]> git.kernelconcepts.de Git - karo-tx-linux.git/commitdiff
FS-Cache: The operation cancellation method needs calling in more places
authorDavid Howells <dhowells@redhat.com>
Tue, 24 Feb 2015 10:05:29 +0000 (10:05 +0000)
committerDavid Howells <dhowells@redhat.com>
Thu, 2 Apr 2015 13:28:53 +0000 (14:28 +0100)
Any time an incomplete operation is cancelled, the operation cancellation
function needs to be called to clean up.  This is currently being passed
directly to some of the functions that might want to call it, but not all.

Instead, pass the cancellation method pointer to the fscache_operation_init()
and have that cache it in the operation struct.  Further, plug in a dummy
cancellation handler if the caller declines to set one as this allows us to
call the function unconditionally (the extra overhead isn't worth bothering
about as we don't expect to be calling this typically).

The cancellation method must thence be called everywhere the CANCELLED state
is set.  Note that we call it *before* setting the CANCELLED state such that
the method can use the old state value to guide its operation.

fscache_do_cancel_retrieval() needs moving higher up in the sources so that
the init function can use it now.

Without this, the following oops may be seen:

FS-Cache: Assertion failed
FS-Cache: 3 == 0 is false
------------[ cut here ]------------
kernel BUG at ../fs/fscache/page.c:261!
...
RIP: 0010:[<ffffffffa0089c1b>]  fscache_release_retrieval_op+0x77/0x100
 [<ffffffffa008853d>] fscache_put_operation+0x114/0x2da
 [<ffffffffa008b8c2>] __fscache_read_or_alloc_pages+0x358/0x3b3
 [<ffffffffa00b761f>] __nfs_readpages_from_fscache+0x59/0xbf [nfs]
 [<ffffffffa00b06c5>] nfs_readpages+0x10c/0x185 [nfs]
 [<ffffffff81124925>] ? alloc_pages_current+0x119/0x13e
 [<ffffffff810ee5fd>] ? __page_cache_alloc+0xfb/0x10a
 [<ffffffff810f87f8>] __do_page_cache_readahead+0x188/0x22c
 [<ffffffff810f8b3a>] ondemand_readahead+0x29e/0x2af
 [<ffffffff810f8c92>] page_cache_sync_readahead+0x38/0x3a
 [<ffffffff810ef337>] generic_file_read_iter+0x1a2/0x55a
 [<ffffffffa00a9dff>] ? nfs_revalidate_mapping+0xd6/0x288 [nfs]
 [<ffffffffa00a6a23>] nfs_file_read+0x49/0x70 [nfs]
 [<ffffffff811363be>] new_sync_read+0x78/0x9c
 [<ffffffff81137164>] __vfs_read+0x13/0x38
 [<ffffffff8113721e>] vfs_read+0x95/0x121
 [<ffffffff811372f6>] SyS_read+0x4c/0x8a
 [<ffffffff81557a52>] system_call_fastpath+0x12/0x17

The assertion is showing that the remaining number of pages (n_pages) is not 0
when the operation is being released.

Signed-off-by: David Howells <dhowells@redhat.com>
Reviewed-by: Steve Dickson <steved@redhat.com>
Acked-by: Jeff Layton <jeff.layton@primarydata.com>
fs/fscache/cookie.c
fs/fscache/internal.h
fs/fscache/object.c
fs/fscache/operation.c
fs/fscache/page.c
include/linux/fscache-cache.h

index 8de22164f5fb1f9a49d62559e42880115789bfda..d403c69bee0829fba9187114bc318dd810e8d179 100644 (file)
@@ -672,7 +672,7 @@ int __fscache_check_consistency(struct fscache_cookie *cookie)
        if (!op)
                return -ENOMEM;
 
