]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blobdiff - fs/locks.c
HID: wacom: drop WACOM_PKGLEN_STATUS
[mirror_ubuntu-artful-kernel.git] / fs / locks.c
index 59e2f905e4ffea324dbf44faf1b666974adc6c23..4d0d41163a50fedbe7c9b7d81911c5d5f2fbec6f 100644 (file)
@@ -157,14 +157,11 @@ static int target_leasetype(struct file_lock *fl)
 int leases_enable = 1;
 int lease_break_time = 45;
 
-#define for_each_lock(inode, lockp) \
-       for (lockp = &inode->i_flock; *lockp != NULL; lockp = &(*lockp)->fl_next)
-
 /*
  * The global file_lock_list is only used for displaying /proc/locks, so we
  * keep a list on each CPU, with each list protected by its own spinlock via
  * the file_lock_lglock. Note that alterations to the list also require that
- * the relevant i_lock is held.
+ * the relevant flc_lock is held.
  */
 DEFINE_STATIC_LGLOCK(file_lock_lglock);
 static DEFINE_PER_CPU(struct hlist_head, file_lock_list);
@@ -192,21 +189,68 @@ static DEFINE_HASHTABLE(blocked_hash, BLOCKED_HASH_BITS);
  * contrast to those that are acting as records of acquired locks).
  *
  * Note that when we acquire this lock in order to change the above fields,
- * we often hold the i_lock as well. In certain cases, when reading the fields
+ * we often hold the flc_lock as well. In certain cases, when reading the fields
  * protected by this lock, we can skip acquiring it iff we already hold the
- * i_lock.
+ * flc_lock.
  *
  * In particular, adding an entry to the fl_block list requires that you hold
- * both the i_lock and the blocked_lock_lock (acquired in that order). Deleting
- * an entry from the list however only requires the file_lock_lock.
+ * both the flc_lock and the blocked_lock_lock (acquired in that order).
+ * Deleting an entry from the list however only requires the file_lock_lock.
  */
 static DEFINE_SPINLOCK(blocked_lock_lock);
 
+static struct kmem_cache *flctx_cache __read_mostly;
 static struct kmem_cache *filelock_cache __read_mostly;
 
+static struct file_lock_context *
+locks_get_lock_context(struct inode *inode)
+{
+       struct file_lock_context *new;
+
+       if (likely(inode->i_flctx))
+               goto out;
+
+       new = kmem_cache_alloc(flctx_cache, GFP_KERNEL);
+       if (!new)
+               goto out;
+
+       spin_lock_init(&new->flc_lock);
+       INIT_LIST_HEAD(&new->flc_flock);
+       INIT_LIST_HEAD(&new->flc_posix);
+       INIT_LIST_HEAD(&new->flc_lease);
+
+       /*
+        * Assign the pointer if it's not already assigned. If it is, then
+        * free the context we just allocated.
+        */
+       spin_lock(&inode->i_lock);
+       if (likely(!inode->i_flctx)) {
+               inode->i_flctx = new;
+               new = NULL;
+       }
+       spin_unlock(&inode->i_lock);
+
+       if (new)
+               kmem_cache_free(flctx_cache, new);
+out:
+       return inode->i_flctx;
+}
+
+void
+locks_free_lock_context(struct file_lock_context *ctx)
+{
+       if (ctx) {
+               WARN_ON_ONCE(!list_empty(&ctx->flc_flock));
+               WARN_ON_ONCE(!list_empty(&ctx->flc_posix));
+               WARN_ON_ONCE(!list_empty(&ctx->flc_lease));
+               kmem_cache_free(flctx_cache, ctx);
+       }
+}
+
 static void locks_init_lock_heads(struct file_lock *fl)
 {
        INIT_HLIST_NODE(&fl->fl_link);
+       INIT_LIST_HEAD(&fl->fl_list);
        INIT_LIST_HEAD(&fl->fl_block);
        init_waitqueue_head(&fl->fl_wait);
 }
@@ -243,6 +287,7 @@ EXPORT_SYMBOL_GPL(locks_release_private);
 void locks_free_lock(struct file_lock *fl)
 {
        BUG_ON(waitqueue_active(&fl->fl_wait));
+       BUG_ON(!list_empty(&fl->fl_list));
        BUG_ON(!list_empty(&fl->fl_block));
        BUG_ON(!hlist_unhashed(&fl->fl_link));
 
@@ -257,8 +302,8 @@ locks_dispose_list(struct list_head *dispose)
        struct file_lock *fl;
 
        while (!list_empty(dispose)) {
-               fl = list_first_entry(dispose, struct file_lock, fl_block);
-               list_del_init(&fl->fl_block);
+               fl = list_first_entry(dispose, struct file_lock, fl_list);
+               list_del_init(&fl->fl_list);
                locks_free_lock(fl);
        }
 }
@@ -513,7 +558,7 @@ static int posix_same_owner(struct file_lock *fl1, struct file_lock *fl2)
        return fl1->fl_owner == fl2->fl_owner;
 }
 
