linux/fs/pipe.c

1212 lines
27 KiB
C
Raw Normal View History

/*
* linux/fs/pipe.c
*
* Copyright (C) 1991, 1992, 1999 Linus Torvalds
*/
#include <linux/mm.h>
#include <linux/file.h>
#include <linux/poll.h>
#include <linux/slab.h>
#include <linux/module.h>
#include <linux/init.h>
#include <linux/fs.h>
#include <linux/log2.h>
#include <linux/mount.h>
#include <linux/magic.h>
#include <linux/pipe_fs_i.h>
#include <linux/uio.h>
#include <linux/highmem.h>
#include <linux/pagemap.h>
#include <linux/audit.h>
#include <linux/syscalls.h>
#include <linux/fcntl.h>
pipe: account to kmemcg Pipes can consume a significant amount of system memory, hence they should be accounted to kmemcg. This patch marks pipe_inode_info and anonymous pipe buffer page allocations as __GFP_ACCOUNT so that they would be charged to kmemcg. Note, since a pipe buffer page can be "stolen" and get reused for other purposes, including mapping to userspace, we clear PageKmemcg thus resetting page->_mapcount and uncharge it in anon_pipe_buf_steal, which is introduced by this patch. A note regarding anon_pipe_buf_steal implementation. We allow to steal the page if its ref count equals 1. It looks racy, but it is correct for anonymous pipe buffer pages, because: - We lock out all other pipe users, because ->steal is called with pipe_lock held, so the page can't be spliced to another pipe from under us. - The page is not on LRU and it never was. - Thus a parallel thread can access it only by PFN. Although this is quite possible (e.g. see page_idle_get_page and balloon_page_isolate) this is not dangerous, because all such functions do is increase page ref count, check if the page is the one they are looking for, and decrease ref count if it isn't. Since our page is clean except for PageKmemcg mark, which doesn't conflict with other _mapcount users, the worst that can happen is we see page_count > 2 due to a transient ref, in which case we false-positively abort ->steal, which is still fine, because ->steal is not guaranteed to succeed. Link: http://lkml.kernel.org/r/20160527150313.GD26059@esperanza Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Minchan Kim <minchan@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-07-27 06:24:33 +08:00
#include <linux/memcontrol.h>
#include <asm/uaccess.h>
#include <asm/ioctls.h>
#include "internal.h"
/*
* The max size that a non-root user is allowed to grow the pipe. Can
* be set by root in /proc/sys/fs/pipe-max-size
*/
unsigned int pipe_max_size = 1048576;
/*
* Minimum pipe size, as required by POSIX
*/
unsigned int pipe_min_size = PAGE_SIZE;
pipe: limit the per-user amount of pages allocated in pipes On no-so-small systems, it is possible for a single process to cause an OOM condition by filling large pipes with data that are never read. A typical process filling 4000 pipes with 1 MB of data will use 4 GB of memory. On small systems it may be tricky to set the pipe max size to prevent this from happening. This patch makes it possible to enforce a per-user soft limit above which new pipes will be limited to a single page, effectively limiting them to 4 kB each, as well as a hard limit above which no new pipes may be created for this user. This has the effect of protecting the system against memory abuse without hurting other users, and still allowing pipes to work correctly though with less data at once. The limit are controlled by two new sysctls : pipe-user-pages-soft, and pipe-user-pages-hard. Both may be disabled by setting them to zero. The default soft limit allows the default number of FDs per process (1024) to create pipes of the default size (64kB), thus reaching a limit of 64MB before starting to create only smaller pipes. With 256 processes limited to 1024 FDs each, this results in 1024*64kB + (256*1024 - 1024) * 4kB = 1084 MB of memory allocated for a user. The hard limit is disabled by default to avoid breaking existing applications that make intensive use of pipes (eg: for splicing). Reported-by: socketpair@gmail.com Reported-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Mitigates: CVE-2013-4312 (Linux 2.0+) Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-01-18 23:36:09 +08:00
/* Maximum allocatable pages per user. Hard limit is unset by default, soft
* matches default values.
*/
unsigned long pipe_user_pages_hard;
unsigned long pipe_user_pages_soft = PIPE_DEF_BUFFERS * INR_OPEN_CUR;
/*
* We use a start+len construction, which provides full use of the
* allocated memory.
* -- Florian Coosmann (FGC)
*
* Reads with count = 0 should always return 0.
* -- Julian Bradfield 1999-06-07.
*
* FIFOs and Pipes now generate SIGIO for both readers and writers.
* -- Jeremy Elson <jelson@circlemud.org> 2001-08-16
*
* pipe_read & write cleanup
* -- Manfred Spraul <manfred@colorfullife.com> 2002-05-09
*/
static void pipe_lock_nested(struct pipe_inode_info *pipe, int subclass)
{
if (pipe->files)
mutex_lock_nested(&pipe->mutex, subclass);
}
void pipe_lock(struct pipe_inode_info *pipe)
{
/*
* pipe_lock() nests non-pipe inode locks (for writing to a file)
*/
pipe_lock_nested(pipe, I_MUTEX_PARENT);
}
EXPORT_SYMBOL(pipe_lock);
void pipe_unlock(struct pipe_inode_info *pipe)
{
if (pipe->files)
mutex_unlock(&pipe->mutex);
}
EXPORT_SYMBOL(pipe_unlock);
static inline void __pipe_lock(struct pipe_inode_info *pipe)
{
mutex_lock_nested(&pipe->mutex, I_MUTEX_PARENT);
}
static inline void __pipe_unlock(struct pipe_inode_info *pipe)
{
mutex_unlock(&pipe->mutex);
}
void pipe_double_lock(struct pipe_inode_info *pipe1,
struct pipe_inode_info *pipe2)
{
BUG_ON(pipe1 == pipe2);
if (pipe1 < pipe2) {
pipe_lock_nested(pipe1, I_MUTEX_PARENT);
pipe_lock_nested(pipe2, I_MUTEX_CHILD);
} else {
pipe_lock_nested(pipe2, I_MUTEX_PARENT);
pipe_lock_nested(pipe1, I_MUTEX_CHILD);
}
}
/* Drop the inode semaphore and wait for a pipe event, atomically */
void pipe_wait(struct pipe_inode_info *pipe)
{
DEFINE_WAIT(wait);
/*
* Pipes are system-local resources, so sleeping on them
* is considered a noninteractive wait:
*/
prepare_to_wait(&pipe->wait, &wait, TASK_INTERRUPTIBLE);
pipe_unlock(pipe);
schedule();
finish_wait(&pipe->wait, &wait);
pipe_lock(pipe);
}
static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
struct pipe_buffer *buf)
{
struct page *page = buf->page;
/*
* If nobody else uses this page, and we don't already have a
* temporary page, let's keep track of it as a one-deep
* allocation cache. (Otherwise just release our reference to it)
*/
if (page_count(page) == 1 && !pipe->tmp_page)
pipe->tmp_page = page;
else
mm, fs: get rid of PAGE_CACHE_* and page_cache_{get,release} macros PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} macros were introduced *long* time ago with promise that one day it will be possible to implement page cache with bigger chunks than PAGE_SIZE. This promise never materialized. And unlikely will. We have many places where PAGE_CACHE_SIZE assumed to be equal to PAGE_SIZE. And it's constant source of confusion on whether PAGE_CACHE_* or PAGE_* constant should be used in a particular case, especially on the border between fs and mm. Global switching to PAGE_CACHE_SIZE != PAGE_SIZE would cause to much breakage to be doable. Let's stop pretending that pages in page cache are special. They are not. The changes are pretty straight-forward: - <foo> << (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - <foo> >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} -> PAGE_{SIZE,SHIFT,MASK,ALIGN}; - page_cache_get() -> get_page(); - page_cache_release() -> put_page(); This patch contains automated changes generated with coccinelle using script below. For some reason, coccinelle doesn't patch header files. I've called spatch for them manually. The only adjustment after coccinelle is revert of changes to PAGE_CAHCE_ALIGN definition: we are going to drop it later. There are few places in the code where coccinelle didn't reach. I'll fix them manually in a separate patch. Comments and documentation also will be addressed with the separate patch. virtual patch @@ expression E; @@ - E << (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ expression E; @@ - E >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ @@ - PAGE_CACHE_SHIFT + PAGE_SHIFT @@ @@ - PAGE_CACHE_SIZE + PAGE_SIZE @@ @@ - PAGE_CACHE_MASK + PAGE_MASK @@ expression E; @@ - PAGE_CACHE_ALIGN(E) + PAGE_ALIGN(E) @@ expression E; @@ - page_cache_get(E) + get_page(E) @@ expression E; @@ - page_cache_release(E) + put_page(E) Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Acked-by: Michal Hocko <mhocko@suse.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-04-01 20:29:47 +08:00
put_page(page);
}
pipe: account to kmemcg Pipes can consume a significant amount of system memory, hence they should be accounted to kmemcg. This patch marks pipe_inode_info and anonymous pipe buffer page allocations as __GFP_ACCOUNT so that they would be charged to kmemcg. Note, since a pipe buffer page can be "stolen" and get reused for other purposes, including mapping to userspace, we clear PageKmemcg thus resetting page->_mapcount and uncharge it in anon_pipe_buf_steal, which is introduced by this patch. A note regarding anon_pipe_buf_steal implementation. We allow to steal the page if its ref count equals 1. It looks racy, but it is correct for anonymous pipe buffer pages, because: - We lock out all other pipe users, because ->steal is called with pipe_lock held, so the page can't be spliced to another pipe from under us. - The page is not on LRU and it never was. - Thus a parallel thread can access it only by PFN. Although this is quite possible (e.g. see page_idle_get_page and balloon_page_isolate) this is not dangerous, because all such functions do is increase page ref count, check if the page is the one they are looking for, and decrease ref count if it isn't. Since our page is clean except for PageKmemcg mark, which doesn't conflict with other _mapcount users, the worst that can happen is we see page_count > 2 due to a transient ref, in which case we false-positively abort ->steal, which is still fine, because ->steal is not guaranteed to succeed. Link: http://lkml.kernel.org/r/20160527150313.GD26059@esperanza Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Minchan Kim <minchan@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-07-27 06:24:33 +08:00
static int anon_pipe_buf_steal(struct pipe_inode_info *pipe,
struct pipe_buffer *buf)
{
struct page *page = buf->page;
if (page_count(page) == 1) {
mm: memcontrol: only mark charged pages with PageKmemcg To distinguish non-slab pages charged to kmemcg we mark them PageKmemcg, which sets page->_mapcount to -512. Currently, we set/clear PageKmemcg in __alloc_pages_nodemask()/free_pages_prepare() for any page allocated with __GFP_ACCOUNT, including those that aren't actually charged to any cgroup, i.e. allocated from the root cgroup context. To avoid overhead in case cgroups are not used, we only do that if memcg_kmem_enabled() is true. The latter is set iff there are kmem-enabled memory cgroups (online or offline). The root cgroup is not considered kmem-enabled. As a result, if a page is allocated with __GFP_ACCOUNT for the root cgroup when there are kmem-enabled memory cgroups and is freed after all kmem-enabled memory cgroups were removed, e.g. # no memory cgroups has been created yet, create one mkdir /sys/fs/cgroup/memory/test # run something allocating pages with __GFP_ACCOUNT, e.g. # a program using pipe dmesg | tail # remove the memory cgroup rmdir /sys/fs/cgroup/memory/test we'll get bad page state bug complaining about page->_mapcount != -1: BUG: Bad page state in process swapper/0 pfn:1fd945c page:ffffea007f651700 count:0 mapcount:-511 mapping: (null) index:0x0 flags: 0x1000000000000000() To avoid that, let's mark with PageKmemcg only those pages that are actually charged to and hence pin a non-root memory cgroup. Fixes: 4949148ad433 ("mm: charge/uncharge kmemcg from generic page allocator paths") Reported-and-tested-by: Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-08-09 04:03:12 +08:00
if (memcg_kmem_enabled())
pipe: account to kmemcg Pipes can consume a significant amount of system memory, hence they should be accounted to kmemcg. This patch marks pipe_inode_info and anonymous pipe buffer page allocations as __GFP_ACCOUNT so that they would be charged to kmemcg. Note, since a pipe buffer page can be "stolen" and get reused for other purposes, including mapping to userspace, we clear PageKmemcg thus resetting page->_mapcount and uncharge it in anon_pipe_buf_steal, which is introduced by this patch. A note regarding anon_pipe_buf_steal implementation. We allow to steal the page if its ref count equals 1. It looks racy, but it is correct for anonymous pipe buffer pages, because: - We lock out all other pipe users, because ->steal is called with pipe_lock held, so the page can't be spliced to another pipe from under us. - The page is not on LRU and it never was. - Thus a parallel thread can access it only by PFN. Although this is quite possible (e.g. see page_idle_get_page and balloon_page_isolate) this is not dangerous, because all such functions do is increase page ref count, check if the page is the one they are looking for, and decrease ref count if it isn't. Since our page is clean except for PageKmemcg mark, which doesn't conflict with other _mapcount users, the worst that can happen is we see page_count > 2 due to a transient ref, in which case we false-positively abort ->steal, which is still fine, because ->steal is not guaranteed to succeed. Link: http://lkml.kernel.org/r/20160527150313.GD26059@esperanza Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Minchan Kim <minchan@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-07-27 06:24:33 +08:00
memcg_kmem_uncharge(page, 0);
__SetPageLocked(page);
return 0;
}
return 1;
}
/**
* generic_pipe_buf_steal - attempt to take ownership of a &pipe_buffer
* @pipe: the pipe that the buffer belongs to
* @buf: the buffer to attempt to steal
*
* Description:
* This function attempts to steal the &struct page attached to
* @buf. If successful, this function returns 0 and returns with
* the page locked. The caller may then reuse the page for whatever
* he wishes; the typical use is insertion into a different file
* page cache.
*/
int generic_pipe_buf_steal(struct pipe_inode_info *pipe,
struct pipe_buffer *buf)
{
struct page *page = buf->page;
/*
* A reference of one is golden, that means that the owner of this
* page is the only one holding a reference to it. lock the page
* and return OK.
*/
if (page_count(page) == 1) {
lock_page(page);
return 0;
}
return 1;
}
EXPORT_SYMBOL(generic_pipe_buf_steal);
/**
* generic_pipe_buf_get - get a reference to a &struct pipe_buffer
* @pipe: the pipe that the buffer belongs to
* @buf: the buffer to get a reference to
*
* Description:
* This function grabs an extra reference to @buf. It's used in
* in the tee() system call, when we duplicate the buffers in one
* pipe into another.
*/
void generic_pipe_buf_get(struct pipe_inode_info *pipe, struct pipe_buffer *buf)
{
mm, fs: get rid of PAGE_CACHE_* and page_cache_{get,release} macros PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} macros were introduced *long* time ago with promise that one day it will be possible to implement page cache with bigger chunks than PAGE_SIZE. This promise never materialized. And unlikely will. We have many places where PAGE_CACHE_SIZE assumed to be equal to PAGE_SIZE. And it's constant source of confusion on whether PAGE_CACHE_* or PAGE_* constant should be used in a particular case, especially on the border between fs and mm. Global switching to PAGE_CACHE_SIZE != PAGE_SIZE would cause to much breakage to be doable. Let's stop pretending that pages in page cache are special. They are not. The changes are pretty straight-forward: - <foo> << (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - <foo> >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} -> PAGE_{SIZE,SHIFT,MASK,ALIGN}; - page_cache_get() -> get_page(); - page_cache_release() -> put_page(); This patch contains automated changes generated with coccinelle using script below. For some reason, coccinelle doesn't patch header files. I've called spatch for them manually. The only adjustment after coccinelle is revert of changes to PAGE_CAHCE_ALIGN definition: we are going to drop it later. There are few places in the code where coccinelle didn't reach. I'll fix them manually in a separate patch. Comments and documentation also will be addressed with the separate patch. virtual patch @@ expression E; @@ - E << (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ expression E; @@ - E >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ @@ - PAGE_CACHE_SHIFT + PAGE_SHIFT @@ @@ - PAGE_CACHE_SIZE + PAGE_SIZE @@ @@ - PAGE_CACHE_MASK + PAGE_MASK @@ expression E; @@ - PAGE_CACHE_ALIGN(E) + PAGE_ALIGN(E) @@ expression E; @@ - page_cache_get(E) + get_page(E) @@ expression E; @@ - page_cache_release(E) + put_page(E) Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Acked-by: Michal Hocko <mhocko@suse.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-04-01 20:29:47 +08:00
get_page(buf->page);
}
EXPORT_SYMBOL(generic_pipe_buf_get);
/**
* generic_pipe_buf_confirm - verify contents of the pipe buffer
* @info: the pipe that the buffer belongs to
* @buf: the buffer to confirm
*
* Description:
* This function does nothing, because the generic pipe code uses
* pages that are always good when inserted into the pipe.
*/
int generic_pipe_buf_confirm(struct pipe_inode_info *info,
struct pipe_buffer *buf)
{
return 0;
}
EXPORT_SYMBOL(generic_pipe_buf_confirm);
/**
* generic_pipe_buf_release - put a reference to a &struct pipe_buffer
* @pipe: the pipe that the buffer belongs to
* @buf: the buffer to put a reference to
*
* Description:
* This function releases a reference to @buf.
