Whamcloud - gitweb
LU-3569 ofd: packing ost_idx in IDIF
[fs/lustre-release.git] / lustre / lod / lod_lov.c
index 14a2e79..e60ba8b 100644 (file)
@@ -31,9 +31,6 @@
  * Author: Alex Zhuravlev <alexey.zhuravlev@intel.com> 
  */
 
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
 #define DEBUG_SUBSYSTEM S_MDS
 
 #include <obd_class.h>
@@ -250,7 +247,7 @@ int lod_add_device(const struct lu_env *env, struct lod_device *lod,
                spin_lock(&imp->imp_lock);
                imp->imp_server_timeout = 1;
                spin_unlock(&imp->imp_lock);
-               imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
+               imp->imp_client->cli_request_portal = OUT_PORTAL;
                CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
                      obd->obd_name);
                ltd = &lod->lod_mdt_descs;
@@ -504,9 +501,8 @@ int lod_generate_and_set_lovea(const struct lu_env *env,
        ENTRY;
 
        LASSERT(lo);
-       LASSERT(lo->ldo_stripenr > 0);
 
-       magic = lo->ldo_pool ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
+       magic = lo->ldo_pool != NULL ? LOV_MAGIC_V3 : LOV_MAGIC_V1;
        lmm_size = lov_mds_md_size(lo->ldo_stripenr, magic);
        if (info->lti_ea_store_size < lmm_size) {
                rc = lod_ea_store_resize(info, lmm_size);
@@ -514,14 +510,19 @@ int lod_generate_and_set_lovea(const struct lu_env *env,
                        RETURN(rc);
        }
 
+       if (lo->ldo_pattern == 0) /* default striping */
+               lo->ldo_pattern = LOV_PATTERN_RAID0;
+
        lmm = info->lti_ea_store;
 
        lmm->lmm_magic = cpu_to_le32(magic);
-       lmm->lmm_pattern = cpu_to_le32(LOV_PATTERN_RAID0);
+       lmm->lmm_pattern = cpu_to_le32(lo->ldo_pattern);
        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);
+       if (lo->ldo_pattern & LOV_PATTERN_F_RELEASED)
+               lmm->lmm_stripe_count = cpu_to_le16(lo->ldo_released_stripenr);
        lmm->lmm_layout_gen = 0;
        if (magic == LOV_MAGIC_V1) {
                objs = &lmm->lmm_objects[0];
@@ -550,6 +551,7 @@ int lod_generate_and_set_lovea(const struct lu_env *env,
                objs[i].l_ost_gen    = cpu_to_le32(0);
                rc = lod_fld_lookup(env, lod, fid, &index, LU_SEQ_RANGE_OST);
                if (rc < 0) {
+                       lod_object_free_striping(env, lo);
                        CERROR("%s: Can not locate "DFID": rc = %d\n",
                               lod2obd(lod)->obd_name, PFID(fid), rc);
                        RETURN(rc);
@@ -662,6 +664,30 @@ int lod_store_def_striping(const struct lu_env *env, struct dt_object *dt,
        RETURN(rc);
 }
 
+static int validate_lod_and_idx(struct lod_device *md, int idx)
+{
+       if (unlikely(idx >= md->lod_ost_descs.ltd_tgts_size ||
+                    !cfs_bitmap_check(md->lod_ost_bitmap, idx))) {
+               CERROR("%s: bad idx: %d of %d\n", lod2obd(md)->obd_name, idx,
+                      md->lod_ost_descs.ltd_tgts_size);
+               return -EINVAL;
+       }
+
+       if (unlikely(OST_TGT(md, idx) == NULL)) {
+               CERROR("%s: bad lod_tgt_desc for idx: %d\n",
+                      lod2obd(md)->obd_name, idx);
+               return -EINVAL;
+       }
+
+       if (unlikely(OST_TGT(md, idx)->ltd_ost == NULL)) {
+               CERROR("%s: invalid lod device, for idx: %d\n",
+                      lod2obd(md)->obd_name , idx);
+               return -EINVAL;
+       }
+
+       return 0;
+}
+
 /*
  * allocate array of objects pointers, find/create objects
  * stripenr and other fields should be initialized by this moment
@@ -670,7 +696,7 @@ int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
                           struct lov_ost_data_v1 *objs)
 {
        struct lod_thread_info  *info = lod_env_info(env);
-       struct lod_device       *md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
+       struct lod_device       *md;
        struct lu_object        *o, *n;
        struct lu_device        *nd;
        struct dt_object       **stripe;
@@ -678,7 +704,8 @@ int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
        int                      i, idx, rc = 0;
        ENTRY;
 
