LU-1943 zfs-osd: different fixes for zfs

- echo_create_md_object() to mark la_mode is valid

- ofd_statfs() to set avail/free blocks to 2 when the counters
  are converted to 4K units. otherwise with zfs 2 blocks of size
  128K can be > 0.01% leading to failures in sanity/27*

- zfs-osd to initialize LMA

- zfs-osd's osd_attr_get() to calculate size of directory from
  blocks and do not use cached size (as we don't control it as
  with regular files)

- zfs-osd to track enabled/disabled status for SA and follow it

- take from orion additional locking in osd_object_sa_dirty_add()
  and osd_object_sa_dirty_rele()

- lu_cache_shrink() do not purge objects if __GFP_FS is not set
  this should prevent deadlock when used with zfs backend

- ost_checksum_bulk() to allocate new page to simulate corruption,
  otherwise it corrupts zfs's internal buffers which do not rely
  on PageUptoDate flag

- t-f to modprobe zfs module if osd-zfs is in use

Signed-off-by: Alex Zhuravlev <alexey.zhuravlev@intel.com>
Change-Id: I027483162735cd0fb4be9fa8bd12c2f2dc43b44a
Reviewed-on: http://review.whamcloud.com/4141
Tested-by: Hudson
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Tested-by: Maloo <whamcloud.maloo@gmail.com>
Reviewed-by: Mike Pershin <tappro@whamcloud.com>

diff --git a/lustre/obdclass/lu_object.c b/lustre/obdclass/lu_object.c
index 9eaa1ae..1f9eb53 100644 (file)
--- a/lustre/obdclass/lu_object.c
+++ b/lustre/obdclass/lu_object.c
@@ -1797,6 +1797,24 @@ static void lu_site_stats_get(cfs_hash_t *hs,
 
 #ifdef __KERNEL__
 
 
 #ifdef __KERNEL__
 

+/*
+ * There exists a potential lock inversion deadlock scenario when using
+ * Lustre on top of ZFS. This occurs between one of ZFS's
+ * buf_hash_table.ht_lock's, and Lustre's lu_sites_guard lock. Essentially,
+ * thread A will take the lu_sites_guard lock and sleep on the ht_lock,
+ * while thread B will take the ht_lock and sleep on the lu_sites_guard
+ * lock. Obviously neither thread will wake and drop their respective hold
+ * on their lock.
+ *
+ * To prevent this from happening we must ensure the lu_sites_guard lock is
+ * not taken while down this code path. ZFS reliably does not set the
+ * __GFP_FS bit in its code paths, so this can be used to determine if it
+ * is safe to take the lu_sites_guard lock.
+ *
+ * Ideally we should accurately return the remaining number of cached
+ * objects without taking the  lu_sites_guard lock, but this is not
+ * possible in the current implementation.
+ */

 static int lu_cache_shrink(SHRINKER_ARGS(sc, nr_to_scan, gfp_mask))
 {
         lu_site_stats_t stats;
 static int lu_cache_shrink(SHRINKER_ARGS(sc, nr_to_scan, gfp_mask))
 {
         lu_site_stats_t stats;


@@ -1806,12 +1824,26 @@ static int lu_cache_shrink(SHRINKER_ARGS(sc, nr_to_scan, gfp_mask))
         int remain = shrink_param(sc, nr_to_scan);
         CFS_LIST_HEAD(splice);
 
         int remain = shrink_param(sc, nr_to_scan);
         CFS_LIST_HEAD(splice);
 

-        if (remain != 0) {
-                if (!(shrink_param(sc, gfp_mask) & __GFP_FS))
+       if (!(shrink_param(sc, gfp_mask) & __GFP_FS)) {
+               if (remain != 0)

                         return -1;
                         return -1;

-                CDEBUG(D_INODE, "Shrink %d objects\n", remain);
+               else
+                       /* We must not take the lu_sites_guard lock when
+                        * __GFP_FS is *not* set because of the deadlock
+                        * possibility detailed above. Additionally,
+                        * since we cannot determine the number of
+                        * objects in the cache without taking this
+                        * lock, we're in a particularly tough spot. As
+                        * a result, we'll just lie and say our cache is
+                        * empty. This _should_ be ok, as we can't
+                        * reclaim objects when __GFP_FS is *not* set
+                        * anyways.
+                        */
+                       return 0;

