Whamcloud - gitweb
LU-7991 quota: project quota against ZFS backend
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_quota.c
index 5abfafa..2ff5770 100644 (file)
  * GPL HEADER END
  */
 /*
- * Copyright (c) 2011, 2012, Whamcloud, Inc.
+ * Copyright (c) 2012, 2015, 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"
 
 /**
 static inline int fid2type(const struct lu_fid *fid)
 {
        LASSERT(fid_is_acct(fid));
-       if (fid_oid(fid) == ACCT_GROUP_OID)
+       switch (fid_oid(fid)) {
+       case ACCT_USER_OID:
+               return USRQUOTA;
+       case ACCT_GROUP_OID:
                return GRPQUOTA;
-       return USRQUOTA;
-}
+       case ACCT_PROJECT_OID:
+               return PRJQUOTA;
+       }
 
-static inline int obj2type(struct dt_object *obj)
-{
-       return fid2type(lu_object_fid(&obj->do_lu));
+       LASSERTF(0, "invalid fid for quota type: %u", fid_oid(fid));
+       return USRQUOTA;
 }
 
 /**
@@ -74,7 +77,26 @@ 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)];
+       switch (fid2type(fid)) {
+       case USRQUOTA:
+               id->oii_ino =
+                       le32_to_cpu(LDISKFS_SB(sb)->s_es->s_usr_quota_inum);
+               break;
+       case GRPQUOTA:
+               id->oii_ino =
+                       le32_to_cpu(LDISKFS_SB(sb)->s_es->s_grp_quota_inum);
+               break;
+       case PRJQUOTA:
+ #ifdef HAVE_PROJECT_QUOTA
+               if (LDISKFS_HAS_RO_COMPAT_FEATURE(sb,
+                                       LDISKFS_FEATURE_RO_COMPAT_PROJECT))
+                       id->oii_ino =
+                               le32_to_cpu(LDISKFS_SB(sb)->s_es->s_prj_quota_inum);
+               else
+ #endif
+                       RETURN(-ENOENT);
+               break;
+       }
        if (!ldiskfs_valid_inum(sb, id->oii_ino))
                RETURN(-ENOENT);
        RETURN(0);
@@ -86,9 +108,8 @@ int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
  * \param env   - is the environment passed by the caller
  * \param dtobj - is the accounting object
  * \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
+ * \param dtkey - is the id 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 +117,65 @@ 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);
+#if defined(HAVE_DQUOT_QC_DQBLK)
+       struct qc_dqblk         *dqblk = &info->oti_qdq;
+#elif defined(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 lquota_acct_rec  *rec = (struct lquota_acct_rec *)dtrec;
        __u64                    id = *((__u64 *)dtkey);
        int                      rc;
+#ifdef HAVE_DQUOT_KQID
+       struct kqid              qid;
+#endif
+       int type;
 
        ENTRY;
 
-       memset((void *)dqblk, 0, sizeof(struct obd_dqblk));
-       rc = sb->s_qcop->get_dqblk(sb, obj2type(dtobj), (qid_t) id, dqblk);
+       type = fid2type(lu_object_fid(&dtobj->do_lu));
+       memset(dqblk, 0, sizeof(*dqblk));
+#ifdef HAVE_DQUOT_KQID
+       qid = make_kqid(&init_user_ns, type, id);
+       rc = sb->s_qcop->get_dqblk(sb, qid, dqblk);
+#else
+       rc = sb->s_qcop->get_dqblk(sb, type, (qid_t) id, dqblk);
+#endif
        if (rc)
                RETURN(rc);
+#if defined(HAVE_DQUOT_QC_DQBLK)
+       rec->bspace = dqblk->d_space;
+       rec->ispace = dqblk->d_ino_count;
+#elif defined(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);
@@ -142,14 +184,13 @@ 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;
-       CFS_INIT_LIST_HEAD(&it->oiq_list);
+       INIT_LIST_HEAD(&it->oiq_list);
 
