Whamcloud - gitweb
LU-847 quota: client retrieve quota usage directly
[fs/lustre-release.git] / lustre / llite / dir.c
index 7466dd8..dcd4331 100644 (file)
@@ -179,15 +179,7 @@ static int ll_dir_readpage(struct file *file, struct page *page0)
 
                 hash = fd->fd_dir.lfd_next;
         } else {
-                struct ll_inode_info *lli = ll_i2info(inode);
-
-                cfs_spin_lock(&lli->lli_sa_lock);
-                if (lli->lli_sai)
-                        LASSERT(lli->lli_sai->sai_pid == cfs_curproc_pid());
-                else
-                        LASSERT(lli->lli_opendir_pid == cfs_curproc_pid());
-                hash = lli->lli_sa_pos;
-                cfs_spin_unlock(&lli->lli_sa_lock);
+                hash = ll_i2info(inode)->lli_sa_pos;
         }
         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) hash "LPU64"\n",
                inode->i_ino, inode->i_generation, inode, hash);
@@ -287,19 +279,14 @@ static void ll_check_page(struct inode *dir, struct page *page)
         SetPageChecked(page);
 }
 
-static void ll_release_page(struct page *page, __u64 hash,
-                            __u64 start, __u64 end)
+void ll_release_page(struct page *page, int remove)
 {
         kunmap(page);
-        lock_page(page);
-        if (likely(page->mapping != NULL)) {
-                ll_truncate_complete_page(page);
-                unlock_page(page);
-        } else {
+        if (remove) {
+                lock_page(page);
+                if (likely(page->mapping != NULL))
+                        truncate_complete_page(page->mapping, page);
                 unlock_page(page);
-                CWARN("NULL mapping page %p, truncated by others: "
-                      "hash("LPX64") | start("LPX64") | end("LPX64")\n",
-                      page, hash, start, end);
         }
         page_cache_release(page);
 }
@@ -352,19 +339,18 @@ static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
                                  LPX64",hash = "LPX64"\n", *start, *end, *hash);
                         CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash "LPU64"\n",
                                offset, *start, *end, *hash);
-                        if (*hash > *end || (*end != *start && *hash == *end)) {
+                        if (*hash > *end) {
+                                ll_release_page(page, 0);
+                                page = NULL;
+                        } else if (*end != *start && *hash == *end) {
                                 /*
                                  * upon hash collision, remove this page,
                                  * otherwise put page reference, and
                                  * ll_get_dir_page() will issue RPC to fetch
                                  * the page we want.
                                  */
-                                if (dp->ldp_flags & cpu_to_le32(LDF_COLLIDE)) {
-                                        ll_release_page(page, *hash, *start, *end);
-                                } else {
-                                        cfs_kunmap(page);
-                                        page_cache_release(page);
-                                }
+                                ll_release_page(page,
+                                    le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
                                 page = NULL;
                         }
                 } else {
@@ -380,7 +366,7 @@ static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
 }
 
 struct page *ll_get_dir_page(struct file *filp, struct inode *dir, __u64 hash,
-                             int exact, struct ll_dir_chain *chain)
+                             struct ll_dir_chain *chain)
 {
         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
         struct address_space *mapping = dir->i_mapping;
@@ -438,9 +424,7 @@ struct page *ll_get_dir_page(struct file *filp, struct inode *dir, __u64 hash,
                 CERROR("dir page locate: "DFID" at "LPU64": rc %ld\n",
                        PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
                 GOTO(out_unlock, page);
-        }
-
-        if (page != NULL) {
+        } else if (page != NULL) {
                 /*
                  * XXX nikita: not entirely correct handling of a corner case:
                  * suppose hash chain of entries with hash value HASH crosses
@@ -455,20 +439,7 @@ struct page *ll_get_dir_page(struct file *filp, struct inode *dir, __u64 hash,
                  * it as an "overflow" page. 1. invalidate all pages at
                  * once. 2. use HASH|1 as an index for P1.
                  */
-                if (exact && lhash != start) {
-                        /*
-                         * readdir asked for a page starting _exactly_ from
-                         * given hash, but cache contains stale page, with
-                         * entries with smaller hash values. Stale page should
-                         * be invalidated, and new one fetched.
-                         */
-                        CDEBUG(D_OTHER, "Stale readpage page %p: "
-                               "start = "LPX64",end = "LPX64"hash ="LPX64"\n",
-                               page, start, end, lhash);
-                        ll_release_page(page, lhash, start, end);
-                } else {
-                        GOTO(hash_collision, page);
-                }
+                GOTO(hash_collision, page);
         }
 
