Whamcloud - gitweb
Branch: b1_4
authoradilger <adilger>
Mon, 25 Oct 2004 06:27:28 +0000 (06:27 +0000)
committeradilger <adilger>
Mon, 25 Oct 2004 06:27:28 +0000 (06:27 +0000)
- reorganization of lov code
- fixups for compiling on 2.6

21 files changed:
lustre/ChangeLog
lustre/configure.in
lustre/include/linux/lustre_net.h
lustre/include/linux/lvfs_linux.h
lustre/include/linux/obd_lov.h
lustre/include/lustre/lustre_user.h
lustre/kernel_patches/patches/export-zap-page-range.patch [new file with mode: 0644]
lustre/liblustre/rw.c
lustre/llite/file.c
lustre/llite/llite_lib.c
lustre/lov/Makefile.in
lustre/lov/Makefile.mk
lustre/lov/autoMakefile.am
lustre/lov/lov_internal.h
lustre/lov/lov_merge.c [new file with mode: 0644]
lustre/lov/lov_obd.c
lustre/lov/lov_offset.c [new file with mode: 0644]
lustre/lov/lov_pack.c
lustre/lov/lov_qos.c [new file with mode: 0644]
lustre/lov/lov_request.c [new file with mode: 0644]
lustre/ptlrpc/client.c

index 642f790..7aed4ee 100644 (file)
@@ -1,4 +1,10 @@
 tbd         Cluster File Systems, Inc. <info@clusterfs.com>
+       * version 1.3.4
+       * bug fixes
+       * miscellania
+       - reorganization of lov code
+
+tbd         Cluster File Systems, Inc. <info@clusterfs.com>
        * version 1.2.8
        * bug fixes
        - allocate qswnal tx descriptors singly to avoid fragmentation (4504)
index 5fbcd65..bcf65df 100644 (file)
@@ -5,7 +5,7 @@
 
 AC_INIT
 AC_CANONICAL_SYSTEM
-AM_INIT_AUTOMAKE(lustre, 1.2.7.4)
+AM_INIT_AUTOMAKE(lustre, 1.3.4)
 # AM_MAINTAINER_MODE
 
 # Four main targets: lustre kernel modules, utilities, tests, and liblustre
index 3b6617d..7e612eb 100644 (file)
@@ -201,7 +201,7 @@ struct ptlrpc_request_set {
         wait_queue_head_t *set_wakeup_ptr;
         struct list_head  set_requests;
         set_interpreter_func    set_interpret; /* completion callback */
-        union ptlrpc_async_args set_args; /* completion context */
+        void             *set_arg; /* completion context */
         /* locked so that any old caller can communicate requests to
          * the set holder who can then fold them into the lock-free set */
         spinlock_t        set_new_req_lock;
index 71fc431..cf73d47 100644 (file)
@@ -8,11 +8,8 @@
 #include <linux/module.h>
 #include <linux/fs.h>
 #include <linux/sched.h>
-#include <linux/jbd.h>
 
 #include <linux/lvfs.h>
-/* we have made EXT3_IOC_SETFLAGS a Lustre constant */
-#include <linux/ext3_fs.h>
 
 #define l_file file
 #define l_dentry dentry
index cf3ccec..9692a9b 100644 (file)
@@ -7,21 +7,6 @@
 
 #define OBD_LOV_DEVICENAME "lov"
 
-struct lov_brw_async_args {
-        struct lov_stripe_md  *aa_lsm;
-        struct obdo           *aa_obdos;
-        struct obdo           *aa_oa;
-        struct brw_page       *aa_ioarr;
-        obd_count              aa_oa_bufs;
-};
-
-struct lov_getattr_async_args {
-        struct lov_stripe_md  *aa_lsm;
-        struct obdo           *aa_oa;
-        struct obdo           *aa_obdos;
-        struct lov_obd        *aa_lov;
-};
-
 static inline int lov_stripe_md_size(int stripes)
 {
         return sizeof(struct lov_stripe_md) + stripes*sizeof(struct lov_oinfo);
index 38ea2f2..0286b49 100644 (file)
 /* for statfs() */
 #define LL_SUPER_MAGIC 0x0BD00BD0
 
+#ifndef EXT3_IOC_GETFLAGS
+#define EXT3_IOC_GETFLAGS               _IOR('f', 1, long)
+#define EXT3_IOC_SETFLAGS               _IOW('f', 2, long)
+#define EXT3_IOC_GETVERSION             _IOR('f', 3, long)
+#define EXT3_IOC_SETVERSION             _IOW('f', 4, long)
+#define EXT3_IOC_GETVERSION_OLD         _IOR('v', 1, long)
+#define EXT3_IOC_SETVERSION_OLD         _IOW('v', 2, long)
+#endif
+
 #define LL_IOC_GETFLAGS                 _IOR ('f', 151, long)
 #define LL_IOC_SETFLAGS                 _IOW ('f', 152, long)
 #define LL_IOC_CLRFLAGS                 _IOW ('f', 153, long)
diff --git a/lustre/kernel_patches/patches/export-zap-page-range.patch b/lustre/kernel_patches/patches/export-zap-page-range.patch
new file mode 100644 (file)
index 0000000..9b9d48f
--- /dev/null
@@ -0,0 +1,12 @@
+Index: linux-2.4.24-l36mmap/mm/memory.c
+===================================================================
+--- linux-2.4.24-l36mmap.orig/mm/memory.c      2004-05-27 17:44:13.000000000 -0700
++++ linux-2.4.24-l36mmap/mm/memory.c   2004-05-27 17:45:07.000000000 -0700
+@@ -411,6 +411,7 @@
+               mm->rss = 0;
+       spin_unlock(&mm->page_table_lock);
+ }
++EXPORT_SYMBOL_GPL(zap_page_range);
+ /*
+  * Do a quick page-table lookup for a single page. 
index 3104ff0..43e75c5 100644 (file)
@@ -241,8 +241,6 @@ int llu_extent_lock(struct ll_file_data *fd, struct inode *inode,
             policy->l_extent.end == OBD_OBJECT_EOF)
                 lli->lli_st_size = lov_merge_size(lsm, 1);
 
-        //inode->i_mtime = lov_merge_mtime(lsm, inode->i_mtime);
-
         RETURN(rc);
 }
 
index ef1f0a5..b0e16c3 100644 (file)
@@ -361,7 +361,7 @@ void ll_pgcache_remove_extent(struct inode *inode, struct lov_stripe_md *lsm,
         if (end < tmpex.l_extent.end >> PAGE_CACHE_SHIFT)
                 end = ~0;
 
-        i = (inode->i_size + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
+        i = inode->i_size ? (inode->i_size - 1) >> PAGE_CACHE_SHIFT : 0;
         if (i < end)
                 end = i;
 
@@ -650,7 +650,8 @@ int ll_glimpse_size(struct inode *inode)
 
         inode->i_size = lov_merge_size(lli->lli_smd, 0);
         inode->i_blocks = lov_merge_blocks(lli->lli_smd);
-        inode->i_mtime = lov_merge_mtime(lli->lli_smd, inode->i_mtime);
+        LTIME_S(inode->i_mtime) = lov_merge_mtime(lli->lli_smd,
+                                                  LTIME_S(inode->i_mtime));
 
         CDEBUG(D_DLMTRACE, "glimpse: size: %llu, blocks: %lu\n",
                inode->i_size, inode->i_blocks);
@@ -700,8 +701,9 @@ int ll_extent_lock(struct ll_file_data *fd, struct inode *inode,
                 inode->i_size = lov_merge_size(lsm, 1);
                 up(&inode->i_sem);
         }
-
-        //inode->i_mtime = lov_merge_mtime(lsm, inode->i_mtime);
+        if (rc > 0)
+                LTIME_S(inode->i_mtime) =
+                        lov_merge_mtime(lsm, LTIME_S(inode->i_mtime));
 
         RETURN(rc);
 }
index b61b721..57a5a2d 100644 (file)
@@ -150,12 +150,6 @@ int lustre_common_fill_super(struct super_block *sb, char *mdc, char *osc)
         }
         sbi->ll_osc_exp = class_conn2export(&osc_conn);
 
-        /* need to do a statfs to initialize the per-OSC osfs cache -
-         * that is used by the OSC IO code to know the blocksize */
-        err = obd_statfs(obd, &osfs, jiffies - HZ);
-        if (err)
-                GOTO(out_mdc, err);
-
         err = mdc_getstatus(sbi->ll_mdc_exp, &rootfid);
         if (err) {
                 CERROR("cannot mds_connect: rc = %d\n", err);
index 52dba75..aebee3e 100644 (file)
@@ -1,4 +1,4 @@
 MODULES := lov
-lov-objs := lov_log.o lov_obd.o lov_pack.o lproc_lov.o
+lov-objs := lov_log.o lov_obd.o lov_pack.o lproc_lov.o lov_offset.o lov_merge.o lov_request.o lov_qos.o
 
 @INCLUDE_RULES@
index 48c14bb..bdef333 100644 (file)
@@ -6,4 +6,4 @@
 include $(src)/../portals/Kernelenv
 
 obj-y += lov.o
-lov-objs := lov_obd.o lov_pack.o lproc_lov.o lov_log.o
+lov-objs := lov_obd.o lov_pack.o lproc_lov.o lov_log.o lov_offset.o lov_merge.o lov_request.o lov_qos.o
index 2847d56..e6854ec 100644 (file)
@@ -5,7 +5,7 @@
 
 if LIBLUSTRE
 noinst_LIBRARIES = liblov.a
-liblov_a_SOURCES = lov_log.c lov_obd.c lov_pack.c lov_internal.h
+liblov_a_SOURCES = lov_log.c lov_obd.c lov_pack.c lov_request.c lov_offset.c lov_qos.c lov_merge.c lov_internal.h
 liblov_a_CPPFLAGS = $(LLCPPFLAGS)
 liblov_a_CFLAGS = $(LLCFLAGS)
 endif
index cefe534..e6ecc5e 100644 (file)
 
 #include <lustre/lustre_user.h>
 
+struct lov_lock_handles {
+        struct portals_handle   llh_handle;
+        atomic_t                llh_refcount;
+        int                     llh_stripe_count;
+        struct lustre_handle    llh_handles[0];
+};
+
+struct lov_request {
+        struct list_head         rq_link;
+        struct ldlm_extent       rq_extent;
+        int                      rq_idx;        /* index in lov->tgts array */
+        int                      rq_stripe;     /* stripe number */
+        int                      rq_complete;
+        int                      rq_rc;
+        int                      rq_buflen;     /* length of sub_md */
+        struct obdo             *rq_oa;
+        struct lov_stripe_md    *rq_md;
+        obd_count                rq_oabufs;
+        obd_count                rq_pgaidx;
+};
+
+struct lov_request_set {
+        atomic_t                 set_refcount;
+        struct obd_export       *set_exp;
+        int                      set_count;
+        int                      set_completes;
+        int                      set_success;
+        struct llog_cookie      *set_cookies;
+        int                      set_cookie_sent;
+        struct lov_stripe_md    *set_md;
+        struct obdo             *set_oa;
+        struct obd_trans_info   *set_oti;
+        obd_count                set_oabufs;
+        struct brw_page         *set_pga;
+        struct lov_lock_handles *set_lockh;
+        struct list_head         set_list;
+};
+
 #define LAP_MAGIC 8200
 
 struct lov_async_page {
@@ -20,10 +58,139 @@ struct lov_async_page {
         obd_off                         lap_sub_offset;
         void                            *lap_sub_cookie;
         struct obd_async_page_ops       *lap_caller_ops;
-        struct obd_async_page_ops       *lap_caller_data;
+        void                            *lap_caller_data;
         obd_id                          lap_loi_id;
 };
 
+#define LAP_FROM_COOKIE(c)                                                      \
+        (LASSERT(((struct lov_async_page *)(c))->lap_magic == LAP_MAGIC),       \
+         (struct lov_async_page *)(c))
+
+static inline void lov_llh_addref(void *llhp)
+{
+        struct lov_lock_handles *llh = llhp;
+        atomic_inc(&llh->llh_refcount);
+        CDEBUG(D_INFO, "GETting llh %p : new refcount %d\n", llh,
+               atomic_read(&llh->llh_refcount));
+}
+
+static inline struct lov_lock_handles *lov_llh_new(struct lov_stripe_md *lsm)
+{
+        struct lov_lock_handles *llh;
+
+        OBD_ALLOC(llh, sizeof *llh +
+                  sizeof(*llh->llh_handles) * lsm->lsm_stripe_count);
+        if (llh == NULL) 
+                return NULL;
+        atomic_set(&llh->llh_refcount, 2);
+        llh->llh_stripe_count = lsm->lsm_stripe_count;
+        INIT_LIST_HEAD(&llh->llh_handle.h_link);
+        class_handle_hash(&llh->llh_handle, lov_llh_addref);
+        return llh;
+}
+
+static inline struct lov_lock_handles *
+lov_handle2llh(struct lustre_handle *handle)
+{
+        LASSERT(handle != NULL);
+        return(class_handle2object(handle->cookie));
+}
+
+static inline void lov_llh_put(struct lov_lock_handles *llh)
+{
+        CDEBUG(D_INFO, "PUTting llh %p : new refcount %d\n", llh,
+               atomic_read(&llh->llh_refcount) - 1);
+        LASSERT(atomic_read(&llh->llh_refcount) > 0 &&
+                atomic_read(&llh->llh_refcount) < 0x5a5a);
+        if (atomic_dec_and_test(&llh->llh_refcount)) {
+                class_handle_unhash(&llh->llh_handle);
+                LASSERT(list_empty(&llh->llh_handle.h_link));
+                OBD_FREE(llh, sizeof *llh +
+                         sizeof(*llh->llh_handles) * llh->llh_stripe_count);
+        }
+}
+
+/* lov_merge.c */
+void lov_merge_attrs(struct obdo *tgt, struct obdo *src, obd_flag valid,
+                     struct lov_stripe_md *lsm, int stripeno, int *set);
+
+/* lov_offset.c */
+obd_size lov_stripe_size(struct lov_stripe_md *lsm, obd_size ost_size, 
+                         int stripeno);
+int lov_stripe_offset(struct lov_stripe_md *lsm, obd_off lov_off,
+                      int stripeno, obd_off *obd_off);
+obd_off lov_size_to_stripe(struct lov_stripe_md *lsm, obd_off file_size,
+                           int stripeno);
+int lov_stripe_intersects(struct lov_stripe_md *lsm, int stripeno,
+                          obd_off start, obd_off end,
+                          obd_off *obd_start, obd_off *obd_end);
+int lov_stripe_number(struct lov_stripe_md *lsm, obd_off lov_off);
+
+/* lov_qos.c */
+void qos_shrink_lsm(struct lov_request_set *set);
+int qos_prep_create(struct lov_obd *lov, struct lov_request_set *set, 
+                    int newea);
+
+/* lov_request.c */
+void lov_set_add_req(struct lov_request *req, struct lov_request_set *set);
+int lov_update_common_set(struct lov_request_set *set, 
+                          struct lov_request *req, int rc);
+int lov_prep_create_set(struct obd_export *exp, struct lov_stripe_md **ea, 
+                        struct obdo *src_oa, struct obd_trans_info *oti,
+                        struct lov_request_set **reqset);
+int lov_update_create_set(struct lov_request_set *set,
+                          struct lov_request *req, int rc);
+int lov_fini_create_set(struct lov_request_set *set, struct lov_stripe_md **ea);
+int lov_prep_brw_set(struct obd_export *exp, struct obdo *src_oa, 
+                     struct lov_stripe_md *lsm, obd_count oa_bufs,
+                     struct brw_page *pga, struct obd_trans_info *oti,
+                     struct lov_request_set **reqset);
+int lov_fini_brw_set(struct lov_request_set *set);
+int lov_prep_getattr_set(struct obd_export *exp, struct obdo *src_oa, 
+                         struct lov_stripe_md *lsm, 
+                         struct lov_request_set **reqset);
+int lov_fini_getattr_set(struct lov_request_set *set);
+int lov_prep_destroy_set(struct obd_export *exp, struct obdo *src_oa,
+                         struct lov_stripe_md *lsm, 
+                         struct obd_trans_info *oti, 
+                         struct lov_request_set **reqset);
+int lov_update_destroy_set(struct lov_request_set *set,
+                           struct lov_request *req, int rc);
+int lov_fini_destroy_set(struct lov_request_set *set);
+int lov_prep_setattr_set(struct obd_export *exp, struct obdo *src_oa,
+                         struct lov_stripe_md *lsm, struct obd_trans_info *oti,
+                         struct lov_request_set **reqset);
+int lov_fini_setattr_set(struct lov_request_set *set);
+int lov_prep_punch_set(struct obd_export *exp, struct obdo *src_oa,
+                       struct lov_stripe_md *lsm, obd_off start,
+                       obd_off end, struct obd_trans_info *oti,
+                       struct lov_request_set **reqset);
+int lov_update_punch_set(struct lov_request_set *set, struct lov_request *req,
+                         int rc);
+int lov_fini_punch_set(struct lov_request_set *set);
+int lov_prep_sync_set(struct obd_export *exp, struct obdo *src_oa,
+                      struct lov_stripe_md *lsm, obd_off start,
+                      obd_off end, struct lov_request_set **reqset);
+int lov_fini_sync_set(struct lov_request_set *set);
+int lov_prep_enqueue_set(struct obd_export *exp, struct lov_stripe_md *lsm, 
+                         ldlm_policy_data_t *policy, __u32 mode,
+                         struct lustre_handle *lockh,
+                         struct lov_request_set **reqset);
+int lov_update_enqueue_set(struct lov_request_set *set, 
+                           struct lov_request *req, int rc, int flags);
+int lov_fini_enqueue_set(struct lov_request_set *set, __u32 mode);
+int lov_prep_match_set(struct obd_export *exp, struct lov_stripe_md *lsm,
+                       ldlm_policy_data_t *policy, __u32 mode,
+                       struct lustre_handle *lockh,
+                       struct lov_request_set **reqset);
+int lov_update_match_set(struct lov_request_set *set, struct lov_request *req,
+                         int rc);
+int lov_fini_match_set(struct lov_request_set *set, __u32 mode, int flags); 
+int lov_prep_cancel_set(struct obd_export *exp, struct lov_stripe_md *lsm,
+                        __u32 mode, struct lustre_handle *lockh,
+                        struct lov_request_set **reqset);
+int lov_fini_cancel_set(struct lov_request_set *set);
+
 /* lov_obd.c */
 int lov_get_stripecnt(struct lov_obd *lov, int stripe_count);
 int lov_alloc_memmd(struct lov_stripe_md **lsmp, int stripe_count, int pattern);
diff --git a/lustre/lov/lov_merge.c b/lustre/lov/lov_merge.c
new file mode 100644 (file)
index 0000000..e4e40bd
--- /dev/null
@@ -0,0 +1,142 @@
+/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
+ * vim:expandtab:shiftwidth=8:tabstop=8:
+ *
+ * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
+ *
+ *   This file is part of Lustre, http://www.lustre.org.
+ *
+ *   Lustre is free software; you can redistribute it and/or
+ *   modify it under the terms of version 2 of the GNU General Public
+ *   License as published by the Free Software Foundation.
+ *
+ *   Lustre is distributed in the hope that it will be useful,
+ *   but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *   GNU General Public License for more details.
+ *
+ *   You should have received a copy of the GNU General Public License
+ *   along with Lustre; if not, write to the Free Software
+ *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#ifndef EXPORT_SYMTAB
+# define EXPORT_SYMTAB
+#endif
+#define DEBUG_SUBSYSTEM S_LOV
+
+#ifdef __KERNEL__
+#include <asm/div64.h>
+#else
+#include <liblustre.h>
+#endif
+
+#include <linux/obd_class.h>
+#include <linux/obd_lov.h>
+
+#include "lov_internal.h"
+
+/* Merge rss if kms == 0
+ *
+ * Even when merging RSS, we will take the KMS value if it's larger.
+ * This prevents getattr from stomping on dirty cached pages which
+ * extend the file size. */
+__u64 lov_merge_size(struct lov_stripe_md *lsm, int kms)
+{
+        struct lov_oinfo *loi;
+        __u64 size = 0;
+        int i;
+
+        for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
+             i++, loi++) {
+                obd_size lov_size, tmpsize;
+
+                tmpsize = loi->loi_kms;
+                if (kms == 0 && loi->loi_rss > tmpsize)
+                        tmpsize = loi->loi_rss;
+
+                lov_size = lov_stripe_size(lsm, tmpsize, i);
+                if (lov_size > size)
+                        size = lov_size;
+        }
+
+        return size;
+}
+EXPORT_SYMBOL(lov_merge_size);
+
+/* Merge blocks */
+__u64 lov_merge_blocks(struct lov_stripe_md *lsm)
+{
+        struct lov_oinfo *loi;
+        __u64 blocks = 0;
+        int i;
+
+        for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++, loi++)
+                blocks += loi->loi_blocks;
+        return blocks;
+}
+EXPORT_SYMBOL(lov_merge_blocks);
+
+__u64 lov_merge_mtime(struct lov_stripe_md *lsm, __u64 current_time)
+{
+        struct lov_oinfo *loi;
+        int i;
+
+        for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++, loi++)
+                if (loi->loi_mtime > current_time)
+                        current_time = loi->loi_mtime;
+        return current_time;
+}
+EXPORT_SYMBOL(lov_merge_mtime);
+
+int lov_increase_kms(struct obd_export *exp, struct lov_stripe_md *lsm,
+                      obd_off size)
+{
+        struct lov_oinfo *loi;
+        int stripe = 0;
+        __u64 kms;
+        ENTRY;
+
+        if (size > 0)
+                stripe = lov_stripe_number(lsm, size - 1);
+        kms = lov_size_to_stripe(lsm, size, stripe);
+        loi = &(lsm->lsm_oinfo[stripe]);
+
+        CDEBUG(D_INODE, "stripe %d KMS %sincreasing "LPU64"->"LPU64"\n",
+               stripe, kms > loi->loi_kms ? "" : "not ", loi->loi_kms, kms);
+        if (kms > loi->loi_kms)
+                loi->loi_kms = kms;
+
+        RETURN(0);
+}
+EXPORT_SYMBOL(lov_increase_kms);
+
+void lov_merge_attrs(struct obdo *tgt, struct obdo *src, obd_flag valid,
+                     struct lov_stripe_md *lsm, int stripeno, int *set)
+{
+        valid &= src->o_valid;
+
+        if (*set) {
+                if (valid & OBD_MD_FLSIZE) {
+                        /* this handles sparse files properly */
+                        obd_size lov_size;
+
+                        lov_size = lov_stripe_size(lsm, src->o_size, stripeno);
+                        if (lov_size > tgt->o_size)
+                                tgt->o_size = lov_size;
+                }
+                if (valid & OBD_MD_FLBLOCKS)
+                        tgt->o_blocks += src->o_blocks;
+                if (valid & OBD_MD_FLBLKSZ)
+                        tgt->o_blksize += src->o_blksize;
+                if (valid & OBD_MD_FLCTIME && tgt->o_ctime < src->o_ctime)
+                        tgt->o_ctime = src->o_ctime;
+                if (valid & OBD_MD_FLMTIME && tgt->o_mtime < src->o_mtime)
+                        tgt->o_mtime = src->o_mtime;
+        } else {
+                memcpy(tgt, src, sizeof(*tgt));
+                tgt->o_id = lsm->lsm_object_id;
+                if (valid & OBD_MD_FLSIZE)
+                        tgt->o_size = lov_stripe_size(lsm,src->o_size,stripeno);
+                *set = 1;
+        }
+}
index 977ac2c..5ca8bf3 100644 (file)
 