         }
 
         }
 

+       CDEBUG(D_INODE, "Shrink %d objects\n", remain);
+

         cfs_mutex_lock(&lu_sites_guard);
         cfs_list_for_each_entry_safe(s, tmp, &lu_sites, ls_linkage) {
                 if (shrink_param(sc, nr_to_scan) != 0) {
         cfs_mutex_lock(&lu_sites_guard);
         cfs_list_for_each_entry_safe(s, tmp, &lu_sites, ls_linkage) {
                 if (shrink_param(sc, nr_to_scan) != 0) {

diff --git a/lustre/obdecho/echo_client.c b/lustre/obdecho/echo_client.c
index 3ee8138..b51d57e 100644 (file)
--- a/lustre/obdecho/echo_client.c
+++ b/lustre/obdecho/echo_client.c
@@ -1606,7 +1606,7 @@ static int echo_create_md_object(const struct lu_env *env,
         }
 
         ma->ma_attr.la_mode = mode;
         }
 
         ma->ma_attr.la_mode = mode;

-        ma->ma_attr.la_valid = LA_CTIME;
+       ma->ma_attr.la_valid = LA_CTIME | LA_MODE;

         ma->ma_attr.la_ctime = cfs_time_current_64();
 
         if (name != NULL) {
         ma->ma_attr.la_ctime = cfs_time_current_64();
 
         if (name != NULL) {

diff --git a/lustre/ofd/ofd_obd.c b/lustre/ofd/ofd_obd.c
index 4895d31..1f8e917 100644 (file)
--- a/lustre/ofd/ofd_obd.c
+++ b/lustre/ofd/ofd_obd.c
@@ -729,10 +729,6 @@ static int ofd_statfs(const struct lu_env *env,  struct obd_export *exp,
               osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
               osfs->os_files, osfs->os_ffree, osfs->os_state);
 
               osfs->os_blocks, osfs->os_bfree, osfs->os_bavail,
               osfs->os_files, osfs->os_ffree, osfs->os_state);
 

-       if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
-                                ofd->ofd_lut.lut_lsd.lsd_ost_index))
-               osfs->os_bfree = osfs->os_bavail = 2;
-

        if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
                                 ofd->ofd_lut.lut_lsd.lsd_ost_index))
                osfs->os_ffree = 0;
        if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOINO,
                                 ofd->ofd_lut.lut_lsd.lsd_ost_index))
                osfs->os_ffree = 0;


@@ -753,6 +749,10 @@ static int ofd_statfs(const struct lu_env *env,  struct obd_export *exp,
                osfs->os_bsize    = 1 << COMPAT_BSIZE_SHIFT;
        }
 
                osfs->os_bsize    = 1 << COMPAT_BSIZE_SHIFT;
        }
 

+       if (OBD_FAIL_CHECK_VALUE(OBD_FAIL_OST_ENOSPC,
+                                ofd->ofd_lut.lut_lsd.lsd_ost_index))
+               osfs->os_bfree = osfs->os_bavail = 2;
+

        EXIT;
 out:
        return rc;
        EXIT;
 out:
        return rc;

diff --git a/lustre/osd-zfs/osd_handler.c b/lustre/osd-zfs/osd_handler.c
index bba18f7..7b85f40 100644 (file)
--- a/lustre/osd-zfs/osd_handler.c
+++ b/lustre/osd-zfs/osd_handler.c
@@ -505,9 +505,17 @@ static int osd_shutdown(const struct lu_env *env, struct osd_device *o)
        RETURN(0);
 }
 
        RETURN(0);
 }
 

+static void osd_xattr_changed_cb(void *arg, uint64_t newval)
+{
+       struct osd_device *osd = arg;
+
+       osd->od_xattr_in_sa = (newval == ZFS_XATTR_SA);
+}
+

 static int osd_mount(const struct lu_env *env,
                     struct osd_device *o, struct lustre_cfg *cfg)
 {
 static int osd_mount(const struct lu_env *env,
                     struct osd_device *o, struct lustre_cfg *cfg)
 {

