Whamcloud - gitweb
LU-6836 tests: Skip sanity-quota test 4a for ZFS
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_quota.c
index e636c6d..368a74a 100644 (file)
  * GPL HEADER END
  */
 /*
- * Copyright (c) 2011, 2012, Whamcloud, Inc.
+ * Copyright (c) 2012, 2014, Intel Corporation.
  * Use is subject to license terms.
  *
  * Author: Johann Lombardi <johann@whamcloud.com>
  * Author: Niu    Yawei    <niu@whamcloud.com>
  */
 
-#include <lquota.h>
+#include <lustre_quota.h>
 #include "osd_internal.h"
 
 /**
@@ -65,6 +65,10 @@ int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
                        const struct lu_fid *fid, struct osd_inode_id *id)
 {
        struct super_block *sb = osd_sb(osd);
+        unsigned long qf_inums[2] = {
+               le32_to_cpu(LDISKFS_SB(sb)->s_es->s_usr_quota_inum),
+               le32_to_cpu(LDISKFS_SB(sb)->s_es->s_grp_quota_inum)
+       };
 
        ENTRY;
        LASSERT(fid_is_acct(fid));
@@ -74,7 +78,7 @@ int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
                RETURN(-ENOENT);
 
        id->oii_gen = OSD_OII_NOGEN;
-       id->oii_ino = LDISKFS_SB(sb)->s_qf_inums[fid2type(fid)];
+       id->oii_ino = qf_inums[fid2type(fid)];
        if (!ldiskfs_valid_inum(sb, id->oii_ino))
                RETURN(-ENOENT);
        RETURN(0);
@@ -88,7 +92,6 @@ int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
  * \param dtrec - is the record to fill with space usage information
  * \param dtkey - is the id the of the user or group for which we would
  *                like to access disk usage.
- * \param capa - is the capability, not used.
  *
  * \retval +ve - success : exact match
  * \retval -ve - failure
@@ -96,44 +99,58 @@ int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
 static int osd_acct_index_lookup(const struct lu_env *env,
                                 struct dt_object *dtobj,
                                 struct dt_rec *dtrec,
-                                const struct dt_key *dtkey,
-                                struct lustre_capa *capa)
+                                const struct dt_key *dtkey)
 {
        struct osd_thread_info  *info = osd_oti_get(env);
+#ifdef HAVE_DQUOT_FS_DISK_QUOTA
+       struct fs_disk_quota    *dqblk = &info->oti_fdq;
+#else
        struct if_dqblk         *dqblk = &info->oti_dqblk;
+#endif
        struct super_block      *sb = osd_sb(osd_obj2dev(osd_dt_obj(dtobj)));
-       struct acct_rec         *rec = (struct acct_rec *)dtrec;
+       struct lquota_acct_rec  *rec = (struct lquota_acct_rec *)dtrec;
        __u64                    id = *((__u64 *)dtkey);
        int                      rc;
+#ifdef HAVE_DQUOT_KQID
+       struct kqid             qid;
+#endif
 
        ENTRY;
 
        memset((void *)dqblk, 0, sizeof(struct obd_dqblk));
+#ifdef HAVE_DQUOT_KQID
+       qid = make_kqid(&init_user_ns, obj2type(dtobj), id);
+       rc = sb->s_qcop->get_dqblk(sb, qid, dqblk);
+#else
        rc = sb->s_qcop->get_dqblk(sb, obj2type(dtobj), (qid_t) id, dqblk);
+#endif
        if (rc)
                RETURN(rc);
+#ifdef HAVE_DQUOT_FS_DISK_QUOTA
+       rec->bspace = dqblk->d_bcount;
+       rec->ispace = dqblk->d_icount;
+#else
        rec->bspace = dqblk->dqb_curspace;
        rec->ispace = dqblk->dqb_curinodes;
+#endif
        RETURN(+1);
 }
 
 #define QUOTA_IT_READ_ERROR(it, rc)                                    \
        CERROR("%s: Error while trying to read quota information, "    \
               "failed with %d\n",                                     \
-              it->oiq_obj->oo_dt.do_lu.lo_dev->ld_obd->obd_name, rc); \
+              osd_dev(it->oiq_obj->oo_dt.do_lu.lo_dev)->od_svname, rc); \
 
