2023-04-09 17:41:33

by Vincent Donnefort

[permalink] [raw]
Subject: [PATCH v3 0/2] Introducing trace buffer mapping by user-space

The tracing ring-buffers can be stored on disk or sent to network without any
copy via splice. However the later doesn't allow real time processing of the
traces. A solution is to give access to userspace to the ring-buffer pages
directly via a mapping. A piece of software can now become a reader of the
ring-buffer, and drive a consuming or non-consuming read in a similar fashion to
what trace and trace_pipe offer.

Attached to this cover letter an example of consuming read for a ring-buffer,
using libtracefs.

Vincent

v2 -> v3:
* Remove data page list (for non-consuming read)
** Implies removing order > 0 meta-page
* Add a new meta page field ->read
* Rename ring_buffer_meta_page_header into ring_buffer_meta_header

v1 -> v2:
* Hide data_pages from the userspace struct
* Fix META_PAGE_MAX_PAGES
* Support for order > 0 meta-page
* Add missing page->mapping.

Vincent Donnefort (2):
ring-buffer: Introducing ring-buffer mapping functions
tracing: Allow user-space mapping of the ring-buffer

include/linux/ring_buffer.h | 7 +
include/uapi/linux/trace_mmap.h | 26 +++
kernel/trace/ring_buffer.c | 302 +++++++++++++++++++++++++++++++-
kernel/trace/trace.c | 72 +++++++-
4 files changed, 400 insertions(+), 7 deletions(-)
create mode 100644 include/uapi/linux/trace_mmap.h

--

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <stdarg.h>
#include <signal.h>
#include <errno.h>
#include <unistd.h>
#include <tracefs.h>
#include <kbuffer.h>
#include <event-parse.h>

#include <asm/types.h>
#include <sys/mman.h>
#include <sys/ioctl.h>

#define TRACE_MMAP_IOCTL_GET_READER_PAGE _IO('T', 0x1)

struct ring_buffer_meta_header {
__u64 entries;
__u64 overrun;
__u32 pages_touched;
__u32 meta_page_size;
__u32 nr_data_pages; /* Number of pages in the ring-buffer */
__u32 reader_page; /* Reader page ID, from 0 to nr_data_pages - 1 */
__u32 read; /* Number of bytes read on the current reader page */
};

/* Need to access private struct to save counters */
struct kbuffer {
unsigned long long timestamp;
long long lost_events;
unsigned long flags;
void *subbuffer;
void *data;
unsigned int index;
unsigned int curr;
unsigned int next;
unsigned int size;
unsigned int start;
unsigned int first;

unsigned int (*read_4)(void *ptr);
unsigned long long (*read_8)(void *ptr);
unsigned long long (*read_long)(struct kbuffer *kbuf, void *ptr);
int (*next_event)(struct kbuffer *kbuf);
};

static char *argv0;
static bool need_exit;

static char *get_this_name(void)
{
static char *this_name;
char *arg;
char *p;

if (this_name)
return this_name;

arg = argv0;
p = arg+strlen(arg);

while (p >= arg && *p != '/')
p--;
p++;

this_name = p;
return p;
}

static void __vdie(const char *fmt, va_list ap, int err)
{
int ret = errno;
char *p = get_this_name();

if (err && errno)
perror(p);
else
ret = -1;

fprintf(stderr, " ");
vfprintf(stderr, fmt, ap);

fprintf(stderr, "\n");
exit(ret);
}

void pdie(const char *fmt, ...)
{
va_list ap;

va_start(ap, fmt);
__vdie(fmt, ap, 1);
va_end(ap);
}

static void read_page(struct tep_handle *tep, struct kbuffer *kbuf,
void *data, int page)
{
static struct trace_seq seq;
struct tep_record record;

if (seq.buffer)
trace_seq_reset(&seq);
else
trace_seq_init(&seq);