*/
void generic_pipe_buf_release(struct pipe_inode_info *pipe,
struct pipe_buffer *buf)
{
mm, fs: get rid of PAGE_CACHE_* and page_cache_{get,release} macros PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} macros were introduced *long* time ago with promise that one day it will be possible to implement page cache with bigger chunks than PAGE_SIZE. This promise never materialized. And unlikely will. We have many places where PAGE_CACHE_SIZE assumed to be equal to PAGE_SIZE. And it's constant source of confusion on whether PAGE_CACHE_* or PAGE_* constant should be used in a particular case, especially on the border between fs and mm. Global switching to PAGE_CACHE_SIZE != PAGE_SIZE would cause to much breakage to be doable. Let's stop pretending that pages in page cache are special. They are not. The changes are pretty straight-forward: - <foo> << (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - <foo> >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) -> <foo>; - PAGE_CACHE_{SIZE,SHIFT,MASK,ALIGN} -> PAGE_{SIZE,SHIFT,MASK,ALIGN}; - page_cache_get() -> get_page(); - page_cache_release() -> put_page(); This patch contains automated changes generated with coccinelle using script below. For some reason, coccinelle doesn't patch header files. I've called spatch for them manually. The only adjustment after coccinelle is revert of changes to PAGE_CAHCE_ALIGN definition: we are going to drop it later. There are few places in the code where coccinelle didn't reach. I'll fix them manually in a separate patch. Comments and documentation also will be addressed with the separate patch. virtual patch @@ expression E; @@ - E << (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ expression E; @@ - E >> (PAGE_CACHE_SHIFT - PAGE_SHIFT) + E @@ @@ - PAGE_CACHE_SHIFT + PAGE_SHIFT @@ @@ - PAGE_CACHE_SIZE + PAGE_SIZE @@ @@ - PAGE_CACHE_MASK + PAGE_MASK @@ expression E; @@ - PAGE_CACHE_ALIGN(E) + PAGE_ALIGN(E) @@ expression E; @@ - page_cache_get(E) + get_page(E) @@ expression E; @@ - page_cache_release(E) + put_page(E) Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Acked-by: Michal Hocko <mhocko@suse.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-04-01 20:29:47 +08:00
put_page(buf->page);
}
EXPORT_SYMBOL(generic_pipe_buf_release);
static const struct pipe_buf_operations anon_pipe_buf_ops = {
.can_merge = 1,
.confirm = generic_pipe_buf_confirm,
.release = anon_pipe_buf_release,
pipe: account to kmemcg Pipes can consume a significant amount of system memory, hence they should be accounted to kmemcg. This patch marks pipe_inode_info and anonymous pipe buffer page allocations as __GFP_ACCOUNT so that they would be charged to kmemcg. Note, since a pipe buffer page can be "stolen" and get reused for other purposes, including mapping to userspace, we clear PageKmemcg thus resetting page->_mapcount and uncharge it in anon_pipe_buf_steal, which is introduced by this patch. A note regarding anon_pipe_buf_steal implementation. We allow to steal the page if its ref count equals 1. It looks racy, but it is correct for anonymous pipe buffer pages, because: - We lock out all other pipe users, because ->steal is called with pipe_lock held, so the page can't be spliced to another pipe from under us. - The page is not on LRU and it never was. - Thus a parallel thread can access it only by PFN. Although this is quite possible (e.g. see page_idle_get_page and balloon_page_isolate) this is not dangerous, because all such functions do is increase page ref count, check if the page is the one they are looking for, and decrease ref count if it isn't. Since our page is clean except for PageKmemcg mark, which doesn't conflict with other _mapcount users, the worst that can happen is we see page_count > 2 due to a transient ref, in which case we false-positively abort ->steal, which is still fine, because ->steal is not guaranteed to succeed. Link: http://lkml.kernel.org/r/20160527150313.GD26059@esperanza Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Minchan Kim <minchan@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-07-27 06:24:33 +08:00
.steal = anon_pipe_buf_steal,
.get = generic_pipe_buf_get,
};
pipes: add a "packetized pipe" mode for writing The actual internal pipe implementation is already really about individual packets (called "pipe buffers"), and this simply exposes that as a special packetized mode. When we are in the packetized mode (marked by O_DIRECT as suggested by Alan Cox), a write() on a pipe will not merge the new data with previous writes, so each write will get a pipe buffer of its own. The pipe buffer is then marked with the PIPE_BUF_FLAG_PACKET flag, which in turn will tell the reader side to break the read at that boundary (and throw away any partial packet contents that do not fit in the read buffer). End result: as long as you do writes less than PIPE_BUF in size (so that the pipe doesn't have to split them up), you can now treat the pipe as a packet interface, where each read() system call will read one packet at a time. You can just use a sufficiently big read buffer (PIPE_BUF is sufficient, since bigger than that doesn't guarantee atomicity anyway), and the return value of the read() will naturally give you the size of the packet. NOTE! We do not support zero-sized packets, and zero-sized reads and writes to a pipe continue to be no-ops. Also note that big packets will currently be split at write time, but that the size at which that happens is not really specified (except that it's bigger than PIPE_BUF). Currently that limit is the system page size, but we might want to explicitly support bigger packets some day. The main user for this is going to be the autofs packet interface, allowing us to stop having to care so deeply about exact packet sizes (which have had bugs with 32/64-bit compatibility modes). But user space can create packetized pipes with "pipe2(fd, O_DIRECT)", which will fail with an EINVAL on kernels that do not support this interface. Tested-by: Michael Tokarev <mjt@tls.msk.ru> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Cc: David Miller <davem@davemloft.net> Cc: Ian Kent <raven@themaw.net> Cc: Thomas Meyer <thomas@m3y3r.de> Cc: stable@kernel.org # needed for systemd/autofs interaction fix Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-04-30 04:12:42 +08:00
static const struct pipe_buf_operations packet_pipe_buf_ops = {
.can_merge = 0,
.confirm = generic_pipe_buf_confirm,
.release = anon_pipe_buf_release,
pipe: account to kmemcg Pipes can consume a significant amount of system memory, hence they should be accounted to kmemcg. This patch marks pipe_inode_info and anonymous pipe buffer page allocations as __GFP_ACCOUNT so that they would be charged to kmemcg. Note, since a pipe buffer page can be "stolen" and get reused for other purposes, including mapping to userspace, we clear PageKmemcg thus resetting page->_mapcount and uncharge it in anon_pipe_buf_steal, which is introduced by this patch. A note regarding anon_pipe_buf_steal implementation. We allow to steal the page if its ref count equals 1. It looks racy, but it is correct for anonymous pipe buffer pages, because: - We lock out all other pipe users, because ->steal is called with pipe_lock held, so the page can't be spliced to another pipe from under us. - The page is not on LRU and it never was. - Thus a parallel thread can access it only by PFN. Although this is quite possible (e.g. see page_idle_get_page and balloon_page_isolate) this is not dangerous, because all such functions do is increase page ref count, check if the page is the one they are looking for, and decrease ref count if it isn't. Since our page is clean except for PageKmemcg mark, which doesn't conflict with other _mapcount users, the worst that can happen is we see page_count > 2 due to a transient ref, in which case we false-positively abort ->steal, which is still fine, because ->steal is not guaranteed to succeed. Link: http://lkml.kernel.org/r/20160527150313.GD26059@esperanza Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Minchan Kim <minchan@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-07-27 06:24:33 +08:00
.steal = anon_pipe_buf_steal,
pipes: add a "packetized pipe" mode for writing The actual internal pipe implementation is already really about individual packets (called "pipe buffers"), and this simply exposes that as a special packetized mode. When we are in the packetized mode (marked by O_DIRECT as suggested by Alan Cox), a write() on a pipe will not merge the new data with previous writes, so each write will get a pipe buffer of its own. The pipe buffer is then marked with the PIPE_BUF_FLAG_PACKET flag, which in turn will tell the reader side to break the read at that boundary (and throw away any partial packet contents that do not fit in the read buffer). End result: as long as you do writes less than PIPE_BUF in size (so that the pipe doesn't have to split them up), you can now treat the pipe as a packet interface, where each read() system call will read one packet at a time. You can just use a sufficiently big read buffer (PIPE_BUF is sufficient, since bigger than that doesn't guarantee atomicity anyway), and the return value of the read() will naturally give you the size of the packet. NOTE! We do not support zero-sized packets, and zero-sized reads and writes to a pipe continue to be no-ops. Also note that big packets will currently be split at write time, but that the size at which that happens is not really specified (except that it's bigger than PIPE_BUF). Currently that limit is the system page size, but we might want to explicitly support bigger packets some day. The main user for this is going to be the autofs packet interface, allowing us to stop having to care so deeply about exact packet sizes (which have had bugs with 32/64-bit compatibility modes). But user space can create packetized pipes with "pipe2(fd, O_DIRECT)", which will fail with an EINVAL on kernels that do not support this interface. Tested-by: Michael Tokarev <mjt@tls.msk.ru> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Cc: David Miller <davem@davemloft.net> Cc: Ian Kent <raven@themaw.net> Cc: Thomas Meyer <thomas@m3y3r.de> Cc: stable@kernel.org # needed for systemd/autofs interaction fix Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-04-30 04:12:42 +08:00
.get = generic_pipe_buf_get,
};
static ssize_t
pipe_read(struct kiocb *iocb, struct iov_iter *to)
{
size_t total_len = iov_iter_count(to);
struct file *filp = iocb->ki_filp;
struct pipe_inode_info *pipe = filp->private_data;
int do_wakeup;
ssize_t ret;
/* Null read succeeds. */
if (unlikely(total_len == 0))
return 0;
do_wakeup = 0;
ret = 0;
__pipe_lock(pipe);
for (;;) {
int bufs = pipe->nrbufs;
if (bufs) {
int curbuf = pipe->curbuf;
struct pipe_buffer *buf = pipe->bufs + curbuf;
size_t chars = buf->len;
size_t written;
int error;
if (chars > total_len)
chars = total_len;
error = pipe_buf_confirm(pipe, buf);
if (error) {
if (!ret)
ret = error;
break;
}
written = copy_page_to_iter(buf->page, buf->offset, chars, to);
if (unlikely(written < chars)) {
if (!ret)
ret = -EFAULT;
break;
}
ret += chars;
buf->offset += chars;
buf->len -= chars;
pipes: add a "packetized pipe" mode for writing The actual internal pipe implementation is already really about individual packets (called "pipe buffers"), and this simply exposes that as a special packetized mode. When we are in the packetized mode (marked by O_DIRECT as suggested by Alan Cox), a write() on a pipe will not merge the new data with previous writes, so each write will get a pipe buffer of its own. The pipe buffer is then marked with the PIPE_BUF_FLAG_PACKET flag, which in turn will tell the reader side to break the read at that boundary (and throw away any partial packet contents that do not fit in the read buffer). End result: as long as you do writes less than PIPE_BUF in size (so that the pipe doesn't have to split them up), you can now treat the pipe as a packet interface, where each read() system call will read one packet at a time. You can just use a sufficiently big read buffer (PIPE_BUF is sufficient, since bigger than that doesn't guarantee atomicity anyway), and the return value of the read() will naturally give you the size of the packet. NOTE! We do not support zero-sized packets, and zero-sized reads and writes to a pipe continue to be no-ops. Also note that big packets will currently be split at write time, but that the size at which that happens is not really specified (except that it's bigger than PIPE_BUF). Currently that limit is the system page size, but we might want to explicitly support bigger packets some day. The main user for this is going to be the autofs packet interface, allowing us to stop having to care so deeply about exact packet sizes (which have had bugs with 32/64-bit compatibility modes). But user space can create packetized pipes with "pipe2(fd, O_DIRECT)", which will fail with an EINVAL on kernels that do not support this interface. Tested-by: Michael Tokarev <mjt@tls.msk.ru> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Cc: David Miller <davem@davemloft.net> Cc: Ian Kent <raven@themaw.net> Cc: Thomas Meyer <thomas@m3y3r.de> Cc: stable@kernel.org # needed for systemd/autofs interaction fix Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-04-30 04:12:42 +08:00
/* Was it a packet buffer? Clean up and exit */
if (buf->flags & PIPE_BUF_FLAG_PACKET) {
total_len = chars;
buf->len = 0;
}
if (!buf->len) {
pipe_buf_release(pipe, buf);
curbuf = (curbuf + 1) & (pipe->buffers - 1);
pipe->curbuf = curbuf;
pipe->nrbufs = --bufs;
do_wakeup = 1;
}
total_len -= chars;
if (!total_len)
break; /* common path: read succeeded */
}
if (bufs) /* More to do? */
continue;
if (!pipe->writers)
break;
if (!pipe->waiting_writers) {
/* syscall merging: Usually we must not sleep
* if O_NONBLOCK is set, or if we got some data.
* But if a writer sleeps in kernel space, then
* we can wait for that data without violating POSIX.
*/
if (ret)
break;
if (filp->f_flags & O_NONBLOCK) {
ret = -EAGAIN;
break;
}
}
if (signal_pending(current)) {
if (!ret)
ret = -ERESTARTSYS;
break;
}
if (do_wakeup) {
wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
}
pipe_wait(pipe);
}
__pipe_unlock(pipe);
/* Signal writers asynchronously that there is more room. */
if (do_wakeup) {
wake_up_interruptible_sync_poll(&pipe->wait, POLLOUT | POLLWRNORM);
kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
}
if (ret > 0)
file_accessed(filp);
return ret;
}
pipes: add a "packetized pipe" mode for writing The actual internal pipe implementation is already really about individual packets (called "pipe buffers"), and this simply exposes that as a special packetized mode. When we are in the packetized mode (marked by O_DIRECT as suggested by Alan Cox), a write() on a pipe will not merge the new data with previous writes, so each write will get a pipe buffer of its own. The pipe buffer is then marked with the PIPE_BUF_FLAG_PACKET flag, which in turn will tell the reader side to break the read at that boundary (and throw away any partial packet contents that do not fit in the read buffer). End result: as long as you do writes less than PIPE_BUF in size (so that the pipe doesn't have to split them up), you can now treat the pipe as a packet interface, where each read() system call will read one packet at a time. You can just use a sufficiently big read buffer (PIPE_BUF is sufficient, since bigger than that doesn't guarantee atomicity anyway), and the return value of the read() will naturally give you the size of the packet. NOTE! We do not support zero-sized packets, and zero-sized reads and writes to a pipe continue to be no-ops. Also note that big packets will currently be split at write time, but that the size at which that happens is not really specified (except that it's bigger than PIPE_BUF). Currently that limit is the system page size, but we might want to explicitly support bigger packets some day. The main user for this is going to be the autofs packet interface, allowing us to stop having to care so deeply about exact packet sizes (which have had bugs with 32/64-bit compatibility modes). But user space can create packetized pipes with "pipe2(fd, O_DIRECT)", which will fail with an EINVAL on kernels that do not support this interface. Tested-by: Michael Tokarev <mjt@tls.msk.ru> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Cc: David Miller <davem@davemloft.net> Cc: Ian Kent <raven@themaw.net> Cc: Thomas Meyer <thomas@m3y3r.de> Cc: stable@kernel.org # needed for systemd/autofs interaction fix Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-04-30 04:12:42 +08:00
static inline int is_packetized(struct file *file)
{
return (file->f_flags & O_DIRECT) != 0;
}
static ssize_t
pipe_write(struct kiocb *iocb, struct iov_iter *from)
{
struct file *filp = iocb->ki_filp;
struct pipe_inode_info *pipe = filp->private_data;
ssize_t ret = 0;
int do_wakeup = 0;
size_t total_len = iov_iter_count(from);
ssize_t chars;
/* Null write succeeds. */
if (unlikely(total_len == 0))
return 0;
__pipe_lock(pipe);
if (!pipe->readers) {
send_sig(SIGPIPE, current, 0);
ret = -EPIPE;
goto out;
}
/* We try to merge small writes */
chars = total_len & (PAGE_SIZE-1); /* size of the last buffer */
if (pipe->nrbufs && chars != 0) {
int lastbuf = (pipe->curbuf + pipe->nrbufs - 1) &
(pipe->buffers - 1);
struct pipe_buffer *buf = pipe->bufs + lastbuf;
int offset = buf->offset + buf->len;
if (buf->ops->can_merge && offset + chars <= PAGE_SIZE) {
ret = pipe_buf_confirm(pipe, buf);
if (ret)
goto out;
ret = copy_page_from_iter(buf->page, offset, chars, from);
if (unlikely(ret < chars)) {
ret = -EFAULT;
goto out;
}
do_wakeup = 1;
buf->len += ret;
if (!iov_iter_count(from))
goto out;
}
}
for (;;) {
int bufs;
if (!pipe->readers) {
send_sig(SIGPIPE, current, 0);
if (!ret)
ret = -EPIPE;
break;
}
bufs = pipe->nrbufs;
if (bufs < pipe->buffers) {
int newbuf = (pipe->curbuf + bufs) & (pipe->buffers-1);
struct pipe_buffer *buf = pipe->bufs + newbuf;
struct page *page = pipe->tmp_page;
int copied;
if (!page) {
pipe: account to kmemcg Pipes can consume a significant amount of system memory, hence they should be accounted to kmemcg. This patch marks pipe_inode_info and anonymous pipe buffer page allocations as __GFP_ACCOUNT so that they would be charged to kmemcg. Note, since a pipe buffer page can be "stolen" and get reused for other purposes, including mapping to userspace, we clear PageKmemcg thus resetting page->_mapcount and uncharge it in anon_pipe_buf_steal, which is introduced by this patch. A note regarding anon_pipe_buf_steal implementation. We allow to steal the page if its ref count equals 1. It looks racy, but it is correct for anonymous pipe buffer pages, because: - We lock out all other pipe users, because ->steal is called with pipe_lock held, so the page can't be spliced to another pipe from under us. - The page is not on LRU and it never was. - Thus a parallel thread can access it only by PFN. Although this is quite possible (e.g. see page_idle_get_page and balloon_page_isolate) this is not dangerous, because all such functions do is increase page ref count, check if the page is the one they are looking for, and decrease ref count if it isn't. Since our page is clean except for PageKmemcg mark, which doesn't conflict with other _mapcount users, the worst that can happen is we see page_count > 2 due to a transient ref, in which case we false-positively abort ->steal, which is still fine, because ->steal is not guaranteed to succeed. Link: http://lkml.kernel.org/r/20160527150313.GD26059@esperanza Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Minchan Kim <minchan@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-07-27 06:24:33 +08:00
page = alloc_page(GFP_HIGHUSER | __GFP_ACCOUNT);
if (unlikely(!page)) {
ret = ret ? : -ENOMEM;
break;
}
pipe->tmp_page = page;
}
/* Always wake up, even if the copy fails. Otherwise
* we lock up (O_NONBLOCK-)readers that sleep due to
* syscall merging.
* FIXME! Is this really true?
*/
do_wakeup = 1;
copied = copy_page_from_iter(page, 0, PAGE_SIZE, from);
if (unlikely(copied < PAGE_SIZE && iov_iter_count(from))) {
if (!ret)
ret = -EFAULT;
break;
}
ret += copied;
/* Insert it into the buffer array */
buf->page = page;
buf->ops = &anon_pipe_buf_ops;
buf->offset = 0;
buf->len = copied;
pipes: add a "packetized pipe" mode for writing The actual internal pipe implementation is already really about individual packets (called "pipe buffers"), and this simply exposes that as a special packetized mode. When we are in the packetized mode (marked by O_DIRECT as suggested by Alan Cox), a write() on a pipe will not merge the new data with previous writes, so each write will get a pipe buffer of its own. The pipe buffer is then marked with the PIPE_BUF_FLAG_PACKET flag, which in turn will tell the reader side to break the read at that boundary (and throw away any partial packet contents that do not fit in the read buffer). End result: as long as you do writes less than PIPE_BUF in size (so that the pipe doesn't have to split them up), you can now treat the pipe as a packet interface, where each read() system call will read one packet at a time. You can just use a sufficiently big read buffer (PIPE_BUF is sufficient, since bigger than that doesn't guarantee atomicity anyway), and the return value of the read() will naturally give you the size of the packet. NOTE! We do not support zero-sized packets, and zero-sized reads and writes to a pipe continue to be no-ops. Also note that big packets will currently be split at write time, but that the size at which that happens is not really specified (except that it's bigger than PIPE_BUF). Currently that limit is the system page size, but we might want to explicitly support bigger packets some day. The main user for this is going to be the autofs packet interface, allowing us to stop having to care so deeply about exact packet sizes (which have had bugs with 32/64-bit compatibility modes). But user space can create packetized pipes with "pipe2(fd, O_DIRECT)", which will fail with an EINVAL on kernels that do not support this interface. Tested-by: Michael Tokarev <mjt@tls.msk.ru> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Cc: David Miller <davem@davemloft.net> Cc: Ian Kent <raven@themaw.net> Cc: Thomas Meyer <thomas@m3y3r.de> Cc: stable@kernel.org # needed for systemd/autofs interaction fix Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-04-30 04:12:42 +08:00
buf->flags = 0;
if (is_packetized(filp)) {
buf->ops = &packet_pipe_buf_ops;
buf->flags = PIPE_BUF_FLAG_PACKET;
}
pipe->nrbufs = ++bufs;
pipe->tmp_page = NULL;
if (!iov_iter_count(from))
break;
}
if (bufs < pipe->buffers)
continue;
if (filp->f_flags & O_NONBLOCK) {
if (!ret)
ret = -EAGAIN;
break;
}
if (signal_pending(current)) {
if (!ret)
ret = -ERESTARTSYS;
break;
}
if (do_wakeup) {
wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
do_wakeup = 0;
}
pipe->waiting_writers++;
pipe_wait(pipe);
pipe->waiting_writers--;
}
out:
__pipe_unlock(pipe);
if (do_wakeup) {
wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLRDNORM);
kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
}
if (ret > 0 && sb_start_write_trylock(file_inode(filp)->i_sb)) {
int err = file_update_time(filp);
if (err)
ret = err;
sb_end_write(file_inode(filp)->i_sb);
}
return ret;
}
static long pipe_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
{
struct pipe_inode_info *pipe = filp->private_data;
int count, buf, nrbufs;
switch (cmd) {
case FIONREAD:
__pipe_lock(pipe);
count = 0;
buf = pipe->curbuf;
nrbufs = pipe->nrbufs;
while (--nrbufs >= 0) {
count += pipe->bufs[buf].len;
buf = (buf+1) & (pipe->buffers - 1);
}
__pipe_unlock(pipe);
return put_user(count, (int __user *)arg);
default:
return -ENOIOCTLCMD;
}
}
/* No kernel lock held - fine */
static unsigned int
pipe_poll(struct file *filp, poll_table *wait)
{
unsigned int mask;
struct pipe_inode_info *pipe = filp->private_data;
int nrbufs;
poll_wait(filp, &pipe->wait, wait);
/* Reading only -- no need for acquiring the semaphore. */
nrbufs = pipe->nrbufs;
mask = 0;
if (filp->f_mode & FMODE_READ) {
mask = (nrbufs > 0) ? POLLIN | POLLRDNORM : 0;
if (!pipe->writers && filp->f_version != pipe->w_counter)
mask |= POLLHUP;
}
if (filp->f_mode & FMODE_WRITE) {
mask |= (nrbufs < pipe->buffers) ? POLLOUT | POLLWRNORM : 0;
/*
* Most Unices do not set POLLERR for FIFOs but on Linux they
* behave exactly like pipes for poll().