-/* Must be called with the i_lock held! */
+/* Must be called with the flc_lock held! */
 static void locks_insert_global_locks(struct file_lock *fl)
 {
        lg_local_lock(&file_lock_lglock);
@@ -522,12 +567,12 @@ static void locks_insert_global_locks(struct file_lock *fl)
        lg_local_unlock(&file_lock_lglock);
 }
 
-/* Must be called with the i_lock held! */
+/* Must be called with the flc_lock held! */
 static void locks_delete_global_locks(struct file_lock *fl)
 {
        /*
         * Avoid taking lock if already unhashed. This is safe since this check
-        * is done while holding the i_lock, and new insertions into the list
+        * is done while holding the flc_lock, and new insertions into the list
         * also require that it be held.
         */
        if (hlist_unhashed(&fl->fl_link))
@@ -579,10 +624,10 @@ static void locks_delete_block(struct file_lock *waiter)
  * the order they blocked. The documentation doesn't require this but
  * it seems like the reasonable thing to do.
  *
- * Must be called with both the i_lock and blocked_lock_lock held. The fl_block
- * list itself is protected by the blocked_lock_lock, but by ensuring that the
- * i_lock is also held on insertions we can avoid taking the blocked_lock_lock
- * in some cases when we see that the fl_block list is empty.
+ * Must be called with both the flc_lock and blocked_lock_lock held. The
+ * fl_block list itself is protected by the blocked_lock_lock, but by ensuring
+ * that the flc_lock is also held on insertions we can avoid taking the
+ * blocked_lock_lock in some cases when we see that the fl_block list is empty.
  */
 static void __locks_insert_block(struct file_lock *blocker,
                                        struct file_lock *waiter)
@@ -594,7 +639,7 @@ static void __locks_insert_block(struct file_lock *blocker,
                locks_insert_global_blocked(waiter);
 }
 
-/* Must be called with i_lock held. */
+/* Must be called with flc_lock held. */
 static void locks_insert_block(struct file_lock *blocker,
                                        struct file_lock *waiter)
 {
@@ -606,15 +651,15 @@ static void locks_insert_block(struct file_lock *blocker,
 /*
  * Wake up processes blocked waiting for blocker.
  *
- * Must be called with the inode->i_lock held!
+ * Must be called with the inode->flc_lock held!
  */
 static void locks_wake_up_blocks(struct file_lock *blocker)
 {
        /*
         * Avoid taking global lock if list is empty. This is safe since new
-        * blocked requests are only added to the list under the i_lock, and
-        * the i_lock is always held here. Note that removal from the fl_block
-        * list does not require the i_lock, so we must recheck list_empty()
+        * blocked requests are only added to the list under the flc_lock, and
+        * the flc_lock is always held here. Note that removal from the fl_block
+        * list does not require the flc_lock, so we must recheck list_empty()
         * after acquiring the blocked_lock_lock.
         */
        if (list_empty(&blocker->fl_block))
@@ -635,63 +680,36 @@ static void locks_wake_up_blocks(struct file_lock *blocker)
        spin_unlock(&blocked_lock_lock);
 }
 
-/* Insert file lock fl into an inode's lock list at the position indicated
- * by pos. At the same time add the lock to the global file lock list.
- *
- * Must be called with the i_lock held!
- */
-static void locks_insert_lock(struct file_lock **pos, struct file_lock *fl)
+static void
+locks_insert_lock_ctx(struct file_lock *fl, int *counter,
+                     struct list_head *before)
 {
        fl->fl_nspid = get_pid(task_tgid(current));
-
-       /* insert into file's list */
-       fl->fl_next = *pos;
-       *pos = fl;
-
+       list_add_tail(&fl->fl_list, before);
+       ++*counter;
        locks_insert_global_locks(fl);
 }
 
-/**
- * locks_delete_lock - Delete a lock and then free it.
- * @thisfl_p: pointer that points to the fl_next field of the previous
- *           inode->i_flock list entry
- *
- * Unlink a lock from all lists and free the namespace reference, but don't
- * free it yet. Wake up processes that are blocked waiting for this lock and
- * notify the FS that the lock has been cleared.
- *
- * Must be called with the i_lock held!
- */
-static void locks_unlink_lock(struct file_lock **thisfl_p)
+static void
+locks_unlink_lock_ctx(struct file_lock *fl, int *counter)
 {
-       struct file_lock *fl = *thisfl_p;
-
        locks_delete_global_locks(fl);
-
-       *thisfl_p = fl->fl_next;
-       fl->fl_next = NULL;
-
+       list_del_init(&fl->fl_list);
+       --*counter;
        if (fl->fl_nspid) {
                put_pid(fl->fl_nspid);
                fl->fl_nspid = NULL;
        }
-
        locks_wake_up_blocks(fl);
 }
 
