Merge to Fedora kernel-2.6.17-1.2142_FC4 patched with stable patch-2.6.17.13-vs2...
[linux-2.6.git] / fs / locks.c
index 0204acb..7ec121c 100644 (file)
 #include <linux/security.h>
 #include <linux/slab.h>
 #include <linux/smp_lock.h>
+#include <linux/syscalls.h>
 #include <linux/time.h>
+#include <linux/rcupdate.h>
+#include <linux/vs_limit.h>
 
 #include <asm/semaphore.h>
 #include <asm/uaccess.h>
@@ -137,36 +140,43 @@ int lease_break_time = 45;
 #define for_each_lock(inode, lockp) \
        for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
 
-LIST_HEAD(file_lock_list);
-
-EXPORT_SYMBOL(file_lock_list);
-
+static LIST_HEAD(file_lock_list);
 static LIST_HEAD(blocked_list);
 
-static kmem_cache_t *filelock_cache;
+static kmem_cache_t *filelock_cache __read_mostly;
 
 /* Allocate an empty lock structure. */
 static struct file_lock *locks_alloc_lock(void)
 {
+       if (!vx_locks_avail(1))
+               return NULL;
        return kmem_cache_alloc(filelock_cache, SLAB_KERNEL);
 }
 
-/* Free a lock which is not in use. */
-static inline void locks_free_lock(struct file_lock *fl)
+static void locks_release_private(struct file_lock *fl)
 {
-       if (fl == NULL) {
-               BUG();
-               return;
+       if (fl->fl_ops) {
+               if (fl->fl_ops->fl_release_private)
+                       fl->fl_ops->fl_release_private(fl);
+               fl->fl_ops = NULL;
+       }
+       if (fl->fl_lmops) {
+               if (fl->fl_lmops->fl_release_private)
+                       fl->fl_lmops->fl_release_private(fl);
+               fl->fl_lmops = NULL;
        }
-       if (waitqueue_active(&fl->fl_wait))
-               panic("Attempting to free lock with active wait queue");
 
-       if (!list_empty(&fl->fl_block))
-               panic("Attempting to free lock with active block list");
+}
 
-       if (!list_empty(&fl->fl_link))
-               panic("Attempting to free lock on active lock list");
+/* Free a lock which is not in use. */
+static void locks_free_lock(struct file_lock *fl)
+{
+       BUG_ON(waitqueue_active(&fl->fl_wait));
+       BUG_ON(!list_empty(&fl->fl_block));
+       BUG_ON(!list_empty(&fl->fl_link));
 
+       vx_locks_dec(fl);
+       locks_release_private(fl);
        kmem_cache_free(filelock_cache, fl);
 }
 
@@ -183,9 +193,9 @@ void locks_init_lock(struct file_lock *fl)
        fl->fl_flags = 0;
        fl->fl_type = 0;
        fl->fl_start = fl->fl_end = 0;
-       fl->fl_notify = NULL;
-       fl->fl_insert = NULL;
-       fl->fl_remove = NULL;
+       fl->fl_ops = NULL;
+       fl->fl_lmops = NULL;
+       fl->fl_xid = -1;
 }
 
 EXPORT_SYMBOL(locks_init_lock);
@@ -205,22 +215,47 @@ static void init_once(void *foo, kmem_cache_t *cache, unsigned long flags)
        locks_init_lock(lock);
 }
 
+static void locks_copy_private(struct file_lock *new, struct file_lock *fl)
+{
+       if (fl->fl_ops) {
+               if (fl->fl_ops->fl_copy_lock)
+                       fl->fl_ops->fl_copy_lock(new, fl);
+               new->fl_ops = fl->fl_ops;
+       }
+       if (fl->fl_lmops) {
+               if (fl->fl_lmops->fl_copy_lock)
+                       fl->fl_lmops->fl_copy_lock(new, fl);
+               new->fl_lmops = fl->fl_lmops;
+       }
+}
+
 /*
  * Initialize a new lock from an existing file_lock structure.
  */
-void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
+static void __locks_copy_lock(struct file_lock *new, const struct file_lock *fl)
 {
        new->fl_owner = fl->fl_owner;
        new->fl_pid = fl->fl_pid;
-       new->fl_file = fl->fl_file;
+       new->fl_file = NULL;
        new->fl_flags = fl->fl_flags;
        new->fl_type = fl->fl_type;
        new->fl_start = fl->fl_start;
        new->fl_end = fl->fl_end;
-       new->fl_notify = fl->fl_notify;
-       new->fl_insert = fl->fl_insert;
-       new->fl_remove = fl->fl_remove;
-       new->fl_u = fl->fl_u;
+       new->fl_ops = NULL;
+       new->fl_lmops = NULL;
+}
+
+void locks_copy_lock(struct file_lock *new, struct file_lock *fl)
+{
+       locks_release_private(new);
+
+       __locks_copy_lock(new, fl);
+       new->fl_file = fl->fl_file;
+       new->fl_ops = fl->fl_ops;
+       new->fl_lmops = fl->fl_lmops;
+       new->fl_xid = fl->fl_xid;
+
+       locks_copy_private(new, fl);
 }
 
 EXPORT_SYMBOL(locks_copy_lock);
@@ -257,6 +292,9 @@ static int flock_make_lock(struct file *filp, struct file_lock **lock,
        fl->fl_flags = FL_FLOCK;
        fl->fl_type = type;
        fl->fl_end = OFFSET_MAX;
+
+       fl->fl_xid = filp->f_xid;
+       vx_locks_inc(fl);
        