-       fscache_operation_init(op, NULL, NULL);
+       fscache_operation_init(op, NULL, NULL, NULL);
        op->flags = FSCACHE_OP_MYTHREAD |
                (1 << FSCACHE_OP_WAITING) |
                (1 << FSCACHE_OP_UNUSE_COOKIE);
@@ -696,8 +696,7 @@ int __fscache_check_consistency(struct fscache_cookie *cookie)
        /* the work queue now carries its own ref on the object */
        spin_unlock(&cookie->lock);
 
-       ret = fscache_wait_for_operation_activation(object, op,
-                                                   NULL, NULL, NULL);
+       ret = fscache_wait_for_operation_activation(object, op, NULL, NULL);
        if (ret == 0) {
                /* ask the cache to honour the operation */
                ret = object->cache->ops->check_consistency(op);
index a63225116db644e1a04f37d5ccf8533d6151e099..97ec4511095765ffa96f98538aedc74cae8fbadf 100644 (file)
@@ -124,9 +124,7 @@ extern int fscache_submit_exclusive_op(struct fscache_object *,
                                       struct fscache_operation *);
 extern int fscache_submit_op(struct fscache_object *,
                             struct fscache_operation *);
-extern int fscache_cancel_op(struct fscache_operation *,
-                            void (*)(struct fscache_operation *),
-                            bool);
+extern int fscache_cancel_op(struct fscache_operation *, bool);
 extern void fscache_cancel_all_ops(struct fscache_object *);
 extern void fscache_abort_object(struct fscache_object *);
 extern void fscache_start_operations(struct fscache_object *);
@@ -139,8 +137,7 @@ extern int fscache_wait_for_deferred_lookup(struct fscache_cookie *);
 extern int fscache_wait_for_operation_activation(struct fscache_object *,
                                                 struct fscache_operation *,
                                                 atomic_t *,
-                                                atomic_t *,
-                                                void (*)(struct fscache_operation *));
+                                                atomic_t *);
 extern void fscache_invalidate_writes(struct fscache_cookie *);
 
 /*
index 40049f7505f0cb4551b8616948665db4e273da87..9e792e30f4db47b38c6db644487c440a2e12febb 100644 (file)
@@ -961,7 +961,8 @@ static const struct fscache_state *_fscache_invalidate_object(struct fscache_obj
        if (!op)
                goto nomem;
 
-       fscache_operation_init(op, object->cache->ops->invalidate_object, NULL);
+       fscache_operation_init(op, object->cache->ops->invalidate_object,
+                              NULL, NULL);
        op->flags = FSCACHE_OP_ASYNC |
                (1 << FSCACHE_OP_EXCLUSIVE) |
                (1 << FSCACHE_OP_UNUSE_COOKIE);
index c76c097307682d195a5b3e9c827b36698deed5d1..57d4abb68656fd61915e9a5663bd5f8bcb3952bc 100644 (file)
 atomic_t fscache_op_debug_id;
 EXPORT_SYMBOL(fscache_op_debug_id);
 
