is mounted */
 const char *data_file = "/sys/kernel/debug/tracing/user_events_data";
 const char *status_file = "/sys/kernel/debug/tracing/user_events_status";
 
-static int event_status(char **status)
+static int event_status(long **status)
 {
 	int fd = open(status_file, O_RDONLY);
 
@@ -33,7 +41,8 @@ static int event_status(char **status)
 	return 0;
 }
 
-static int event_reg(int fd, const char *command, int *status, int *write)
+static int event_reg(int fd, const char *command, long *index, long *mask,
+		     int *write)
 {
 	struct user_reg reg = {0};
 
@@ -43,7 +52,8 @@ static int event_reg(int fd, const char *command, int *status, int *write)
 	if (ioctl(fd, DIAG_IOCSREG, &reg) == -1)
 		return -1;
 
-	*status = reg.status_index;
+	*index = reg.status_bit / __BITS_PER_LONG;
+	*mask = endian_swap(1L << (reg.status_bit % __BITS_PER_LONG));
 	*write = reg.write_index;
 
 	return 0;
@@ -51,8 +61,9 @@ static int event_reg(int fd, const char *command, int *status, int *write)
 
 int main(int argc, char **argv)
 {
-	int data_fd, status, write;
-	char *status_page;
+	int data_fd, write;
+	long index, mask;
+	long *status_page;
 	struct iovec io[2];
 	__u32 count = 0;
 
@@ -61,7 +72,7 @@ int main(int argc, char **argv)
 
 	data_fd = open(data_file, O_RDWR);
 
-	if (event_reg(data_fd, "test u32 count", &status, &write) == -1)
+	if (event_reg(data_fd, "test u32 count", &index, &mask, &write) == -1)
 		return errno;
 
 	/* Setup iovec */
@@ -75,7 +86,7 @@ int main(int argc, char **argv)
 	getchar();
 
 	/* Check if anyone is listening */
-	if (status_page[status]) {
+	if (status_page[index] & mask) {
 		/* Yep, trace out our data */
 		writev(data_fd, (const struct iovec *)io, 2);
 
tools/testing/selftests/user_events/ftrace_test.c +39 -8

120 lines not on this page · raw message

tools/testing/selftests/user_events/perf_test.c +8 -3
index 26851d51d6bb..8b4c7879d5a7 100644
@@ -35,6 +35,11 @@ static long perf_event_open(struct perf_event_attr *pe, pid_t pid,
 	return syscall(__NR_perf_event_open, pe, pid, cpu, group_fd, flags);
 }
 
+static inline int status_check(char *status_page, int status_bit)
+{
+	return status_page[status_bit >> 3] & (1 << (status_bit & 7));
+}
+
 static int get_id(void)
 {
 	FILE *fp = fopen(id_file, "r");
@@ -120,8 +125,8 @@ TEST_F(user, perf_write) {
 	/* Register should work */
 	ASSERT_EQ(0, ioctl(self->data_fd, DIAG_IOCSREG, &reg));
 	ASSERT_EQ(0, reg.write_index);
-	ASSERT_NE(0, reg.status_index);
-	ASSERT_EQ(0, status_page[reg.status_index]);
+	ASSERT_NE(0, reg.status_bit);
+	ASSERT_EQ(0, status_check(status_page, reg.status_bit));
 
 	/* Id should be there */
 	id = get_id();
@@ -144,7 +149,7 @@ TEST_F(user, perf_write) {
 	ASSERT_NE(MAP_FAILED, perf_page);
 
 	/* Status should be updated */
-	ASSERT_EQ(EVENT_STATUS_PERF, status_page[reg.status_index]);
+	ASSERT_NE(0, status_check(status_page, reg.status_bit));
 
 	event.index = reg.write_index;
 	event.field1 = 0xc001;
Signature · 1 line
2.25.1
maintainer
#7 lore raw

Re: [PATCH v2 6/7] tracing/user_events: Use bits vs bytes for enabled status page data