while ((record.data = kbuffer_read_event(kbuf, &record.ts))) {
kbuffer_next_event(kbuf, NULL);
tep_print_event(tep, &seq, &record,
"%s-%d %9d\t%s\n", TEP_PRINT_COMM,
TEP_PRINT_PID, TEP_PRINT_TIME, TEP_PRINT_NAME);
trace_seq_do_printf(&seq);
trace_seq_reset(&seq);
}
}

static int next_reader_page(int fd, struct ring_buffer_meta_header *meta,
struct kbuffer *kbuf)
{
int prev_reader_page = meta->reader_page;

if (ioctl(fd, TRACE_MMAP_IOCTL_GET_READER_PAGE) < 0)
pdie("ioctl");

return meta->reader_page;
}

static void signal_handler(int unused)
{
printf("Exit!\n");
need_exit = true;
}

int main(int argc, char **argv)
{
int page_size, meta_len, data_len, page, fd, start = -1;
unsigned long prev_read, prev_entries;
struct ring_buffer_meta_header *map;
unsigned long long prev_timestamp;
struct tep_handle *tep;
struct kbuffer *kbuf;
void *meta, *data;
char *buf, path[32];
int cpu;

if (argc != 2)
return -EINVAL;

argv0 = argv[0];
cpu = atoi(argv[1]);
snprintf(path, 32, "per_cpu/cpu%d/trace_pipe_raw", cpu);

tep = tracefs_local_events(NULL);
kbuf = tep_kbuffer(tep);
page_size = getpagesize();

fd = tracefs_instance_file_open(NULL, path, O_RDONLY);
if (fd < 0)
pdie("raw");

meta = mmap(NULL, page_size, PROT_READ, MAP_SHARED, fd, 0);
if (meta == MAP_FAILED)
pdie("mmap");
map = (struct ring_buffer_meta_header *)meta;
meta_len = map->meta_page_size;

printf("entries: %llu\n", map->entries);
printf("overrun: %llu\n", map->overrun);
printf("pages_touched: %u\n", map->pages_touched);
printf("reader_page: %u\n", map->reader_page);
printf("nr_data_pages: %u\n\n", map->nr_data_pages);

data_len = page_size * map->nr_data_pages;
data = mmap(NULL, data_len, PROT_READ, MAP_SHARED, fd, meta_len);
if (data == MAP_FAILED)
pdie("mmap data");

signal(SIGINT, signal_handler);

page = map->reader_page;
again:
prev_read = map->read;

do {
kbuffer_load_subbuffer(kbuf, data + page_size * page);

kbuf->curr = kbuf->index = kbuf->next = prev_read;
kbuffer_next_event(kbuf, NULL);

if (page != start)
printf("READER PAGE: %d\n", map->reader_page);
else
kbuf->timestamp = prev_timestamp;

start = page;

read_page(tep, kbuf, data, page);
} while ((page = next_reader_page(fd, meta, kbuf)) != start);

prev_timestamp = kbuf->timestamp;
prev_entries = map->entries;

while (prev_entries == *(volatile __u64 *)&map->entries && !need_exit)
usleep(100000);

if (!need_exit)
goto again;

munmap(data, data_len);
munmap(meta, page_size);
close(fd);

return 0;
}

--
2.40.0.577.gac1e443424-goog


2023-04-09 17:42:13

by Vincent Donnefort

[permalink] [raw]
Subject: [PATCH v3 1/2] ring-buffer: Introducing ring-buffer mapping functions

In preparation for allowing the user-space to map a ring-buffer, add
a set of mapping functions:

ring_buffer_{map,unmap}()
ring_buffer_map_fault()

And controls on the ring-buffer:

ring_buffer_get_reader_page() /* swap reader and head */

Mapping the ring-buffer also involves:

A unique ID for each page of the ring-buffer, as currently the pages
are only identified through their in-kernel VA.

A meta-page, where are stored statistics about the ring-buffer and
a page IDs list, ordered. A field gives what page is the reader
one and one to gives where the ring-buffer starts in the list of data
pages.