         page = read_cache_page(mapping, hash_x_index(hash, hash64),
@@ -525,7 +496,7 @@ out_unlock:
         return page;
 
 fail:
-        ll_put_page(page);
+        ll_release_page(page, 1);
         page = ERR_PTR(-EIO);
         goto out_unlock;
 }
@@ -542,7 +513,6 @@ int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
         struct page          *page;
         struct ll_dir_chain   chain;
         int                   done;
-        int                   shift;
         int                   rc;
         ENTRY;
 
@@ -558,11 +528,10 @@ int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 
         rc    = 0;
         done  = 0;
-        shift = 0;
         ll_dir_chain_init(&chain);
 
         fd->fd_dir.lfd_next = pos;
-        page = ll_get_dir_page(filp, inode, pos, 0, &chain);
+        page = ll_get_dir_page(filp, inode, pos, &chain);
 
         while (rc == 0 && !done) {
                 struct lu_dirpage *dp;
@@ -619,7 +588,6 @@ int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
                                                lhash, ino, type);
                         }
                         next = le64_to_cpu(dp->ldp_hash_end);
-                        ll_put_page(page);
                         if (!done) {
                                 pos = next;
                                 if (pos == MDS_DIR_END_OFF) {
@@ -627,21 +595,29 @@ int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
                                          * End of directory reached.
                                          */
                                         done = 1;
+                                        ll_release_page(page, 0);
                                 } else if (1 /* chain is exhausted*/) {
                                         /*
                                          * Normal case: continue to the next
                                          * page.
                                          */
+                                        ll_release_page(page,
+                                            le32_to_cpu(dp->ldp_flags) &
+                                                        LDF_COLLIDE);
                                         fd->fd_dir.lfd_next = pos;
                                         page = ll_get_dir_page(filp, inode, pos,
-                                                               1, &chain);
+                                                               &chain);
                                 } else {
                                         /*
                                          * go into overflow page.
                                          */
+                                        LASSERT(le32_to_cpu(dp->ldp_flags) &
+                                                LDF_COLLIDE);
+                                        ll_release_page(page, 1);
                                 }
                         } else {
                                 pos = hash;
+                                ll_release_page(page, 0);
                         }
                 } else {
                         rc = PTR_ERR(page);
@@ -738,7 +714,7 @@ int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
                         lum_size = sizeof(struct lov_user_md_v3);
                         break;
                         }
-               default: {
+                default: {
                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
                                         " %#08x != %#08x nor %#08x\n",
                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
@@ -820,8 +796,8 @@ int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
                                      0, lmmsize, LUSTRE_OPC_ANY,
                                      NULL);
-        if (op_data == NULL)
-                RETURN(-ENOMEM);
+        if (IS_ERR(op_data))
+                RETURN(PTR_ERR(op_data));
 
         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
@@ -885,8 +861,8 @@ int ll_get_mdt_idx(struct inode *inode)
 
         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
                                      0, LUSTRE_OPC_ANY, NULL);
-        if (op_data == NULL)
-                RETURN(-ENOMEM);
+        if (IS_ERR(op_data))
+                RETURN(PTR_ERR(op_data));
 
         op_data->op_valid |= OBD_MD_MDTIDX;
         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
@@ -916,6 +892,142 @@ static int copy_and_ioctl(int cmd, struct obd_export *exp, void *data, int len)
         return rc;
 }
 