+       struct dsl_dataset *ds;

        char      *dev  = lustre_cfg_string(cfg, 1);
        dmu_buf_t *rootdb;
        int        rc;
        char      *dev  = lustre_cfg_string(cfg, 1);
        dmu_buf_t *rootdb;
        int        rc;


@@ -529,6 +537,13 @@ static int osd_mount(const struct lu_env *env,
                RETURN(rc);
        }
 
                RETURN(rc);
        }
 

+       ds = dmu_objset_ds(o->od_objset.os);
+       LASSERT(ds);
+       rc = dsl_prop_register(ds, "xattr", osd_xattr_changed_cb, o);
+       if (rc)
+               CERROR("%s: cat not register xattr callback, ignore: %d\n",
+                      o->od_svname, rc);
+

        rc = __osd_obj2dbuf(env, o->od_objset.os, o->od_objset.root,
                                &rootdb, root_tag);
        if (rc) {
        rc = __osd_obj2dbuf(env, o->od_objset.os, o->od_objset.root,
                                &rootdb, root_tag);
        if (rc) {


@@ -668,6 +683,7 @@ static struct lu_device *osd_device_fini(const struct lu_env *env,
                                         struct lu_device *d)
 {
        struct osd_device *o = osd_dev(d);
                                         struct lu_device *d)
 {
        struct osd_device *o = osd_dev(d);

+       struct dsl_dataset *ds;

        int                rc;
        ENTRY;
 
        int                rc;
        ENTRY;
 


@@ -675,6 +691,11 @@ static struct lu_device *osd_device_fini(const struct lu_env *env,
        osd_oi_fini(env, o);
 
        if (o->od_objset.os) {
        osd_oi_fini(env, o);
 
        if (o->od_objset.os) {

+               ds = dmu_objset_ds(o->od_objset.os);
+               rc = dsl_prop_unregister(ds, "xattr", osd_xattr_changed_cb, o);
+               if (rc)
+                       CERROR("%s: dsl_prop_unregister xattr error %d\n",
+                               o->od_svname, rc);

                arc_remove_prune_callback(o->arc_prune_cb);
                o->arc_prune_cb = NULL;
                osd_sync(env, lu2dt_dev(d));
                arc_remove_prune_callback(o->arc_prune_cb);
                o->arc_prune_cb = NULL;
                osd_sync(env, lu2dt_dev(d));

diff --git a/lustre/osd-zfs/osd_internal.h b/lustre/osd-zfs/osd_internal.h
index cb9b8a5..d3025b7 100644 (file)
--- a/lustre/osd-zfs/osd_internal.h
+++ b/lustre/osd-zfs/osd_internal.h
@@ -147,6 +147,7 @@ struct osd_thread_info {
 
        char                     oti_str[64];
        char                     oti_key[MAXNAMELEN + 1];
 
        char                     oti_str[64];
        char                     oti_key[MAXNAMELEN + 1];

+       struct lustre_mdt_attrs oti_mdt_attrs;

 
        struct lu_attr           oti_la;
        struct osa_attr          oti_osa;
 
        struct lu_attr           oti_la;
        struct osa_attr          oti_osa;


@@ -215,6 +216,7 @@ struct osd_device {
        uint64_t                 od_ost_compat_grp0;
 
        unsigned int             od_rdonly:1,
        uint64_t                 od_ost_compat_grp0;
 
        unsigned int             od_rdonly:1,

+                                od_xattr_in_sa:1,

                                 od_quota_iused_est:1;
        char                     od_mntdev[128];
        char                     od_svname[128];
                                 od_quota_iused_est:1;
        char                     od_mntdev[128];
        char                     od_svname[128];


@@ -409,6 +411,31 @@ int osd_xattr_del(const struct lu_env *env, struct dt_object *dt,
                  struct lustre_capa *capa);
 int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
                   struct lu_buf *lb, struct lustre_capa *capa);
                  struct lustre_capa *capa);
 int osd_xattr_list(const struct lu_env *env, struct dt_object *dt,
                   struct lu_buf *lb, struct lustre_capa *capa);