        *lock = fl;
        return 0;
@@ -301,29 +339,29 @@ static int flock_to_posix_lock(struct file *filp, struct file_lock *fl,
        /* POSIX-1996 leaves the case l->l_len < 0 undefined;
           POSIX-2001 defines it. */
        start += l->l_start;
-       end = start + l->l_len - 1;
-       if (l->l_len < 0) {
+       if (start < 0)
+               return -EINVAL;
+       fl->fl_end = OFFSET_MAX;
+       if (l->l_len > 0) {
+               end = start + l->l_len - 1;
+               fl->fl_end = end;
+       } else if (l->l_len < 0) {
                end = start - 1;
+               fl->fl_end = end;
                start += l->l_len;
+               if (start < 0)
+                       return -EINVAL;
        }
-
-       if (start < 0)
-               return -EINVAL;
-       if (l->l_len > 0 && end < 0)
-               return -EOVERFLOW;
-
        fl->fl_start = start;   /* we record the absolute position */
-       fl->fl_end = end;
-       if (l->l_len == 0)
-               fl->fl_end = OFFSET_MAX;
+       if (fl->fl_end < fl->fl_start)
+               return -EOVERFLOW;
        
        fl->fl_owner = current->files;
        fl->fl_pid = current->tgid;
        fl->fl_file = filp;
        fl->fl_flags = FL_POSIX;
-       fl->fl_notify = NULL;
-       fl->fl_insert = NULL;
-       fl->fl_remove = NULL;
+       fl->fl_ops = NULL;
+       fl->fl_lmops = NULL;
 
        return assign_type(fl, l->l_type);
 }
@@ -348,22 +386,28 @@ static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
                return -EINVAL;
        }
 
-       if (((start += l->l_start) < 0) || (l->l_len < 0))
+       start += l->l_start;
+       if (start < 0)
                return -EINVAL;
-       fl->fl_end = start + l->l_len - 1;
-       if (l->l_len > 0 && fl->fl_end < 0)
-               return -EOVERFLOW;
+       fl->fl_end = OFFSET_MAX;
+       if (l->l_len > 0) {
+               fl->fl_end = start + l->l_len - 1;
+       } else if (l->l_len < 0) {
+               fl->fl_end = start - 1;
+               start += l->l_len;
+               if (start < 0)
+                       return -EINVAL;
+       }
        fl->fl_start = start;   /* we record the absolute position */
-       if (l->l_len == 0)
-               fl->fl_end = OFFSET_MAX;
+       if (fl->fl_end < fl->fl_start)
+               return -EOVERFLOW;
        
        fl->fl_owner = current->files;
        fl->fl_pid = current->tgid;
        fl->fl_file = filp;
        fl->fl_flags = FL_POSIX;
-       fl->fl_notify = NULL;
-       fl->fl_insert = NULL;
-       fl->fl_remove = NULL;
+       fl->fl_ops = NULL;
+       fl->fl_lmops = NULL;
 
        switch (l->l_type) {
        case F_RDLCK:
@@ -379,30 +423,73 @@ static int flock64_to_posix_lock(struct file *filp, struct file_lock *fl,
 }
 #endif
 
-/* Allocate a file_lock initialised to this type of lease */
-static int lease_alloc(struct file *filp, int type, struct file_lock **flp)
+/* default lease lock manager operations */
+static void lease_break_callback(struct file_lock *fl)
 {
-       struct file_lock *fl = locks_alloc_lock();
-       if (fl == NULL)
-               return -ENOMEM;
+       kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
+}
+
+static void lease_release_private_callback(struct file_lock *fl)
+{
+       if (!fl->fl_file)
+               return;
+
+       f_delown(fl->fl_file);
+       fl->fl_file->f_owner.signum = 0;
+}
+
+static int lease_mylease_callback(struct file_lock *fl, struct file_lock *try)
+{
+       return fl->fl_file == try->fl_file;
+}
+
+static struct lock_manager_operations lease_manager_ops = {
+       .fl_break = lease_break_callback,
+       .fl_release_private = lease_release_private_callback,
+       .fl_mylease = lease_mylease_callback,
+       .fl_change = lease_modify,
+};
+
+/*
+ * Initialize a lease, use the default lock manager operations
+ */
+static int lease_init(struct file *filp, int type, struct file_lock *fl)
+ {
+       if (assign_type(fl, type) != 0)
+               return -EINVAL;
 
        fl->fl_owner = current->files;
        fl->fl_pid = current->tgid;
+       fl->fl_xid = vx_current_xid();
 
        fl->fl_file = filp;
        fl->fl_flags = FL_LEASE;
-       if (assign_type(fl, type) != 0) {
-               locks_free_lock(fl);
-               return -EINVAL;
-       }
        fl->fl_start = 0;
        fl->fl_end = OFFSET_MAX;
-       fl->fl_notify = NULL;
-       fl->fl_insert = NULL;
-       fl->fl_remove = NULL;
+       fl->fl_ops = NULL;
+       fl->fl_lmops = &lease_manager_ops;
+       return 0;
+}
+
+/* Allocate a file_lock initialised to this type of lease */
+static int lease_alloc(struct file *filp, int type, struct file_lock **flp)
+{
+       struct file_lock *fl = locks_alloc_lock();
+       int error = -ENOMEM;
 
+       if (fl == NULL)
+               goto out;
+
+       fl->fl_xid = vx_current_xid();
+       vx_locks_inc(fl);
+       error = lease_init(filp, type, fl);
+       if (error) {
+               locks_free_lock(fl);
+               fl = NULL;
+       }
+out:
        *flp = fl;
-       return 0;
+       return error;
 }
 
 /* Check if two locks overlap each other.
@@ -414,20 +501,20 @@ static inline int locks_overlap(struct file_lock *fl1, struct file_lock *fl2)
 }
 
 /*
- * Check whether two locks have the same owner.  The apparently superfluous
- * check for fl_pid enables us to distinguish between locks set by lockd.
+ * Check whether two locks have the same owner.
  */
-static inline int
-posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
+static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
 {
-       return (fl1->fl_owner == fl2->fl_owner) &&
-               (fl1->fl_pid == fl2->fl_pid);
+       if (fl1->fl_lmops && fl1->fl_lmops->fl_compare_owner)
+               return fl2->fl_lmops == fl1->fl_lmops &&
+                       fl1->fl_lmops->fl_compare_owner(fl1, fl2);
+       return fl1->fl_owner == fl2->fl_owner;
 }
 
