Whamcloud - gitweb
LU-9679 modules: convert MIN/MAX to kernel style
[fs/lustre-release.git] / lustre / mdc / mdc_locks.c
index 66694f5..574d09b 100644 (file)
@@ -197,11 +197,12 @@ static inline void mdc_clear_replay_flag(struct ptlrpc_request *req, int rc)
                spin_lock(&req->rq_lock);
                req->rq_replay = 0;
                spin_unlock(&req->rq_lock);
-        }
-        if (rc && req->rq_transno != 0) {
-                DEBUG_REQ(D_ERROR, req, "transno returned on error rc %d", rc);
-                LBUG();
-        }
+       }
+       if (rc && req->rq_transno != 0) {
+               DEBUG_REQ(D_ERROR, req, "transno returned on error: rc = %d",
+                         rc);
+               LBUG();
+       }
 }
 
 /* Save a large LOV EA into the request buffer so that it is available
@@ -220,7 +221,7 @@ int mdc_save_lovea(struct ptlrpc_request *req,
                   void *data, u32 size)
 {
        struct req_capsule *pill = &req->rq_pill;
-       void *lmm;
+       struct lov_user_md *lmm;
        int rc = 0;
 
        if (req_capsule_get_size(pill, field, RCL_CLIENT) < size) {
@@ -237,8 +238,12 @@ int mdc_save_lovea(struct ptlrpc_request *req,
 
        req_capsule_set_size(pill, field, RCL_CLIENT, size);
        lmm = req_capsule_client_get(pill, field);
-       if (lmm)
+       if (lmm) {
                memcpy(lmm, data, size);
+               /* overwrite layout generation returned from the MDS */
+               lmm->lmm_stripe_offset =
+                 (typeof(lmm->lmm_stripe_offset))LOV_OFFSET_DEFAULT;
+       }
 
        return rc;
 }
@@ -252,7 +257,8 @@ mdc_intent_open_pack(struct obd_export *exp, struct lookup_intent *it,
        struct ldlm_intent      *lit;
        const void              *lmm = op_data->op_data;
        __u32                    lmmsize = op_data->op_data_size;
-       struct list_head         cancels = LIST_HEAD_INIT(cancels);
+       __u32                    mdt_md_capsule_size;
+       LIST_HEAD(cancels);
        int                      count = 0;
        enum ldlm_mode           mode;
        int                      rc;
@@ -260,6 +266,8 @@ mdc_intent_open_pack(struct obd_export *exp, struct lookup_intent *it,
 
        ENTRY;
 
+       mdt_md_capsule_size = obddev->u.cli.cl_default_mds_easize;
+
        it->it_create_mode = (it->it_create_mode & ~S_IFMT) | S_IFREG;
 
        /* XXX: openlock is not cancelled for cross-refs. */
@@ -314,11 +322,21 @@ mdc_intent_open_pack(struct obd_export *exp, struct lookup_intent *it,
 
        req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX_NAME,
                             RCL_CLIENT, op_data->op_file_secctx_name != NULL ?
-                            strlen(op_data->op_file_secctx_name) + 1 : 0);
+                            op_data->op_file_secctx_name_size : 0);
 
        req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX, RCL_CLIENT,
                             op_data->op_file_secctx_size);
 
+       /* get SELinux policy info if any */
+       rc = sptlrpc_get_sepol(req);
+       if (rc < 0) {
+               ptlrpc_request_free(req);
+               RETURN(ERR_PTR(rc));
+       }
+       req_capsule_set_size(&req->rq_pill, &RMF_SELINUX_POL, RCL_CLIENT,
+                            strlen(req->rq_sepol) ?
+                            strlen(req->rq_sepol) + 1 : 0);
+
        rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
        if (rc < 0) {
                ptlrpc_request_free(req);
@@ -338,9 +356,33 @@ mdc_intent_open_pack(struct obd_export *exp, struct lookup_intent *it,
                       lmmsize);
 
        req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
-                            obddev->u.cli.cl_max_mds_easize);
+                            mdt_md_capsule_size);
        req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
 