+void __osd_xattr_declare_set(const struct lu_env *env, struct osd_object *obj,
+                       int vallen, const char *name, struct osd_thandle *oh);
+int __osd_sa_xattr_set(const struct lu_env *env, struct osd_object *obj,
+                      const struct lu_buf *buf, const char *name, int fl,
+                      struct osd_thandle *oh);;
+int __osd_xattr_set(const struct lu_env *env, struct osd_object *obj,
+                   const struct lu_buf *buf, const char *name, int fl,
+                   struct osd_thandle *oh);
+static inline int
+osd_xattr_set_internal(const struct lu_env *env, struct osd_object *obj,
+                      const struct lu_buf *buf, const char *name, int fl,
+                      struct osd_thandle *oh, struct lustre_capa *capa)
+{
+       int rc;
+
+       if (osd_obj2dev(obj)->od_xattr_in_sa) {
+               rc = __osd_sa_xattr_set(env, obj, buf, name, fl, oh);
+               if (rc == -EFBIG)
+                       rc = __osd_xattr_set(env, obj, buf, name, fl, oh);
+       } else {
+               rc = __osd_xattr_set(env, obj, buf, name, fl, oh);
+       }
+
+       return rc;
+}

 
 #endif
 #endif /* _OSD_INTERNAL_H */
 
 #endif
 #endif /* _OSD_INTERNAL_H */

diff --git a/lustre/osd-zfs/osd_object.c b/lustre/osd-zfs/osd_object.c
index e88f4e1..bb10a1f 100644 (file)
--- a/lustre/osd-zfs/osd_object.c
+++ b/lustre/osd-zfs/osd_object.c
@@ -125,8 +125,10 @@ osd_object_sa_dirty_add(struct osd_object *obj, struct osd_thandle *oh)
                return;
 
        cfs_down(&oh->ot_sa_lock);
                return;
 
        cfs_down(&oh->ot_sa_lock);

+       cfs_write_lock(&obj->oo_attr_lock);

        if (likely(cfs_list_empty(&obj->oo_sa_linkage)))
                cfs_list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);
        if (likely(cfs_list_empty(&obj->oo_sa_linkage)))
                cfs_list_add(&obj->oo_sa_linkage, &oh->ot_sa_list);

+       cfs_write_unlock(&obj->oo_attr_lock);

        cfs_up(&oh->ot_sa_lock);
 }
 
        cfs_up(&oh->ot_sa_lock);
 }
 


@@ -142,7 +144,9 @@ void osd_object_sa_dirty_rele(struct osd_thandle *oh)
                obj = cfs_list_entry(oh->ot_sa_list.next,
                                     struct osd_object, oo_sa_linkage);
                sa_spill_rele(obj->oo_sa_hdl);
                obj = cfs_list_entry(oh->ot_sa_list.next,
                                     struct osd_object, oo_sa_linkage);
                sa_spill_rele(obj->oo_sa_hdl);

+               cfs_write_lock(&obj->oo_attr_lock);

                cfs_list_del_init(&obj->oo_sa_linkage);
                cfs_list_del_init(&obj->oo_sa_linkage);

+               cfs_write_unlock(&obj->oo_attr_lock);

        }
        cfs_up(&oh->ot_sa_lock);
 }
        }
        cfs_up(&oh->ot_sa_lock);
 }


@@ -731,6 +735,10 @@ static int osd_attr_get(const struct lu_env *env,
         * from within sa_object_size() can block on a mutex, so
         * we can't call sa_object_size() holding rwlock */
        sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);
         * from within sa_object_size() can block on a mutex, so
         * we can't call sa_object_size() holding rwlock */
        sa_object_size(obj->oo_sa_hdl, &blksize, &blocks);

+       /* we do not control size of indices, so always calculate
+        * it from number of blocks reported by DMU */
+       if (S_ISDIR(attr->la_mode))
+               attr->la_size = 512 * blocks;

        /* Block size may be not set; suggest maximal I/O transfers. */
        if (blksize == 0)
                blksize = 1ULL << SPA_MAXBLOCKSHIFT;
        /* Block size may be not set; suggest maximal I/O transfers. */
        if (blksize == 0)
                blksize = 1ULL << SPA_MAXBLOCKSHIFT;