 #include "lov_internal.h"
 
-static int lov_stripe_offset(struct lov_stripe_md *lsm, obd_off lov_off,
-                             int stripeno, obd_off *obd_off);
-
-struct lov_lock_handles {
-        struct portals_handle llh_handle;
-        atomic_t llh_refcount;
-        int llh_stripe_count;
-        struct lustre_handle llh_handles[0];
-};
-
-static void lov_llh_addref(void *llhp)
-{
-        struct lov_lock_handles *llh = llhp;
-
-        atomic_inc(&llh->llh_refcount);
-        CDEBUG(D_INFO, "GETting llh %p : new refcount %d\n", llh,
-               atomic_read(&llh->llh_refcount));
-}
-
-static struct lov_lock_handles *lov_llh_new(struct lov_stripe_md *lsm)
-{
-        struct lov_lock_handles *llh;
-
-        OBD_ALLOC(llh, sizeof *llh +
-                  sizeof(*llh->llh_handles) * lsm->lsm_stripe_count);
-        if (llh == NULL) {
-                CERROR("out of memory\n");
-                return NULL;
-        }
-        atomic_set(&llh->llh_refcount, 2);
-        llh->llh_stripe_count = lsm->lsm_stripe_count;
-        INIT_LIST_HEAD(&llh->llh_handle.h_link);
-        class_handle_hash(&llh->llh_handle, lov_llh_addref);
-        return llh;
-}
-
-static struct lov_lock_handles *lov_handle2llh(struct lustre_handle *handle)
-{
-        ENTRY;
-        LASSERT(handle != NULL);
-        RETURN(class_handle2object(handle->cookie));
-}
-
-static void lov_llh_put(struct lov_lock_handles *llh)
-{
-        CDEBUG(D_INFO, "PUTting llh %p : new refcount %d\n", llh,
-               atomic_read(&llh->llh_refcount) - 1);
-        LASSERT(atomic_read(&llh->llh_refcount) > 0 &&
-                atomic_read(&llh->llh_refcount) < 0x5a5a);
-        if (atomic_dec_and_test(&llh->llh_refcount)) {
-                LASSERT(list_empty(&llh->llh_handle.h_link));
-                OBD_FREE(llh, sizeof *llh +
-                         sizeof(*llh->llh_handles) * llh->llh_stripe_count);
-        }
-}
-
-static void lov_llh_destroy(struct lov_lock_handles *llh)
-{
-        class_handle_unhash(&llh->llh_handle);
-        lov_llh_put(llh);
-}
-
 /* obd methods */
 static int lov_connect(struct lustre_handle *conn, struct obd_device *obd,
                        struct obd_uuid *cluuid)
@@ -391,6 +329,14 @@ static int lov_setup(struct obd_device *obd, obd_count len, void *buf)
                 RETURN(-EINVAL);
         }
 
+        if (desc->ld_default_stripe_size < PTLRPC_MAX_BRW_SIZE) {
+                CWARN("Increasing default_stripe_size "LPU64" to %u\n",
+                      desc->ld_default_stripe_size, PTLRPC_MAX_BRW_SIZE);
+                CWARN("Please update config and run --write-conf on MDS\n");
+
+                desc->ld_default_stripe_size = PTLRPC_MAX_BRW_SIZE;
+        }
+
         /* Because of 64-bit divide/mod operations only work with a 32-bit
          * divisor in a 32-bit kernel, we cannot support a stripe width
          * of 4GB or larger on 32-bit CPUs.
@@ -450,60 +396,6 @@ static int lov_cleanup(struct obd_device *obd, int flags)
         RETURN(0);
 }
 
-
-/* compute object size given "stripeno" and the ost size */
-static obd_size lov_stripe_size(struct lov_stripe_md *lsm, obd_size ost_size,
-                                int stripeno)
-{
-        unsigned long ssize  = lsm->lsm_stripe_size;
-        unsigned long swidth = ssize * lsm->lsm_stripe_count;
-        unsigned long stripe_size;
-        obd_size lov_size;
-
-        if (ost_size == 0)
-                return 0;
-
-        /* do_div(a, b) returns a % b, and a = a / b */
-        stripe_size = do_div(ost_size, ssize);
-        if (stripe_size)
-                lov_size = ost_size * swidth + stripeno * ssize + stripe_size;
-        else
-                lov_size = (ost_size - 1) * swidth + (stripeno + 1) * ssize;
-
-        return lov_size;
-}
-
-static void lov_merge_attrs(struct obdo *tgt, struct obdo *src, obd_flag valid,
-                            struct lov_stripe_md *lsm, int stripeno, int *set)
-{
-        valid &= src->o_valid;
-
-        if (*set) {
-                if (valid & OBD_MD_FLSIZE) {
-                        /* this handles sparse files properly */
-                        obd_size lov_size;
-
-                        lov_size = lov_stripe_size(lsm, src->o_size, stripeno);
-                        if (lov_size > tgt->o_size)
-                                tgt->o_size = lov_size;
-                }
-                if (valid & OBD_MD_FLBLOCKS)
-                        tgt->o_blocks += src->o_blocks;
-                if (valid & OBD_MD_FLBLKSZ)
-                        tgt->o_blksize += src->o_blksize;
-                if (valid & OBD_MD_FLCTIME && tgt->o_ctime < src->o_ctime)
-                        tgt->o_ctime = src->o_ctime;
-                if (valid & OBD_MD_FLMTIME && tgt->o_mtime < src->o_mtime)
-                        tgt->o_mtime = src->o_mtime;
-        } else {
-                memcpy(tgt, src, sizeof(*tgt));
-                tgt->o_id = lsm->lsm_object_id;
-                if (valid & OBD_MD_FLSIZE)
-                        tgt->o_size = lov_stripe_size(lsm,src->o_size,stripeno);
-                *set = 1;
-        }
-}
-
 #ifndef log2
 #define log2(n) ffz(~(n))
 #endif
@@ -565,23 +457,58 @@ static int lov_clear_orphans(struct obd_export *export, struct obdo *src_oa,
         RETURN(rc);
 }
 
-#define LOV_CREATE_RESEED_INTERVAL 1000
+static int lov_recreate(struct obd_export *exp, struct obdo *src_oa,
+                        struct lov_stripe_md **ea, struct obd_trans_info *oti)
+{
+        struct lov_stripe_md *obj_mdp, *lsm;
+        struct lov_obd *lov = &exp->exp_obd->u.lov;
+        unsigned ost_idx;
+        int rc, i;
+        ENTRY;
+
+        LASSERT(src_oa->o_valid & OBD_MD_FLFLAGS &&
+                src_oa->o_flags & OBD_FL_RECREATE_OBJS);
+
+        OBD_ALLOC(obj_mdp, sizeof(*obj_mdp));
+        if (obj_mdp == NULL)
+                RETURN(-ENOMEM);
+
+        ost_idx = src_oa->o_nlink;
+        lsm = *ea;
+        if (lsm == NULL)
+                GOTO(out, rc = -EINVAL);
+        if (ost_idx >= lov->desc.ld_tgt_count)
+                GOTO(out, rc = -EINVAL);
+
+        for (i = 0; i < lsm->lsm_stripe_count; i++) {
+                if (lsm->lsm_oinfo[i].loi_ost_idx == ost_idx) {
+                        if (lsm->lsm_oinfo[i].loi_id != src_oa->o_id)
+                                GOTO(out, rc = -EINVAL);
+                        break;
+                }
+        }
+        if (i == lsm->lsm_stripe_count)
+                GOTO(out, rc = -EINVAL);
+
+        rc = obd_create(lov->tgts[ost_idx].ltd_exp, src_oa, &obj_mdp, oti);
+out:
+        OBD_FREE(obj_mdp, sizeof(*obj_mdp));
+        RETURN(rc);
+}
 
 /* the LOV expects oa->o_id to be set to the LOV object id */
 static int lov_create(struct obd_export *exp, struct obdo *src_oa,
                       struct lov_stripe_md **ea, struct obd_trans_info *oti)
 {
-        static int ost_start_idx, ost_start_count;
         struct lov_obd *lov;
-        struct lov_stripe_md *lsm;
-        struct lov_oinfo *loi = NULL;
-        struct obdo *tmp_oa, *ret_oa;
-        struct llog_cookie *cookies = NULL;
-        unsigned ost_count, ost_idx;
-        int set = 0, obj_alloc = 0, cookie_sent = 0, rc = 0, i;
+        struct lov_request_set *set = NULL;
+        struct list_head *pos;
+        int rc = 0;
         ENTRY;
 
         LASSERT(ea != NULL);
+        if (exp == NULL)
+                RETURN(-EINVAL);
 
         if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
             src_oa->o_flags == OBD_FL_DELORPHAN) {
@@ -589,425 +516,142 @@ static int lov_create(struct obd_export *exp, struct obdo *src_oa,
                 RETURN(rc);
         }
 
-        if (exp == NULL)
-                RETURN(-EINVAL);
-
         lov = &exp->exp_obd->u.lov;
-
         if (!lov->desc.ld_active_tgt_count)
                 RETURN(-EIO);
 
         /* Recreate a specific object id at the given OST index */
-        if (src_oa->o_valid & OBD_MD_FLFLAGS && src_oa->o_flags &
-                                                OBD_FL_RECREATE_OBJS) {
-                 struct lov_stripe_md obj_md;
-                 struct lov_stripe_md *obj_mdp = &obj_md;
-
-                 ost_idx = src_oa->o_nlink;
-                 lsm = *ea;
-                 if (lsm == NULL)
-                        RETURN(-EINVAL);
-                 if (ost_idx >= lov->desc.ld_tgt_count)
-                         RETURN(-EINVAL);
-                 for (i = 0; i < lsm->lsm_stripe_count; i++) {
-                         if (lsm->lsm_oinfo[i].loi_ost_idx == ost_idx) {
-                                 if (lsm->lsm_oinfo[i].loi_id != src_oa->o_id)
-                                         RETURN(-EINVAL);
-                                 break;
-                         }
-                 }
-                 if (i == lsm->lsm_stripe_count)
-                         RETURN(-EINVAL);
-
-                 rc = obd_create(lov->tgts[ost_idx].ltd_exp, src_oa,
-                                 &obj_mdp, oti);
+        if ((src_oa->o_valid & OBD_MD_FLFLAGS) &&
+            (src_oa->o_flags & OBD_FL_RECREATE_OBJS)) {
+                 rc = lov_recreate(exp, src_oa, ea, oti);
                  RETURN(rc);
         }
 
-        ret_oa = obdo_alloc();
-        if (!ret_oa)
-                RETURN(-ENOMEM);
-
-        tmp_oa = obdo_alloc();
-        if (!tmp_oa)
-                GOTO(out_oa, rc = -ENOMEM);
-
-        lsm = *ea;
-        if (lsm == NULL) {
-                int stripes;
-                ost_count = lov_get_stripecnt(lov, 0);
-
-                /* If the MDS file was truncated up to some size, stripe over
-                 * enough OSTs to allow the file to be created at that size. */
-                if (src_oa->o_valid & OBD_MD_FLSIZE) {
-                        stripes=((src_oa->o_size+LUSTRE_STRIPE_MAXBYTES)>>12)-1;
-                        do_div(stripes, (__u32)(LUSTRE_STRIPE_MAXBYTES >> 12));
-
-                        if (stripes > lov->desc.ld_active_tgt_count)
-                                RETURN(-EFBIG);
-                        if (stripes < ost_count)
-                                stripes = ost_count;
-                } else {
-                        stripes = ost_count;
-                }
-
-                rc = lov_alloc_memmd(&lsm, stripes, lov->desc.ld_pattern ?
-                                     lov->desc.ld_pattern : LOV_PATTERN_RAID0);
-                if (rc < 0)
-                        GOTO(out_tmp, rc);
-
-                rc = 0;
-        }
-
-        ost_count = lov->desc.ld_tgt_count;
-
-        LASSERT(src_oa->o_valid & OBD_MD_FLID);
-        lsm->lsm_object_id = src_oa->o_id;
-        if (!lsm->lsm_stripe_size)
-                lsm->lsm_stripe_size = lov->desc.ld_default_stripe_size;
-        if (!lsm->lsm_pattern) {
-                lsm->lsm_pattern = lov->desc.ld_pattern ?
-                        lov->desc.ld_pattern : LOV_PATTERN_RAID0;
-        }
-
-        if (*ea == NULL || lsm->lsm_oinfo[0].loi_ost_idx >= ost_count) {
-                if (--ost_start_count <= 0) {
-                        ost_start_idx = ll_insecure_random_int();
-                        ost_start_count = LOV_CREATE_RESEED_INTERVAL;
-                } else if (lsm->lsm_stripe_count >=
-                           lov->desc.ld_active_tgt_count) {
-                        /* If we allocate from all of the stripes, make the
-                         * next file start on the next OST. */
-                        ++ost_start_idx;
-                }
-                ost_idx = ost_start_idx % ost_count;
-        } else {
-                ost_idx = lsm->lsm_oinfo[0].loi_ost_idx;
-        }
-
-        CDEBUG(D_INODE, "allocating %d subobjs for objid "LPX64" at idx %d\n",
-               lsm->lsm_stripe_count, lsm->lsm_object_id, ost_idx);
-
-        /* XXX LOV STACKING: need to figure out how many real OSCs */
-        if (oti && (src_oa->o_valid & OBD_MD_FLCOOKIE)) {
-                oti_alloc_cookies(oti, lsm->lsm_stripe_count);
-                if (!oti->oti_logcookies)
-                        GOTO(out_cleanup, rc = -ENOMEM);
-                cookies = oti->oti_logcookies;
-        }
-
-        loi = lsm->lsm_oinfo;
-        for (i = 0; i < ost_count; i++, ost_idx = (ost_idx + 1) % ost_count) {
-                struct lov_stripe_md obj_md;
-                struct lov_stripe_md *obj_mdp = &obj_md;
-                int err;
-
-                ++ost_start_idx;
-                if (lov->tgts[ost_idx].active == 0) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", ost_idx);
-                        continue;
-                }
-
-                /* create data objects with "parent" OA */
-                memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
-
-                /* XXX When we start creating objects on demand, we need to
-                 *     make sure that we always create the object on the
-                 *     stripe which holds the existing file size.
-                 */
-                if (src_oa->o_valid & OBD_MD_FLSIZE) {
-                        if (lov_stripe_offset(lsm, src_oa->o_size, i,
-                                              &tmp_oa->o_size) < 0 &&
-                            tmp_oa->o_size)
-                                tmp_oa->o_size--;
-
-                        CDEBUG(D_INODE, "stripe %d has size "LPU64"/"LPU64"\n",
-                               i, tmp_oa->o_size, src_oa->o_size);
-                }
+        rc = lov_prep_create_set(exp, ea, src_oa, oti, &set);
+        if (rc)
+                RETURN(rc);
 
+        list_for_each (pos, &set->set_list) {
+                struct lov_request *req = 
+                        list_entry(pos, struct lov_request, rq_link);
 
                 /* XXX: LOV STACKING: use real "obj_mdp" sub-data */
-                err = obd_create(lov->tgts[ost_idx].ltd_exp, tmp_oa, &obj_mdp,
-                                 oti);
-                if (err) {
-                        if (lov->tgts[ost_idx].active) {
-                                CERROR("error creating objid "LPX64" sub-object"
-                                       " on OST idx %d/%d: rc = %d\n",
-                                       src_oa->o_id, ost_idx,
-                                       lsm->lsm_stripe_count, err);
-                                if (err > 0) {
-                                        CERROR("obd_create returned invalid "
-                                               "err %d\n", err);
-                                        err = -EIO;
-                                }
-                        }
-                        if (!rc)
-                                rc = err;
-                        continue;
-                }
-                if (oti->oti_objid)
-                        oti->oti_objid[ost_idx] = tmp_oa->o_id;
-                loi->loi_id = tmp_oa->o_id;
-                loi->loi_ost_idx = ost_idx;
-                CDEBUG(D_INODE, "objid "LPX64" has subobj "LPX64" at idx %d\n",
-                       lsm->lsm_object_id, loi->loi_id, ost_idx);
-
-                lov_merge_attrs(ret_oa, tmp_oa, tmp_oa->o_valid, lsm,
-                                obj_alloc, &set);
-                loi_init(loi);
-
-                if (cookies)
-                        ++oti->oti_logcookies;
-                if (tmp_oa->o_valid & OBD_MD_FLCOOKIE)
-                        ++cookie_sent;
-                ++obj_alloc;
-                ++loi;
-
-                /* If we have allocated enough objects, we are OK */
-                if (obj_alloc == lsm->lsm_stripe_count)
-                        GOTO(out_done, rc = 0);
-        }
-
-        if (obj_alloc == 0) {
-                if (rc == 0)
-                        rc = -EIO;
-                GOTO(out_cleanup, rc);
-        }
-
-        /* If we were passed specific striping params, then a failure to
-         * meet those requirements is an error, since we can't reallocate
-         * that memory (it might be part of a larger array or something).
-         *
-         * We can only get here if lsm_stripe_count was originally > 1.
-         */
-        if (*ea != NULL) {
-                CERROR("can't lstripe objid "LPX64": have %u want %u, rc %d\n",
-                       lsm->lsm_object_id, obj_alloc, lsm->lsm_stripe_count,rc);
-                if (rc == 0)
-                        rc = -EFBIG;
-                GOTO(out_cleanup, rc);
-        } else {
-                struct lov_stripe_md *lsm_new;
-                /* XXX LOV STACKING call into osc for sizes */
-                unsigned oldsize, newsize;
-
-                if (oti && cookies && cookie_sent) {
-                        oldsize = lsm->lsm_stripe_count * sizeof(*cookies);
-                        newsize = obj_alloc * sizeof(*cookies);
-
-                        oti_alloc_cookies(oti, obj_alloc);
-                        if (oti->oti_logcookies) {
-                                memcpy(oti->oti_logcookies, cookies, newsize);
-                                OBD_FREE(cookies, oldsize);
-                                cookies = oti->oti_logcookies;
-                        } else {
-                                CWARN("'leaking' %d bytes\n", oldsize-newsize);
-                        }
-                }
-
-                CWARN("using fewer stripes for object "LPX64": old %u new %u\n",
-                      lsm->lsm_object_id, lsm->lsm_stripe_count, obj_alloc);
-                oldsize = lov_stripe_md_size(lsm->lsm_stripe_count);
-                newsize = lov_stripe_md_size(obj_alloc);
-                OBD_ALLOC(lsm_new, newsize);
-                if (lsm_new != NULL) {
-                        memcpy(lsm_new, lsm, newsize);
-                        lsm_new->lsm_stripe_count = obj_alloc;
-                        OBD_FREE(lsm, oldsize);
-                        lsm = lsm_new;
-                } else {
-                        CWARN("'leaking' %d bytes\n", oldsize - newsize);
-                }
-                rc = 0;
-        }
-        EXIT;
- out_done:
-        *ea = lsm;
-        if (src_oa->o_valid & OBD_MD_FLSIZE &&
-            ret_oa->o_size != src_oa->o_size) {
-                CERROR("original size "LPU64" isn't new object size "LPU64"\n",
-                       src_oa->o_size, ret_oa->o_size);
-                LBUG();
-        }
-        ret_oa->o_id = src_oa->o_id;
-        memcpy(src_oa, ret_oa, sizeof(*src_oa));
-
- out_tmp:
-        obdo_free(tmp_oa);
- out_oa:
-        obdo_free(ret_oa);
-        if (oti && cookies) {
-                oti->oti_logcookies = cookies;
-                if (!cookie_sent) {
-                        oti_free_cookies(oti);
-                        src_oa->o_valid &= ~OBD_MD_FLCOOKIE;
-                } else {
-                        src_oa->o_valid |= OBD_MD_FLCOOKIE;
-                }
+                rc = obd_create(lov->tgts[req->rq_idx].ltd_exp, 
+                                req->rq_oa, &req->rq_md, oti);
+                lov_update_create_set(set, req, rc);
         }
+        rc = lov_fini_create_set(set, ea);
         RETURN(rc);
-
- out_cleanup:
-        while (obj_alloc-- > 0) {
-                struct obd_export *sub_exp;
-                int err;
-
-                --loi;
-                sub_exp = lov->tgts[loi->loi_ost_idx].ltd_exp;
-                /* destroy already created objects here */
-                memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
-                tmp_oa->o_id = loi->loi_id;
-
-                err = obd_destroy(sub_exp, tmp_oa, NULL, oti);
-                if (err)
-                        CERROR("Failed to uncreate objid "LPX64" subobj "LPX64
-                               " on OST idx %d: rc = %d\n", src_oa->o_id,
-                               loi->loi_id, loi->loi_ost_idx, err);
-        }
-        if (*ea == NULL)
-                obd_free_memmd(exp, &lsm);
-        goto out_tmp;
 }
 