+       if (!(it->it_op & IT_CREAT) && it->it_op & IT_OPEN &&
+           req_capsule_has_field(&req->rq_pill, &RMF_FILE_SECCTX_NAME,
+                                 RCL_CLIENT) &&
+           op_data->op_file_secctx_name_size > 0 &&
+           op_data->op_file_secctx_name != NULL) {
+               char *secctx_name;
+
+               secctx_name = req_capsule_client_get(&req->rq_pill,
+                                                    &RMF_FILE_SECCTX_NAME);
+               memcpy(secctx_name, op_data->op_file_secctx_name,
+                      op_data->op_file_secctx_name_size);
+               req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX,
+                                    RCL_SERVER,
+                                    obddev->u.cli.cl_max_mds_easize);
+
+               CDEBUG(D_SEC, "packed '%.*s' as security xattr name\n",
+                      op_data->op_file_secctx_name_size,
+                      op_data->op_file_secctx_name);
+
+       } else {
+               req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX,
+                                    RCL_SERVER, 0);
+       }
+
        /**
         * Inline buffer for possible data from Data-on-MDT files.
         */
@@ -353,7 +395,7 @@ mdc_intent_open_pack(struct obd_export *exp, struct lookup_intent *it,
                                      lustre_msg_early_size());
        /* Estimate free space for DoM files in repbuf */
        repsize_estimate = repsize - (req->rq_replen -
-                          obddev->u.cli.cl_max_mds_easize +
+                          mdt_md_capsule_size +
                           sizeof(struct lov_comp_md_v1) +
                           sizeof(struct lov_comp_md_entry_v1) +
                           lov_mds_md_size(0, LOV_MAGIC_V3));
@@ -394,7 +436,7 @@ mdc_intent_getxattr_pack(struct obd_export *exp,
        struct ptlrpc_request   *req;
        struct ldlm_intent      *lit;
        int                     rc, count = 0;
-       struct list_head        cancels = LIST_HEAD_INIT(cancels);
+       LIST_HEAD(cancels);
        u32 ea_vals_buf_size = GA_DEFAULT_EA_VAL_LEN * GA_DEFAULT_EA_NUM;
 
        ENTRY;
@@ -404,6 +446,16 @@ mdc_intent_getxattr_pack(struct obd_export *exp,
        if (req == NULL)
                RETURN(ERR_PTR(-ENOMEM));
 
+       /* get SELinux policy info if any */
+       rc = sptlrpc_get_sepol(req);
+       if (rc < 0) {
+               ptlrpc_request_free(req);
+               RETURN(ERR_PTR(rc));
+       }
+       req_capsule_set_size(&req->rq_pill, &RMF_SELINUX_POL, RCL_CLIENT,
+                            strlen(req->rq_sepol) ?
+                            strlen(req->rq_sepol) + 1 : 0);
+
        rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
        if (rc) {
                ptlrpc_request_free(req);
@@ -413,6 +465,8 @@ mdc_intent_getxattr_pack(struct obd_export *exp,
        /* pack the intent */
        lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
        lit->opc = IT_GETXATTR;
+       CDEBUG(D_INFO, "%s: get xattrs for "DFID"\n",
+              exp->exp_obd->obd_name, PFID(&op_data->op_fid1));
 
 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
        /* If the supplied buffer is too small then the server will
@@ -432,6 +486,9 @@ mdc_intent_getxattr_pack(struct obd_export *exp,
        mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
                      ea_vals_buf_size, -1, 0);
 
+       /* get SELinux policy info if any */
+       mdc_file_sepol_pack(req);
+
        req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_SERVER,
                             GA_DEFAULT_EA_NAME_LEN * GA_DEFAULT_EA_NUM);
 
@@ -452,44 +509,77 @@ static struct ptlrpc_request *
 mdc_intent_getattr_pack(struct obd_export *exp, struct lookup_intent *it,
                        struct md_op_data *op_data, __u32 acl_bufsize)
 {
-       struct ptlrpc_request   *req;
-       struct obd_device       *obddev = class_exp2obd(exp);
-       u64                      valid = OBD_MD_FLGETATTR | OBD_MD_FLEASIZE |
-                                        OBD_MD_FLMODEASIZE | OBD_MD_FLDIREA |
-                                        OBD_MD_MEA | OBD_MD_FLACL;
-       struct ldlm_intent      *lit;
-       int                      rc;
-       __u32                    easize;
+       struct ptlrpc_request *req;
+       struct obd_device *obddev = class_exp2obd(exp);
+       u64 valid = OBD_MD_FLGETATTR | OBD_MD_FLEASIZE | OBD_MD_FLMODEASIZE |
+                   OBD_MD_FLDIREA | OBD_MD_MEA | OBD_MD_FLACL |
+                   OBD_MD_DEFAULT_MEA;
+       struct ldlm_intent *lit;
+       __u32 easize;
+       bool have_secctx = false;
+       int rc;
+
        ENTRY;
 