+static void fscache_operation_dummy_cancel(struct fscache_operation *op)
+{
+}
+
 /**
  * fscache_operation_init - Do basic initialisation of an operation
  * @op: The operation to initialise
@@ -30,6 +34,7 @@ EXPORT_SYMBOL(fscache_op_debug_id);
  */
 void fscache_operation_init(struct fscache_operation *op,
                            fscache_operation_processor_t processor,
+                           fscache_operation_cancel_t cancel,
                            fscache_operation_release_t release)
 {
        INIT_WORK(&op->work, fscache_op_work_func);
@@ -37,6 +42,7 @@ void fscache_operation_init(struct fscache_operation *op,
        op->state = FSCACHE_OP_ST_INITIALISED;
        op->debug_id = atomic_inc_return(&fscache_op_debug_id);
        op->processor = processor;
+       op->cancel = cancel ?: fscache_operation_dummy_cancel;
        op->release = release;
        INIT_LIST_HEAD(&op->pend_link);
        fscache_stat(&fscache_n_op_initialised);
@@ -164,9 +170,11 @@ int fscache_submit_exclusive_op(struct fscache_object *object,
        flags = READ_ONCE(object->flags);
        if (unlikely(!(flags & BIT(FSCACHE_OBJECT_IS_LIVE)))) {
                fscache_stat(&fscache_n_op_rejected);
+               op->cancel(op);
                op->state = FSCACHE_OP_ST_CANCELLED;
                ret = -ENOBUFS;
        } else if (unlikely(fscache_cache_is_broken(object))) {
+               op->cancel(op);
                op->state = FSCACHE_OP_ST_CANCELLED;
                ret = -EIO;
        } else if (flags & BIT(FSCACHE_OBJECT_IS_AVAILABLE)) {
@@ -200,10 +208,12 @@ int fscache_submit_exclusive_op(struct fscache_object *object,
                fscache_stat(&fscache_n_op_pend);
                ret = 0;
        } else if (flags & BIT(FSCACHE_OBJECT_KILLED_BY_CACHE)) {
+               op->cancel(op);
                op->state = FSCACHE_OP_ST_CANCELLED;
                ret = -ENOBUFS;
        } else {
                fscache_report_unexpected_submission(object, op, ostate);
+               op->cancel(op);
                op->state = FSCACHE_OP_ST_CANCELLED;
                ret = -ENOBUFS;
        }
@@ -245,9 +255,11 @@ int fscache_submit_op(struct fscache_object *object,
        flags = READ_ONCE(object->flags);
        if (unlikely(!(flags & BIT(FSCACHE_OBJECT_IS_LIVE)))) {
                fscache_stat(&fscache_n_op_rejected);
+               op->cancel(op);
                op->state = FSCACHE_OP_ST_CANCELLED;
                ret = -ENOBUFS;
        } else if (unlikely(fscache_cache_is_broken(object))) {
+               op->cancel(op);
                op->state = FSCACHE_OP_ST_CANCELLED;
                ret = -EIO;
        } else if (flags & BIT(FSCACHE_OBJECT_IS_AVAILABLE)) {
@@ -276,11 +288,13 @@ int fscache_submit_op(struct fscache_object *object,
                fscache_stat(&fscache_n_op_pend);
                ret = 0;
        } else if (flags & BIT(FSCACHE_OBJECT_KILLED_BY_CACHE)) {
+               op->cancel(op);
                op->state = FSCACHE_OP_ST_CANCELLED;
                ret = -ENOBUFS;
        } else {
                fscache_report_unexpected_submission(object, op, ostate);
                ASSERT(!fscache_object_is_active(object));
+               op->cancel(op);
                op->state = FSCACHE_OP_ST_CANCELLED;
                ret = -ENOBUFS;
        }
@@ -335,7 +349,6 @@ void fscache_start_operations(struct fscache_object *object)
  * cancel an operation that's pending on an object
  */
 int fscache_cancel_op(struct fscache_operation *op,
-                     void (*do_cancel)(struct fscache_operation *),
                      bool cancel_in_progress_op)
 {
        struct fscache_object *object = op->object;
@@ -355,9 +368,9 @@ int fscache_cancel_op(struct fscache_operation *op,
                ASSERT(!list_empty(&op->pend_link));
                list_del_init(&op->pend_link);
                put = true;
+
                fscache_stat(&fscache_n_op_cancelled);
-               if (do_cancel)
-                       do_cancel(op);
+               op->cancel(op);
                op->state = FSCACHE_OP_ST_CANCELLED;
                if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags))
                        object->n_exclusive--;
@@ -373,8 +386,7 @@ int fscache_cancel_op(struct fscache_operation *op,
                        fscache_start_operations(object);
 
                fscache_stat(&fscache_n_op_cancelled);
-               if (do_cancel)
-                       do_cancel(op);
+               op->cancel(op);
                op->state = FSCACHE_OP_ST_CANCELLED;
                if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags))
                        object->n_exclusive--;
@@ -408,6 +420,7 @@ void fscache_cancel_all_ops(struct fscache_object *object)
                list_del_init(&op->pend_link);
 