*/
if (!pipe->readers)
mask |= POLLERR;
}
return mask;
}
vfs: fix subtle use-after-free of pipe_inode_info The pipe code was trying (and failing) to be very careful about freeing the pipe info only after the last access, with a pattern like: spin_lock(&inode->i_lock); if (!--pipe->files) { inode->i_pipe = NULL; kill = 1; } spin_unlock(&inode->i_lock); __pipe_unlock(pipe); if (kill) free_pipe_info(pipe); where the final freeing is done last. HOWEVER. The above is actually broken, because while the freeing is done at the end, if we have two racing processes releasing the pipe inode info, the one that *doesn't* free it will decrement the ->files count, and unlock the inode i_lock, but then still use the "pipe_inode_info" afterwards when it does the "__pipe_unlock(pipe)". This is *very* hard to trigger in practice, since the race window is very small, and adding debug options seems to just hide it by slowing things down. Simon originally reported this way back in July as an Oops in kmem_cache_allocate due to a single bit corruption (due to the final "spin_unlock(pipe->mutex.wait_lock)" incrementing a field in a different allocation that had re-used the free'd pipe-info), it's taken this long to figure out. Since the 'pipe->files' accesses aren't even protected by the pipe lock (we very much use the inode lock for that), the simple solution is to just drop the pipe lock early. And since there were two users of this pattern, create a helper function for it. Introduced commit ba5bb147330a ("pipe: take allocation and freeing of pipe_inode_info out of ->i_mutex"). Reported-by: Simon Kirby <sim@hostway.ca> Reported-by: Ian Applegate <ia@cloudflare.com> Acked-by: Al Viro <viro@zeniv.linux.org.uk> Cc: stable@kernel.org # v3.10+ Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-12-03 01:44:51 +08:00
static void put_pipe_info(struct inode *inode, struct pipe_inode_info *pipe)
{
int kill = 0;
spin_lock(&inode->i_lock);
if (!--pipe->files) {
inode->i_pipe = NULL;
kill = 1;
}
spin_unlock(&inode->i_lock);
if (kill)
free_pipe_info(pipe);
}
static int
pipe_release(struct inode *inode, struct file *file)
{
vfs: fix subtle use-after-free of pipe_inode_info The pipe code was trying (and failing) to be very careful about freeing the pipe info only after the last access, with a pattern like: spin_lock(&inode->i_lock); if (!--pipe->files) { inode->i_pipe = NULL; kill = 1; } spin_unlock(&inode->i_lock); __pipe_unlock(pipe); if (kill) free_pipe_info(pipe); where the final freeing is done last. HOWEVER. The above is actually broken, because while the freeing is done at the end, if we have two racing processes releasing the pipe inode info, the one that *doesn't* free it will decrement the ->files count, and unlock the inode i_lock, but then still use the "pipe_inode_info" afterwards when it does the "__pipe_unlock(pipe)". This is *very* hard to trigger in practice, since the race window is very small, and adding debug options seems to just hide it by slowing things down. Simon originally reported this way back in July as an Oops in kmem_cache_allocate due to a single bit corruption (due to the final "spin_unlock(pipe->mutex.wait_lock)" incrementing a field in a different allocation that had re-used the free'd pipe-info), it's taken this long to figure out. Since the 'pipe->files' accesses aren't even protected by the pipe lock (we very much use the inode lock for that), the simple solution is to just drop the pipe lock early. And since there were two users of this pattern, create a helper function for it. Introduced commit ba5bb147330a ("pipe: take allocation and freeing of pipe_inode_info out of ->i_mutex"). Reported-by: Simon Kirby <sim@hostway.ca> Reported-by: Ian Applegate <ia@cloudflare.com> Acked-by: Al Viro <viro@zeniv.linux.org.uk> Cc: stable@kernel.org # v3.10+ Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-12-03 01:44:51 +08:00
struct pipe_inode_info *pipe = file->private_data;
__pipe_lock(pipe);
if (file->f_mode & FMODE_READ)
pipe->readers--;
if (file->f_mode & FMODE_WRITE)
pipe->writers--;
if (pipe->readers || pipe->writers) {
wake_up_interruptible_sync_poll(&pipe->wait, POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM | POLLERR | POLLHUP);
kill_fasync(&pipe->fasync_readers, SIGIO, POLL_IN);
kill_fasync(&pipe->fasync_writers, SIGIO, POLL_OUT);
}
__pipe_unlock(pipe);
vfs: fix subtle use-after-free of pipe_inode_info The pipe code was trying (and failing) to be very careful about freeing the pipe info only after the last access, with a pattern like: spin_lock(&inode->i_lock); if (!--pipe->files) { inode->i_pipe = NULL; kill = 1; } spin_unlock(&inode->i_lock); __pipe_unlock(pipe); if (kill) free_pipe_info(pipe); where the final freeing is done last. HOWEVER. The above is actually broken, because while the freeing is done at the end, if we have two racing processes releasing the pipe inode info, the one that *doesn't* free it will decrement the ->files count, and unlock the inode i_lock, but then still use the "pipe_inode_info" afterwards when it does the "__pipe_unlock(pipe)". This is *very* hard to trigger in practice, since the race window is very small, and adding debug options seems to just hide it by slowing things down. Simon originally reported this way back in July as an Oops in kmem_cache_allocate due to a single bit corruption (due to the final "spin_unlock(pipe->mutex.wait_lock)" incrementing a field in a different allocation that had re-used the free'd pipe-info), it's taken this long to figure out. Since the 'pipe->files' accesses aren't even protected by the pipe lock (we very much use the inode lock for that), the simple solution is to just drop the pipe lock early. And since there were two users of this pattern, create a helper function for it. Introduced commit ba5bb147330a ("pipe: take allocation and freeing of pipe_inode_info out of ->i_mutex"). Reported-by: Simon Kirby <sim@hostway.ca> Reported-by: Ian Applegate <ia@cloudflare.com> Acked-by: Al Viro <viro@zeniv.linux.org.uk> Cc: stable@kernel.org # v3.10+ Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-12-03 01:44:51 +08:00
put_pipe_info(inode, pipe);
return 0;
}
static int
pipe_fasync(int fd, struct file *filp, int on)
{
struct pipe_inode_info *pipe = filp->private_data;
int retval = 0;
__pipe_lock(pipe);
if (filp->f_mode & FMODE_READ)
retval = fasync_helper(fd, filp, on, &pipe->fasync_readers);
if ((filp->f_mode & FMODE_WRITE) && retval >= 0) {
retval = fasync_helper(fd, filp, on, &pipe->fasync_writers);
if (retval < 0 && (filp->f_mode & FMODE_READ))
/* this can happen only if on == T */
fasync_helper(-1, filp, 0, &pipe->fasync_readers);
}
__pipe_unlock(pipe);
return retval;
}
static unsigned long account_pipe_buffers(struct user_struct *user,
pipe: limit the per-user amount of pages allocated in pipes On no-so-small systems, it is possible for a single process to cause an OOM condition by filling large pipes with data that are never read. A typical process filling 4000 pipes with 1 MB of data will use 4 GB of memory. On small systems it may be tricky to set the pipe max size to prevent this from happening. This patch makes it possible to enforce a per-user soft limit above which new pipes will be limited to a single page, effectively limiting them to 4 kB each, as well as a hard limit above which no new pipes may be created for this user. This has the effect of protecting the system against memory abuse without hurting other users, and still allowing pipes to work correctly though with less data at once. The limit are controlled by two new sysctls : pipe-user-pages-soft, and pipe-user-pages-hard. Both may be disabled by setting them to zero. The default soft limit allows the default number of FDs per process (1024) to create pipes of the default size (64kB), thus reaching a limit of 64MB before starting to create only smaller pipes. With 256 processes limited to 1024 FDs each, this results in 1024*64kB + (256*1024 - 1024) * 4kB = 1084 MB of memory allocated for a user. The hard limit is disabled by default to avoid breaking existing applications that make intensive use of pipes (eg: for splicing). Reported-by: socketpair@gmail.com Reported-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Mitigates: CVE-2013-4312 (Linux 2.0+) Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-01-18 23:36:09 +08:00
unsigned long old, unsigned long new)
{
return atomic_long_add_return(new - old, &user->pipe_bufs);
pipe: limit the per-user amount of pages allocated in pipes On no-so-small systems, it is possible for a single process to cause an OOM condition by filling large pipes with data that are never read. A typical process filling 4000 pipes with 1 MB of data will use 4 GB of memory. On small systems it may be tricky to set the pipe max size to prevent this from happening. This patch makes it possible to enforce a per-user soft limit above which new pipes will be limited to a single page, effectively limiting them to 4 kB each, as well as a hard limit above which no new pipes may be created for this user. This has the effect of protecting the system against memory abuse without hurting other users, and still allowing pipes to work correctly though with less data at once. The limit are controlled by two new sysctls : pipe-user-pages-soft, and pipe-user-pages-hard. Both may be disabled by setting them to zero. The default soft limit allows the default number of FDs per process (1024) to create pipes of the default size (64kB), thus reaching a limit of 64MB before starting to create only smaller pipes. With 256 processes limited to 1024 FDs each, this results in 1024*64kB + (256*1024 - 1024) * 4kB = 1084 MB of memory allocated for a user. The hard limit is disabled by default to avoid breaking existing applications that make intensive use of pipes (eg: for splicing). Reported-by: socketpair@gmail.com Reported-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Mitigates: CVE-2013-4312 (Linux 2.0+) Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-01-18 23:36:09 +08:00
}
static bool too_many_pipe_buffers_soft(unsigned long user_bufs)
pipe: limit the per-user amount of pages allocated in pipes On no-so-small systems, it is possible for a single process to cause an OOM condition by filling large pipes with data that are never read. A typical process filling 4000 pipes with 1 MB of data will use 4 GB of memory. On small systems it may be tricky to set the pipe max size to prevent this from happening. This patch makes it possible to enforce a per-user soft limit above which new pipes will be limited to a single page, effectively limiting them to 4 kB each, as well as a hard limit above which no new pipes may be created for this user. This has the effect of protecting the system against memory abuse without hurting other users, and still allowing pipes to work correctly though with less data at once. The limit are controlled by two new sysctls : pipe-user-pages-soft, and pipe-user-pages-hard. Both may be disabled by setting them to zero. The default soft limit allows the default number of FDs per process (1024) to create pipes of the default size (64kB), thus reaching a limit of 64MB before starting to create only smaller pipes. With 256 processes limited to 1024 FDs each, this results in 1024*64kB + (256*1024 - 1024) * 4kB = 1084 MB of memory allocated for a user. The hard limit is disabled by default to avoid breaking existing applications that make intensive use of pipes (eg: for splicing). Reported-by: socketpair@gmail.com Reported-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Mitigates: CVE-2013-4312 (Linux 2.0+) Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-01-18 23:36:09 +08:00
{
return pipe_user_pages_soft && user_bufs >= pipe_user_pages_soft;
pipe: limit the per-user amount of pages allocated in pipes On no-so-small systems, it is possible for a single process to cause an OOM condition by filling large pipes with data that are never read. A typical process filling 4000 pipes with 1 MB of data will use 4 GB of memory. On small systems it may be tricky to set the pipe max size to prevent this from happening. This patch makes it possible to enforce a per-user soft limit above which new pipes will be limited to a single page, effectively limiting them to 4 kB each, as well as a hard limit above which no new pipes may be created for this user. This has the effect of protecting the system against memory abuse without hurting other users, and still allowing pipes to work correctly though with less data at once. The limit are controlled by two new sysctls : pipe-user-pages-soft, and pipe-user-pages-hard. Both may be disabled by setting them to zero. The default soft limit allows the default number of FDs per process (1024) to create pipes of the default size (64kB), thus reaching a limit of 64MB before starting to create only smaller pipes. With 256 processes limited to 1024 FDs each, this results in 1024*64kB + (256*1024 - 1024) * 4kB = 1084 MB of memory allocated for a user. The hard limit is disabled by default to avoid breaking existing applications that make intensive use of pipes (eg: for splicing). Reported-by: socketpair@gmail.com Reported-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Mitigates: CVE-2013-4312 (Linux 2.0+) Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-01-18 23:36:09 +08:00
}
static bool too_many_pipe_buffers_hard(unsigned long user_bufs)
pipe: limit the per-user amount of pages allocated in pipes On no-so-small systems, it is possible for a single process to cause an OOM condition by filling large pipes with data that are never read. A typical process filling 4000 pipes with 1 MB of data will use 4 GB of memory. On small systems it may be tricky to set the pipe max size to prevent this from happening. This patch makes it possible to enforce a per-user soft limit above which new pipes will be limited to a single page, effectively limiting them to 4 kB each, as well as a hard limit above which no new pipes may be created for this user. This has the effect of protecting the system against memory abuse without hurting other users, and still allowing pipes to work correctly though with less data at once. The limit are controlled by two new sysctls : pipe-user-pages-soft, and pipe-user-pages-hard. Both may be disabled by setting them to zero. The default soft limit allows the default number of FDs per process (1024) to create pipes of the default size (64kB), thus reaching a limit of 64MB before starting to create only smaller pipes. With 256 processes limited to 1024 FDs each, this results in 1024*64kB + (256*1024 - 1024) * 4kB = 1084 MB of memory allocated for a user. The hard limit is disabled by default to avoid breaking existing applications that make intensive use of pipes (eg: for splicing). Reported-by: socketpair@gmail.com Reported-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Mitigates: CVE-2013-4312 (Linux 2.0+) Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-01-18 23:36:09 +08:00
{
return pipe_user_pages_hard && user_bufs >= pipe_user_pages_hard;
pipe: limit the per-user amount of pages allocated in pipes On no-so-small systems, it is possible for a single process to cause an OOM condition by filling large pipes with data that are never read. A typical process filling 4000 pipes with 1 MB of data will use 4 GB of memory. On small systems it may be tricky to set the pipe max size to prevent this from happening. This patch makes it possible to enforce a per-user soft limit above which new pipes will be limited to a single page, effectively limiting them to 4 kB each, as well as a hard limit above which no new pipes may be created for this user. This has the effect of protecting the system against memory abuse without hurting other users, and still allowing pipes to work correctly though with less data at once. The limit are controlled by two new sysctls : pipe-user-pages-soft, and pipe-user-pages-hard. Both may be disabled by setting them to zero. The default soft limit allows the default number of FDs per process (1024) to create pipes of the default size (64kB), thus reaching a limit of 64MB before starting to create only smaller pipes. With 256 processes limited to 1024 FDs each, this results in 1024*64kB + (256*1024 - 1024) * 4kB = 1084 MB of memory allocated for a user. The hard limit is disabled by default to avoid breaking existing applications that make intensive use of pipes (eg: for splicing). Reported-by: socketpair@gmail.com Reported-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Mitigates: CVE-2013-4312 (Linux 2.0+) Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-01-18 23:36:09 +08:00
}
struct pipe_inode_info *alloc_pipe_info(void)
{
struct pipe_inode_info *pipe;
unsigned long pipe_bufs = PIPE_DEF_BUFFERS;
struct user_struct *user = get_current_user();
unsigned long user_bufs;
pipe: account to kmemcg Pipes can consume a significant amount of system memory, hence they should be accounted to kmemcg. This patch marks pipe_inode_info and anonymous pipe buffer page allocations as __GFP_ACCOUNT so that they would be charged to kmemcg. Note, since a pipe buffer page can be "stolen" and get reused for other purposes, including mapping to userspace, we clear PageKmemcg thus resetting page->_mapcount and uncharge it in anon_pipe_buf_steal, which is introduced by this patch. A note regarding anon_pipe_buf_steal implementation. We allow to steal the page if its ref count equals 1. It looks racy, but it is correct for anonymous pipe buffer pages, because: - We lock out all other pipe users, because ->steal is called with pipe_lock held, so the page can't be spliced to another pipe from under us. - The page is not on LRU and it never was. - Thus a parallel thread can access it only by PFN. Although this is quite possible (e.g. see page_idle_get_page and balloon_page_isolate) this is not dangerous, because all such functions do is increase page ref count, check if the page is the one they are looking for, and decrease ref count if it isn't. Since our page is clean except for PageKmemcg mark, which doesn't conflict with other _mapcount users, the worst that can happen is we see page_count > 2 due to a transient ref, in which case we false-positively abort ->steal, which is still fine, because ->steal is not guaranteed to succeed. Link: http://lkml.kernel.org/r/20160527150313.GD26059@esperanza Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Minchan Kim <minchan@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-07-27 06:24:33 +08:00
pipe = kzalloc(sizeof(struct pipe_inode_info), GFP_KERNEL_ACCOUNT);
if (pipe == NULL)
goto out_free_uid;
pipe: cap initial pipe capacity according to pipe-max-size limit This is a patch that provides behavior that is more consistent, and probably less surprising to users. I consider the change optional, and welcome opinions about whether it should be applied. By default, pipes are created with a capacity of 64 kiB. However, /proc/sys/fs/pipe-max-size may be set smaller than this value. In this scenario, an unprivileged user could thus create a pipe whose initial capacity exceeds the limit. Therefore, it seems logical to cap the initial pipe capacity according to the value of pipe-max-size. The test program shown earlier in this patch series can be used to demonstrate the effect of the change brought about with this patch: # cat /proc/sys/fs/pipe-max-size 1048576 # sudo -u mtk ./test_F_SETPIPE_SZ 1 Initial pipe capacity: 65536 # echo 10000 > /proc/sys/fs/pipe-max-size # cat /proc/sys/fs/pipe-max-size 16384 # sudo -u mtk ./test_F_SETPIPE_SZ 1 Initial pipe capacity: 16384 # ./test_F_SETPIPE_SZ 1 Initial pipe capacity: 65536 The last two executions of 'test_F_SETPIPE_SZ' show that pipe-max-size caps the initial allocation for a new pipe for unprivileged users, but not for privileged users. Link: http://lkml.kernel.org/r/31dc7064-2a17-9c5b-1df1-4e3012ee992c@gmail.com Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com> Reviewed-by: Vegard Nossum <vegard.nossum@oracle.com> Cc: Willy Tarreau <w@1wt.eu> Cc: <socketpair@gmail.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Jens Axboe <axboe@fb.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-12 04:53:43 +08:00
if (pipe_bufs * PAGE_SIZE > pipe_max_size && !capable(CAP_SYS_RESOURCE))
pipe_bufs = pipe_max_size >> PAGE_SHIFT;
user_bufs = account_pipe_buffers(user, 0, pipe_bufs);
pipe: fix limit checking in alloc_pipe_info() The limit checking in alloc_pipe_info() (used by pipe(2) and when opening a FIFO) has the following problems: (1) When checking capacity required for the new pipe, the checks against the limit in /proc/sys/fs/pipe-user-pages-{soft,hard} are made against existing consumption, and exclude the memory required for the new pipe capacity. As a consequence: (1) the memory allocation throttling provided by the soft limit does not kick in quite as early as it should, and (2) the user can overrun the hard limit. (2) As currently implemented, accounting and checking against the limits is done as follows: (a) Test whether the user has exceeded the limit. (b) Make new pipe buffer allocation. (c) Account new allocation against the limits. This is racey. Multiple processes may pass point (a) simultaneously, and then allocate pipe buffers that are accounted for only in step (c). The race means that the user's pipe buffer allocation could be pushed over the limit (by an arbitrary amount, depending on how unlucky we were in the race). [Thanks to Vegard Nossum for spotting this point, which I had missed.] This patch addresses the above problems as follows: * Alter the checks against limits to include the memory required for the new pipe. * Re-order the accounting step so that it precedes the buffer allocation. If the accounting step determines that a limit has been reached, revert the accounting and cause the operation to fail. Link: http://lkml.kernel.org/r/8ff3e9f9-23f6-510c-644f-8e70cd1c0bd9@gmail.com Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com> Reviewed-by: Vegard Nossum <vegard.nossum@oracle.com> Cc: Willy Tarreau <w@1wt.eu> Cc: <socketpair@gmail.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Jens Axboe <axboe@fb.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-12 04:53:37 +08:00
if (too_many_pipe_buffers_soft(user_bufs)) {
user_bufs = account_pipe_buffers(user, pipe_bufs, 1);