-/*
- * Unlink a lock from all lists and free it.
- *
- * Must be called with i_lock held!
- */
-static void locks_delete_lock(struct file_lock **thisfl_p,
-                             struct list_head *dispose)
+static void
+locks_delete_lock_ctx(struct file_lock *fl, int *counter,
+                     struct list_head *dispose)
 {
-       struct file_lock *fl = *thisfl_p;
-
-       locks_unlink_lock(thisfl_p);
+       locks_unlink_lock_ctx(fl, counter);
        if (dispose)
-               list_add(&fl->fl_block, dispose);
+               list_add(&fl->fl_list, dispose);
        else
                locks_free_lock(fl);
 }
@@ -746,22 +764,27 @@ void
 posix_test_lock(struct file *filp, struct file_lock *fl)
 {
        struct file_lock *cfl;
+       struct file_lock_context *ctx;
        struct inode *inode = file_inode(filp);
 
-       spin_lock(&inode->i_lock);
-       for (cfl = file_inode(filp)->i_flock; cfl; cfl = cfl->fl_next) {
-               if (!IS_POSIX(cfl))
-                       continue;
-               if (posix_locks_conflict(fl, cfl))
-                       break;
-       }
-       if (cfl) {
-               locks_copy_conflock(fl, cfl);
-               if (cfl->fl_nspid)
-                       fl->fl_pid = pid_vnr(cfl->fl_nspid);
-       } else
+       ctx = inode->i_flctx;
+       if (!ctx || list_empty_careful(&ctx->flc_posix)) {
                fl->fl_type = F_UNLCK;
-       spin_unlock(&inode->i_lock);
+               return;
+       }
+
+       spin_lock(&ctx->flc_lock);
+       list_for_each_entry(cfl, &ctx->flc_posix, fl_list) {
+               if (posix_locks_conflict(fl, cfl)) {
+                       locks_copy_conflock(fl, cfl);
+                       if (cfl->fl_nspid)
+                               fl->fl_pid = pid_vnr(cfl->fl_nspid);
+                       goto out;
+               }
+       }
+       fl->fl_type = F_UNLCK;
+out:
+       spin_unlock(&ctx->flc_lock);
        return;
 }
 EXPORT_SYMBOL(posix_test_lock);
@@ -845,34 +868,34 @@ static int posix_locks_deadlock(struct file_lock *caller_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 = file_inode(filp);
+       struct file_lock *fl;
+       struct file_lock_context *ctx;
+       struct inode *inode = file_inode(filp);
        int error = 0;
-       int found = 0;
+       bool found = false;
        LIST_HEAD(dispose);
 
+       ctx = locks_get_lock_context(inode);
+       if (!ctx)
+               return -ENOMEM;
+
        if (!(request->fl_flags & FL_ACCESS) && (request->fl_type != F_UNLCK)) {
                new_fl = locks_alloc_lock();
                if (!new_fl)
                        return -ENOMEM;
        }
 
-       spin_lock(&inode->i_lock);
+       spin_lock(&ctx->flc_lock);
        if (request->fl_flags & FL_ACCESS)
                goto find_conflict;
 
-       for_each_lock(inode, before) {
-               struct file_lock *fl = *before;
-               if (IS_POSIX(fl))
-                       break;
-               if (IS_LEASE(fl))
-                       continue;
+       list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
                if (filp != fl->fl_file)
                        continue;
                if (request->fl_type == fl->fl_type)
                        goto out;
-               found = 1;
-               locks_delete_lock(before, &dispose);
+               found = true;
+               locks_delete_lock_ctx(fl, &ctx->flc_flock_cnt, &dispose);
                break;
        }
 
@@ -887,18 +910,13 @@ static int flock_lock_file(struct file *filp, struct file_lock *request)
         * give it the opportunity to lock the file.
         */
        if (found) {
-               spin_unlock(&inode->i_lock);
+               spin_unlock(&ctx->flc_lock);
                cond_resched();
-               spin_lock(&inode->i_lock);
+               spin_lock(&ctx->flc_lock);
        }
 
 find_conflict:
-       for_each_lock(inode, before) {
-               struct file_lock *fl = *before;
-               if (IS_POSIX(fl))
-                       break;
-               if (IS_LEASE(fl))
-                       continue;
+       list_for_each_entry(fl, &ctx->flc_flock, fl_list) {
                if (!flock_locks_conflict(request, fl))
                        continue;
                error = -EAGAIN;
@@ -911,12 +929,12 @@ find_conflict:
        if (request->fl_flags & FL_ACCESS)
                goto out;
        locks_copy_lock(new_fl, request);
-       locks_insert_lock(before, new_fl);
+       locks_insert_lock_ctx(new_fl, &ctx->flc_flock_cnt, &ctx->flc_flock);
        new_fl = NULL;
        error = 0;
 
 out:
-       spin_unlock(&inode->i_lock);
+       spin_unlock(&ctx->flc_lock);
        if (new_fl)
                locks_free_lock(new_fl);
        locks_dispose_list(&dispose);
@@ -925,16 +943,20 @@ out:
 
 static int __posix_lock_file(struct inode *inode, struct file_lock *request, struct file_lock *conflock)
 {
-       struct file_lock *fl;
+       struct file_lock *fl, *tmp;
        struct file_lock *new_fl = NULL;
        struct file_lock *new_fl2 = NULL;
        struct file_lock *left = NULL;
        struct file_lock *right = NULL;
-       struct file_lock **before;
+       struct file_lock_context *ctx;
        int error;
        bool added = false;
        LIST_HEAD(dispose);
 