-#define lsm_bad_magic(LSMP)                                     \
-({                                                              \
-        struct lov_stripe_md *_lsm__ = (LSMP);                  \
-        int _ret__ = 0;                                         \
-        if (!_lsm__) {                                          \
-                CERROR("LOV requires striping ea\n");           \
-                _ret__ = 1;                                     \
-        } else if (_lsm__->lsm_magic != LOV_MAGIC) {            \
-                CERROR("LOV striping magic bad %#x != %#x\n",   \
-                       _lsm__->lsm_magic, LOV_MAGIC);           \
-                _ret__ = 1;                                     \
-        }                                                       \
-        _ret__;                                                 \
-})
+#define ASSERT_LSM_MAGIC(lsmp)                                  \
+do {                                                            \
+        LASSERT((lsmp) != NULL);                                \
+        LASSERTF((lsmp)->lsm_magic == LOV_MAGIC, "%p, %x",      \
+                 (lsmp), (lsmp)->lsm_magic);                    \
+} while (0)
 
 static int lov_destroy(struct obd_export *exp, struct obdo *oa,
                        struct lov_stripe_md *lsm, struct obd_trans_info *oti)
 {
-        struct obdo tmp;
+        struct lov_request_set *set;
+        struct lov_request *req;
+        struct list_head *pos;
         struct lov_obd *lov;
-        struct lov_oinfo *loi;
-        int rc = 0, i;
+        int rc = 0;
         ENTRY;
 
-        if (lsm_bad_magic(lsm))
-                RETURN(-EINVAL);
+        ASSERT_LSM_MAGIC(lsm);
 
         if (!exp || !exp->exp_obd)
                 RETURN(-ENODEV);
 
         lov = &exp->exp_obd->u.lov;
-        for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) {
+        rc = lov_prep_destroy_set(exp, oa, lsm, oti, &set);
+        if (rc)
+                RETURN(rc);
+
+        list_for_each (pos, &set->set_list) {
                 int err;
-                if (lov->tgts[loi->loi_ost_idx].active == 0) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
-                        /* Orphan clean up will (someday) fix this up. */
-                        if (oti != NULL && oa->o_valid & OBD_MD_FLCOOKIE)
-                                oti->oti_logcookies++;
-                        continue;
-                }
+                req = list_entry(pos, struct lov_request, rq_link);
 
-                memcpy(&tmp, oa, sizeof(tmp));
-                tmp.o_id = loi->loi_id;
-                err = obd_destroy(lov->tgts[loi->loi_ost_idx].ltd_exp, &tmp,
-                                  NULL, oti);
-                if (err && lov->tgts[loi->loi_ost_idx].active) {
+                /* XXX update the cookie position */
+                oti->oti_logcookies = set->set_cookies + req->rq_stripe;
+                rc = obd_destroy(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa,
+                                 NULL, oti);
+                err = lov_update_common_set(set, req, rc);
+                if (rc) {
                         CERROR("error: destroying objid "LPX64" subobj "
-                               LPX64" on OST idx %d: rc = %d\n",
-                               oa->o_id, loi->loi_id, loi->loi_ost_idx, err);
+                               LPX64" on OST idx %d: rc = %d\n", 
+                               set->set_oa->o_id, req->rq_oa->o_id, 
+                               req->rq_idx, rc);
                         if (!rc)
                                 rc = err;
                 }
         }
+        lov_fini_destroy_set(set);
         RETURN(rc);
 }
 
 static int lov_getattr(struct obd_export *exp, struct obdo *oa,
                        struct lov_stripe_md *lsm)
 {
-        struct obdo tmp;
+        struct lov_request_set *set;
+        struct lov_request *req;
+        struct list_head *pos;
         struct lov_obd *lov;
-        struct lov_oinfo *loi;
-        int i, rc = 0, set = 0;
+        int err = 0, rc = 0;
         ENTRY;
 
-        if (lsm_bad_magic(lsm))
-                RETURN(-EINVAL);
+        ASSERT_LSM_MAGIC(lsm);
 
         if (!exp || !exp->exp_obd)
                 RETURN(-ENODEV);
 
         lov = &exp->exp_obd->u.lov;
+        
+        rc = lov_prep_getattr_set(exp, oa, lsm, &set);
+        if (rc)
+                RETURN(rc);
 
-        CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
-               lsm->lsm_object_id, lsm->lsm_stripe_count, lsm->lsm_stripe_size);
-        for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) {
-                int err;
-
-                if (lov->tgts[loi->loi_ost_idx].active == 0) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
-                        continue;
-                }
-
+        list_for_each (pos, &set->set_list) {
+                req = list_entry(pos, struct lov_request, rq_link);
+                
                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
-                       "%u\n", oa->o_id, i, loi->loi_id, loi->loi_ost_idx);
-                /* create data objects with "parent" OA */
-                memcpy(&tmp, oa, sizeof(tmp));
-                tmp.o_id = loi->loi_id;
+                       "%u\n", oa->o_id, req->rq_stripe, req->rq_oa->o_id, 
+                       req->rq_idx);
 
-                err = obd_getattr(lov->tgts[loi->loi_ost_idx].ltd_exp, &tmp,
-                                  NULL);
+                rc = obd_getattr(lov->tgts[req->rq_idx].ltd_exp, 
+                                 req->rq_oa, NULL);
+                err = lov_update_common_set(set, req, rc);
                 if (err) {
-                        if (lov->tgts[loi->loi_ost_idx].active) {
-                                CERROR("error: getattr objid "LPX64" subobj "
-                                       LPX64" on OST idx %d: rc = %d\n",
-                                       oa->o_id, loi->loi_id, loi->loi_ost_idx,
-                                       err);
-                                RETURN(err);
-                        }
-                } else {
-                        lov_merge_attrs(oa, &tmp, tmp.o_valid, lsm, i, &set);
+                        CERROR("error: getattr objid "LPX64" subobj "
+                               LPX64" on OST idx %d: rc = %d\n",
+                               set->set_oa->o_id, req->rq_oa->o_id, 
+                               req->rq_idx, err);
+                        break;
                 }
         }
-        if (!set)
-                rc = -EIO;
+        
+        rc = lov_fini_getattr_set(set);
+        if (err)
+                rc = err;
         RETURN(rc);
 }
 
 static int lov_getattr_interpret(struct ptlrpc_request_set *rqset, void *data,
                                  int rc)
 {
-        struct lov_getattr_async_args *aa = data;
-        struct lov_stripe_md *lsm = aa->aa_lsm;
-        struct obdo          *oa = aa->aa_oa;
-        struct obdo          *obdos = aa->aa_obdos;
-        struct lov_oinfo     *loi;
-        int                   i;
-        int                   set = 0;
+        struct lov_request_set *lovset = (struct lov_request_set *)data;
         ENTRY;
 
-        if (rc == 0) {
-                /* NB all stripe requests succeeded to get here */
-
-                for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
-                     i++, loi++) {
-                        if (obdos[i].o_valid == 0)      /* inactive stripe */
-                                continue;
-
-                        lov_merge_attrs(oa, &obdos[i], obdos[i].o_valid, lsm,
-                                        i, &set);
-                }
-
-                if (!set) {
-                        CERROR ("No stripes had valid attrs\n");
-                        rc = -EIO;
-                }
+        /* don't do attribute merge if this aysnc op failed */
+        if (rc) {
+                lovset->set_completes = 0;
+                lov_fini_getattr_set(lovset);
+        } else {
+                rc = lov_fini_getattr_set(lovset);
         }
-
-        OBD_FREE (obdos, lsm->lsm_stripe_count * sizeof (*obdos));
         RETURN (rc);
 }
 
@@ -1015,84 +659,67 @@ static int lov_getattr_async(struct obd_export *exp, struct obdo *oa,
                               struct lov_stripe_md *lsm,
                               struct ptlrpc_request_set *rqset)
 {
-        struct obdo *obdos;
+        struct lov_request_set *lovset;
         struct lov_obd *lov;
-        struct lov_oinfo *loi;
-        struct lov_getattr_async_args *aa;
-        int i, rc = 0, set = 0;
+        struct list_head *pos;
+        struct lov_request *req;
+        int rc = 0;
         ENTRY;
 
-        if (lsm_bad_magic(lsm))
-                RETURN(-EINVAL);
+        ASSERT_LSM_MAGIC(lsm);
 
         if (!exp || !exp->exp_obd)
                 RETURN(-ENODEV);
 
         lov = &exp->exp_obd->u.lov;
 
-        OBD_ALLOC (obdos, lsm->lsm_stripe_count * sizeof (*obdos));
-        if (obdos == NULL)
-                RETURN(-ENOMEM);
+        rc = lov_prep_getattr_set(exp, oa, lsm, &lovset);
+        if (rc)
+                RETURN(rc);
 
         CDEBUG(D_INFO, "objid "LPX64": %ux%u byte stripes\n",
                lsm->lsm_object_id, lsm->lsm_stripe_count, lsm->lsm_stripe_size);
-        for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) {
-                int err;
-
-                if (lov->tgts[loi->loi_ost_idx].active == 0) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
-                        /* leaves obdos[i].obd_valid unset */
-                        continue;
-                }
 
+        list_for_each (pos, &lovset->set_list) {
+                req = list_entry(pos, struct lov_request, rq_link);
+                
                 CDEBUG(D_INFO, "objid "LPX64"[%d] has subobj "LPX64" at idx "
-                       "%u\n", oa->o_id, i, loi->loi_id, loi->loi_ost_idx);
-                /* create data objects with "parent" OA */
-                memcpy(&obdos[i], oa, sizeof(obdos[i]));
-                obdos[i].o_id = loi->loi_id;
-
-                err = obd_getattr_async(lov->tgts[loi->loi_ost_idx].ltd_exp,
-                                         &obdos[i], NULL, rqset);
-                if (err) {
+                       "%u\n", oa->o_id, req->rq_stripe, req->rq_oa->o_id, 
+                       req->rq_idx);
+                rc = obd_getattr_async(lov->tgts[req->rq_idx].ltd_exp,
+                                       req->rq_oa, NULL, rqset);
+                if (rc) {
                         CERROR("error: getattr objid "LPX64" subobj "
                                LPX64" on OST idx %d: rc = %d\n",
-                               oa->o_id, loi->loi_id, loi->loi_ost_idx,
-                               err);
-                        GOTO(out_obdos, rc = err);
+                               lovset->set_oa->o_id, req->rq_oa->o_id, 
+                               req->rq_idx, rc);
+                        GOTO(out, rc);
                 }
-                set = 1;
+                lov_update_common_set(lovset, req, rc);
         }
-        if (!set)
-                GOTO (out_obdos, rc = -EIO);
-
+        
+        LASSERT(rc == 0);
         LASSERT (rqset->set_interpret == NULL);
         rqset->set_interpret = lov_getattr_interpret;
-        LASSERT (sizeof (rqset->set_args) >= sizeof (*aa));
-        aa = (struct lov_getattr_async_args *)&rqset->set_args;
-        aa->aa_lsm = lsm;
-        aa->aa_oa = oa;
-        aa->aa_obdos = obdos;
-        aa->aa_lov = lov;
-        GOTO(out, rc = 0);
-
-out_obdos:
-        OBD_FREE (obdos, lsm->lsm_stripe_count * sizeof (*obdos));
+        rqset->set_arg = (void *)lovset;
+        RETURN(rc);
 out:
+        LASSERT(rc);
+        lov_fini_getattr_set(lovset);
         RETURN(rc);
 }
 
-
 static int lov_setattr(struct obd_export *exp, struct obdo *src_oa,
                        struct lov_stripe_md *lsm, struct obd_trans_info *oti)
 {
-        struct obdo *tmp_oa, *ret_oa;
+        struct lov_request_set *set;
         struct lov_obd *lov;
-        struct lov_oinfo *loi;
-        int rc = 0, i, set = 0;
+        struct list_head *pos;
+        struct lov_request *req;
+        int err = 0, rc = 0;
         ENTRY;
 
-        if (lsm_bad_magic(lsm))
-                RETURN(-EINVAL);
+        ASSERT_LSM_MAGIC(lsm);
 
         if (!exp || !exp->exp_obd)
                 RETURN(-ENODEV);
@@ -1102,245 +729,30 @@ static int lov_setattr(struct obd_export *exp, struct obdo *src_oa,
                                       OBD_MD_FLATIME | OBD_MD_FLMTIME |
                                       OBD_MD_FLCTIME | OBD_MD_FLFLAGS |
                                       OBD_MD_FLSIZE)));
-        ret_oa = obdo_alloc();
-        if (!ret_oa)
-                RETURN(-ENOMEM);
-
-        tmp_oa = obdo_alloc();
-        if (!tmp_oa)
-                GOTO(out_oa, rc = -ENOMEM);
-
         lov = &exp->exp_obd->u.lov;
-        for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) {
-                int err;
-
-                if (lov->tgts[loi->loi_ost_idx].active == 0) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
-                        continue;
-                }
-
-                memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
-                tmp_oa->o_id = loi->loi_id;
-
-                if (src_oa->o_valid & OBD_MD_FLSIZE) {
-                        if (lov_stripe_offset(lsm, src_oa->o_size, i,
-                                              &tmp_oa->o_size) < 0 &&
-                            tmp_oa->o_size)
-                                tmp_oa->o_size--;
-
-                        CDEBUG(D_INODE, "stripe %d has size "LPU64"/"LPU64"\n",
-                               i, tmp_oa->o_size, src_oa->o_size);
-                }
-                if (src_oa->o_valid & OBD_MD_FLMTIME)
-                        loi->loi_mtime = src_oa->o_mtime;
+        rc = lov_prep_setattr_set(exp, src_oa, lsm, NULL, &set);
+        if (rc)
+                RETURN(rc);
 
-                err = obd_setattr(lov->tgts[loi->loi_ost_idx].ltd_exp, tmp_oa,
-                                  NULL, NULL);
+        list_for_each (pos, &set->set_list) {
+                req = list_entry(pos, struct lov_request, rq_link);
+                
+                rc = obd_setattr(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa,
+                                 NULL, NULL);
+                err = lov_update_common_set(set, req, rc);
                 if (err) {
-                        if (lov->tgts[loi->loi_ost_idx].active) {
-                                CERROR("error: setattr objid "LPX64" subobj "
-                                       LPX64" on OST idx %d: rc = %d\n",
-                                       src_oa->o_id, loi->loi_id,
-                                       loi->loi_ost_idx, err);
-                                if (!rc)
-                                        rc = err;
-                        }
-                        continue;
-                }
-                lov_merge_attrs(ret_oa, tmp_oa, tmp_oa->o_valid, lsm, i, &set);
-        }
-        if (!set && !rc)
-                rc = -EIO;
-
-        ret_oa->o_id = src_oa->o_id;
-        memcpy(src_oa, ret_oa, sizeof(*src_oa));
-        GOTO(out_tmp, rc);
-out_tmp:
-        obdo_free(tmp_oa);
-out_oa:
-        obdo_free(ret_oa);
-        return rc;
-}
-
-/* we have an offset in file backed by an lov and want to find out where
- * that offset lands in our given stripe of the file.  for the easy
- * case where the offset is within the stripe, we just have to scale the
- * offset down to make it relative to the stripe instead of the lov.
- *
- * the harder case is what to do when the offset doesn't intersect the
- * stripe.  callers will want start offsets clamped ahead to the start
- * of the nearest stripe in the file.  end offsets similarly clamped to the
- * nearest ending byte of a stripe in the file:
- *
- * all this function does is move offsets to the nearest region of the
- * stripe, and it does its work "mod" the full length of all the stripes.
- * consider a file with 3 stripes:
- *
- *             S                                              E
- * ---------------------------------------------------------------------
- * |    0    |     1     |     2     |    0    |     1     |     2     |
- * ---------------------------------------------------------------------
- *
- * to find stripe 1's offsets for S and E, it divides by the full stripe
- * width and does its math in the context of a single set of stripes:
- *
- *             S         E
- * -----------------------------------
- * |    0    |     1     |     2     |
- * -----------------------------------
- *
- * it'll notice that E is outside stripe 1 and clamp it to the end of the
- * stripe, then multiply it back out by lov_off to give the real offsets in
- * the stripe:
- *
- *   S                   E
- * ---------------------------------------------------------------------
- * |    1    |     1     |     1     |    1    |     1     |     1     |
- * ---------------------------------------------------------------------
- *
- * it would have done similarly and pulled S forward to the start of a 1
- * stripe if, say, S had landed in a 0 stripe.
- *
- * this rounding isn't always correct.  consider an E lov offset that lands
- * on a 0 stripe, the "mod stripe width" math will pull it forward to the
- * start of a 1 stripe, when in fact it wanted to be rounded back to the end
- * of a previous 1 stripe.  this logic is handled by callers and this is why:
- *
- * this function returns < 0 when the offset was "before" the stripe and
- * was moved forward to the start of the stripe in question;  0 when it
- * falls in the stripe and no shifting was done; > 0 when the offset
- * was outside the stripe and was pulled back to its final byte. */
-static int lov_stripe_offset(struct lov_stripe_md *lsm, obd_off lov_off,
-                             int stripeno, obd_off *obd_off)
-{
-        unsigned long ssize  = lsm->lsm_stripe_size;
-        unsigned long swidth = ssize * lsm->lsm_stripe_count;
-        unsigned long stripe_off, this_stripe;
-        int ret = 0;
-
-        if (lov_off == OBD_OBJECT_EOF) {
-                *obd_off = OBD_OBJECT_EOF;
-                return 0;
-        }
-
-        /* do_div(a, b) returns a % b, and a = a / b */
-        stripe_off = do_div(lov_off, swidth);
-
-        this_stripe = stripeno * ssize;
-        if (stripe_off < this_stripe) {
-                stripe_off = 0;
-                ret = -1;
-        } else {
-                stripe_off -= this_stripe;
-
-                if (stripe_off >= ssize) {
-                        stripe_off = ssize;
-                        ret = 1;
-                }
-        }
-
-        *obd_off = lov_off * ssize + stripe_off;
-        return ret;
-}
-
-/* Given a whole-file size and a stripe number, give the file size which
- * corresponds to the individual object of that stripe.
- *
- * This behaves basically in the same was as lov_stripe_offset, except that
- * file sizes falling before the beginning of a stripe are clamped to the end
- * of the previous stripe, not the beginning of the next:
- *
- *                                               S
- * ---------------------------------------------------------------------
- * |    0    |     1     |     2     |    0    |     1     |     2     |
- * ---------------------------------------------------------------------
- *
- * if clamped to stripe 2 becomes:
- *
- *                                   S
- * ---------------------------------------------------------------------
- * |    0    |     1     |     2     |    0    |     1     |     2     |
- * ---------------------------------------------------------------------
- */
-static obd_off lov_size_to_stripe(struct lov_stripe_md *lsm, obd_off file_size,
-                                  int stripeno)
-{
-        unsigned long ssize  = lsm->lsm_stripe_size;
-        unsigned long swidth = ssize * lsm->lsm_stripe_count;
-        unsigned long stripe_off, this_stripe;
-
-        if (file_size == OBD_OBJECT_EOF)
-                return OBD_OBJECT_EOF;
-
-        /* do_div(a, b) returns a % b, and a = a / b */
-        stripe_off = do_div(file_size, swidth);
-
-        this_stripe = stripeno * ssize;
-        if (stripe_off < this_stripe) {
-                /* Move to end of previous stripe, or zero */
-                if (file_size > 0) {
-                        file_size--;
-                        stripe_off = ssize;
-                } else {
-                        stripe_off = 0;
-                }
-        } else {
-                stripe_off -= this_stripe;
-
-                if (stripe_off >= ssize) {
-                        /* Clamp to end of this stripe */
-                        stripe_off = ssize;
+                        CERROR("error: setattr objid "LPX64" subobj "
+                               LPX64" on OST idx %d: rc = %d\n",
+                               set->set_oa->o_id, req->rq_oa->o_id,
+                               req->rq_idx, err);
+                        if (!rc)
+                                rc = err;
                 }
         }
-
-        return (file_size * ssize + stripe_off);
-}
-
-/* given an extent in an lov and a stripe, calculate the extent of the stripe
- * that is contained within the lov extent.  this returns true if the given
- * stripe does intersect with the lov extent. */
-static int lov_stripe_intersects(struct lov_stripe_md *lsm, int stripeno,
-                                 obd_off start, obd_off end,
-                                 obd_off *obd_start, obd_off *obd_end)
-{
-        int start_side, end_side;
-
-        start_side = lov_stripe_offset(lsm, start, stripeno, obd_start);
-        end_side = lov_stripe_offset(lsm, end, stripeno, obd_end);
-
-        CDEBUG(D_INODE, "["LPU64"->"LPU64"] -> [(%d) "LPU64"->"LPU64" (%d)]\n",
-               start, end, start_side, *obd_start, *obd_end, end_side);
-
-        /* this stripe doesn't intersect the file extent when neither
-         * start or the end intersected the stripe and obd_start and
-         * obd_end got rounded up to the save value. */
-        if (start_side != 0 && end_side != 0 && *obd_start == *obd_end)
-                return 0;
-
-        /* as mentioned in the lov_stripe_offset commentary, end
-         * might have been shifted in the wrong direction.  This
-         * happens when an end offset is before the stripe when viewed
-         * through the "mod stripe size" math. we detect it being shifted
-         * in the wrong direction and touch it up.
-         * interestingly, this can't underflow since end must be > start
-         * if we passed through the previous check.
-         * (should we assert for that somewhere?) */
-        if (end_side != 0)
-                (*obd_end)--;
-
-        return 1;
-}
-
-/* compute which stripe number "lov_off" will be written into */
-static int lov_stripe_number(struct lov_stripe_md *lsm, obd_off lov_off)
-{
-        unsigned long ssize  = lsm->lsm_stripe_size;
-        unsigned long swidth = ssize * lsm->lsm_stripe_count;
-        unsigned long stripe_off;
-
-        stripe_off = do_div(lov_off, swidth);
-
-        return stripe_off / ssize;
+        err = lov_fini_setattr_set(set);
+        if (!rc)
+                rc = err;
+        RETURN(rc);
 }
 
