Commit 0e3125c7 authored by Neil Horman's avatar Neil Horman Committed by David S. Miller

packet: Enhance AF_PACKET implementation to not require high order contiguous...

packet: Enhance AF_PACKET implementation to not require high order contiguous memory allocation (v4)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Version 4 of this patch.

Change notes:
1) Removed extra memset.  Didn't think kcalloc added a GFP_ZERO the way kzalloc did :)

Summary:
It was shown to me recently that systems under high load were driven very deep
into swap when tcpdump was run.  The reason this happened was because the
AF_PACKET protocol has a SET_RINGBUFFER socket option that allows the user space
application to specify how many entries an AF_PACKET socket will have and how
large each entry will be.  It seems the default setting for tcpdump is to set
the ring buffer to 32 entries of 64 Kb each, which implies 32 order 5
allocation.  Thats difficult under good circumstances, and horrid under memory
pressure.

I thought it would be good to make that a bit more usable.  I was going to do a
simple conversion of the ring buffer from contigous pages to iovecs, but
unfortunately, the metadata which AF_PACKET places in these buffers can easily
span a page boundary, and given that these buffers get mapped into user space,
and the data layout doesn't easily allow for a change to padding between frames
to avoid that, a simple iovec change is just going to break user space ABI
consistency.

So I've done this, I've added a three tiered mechanism to the af_packet set_ring
socket option.  It attempts to allocate memory in the following order:

1) Using __get_free_pages with GFP_NORETRY set, so as to fail quickly without
digging into swap

2) Using vmalloc

3) Using __get_free_pages with GFP_NORETRY clear, causing us to try as hard as
needed to get the memory

The effect is that we don't disturb the system as much when we're under load,
while still being able to conduct tcpdumps effectively.