+       ctx = locks_get_lock_context(inode);
+       if (!ctx)
+               return -ENOMEM;
+
        /*
         * We may need two file_lock structures for this operation,
         * so we get them in advance to avoid races.
@@ -948,15 +970,14 @@ static int __posix_lock_file(struct inode *inode, struct file_lock *request, str
                new_fl2 = locks_alloc_lock();
        }
 
-       spin_lock(&inode->i_lock);
+       spin_lock(&ctx->flc_lock);
        /*
         * New lock request. Walk all POSIX locks and look for conflicts. If
         * there are any, either return error or put the request on the
         * blocker's list of waiters and the global blocked_hash.
         */
        if (request->fl_type != F_UNLCK) {
-               for_each_lock(inode, before) {
-                       fl = *before;
+               list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
                        if (!IS_POSIX(fl))
                                continue;
                        if (!posix_locks_conflict(request, fl))
@@ -986,29 +1007,25 @@ static int __posix_lock_file(struct inode *inode, struct file_lock *request, str
        if (request->fl_flags & FL_ACCESS)
                goto out;
 
-       /*
-        * Find the first old lock with the same owner as the new lock.
-        */
-       
-       before = &inode->i_flock;
-
-       /* First skip locks owned by other processes.  */
-       while ((fl = *before) && (!IS_POSIX(fl) ||
-                                 !posix_same_owner(request, fl))) {
-               before = &fl->fl_next;
+       /* Find the first old lock with the same owner as the new lock */
+       list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
+               if (posix_same_owner(request, fl))
+                       break;
        }
 
        /* Process locks with this owner. */
-       while ((fl = *before) && posix_same_owner(request, fl)) {
-               /* Detect adjacent or overlapping regions (if same lock type)
-                */
+       list_for_each_entry_safe_from(fl, tmp, &ctx->flc_posix, fl_list) {
+               if (!posix_same_owner(request, fl))
+                       break;
+
+               /* 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;
+                               continue;
                        /* If the next lock in the list has entirely bigger
                         * addresses than the new one, insert the lock here.
                         */
@@ -1029,18 +1046,18 @@ static int __posix_lock_file(struct inode *inode, struct file_lock *request, str
                        else
                                request->fl_end = fl->fl_end;
                        if (added) {
-                               locks_delete_lock(before, &dispose);
+                               locks_delete_lock_ctx(fl, &ctx->flc_posix_cnt,
+                                                       &dispose);
                                continue;
                        }
                        request = fl;
                        added = true;
-               }
-               else {
+               } else {
                        /* Processing for different lock types is a bit
                         * more complex.
                         */
                        if (fl->fl_end < request->fl_start)
-                               goto next_lock;
+                               continue;
                        if (fl->fl_start > request->fl_end)
                                break;
                        if (request->fl_type == F_UNLCK)
@@ -1059,7 +1076,8 @@ static int __posix_lock_file(struct inode *inode, struct file_lock *request, str
                                 * one (This may happen several times).
                                 */
                                if (added) {
-                                       locks_delete_lock(before, &dispose);
+                                       locks_delete_lock_ctx(fl,
+                                               &ctx->flc_posix_cnt, &dispose);
                                        continue;
                                }
                                /*
@@ -1075,15 +1093,13 @@ static int __posix_lock_file(struct inode *inode, struct file_lock *request, str
                                locks_copy_lock(new_fl, request);
                                request = new_fl;
                                new_fl = NULL;
-                               locks_delete_lock(before, &dispose);
-                               locks_insert_lock(before, request);
+                               locks_insert_lock_ctx(request,
+                                       &ctx->flc_posix_cnt, &fl->fl_list);
+                               locks_delete_lock_ctx(fl,
+                                       &ctx->flc_posix_cnt, &dispose);
                                added = true;
                        }
                }
-               /* Go on to next lock.
-                */
-       next_lock:
-               before = &fl->fl_next;
        }
 
        /*
@@ -1108,7 +1124,8 @@ static int __posix_lock_file(struct inode *inode, struct file_lock *request, str
                        goto out;
                }
                locks_copy_lock(new_fl, request);
-               locks_insert_lock(before, new_fl);
+               locks_insert_lock_ctx(new_fl, &ctx->flc_posix_cnt,
+                                       &fl->fl_list);
                new_fl = NULL;
        }
        if (right) {
@@ -1119,7 +1136,8 @@ static int __posix_lock_file(struct inode *inode, struct file_lock *request, str
                        left = new_fl2;
                        new_fl2 = NULL;
                        locks_copy_lock(left, right);
-                       locks_insert_lock(before, left);
+                       locks_insert_lock_ctx(left, &ctx->flc_posix_cnt,
+                                               &fl->fl_list);
                }
                right->fl_start = request->fl_end + 1;
                locks_wake_up_blocks(right);
@@ -1129,7 +1147,7 @@ static int __posix_lock_file(struct inode *inode, struct file_lock *request, str
                locks_wake_up_blocks(left);
        }
  out:
-       spin_unlock(&inode->i_lock);
+       spin_unlock(&ctx->flc_lock);
        /*
         * Free any unused locks.
         */
@@ -1199,22 +1217,29 @@ EXPORT_SYMBOL(posix_lock_file_wait);
  */
 int locks_mandatory_locked(struct file *file)
 {
+       int ret;
        struct inode *inode = file_inode(file);
+       struct file_lock_context *ctx;
        struct file_lock *fl;
 