Quoted 14 lines · On Mon, 25 Apr 2022 11:46:30 -0700 Beau Belgrave <[email protected]>…
kernel/trace/trace_events_user.c +2 -1
index 2bcae7abfa81..4afc41e321ac 100644
@@ -40,17 +40,26 @@
  */
 #define MAX_PAGE_ORDER 0
 #define MAX_PAGES (1 << MAX_PAGE_ORDER)
-#define MAX_EVENTS (MAX_PAGES * PAGE_SIZE)
+#define MAX_BYTES (MAX_PAGES * PAGE_SIZE)
+#define MAX_EVENTS (MAX_BYTES * 8)
 
 /* Limit how long of an event name plus args within the subsystem. */
 #define MAX_EVENT_DESC 512

kernel/trace/trace_events_user.c line 57

Quoted 10 lines
kernel/trace/trace_events_user.c +7 -0
 #define EVENT_NAME(user_event) ((user_event)->tracepoint.name)
 #define MAX_FIELD_ARRAY_SIZE 1024
 
+#define STATUS_BYTE(bit) ((bit) >> 3)
+#define STATUS_MASK(bit) (1 << ((bit) & 7))
+
+/* Internal bits to keep track of connected probes */
+#define EVENT_STATUS_FTRACE (1 << 0)
+#define EVENT_STATUS_PERF (1 << 1)
+#define EVENT_STATUS_OTHER (1 << 7)

Did you mean to shift STATUS_OTHER by 7?

Is EVENT_STATUS_OTHER suppose to be one of the flags within the 3 bits of the 7 in STATUS_MASK?

-- Steve

Quoted 15 lines
kernel/trace/trace_events_user.c +3 -1
+
 static char *register_page_data;
 
 static DEFINE_MUTEX(reg_mutex);
-static DEFINE_HASHTABLE(register_table, 4);
+static DEFINE_HASHTABLE(register_table, 8);
 static DECLARE_BITMAP(page_bitmap, MAX_EVENTS);
 
 /*
@@ -72,6 +81,7 @@ struct user_event {
 	int index;
 	int flags;
 	int min_size;
+	char status;
 };
#8 lore raw

Re: [PATCH v2 6/7] tracing/user_events: Use bits vs bytes for enabled status page data

Quoted 20 lines · Steven Rostedt
Quoted 17 lines · On Mon, 25 Apr 2022 11:46:30 -0700 Beau Belgrave <[email protected]>…
kernel/trace/trace_events_user.c +2 -1
index 2bcae7abfa81..4afc41e321ac 100644
@@ -40,17 +40,26 @@
  */
 #define MAX_PAGE_ORDER 0
 #define MAX_PAGES (1 << MAX_PAGE_ORDER)
-#define MAX_EVENTS (MAX_PAGES * PAGE_SIZE)
+#define MAX_BYTES (MAX_PAGES * PAGE_SIZE)
+#define MAX_EVENTS (MAX_BYTES * 8)
 
 /* Limit how long of an event name plus args within the subsystem. */
 #define MAX_EVENT_DESC 512
 #define EVENT_NAME(user_event) ((user_event)->tracepoint.name)
 #define MAX_FIELD_ARRAY_SIZE 1024
 
Quoted 10 lines

kernel/trace/trace_events_user.c line 57

Quoted 7 lines
kernel/trace/trace_events_user.c +7 -0
+#define STATUS_BYTE(bit) ((bit) >> 3)
+#define STATUS_MASK(bit) (1 << ((bit) & 7))
+
+/* Internal bits to keep track of connected probes */
+#define EVENT_STATUS_FTRACE (1 << 0)
+#define EVENT_STATUS_PERF (1 << 1)
+#define EVENT_STATUS_OTHER (1 << 7)

Did you mean to shift STATUS_OTHER by 7?

Yes, it should be the value 128.

Quoted 3 lines

Is EVENT_STATUS_OTHER suppose to be one of the flags within the 3 bits of the 7 in STATUS_MASK?

My thought was that STATUS_OTHER would stay on the highest bit. Then when we have other systems they would slot into (1 << 2), etc.

This may not be as important now since the byte is never given back to the user and is only used when printing out status via the user_events_status file in text form.

