cachefiles: introduce object ondemand state
authorJia Zhu <zhujia.zj@bytedance.com>
Mon, 20 Nov 2023 04:14:18 +0000 (12:14 +0800)
committerChristian Brauner <brauner@kernel.org>
Sat, 25 Nov 2023 15:03:54 +0000 (16:03 +0100)
Previously, @ondemand_id field was used not only to identify ondemand
state of the object, but also to represent the index of the xarray.
This commit introduces @state field to decouple the role of @ondemand_id
and adds helpers to access it.

Signed-off-by: Jia Zhu <zhujia.zj@bytedance.com>
Link: https://lore.kernel.org/r/20231120041422.75170-2-zhujia.zj@bytedance.com
Reviewed-by: Jingbo Xu <jefflexu@linux.alibaba.com>
Reviewed-by: David Howells <dhowells@redhat.com>
Signed-off-by: Christian Brauner <brauner@kernel.org>
fs/cachefiles/internal.h
fs/cachefiles/ondemand.c

index 2ad58c465208480e42106393ae01fd2f317389ee..00beedeaec1837a04db7265d3985a6f4d1196b74 100644 (file)
@@ -44,6 +44,11 @@ struct cachefiles_volume {
        struct dentry                   *fanout[256];   /* Fanout subdirs */
 };
 
+enum cachefiles_object_state {
+       CACHEFILES_ONDEMAND_OBJSTATE_CLOSE, /* Anonymous fd closed by daemon or initial state */
+       CACHEFILES_ONDEMAND_OBJSTATE_OPEN, /* Anonymous fd associated with object is available */
+};
+
 /*
  * Backing file state.
  */
@@ -62,6 +67,7 @@ struct cachefiles_object {
 #define CACHEFILES_OBJECT_USING_TMPFILE        0               /* Have an unlinked tmpfile */
 #ifdef CONFIG_CACHEFILES_ONDEMAND
        int                             ondemand_id;
+       enum cachefiles_object_state    state;
 #endif
 };
 
@@ -296,6 +302,21 @@ extern void cachefiles_ondemand_clean_object(struct cachefiles_object *object);
 extern int cachefiles_ondemand_read(struct cachefiles_object *object,
                                    loff_t pos, size_t len);
 
+#define CACHEFILES_OBJECT_STATE_FUNCS(_state, _STATE)  \
+static inline bool                                                             \
+cachefiles_ondemand_object_is_##_state(const struct cachefiles_object *object) \
+{                                                                                              \
+       return object->state == CACHEFILES_ONDEMAND_OBJSTATE_##_STATE; \
+}                                                                                              \
+                                                                                               \
+static inline void                                                             \
+cachefiles_ondemand_set_object_##_state(struct cachefiles_object *object) \
+{                                                                                              \
+       object->state = CACHEFILES_ONDEMAND_OBJSTATE_##_STATE; \
+}
+
+CACHEFILES_OBJECT_STATE_FUNCS(open, OPEN);
+CACHEFILES_OBJECT_STATE_FUNCS(close, CLOSE);
 #else
 static inline ssize_t cachefiles_ondemand_daemon_read(struct cachefiles_cache *cache,
                                        char __user *_buffer, size_t buflen)
index 0254ed39f68ceb75dcb2ec840af88b136ac21fe3..90456b8a4b3e09ee3a36fea3ca13ed188baca559 100644 (file)
@@ -15,6 +15,7 @@ static int cachefiles_ondemand_fd_release(struct inode *inode,
 
        xa_lock(&cache->reqs);
        object->ondemand_id = CACHEFILES_ONDEMAND_ID_CLOSED;
+       cachefiles_ondemand_set_object_close(object);
 
        /*
         * Flush all pending READ requests since their completion depends on
@@ -176,6 +177,8 @@ int cachefiles_ondemand_copen(struct cachefiles_cache *cache, char *args)
                set_bit(FSCACHE_COOKIE_NO_DATA_TO_READ, &cookie->flags);
        trace_cachefiles_ondemand_copen(req->object, id, size);
 
+       cachefiles_ondemand_set_object_open(req->object);
+
 out:
        complete(&req->done);
        return ret;
@@ -363,7 +366,8 @@ static int cachefiles_ondemand_send_req(struct cachefiles_object *object,
                /* coupled with the barrier in cachefiles_flush_reqs() */
                smp_mb();
 
-               if (opcode != CACHEFILES_OP_OPEN && object->ondemand_id <= 0) {
+               if (opcode != CACHEFILES_OP_OPEN &&
+                       !cachefiles_ondemand_object_is_open(object)) {
                        WARN_ON_ONCE(object->ondemand_id == 0);
                        xas_unlock(&xas);
                        ret = -EIO;
@@ -430,18 +434,11 @@ static int cachefiles_ondemand_init_close_req(struct cachefiles_req *req,
                                              void *private)
 {
        struct cachefiles_object *object = req->object;
-       int object_id = object->ondemand_id;
 
-       /*
-        * It's possible that object id is still 0 if the cookie looking up
-        * phase failed before OPEN request has ever been sent. Also avoid
-        * sending CLOSE request for CACHEFILES_ONDEMAND_ID_CLOSED, which means
-        * anon_fd has already been closed.
-        */
-       if (object_id <= 0)
+       if (!cachefiles_ondemand_object_is_open(object))
                return -ENOENT;
 
-       req->msg.object_id = object_id;
+       req->msg.object_id = object->ondemand_id;
        trace_cachefiles_ondemand_close(object, &req->msg);
        return 0;
 }
@@ -460,7 +457,7 @@ static int cachefiles_ondemand_init_read_req(struct cachefiles_req *req,
        int object_id = object->ondemand_id;
 
        /* Stop enqueuing requests when daemon has closed anon_fd. */
-       if (object_id <= 0) {
+       if (!cachefiles_ondemand_object_is_open(object)) {
                WARN_ON_ONCE(object_id == 0);
                pr_info_once("READ: anonymous fd closed prematurely.\n");
                return -EIO;
@@ -485,7 +482,7 @@ int cachefiles_ondemand_init_object(struct cachefiles_object *object)
         * creating a new tmpfile as the cache file. Reuse the previously
         * allocated object ID if any.
         */
-       if (object->ondemand_id > 0)
+       if (cachefiles_ondemand_object_is_open(object))
                return 0;
 
        volume_key_size = volume->key[0] + 1;