+       ctx = inode->i_flctx;
+       if (!ctx || list_empty_careful(&ctx->flc_posix))
+               return 0;
+
        /*
         * Search the lock list for this inode for any POSIX locks.
         */
-       spin_lock(&inode->i_lock);
-       for (fl = inode->i_flock; fl != NULL; fl = fl->fl_next) {
-               if (!IS_POSIX(fl))
-                       continue;
+       spin_lock(&ctx->flc_lock);
+       ret = 0;
+       list_for_each_entry(fl, &ctx->flc_posix, fl_list) {
                if (fl->fl_owner != current->files &&
-                   fl->fl_owner != file)
+                   fl->fl_owner != file) {
+                       ret = -EAGAIN;
                        break;
+               }
        }
-       spin_unlock(&inode->i_lock);
-       return fl ? -EAGAIN : 0;
+       spin_unlock(&ctx->flc_lock);
+       return ret;
 }
 
 /**
@@ -1294,9 +1319,9 @@ static void lease_clear_pending(struct file_lock *fl, int arg)
 }
 
 /* We already had a lease on this file; just change its type */
-int lease_modify(struct file_lock **before, int arg, struct list_head *dispose)
+int lease_modify(struct file_lock *fl, int arg, struct list_head *dispose)
 {
-       struct file_lock *fl = *before;
+       struct file_lock_context *flctx;
        int error = assign_type(fl, arg);
 
        if (error)
@@ -1306,6 +1331,7 @@ int lease_modify(struct file_lock **before, int arg, struct list_head *dispose)
        if (arg == F_UNLCK) {
                struct file *filp = fl->fl_file;
 
+               flctx = file_inode(filp)->i_flctx;
                f_delown(filp);
                filp->f_owner.signum = 0;
                fasync_helper(0, fl->fl_file, 0, &fl->fl_fasync);
@@ -1313,7 +1339,7 @@ int lease_modify(struct file_lock **before, int arg, struct list_head *dispose)
                        printk(KERN_ERR "locks_delete_lock: fasync == %p\n", fl->fl_fasync);
                        fl->fl_fasync = NULL;
                }
-               locks_delete_lock(before, dispose);
+               locks_delete_lock_ctx(fl, &flctx->flc_lease_cnt, dispose);
        }
        return 0;
 }
@@ -1329,20 +1355,17 @@ static bool past_time(unsigned long then)
 
 static void time_out_leases(struct inode *inode, struct list_head *dispose)
 {
-       struct file_lock **before;
-       struct file_lock *fl;
+       struct file_lock_context *ctx = inode->i_flctx;
+       struct file_lock *fl, *tmp;
 
-       lockdep_assert_held(&inode->i_lock);
+       lockdep_assert_held(&ctx->flc_lock);
 
-       before = &inode->i_flock;
-       while ((fl = *before) && IS_LEASE(fl) && lease_breaking(fl)) {
+       list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list) {
                trace_time_out_leases(inode, fl);
                if (past_time(fl->fl_downgrade_time))
-                       lease_modify(before, F_RDLCK, dispose);
+                       lease_modify(fl, F_RDLCK, dispose);
                if (past_time(fl->fl_break_time))
-                       lease_modify(before, F_UNLCK, dispose);
-               if (fl == *before)      /* lease_modify may have freed fl */
-                       before = &fl->fl_next;
+                       lease_modify(fl, F_UNLCK, dispose);
        }
 }
 
@@ -1356,11 +1379,12 @@ static bool leases_conflict(struct file_lock *lease, struct file_lock *breaker)
 static bool
 any_leases_conflict(struct inode *inode, struct file_lock *breaker)
 {
+       struct file_lock_context *ctx = inode->i_flctx;
        struct file_lock *fl;
 
-       lockdep_assert_held(&inode->i_lock);
+       lockdep_assert_held(&ctx->flc_lock);
 
-       for (fl = inode->i_flock ; fl && IS_LEASE(fl); fl = fl->fl_next) {
+       list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
                if (leases_conflict(fl, breaker))
                        return true;
        }
@@ -1384,7 +1408,8 @@ int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
 {
        int error = 0;
        struct file_lock *new_fl;
-       struct file_lock *fl, **before;
+       struct file_lock_context *ctx = inode->i_flctx;
+       struct file_lock *fl;
        unsigned long break_time;
        int want_write = (mode & O_ACCMODE) != O_RDONLY;
        LIST_HEAD(dispose);
@@ -1394,7 +1419,13 @@ int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
                return PTR_ERR(new_fl);
        new_fl->fl_flags = type;
 
