performance/stat-prefetch: Add support code to implement lookup-behind.

- change sp_lookup and sp_lookup_cbk to support sending lookups when needed.
    lookups might need to be sent in fops like open, chmod, chown etc which
    operate on a path, since the actual lookup sent to stat-prefetch by its
    parent xlators is not propagated down the xlator tree if the path is
    cached.

Signed-off-by: Anand V. Avati <avati@dev.gluster.com>

BUG: 221 (stat prefetch implementation)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=221

2 files changed, 199 insertions, 55 deletions

diff --git a/xlators/performance/stat-prefetch/src/stat-prefetch.c b/xlators/performance/stat-prefetch/src/stat-prefetch.c
index 883345a2d3b..b96e92c6e2c 100644
--- a/xlators/performance/stat-prefetch/src/stat-prefetch.c
+++ b/xlators/performance/stat-prefetch/src/stat-prefetch.c
@@ -18,9 +18,6 @@
 */
 
 #include "stat-prefetch.h"
-#include "locking.h"
-#include "inode.h"
-#include <libgen.h>
 
 #define GF_SP_CACHE_BUCKETS 4096
 
@@ -374,53 +371,123 @@ sp_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                int32_t op_ret, int32_t op_errno, inode_t *inode,
                struct stat *buf, dict_t *dict, struct stat *postparent)
 {
-        struct stat *stbuf = NULL;
-        int32_t      ret   = -1;
-        uint64_t     value = 0;
+        sp_inode_ctx_t      *inode_ctx   = NULL;
+        uint64_t             value       = 0;
+        int                  ret         = 0;
+        struct list_head     waiting_ops = {0, };
+        call_stub_t         *stub        = NULL, *tmp = NULL;
+        sp_local_t          *local       = NULL;
+        sp_cache_t          *cache       = NULL;
 
-        if (op_ret == -1) {
-                goto out;
+        INIT_LIST_HEAD (&waiting_ops);
+
+        local = frame->local;
+        if (local == NULL) {
+                op_ret = -1;
+                op_errno = EINVAL;
+        } else if (op_ret == -1) {
+                cache = sp_get_cache_inode (this, local->loc.parent,
+                                            frame->root->pid);
+
+                if (cache) {
+                        sp_cache_remove_entry (cache, (char *)local->loc.name,
+                                               0);
+                }
         }
 
-        if (S_ISDIR (buf->st_mode)) {
-                LOCK (&inode->lock);
+        ret = inode_ctx_get (inode, this, &value);
+        if (ret == 0) {
+                inode_ctx = (sp_inode_ctx_t *)(long)value; 
+                if (inode_ctx == NULL) {
+                        op_ret = -1;
+                        op_errno = EINVAL;
+                        goto out;
+                }
+
+                LOCK (&inode_ctx->lock);
                 {
-                        ret = __inode_ctx_get (inode, this, &value);
-                        if (ret == 0) {
-                                stbuf = (struct stat *)(long)value;
+                        inode_ctx->op_ret = op_ret;
+                        inode_ctx->op_errno = op_errno;
+                        inode_ctx->looked_up = 1;
+                        inode_ctx->lookup_in_progress = 0;
+                        if (S_ISDIR (buf->st_mode) && (op_ret == 0)) {
+                                memcpy (&inode_ctx->stbuf, buf,
+                                        sizeof (*buf));
                         }
+                        list_splice_init (&inode_ctx->waiting_ops,
+                                          &waiting_ops);
+                }
+                UNLOCK (&inode_ctx->lock);
 
-                        if (stbuf == NULL) {
-                                stbuf = CALLOC (1, sizeof (*stbuf));
-                                if (stbuf == NULL) {
-                                        op_ret = -1;
-                                        op_errno = ENOMEM;
-                                        gf_log (this->name, GF_LOG_ERROR,
-                                                "out of memory");
-                                        goto unlock;
-                                }
+                list_for_each_entry_safe (stub, tmp, &waiting_ops, list) {
+                        list_del_init (&stub->list);
+                        call_resume (stub);
+                }
+        } else {
+                op_errno = EINVAL;
+                op_ret = -1;
+        }
 