 /* Remove waiter from blocker's block list.
  * When blocker ends up pointing to itself then the list is empty.
  */
-static inline void __locks_delete_block(struct file_lock *waiter)
+static void __locks_delete_block(struct file_lock *waiter)
 {
        list_del_init(&waiter->fl_block);
        list_del_init(&waiter->fl_link);
@@ -451,15 +538,11 @@ static void locks_delete_block(struct file_lock *waiter)
 static void locks_insert_block(struct file_lock *blocker, 
                               struct file_lock *waiter)
 {
-       if (!list_empty(&waiter->fl_block)) {
-               printk(KERN_ERR "locks_insert_block: removing duplicated lock "
-                       "(pid=%d %Ld-%Ld type=%d)\n", waiter->fl_pid,
-                       waiter->fl_start, waiter->fl_end, waiter->fl_type);
-               __locks_delete_block(waiter);
-       }
+       BUG_ON(!list_empty(&waiter->fl_block));
        list_add_tail(&waiter->fl_block, &blocker->fl_block);
        waiter->fl_next = blocker;
-       list_add(&waiter->fl_link, &blocked_list);
+       if (IS_POSIX(blocker))
+               list_add(&waiter->fl_link, &blocked_list);
 }
 
 /* Wake up processes blocked waiting for blocker.
@@ -472,8 +555,8 @@ static void locks_wake_up_blocks(struct file_lock *blocker)
                struct file_lock *waiter = list_entry(blocker->fl_block.next,
                                struct file_lock, fl_block);
                __locks_delete_block(waiter);
-               if (waiter->fl_notify)
-                       waiter->fl_notify(waiter);
+               if (waiter->fl_lmops && waiter->fl_lmops->fl_notify)
+                       waiter->fl_lmops->fl_notify(waiter);
                else
                        wake_up(&waiter->fl_wait);
        }
@@ -490,8 +573,8 @@ static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
        fl->fl_next = *pos;
        *pos = fl;
 
-       if (fl->fl_insert)
-               fl->fl_insert(fl);
+       if (fl->fl_ops && fl->fl_ops->fl_insert)
+               fl->fl_ops->fl_insert(fl);
 }
 
 /*
@@ -514,8 +597,8 @@ static void locks_delete_lock(struct file_lock **thisfl_p)
                fl->fl_fasync = NULL;
        }
 
-       if (fl->fl_remove)
-               fl->fl_remove(fl);
+       if (fl->fl_ops && fl->fl_ops->fl_remove)
+               fl->fl_ops->fl_remove(fl);
 
        locks_wake_up_blocks(fl);
        locks_free_lock(fl);
@@ -594,8 +677,9 @@ static int locks_block_on_timeout(struct file_lock *blocker, struct file_lock *w
        return result;
 }
 
-struct file_lock *
-posix_test_lock(struct file *filp, struct file_lock *fl)
+int
+posix_test_lock(struct file *filp, struct file_lock *fl,
+               struct file_lock *conflock)
 {
        struct file_lock *cfl;
 
@@ -606,9 +690,13 @@ posix_test_lock(struct file *filp, struct file_lock *fl)
                if (posix_locks_conflict(cfl, fl))
                        break;
        }
+       if (cfl) {
+               __locks_copy_lock(conflock, cfl);
+               unlock_kernel();
+               return 1;
+       }
        unlock_kernel();
-
-       return (cfl);
+       return 0;
 }
 
 EXPORT_SYMBOL(posix_test_lock);
@@ -631,24 +719,15 @@ int posix_locks_deadlock(struct file_lock *caller_fl,
                                struct file_lock *block_fl)
 {
        struct list_head *tmp;
-       fl_owner_t caller_owner, blocked_owner;
-       unsigned int     caller_pid, blocked_pid;
-
-       caller_owner = caller_fl->fl_owner;
-       caller_pid = caller_fl->fl_pid;
-       blocked_owner = block_fl->fl_owner;
-       blocked_pid = block_fl->fl_pid;
 
 next_task:
-       if (caller_owner == blocked_owner && caller_pid == blocked_pid)
+       if (posix_same_owner(caller_fl, block_fl))
                return 1;
        list_for_each(tmp, &blocked_list) {
                struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
-               if ((fl->fl_owner == blocked_owner)
-                   && (fl->fl_pid == blocked_pid)) {
+               if (posix_same_owner(fl, block_fl)) {
                        fl = fl->fl_next;
-                       blocked_owner = fl->fl_owner;
-                       blocked_pid = fl->fl_pid;
+                       block_fl = fl;
                        goto next_task;
                }
        }
@@ -661,8 +740,9 @@ EXPORT_SYMBOL(posix_locks_deadlock);
  * at the head of the list, but that's secret knowledge known only to
  * flock_lock_file and posix_lock_file.
  */
-static int flock_lock_file(struct file *filp, struct file_lock *new_fl)
+static int flock_lock_file(struct file *filp, struct file_lock *request)
 {
+       struct file_lock *new_fl = NULL;
        struct file_lock **before;
        struct inode * inode = filp->f_dentry->d_inode;
        int error = 0;
@@ -677,17 +757,20 @@ static int flock_lock_file(struct file *filp, struct file_lock *new_fl)
                        continue;
                if (filp != fl->fl_file)
                        continue;
-               if (new_fl->fl_type == fl->fl_type)
+               if (request->fl_type == fl->fl_type)
                        goto out;
                found = 1;
                locks_delete_lock(before);
                break;
        }
-       unlock_kernel();
 
-       if (new_fl->fl_type == F_UNLCK)
-               return 0;
+       if (request->fl_type == F_UNLCK)
+               goto out;
 
+       error = -ENOMEM;
+       new_fl = locks_alloc_lock();
+       if (new_fl == NULL)
+               goto out;
        /*
         * If a higher-priority process was blocked on the old file lock,
         * give it the opportunity to lock the file.
@@ -695,32 +778,34 @@ static int flock_lock_file(struct file *filp, struct file_lock *new_fl)
        if (found)
                cond_resched();
 