The linear mapping exposes the meta-page, and each page of the
ring-buffer, ordered following their unique ID, assigned during the
first mapping.

Once mapped, no page can get in or out of the ring-buffer: the buffer
size will remain unmodified and the splice enabling functions will in
reality simply memcpy the data instead of swapping pages.

Signed-off-by: Vincent Donnefort <[email protected]>

diff --git a/include/linux/ring_buffer.h b/include/linux/ring_buffer.h
index 782e14f62201..5b3563507842 100644
--- a/include/linux/ring_buffer.h
+++ b/include/linux/ring_buffer.h
@@ -6,6 +6,8 @@
#include <linux/seq_file.h>
#include <linux/poll.h>

+#include <uapi/linux/trace_mmap.h>
+
struct trace_buffer;
struct ring_buffer_iter;

@@ -211,4 +213,9 @@ int trace_rb_cpu_prepare(unsigned int cpu, struct hlist_node *node);
#define trace_rb_cpu_prepare NULL
#endif

+int ring_buffer_map(struct trace_buffer *buffer, int cpu);
+int ring_buffer_unmap(struct trace_buffer *buffer, int cpu);
+struct page *ring_buffer_map_fault(struct trace_buffer *buffer, int cpu,
+ unsigned long pgoff);
+int ring_buffer_get_reader_page(struct trace_buffer *buffer, int cpu);
#endif /* _LINUX_RING_BUFFER_H */
diff --git a/include/uapi/linux/trace_mmap.h b/include/uapi/linux/trace_mmap.h
new file mode 100644
index 000000000000..b89e0c8562b5
--- /dev/null
+++ b/include/uapi/linux/trace_mmap.h
@@ -0,0 +1,24 @@
+/* SPDX-License-Identifier: GPL-2.0 WITH Linux-syscall-note */
+#ifndef _UAPI_TRACE_MMAP_H_
+#define _UAPI_TRACE_MMAP_H_
+
+#include <asm/bitsperlong.h>
+
+#include <linux/types.h>
+
+struct ring_buffer_meta_header {
+#if __BITS_PER_LONG == 64
+ __u64 entries;
+ __u64 overrun;
+#else
+ __u32 entries;
+ __u32 overrun;
+#endif
+ __u32 pages_touched;
+ __u32 meta_page_size;
+ __u32 nr_data_pages; /* Number of pages in the ring-buffer */
+ __u32 reader_page; /* Reader page ID, from 0 to nr_data_pages - 1 */
+ __u32 read; /* Number of bytes read on the current reader page */
+};
+
+#endif /* _UAPI_TRACE_MMAP_H_ */
diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
index af50d931b020..529702af7615 100644
--- a/kernel/trace/ring_buffer.c
+++ b/kernel/trace/ring_buffer.c
@@ -332,6 +332,7 @@ struct buffer_page {
local_t entries; /* entries on this page */
unsigned long real_end; /* real end of data */
struct buffer_data_page *page; /* Actual data page */
+ u32 id; /* ID for external mapping */
};

/*
@@ -529,6 +530,12 @@ struct ring_buffer_per_cpu {
rb_time_t before_stamp;
u64 event_stamp[MAX_NEST];
u64 read_stamp;
+
+ int mapped;
+ struct mutex mapping_lock;
+ unsigned long *page_ids; /* ID to addr */
+ struct ring_buffer_meta_header *meta_page;
+
/* ring buffer pages to update, > 0 to add, < 0 to remove */
long nr_pages_to_update;
struct list_head new_pages; /* new pages to add */
@@ -1567,6 +1574,13 @@ static void rb_tail_page_update(struct ring_buffer_per_cpu *cpu_buffer,
/* Again, either we update tail_page or an interrupt does */
(void)cmpxchg(&cpu_buffer->tail_page, tail_page, next_page);
}
+
+ if (READ_ONCE(cpu_buffer->mapped)) {
+ /* Ensure the meta_page is ready */
+ smp_rmb();
+ WRITE_ONCE(cpu_buffer->meta_page->pages_touched,
+ local_read(&cpu_buffer->pages_touched));
+ }
}

