Whamcloud - gitweb
New tag 2.15.63
[fs/lustre-release.git] / lustre / osd-ldiskfs / osd_quota.c
index f0fc100..d8f0fb4 100644 (file)
@@ -28,6 +28,7 @@
  * Author: Niu    Yawei    <niu@whamcloud.com>
  */
 
+#include <dt_object.h>
 #include <lustre_quota.h>
 #include "osd_internal.h"
 
@@ -47,7 +48,7 @@ static inline int fid2type(const struct lu_fid *fid)
                return PRJQUOTA;
        }
 
-       LASSERTF(0, "invalid fid for quota type: %u", fid_oid(fid));
+       LASSERTF(0, "invalid fid for quota type: %u\n", fid_oid(fid));
        return USRQUOTA;
 }
 
@@ -75,6 +76,14 @@ int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
        if (!ldiskfs_has_feature_quota(sb))
                RETURN(-ENOENT);
 
+       /**
+        * ldiskfs won't load quota inodes on RO mount,
+        * So disable it in osd-ldiskfs to keep same behavior
+        * like lower layer to avoid further confusions.
+        */
+       if (osd->od_dt_dev.dd_rdonly)
+               RETURN(-ENOENT);
+
        id->oii_gen = OSD_OII_NOGEN;
        switch (fid2type(fid)) {
        case USRQUOTA:
@@ -86,17 +95,23 @@ int osd_acct_obj_lookup(struct osd_thread_info *info, struct osd_device *osd,
                        le32_to_cpu(LDISKFS_SB(sb)->s_es->s_grp_quota_inum);
                break;
        case PRJQUOTA:
- #ifdef HAVE_PROJECT_QUOTA
-               if (ldiskfs_has_feature_project(sb))
-                       id->oii_ino =
-                               le32_to_cpu(LDISKFS_SB(sb)->s_es->s_prj_quota_inum);
-               else
- #endif
+#ifdef HAVE_PROJECT_QUOTA
+               if (ldiskfs_has_feature_project(sb)) {
+                       __le32 prj_quota;
+
+                       prj_quota = LDISKFS_SB(sb)->s_es->s_prj_quota_inum;
+                       id->oii_ino = le32_to_cpu(prj_quota);
+               } else
+#endif
                        RETURN(-ENOENT);
                break;
        }
-       if (!ldiskfs_valid_inum(sb, id->oii_ino))
+
+       if (!ldiskfs_valid_inum(sb, id->oii_ino) &&
+           id->oii_ino != LDISKFS_USR_QUOTA_INO &&
+           id->oii_ino != LDISKFS_GRP_QUOTA_INO)
                RETURN(-ENOENT);
+
        RETURN(0);
 }
 
@@ -117,44 +132,33 @@ static int osd_acct_index_lookup(const struct lu_env *env,
                                 struct dt_rec *dtrec,
                                 const struct dt_key *dtkey)
 {
-       struct osd_thread_info  *info = osd_oti_get(env);
+       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;
+       struct qc_dqblk *dqblk = &info->oti_qdq;
 #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;
+       struct fs_disk_quota *dqblk = &info->oti_fdq;
 #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;
+       struct kqid qid;
        int type;
 
        ENTRY;
 
        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)
+#else
        rec->bspace = dqblk->d_bcount;
        rec->ispace = dqblk->d_icount;
-#else
-       rec->bspace = dqblk->dqb_curspace;
-       rec->ispace = dqblk->dqb_curinodes;
 #endif
        RETURN(+1);
 }
@@ -162,7 +166,7 @@ static int osd_acct_index_lookup(const struct lu_env *env,
 #define QUOTA_IT_READ_ERROR(it, rc)                                    \
        CERROR("%s: Error while trying to read quota information, "    \
               "failed with %d\n",                                     \
-              osd_dev(it->oiq_obj->oo_dt.do_lu.lo_dev)->od_svname, rc); \
+              osd_dev(it->oiq_obj->oo_dt.do_lu.lo_dev)->od_svname, rc) \
 