-       lock_kernel();
        for_each_lock(inode, before) {
                struct file_lock *fl = *before;
                if (IS_POSIX(fl))
                        break;
                if (IS_LEASE(fl))
                        continue;
-               if (!flock_locks_conflict(new_fl, fl))
+               if (!flock_locks_conflict(request, fl))
                        continue;
                error = -EAGAIN;
-               if (new_fl->fl_flags & FL_SLEEP) {
-                       locks_insert_block(fl, new_fl);
-               }
+               if (request->fl_flags & FL_SLEEP)
+                       locks_insert_block(fl, request);
                goto out;
        }
+       locks_copy_lock(new_fl, request);
+       vx_locks_inc(new_fl);
        locks_insert_lock(&inode->i_flock, new_fl);
+       new_fl = NULL;
        error = 0;
 
 out:
        unlock_kernel();
+       if (new_fl)
+               locks_free_lock(new_fl);
        return error;
 }
 
-EXPORT_SYMBOL(posix_lock_file);
-
-static int __posix_lock_file(struct inode *inode, struct file_lock *request)
+static int __posix_lock_file_conf(struct inode *inode, struct file_lock *request,
+       struct file_lock *conflock, xid_t xid)
 {
        struct file_lock *fl;
        struct file_lock *new_fl, *new_fl2;
@@ -734,7 +819,11 @@ static int __posix_lock_file(struct inode *inode, struct file_lock *request)
         * so we get them in advance to avoid races.
         */
        new_fl = locks_alloc_lock();
+       new_fl->fl_xid = xid;
+       vx_locks_inc(new_fl);
        new_fl2 = locks_alloc_lock();
+       new_fl2->fl_xid = xid;
+       vx_locks_inc(new_fl2);
 
        lock_kernel();
        if (request->fl_type != F_UNLCK) {
@@ -744,6 +833,8 @@ static int __posix_lock_file(struct inode *inode, struct file_lock *request)
                                continue;
                        if (!posix_locks_conflict(request, fl))
                                continue;
+                       if (conflock)
+                               locks_copy_lock(conflock, fl);
                        error = -EAGAIN;
                        if (!(request->fl_flags & FL_SLEEP))
                                goto out;
@@ -785,12 +876,16 @@ static int __posix_lock_file(struct inode *inode, struct file_lock *request)
                /* Detect adjacent or overlapping regions (if same lock type)
                 */
                if (request->fl_type == fl->fl_type) {
+                       /* In all comparisons of start vs end, use
+                        * "start - 1" rather than "end + 1". If end
+                        * is OFFSET_MAX, end + 1 will become negative.
+                        */
                        if (fl->fl_end < request->fl_start - 1)
                                goto next_lock;
                        /* If the next lock in the list has entirely bigger
                         * addresses than the new one, insert the lock here.
                         */
-                       if (fl->fl_start > request->fl_end + 1)
+                       if (fl->fl_start - 1 > request->fl_end)
                                break;
 
                        /* If we come here, the new and old lock are of the
@@ -849,7 +944,8 @@ static int __posix_lock_file(struct inode *inode, struct file_lock *request)
                                fl->fl_start = request->fl_start;
                                fl->fl_end = request->fl_end;
                                fl->fl_type = request->fl_type;
-                               fl->fl_u = request->fl_u;
+                               locks_release_private(fl);
+                               locks_copy_private(fl, request);
                                request = fl;
                                added = 1;
                        }
@@ -908,8 +1004,54 @@ static int __posix_lock_file(struct inode *inode, struct file_lock *request)
  */
 int posix_lock_file(struct file *filp, struct file_lock *fl)
 {
-       return __posix_lock_file(filp->f_dentry->d_inode, fl);
+       return __posix_lock_file_conf(filp->f_dentry->d_inode,
+               fl, NULL, filp->f_xid);
+}
+EXPORT_SYMBOL(posix_lock_file);
+
+/**
+ * posix_lock_file_conf - Apply a POSIX-style lock to a file
+ * @filp: The file to apply the lock to
+ * @fl: The lock to be applied
+ * @conflock: Place to return a copy of the conflicting lock, if found.
+ *
+ * Except for the conflock parameter, acts just like posix_lock_file.
+ */
+int posix_lock_file_conf(struct file *filp, struct file_lock *fl,
+                       struct file_lock *conflock)
+{
+       return __posix_lock_file_conf(filp->f_dentry->d_inode,
+               fl, conflock, filp->f_xid);
 }
+EXPORT_SYMBOL(posix_lock_file_conf);
+
+/**
+ * posix_lock_file_wait - Apply a POSIX-style lock to a file
+ * @filp: The file to apply the lock to
+ * @fl: The lock to be applied
+ *
+ * Add a POSIX style lock to a file.
+ * We merge adjacent & overlapping locks whenever possible.
+ * POSIX locks are sorted by owner task, then by starting address
+ */
+int posix_lock_file_wait(struct file *filp, struct file_lock *fl)
+{
+       int error;
+       might_sleep ();
+       for (;;) {
+               error = posix_lock_file(filp, fl);
+               if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
+                       break;
+               error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
+               if (!error)
+                       continue;
+
+               locks_delete_block(fl);
+               break;
+       }
+       return error;
+}
+EXPORT_SYMBOL(posix_lock_file_wait);
 
 /**
  * locks_mandatory_locked - Check for an active lock
@@ -947,7 +1089,7 @@ int locks_mandatory_locked(struct inode *inode)
  * @count:      length of area to check
  *
  * Searches the inode's list of locks to find any POSIX locks which conflict.
- * This function is called from locks_verify_area() and
+ * This function is called from rw_verify_area() and
  * locks_verify_truncate().
  */
 int locks_mandatory_area(int read_write, struct inode *inode,
@@ -969,7 +1111,7 @@ int locks_mandatory_area(int read_write, struct inode *inode,
        fl.fl_end = offset + count - 1;
 
        for (;;) {
-               error = __posix_lock_file(inode, &fl);
+               error = __posix_lock_file_conf(inode, &fl, NULL, filp->f_xid);
                if (error != -EAGAIN)
                        break;
                if (!(fl.fl_flags & FL_SLEEP))
@@ -994,7 +1136,7 @@ int locks_mandatory_area(int read_write, struct inode *inode,
 EXPORT_SYMBOL(locks_mandatory_area);
 