static int rb_check_bpage(struct ring_buffer_per_cpu *cpu_buffer,
@@ -1735,6 +1749,7 @@ rb_allocate_cpu_buffer(struct trace_buffer *buffer, long nr_pages, int cpu)
init_irq_work(&cpu_buffer->irq_work.work, rb_wake_up_waiters);
init_waitqueue_head(&cpu_buffer->irq_work.waiters);
init_waitqueue_head(&cpu_buffer->irq_work.full_waiters);
+ mutex_init(&cpu_buffer->mapping_lock);

bpage = kzalloc_node(ALIGN(sizeof(*bpage), cache_line_size()),
GFP_KERNEL, cpu_to_node(cpu));
@@ -2173,7 +2188,6 @@ int ring_buffer_resize(struct trace_buffer *buffer, unsigned long size,
/* prevent another thread from changing buffer sizes */
mutex_lock(&buffer->mutex);

-
if (cpu_id == RING_BUFFER_ALL_CPUS) {
/*
* Don't succeed if resizing is disabled, as a reader might be
@@ -2523,6 +2537,13 @@ rb_handle_head_page(struct ring_buffer_per_cpu *cpu_buffer,
local_sub(BUF_PAGE_SIZE, &cpu_buffer->entries_bytes);
local_inc(&cpu_buffer->pages_lost);

+ if (READ_ONCE(cpu_buffer->mapped)) {
+ /* Ensure the meta_page is ready */
+ smp_rmb();
+ WRITE_ONCE(cpu_buffer->meta_page->overrun,
+ local_read(&cpu_buffer->overrun));
+ }
+
/*
* The entries will be zeroed out when we move the
* tail page.
@@ -3179,6 +3200,14 @@ static inline void rb_event_discard(struct ring_buffer_event *event)
static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer)
{
local_inc(&cpu_buffer->entries);
+
+ if (READ_ONCE(cpu_buffer->mapped)) {
+ /* Ensure the meta_page is ready */
+ smp_rmb();
+ WRITE_ONCE(cpu_buffer->meta_page->entries,
+ local_read(&cpu_buffer->entries));
+ }
+
rb_end_commit(cpu_buffer);
}

@@ -3482,7 +3511,7 @@ static void check_buffer(struct ring_buffer_per_cpu *cpu_buffer,
return;

/*
- * If this interrupted another event,
+ * If this interrupted another event,
*/
if (atomic_inc_return(this_cpu_ptr(&checking)) != 1)
goto out;
@@ -4651,6 +4680,11 @@ rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
cpu_buffer->reader_page = reader;
cpu_buffer->reader_page->read = 0;

+ if (cpu_buffer->mapped) {
+ WRITE_ONCE(cpu_buffer->meta_page->read, 0);
+ WRITE_ONCE(cpu_buffer->meta_page->reader_page, reader->id);
+ }
+
if (overwrite != cpu_buffer->last_overrun) {
cpu_buffer->lost_events = overwrite - cpu_buffer->last_overrun;
cpu_buffer->last_overrun = overwrite;
@@ -4717,6 +4751,10 @@ static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer)

length = rb_event_length(event);
cpu_buffer->reader_page->read += length;
+
+ if (cpu_buffer->mapped)
+ WRITE_ONCE(cpu_buffer->meta_page->read,
+ cpu_buffer->reader_page->read);
}

static void rb_advance_iter(struct ring_buffer_iter *iter)
@@ -5285,6 +5323,13 @@ rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer)
cpu_buffer->lost_events = 0;
cpu_buffer->last_overrun = 0;

+ if (cpu_buffer->mapped) {
+ WRITE_ONCE(cpu_buffer->meta_page->entries, 0);
+ WRITE_ONCE(cpu_buffer->meta_page->pages_touched, 0);
+ WRITE_ONCE(cpu_buffer->meta_page->overrun, 0);
+ WRITE_ONCE(cpu_buffer->meta_page->read, 0);
+ }
+
rb_head_page_activate(cpu_buffer);
}