        /* LUSTRE_DQTREEOFF is the initial offset where the tree can be found */
        it->oiq_blk[0] = LUSTRE_DQTREEOFF;
@@ -171,12 +212,15 @@ static void osd_it_acct_fini(const struct lu_env *env, struct dt_it *di)
        struct osd_quota_leaf *leaf, *tmp;
        ENTRY;
 
-       lu_object_put(env, &it->oiq_obj->oo_dt.do_lu);
+       osd_object_put(env, it->oiq_obj);
 
-       cfs_list_for_each_entry_safe(leaf, tmp, &it->oiq_list, oql_link) {
-               cfs_list_del_init(&leaf->oql_link);
+       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;
 }
 
@@ -197,12 +241,13 @@ static int osd_it_acct_get(const struct lu_env *env, struct dt_it *di,
        struct osd_it_quota     *it = (struct osd_it_quota *)di;
        const struct lu_fid     *fid =
                                lu_object_fid(&it->oiq_obj->oo_dt.do_lu);
-       int                      type = fid2type(fid);
+       int                      type;
        qid_t                    dqid = *(qid_t *)key;
        loff_t                   offset;
        int                      rc;
 
        ENTRY;
+       type = fid2type(fid);
 
        offset = find_tree_dqentry(env, it->oiq_obj, type, dqid,
                                   LUSTRE_DQTREEOFF, 0, it);
@@ -241,9 +286,9 @@ static int osd_it_add_processed(struct osd_it_quota *it, int depth)
        OBD_ALLOC_PTR(leaf);
        if (leaf == NULL)
                RETURN(-ENOMEM);
-       CFS_INIT_LIST_HEAD(&leaf->oql_link);
+       INIT_LIST_HEAD(&leaf->oql_link);
        leaf->oql_blk = it->oiq_blk[depth];
-       cfs_list_add_tail(&leaf->oql_link, &it->oiq_list);
+       list_add_tail(&leaf->oql_link, &it->oiq_list);
        RETURN(0);
 }
 
@@ -261,12 +306,14 @@ static int osd_it_acct_next(const struct lu_env *env, struct dt_it *di)
        struct osd_it_quota     *it = (struct osd_it_quota *)di;
        const struct lu_fid     *fid =
                                lu_object_fid(&it->oiq_obj->oo_dt.do_lu);
-       int                      type = fid2type(fid);
+       int                      type;
        int                      depth, rc;
        uint                     index;
 
        ENTRY;
 
+       type = fid2type(fid);
+
        /* 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.
         */
@@ -358,8 +405,7 @@ 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);
 }
 
@@ -415,37 +461,77 @@ const struct dt_index_operations osd_acct_index_ops = {
        }
 };
 
-static inline int osd_qid_type(struct osd_thandle *oh, int i)
+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)
 {
-       return (oh->ot_id_type & (1 << i)) ? GRPQUOTA : USRQUOTA;
+#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 void osd_qid_set_type(struct osd_thandle *oh, int i, int type)
+static inline int osd_qid_type(struct osd_thandle *oh, int i)
 {
-       oh->ot_id_type |= ((type == GRPQUOTA) ? (1 << i) : 0);
+       return oh->ot_id_types[i];
 }
 
 /**
  * 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  allocated - dquot entry in quota accounting file has been allocated
- * \param  flags  - if the operation is write, return no user quota, no
- *                  group quota, or sync commit flags to the caller
+ * \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
+ * \retval 0       - success
+ * \retval -ve     - failure
  */
 int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
-                    struct lquota_id_info *qi, bool allocated, int *flags)
+                   struct lquota_id_info *qi, struct osd_object *obj,
+                   bool enforce, int *flags)
 {
-       struct osd_thread_info  *info = osd_oti_get(env);
-       struct osd_device       *dev = info->oti_dev;
-       struct qsd_instance     *qsd = dev->od_quota_slave;
-       int                      i, rc;
+       struct osd_device       *dev;
+       struct qsd_instance     *qsd;
+       struct inode            *inode = NULL;
+       int                      i, rc = 0, crd;
        bool                     found = false;
        ENTRY;
 
