[PATCH] lightweight robust futexes: core

Add the core infrastructure for robust futexes: structure definitions, the new
syscalls and the do_exit() based cleanup mechanism.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Arjan van de Ven <arjan@infradead.org>
Acked-by: Ulrich Drepper <drepper@redhat.com>
Cc: Michael Kerrisk <mtk-manpages@gmx.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

3 files changed, 100 insertions, 1 deletions

diff --git a/include/linux/futex.h b/include/linux/futex.h
index 10f96c31971e..20face6b798d 100644
--- a/include/linux/futex.h
+++ b/include/linux/futex.h
@@ -1,6 +1,8 @@
 #ifndef _LINUX_FUTEX_H
 #define _LINUX_FUTEX_H
 
+#include <linux/sched.h>
+
 /* Second argument to futex syscall */
 
 
@@ -11,10 +13,103 @@
 #define FUTEX_CMP_REQUEUE       4
 #define FUTEX_WAKE_OP           5
 
+/*
+ * Support for robust futexes: the kernel cleans up held futexes at
+ * thread exit time.
+ */
+
+/*
+ * Per-lock list entry - embedded in user-space locks, somewhere close
+ * to the futex field. (Note: user-space uses a double-linked list to
+ * achieve O(1) list add and remove, but the kernel only needs to know
+ * about the forward link)
+ *
+ * NOTE: this structure is part of the syscall ABI, and must not be
+ * changed.
+ */
+struct robust_list {
+        struct robust_list __user *next;
+};
+
+/*
+ * Per-thread list head:
+ *
+ * NOTE: this structure is part of the syscall ABI, and must only be
+ * changed if the change is first communicated with the glibc folks.
+ * (When an incompatible change is done, we'll increase the structure
+ *  size, which glibc will detect)
+ */
+struct robust_list_head {
+        /*
+         * The head of the list. Points back to itself if empty:
+         */
+        struct robust_list list;
+
+        /*
+         * This relative offset is set by user-space, it gives the kernel
+         * the relative position of the futex field to examine. This way
+         * we keep userspace flexible, to freely shape its data-structure,
+         * without hardcoding any particular offset into the kernel:
+         */
+        long futex_offset;
+
+        /*
+         * The death of the thread may race with userspace setting
+         * up a lock's links. So to handle this race, userspace first
+         * sets this field to the address of the to-be-taken lock,
+         * then does the lock acquire, and then adds itself to the
+         * list, and then clears this field. Hence the kernel will
+         * always have full knowledge of all locks that the thread
+         * _might_ have taken. We check the owner TID in any case,
+         * so only truly owned locks will be handled.
+         */
+        struct robust_list __user *list_op_pending;
+};
+
+/*
+ * Are there any waiters for this robust futex:
+ */
+#define FUTEX_WAITERS           0x80000000
+
+/*
+ * The kernel signals via this bit that a thread holding a futex
+ * has exited without unlocking the futex. The kernel also does
+ * a FUTEX_WAKE on such futexes, after setting the bit, to wake
+ * up any possible waiters:
+ */
+#define FUTEX_OWNER_DIED        0x40000000
+
+/*
+ * Reserved bit:
+ */
+#define FUTEX_OWNER_PENDING     0x20000000
+
+/*
+ * The rest of the robust-futex field is for the TID:
+ */
+#define FUTEX_TID_MASK          0x1fffffff
+
+/*
+ * A limit of one million locks held per thread (!) ought to be enough
+ * for some time. This also protects against a deliberately circular
+ * list. Not worth introducing an rlimit for this:
+ */
+#define ROBUST_LIST_LIMIT       1048576
+
 long do_futex(unsigned long uaddr, int op, int val,
                 unsigned long timeout, unsigned long uaddr2, int val2,
                 int val3);
 
+extern int handle_futex_death(unsigned int *uaddr, struct task_struct *curr);
+
+#ifdef CONFIG_FUTEX
+extern void exit_robust_list(struct task_struct *curr);
+#else
+static inline void exit_robust_list(struct task_struct *curr)
+{
+}
+#endif
+
 #define FUTEX_OP_SET            0       /* *(int *)UADDR2 = OPARG; */
 #define FUTEX_OP_ADD            1       /* *(int *)UADDR2 += OPARG; */
 #define FUTEX_OP_OR             2       /* *(int *)UADDR2 |= OPARG; */
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 036d14d2bf90..fd4848f2d750 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -35,6 +35,7 @@
 #include <linux/topology.h>
 #include <linux/seccomp.h>
 #include <linux/rcupdate.h>
+#include <linux/futex.h>
 
 #include <linux/auxvec.h>       /* For AT_VECTOR_SIZE */
 
@@ -872,6 +873,8 @@ struct task_struct {
         int cpuset_mems_generation;
         int cpuset_mem_spread_rotor;
 #endif
+        struct robust_list_head __user *robust_list;
+
         atomic_t fs_excl;       /* holding fs exclusive resources */
         struct rcu_head rcu;
 };
diff --git a/include/linux/threads.h b/include/linux/threads.h
index b59738ac6197..e646bcdf2614 100644
--- a/include/linux/threads.h
+++ b/include/linux/threads.h
@@ -28,7 +28,8 @@
 #define PID_MAX_DEFAULT (CONFIG_BASE_SMALL ? 0x1000 : 0x8000)
 
 /*
- * A maximum of 4 million PIDs should be enough for a while:
+ * A maximum of 4 million PIDs should be enough for a while.
+ * [NOTE: PID/TIDs are limited to 2^29 ~= 500+ million, see futex.h.]
  */
 #define PID_MAX_LIMIT (CONFIG_BASE_SMALL ? PAGE_SIZE * 8 : \
         (sizeof(long) > 4 ? 4 * 1024 * 1024 : PID_MAX_DEFAULT))