-       spin_lock(&inode->i_lock);
+       /* typically we will check that ctx is non-NULL before calling */
+       if (!ctx) {
+               WARN_ON_ONCE(1);
+               return error;
+       }
+
+       spin_lock(&ctx->flc_lock);
 
        time_out_leases(inode, &dispose);
 
@@ -1408,9 +1439,7 @@ int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
                        break_time++;   /* so that 0 means no break time */
        }
 
-       for (before = &inode->i_flock;
-                       ((fl = *before) != NULL) && IS_LEASE(fl);
-                       before = &fl->fl_next) {
+       list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
                if (!leases_conflict(fl, new_fl))
                        continue;
                if (want_write) {
@@ -1419,17 +1448,17 @@ int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
                        fl->fl_flags |= FL_UNLOCK_PENDING;
                        fl->fl_break_time = break_time;
                } else {
-                       if (lease_breaking(inode->i_flock))
+                       if (lease_breaking(fl))
                                continue;
                        fl->fl_flags |= FL_DOWNGRADE_PENDING;
                        fl->fl_downgrade_time = break_time;
                }
                if (fl->fl_lmops->lm_break(fl))
-                       locks_delete_lock(before, &dispose);
+                       locks_delete_lock_ctx(fl, &ctx->flc_lease_cnt,
+                                               &dispose);
        }
 
-       fl = inode->i_flock;
-       if (!fl || !IS_LEASE(fl))
+       if (list_empty(&ctx->flc_lease))
                goto out;
 
        if (mode & O_NONBLOCK) {
@@ -1439,18 +1468,19 @@ int __break_lease(struct inode *inode, unsigned int mode, unsigned int type)
        }
 
 restart:
-       break_time = inode->i_flock->fl_break_time;
+       fl = list_first_entry(&ctx->flc_lease, struct file_lock, fl_list);
+       break_time = fl->fl_break_time;
        if (break_time != 0)
                break_time -= jiffies;
        if (break_time == 0)
                break_time++;
-       locks_insert_block(inode->i_flock, new_fl);
+       locks_insert_block(fl, new_fl);
        trace_break_lease_block(inode, new_fl);
-       spin_unlock(&inode->i_lock);
+       spin_unlock(&ctx->flc_lock);
        locks_dispose_list(&dispose);
        error = wait_event_interruptible_timeout(new_fl->fl_wait,
                                                !new_fl->fl_next, break_time);
-       spin_lock(&inode->i_lock);
+       spin_lock(&ctx->flc_lock);
        trace_break_lease_unblock(inode, new_fl);
        locks_delete_block(new_fl);
        if (error >= 0) {
@@ -1462,12 +1492,10 @@ restart:
                        time_out_leases(inode, &dispose);
                if (any_leases_conflict(inode, new_fl))
                        goto restart;
-
                error = 0;
        }
-
 out:
-       spin_unlock(&inode->i_lock);
+       spin_unlock(&ctx->flc_lock);
        locks_dispose_list(&dispose);
        locks_free_lock(new_fl);
        return error;
@@ -1487,14 +1515,18 @@ EXPORT_SYMBOL(__break_lease);
 void lease_get_mtime(struct inode *inode, struct timespec *time)
 {
        bool has_lease = false;
-       struct file_lock *flock;
+       struct file_lock_context *ctx = inode->i_flctx;
+       struct file_lock *fl;
 
-       if (inode->i_flock) {
-               spin_lock(&inode->i_lock);
-               flock = inode->i_flock;
-               if (flock && IS_LEASE(flock) && (flock->fl_type == F_WRLCK))
-                       has_lease = true;
-               spin_unlock(&inode->i_lock);
+       if (ctx && !list_empty_careful(&ctx->flc_lease)) {
+               spin_lock(&ctx->flc_lock);
+               if (!list_empty(&ctx->flc_lease)) {
+                       fl = list_first_entry(&ctx->flc_lease,
+                                               struct file_lock, fl_list);
+                       if (fl->fl_type == F_WRLCK)
+                               has_lease = true;
+               }
+               spin_unlock(&ctx->flc_lock);
        }
 
        if (has_lease)
@@ -1532,20 +1564,22 @@ int fcntl_getlease(struct file *filp)
 {
        struct file_lock *fl;
        struct inode *inode = file_inode(filp);
+       struct file_lock_context *ctx = inode->i_flctx;
        int type = F_UNLCK;
        LIST_HEAD(dispose);
 
-       spin_lock(&inode->i_lock);
-       time_out_leases(file_inode(filp), &dispose);
-       for (fl = file_inode(filp)->i_flock; fl && IS_LEASE(fl);
-                       fl = fl->fl_next) {
-               if (fl->fl_file == filp) {
+       if (ctx && !list_empty_careful(&ctx->flc_lease)) {
+               spin_lock(&ctx->flc_lock);
+               time_out_leases(file_inode(filp), &dispose);
+               list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
+                       if (fl->fl_file != filp)
+                               continue;
                        type = target_leasetype(fl);
                        break;
                }
+               spin_unlock(&ctx->flc_lock);
+               locks_dispose_list(&dispose);
        }
-       spin_unlock(&inode->i_lock);
-       locks_dispose_list(&dispose);
        return type;
 }
 
