Whamcloud - gitweb
LU-3268 lod: support BE servers in lod_verify_striping()
[fs/lustre-release.git] / lustre / lod / lod_lov.c
index 81a5ed1..5426554 100644 (file)
@@ -23,7 +23,7 @@
  * Copyright  2009 Sun Microsystems, Inc. All rights reserved
  * Use is subject to license terms.
  *
- * Copyright (c) 2012, Intel Corporation.
+ * Copyright (c) 2012, 2013, Intel Corporation.
  */
 /*
  * lustre/lod/lod_lov.c
@@ -171,9 +171,12 @@ out:
  * \param osp - is the name of OSP device name about to be added
  * \param index - is the OSP index
  * \param gen - is the generation number
+ * \param tgt_index - is the group of the OSP.
+ * \param type - is the type of device (mdc or osc)
  */
 int lod_add_device(const struct lu_env *env, struct lod_device *lod,
-                  char *osp, unsigned index, unsigned gen, int active)
+                  char *osp, unsigned index, unsigned gen, int tgt_index,
+                  char *type, int active)
 {
        struct obd_connect_data *data = NULL;
        struct obd_export       *exp = NULL;
@@ -182,7 +185,7 @@ int lod_add_device(const struct lu_env *env, struct lod_device *lod,
        struct dt_device        *d;
        int                      rc;
        struct lod_tgt_desc     *tgt_desc;
-       struct lod_tgt_descs    *ltd = &lod->lod_ost_descs;
+       struct lod_tgt_descs    *ltd;
        struct obd_uuid         obd_uuid;
        ENTRY;
 
@@ -211,6 +214,48 @@ int lod_add_device(const struct lu_env *env, struct lod_device *lod,
        data->ocd_version = LUSTRE_VERSION_CODE;
        data->ocd_index = index;
 
+       if (strcmp(LUSTRE_OSC_NAME, type) == 0) {
+               data->ocd_connect_flags |= OBD_CONNECT_AT |
+                                          OBD_CONNECT_FULL20 |
+                                          OBD_CONNECT_INDEX |
+#ifdef HAVE_LRU_RESIZE_SUPPORT
+                                          OBD_CONNECT_LRU_RESIZE |
+#endif
+                                          OBD_CONNECT_MDS |
+                                          OBD_CONNECT_OSS_CAPA |
+                                          OBD_CONNECT_REQPORTAL |
+                                          OBD_CONNECT_SKIP_ORPHAN |
+                                          OBD_CONNECT_FID |
+                                          OBD_CONNECT_LVB_TYPE |
+                                          OBD_CONNECT_VERSION |
+                                          OBD_CONNECT_PINGLESS;
+
+               data->ocd_group = tgt_index;
+               ltd = &lod->lod_ost_descs;
+       } else {
+               struct obd_import *imp = obd->u.cli.cl_import;
+
+               data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
+               data->ocd_connect_flags |= OBD_CONNECT_ACL |
+                                          OBD_CONNECT_MDS_CAPA |
+                                          OBD_CONNECT_OSS_CAPA |
+                                          OBD_CONNECT_IBITS |
+                                          OBD_CONNECT_MDS_MDS |
+                                          OBD_CONNECT_FID |
+                                          OBD_CONNECT_AT |
+                                          OBD_CONNECT_FULL20;
+               /* XXX set MDS-MDS flags, remove this when running this
+                * on client*/
+               data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS;
+               spin_lock(&imp->imp_lock);
+               imp->imp_server_timeout = 1;
+               spin_unlock(&imp->imp_lock);
+               imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
+               CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
+                     obd->obd_name);
+               ltd = &lod->lod_mdt_descs;
+       }
+
        rc = obd_connect(env, &exp, obd, &obd->obd_uuid, data, NULL);
        OBD_FREE_PTR(data);
        if (rc) {
@@ -220,7 +265,7 @@ int lod_add_device(const struct lu_env *env, struct lod_device *lod,
        }
 