-                                ret = __inode_ctx_put (inode, this,
-                                                       (long)stbuf);
-                                if (ret == -1) {
-                                        op_ret = -1;
-                                
-                                        /* FIXME: EINVAL is not correct */ 
-                                        op_errno = EINVAL;
-                                        FREE (stbuf);
-                                        goto unlock;
-                                }
+out:
+        if ((local != NULL) && (local->is_lookup)) {
+                SP_STACK_UNWIND (frame, op_ret, op_errno, inode, buf, dict);
+        }
+
+        return 0;
+}
+
+void
+sp_inode_ctx_free (xlator_t *this, sp_inode_ctx_t *ctx)
+{
+        call_stub_t *stub = NULL, *tmp = NULL;
+
+        if (ctx == NULL) {
+                goto out;
+        }
+
+        LOCK (&ctx->lock);
+        {
+                if (!list_empty (&ctx->waiting_ops)) {
+                        gf_log (this->name, GF_LOG_CRITICAL, "inode ctx is "
+                                "being freed even when there are file "
+                                "operations waiting for lookup-behind to "
+                                "complete. The operations in the waiting list "
+                                "are:");
+                        list_for_each_entry_safe (stub, tmp, &ctx->waiting_ops,
+                                                  list) {
+                                gf_log (this->name, GF_LOG_CRITICAL,
+                                        "OP (%d)", stub->fop);
+
+                                list_del_init (&stub->list);
+                                call_stub_destroy (stub);
                         }
-                        
-                        memcpy (stbuf, buf, sizeof (*stbuf));
                 }
-        unlock:
-                UNLOCK (&inode->lock);
         }
+        UNLOCK (&ctx->lock);
+
+        LOCK_DESTROY (&ctx->lock);
+        FREE (ctx);
 
 out:
-	SP_STACK_UNWIND (frame, op_ret, op_errno, inode, buf, dict);
-        return 0;
+        return;
+}
+
+
+sp_inode_ctx_t *
+sp_inode_ctx_init ()
+{
+        sp_inode_ctx_t *inode_ctx = NULL;
+
+        inode_ctx = CALLOC (1, sizeof (*inode_ctx));
+        if (inode_ctx == NULL) {
+                goto out;
+        }
+
+        LOCK_INIT (&inode_ctx->lock);
+        INIT_LIST_HEAD (&inode_ctx->waiting_ops);
+
+out:
+        return inode_ctx;
 }
 
 
@@ -520,19 +587,48 @@ sp_is_empty (dict_t *this, char *key, data_t *value, void *data)
         }
 }
 
-
+/* 
+ * TODO: implement sending lookups for every fop done on this path. As of now
+ * lookup on the path is sent only for the first fop on this path.
+ */
 int32_t
 sp_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
 {
-        gf_dirent_t   dirent;
-        char          entry_cached    = 0; 
-        uint64_t      value           = 0;
-        char          xattr_req_empty = 1;
-        sp_cache_t   *cache           = NULL;
-        struct stat  *postparent      = NULL, *buf = NULL;
-        int32_t       ret             = -1, op_ret = -1, op_errno = EINVAL; 
+        gf_dirent_t     dirent;
+        char            entry_cached    = 0; 
+        uint64_t        value           = 0;
+        char            xattr_req_empty = 1;
+        sp_cache_t     *cache           = NULL;
+        struct stat     postparent      = {0, }, buf = {0, };
+        int32_t         ret             = -1, op_ret = -1, op_errno = EINVAL; 
+        sp_inode_ctx_t *inode_ctx       = NULL;
+        sp_local_t     *local           = NULL;
+
+        if (loc == NULL || loc->inode == NULL) {
+                goto unwind;
+        }
+
+        LOCK (&loc->inode->lock);
+        {
+                ret = __inode_ctx_get (loc->inode, this, &value);
+                if (ret == 0) {
+                        inode_ctx = (sp_inode_ctx_t *)(long)value;
+                } else {
+                        inode_ctx = sp_inode_ctx_init ();
+                        if (inode_ctx != NULL) {
+                                ret = __inode_ctx_put (loc->inode, this,
+                                                       (long)inode_ctx);
+                                if (ret == -1) {
+                                        sp_inode_ctx_free (this, inode_ctx);
+                                        inode_ctx = NULL;
+                                }
+                        }
+                }
+        }
+        UNLOCK (&loc->inode->lock);
 
-        if (loc == NULL) {
+        if (inode_ctx == NULL) {
+                op_errno = ENOMEM;
                 goto unwind;
         }
 