pipe: fix limit checking in alloc_pipe_info() The limit checking in alloc_pipe_info() (used by pipe(2) and when opening a FIFO) has the following problems: (1) When checking capacity required for the new pipe, the checks against the limit in /proc/sys/fs/pipe-user-pages-{soft,hard} are made against existing consumption, and exclude the memory required for the new pipe capacity. As a consequence: (1) the memory allocation throttling provided by the soft limit does not kick in quite as early as it should, and (2) the user can overrun the hard limit. (2) As currently implemented, accounting and checking against the limits is done as follows: (a) Test whether the user has exceeded the limit. (b) Make new pipe buffer allocation. (c) Account new allocation against the limits. This is racey. Multiple processes may pass point (a) simultaneously, and then allocate pipe buffers that are accounted for only in step (c). The race means that the user's pipe buffer allocation could be pushed over the limit (by an arbitrary amount, depending on how unlucky we were in the race). [Thanks to Vegard Nossum for spotting this point, which I had missed.] This patch addresses the above problems as follows: * Alter the checks against limits to include the memory required for the new pipe. * Re-order the accounting step so that it precedes the buffer allocation. If the accounting step determines that a limit has been reached, revert the accounting and cause the operation to fail. Link: http://lkml.kernel.org/r/8ff3e9f9-23f6-510c-644f-8e70cd1c0bd9@gmail.com Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com> Reviewed-by: Vegard Nossum <vegard.nossum@oracle.com> Cc: Willy Tarreau <w@1wt.eu> Cc: <socketpair@gmail.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Jens Axboe <axboe@fb.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-12 04:53:37 +08:00
pipe_bufs = 1;
}
pipe: limit the per-user amount of pages allocated in pipes On no-so-small systems, it is possible for a single process to cause an OOM condition by filling large pipes with data that are never read. A typical process filling 4000 pipes with 1 MB of data will use 4 GB of memory. On small systems it may be tricky to set the pipe max size to prevent this from happening. This patch makes it possible to enforce a per-user soft limit above which new pipes will be limited to a single page, effectively limiting them to 4 kB each, as well as a hard limit above which no new pipes may be created for this user. This has the effect of protecting the system against memory abuse without hurting other users, and still allowing pipes to work correctly though with less data at once. The limit are controlled by two new sysctls : pipe-user-pages-soft, and pipe-user-pages-hard. Both may be disabled by setting them to zero. The default soft limit allows the default number of FDs per process (1024) to create pipes of the default size (64kB), thus reaching a limit of 64MB before starting to create only smaller pipes. With 256 processes limited to 1024 FDs each, this results in 1024*64kB + (256*1024 - 1024) * 4kB = 1084 MB of memory allocated for a user. The hard limit is disabled by default to avoid breaking existing applications that make intensive use of pipes (eg: for splicing). Reported-by: socketpair@gmail.com Reported-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Mitigates: CVE-2013-4312 (Linux 2.0+) Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-01-18 23:36:09 +08:00
if (too_many_pipe_buffers_hard(user_bufs))
pipe: fix limit checking in alloc_pipe_info() The limit checking in alloc_pipe_info() (used by pipe(2) and when opening a FIFO) has the following problems: (1) When checking capacity required for the new pipe, the checks against the limit in /proc/sys/fs/pipe-user-pages-{soft,hard} are made against existing consumption, and exclude the memory required for the new pipe capacity. As a consequence: (1) the memory allocation throttling provided by the soft limit does not kick in quite as early as it should, and (2) the user can overrun the hard limit. (2) As currently implemented, accounting and checking against the limits is done as follows: (a) Test whether the user has exceeded the limit. (b) Make new pipe buffer allocation. (c) Account new allocation against the limits. This is racey. Multiple processes may pass point (a) simultaneously, and then allocate pipe buffers that are accounted for only in step (c). The race means that the user's pipe buffer allocation could be pushed over the limit (by an arbitrary amount, depending on how unlucky we were in the race). [Thanks to Vegard Nossum for spotting this point, which I had missed.] This patch addresses the above problems as follows: * Alter the checks against limits to include the memory required for the new pipe. * Re-order the accounting step so that it precedes the buffer allocation. If the accounting step determines that a limit has been reached, revert the accounting and cause the operation to fail. Link: http://lkml.kernel.org/r/8ff3e9f9-23f6-510c-644f-8e70cd1c0bd9@gmail.com Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com> Reviewed-by: Vegard Nossum <vegard.nossum@oracle.com> Cc: Willy Tarreau <w@1wt.eu> Cc: <socketpair@gmail.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Jens Axboe <axboe@fb.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-12 04:53:37 +08:00
goto out_revert_acct;
pipe->bufs = kcalloc(pipe_bufs, sizeof(struct pipe_buffer),
GFP_KERNEL_ACCOUNT);
if (pipe->bufs) {
init_waitqueue_head(&pipe->wait);
pipe->r_counter = pipe->w_counter = 1;
pipe->buffers = pipe_bufs;
pipe->user = user;
mutex_init(&pipe->mutex);
return pipe;
}
pipe: fix limit checking in alloc_pipe_info() The limit checking in alloc_pipe_info() (used by pipe(2) and when opening a FIFO) has the following problems: (1) When checking capacity required for the new pipe, the checks against the limit in /proc/sys/fs/pipe-user-pages-{soft,hard} are made against existing consumption, and exclude the memory required for the new pipe capacity. As a consequence: (1) the memory allocation throttling provided by the soft limit does not kick in quite as early as it should, and (2) the user can overrun the hard limit. (2) As currently implemented, accounting and checking against the limits is done as follows: (a) Test whether the user has exceeded the limit. (b) Make new pipe buffer allocation. (c) Account new allocation against the limits. This is racey. Multiple processes may pass point (a) simultaneously, and then allocate pipe buffers that are accounted for only in step (c). The race means that the user's pipe buffer allocation could be pushed over the limit (by an arbitrary amount, depending on how unlucky we were in the race). [Thanks to Vegard Nossum for spotting this point, which I had missed.] This patch addresses the above problems as follows: * Alter the checks against limits to include the memory required for the new pipe. * Re-order the accounting step so that it precedes the buffer allocation. If the accounting step determines that a limit has been reached, revert the accounting and cause the operation to fail. Link: http://lkml.kernel.org/r/8ff3e9f9-23f6-510c-644f-8e70cd1c0bd9@gmail.com Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com> Reviewed-by: Vegard Nossum <vegard.nossum@oracle.com> Cc: Willy Tarreau <w@1wt.eu> Cc: <socketpair@gmail.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Jens Axboe <axboe@fb.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-12 04:53:37 +08:00
out_revert_acct:
(void) account_pipe_buffers(user, pipe_bufs, 0);
kfree(pipe);
out_free_uid:
free_uid(user);
return NULL;
}
void free_pipe_info(struct pipe_inode_info *pipe)
{
int i;
(void) account_pipe_buffers(pipe->user, pipe->buffers, 0);
pipe: limit the per-user amount of pages allocated in pipes On no-so-small systems, it is possible for a single process to cause an OOM condition by filling large pipes with data that are never read. A typical process filling 4000 pipes with 1 MB of data will use 4 GB of memory. On small systems it may be tricky to set the pipe max size to prevent this from happening. This patch makes it possible to enforce a per-user soft limit above which new pipes will be limited to a single page, effectively limiting them to 4 kB each, as well as a hard limit above which no new pipes may be created for this user. This has the effect of protecting the system against memory abuse without hurting other users, and still allowing pipes to work correctly though with less data at once. The limit are controlled by two new sysctls : pipe-user-pages-soft, and pipe-user-pages-hard. Both may be disabled by setting them to zero. The default soft limit allows the default number of FDs per process (1024) to create pipes of the default size (64kB), thus reaching a limit of 64MB before starting to create only smaller pipes. With 256 processes limited to 1024 FDs each, this results in 1024*64kB + (256*1024 - 1024) * 4kB = 1084 MB of memory allocated for a user. The hard limit is disabled by default to avoid breaking existing applications that make intensive use of pipes (eg: for splicing). Reported-by: socketpair@gmail.com Reported-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Mitigates: CVE-2013-4312 (Linux 2.0+) Suggested-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Willy Tarreau <w@1wt.eu> Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2016-01-18 23:36:09 +08:00
free_uid(pipe->user);
for (i = 0; i < pipe->buffers; i++) {
struct pipe_buffer *buf = pipe->bufs + i;
if (buf->ops)
pipe_buf_release(pipe, buf);
}
if (pipe->tmp_page)
__free_page(pipe->tmp_page);
kfree(pipe->bufs);
kfree(pipe);
}
static struct vfsmount *pipe_mnt __read_mostly;
/*
* pipefs_dname() is called from d_path().
*/
static char *pipefs_dname(struct dentry *dentry, char *buffer, int buflen)
{
return dynamic_dname(dentry, buffer, buflen, "pipe:[%lu]",
d_inode(dentry)->i_ino);
}
static const struct dentry_operations pipefs_dentry_operations = {
.d_dname = pipefs_dname,
};
static struct inode * get_pipe_inode(void)
{
struct inode *inode = new_inode_pseudo(pipe_mnt->mnt_sb);
struct pipe_inode_info *pipe;
if (!inode)
goto fail_inode;
inode->i_ino = get_next_ino();
pipe = alloc_pipe_info();
if (!pipe)
goto fail_iput;
inode->i_pipe = pipe;
pipe->files = 2;
pipe->readers = pipe->writers = 1;
inode->i_fop = &pipefifo_fops;
/*
* Mark the inode dirty from the very beginning,
* that way it will never be moved to the dirty
* list because "mark_inode_dirty()" will think
* that it already _is_ on the dirty list.
*/
inode->i_state = I_DIRTY;
inode->i_mode = S_IFIFO | S_IRUSR | S_IWUSR;
inode->i_uid = current_fsuid();
inode->i_gid = current_fsgid();
inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
return inode;
fail_iput:
iput(inode);
fail_inode:
return NULL;
}
int create_pipe_files(struct file **res, int flags)
{
int err;
struct inode *inode = get_pipe_inode();
struct file *f;
struct path path;
static struct qstr name = { .name = "" };
if (!inode)
return -ENFILE;
err = -ENOMEM;
path.dentry = d_alloc_pseudo(pipe_mnt->mnt_sb, &name);
if (!path.dentry)
goto err_inode;
path.mnt = mntget(pipe_mnt);
d_instantiate(path.dentry, inode);
f = alloc_file(&path, FMODE_WRITE, &pipefifo_fops);
if (IS_ERR(f)) {
err = PTR_ERR(f);
goto err_dentry;
}
pipes: add a "packetized pipe" mode for writing The actual internal pipe implementation is already really about individual packets (called "pipe buffers"), and this simply exposes that as a special packetized mode. When we are in the packetized mode (marked by O_DIRECT as suggested by Alan Cox), a write() on a pipe will not merge the new data with previous writes, so each write will get a pipe buffer of its own. The pipe buffer is then marked with the PIPE_BUF_FLAG_PACKET flag, which in turn will tell the reader side to break the read at that boundary (and throw away any partial packet contents that do not fit in the read buffer). End result: as long as you do writes less than PIPE_BUF in size (so that the pipe doesn't have to split them up), you can now treat the pipe as a packet interface, where each read() system call will read one packet at a time. You can just use a sufficiently big read buffer (PIPE_BUF is sufficient, since bigger than that doesn't guarantee atomicity anyway), and the return value of the read() will naturally give you the size of the packet. NOTE! We do not support zero-sized packets, and zero-sized reads and writes to a pipe continue to be no-ops. Also note that big packets will currently be split at write time, but that the size at which that happens is not really specified (except that it's bigger than PIPE_BUF). Currently that limit is the system page size, but we might want to explicitly support bigger packets some day. The main user for this is going to be the autofs packet interface, allowing us to stop having to care so deeply about exact packet sizes (which have had bugs with 32/64-bit compatibility modes). But user space can create packetized pipes with "pipe2(fd, O_DIRECT)", which will fail with an EINVAL on kernels that do not support this interface. Tested-by: Michael Tokarev <mjt@tls.msk.ru> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Cc: David Miller <davem@davemloft.net> Cc: Ian Kent <raven@themaw.net> Cc: Thomas Meyer <thomas@m3y3r.de> Cc: stable@kernel.org # needed for systemd/autofs interaction fix Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-04-30 04:12:42 +08:00
f->f_flags = O_WRONLY | (flags & (O_NONBLOCK | O_DIRECT));
f->private_data = inode->i_pipe;
res[0] = alloc_file(&path, FMODE_READ, &pipefifo_fops);
if (IS_ERR(res[0])) {
err = PTR_ERR(res[0]);
goto err_file;
}
path_get(&path);
res[0]->private_data = inode->i_pipe;
res[0]->f_flags = O_RDONLY | (flags & O_NONBLOCK);
res[1] = f;
return 0;
err_file:
put_filp(f);
err_dentry:
free_pipe_info(inode->i_pipe);
path_put(&path);
return err;
err_inode:
free_pipe_info(inode->i_pipe);
iput(inode);
return err;
}
static int __do_pipe_flags(int *fd, struct file **files, int flags)
{
int error;
int fdw, fdr;
pipes: add a "packetized pipe" mode for writing The actual internal pipe implementation is already really about individual packets (called "pipe buffers"), and this simply exposes that as a special packetized mode. When we are in the packetized mode (marked by O_DIRECT as suggested by Alan Cox), a write() on a pipe will not merge the new data with previous writes, so each write will get a pipe buffer of its own. The pipe buffer is then marked with the PIPE_BUF_FLAG_PACKET flag, which in turn will tell the reader side to break the read at that boundary (and throw away any partial packet contents that do not fit in the read buffer). End result: as long as you do writes less than PIPE_BUF in size (so that the pipe doesn't have to split them up), you can now treat the pipe as a packet interface, where each read() system call will read one packet at a time. You can just use a sufficiently big read buffer (PIPE_BUF is sufficient, since bigger than that doesn't guarantee atomicity anyway), and the return value of the read() will naturally give you the size of the packet. NOTE! We do not support zero-sized packets, and zero-sized reads and writes to a pipe continue to be no-ops. Also note that big packets will currently be split at write time, but that the size at which that happens is not really specified (except that it's bigger than PIPE_BUF). Currently that limit is the system page size, but we might want to explicitly support bigger packets some day. The main user for this is going to be the autofs packet interface, allowing us to stop having to care so deeply about exact packet sizes (which have had bugs with 32/64-bit compatibility modes). But user space can create packetized pipes with "pipe2(fd, O_DIRECT)", which will fail with an EINVAL on kernels that do not support this interface. Tested-by: Michael Tokarev <mjt@tls.msk.ru> Cc: Alan Cox <alan@lxorguk.ukuu.org.uk> Cc: David Miller <davem@davemloft.net> Cc: Ian Kent <raven@themaw.net> Cc: Thomas Meyer <thomas@m3y3r.de> Cc: stable@kernel.org # needed for systemd/autofs interaction fix Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-04-30 04:12:42 +08:00
if (flags & ~(O_CLOEXEC | O_NONBLOCK | O_DIRECT))
flag parameters: pipe This patch introduces the new syscall pipe2 which is like pipe but it also takes an additional parameter which takes a flag value. This patch implements the handling of O_CLOEXEC for the flag. I did not add support for the new syscall for the architectures which have a special sys_pipe implementation. I think the maintainers of those archs have the chance to go with the unified implementation but that's up to them. The implementation introduces do_pipe_flags. I did that instead of changing all callers of do_pipe because some of the callers are written in assembler. I would probably screw up changing the assembly code. To avoid breaking code do_pipe is now a small wrapper around do_pipe_flags. Once all callers are changed over to do_pipe_flags the old do_pipe function can be removed. The following test must be adjusted for architectures other than x86 and x86-64 and in case the syscall numbers changed. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ #include <fcntl.h> #include <stdio.h> #include <unistd.h> #include <sys/syscall.h> #ifndef __NR_pipe2 # ifdef __x86_64__ # define __NR_pipe2 293 # elif defined __i386__ # define __NR_pipe2 331 # else # error "need __NR_pipe2" # endif #endif int main (void) { int fd[2]; if (syscall (__NR_pipe2, fd, 0) != 0) { puts ("pipe2(0) failed"); return 1; } for (int i = 0; i < 2; ++i) { int coe = fcntl (fd[i], F_GETFD); if (coe == -1) { puts ("fcntl failed"); return 1; } if (coe & FD_CLOEXEC) { printf ("pipe2(0) set close-on-exit for fd[%d]\n", i); return 1; } } close (fd[0]); close (fd[1]); if (syscall (__NR_pipe2, fd, O_CLOEXEC) != 0) { puts ("pipe2(O_CLOEXEC) failed"); return 1; } for (int i = 0; i < 2; ++i) { int coe = fcntl (fd[i], F_GETFD); if (coe == -1) { puts ("fcntl failed"); return 1; } if ((coe & FD_CLOEXEC) == 0) { printf ("pipe2(O_CLOEXEC) does not set close-on-exit for fd[%d]\n", i); return 1; } } close (fd[0]); close (fd[1]); puts ("OK"); return 0; } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Signed-off-by: Ulrich Drepper <drepper@redhat.com> Acked-by: Davide Libenzi <davidel@xmailserver.org> Cc: Michael Kerrisk <mtk.manpages@googlemail.com> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:29:30 +08:00
return -EINVAL;
error = create_pipe_files(files, flags);
if (error)
return error;
flag parameters: pipe This patch introduces the new syscall pipe2 which is like pipe but it also takes an additional parameter which takes a flag value. This patch implements the handling of O_CLOEXEC for the flag. I did not add support for the new syscall for the architectures which have a special sys_pipe implementation. I think the maintainers of those archs have the chance to go with the unified implementation but that's up to them. The implementation introduces do_pipe_flags. I did that instead of changing all callers of do_pipe because some of the callers are written in assembler. I would probably screw up changing the assembly code. To avoid breaking code do_pipe is now a small wrapper around do_pipe_flags. Once all callers are changed over to do_pipe_flags the old do_pipe function can be removed. The following test must be adjusted for architectures other than x86 and x86-64 and in case the syscall numbers changed. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ #include <fcntl.h> #include <stdio.h> #include <unistd.h> #include <sys/syscall.h> #ifndef __NR_pipe2 # ifdef __x86_64__ # define __NR_pipe2 293 # elif defined __i386__ # define __NR_pipe2 331 # else # error "need __NR_pipe2" # endif #endif int main (void) { int fd[2]; if (syscall (__NR_pipe2, fd, 0) != 0) { puts ("pipe2(0) failed"); return 1; } for (int i = 0; i < 2; ++i) { int coe = fcntl (fd[i], F_GETFD); if (coe == -1) { puts ("fcntl failed"); return 1; } if (coe & FD_CLOEXEC) { printf ("pipe2(0) set close-on-exit for fd[%d]\n", i); return 1; } } close (fd[0]); close (fd[1]); if (syscall (__NR_pipe2, fd, O_CLOEXEC) != 0) { puts ("pipe2(O_CLOEXEC) failed"); return 1; } for (int i = 0; i < 2; ++i) { int coe = fcntl (fd[i], F_GETFD); if (coe == -1) { puts ("fcntl failed"); return 1; } if ((coe & FD_CLOEXEC) == 0) { printf ("pipe2(O_CLOEXEC) does not set close-on-exit for fd[%d]\n", i); return 1; } } close (fd[0]); close (fd[1]); puts ("OK"); return 0; } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Signed-off-by: Ulrich Drepper <drepper@redhat.com> Acked-by: Davide Libenzi <davidel@xmailserver.org> Cc: Michael Kerrisk <mtk.manpages@googlemail.com> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:29:30 +08:00
error = get_unused_fd_flags(flags);
if (error < 0)
goto err_read_pipe;
fdr = error;
flag parameters: pipe This patch introduces the new syscall pipe2 which is like pipe but it also takes an additional parameter which takes a flag value. This patch implements the handling of O_CLOEXEC for the flag. I did not add support for the new syscall for the architectures which have a special sys_pipe implementation. I think the maintainers of those archs have the chance to go with the unified implementation but that's up to them. The implementation introduces do_pipe_flags. I did that instead of changing all callers of do_pipe because some of the callers are written in assembler. I would probably screw up changing the assembly code. To avoid breaking code do_pipe is now a small wrapper around do_pipe_flags. Once all callers are changed over to do_pipe_flags the old do_pipe function can be removed. The following test must be adjusted for architectures other than x86 and x86-64 and in case the syscall numbers changed. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ #include <fcntl.h> #include <stdio.h> #include <unistd.h> #include <sys/syscall.h> #ifndef __NR_pipe2 # ifdef __x86_64__ # define __NR_pipe2 293 # elif defined __i386__ # define __NR_pipe2 331 # else # error "need __NR_pipe2" # endif #endif int main (void) { int fd[2]; if (syscall (__NR_pipe2, fd, 0) != 0) { puts ("pipe2(0) failed"); return 1; } for (int i = 0; i < 2; ++i) { int coe = fcntl (fd[i], F_GETFD); if (coe == -1) { puts ("fcntl failed"); return 1; } if (coe & FD_CLOEXEC) { printf ("pipe2(0) set close-on-exit for fd[%d]\n", i); return 1; } } close (fd[0]); close (fd[1]); if (syscall (__NR_pipe2, fd, O_CLOEXEC) != 0) { puts ("pipe2(O_CLOEXEC) failed"); return 1; } for (int i = 0; i < 2; ++i) { int coe = fcntl (fd[i], F_GETFD); if (coe == -1) { puts ("fcntl failed"); return 1; } if ((coe & FD_CLOEXEC) == 0) { printf ("pipe2(O_CLOEXEC) does not set close-on-exit for fd[%d]\n", i); return 1; } } close (fd[0]); close (fd[1]); puts ("OK"); return 0; } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Signed-off-by: Ulrich Drepper <drepper@redhat.com> Acked-by: Davide Libenzi <davidel@xmailserver.org> Cc: Michael Kerrisk <mtk.manpages@googlemail.com> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:29:30 +08:00
error = get_unused_fd_flags(flags);
if (error < 0)
goto err_fdr;
fdw = error;
audit_fd_pair(fdr, fdw);
fd[0] = fdr;
fd[1] = fdw;
return 0;
err_fdr:
put_unused_fd(fdr);
err_read_pipe:
fput(files[0]);
fput(files[1]);
return error;
}
int do_pipe_flags(int *fd, int flags)
{
struct file *files[2];
int error = __do_pipe_flags(fd, files, flags);
if (!error) {
fd_install(fd[0], files[0]);
fd_install(fd[1], files[1]);
}
return error;
}
/*
* sys_pipe() is the normal C calling standard for creating
* a pipe. It's not the way Unix traditionally does this, though.