+static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
+{
+        int cmd = qctl->qc_cmd;
+        int type = qctl->qc_type;
+        int id = qctl->qc_id;
+        int valid = qctl->qc_valid;
+        int rc = 0;
+        ENTRY;
+
+        switch (cmd) {
+        case LUSTRE_Q_INVALIDATE:
+        case LUSTRE_Q_FINVALIDATE:
+        case Q_QUOTAON:
+        case Q_QUOTAOFF:
+        case Q_SETQUOTA:
+        case Q_SETINFO:
+                if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
+                    sbi->ll_flags & LL_SBI_RMT_CLIENT)
+                        RETURN(-EPERM);
+                break;
+        case Q_GETQUOTA:
+                if (((type == USRQUOTA && cfs_curproc_euid() != id) ||
+                     (type == GRPQUOTA && !in_egroup_p(id))) &&
+                    (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
+                     sbi->ll_flags & LL_SBI_RMT_CLIENT))
+                        RETURN(-EPERM);
+                break;
+        case Q_GETINFO:
+                break;
+        default:
+                CERROR("unsupported quotactl op: %#x\n", cmd);
+                RETURN(-ENOTTY);
+        }
+
+        if (valid != QC_GENERAL) {
+                if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
+                        RETURN(-EOPNOTSUPP);
+
+                if (cmd == Q_GETINFO)
+                        qctl->qc_cmd = Q_GETOINFO;
+                else if (cmd == Q_GETQUOTA)
+                        qctl->qc_cmd = Q_GETOQUOTA;
+                else
+                        RETURN(-EINVAL);
+
+                switch (valid) {
+                case QC_MDTIDX:
+                        rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
+                                           sizeof(*qctl), qctl, NULL);
+                        break;
+                case QC_OSTIDX:
+                        rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
+                                           sizeof(*qctl), qctl, NULL);
+                        break;
+                case QC_UUID:
+                        rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
+                                           sizeof(*qctl), qctl, NULL);
+                        if (rc == -EAGAIN)
+                                rc = obd_iocontrol(OBD_IOC_QUOTACTL,
+                                                   sbi->ll_dt_exp,
+                                                   sizeof(*qctl), qctl, NULL);
+                        break;
+                default:
+                        rc = -EINVAL;
+                        break;
+                }
+
+                if (rc)
+                        RETURN(rc);
+
+                qctl->qc_cmd = cmd;
+        } else {
+                struct obd_quotactl *oqctl;
+
+                OBD_ALLOC_PTR(oqctl);
+                if (oqctl == NULL)
+                        RETURN(-ENOMEM);
+
+                QCTL_COPY(oqctl, qctl);
+                rc = obd_quotactl(sbi->ll_md_exp, oqctl);
+                if (rc) {
+                        if (rc != -EALREADY && cmd == Q_QUOTAON) {
+                                oqctl->qc_cmd = Q_QUOTAOFF;
+                                obd_quotactl(sbi->ll_md_exp, oqctl);
+                        }
+                        OBD_FREE_PTR(oqctl);
+                        RETURN(rc);
+                }
+                /* If QIF_SPACE is not set, client should collect the
+                 * space usage from OSSs by itself */
+                if (cmd == Q_GETQUOTA &&
+                    !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
+                    !oqctl->qc_dqblk.dqb_curspace) {
+                        struct obd_quotactl *oqctl_tmp;
+
+                        OBD_ALLOC_PTR(oqctl_tmp);
+                        if (oqctl_tmp == NULL)
+                                GOTO(out, rc = -ENOMEM);
+
+                        oqctl_tmp->qc_cmd = Q_GETOQUOTA;
+                        oqctl_tmp->qc_id = oqctl->qc_id;
+                        oqctl_tmp->qc_type = oqctl->qc_type;
+
+                        /* collect space usage from OSTs */
+                        oqctl_tmp->qc_dqblk.dqb_curspace = 0;
+                        rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
+                        if (!rc || rc == -EREMOTEIO) {
+                                oqctl->qc_dqblk.dqb_curspace =
+                                        oqctl_tmp->qc_dqblk.dqb_curspace;
+                                oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
+                        }
+
+                        /* collect space & inode usage from MDTs */
+                        oqctl_tmp->qc_dqblk.dqb_curspace = 0;
+                        oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
+                        rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
+                        if (!rc || rc == -EREMOTEIO) {
+                                oqctl->qc_dqblk.dqb_curspace +=
+                                        oqctl_tmp->qc_dqblk.dqb_curspace;
+                                oqctl->qc_dqblk.dqb_curinodes =
+                                        oqctl_tmp->qc_dqblk.dqb_curinodes;
+                                oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
+                        } else {
+                                oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
+                        }
+
+                        OBD_FREE_PTR(oqctl_tmp);
+                }
+out:
+                QCTL_COPY(qctl, oqctl);
+                OBD_FREE_PTR(oqctl);
+        }
+
+        RETURN(rc);
+}
+
 static int ll_dir_ioctl(struct inode *inode, struct file *file,
                         unsigned int cmd, unsigned long arg)
 {
@@ -979,8 +1091,8 @@ static int ll_dir_ioctl(struct inode *inode, struct file *file,
 