 /* FIXME: maybe we'll just make one node the authoritative attribute node, then
@@ -1350,101 +762,81 @@ static int lov_punch(struct obd_export *exp, struct obdo *oa,
                      struct lov_stripe_md *lsm,
                      obd_off start, obd_off end, struct obd_trans_info *oti)
 {
-        struct obdo tmp;
+        struct lov_request_set *set;
         struct lov_obd *lov;
-        struct lov_oinfo *loi;
-        int rc = 0, i;
+        struct list_head *pos;
+        struct lov_request *req;
+        int err = 0, rc = 0;
         ENTRY;
 
-        if (lsm_bad_magic(lsm))
-                RETURN(-EINVAL);
+        ASSERT_LSM_MAGIC(lsm);
 
         if (!exp || !exp->exp_obd)
                 RETURN(-ENODEV);
 
         lov = &exp->exp_obd->u.lov;
-        for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) {
-                obd_off starti, endi;
-                int err;
-
-                if (lov->tgts[loi->loi_ost_idx].active == 0) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
-                        continue;
-                }
-
-                if (!lov_stripe_intersects(lsm, i, start, end, &starti, &endi))
-                        continue;
+        rc = lov_prep_punch_set(exp, oa, lsm, start, end, oti, &set);
+        if (rc)
+                RETURN(rc);
 
-                /* create data objects with "parent" OA */
-                memcpy(&tmp, oa, sizeof(tmp));
-                tmp.o_id = loi->loi_id;
+        list_for_each (pos, &set->set_list) {
+                req = list_entry(pos, struct lov_request, rq_link);
 
-                err = obd_punch(lov->tgts[loi->loi_ost_idx].ltd_exp, &tmp, NULL,
-                                starti, endi, NULL);
+                rc = obd_punch(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa, 
+                               NULL, req->rq_extent.start, 
+                               req->rq_extent.end, NULL);
+                err = lov_update_punch_set(set, req, rc);
                 if (err) {
-                        if (lov->tgts[loi->loi_ost_idx].active) {
-                                CERROR("error: punch objid "LPX64" subobj "LPX64
-                                       " on OST idx %d: rc = %d\n", oa->o_id,
-                                       loi->loi_id, loi->loi_ost_idx, err);
-                        }
+                        CERROR("error: punch objid "LPX64" subobj "LPX64
+                               " on OST idx %d: rc = %d\n", set->set_oa->o_id,
+                               req->rq_oa->o_id, req->rq_idx, rc);
                         if (!rc)
                                 rc = err;
-                } else {
-                        loi->loi_kms = loi->loi_rss = starti;
                 }
         }
+        err = lov_fini_punch_set(set);
+        if (!rc)
+                rc = err;
         RETURN(rc);
 }
 
 static int lov_sync(struct obd_export *exp, struct obdo *oa,
                     struct lov_stripe_md *lsm, obd_off start, obd_off end)
 {
-        struct obdo *tmp;
+        struct lov_request_set *set;
         struct lov_obd *lov;
-        struct lov_oinfo *loi;
-        int rc = 0, i;
+        struct list_head *pos;
+        struct lov_request *req;
+        int err = 0, rc = 0;
         ENTRY;
 
-        if (lsm_bad_magic(lsm))
-                RETURN(-EINVAL);
+        ASSERT_LSM_MAGIC(lsm);
 
         if (!exp->exp_obd)
                 RETURN(-ENODEV);
 
-        tmp = obdo_alloc();
-        if (!tmp)
-                RETURN(-ENOMEM);
-
-        lov = &exp->exp_obd->u.lov;
-        for (i = 0,loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++) {
-                obd_off starti, endi;
-                int err;
-
-                if (lov->tgts[loi->loi_ost_idx].active == 0) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
-                        continue;
-                }
-
-                if (!lov_stripe_intersects(lsm, i, start, end, &starti, &endi))
-                        continue;
+        lov = &exp->exp_obd->u.lov;
+        rc = lov_prep_sync_set(exp, oa, lsm, start, end, &set);
+        if (rc)
+                RETURN(rc);
 
-                memcpy(tmp, oa, sizeof(*tmp));
-                tmp->o_id = loi->loi_id;
+        list_for_each (pos, &set->set_list) {
+                req = list_entry(pos, struct lov_request, rq_link);
 
-                err = obd_sync(lov->tgts[loi->loi_ost_idx].ltd_exp, tmp, NULL,
-                               starti, endi);
+                rc = obd_sync(lov->tgts[req->rq_idx].ltd_exp, req->rq_oa, 
+                              NULL, req->rq_extent.start, req->rq_extent.end);
+                err = lov_update_common_set(set, req, rc);
                 if (err) {
-                        if (lov->tgts[loi->loi_ost_idx].active) {
-                                CERROR("error: fsync objid "LPX64" subobj "LPX64
-                                       " on OST idx %d: rc = %d\n", oa->o_id,
-                                       loi->loi_id, loi->loi_ost_idx, err);
-                        }
+                        CERROR("error: fsync objid "LPX64" subobj "LPX64
+                               " on OST idx %d: rc = %d\n", set->set_oa->o_id,
+                               req->rq_oa->o_id, req->rq_idx, rc);
                         if (!rc)
                                 rc = err;
                 }
         }
-
-        obdo_free(tmp);
+        err = lov_fini_sync_set(set);
+        if (!rc)
+                rc = err;
         RETURN(rc);
 }
 
@@ -1459,12 +851,11 @@ static int lov_brw_check(struct lov_obd *lov, struct obdo *oa,
         for (i = 0; i < oa_bufs; i++) {
                 int stripe = lov_stripe_number(lsm, pga[i].off);
                 int ost = lsm->lsm_oinfo[stripe].loi_ost_idx;
-                struct ldlm_extent ext, subext;
-                ext.start = pga[i].off;
-                ext.end = pga[i].off + pga[i].count;
+                obd_off start, end;
 
-                if (!lov_stripe_intersects(lsm, i, ext.start, ext.end,
-                                           &subext.start, &subext.end))
+                if (!lov_stripe_intersects(lsm, i, pga[i].off,
+                                           pga[i].off + pga[i].count,
+                                           &start, &end))
                         continue;
 
                 if (lov->tgts[ost].active == 0) {
@@ -1483,137 +874,57 @@ static int lov_brw(int cmd, struct obd_export *exp, struct obdo *src_oa,
                    struct lov_stripe_md *lsm, obd_count oa_bufs,
                    struct brw_page *pga, struct obd_trans_info *oti)
 {
-        struct {
-                int bufct;
-                int index;
-                int subcount;
-                struct lov_stripe_md lsm;
-                int ost_idx;
-        } *stripeinfo, *si, *si_last;
-        struct obdo *ret_oa = NULL, *tmp_oa = NULL;
-        struct lov_obd *lov;
-        struct brw_page *ioarr;
-        struct lov_oinfo *loi;
-        int rc = 0, i, *where, stripe_count = lsm->lsm_stripe_count, set = 0;
+        struct lov_request_set *set;
+        struct lov_request *req;
+        struct list_head *pos;
+        struct lov_obd *lov = &exp->exp_obd->u.lov;
+        int err, rc = 0;
         ENTRY;
 
-        if (lsm_bad_magic(lsm))
-                RETURN(-EINVAL);
-
-        lov = &exp->exp_obd->u.lov;
+        ASSERT_LSM_MAGIC(lsm);
 
         if (cmd == OBD_BRW_CHECK) {
                 rc = lov_brw_check(lov, src_oa, lsm, oa_bufs, pga);
                 RETURN(rc);
         }
 
-        OBD_ALLOC(stripeinfo, stripe_count * sizeof(*stripeinfo));
-        if (!stripeinfo)
-                RETURN(-ENOMEM);
-
-        OBD_ALLOC(where, sizeof(*where) * oa_bufs);
-        if (!where)
-                GOTO(out_sinfo, rc = -ENOMEM);
-
-        OBD_ALLOC(ioarr, sizeof(*ioarr) * oa_bufs);
-        if (!ioarr)
-                GOTO(out_where, rc = -ENOMEM);
-
-        if (src_oa) {
-                ret_oa = obdo_alloc();
-                if (!ret_oa)
-                        GOTO(out_ioarr, rc = -ENOMEM);
-
-                tmp_oa = obdo_alloc();
-                if (!tmp_oa)
-                        GOTO(out_oa, rc = -ENOMEM);
-        }
-
-        for (i = 0; i < oa_bufs; i++) {
-                where[i] = lov_stripe_number(lsm, pga[i].off);
-                stripeinfo[where[i]].bufct++;
-        }
-
-        for (i = 0, loi = lsm->lsm_oinfo, si_last = si = stripeinfo;
-             i < stripe_count; i++, loi++, si_last = si, si++) {
-                if (i > 0)
-                        si->index = si_last->index + si_last->bufct;
-                si->lsm.lsm_object_id = loi->loi_id;
-                si->ost_idx = loi->loi_ost_idx;
-        }
-
-        for (i = 0; i < oa_bufs; i++) {
-                int which = where[i];
-                int shift;
-
-                shift = stripeinfo[which].index + stripeinfo[which].subcount;
-                LASSERT(shift < oa_bufs);
-                ioarr[shift] = pga[i];
-                lov_stripe_offset(lsm, pga[i].off, which, &ioarr[shift].off);
-                stripeinfo[which].subcount++;
-        }
-
-        for (i = 0, si = stripeinfo; i < stripe_count; i++, si++) {
-                int shift = si->index;
-
-                if (lov->tgts[si->ost_idx].active == 0) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", si->ost_idx);
-                        GOTO(out_oa, rc = -EIO);
-                }
-
-                if (si->bufct) {
-                        LASSERT(shift < oa_bufs);
-                        if (src_oa)
-                                memcpy(tmp_oa, src_oa, sizeof(*tmp_oa));
-
-                        tmp_oa->o_id = si->lsm.lsm_object_id;
-                        rc = obd_brw(cmd, lov->tgts[si->ost_idx].ltd_exp,
-                                     tmp_oa, &si->lsm, si->bufct,
-                                     &ioarr[shift], oti);
-                        if (rc)
-                                GOTO(out_oa, rc);
+        rc = lov_prep_brw_set(exp, src_oa, lsm, oa_bufs, pga, oti, &set);
+        if (rc)
+                RETURN(rc);
 
-                        lov_merge_attrs(ret_oa, tmp_oa, tmp_oa->o_valid, lsm,
-                                        i, &set);
-                }
+        list_for_each (pos, &set->set_list) {
+                struct obd_export *sub_exp;
+                struct brw_page *sub_pga;
+                req = list_entry(pos, struct lov_request, rq_link);
+                
+                sub_exp = lov->tgts[req->rq_idx].ltd_exp;
+                sub_pga = set->set_pga + req->rq_pgaidx;
+                rc = obd_brw(cmd, sub_exp, req->rq_oa, req->rq_md, 
+                             req->rq_oabufs, sub_pga, oti);
+                if (rc)
+                        break;
+                lov_update_common_set(set, req, rc);
         }
 
-        ret_oa->o_id = src_oa->o_id;
-        memcpy(src_oa, ret_oa, sizeof(*src_oa));
-
-        GOTO(out_oa, rc);
- out_oa:
-        if (tmp_oa)
-                obdo_free(tmp_oa);
-        if (ret_oa)
-                obdo_free(ret_oa);
- out_ioarr:
-        OBD_FREE(ioarr, sizeof(*ioarr) * oa_bufs);
- out_where:
-        OBD_FREE(where, sizeof(*where) * oa_bufs);
- out_sinfo:
-        OBD_FREE(stripeinfo, stripe_count * sizeof(*stripeinfo));
-        return rc;
+        err = lov_fini_brw_set(set);
+        if (!rc)
+                rc = err;
+        RETURN(rc);
 }
 
-static int lov_brw_interpret(struct ptlrpc_request_set *set, void *data, int rc)
+static int lov_brw_interpret(struct ptlrpc_request_set *reqset, void *data,
+                             int rc)
 {
-        struct lov_brw_async_args *aa = data;
-        struct lov_stripe_md *lsm = aa->aa_lsm;
-        struct obdo          *obdos = aa->aa_obdos;
-        struct lov_oinfo     *loi;
-        int i = 0;
+        struct lov_request_set *lovset = (struct lov_request_set *)data;
         ENTRY;
-
-        for (loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++, loi++) {
-                if (obdos[i].o_valid & OBD_MD_FLBLOCKS)
-                        loi->loi_blocks = obdos[i].o_blocks;
-                if (obdos[i].o_valid & OBD_MD_FLMTIME)
-                        loi->loi_mtime = obdos[i].o_mtime;
+        
+        if (rc) {
+                lovset->set_completes = 0;
+                lov_fini_brw_set(lovset);
+        } else {
+                rc = lov_fini_brw_set(lovset);
         }
-
-        OBD_FREE(obdos, lsm->lsm_stripe_count * sizeof(*obdos));
-        OBD_FREE(aa->aa_ioarr, sizeof(*aa->aa_ioarr) * aa->aa_oa_bufs);
+                
         RETURN(rc);
 }
 
@@ -1622,154 +933,64 @@ static int lov_brw_async(int cmd, struct obd_export *exp, struct obdo *oa,
                          struct brw_page *pga, struct ptlrpc_request_set *set,
                          struct obd_trans_info *oti)
 {
-        struct {
-                int bufct;
-                int index;
-                int subcount;
-                struct lov_stripe_md lsm;
-                int ost_idx;
-        } *stripeinfo, *si, *si_last;
-        struct lov_obd *lov;
-        struct brw_page *ioarr;
-        struct obdo *obdos = NULL;
-        struct lov_oinfo *loi;
-        struct lov_brw_async_args *aa;
-        int rc = 0, i, *where, stripe_count = lsm->lsm_stripe_count;
+        struct lov_request_set *lovset;
+        struct lov_request *req;
+        struct list_head *pos;
+        struct lov_obd *lov = &exp->exp_obd->u.lov;
+        int rc = 0;
         ENTRY;
 
-        if (lsm_bad_magic(lsm))
-                RETURN(-EINVAL);
-
-        lov = &exp->exp_obd->u.lov;
+        ASSERT_LSM_MAGIC(lsm);
 
         if (cmd == OBD_BRW_CHECK) {
                 rc = lov_brw_check(lov, oa, lsm, oa_bufs, pga);
                 RETURN(rc);
         }
 
-        OBD_ALLOC(stripeinfo, stripe_count * sizeof(*stripeinfo));
-        if (!stripeinfo)
-                RETURN(-ENOMEM);
-
-        OBD_ALLOC(where, sizeof(*where) * oa_bufs);
-        if (!where)
-                GOTO(out_sinfo, rc = -ENOMEM);
-
-        if (oa) {
-                OBD_ALLOC(obdos, sizeof(*obdos) * stripe_count);
-                if (!obdos)
-                        GOTO(out_where, rc = -ENOMEM);
-        }
-
-        OBD_ALLOC(ioarr, sizeof(*ioarr) * oa_bufs);
-        if (!ioarr)
-                GOTO(out_obdos, rc = -ENOMEM);
-
-        for (i = 0; i < oa_bufs; i++) {
-                where[i] = lov_stripe_number(lsm, pga[i].off);
-                stripeinfo[where[i]].bufct++;
-        }
-
-        for (i = 0, loi = lsm->lsm_oinfo, si_last = si = stripeinfo;
-             i < stripe_count; i++, loi++, si_last = si, si++) {
-                if (i > 0)
-                        si->index = si_last->index + si_last->bufct;
-                si->lsm.lsm_object_id = loi->loi_id;
-                si->ost_idx = loi->loi_ost_idx;
-
-                if (oa) {
-                        memcpy(&obdos[i], oa, sizeof(*obdos));
-                        obdos[i].o_id = si->lsm.lsm_object_id;
-                }
-        }
-
-        for (i = 0; i < oa_bufs; i++) {
-                int which = where[i];
-                int shift;
-
-                shift = stripeinfo[which].index + stripeinfo[which].subcount;
-                LASSERT(shift < oa_bufs);
-                ioarr[shift] = pga[i];
-                lov_stripe_offset(lsm, pga[i].off, which, &ioarr[shift].off);
-                stripeinfo[which].subcount++;
-        }
-
-        for (i = 0, si = stripeinfo; i < stripe_count; i++, si++) {
-                int shift = si->index;
-
-                if (si->bufct == 0)
-                        continue;
-
-                if (lov->tgts[si->ost_idx].active == 0) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", si->ost_idx);
-                        GOTO(out_ioarr, rc = -EIO);
-                }
-
-                LASSERT(shift < oa_bufs);
+        rc = lov_prep_brw_set(exp, oa, lsm, oa_bufs, pga, oti, &lovset);
+        if (rc)
+                RETURN(rc);
 
-                rc = obd_brw_async(cmd, lov->tgts[si->ost_idx].ltd_exp,
-                                   &obdos[i], &si->lsm, si->bufct,
-                                   &ioarr[shift], set, oti);
+        list_for_each (pos, &lovset->set_list) {
+                struct obd_export *sub_exp;
+                struct brw_page *sub_pga;
+                req = list_entry(pos, struct lov_request, rq_link);
+                
+                sub_exp = lov->tgts[req->rq_idx].ltd_exp;
+                sub_pga = lovset->set_pga + req->rq_pgaidx;
+                rc = obd_brw_async(cmd, sub_exp, req->rq_oa, req->rq_md,
+                                   req->rq_oabufs, sub_pga, set, oti);
                 if (rc)
-                        GOTO(out_ioarr, rc);
+                        GOTO(out, rc);
+                lov_update_common_set(lovset, req, rc);
         }
         LASSERT(rc == 0);
         LASSERT(set->set_interpret == NULL);
         set->set_interpret = (set_interpreter_func)lov_brw_interpret;
-        LASSERT(sizeof(set->set_args) >= sizeof(struct lov_brw_async_args));
-        aa = (struct lov_brw_async_args *)&set->set_args;
-        aa->aa_lsm = lsm;
-        aa->aa_obdos = obdos;
-        aa->aa_oa = oa;
-        aa->aa_ioarr = ioarr;
-        aa->aa_oa_bufs = oa_bufs;
-
-        /* Don't free ioarr or obdos - that's done in lov_brw_interpret */
-        GOTO(out_where, rc);
-
- out_ioarr:
-        OBD_FREE(ioarr, sizeof(*ioarr) * oa_bufs);
- out_obdos:
-        OBD_FREE(obdos, stripe_count * sizeof(*obdos));
- out_where:
-        OBD_FREE(where, sizeof(*where) * oa_bufs);
- out_sinfo:
-        OBD_FREE(stripeinfo, stripe_count * sizeof(*stripeinfo));
-        return rc;
+        set->set_arg = (void *)lovset;
+        
+        RETURN(rc);
+out:
+        lov_fini_brw_set(lovset);
+        RETURN(rc);
 }
 
-struct lov_async_page *lap_from_cookie(void *cookie)
-{
-        struct lov_async_page *lap = cookie;
-        if (lap->lap_magic != LAP_MAGIC)
-                return ERR_PTR(-EINVAL);
-        return lap;
-};
-
 static int lov_ap_make_ready(void *data, int cmd)
 {
-        struct lov_async_page *lap = lap_from_cookie(data);
-        /* XXX should these assert? */
-        if (IS_ERR(lap))
-                return -EINVAL;
+        struct lov_async_page *lap = LAP_FROM_COOKIE(data);
 
         return lap->lap_caller_ops->ap_make_ready(lap->lap_caller_data, cmd);
 }
 static int lov_ap_refresh_count(void *data, int cmd)
 {
-        struct lov_async_page *lap = lap_from_cookie(data);
-        if (IS_ERR(lap))
-                return -EINVAL;
+        struct lov_async_page *lap = LAP_FROM_COOKIE(data);
 
         return lap->lap_caller_ops->ap_refresh_count(lap->lap_caller_data,
                                                      cmd);
 }
 static void lov_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
 {
-        struct lov_async_page *lap = lap_from_cookie(data);
-        /* XXX should these assert? */
-        if (IS_ERR(lap))
-                return;
+        struct lov_async_page *lap = LAP_FROM_COOKIE(data);
 
         lap->lap_caller_ops->ap_fill_obdo(lap->lap_caller_data, cmd, oa);
         /* XXX woah, shouldn't we be altering more here?  size? */
@@ -1778,9 +999,7 @@ static void lov_ap_fill_obdo(void *data, int cmd, struct obdo *oa)
 
 static void lov_ap_completion(void *data, int cmd, struct obdo *oa, int rc)
 {
-        struct lov_async_page *lap = lap_from_cookie(data);
-        if (IS_ERR(lap))
-                return;
+        struct lov_async_page *lap = LAP_FROM_COOKIE(data);
 
         /* in a raid1 regime this would down a count of many ios
          * in flight, onl calling the caller_ops completion when all
@@ -1805,8 +1024,7 @@ int lov_prep_async_page(struct obd_export *exp, struct lov_stripe_md *lsm,
         int rc;
         ENTRY;
 
-        if (lsm_bad_magic(lsm))
-                RETURN(-EINVAL);
+        ASSERT_LSM_MAGIC(lsm);
         LASSERT(loi == NULL);
 
         OBD_ALLOC(lap, sizeof(*lap));
@@ -1851,12 +1069,9 @@ static int lov_queue_async_io(struct obd_export *exp,
 
         LASSERT(loi == NULL);
 
-        if (lsm_bad_magic(lsm))
-                RETURN(-EINVAL);
+        ASSERT_LSM_MAGIC(lsm);
 
