fuse: add --lru-limit option

The inode LRU mechanism is moot in fuse xlator (ie. there is no
limit for the LRU list), as fuse inodes are referenced from
kernel context, and thus they can only be dropped on request of
the kernel. This might results in a high number of passive
inodes which are useless for the glusterfs client, causing a
significant memory overhead.

This change tries to remedy this by extending the LRU semantics
and allowing to set a finite limit on the fuse inode LRU.

A brief history of problem:

When gluster's inode table was designed, fuse didn't have any
'invalidate' method, which means, userspace application could
never ask kernel to send a 'forget()' fop, instead had to wait
for kernel to send it based on kernel's parameters. Inode table
remembers the number of times kernel has cached the inode based
on the 'nlookup' parameter. And 'nlookup' field is not used by
no other entry points (like server-protocol, gfapi etc).

Hence the inode_table of fuse module always has to have lru-limit
as '0', which means no limit. GlusterFS always had to keep all
inodes in memory as kernel would have had a reference to it.
Again, the reason for this is, kernel's glusterfs inode reference
was pointer of 'inode_t' structure in glusterfs. As it is a
pointer, we could never free it (to prevent segfault, or memory
corruption).

Solution:

In the inode table, handle the prune case of inodes with 'nlookup'
differently, and call a 'invalidator' method, which in this case is
fuse_invalidate(), and it sends the request to kernel for getting
the forget request.

When the kernel sends the forget, it means, it has dropped all
the reference to the inode, and it will send the forget with the
'nlookup' parameter too. We just need to make sure to reduce the
'nlookup' value we have when we get forget. That automatically
cause the relevant prune to happen.

Credits: Csaba Henk, Xavier Hernandez, Raghavendra Gowdappa, Nithya B

fixes: bz#1623107
Change-Id: Ifee0737b23b12b1426c224ec5b8f591f487d83a2
Signed-off-by: Amar Tumballi <amarts@redhat.com>

3 files changed, 86 insertions, 51 deletions

diff --git a/xlators/mount/fuse/src/fuse-bridge.c b/xlators/mount/fuse/src/fuse-bridge.c
index efb390a9c54..374f20e106f 100644
--- a/xlators/mount/fuse/src/fuse-bridge.c
+++ b/xlators/mount/fuse/src/fuse-bridge.c
@@ -215,8 +215,8 @@ check_and_dump_fuse_W(fuse_private_t *priv, struct iovec *iov_out, int count,
     struct fuse_out_header *fouh = NULL;
 
     if (res == -1) {
-        gf_log("glusterfs-fuse", GF_LOG_ERROR,
-               "writing to fuse device failed: %s", strerror(errno));
+        gf_log_callingfn("glusterfs-fuse", GF_LOG_ERROR,
+                         "writing to fuse device failed: %s", strerror(errno));
         return errno;
     }
 
@@ -311,29 +311,29 @@ send_fuse_data(xlator_t *this, fuse_in_header_t *finh, void *data, size_t size)
 #define send_fuse_obj(this, finh, obj)                                         \
     send_fuse_data(this, finh, obj, sizeof(*(obj)))
 
-#if FUSE_KERNEL_MINOR_VERSION >= 11
 static void
 fuse_invalidate_entry(xlator_t *this, uint64_t fuse_ino)
 {
+#if FUSE_KERNEL_MINOR_VERSION >= 11
     struct fuse_out_header *fouh = NULL;
     struct fuse_notify_inval_entry_out *fnieo = NULL;
     fuse_private_t *priv = NULL;
     dentry_t *dentry = NULL;
+    dentry_t *tmp = NULL;
     inode_t *inode = NULL;
     size_t nlen = 0;
     fuse_invalidate_node_t *node = NULL;
+    char gfid_str[UUID_CANONICAL_FORM_LEN + 1];
 
     priv = this->private;
-
     if (!priv->reverse_fuse_thread_started)
         return;
 
-    inode = fuse_ino_to_inode(fuse_ino, this);
-    if (inode == NULL) {
+    inode = (inode_t *)(unsigned long)fuse_ino;
+    if (inode == NULL)
         return;
-    }
 
-    list_for_each_entry(dentry, &inode->dentry_list, inode_list)
+    list_for_each_entry_safe(dentry, tmp, &inode->dentry_list, inode_list)
     {
         node = GF_CALLOC(1, sizeof(*node), gf_fuse_mt_invalidate_node_t);
         if (node == NULL)
@@ -347,38 +347,41 @@ fuse_invalidate_entry(xlator_t *this, uint64_t fuse_ino)
         fouh->unique = 0;
         fouh->error = FUSE_NOTIFY_INVAL_ENTRY;
 
-        nlen = strlen(dentry->name);
-        fouh->len = sizeof(*fouh) + sizeof(*fnieo) + nlen + 1;
-        fnieo->parent = inode_to_fuse_nodeid(dentry->parent);
-
-        fnieo->namelen = nlen;
-        strcpy(node->inval_buf + sizeof(*fouh) + sizeof(*fnieo), dentry->name);
+        if (dentry->name) {
+            nlen = strlen(dentry->name);
+            fouh->len = sizeof(*fouh) + sizeof(*fnieo) + nlen + 1;
+            fnieo->parent = inode_to_fuse_nodeid(dentry->parent);
 