        LASSERT(obd->obd_lu_dev);
-       LASSERT(obd->obd_lu_dev->ld_site = lod->lod_dt_dev.dd_lu_dev.ld_site);
+       LASSERT(obd->obd_lu_dev->ld_site == lod->lod_dt_dev.dd_lu_dev.ld_site);
 
        ldev = obd->obd_lu_dev;
        d = lu2dt_dev(ldev);
@@ -273,27 +318,29 @@ int lod_add_device(const struct lu_env *env, struct lod_device *lod,
                }
        }
 
-       /* pool and qos are not supported for MDS stack yet */
-       rc = lod_ost_pool_add(&lod->lod_pool_info, index,
-                             lod->lod_osts_size);
-       if (rc) {
-               CERROR("%s: can't set up pool, failed with %d\n",
-                      obd->obd_name, rc);
-               GOTO(out_mutex, rc);
-       }
+       if (!strcmp(LUSTRE_OSC_NAME, type)) {
+               /* pool and qos are not supported for MDS stack yet */
+               rc = lod_ost_pool_add(&lod->lod_pool_info, index,
+                                     lod->lod_osts_size);
+               if (rc) {
+                       CERROR("%s: can't set up pool, failed with %d\n",
+                              obd->obd_name, rc);
+                       GOTO(out_mutex, rc);
+               }
 
-       rc = qos_add_tgt(lod, tgt_desc);
-       if (rc) {
-               CERROR("%s: qos_add_tgt failed with %d\n",
-                       obd->obd_name, rc);
-               GOTO(out_pool, rc);
-       }
+               rc = qos_add_tgt(lod, tgt_desc);
+               if (rc) {
+                       CERROR("%s: qos_add_tgt failed with %d\n",
+                               obd->obd_name, rc);
+                       GOTO(out_pool, rc);
+               }
 
-       /* The new OST is now a full citizen */
-       if (index >= lod->lod_desc.ld_tgt_count)
-               lod->lod_desc.ld_tgt_count = index + 1;
-       if (active)
-               lod->lod_desc.ld_active_tgt_count++;
+               /* The new OST is now a full citizen */
+               if (index >= lod->lod_desc.ld_tgt_count)
+                       lod->lod_desc.ld_tgt_count = index + 1;
+               if (active)
+                       lod->lod_desc.ld_active_tgt_count++;
+       }
 
        LTD_TGT(ltd, index) = tgt_desc;
        cfs_bitmap_set(ltd->ltd_tgt_bitmap, index);
@@ -376,7 +423,10 @@ int lod_del_device(const struct lu_env *env, struct lod_device *lod,
 
        CDEBUG(D_CONFIG, "osp:%s idx:%d gen:%d\n", osp, idx, gen);
 
-       obd = class_name2obd(osp);
+       obd_str2uuid(&uuid, osp);
+
+       obd = class_find_client_obd(&uuid, LUSTRE_OSP_NAME,
+                                  &lod->lod_dt_dev.dd_lu_dev.ld_obd->obd_uuid);
        if (obd == NULL) {
                CERROR("can't find %s device\n", osp);
                RETURN(-EINVAL);
@@ -450,6 +500,7 @@ int lod_generate_and_set_lovea(const struct lu_env *env,
        struct lov_ost_data_v1  *objs;
        __u32                    magic;
        int                      i, rc, lmm_size;
+       int                      cplen = 0;
        ENTRY;
 