@@ -1578,9 +1612,10 @@ check_conflicting_open(const struct dentry *dentry, const long arg)
 static int
 generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **priv)
 {
-       struct file_lock *fl, **before, **my_before = NULL, *lease;
+       struct file_lock *fl, *my_fl = NULL, *lease;
        struct dentry *dentry = filp->f_path.dentry;
        struct inode *inode = dentry->d_inode;
+       struct file_lock_context *ctx;
        bool is_deleg = (*flp)->fl_flags & FL_DELEG;
        int error;
        LIST_HEAD(dispose);
@@ -1588,6 +1623,10 @@ generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **pr
        lease = *flp;
        trace_generic_add_lease(inode, lease);
 
+       ctx = locks_get_lock_context(inode);
+       if (!ctx)
+               return -ENOMEM;
+
        /*
         * In the delegation case we need mutual exclusion with
         * a number of operations that take the i_mutex.  We trylock
@@ -1606,7 +1645,7 @@ generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **pr
                return -EINVAL;
        }
 
-       spin_lock(&inode->i_lock);
+       spin_lock(&ctx->flc_lock);
        time_out_leases(inode, &dispose);
        error = check_conflicting_open(dentry, arg);
        if (error)
@@ -1621,13 +1660,12 @@ generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **pr
         * except for this filp.
         */
        error = -EAGAIN;
-       for (before = &inode->i_flock;
-                       ((fl = *before) != NULL) && IS_LEASE(fl);
-                       before = &fl->fl_next) {
+       list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
                if (fl->fl_file == filp) {
-                       my_before = before;
+                       my_fl = fl;
                        continue;
                }
+
                /*
                 * No exclusive leases if someone else has a lease on
                 * this file:
@@ -1642,9 +1680,8 @@ generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **pr
                        goto out;
        }
 
-       if (my_before != NULL) {
-               lease = *my_before;
-               error = lease->fl_lmops->lm_change(my_before, arg, &dispose);
+       if (my_fl != NULL) {
+               error = lease->fl_lmops->lm_change(my_fl, arg, &dispose);
                if (error)
                        goto out;
                goto out_setup;
@@ -1654,7 +1691,7 @@ generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **pr
        if (!leases_enable)
                goto out;
 
-       locks_insert_lock(before, lease);
+       locks_insert_lock_ctx(lease, &ctx->flc_lease_cnt, &ctx->flc_lease);
        /*
         * The check in break_lease() is lockless. It's possible for another
         * open to race in after we did the earlier check for a conflicting
@@ -1666,45 +1703,49 @@ generic_add_lease(struct file *filp, long arg, struct file_lock **flp, void **pr
         */
        smp_mb();
        error = check_conflicting_open(dentry, arg);
-       if (error)
-               goto out_unlink;
+       if (error) {
+               locks_unlink_lock_ctx(lease, &ctx->flc_lease_cnt);
+               goto out;
+       }
 
 out_setup:
        if (lease->fl_lmops->lm_setup)
                lease->fl_lmops->lm_setup(lease, priv);
 out:
-       spin_unlock(&inode->i_lock);
+       spin_unlock(&ctx->flc_lock);
        locks_dispose_list(&dispose);
        if (is_deleg)
                mutex_unlock(&inode->i_mutex);
-       if (!error && !my_before)
+       if (!error && !my_fl)
                *flp = NULL;
        return error;
-out_unlink:
-       locks_unlink_lock(before);
-       goto out;
 }
 
 static int generic_delete_lease(struct file *filp)
 {
        int error = -EAGAIN;
-       struct file_lock *fl, **before;
+       struct file_lock *fl, *victim = NULL;
        struct dentry *dentry = filp->f_path.dentry;
        struct inode *inode = dentry->d_inode;
+       struct file_lock_context *ctx = inode->i_flctx;
        LIST_HEAD(dispose);
 
-       spin_lock(&inode->i_lock);
-       time_out_leases(inode, &dispose);
-       for (before = &inode->i_flock;
-                       ((fl = *before) != NULL) && IS_LEASE(fl);
-                       before = &fl->fl_next) {
-               if (fl->fl_file == filp)
+       if (!ctx) {
+               trace_generic_delete_lease(inode, NULL);
+               return error;
+       }
+
+       spin_lock(&ctx->flc_lock);
+       list_for_each_entry(fl, &ctx->flc_lease, fl_list) {
+               if (fl->fl_file == filp) {
+                       victim = fl;
                        break;
+               }
        }
        trace_generic_delete_lease(inode, fl);
-       if (fl && IS_LEASE(fl))
-               error = fl->fl_lmops->lm_change(before, F_UNLCK, &dispose);
-       spin_unlock(&inode->i_lock);
+       if (victim)
+               error = fl->fl_lmops->lm_change(victim, F_UNLCK, &dispose);
+       spin_unlock(&ctx->flc_lock);
        locks_dispose_list(&dispose);
        return error;
 }
@@ -2171,7 +2212,7 @@ again:
         */
        /*
         * we need that spin_lock here - it prevents reordering between
-        * update of inode->i_flock and check for it done in close().
+        * update of i_flctx->flc_posix and check for it done in close().
         * rcu_read_lock() wouldn't do.
         */
        spin_lock(&current->files->file_lock);
@@ -2331,13 +2372,14 @@ out:
 void locks_remove_posix(struct file *filp, fl_owner_t owner)
 {
        struct file_lock lock;
+       struct file_lock_context *ctx = file_inode(filp)->i_flctx;
 