 /**
  * Initialize osd Iterator for given osd index object.
@@ -174,9 +178,9 @@ static struct dt_it *osd_it_acct_init(const struct lu_env *env,
                                      struct dt_object *dt,
                                      __u32 attr)
 {
-       struct osd_it_quota     *it;
-       struct lu_object        *lo = &dt->do_lu;
-       struct osd_object       *obj = osd_dt_obj(dt);
+       struct osd_it_quota *it;
+       struct lu_object *lo = &dt->do_lu;
+       struct osd_object *obj = osd_dt_obj(dt);
 
        ENTRY;
 
@@ -193,9 +197,11 @@ static struct dt_it *osd_it_acct_init(const struct lu_env *env,
        /* LUSTRE_DQTREEOFF is the initial offset where the tree can be found */
        it->oiq_blk[0] = LUSTRE_DQTREEOFF;
 
-       /* NB: we don't need to store the tree depth since it is always
+       /*
+        * NB: we don't need to store the tree depth since it is always
         * equal to LUSTRE_DQTREEDEPTH - 1 (root has depth = 0) for a leaf
-        * block. */
+        * block.
+        */
        RETURN((struct dt_it *)it);
 }
 
@@ -208,6 +214,7 @@ 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;
 
        osd_object_put(env, it->oiq_obj);
@@ -236,13 +243,12 @@ static void osd_it_acct_fini(const struct lu_env *env, struct dt_it *di)
 static int osd_it_acct_get(const struct lu_env *env, struct dt_it *di,
                           const struct dt_key *key)
 {
-       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;
-       qid_t                    dqid = *(qid_t *)key;
-       loff_t                   offset;
-       int                      rc;
+       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;
+       qid_t dqid = *(qid_t *)key;
+       loff_t offset;
+       int rc;
 
        ENTRY;
        type = fid2type(fid);
@@ -301,18 +307,18 @@ static int osd_it_add_processed(struct osd_it_quota *it, int depth)
  */
 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;
-       int                      depth, rc;
-       uint                     index;
+       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;
+       int depth, rc;
+       uint index;
 
        ENTRY;
 
        type = fid2type(fid);
 
-       /* Let's first check if there are any remaining valid entry in the
+       /*
+        * 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;
@@ -325,8 +331,10 @@ static int osd_it_acct_next(const struct lu_env *env, struct dt_it *di)
                        QUOTA_IT_READ_ERROR(it, rc);
                        RETURN(rc);
                } else if (rc == 0) {
-                       /* Found on entry, @it is already updated to the
-                        * new position in walk_block_dqentry(). */
+                       /*
+                        * 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);
@@ -340,12 +348,16 @@ static int osd_it_acct_next(const struct lu_env *env, struct dt_it *di)
        }
        rc = 1;
 
-       /* We have consumed all the entries of the current leaf block, move on
-        * to the next one. */
+       /*
+        * We have consumed all the entries of the current leaf block, move on
+        * to the next one.
+        */
        depth--;
 
-       /* We keep searching as long as walk_tree_dqentry() returns +1
-        * (= no valid entry found). */
+       /*
+        * We keep searching as long as walk_tree_dqentry() returns +1
+        * (= no valid entry found).
+        */
        for (; depth >= 0 && rc > 0; depth--) {
                index = it->oiq_index[depth];
                if (++index > 0xff)
@@ -397,9 +409,9 @@ static int osd_it_acct_rec(const struct lu_env *env,
                           const struct dt_it *di,
                           struct dt_rec *dtrec, __u32 attr)
 {
-       struct osd_it_quota     *it = (struct osd_it_quota *)di;
-       const struct dt_key     *key = osd_it_acct_key(env, di);
-       int                      rc;
+       struct osd_it_quota *it = (struct osd_it_quota *)di;
+       const struct dt_key *key = osd_it_acct_key(env, di);
+       int rc;
 
        ENTRY;
 
@@ -439,10 +451,11 @@ static int osd_it_acct_load(const struct lu_env *env,
 
        ENTRY;
 