@@ -1079,6 +1087,9 @@ static int osd_declare_object_create(const struct lu_env *env,
 
        dmu_tx_hold_sa_create(oh->ot_tx, ZFS_SA_BASE_ATTR_SIZE);
 
 
        dmu_tx_hold_sa_create(oh->ot_tx, ZFS_SA_BASE_ATTR_SIZE);
 

+       __osd_xattr_declare_set(env, obj, sizeof(struct lustre_mdt_attrs),
+                               XATTR_NAME_LMA, oh);
+

        RETURN(osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
                                 false, NULL, false));
 }
        RETURN(osd_declare_quota(env, osd, attr->la_uid, attr->la_gid, 1, oh,
                                 false, NULL, false));
 }


@@ -1361,6 +1372,24 @@ static osd_obj_type_f osd_create_type_f(enum dt_format_type type)
 /*
  * Primitives for directory (i.e. ZAP) handling
  */
 /*
  * Primitives for directory (i.e. ZAP) handling
  */

+static inline int osd_init_lma(const struct lu_env *env, struct osd_object *obj,
+                              const struct lu_fid *fid, struct osd_thandle *oh)
+{
+       struct osd_thread_info  *info = osd_oti_get(env);
+       struct lustre_mdt_attrs *lma = &info->oti_mdt_attrs;
+       struct lu_buf            buf;
+       int rc;
+
+       lustre_lma_init(lma, fid);
+       lustre_lma_swab(lma);
+       buf.lb_buf = lma;
+       buf.lb_len = sizeof(*lma);
+
+       rc = osd_xattr_set_internal(env, obj, &buf, XATTR_NAME_LMA,
+                                   LU_XATTR_CREATE, oh, BYPASS_CAPA);
+
+       return rc;
+}

 
 /*
  * Concurrency: @dt is write locked.
 
 /*
  * Concurrency: @dt is write locked.


@@ -1436,6 +1465,14 @@ static int osd_object_create(const struct lu_env *env, struct dt_object *dt,
        LASSERT(ergo(rc == 0, dt_object_exists(dt)));
        LASSERT(osd_invariant(obj));
 
        LASSERT(ergo(rc == 0, dt_object_exists(dt)));
        LASSERT(osd_invariant(obj));
 

+       rc = osd_init_lma(env, obj, fid, oh);
+       if (rc) {
+               CERROR("%s: can not set LMA on "DFID": rc = %d\n",
+                      osd->od_svname, PFID(fid), rc);
+               /* ignore errors during LMA initialization */
+               rc = 0;
+       }
+

 out:
        cfs_up(&obj->oo_guard);
        RETURN(rc);
 out:
        cfs_up(&obj->oo_guard);
        RETURN(rc);

diff --git a/lustre/osd-zfs/osd_xattr.c b/lustre/osd-zfs/osd_xattr.c
index 9ea4fd1..94372f2 100644 (file)
--- a/lustre/osd-zfs/osd_xattr.c
+++ b/lustre/osd-zfs/osd_xattr.c
@@ -435,7 +435,7 @@ int __osd_sa_xattr_set(const struct lu_env *env, struct osd_object *obj,
        return rc;
 }
 
        return rc;
 }
 

-static int
+int

 __osd_xattr_set(const struct lu_env *env, struct osd_object *obj,
                const struct lu_buf *buf, const char *name, int fl,
                struct osd_thandle *oh)
 __osd_xattr_set(const struct lu_env *env, struct osd_object *obj,
                const struct lu_buf *buf, const char *name, int fl,
                struct osd_thandle *oh)


@@ -549,8 +549,8 @@ out:
 }
 
 int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
 }
 
 int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,

-               const struct lu_buf *buf, const char *name, int fl,
-               struct thandle *handle, struct lustre_capa *capa)
+                 const struct lu_buf *buf, const char *name, int fl,
+                 struct thandle *handle, struct lustre_capa *capa)

 {
        struct osd_object  *obj = osd_dt_obj(dt);
        struct osd_thandle *oh;
 {
        struct osd_object  *obj = osd_dt_obj(dt);
        struct osd_thandle *oh;