*/
SYSCALL_DEFINE2(pipe2, int __user *, fildes, int, flags)
{
struct file *files[2];
int fd[2];
int error;
error = __do_pipe_flags(fd, files, flags);
if (!error) {
if (unlikely(copy_to_user(fildes, fd, sizeof(fd)))) {
fput(files[0]);
fput(files[1]);
put_unused_fd(fd[0]);
put_unused_fd(fd[1]);
error = -EFAULT;
} else {
fd_install(fd[0], files[0]);
fd_install(fd[1], files[1]);
}
}
return error;
}
SYSCALL_DEFINE1(pipe, int __user *, fildes)
flag parameters: pipe This patch introduces the new syscall pipe2 which is like pipe but it also takes an additional parameter which takes a flag value. This patch implements the handling of O_CLOEXEC for the flag. I did not add support for the new syscall for the architectures which have a special sys_pipe implementation. I think the maintainers of those archs have the chance to go with the unified implementation but that's up to them. The implementation introduces do_pipe_flags. I did that instead of changing all callers of do_pipe because some of the callers are written in assembler. I would probably screw up changing the assembly code. To avoid breaking code do_pipe is now a small wrapper around do_pipe_flags. Once all callers are changed over to do_pipe_flags the old do_pipe function can be removed. The following test must be adjusted for architectures other than x86 and x86-64 and in case the syscall numbers changed. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ #include <fcntl.h> #include <stdio.h> #include <unistd.h> #include <sys/syscall.h> #ifndef __NR_pipe2 # ifdef __x86_64__ # define __NR_pipe2 293 # elif defined __i386__ # define __NR_pipe2 331 # else # error "need __NR_pipe2" # endif #endif int main (void) { int fd[2]; if (syscall (__NR_pipe2, fd, 0) != 0) { puts ("pipe2(0) failed"); return 1; } for (int i = 0; i < 2; ++i) { int coe = fcntl (fd[i], F_GETFD); if (coe == -1) { puts ("fcntl failed"); return 1; } if (coe & FD_CLOEXEC) { printf ("pipe2(0) set close-on-exit for fd[%d]\n", i); return 1; } } close (fd[0]); close (fd[1]); if (syscall (__NR_pipe2, fd, O_CLOEXEC) != 0) { puts ("pipe2(O_CLOEXEC) failed"); return 1; } for (int i = 0; i < 2; ++i) { int coe = fcntl (fd[i], F_GETFD); if (coe == -1) { puts ("fcntl failed"); return 1; } if ((coe & FD_CLOEXEC) == 0) { printf ("pipe2(O_CLOEXEC) does not set close-on-exit for fd[%d]\n", i); return 1; } } close (fd[0]); close (fd[1]); puts ("OK"); return 0; } ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Signed-off-by: Ulrich Drepper <drepper@redhat.com> Acked-by: Davide Libenzi <davidel@xmailserver.org> Cc: Michael Kerrisk <mtk.manpages@googlemail.com> Cc: <linux-arch@vger.kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-24 12:29:30 +08:00
{
return sys_pipe2(fildes, 0);
}
static int wait_for_partner(struct pipe_inode_info *pipe, unsigned int *cnt)
{
int cur = *cnt;
while (cur == *cnt) {
pipe_wait(pipe);
if (signal_pending(current))
break;
}
return cur == *cnt ? -ERESTARTSYS : 0;
}
static void wake_up_partner(struct pipe_inode_info *pipe)
{
wake_up_interruptible(&pipe->wait);
}
static int fifo_open(struct inode *inode, struct file *filp)
{
struct pipe_inode_info *pipe;
bool is_pipe = inode->i_sb->s_magic == PIPEFS_MAGIC;
int ret;
filp->f_version = 0;
spin_lock(&inode->i_lock);
if (inode->i_pipe) {
pipe = inode->i_pipe;
pipe->files++;
spin_unlock(&inode->i_lock);
} else {
spin_unlock(&inode->i_lock);
pipe = alloc_pipe_info();
if (!pipe)
return -ENOMEM;
pipe->files = 1;
spin_lock(&inode->i_lock);
if (unlikely(inode->i_pipe)) {
inode->i_pipe->files++;
spin_unlock(&inode->i_lock);
free_pipe_info(pipe);
pipe = inode->i_pipe;
} else {
inode->i_pipe = pipe;
spin_unlock(&inode->i_lock);
}
}
filp->private_data = pipe;
/* OK, we have a pipe and it's pinned down */
__pipe_lock(pipe);
/* We can only do regular read/write on fifos */
filp->f_mode &= (FMODE_READ | FMODE_WRITE);
switch (filp->f_mode) {
case FMODE_READ:
/*
* O_RDONLY
* POSIX.1 says that O_NONBLOCK means return with the FIFO
* opened, even when there is no process writing the FIFO.
*/
pipe->r_counter++;
if (pipe->readers++ == 0)
wake_up_partner(pipe);
if (!is_pipe && !pipe->writers) {
if ((filp->f_flags & O_NONBLOCK)) {
/* suppress POLLHUP until we have
* seen a writer */
filp->f_version = pipe->w_counter;
} else {
if (wait_for_partner(pipe, &pipe->w_counter))
goto err_rd;
}
}
break;
case FMODE_WRITE:
/*
* O_WRONLY
* POSIX.1 says that O_NONBLOCK means return -1 with
* errno=ENXIO when there is no process reading the FIFO.
*/
ret = -ENXIO;
if (!is_pipe && (filp->f_flags & O_NONBLOCK) && !pipe->readers)
goto err;
pipe->w_counter++;
if (!pipe->writers++)
wake_up_partner(pipe);
if (!is_pipe && !pipe->readers) {
if (wait_for_partner(pipe, &pipe->r_counter))
goto err_wr;
}
break;
case FMODE_READ | FMODE_WRITE:
/*
* O_RDWR
* POSIX.1 leaves this case "undefined" when O_NONBLOCK is set.
* This implementation will NEVER block on a O_RDWR open, since
* the process can at least talk to itself.
*/
pipe->readers++;
pipe->writers++;
pipe->r_counter++;
pipe->w_counter++;
if (pipe->readers == 1 || pipe->writers == 1)
wake_up_partner(pipe);
break;
default:
ret = -EINVAL;
goto err;
}
/* Ok! */
__pipe_unlock(pipe);
return 0;
err_rd:
if (!--pipe->readers)
wake_up_interruptible(&pipe->wait);
ret = -ERESTARTSYS;
goto err;
err_wr:
if (!--pipe->writers)
wake_up_interruptible(&pipe->wait);
ret = -ERESTARTSYS;
goto err;
err:
__pipe_unlock(pipe);
vfs: fix subtle use-after-free of pipe_inode_info The pipe code was trying (and failing) to be very careful about freeing the pipe info only after the last access, with a pattern like: spin_lock(&inode->i_lock); if (!--pipe->files) { inode->i_pipe = NULL; kill = 1; } spin_unlock(&inode->i_lock); __pipe_unlock(pipe); if (kill) free_pipe_info(pipe); where the final freeing is done last. HOWEVER. The above is actually broken, because while the freeing is done at the end, if we have two racing processes releasing the pipe inode info, the one that *doesn't* free it will decrement the ->files count, and unlock the inode i_lock, but then still use the "pipe_inode_info" afterwards when it does the "__pipe_unlock(pipe)". This is *very* hard to trigger in practice, since the race window is very small, and adding debug options seems to just hide it by slowing things down. Simon originally reported this way back in July as an Oops in kmem_cache_allocate due to a single bit corruption (due to the final "spin_unlock(pipe->mutex.wait_lock)" incrementing a field in a different allocation that had re-used the free'd pipe-info), it's taken this long to figure out. Since the 'pipe->files' accesses aren't even protected by the pipe lock (we very much use the inode lock for that), the simple solution is to just drop the pipe lock early. And since there were two users of this pattern, create a helper function for it. Introduced commit ba5bb147330a ("pipe: take allocation and freeing of pipe_inode_info out of ->i_mutex"). Reported-by: Simon Kirby <sim@hostway.ca> Reported-by: Ian Applegate <ia@cloudflare.com> Acked-by: Al Viro <viro@zeniv.linux.org.uk> Cc: stable@kernel.org # v3.10+ Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-12-03 01:44:51 +08:00
put_pipe_info(inode, pipe);
return ret;
}
const struct file_operations pipefifo_fops = {
.open = fifo_open,
.llseek = no_llseek,
.read_iter = pipe_read,
.write_iter = pipe_write,
.poll = pipe_poll,
.unlocked_ioctl = pipe_ioctl,
.release = pipe_release,
.fasync = pipe_fasync,
};
pipe: relocate round_pipe_size() above pipe_set_size() Patch series "pipe: fix limit handling", v2. When changing a pipe's capacity with fcntl(F_SETPIPE_SZ), various limits defined by /proc/sys/fs/pipe-* files are checked to see if unprivileged users are exceeding limits on memory consumption. While documenting and testing the operation of these limits I noticed that, as currently implemented, these checks have a number of problems: (1) When increasing the pipe capacity, the checks against the limits in /proc/sys/fs/pipe-user-pages-{soft,hard} are made against existing consumption, and exclude the memory required for the increased pipe capacity. The new increase in pipe capacity can then push the total memory used by the user for pipes (possibly far) over a limit. This can also trigger the problem described next. (2) The limit checks are performed even when the new pipe capacity is less than the existing pipe capacity. This can lead to problems if a user sets a large pipe capacity, and then the limits are lowered, with the result that the user will no longer be able to decrease the pipe capacity. (3) As currently implemented, accounting and checking against the limits is done as follows: (a) Test whether the user has exceeded the limit. (b) Make new pipe buffer allocation. (c) Account new allocation against the limits. This is racey. Multiple processes may pass point (a) simultaneously, and then allocate pipe buffers that are accounted for only in step (c). The race means that the user's pipe buffer allocation could be pushed over the limit (by an arbitrary amount, depending on how unlucky we were in the race). [Thanks to Vegard Nossum for spotting this point, which I had missed.] This patch series addresses these three problems. This patch (of 8): This is a minor preparatory patch. After subsequent patches, round_pipe_size() will be called from pipe_set_size(), so place round_pipe_size() above pipe_set_size(). Link: http://lkml.kernel.org/r/91a91fdb-a959-ba7f-b551-b62477cc98a1@gmail.com Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com> Reviewed-by: Vegard Nossum <vegard.nossum@oracle.com> Cc: Willy Tarreau <w@1wt.eu> Cc: <socketpair@gmail.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Jens Axboe <axboe@fb.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-12 04:53:22 +08:00
/*
* Currently we rely on the pipe array holding a power-of-2 number
* of pages.
*/
static inline unsigned int round_pipe_size(unsigned int size)
{
unsigned long nr_pages;
nr_pages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
return roundup_pow_of_two(nr_pages) << PAGE_SHIFT;
}
/*
* Allocate a new array of pipe buffers and copy the info over. Returns the
* pipe size if successful, or return -ERROR on error.
*/
static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg)
{
struct pipe_buffer *bufs;
unsigned int size, nr_pages;
unsigned long user_bufs;
pipe: fix limit checking in pipe_set_size() The limit checking in pipe_set_size() (used by fcntl(F_SETPIPE_SZ)) has the following problems: (1) When increasing the pipe capacity, the checks against the limits in /proc/sys/fs/pipe-user-pages-{soft,hard} are made against existing consumption, and exclude the memory required for the increased pipe capacity. The new increase in pipe capacity can then push the total memory used by the user for pipes (possibly far) over a limit. This can also trigger the problem described next. (2) The limit checks are performed even when the new pipe capacity is less than the existing pipe capacity. This can lead to problems if a user sets a large pipe capacity, and then the limits are lowered, with the result that the user will no longer be able to decrease the pipe capacity. (3) As currently implemented, accounting and checking against the limits is done as follows: (a) Test whether the user has exceeded the limit. (b) Make new pipe buffer allocation. (c) Account new allocation against the limits. This is racey. Multiple processes may pass point (a) simultaneously, and then allocate pipe buffers that are accounted for only in step (c). The race means that the user's pipe buffer allocation could be pushed over the limit (by an arbitrary amount, depending on how unlucky we were in the race). [Thanks to Vegard Nossum for spotting this point, which I had missed.] This patch addresses the above problems as follows: * Perform checks against the limits only when increasing a pipe's capacity; an unprivileged user can always decrease a pipe's capacity. * Alter the checks against limits to include the memory required for the new pipe capacity. * Re-order the accounting step so that it precedes the buffer allocation. If the accounting step determines that a limit has been reached, revert the accounting and cause the operation to fail. The program below can be used to demonstrate problems 1 and 2, and the effect of the fix. The program takes one or more command-line arguments. The first argument specifies the number of pipes that the program should create. The remaining arguments are, alternately, pipe capacities that should be set using fcntl(F_SETPIPE_SZ), and sleep intervals (in seconds) between the fcntl() operations. (The sleep intervals allow the possibility to change the limits between fcntl() operations.) Problem 1 ========= Using the test program on an unpatched kernel, we first set some limits: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # 40.96 MB Then show that we can set a pipe with capacity (100MB) that is over the hard limit # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes F_SETPIPE_SZ returned 134217728 Now set the capacity to 100MB twice. The second call fails (which is probably surprising to most users, since it seems like a no-op): # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 0 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes F_SETPIPE_SZ returned 134217728 Loop 2: set pipe capacity to 100000000 bytes Loop 2, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted With a patched kernel, setting a capacity over the limit fails at the first attempt: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes Loop 1, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted There is a small chance that the change to fix this problem could break user-space, since there are cases where fcntl(F_SETPIPE_SZ) calls that previously succeeded might fail. However, the chances are small, since (a) the pipe-user-pages-{soft,hard} limits are new (in 4.5), and the default soft/hard limits are high/unlimited. Therefore, it seems warranted to make these limits operate more precisely (and behave more like what users probably expect). Problem 2 ========= Running the test program on an unpatched kernel, we first set some limits: # getconf PAGESIZE 4096 # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # 40.96 MB Now perform two fcntl(F_SETPIPE_SZ) operations on a single pipe, first setting a pipe capacity (10MB), sleeping for a few seconds, during which time the hard limit is lowered, and then set pipe capacity to a smaller amount (5MB): # sudo -u mtk ./test_F_SETPIPE_SZ 1 10000000 15 5000000 & [1] 748 # Initial pipe capacity: 65536 Loop 1: set pipe capacity to 10000000 bytes F_SETPIPE_SZ returned 16777216 Sleeping 15 seconds # echo 1000 > /proc/sys/fs/pipe-user-pages-hard # 4.096 MB # Loop 2: set pipe capacity to 5000000 bytes Loop 2, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted In this case, the user should be able to lower the limit. With a kernel that has the patch below, the second fcntl() succeeds: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # sudo -u mtk ./test_F_SETPIPE_SZ 1 10000000 15 5000000 & [1] 3215 # Initial pipe capacity: 65536 # Loop 1: set pipe capacity to 10000000 bytes F_SETPIPE_SZ returned 16777216 Sleeping 15 seconds # echo 1000 > /proc/sys/fs/pipe-user-pages-hard # Loop 2: set pipe capacity to 5000000 bytes F_SETPIPE_SZ returned 8388608 8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x--- /* test_F_SETPIPE_SZ.c (C) 2016, Michael Kerrisk; licensed under GNU GPL version 2 or later Test operation of fcntl(F_SETPIPE_SZ) for setting pipe capacity and interactions with limits defined by /proc/sys/fs/pipe-* files. */ #define _GNU_SOURCE #include <stdio.h> #include <stdlib.h> #include <fcntl.h> #include <unistd.h> int main(int argc, char *argv[]) { int (*pfd)[2]; int npipes; int pcap, rcap; int j, p, s, stime, loop; if (argc < 2) { fprintf(stderr, "Usage: %s num-pipes " "[pipe-capacity sleep-time]...\n", argv[0]); exit(EXIT_FAILURE); } npipes = atoi(argv[1]); pfd = calloc(npipes, sizeof (int [2])); if (pfd == NULL) { perror("calloc"); exit(EXIT_FAILURE); } for (j = 0; j < npipes; j++) { if (pipe(pfd[j]) == -1) { fprintf(stderr, "Loop %d: pipe() failed: ", j); perror("pipe"); exit(EXIT_FAILURE); } } printf("Initial pipe capacity: %d\n", fcntl(pfd[0][0], F_GETPIPE_SZ)); for (j = 2; j < argc; j += 2 ) { loop = j / 2; pcap = atoi(argv[j]); printf(" Loop %d: set pipe capacity to %d bytes\n", loop, pcap); for (p = 0; p < npipes; p++) { s = fcntl(pfd[p][0], F_SETPIPE_SZ, pcap); if (s == -1) { fprintf(stderr, " Loop %d, pipe %d: F_SETPIPE_SZ " "failed: ", loop, p); perror("fcntl"); exit(EXIT_FAILURE); } if (p == 0) { printf(" F_SETPIPE_SZ returned %d\n", s); rcap = s; } else { if (s != rcap) { fprintf(stderr, " Loop %d, pipe %d: F_SETPIPE_SZ " "unexpected return: %d\n", loop, p, s); exit(EXIT_FAILURE); } } stime = (j + 1 < argc) ? atoi(argv[j + 1]) : 0; if (stime > 0) { printf(" Sleeping %d seconds\n", stime); sleep(stime); } } } exit(EXIT_SUCCESS); } 8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x--- Patch history: v2 * Switch order of test in 'if' statement to avoid function call (to capability()) in normal path. [This is a fix to a preexisting wart in the code. Thanks to Willy Tarreau] * Perform (size > pipe_max_size) check before calling account_pipe_buffers(). [Thanks to Vegard Nossum] Quoting Vegard: The potential problem happens if the user passes a very large number which will overflow pipe->user->pipe_bufs. On 32-bit, sizeof(int) == sizeof(long), so if they pass arg = INT_MAX then round_pipe_size() returns INT_MAX. Although it's true that the accounting is done in terms of pages and not bytes, so you'd need on the order of (1 << 13) = 8192 processes hitting the limit at the same time in order to make it overflow, which seems a bit unlikely. (See https://lkml.org/lkml/2016/8/12/215 for another discussion on the limit checking) Link: http://lkml.kernel.org/r/1e464945-536b-2420-798b-e77b9c7e8593@gmail.com Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com> Reviewed-by: Vegard Nossum <vegard.nossum@oracle.com> Cc: Willy Tarreau <w@1wt.eu> Cc: <socketpair@gmail.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Jens Axboe <axboe@fb.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-12 04:53:31 +08:00
long ret = 0;
size = round_pipe_size(arg);
nr_pages = size >> PAGE_SHIFT;
if (!nr_pages)
return -EINVAL;