-        lap = lap_from_cookie(cookie);
-        if (IS_ERR(lap))
-                RETURN(PTR_ERR(lap));
+        lap = LAP_FROM_COOKIE(cookie);
 
         loi = &lsm->lsm_oinfo[lap->lap_stripe];
         rc = obd_queue_async_io(lov->tgts[loi->loi_ost_idx].ltd_exp, lsm,
@@ -1876,12 +1091,9 @@ static int lov_set_async_flags(struct obd_export *exp,
 
         LASSERT(loi == NULL);
 
-        if (lsm_bad_magic(lsm))
-                RETURN(-EINVAL);
+        ASSERT_LSM_MAGIC(lsm);
 
-        lap = lap_from_cookie(cookie);
-        if (IS_ERR(lap))
-                RETURN(PTR_ERR(lap));
+        lap = LAP_FROM_COOKIE(cookie);
 
         loi = &lsm->lsm_oinfo[lap->lap_stripe];
         rc = obd_set_async_flags(lov->tgts[loi->loi_ost_idx].ltd_exp,
@@ -1902,12 +1114,9 @@ static int lov_queue_group_io(struct obd_export *exp,
 
         LASSERT(loi == NULL);
 
-        if (lsm_bad_magic(lsm))
-                RETURN(-EINVAL);
+        ASSERT_LSM_MAGIC(lsm);
 
-        lap = lap_from_cookie(cookie);
-        if (IS_ERR(lap))
-                RETURN(PTR_ERR(lap));
+        lap = LAP_FROM_COOKIE(cookie);
 
         loi = &lsm->lsm_oinfo[lap->lap_stripe];
         rc = obd_queue_group_io(lov->tgts[loi->loi_ost_idx].ltd_exp, lsm, loi,
@@ -1929,8 +1138,7 @@ static int lov_trigger_group_io(struct obd_export *exp,
 
         LASSERT(loi == NULL);
 
-        if (lsm_bad_magic(lsm))
-                RETURN(-EINVAL);
+        ASSERT_LSM_MAGIC(lsm);
 
         for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
              i++, loi++) {
@@ -1952,12 +1160,9 @@ static int lov_teardown_async_page(struct obd_export *exp,
 
         LASSERT(loi == NULL);
 
-        if (lsm_bad_magic(lsm))
-                RETURN(-EINVAL);
+        ASSERT_LSM_MAGIC(lsm);
 
-        lap = lap_from_cookie(cookie);
-        if (IS_ERR(lap))
-                RETURN(PTR_ERR(lap));
+        lap = LAP_FROM_COOKIE(cookie);
 
         loi = &lsm->lsm_oinfo[lap->lap_stripe];
         rc = obd_teardown_async_page(lov->tgts[loi->loi_ost_idx].ltd_exp,
@@ -1977,18 +1182,16 @@ static int lov_enqueue(struct obd_export *exp, struct lov_stripe_md *lsm,
                        void *data,__u32 lvb_len, void *lvb_swabber,
                        struct lustre_handle *lockh)
 {
-        struct lov_lock_handles *lov_lockh = NULL;
+        struct lov_request_set *set;
+        struct lov_request *req;
+        struct list_head *pos;
         struct lustre_handle *lov_lockhp;
         struct lov_obd *lov;
-        struct lov_oinfo *loi;
-        char submd_buf[sizeof(struct lov_stripe_md) + sizeof(struct lov_oinfo)];
-        struct lov_stripe_md *submd = (void *)submd_buf;
         ldlm_error_t rc;
-        int i, save_flags = *flags, all_skipped = 1;
+        int save_flags = *flags;
         ENTRY;
 
-        if (lsm_bad_magic(lsm))
-                RETURN(-EINVAL);
+        ASSERT_LSM_MAGIC(lsm);
 
         /* we should never be asked to replay a lock this way. */
         LASSERT((*flags & LDLM_FL_REPLAY) == 0);
@@ -1996,229 +1199,74 @@ static int lov_enqueue(struct obd_export *exp, struct lov_stripe_md *lsm,
         if (!exp || !exp->exp_obd)
                 RETURN(-ENODEV);
 
-        if (lsm->lsm_stripe_count > 1) {
-                lov_lockh = lov_llh_new(lsm);
-                if (lov_lockh == NULL)
-                        RETURN(-ENOMEM);
-
-                lockh->cookie = lov_lockh->llh_handle.h_cookie;
-                lov_lockhp = lov_lockh->llh_handles;
-        } else {
-                lov_lockhp = lockh;
-        }
-
         lov = &exp->exp_obd->u.lov;
-        for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
-             i++, loi++, lov_lockhp++) {
-                ldlm_policy_data_t sub_ext;
-
-                if (!lov_stripe_intersects(lsm, i, policy->l_extent.start,
-                                           policy->l_extent.end,
-                                           &sub_ext.l_extent.start,
-                                           &sub_ext.l_extent.end))
-                        continue;
+        rc = lov_prep_enqueue_set(exp, lsm, policy, mode, lockh, &set);
+        if (rc)
+                RETURN(rc);
 
-                if (lov->tgts[loi->loi_ost_idx].active == 0) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
-                        continue;
-                }
+        list_for_each (pos, &set->set_list) {
+                ldlm_policy_data_t sub_policy;
+                req = list_entry(pos, struct lov_request, rq_link);
+                lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
+                LASSERT(lov_lockhp);
 
-                all_skipped = 0;
-
-                /* XXX LOV STACKING: submd should be from the subobj */
-                submd->lsm_object_id = loi->loi_id;
-                submd->lsm_stripe_count = 0;
-                submd->lsm_oinfo->loi_kms_valid = loi->loi_kms_valid;
-                submd->lsm_oinfo->loi_rss = loi->loi_rss;
-                submd->lsm_oinfo->loi_kms = loi->loi_kms;
-                submd->lsm_oinfo->loi_blocks = loi->loi_blocks;
-                submd->lsm_oinfo->loi_mtime = loi->loi_mtime;
-                /* XXX submd is not fully initialized here */
                 *flags = save_flags;
-                rc = obd_enqueue(lov->tgts[loi->loi_ost_idx].ltd_exp, submd,
-                                 type, &sub_ext, mode, flags, bl_cb, cp_cb,
-                                 gl_cb, data, lvb_len, lvb_swabber, lov_lockhp);
-
-                /* XXX FIXME: This unpleasantness doesn't belong here at *all*.
-                 * It belongs in the OSC, except that the OSC doesn't have
-                 * access to the real LOI -- it gets a copy, that we created
-                 * above, and that copy can be arbitrarily out of date.
-                 *
-                 * The LOV API is due for a serious rewriting anyways, and this
-                 * can be addressed then. */
-                if (rc == ELDLM_OK) {
-                        struct ldlm_lock *lock = ldlm_handle2lock(lov_lockhp);
-                        __u64 tmp = submd->lsm_oinfo->loi_rss;
-
-                        LASSERT(lock != NULL);
-                        loi->loi_rss = tmp;
-                        loi->loi_mtime = submd->lsm_oinfo->loi_mtime;
-                        loi->loi_blocks = submd->lsm_oinfo->loi_blocks;
-                        /* Extend KMS up to the end of this lock and no further
-                         * A lock on [x,y] means a KMS of up to y + 1 bytes! */
-                        if (tmp > lock->l_policy_data.l_extent.end)
-                                tmp = lock->l_policy_data.l_extent.end + 1;
-                        if (tmp >= loi->loi_kms) {
-                                LDLM_DEBUG(lock, "acquired set stripe %d rss="
-                                           LPU64", kms="LPU64"\n", i,
-                                           loi->loi_rss, tmp);
-                                loi->loi_kms = tmp;
-                                loi->loi_kms_valid = 1;
-                        } else {
-                                LDLM_DEBUG(lock, "acquired, set stripe %d rss="
-                                           LPU64"; leaving kms="LPU64", end="
-                                           LPU64"\n", i, loi->loi_rss,
-                                           loi->loi_kms,
-                                           lock->l_policy_data.l_extent.end);
-                        }
-                        ldlm_lock_allow_match(lock);
-                        LDLM_LOCK_PUT(lock);
-                } else if (rc == ELDLM_LOCK_ABORTED &&
-                           save_flags & LDLM_FL_HAS_INTENT) {
-                        memset(lov_lockhp, 0, sizeof(*lov_lockhp));
-                        loi->loi_rss = submd->lsm_oinfo->loi_rss;
-                        loi->loi_mtime = submd->lsm_oinfo->loi_mtime;
-                        loi->loi_blocks = submd->lsm_oinfo->loi_blocks;
-                        CDEBUG(D_DLMTRACE, "glimpsed, set stripe %d rss="LPU64
-                               "; leaving kms="LPU64"\n", i, loi->loi_rss,
-                               loi->loi_kms);
-                } else {
-                        memset(lov_lockhp, 0, sizeof(*lov_lockhp));
-                        if (lov->tgts[loi->loi_ost_idx].active) {
-                                CERROR("error: enqueue objid "LPX64" subobj "
-                                       LPX64" stripe %d idx %d: rc = %d\n",
-                                       lsm->lsm_object_id, loi->loi_id,
-                                       i, loi->loi_ost_idx, rc);
-                                GOTO(out_locks, rc);
-                        }
-                }
-        }
-        if (all_skipped)
-                GOTO(out_lockh, rc = -EIO);
-
-        if (lsm->lsm_stripe_count > 1)
-                lov_llh_put(lov_lockh);
-        RETURN(ELDLM_OK);
-
- out_locks:
-        while (loi--, lov_lockhp--, i-- > 0) {
-                struct lov_stripe_md submd;
-                int err;
-
-                if (lov_lockhp->cookie == 0)
-                        continue;
-
-                /* XXX LOV STACKING: submd should be from the subobj */
-                submd.lsm_object_id = loi->loi_id;
-                submd.lsm_stripe_count = 0;
-                err = obd_cancel(lov->tgts[loi->loi_ost_idx].ltd_exp, &submd,
-                                 mode, lov_lockhp);
-                if (err && lov->tgts[loi->loi_ost_idx].active) {
-                        CERROR("error: cancelling objid "LPX64" on OST "
-                               "idx %d after enqueue error: rc = %d\n",
-                               loi->loi_id, loi->loi_ost_idx, err);
-                }
+                sub_policy.l_extent.start = req->rq_extent.start;
+                sub_policy.l_extent.end = req->rq_extent.end;
+
+                rc = obd_enqueue(lov->tgts[req->rq_idx].ltd_exp, req->rq_md,
+                                 type, &sub_policy, mode, flags, bl_cb,
+                                 cp_cb, gl_cb, data, lvb_len, lvb_swabber,
+                                 lov_lockhp);
+                rc = lov_update_enqueue_set(set, req, rc, save_flags);
+                if (rc != ELDLM_OK)
+                        break;
         }
 
-out_lockh:
-        if (lsm->lsm_stripe_count > 1) {
-                lov_llh_destroy(lov_lockh);
-                lov_llh_put(lov_lockh);
-        }
-        return rc;
+        lov_fini_enqueue_set(set, mode);
+        RETURN(rc);
 }
 
 static int lov_match(struct obd_export *exp, struct lov_stripe_md *lsm,
                      __u32 type, ldlm_policy_data_t *policy, __u32 mode,
                      int *flags, void *data, struct lustre_handle *lockh)
 {
-        struct lov_lock_handles *lov_lockh = NULL;
+        struct lov_request_set *set;
+        struct lov_request *req;
+        struct list_head *pos;
+        struct lov_obd *lov = &exp->exp_obd->u.lov;
         struct lustre_handle *lov_lockhp;
-        struct lov_obd *lov;
-        struct lov_oinfo *loi;
-        struct lov_stripe_md submd;
-        ldlm_error_t rc = 0;
-        int i;
+        int lov_flags, rc = 0;
         ENTRY;
 
-        if (lsm_bad_magic(lsm))
-                RETURN(-EINVAL);
+        ASSERT_LSM_MAGIC(lsm);
 
         if (!exp || !exp->exp_obd)
                 RETURN(-ENODEV);
 
-        if (lsm->lsm_stripe_count > 1) {
-                lov_lockh = lov_llh_new(lsm);
-                if (lov_lockh == NULL)
-                        RETURN(-ENOMEM);
-
-                lockh->cookie = lov_lockh->llh_handle.h_cookie;
-                lov_lockhp = lov_lockh->llh_handles;
-        } else {
-                lov_lockhp = lockh;
-        }
-
         lov = &exp->exp_obd->u.lov;
-        for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
-             i++, loi++, lov_lockhp++) {
-                ldlm_policy_data_t sub_ext;
-                int lov_flags;
-
-                if (!lov_stripe_intersects(lsm, i, policy->l_extent.start,
-                                           policy->l_extent.end,
-                                           &sub_ext.l_extent.start,
-                                           &sub_ext.l_extent.end))
-                        continue;
+        rc = lov_prep_match_set(exp, lsm, policy, mode, lockh, &set);
+        if (rc)
+                RETURN(rc);
 
-                if (lov->tgts[loi->loi_ost_idx].active == 0) {
-                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
-                        rc = -EIO;
-                        break;
-                }
+        list_for_each (pos, &set->set_list) {
+                ldlm_policy_data_t sub_policy;
+                req = list_entry(pos, struct lov_request, rq_link);
+                lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
+                LASSERT(lov_lockhp);
 
-                /* XXX LOV STACKING: submd should be from the subobj */
-                submd.lsm_object_id = loi->loi_id;
-                submd.lsm_stripe_count = 0;
+                sub_policy.l_extent.start = req->rq_extent.start;
+                sub_policy.l_extent.end = req->rq_extent.end;
                 lov_flags = *flags;
-                /* XXX submd is not fully initialized here */
-                rc = obd_match(lov->tgts[loi->loi_ost_idx].ltd_exp, &submd,
-                               type, &sub_ext, mode, &lov_flags, data,
+
+                rc = obd_match(lov->tgts[req->rq_idx].ltd_exp, req->rq_md,
+                               type, &sub_policy, mode, &lov_flags, data,
                                lov_lockhp);
+                rc = lov_update_match_set(set, req, rc);
                 if (rc != 1)
                         break;
         }
-        if (rc == 1) {
-                if (lsm->lsm_stripe_count > 1) {
-                        if (*flags & LDLM_FL_TEST_LOCK)
-                                lov_llh_destroy(lov_lockh);
-                        lov_llh_put(lov_lockh);
-                }
-                RETURN(1);
-        }
-
-        while (loi--, lov_lockhp--, i-- > 0) {
-                struct lov_stripe_md submd;
-                int err;
-
-                if (lov_lockhp->cookie == 0)
-                        continue;
-
-                /* XXX LOV STACKING: submd should be from the subobj */
-                submd.lsm_object_id = loi->loi_id;
-                submd.lsm_stripe_count = 0;
-                err = obd_cancel(lov->tgts[loi->loi_ost_idx].ltd_exp, &submd,
-                                 mode, lov_lockhp);
-                if (err && lov->tgts[loi->loi_ost_idx].active) {
-                        CERROR("error: cancelling objid "LPX64" on OST "
-                               "idx %d after match failure: rc = %d\n",
-                               loi->loi_id, loi->loi_ost_idx, err);
-                }
-        }
-
-        if (lsm->lsm_stripe_count > 1) {
-                lov_llh_destroy(lov_lockh);
-                lov_llh_put(lov_lockh);
-        }
+        lov_fini_match_set(set, mode, *flags);
         RETURN(rc);
 }
 
@@ -2231,8 +1279,7 @@ static int lov_change_cbdata(struct obd_export *exp,
         int rc = 0, i;
         ENTRY;
 
-        if (lsm_bad_magic(lsm))
-                RETURN(-EINVAL);
+        ASSERT_LSM_MAGIC(lsm);
 
         if (!exp || !exp->exp_obd)
                 RETURN(-ENODEV);
@@ -2254,66 +1301,43 @@ static int lov_change_cbdata(struct obd_export *exp,
 static int lov_cancel(struct obd_export *exp, struct lov_stripe_md *lsm,
                       __u32 mode, struct lustre_handle *lockh)
 {
-        struct lov_lock_handles *lov_lockh = NULL;
+        struct lov_request_set *set;
+        struct lov_request *req;
+        struct list_head *pos;
+        struct lov_obd *lov = &exp->exp_obd->u.lov;
         struct lustre_handle *lov_lockhp;
-        struct lov_obd *lov;
-        struct lov_oinfo *loi;
-        int rc = 0, i;
+        int err = 0, rc = 0;
         ENTRY;
 
-        if (lsm_bad_magic(lsm))
-                RETURN(-EINVAL);
+        ASSERT_LSM_MAGIC(lsm);
 
         if (!exp || !exp->exp_obd)
                 RETURN(-ENODEV);
 
         LASSERT(lockh);
-        if (lsm->lsm_stripe_count > 1) {
-                lov_lockh = lov_handle2llh(lockh);
-                if (!lov_lockh) {
-                        CERROR("LOV: invalid lov lock handle %p\n", lockh);
-                        RETURN(-EINVAL);
-                }
-
-                lov_lockhp = lov_lockh->llh_handles;
-        } else {
-                lov_lockhp = lockh;
-        }
-
         lov = &exp->exp_obd->u.lov;
-        for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
-             i++, loi++, lov_lockhp++) {
-                struct lov_stripe_md submd;
-                int err;
+        rc = lov_prep_cancel_set(exp, lsm, mode, lockh, &set);
+        if (rc)
+                RETURN(rc);
 
-                if (lov_lockhp->cookie == 0) {
-                        CDEBUG(D_HA, "lov idx %d subobj "LPX64" no lock?\n",
-                               loi->loi_ost_idx, loi->loi_id);
-                        continue;
-                }
+        list_for_each (pos, &set->set_list) {
+                req = list_entry(pos, struct lov_request, rq_link);
+                lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
 
-                /* XXX LOV STACKING: submd should be from the subobj */
-                submd.lsm_object_id = loi->loi_id;
-                submd.lsm_stripe_count = 0;
-                err = obd_cancel(lov->tgts[loi->loi_ost_idx].ltd_exp, &submd,
-                                 mode, lov_lockhp);
-                if (err) {
-                        if (lov->tgts[loi->loi_ost_idx].active) {
-                                CERROR("error: cancel objid "LPX64" subobj "
-                                       LPX64" on OST idx %d: rc = %d\n",
-                                       lsm->lsm_object_id,
-                                       loi->loi_id, loi->loi_ost_idx, err);
-                                if (!rc)
-                                        rc = err;
-                        }
+                rc = obd_cancel(lov->tgts[req->rq_idx].ltd_exp, req->rq_md,
+                                mode, lov_lockhp);
+                rc = lov_update_common_set(set, req, rc);
+                if (rc) {
+                        CERROR("error: cancel objid "LPX64" subobj "
+                               LPX64" on OST idx %d: rc = %d\n",
+                               lsm->lsm_object_id,
+                               req->rq_md->lsm_object_id, req->rq_idx, rc);
+                        err = rc;
                 }
         }
-
-        if (lsm->lsm_stripe_count > 1)
-                lov_llh_destroy(lov_lockh);
-        if (lov_lockh != NULL)
-                lov_llh_put(lov_lockh);
-        RETURN(rc);
+        lov_fini_cancel_set(set);
+        RETURN(err);
 }
 
 static int lov_cancel_unused(struct obd_export *exp,
@@ -2324,8 +1348,7 @@ static int lov_cancel_unused(struct obd_export *exp,
         int rc = 0, i;
         ENTRY;
 
-        if (lsm_bad_magic(lsm))
-                RETURN(-EINVAL);
+        ASSERT_LSM_MAGIC(lsm);
 
         if (!exp || !exp->exp_obd)
                 RETURN(-ENODEV);
@@ -2574,8 +1597,8 @@ static int lov_get_info(struct obd_export *exp, __u32 keylen,
                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
                         if (!lov->tgts[i].active)
                                 continue;
-                        rc = obd_get_info(lov->tgts[i].ltd_exp, keylen, key,
-                                          &size, &(ids[i]));
+                        rc = obd_get_info(lov->tgts[i].ltd_exp,
+                                          keylen, key, &size, &(ids[i]));
                         if (rc != 0)
                                 RETURN(rc);
                 }
@@ -2595,7 +1618,7 @@ static int lov_set_info(struct obd_export *exp, obd_count keylen,
 {
         struct obd_device *obddev = class_exp2obd(exp);
         struct lov_obd *lov = &obddev->u.lov;
-        int i, rc = 0;
+        int i, rc = 0, err;
         ENTRY;
 
 #define KEY_IS(str) \
@@ -2605,22 +1628,18 @@ static int lov_set_info(struct obd_export *exp, obd_count keylen,
                 if (vallen != lov->desc.ld_tgt_count)
                         RETURN(-EINVAL);
                 for (i = 0; i < lov->desc.ld_tgt_count; i++) {
-                        int er;
-
                         /* initialize all OSCs, even inactive ones */
 
-                        er = obd_set_info(lov->tgts[i].ltd_exp, keylen, key,
-                                          sizeof(obd_id), ((obd_id*)val) + i);
+                        err = obd_set_info(lov->tgts[i].ltd_exp,
+                                          keylen, key, sizeof(obd_id),
+                                          ((obd_id*)val) + i);
                         if (!rc)
-                                rc = er;
+                                rc = err;
                 }
                 RETURN(rc);
         }
 
-        if (KEY_IS("growth_count")) {
-                if (vallen != sizeof(int))
-                        RETURN(-EINVAL);
-        } else if (KEY_IS("mds_conn") || KEY_IS("unlinked")) {
+        if (KEY_IS("mds_conn") || KEY_IS("unlinked")) {
                 if (vallen != 0)
                         RETURN(-EINVAL);
         } else {
@@ -2628,78 +1647,20 @@ static int lov_set_info(struct obd_export *exp, obd_count keylen,
         }
 
         for (i = 0; i < lov->desc.ld_tgt_count; i++) {
-                int er;
-
                 if (val && !obd_uuid_equals(val, &lov->tgts[i].uuid))
                         continue;
 
                 if (!val && !lov->tgts[i].active)
                         continue;
 