        LASSERT(lo);
@@ -467,8 +518,8 @@ int lod_generate_and_set_lovea(const struct lu_env *env,
 
        lmm->lmm_magic = cpu_to_le32(magic);
        lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
-       lmm->lmm_object_id = cpu_to_le64(fid_ver_oid(fid));
-       lmm->lmm_object_seq = cpu_to_le64(fid_seq(fid));
+       fid_to_lmm_oi(fid, &lmm->lmm_oi);
+       lmm_oi_cpu_to_le(&lmm->lmm_oi, &lmm->lmm_oi);
        lmm->lmm_stripe_size = cpu_to_le32(lo->ldo_stripe_size);
        lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_stripenr);
        lmm->lmm_layout_gen = 0;
@@ -476,7 +527,10 @@ int lod_generate_and_set_lovea(const struct lu_env *env,
                objs = &lmm->lmm_objects[0];
        } else {
                struct lov_mds_md_v3 *v3 = (struct lov_mds_md_v3 *) lmm;
-               strncpy(v3->lmm_pool_name, lo->ldo_pool, LOV_MAXPOOLNAME);
+               cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool,
+                               sizeof(v3->lmm_pool_name));
+               if (cplen >= sizeof(v3->lmm_pool_name))
+                       RETURN(-E2BIG);
                objs = &v3->lmm_objects[0];
        }
 
@@ -489,11 +543,10 @@ int lod_generate_and_set_lovea(const struct lu_env *env,
                LASSERT(lo->ldo_stripe[i]);
                fid = lu_object_fid(&lo->ldo_stripe[i]->do_lu);
 
-               rc = fid_ostid_pack(fid, &info->lti_ostid);
+               rc = fid_to_ostid(fid, &info->lti_ostid);
                LASSERT(rc == 0);
 
-               objs[i].l_object_id  = cpu_to_le64(info->lti_ostid.oi_id);
-               objs[i].l_object_seq = cpu_to_le64(info->lti_ostid.oi_seq);
+               ostid_cpu_to_le(&info->lti_ostid, &objs[i].l_ost_oi);
                objs[i].l_ost_gen    = cpu_to_le32(0);
                rc = lod_fld_lookup(env, lod, fid, &index, LU_SEQ_RANGE_OST);
                if (rc < 0) {
@@ -562,6 +615,7 @@ int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt,
        struct dt_object        *next = dt_object_child(dt);
        struct lov_user_md_v3   *v3;
        int                      rc;
+       int                      cplen = 0;
        ENTRY;
 
        LASSERT(S_ISDIR(dt->do_lu.lo_header->loh_attr));
@@ -586,13 +640,17 @@ int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt,
 
        v3->lmm_magic = cpu_to_le32(LOV_MAGIC_V3);
        v3->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
-       v3->lmm_object_id = 0;
-       v3->lmm_object_seq = 0;
        v3->lmm_stripe_size = cpu_to_le32(lo->ldo_def_stripe_size);
        v3->lmm_stripe_count = cpu_to_le16(lo->ldo_def_stripenr);
        v3->lmm_stripe_offset = cpu_to_le16(lo->ldo_def_stripe_offset);
-       if (lo->ldo_pool)
-               strncpy(v3->lmm_pool_name, lo->ldo_pool, LOV_MAXPOOLNAME);
+       if (lo->ldo_pool) {
+               cplen = strlcpy(v3->lmm_pool_name, lo->ldo_pool,
+                               sizeof(v3->lmm_pool_name));
+               if (cplen >= sizeof(v3->lmm_pool_name)) {
+                       OBD_FREE_PTR(v3);
+                       RETURN(-E2BIG);
+               }
+       }
 
        info->lti_buf.lb_buf = v3;
        info->lti_buf.lb_len = sizeof(*v3);
@@ -615,7 +673,9 @@ int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
        struct lod_device       *md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
        struct lu_object        *o, *n;
        struct lu_device        *nd;
-       int                     i, idx, rc = 0;
+       struct dt_object       **stripe;
+       int                      stripe_len;
+       int                      i, idx, rc = 0;
        ENTRY;
 
        LASSERT(lo);
@@ -623,17 +683,17 @@ int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
        LASSERT(lo->ldo_stripenr > 0);
        LASSERT(lo->ldo_stripe_size > 0);
 