Quoted 2 lines

-- Steve

Thanks,
-Beau

maintainer
#9 lore raw

Re: [PATCH v2 6/7] tracing/user_events: Use bits vs bytes for enabled status page data

Quoted 22 lines · On Tue, 26 Jul 2022 17:02:49 -0700 Beau Belgrave <[email protected]>…
Quoted 5 lines
Quoted 5 lines

/* Limit how long of an event name plus args within the subsystem. */ #define MAX_EVENT_DESC 512
#define EVENT_NAME(user_event) ((user_event)->tracepoint.name) #define MAX_FIELD_ARRAY_SIZE 1024

Quoted 10 lines

kernel/trace/trace_events_user.c line 57

Quoted 7 lines
kernel/trace/trace_events_user.c +7 -0
+#define STATUS_BYTE(bit) ((bit) >> 3)
+#define STATUS_MASK(bit) (1 << ((bit) & 7))
+
+/* Internal bits to keep track of connected probes */
+#define EVENT_STATUS_FTRACE (1 << 0)
+#define EVENT_STATUS_PERF (1 << 1)
+#define EVENT_STATUS_OTHER (1 << 7)  

Did you mean to shift STATUS_OTHER by 7?

Yes, it should be the value 128.

Quoted 3 lines

Is EVENT_STATUS_OTHER suppose to be one of the flags within the 3 bits of the 7 in STATUS_MASK?

Quoted 6 lines

My thought was that STATUS_OTHER would stay on the highest bit. Then when we have other systems they would slot into (1 << 2), etc.

This may not be as important now since the byte is never given back to the user and is only used when printing out status via the user_events_status file in text form.

So, it is confusing because of STATUS_MASK() is bits 0,1,2 and we are only using bits 0 and 1, with a OTHER bit at bit 7. And it would be good to use the BIT() macro.

Is STATUS_OTHER suppose to be part of STATUS_MASK()?

-- Steve

#10 lore raw

Re: [PATCH v2 6/7] tracing/user_events: Use bits vs bytes for enabled status page data

Quoted 28 lines · Steven Rostedt
Quoted 25 lines · On Tue, 26 Jul 2022 17:02:49 -0700 Beau Belgrave <[email protected]>…
Quoted 5 lines
Quoted 5 lines

/* Limit how long of an event name plus args within the subsystem. */ #define MAX_EVENT_DESC 512
#define EVENT_NAME(user_event) ((user_event)->tracepoint.name) #define MAX_FIELD_ARRAY_SIZE 1024

Quoted 10 lines

kernel/trace/trace_events_user.c line 57

Quoted 7 lines
kernel/trace/trace_events_user.c +7 -0
+#define STATUS_BYTE(bit) ((bit) >> 3)
+#define STATUS_MASK(bit) (1 << ((bit) & 7))
+
+/* Internal bits to keep track of connected probes */
+#define EVENT_STATUS_FTRACE (1 << 0)
+#define EVENT_STATUS_PERF (1 << 1)
+#define EVENT_STATUS_OTHER (1 << 7)  

Did you mean to shift STATUS_OTHER by 7?

Yes, it should be the value 128.

Quoted 3 lines

Is EVENT_STATUS_OTHER suppose to be one of the flags within the 3 bits of the 7 in STATUS_MASK?

My thought was that STATUS_OTHER would stay on the highest bit. Then when we have other systems they would slot into (1 << 2), etc.

Quoted 8 lines
Quoted 3 lines

This may not be as important now since the byte is never given back to the user and is only used when printing out status via the user_events_status file in text form.

So, it is confusing because of STATUS_MASK() is bits 0,1,2 and we are only using bits 0 and 1, with a OTHER bit at bit 7. And it would be good to use the BIT() macro.

Ah, I see the confusion. Sorry.

EVENT_STATUS_* are internal bits that aren't used with STATUS_MASK or STATUS_BYTE. It's only used to set and check the user event status byte for checking if anything is attached and outputting which probe is connected within the kernel side.

STATUS_BYTE and STATUS_MASK take a bit in a bitmap and figure out which byte in the status mapping should be used and which bit in that byte should be set/reset (mask) when it's enabled/disabled via a probe. Both the user and kernel need to align on this logic.

IE: Bits above the lower 3 of the index/bit of the event to enable is the byte and the lower 3 bits (& 7) is the actual bit to set.

For example if the user_event with the index 1024 is enabled, we need to figure out which byte and bit represents that event when a probe is attached.