-        req = ptlrpc_request_alloc(class_exp2cliimp(exp),
-                                   &RQF_LDLM_INTENT_GETATTR);
-        if (req == NULL)
-                RETURN(ERR_PTR(-ENOMEM));
+       req = ptlrpc_request_alloc(class_exp2cliimp(exp),
+                                  &RQF_LDLM_INTENT_GETATTR);
+       if (req == NULL)
+               RETURN(ERR_PTR(-ENOMEM));
 
-        req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
-                             op_data->op_namelen + 1);
+       /* send name of security xattr to get upon intent */
+       if (it->it_op & (IT_LOOKUP | IT_GETATTR) &&
+           req_capsule_has_field(&req->rq_pill, &RMF_FILE_SECCTX_NAME,
+                                 RCL_CLIENT) &&
+           op_data->op_file_secctx_name_size > 0 &&
+           op_data->op_file_secctx_name != NULL) {
+               have_secctx = true;
+               req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX_NAME,
+                                    RCL_CLIENT,
+                                    op_data->op_file_secctx_name_size);
+       }
 
-        rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
-        if (rc) {
-                ptlrpc_request_free(req);
-                RETURN(ERR_PTR(rc));
-        }
+       req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
+                            op_data->op_namelen + 1);
+
+       rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
+       if (rc) {
+               ptlrpc_request_free(req);
+               RETURN(ERR_PTR(rc));
+       }
 
         /* pack the intent */
-        lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
-        lit->opc = (__u64)it->it_op;
+       lit = req_capsule_client_get(&req->rq_pill, &RMF_LDLM_INTENT);
+       lit->opc = (__u64)it->it_op;
 
-       if (obddev->u.cli.cl_default_mds_easize > 0)
-               easize = obddev->u.cli.cl_default_mds_easize;
-       else
-               easize = obddev->u.cli.cl_max_mds_easize;
+       easize = obddev->u.cli.cl_default_mds_easize;
 
        /* pack the intended request */
        mdc_getattr_pack(req, valid, it->it_flags, op_data, easize);
 
        req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER, easize);
        req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
+       req_capsule_set_size(&req->rq_pill, &RMF_DEFAULT_MDT_MD, RCL_SERVER,
+                            sizeof(struct lmv_user_md));
+
+       if (have_secctx) {
+               char *secctx_name;
+
+               secctx_name = req_capsule_client_get(&req->rq_pill,
+                                                    &RMF_FILE_SECCTX_NAME);
+               memcpy(secctx_name, op_data->op_file_secctx_name,
+                      op_data->op_file_secctx_name_size);
+
+               req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX,
+                                    RCL_SERVER, easize);
+
+               CDEBUG(D_SEC, "packed '%.*s' as security xattr name\n",
+                      op_data->op_file_secctx_name_size,
+                      op_data->op_file_secctx_name);
+       } else {
+               req_capsule_set_size(&req->rq_pill, &RMF_FILE_SECCTX,
+                                    RCL_SERVER, 0);
+       }
+
        ptlrpc_request_set_replen(req);
        RETURN(req);
 }
@@ -499,10 +589,11 @@ static struct ptlrpc_request *mdc_intent_layout_pack(struct obd_export *exp,
                                                     struct md_op_data *op_data)
 {
        struct obd_device     *obd = class_exp2obd(exp);
+       LIST_HEAD(cancels);
        struct ptlrpc_request *req;
        struct ldlm_intent    *lit;
        struct layout_intent  *layout;
-       int rc;
+       int count = 0, rc;
        ENTRY;
 
        req = ptlrpc_request_alloc(class_exp2cliimp(exp),
@@ -510,8 +601,15 @@ static struct ptlrpc_request *mdc_intent_layout_pack(struct obd_export *exp,
        if (req == NULL)
                RETURN(ERR_PTR(-ENOMEM));
 
+       if (fid_is_sane(&op_data->op_fid2) && (it->it_op & IT_LAYOUT) &&
+           (it->it_flags & FMODE_WRITE)) {
+               count = mdc_resource_get_unused(exp, &op_data->op_fid2,
+                                               &cancels, LCK_EX,
+                                               MDS_INODELOCK_LAYOUT);
+       }
+
        req_capsule_set_size(&req->rq_pill, &RMF_EADATA, RCL_CLIENT, 0);
-       rc = ldlm_prep_enqueue_req(exp, req, NULL, 0);
+       rc = ldlm_prep_enqueue_req(exp, req, &cancels, count);
        if (rc) {
                ptlrpc_request_free(req);
                RETURN(ERR_PTR(rc));
@@ -623,7 +721,7 @@ static int mdc_finish_enqueue(struct obd_export *exp,
            (!it_disposition(it, DISP_OPEN_OPEN) || it->it_status != 0))
                mdc_clear_replay_flag(req, it->it_status);
 