                ASSERTCMP(op->state, ==, FSCACHE_OP_ST_PENDING);
+               op->cancel(op);
                op->state = FSCACHE_OP_ST_CANCELLED;
 
                if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags))
@@ -440,8 +453,12 @@ void fscache_op_complete(struct fscache_operation *op, bool cancelled)
 
        spin_lock(&object->lock);
 
-       op->state = cancelled ?
-               FSCACHE_OP_ST_CANCELLED : FSCACHE_OP_ST_COMPLETE;
+       if (!cancelled) {
+               op->state = FSCACHE_OP_ST_COMPLETE;
+       } else {
+               op->cancel(op);
+               op->state = FSCACHE_OP_ST_CANCELLED;
+       }
 
        if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags))
                object->n_exclusive--;
index 7db9752252ef1086c3616a9349405502f34f5003..d1b23a67c03123e16854d254077acf51766c2826 100644 (file)
@@ -213,7 +213,7 @@ int __fscache_attr_changed(struct fscache_cookie *cookie)
                return -ENOMEM;
        }
 
-       fscache_operation_init(op, fscache_attr_changed_op, NULL);
+       fscache_operation_init(op, fscache_attr_changed_op, NULL, NULL);
        op->flags = FSCACHE_OP_ASYNC |
                (1 << FSCACHE_OP_EXCLUSIVE) |
                (1 << FSCACHE_OP_UNUSE_COOKIE);
@@ -248,6 +248,17 @@ nobufs:
 }
 EXPORT_SYMBOL(__fscache_attr_changed);
 
+/*
+ * Handle cancellation of a pending retrieval op
+ */
+static void fscache_do_cancel_retrieval(struct fscache_operation *_op)
+{
+       struct fscache_retrieval *op =
+               container_of(_op, struct fscache_retrieval, op);
+
+       atomic_set(&op->n_pages, 0);
+}
+
 /*
  * release a retrieval op reference
  */
@@ -285,7 +296,9 @@ static struct fscache_retrieval *fscache_alloc_retrieval(
                return NULL;
        }
 
-       fscache_operation_init(&op->op, NULL, fscache_release_retrieval_op);
+       fscache_operation_init(&op->op, NULL,
+                              fscache_do_cancel_retrieval,
+                              fscache_release_retrieval_op);
        op->op.flags    = FSCACHE_OP_MYTHREAD |
                (1UL << FSCACHE_OP_WAITING) |
                (1UL << FSCACHE_OP_UNUSE_COOKIE);
@@ -329,25 +342,13 @@ int fscache_wait_for_deferred_lookup(struct fscache_cookie *cookie)
        return 0;
 }
 
-/*
- * Handle cancellation of a pending retrieval op
- */
-static void fscache_do_cancel_retrieval(struct fscache_operation *_op)
-{
-       struct fscache_retrieval *op =
-               container_of(_op, struct fscache_retrieval, op);
-
-       atomic_set(&op->n_pages, 0);
-}
-
 /*
  * wait for an object to become active (or dead)
  */
 int fscache_wait_for_operation_activation(struct fscache_object *object,
                                          struct fscache_operation *op,
                                          atomic_t *stat_op_waits,
-                                         atomic_t *stat_object_dead,
-                                         void (*do_cancel)(struct fscache_operation *))
+                                         atomic_t *stat_object_dead)
 {
        int ret;
 
@@ -359,7 +360,7 @@ int fscache_wait_for_operation_activation(struct fscache_object *object,
                fscache_stat(stat_op_waits);
        if (wait_on_bit(&op->flags, FSCACHE_OP_WAITING,
                        TASK_INTERRUPTIBLE) != 0) {
-               ret = fscache_cancel_op(op, do_cancel, false);
+               ret = fscache_cancel_op(op, false);
                if (ret == 0)
                        return -ERESTARTSYS;
 