-       LASSERT(lo);
+       LASSERT(lo != NULL);
+       md = lu2lod_dev(lo->ldo_obj.do_lu.lo_dev);
        LASSERT(lo->ldo_stripe == NULL);
        LASSERT(lo->ldo_stripenr > 0);
        LASSERT(lo->ldo_stripe_size > 0);
@@ -690,22 +717,20 @@ int lod_initialize_objects(const struct lu_env *env, struct lod_object *lo,
 
        for (i = 0; i < lo->ldo_stripenr; i++) {
                ostid_le_to_cpu(&objs[i].l_ost_oi, &info->lti_ostid);
-               idx = le64_to_cpu(objs[i].l_ost_idx);
+               idx = le32_to_cpu(objs[i].l_ost_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));
-               /*
-                * XXX: assertion is left for testing, to make
-                * sure we never process requests till configuration
-                * is completed. to be changed to -EINVAL
-                */
-
                lod_getref(&md->lod_ost_descs);
-               LASSERT(cfs_bitmap_check(md->lod_ost_bitmap, idx));
-               LASSERT(OST_TGT(md,idx));
-               LASSERTF(OST_TGT(md,idx)->ltd_ost, "idx %d\n", idx);
+
+               rc = validate_lod_and_idx(md, idx);
+               if (unlikely(rc != 0)) {
+                       lod_putref(md, &md->lod_ost_descs);
+                       GOTO(out, rc);
+               }
+
                nd = &OST_TGT(md,idx)->ltd_ost->dd_lu_dev;
                lod_putref(md, &md->lod_ost_descs);
 
@@ -746,6 +771,7 @@ int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
        struct lov_mds_md_v1    *lmm;
        struct lov_ost_data_v1  *objs;
        __u32                    magic;
+       __u32                    pattern;
        int                      rc = 0;
        ENTRY;
 
@@ -755,15 +781,20 @@ int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
 
        lmm = (struct lov_mds_md_v1 *) buf->lb_buf;
        magic = le32_to_cpu(lmm->lmm_magic);
+       pattern = le32_to_cpu(lmm->lmm_pattern);
 
        if (magic != LOV_MAGIC_V1 && magic != LOV_MAGIC_V3)
                GOTO(out, rc = -EINVAL);
-       if (le32_to_cpu(lmm->lmm_pattern) != LOV_PATTERN_RAID0)
+       if (lov_pattern(pattern) != LOV_PATTERN_RAID0)
                GOTO(out, rc = -EINVAL);
 
+       lo->ldo_pattern = pattern;
        lo->ldo_stripe_size = le32_to_cpu(lmm->lmm_stripe_size);
-       lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
        lo->ldo_layout_gen = le16_to_cpu(lmm->lmm_layout_gen);
+       lo->ldo_stripenr = le16_to_cpu(lmm->lmm_stripe_count);
+       /* released file stripenr fixup. */
+       if (pattern & LOV_PATTERN_F_RELEASED)
+               lo->ldo_stripenr = 0;
 
        LASSERT(buf->lb_len >= lov_mds_md_size(lo->ldo_stripenr, magic));
 
