LU-2910 clio: restore iov when restart io

The iovector needs be restored on restarted io. This patch also
added some LASSERT and debug message to make it easier to debug
this kind of problem later.

Test-Parameters: testlist=racer
Signed-off-by: Niu Yawei <yawei.niu@intel.com>
Change-Id: I5d737cfa083ae3b3d9f040a2dc36d6b8693b548b
Reviewed-on: http://review.whamcloud.com/5652
Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com>
Tested-by: Hudson
Tested-by: Maloo <whamcloud.maloo@gmail.com>
Reviewed-by: Bobi Jam <bobijam@gmail.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>

diff --git a/lustre/lclient/lcommon_cl.c b/lustre/lclient/lcommon_cl.c
index 41ebae9..a6d5d58 100644 (file)
--- a/lustre/lclient/lcommon_cl.c
+++ b/lustre/lclient/lcommon_cl.c
@@ -789,6 +789,9 @@ void ccc_io_update_iov(const struct lu_env *env,
         }
 
         cio->cui_nrsegs = i + 1;
+       LASSERTF(cio->cui_tot_nrsegs >= cio->cui_nrsegs,
+                "tot_nrsegs: %lu, nrsegs: %lu\n",
+                cio->cui_tot_nrsegs, cio->cui_nrsegs);
 }
 
 int ccc_io_one_lock(const struct lu_env *env, struct cl_io *io,
@@ -810,32 +813,39 @@ void ccc_io_advance(const struct lu_env *env,
                     const struct cl_io_slice *ios,
                     size_t nob)
 {
-        struct ccc_io    *cio = cl2ccc_io(env, ios);
-        struct cl_io     *io  = ios->cis_io;
-        struct cl_object *obj = ios->cis_io->ci_obj;
+       struct ccc_io    *cio = cl2ccc_io(env, ios);
+       struct cl_io     *io  = ios->cis_io;
+       struct cl_object *obj = ios->cis_io->ci_obj;
 
-        CLOBINVRNT(env, obj, ccc_object_invariant(obj));
+       CLOBINVRNT(env, obj, ccc_object_invariant(obj));
 
-        if (cl_is_normalio(env, io) && io->ci_continue) {
-                /* update the iov */
-                LASSERT(cio->cui_tot_nrsegs >= cio->cui_nrsegs);
-                LASSERT(cio->cui_tot_count  >= nob);
+       if (!cl_is_normalio(env, io))
+               return;
 
-                cio->cui_iov        += cio->cui_nrsegs;
-                cio->cui_tot_nrsegs -= cio->cui_nrsegs;
-                cio->cui_tot_count  -= nob;
+       LASSERT(cio->cui_tot_nrsegs >= cio->cui_nrsegs);
+       LASSERT(cio->cui_tot_count  >= nob);
 
-                if (cio->cui_iov_olen) {
-                        struct iovec *iv;
+       cio->cui_iov        += cio->cui_nrsegs;
+       cio->cui_tot_nrsegs -= cio->cui_nrsegs;
+       cio->cui_tot_count  -= nob;
 
-                        cio->cui_iov--;
-                        cio->cui_tot_nrsegs++;
-                        iv = &cio->cui_iov[0];
-                        iv->iov_base += iv->iov_len;
-                        LASSERT(cio->cui_iov_olen > iv->iov_len);
-                        iv->iov_len = cio->cui_iov_olen - iv->iov_len;
-                }
-        }
+       /* update the iov */
+       if (cio->cui_iov_olen > 0) {
+               struct iovec *iv;
+
+               cio->cui_iov--;
+               cio->cui_tot_nrsegs++;
+               iv = &cio->cui_iov[0];
+               if (io->ci_continue) {
+                       iv->iov_base += iv->iov_len;
+                       LASSERT(cio->cui_iov_olen > iv->iov_len);
+                       iv->iov_len = cio->cui_iov_olen - iv->iov_len;
+               } else {
+                       /* restore the iov_len, in case of restart io. */
+                       iv->iov_len = cio->cui_iov_olen;
+               }
+               cio->cui_iov_olen = 0;
+       }
 }
 
 /**

diff --git a/lustre/llite/file.c b/lustre/llite/file.c
index d763cf7..723b8e8 100644 (file)
--- a/lustre/llite/file.c
+++ b/lustre/llite/file.c
@@ -920,8 +920,16 @@ restart:
         GOTO(out, result);
 out:
         cl_io_fini(env, io);
-       if (result == 0 && io->ci_need_restart) /* need to restart whole IO */
+       /* If any bit been read/written (result != 0), we just return
+        * short read/write instead of restart io. */
+       if (result == 0 && io->ci_need_restart) {
+               CDEBUG(D_VFSTRACE, "Restart %s on %s from %lld, count:%zd\n",
+                      iot == CIT_READ ? "read" : "write",
+                      file->f_dentry->d_name.name, *ppos, count);
+               LASSERTF(io->u.ci_rw.crw_count == count, "%zd != %zd\n",
+                        io->u.ci_rw.crw_count, count);
                goto restart;
+       }
 
         if (iot == CIT_READ) {
                 if (result >= 0)