pipe: fix limit checking in pipe_set_size() The limit checking in pipe_set_size() (used by fcntl(F_SETPIPE_SZ)) has the following problems: (1) When increasing the pipe capacity, the checks against the limits in /proc/sys/fs/pipe-user-pages-{soft,hard} are made against existing consumption, and exclude the memory required for the increased pipe capacity. The new increase in pipe capacity can then push the total memory used by the user for pipes (possibly far) over a limit. This can also trigger the problem described next. (2) The limit checks are performed even when the new pipe capacity is less than the existing pipe capacity. This can lead to problems if a user sets a large pipe capacity, and then the limits are lowered, with the result that the user will no longer be able to decrease the pipe capacity. (3) As currently implemented, accounting and checking against the limits is done as follows: (a) Test whether the user has exceeded the limit. (b) Make new pipe buffer allocation. (c) Account new allocation against the limits. This is racey. Multiple processes may pass point (a) simultaneously, and then allocate pipe buffers that are accounted for only in step (c). The race means that the user's pipe buffer allocation could be pushed over the limit (by an arbitrary amount, depending on how unlucky we were in the race). [Thanks to Vegard Nossum for spotting this point, which I had missed.] This patch addresses the above problems as follows: * Perform checks against the limits only when increasing a pipe's capacity; an unprivileged user can always decrease a pipe's capacity. * Alter the checks against limits to include the memory required for the new pipe capacity. * Re-order the accounting step so that it precedes the buffer allocation. If the accounting step determines that a limit has been reached, revert the accounting and cause the operation to fail. The program below can be used to demonstrate problems 1 and 2, and the effect of the fix. The program takes one or more command-line arguments. The first argument specifies the number of pipes that the program should create. The remaining arguments are, alternately, pipe capacities that should be set using fcntl(F_SETPIPE_SZ), and sleep intervals (in seconds) between the fcntl() operations. (The sleep intervals allow the possibility to change the limits between fcntl() operations.) Problem 1 ========= Using the test program on an unpatched kernel, we first set some limits: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # 40.96 MB Then show that we can set a pipe with capacity (100MB) that is over the hard limit # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes F_SETPIPE_SZ returned 134217728 Now set the capacity to 100MB twice. The second call fails (which is probably surprising to most users, since it seems like a no-op): # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 0 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes F_SETPIPE_SZ returned 134217728 Loop 2: set pipe capacity to 100000000 bytes Loop 2, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted With a patched kernel, setting a capacity over the limit fails at the first attempt: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes Loop 1, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted There is a small chance that the change to fix this problem could break user-space, since there are cases where fcntl(F_SETPIPE_SZ) calls that previously succeeded might fail. However, the chances are small, since (a) the pipe-user-pages-{soft,hard} limits are new (in 4.5), and the default soft/hard limits are high/unlimited. Therefore, it seems warranted to make these limits operate more precisely (and behave more like what users probably expect). Problem 2 ========= Running the test program on an unpatched kernel, we first set some limits: # getconf PAGESIZE 4096 # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # 40.96 MB Now perform two fcntl(F_SETPIPE_SZ) operations on a single pipe, first setting a pipe capacity (10MB), sleeping for a few seconds, during which time the hard limit is lowered, and then set pipe capacity to a smaller amount (5MB): # sudo -u mtk ./test_F_SETPIPE_SZ 1 10000000 15 5000000 & [1] 748 # Initial pipe capacity: 65536 Loop 1: set pipe capacity to 10000000 bytes F_SETPIPE_SZ returned 16777216 Sleeping 15 seconds # echo 1000 > /proc/sys/fs/pipe-user-pages-hard # 4.096 MB # Loop 2: set pipe capacity to 5000000 bytes Loop 2, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted In this case, the user should be able to lower the limit. With a kernel that has the patch below, the second fcntl() succeeds: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # sudo -u mtk ./test_F_SETPIPE_SZ 1 10000000 15 5000000 & [1] 3215 # Initial pipe capacity: 65536 # Loop 1: set pipe capacity to 10000000 bytes F_SETPIPE_SZ returned 16777216 Sleeping 15 seconds # echo 1000 > /proc/sys/fs/pipe-user-pages-hard # Loop 2: set pipe capacity to 5000000 bytes F_SETPIPE_SZ returned 8388608 8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x--- /* test_F_SETPIPE_SZ.c (C) 2016, Michael Kerrisk; licensed under GNU GPL version 2 or later Test operation of fcntl(F_SETPIPE_SZ) for setting pipe capacity and interactions with limits defined by /proc/sys/fs/pipe-* files. */ #define _GNU_SOURCE #include <stdio.h> #include <stdlib.h> #include <fcntl.h> #include <unistd.h> int main(int argc, char *argv[]) { int (*pfd)[2]; int npipes; int pcap, rcap; int j, p, s, stime, loop; if (argc < 2) { fprintf(stderr, "Usage: %s num-pipes " "[pipe-capacity sleep-time]...\n", argv[0]); exit(EXIT_FAILURE); } npipes = atoi(argv[1]); pfd = calloc(npipes, sizeof (int [2])); if (pfd == NULL) { perror("calloc"); exit(EXIT_FAILURE); } for (j = 0; j < npipes; j++) { if (pipe(pfd[j]) == -1) { fprintf(stderr, "Loop %d: pipe() failed: ", j); perror("pipe"); exit(EXIT_FAILURE); } } printf("Initial pipe capacity: %d\n", fcntl(pfd[0][0], F_GETPIPE_SZ)); for (j = 2; j < argc; j += 2 ) { loop = j / 2; pcap = atoi(argv[j]); printf(" Loop %d: set pipe capacity to %d bytes\n", loop, pcap); for (p = 0; p < npipes; p++) { s = fcntl(pfd[p][0], F_SETPIPE_SZ, pcap); if (s == -1) { fprintf(stderr, " Loop %d, pipe %d: F_SETPIPE_SZ " "failed: ", loop, p); perror("fcntl"); exit(EXIT_FAILURE); } if (p == 0) { printf(" F_SETPIPE_SZ returned %d\n", s); rcap = s; } else { if (s != rcap) { fprintf(stderr, " Loop %d, pipe %d: F_SETPIPE_SZ " "unexpected return: %d\n", loop, p, s); exit(EXIT_FAILURE); } } stime = (j + 1 < argc) ? atoi(argv[j + 1]) : 0; if (stime > 0) { printf(" Sleeping %d seconds\n", stime); sleep(stime); } } } exit(EXIT_SUCCESS); } 8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x--- Patch history: v2 * Switch order of test in 'if' statement to avoid function call (to capability()) in normal path. [This is a fix to a preexisting wart in the code. Thanks to Willy Tarreau] * Perform (size > pipe_max_size) check before calling account_pipe_buffers(). [Thanks to Vegard Nossum] Quoting Vegard: The potential problem happens if the user passes a very large number which will overflow pipe->user->pipe_bufs. On 32-bit, sizeof(int) == sizeof(long), so if they pass arg = INT_MAX then round_pipe_size() returns INT_MAX. Although it's true that the accounting is done in terms of pages and not bytes, so you'd need on the order of (1 << 13) = 8192 processes hitting the limit at the same time in order to make it overflow, which seems a bit unlikely. (See https://lkml.org/lkml/2016/8/12/215 for another discussion on the limit checking) Link: http://lkml.kernel.org/r/1e464945-536b-2420-798b-e77b9c7e8593@gmail.com Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com> Reviewed-by: Vegard Nossum <vegard.nossum@oracle.com> Cc: Willy Tarreau <w@1wt.eu> Cc: <socketpair@gmail.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Jens Axboe <axboe@fb.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-12 04:53:31 +08:00
/*
* If trying to increase the pipe capacity, check that an
* unprivileged user is not trying to exceed various limits
* (soft limit check here, hard limit check just below).
* Decreasing the pipe capacity is always permitted, even
* if the user is currently over a limit.
*/
if (nr_pages > pipe->buffers &&
size > pipe_max_size && !capable(CAP_SYS_RESOURCE))
return -EPERM;
user_bufs = account_pipe_buffers(pipe->user, pipe->buffers, nr_pages);
pipe: fix limit checking in pipe_set_size() The limit checking in pipe_set_size() (used by fcntl(F_SETPIPE_SZ)) has the following problems: (1) When increasing the pipe capacity, the checks against the limits in /proc/sys/fs/pipe-user-pages-{soft,hard} are made against existing consumption, and exclude the memory required for the increased pipe capacity. The new increase in pipe capacity can then push the total memory used by the user for pipes (possibly far) over a limit. This can also trigger the problem described next. (2) The limit checks are performed even when the new pipe capacity is less than the existing pipe capacity. This can lead to problems if a user sets a large pipe capacity, and then the limits are lowered, with the result that the user will no longer be able to decrease the pipe capacity. (3) As currently implemented, accounting and checking against the limits is done as follows: (a) Test whether the user has exceeded the limit. (b) Make new pipe buffer allocation. (c) Account new allocation against the limits. This is racey. Multiple processes may pass point (a) simultaneously, and then allocate pipe buffers that are accounted for only in step (c). The race means that the user's pipe buffer allocation could be pushed over the limit (by an arbitrary amount, depending on how unlucky we were in the race). [Thanks to Vegard Nossum for spotting this point, which I had missed.] This patch addresses the above problems as follows: * Perform checks against the limits only when increasing a pipe's capacity; an unprivileged user can always decrease a pipe's capacity. * Alter the checks against limits to include the memory required for the new pipe capacity. * Re-order the accounting step so that it precedes the buffer allocation. If the accounting step determines that a limit has been reached, revert the accounting and cause the operation to fail. The program below can be used to demonstrate problems 1 and 2, and the effect of the fix. The program takes one or more command-line arguments. The first argument specifies the number of pipes that the program should create. The remaining arguments are, alternately, pipe capacities that should be set using fcntl(F_SETPIPE_SZ), and sleep intervals (in seconds) between the fcntl() operations. (The sleep intervals allow the possibility to change the limits between fcntl() operations.) Problem 1 ========= Using the test program on an unpatched kernel, we first set some limits: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # 40.96 MB Then show that we can set a pipe with capacity (100MB) that is over the hard limit # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes F_SETPIPE_SZ returned 134217728 Now set the capacity to 100MB twice. The second call fails (which is probably surprising to most users, since it seems like a no-op): # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 0 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes F_SETPIPE_SZ returned 134217728 Loop 2: set pipe capacity to 100000000 bytes Loop 2, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted With a patched kernel, setting a capacity over the limit fails at the first attempt: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes Loop 1, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted There is a small chance that the change to fix this problem could break user-space, since there are cases where fcntl(F_SETPIPE_SZ) calls that previously succeeded might fail. However, the chances are small, since (a) the pipe-user-pages-{soft,hard} limits are new (in 4.5), and the default soft/hard limits are high/unlimited. Therefore, it seems warranted to make these limits operate more precisely (and behave more like what users probably expect). Problem 2 ========= Running the test program on an unpatched kernel, we first set some limits: # getconf PAGESIZE 4096 # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # 40.96 MB Now perform two fcntl(F_SETPIPE_SZ) operations on a single pipe, first setting a pipe capacity (10MB), sleeping for a few seconds, during which time the hard limit is lowered, and then set pipe capacity to a smaller amount (5MB): # sudo -u mtk ./test_F_SETPIPE_SZ 1 10000000 15 5000000 & [1] 748 # Initial pipe capacity: 65536 Loop 1: set pipe capacity to 10000000 bytes F_SETPIPE_SZ returned 16777216 Sleeping 15 seconds # echo 1000 > /proc/sys/fs/pipe-user-pages-hard # 4.096 MB # Loop 2: set pipe capacity to 5000000 bytes Loop 2, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted In this case, the user should be able to lower the limit. With a kernel that has the patch below, the second fcntl() succeeds: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # sudo -u mtk ./test_F_SETPIPE_SZ 1 10000000 15 5000000 & [1] 3215 # Initial pipe capacity: 65536 # Loop 1: set pipe capacity to 10000000 bytes F_SETPIPE_SZ returned 16777216 Sleeping 15 seconds # echo 1000 > /proc/sys/fs/pipe-user-pages-hard # Loop 2: set pipe capacity to 5000000 bytes F_SETPIPE_SZ returned 8388608 8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x--- /* test_F_SETPIPE_SZ.c (C) 2016, Michael Kerrisk; licensed under GNU GPL version 2 or later Test operation of fcntl(F_SETPIPE_SZ) for setting pipe capacity and interactions with limits defined by /proc/sys/fs/pipe-* files. */ #define _GNU_SOURCE #include <stdio.h> #include <stdlib.h> #include <fcntl.h> #include <unistd.h> int main(int argc, char *argv[]) { int (*pfd)[2]; int npipes; int pcap, rcap; int j, p, s, stime, loop; if (argc < 2) { fprintf(stderr, "Usage: %s num-pipes " "[pipe-capacity sleep-time]...\n", argv[0]); exit(EXIT_FAILURE); } npipes = atoi(argv[1]); pfd = calloc(npipes, sizeof (int [2])); if (pfd == NULL) { perror("calloc"); exit(EXIT_FAILURE); } for (j = 0; j < npipes; j++) { if (pipe(pfd[j]) == -1) { fprintf(stderr, "Loop %d: pipe() failed: ", j); perror("pipe"); exit(EXIT_FAILURE); } } printf("Initial pipe capacity: %d\n", fcntl(pfd[0][0], F_GETPIPE_SZ)); for (j = 2; j < argc; j += 2 ) { loop = j / 2; pcap = atoi(argv[j]); printf(" Loop %d: set pipe capacity to %d bytes\n", loop, pcap); for (p = 0; p < npipes; p++) { s = fcntl(pfd[p][0], F_SETPIPE_SZ, pcap); if (s == -1) { fprintf(stderr, " Loop %d, pipe %d: F_SETPIPE_SZ " "failed: ", loop, p); perror("fcntl"); exit(EXIT_FAILURE); } if (p == 0) { printf(" F_SETPIPE_SZ returned %d\n", s); rcap = s; } else { if (s != rcap) { fprintf(stderr, " Loop %d, pipe %d: F_SETPIPE_SZ " "unexpected return: %d\n", loop, p, s); exit(EXIT_FAILURE); } } stime = (j + 1 < argc) ? atoi(argv[j + 1]) : 0; if (stime > 0) { printf(" Sleeping %d seconds\n", stime); sleep(stime); } } } exit(EXIT_SUCCESS); } 8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x--- Patch history: v2 * Switch order of test in 'if' statement to avoid function call (to capability()) in normal path. [This is a fix to a preexisting wart in the code. Thanks to Willy Tarreau] * Perform (size > pipe_max_size) check before calling account_pipe_buffers(). [Thanks to Vegard Nossum] Quoting Vegard: The potential problem happens if the user passes a very large number which will overflow pipe->user->pipe_bufs. On 32-bit, sizeof(int) == sizeof(long), so if they pass arg = INT_MAX then round_pipe_size() returns INT_MAX. Although it's true that the accounting is done in terms of pages and not bytes, so you'd need on the order of (1 << 13) = 8192 processes hitting the limit at the same time in order to make it overflow, which seems a bit unlikely. (See https://lkml.org/lkml/2016/8/12/215 for another discussion on the limit checking) Link: http://lkml.kernel.org/r/1e464945-536b-2420-798b-e77b9c7e8593@gmail.com Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com> Reviewed-by: Vegard Nossum <vegard.nossum@oracle.com> Cc: Willy Tarreau <w@1wt.eu> Cc: <socketpair@gmail.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Jens Axboe <axboe@fb.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-12 04:53:31 +08:00
if (nr_pages > pipe->buffers &&
(too_many_pipe_buffers_hard(user_bufs) ||
too_many_pipe_buffers_soft(user_bufs)) &&
pipe: fix limit checking in pipe_set_size() The limit checking in pipe_set_size() (used by fcntl(F_SETPIPE_SZ)) has the following problems: (1) When increasing the pipe capacity, the checks against the limits in /proc/sys/fs/pipe-user-pages-{soft,hard} are made against existing consumption, and exclude the memory required for the increased pipe capacity. The new increase in pipe capacity can then push the total memory used by the user for pipes (possibly far) over a limit. This can also trigger the problem described next. (2) The limit checks are performed even when the new pipe capacity is less than the existing pipe capacity. This can lead to problems if a user sets a large pipe capacity, and then the limits are lowered, with the result that the user will no longer be able to decrease the pipe capacity. (3) As currently implemented, accounting and checking against the limits is done as follows: (a) Test whether the user has exceeded the limit. (b) Make new pipe buffer allocation. (c) Account new allocation against the limits. This is racey. Multiple processes may pass point (a) simultaneously, and then allocate pipe buffers that are accounted for only in step (c). The race means that the user's pipe buffer allocation could be pushed over the limit (by an arbitrary amount, depending on how unlucky we were in the race). [Thanks to Vegard Nossum for spotting this point, which I had missed.] This patch addresses the above problems as follows: * Perform checks against the limits only when increasing a pipe's capacity; an unprivileged user can always decrease a pipe's capacity. * Alter the checks against limits to include the memory required for the new pipe capacity. * Re-order the accounting step so that it precedes the buffer allocation. If the accounting step determines that a limit has been reached, revert the accounting and cause the operation to fail. The program below can be used to demonstrate problems 1 and 2, and the effect of the fix. The program takes one or more command-line arguments. The first argument specifies the number of pipes that the program should create. The remaining arguments are, alternately, pipe capacities that should be set using fcntl(F_SETPIPE_SZ), and sleep intervals (in seconds) between the fcntl() operations. (The sleep intervals allow the possibility to change the limits between fcntl() operations.) Problem 1 ========= Using the test program on an unpatched kernel, we first set some limits: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # 40.96 MB Then show that we can set a pipe with capacity (100MB) that is over the hard limit # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes F_SETPIPE_SZ returned 134217728 Now set the capacity to 100MB twice. The second call fails (which is probably surprising to most users, since it seems like a no-op): # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 0 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes F_SETPIPE_SZ returned 134217728 Loop 2: set pipe capacity to 100000000 bytes Loop 2, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted With a patched kernel, setting a capacity over the limit fails at the first attempt: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes Loop 1, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted There is a small chance that the change to fix this problem could break user-space, since there are cases where fcntl(F_SETPIPE_SZ) calls that previously succeeded might fail. However, the chances are small, since (a) the pipe-user-pages-{soft,hard} limits are new (in 4.5), and the default soft/hard limits are high/unlimited. Therefore, it seems warranted to make these limits operate more precisely (and behave more like what users probably expect). Problem 2 ========= Running the test program on an unpatched kernel, we first set some limits: # getconf PAGESIZE 4096 # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # 40.96 MB Now perform two fcntl(F_SETPIPE_SZ) operations on a single pipe, first setting a pipe capacity (10MB), sleeping for a few seconds, during which time the hard limit is lowered, and then set pipe capacity to a smaller amount (5MB): # sudo -u mtk ./test_F_SETPIPE_SZ 1 10000000 15 5000000 & [1] 748 # Initial pipe capacity: 65536 Loop 1: set pipe capacity to 10000000 bytes F_SETPIPE_SZ returned 16777216 Sleeping 15 seconds # echo 1000 > /proc/sys/fs/pipe-user-pages-hard # 4.096 MB # Loop 2: set pipe capacity to 5000000 bytes Loop 2, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted In this case, the user should be able to lower the limit. With a kernel that has the patch below, the second fcntl() succeeds: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # sudo -u mtk ./test_F_SETPIPE_SZ 1 10000000 15 5000000 & [1] 3215 # Initial pipe capacity: 65536 # Loop 1: set pipe capacity to 10000000 bytes F_SETPIPE_SZ returned 16777216 Sleeping 15 seconds # echo 1000 > /proc/sys/fs/pipe-user-pages-hard # Loop 2: set pipe capacity to 5000000 bytes F_SETPIPE_SZ returned 8388608 8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x--- /* test_F_SETPIPE_SZ.c (C) 2016, Michael Kerrisk; licensed under GNU GPL version 2 or later Test operation of fcntl(F_SETPIPE_SZ) for setting pipe capacity and interactions with limits defined by /proc/sys/fs/pipe-* files. */ #define _GNU_SOURCE #include <stdio.h> #include <stdlib.h> #include <fcntl.h> #include <unistd.h> int main(int argc, char *argv[]) { int (*pfd)[2]; int npipes; int pcap, rcap; int j, p, s, stime, loop; if (argc < 2) { fprintf(stderr, "Usage: %s num-pipes " "[pipe-capacity sleep-time]...\n", argv[0]); exit(EXIT_FAILURE); } npipes = atoi(argv[1]); pfd = calloc(npipes, sizeof (int [2])); if (pfd == NULL) { perror("calloc"); exit(EXIT_FAILURE); } for (j = 0; j < npipes; j++) { if (pipe(pfd[j]) == -1) { fprintf(stderr, "Loop %d: pipe() failed: ", j); perror("pipe"); exit(EXIT_FAILURE); } } printf("Initial pipe capacity: %d\n", fcntl(pfd[0][0], F_GETPIPE_SZ)); for (j = 2; j < argc; j += 2 ) { loop = j / 2; pcap = atoi(argv[j]); printf(" Loop %d: set pipe capacity to %d bytes\n", loop, pcap); for (p = 0; p < npipes; p++) { s = fcntl(pfd[p][0], F_SETPIPE_SZ, pcap); if (s == -1) { fprintf(stderr, " Loop %d, pipe %d: F_SETPIPE_SZ " "failed: ", loop, p); perror("fcntl"); exit(EXIT_FAILURE); } if (p == 0) { printf(" F_SETPIPE_SZ returned %d\n", s); rcap = s; } else { if (s != rcap) { fprintf(stderr, " Loop %d, pipe %d: F_SETPIPE_SZ " "unexpected return: %d\n", loop, p, s); exit(EXIT_FAILURE); } } stime = (j + 1 < argc) ? atoi(argv[j + 1]) : 0; if (stime > 0) { printf(" Sleeping %d seconds\n", stime); sleep(stime); } } } exit(EXIT_SUCCESS); } 8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x--- Patch history: v2 * Switch order of test in 'if' statement to avoid function call (to capability()) in normal path. [This is a fix to a preexisting wart in the code. Thanks to Willy Tarreau] * Perform (size > pipe_max_size) check before calling account_pipe_buffers(). [Thanks to Vegard Nossum] Quoting Vegard: The potential problem happens if the user passes a very large number which will overflow pipe->user->pipe_bufs. On 32-bit, sizeof(int) == sizeof(long), so if they pass arg = INT_MAX then round_pipe_size() returns INT_MAX. Although it's true that the accounting is done in terms of pages and not bytes, so you'd need on the order of (1 << 13) = 8192 processes hitting the limit at the same time in order to make it overflow, which seems a bit unlikely. (See https://lkml.org/lkml/2016/8/12/215 for another discussion on the limit checking) Link: http://lkml.kernel.org/r/1e464945-536b-2420-798b-e77b9c7e8593@gmail.com Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com> Reviewed-by: Vegard Nossum <vegard.nossum@oracle.com> Cc: Willy Tarreau <w@1wt.eu> Cc: <socketpair@gmail.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Jens Axboe <axboe@fb.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-12 04:53:31 +08:00
!capable(CAP_SYS_RESOURCE) && !capable(CAP_SYS_ADMIN)) {
ret = -EPERM;
goto out_revert_acct;
}
/*
* We can shrink the pipe, if arg >= pipe->nrbufs. Since we don't
* expect a lot of shrink+grow operations, just free and allocate
* again like we would do for growing. If the pipe currently
* contains more buffers than arg, then return busy.