-       /* LU-8999 - If it is called to resume the iteration, calling
+       /*
+        * LU-8999 - If it is called to resume the iteration, calling
         * osd_it_acct_get could change the block orders in the lower level
         * of the quota tree, which are saved in osd_it_quota->oiq_blk.
-        * */
+        */
        if (it->oiq_id != 0 && it->oiq_id == hash)
                RETURN(1);
 
@@ -454,8 +467,8 @@ static int osd_it_acct_load(const struct lu_env *env,
  * Index and Iterator operations for accounting objects
  */
 const struct dt_index_operations osd_acct_index_ops = {
-       .dio_lookup     = osd_acct_index_lookup,
-       .dio_it         = {
+       .dio_lookup = osd_acct_index_lookup,
+       .dio_it = {
                .init           = osd_it_acct_init,
                .fini           = osd_it_acct_fini,
                .get            = osd_it_acct_get,
@@ -476,7 +489,7 @@ static inline void osd_quota_swab(char *ptr, size_t size)
        LASSERT((size & (sizeof(__u64) - 1)) == 0);
 
        for (offset = 0; offset < size; offset += sizeof(__u64))
-            __swab64s((__u64 *)(ptr + offset));
+               __swab64s((__u64 *)(ptr + offset));
 }
 
 const struct dt_rec *osd_quota_pack(struct osd_object *obj,
@@ -507,8 +520,6 @@ void osd_quota_unpack(struct osd_object *obj, const struct dt_rec *rec)
        descr = obj->oo_dir->od_container.ic_descr;
 
        osd_quota_swab((char *)rec, descr->id_rec_size);
-#else
-       return;
 #endif
 }
 
@@ -534,14 +545,25 @@ static inline int osd_qid_type(struct osd_thandle *oh, int i)
  */
 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)
+                   bool enforce, enum osd_quota_local_flags *local_flags)
 {
-       struct osd_device       *dev;
-       struct qsd_instance     *qsd;
-       struct inode            *inode = NULL;
-       int                      i, rc = 0, crd;
-       bool                     found = false;
+       struct osd_device *dev;
+       struct qsd_instance *qsd;
+       struct lu_fid fid = { 0 };
+       struct inode *inode = NULL;
+       unsigned long long ino =  0;
+       int i, rc = 0, crd;
+       __u8 res = qi->lqi_is_blk ? LQUOTA_RES_DT : LQUOTA_RES_MD;
+       bool found = false;
+
        ENTRY;
+       if (obj) {
+               fid = *lu_object_fid(&obj->oo_dt.do_lu);
+               inode = obj->oo_inode;
+               ino = inode ? inode->i_ino : 0;
+       }
+       CDEBUG(D_QUOTA, "fid="DFID" ino=%llu type=%u, id=%llu\n",
+              PFID(&fid), ino, qi->lqi_type, qi->lqi_id.qid_uid);
 
        LASSERT(oh != NULL);
        LASSERTF(oh->ot_id_cnt <= OSD_MAX_UGID_CNT, "count=%d\n",
@@ -550,10 +572,14 @@ int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
        dev = osd_dt_dev(oh->ot_super.th_dev);
        LASSERT(dev != NULL);
 
-       qsd = dev->od_quota_slave;
+       if (res == LQUOTA_RES_DT)
+               qsd = dev->od_quota_slave_dt;
+       else
+               qsd = dev->od_quota_slave_md;
 
        for (i = 0; i < oh->ot_id_cnt; i++) {
                if (oh->ot_id_array[i] == qi->lqi_id.qid_uid &&
+                   oh->ot_id_res[i] == res &&
                    oh->ot_id_types[i] == qi->lqi_type) {
                        found = true;
                        break;
@@ -563,39 +589,34 @@ int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
        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);
+                       rc = -EOVERFLOW;
+                       CERROR("%s: too many qids %u > %u on "DFID": rc = %d\n",
+                              osd_name(dev), i + 1, OSD_MAX_UGID_CNT,
+                              PFID(&fid), rc);
+                       RETURN(rc);
                }
 