@@ -567,10 +567,7 @@ int osd_xattr_set(const struct lu_env *env, struct dt_object *dt,
        cfs_down(&obj->oo_guard);
        CDEBUG(D_INODE, "Setting xattr %s with size %d\n",
                name, (int)buf->lb_len);
        cfs_down(&obj->oo_guard);
        CDEBUG(D_INODE, "Setting xattr %s with size %d\n",
                name, (int)buf->lb_len);

-       rc = __osd_sa_xattr_set(env, obj, buf, name, fl, oh);
-       /* place xattr in dnode if SA is full */
-       if (rc == -EFBIG)
-               rc = __osd_xattr_set(env, obj, buf, name, fl, oh);
+       rc = osd_xattr_set_internal(env, obj, buf, name, fl, oh, capa);

        cfs_up(&obj->oo_guard);
 
        RETURN(rc);
        cfs_up(&obj->oo_guard);
 
        RETURN(rc);

diff --git a/lustre/ost/ost_handler.c b/lustre/ost/ost_handler.c
index 37c0304..61e0ba3 100644 (file)
--- a/lustre/ost/ost_handler.c
+++ b/lustre/ost/ost_handler.c
@@ -555,9 +555,20 @@ static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc,
                    OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
                        int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
                        int len = desc->bd_iov[i].kiov_len;
                    OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
                        int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
                        int len = desc->bd_iov[i].kiov_len;

+                       struct page *np = cfs_alloc_page(CFS_ALLOC_STD);

                        char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
                        char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;

-                       memcpy(ptr, "bad3", min(4, len));
-                       kunmap(desc->bd_iov[i].kiov_page);
+
+                       if (np) {
+                               char *ptr2 = kmap(np) + off;
+
+                               memcpy(ptr2, ptr, len);
+                               memcpy(ptr2, "bad3", min(4, len));
+                               kunmap(np);
+                               cfs_page_unpin(desc->bd_iov[i].kiov_page);
+                               desc->bd_iov[i].kiov_page = np;
+                       } else {
+                               CERROR("can't alloc page for corruption\n");
+                       }

                }
                cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
                                  desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,
                }
                cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
                                  desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,


@@ -569,11 +580,20 @@ static __u32 ost_checksum_bulk(struct ptlrpc_bulk_desc *desc, int opc,
                    OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
                        int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
                        int len = desc->bd_iov[i].kiov_len;
                    OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
                        int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
                        int len = desc->bd_iov[i].kiov_len;

+                       struct page *np = cfs_alloc_page(CFS_ALLOC_STD);

                        char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
                        char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;

-                       memcpy(ptr, "bad4", min(4, len));
-                       kunmap(desc->bd_iov[i].kiov_page);
-                       /* nobody should use corrupted page again */
-                       ClearPageUptodate(desc->bd_iov[i].kiov_page);
+
+                       if (np) {
+                               char *ptr2 = kmap(np) + off;
+
+                               memcpy(ptr2, ptr, len);
+                               memcpy(ptr2, "bad4", min(4, len));
+                               kunmap(np);
+                               cfs_page_unpin(desc->bd_iov[i].kiov_page);
+                               desc->bd_iov[i].kiov_page = np;
+                       } else {
+                               CERROR("can't alloc page for corruption\n");
+                       }

                }
        }
 
                }
        }
 

diff --git a/lustre/tests/conf-sanity.sh b/lustre/tests/conf-sanity.sh
index f1ec2de..a75f054 100644 (file)
--- a/lustre/tests/conf-sanity.sh
+++ b/lustre/tests/conf-sanity.sh
@@ -19,6 +19,10 @@ if [ "$FAILURE_MODE" = "HARD" ]; then
        ALWAYS_EXCEPT="$ALWAYS_EXCEPT $CONFIG_EXCEPTIONS"
 fi
 
        ALWAYS_EXCEPT="$ALWAYS_EXCEPT $CONFIG_EXCEPTIONS"
 fi
 

+# LU-2059
+ALWAYS_EXCEPT="$ALWAYS_EXCEPT 5d 19b 21b 27a"
+
+

 SRCDIR=`dirname $0`
 PATH=$PWD/$SRCDIR:$SRCDIR:$SRCDIR/../utils:$PATH
 
 SRCDIR=`dirname $0`
 PATH=$PWD/$SRCDIR:$SRCDIR:$SRCDIR/../utils:$PATH
 


@@ -558,6 +562,11 @@ is_blkdev () {
 #
 
 test_17() {
 #
 
 test_17() {

+       if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then
+               skip "Only applicable to ldiskfs-based MDTs"
+               return
+       fi
+

         setup
         check_mount || return 41
         cleanup || return $?
         setup
         check_mount || return 41
         cleanup || return $?