@@ -5489,6 +5534,11 @@ int ring_buffer_swap_cpu(struct trace_buffer *buffer_a,
cpu_buffer_a = buffer_a->buffers[cpu];
cpu_buffer_b = buffer_b->buffers[cpu];

+ if (READ_ONCE(cpu_buffer_a->mapped) || READ_ONCE(cpu_buffer_b->mapped)) {
+ ret = -EBUSY;
+ goto out;
+ }
+
/* At least make sure the two buffers are somewhat the same */
if (cpu_buffer_a->nr_pages != cpu_buffer_b->nr_pages)
goto out;
@@ -5722,7 +5772,8 @@ int ring_buffer_read_page(struct trace_buffer *buffer,
* Otherwise, we can simply swap the page with the one passed in.
*/
if (read || (len < (commit - read)) ||
- cpu_buffer->reader_page == cpu_buffer->commit_page) {
+ cpu_buffer->reader_page == cpu_buffer->commit_page ||
+ cpu_buffer->mapped) {
struct buffer_data_page *rpage = cpu_buffer->reader_page->page;
unsigned int rpos = read;
unsigned int pos = 0;
@@ -5839,6 +5890,251 @@ int ring_buffer_read_page(struct trace_buffer *buffer,
}
EXPORT_SYMBOL_GPL(ring_buffer_read_page);