        /*
         * If there are no locks held on this file, we don't need to call
         * posix_lock_file().  Another process could be setting a lock on this
         * file at the same time, but we wouldn't remove that lock anyway.
         */
-       if (!file_inode(filp)->i_flock)
+       if (!ctx || list_empty(&ctx->flc_posix))
                return;
 
        lock.fl_type = F_UNLCK;
@@ -2358,67 +2400,67 @@ void locks_remove_posix(struct file *filp, fl_owner_t owner)
 
 EXPORT_SYMBOL(locks_remove_posix);
 
+/* The i_flctx must be valid when calling into here */
+static void
+locks_remove_flock(struct file *filp)
+{
+       struct file_lock fl = {
+               .fl_owner = filp,
+               .fl_pid = current->tgid,
+               .fl_file = filp,
+               .fl_flags = FL_FLOCK,
+               .fl_type = F_UNLCK,
+               .fl_end = OFFSET_MAX,
+       };
+       struct file_lock_context *flctx = file_inode(filp)->i_flctx;
+
+       if (list_empty(&flctx->flc_flock))
+               return;
+
+       if (filp->f_op->flock)
+               filp->f_op->flock(filp, F_SETLKW, &fl);
+       else
+               flock_lock_file(filp, &fl);
+
+       if (fl.fl_ops && fl.fl_ops->fl_release_private)
+               fl.fl_ops->fl_release_private(&fl);
+}
+
+/* The i_flctx must be valid when calling into here */
+static void
+locks_remove_lease(struct file *filp)
+{
+       struct inode *inode = file_inode(filp);
+       struct file_lock_context *ctx = inode->i_flctx;
+       struct file_lock *fl, *tmp;
+       LIST_HEAD(dispose);
+
+       if (list_empty(&ctx->flc_lease))
+               return;
+
+       spin_lock(&ctx->flc_lock);
+       list_for_each_entry_safe(fl, tmp, &ctx->flc_lease, fl_list)
+               lease_modify(fl, F_UNLCK, &dispose);
+       spin_unlock(&ctx->flc_lock);
+       locks_dispose_list(&dispose);
+}
+
 /*
  * This function is called on the last close of an open file.
  */
 void locks_remove_file(struct file *filp)
 {
-       struct inode * inode = file_inode(filp);
-       struct file_lock *fl;
-       struct file_lock **before;
-       LIST_HEAD(dispose);
-
-       if (!inode->i_flock)
+       if (!file_inode(filp)->i_flctx)
                return;
 
+       /* remove any OFD locks */
        locks_remove_posix(filp, filp);
 
-       if (filp->f_op->flock) {
-               struct file_lock fl = {
-                       .fl_owner = filp,
-                       .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);
-       }
-
-       spin_lock(&inode->i_lock);
-       before = &inode->i_flock;
-
-       while ((fl = *before) != NULL) {
-               if (fl->fl_file == filp) {
-                       if (IS_LEASE(fl)) {
-                               lease_modify(before, F_UNLCK, &dispose);
-                               continue;
-                       }
-
-                       /*
-                        * There's a leftover lock on the list of a type that
-                        * we didn't expect to see. Most likely a classic
-                        * POSIX lock that ended up not getting released
-                        * properly, or that raced onto the list somehow. Log
-                        * some info about it and then just remove it from
-                        * the list.
-                        */
-                       WARN(!IS_FLOCK(fl),
-                               "leftover lock: dev=%u:%u ino=%lu type=%hhd flags=0x%x start=%lld end=%lld\n",
-                               MAJOR(inode->i_sb->s_dev),
-                               MINOR(inode->i_sb->s_dev), inode->i_ino,
-                               fl->fl_type, fl->fl_flags,
-                               fl->fl_start, fl->fl_end);
+       /* remove flock locks */
+       locks_remove_flock(filp);
 
-                       locks_delete_lock(before, &dispose);
-                       continue;
-               }
-               before = &fl->fl_next;
-       }
-       spin_unlock(&inode->i_lock);
-       locks_dispose_list(&dispose);
+       /* remove any leases */
+       locks_remove_lease(filp);
 }
 
 /**
@@ -2621,6 +2663,9 @@ static int __init filelock_init(void)
 {
        int i;
 
+       flctx_cache = kmem_cache_create("file_lock_ctx",
+                       sizeof(struct file_lock_context), 0, SLAB_PANIC, NULL);
+
        filelock_cache = kmem_cache_create("file_lock_cache",
                        sizeof(struct file_lock), 0, SLAB_PANIC, NULL);