-               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) {
+                       /* root ID should be always present in the quota file */
                        crd = 1;
                } else {
-                       /* used space for this ID could be dropped to zero,
+                       /* can't rely on the current state as it can change
+                        * by the execution.
+                        * if used space for this ID could be dropped to zero,
                         * reserve extra credits for removing ID entry from
-                        * the quota file */
+                        * 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;
+                               crd = LDISKFS_QUOTA_INIT_BLOCKS(osd_sb(dev));
                }
 
                osd_trans_declare_op(env, oh, OSD_OT_QUOTA, crd);
 
                oh->ot_id_array[i] = qi->lqi_id.qid_uid;
                oh->ot_id_types[i] = qi->lqi_type;
+               oh->ot_id_res[i] = res;
                oh->ot_id_cnt++;
        }
 
@@ -605,7 +626,8 @@ int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
 
        /* check quota */
        if (enforce)
-               rc = qsd_op_begin(env, qsd, oh->ot_quota_trans, qi, flags);
+               rc = qsd_op_begin(env, qsd, oh->ot_quota_trans, qi,
+                                 local_flags);
        RETURN(rc);
 }
 
@@ -628,56 +650,68 @@ int osd_declare_qid(const struct lu_env *env, struct osd_thandle *oh,
  */
 int osd_declare_inode_qid(const struct lu_env *env, qid_t uid, qid_t gid,
                          __u32 projid, long long space, struct osd_thandle *oh,
-                         struct osd_object *obj, int *flags,
+                         struct osd_object *obj,
+                         enum osd_quota_local_flags *local_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;
+       struct osd_thread_info *info = osd_oti_get(env);
+       struct lquota_id_info *qi = &info->oti_qi;
        int rcu, rcg, rcp = 0; /* user & group & project rc */
        struct thandle *th = &oh->ot_super;
        bool force = !!(osd_qid_declare_flags & OSD_QID_FORCE) ||
                        th->th_ignore_quota;
        ENTRY;
 
+       /* very fast path for special files like llog */
+       if (uid == 0 && gid == 0 && projid == 0)
+               return 0;
+
        /* let's start with user quota */
        qi->lqi_id.qid_uid = uid;
-       qi->lqi_type       = USRQUOTA;
-       qi->lqi_space      = space;
-       qi->lqi_is_blk     = !!(osd_qid_declare_flags & OSD_QID_BLK);
-       rcu = osd_declare_qid(env, oh, qi, obj, true, flags);
+       qi->lqi_type = USRQUOTA;
+       qi->lqi_space = space;
+       qi->lqi_is_blk = !!(osd_qid_declare_flags & OSD_QID_BLK);
+       rcu = osd_declare_qid(env, oh, qi, obj, true, local_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 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))
+        * When force is set to true, we also want to proceed with the gid
+        */
+       if (rcu && (rcu != -EDQUOT || local_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);
+       qi->lqi_type = GRPQUOTA;
+       rcg = osd_declare_qid(env, oh, qi, obj, true, local_flags);
 
        if (force && (rcg == -EDQUOT || rcg == -EINPROGRESS))
                /* as before, ignore EDQUOT & EINPROGRESS for root */
                rcg = 0;
 
 #ifdef HAVE_PROJECT_QUOTA
-       if (rcg && (rcg != -EDQUOT || flags == NULL))
+       if (rcg && (rcg != -EDQUOT || local_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);
+       rcp = osd_declare_qid(env, oh, qi, obj, true, local_flags);
 
-       if (force && (rcp == -EDQUOT || rcp == -EINPROGRESS))
+       if (local_flags && *local_flags & QUOTA_FL_ROOT_PRJQUOTA)
+               force = th->th_ignore_quota;
+       if (force && (rcp == -EDQUOT || rcp == -EINPROGRESS)) {
+               CDEBUG(D_QUOTA, "forced to ignore quota flags = %#x\n",
+                      local_flags ? *local_flags : -1);
                /* as before, ignore EDQUOT & EINPROGRESS for root */
                rcp = 0;
+       }
 #endif
 
        RETURN(rcu ? rcu : (rcg ? rcg : rcp));