@@ -554,9 +650,10 @@ sp_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
                 ret = sp_cache_get_entry (cache, (char *)loc->name, &dirent);
                 if (ret == 0) {
                         ret = inode_ctx_get (loc->parent, this, &value);
-                        if (ret == 0) {
-                                postparent = (void *)(long)value;
-                                buf = &dirent.d_stat;
+                        if ((ret == 0) && (value != 0)) {
+                                inode_ctx = (void *)(long)value;
+                                postparent = inode_ctx->stbuf;
+                                buf = dirent.d_stat;
                                 op_ret = 0;
                                 op_errno = 0;
                                 entry_cached = 1;
@@ -567,10 +664,15 @@ sp_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
                 if (cache) {
                         ret = sp_cache_get_entry (cache, ".", &dirent);
                         if (ret == 0) {
-                                buf = &dirent.d_stat;
-                                op_ret = 0;
-                                op_errno = 0;
-                                entry_cached = 1;
+                                ret = inode_ctx_get (loc->parent, this, &value);
+                                if ((ret == 0) && (value != 0)) {
+                                        inode_ctx = (void *)(long)value;
+                                        postparent = inode_ctx->stbuf;
+                                        buf = dirent.d_stat;
+                                        op_ret = 0;
+                                        op_errno = 0;
+                                        entry_cached = 1;
+                                }
                         }
                 }
         }
@@ -585,6 +687,33 @@ wind:
                         cache->miss++;
                 }
 
+                LOCK (&inode_ctx->lock);
+                {
+                        if (!(inode_ctx->looked_up
+                              || inode_ctx->lookup_in_progress)) {
+                                inode_ctx->lookup_in_progress = 1;
+                        }
+                }
+                UNLOCK (&inode_ctx->lock);
+
+                op_errno = ENOMEM;  
+                local = CALLOC (1, sizeof (*local));
+                GF_VALIDATE_OR_GOTO_WITH_ERROR (this->name, local, unwind,
+                                                ENOMEM);
+                op_errno = 0;
+                
+                frame->local = local;
+
+                ret = loc_copy (&local->loc, loc);
+                if (ret == -1) {
+                        op_errno = errno;
+                        gf_log (this->name, GF_LOG_ERROR, "%s",
+                                strerror (op_errno));
+                        goto unwind;
+                }
+
+                local->is_lookup = 1;
+
                 STACK_WIND (frame, sp_lookup_cbk, FIRST_CHILD(this),
                             FIRST_CHILD(this)->fops->lookup, loc, xattr_req);
      
@@ -592,7 +721,7 @@ wind:
         }
 
 unwind:
-	SP_STACK_UNWIND (frame, op_ret, op_errno, loc->inode, buf, postparent,
+	SP_STACK_UNWIND (frame, op_ret, op_errno, loc->inode, &buf, &postparent,
                          NULL);
         return 0;
 
diff --git a/xlators/performance/stat-prefetch/src/stat-prefetch.h b/xlators/performance/stat-prefetch/src/stat-prefetch.h
index dd6b067c5bd..81f504a7f31 100644
--- a/xlators/performance/stat-prefetch/src/stat-prefetch.h
+++ b/xlators/performance/stat-prefetch/src/stat-prefetch.h
@@ -25,11 +25,15 @@
 #include "config.h"
 #endif
 
+#include "locking.h"
+#include "inode.h"
 #include "glusterfs.h"
 #include "dict.h"
 #include "xlator.h"
 #include "rbthash.h"
 #include "hashfn.h"
+#include "call-stub.h"
+#include <libgen.h>
 
 struct sp_cache {
         rbthash_table_t *table;
@@ -58,9 +62,20 @@ typedef struct sp_fd_ctx sp_fd_ctx_t;
 struct sp_local {
         loc_t  loc;
         fd_t  *fd;
+        char   is_lookup;
 };
 typedef struct sp_local sp_local_t;
 
+struct sp_inode_ctx {
+        char             looked_up;
+        char             lookup_in_progress;
+        int32_t          op_ret;
+        int32_t          op_errno;
+        struct stat      stbuf;  
+        gf_lock_t        lock;
+        struct list_head waiting_ops;
+};
+typedef struct sp_inode_ctx sp_inode_ctx_t;
 
 void sp_local_free (sp_local_t *local);