@@ -453,9 +539,14 @@ int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
        LASSERTF(oh->ot_id_cnt <= OSD_MAX_UGID_CNT, "count=%d\n",
                 oh->ot_id_cnt);
 
+       dev = osd_dt_dev(oh->ot_super.th_dev);
+       LASSERT(dev != NULL);
+
+       qsd = dev->od_quota_slave;
+
        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) {
+                   oh->ot_id_types[i] == qi->lqi_type) {
                        found = true;
                        break;
                }
@@ -468,11 +559,35 @@ int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
                        RETURN(-EOVERFLOW);
                }
 
-               oh->ot_credits += (allocated || qi->lqi_id.qid_uid == 0) ?
-                       1 : LDISKFS_QUOTA_INIT_BLOCKS(osd_sb(dev));
+               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 ||
+#ifdef HAVE_EXT4_INFO_DQUOT
+                                LDISKFS_I(inode)->i_dquot[qi->lqi_type] == NULL)
+#else
+                                inode->i_dquot[qi->lqi_type] == NULL)
+#endif
+                               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_types[i] = qi->lqi_type;
                oh->ot_id_cnt++;
        }
 
@@ -481,7 +596,8 @@ int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
                RETURN(0);
 
        /* check quota */
-       rc = qsd_op_begin(env, qsd, oh->ot_quota_trans, qi, flags);
+       if (enforce)
+               rc = qsd_op_begin(env, qsd, oh->ot_quota_trans, qi, flags);
        RETURN(rc);
 }
 
@@ -493,31 +609,32 @@ int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
  * \param  gid    - group id of the inode
  * \param  space  - how many blocks/inodes will be consumed/released
  * \param  oh     - osd transaction handle
- * \param  is_blk - block quota or inode quota?
- * \param  allocated - dquot entry in quota accounting file has been allocated
+ * \param  obj    - osd object, could be NULL when it's under create
  * \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
+ * \param osd_qid_flags - indicate this is a inode/block accounting
+ *                     and whether changes are performed by root user
  *
  * \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,
-                         bool is_blk, bool allocated, int *flags, bool force)
+                         __u32 projid, long long space, struct osd_thandle *oh,
+                         struct osd_object *obj, int *flags,
+                         enum osd_qid_declare_flags osd_qid_declare_flags)
 {
        struct osd_thread_info  *info = osd_oti_get(env);
        struct lquota_id_info   *qi = &info->oti_qi;
-       int                      rcu, rcg; /* user & group rc */
+       int rcu, rcg, rcp = 0; /* user & group & project rc */
+       bool force = !!(osd_qid_declare_flags & OSD_QID_FORCE);
        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, allocated, flags);
+       qi->lqi_is_blk     = !!(osd_qid_declare_flags & OSD_QID_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 */
@@ -533,12 +650,25 @@ int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
        /* and now group quota */
        qi->lqi_id.qid_gid = gid;
        qi->lqi_type       = GRPQUOTA;
-       rcg = osd_declare_qid(env, oh, qi, allocated, flags);
+       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);
-}
+#ifdef HAVE_PROJECT_QUOTA
+       if (rcg && (rcg != -EDQUOT || flags == NULL))
+               RETURN(rcg);
+
+       /* and now project quota */
+       qi->lqi_id.qid_projid = projid;
+       qi->lqi_type = PRJQUOTA;
+       rcp = osd_declare_qid(env, oh, qi, obj, true, flags);
 
+       if (force && (rcp == -EDQUOT || rcp == -EINPROGRESS))
+               /* as before, ignore EDQUOT & EINPROGRESS for root */
+               rcp = 0;
+#endif
+
+       RETURN(rcu ? rcu : (rcg ? rcg : rcp));
+}