*/
pipe: fix limit checking in pipe_set_size() The limit checking in pipe_set_size() (used by fcntl(F_SETPIPE_SZ)) has the following problems: (1) When increasing the pipe capacity, the checks against the limits in /proc/sys/fs/pipe-user-pages-{soft,hard} are made against existing consumption, and exclude the memory required for the increased pipe capacity. The new increase in pipe capacity can then push the total memory used by the user for pipes (possibly far) over a limit. This can also trigger the problem described next. (2) The limit checks are performed even when the new pipe capacity is less than the existing pipe capacity. This can lead to problems if a user sets a large pipe capacity, and then the limits are lowered, with the result that the user will no longer be able to decrease the pipe capacity. (3) As currently implemented, accounting and checking against the limits is done as follows: (a) Test whether the user has exceeded the limit. (b) Make new pipe buffer allocation. (c) Account new allocation against the limits. This is racey. Multiple processes may pass point (a) simultaneously, and then allocate pipe buffers that are accounted for only in step (c). The race means that the user's pipe buffer allocation could be pushed over the limit (by an arbitrary amount, depending on how unlucky we were in the race). [Thanks to Vegard Nossum for spotting this point, which I had missed.] This patch addresses the above problems as follows: * Perform checks against the limits only when increasing a pipe's capacity; an unprivileged user can always decrease a pipe's capacity. * Alter the checks against limits to include the memory required for the new pipe capacity. * Re-order the accounting step so that it precedes the buffer allocation. If the accounting step determines that a limit has been reached, revert the accounting and cause the operation to fail. The program below can be used to demonstrate problems 1 and 2, and the effect of the fix. The program takes one or more command-line arguments. The first argument specifies the number of pipes that the program should create. The remaining arguments are, alternately, pipe capacities that should be set using fcntl(F_SETPIPE_SZ), and sleep intervals (in seconds) between the fcntl() operations. (The sleep intervals allow the possibility to change the limits between fcntl() operations.) Problem 1 ========= Using the test program on an unpatched kernel, we first set some limits: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # 40.96 MB Then show that we can set a pipe with capacity (100MB) that is over the hard limit # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes F_SETPIPE_SZ returned 134217728 Now set the capacity to 100MB twice. The second call fails (which is probably surprising to most users, since it seems like a no-op): # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 0 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes F_SETPIPE_SZ returned 134217728 Loop 2: set pipe capacity to 100000000 bytes Loop 2, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted With a patched kernel, setting a capacity over the limit fails at the first attempt: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes Loop 1, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted There is a small chance that the change to fix this problem could break user-space, since there are cases where fcntl(F_SETPIPE_SZ) calls that previously succeeded might fail. However, the chances are small, since (a) the pipe-user-pages-{soft,hard} limits are new (in 4.5), and the default soft/hard limits are high/unlimited. Therefore, it seems warranted to make these limits operate more precisely (and behave more like what users probably expect). Problem 2 ========= Running the test program on an unpatched kernel, we first set some limits: # getconf PAGESIZE 4096 # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # 40.96 MB Now perform two fcntl(F_SETPIPE_SZ) operations on a single pipe, first setting a pipe capacity (10MB), sleeping for a few seconds, during which time the hard limit is lowered, and then set pipe capacity to a smaller amount (5MB): # sudo -u mtk ./test_F_SETPIPE_SZ 1 10000000 15 5000000 & [1] 748 # Initial pipe capacity: 65536 Loop 1: set pipe capacity to 10000000 bytes F_SETPIPE_SZ returned 16777216 Sleeping 15 seconds # echo 1000 > /proc/sys/fs/pipe-user-pages-hard # 4.096 MB # Loop 2: set pipe capacity to 5000000 bytes Loop 2, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted In this case, the user should be able to lower the limit. With a kernel that has the patch below, the second fcntl() succeeds: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # sudo -u mtk ./test_F_SETPIPE_SZ 1 10000000 15 5000000 & [1] 3215 # Initial pipe capacity: 65536 # Loop 1: set pipe capacity to 10000000 bytes F_SETPIPE_SZ returned 16777216 Sleeping 15 seconds # echo 1000 > /proc/sys/fs/pipe-user-pages-hard # Loop 2: set pipe capacity to 5000000 bytes F_SETPIPE_SZ returned 8388608 8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x--- /* test_F_SETPIPE_SZ.c (C) 2016, Michael Kerrisk; licensed under GNU GPL version 2 or later Test operation of fcntl(F_SETPIPE_SZ) for setting pipe capacity and interactions with limits defined by /proc/sys/fs/pipe-* files. */ #define _GNU_SOURCE #include <stdio.h> #include <stdlib.h> #include <fcntl.h> #include <unistd.h> int main(int argc, char *argv[]) { int (*pfd)[2]; int npipes; int pcap, rcap; int j, p, s, stime, loop; if (argc < 2) { fprintf(stderr, "Usage: %s num-pipes " "[pipe-capacity sleep-time]...\n", argv[0]); exit(EXIT_FAILURE); } npipes = atoi(argv[1]); pfd = calloc(npipes, sizeof (int [2])); if (pfd == NULL) { perror("calloc"); exit(EXIT_FAILURE); } for (j = 0; j < npipes; j++) { if (pipe(pfd[j]) == -1) { fprintf(stderr, "Loop %d: pipe() failed: ", j); perror("pipe"); exit(EXIT_FAILURE); } } printf("Initial pipe capacity: %d\n", fcntl(pfd[0][0], F_GETPIPE_SZ)); for (j = 2; j < argc; j += 2 ) { loop = j / 2; pcap = atoi(argv[j]); printf(" Loop %d: set pipe capacity to %d bytes\n", loop, pcap); for (p = 0; p < npipes; p++) { s = fcntl(pfd[p][0], F_SETPIPE_SZ, pcap); if (s == -1) { fprintf(stderr, " Loop %d, pipe %d: F_SETPIPE_SZ " "failed: ", loop, p); perror("fcntl"); exit(EXIT_FAILURE); } if (p == 0) { printf(" F_SETPIPE_SZ returned %d\n", s); rcap = s; } else { if (s != rcap) { fprintf(stderr, " Loop %d, pipe %d: F_SETPIPE_SZ " "unexpected return: %d\n", loop, p, s); exit(EXIT_FAILURE); } } stime = (j + 1 < argc) ? atoi(argv[j + 1]) : 0; if (stime > 0) { printf(" Sleeping %d seconds\n", stime); sleep(stime); } } } exit(EXIT_SUCCESS); } 8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x--- Patch history: v2 * Switch order of test in 'if' statement to avoid function call (to capability()) in normal path. [This is a fix to a preexisting wart in the code. Thanks to Willy Tarreau] * Perform (size > pipe_max_size) check before calling account_pipe_buffers(). [Thanks to Vegard Nossum] Quoting Vegard: The potential problem happens if the user passes a very large number which will overflow pipe->user->pipe_bufs. On 32-bit, sizeof(int) == sizeof(long), so if they pass arg = INT_MAX then round_pipe_size() returns INT_MAX. Although it's true that the accounting is done in terms of pages and not bytes, so you'd need on the order of (1 << 13) = 8192 processes hitting the limit at the same time in order to make it overflow, which seems a bit unlikely. (See https://lkml.org/lkml/2016/8/12/215 for another discussion on the limit checking) Link: http://lkml.kernel.org/r/1e464945-536b-2420-798b-e77b9c7e8593@gmail.com Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com> Reviewed-by: Vegard Nossum <vegard.nossum@oracle.com> Cc: Willy Tarreau <w@1wt.eu> Cc: <socketpair@gmail.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Jens Axboe <axboe@fb.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-12 04:53:31 +08:00
if (nr_pages < pipe->nrbufs) {
ret = -EBUSY;
goto out_revert_acct;
}
pipe: account to kmemcg Pipes can consume a significant amount of system memory, hence they should be accounted to kmemcg. This patch marks pipe_inode_info and anonymous pipe buffer page allocations as __GFP_ACCOUNT so that they would be charged to kmemcg. Note, since a pipe buffer page can be "stolen" and get reused for other purposes, including mapping to userspace, we clear PageKmemcg thus resetting page->_mapcount and uncharge it in anon_pipe_buf_steal, which is introduced by this patch. A note regarding anon_pipe_buf_steal implementation. We allow to steal the page if its ref count equals 1. It looks racy, but it is correct for anonymous pipe buffer pages, because: - We lock out all other pipe users, because ->steal is called with pipe_lock held, so the page can't be spliced to another pipe from under us. - The page is not on LRU and it never was. - Thus a parallel thread can access it only by PFN. Although this is quite possible (e.g. see page_idle_get_page and balloon_page_isolate) this is not dangerous, because all such functions do is increase page ref count, check if the page is the one they are looking for, and decrease ref count if it isn't. Since our page is clean except for PageKmemcg mark, which doesn't conflict with other _mapcount users, the worst that can happen is we see page_count > 2 due to a transient ref, in which case we false-positively abort ->steal, which is still fine, because ->steal is not guaranteed to succeed. Link: http://lkml.kernel.org/r/20160527150313.GD26059@esperanza Signed-off-by: Vladimir Davydov <vdavydov@virtuozzo.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Michal Hocko <mhocko@kernel.org> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Minchan Kim <minchan@kernel.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-07-27 06:24:33 +08:00
bufs = kcalloc(nr_pages, sizeof(*bufs),
GFP_KERNEL_ACCOUNT | __GFP_NOWARN);
pipe: fix limit checking in pipe_set_size() The limit checking in pipe_set_size() (used by fcntl(F_SETPIPE_SZ)) has the following problems: (1) When increasing the pipe capacity, the checks against the limits in /proc/sys/fs/pipe-user-pages-{soft,hard} are made against existing consumption, and exclude the memory required for the increased pipe capacity. The new increase in pipe capacity can then push the total memory used by the user for pipes (possibly far) over a limit. This can also trigger the problem described next. (2) The limit checks are performed even when the new pipe capacity is less than the existing pipe capacity. This can lead to problems if a user sets a large pipe capacity, and then the limits are lowered, with the result that the user will no longer be able to decrease the pipe capacity. (3) As currently implemented, accounting and checking against the limits is done as follows: (a) Test whether the user has exceeded the limit. (b) Make new pipe buffer allocation. (c) Account new allocation against the limits. This is racey. Multiple processes may pass point (a) simultaneously, and then allocate pipe buffers that are accounted for only in step (c). The race means that the user's pipe buffer allocation could be pushed over the limit (by an arbitrary amount, depending on how unlucky we were in the race). [Thanks to Vegard Nossum for spotting this point, which I had missed.] This patch addresses the above problems as follows: * Perform checks against the limits only when increasing a pipe's capacity; an unprivileged user can always decrease a pipe's capacity. * Alter the checks against limits to include the memory required for the new pipe capacity. * Re-order the accounting step so that it precedes the buffer allocation. If the accounting step determines that a limit has been reached, revert the accounting and cause the operation to fail. The program below can be used to demonstrate problems 1 and 2, and the effect of the fix. The program takes one or more command-line arguments. The first argument specifies the number of pipes that the program should create. The remaining arguments are, alternately, pipe capacities that should be set using fcntl(F_SETPIPE_SZ), and sleep intervals (in seconds) between the fcntl() operations. (The sleep intervals allow the possibility to change the limits between fcntl() operations.) Problem 1 ========= Using the test program on an unpatched kernel, we first set some limits: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # 40.96 MB Then show that we can set a pipe with capacity (100MB) that is over the hard limit # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes F_SETPIPE_SZ returned 134217728 Now set the capacity to 100MB twice. The second call fails (which is probably surprising to most users, since it seems like a no-op): # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 0 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes F_SETPIPE_SZ returned 134217728 Loop 2: set pipe capacity to 100000000 bytes Loop 2, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted With a patched kernel, setting a capacity over the limit fails at the first attempt: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes Loop 1, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted There is a small chance that the change to fix this problem could break user-space, since there are cases where fcntl(F_SETPIPE_SZ) calls that previously succeeded might fail. However, the chances are small, since (a) the pipe-user-pages-{soft,hard} limits are new (in 4.5), and the default soft/hard limits are high/unlimited. Therefore, it seems warranted to make these limits operate more precisely (and behave more like what users probably expect). Problem 2 ========= Running the test program on an unpatched kernel, we first set some limits: # getconf PAGESIZE 4096 # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # 40.96 MB Now perform two fcntl(F_SETPIPE_SZ) operations on a single pipe, first setting a pipe capacity (10MB), sleeping for a few seconds, during which time the hard limit is lowered, and then set pipe capacity to a smaller amount (5MB): # sudo -u mtk ./test_F_SETPIPE_SZ 1 10000000 15 5000000 & [1] 748 # Initial pipe capacity: 65536 Loop 1: set pipe capacity to 10000000 bytes F_SETPIPE_SZ returned 16777216 Sleeping 15 seconds # echo 1000 > /proc/sys/fs/pipe-user-pages-hard # 4.096 MB # Loop 2: set pipe capacity to 5000000 bytes Loop 2, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted In this case, the user should be able to lower the limit. With a kernel that has the patch below, the second fcntl() succeeds: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # sudo -u mtk ./test_F_SETPIPE_SZ 1 10000000 15 5000000 & [1] 3215 # Initial pipe capacity: 65536 # Loop 1: set pipe capacity to 10000000 bytes F_SETPIPE_SZ returned 16777216 Sleeping 15 seconds # echo 1000 > /proc/sys/fs/pipe-user-pages-hard # Loop 2: set pipe capacity to 5000000 bytes F_SETPIPE_SZ returned 8388608 8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x--- /* test_F_SETPIPE_SZ.c (C) 2016, Michael Kerrisk; licensed under GNU GPL version 2 or later Test operation of fcntl(F_SETPIPE_SZ) for setting pipe capacity and interactions with limits defined by /proc/sys/fs/pipe-* files. */ #define _GNU_SOURCE #include <stdio.h> #include <stdlib.h> #include <fcntl.h> #include <unistd.h> int main(int argc, char *argv[]) { int (*pfd)[2]; int npipes; int pcap, rcap; int j, p, s, stime, loop; if (argc < 2) { fprintf(stderr, "Usage: %s num-pipes " "[pipe-capacity sleep-time]...\n", argv[0]); exit(EXIT_FAILURE); } npipes = atoi(argv[1]); pfd = calloc(npipes, sizeof (int [2])); if (pfd == NULL) { perror("calloc"); exit(EXIT_FAILURE); } for (j = 0; j < npipes; j++) { if (pipe(pfd[j]) == -1) { fprintf(stderr, "Loop %d: pipe() failed: ", j); perror("pipe"); exit(EXIT_FAILURE); } } printf("Initial pipe capacity: %d\n", fcntl(pfd[0][0], F_GETPIPE_SZ)); for (j = 2; j < argc; j += 2 ) { loop = j / 2; pcap = atoi(argv[j]); printf(" Loop %d: set pipe capacity to %d bytes\n", loop, pcap); for (p = 0; p < npipes; p++) { s = fcntl(pfd[p][0], F_SETPIPE_SZ, pcap); if (s == -1) { fprintf(stderr, " Loop %d, pipe %d: F_SETPIPE_SZ " "failed: ", loop, p); perror("fcntl"); exit(EXIT_FAILURE); } if (p == 0) { printf(" F_SETPIPE_SZ returned %d\n", s); rcap = s; } else { if (s != rcap) { fprintf(stderr, " Loop %d, pipe %d: F_SETPIPE_SZ " "unexpected return: %d\n", loop, p, s); exit(EXIT_FAILURE); } } stime = (j + 1 < argc) ? atoi(argv[j + 1]) : 0; if (stime > 0) { printf(" Sleeping %d seconds\n", stime); sleep(stime); } } } exit(EXIT_SUCCESS); } 8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x--- Patch history: v2 * Switch order of test in 'if' statement to avoid function call (to capability()) in normal path. [This is a fix to a preexisting wart in the code. Thanks to Willy Tarreau] * Perform (size > pipe_max_size) check before calling account_pipe_buffers(). [Thanks to Vegard Nossum] Quoting Vegard: The potential problem happens if the user passes a very large number which will overflow pipe->user->pipe_bufs. On 32-bit, sizeof(int) == sizeof(long), so if they pass arg = INT_MAX then round_pipe_size() returns INT_MAX. Although it's true that the accounting is done in terms of pages and not bytes, so you'd need on the order of (1 << 13) = 8192 processes hitting the limit at the same time in order to make it overflow, which seems a bit unlikely. (See https://lkml.org/lkml/2016/8/12/215 for another discussion on the limit checking) Link: http://lkml.kernel.org/r/1e464945-536b-2420-798b-e77b9c7e8593@gmail.com Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com> Reviewed-by: Vegard Nossum <vegard.nossum@oracle.com> Cc: Willy Tarreau <w@1wt.eu> Cc: <socketpair@gmail.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Jens Axboe <axboe@fb.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-12 04:53:31 +08:00
if (unlikely(!bufs)) {
ret = -ENOMEM;
goto out_revert_acct;
}
/*
* The pipe array wraps around, so just start the new one at zero
* and adjust the indexes.