@@ -1191,6 +1200,11 @@ cleanup_32() {
 }
 
 test_32a() {
 }
 
 test_32a() {

+       if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then
+               skip "Only applicable to ldiskfs-based MDTs"
+               return
+       fi
+

        client_only && skip "client only testing" && return 0
        [ "$NETTYPE" = "tcp" ] || { skip "NETTYPE != tcp" && return 0; }
        [ -z "$TUNEFS" ] && skip_env "No tunefs" && return 0
        client_only && skip "client only testing" && return 0
        [ "$NETTYPE" = "tcp" ] || { skip "NETTYPE != tcp" && return 0; }
        [ -z "$TUNEFS" ] && skip_env "No tunefs" && return 0


@@ -1258,6 +1272,11 @@ test_32a() {
 run_test 32a "Upgrade from 1.8 (not live)"
 
 test_32b() {
 run_test 32a "Upgrade from 1.8 (not live)"
 
 test_32b() {

+       if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then
+               skip "Only applicable to ldiskfs-based MDTs"
+               return
+       fi
+

        client_only && skip "client only testing" && return 0
        [ "$NETTYPE" = "tcp" ] || { skip "NETTYPE != tcp" && return 0; }
        [ -z "$TUNEFS" ] && skip_env "No tunefs" && return
        client_only && skip "client only testing" && return 0
        [ "$NETTYPE" = "tcp" ] || { skip "NETTYPE != tcp" && return 0; }
        [ -z "$TUNEFS" ] && skip_env "No tunefs" && return


@@ -1691,6 +1710,11 @@ test_37() {
 run_test 37 "verify set tunables works for symlink device"
 
 test_38() { # bug 14222
 run_test 37 "verify set tunables works for symlink device"
 
 test_38() { # bug 14222

+       if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then
+               skip "Only applicable to ldiskfs-based MDTs"
+               return
+       fi
+

        setup
        # like runtests
        COUNT=10
        setup
        # like runtests
        COUNT=10


@@ -2441,6 +2465,11 @@ diff_files_xattrs()
 }
 
 test_52() {
 }
 
 test_52() {

+       if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then
+               skip "Only applicable to ldiskfs-based MDTs"
+               return
+       fi
+

        start_mds
        [ $? -eq 0 ] || { error "Unable to start MDS"; return 1; }
        start_ost
        start_mds
        [ $? -eq 0 ] || { error "Unable to start MDS"; return 1; }
        start_ost


@@ -2646,6 +2675,11 @@ test_53b() {
 run_test 53b "check MDT thread count params"
 
 test_54a() {
 run_test 53b "check MDT thread count params"
 
 test_54a() {

+       if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then
+               skip "Only applicable to ldiskfs-based MDTs"
+               return
+       fi
+

     do_rpc_nodes $(facet_host ost1) run_llverdev $(ostdevname 1) -p
     [ $? -eq 0 ] || error "llverdev failed!"
     reformat_and_config
     do_rpc_nodes $(facet_host ost1) run_llverdev $(ostdevname 1) -p
     [ $? -eq 0 ] || error "llverdev failed!"
     reformat_and_config


@@ -2653,6 +2687,11 @@ test_54a() {
 run_test 54a "test llverdev and partial verify of device"
 
 test_54b() {
 run_test 54a "test llverdev and partial verify of device"
 
 test_54b() {

+       if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then
+               skip "Only applicable to ldiskfs-based MDTs"
+               return
+       fi
+

     setup
     run_llverfs $MOUNT -p
     [ $? -eq 0 ] || error "llverfs failed!"
     setup
     run_llverfs $MOUNT -p
     [ $? -eq 0 ] || error "llverfs failed!"