 /* We already had a lease on this file; just change its type */
-static int lease_modify(struct file_lock **before, int arg)
+int lease_modify(struct file_lock **before, int arg)
 {
        struct file_lock *fl = *before;
        int error = assign_type(fl, arg);
@@ -1002,16 +1144,13 @@ static int lease_modify(struct file_lock **before, int arg)
        if (error)
                return error;
        locks_wake_up_blocks(fl);
-       if (arg == F_UNLCK) {
-               struct file *filp = fl->fl_file;
-
-               f_delown(filp);
-               filp->f_owner.signum = 0;
+       if (arg == F_UNLCK)
                locks_delete_lock(before);
-       }
        return 0;
 }
 
+EXPORT_SYMBOL(lease_modify);
+
 static void time_out_leases(struct inode *inode)
 {
        struct file_lock **before;
@@ -1024,7 +1163,6 @@ static void time_out_leases(struct inode *inode)
                        before = &fl->fl_next;
                        continue;
                }
-               printk(KERN_INFO "lease broken - owner pid = %d\n", fl->fl_pid);
                lease_modify(before, fl->fl_type & ~F_INPROGRESS);
                if (fl == *before)      /* lease_modify may have freed fl */
                        before = &fl->fl_next;
@@ -1095,7 +1233,8 @@ int __break_lease(struct inode *inode, unsigned int mode)
                if (fl->fl_type != future) {
                        fl->fl_type = future;
                        fl->fl_break_time = break_time;
-                       kill_fasync(&fl->fl_fasync, SIGIO, POLL_MSG);
+                       /* lease must have lmops break callback */
+                       fl->fl_lmops->fl_break(fl);
                }
        }
 
@@ -1146,7 +1285,7 @@ void lease_get_mtime(struct inode *inode, struct timespec *time)
 {
        struct file_lock *flock = inode->i_flock;
        if (flock && IS_LEASE(flock) && (flock->fl_type & F_WRLCK))
-               *time = CURRENT_TIME;
+               *time = current_fs_time(inode->i_sb);
        else
                *time = inode->i_mtime;
 }
@@ -1195,45 +1334,38 @@ int fcntl_getlease(struct file *filp)
 }
 
 /**
- *     fcntl_setlease  -       sets a lease on an open file
- *     @fd: open file descriptor
+ *     __setlease      -       sets a lease on an open file
  *     @filp: file pointer
  *     @arg: type of lease to obtain
+ *     @flp: input - file_lock to use, output - file_lock inserted
  *
- *     Call this fcntl to establish a lease on the file.
- *     Note that you also need to call %F_SETSIG to
- *     receive a signal when the lease is broken.
+ *     The (input) flp->fl_lmops->fl_break function is required
+ *     by break_lease().
+ *
+ *     Called with kernel lock held.
  */
-int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
+static int __setlease(struct file *filp, long arg, struct file_lock **flp)
 {
-       struct file_lock *fl, **before, **my_before = NULL;
-       struct dentry *dentry;
-       struct inode *inode;
+       struct file_lock *fl, **before, **my_before = NULL, *lease;
+       struct dentry *dentry = filp->f_dentry;
+       struct inode *inode = dentry->d_inode;
        int error, rdlease_count = 0, wrlease_count = 0;
 
-       dentry = filp->f_dentry;
-       inode = dentry->d_inode;
-
-       if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
-               return -EACCES;
-       if (!S_ISREG(inode->i_mode))
-               return -EINVAL;
-       error = security_file_lock(filp, arg);
-       if (error)
-               return error;
+       time_out_leases(inode);
 
-       lock_kernel();
+       error = -EINVAL;
+       if (!flp || !(*flp) || !(*flp)->fl_lmops || !(*flp)->fl_lmops->fl_break)
+               goto out;
 
-       time_out_leases(inode);
+       lease = *flp;
 
-       /*
-        * FIXME: What about F_RDLCK and files open for writing?
-        */
        error = -EAGAIN;
+       if ((arg == F_RDLCK) && (atomic_read(&inode->i_writecount) > 0))
+               goto out;
        if ((arg == F_WRLCK)
            && ((atomic_read(&dentry->d_count) > 1)
                || (atomic_read(&inode->i_count) > 1)))
-               goto out_unlock;
+               goto out;
 
        /*
         * At this point, we know that if there is an exclusive
@@ -1246,7 +1378,7 @@ int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
        for (before = &inode->i_flock;
                        ((fl = *before) != NULL) && IS_LEASE(fl);
                        before = &fl->fl_next) {
-               if (fl->fl_file == filp)
+               if (lease->fl_lmops->fl_mylease(fl, lease))
                        my_before = before;
                else if (fl->fl_type == (F_INPROGRESS | F_UNLCK))
                        /*
@@ -1261,39 +1393,146 @@ int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
 
        if ((arg == F_RDLCK && (wrlease_count > 0)) ||
            (arg == F_WRLCK && ((rdlease_count + wrlease_count) > 0)))
-               goto out_unlock;
+               goto out;
 
        if (my_before != NULL) {
-               error = lease_modify(my_before, arg);
-               goto out_unlock;
+               *flp = *my_before;
+               error = lease->fl_lmops->fl_change(my_before, arg);
+               goto out;
        }
 
        error = 0;
        if (arg == F_UNLCK)
-               goto out_unlock;
+               goto out;
 
        error = -EINVAL;
        if (!leases_enable)
-               goto out_unlock;
+               goto out;
 