I got into detail of this in the documentation for both a byte and long wise checking of these values.

Hope that helps explain it.

Quoted 3 lines

Is STATUS_OTHER suppose to be part of STATUS_MASK()?

-- Steve

Thanks,
-Beau

maintainer
#11 lore raw

Re: [PATCH v2 6/7] tracing/user_events: Use bits vs bytes for enabled status page data

Quoted 15 lines · On Tue, 26 Jul 2022 19:01:47 -0700 Beau Belgrave <[email protected]>…

Ah, I see the confusion. Sorry.

EVENT_STATUS_* are internal bits that aren't used with STATUS_MASK or STATUS_BYTE. It's only used to set and check the user event status byte for checking if anything is attached and outputting which probe is connected within the kernel side.

STATUS_BYTE and STATUS_MASK take a bit in a bitmap and figure out which byte in the status mapping should be used and which bit in that byte should be set/reset (mask) when it's enabled/disabled via a probe. Both the user and kernel need to align on this logic.

IE: Bits above the lower 3 of the index/bit of the event to enable is the byte and the lower 3 bits (& 7) is the actual bit to set.

Quoted 8 lines

For example if the user_event with the index 1024 is enabled, we need to figure out which byte and bit represents that event when a probe is attached.

I got into detail of this in the documentation for both a byte and long wise checking of these values.

Hope that helps explain it.

Yes, but that should be in the comments above the code.

-- Steve

#12 lore raw

Re: [PATCH v2 6/7] tracing/user_events: Use bits vs bytes for enabled status page data

Quoted 22 lines · Steven Rostedt
Quoted 19 lines · On Tue, 26 Jul 2022 19:01:47 -0700 Beau Belgrave <[email protected]>…

Ah, I see the confusion. Sorry.

EVENT_STATUS_* are internal bits that aren't used with STATUS_MASK or STATUS_BYTE. It's only used to set and check the user event status byte for checking if anything is attached and outputting which probe is connected within the kernel side.

STATUS_BYTE and STATUS_MASK take a bit in a bitmap and figure out which byte in the status mapping should be used and which bit in that byte should be set/reset (mask) when it's enabled/disabled via a probe. Both the user and kernel need to align on this logic.

IE: Bits above the lower 3 of the index/bit of the event to enable is the byte and the lower 3 bits (& 7) is the actual bit to set.

For example if the user_event with the index 1024 is enabled, we need to figure out which byte and bit represents that event when a probe is attached.

Quoted 7 lines
Quoted 4 lines

I got into detail of this in the documentation for both a byte and long wise checking of these values.

Hope that helps explain it.

Yes, but that should be in the comments above the code.

Will do, I will also change to use the BIT() macro as you suggested.

Quoted 1 line

-- Steve

Thanks,
-Beau

#13 lore raw

PATCH v2 7/7 tracing/user_events: Update ABI documentation to align to bits vs bytes

In reply to Beau Belgrave

Update the documentation to reflect the new ABI requirements and how to use the byte index with the mask properly to check event status.

---
 Documentation/trace/user_events.rst | 86 +++++++++++++++++++----------
 1 file changed, 58 insertions(+), 28 deletions(-)
Documentation/trace/user_events.rst +58 -28
index c180936f49fc..9f181f342a70 100644
@@ -20,14 +20,14 @@ dynamic_events is the same as the ioctl with the u: prefix applied.
 
 Typically programs will register a set of events that they wish to expose to
 tools that can read trace_events (such as ftrace and perf). The registration
-process gives back two ints to the program for each event. The first int is the
-status index. This index describes which byte in the
+process gives back two ints to the program for each event. The first int is
+the status bit. This describes which bit in little-endian format in the
 /sys/kernel/debug/tracing/user_events_status file represents this event. The
-second int is the write index. This index describes the data when a write() or
+second int is the write index which describes the data when a write() or
 writev() is called on the /sys/kernel/debug/tracing/user_events_data file.
 
-The structures referenced in this document are contained with the
-/include/uap/linux/user_events.h file in the source tree.
+The structures referenced in this document are contained within the
+/include/uapi/linux/user_events.h file in the source tree.
 