 /**
  * Initialize osd Iterator for given osd index object.
  *
  * \param  dt    - osd index object
  * \param  attr  - not used
- * \param  capa  - BYPASS_CAPA
  */
 static struct dt_it *osd_it_acct_init(const struct lu_env *env,
                                      struct dt_object *dt,
-                                     __u32 attr, struct lustre_capa *capa)
+                                     __u32 attr)
 {
-       struct osd_thread_info  *info = osd_oti_get(env);
        struct osd_it_quota     *it;
        struct lu_object        *lo = &dt->do_lu;
        struct osd_object       *obj = osd_dt_obj(dt);
@@ -142,13 +159,13 @@ static struct dt_it *osd_it_acct_init(const struct lu_env *env,
 
        LASSERT(lu_object_exists(lo));
 
-       if (info == NULL)
+       OBD_ALLOC_PTR(it);
+       if (it == NULL)
                RETURN(ERR_PTR(-ENOMEM));
 
-       it = &info->oti_it_quota;
-       memset(it, 0, sizeof(*it));
        lu_object_get(lo);
        it->oiq_obj = obj;
+       INIT_LIST_HEAD(&it->oiq_list);
 
        /* LUSTRE_DQTREEOFF is the initial offset where the tree can be found */
        it->oiq_blk[0] = LUSTRE_DQTREEOFF;
@@ -167,9 +184,18 @@ static struct dt_it *osd_it_acct_init(const struct lu_env *env,
 static void osd_it_acct_fini(const struct lu_env *env, struct dt_it *di)
 {
        struct osd_it_quota *it = (struct osd_it_quota *)di;
-
+       struct osd_quota_leaf *leaf, *tmp;
        ENTRY;
+
        lu_object_put(env, &it->oiq_obj->oo_dt.do_lu);
+
+       list_for_each_entry_safe(leaf, tmp, &it->oiq_list, oql_link) {
+               list_del_init(&leaf->oql_link);
+               OBD_FREE_PTR(leaf);
+       }
+
+       OBD_FREE_PTR(it);
+
        EXIT;
 }
 
@@ -227,6 +253,19 @@ static void osd_it_acct_put(const struct lu_env *env, struct dt_it *di)
        return;
 }
 
+static int osd_it_add_processed(struct osd_it_quota *it, int depth)
+{
+       struct osd_quota_leaf *leaf;
+
+       OBD_ALLOC_PTR(leaf);
+       if (leaf == NULL)
+               RETURN(-ENOMEM);
+       INIT_LIST_HEAD(&leaf->oql_link);
+       leaf->oql_blk = it->oiq_blk[depth];
+       list_add_tail(&leaf->oql_link, &it->oiq_list);
+       RETURN(0);
+}
+
 /**
  * Move on to the next valid entry.
  *
@@ -250,8 +289,8 @@ static int osd_it_acct_next(const struct lu_env *env, struct dt_it *di)
        /* Let's first check if there are any remaining valid entry in the
         * current leaf block. Start with the next entry after the current one.
         */
-       depth = LUSTRE_DQTREEDEPTH - 1;
-       index = GETIDINDEX(it->oiq_id, depth);
+       depth = LUSTRE_DQTREEDEPTH;
+       index = it->oiq_index[depth];
        if (++index < LUSTRE_DQSTRINBLK) {
                /* Search for the next valid entry from current index */
                rc = walk_block_dqentry(env, it->oiq_obj, type,
@@ -263,7 +302,15 @@ static int osd_it_acct_next(const struct lu_env *env, struct dt_it *di)
                        /* Found on entry, @it is already updated to the
                         * new position in walk_block_dqentry(). */
                        RETURN(0);
+               } else {
+                       rc = osd_it_add_processed(it, depth);
+                       if (rc)
+                               RETURN(rc);
                }
+       } else {
+               rc = osd_it_add_processed(it, depth);
+               if (rc)
+                       RETURN(rc);
        }
        rc = 1;
 
