LU-8726 osd-ldiskfs: bypass read for benchmarking

[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_io.c

diff --git a/lustre/osd-ldiskfs/osd_io.c b/lustre/osd-ldiskfs/osd_io.c
index 5e1a31b..3434634 100644 (file)
--- a/lustre/osd-ldiskfs/osd_io.c
+++ b/lustre/osd-ldiskfs/osd_io.c
@@ -23,7 +23,7 @@
  * Copyright (c) 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2012, 2015, Intel Corporation.
+ * Copyright (c) 2012, 2016, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
@@ -141,16 +141,17 @@ void osd_fini_iobuf(struct osd_device *d, struct osd_iobuf *iobuf)
 #define __REQ_WRITE BIO_RW
 #endif
 
+#ifdef HAVE_BIO_ENDIO_USES_ONE_ARG
+static void dio_complete_routine(struct bio *bio)
+{
+       int error = bio->bi_error;
+#else
 static void dio_complete_routine(struct bio *bio, int error)
 {
+#endif
        struct osd_iobuf *iobuf = bio->bi_private;
-#ifdef HAVE_BVEC_ITER
-       struct bvec_iter iter;
-       struct bio_vec bvl;
-#else
        int iter;
        struct bio_vec *bvl;
-#endif
 
         /* CAVEAT EMPTOR: possibly in IRQ context
          * DO NOT record procfs stats here!!! */
@@ -166,10 +167,16 @@ static void dio_complete_routine(struct bio *bio, int error)
                       "IO - you will probably have to reboot this node.\n");
                CERROR("bi_next: %p, bi_flags: %lx, bi_rw: %lu, bi_vcnt: %d, "
                       "bi_idx: %d, bi->size: %d, bi_end_io: %p, bi_cnt: %d, "
-                      "bi_private: %p\n", bio->bi_next, bio->bi_flags,
+                      "bi_private: %p\n", bio->bi_next,
+                       (unsigned long)bio->bi_flags,
                        bio->bi_rw, bio->bi_vcnt, bio_idx(bio),
                        bio_sectors(bio) << 9, bio->bi_end_io,
-                       atomic_read(&bio->bi_cnt), bio->bi_private);
+#ifdef HAVE_BI_CNT
+                       atomic_read(&bio->bi_cnt),
+#else
+                       atomic_read(&bio->__bi_cnt),
+#endif
+                       bio->bi_private);
                return;
        }
 
@@ -880,17 +887,25 @@ static int osd_ldiskfs_map_inode_pages(struct inode *inode, struct page **page,
        struct page *fp = NULL;
        int clen = 0;
        pgoff_t max_page_index;
+       handle_t *handle = NULL;
 
        max_page_index = inode->i_sb->s_maxbytes >> PAGE_SHIFT;
 
        CDEBUG(D_OTHER, "inode %lu: map %d pages from %lu\n",
                inode->i_ino, pages, (*page)->index);
 
+       if (create) {
+               create = LDISKFS_GET_BLOCKS_CREATE;
+               handle = ldiskfs_journal_current_handle();
+               LASSERT(handle != NULL);
+               rc = osd_attach_jinode(inode);
+               if (rc)
+                       return rc;
+       }
        /* pages are sorted already. so, we just have to find
         * contig. space and process them properly */
        while (i < pages) {
                long blen, total = 0;
-               handle_t *handle = NULL;
                struct ldiskfs_map_blocks map = { 0 };
 
                if (fp == NULL) { /* start new extent */
@@ -910,11 +925,6 @@ static int osd_ldiskfs_map_inode_pages(struct inode *inode, struct page **page,
                /* process found extent */
                map.m_lblk = fp->index * blocks_per_page;
                map.m_len = blen = clen * blocks_per_page;
-               if (create) {
-                       create = LDISKFS_GET_BLOCKS_CREATE;
-                       handle = ldiskfs_journal_current_handle();
-                       LASSERT(handle != NULL);
-               }
 cont_map:
                rc = ldiskfs_map_blocks(handle, inode, &map, create);
                if (rc >= 0) {
@@ -962,9 +972,9 @@ static int osd_write_prep(const struct lu_env *env, struct dt_object *dt,
         struct osd_iobuf       *iobuf = &oti->oti_iobuf;
         struct inode           *inode = osd_dt_obj(dt)->oo_inode;
         struct osd_device      *osd   = osd_obj2dev(osd_dt_obj(dt));
-        struct timeval          start;
-        struct timeval          end;
-        unsigned long           timediff;
+       ktime_t start;
+       ktime_t end;
+       s64 timediff;
         ssize_t                 isize;
         __s64                   maxidx;
         int                     rc = 0;
@@ -985,7 +995,7 @@ static int osd_write_prep(const struct lu_env *env, struct dt_object *dt,
         if (isize > osd->od_readcache_max_filesize)
                 cache = 0;
 
-       do_gettimeofday(&start);
+       start = ktime_get();
        for (i = 0; i < npages; i++) {
 
                if (cache == 0)
@@ -1018,8 +1028,8 @@ static int osd_write_prep(const struct lu_env *env, struct dt_object *dt,
                        kunmap(lnb[i].lnb_page);
                }
        }
-       do_gettimeofday(&end);
-       timediff = cfs_timeval_sub(&end, &start, NULL);
+       end = ktime_get();
+       timediff = ktime_us_delta(end, start);
        lprocfs_counter_add(osd->od_stats, LPROC_OSD_GET_PAGE, timediff);
 
         if (iobuf->dr_npages) {
@@ -1311,9 +1321,9 @@ static int osd_read_prep(const struct lu_env *env, struct dt_object *dt,
         struct osd_iobuf *iobuf = &oti->oti_iobuf;
         struct inode *inode = osd_dt_obj(dt)->oo_inode;
         struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
-        struct timeval start, end;
-        unsigned long timediff;
        int rc = 0, i, cache = 0, cache_hits = 0, cache_misses = 0;
+       ktime_t start, end;
+       s64 timediff;
        loff_t isize;
 
         LASSERT(inode);
@@ -1329,7 +1339,7 @@ static int osd_read_prep(const struct lu_env *env, struct dt_object *dt,
        if (isize > osd->od_readcache_max_filesize)
                cache = 0;
 
-       do_gettimeofday(&start);
+       start = ktime_get();
        for (i = 0; i < npages; i++) {
 
                if (isize <= lnb[i].lnb_file_offset)
@@ -1342,6 +1352,10 @@ static int osd_read_prep(const struct lu_env *env, struct dt_object *dt,
                else
                        lnb[i].lnb_rc = lnb[i].lnb_len;
 
+               /* Bypass disk read if fail_loc is set properly */
+               if (OBD_FAIL_CHECK(OBD_FAIL_OST_FAKE_RW))
+                       SetPageUptodate(lnb[i].lnb_page);
+
                if (PageUptodate(lnb[i].lnb_page)) {
                        cache_hits++;
                } else {
@@ -1353,8 +1367,8 @@ static int osd_read_prep(const struct lu_env *env, struct dt_object *dt,
                        generic_error_remove_page(inode->i_mapping,
                                                  lnb[i].lnb_page);
        }
-       do_gettimeofday(&end);
-       timediff = cfs_timeval_sub(&end, &start, NULL);
+       end = ktime_get();
+       timediff = ktime_us_delta(end, start);
        lprocfs_counter_add(osd->od_stats, LPROC_OSD_GET_PAGE, timediff);
 
        if (cache_hits != 0)