+static void rb_free_page_ids(struct ring_buffer_per_cpu *cpu_buffer)
+{
+ int i;
+
+ for (i = 0; i < cpu_buffer->nr_pages + 1; i++)
+ virt_to_page(cpu_buffer->page_ids[i])->mapping = NULL;
+
+ kfree(cpu_buffer->page_ids);
+ cpu_buffer->page_ids = NULL;
+}
+
+static int rb_alloc_meta_page(struct ring_buffer_per_cpu *cpu_buffer)
+{
+ if (cpu_buffer->meta_page)
+ return 0;
+
+ cpu_buffer->meta_page = page_to_virt(alloc_page(GFP_USER));
+ if (!cpu_buffer->meta_page)
+ return -ENOMEM;
+
+ return 0;
+}
+
+static void rb_free_meta_page(struct ring_buffer_per_cpu *cpu_buffer)
+{
+ unsigned long addr = (unsigned long)cpu_buffer->meta_page;
+
+ virt_to_page(addr)->mapping = NULL;
+ free_page(addr);
+ cpu_buffer->meta_page = NULL;
+}
+
+static void rb_setup_ids_meta_page(struct ring_buffer_per_cpu *cpu_buffer,
+ unsigned long *page_ids)
+{
+ struct ring_buffer_meta_header *meta = cpu_buffer->meta_page;
+ unsigned int nr_data_pages = cpu_buffer->nr_pages + 1;
+ struct buffer_page *first_page, *bpage;
+ int id = 0;
+
+ page_ids[id] = (unsigned long)cpu_buffer->reader_page->page;
+ cpu_buffer->reader_page->id = id++;
+
+ first_page = bpage = rb_set_head_page(cpu_buffer);
+ do {
+ if (id >= nr_data_pages) {
+ WARN_ON(1);
+ break;
+ }
+
+ page_ids[id] = (unsigned long)bpage->page;
+ bpage->id = id;
+
+ rb_inc_page(&bpage);
+ id++;
+ } while (bpage != first_page);
+
+ /* install page ID to kern VA translation */
+ cpu_buffer->page_ids = page_ids;
+
+ meta->entries = 0;
+ meta->overrun = 0;
+ meta->pages_touched = 0;
+ meta->meta_page_size = PAGE_SIZE;
+ meta->nr_data_pages = nr_data_pages;
+ meta->reader_page = cpu_buffer->reader_page->id;
+ meta->read = cpu_buffer->reader_page->read;
+}
+
+static inline struct ring_buffer_per_cpu *
+rb_get_mapped_buffer(struct trace_buffer *buffer, int cpu)
+{
+ struct ring_buffer_per_cpu *cpu_buffer;
+
+ if (!cpumask_test_cpu(cpu, buffer->cpumask))
+ return ERR_PTR(-EINVAL);
+
+ cpu_buffer = buffer->buffers[cpu];
+
+ mutex_lock(&cpu_buffer->mapping_lock);
+
+ if (!cpu_buffer->mapped) {
+ mutex_unlock(&cpu_buffer->mapping_lock);
+ return ERR_PTR(-ENODEV);
+ }
+
+ return cpu_buffer;
+}
+
+static inline void rb_put_mapped_buffer(struct ring_buffer_per_cpu *cpu_buffer)
+{
+ mutex_unlock(&cpu_buffer->mapping_lock);
+}
+
+int ring_buffer_map(struct trace_buffer *buffer, int cpu)
+{
+ struct ring_buffer_per_cpu *cpu_buffer;
+ unsigned long flags, *page_ids;
+ int err = 0;
+
+ if (!cpumask_test_cpu(cpu, buffer->cpumask))
+ return -EINVAL;
+
+ cpu_buffer = buffer->buffers[cpu];
+
+ mutex_lock(&cpu_buffer->mapping_lock);
+
+ if (cpu_buffer->mapped) {
+ WRITE_ONCE(cpu_buffer->mapped, cpu_buffer->mapped + 1);
+ goto unlock;
+ }
+
+ /* prevent another thread from changing buffer sizes */
+ mutex_lock(&buffer->mutex);
+ atomic_inc(&cpu_buffer->resize_disabled);
+ mutex_unlock(&buffer->mutex);
+
+ err = rb_alloc_meta_page(cpu_buffer);
+ if (err) {
+ atomic_dec(&cpu_buffer->resize_disabled);
+ goto unlock;
+ }
+
+ /* page_ids include the reader page while nr_pages does not */
+ page_ids = kzalloc(sizeof(*page_ids) * (cpu_buffer->nr_pages + 1),
+ GFP_KERNEL);
+ if (!page_ids) {
+ rb_free_meta_page(cpu_buffer);
+ atomic_dec(&cpu_buffer->resize_disabled);
+ err = -ENOMEM;
+ goto unlock;
+ }
+
+ /*
+ * Lock all readers to block any page swap until the page IDs are
+ * assigned.
+ */
+ raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
+
+ rb_setup_ids_meta_page(cpu_buffer, page_ids);
+ /*
+ * Ensure the writer will observe the meta-page before
+ * cpu_buffer->mapped.
+ */
+ smp_wmb();
+ WRITE_ONCE(cpu_buffer->mapped, 1);
+
+ /* Init meta_page values unless the writer did it already */
+ cmpxchg(&cpu_buffer->meta_page->entries, 0,
+ local_read(&cpu_buffer->entries));
+ cmpxchg(&cpu_buffer->meta_page->overrun, 0,
+ local_read(&cpu_buffer->overrun));
+ cmpxchg(&cpu_buffer->meta_page->pages_touched, 0,
+ local_read(&cpu_buffer->pages_touched));
+
+ raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
+unlock:
+ mutex_unlock(&cpu_buffer->mapping_lock);
+
+ return err;
+}
+
+int ring_buffer_unmap(struct trace_buffer *buffer, int cpu)
+{
+ struct ring_buffer_per_cpu *cpu_buffer;
+ int err = 0;
+
+ if (!cpumask_test_cpu(cpu, buffer->cpumask))
+ return -EINVAL;
+
+ cpu_buffer = buffer->buffers[cpu];
+
+ mutex_lock(&cpu_buffer->mapping_lock);
+
+ if (!cpu_buffer->mapped) {
+ err = -ENODEV;
+ goto unlock;
+ }
+
+ WRITE_ONCE(cpu_buffer->mapped, cpu_buffer->mapped - 1);
+ if (!cpu_buffer->mapped) {
+ /* Wait the writer and readers to observe !mapped */
+ synchronize_rcu();
+
+ rb_free_page_ids(cpu_buffer);
+ rb_free_meta_page(cpu_buffer);
+ atomic_dec(&cpu_buffer->resize_disabled);
+ }
+
+unlock:
+ mutex_unlock(&cpu_buffer->mapping_lock);
+
+ return err;
+}
+
+/*
+ * +--------------+
+ * | meta page | pgoff=0
+ * +--------------+
+ * | data page1 | page_ids=0
+ * +--------------+
+ * | data page2 | page_ids=1
+ * ...
+ */
+struct page *ring_buffer_map_fault(struct trace_buffer *buffer, int cpu,
+ unsigned long pgoff)
+{
+ struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
+
+ if (!pgoff)
+ return virt_to_page((void *)cpu_buffer->meta_page);
+
+ pgoff--;
+ if (pgoff > cpu_buffer->nr_pages)
+ return NULL;
+
+ return virt_to_page(cpu_buffer->page_ids[pgoff]);
+}
+
+int ring_buffer_get_reader_page(struct trace_buffer *buffer, int cpu)
+{
+ struct ring_buffer_per_cpu *cpu_buffer;
+ unsigned long flags;
+
+ cpu_buffer = rb_get_mapped_buffer(buffer, cpu);
+ if (IS_ERR(cpu_buffer))
+ return (int)PTR_ERR(cpu_buffer);
+
+ raw_spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
+ if (!rb_per_cpu_empty(cpu_buffer)) {
+ unsigned int reader_size;
+
+ WARN_ON(!rb_get_reader_page(cpu_buffer));
+
+ reader_size = rb_page_size(cpu_buffer->reader_page);
+ cpu_buffer->reader_page->read = reader_size;
+ cpu_buffer->meta_page->read = reader_size;
+ }
+ raw_spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
+
+ rb_put_mapped_buffer(cpu_buffer);
+
+ return 0;
+}
+
/*
* We only allocate new buffers, never free them if the CPU goes down.
* If we were to free the buffer, then the user would lose any trace that was in
--
2.40.0.577.gac1e443424-goog