                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
                                              0, LUSTRE_OPC_ANY, NULL);
-                if (op_data == NULL)
-                        GOTO(out_free, rc = -ENOMEM);
+                if (IS_ERR(op_data))
+                        GOTO(out_free, rc = PTR_ERR(op_data));
 
                 op_data->op_valid = OBD_MD_FLID;
                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
@@ -1287,9 +1399,62 @@ out_free:
                 OBD_FREE_PTR(check);
                 RETURN(rc);
         }
-        case OBD_IOC_QUOTACTL: {
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0)
+        case LL_IOC_QUOTACTL_18: {
+                /* copy the old 1.x quota struct for internal use, then copy
+                 * back into old format struct.  For 1.8 compatibility. */
+                struct if_quotactl_18 *qctl_18;
+                struct if_quotactl *qctl_20;
+
+                OBD_ALLOC_PTR(qctl_18);
+                if (!qctl_18)
+                        RETURN(-ENOMEM);
+
+                OBD_ALLOC_PTR(qctl_20);
+                if (!qctl_20)
+                        GOTO(out_quotactl_18, rc = -ENOMEM);
+
+                if (cfs_copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
+                        GOTO(out_quotactl_20, rc = -ENOMEM);
+
+                QCTL_COPY(qctl_20, qctl_18);
+                qctl_20->qc_idx = 0;
+
+                /* XXX: dqb_valid was borrowed as a flag to mark that
+                 *      only mds quota is wanted */
+                if (qctl_18->qc_cmd == Q_GETQUOTA &&
+                    qctl_18->qc_dqblk.dqb_valid) {
+                        qctl_20->qc_valid = QC_MDTIDX;
+                        qctl_20->qc_dqblk.dqb_valid = 0;
+                } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
+                        qctl_20->qc_valid = QC_UUID;
+                        qctl_20->obd_uuid = qctl_18->obd_uuid;
+                } else {
+                        qctl_20->qc_valid = QC_GENERAL;
+                }
+
+                rc = quotactl_ioctl(sbi, qctl_20);
+
+                if (rc == 0) {
+                        QCTL_COPY(qctl_18, qctl_20);
+                        qctl_18->obd_uuid = qctl_20->obd_uuid;
+
+                        if (cfs_copy_to_user((void *)arg, qctl_18,
+                                             sizeof(*qctl_18)))
+                                rc = -EFAULT;
+                }
+
+        out_quotactl_20:
+                OBD_FREE_PTR(qctl_20);
+        out_quotactl_18:
+                OBD_FREE_PTR(qctl_18);
+                RETURN(rc);
+        }
+#else
+#warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
+#endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0) */
+        case LL_IOC_QUOTACTL: {
                 struct if_quotactl *qctl;
-                int cmd, type, id, valid;
 
                 OBD_ALLOC_PTR(qctl);
                 if (!qctl)
@@ -1298,100 +1463,9 @@ out_free:
                 if (cfs_copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
                         GOTO(out_quotactl, rc = -EFAULT);
 
-                cmd = qctl->qc_cmd;
-                type = qctl->qc_type;
-                id = qctl->qc_id;
-                valid = qctl->qc_valid;
-
-                switch (cmd) {
-                case LUSTRE_Q_INVALIDATE:
-                case LUSTRE_Q_FINVALIDATE:
-                case Q_QUOTAON:
-                case Q_QUOTAOFF:
-                case Q_SETQUOTA:
-                case Q_SETINFO:
-                        if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
-                            sbi->ll_flags & LL_SBI_RMT_CLIENT)
-                                GOTO(out_quotactl, rc = -EPERM);
-                        break;
-                case Q_GETQUOTA:
-                        if (((type == USRQUOTA && cfs_curproc_euid() != id) ||
-                             (type == GRPQUOTA && !in_egroup_p(id))) &&
-                            (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
-                             sbi->ll_flags & LL_SBI_RMT_CLIENT))
-                                GOTO(out_quotactl, rc = -EPERM);
-                        break;
-                case Q_GETINFO:
-                        break;
-                default:
-                        CERROR("unsupported quotactl op: %#x\n", cmd);
-                        GOTO(out_quotactl, rc = -ENOTTY);
-                }
-
-                if (valid != QC_GENERAL) {
-                        if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
-                                GOTO(out_quotactl, rc = -EOPNOTSUPP);
+                rc = quotactl_ioctl(sbi, qctl);
 