@@ -2667,6 +2706,11 @@ lov_objid_size()
 }
 
 test_55() {
 }
 
 test_55() {

+       if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then
+               skip "Only applicable to ldiskfs-based MDTs"
+               return
+       fi
+

        local mdsdev=$(mdsdevname 1)
        local mdsvdev=$(mdsvdevname 1)
 
        local mdsdev=$(mdsdevname 1)
        local mdsvdev=$(mdsvdevname 1)
 


@@ -2742,8 +2786,8 @@ count_osts() {
 }
 
 test_58() { # bug 22658
 }
 
 test_58() { # bug 22658

-       if [ $(facet_fstype mds) == zfs ]; then
-               skip "Does not work with ZFS-based MDTs yet"
+       if [ $(facet_fstype mds) != ldiskfs ]; then
+               skip "Only applicable to ldiskfs-based MDTs"

                return
        fi
        setup_noconfig
                return
        fi
        setup_noconfig


@@ -2890,6 +2934,11 @@ test_61() { # LU-80
 run_test 61 "large xattr"
 
 test_62() {
 run_test 61 "large xattr"
 
 test_62() {

+       if [ $(facet_fstype $SINGLEMDS) != ldiskfs ]; then
+               skip "Only applicable to ldiskfs-based MDTs"
+               return
+       fi
+

        # MRP-118
        local mdsdev=$(mdsdevname 1)
        local ostdev=$(ostdevname 1)
        # MRP-118
        local mdsdev=$(mdsdevname 1)
        local ostdev=$(ostdevname 1)

diff --git a/lustre/tests/sanity.sh b/lustre/tests/sanity.sh
index c31107c..10b1621 100644 (file)
--- a/lustre/tests/sanity.sh
+++ b/lustre/tests/sanity.sh
@@ -3985,6 +3985,11 @@ run_test 56w "check lfs_migrate -c stripe_count works"
 
 test_57a() {
        # note test will not do anything if MDS is not local
 
 test_57a() {
        # note test will not do anything if MDS is not local

+       if [ "$(facet_type_fstype MDS)" != ldiskfs ]; then
+               skip "Only applicable to ldiskfs-based MDTs"
+               return
+       fi
+

        remote_mds_nodsh && skip "remote MDS with nodsh" && return
        local MNTDEV="osd*.*MDT*.mntdev"
        DEV=$(do_facet $SINGLEMDS lctl get_param -n $MNTDEV)
        remote_mds_nodsh && skip "remote MDS with nodsh" && return
        local MNTDEV="osd*.*MDT*.mntdev"
        DEV=$(do_facet $SINGLEMDS lctl get_param -n $MNTDEV)


@@ -4000,6 +4005,11 @@ test_57a() {
 run_test 57a "verify MDS filesystem created with large inodes =="
 
 test_57b() {
 run_test 57a "verify MDS filesystem created with large inodes =="
 
 test_57b() {

+       if [ "$(facet_type_fstype MDS)" != ldiskfs ]; then
+               skip "Only applicable to ldiskfs-based MDTs"
+               return
+       fi
+

        remote_mds_nodsh && skip "remote MDS with nodsh" && return
        local dir=$DIR/d57b
 
        remote_mds_nodsh && skip "remote MDS with nodsh" && return
        local dir=$DIR/d57b
 

diff --git a/lustre/tests/test-framework.sh b/lustre/tests/test-framework.sh
index 9afb3ea..aa17bb8 100644 (file)
--- a/lustre/tests/test-framework.sh
+++ b/lustre/tests/test-framework.sh
@@ -455,6 +455,7 @@ load_modules_local() {
         grep -q -w jbd2 $SYMLIST || { modprobe jbd2 2>/dev/null || true; }
                [ "$LQUOTA" != "no" ] && load_module quota/lquota $LQUOTAOPTS
                if [[ $(node_fstypes $HOSTNAME) == *zfs* ]]; then
         grep -q -w jbd2 $SYMLIST || { modprobe jbd2 2>/dev/null || true; }
                [ "$LQUOTA" != "no" ] && load_module quota/lquota $LQUOTAOPTS
                if [[ $(node_fstypes $HOSTNAME) == *zfs* ]]; then

+                       modprobe zfs

                        load_module osd-zfs/osd_zfs
                fi
                load_module mgs/mgs
                        load_module osd-zfs/osd_zfs
                fi
                load_module mgs/mgs