        error = lease_alloc(filp, arg, &fl);
        if (error)
+               goto out;
+
+       locks_copy_lock(fl, lease);
+
+       locks_insert_lock(before, fl);
+
+       *flp = fl;
+out:
+       return error;
+}
+
+ /**
+ *     setlease        -       sets a lease on an open file
+ *     @filp: file pointer
+ *     @arg: type of lease to obtain
+ *     @lease: file_lock to use
+ *
+ *     Call this to establish a lease on the file.
+ *     The fl_lmops fl_break function is required by break_lease
+ */
+
+int setlease(struct file *filp, long arg, struct file_lock **lease)
+{
+       struct dentry *dentry = filp->f_dentry;
+       struct inode *inode = dentry->d_inode;
+       int error;
+
+       if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
+               return -EACCES;
+       if (!S_ISREG(inode->i_mode))
+               return -EINVAL;
+       error = security_file_lock(filp, arg);
+       if (error)
+               return error;
+
+       lock_kernel();
+       error = __setlease(filp, arg, lease);
+       unlock_kernel();
+
+       return error;
+}
+
+EXPORT_SYMBOL(setlease);
+
+/**
+ *     fcntl_setlease  -       sets a lease on an open file
+ *     @fd: open file descriptor
+ *     @filp: file pointer
+ *     @arg: type of lease to obtain
+ *
+ *     Call this fcntl to establish a lease on the file.
+ *     Note that you also need to call %F_SETSIG to
+ *     receive a signal when the lease is broken.
+ */
+int fcntl_setlease(unsigned int fd, struct file *filp, long arg)
+{
+       struct file_lock fl, *flp = &fl;
+       struct dentry *dentry = filp->f_dentry;
+       struct inode *inode = dentry->d_inode;
+       int error;
+
+       if ((current->fsuid != inode->i_uid) && !capable(CAP_LEASE))
+               return -EACCES;
+       if (!S_ISREG(inode->i_mode))
+               return -EINVAL;
+       error = security_file_lock(filp, arg);
+       if (error)
+               return error;
+
+       locks_init_lock(&fl);
+       error = lease_init(filp, arg, &fl);
+       if (error)
+               return error;
+
+       lock_kernel();
+
+       error = __setlease(filp, arg, &flp);
+       if (error || arg == F_UNLCK)
                goto out_unlock;
 
-       error = fasync_helper(fd, filp, 1, &fl->fl_fasync);
+       error = fasync_helper(fd, filp, 1, &flp->fl_fasync);
        if (error < 0) {
-               locks_free_lock(fl);
+               /* remove lease just inserted by __setlease */
+               flp->fl_type = F_UNLCK | F_INPROGRESS;
+               flp->fl_break_time = jiffies- 10;
+               time_out_leases(inode);
                goto out_unlock;
        }
 
-       locks_insert_lock(before, fl);
-
        error = f_setown(filp, current->pid, 0);
 out_unlock:
        unlock_kernel();
        return error;
 }
 
+/**
+ * flock_lock_file_wait - Apply a FLOCK-style lock to a file
+ * @filp: The file to apply the lock to
+ * @fl: The lock to be applied
+ *
+ * Add a FLOCK style lock to a file.
+ */
+int flock_lock_file_wait(struct file *filp, struct file_lock *fl)
+{
+       int error;
+       might_sleep();
+       for (;;) {
+               error = flock_lock_file(filp, fl);
+               if ((error != -EAGAIN) || !(fl->fl_flags & FL_SLEEP))
+                       break;
+               error = wait_event_interruptible(fl->fl_wait, !fl->fl_next);
+               if (!error)
+                       continue;
+
+               locks_delete_block(fl);
+               break;
+       }
+       return error;
+}
+
+EXPORT_SYMBOL(flock_lock_file_wait);
+
 /**
  *     sys_flock: - flock() system call.
  *     @fd: the file descriptor to lock.
@@ -1342,22 +1581,15 @@ asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
        if (error)
                goto out_free;
 
-       for (;;) {
-               error = flock_lock_file(filp, lock);
-               if ((error != -EAGAIN) || !can_sleep)
-                       break;
-               error = wait_event_interruptible(lock->fl_wait, !lock->fl_next);
-               if (!error)
-                       continue;
-
-               locks_delete_block(lock);
-               break;
-       }
+       if (filp->f_op && filp->f_op->flock)
+               error = filp->f_op->flock(filp,
+                                         (can_sleep) ? F_SETLKW : F_SETLK,
+                                         lock);
+       else
+               error = flock_lock_file_wait(filp, lock);
 
  out_free:
-       if (list_empty(&lock->fl_link)) {
-               locks_free_lock(lock);
-       }
+       locks_free_lock(lock);
 
  out_putf:
        fput(filp);
@@ -1370,7 +1602,7 @@ asmlinkage long sys_flock(unsigned int fd, unsigned int cmd)
  */
 int fcntl_getlk(struct file *filp, struct flock __user *l)
 {
-       struct file_lock *fl, file_lock;
+       struct file_lock *fl, cfl, file_lock;
        struct flock flock;
        int error;
 
@@ -1387,15 +1619,14 @@ int fcntl_getlk(struct file *filp, struct flock __user *l)
 
        if (filp->f_op && filp->f_op->lock) {
                error = filp->f_op->lock(filp, F_GETLK, &file_lock);
+               if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
+                       file_lock.fl_ops->fl_release_private(&file_lock);
                if (error < 0)
                        goto out;
-               else if (error == LOCK_USE_CLNT)
-                 /* Bypass for NFS with no locking - 2.0.36 compat */
-                 fl = posix_test_lock(filp, &file_lock);
                else
                  fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
        } else {
-               fl = posix_test_lock(filp, &file_lock);
+               fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
        }
  
        flock.l_type = F_UNLCK;
@@ -1422,7 +1653,6 @@ int fcntl_getlk(struct file *filp, struct flock __user *l)
        error = -EFAULT;
        if (!copy_to_user(l, &flock, sizeof(flock)))
                error = 0;
-  
 out:
        return error;
 }
@@ -1430,7 +1660,8 @@ out:
 /* Apply the lock described by l to an open file descriptor.
  * This implements both the F_SETLK and F_SETLKW commands of fcntl().
  */
-int fcntl_setlk(struct file *filp, unsigned int cmd, struct flock __user *l)
+int fcntl_setlk(unsigned int fd, struct file *filp, unsigned int cmd,
+               struct flock __user *l)
 {
        struct file_lock *file_lock = locks_alloc_lock();
        struct flock flock;
@@ -1440,6 +1671,9 @@ int fcntl_setlk(struct file *filp, unsigned int cmd, struct flock __user *l)
        if (file_lock == NULL)
                return -ENOLCK;
 