@@ -274,7 +321,7 @@ static int osd_it_acct_next(const struct lu_env *env, struct dt_it *di)
        /* We keep searching as long as walk_tree_dqentry() returns +1
         * (= no valid entry found). */
        for (; depth >= 0 && rc > 0; depth--) {
-               index = GETIDINDEX(it->oiq_id, depth);
+               index = it->oiq_index[depth];
                if (++index > 0xff)
                        continue;
                rc = walk_tree_dqentry(env, it->oiq_obj, type,
@@ -330,8 +377,7 @@ static int osd_it_acct_rec(const struct lu_env *env,
 
        ENTRY;
 
-       rc = osd_acct_index_lookup(env, &it->oiq_obj->oo_dt, dtrec, key,
-                                  BYPASS_CAPA);
+       rc = osd_acct_index_lookup(env, &it->oiq_obj->oo_dt, dtrec, key);
        RETURN(rc > 0 ? 0 : rc);
 }
 
@@ -387,3 +433,287 @@ const struct dt_index_operations osd_acct_index_ops = {
        }
 };
 
+static inline void osd_quota_swab(char *ptr, size_t size)
+{
+       int offset;
+
+       LASSERT((size & (sizeof(__u64) - 1)) == 0);
+
+       for (offset = 0; offset < size; offset += sizeof(__u64))
+            __swab64s((__u64 *)(ptr + offset));
+}
+
+const struct dt_rec *osd_quota_pack(struct osd_object *obj,
+                                   const struct dt_rec *rec,
+                                   union lquota_rec *quota_rec)
+{
+#ifdef __BIG_ENDIAN
+       struct iam_descr        *descr;
+
+       LASSERT(obj->oo_dir != NULL);
+       descr = obj->oo_dir->od_container.ic_descr;
+
+       memcpy(quota_rec, rec, descr->id_rec_size);
+
+       osd_quota_swab((char *)quota_rec, descr->id_rec_size);
+       return (const struct dt_rec *)quota_rec;
+#else
+       return rec;
+#endif
+}
+
+void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec)
+{
+#ifdef __BIG_ENDIAN
+       struct iam_descr *descr;
+
+       LASSERT(obj->oo_dir != NULL);
+       descr = obj->oo_dir->od_container.ic_descr;
+
+       osd_quota_swab((char *)rec, descr->id_rec_size);
+#else
+       return;
+#endif
+}
+
+static inline int osd_qid_type(struct osd_thandle *oh, int i)
+{
+       return (oh->ot_id_type & (1 << i)) ? GRPQUOTA : USRQUOTA;
+}
+
+static inline void osd_qid_set_type(struct osd_thandle *oh, int i, int type)
+{
+       oh->ot_id_type |= ((type == GRPQUOTA) ? (1 << i) : 0);
+}
+
+/**
+ * Reserve journal credits for quota files update first, then call
+ * ->op_begin() to perform quota enforcement.
+ *
+ * \param  env     - the environment passed by the caller
+ * \param  oh      - osd transaction handle
+ * \param  qi      - quota id & space required for this operation
+ * \param  obj     - osd object, could be NULL when it's under create
+ * \param  enforce - whether to perform quota enforcement
+ * \param  flags   - if the operation is write, return no user quota, no
+ *                   group quota, or sync commit flags to the caller
+ *
+ * \retval 0       - success
+ * \retval -ve     - failure
+ */
+int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
+                   struct lquota_id_info *qi, struct osd_object *obj,
+                   bool enforce, int *flags)
+{
+       struct osd_device       *dev = osd_dt_dev(oh->ot_super.th_dev);
+       struct qsd_instance     *qsd = dev->od_quota_slave;
+       struct inode            *inode = NULL;
+       int                      i, rc = 0, crd;
+       bool                     found = false;
+       ENTRY;
+
+       LASSERT(oh != NULL);
+       LASSERTF(oh->ot_id_cnt <= OSD_MAX_UGID_CNT, "count=%d\n",
+                oh->ot_id_cnt);
+
+       for (i = 0; i < oh->ot_id_cnt; i++) {
+               if (oh->ot_id_array[i] == qi->lqi_id.qid_uid &&
+                   osd_qid_type(oh, i) == qi->lqi_type) {
+                       found = true;
+                       break;
+               }
+       }
+
+       if (!found) {