-       i = sizeof(struct dt_object *) * lo->ldo_stripenr;
-       OBD_ALLOC(lo->ldo_stripe, i);
-       if (lo->ldo_stripe == NULL)
-               GOTO(out, rc = -ENOMEM);
-       lo->ldo_stripes_allocated = lo->ldo_stripenr;
+       stripe_len = lo->ldo_stripenr;
+       OBD_ALLOC(stripe, sizeof(stripe[0]) * stripe_len);
+       if (stripe == NULL)
+               RETURN(-ENOMEM);
 
        for (i = 0; i < lo->ldo_stripenr; i++) {
-               info->lti_ostid.oi_id = le64_to_cpu(objs[i].l_object_id);
-               info->lti_ostid.oi_seq = le64_to_cpu(objs[i].l_object_seq);
+               ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
                idx = le64_to_cpu(objs[i].l_ost_idx);
-               fid_ostid_unpack(&info->lti_fid, &info->lti_ostid, idx);
+               rc = ostid_to_fid(&info->lti_fid, &info->lti_ostid, idx);
+               if (rc != 0)
+                       GOTO(out, rc);
                LASSERTF(fid_is_sane(&info->lti_fid), ""DFID" insane!\n",
                         PFID(&info->lti_fid));
                /*
@@ -649,6 +709,9 @@ int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
                nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
                lod_putref(md, &md->lod_ost_descs);
 
+               /* In the function below, .hs_keycmp resolves to
+                * u_obj_hop_keycmp() */
+               /* coverity[overrun-buffer-val] */
                o = lu_object_find_at(env, nd, &info->lti_fid, NULL);
                if (IS_ERR(o))
                        GOTO(out, rc = PTR_ERR(o));
@@ -656,10 +719,21 @@ int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
                n = lu_object_locate(o->lo_header, nd->ld_type);
                LASSERT(n);
 
-               lo->ldo_stripe[i] = container_of(n, struct dt_object, do_lu);
+               stripe[i] = container_of(n, struct dt_object, do_lu);
        }
 
 out:
+       if (rc != 0) {
+               for (i = 0; i < stripe_len; i++)
+                       if (stripe[i] != NULL)
+                               lu_object_put(env, &stripe[i]->do_lu);
+
+               OBD_FREE(stripe, sizeof(stripe[0]) * stripe_len);
+       } else {
+               lo->ldo_stripe = stripe;
+               lo->ldo_stripes_allocated = stripe_len;
+       }
+
        RETURN(rc);
 }
 
@@ -761,92 +835,131 @@ int lod_verify_striping(struct lod_device *d, const struct lu_buf *buf,
                        int specific)
 {
        struct lov_user_md_v1   *lum;
-       struct lov_user_md_v3   *v3 = NULL;
+       struct lov_user_md_v3   *lum3;
        struct pool_desc        *pool = NULL;
-       int                      rc;
+       __u32                    magic;
+       __u32                    stripe_size;
+       __u16                    stripe_count;
+       __u16                    stripe_offset;
+       size_t                   lum_size;
+       int                      rc = 0;
        ENTRY;
 
        lum = buf->lb_buf;
 
-       if (lum->lmm_magic != LOV_USER_MAGIC_V1 &&
-           lum->lmm_magic != LOV_USER_MAGIC_V3 &&
-           lum->lmm_magic != LOV_MAGIC_V1_DEF &&
-           lum->lmm_magic != LOV_MAGIC_V3_DEF) {
-               CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n",
-                      lum->lmm_magic);
-               RETURN(-EINVAL);
+       LASSERT(sizeof(*lum) < sizeof(*lum3));
+
+       if (buf->lb_len < sizeof(*lum)) {
+               CDEBUG(D_IOCTL, "buf len %zd too small for lov_user_md\n",
+                      buf->lb_len);
+               GOTO(out, rc = -EINVAL);
+       }
+
+       magic = le32_to_cpu(lum->lmm_magic);
+       if (magic != LOV_USER_MAGIC_V1 &&
+           magic != LOV_USER_MAGIC_V3 &&
+           magic != LOV_MAGIC_V1_DEF &&
+           magic != LOV_MAGIC_V3_DEF) {
+               CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#x\n", magic);
+               GOTO(out, rc = -EINVAL);
        }
 