+       file_lock->fl_xid = filp->f_xid;
+       vx_locks_inc(file_lock);
+
        /*
         * This might block, so we do it before checking the inode.
         */
@@ -1459,6 +1693,7 @@ int fcntl_setlk(struct file *filp, unsigned int cmd, struct flock __user *l)
                goto out;
        }
 
+again:
        error = flock_to_posix_lock(filp, file_lock, &flock);
        if (error)
                goto out;
@@ -1487,26 +1722,33 @@ int fcntl_setlk(struct file *filp, unsigned int cmd, struct flock __user *l)
        if (error)
                goto out;
 
-       if (filp->f_op && filp->f_op->lock != NULL) {
+       if (filp->f_op && filp->f_op->lock != NULL)
                error = filp->f_op->lock(filp, cmd, file_lock);
-               if (error < 0)
-                       goto out;
-       }
+       else {
+               for (;;) {
+                       error = posix_lock_file(filp, file_lock);
+                       if ((error != -EAGAIN) || (cmd == F_SETLK))
+                               break;
+                       error = wait_event_interruptible(file_lock->fl_wait,
+                                       !file_lock->fl_next);
+                       if (!error)
+                               continue;
 
-       for (;;) {
-               error = __posix_lock_file(inode, file_lock);
-               if ((error != -EAGAIN) || (cmd == F_SETLK))
+                       locks_delete_block(file_lock);
                        break;
-               error = wait_event_interruptible(file_lock->fl_wait,
-                               !file_lock->fl_next);
-               if (!error)
-                       continue;
+               }
+       }
 
-               locks_delete_block(file_lock);
-               break;
+       /*
+        * Attempt to detect a close/fcntl race and recover by
+        * releasing the lock that was just acquired.
+        */
+       if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
+               flock.l_type = F_UNLCK;
+               goto again;
        }
 
- out:
+out:
        locks_free_lock(file_lock);
        return error;
 }
@@ -1517,7 +1759,7 @@ int fcntl_setlk(struct file *filp, unsigned int cmd, struct flock __user *l)
  */
 int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
 {
-       struct file_lock *fl, file_lock;
+       struct file_lock *fl, cfl, file_lock;
        struct flock64 flock;
        int error;
 
@@ -1534,15 +1776,14 @@ int fcntl_getlk64(struct file *filp, struct flock64 __user *l)
 
        if (filp->f_op && filp->f_op->lock) {
                error = filp->f_op->lock(filp, F_GETLK, &file_lock);
+               if (file_lock.fl_ops && file_lock.fl_ops->fl_release_private)
+                       file_lock.fl_ops->fl_release_private(&file_lock);
                if (error < 0)
                        goto out;
-               else if (error == LOCK_USE_CLNT)
-                 /* Bypass for NFS with no locking - 2.0.36 compat */
-                 fl = posix_test_lock(filp, &file_lock);
                else
                  fl = (file_lock.fl_type == F_UNLCK ? NULL : &file_lock);
        } else {
-               fl = posix_test_lock(filp, &file_lock);
+               fl = (posix_test_lock(filp, &file_lock, &cfl) ? &cfl : NULL);
        }
  
        flock.l_type = F_UNLCK;
@@ -1565,7 +1806,8 @@ out:
 /* Apply the lock described by l to an open file descriptor.
  * This implements both the F_SETLK and F_SETLKW commands of fcntl().
  */
-int fcntl_setlk64(struct file *filp, unsigned int cmd, struct flock64 __user *l)
+int fcntl_setlk64(unsigned int fd, struct file *filp, unsigned int cmd,
+               struct flock64 __user *l)
 {
        struct file_lock *file_lock = locks_alloc_lock();
        struct flock64 flock;
@@ -1575,6 +1817,9 @@ int fcntl_setlk64(struct file *filp, unsigned int cmd, struct flock64 __user *l)
        if (file_lock == NULL)
                return -ENOLCK;
 
+       file_lock->fl_xid = filp->f_xid;
+       vx_locks_inc(file_lock);
+
        /*
         * This might block, so we do it before checking the inode.
         */
@@ -1594,6 +1839,7 @@ int fcntl_setlk64(struct file *filp, unsigned int cmd, struct flock64 __user *l)
                goto out;
        }
 
+again:
        error = flock64_to_posix_lock(filp, file_lock, &flock);
        if (error)
                goto out;
@@ -1622,23 +1868,30 @@ int fcntl_setlk64(struct file *filp, unsigned int cmd, struct flock64 __user *l)
        if (error)
                goto out;
 
-       if (filp->f_op && filp->f_op->lock != NULL) {
+       if (filp->f_op && filp->f_op->lock != NULL)
                error = filp->f_op->lock(filp, cmd, file_lock);
-               if (error < 0)
-                       goto out;
-       }
+       else {
+               for (;;) {
+                       error = posix_lock_file(filp, file_lock);
+                       if ((error != -EAGAIN) || (cmd == F_SETLK64))
+                               break;
+                       error = wait_event_interruptible(file_lock->fl_wait,
+                                       !file_lock->fl_next);
+                       if (!error)
+                               continue;
 
-       for (;;) {
-               error = __posix_lock_file(inode, file_lock);
-               if ((error != -EAGAIN) || (cmd == F_SETLK64))
+                       locks_delete_block(file_lock);
                        break;
-               error = wait_event_interruptible(file_lock->fl_wait,
-                               !file_lock->fl_next);
-               if (!error)
-                       continue;
+               }
+       }
 