@@ -380,7 +381,7 @@ check_if_dead:
        if (unlikely(fscache_object_is_dying(object) ||
                     fscache_cache_is_broken(object))) {
                enum fscache_operation_state state = op->state;
-               fscache_cancel_op(op, do_cancel, true);
+               fscache_cancel_op(op, true);
                if (stat_object_dead)
                        fscache_stat(stat_object_dead);
                _leave(" = -ENOBUFS [obj dead %d]", state);
@@ -464,8 +465,7 @@ int __fscache_read_or_alloc_page(struct fscache_cookie *cookie,
        ret = fscache_wait_for_operation_activation(
                object, &op->op,
                __fscache_stat(&fscache_n_retrieval_op_waits),
-               __fscache_stat(&fscache_n_retrievals_object_dead),
-               fscache_do_cancel_retrieval);
+               __fscache_stat(&fscache_n_retrievals_object_dead));
        if (ret < 0)
                goto error;
 
@@ -595,8 +595,7 @@ int __fscache_read_or_alloc_pages(struct fscache_cookie *cookie,
        ret = fscache_wait_for_operation_activation(
                object, &op->op,
                __fscache_stat(&fscache_n_retrieval_op_waits),
-               __fscache_stat(&fscache_n_retrievals_object_dead),
-               fscache_do_cancel_retrieval);
+               __fscache_stat(&fscache_n_retrievals_object_dead));
        if (ret < 0)
                goto error;
 
@@ -702,8 +701,7 @@ int __fscache_alloc_page(struct fscache_cookie *cookie,
        ret = fscache_wait_for_operation_activation(
                object, &op->op,
                __fscache_stat(&fscache_n_alloc_op_waits),
-               __fscache_stat(&fscache_n_allocs_object_dead),
-               fscache_do_cancel_retrieval);
+               __fscache_stat(&fscache_n_allocs_object_dead));
        if (ret < 0)
                goto error;
 
@@ -946,7 +944,7 @@ int __fscache_write_page(struct fscache_cookie *cookie,
        if (!op)
                goto nomem;
 
-       fscache_operation_init(&op->op, fscache_write_op,
+       fscache_operation_init(&op->op, fscache_write_op, NULL,
                               fscache_release_write_op);
        op->op.flags = FSCACHE_OP_ASYNC |
                (1 << FSCACHE_OP_WAITING) |
index 0e26d49972e3369ee707bf80b0656393fde681d7..69c6eb10d8589721e6bdb555233f9fde449df46f 100644 (file)
@@ -74,6 +74,7 @@ extern wait_queue_head_t fscache_cache_cleared_wq;
  */
 typedef void (*fscache_operation_release_t)(struct fscache_operation *op);
 typedef void (*fscache_operation_processor_t)(struct fscache_operation *op);
+typedef void (*fscache_operation_cancel_t)(struct fscache_operation *op);
 
 enum fscache_operation_state {
        FSCACHE_OP_ST_BLANK,            /* Op is not yet submitted */
@@ -109,6 +110,9 @@ struct fscache_operation {
         *   the op in a non-pool thread */
        fscache_operation_processor_t processor;
 
+       /* Operation cancellation cleanup (optional) */
+       fscache_operation_cancel_t cancel;
+
        /* operation releaser */
        fscache_operation_release_t release;
 };
@@ -121,6 +125,7 @@ extern void fscache_op_complete(struct fscache_operation *, bool);
 extern void fscache_put_operation(struct fscache_operation *);
 extern void fscache_operation_init(struct fscache_operation *,
                                   fscache_operation_processor_t,
+                                  fscache_operation_cancel_t,
                                   fscache_operation_release_t);
 
 /*