+               /* we need to account for credits for this new ID */
+               if (i >= OSD_MAX_UGID_CNT) {
+                       CERROR("Too many(%d) trans qids!\n", i + 1);
+                       RETURN(-EOVERFLOW);
+               }
+
+               if (obj != NULL)
+                       inode = obj->oo_inode;
+
+               /* root ID entry should be always present in the quota file */
+               if (qi->lqi_id.qid_uid == 0) {
+                       crd = 1;
+               } else {
+                       /* used space for this ID could be dropped to zero,
+                        * reserve extra credits for removing ID entry from
+                        * the quota file */
+                       if (qi->lqi_space < 0)
+                               crd = LDISKFS_QUOTA_DEL_BLOCKS(osd_sb(dev));
+                       /* reserve credits for adding ID entry to the quota
+                        * file if the i_dquot isn't initialized yet. */
+                       else if (inode == NULL ||
+                                inode->i_dquot[qi->lqi_type] == NULL)
+                               crd = LDISKFS_QUOTA_INIT_BLOCKS(osd_sb(dev));
+                       else
+                               crd = 1;
+               }
+
+               osd_trans_declare_op(env, oh, OSD_OT_QUOTA, crd);
+
+               oh->ot_id_array[i] = qi->lqi_id.qid_uid;
+               osd_qid_set_type(oh, i, qi->lqi_type);
+               oh->ot_id_cnt++;
+       }
+
+       if (unlikely(qsd == NULL))
+               /* quota slave instance hasn't been allocated yet */
+               RETURN(0);
+
+       /* check quota */
+       if (enforce)
+               rc = qsd_op_begin(env, qsd, oh->ot_quota_trans, qi, flags);
+       RETURN(rc);
+}
+
+/**
+ * Wrapper for osd_declare_qid()
+ *
+ * \param  env    - the environment passed by the caller
+ * \param  uid    - user id of the inode
+ * \param  gid    - group id of the inode
+ * \param  space  - how many blocks/inodes will be consumed/released
+ * \param  oh     - osd transaction handle
+ * \param  obj    - osd object, could be NULL when it's under create
+ * \param  is_blk - block quota or inode quota?
+ * \param  flags  - if the operation is write, return no user quota, no
+ *                  group quota, or sync commit flags to the caller
+ * \param force   - set to 1 when changes are performed by root user and thus
+ *                  can't failed with EDQUOT
+ *
+ * \retval 0      - success
+ * \retval -ve    - failure
+ */
+int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
+                         long long space, struct osd_thandle *oh,
+                         struct osd_object *obj, bool is_blk, int *flags,
+                         bool force)
+{
+       struct osd_thread_info  *info = osd_oti_get(env);
+       struct lquota_id_info   *qi = &info->oti_qi;
+       int                      rcu, rcg; /* user & group rc */
+       ENTRY;
+
+       /* let's start with user quota */
+       qi->lqi_id.qid_uid = uid;
+       qi->lqi_type       = USRQUOTA;
+       qi->lqi_space      = space;
+       qi->lqi_is_blk     = is_blk;
+       rcu = osd_declare_qid(env, oh, qi, obj, true, flags);
+
+       if (force && (rcu == -EDQUOT || rcu == -EINPROGRESS))
+               /* ignore EDQUOT & EINPROGRESS when changes are done by root */
+               rcu = 0;
+
+       /* For non-fatal error, we want to continue to get the noquota flags
+        * for group id. This is only for commit write, which has @flags passed
+        * in. See osd_declare_write_commit().
+        * When force is set to true, we also want to proceed with the gid */
+       if (rcu && (rcu != -EDQUOT || flags == NULL))
+               RETURN(rcu);
+
+       /* and now group quota */
+       qi->lqi_id.qid_gid = gid;
+       qi->lqi_type       = GRPQUOTA;
