LU-10239 osc: limit chunk number of write submit

[fs/lustre-release.git] / lustre / osc / osc_internal.h

diff --git a/lustre/osc/osc_internal.h b/lustre/osc/osc_internal.h
index 20aadc8..62b5c2b 100644 (file)
--- a/lustre/osc/osc_internal.h
+++ b/lustre/osc/osc_internal.h
@@ -23,7 +23,7 @@
  * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2016, Intel Corporation.
+ * Copyright (c) 2011, 2017, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -35,6 +35,7 @@
 
 #define OAP_MAGIC 8675309
 
+#include <libcfs/linux/linux-mem.h>
 #include <lustre_osc.h>
 
 extern atomic_t osc_pool_req_count;
@@ -89,12 +90,7 @@ unsigned long osc_ldlm_weigh_ast(struct ldlm_lock *dlmlock);
 int osc_cleanup(struct obd_device *obd);
 int osc_setup(struct obd_device *obd, struct lustre_cfg *lcfg);
 
-#ifdef CONFIG_PROC_FS
-extern struct lprocfs_vars lprocfs_osc_obd_vars[];
-int lproc_osc_attach_seqstat(struct obd_device *dev);
-#else
-static inline int lproc_osc_attach_seqstat(struct obd_device *dev) {return 0;}
-#endif
+int osc_tunables_init(struct obd_device *obd);
 
 extern struct lu_device_type osc_device_type;
 
@@ -159,6 +155,8 @@ int osc_quotactl(struct obd_device *unused, struct obd_export *exp,
 void osc_inc_unstable_pages(struct ptlrpc_request *req);
 void osc_dec_unstable_pages(struct ptlrpc_request *req);
 bool osc_over_unstable_soft_limit(struct client_obd *cli);
+void osc_page_touch_at(const struct lu_env *env, struct cl_object *obj,
+                      pgoff_t idx, size_t to);
 
 struct ldlm_lock *osc_obj_dlmlock_at_pgoff(const struct lu_env *env,
                                           struct osc_object *obj,
@@ -175,6 +173,35 @@ extern unsigned long osc_cache_shrink_count(struct shrinker *sk,
                                            struct shrink_control *sc);
 extern unsigned long osc_cache_shrink_scan(struct shrinker *sk,
                                           struct shrink_control *sc);
+static inline unsigned int osc_max_write_chunks(const struct client_obd *cli)
+{
+       /*
+        * LU-8135:
+        *
+        * The maximum size of a single transaction is about 64MB in ZFS.
+        * #define DMU_MAX_ACCESS (64 * 1024 * 1024)
+        *
+        * Since ZFS is a copy-on-write file system, a single dirty page in
+        * a chunk will result in the rewrite of the whole chunk, therefore
+        * an RPC shouldn't be allowed to contain too many chunks otherwise
+        * it will make transaction size much bigger than 64MB, especially
+        * with big block size for ZFS.
+        *
+        * This piece of code is to make sure that OSC won't send write RPCs
+        * with too many chunks. The maximum chunk size that an RPC can cover
+        * is set to PTLRPC_MAX_BRW_SIZE, which is defined to 16MB. Ideally
+        * OST should tell the client what the biggest transaction size is,
+        * but it's good enough for now.
+        *
+        * This limitation doesn't apply to ldiskfs, which allows as many
+        * chunks in one RPC as we want. However, it won't have any benefits
+        * to have too many discontiguous pages in one RPC.
+        *
+        * An osc_extent won't cover over a RPC size, so the chunks in an
+        * osc_extent won't bigger than PTLRPC_MAX_BRW_SIZE >> chunkbits.
+        */
+       return PTLRPC_MAX_BRW_SIZE >> cli->cl_chunkbits;
+}
 
 static inline void osc_set_io_portal(struct ptlrpc_request *req)
 {