@@ -775,7 +806,8 @@ int lod_parse_striping(const struct lu_env *env, struct lod_object *lo,
                objs = &lmm->lmm_objects[0];
        }
 
-       rc = lod_initialize_objects(env, lo, objs);
+       if (lo->ldo_stripenr > 0)
+               rc = lod_initialize_objects(env, lo, objs);
 
 out:
        RETURN(rc);
@@ -798,13 +830,8 @@ int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
         */
        dt_write_lock(env, next, 0);
        /* already initialized? */
-       if (lo->ldo_stripe) {
-               int i;
-               /* check validity */
-               for (i = 0; i < lo->ldo_stripenr; i++)
-                       LASSERTF(lo->ldo_stripe[i], "stripe %d is NULL\n", i);
+       if (lo->ldo_stripe != NULL)
                GOTO(out, rc = 0);
-       }
 
        if (!dt_object_exists(next))
                GOTO(out, rc = 0);
@@ -813,8 +840,6 @@ int lod_load_striping(const struct lu_env *env, struct lod_object *lo)
        if (!(lu_object_attr(lod2lu_obj(lo)) & S_IFREG))
                GOTO(out, rc = 0);
 
-       LASSERT(lo->ldo_stripenr == 0);
-
        rc = lod_get_lov_ea(env, lo);
        if (rc <= 0)
                GOTO(out, rc);
@@ -835,87 +860,121 @@ 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);
        }
 
-       if ((specific && lum->lmm_pattern != LOV_PATTERN_RAID0) ||
-           (specific == 0 && lum->lmm_pattern != 0)) {
+       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 && 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)
-               /* In the function below, .hs_keycmp resolves to
-                * pool_hashkey_keycmp() */
-               /* coverity[overrun-buffer-val] */
-               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)
@@ -1017,11 +1076,6 @@ int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
        lod->lod_qos.lq_prio_free = 232;
        /* Default threshold for rr (roughly 17%) */
        lod->lod_qos.lq_threshold_rr = 43;
-       /* Init statfs fields */
-       OBD_ALLOC_PTR(lod->lod_qos.lq_statfs_data);
-       if (NULL == lod->lod_qos.lq_statfs_data)
-               RETURN(-ENOMEM);
-       cfs_waitq_init(&lod->lod_qos.lq_statfs_waitq);
 
        /* Set up OST pool environment */
        lod->lod_pools_hash_body = cfs_hash_create("POOLS", HASH_POOLS_CUR_BITS,
@@ -1031,8 +1085,9 @@ int lod_pools_init(struct lod_device *lod, struct lustre_cfg *lcfg)
                                                   CFS_HASH_MAX_THETA,
                                                   &pool_hash_operations,
                                                   CFS_HASH_DEFAULT);
-       if (!lod->lod_pools_hash_body)
-               GOTO(out_statfs, rc = -ENOMEM);
+       if (lod->lod_pools_hash_body == NULL)
+               RETURN(-ENOMEM);
+
        CFS_INIT_LIST_HEAD(&lod->lod_pool_list);
        lod->lod_pool_count = 0;
        rc = lod_ost_pool_init(&lod->lod_pool_info, 0);
@@ -1048,8 +1103,7 @@ out_pool_info:
        lod_ost_pool_free(&lod->lod_pool_info);
 out_hash:
        cfs_hash_putref(lod->lod_pools_hash_body);
-out_statfs:
-       OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);
+
        return rc;
 }
 
@@ -1073,7 +1127,7 @@ int lod_pools_fini(struct lod_device *lod)
        cfs_hash_putref(lod->lod_pools_hash_body);
        lod_ost_pool_free(&(lod->lod_qos.lq_rr.lqr_pool));
        lod_ost_pool_free(&lod->lod_pool_info);
-       OBD_FREE_PTR(lod->lod_qos.lq_statfs_data);
+
        RETURN(0);
 }