+       rcg = osd_declare_qid(env, oh, qi, obj, true, flags);
+
+       if (force && (rcg == -EDQUOT || rcg == -EINPROGRESS))
+               /* as before, ignore EDQUOT & EINPROGRESS for root */
+               rcg = 0;
+
+       RETURN(rcu ? rcu : rcg);
+}
+
+int osd_quota_migration(const struct lu_env *env, struct dt_object *dt)
+{
+       struct osd_thread_info  *oti = osd_oti_get(env);
+       struct osd_device       *osd = osd_obj2dev(osd_dt_obj(dt));
+       struct dt_object        *root, *parent = NULL, *admin = NULL;
+       dt_obj_version_t         version;
+       char                    *fname, *fnames[] = {ADMIN_USR, ADMIN_GRP};
+       int                      rc, i;
+       ENTRY;
+
+       /* not newly created global index */
+       version = dt_version_get(env, dt);
+       if (version != 0)
+               RETURN(0);
+
+       /* locate root */
+       rc = dt_root_get(env, &osd->od_dt_dev, &oti->oti_fid);
+       if (rc) {
+               CERROR("%s: Can't get root FID, rc:%d\n", osd->od_svname, rc);
+               RETURN(rc);
+       }
+
+       root = dt_locate(env, &osd->od_dt_dev, &oti->oti_fid);
+       if (IS_ERR(root)) {
+               CERROR("%s: Failed to locate root "DFID", rc:%ld\n",
+                      osd->od_svname, PFID(&oti->oti_fid), PTR_ERR(root));
+               RETURN(PTR_ERR(root));
+       }
+
+       /* locate /OBJECTS */
+       rc = dt_lookup_dir(env, root, OBJECTS, &oti->oti_fid);
+       if (rc == -ENOENT) {
+               GOTO(out, rc = 0);
+       } else if (rc) {
+               CERROR("%s: Failed to lookup %s, rc:%d\n",
+                      osd->od_svname, OBJECTS, rc);
+               GOTO(out, rc);
+       }
+
+       parent = dt_locate(env, &osd->od_dt_dev, &oti->oti_fid);
+       if (IS_ERR(parent)) {
+               CERROR("%s: Failed to locate %s "DFID", rc:%ld\n",
+                      osd->od_svname, OBJECTS, PFID(&oti->oti_fid),
+                      PTR_ERR(parent));
+               GOTO(out, rc = PTR_ERR(parent));
+       }
+
+       /* locate quota admin files */
+       for (i = 0; i < 2; i++) {
+               fname = fnames[i];
+               rc = dt_lookup_dir(env, parent, fname, &oti->oti_fid);
+               if (rc == -ENOENT) {
+                       rc = 0;
+                       continue;
+               } else if (rc) {
+                       CERROR("%s: Failed to lookup %s, rc:%d\n",
+                              osd->od_svname, fname, rc);
+                       GOTO(out, rc);
+               }
+
+               admin = dt_locate(env, &osd->od_dt_dev, &oti->oti_fid);
+               if (IS_ERR(admin)) {
+                       CERROR("%s: Failed to locate %s "DFID", rc:%d\n",
+                              osd->od_svname, fname, PFID(&oti->oti_fid), rc);
+                       GOTO(out, rc = PTR_ERR(admin));
+               }
+
+               if (!dt_object_exists(admin)) {
+                       CERROR("%s: Old admin file %s doesn't exist, but is "
+                              "still referenced in parent directory.\n",
+                              osd->od_svname, fname);
+                       lu_object_put(env, &admin->do_lu);
+                       GOTO(out, rc = -ENOENT);
+               }
+
+               LCONSOLE_WARN("%s: Detected old quota admin file(%s)! If you "
+                             "want to keep the old quota limits settings, "
+                             "please upgrade to lower version(2.5) first to "
+                             "convert them into new format.\n",
+                             osd->od_svname, fname);
+
+               lu_object_put(env, &admin->do_lu);
+               GOTO(out, rc = -EINVAL);
+       }
+out:
+       if (parent && !IS_ERR(parent))
+               lu_object_put(env, &parent->do_lu);
+       lu_object_put(env, &root->do_lu);
+       RETURN(rc);
+}