-                er = obd_set_info(lov->tgts[i].ltd_exp, keylen, key, vallen,
-                                   val);
+                err = obd_set_info(lov->tgts[i].ltd_exp,
+                                  keylen, key, vallen, val);
                 if (!rc)
-                        rc = er;
+                        rc = err;
         }
         RETURN(rc);
 #undef KEY_IS
-
-}
-
-/* Merge rss if @kms_only == 0
- *
- * Even when merging RSS, we will take the KMS value if it's larger.
- * This prevents getattr from stomping on dirty cached pages which
- * extend the file size. */
-__u64 lov_merge_size(struct lov_stripe_md *lsm, int kms_only)
-{
-        struct lov_oinfo *loi;
-        __u64 size = 0;
-        int i;
-
-        for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count;
-             i++, loi++) {
-                obd_size lov_size, tmpsize;
-
-                tmpsize = loi->loi_kms;
-                if (kms_only == 0 && loi->loi_rss > tmpsize)
-                        tmpsize = loi->loi_rss;
-
-                lov_size = lov_stripe_size(lsm, tmpsize, i);
-                if (lov_size > size)
-                        size = lov_size;
-        }
-
-        return size;
-}
-EXPORT_SYMBOL(lov_merge_size);
-
-/* Merge blocks */
-__u64 lov_merge_blocks(struct lov_stripe_md *lsm)
-{
-        struct lov_oinfo *loi;
-        __u64 blocks = 0;
-        int i;
-
-        for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++){
-                blocks += loi->loi_blocks;
-        }
-        return blocks;
-}
-EXPORT_SYMBOL(lov_merge_blocks);
-
-__u64 lov_merge_mtime(struct lov_stripe_md *lsm, __u64 current_time)
-{
-        struct lov_oinfo *loi;
-        int i;
-
-        for (i = 0, loi = lsm->lsm_oinfo; i < lsm->lsm_stripe_count; i++,loi++){
-                if (loi->loi_mtime > current_time)
-                        current_time = loi->loi_mtime;
-        }
-        return current_time;
 }
-EXPORT_SYMBOL(lov_merge_mtime);
 
 #if 0
 struct lov_multi_wait {
@@ -2720,8 +1681,7 @@ int lov_complete_many(struct obd_export *exp, struct lov_stripe_md *lsm,
         int rc = 0, i;
         ENTRY;
 
-        if (lsm_bad_magic(lsm))
-                RETURN(-EINVAL);
+        ASSERT_LSM_MAGIC(lsm);
 
         if (!exp || !exp->exp_obd)
                 RETURN(-ENODEV);
@@ -2792,25 +1752,7 @@ int lov_complete_many(struct obd_export *exp, struct lov_stripe_md *lsm,
 #endif
 
 int lov_increase_kms(struct obd_export *exp, struct lov_stripe_md *lsm,
-                      obd_off size)
-{
-        struct lov_oinfo *loi;
-        int stripe = 0;
-        __u64 kms;
-        ENTRY;
-
-        if (size > 0)
-                stripe = lov_stripe_number(lsm, size - 1);
-        kms = lov_size_to_stripe(lsm, size, stripe);
-        loi = &(lsm->lsm_oinfo[stripe]);
-
-        CDEBUG(D_INODE, "stripe %d KMS %sincreasing "LPU64"->"LPU64"\n",
-               stripe, kms > loi->loi_kms ? "" : "not ", loi->loi_kms, kms);
-        if (kms > loi->loi_kms)
-                loi->loi_kms = kms;
-        RETURN(0);
-}
-EXPORT_SYMBOL(lov_increase_kms);
+                      obd_off size);
 
 struct obd_ops lov_obd_ops = {
         .o_owner               = THIS_MODULE,
diff --git a/lustre/lov/lov_offset.c b/lustre/lov/lov_offset.c
new file mode 100644 (file)
index 0000000..66fad27
--- /dev/null
@@ -0,0 +1,240 @@
+/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
+ * vim:expandtab:shiftwidth=8:tabstop=8:
+ *
+ * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
+ *
+ *   This file is part of Lustre, http://www.lustre.org.
+ *
+ *   Lustre is free software; you can redistribute it and/or
+ *   modify it under the terms of version 2 of the GNU General Public
+ *   License as published by the Free Software Foundation.
+ *
+ *   Lustre is distributed in the hope that it will be useful,
+ *   but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *   GNU General Public License for more details.
+ *
+ *   You should have received a copy of the GNU General Public License
+ *   along with Lustre; if not, write to the Free Software
+ *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#ifndef EXPORT_SYMTAB
+# define EXPORT_SYMTAB
+#endif
+#define DEBUG_SUBSYSTEM S_LOV
+
+#ifdef __KERNEL__
+#include <asm/div64.h>
+#else
+#include <liblustre.h>
+#endif
+
+#include <linux/obd_class.h>
+#include <linux/obd_lov.h>
+
+#include "lov_internal.h"
+
+/* compute object size given "stripeno" and the ost size */
+obd_size lov_stripe_size(struct lov_stripe_md *lsm, obd_size ost_size,
+                         int stripeno)
+{
+        unsigned long ssize  = lsm->lsm_stripe_size;
+        unsigned long swidth = ssize * lsm->lsm_stripe_count;
+        unsigned long stripe_size;
+        obd_size lov_size;
+        ENTRY;
+
+        if (ost_size == 0)
+                RETURN(0);
+
+        /* do_div(a, b) returns a % b, and a = a / b */
+        stripe_size = do_div(ost_size, ssize);
+        if (stripe_size)
+                lov_size = ost_size * swidth + stripeno * ssize + stripe_size;
+        else
+                lov_size = (ost_size - 1) * swidth + (stripeno + 1) * ssize;
+
+        RETURN(lov_size);
+}
+
+/* we have an offset in file backed by an lov and want to find out where
+ * that offset lands in our given stripe of the file.  for the easy
+ * case where the offset is within the stripe, we just have to scale the
+ * offset down to make it relative to the stripe instead of the lov.
+ *
+ * the harder case is what to do when the offset doesn't intersect the
+ * stripe.  callers will want start offsets clamped ahead to the start
+ * of the nearest stripe in the file.  end offsets similarly clamped to the
+ * nearest ending byte of a stripe in the file:
+ *
+ * all this function does is move offsets to the nearest region of the
+ * stripe, and it does its work "mod" the full length of all the stripes.
+ * consider a file with 3 stripes:
+ *
+ *             S                                              E
+ * ---------------------------------------------------------------------
+ * |    0    |     1     |     2     |    0    |     1     |     2     |
+ * ---------------------------------------------------------------------
+ *
+ * to find stripe 1's offsets for S and E, it divides by the full stripe
+ * width and does its math in the context of a single set of stripes:
+ *
+ *             S         E
+ * -----------------------------------
+ * |    0    |     1     |     2     |
+ * -----------------------------------
+ *
+ * it'll notice that E is outside stripe 1 and clamp it to the end of the
+ * stripe, then multiply it back out by lov_off to give the real offsets in
+ * the stripe:
+ *
+ *   S                   E
+ * ---------------------------------------------------------------------
+ * |    1    |     1     |     1     |    1    |     1     |     1     |
+ * ---------------------------------------------------------------------
+ *
+ * it would have done similarly and pulled S forward to the start of a 1
+ * stripe if, say, S had landed in a 0 stripe.
+ *
+ * this rounding isn't always correct.  consider an E lov offset that lands
+ * on a 0 stripe, the "mod stripe width" math will pull it forward to the
+ * start of a 1 stripe, when in fact it wanted to be rounded back to the end
+ * of a previous 1 stripe.  this logic is handled by callers and this is why:
+ *
+ * this function returns < 0 when the offset was "before" the stripe and
+ * was moved forward to the start of the stripe in question;  0 when it
+ * falls in the stripe and no shifting was done; > 0 when the offset
+ * was outside the stripe and was pulled back to its final byte. */
+int lov_stripe_offset(struct lov_stripe_md *lsm, obd_off lov_off,
+                      int stripeno, obd_off *obd_off)
+{
+        unsigned long ssize  = lsm->lsm_stripe_size;
+        unsigned long swidth = ssize * lsm->lsm_stripe_count;
+        unsigned long stripe_off, this_stripe;
+        int ret = 0;
+
+        if (lov_off == OBD_OBJECT_EOF) {
+                *obd_off = OBD_OBJECT_EOF;
+                return 0;
+        }
+
+        /* do_div(a, b) returns a % b, and a = a / b */
+        stripe_off = do_div(lov_off, swidth);
+
+        this_stripe = stripeno * ssize;
+        if (stripe_off < this_stripe) {
+                stripe_off = 0;
+                ret = -1;
+        } else {
+                stripe_off -= this_stripe;
+
+                if (stripe_off >= ssize) {
+                        stripe_off = ssize;
+                        ret = 1;
+                }
+        }
+
+        *obd_off = lov_off * ssize + stripe_off;
+        return ret;
+}
+
+/* Given a whole-file size and a stripe number, give the file size which
+ * corresponds to the individual object of that stripe.
+ *
+ * This behaves basically in the same was as lov_stripe_offset, except that
+ * file sizes falling before the beginning of a stripe are clamped to the end
+ * of the previous stripe, not the beginning of the next:
+ *
+ *                                               S
+ * ---------------------------------------------------------------------
+ * |    0    |     1     |     2     |    0    |     1     |     2     |
+ * ---------------------------------------------------------------------
+ *
+ * if clamped to stripe 2 becomes:
+ *
+ *                                   S
+ * ---------------------------------------------------------------------
+ * |    0    |     1     |     2     |    0    |     1     |     2     |
+ * ---------------------------------------------------------------------
+ */
+obd_off lov_size_to_stripe(struct lov_stripe_md *lsm, obd_off file_size,
+                           int stripeno)
+{
+        unsigned long ssize  = lsm->lsm_stripe_size;
+        unsigned long swidth = ssize * lsm->lsm_stripe_count;
+        unsigned long stripe_off, this_stripe;
+
+        if (file_size == OBD_OBJECT_EOF)
+                return OBD_OBJECT_EOF;
+
+        /* do_div(a, b) returns a % b, and a = a / b */
+        stripe_off = do_div(file_size, swidth);
+
+        this_stripe = stripeno * ssize;
+        if (stripe_off < this_stripe) {
+                /* Move to end of previous stripe, or zero */
+                if (file_size > 0) {
+                        file_size--;
+                        stripe_off = ssize;
+                } else {
+                        stripe_off = 0;
+                }
+        } else {
+                stripe_off -= this_stripe;
+
+                if (stripe_off >= ssize) {
+                        /* Clamp to end of this stripe */
+                        stripe_off = ssize;
+                }
+        }
+
+        return (file_size * ssize + stripe_off);
+}
+
+/* given an extent in an lov and a stripe, calculate the extent of the stripe
+ * that is contained within the lov extent.  this returns true if the given
+ * stripe does intersect with the lov extent. */
+int lov_stripe_intersects(struct lov_stripe_md *lsm, int stripeno,
+                          obd_off start, obd_off end,
+                          obd_off *obd_start, obd_off *obd_end)
+{
+        int start_side, end_side;
+
+        start_side = lov_stripe_offset(lsm, start, stripeno, obd_start);
+        end_side = lov_stripe_offset(lsm, end, stripeno, obd_end);
+
+        CDEBUG(D_INODE, "["LPU64"->"LPU64"] -> [(%d) "LPU64"->"LPU64" (%d)]\n",
+               start, end, start_side, *obd_start, *obd_end, end_side);
+
+        /* this stripe doesn't intersect the file extent when neither
+         * start or the end intersected the stripe and obd_start and
+         * obd_end got rounded up to the save value. */
+        if (start_side != 0 && end_side != 0 && *obd_start == *obd_end)
+                return 0;
+
+        /* as mentioned in the lov_stripe_offset commentary, end
+         * might have been shifted in the wrong direction.  This
+         * happens when an end offset is before the stripe when viewed
+         * through the "mod stripe size" math. we detect it being shifted
+         * in the wrong direction and touch it up.
+         * interestingly, this can't underflow since end must be > start
+         * if we passed through the previous check.
+         * (should we assert for that somewhere?) */
+        if (end_side != 0)
+                (*obd_end)--;
+
+        return 1;
+}
+
+/* compute which stripe number "lov_off" will be written into */
+int lov_stripe_number(struct lov_stripe_md *lsm, obd_off lov_off)
+{
+        unsigned long ssize  = lsm->lsm_stripe_size;
+        unsigned long swidth = ssize * lsm->lsm_stripe_count;
+        unsigned long stripe_off;
+
+        stripe_off = do_div(lov_off, swidth);
+
+        return stripe_off / ssize;
+}
index 419f3ba..164bcec 100644 (file)
@@ -295,7 +295,7 @@ int lov_alloc_memmd(struct lov_stripe_md **lsmp, int stripe_count, int pattern)
         (*lsmp)->lsm_magic = LOV_MAGIC;
         (*lsmp)->lsm_stripe_count = stripe_count;
         (*lsmp)->lsm_maxbytes = LUSTRE_STRIPE_MAXBYTES * stripe_count;
-        (*lsmp)->lsm_xfersize = PTLRPC_MTU * stripe_count;
+        (*lsmp)->lsm_xfersize = PTLRPC_MAX_BRW_SIZE * stripe_count;
         (*lsmp)->lsm_pattern = pattern;
         (*lsmp)->lsm_oinfo[0].loi_ost_idx = ~0;
 