-               locks_delete_block(file_lock);
-               break;
+       /*
+        * Attempt to detect a close/fcntl race and recover by
+        * releasing the lock that was just acquired.
+        */
+       if (!error && fcheck(fd) != filp && flock.l_type != F_UNLCK) {
+               flock.l_type = F_UNLCK;
+               goto again;
        }
 
 out:
@@ -1672,10 +1925,12 @@ void locks_remove_posix(struct file *filp, fl_owner_t owner)
        lock.fl_owner = owner;
        lock.fl_pid = current->tgid;
        lock.fl_file = filp;
+       lock.fl_ops = NULL;
+       lock.fl_lmops = NULL;
 
        if (filp->f_op && filp->f_op->lock != NULL) {
                filp->f_op->lock(filp, F_SETLK, &lock);
-               /* Ignore any error -- we must remove the locks anyway */
+               goto out;
        }
 
        /* Can't use posix_lock_file here; we need to remove it no matter
@@ -1684,13 +1939,16 @@ void locks_remove_posix(struct file *filp, fl_owner_t owner)
        lock_kernel();
        while (*before != NULL) {
                struct file_lock *fl = *before;
-               if (IS_POSIX(fl) && (fl->fl_owner == owner)) {
+               if (IS_POSIX(fl) && posix_same_owner(fl, &lock)) {
                        locks_delete_lock(before);
                        continue;
                }
                before = &fl->fl_next;
        }
        unlock_kernel();
+out:
+       if (lock.fl_ops && lock.fl_ops->fl_release_private)
+               lock.fl_ops->fl_release_private(&lock);
 }
 
 EXPORT_SYMBOL(locks_remove_posix);
@@ -1707,6 +1965,19 @@ void locks_remove_flock(struct file *filp)
        if (!inode->i_flock)
                return;
 
+       if (filp->f_op && filp->f_op->flock) {
+               struct file_lock fl = {
+                       .fl_pid = current->tgid,
+                       .fl_file = filp,
+                       .fl_flags = FL_FLOCK,
+                       .fl_type = F_UNLCK,
+                       .fl_end = OFFSET_MAX,
+               };
+               filp->f_op->flock(filp, F_SETLKW, &fl);
+               if (fl.fl_ops && fl.fl_ops->fl_release_private)
+                       fl.fl_ops->fl_release_private(&fl);
+       }
+
        lock_kernel();
        before = &inode->i_flock;
 
@@ -1720,9 +1991,7 @@ void locks_remove_flock(struct file *filp)
                                lease_modify(before, F_UNLCK);
                                continue;
                        }
-                       /* FL_POSIX locks of this process have already been
-                        * removed in filp_close->locks_remove_posix.
-                        */
+                       /* What? */
                        BUG();
                }
                before = &fl->fl_next;
@@ -1730,21 +1999,6 @@ void locks_remove_flock(struct file *filp)
        unlock_kernel();
 }
 
-/**
- *     posix_block_lock - blocks waiting for a file lock
- *     @blocker: the lock which is blocking
- *     @waiter: the lock which conflicts and has to wait
- *
- * lockd needs to block waiting for locks.
- */
-void
-posix_block_lock(struct file_lock *blocker, struct file_lock *waiter)
-{
-       locks_insert_block(blocker, waiter);
-}
-
-EXPORT_SYMBOL(posix_block_lock);
-
 /**
  *     posix_unblock_lock - stop waiting for a file lock
  *      @filp:   how the file was opened
@@ -1752,22 +2006,18 @@ EXPORT_SYMBOL(posix_block_lock);
  *
  *     lockd needs to block waiting for locks.
  */
-void
+int
 posix_unblock_lock(struct file *filp, struct file_lock *waiter)
 {
-       /* 
-        * A remote machine may cancel the lock request after it's been
-        * granted locally.  If that happens, we need to delete the lock.
-        */
+       int status = 0;
+
        lock_kernel();
-       if (waiter->fl_next) {
+       if (waiter->fl_next)
                __locks_delete_block(waiter);
-               unlock_kernel();
-       } else {
-               unlock_kernel();
-               waiter->fl_type = F_UNLCK;
-               posix_lock_file(filp, waiter);
-       }
+       else
+               status = -ENOENT;
+       unlock_kernel();
+       return status;
 }
 
 EXPORT_SYMBOL(posix_unblock_lock);
@@ -1880,6 +2130,10 @@ int get_locks_status(char *buffer, char **start, off_t offset, int length)
        list_for_each(tmp, &file_lock_list) {
                struct list_head *btmp;
                struct file_lock *fl = list_entry(tmp, struct file_lock, fl_link);
+
+               if (!vx_check(fl->fl_xid, VX_IDENT|VX_WATCH))
+                       continue;
+
                lock_get_status(q, fl, ++i, "");
                move_lock_status(&q, &pos, offset);
 
@@ -2004,21 +2258,29 @@ void steal_locks(fl_owner_t from)
 {
        struct files_struct *files = current->files;
        int i, j;
+       struct fdtable *fdt;
 
        if (from == files)
                return;
 
        lock_kernel();
        j = 0;
+
+       /*
+        * We are not taking a ref to the file structures, so
+        * we need to acquire ->file_lock.
+        */
+       spin_lock(&files->file_lock);
+       fdt = files_fdtable(files);
        for (;;) {
                unsigned long set;
                i = j * __NFDBITS;
-               if (i >= files->max_fdset || i >= files->max_fds)
+               if (i >= fdt->max_fdset || i >= fdt->max_fds)
                        break;
-               set = files->open_fds->fds_bits[j++];
+               set = fdt->open_fds->fds_bits[j++];
                while (set) {
                        if (set & 1) {
-                               struct file *file = files->fd[i];
+                               struct file *file = fdt->fd[i];
                                if (file)
                                        __steal_locks(file, from);
                        }
@@ -2026,6 +2288,7 @@ void steal_locks(fl_owner_t from)
                        set >>= 1;
                }
        }
+       spin_unlock(&files->file_lock);
        unlock_kernel();
 }
 EXPORT_SYMBOL(steal_locks);
@@ -2038,4 +2301,4 @@ static int __init filelock_init(void)
        return 0;
 }
 
-module_init(filelock_init)
+core_initcall(filelock_init);