*/
if (pipe->nrbufs) {
unsigned int tail;
unsigned int head;
tail = pipe->curbuf + pipe->nrbufs;
if (tail < pipe->buffers)
tail = 0;
else
tail &= (pipe->buffers - 1);
head = pipe->nrbufs - tail;
if (head)
memcpy(bufs, pipe->bufs + pipe->curbuf, head * sizeof(struct pipe_buffer));
if (tail)
memcpy(bufs + head, pipe->bufs, tail * sizeof(struct pipe_buffer));
}
pipe->curbuf = 0;
kfree(pipe->bufs);
pipe->bufs = bufs;
pipe->buffers = nr_pages;
return nr_pages * PAGE_SIZE;
pipe: fix limit checking in pipe_set_size() The limit checking in pipe_set_size() (used by fcntl(F_SETPIPE_SZ)) has the following problems: (1) When increasing the pipe capacity, the checks against the limits in /proc/sys/fs/pipe-user-pages-{soft,hard} are made against existing consumption, and exclude the memory required for the increased pipe capacity. The new increase in pipe capacity can then push the total memory used by the user for pipes (possibly far) over a limit. This can also trigger the problem described next. (2) The limit checks are performed even when the new pipe capacity is less than the existing pipe capacity. This can lead to problems if a user sets a large pipe capacity, and then the limits are lowered, with the result that the user will no longer be able to decrease the pipe capacity. (3) As currently implemented, accounting and checking against the limits is done as follows: (a) Test whether the user has exceeded the limit. (b) Make new pipe buffer allocation. (c) Account new allocation against the limits. This is racey. Multiple processes may pass point (a) simultaneously, and then allocate pipe buffers that are accounted for only in step (c). The race means that the user's pipe buffer allocation could be pushed over the limit (by an arbitrary amount, depending on how unlucky we were in the race). [Thanks to Vegard Nossum for spotting this point, which I had missed.] This patch addresses the above problems as follows: * Perform checks against the limits only when increasing a pipe's capacity; an unprivileged user can always decrease a pipe's capacity. * Alter the checks against limits to include the memory required for the new pipe capacity. * Re-order the accounting step so that it precedes the buffer allocation. If the accounting step determines that a limit has been reached, revert the accounting and cause the operation to fail. The program below can be used to demonstrate problems 1 and 2, and the effect of the fix. The program takes one or more command-line arguments. The first argument specifies the number of pipes that the program should create. The remaining arguments are, alternately, pipe capacities that should be set using fcntl(F_SETPIPE_SZ), and sleep intervals (in seconds) between the fcntl() operations. (The sleep intervals allow the possibility to change the limits between fcntl() operations.) Problem 1 ========= Using the test program on an unpatched kernel, we first set some limits: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # 40.96 MB Then show that we can set a pipe with capacity (100MB) that is over the hard limit # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes F_SETPIPE_SZ returned 134217728 Now set the capacity to 100MB twice. The second call fails (which is probably surprising to most users, since it seems like a no-op): # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 0 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes F_SETPIPE_SZ returned 134217728 Loop 2: set pipe capacity to 100000000 bytes Loop 2, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted With a patched kernel, setting a capacity over the limit fails at the first attempt: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes Loop 1, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted There is a small chance that the change to fix this problem could break user-space, since there are cases where fcntl(F_SETPIPE_SZ) calls that previously succeeded might fail. However, the chances are small, since (a) the pipe-user-pages-{soft,hard} limits are new (in 4.5), and the default soft/hard limits are high/unlimited. Therefore, it seems warranted to make these limits operate more precisely (and behave more like what users probably expect). Problem 2 ========= Running the test program on an unpatched kernel, we first set some limits: # getconf PAGESIZE 4096 # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # 40.96 MB Now perform two fcntl(F_SETPIPE_SZ) operations on a single pipe, first setting a pipe capacity (10MB), sleeping for a few seconds, during which time the hard limit is lowered, and then set pipe capacity to a smaller amount (5MB): # sudo -u mtk ./test_F_SETPIPE_SZ 1 10000000 15 5000000 & [1] 748 # Initial pipe capacity: 65536 Loop 1: set pipe capacity to 10000000 bytes F_SETPIPE_SZ returned 16777216 Sleeping 15 seconds # echo 1000 > /proc/sys/fs/pipe-user-pages-hard # 4.096 MB # Loop 2: set pipe capacity to 5000000 bytes Loop 2, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted In this case, the user should be able to lower the limit. With a kernel that has the patch below, the second fcntl() succeeds: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # sudo -u mtk ./test_F_SETPIPE_SZ 1 10000000 15 5000000 & [1] 3215 # Initial pipe capacity: 65536 # Loop 1: set pipe capacity to 10000000 bytes F_SETPIPE_SZ returned 16777216 Sleeping 15 seconds # echo 1000 > /proc/sys/fs/pipe-user-pages-hard # Loop 2: set pipe capacity to 5000000 bytes F_SETPIPE_SZ returned 8388608 8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x--- /* test_F_SETPIPE_SZ.c (C) 2016, Michael Kerrisk; licensed under GNU GPL version 2 or later Test operation of fcntl(F_SETPIPE_SZ) for setting pipe capacity and interactions with limits defined by /proc/sys/fs/pipe-* files. */ #define _GNU_SOURCE #include <stdio.h> #include <stdlib.h> #include <fcntl.h> #include <unistd.h> int main(int argc, char *argv[]) { int (*pfd)[2]; int npipes; int pcap, rcap; int j, p, s, stime, loop; if (argc < 2) { fprintf(stderr, "Usage: %s num-pipes " "[pipe-capacity sleep-time]...\n", argv[0]); exit(EXIT_FAILURE); } npipes = atoi(argv[1]); pfd = calloc(npipes, sizeof (int [2])); if (pfd == NULL) { perror("calloc"); exit(EXIT_FAILURE); } for (j = 0; j < npipes; j++) { if (pipe(pfd[j]) == -1) { fprintf(stderr, "Loop %d: pipe() failed: ", j); perror("pipe"); exit(EXIT_FAILURE); } } printf("Initial pipe capacity: %d\n", fcntl(pfd[0][0], F_GETPIPE_SZ)); for (j = 2; j < argc; j += 2 ) { loop = j / 2; pcap = atoi(argv[j]); printf(" Loop %d: set pipe capacity to %d bytes\n", loop, pcap); for (p = 0; p < npipes; p++) { s = fcntl(pfd[p][0], F_SETPIPE_SZ, pcap); if (s == -1) { fprintf(stderr, " Loop %d, pipe %d: F_SETPIPE_SZ " "failed: ", loop, p); perror("fcntl"); exit(EXIT_FAILURE); } if (p == 0) { printf(" F_SETPIPE_SZ returned %d\n", s); rcap = s; } else { if (s != rcap) { fprintf(stderr, " Loop %d, pipe %d: F_SETPIPE_SZ " "unexpected return: %d\n", loop, p, s); exit(EXIT_FAILURE); } } stime = (j + 1 < argc) ? atoi(argv[j + 1]) : 0; if (stime > 0) { printf(" Sleeping %d seconds\n", stime); sleep(stime); } } } exit(EXIT_SUCCESS); } 8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x--- Patch history: v2 * Switch order of test in 'if' statement to avoid function call (to capability()) in normal path. [This is a fix to a preexisting wart in the code. Thanks to Willy Tarreau] * Perform (size > pipe_max_size) check before calling account_pipe_buffers(). [Thanks to Vegard Nossum] Quoting Vegard: The potential problem happens if the user passes a very large number which will overflow pipe->user->pipe_bufs. On 32-bit, sizeof(int) == sizeof(long), so if they pass arg = INT_MAX then round_pipe_size() returns INT_MAX. Although it's true that the accounting is done in terms of pages and not bytes, so you'd need on the order of (1 << 13) = 8192 processes hitting the limit at the same time in order to make it overflow, which seems a bit unlikely. (See https://lkml.org/lkml/2016/8/12/215 for another discussion on the limit checking) Link: http://lkml.kernel.org/r/1e464945-536b-2420-798b-e77b9c7e8593@gmail.com Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com> Reviewed-by: Vegard Nossum <vegard.nossum@oracle.com> Cc: Willy Tarreau <w@1wt.eu> Cc: <socketpair@gmail.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Jens Axboe <axboe@fb.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-12 04:53:31 +08:00
out_revert_acct:
(void) account_pipe_buffers(pipe->user, nr_pages, pipe->buffers);
pipe: fix limit checking in pipe_set_size() The limit checking in pipe_set_size() (used by fcntl(F_SETPIPE_SZ)) has the following problems: (1) When increasing the pipe capacity, the checks against the limits in /proc/sys/fs/pipe-user-pages-{soft,hard} are made against existing consumption, and exclude the memory required for the increased pipe capacity. The new increase in pipe capacity can then push the total memory used by the user for pipes (possibly far) over a limit. This can also trigger the problem described next. (2) The limit checks are performed even when the new pipe capacity is less than the existing pipe capacity. This can lead to problems if a user sets a large pipe capacity, and then the limits are lowered, with the result that the user will no longer be able to decrease the pipe capacity. (3) As currently implemented, accounting and checking against the limits is done as follows: (a) Test whether the user has exceeded the limit. (b) Make new pipe buffer allocation. (c) Account new allocation against the limits. This is racey. Multiple processes may pass point (a) simultaneously, and then allocate pipe buffers that are accounted for only in step (c). The race means that the user's pipe buffer allocation could be pushed over the limit (by an arbitrary amount, depending on how unlucky we were in the race). [Thanks to Vegard Nossum for spotting this point, which I had missed.] This patch addresses the above problems as follows: * Perform checks against the limits only when increasing a pipe's capacity; an unprivileged user can always decrease a pipe's capacity. * Alter the checks against limits to include the memory required for the new pipe capacity. * Re-order the accounting step so that it precedes the buffer allocation. If the accounting step determines that a limit has been reached, revert the accounting and cause the operation to fail. The program below can be used to demonstrate problems 1 and 2, and the effect of the fix. The program takes one or more command-line arguments. The first argument specifies the number of pipes that the program should create. The remaining arguments are, alternately, pipe capacities that should be set using fcntl(F_SETPIPE_SZ), and sleep intervals (in seconds) between the fcntl() operations. (The sleep intervals allow the possibility to change the limits between fcntl() operations.) Problem 1 ========= Using the test program on an unpatched kernel, we first set some limits: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # 40.96 MB Then show that we can set a pipe with capacity (100MB) that is over the hard limit # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes F_SETPIPE_SZ returned 134217728 Now set the capacity to 100MB twice. The second call fails (which is probably surprising to most users, since it seems like a no-op): # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 0 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes F_SETPIPE_SZ returned 134217728 Loop 2: set pipe capacity to 100000000 bytes Loop 2, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted With a patched kernel, setting a capacity over the limit fails at the first attempt: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # sudo -u mtk ./test_F_SETPIPE_SZ 1 100000000 Initial pipe capacity: 65536 Loop 1: set pipe capacity to 100000000 bytes Loop 1, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted There is a small chance that the change to fix this problem could break user-space, since there are cases where fcntl(F_SETPIPE_SZ) calls that previously succeeded might fail. However, the chances are small, since (a) the pipe-user-pages-{soft,hard} limits are new (in 4.5), and the default soft/hard limits are high/unlimited. Therefore, it seems warranted to make these limits operate more precisely (and behave more like what users probably expect). Problem 2 ========= Running the test program on an unpatched kernel, we first set some limits: # getconf PAGESIZE 4096 # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # 40.96 MB Now perform two fcntl(F_SETPIPE_SZ) operations on a single pipe, first setting a pipe capacity (10MB), sleeping for a few seconds, during which time the hard limit is lowered, and then set pipe capacity to a smaller amount (5MB): # sudo -u mtk ./test_F_SETPIPE_SZ 1 10000000 15 5000000 & [1] 748 # Initial pipe capacity: 65536 Loop 1: set pipe capacity to 10000000 bytes F_SETPIPE_SZ returned 16777216 Sleeping 15 seconds # echo 1000 > /proc/sys/fs/pipe-user-pages-hard # 4.096 MB # Loop 2: set pipe capacity to 5000000 bytes Loop 2, pipe 0: F_SETPIPE_SZ failed: fcntl: Operation not permitted In this case, the user should be able to lower the limit. With a kernel that has the patch below, the second fcntl() succeeds: # echo 0 > /proc/sys/fs/pipe-user-pages-soft # echo 1000000000 > /proc/sys/fs/pipe-max-size # echo 10000 > /proc/sys/fs/pipe-user-pages-hard # sudo -u mtk ./test_F_SETPIPE_SZ 1 10000000 15 5000000 & [1] 3215 # Initial pipe capacity: 65536 # Loop 1: set pipe capacity to 10000000 bytes F_SETPIPE_SZ returned 16777216 Sleeping 15 seconds # echo 1000 > /proc/sys/fs/pipe-user-pages-hard # Loop 2: set pipe capacity to 5000000 bytes F_SETPIPE_SZ returned 8388608 8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x--- /* test_F_SETPIPE_SZ.c (C) 2016, Michael Kerrisk; licensed under GNU GPL version 2 or later Test operation of fcntl(F_SETPIPE_SZ) for setting pipe capacity and interactions with limits defined by /proc/sys/fs/pipe-* files. */ #define _GNU_SOURCE #include <stdio.h> #include <stdlib.h> #include <fcntl.h> #include <unistd.h> int main(int argc, char *argv[]) { int (*pfd)[2]; int npipes; int pcap, rcap; int j, p, s, stime, loop; if (argc < 2) { fprintf(stderr, "Usage: %s num-pipes " "[pipe-capacity sleep-time]...\n", argv[0]); exit(EXIT_FAILURE); } npipes = atoi(argv[1]); pfd = calloc(npipes, sizeof (int [2])); if (pfd == NULL) { perror("calloc"); exit(EXIT_FAILURE); } for (j = 0; j < npipes; j++) { if (pipe(pfd[j]) == -1) { fprintf(stderr, "Loop %d: pipe() failed: ", j); perror("pipe"); exit(EXIT_FAILURE); } } printf("Initial pipe capacity: %d\n", fcntl(pfd[0][0], F_GETPIPE_SZ)); for (j = 2; j < argc; j += 2 ) { loop = j / 2; pcap = atoi(argv[j]); printf(" Loop %d: set pipe capacity to %d bytes\n", loop, pcap); for (p = 0; p < npipes; p++) { s = fcntl(pfd[p][0], F_SETPIPE_SZ, pcap); if (s == -1) { fprintf(stderr, " Loop %d, pipe %d: F_SETPIPE_SZ " "failed: ", loop, p); perror("fcntl"); exit(EXIT_FAILURE); } if (p == 0) { printf(" F_SETPIPE_SZ returned %d\n", s); rcap = s; } else { if (s != rcap) { fprintf(stderr, " Loop %d, pipe %d: F_SETPIPE_SZ " "unexpected return: %d\n", loop, p, s); exit(EXIT_FAILURE); } } stime = (j + 1 < argc) ? atoi(argv[j + 1]) : 0; if (stime > 0) { printf(" Sleeping %d seconds\n", stime); sleep(stime); } } } exit(EXIT_SUCCESS); } 8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x---8x--- Patch history: v2 * Switch order of test in 'if' statement to avoid function call (to capability()) in normal path. [This is a fix to a preexisting wart in the code. Thanks to Willy Tarreau] * Perform (size > pipe_max_size) check before calling account_pipe_buffers(). [Thanks to Vegard Nossum] Quoting Vegard: The potential problem happens if the user passes a very large number which will overflow pipe->user->pipe_bufs. On 32-bit, sizeof(int) == sizeof(long), so if they pass arg = INT_MAX then round_pipe_size() returns INT_MAX. Although it's true that the accounting is done in terms of pages and not bytes, so you'd need on the order of (1 << 13) = 8192 processes hitting the limit at the same time in order to make it overflow, which seems a bit unlikely. (See https://lkml.org/lkml/2016/8/12/215 for another discussion on the limit checking) Link: http://lkml.kernel.org/r/1e464945-536b-2420-798b-e77b9c7e8593@gmail.com Signed-off-by: Michael Kerrisk <mtk.manpages@gmail.com> Reviewed-by: Vegard Nossum <vegard.nossum@oracle.com> Cc: Willy Tarreau <w@1wt.eu> Cc: <socketpair@gmail.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Jens Axboe <axboe@fb.com> Cc: Al Viro <viro@zeniv.linux.org.uk> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-10-12 04:53:31 +08:00
return ret;
}
/*
* This should work even if CONFIG_PROC_FS isn't set, as proc_dointvec_minmax
* will return an error.
*/
int pipe_proc_fn(struct ctl_table *table, int write, void __user *buf,
size_t *lenp, loff_t *ppos)
{
int ret;
ret = proc_dointvec_minmax(table, write, buf, lenp, ppos);
if (ret < 0 || !write)
return ret;
pipe_max_size = round_pipe_size(pipe_max_size);
return ret;
}
/*
* After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
* location, so checking ->i_pipe is not enough to verify that this is a
* pipe.
*/
struct pipe_inode_info *get_pipe_info(struct file *file)
{
return file->f_op == &pipefifo_fops ? file->private_data : NULL;
}
long pipe_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
{
struct pipe_inode_info *pipe;
long ret;
pipe = get_pipe_info(file);
if (!pipe)
return -EBADF;
__pipe_lock(pipe);
switch (cmd) {
case F_SETPIPE_SZ:
ret = pipe_set_size(pipe, arg);
break;
case F_GETPIPE_SZ:
ret = pipe->buffers * PAGE_SIZE;
break;
default:
ret = -EINVAL;
break;
}
__pipe_unlock(pipe);
return ret;
}
static const struct super_operations pipefs_ops = {
.destroy_inode = free_inode_nonrcu,
.statfs = simple_statfs,
};
/*
* pipefs should _never_ be mounted by userland - too much of security hassle,
* no real gain from having the whole whorehouse mounted. So we don't need
* any operations on the root directory. However, we need a non-trivial
* d_name - pipe: will go nicely and kill the special-casing in procfs.
*/
static struct dentry *pipefs_mount(struct file_system_type *fs_type,
int flags, const char *dev_name, void *data)
{
return mount_pseudo(fs_type, "pipe:", &pipefs_ops,
&pipefs_dentry_operations, PIPEFS_MAGIC);
}
static struct file_system_type pipe_fs_type = {
.name = "pipefs",
.mount = pipefs_mount,
.kill_sb = kill_anon_super,
};
static int __init init_pipe_fs(void)
{
int err = register_filesystem(&pipe_fs_type);
if (!err) {
pipe_mnt = kern_mount(&pipe_fs_type);
if (IS_ERR(pipe_mnt)) {
err = PTR_ERR(pipe_mnt);
unregister_filesystem(&pipe_fs_type);
}
}
return err;
}
fs_initcall(init_pipe_fs);