diff --git a/lustre/lov/lov_qos.c b/lustre/lov/lov_qos.c
new file mode 100644 (file)
index 0000000..0f6f012
--- /dev/null
@@ -0,0 +1,186 @@
+/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
+ * vim:expandtab:shiftwidth=8:tabstop=8:
+ *
+ * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
+ *
+ *   This file is part of Lustre, http://www.lustre.org.
+ *
+ *   Lustre is free software; you can redistribute it and/or
+ *   modify it under the terms of version 2 of the GNU General Public
+ *   License as published by the Free Software Foundation.
+ *
+ *   Lustre is distributed in the hope that it will be useful,
+ *   but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *   GNU General Public License for more details.
+ *
+ *   You should have received a copy of the GNU General Public License
+ *   along with Lustre; if not, write to the Free Software
+ *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#ifndef EXPORT_SYMTAB
+# define EXPORT_SYMTAB
+#endif
+#define DEBUG_SUBSYSTEM S_LOV
+
+#ifdef __KERNEL__
+#else
+#include <liblustre.h>
+#endif
+
+#include <linux/obd_class.h>
+#include <linux/obd_lov.h>
+
+#include "lov_internal.h"
+
+void qos_shrink_lsm(struct lov_request_set *set)
+{
+        struct lov_stripe_md *lsm = set->set_md;
+        struct lov_stripe_md *lsm_new;
+        /* XXX LOV STACKING call into osc for sizes */
+        unsigned oldsize, newsize;
+
+        if (set->set_oti && set->set_cookies && set->set_cookie_sent) {
+                struct llog_cookie *cookies;
+                oldsize = lsm->lsm_stripe_count * sizeof(*cookies);
+                newsize = set->set_count * sizeof(*cookies);
+
+                cookies = set->set_cookies;
+                oti_alloc_cookies(set->set_oti, set->set_count);
+                if (set->set_oti->oti_logcookies) {
+                        memcpy(set->set_oti->oti_logcookies, cookies, newsize);
+                        OBD_FREE(cookies, oldsize);
+                        set->set_cookies = set->set_oti->oti_logcookies;
+                } else {
+                        CWARN("'leaking' %d bytes\n", oldsize - newsize);
+                }
+        }
+
+        CWARN("using fewer stripes for object "LPX64": old %u new %u\n",
+              lsm->lsm_object_id, lsm->lsm_stripe_count, set->set_count);
+
+        oldsize = lov_stripe_md_size(lsm->lsm_stripe_count);
+        newsize = lov_stripe_md_size(set->set_count);
+        OBD_ALLOC(lsm_new, newsize);
+        if (lsm_new != NULL) {
+                memcpy(lsm_new, lsm, newsize);
+                lsm_new->lsm_stripe_count = set->set_count;
+                OBD_FREE(lsm, oldsize);
+                set->set_md = lsm_new;
+        } else {
+                CWARN("'leaking' %d bytes\n", oldsize - newsize);
+        }
+}
+
+#define LOV_CREATE_RESEED_INTERVAL 1000
+/* FIXME use real qos data to prepare the lov create request */
+int qos_prep_create(struct lov_obd *lov, struct lov_request_set *set, int newea)
+{
+        static int ost_start_idx, ost_start_count;
+        unsigned ost_idx, ost_count = lov->desc.ld_tgt_count;
+        struct lov_stripe_md *lsm = set->set_md;
+        struct obdo *src_oa = set->set_oa;
+        int i, rc = 0;
+        ENTRY;
+
+        LASSERT(src_oa->o_valid & OBD_MD_FLID);
+        
+        lsm->lsm_object_id = src_oa->o_id;
+        if (!lsm->lsm_stripe_size)
+                lsm->lsm_stripe_size = lov->desc.ld_default_stripe_size;
+        if (!lsm->lsm_pattern) {
+                lsm->lsm_pattern = lov->desc.ld_pattern ?
+                        lov->desc.ld_pattern : LOV_PATTERN_RAID0;
+        }
+
+        if (newea || lsm->lsm_oinfo[0].loi_ost_idx >= ost_count) {
+                if (--ost_start_count <= 0) {
+                        ost_start_idx = ll_insecure_random_int();
+                        ost_start_count = LOV_CREATE_RESEED_INTERVAL;
+                } else if (lsm->lsm_stripe_count >=
+                           lov->desc.ld_active_tgt_count) {
+                        /* If we allocate from all of the stripes, make the
+                         * next file start on the next OST. */
+                        ++ost_start_idx;
+                }
+                ost_idx = ost_start_idx % ost_count;
+        } else {
+                ost_idx = lsm->lsm_oinfo[0].loi_ost_idx;
+        }
+
+        CDEBUG(D_INODE, "allocating %d subobjs for objid "LPX64" at idx %d\n",
+               lsm->lsm_stripe_count, lsm->lsm_object_id, ost_idx);
+
+        for (i = 0; i < ost_count; i++, ost_idx = (ost_idx + 1) % ost_count) {
+                struct lov_request *req;
+                
+                ++ost_start_idx;
+                if (lov->tgts[ost_idx].active == 0) {
+                        CDEBUG(D_HA, "lov idx %d inactive\n", ost_idx);
+                        continue;
+                }
+
+                OBD_ALLOC(req, sizeof(*req));
+                if (req == NULL)
+                        GOTO(out, rc = -ENOMEM);
+                
+                req->rq_buflen = sizeof(*req->rq_md);
+                OBD_ALLOC(req->rq_md, req->rq_buflen);
+                if (req->rq_md == NULL)
+                        GOTO(out, rc = -ENOMEM);
+                
+                req->rq_oa = obdo_alloc();
+                if (req->rq_oa == NULL)
+                        GOTO(out, rc = -ENOMEM);
+                
+                req->rq_idx = ost_idx;
+                req->rq_stripe = i;
+                /* create data objects with "parent" OA */
+                memcpy(req->rq_oa, src_oa, sizeof(*req->rq_oa));
+
+                /* XXX When we start creating objects on demand, we need to
+                 *     make sure that we always create the object on the
+                 *     stripe which holds the existing file size.
+                 */
+                if (src_oa->o_valid & OBD_MD_FLSIZE) {
+                        if (lov_stripe_offset(lsm, src_oa->o_size, i,
+                                              &req->rq_oa->o_size) < 0 &&
+                            req->rq_oa->o_size)
+                                req->rq_oa->o_size--;
+
+                        CDEBUG(D_INODE, "stripe %d has size "LPU64"/"LPU64"\n",
+                               i, req->rq_oa->o_size, src_oa->o_size);
+                }
+
+                lov_set_add_req(req, set);
+
+                /* If we have allocated enough objects, we are OK */
+                if (set->set_count == lsm->lsm_stripe_count)
+                        GOTO(out, rc = 0);
+        }
+        
+        if (set->set_count == 0)
+                GOTO(out, rc = -EIO);
+        
+        /* If we were passed specific striping params, then a failure to
+         * meet those requirements is an error, since we can't reallocate
+         * that memory (it might be part of a larger array or something).
+         *
+         * We can only get here if lsm_stripe_count was originally > 1.
+         */
+        if (!newea) {
+                CERROR("can't lstripe objid "LPX64": have %u want %u, rc %d\n",
+                       lsm->lsm_object_id, set->set_count, 
+                       lsm->lsm_stripe_count, rc);
+                rc = rc ? rc : -EFBIG;
+        } else {
+                qos_shrink_lsm(set);
+                rc = 0;
+        }
+out:
+        RETURN(rc);
+}
+
+
+
diff --git a/lustre/lov/lov_request.c b/lustre/lov/lov_request.c
new file mode 100644 (file)
index 0000000..0f59335
--- /dev/null
@@ -0,0 +1,1293 @@
+/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
+ * vim:expandtab:shiftwidth=8:tabstop=8:
+ *
+ * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
+ *
+ *   This file is part of Lustre, http://www.lustre.org.
+ *
+ *   Lustre is free software; you can redistribute it and/or
+ *   modify it under the terms of version 2 of the GNU General Public
+ *   License as published by the Free Software Foundation.
+ *
+ *   Lustre is distributed in the hope that it will be useful,
+ *   but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *   GNU General Public License for more details.
+ *
+ *   You should have received a copy of the GNU General Public License
+ *   along with Lustre; if not, write to the Free Software
+ *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#ifndef EXPORT_SYMTAB
+# define EXPORT_SYMTAB
+#endif
+#define DEBUG_SUBSYSTEM S_LOV
+
+#ifdef __KERNEL__
+#include <asm/div64.h>
+#else
+#include <liblustre.h>
+#endif
+
+#include <linux/obd_class.h>
+#include <linux/obd_lov.h>
+#include <linux/lustre_idl.h>
+
+#include "lov_internal.h"
+
+static void lov_init_set(struct lov_request_set *set)
+{
+        set->set_count = 0;
+        set->set_completes = 0;
+        set->set_success = 0;
+        INIT_LIST_HEAD(&set->set_list);
+        atomic_set(&set->set_refcount, 1);
+}
+
+static void lov_finish_set(struct lov_request_set *set)
+{
+        struct list_head *pos, *n;
+        ENTRY;
+
+        LASSERT(set);
+        list_for_each_safe(pos, n, &set->set_list) {
+                struct lov_request *req = list_entry(pos, struct lov_request,
+                                                     rq_link);
+                list_del_init(&req->rq_link);
+
+                if (req->rq_oa)
+                        obdo_free(req->rq_oa);
+                if (req->rq_md)
+                        OBD_FREE(req->rq_md, req->rq_buflen);
+                OBD_FREE(req, sizeof(*req));
+        }
+
+        if (set->set_pga) {
+                int len = set->set_oabufs * sizeof(*set->set_pga);
+                OBD_FREE(set->set_pga, len);
+        }
+        if (set->set_lockh)
+                lov_llh_put(set->set_lockh);
+
+        OBD_FREE(set, sizeof(*set));
+        EXIT;
+}
+
+static void lov_update_set(struct lov_request_set *set,
+                           struct lov_request *req, int rc)
+{
+        req->rq_complete = 1;
+        req->rq_rc = rc;
+
+        set->set_completes++;
+        if (rc == 0)
+                set->set_success++;
+}
+
+int lov_update_common_set(struct lov_request_set *set,
+                          struct lov_request *req, int rc)
+{
+        struct lov_obd *lov = &set->set_exp->exp_obd->u.lov;
+        ENTRY;
+
+        lov_update_set(set, req, rc);
+
+        /* grace error on inactive ost */
+        if (rc && !lov->tgts[req->rq_idx].active)
+                rc = 0;
+
+        /* FIXME in raid1 regime, should return 0 */
+        RETURN(rc);
+}
+
+void lov_set_add_req(struct lov_request *req, struct lov_request_set *set)
+{
+        list_add_tail(&req->rq_link, &set->set_list);
+        set->set_count++;
+}
+
+int lov_update_enqueue_set(struct lov_request_set *set,
+                           struct lov_request *req, int rc, int flags)
+{
+        struct lustre_handle *lov_lockhp;
+        struct lov_oinfo *loi;
+        ENTRY;
+
+        lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
+        loi = &set->set_md->lsm_oinfo[req->rq_stripe];
+
+        /* XXX FIXME: This unpleasantness doesn't belong here at *all*.
+         * It belongs in the OSC, except that the OSC doesn't have
+         * access to the real LOI -- it gets a copy, that we created
+         * above, and that copy can be arbitrarily out of date.
+         *
+         * The LOV API is due for a serious rewriting anyways, and this
+         * can be addressed then. */
+        if (rc == ELDLM_OK) {
+                struct ldlm_lock *lock = ldlm_handle2lock(lov_lockhp);
+                __u64 tmp = req->rq_md->lsm_oinfo->loi_rss;
+
+                LASSERT(lock != NULL);
+                loi->loi_rss = tmp;
+                loi->loi_blocks = req->rq_md->lsm_oinfo->loi_blocks;
+                /* Extend KMS up to the end of this lock and no further
+                 * A lock on [x,y] means a KMS of up to y + 1 bytes! */
+                if (tmp > lock->l_policy_data.l_extent.end)
+                        tmp = lock->l_policy_data.l_extent.end + 1;
+                if (tmp >= loi->loi_kms) {
+                        CDEBUG(D_INODE, "lock acquired, setting rss="
+                               LPU64", kms="LPU64"\n", loi->loi_rss, tmp);
+                        loi->loi_kms = tmp;
+                        loi->loi_kms_valid = 1;
+                } else {
+                        CDEBUG(D_INODE, "lock acquired, setting rss="
+                               LPU64"; leaving kms="LPU64", end="LPU64
+                               "\n", loi->loi_rss, loi->loi_kms,
+                               lock->l_policy_data.l_extent.end);
+                }
+                ldlm_lock_allow_match(lock);
+                LDLM_LOCK_PUT(lock);
+        } else if (rc == ELDLM_LOCK_ABORTED && flags & LDLM_FL_HAS_INTENT) {
+                memset(lov_lockhp, 0, sizeof(*lov_lockhp));
+                loi->loi_rss = req->rq_md->lsm_oinfo->loi_rss;
+                loi->loi_blocks = req->rq_md->lsm_oinfo->loi_blocks;
+                CDEBUG(D_INODE, "glimpsed, setting rss="LPU64"; leaving"
+                       " kms="LPU64"\n", loi->loi_rss, loi->loi_kms);
+                rc = ELDLM_OK;
+        } else {
+                struct obd_export *exp = set->set_exp;
+                struct lov_obd *lov = &exp->exp_obd->u.lov;
+
+                memset(lov_lockhp, 0, sizeof(*lov_lockhp));
+                if (lov->tgts[req->rq_idx].active) {
+                        CERROR("error: enqueue objid "LPX64" subobj "
+                                LPX64" on OST idx %d: rc = %d\n",
+                                set->set_md->lsm_object_id, loi->loi_id,
+                                loi->loi_ost_idx, rc);
+                } else {
+                        rc = ELDLM_OK;
+                }
+        }
+        lov_update_set(set, req, rc);
+        RETURN(rc);
+}
+
+static int enqueue_done(struct lov_request_set *set, __u32 mode)
+{
+        struct list_head *pos;
+        struct lov_request *req;
+        struct lustre_handle *lov_lockhp = NULL;
+        struct lov_obd *lov = &set->set_exp->exp_obd->u.lov;
+        int rc = 0;
+        ENTRY;
+
+        LASSERT(set->set_completes);
+        /* enqueue/match success, just return */
+        if (set->set_completes == set->set_success)
+                RETURN(0);
+
+        /* cancel enqueued/matched locks */
+        list_for_each (pos, &set->set_list) {
+                req = list_entry(pos, struct lov_request, rq_link);
+
+                if (!req->rq_complete || req->rq_rc)
+                        continue;
+
+                lov_lockhp = set->set_lockh->llh_handles + req->rq_stripe;
+                LASSERT(lov_lockhp);
+                if (lov_lockhp->cookie == 0)
+                        continue;
+
+                rc = obd_cancel(lov->tgts[req->rq_idx].ltd_exp, req->rq_md,
+                                mode, lov_lockhp);
+                if (rc && lov->tgts[req->rq_idx].active)
+                        CERROR("cancelling obdjid "LPX64" on OST "
+                               "idx %d error: rc = %d\n",
+                               req->rq_md->lsm_object_id, req->rq_idx, rc);
+        }
+        lov_llh_put(set->set_lockh);
+        RETURN(rc);
+}
+
+int lov_fini_enqueue_set(struct lov_request_set *set, __u32 mode)
+{
+        int rc = 0;
+        ENTRY;
+
+        LASSERT(set->set_exp);
+        if (set == NULL)
+                RETURN(0);
+        if (set->set_completes)
+                rc = enqueue_done(set, mode);
+        else
+                lov_llh_put(set->set_lockh);
+
+        if (atomic_dec_and_test(&set->set_refcount))
+                lov_finish_set(set);
+
+        RETURN(rc);
+}
+
+int lov_prep_enqueue_set(struct obd_export *exp, struct lov_stripe_md *lsm,
+                         ldlm_policy_data_t *policy, __u32 mode,
+                         struct lustre_handle *lockh,
+                         struct lov_request_set **reqset)
+{
+        struct lov_obd *lov = &exp->exp_obd->u.lov;
+        struct lov_request_set *set;
+        int i, rc = 0;
+        struct lov_oinfo *loi;
+        ENTRY;
+
+        OBD_ALLOC(set, sizeof(*set));
+        if (set == NULL)
+                RETURN(-ENOMEM);
+        lov_init_set(set);
+
+        set->set_exp = exp;
+        set->set_md = lsm;
+        set->set_lockh = lov_llh_new(lsm);
+        if (set->set_lockh == NULL)
+                GOTO(out_set, rc = -ENOMEM);
+        lockh->cookie = set->set_lockh->llh_handle.h_cookie;
+
+        loi = lsm->lsm_oinfo;
+        for (i = 0; i < lsm->lsm_stripe_count; i++, loi++) {
+                struct lov_request *req;
+                obd_off start, end;
+
+                if (!lov_stripe_intersects(lsm, i, policy->l_extent.start,
+                                           policy->l_extent.end, &start, &end))
+                        continue;
+
+                if (lov->tgts[loi->loi_ost_idx].active == 0) {
+                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
+                        continue;
+                }
+
+                OBD_ALLOC(req, sizeof(*req));
+                if (req == NULL)
+                        GOTO(out_set, rc = -ENOMEM);
+
+                req->rq_buflen = sizeof(*req->rq_md) +
+                        sizeof(struct lov_oinfo);
+                OBD_ALLOC(req->rq_md, req->rq_buflen);
+                if (req->rq_md == NULL)
+                        GOTO(out_set, rc = -ENOMEM);
+
+                req->rq_extent.start = start;
+                req->rq_extent.end = end;
+
+                req->rq_idx = loi->loi_ost_idx;
+                req->rq_stripe = i;
+
+                /* XXX LOV STACKING: submd should be from the subobj */
+                req->rq_md->lsm_object_id = loi->loi_id;
+                req->rq_md->lsm_stripe_count = 0;
+                req->rq_md->lsm_oinfo->loi_kms_valid = loi->loi_kms_valid;
+                req->rq_md->lsm_oinfo->loi_rss = loi->loi_rss;
+                req->rq_md->lsm_oinfo->loi_kms = loi->loi_kms;
+                req->rq_md->lsm_oinfo->loi_blocks = loi->loi_blocks;
+
+                lov_set_add_req(req, set);
+        }
+        if (!set->set_count)
+                GOTO(out_set, rc = -EIO);
+        *reqset = set;
+        RETURN(0);
+out_set:
+        lov_fini_enqueue_set(set, mode);
+        RETURN(rc);
+}
+
+int lov_update_match_set(struct lov_request_set *set, struct lov_request *req,
+                         int rc)
+{
+        int ret = rc;
+        ENTRY;
+
+        if (rc == 1)
+                ret = 0;
+        lov_update_set(set, req, ret);
+        RETURN(rc);
+}
+
+int lov_fini_match_set(struct lov_request_set *set, __u32 mode, int flags)
+{
+        int rc = 0;
+        ENTRY;
+
+        LASSERT(set->set_exp);
+        if (set == NULL)
+                RETURN(0);
+        if (set->set_completes) {
+                if (set->set_count == set->set_success &&
+                    flags & LDLM_FL_TEST_LOCK)
+                        lov_llh_put(set->set_lockh);
+                rc = enqueue_done(set, mode);
+        } else {
+                lov_llh_put(set->set_lockh);
+        }
+
+        if (atomic_dec_and_test(&set->set_refcount))
+                lov_finish_set(set);
+
+        RETURN(rc);
+}
+
+int lov_prep_match_set(struct obd_export *exp, struct lov_stripe_md *lsm,
+                       ldlm_policy_data_t *policy, __u32 mode,
+                       struct lustre_handle *lockh,
+                       struct lov_request_set **reqset)
+{
+        struct lov_obd *lov = &exp->exp_obd->u.lov;
+        struct lov_request_set *set;
+        int i, rc = 0;
+        struct lov_oinfo *loi;
+        ENTRY;
+
+        OBD_ALLOC(set, sizeof(*set));
+        if (set == NULL)
+                RETURN(-ENOMEM);
+        lov_init_set(set);
+
+        set->set_exp = exp;
+        set->set_md = lsm;
+        set->set_lockh = lov_llh_new(lsm);
+        if (set->set_lockh == NULL)
+                GOTO(out_set, rc = -ENOMEM);
+        lockh->cookie = set->set_lockh->llh_handle.h_cookie;
+
+        loi = lsm->lsm_oinfo;
+        for (i = 0; i < lsm->lsm_stripe_count; i++, loi++) {
+                struct lov_request *req;
+                obd_off start, end;
+
+                if (!lov_stripe_intersects(lsm, i, policy->l_extent.start,
+                                           policy->l_extent.end, &start, &end))
+                        continue;
+
+                /* FIXME raid1 should grace this error */
+                if (lov->tgts[loi->loi_ost_idx].active == 0) {
+                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
+                        GOTO(out_set, rc = -EIO);
+                }
+
+                OBD_ALLOC(req, sizeof(*req));
+                if (req == NULL)
+                        GOTO(out_set, rc = -ENOMEM);
+
+                req->rq_buflen = sizeof(*req->rq_md);
+                OBD_ALLOC(req->rq_md, req->rq_buflen);
+                if (req->rq_md == NULL)
+                        GOTO(out_set, rc = -ENOMEM);
+
+                req->rq_extent.start = start;
+                req->rq_extent.end = end;
+
+                req->rq_idx = loi->loi_ost_idx;
+                req->rq_stripe = i;
+
+                /* XXX LOV STACKING: submd should be from the subobj */
+                req->rq_md->lsm_object_id = loi->loi_id;
+                req->rq_md->lsm_stripe_count = 0;
+
+                lov_set_add_req(req, set);
+        }
+        if (!set->set_count)
+                GOTO(out_set, rc = -EIO);
+        *reqset = set;
+        RETURN(rc);
+out_set:
+        lov_fini_match_set(set, mode, 0);
+        RETURN(rc);
+}
+
+int lov_fini_cancel_set(struct lov_request_set *set)
+{
+        int rc = 0;
+        ENTRY;
+
+        LASSERT(set->set_exp);
+        if (set == NULL)
+                RETURN(0);
+
+        if (set->set_lockh)
+                lov_llh_put(set->set_lockh);
+
+        if (atomic_dec_and_test(&set->set_refcount))
+                lov_finish_set(set);
+
+        RETURN(rc);
+}
+
+int lov_prep_cancel_set(struct obd_export *exp, struct lov_stripe_md *lsm,
+                        __u32 mode, struct lustre_handle *lockh,
+                        struct lov_request_set **reqset)
+{
+        struct lov_request_set *set;
+        int i, rc = 0;
+        struct lov_oinfo *loi;
+        ENTRY;
+
+        OBD_ALLOC(set, sizeof(*set));
+        if (set == NULL)
+                RETURN(-ENOMEM);
+        lov_init_set(set);
+
+        set->set_exp = exp;
+        set->set_md = lsm;
+        set->set_lockh = lov_handle2llh(lockh);
+        if (set->set_lockh == NULL) {
+                CERROR("LOV: invalid lov lock handle %p\n", lockh);
+                GOTO(out_set, rc = -EINVAL);
+        }
+        lockh->cookie = set->set_lockh->llh_handle.h_cookie;
+
+        loi = lsm->lsm_oinfo;
+        for (i = 0; i < lsm->lsm_stripe_count; i++, loi++) {
+                struct lov_request *req;
+                struct lustre_handle *lov_lockhp;
+
+                lov_lockhp = set->set_lockh->llh_handles + i;
+                if (lov_lockhp->cookie == 0) {
+                        CDEBUG(D_HA, "lov idx %d subobj "LPX64" no lock?\n",
+                               loi->loi_ost_idx, loi->loi_id);
+                        continue;
+                }
+
+                OBD_ALLOC(req, sizeof(*req));
+                if (req == NULL)
+                        GOTO(out_set, rc = -ENOMEM);
+
+                req->rq_buflen = sizeof(*req->rq_md);
+                OBD_ALLOC(req->rq_md, req->rq_buflen);
+                if (req->rq_md == NULL)
+                        GOTO(out_set, rc = -ENOMEM);
+
+                req->rq_idx = loi->loi_ost_idx;
+                req->rq_stripe = i;
+
+                /* XXX LOV STACKING: submd should be from the subobj */
+                req->rq_md->lsm_object_id = loi->loi_id;
+                req->rq_md->lsm_stripe_count = 0;
+
+                lov_set_add_req(req, set);
+        }
+        if (!set->set_count)
+                GOTO(out_set, rc = -EIO);
+        *reqset = set;
+        RETURN(rc);
+out_set:
+        lov_fini_cancel_set(set);
+        RETURN(rc);
+}
+
+static int create_done(struct obd_export *exp, struct lov_request_set *set,
+                       struct lov_stripe_md **ea)
+{
+        struct lov_obd *lov = &exp->exp_obd->u.lov;
+        struct obd_trans_info *oti = set->set_oti;
+        struct obdo *src_oa = set->set_oa;
+        struct list_head *pos;
+        struct lov_request *req;
+        struct obdo *ret_oa = NULL;
+        int attrset = 0, rc = 0;
+        ENTRY;
+
+        LASSERT(set->set_completes);
+
+        if (!set->set_success)
+                GOTO(cleanup, rc = -EIO);
+        if (*ea == NULL && set->set_count != set->set_success) {
+                set->set_count = set->set_success;
+                qos_shrink_lsm(set);
+        }
+
+        ret_oa = obdo_alloc();
+        if (ret_oa == NULL)
+                GOTO(cleanup, rc = -ENOMEM);
+
+        list_for_each (pos, &set->set_list) {
+                req = list_entry(pos, struct lov_request, rq_link);
+                if (!req->rq_complete || req->rq_rc)
+                        continue;
+                lov_merge_attrs(ret_oa, req->rq_oa, req->rq_oa->o_valid,
+                                set->set_md, req->rq_stripe, &attrset);
+        }
+        if (src_oa->o_valid & OBD_MD_FLSIZE &&
+            ret_oa->o_size != src_oa->o_size) {
+                CERROR("original size "LPU64" isn't new object size "LPU64"\n",
+                       src_oa->o_size, ret_oa->o_size);
+                LBUG();
+        }
+        ret_oa->o_id = src_oa->o_id;