-       if ((specific && lum->lmm_pattern != LOV_PATTERN_RAID0) ||
-           (specific == 0 && lum->lmm_pattern != 0)) {
+       if ((specific && le32_to_cpu(lum->lmm_pattern) != LOV_PATTERN_RAID0) ||
+           (!specific && lum->lmm_pattern != 0)) {
                CDEBUG(D_IOCTL, "bad userland stripe pattern: %#x\n",
-                      lum->lmm_pattern);
-               RETURN(-EINVAL);
+                      le32_to_cpu(lum->lmm_pattern));
+               GOTO(out, rc = -EINVAL);
        }
 
        /* 64kB is the largest common page size we see (ia64), and matches the
         * check in lfs */
-       if (lum->lmm_stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
-               CDEBUG(D_IOCTL, "stripe size %u not multiple of %u, fixing\n",
-                      lum->lmm_stripe_size, LOV_MIN_STRIPE_SIZE);
-               RETURN(-EINVAL);
+       stripe_size = le32_to_cpu(lum->lmm_stripe_size);
+       if (stripe_size & (LOV_MIN_STRIPE_SIZE - 1)) {
+               CDEBUG(D_IOCTL, "stripe size %u not a multiple of %u\n",
+                      stripe_size, LOV_MIN_STRIPE_SIZE);
+               GOTO(out, rc = -EINVAL);
        }
 
        /* an offset of -1 is treated as a "special" valid offset */
-       if (lum->lmm_stripe_offset != (typeof(lum->lmm_stripe_offset))(-1)) {
+       stripe_offset = le16_to_cpu(lum->lmm_stripe_offset);
+       if (stripe_offset != (typeof(stripe_offset))-1) {
                /* if offset is not within valid range [0, osts_size) */
-               if (lum->lmm_stripe_offset >= d->lod_osts_size) {
+               if (stripe_offset >= d->lod_osts_size) {
                        CDEBUG(D_IOCTL, "stripe offset %u >= bitmap size %u\n",
-                              lum->lmm_stripe_offset, d->lod_osts_size);
-                       RETURN(-EINVAL);
+                              stripe_offset, d->lod_osts_size);
+                       GOTO(out, rc = -EINVAL);
                }
 
                /* if lmm_stripe_offset is *not* in bitmap */
-               if (!cfs_bitmap_check(d->lod_ost_bitmap,
-                                     lum->lmm_stripe_offset)) {
+               if (!cfs_bitmap_check(d->lod_ost_bitmap, stripe_offset)) {
                        CDEBUG(D_IOCTL, "stripe offset %u not in bitmap\n",
-                              lum->lmm_stripe_offset);
-                       RETURN(-EINVAL);
+                              stripe_offset);
+                       GOTO(out, rc = -EINVAL);
                }
        }
 
-       if (lum->lmm_magic == LOV_USER_MAGIC_V3)
-               v3 = buf->lb_buf;
+       stripe_count = le16_to_cpu(lum->lmm_stripe_count);
+       if (magic == LOV_USER_MAGIC_V1 || magic == LOV_MAGIC_V1_DEF)
+               lum_size = offsetof(struct lov_user_md_v1,
+                                   lmm_objects[stripe_count]);
+       else if (magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF)
+               lum_size = offsetof(struct lov_user_md_v3,
+                                   lmm_objects[stripe_count]);
+       else
+               LBUG();
+
+       if (specific && buf->lb_len != lum_size) {
+               CDEBUG(D_IOCTL, "invalid buf len %zd for lov_user_md with "
+                      "magic %#x and stripe_count %u\n",
+                      buf->lb_len, magic, stripe_count);
+               GOTO(out, rc = -EINVAL);
+       }
 