-                        if (cmd == Q_GETINFO)
-                                qctl->qc_cmd = Q_GETOINFO;
-                        else if (cmd == Q_GETQUOTA)
-                                qctl->qc_cmd = Q_GETOQUOTA;
-                        else
-                                GOTO(out_quotactl, rc = -EINVAL);
-
-                        switch (valid) {
-                        case QC_MDTIDX:
-                                rc = obd_iocontrol(OBD_IOC_QUOTACTL,
-                                                   sbi->ll_md_exp,
-                                                   sizeof(*qctl), qctl, NULL);
-                                break;
-                        case QC_OSTIDX:
-                                rc = obd_iocontrol(OBD_IOC_QUOTACTL,
-                                                   sbi->ll_dt_exp,
-                                                   sizeof(*qctl), qctl, NULL);
-                                break;
-                        case QC_UUID:
-                                rc = obd_iocontrol(OBD_IOC_QUOTACTL,
-                                                   sbi->ll_md_exp,
-                                                   sizeof(*qctl), qctl, NULL);
-                                if (rc == -EAGAIN)
-                                        rc = obd_iocontrol(OBD_IOC_QUOTACTL,
-                                                           sbi->ll_dt_exp,
-                                                           sizeof(*qctl), qctl,
-                                                           NULL);
-                                break;
-                        default:
-                                rc = -EINVAL;
-                                break;
-                        }
-
-                        if (rc)
-                                GOTO(out_quotactl, rc);
-                        else
-                                qctl->qc_cmd = cmd;
-                } else {
-                        struct obd_quotactl *oqctl;
-
-                        OBD_ALLOC_PTR(oqctl);
-                        if (!oqctl)
-                                GOTO(out_quotactl, rc = -ENOMEM);
-
-                        QCTL_COPY(oqctl, qctl);
-                        rc = obd_quotactl(sbi->ll_md_exp, oqctl);
-                        if (rc) {
-                                if (rc != -EALREADY && cmd == Q_QUOTAON) {
-                                        oqctl->qc_cmd = Q_QUOTAOFF;
-                                        obd_quotactl(sbi->ll_md_exp, oqctl);
-                                }
-                                OBD_FREE_PTR(oqctl);
-                                GOTO(out_quotactl, rc);
-                        } else {
-                                QCTL_COPY(qctl, oqctl);
-                                OBD_FREE_PTR(oqctl);
-                        }
-                }
-
-                if (cfs_copy_to_user((void *)arg, qctl, sizeof(*qctl)))
+                if (rc == 0 && cfs_copy_to_user((void *)arg,qctl,sizeof(*qctl)))
                         rc = -EFAULT;
 
         out_quotactl:
@@ -1425,19 +1499,20 @@ out_free:
         }
 #endif
         case LL_IOC_GETOBDCOUNT: {
-                int count;
+                int count, vallen;
+                struct obd_export *exp;
 
                 if (cfs_copy_from_user(&count, (int *)arg, sizeof(int)))
                         RETURN(-EFAULT);
 
-                if (!count) {
-                        /* get ost count */
-                        struct lov_obd *lov = &sbi->ll_dt_exp->exp_obd->u.lov;
-                        count = lov->desc.ld_tgt_count;
-                } else {
-                        /* get mdt count */
-                        struct lmv_obd *lmv = &sbi->ll_md_exp->exp_obd->u.lmv;
-                        count = lmv->desc.ld_tgt_count;
+                /* get ost count when count is zero, get mdt count otherwise */
+                exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
+                vallen = sizeof(count);
+                rc = obd_get_info(exp, sizeof(KEY_TGT_COUNT), KEY_TGT_COUNT,
+                                  &vallen, &count, NULL);
+                if (rc) {
+                        CERROR("get target count failed: %d\n", rc);
+                        RETURN(rc);
                 }
 
                 if (cfs_copy_to_user((int *)arg, &count, sizeof(int)))
@@ -1539,5 +1614,6 @@ struct file_operations ll_dir_operations = {
         .release  = ll_dir_release,
         .read     = generic_read_dir,
         .readdir  = ll_readdir,
-        .ioctl    = ll_dir_ioctl
+        .ioctl    = ll_dir_ioctl,
+        .fsync    = ll_fsync
 };