2023-04-09 18:26:06

by Vincent Donnefort

[permalink] [raw]
Subject: [PATCH v3 2/2] tracing: Allow user-space mapping of the ring-buffer

Currently, user-space extracts data from the ring-buffer via splice,
which is handy for storage or network sharing. However, due to splice
limitations, it is imposible to do real-time analysis without a copy.

A solution for that problem is to let the user-space map the ring-buffer
directly.

The mapping exposed via the per-CPU file trace_pipe_raw. The first page
is the meta-page and is followed by each page of the ring-buffer,
ordered by their unique page ID. It is therefore easy to translate a
page-ID to an offset in the mapping.

* Meta-page -- include/uapi/linux/trace_mmap.h for a description
* Page ID 0
* Page ID 1
...

The mapper must then do what use to be the kernel job: swap the reader
with the head. This is done with a newly introduced ioctl:
TRACE_MMAP_IOCTL_GET_READER_PAGE.

Signed-off-by: Vincent Donnefort <[email protected]>

diff --git a/include/uapi/linux/trace_mmap.h b/include/uapi/linux/trace_mmap.h
index b89e0c8562b5..70b97c0ea888 100644
--- a/include/uapi/linux/trace_mmap.h
+++ b/include/uapi/linux/trace_mmap.h
@@ -21,4 +21,6 @@ struct ring_buffer_meta_header {
__u32 read; /* Number of bytes read on the current reader page */
};