 **NOTE:** *Both user_events_status and user_events_data are under the tracefs
 filesystem and may be mounted at different paths than above.*
@@ -38,18 +38,18 @@ Registering within a user process is done via ioctl() out to the
 /sys/kernel/debug/tracing/user_events_data file. The command to issue is
 DIAG_IOCSREG.
 
-This command takes a struct user_reg as an argument::
+This command takes a packed struct user_reg as an argument::
 
   struct user_reg {
         u32 size;
         u64 name_args;
-        u32 status_index;
+        u32 status_bit;
         u32 write_index;
   };
 
 The struct user_reg requires two inputs, the first is the size of the structure
 to ensure forward and backward compatibility. The second is the command string
-to issue for registering. Upon success two outputs are set, the status index
+to issue for registering. Upon success two outputs are set, the status bit
 and the write index.
 
 User based events show up under tracefs like any other event under the
@@ -111,15 +111,56 @@ in realtime. This allows user programs to only incur the cost of the write() or
 writev() calls when something is actively attached to the event.
 
 User programs call mmap() on /sys/kernel/debug/tracing/user_events_status to
-check the status for each event that is registered. The byte to check in the
-file is given back after the register ioctl() via user_reg.status_index.
+check the status for each event that is registered. The bit to check in the
+file is given back after the register ioctl() via user_reg.status_bit. The bit
+is always in little-endian format. Programs can check if the bit is set either
+using a byte-wise index with a mask or a long-wise index with a little-endian
+mask.
+
 Currently the size of user_events_status is a single page, however, custom
 kernel configurations can change this size to allow more user based events. In
 all cases the size of the file is a multiple of a page size.
 
-For example, if the register ioctl() gives back a status_index of 3 you would
-check byte 3 of the returned mmap data to see if anything is attached to that
-event.
+For example, if the register ioctl() gives back a status_bit of 3 you would
+check byte 0 (3 / 8) of the returned mmap data and then AND the result with 8
+(1 << (3 % 8)) to see if anything is attached to that event.
+
+A byte-wise index check is performed as follows::
+
+  int index, mask;
+  char *status_page;
+
+  index = status_bit / 8;
+  mask = 1 << (status_bit % 8);
+
+  ...
+
+  if (status_page[index] & mask) {
+        /* Enabled */
+  }
+
+A long-wise index check is performed as follows::
+
+  #include <asm/bitsperlong.h>
+  #include <endian.h>
+
+  #if __BITS_PER_LONG == 64
+  #define endian_swap(x) htole64(x)
+  #else
+  #define endian_swap(x) htole32(x)
+  #endif
+
+  long index, mask, *status_page;
+
+  index = status_bit / __BITS_PER_LONG;
+  mask = 1L << (status_bit % __BITS_PER_LONG);
+  mask = endian_swap(mask);
+
+  ...
+
+  if (status_page[index] & mask) {
+        /* Enabled */
+  }
 
 Administrators can easily check the status of all registered events by reading
 the user_events_status file directly via a terminal. The output is as follows::
@@ -137,7 +178,7 @@ For example, on a system that has a single event the output looks like this::
 
   Active: 1
   Busy: 0
-  Max: 4096
+  Max: 32768
 
 If a user enables the user event via ftrace, the output would change to this::
 
@@ -145,21 +186,10 @@ If a user enables the user event via ftrace, the output would change to this::
 
   Active: 1
   Busy: 1
-  Max: 4096
-
-**NOTE:** *A status index of 0 will never be returned. This allows user
-programs to have an index that can be used on error cases.*
-
-Status Bits
-^^^^^^^^^^^
-The byte being checked will be non-zero if anything is attached. Programs can
-check specific bits in the byte to see what mechanism has been attached.
-
-The following values are defined to aid in checking what has been attached:
-
-**EVENT_STATUS_FTRACE** - Bit set if ftrace has been attached (Bit 0).
+  Max: 32768
 
-**EVENT_STATUS_PERF** - Bit set if perf has been attached (Bit 1).
+**NOTE:** *A status bit of 0 will never be returned. This allows user programs
+to have a bit that can be used on error cases.*
 
 Writing Data
 ------------
Signature · 1 line
2.25.1