-       DEBUG_REQ(D_RPCTRACE, req, "op: %x disposition: %x, status: %d",
+       DEBUG_REQ(D_RPCTRACE, req, "op=%x disposition=%x, status=%d",
                  it->it_op, it->it_disposition, it->it_status);
 
        /* We know what to expect, so we do any byte flipping required here */
@@ -645,6 +743,12 @@ static int mdc_finish_enqueue(struct obd_export *exp,
                        mdc_set_open_replay_data(NULL, NULL, it);
                }
 
+               if (it_disposition(it, DISP_OPEN_CREATE) &&
+                   !it_open_error(DISP_OPEN_CREATE, it)) {
+                       lprocfs_counter_incr(exp->exp_obd->obd_md_stats,
+                                            LPROC_MD_CREATE);
+               }
+
                if (body->mbo_valid & (OBD_MD_FLDIREA | OBD_MD_FLEASIZE)) {
                         void *eadata;
 
@@ -746,7 +850,7 @@ static int mdc_finish_enqueue(struct obd_export *exp,
 
                body = req_capsule_server_get(pill, &RMF_MDT_BODY);
                if (!(body->mbo_valid & OBD_MD_DOM_SIZE)) {
-                       LDLM_ERROR(lock, "%s: DoM lock without size.\n",
+                       LDLM_ERROR(lock, "%s: DoM lock without size.",
                                   exp->exp_obd->obd_name);
                        GOTO(out_lock, rc = -EPROTO);
                }
@@ -762,6 +866,16 @@ out_lock:
        RETURN(rc);
 }
 
+static inline bool mdc_skip_mod_rpc_slot(const struct lookup_intent *it)
+{
+       if (it != NULL &&
+           (it->it_op == IT_GETATTR || it->it_op == IT_LOOKUP ||
+            it->it_op == IT_READDIR ||
+            (it->it_op == IT_LAYOUT && !(it->it_flags & MDS_FMODE_WRITE))))
+               return true;
+       return false;
+}
+
 /* We always reserve enough space in the reply packet for a stripe MD, because
  * we don't know in advance the file type. */
 static int mdc_enqueue_base(struct obd_export *exp,
@@ -773,7 +887,7 @@ static int mdc_enqueue_base(struct obd_export *exp,
                            __u64 extra_lock_flags)
 {
        struct obd_device *obddev = class_exp2obd(exp);
-       struct ptlrpc_request *req = NULL;
+       struct ptlrpc_request *req;
        __u64 flags, saved_flags = extra_lock_flags;
        struct ldlm_res_id res_id;
        static const union ldlm_policy_data lookup_policy = {
@@ -812,8 +926,8 @@ static int mdc_enqueue_base(struct obd_export *exp,
 
        generation = obddev->u.cli.cl_import->imp_generation;
        if (!it || (it->it_op & (IT_OPEN | IT_CREAT)))
-               acl_bufsize = MIN(imp->imp_connect_data.ocd_max_easize,
-                                 XATTR_SIZE_MAX);
+               acl_bufsize = min_t(__u32, imp->imp_connect_data.ocd_max_easize,
+                                   XATTR_SIZE_MAX);
        else
                acl_bufsize = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
 
@@ -824,6 +938,7 @@ resend:
                LASSERTF(einfo->ei_type == LDLM_FLOCK, "lock type %d\n",
                         einfo->ei_type);
                res_id.name[3] = LDLM_FLOCK;
+               req = ldlm_enqueue_pack(exp, 0);
        } else if (it->it_op & IT_OPEN) {
                req = mdc_intent_open_pack(exp, it, op_data, acl_bufsize);
        } else if (it->it_op & (IT_GETATTR | IT_LOOKUP)) {
@@ -851,20 +966,7 @@ resend:
                req->rq_sent = ktime_get_real_seconds() + resends;
         }
 
-       /* It is important to obtain modify RPC slot first (if applicable), so
-        * that threads that are waiting for a modify RPC slot are not polluting
-        * our rpcs in flight counter.
-        * We do not do flock request limiting, though */
-       if (it) {
-               mdc_get_mod_rpc_slot(req, it);
-               rc = obd_get_request_slot(&obddev->u.cli);
-               if (rc != 0) {
-                       mdc_put_mod_rpc_slot(req, it);
-                       mdc_clear_replay_flag(req, 0);
-                       ptlrpc_req_finished(req);
-                       RETURN(rc);
-               }
-       }
+       einfo->ei_enq_slot = !mdc_skip_mod_rpc_slot(it);
 
