LU-18430 llite: disable readahead on file with random fadvise

If a file is advised with random access pattern via posix_fadvise
with POSIX_FADV_RANDOM mode, the client should not trigger
read-ahead on this file.

Add posix_fadvise option 'i' for "multiop".
Add "forceread_pages" stats into llite.*.read_ahead_stats to
account the force read (random) pages.
Add the test case sanity/test_853 to verify it.

Signed-off-by: Qian Yingjin <qian@ddn.com>
Change-Id: Ie02ace55985c92abffb871184f847400e9da03ad
Reviewed-on: https://review.whamcloud.com/c/fs/lustre-release/+/56912
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Alexey Lyashkov <alexey.lyashkov@hpe.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>

diff --git a/lustre/llite/llite_internal.h b/lustre/llite/llite_internal.h
index 8455e2d..eae98ef 100644 (file)
--- a/lustre/llite/llite_internal.h
+++ b/lustre/llite/llite_internal.h
@@ -762,6 +762,7 @@ enum ra_stat {
        RA_STAT_FAILED_FAST_READ,
        RA_STAT_MMAP_RANGE_READ,
        RA_STAT_READAHEAD_PAGES,
+       RA_STAT_FORCEREAD_PAGES,
        _NR_RA_STAT,
 };
 

diff --git a/lustre/llite/lproc_llite.c b/lustre/llite/lproc_llite.c
index 82da954..de36407 100644 (file)
--- a/lustre/llite/lproc_llite.c
+++ b/lustre/llite/lproc_llite.c
@@ -2602,7 +2602,8 @@ static const char *const ra_stat_string[] = {
        [RA_STAT_ASYNC]                 = "async_readahead",
        [RA_STAT_FAILED_FAST_READ]      = "failed_to_fast_read",
        [RA_STAT_MMAP_RANGE_READ]       = "mmap_range_read",
-       [RA_STAT_READAHEAD_PAGES]       = "readahead_pages"
+       [RA_STAT_READAHEAD_PAGES]       = "readahead_pages",
+       [RA_STAT_FORCEREAD_PAGES]       = "forceread_pages"
 };
 
 int ll_debugfs_register_super(struct super_block *sb, const char *name)
@@ -2668,7 +2669,8 @@ int ll_debugfs_register_super(struct super_block *sb, const char *name)
                GOTO(out_stats, err = -ENOMEM);
 
        for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++) {
-               if (id == RA_STAT_READAHEAD_PAGES)
+               if (id == RA_STAT_READAHEAD_PAGES ||
+                   id == RA_STAT_FORCEREAD_PAGES)
                        lprocfs_counter_init(sbi->ll_ra_stats, id,
                                             LPROCFS_TYPE_PAGES |
                                             LPROCFS_CNTR_AVGMINMAX,

diff --git a/lustre/llite/rw.c b/lustre/llite/rw.c
index 62fdd11..9eb2461 100644 (file)
--- a/lustre/llite/rw.c
+++ b/lustre/llite/rw.c
@@ -951,6 +951,10 @@ static int ll_readpages(const struct lu_env *env, struct cl_io *io,
                        count++;
        }
 
+       if (count)
+               ll_ra_stats_add(vvp_object_inode(io->ci_obj),
+                               RA_STAT_FORCEREAD_PAGES, count);
+
        RETURN(count > 0 ? count : ret);
 }
 
@@ -1721,6 +1725,9 @@ int ll_io_read_page(const struct lu_env *env, struct cl_io *io,
        if (file) {
                lfd = file->private_data;
                ras = &lfd->fd_ras;
+
+               if (file->f_mode & FMODE_RANDOM)
+                       io->ci_rand_read = 1;
        }
 
        /* PagePrivate2 is set in ll_io_zero_page() to tell us the vmpage

diff --git a/lustre/tests/multiop.c b/lustre/tests/multiop.c
index 82b91be..4c5a7dc 100644 (file)
--- a/lustre/tests/multiop.c
+++ b/lustre/tests/multiop.c
@@ -83,6 +83,7 @@ char usage[] =
 "       g gid put grouplock\n"
 "       H[num] create HSM released file with num stripes\n"
 "       I  fiemap\n"
+"       i  random fadvise\n"
 "       K  link path to filename\n"
 "       L  link\n"
 "       l  symlink filename to path\n"
@@ -515,6 +516,14 @@ int main(int argc, char **argv)
                case 'I':
                        do_fiemap(fd);
                        break;
+               case 'i':
+                       rc = posix_fadvise(fd, 0, 0, POSIX_FADV_RANDOM);
+                       if (rc) {
+                               save_errno = errno;
+                               perror("fadvise");
+                               exit(save_errno);
+                       }
+                       break;
                case 'j':
                        if (flock(fd, LOCK_EX) == -1)
                                errx(-1, "flock()");

diff --git a/lustre/tests/sanity.sh b/lustre/tests/sanity.sh
index 3546d95..14c2ee1 100755 (executable)
--- a/lustre/tests/sanity.sh
+++ b/lustre/tests/sanity.sh
@@ -33820,6 +33820,35 @@ test_852() {
 }
 run_test 852 "mkdir using intent lock for striped directory"
 
+test_853() {
+       local file=$DIR/$tfile
+       local size=$((PAGE_SIZE * 2))
+
+       dd if=/dev/zero of=$file bs=1M count=100 ||
+               error "failed to write $file"
+       cancel_lru_locks $OSC
+       $LCTL set_param llite.*.read_ahead_stats=clear
+       $MULTIOP $file or1048576c || error "failed to read $file"
+       $LCTL get_param llite.*.read_ahead_stats
+
+       cancel_lru_locks $OSC
+       $LCTL set_param llite.*.read_ahead_stats=clear
+       $MULTIOP $file oir1048576z1048576r${size}c ||
+               error "failed to read $file"
+       $LCTL get_param llite.*.read_ahead_stats
+
+       local ranum=$($LCTL get_param -n llite.*.read_ahead_stats |
+                     get_named_value 'readahead.pages' | calc_sum)
+
+       (( ranum == 0 )) || error "should not trigger readahead"
+
+       local rndnum=$($LCTL get_param -n llite.*.read_ahead_stats |
+                      get_named_value 'forceread.pages' | calc_sum)
+
+       (( rndnum == 2 )) || error "force random read: $rndnum, expected 2"
+}
+run_test 853 "Verify that random fadvise works as expected"
+
 #
 # tests that do cleanup/setup should be run at the end
 #