+#define TRACE_MMAP_IOCTL_GET_READER_PAGE _IO('T', 0x1)
+
#endif /* _UAPI_TRACE_MMAP_H_ */
diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c
index 45551c7b4c36..8a64ca87afaf 100644
--- a/kernel/trace/trace.c
+++ b/kernel/trace/trace.c
@@ -6399,7 +6399,7 @@ static void tracing_set_nop(struct trace_array *tr)
{
if (tr->current_trace == &nop_trace)
return;
-
+
tr->current_trace->enabled--;

if (tr->current_trace->reset)
@@ -8432,15 +8432,21 @@ tracing_buffers_splice_read(struct file *file, loff_t *ppos,
return ret;
}

-/* An ioctl call with cmd 0 to the ring buffer file will wake up all waiters */
static long tracing_buffers_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
{
struct ftrace_buffer_info *info = file->private_data;
struct trace_iterator *iter = &info->iter;

- if (cmd)
- return -ENOIOCTLCMD;
+ if (cmd == TRACE_MMAP_IOCTL_GET_READER_PAGE)
+ return ring_buffer_get_reader_page(iter->array_buffer->buffer,
+ iter->cpu_file);
+ else if (cmd)
+ return -ENOTTY;

+ /*
+ * An ioctl call with cmd 0 to the ring buffer file will wake up all
+ * waiters
+ */
mutex_lock(&trace_types_lock);

iter->wait_index++;
@@ -8453,6 +8459,63 @@ static long tracing_buffers_ioctl(struct file *file, unsigned int cmd, unsigned
return 0;
}

+static vm_fault_t tracing_buffers_mmap_fault(struct vm_fault *vmf)
+{
+ struct ftrace_buffer_info *info = vmf->vma->vm_file->private_data;
+ struct trace_iterator *iter = &info->iter;
+ vm_fault_t ret = VM_FAULT_SIGBUS;
+ struct page *page;
+
+ page = ring_buffer_map_fault(iter->array_buffer->buffer, iter->cpu_file,
+ vmf->pgoff);
+ if (!page)
+ return ret;
+
+ vmf->page = page;
+
+ get_page(vmf->page);
+ vmf->page->mapping = vmf->vma->vm_file->f_mapping;
+ vmf->page->index = vmf->pgoff;
+
+ return 0;
+}
+
+static void tracing_buffers_mmap_close(struct vm_area_struct *vma)
+{
+ struct ftrace_buffer_info *info = vma->vm_file->private_data;
+ struct trace_iterator *iter = &info->iter;
+
+ ring_buffer_unmap(iter->array_buffer->buffer, iter->cpu_file);
+}
+
+static void tracing_buffers_mmap_open(struct vm_area_struct *vma)
+{
+ struct ftrace_buffer_info *info = vma->vm_file->private_data;
+ struct trace_iterator *iter = &info->iter;
+
+ WARN_ON(ring_buffer_map(iter->array_buffer->buffer, iter->cpu_file));
+}
+
+static const struct vm_operations_struct tracing_buffers_vmops = {
+ .open = tracing_buffers_mmap_open,
+ .close = tracing_buffers_mmap_close,
+ .fault = tracing_buffers_mmap_fault,
+};
+
+static int tracing_buffers_mmap(struct file *filp, struct vm_area_struct *vma)
+{
+ struct ftrace_buffer_info *info = filp->private_data;
+ struct trace_iterator *iter = &info->iter;
+
+ if (vma->vm_flags & VM_WRITE)
+ return -EPERM;
+
+ vm_flags_mod(vma, VM_DONTCOPY | VM_DONTDUMP, VM_MAYWRITE);
+ vma->vm_ops = &tracing_buffers_vmops;
+
+ return ring_buffer_map(iter->array_buffer->buffer, iter->cpu_file);
+}
+
static const struct file_operations tracing_buffers_fops = {
.open = tracing_buffers_open,
.read = tracing_buffers_read,
@@ -8461,6 +8524,7 @@ static const struct file_operations tracing_buffers_fops = {
.splice_read = tracing_buffers_splice_read,
.unlocked_ioctl = tracing_buffers_ioctl,
.llseek = no_llseek,
+ .mmap = tracing_buffers_mmap,
};

static ssize_t
--
2.40.0.577.gac1e443424-goog