        /* With Data-on-MDT the glimpse callback is needed too.
         * It is set here in advance but not in mdc_finish_enqueue()
@@ -875,6 +977,7 @@ resend:
 
        rc = ldlm_cli_enqueue(exp, &req, einfo, &res_id, policy, &flags, NULL,
                              0, lvb_type, lockh, 0);
+
        if (!it) {
                /* For flock requests we immediatelly return without further
                   delay and let caller deal with the rest, since rest of
@@ -888,12 +991,10 @@ resend:
                    (einfo->ei_type == LDLM_FLOCK) &&
                    (einfo->ei_mode == LCK_NL))
                        goto resend;
+               ptlrpc_req_finished(req);
                RETURN(rc);
        }
 
-       obd_put_request_slot(&obddev->u.cli);
-       mdc_put_mod_rpc_slot(req, it);
-
        if (rc < 0) {
                CDEBUG(D_INFO,
                      "%s: ldlm_cli_enqueue "DFID":"DFID"=%s failed: rc = %d\n",
@@ -938,8 +1039,8 @@ resend:
            acl_bufsize == LUSTRE_POSIX_ACL_MAX_SIZE_OLD) {
                mdc_clear_replay_flag(req, -ERANGE);
                ptlrpc_req_finished(req);
-               acl_bufsize = MIN(imp->imp_connect_data.ocd_max_easize,
-                                 XATTR_SIZE_MAX);
+               acl_bufsize = min_t(__u32, imp->imp_connect_data.ocd_max_easize,
+                                   XATTR_SIZE_MAX);
                goto resend;
        }
 
@@ -1237,19 +1338,15 @@ static int mdc_intent_getattr_async_interpret(const struct lu_env *env,
        struct obd_export *exp = ga->ga_exp;
        struct md_enqueue_info *minfo = ga->ga_minfo;
        struct ldlm_enqueue_info *einfo = &minfo->mi_einfo;
-       struct lookup_intent *it;
-       struct lustre_handle *lockh;
-       struct obd_device *obddev;
-       struct ldlm_reply *lockrep;
-       __u64 flags = LDLM_FL_HAS_INTENT;
+       struct lookup_intent     *it;
+       struct lustre_handle     *lockh;
+       struct ldlm_reply        *lockrep;
+       __u64                     flags = LDLM_FL_HAS_INTENT;
        ENTRY;
 
         it    = &minfo->mi_it;
         lockh = &minfo->mi_lockh;
 
-        obddev = class_exp2obd(exp);
-
-       obd_put_request_slot(&obddev->u.cli);
         if (OBD_FAIL_CHECK(OBD_FAIL_MDC_GETATTR_ENQUEUE))
                 rc = -ETIMEDOUT;
 
@@ -1286,7 +1383,6 @@ int mdc_intent_getattr_async(struct obd_export *exp,
        struct lookup_intent    *it = &minfo->mi_it;
        struct ptlrpc_request   *req;
        struct mdc_getattr_args *ga;
-       struct obd_device       *obddev = class_exp2obd(exp);
        struct ldlm_res_id       res_id;
        union ldlm_policy_data policy = {
                                .l_inodebits = { MDS_INODELOCK_LOOKUP |
@@ -1307,12 +1403,6 @@ int mdc_intent_getattr_async(struct obd_export *exp,
        if (IS_ERR(req))
                RETURN(PTR_ERR(req));
 
-       rc = obd_get_request_slot(&obddev->u.cli);
-       if (rc != 0) {
-               ptlrpc_req_finished(req);
-               RETURN(rc);
-       }
-
        /* With Data-on-MDT the glimpse callback is needed too.
         * It is set here in advance but not in mdc_finish_enqueue()
         * to avoid possible races. It is safe to have glimpse handler
@@ -1323,13 +1413,11 @@ int mdc_intent_getattr_async(struct obd_export *exp,
        rc = ldlm_cli_enqueue(exp, &req, &minfo->mi_einfo, &res_id, &policy,
                              &flags, NULL, 0, LVB_T_NONE, &minfo->mi_lockh, 1);
        if (rc < 0) {
-               obd_put_request_slot(&obddev->u.cli);
                ptlrpc_req_finished(req);
                RETURN(rc);
        }
 
-       CLASSERT(sizeof(*ga) <= sizeof(req->rq_async_args));
-       ga = ptlrpc_req_async_args(req);
+       ga = ptlrpc_req_async_args(ga, req);
        ga->ga_exp = exp;
        ga->ga_minfo = minfo;