-        pthread_mutex_lock(&priv->invalidate_mutex);
-        {
-            list_add_tail(&node->next, &priv->invalidate_list);
-            pthread_cond_signal(&priv->invalidate_cond);
+            fnieo->namelen = nlen;
+            strcpy((node->inval_buf + sizeof(*fouh) + sizeof(*fnieo)),
+                   dentry->name);
         }
-        pthread_mutex_unlock(&priv->invalidate_mutex);
 
         gf_log("glusterfs-fuse", GF_LOG_TRACE,
-               "INVALIDATE entry: "
-               "%" PRIu64 "/%s",
-               fnieo->parent, dentry->name);
+               "INVALIDATE entry: %" PRIu64 "/%s (gfid:%s)", fnieo->parent,
+               dentry->name, uuid_utoa(inode->gfid));
 
         if (dentry->parent) {
-            fuse_log_eh(this, "Invalidated entry %s (parent: %s)", dentry->name,
-                        uuid_utoa(dentry->parent->gfid));
+            fuse_log_eh(this, "Invalidated entry %s (parent: %s) gfid:%s",
+                        dentry->name, uuid_utoa(dentry->parent->gfid),
+                        uuid_utoa_r(inode->gfid, gfid_str));
         } else {
-            fuse_log_eh(this, "Invalidated entry %s(nodeid: %" PRIu64 ")",
-                        dentry->name, fnieo->parent);
+            fuse_log_eh(this,
+                        "Invalidated entry %s(nodeid: %" PRIu64 ") gfid:%s",
+                        dentry->name, fnieo->parent, uuid_utoa(inode->gfid));
+        }
+
+        pthread_mutex_lock(&priv->invalidate_mutex);
+        {
+            list_add_tail(&node->next, &priv->invalidate_list);
+            pthread_cond_signal(&priv->invalidate_cond);
         }
+        pthread_mutex_unlock(&priv->invalidate_mutex);
     }
 
-    if (inode)
-        inode_unref(inode);
-}
 #endif
+    return;
+}
 
 /*
  * Send an inval inode notification to fuse. This causes an invalidation of the
@@ -399,6 +402,10 @@ fuse_invalidate_inode(xlator_t *this, uint64_t fuse_ino)
     if (!priv->reverse_fuse_thread_started)
         return;
 
+    inode = (inode_t *)(unsigned long)fuse_ino;
+    if (inode == NULL)
+        return;
+
     node = GF_CALLOC(1, sizeof(*node), gf_fuse_mt_invalidate_node_t);
     if (node == NULL)
         return;
@@ -418,7 +425,11 @@ fuse_invalidate_inode(xlator_t *this, uint64_t fuse_ino)
     fniio->off = 0;
     fniio->len = -1;
 
-    inode = fuse_ino_to_inode(fuse_ino, this);
+    fuse_log_eh(this, "Invalidated inode %" PRIu64 " (gfid: %s)", fuse_ino,
+                uuid_utoa(inode->gfid));
+    gf_log("glusterfs-fuse", GF_LOG_TRACE,
+           "INVALIDATE inode: %" PRIu64 "(gfid:%s)", fuse_ino,
+           uuid_utoa(inode->gfid));
 
     pthread_mutex_lock(&priv->invalidate_mutex);
     {
@@ -427,24 +438,22 @@ fuse_invalidate_inode(xlator_t *this, uint64_t fuse_ino)
     }
     pthread_mutex_unlock(&priv->invalidate_mutex);
 
-    gf_log("glusterfs-fuse", GF_LOG_TRACE, "INVALIDATE inode: %" PRIu64,
-           fuse_ino);
-
-    if (inode) {
-        fuse_log_eh(this, "Invalidated inode %" PRIu64 " (gfid: %s)", fuse_ino,
-                    uuid_utoa(inode->gfid));
-    } else {
-        fuse_log_eh(this, "Invalidated inode %" PRIu64, fuse_ino);
-    }
-
-    if (inode)
-        inode_unref(inode);
 #else
     gf_log("glusterfs-fuse", GF_LOG_WARNING,
-           "fuse_invalidate_inode not implemented on OS X due to missing FUSE "
-           "notification");
+           "fuse_invalidate_inode not implemented on this system");
 #endif
+    return;
+}
+
+#if FUSE_KERNEL_MINOR_VERSION >= 11
+/* Need this function for the signature (inode_t *, instead of uint64_t) */
+static int32_t
+fuse_inode_invalidate_fn(xlator_t *this, inode_t *inode)
+{
+    fuse_invalidate_entry(this, (uint64_t)inode);
+    return 0;
 }
+#endif
 
 int
 send_fuse_err(xlator_t *this, fuse_in_header_t *finh, int error)
@@ -707,11 +716,14 @@ do_forget(xlator_t *this, uint64_t unique, uint64_t nodeid, uint64_t nlookup)
 {
     inode_t *fuse_inode = fuse_ino_to_inode(nodeid, this);
 