-       if (v3)
-               pool = lod_find_pool(d, v3->lmm_pool_name);
+       if (!(magic == LOV_USER_MAGIC_V3 || magic == LOV_MAGIC_V3_DEF))
+               goto out;
 
-       if (pool != NULL) {
-               __u16 offs = v3->lmm_stripe_offset;
+       lum3 = buf->lb_buf;
+       if (buf->lb_len < sizeof(*lum3)) {
+               CDEBUG(D_IOCTL, "buf len %zd too small for lov_user_md_v3\n",
+                      buf->lb_len);
+               GOTO(out, rc = -EINVAL);
+       }
 
-               if (offs != (typeof(v3->lmm_stripe_offset))(-1)) {
-                       rc = lod_check_index_in_pool(offs, pool);
-                       if (rc < 0) {
-                               lod_pool_putref(pool);
-                               RETURN(-EINVAL);
-                       }
-               }
+       /* In the function below, .hs_keycmp resolves to
+        * pool_hashkey_keycmp() */
+       /* coverity[overrun-buffer-val] */
+       pool = lod_find_pool(d, lum3->lmm_pool_name);
+       if (pool == NULL)
+               goto out;
+
+       if (stripe_offset != (typeof(stripe_offset))-1) {
+               rc = lod_check_index_in_pool(stripe_offset, pool);
+               if (rc < 0)
+                       GOTO(out, rc = -EINVAL);
+       }
 
-               if (specific && lum->lmm_stripe_count > pool_tgt_count(pool)) {
-                       CDEBUG(D_IOCTL,
-                              "stripe count %u > # OSTs %u in the pool\n",
-                              lum->lmm_stripe_count, pool_tgt_count(pool));
-                       lod_pool_putref(pool);
-                       RETURN(-EINVAL);
-               }
+       if (specific && stripe_count > pool_tgt_count(pool)) {
+               CDEBUG(D_IOCTL,
+                      "stripe count %u > # OSTs %u in the pool\n",
+                      stripe_count, pool_tgt_count(pool));
+               GOTO(out, rc = -EINVAL);
+       }
 
+out:
+       if (pool != NULL)
                lod_pool_putref(pool);
-       }
 
-       RETURN(0);
+       RETURN(rc);
 }
 
 void lod_fix_desc_stripe_size(__u64 *val)
 {
-       if (*val < PTLRPC_MAX_BRW_SIZE) {
-               LCONSOLE_WARN("Increasing default stripe size to min %u\n",
-                             PTLRPC_MAX_BRW_SIZE);
-               *val = PTLRPC_MAX_BRW_SIZE;
+       if (*val < LOV_MIN_STRIPE_SIZE) {
+               if (*val != 0)
+                       LCONSOLE_INFO("Increasing default stripe size to "
+                                     "minimum value %u\n",
+                                     LOV_DEFAULT_STRIPE_SIZE);
+               *val = LOV_DEFAULT_STRIPE_SIZE;
        } else if (*val & (LOV_MIN_STRIPE_SIZE - 1)) {
                *val &= ~(LOV_MIN_STRIPE_SIZE - 1);
                LCONSOLE_WARN("Changing default stripe size to "LPU64" (a "
@@ -985,6 +1098,9 @@ int lod_pools_fini(struct lod_device *lod)
                pool = cfs_list_entry(pos, struct pool_desc, pool_list);
                /* free pool structs */
                CDEBUG(D_INFO, "delete pool %p\n", pool);
+               /* In the function below, .hs_keycmp resolves to
+                * pool_hashkey_keycmp() */
+               /* coverity[overrun-buffer-val] */
                lod_pool_del(obd, pool->pool_name);
        }