lkml.org 
[lkml]   [2011]   [Jun]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    SubjectRe: Change in functionality of futex() system call.
    From
    Date
    On Mon, 2011-06-06 at 18:11 +0200, Peter Zijlstra wrote:
    > On Mon, 2011-06-06 at 17:23 +0200, Eric Dumazet wrote:
    > > Le lundi 06 juin 2011 à 09:28 -0500, David Oliver a écrit :
    > > > Hello,
    > > >
    > > > The functionality of the futex() system call appears to have changed
    > > > between versions 2.6.18 and 2.6.32.28.
    > > >
    > > > Specifically, performing a FUTEX_WAIT on a read-only mapped location
    > > > results in an EFAULT. Although other operations, such as FUTEX_WAKE,
    > > > are only meaningful for writable locations, FUTEX_WAIT is useful for
    > > > processes with read-only access to a memory-mapped file.
    > > >
    > > > The code below illustrates the changed behavior (each of the EXPECT
    > > > operations succeed on the older kernel, the ASSERTs pass in each
    > > > case), assuming the file /tmp/futex_test exists and contains int(42).
    > > >
    > > > With the older kernel, the syscall() suspends until another process
    > > > changes the file and issues a FUTEX_WAKE, whereas the new behavior is
    > > > for an EFAULT error, independent of the file contents.
    > > >
    > > > Let me know if you need further clarification.
    > > >
    > > > Cheers!
    > > >
    > > > David Oliver.
    > > >
    > > >
    > > > #include <errno.h>
    > > > #include <fcntl.h>
    > > > #include <stdint.h>
    > > > typedef uint32_t u32; // for futex.h
    > > > #include <linux/futex.h>
    > > > #include <sys/mman.h>
    > > > #include <sys/syscall.h>
    > > > #include <unistd.h>
    > > > #include "gtest/gtest.h" // test framework to illustrate issue.
    > > >
    > > >
    > > > TEST(Futex, futex_in_read_only_file_is_ok) {
    > > > int fd = open("/tmp/futex_test", O_RDONLY);
    > > > ASSERT_GE(fd, 0);
    > > > int* futex = static_cast<int *>(mmap(0, sizeof(int), PROT_READ,
    > > > MAP_SHARED, fd, 0));
    > > > ASSERT_NE((int *)(0), futex);
    > > >
    > > > int rc = syscall(SYS_futex, futex, FUTEX_WAIT, 42, 0, 0, 0);
    > > >
    > > > EXPECT_NE(-1, rc); // fails.
    > > > if (rc == -1) {
    > > > EXPECT_NE(errno, EFAULT); // fails.
    > > > }
    > > > }
    > > >
    > >
    > > Right you are, this came from commit 7485d0d3758e8e6491a5 (futexes:
    > > Remove rw parameter from get_futex_key()) in 2.6.33
    > >
    > > commit 7485d0d3758e8e6491a5c9468114e74dc050785d
    > > Author: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
    > > Date: Tue Jan 5 16:32:43 2010 +0900
    > >
    > > futexes: Remove rw parameter from get_futex_key()
    > >
    > > Currently, futexes have two problem:
    > >
    > > A) The current futex code doesn't handle private file mappings properly.
    > >
    > > get_futex_key() uses PageAnon() to distinguish file and
    > > anon, which can cause the following bad scenario:
    > >
    > > 1) thread-A call futex(private-mapping, FUTEX_WAIT), it
    > > sleeps on file mapping object.
    > > 2) thread-B writes a variable and it makes it cow.
    > > 3) thread-B calls futex(private-mapping, FUTEX_WAKE), it
    > > wakes up blocked thread on the anonymous page. (but it's nothing)
    > >
    > > B) Current futex code doesn't handle zero page properly.
    > >
    > > Read mode get_user_pages() can return zero page, but current
    > > futex code doesn't handle it at all. Then, zero page makes
    > > infinite loop internally.
    > >
    > > The solution is to use write mode get_user_page() always for
    > > page lookup. It prevents the lookup of both file page of private
    > > mappings and zero page.
    > >
    > > Performance concerns:
    > >
    > > Probaly very little, because glibc always initialize variables
    > > for futex before to call futex(). It means glibc users never see
    > > the overhead of this patch.
    > >
    > > Compatibility concerns:
    > >
    > > This patch has few compatibility issues. After this patch,
    > > FUTEX_WAIT require writable access to futex variables (read-only
    > > mappings makes EFAULT). But practically it's not a problem,
    > > glibc always initalizes variables for futexes explicitly - nobody
    > > uses read-only mappings.
    >
    > Urgh,. maybe something like the below but with more conditionals that
    > enable the extra logic only for FUTEX_WAIT..
    >
    > The idea is to try a RO gup() when the RW gup() fails so as not to slow
    > down the common path of writable anonymous maps and bail when we used
    > the RO path on anonymous memory.
    >
    > ---
    > diff --git a/kernel/futex.c b/kernel/futex.c
    > index fe28dc2..11f2ad1 100644
    > --- a/kernel/futex.c
    > +++ b/kernel/futex.c
    > @@ -234,7 +234,7 @@ get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key)
    > unsigned long address = (unsigned long)uaddr;
    > struct mm_struct *mm = current->mm;
    > struct page *page, *page_head;
    > - int err;
    > + int err, ro = 0;
    >
    > /*
    > * The futex address must be "naturally" aligned.
    > @@ -262,6 +262,10 @@ get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key)
    >
    > again:
    > err = get_user_pages_fast(address, 1, 1, &page);
    > + if (err == -EFAULT) {
    > + err = get_user_pages_fast(address, 1, 0, &page);
    > + ro = 1;
    > + }
    > if (err < 0)
    > return err;
    >
    > @@ -316,6 +320,11 @@ again:
    > * the object not the particular process.
    > */
    > if (PageAnon(page_head)) {
    > + if (ro) {
    > + err = -EFAULT;
    > + goto out;
    > + }
    > +
    > key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */
    > key->private.mm = mm;
    > key->private.address = address;
    > @@ -327,9 +336,10 @@ again:
    >
    > get_futex_key_refs(key);
    >
    > +out:
    > unlock_page(page_head);
    > put_page(page_head);
    > - return 0;
    > + return err;
    > }
    >
    > static inline void put_futex_key(union futex_key *key)
    >

    Hmm, wouldn't that still be susceptible to the zero-page thing if: we
    create a writable private file map of a sparse file, touch a page and
    then remap the thing RO?



    --
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.kernel.org
    More majordomo info at http://vger.kernel.org/majordomo-info.html
    Please read the FAQ at http://www.tux.org/lkml/

    \
     
     \ /
      Last update: 2011-06-06 18:19    [W:0.038 / U:88.200 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site