+        memcpy(src_oa, ret_oa, sizeof(*src_oa));
+        obdo_free(ret_oa);
+
+        *ea = set->set_md;
+        GOTO(done, rc = 0);
+
+cleanup:
+        list_for_each (pos, &set->set_list) {
+                struct obd_export *sub_exp;
+                int err = 0;
+                req = list_entry(pos, struct lov_request, rq_link);
+
+                if (!req->rq_complete || req->rq_rc)
+                        continue;
+
+                sub_exp = lov->tgts[req->rq_idx].ltd_exp,
+                err = obd_destroy(sub_exp, req->rq_oa, NULL, oti);
+                if (err)
+                        CERROR("Failed to uncreate objid "LPX64" subobj "
+                               LPX64" on OST idx %d: rc = %d\n",
+                               set->set_oa->o_id, req->rq_oa->o_id,
+                               req->rq_idx, rc);
+        }
+        if (*ea == NULL)
+                obd_free_memmd(exp, &set->set_md);
+done:
+        if (oti && set->set_cookies) {
+                oti->oti_logcookies = set->set_cookies;
+                if (!set->set_cookie_sent) {
+                        oti_free_cookies(oti);
+                        src_oa->o_valid &= ~OBD_MD_FLCOOKIE;
+                } else {
+                        src_oa->o_valid |= OBD_MD_FLCOOKIE;
+                }
+        }
+        RETURN(rc);
+}
+
+int lov_fini_create_set(struct lov_request_set *set, struct lov_stripe_md **ea)
+{
+        int rc = 0;
+        ENTRY;
+
+        LASSERT(set->set_exp);
+        if (set == NULL)
+                RETURN(0);
+        if (set->set_completes) {
+                rc = create_done(set->set_exp, set, ea);
+                /* FIXME update qos data here */
+        }
+
+        if (atomic_dec_and_test(&set->set_refcount))
+                lov_finish_set(set);
+
+        RETURN(rc);
+}
+
+int lov_update_create_set(struct lov_request_set *set,
+                          struct lov_request *req, int rc)
+{
+        struct obd_trans_info *oti = set->set_oti;
+        struct lov_stripe_md *lsm = set->set_md;
+        struct lov_oinfo *loi;
+        struct lov_obd *lov = &set->set_exp->exp_obd->u.lov;
+        ENTRY;
+
+        req->rq_stripe = set->set_success;
+        loi = &lsm->lsm_oinfo[req->rq_stripe];
+
+        if (rc && lov->tgts[req->rq_idx].active) {
+                CERROR("error creating objid "LPX64" sub-object"
+                       " on OST idx %d/%d: rc = %d\n",
+                       set->set_oa->o_id, req->rq_idx,
+                       lsm->lsm_stripe_count, rc);
+                if (rc > 0) {
+                        CERROR("obd_create returned invalid err %d\n", rc);
+                        rc = -EIO;
+                }
+        }
+        lov_update_set(set, req, rc);
+        if (rc)
+                RETURN(rc);
+
+        if (oti && oti->oti_objid)
+                oti->oti_objid[req->rq_idx] = req->rq_oa->o_id;
+
+        loi->loi_id = req->rq_oa->o_id;
+        loi->loi_ost_idx = req->rq_idx;
+        CDEBUG(D_INODE, "objid "LPX64" has subobj "LPX64"/"LPX64" at idx %d\n",
+               lsm->lsm_object_id, loi->loi_id, loi->loi_id, req->rq_idx);
+        loi_init(loi);
+
+        if (set->set_cookies)
+                ++oti->oti_logcookies;
+        if (req->rq_oa->o_valid & OBD_MD_FLCOOKIE)
+                set->set_cookie_sent++;
+
+        RETURN(0);
+}
+
+int lov_prep_create_set(struct obd_export *exp, struct lov_stripe_md **ea,
+                        struct obdo *src_oa, struct obd_trans_info *oti,
+                        struct lov_request_set **reqset)
+{
+        struct lov_obd *lov = &exp->exp_obd->u.lov;
+        struct lov_request_set *set;
+        int rc = 0, newea = 0;
+        ENTRY;
+
+        OBD_ALLOC(set, sizeof(*set));
+        if (set == NULL)
+                RETURN(-ENOMEM);
+        lov_init_set(set);
+
+        set->set_exp = exp;
+        set->set_md = *ea;
+        set->set_oa = src_oa;
+        set->set_oti = oti;
+
+        if (set->set_md == NULL) {
+                int stripes, stripe_cnt;
+                stripe_cnt = lov_get_stripecnt(lov, 0);
+
+                /* If the MDS file was truncated up to some size, stripe over
+                 * enough OSTs to allow the file to be created at that size. */
+                if (src_oa->o_valid & OBD_MD_FLSIZE) {
+                        stripes=((src_oa->o_size+LUSTRE_STRIPE_MAXBYTES)>>12)-1;
+                        do_div(stripes, (__u32)(LUSTRE_STRIPE_MAXBYTES >> 12));
+
+                        if (stripes > lov->desc.ld_active_tgt_count)
+                                GOTO(out_set, rc = -EFBIG);
+                        if (stripes < stripe_cnt)
+                                stripes = stripe_cnt;
+                } else {
+                        stripes = stripe_cnt;
+                }
+
+                rc = lov_alloc_memmd(&set->set_md, stripes,
+                                     lov->desc.ld_pattern ?
+                                     lov->desc.ld_pattern : LOV_PATTERN_RAID0);
+                if (rc < 0)
+                        goto out_set;
+                newea = 1;
+        }
+
+        rc = qos_prep_create(lov, set, newea);
+        if (rc)
+                goto out_lsm;
+
+        if (oti && (src_oa->o_valid & OBD_MD_FLCOOKIE)) {
+                oti_alloc_cookies(oti, set->set_count);
+                if (!oti->oti_logcookies)
+                        goto out_lsm;
+                set->set_cookies = oti->oti_logcookies;
+        }
+        *reqset = set;
+        RETURN(rc);
+
+out_lsm:
+        if (*ea == NULL)
+                obd_free_memmd(exp, &set->set_md);
+out_set:
+        lov_fini_create_set(set, ea);
+        RETURN(rc);
+}
+
+static int common_attr_done(struct lov_request_set *set)
+{
+        struct list_head *pos;
+        struct lov_request *req;
+        struct obdo *tmp_oa;
+        int rc = 0, attrset = 0;
+        ENTRY;
+
+        if (set->set_oa == NULL)
+                RETURN(0);
+
+        if (!set->set_success)
+                RETURN(-EIO);
+
+        tmp_oa = obdo_alloc();
+        if (tmp_oa == NULL)
+                GOTO(out, rc = -ENOMEM);
+
+        list_for_each (pos, &set->set_list) {
+                req = list_entry(pos, struct lov_request, rq_link);
+
+                if (!req->rq_complete || req->rq_rc)
+                        continue;
+                if (req->rq_oa->o_valid == 0)   /* inactive stripe */
+                        continue;
+                lov_merge_attrs(tmp_oa, req->rq_oa, req->rq_oa->o_valid,
+                                set->set_md, req->rq_stripe, &attrset);
+        }
+        if (!attrset) {
+                CERROR("No stripes had valid attrs\n");
+                rc = -EIO;
+        }
+        tmp_oa->o_id = set->set_oa->o_id;
+        memcpy(set->set_oa, tmp_oa, sizeof(*set->set_oa));
+out:
+        if (tmp_oa)
+                obdo_free(tmp_oa);
+        RETURN(rc);
+
+}
+
+static int brw_done(struct lov_request_set *set)
+{
+        struct lov_stripe_md *lsm = set->set_md;
+        struct lov_oinfo     *loi = NULL;
+        struct list_head *pos;
+        struct lov_request *req;
+        ENTRY;
+                                                                                                                             
+        list_for_each (pos, &set->set_list) {
+                req = list_entry(pos, struct lov_request, rq_link);
+                                                                                                                             
+                if (!req->rq_complete || req->rq_rc)
+                        continue;
+                                                                                                                             
+                loi = &lsm->lsm_oinfo[req->rq_stripe];
+                                                                                                                             
+                if (req->rq_oa->o_valid & OBD_MD_FLBLOCKS)
+                        loi->loi_blocks = req->rq_oa->o_blocks;
+        }
+                                                                                                                             
+        RETURN(0);
+}
+
+int lov_fini_brw_set(struct lov_request_set *set)
+{
+        int rc = 0;
+        ENTRY;
+
+        LASSERT(set->set_exp);
+        if (set == NULL)
+                RETURN(0);
+        if (set->set_completes) {
+                rc = brw_done(set);
+                /* FIXME update qos data here */
+        }
+        if (atomic_dec_and_test(&set->set_refcount))
+                lov_finish_set(set);
+
+        RETURN(rc);
+}
+
+int lov_prep_brw_set(struct obd_export *exp, struct obdo *src_oa,
+                     struct lov_stripe_md *lsm, obd_count oa_bufs,
+                     struct brw_page *pga, struct obd_trans_info *oti,
+                     struct lov_request_set **reqset)
+{
+        struct {
+                obd_count       index;
+                obd_count       count;
+                obd_count       off;
+        } *info = NULL;
+        struct lov_request_set *set;
+        struct lov_oinfo *loi = NULL;
+        struct lov_obd *lov = &exp->exp_obd->u.lov;
+        int rc = 0, i, shift;
+        ENTRY;
+
+        OBD_ALLOC(set, sizeof(*set));
+        if (set == NULL)
+                RETURN(-ENOMEM);
+        lov_init_set(set);
+
+        set->set_exp = exp;
+        set->set_md = lsm;
+        set->set_oa = src_oa;
+        set->set_oti = oti;
+        set->set_oabufs = oa_bufs;
+        OBD_ALLOC(set->set_pga, oa_bufs * sizeof(*set->set_pga));
+        if (!set->set_pga)
+                GOTO(out, rc = -ENOMEM);
+
+        OBD_ALLOC(info, sizeof(*info) * lsm->lsm_stripe_count);
+        if (!info)
+                GOTO(out, rc = -ENOMEM);
+
+        /* calculate the page count for each stripe */
+        for (i = 0; i < oa_bufs; i++) {
+                int stripe = lov_stripe_number(lsm, pga[i].off);
+                info[stripe].count++;
+        }
+
+        /* alloc and initialize lov request */
+        loi = lsm->lsm_oinfo;
+        shift = 0;
+        for (i = 0; i < lsm->lsm_stripe_count; i++, loi++) {
+                struct lov_request *req;
+
+                if (info[i].count == 0)
+                        continue;
+
+                if (lov->tgts[loi->loi_ost_idx].active == 0) {
+                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
+                        GOTO(out, rc = -EIO);
+                }
+
+                OBD_ALLOC(req, sizeof(*req));
+                if (req == NULL)
+                        GOTO(out, rc = -ENOMEM);
+
+                req->rq_oa = obdo_alloc();
+                if (req->rq_oa == NULL)
+                        GOTO(out, rc = -ENOMEM);
+
+                if (src_oa)
+                        memcpy(req->rq_oa, src_oa, sizeof(*req->rq_oa));
+                req->rq_oa->o_id = loi->loi_id;
+
+                req->rq_buflen = sizeof(*req->rq_md);
+                OBD_ALLOC(req->rq_md, req->rq_buflen);
+                if (req->rq_md == NULL)
+                        GOTO(out, rc = -ENOMEM);
+
+                req->rq_idx = loi->loi_ost_idx;
+                req->rq_stripe = i;
+
+                /* XXX LOV STACKING */
+                req->rq_md->lsm_object_id = loi->loi_id;
+                req->rq_md->lsm_object_gr = lsm->lsm_object_gr;
+                req->rq_oabufs = info[i].count;
+                req->rq_pgaidx = shift;
+                shift += req->rq_oabufs;
+
+                /* remember the index for sort brw_page array */
+                info[i].index = req->rq_pgaidx;
+
+                lov_set_add_req(req, set);
+        }
+        if (!set->set_count)
+                GOTO(out, rc = -EIO);
+
+        /* rotate & sort the brw_page array */
+        for (i = 0; i < oa_bufs; i++) {
+                int stripe = lov_stripe_number(lsm, pga[i].off);
+
+                shift = info[stripe].index + info[stripe].off;
+                LASSERT(shift < oa_bufs);
+                set->set_pga[shift] = pga[i];
+                lov_stripe_offset(lsm, pga[i].off, stripe,
+                                  &set->set_pga[shift].off);
+                info[stripe].off++;
+        }
+out:
+        if (info)
+                OBD_FREE(info, sizeof(*info) * lsm->lsm_stripe_count);
+
+        if (rc == 0)
+                *reqset = set;
+        else
+                lov_fini_brw_set(set);
+
+        RETURN(rc);
+}
+
+static int getattr_done(struct lov_request_set *set)
+{
+        return common_attr_done(set);
+}
+
+int lov_fini_getattr_set(struct lov_request_set *set)
+{
+        int rc = 0;
+        ENTRY;
+
+        LASSERT(set->set_exp);
+        if (set == NULL)
+                RETURN(0);
+        if (set->set_completes)
+                rc = getattr_done(set);
+
+        if (atomic_dec_and_test(&set->set_refcount))
+                lov_finish_set(set);
+
+        RETURN(rc);
+}
+
+int lov_prep_getattr_set(struct obd_export *exp, struct obdo *src_oa,
+                         struct lov_stripe_md *lsm,
+                         struct lov_request_set **reqset)
+{
+        struct lov_request_set *set;
+        struct lov_oinfo *loi = NULL;
+        struct lov_obd *lov = &exp->exp_obd->u.lov;
+        int rc = 0, i;
+        ENTRY;
+
+        OBD_ALLOC(set, sizeof(*set));
+        if (set == NULL)
+                RETURN(-ENOMEM);
+        lov_init_set(set);
+
+        set->set_exp = exp;
+        set->set_md = lsm;
+        set->set_oa = src_oa;
+
+        loi = lsm->lsm_oinfo;
+        for (i = 0; i < lsm->lsm_stripe_count; i++, loi++) {
+                struct lov_request *req;
+
+                if (lov->tgts[loi->loi_ost_idx].active == 0) {
+                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
+                        continue;
+                }
+
+                OBD_ALLOC(req, sizeof(*req));
+                if (req == NULL)
+                        GOTO(out_set, rc = -ENOMEM);
+
+                req->rq_stripe = i;
+                req->rq_idx = loi->loi_ost_idx;
+
+                req->rq_oa = obdo_alloc();
+                if (req->rq_oa == NULL)
+                        GOTO(out_set, rc = -ENOMEM);
+                memcpy(req->rq_oa, src_oa, sizeof(*req->rq_oa));
+                req->rq_oa->o_id = loi->loi_id;
+
+                lov_set_add_req(req, set);
+        }
+        if (!set->set_count)
+                GOTO(out_set, rc = -EIO);
+        *reqset = set;
+        RETURN(rc);
+out_set:
+        lov_fini_getattr_set(set);
+        RETURN(rc);
+}
+
+int lov_fini_destroy_set(struct lov_request_set *set)
+{
+        ENTRY;
+
+        LASSERT(set->set_exp);
+        if (set == NULL)
+                RETURN(0);
+        if (set->set_completes) {
+                /* FIXME update qos data here */
+        }
+
+        if (atomic_dec_and_test(&set->set_refcount))
+                lov_finish_set(set);
+
+        RETURN(0);
+}
+
+int lov_prep_destroy_set(struct obd_export *exp, struct obdo *src_oa,
+                         struct lov_stripe_md *lsm,
+                         struct obd_trans_info *oti,
+                         struct lov_request_set **reqset)
+{
+        struct lov_request_set *set;
+        struct lov_oinfo *loi = NULL;
+        struct lov_obd *lov = &exp->exp_obd->u.lov;
+        int rc = 0, cookie_set = 0, i;
+        ENTRY;
+
+        OBD_ALLOC(set, sizeof(*set));
+        if (set == NULL)
+                RETURN(-ENOMEM);
+        lov_init_set(set);
+
+        set->set_exp = exp;
+        set->set_md = lsm;
+        set->set_oa = src_oa;
+        set->set_oti = oti;
+        if (oti != NULL && src_oa->o_valid & OBD_MD_FLCOOKIE)
+                set->set_cookies = oti->oti_logcookies;
+
+        loi = lsm->lsm_oinfo;
+        for (i = 0; i < lsm->lsm_stripe_count; i++, loi++) {
+                struct lov_request *req;
+
+                if (lov->tgts[loi->loi_ost_idx].active == 0) {
+                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
+                        continue;
+                }
+
+                OBD_ALLOC(req, sizeof(*req));
+                if (req == NULL)
+                        GOTO(out_set, rc = -ENOMEM);
+
+                req->rq_stripe = i;
+                req->rq_idx = loi->loi_ost_idx;
+
+                req->rq_oa = obdo_alloc();
+                if (req->rq_oa == NULL)
+                        GOTO(out_set, rc = -ENOMEM);
+                memcpy(req->rq_oa, src_oa, sizeof(*req->rq_oa));
+                req->rq_oa->o_id = loi->loi_id;
+
+                /* Setup the first request's cookie position */
+                if (!cookie_set && set->set_cookies) {
+                        oti->oti_logcookies = set->set_cookies + i;
+                        cookie_set = 1;
+                }
+                lov_set_add_req(req, set);
+        }
+        if (!set->set_count)
+                GOTO(out_set, rc = -EIO);
+        *reqset = set;
+        RETURN(rc);
+out_set:
+        lov_fini_destroy_set(set);
+        RETURN(rc);
+}
+
+static int setattr_done(struct lov_request_set *set)
+{
+        return common_attr_done(set);
+}
+
+int lov_fini_setattr_set(struct lov_request_set *set)
+{
+        int rc = 0;
+        ENTRY;
+
+        LASSERT(set->set_exp);
+        if (set == NULL)
+                RETURN(0);
+        if (set->set_completes) {
+                rc = setattr_done(set);
+                /* FIXME update qos data here */
+        }
+
+        if (atomic_dec_and_test(&set->set_refcount))
+                lov_finish_set(set);
+        RETURN(rc);
+}
+
+int lov_prep_setattr_set(struct obd_export *exp, struct obdo *src_oa,
+                         struct lov_stripe_md *lsm, struct obd_trans_info *oti,
+                         struct lov_request_set **reqset)
+{
+        struct lov_request_set *set;
+        struct lov_oinfo *loi = NULL;
+        struct lov_obd *lov = &exp->exp_obd->u.lov;
+        int rc = 0, i;
+        ENTRY;
+
+        OBD_ALLOC(set, sizeof(*set));
+        if (set == NULL)
+                RETURN(-ENOMEM);
+        lov_init_set(set);
+
+        set->set_exp = exp;
+        set->set_md = lsm;
+        set->set_oa = src_oa;
+
+        loi = lsm->lsm_oinfo;
+        for (i = 0; i < lsm->lsm_stripe_count; i++, loi++) {
+                struct lov_request *req;
+
+                if (lov->tgts[loi->loi_ost_idx].active == 0) {
+                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
+                        continue;
+                }
+
+                OBD_ALLOC(req, sizeof(*req));
+                if (req == NULL)
+                        GOTO(out_set, rc = -ENOMEM);
+                req->rq_stripe = i;
+                req->rq_idx = loi->loi_ost_idx;
+
+                req->rq_oa = obdo_alloc();
+                if (req->rq_oa == NULL)
+                        GOTO(out_set, rc = -ENOMEM);
+                memcpy(req->rq_oa, src_oa, sizeof(*req->rq_oa));
+                req->rq_oa->o_id = loi->loi_id;
+
+                if (src_oa->o_valid & OBD_MD_FLSIZE) {
+                        if (lov_stripe_offset(lsm, src_oa->o_size, i,
+                                              &req->rq_oa->o_size) < 0 &&
+                            req->rq_oa->o_size)
+                                req->rq_oa->o_size--;
+                        CDEBUG(D_INODE, "stripe %d has size "LPU64"/"LPU64"\n",
+                               i, req->rq_oa->o_size, src_oa->o_size);
+                }
+                lov_set_add_req(req, set);
+        }
+        if (!set->set_count)
+                GOTO(out_set, rc = -EIO);
+        *reqset = set;
+        RETURN(rc);
+out_set:
+        lov_fini_setattr_set(set);
+        RETURN(rc);
+}
+
+int lov_update_punch_set(struct lov_request_set *set, struct lov_request *req,
+                         int rc)
+{
+        struct lov_stripe_md *lsm = set->set_md;
+        struct lov_obd *lov = &set->set_exp->exp_obd->u.lov;
+        ENTRY;
+
+        lov_update_set(set, req, rc);
+        if (rc == 0) {
+                struct lov_oinfo *loi = &lsm->lsm_oinfo[req->rq_stripe];
+                loi->loi_kms = loi->loi_rss = req->rq_extent.start;
+        }
+        if (rc && !lov->tgts[req->rq_idx].active)
+                rc = 0;
+        /* FIXME in raid1 regime, should return 0 */
+        RETURN(rc);
+}
+
+int lov_fini_punch_set(struct lov_request_set *set)
+{
+        int rc = 0;
+        ENTRY;
+
+        LASSERT(set->set_exp);
+        if (set == NULL)
+                RETURN(0);
+        if (set->set_completes) {
+                if (!set->set_success)
+                        rc = -EIO;
+                /* FIXME update qos data here */
+        }
+
+        if (atomic_dec_and_test(&set->set_refcount))
+                lov_finish_set(set);
+
+        RETURN(rc);
+}
+
+int lov_prep_punch_set(struct obd_export *exp, struct obdo *src_oa,
+                       struct lov_stripe_md *lsm, obd_off start,
+                       obd_off end, struct obd_trans_info *oti,
+                       struct lov_request_set **reqset)
+{
+        struct lov_request_set *set;
+        struct lov_oinfo *loi = NULL;
+        struct lov_obd *lov = &exp->exp_obd->u.lov;
+        int rc = 0, i;
+        ENTRY;
+
+        OBD_ALLOC(set, sizeof(*set));
+        if (set == NULL)
+                RETURN(-ENOMEM);
+        lov_init_set(set);
+
+        set->set_exp = exp;
+        set->set_md = lsm;
+        set->set_oa = src_oa;
+
+        loi = lsm->lsm_oinfo;
+        for (i = 0; i < lsm->lsm_stripe_count; i++, loi++) {
+                struct lov_request *req;
+                obd_off rs, re;
+
+                if (lov->tgts[loi->loi_ost_idx].active == 0) {
+                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
+                        continue;
+                }
+
+                if (!lov_stripe_intersects(lsm, i, start, end, &rs, &re))
+                        continue;
+
+                OBD_ALLOC(req, sizeof(*req));
+                if (req == NULL)
+                        GOTO(out_set, rc = -ENOMEM);
+                req->rq_stripe = i;
+                req->rq_idx = loi->loi_ost_idx;
+
+                req->rq_oa = obdo_alloc();
+                if (req->rq_oa == NULL)
+                        GOTO(out_set, rc = -ENOMEM);
+                memcpy(req->rq_oa, src_oa, sizeof(*req->rq_oa));
+                req->rq_oa->o_id = loi->loi_id;
+
+                req->rq_extent.start = rs;
+                req->rq_extent.end = re;
+
+                lov_set_add_req(req, set);
+        }
+        if (!set->set_count)
+                GOTO(out_set, rc = -EIO);
+        *reqset = set;
+        RETURN(rc);
+out_set:
+        lov_fini_punch_set(set);
+        RETURN(rc);
+}
+
+int lov_fini_sync_set(struct lov_request_set *set)
+{
+        int rc = 0;
+        ENTRY;
+
+        LASSERT(set->set_exp);
+        if (set == NULL)
+                RETURN(0);
+        if (set->set_completes) {
+                if (!set->set_success)
+                        rc = -EIO;
+                /* FIXME update qos data here */
+        }
+
+        if (atomic_dec_and_test(&set->set_refcount))
+                lov_finish_set(set);
+
+        RETURN(rc);
+}
+
+int lov_prep_sync_set(struct obd_export *exp, struct obdo *src_oa,
+                      struct lov_stripe_md *lsm, obd_off start,
+                      obd_off end, struct lov_request_set **reqset)
+{
+        struct lov_request_set *set;
+        struct lov_oinfo *loi = NULL;
+        struct lov_obd *lov = &exp->exp_obd->u.lov;
+        int rc = 0, i;
+        ENTRY;
+
+        OBD_ALLOC(set, sizeof(*set));
+        if (set == NULL)
+                RETURN(-ENOMEM);
+        lov_init_set(set);
+
+        set->set_exp = exp;
+        set->set_md = lsm;
+        set->set_oa = src_oa;
+
+        loi = lsm->lsm_oinfo;
+        for (i = 0; i < lsm->lsm_stripe_count; i++, loi++) {
+                struct lov_request *req;
+                obd_off rs, re;
+
+                if (lov->tgts[loi->loi_ost_idx].active == 0) {
+                        CDEBUG(D_HA, "lov idx %d inactive\n", loi->loi_ost_idx);
+                        continue;
+                }
+
+                if (!lov_stripe_intersects(lsm, i, start, end, &rs, &re))
+                        continue;
+
+                OBD_ALLOC(req, sizeof(*req));
+                if (req == NULL)
+                        GOTO(out_set, rc = -ENOMEM);
+                req->rq_stripe = i;
+                req->rq_idx = loi->loi_ost_idx;
+
+                req->rq_oa = obdo_alloc();
+                if (req->rq_oa == NULL)
+                        GOTO(out_set, rc = -ENOMEM);
+                memcpy(req->rq_oa, src_oa, sizeof(*req->rq_oa));
+                req->rq_oa->o_id = loi->loi_id;
+
+                req->rq_extent.start = rs;
+                req->rq_extent.end = re;
+
+                lov_set_add_req(req, set);
+        }
+        if (!set->set_count)
+                GOTO(out_set, rc = -EIO);
+        *reqset = set;
+        RETURN(rc);
+out_set:
+        lov_fini_sync_set(set);
+        RETURN(rc);
+}
index bcaa30a..8321e73 100644 (file)
@@ -1005,7 +1005,7 @@ int ptlrpc_set_wait(struct ptlrpc_request_set *set)
         if (set->set_interpret != NULL) {
                 int (*interpreter)(struct ptlrpc_request_set *set,void *,int) =
                         set->set_interpret;
-                rc = interpreter (set, &set->set_args, rc);
+                rc = interpreter (set, set->set_arg, rc);
         }
 
         RETURN(rc);