+    gf_log("fuse", GF_LOG_TRACE,
+           "%" PRIu64 ": FORGET %" PRIu64 "/%" PRIu64 " gfid: (%s)", unique,
+           nodeid, nlookup, uuid_utoa(fuse_inode->gfid));
+
     fuse_log_eh(this, "%" PRIu64 ": FORGET %" PRIu64 "/%" PRIu64 " gfid: (%s)",
                 unique, nodeid, nlookup, uuid_utoa(fuse_inode->gfid));
 
-    inode_forget(fuse_inode, nlookup);
-    inode_unref(fuse_inode);
+    inode_forget_with_unref(fuse_inode, nlookup);
 }
 
 static void
@@ -726,10 +738,6 @@ fuse_forget(xlator_t *this, fuse_in_header_t *finh, void *msg,
         return;
     }
 
-    gf_log("glusterfs-fuse", GF_LOG_TRACE,
-           "%" PRIu64 ": FORGET %" PRIu64 "/%" PRIu64, finh->unique,
-           finh->nodeid, ffi->nlookup);
-
     do_forget(this, finh->unique, finh->nodeid, ffi->nlookup);
 
     GF_FREE(finh);
@@ -4926,7 +4934,9 @@ fuse_thread_proc(void *data)
     fuse_in_header_t *finh = NULL;
     struct iovec iov_in[2];
     void *msg = NULL;
-    const size_t msg0_size = sizeof(*finh) + 128;
+    /* we need 512 extra buffer size for BATCH_FORGET fop. By tests, it is
+       found to be reduces 'REALLOC()' in the loop */
+    const size_t msg0_size = sizeof(*finh) + 512;
     fuse_handler_t **fuse_ops = NULL;
     struct pollfd pfd[2] = {{
         0,
@@ -5258,7 +5268,12 @@ fuse_graph_setup(xlator_t *this, glusterfs_graph_t *graph)
             goto unlock;
         }
 
+#if FUSE_KERNEL_MINOR_VERSION >= 11
+        itable = inode_table_with_invalidator(priv->lru_limit, graph->top,
+                                              fuse_inode_invalidate_fn, this);
+#else
         itable = inode_table_new(0, graph->top);
+#endif
         if (!itable) {
             ret = -1;
             goto unlock;
@@ -5708,6 +5723,8 @@ init(xlator_t *this_xl)
         }
     }
 
+    GF_OPTION_INIT("lru-limit", priv->lru_limit, uint32, cleanup_exit);
+
     GF_OPTION_INIT("event-history", priv->event_history, bool, cleanup_exit);
 
     GF_OPTION_INIT("thin-client", priv->thin_client, bool, cleanup_exit);
@@ -6025,5 +6042,13 @@ struct volume_options options[] = {
         .max = 1000000000,
         .description = "Supported granularity of file attribute times.",
     },
+    {
+        .key = {"lru-limit"},
+        .type = GF_OPTION_TYPE_INT,
+        .default_value = "131072",
+        .min = 0,
+        .description = "makes glusterfs invalidate kernel inodes after "
+                       "reaching this limit (0 means 'unlimited')",
+    },
     {.key = {NULL}},
 };
diff --git a/xlators/mount/fuse/src/fuse-bridge.h b/xlators/mount/fuse/src/fuse-bridge.h
index 318f33b5d61..84022ca1ecc 100644
--- a/xlators/mount/fuse/src/fuse-bridge.h
+++ b/xlators/mount/fuse/src/fuse-bridge.h
@@ -151,6 +151,9 @@ struct fuse_private {
     /* Writeback cache support */
     gf_boolean_t kernel_writeback_cache;
     int attr_times_granularity;
+
+    /* LRU Limit, if not set, default is 128k for now */
+    uint32_t lru_limit;
 };
 typedef struct fuse_private fuse_private_t;
 
diff --git a/xlators/mount/fuse/utils/mount.glusterfs.in b/xlators/mount/fuse/utils/mount.glusterfs.in
index 959f2272e2f..ceeeb100952 100755
--- a/xlators/mount/fuse/utils/mount.glusterfs.in
+++ b/xlators/mount/fuse/utils/mount.glusterfs.in
@@ -249,6 +249,10 @@ start_glusterfs ()
         cmd_line=$(echo "$cmd_line --gid-timeout=$gid_timeout");
     fi
 
+    if [ -n "$lru_limit" ]; then
+        cmd_line=$(echo "$cmd_line --lru-limit=$lru_limit");
+    fi
+
     if [ -n "$bg_qlen" ]; then
         cmd_line=$(echo "$cmd_line --background-qlen=$bg_qlen");
     fi
@@ -485,6 +489,9 @@ with_options()
         "gid-timeout")
             gid_timeout=$value
             ;;
+        "lru-limit")
+            lru_limit=$value
+            ;;
         "background-qlen")
             bg_qlen=$value
             ;;