Tested successfully by me.
Signed-off-by: default avatarNeil Horman <nhorman@tuxdriver.com>
Acked-by: default avatarEric Dumazet <eric.dumazet@gmail.com>
Acked-by: default avatarMaciej Żenczykowski <zenczykowski@gmail.com>
Reported-by: default avatarMaciej Żenczykowski <zenczykowski@gmail.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
parent 020f01eb
...@@ -61,6 +61,7 @@ ...@@ -61,6 +61,7 @@
#include <linux/kernel.h> #include <linux/kernel.h>
#include <linux/kmod.h> #include <linux/kmod.h>
#include <linux/slab.h> #include <linux/slab.h>
#include <linux/vmalloc.h>
#include <net/net_namespace.h> #include <net/net_namespace.h>
#include <net/ip.h> #include <net/ip.h>
#include <net/protocol.h> #include <net/protocol.h>
...@@ -163,8 +164,14 @@ struct packet_mreq_max { ...@@ -163,8 +164,14 @@ struct packet_mreq_max {
static int packet_set_ring(struct sock *sk, struct tpacket_req *req, static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
int closing, int tx_ring); int closing, int tx_ring);
#define PGV_FROM_VMALLOC 1
struct pgv {
char *buffer;
unsigned char flags;
};
struct packet_ring_buffer { struct packet_ring_buffer {
char **pg_vec; struct pgv *pg_vec;
unsigned int head; unsigned int head;
unsigned int frames_per_block; unsigned int frames_per_block;
unsigned int frame_size; unsigned int frame_size;
...@@ -283,7 +290,8 @@ static void *packet_lookup_frame(struct packet_sock *po, ...@@ -283,7 +290,8 @@ static void *packet_lookup_frame(struct packet_sock *po,
pg_vec_pos = position / rb->frames_per_block; pg_vec_pos = position / rb->frames_per_block;
frame_offset = position % rb->frames_per_block; frame_offset = position % rb->frames_per_block;
h.raw = rb->pg_vec[pg_vec_pos] + (frame_offset * rb->frame_size); h.raw = rb->pg_vec[pg_vec_pos].buffer +
(frame_offset * rb->frame_size);
if (status != __packet_get_status(po, h.raw)) if (status != __packet_get_status(po, h.raw))
return NULL; return NULL;
...@@ -2325,37 +2333,74 @@ static const struct vm_operations_struct packet_mmap_ops = { ...@@ -2325,37 +2333,74 @@ static const struct vm_operations_struct packet_mmap_ops = {
.close = packet_mm_close, .close = packet_mm_close,
}; };
static void free_pg_vec(char **pg_vec, unsigned int order, unsigned int len) static void free_pg_vec(struct pgv *pg_vec, unsigned int order,
unsigned int len)
{ {
int i; int i;
for (i = 0; i < len; i++) { for (i = 0; i < len; i++) {
if (likely(pg_vec[i])) if (likely(pg_vec[i].buffer)) {
free_pages((unsigned long) pg_vec[i], order); if (pg_vec[i].flags & PGV_FROM_VMALLOC)
vfree(pg_vec[i].buffer);
else
free_pages((unsigned long)pg_vec[i].buffer,
order);
pg_vec[i].buffer = NULL;
}
} }
kfree(pg_vec); kfree(pg_vec);
} }
static inline char *alloc_one_pg_vec_page(unsigned long order) static inline char *alloc_one_pg_vec_page(unsigned long order,
unsigned char *flags)
{ {
gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP | __GFP_ZERO | __GFP_NOWARN; char *buffer = NULL;
gfp_t gfp_flags = GFP_KERNEL | __GFP_COMP |
__GFP_ZERO | __GFP_NOWARN | __GFP_NORETRY;
buffer = (char *) __get_free_pages(gfp_flags, order);
if (buffer)
return buffer;
/*
* __get_free_pages failed, fall back to vmalloc
*/
*flags |= PGV_FROM_VMALLOC;
buffer = vmalloc((1 << order) * PAGE_SIZE);
return (char *) __get_free_pages(gfp_flags, order); if (buffer)
return buffer;
/*
* vmalloc failed, lets dig into swap here
*/
*flags = 0;
gfp_flags &= ~__GFP_NORETRY;
buffer = (char *)__get_free_pages(gfp_flags, order);
if (buffer)
return buffer;
/*
* complete and utter failure
*/
return NULL;
} }
static char **alloc_pg_vec(struct tpacket_req *req, int order) static struct pgv *alloc_pg_vec(struct tpacket_req *req, int order)
{ {
unsigned int block_nr = req->tp_block_nr; unsigned int block_nr = req->tp_block_nr;
char **pg_vec; struct pgv *pg_vec;
int i; int i;
pg_vec = kzalloc(block_nr * sizeof(char *), GFP_KERNEL); pg_vec = kcalloc(block_nr, sizeof(struct pgv), GFP_KERNEL);
if (unlikely(!pg_vec)) if (unlikely(!pg_vec))
goto out; goto out;
for (i = 0; i < block_nr; i++) { for (i = 0; i < block_nr; i++) {
pg_vec[i] = alloc_one_pg_vec_page(order); pg_vec[i].buffer = alloc_one_pg_vec_page(order,
if (unlikely(!pg_vec[i])) &pg_vec[i].flags);
if (unlikely(!pg_vec[i].buffer))
goto out_free_pgvec; goto out_free_pgvec;
} }
...@@ -2364,6 +2409,7 @@ static char **alloc_pg_vec(struct tpacket_req *req, int order) ...@@ -2364,6 +2409,7 @@ static char **alloc_pg_vec(struct tpacket_req *req, int order)
out_free_pgvec: out_free_pgvec:
free_pg_vec(pg_vec, order, block_nr); free_pg_vec(pg_vec, order, block_nr);
kfree(pg_vec);
pg_vec = NULL; pg_vec = NULL;
goto out; goto out;
} }
...@@ -2371,7 +2417,7 @@ static char **alloc_pg_vec(struct tpacket_req *req, int order) ...@@ -2371,7 +2417,7 @@ static char **alloc_pg_vec(struct tpacket_req *req, int order)
static int packet_set_ring(struct sock *sk, struct tpacket_req *req, static int packet_set_ring(struct sock *sk, struct tpacket_req *req,
int closing, int tx_ring) int closing, int tx_ring)
{ {
char **pg_vec = NULL; struct pgv *pg_vec = NULL;
struct packet_sock *po = pkt_sk(sk); struct packet_sock *po = pkt_sk(sk);
int was_running, order = 0; int was_running, order = 0;
struct packet_ring_buffer *rb; struct packet_ring_buffer *rb;
...@@ -2533,15 +2579,22 @@ static int packet_mmap(struct file *file, struct socket *sock, ...@@ -2533,15 +2579,22 @@ static int packet_mmap(struct file *file, struct socket *sock,
continue; continue;
for (i = 0; i < rb->pg_vec_len; i++) { for (i = 0; i < rb->pg_vec_len; i++) {
struct page *page = virt_to_page(rb->pg_vec[i]); struct page *page;
void *kaddr = rb->pg_vec[i].buffer;
int pg_num; int pg_num;
for (pg_num = 0; pg_num < rb->pg_vec_pages; for (pg_num = 0; pg_num < rb->pg_vec_pages;
pg_num++, page++) { pg_num++) {
if (rb->pg_vec[i].flags & PGV_FROM_VMALLOC)
page = vmalloc_to_page(kaddr);
else
page = virt_to_page(kaddr);
err = vm_insert_page(vma, start, page); err = vm_insert_page(vma, start, page);
if (unlikely(err)) if (unlikely(err))
goto out; goto out;
start += PAGE_SIZE; start += PAGE_SIZE;
kaddr += PAGE_SIZE;
} }
} }
} }
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment