Whamcloud - gitweb
LU-12616 obclass: fix MDS start/stop race
[fs/lustre-release.git] / lustre / mdc / mdc_locks.c
index 32587cf..661cb7f 100644 (file)
@@ -220,7 +220,7 @@ 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 +237,12 @@ 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;
 }
@@ -252,14 +256,17 @@ 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;
+       __u32                    mdt_md_capsule_size;
        struct list_head         cancels = LIST_HEAD_INIT(cancels);
        int                      count = 0;
        enum ldlm_mode           mode;
        int                      rc;
-       int repsize;
+       int repsize, repsize_estimate;
 
        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 +321,21 @@ 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 +355,33 @@ 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.
         */
@@ -351,23 +392,35 @@ mdc_intent_open_pack(struct obd_export *exp, struct lookup_intent *it,
        /* Get real repbuf allocated size as rounded up power of 2 */
        repsize = size_roundup_power2(req->rq_replen +
                                      lustre_msg_early_size());
-
        /* Estimate free space for DoM files in repbuf */
-       repsize -= req->rq_replen - obddev->u.cli.cl_max_mds_easize +
-                  sizeof(struct lov_comp_md_v1) +
-                  sizeof(struct lov_comp_md_entry_v1) +
-                  lov_mds_md_size(0, LOV_MAGIC_V3);
-
-       if (repsize < obddev->u.cli.cl_dom_min_inline_repsize) {
-               repsize = obddev->u.cli.cl_dom_min_inline_repsize - repsize;
+       repsize_estimate = repsize - (req->rq_replen -
+                          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));
+
+       if (repsize_estimate < obddev->u.cli.cl_dom_min_inline_repsize) {
+               repsize = obddev->u.cli.cl_dom_min_inline_repsize -
+                         repsize_estimate + sizeof(struct niobuf_remote);
                req_capsule_set_size(&req->rq_pill, &RMF_NIOBUF_INLINE,
                                     RCL_SERVER,
                                     sizeof(struct niobuf_remote) + repsize);
                ptlrpc_request_set_replen(req);
                CDEBUG(D_INFO, "Increase repbuf by %d bytes, total: %d\n",
                       repsize, req->rq_replen);
+               repsize = size_roundup_power2(req->rq_replen +
+                                             lustre_msg_early_size());
        }
-       return req;
+       /* The only way to report real allocated repbuf size to the server
+        * is the lm_repsize but it must be set prior buffer allocation itself
+        * due to security reasons - it is part of buffer used in signature
+        * calculation (see LU-11414). Therefore the saved size is predicted
+        * value as rq_replen rounded to the next higher power of 2.
+        * Such estimation is safe. Though the final allocated buffer might
+        * be even larger, it is not possible to know that at this point.
+        */
+       req->rq_reqmsg->lm_repsize = repsize;
+       RETURN(req);
 }
 
 #define GA_DEFAULT_EA_NAME_LEN 20
@@ -383,7 +436,7 @@ mdc_intent_getxattr_pack(struct obd_export *exp,
        struct ldlm_intent      *lit;
        int                     rc, count = 0;
        struct list_head        cancels = LIST_HEAD_INIT(cancels);
-       u32 min_buf_size = 0;
+       u32 ea_vals_buf_size = GA_DEFAULT_EA_VAL_LEN * GA_DEFAULT_EA_NUM;
 
        ENTRY;
 
@@ -392,6 +445,16 @@ 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);
@@ -401,6 +464,8 @@ 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
@@ -412,26 +477,25 @@ mdc_intent_getxattr_pack(struct obd_export *exp,
         * of LU-9417 when it would be *more* likely to crash the
         * server. See LU-9856. */
        if (exp->exp_connect_data.ocd_version < OBD_OCD_VERSION(2, 10, 1, 0))
-               min_buf_size = exp->exp_connect_data.ocd_max_easize;
+               ea_vals_buf_size = max_t(u32, ea_vals_buf_size,
+                                        exp->exp_connect_data.ocd_max_easize);
 #endif
 
        /* pack the intended request */
        mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
-                     max_t(u32, min_buf_size,
-                           GA_DEFAULT_EA_VAL_LEN * GA_DEFAULT_EA_NUM),
-                     -1, 0);
+                     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,
-                            max_t(u32, min_buf_size,
-                                GA_DEFAULT_EA_NAME_LEN * GA_DEFAULT_EA_NUM));
+                            GA_DEFAULT_EA_NAME_LEN * GA_DEFAULT_EA_NUM);
 
        req_capsule_set_size(&req->rq_pill, &RMF_EAVALS, RCL_SERVER,
-                            max_t(u32, min_buf_size,
-                                GA_DEFAULT_EA_VAL_LEN * GA_DEFAULT_EA_NUM));
+                            ea_vals_buf_size);
 
        req_capsule_set_size(&req->rq_pill, &RMF_EAVALS_LENS, RCL_SERVER,
-                            max_t(u32, min_buf_size,
-                                sizeof(__u32) * GA_DEFAULT_EA_NUM));
+                            sizeof(u32) * GA_DEFAULT_EA_NUM);
 
        req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, 0);
 
@@ -444,44 +508,77 @@ 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);
 }
@@ -491,10 +588,11 @@ 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);
+       struct list_head cancels = LIST_HEAD_INIT(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),
@@ -502,8 +600,15 @@ 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));
@@ -611,14 +716,14 @@ static int mdc_finish_enqueue(struct obd_export *exp,
          * It's important that we do this first!  Otherwise we might exit the
          * function without doing so, and try to replay a failed create
          * (bug 3440) */
-        if (it->it_op & IT_OPEN && req->rq_replay &&
+       if (it->it_op & IT_OPEN && req->rq_replay &&
            (!it_disposition(it, DISP_OPEN_OPEN) || it->it_status != 0))
                mdc_clear_replay_flag(req, it->it_status);
 
-       DEBUG_REQ(D_RPCTRACE, req, "op: %d 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 */
+       /* We know what to expect, so we do any byte flipping required here */
        if (it_has_reply_body(it)) {
                 body = req_capsule_server_get(pill, &RMF_MDT_BODY);
                 if (body == NULL) {
@@ -679,6 +784,8 @@ static int mdc_finish_enqueue(struct obd_export *exp,
                /* maybe the lock was granted right away and layout
                 * is packed into RMF_DLM_LVB of req */
                lvb_len = req_capsule_get_size(pill, &RMF_DLM_LVB, RCL_SERVER);
+               CDEBUG(D_INFO, "%s: layout return lvb %d transno %lld\n",
+                      class_exp2obd(exp)->obd_name, lvb_len, req->rq_transno);
                if (lvb_len > 0) {
                        lvb_data = req_capsule_server_sized_get(pill,
                                                        &RMF_DLM_LVB, lvb_len);
@@ -736,7 +843,7 @@ 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);
                }
@@ -802,7 +909,8 @@ 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 = imp->imp_connect_data.ocd_max_easize;
+               acl_bufsize = MIN(imp->imp_connect_data.ocd_max_easize,
+                                 XATTR_SIZE_MAX);
        else
                acl_bufsize = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
 
@@ -924,10 +1032,11 @@ resend:
 
        if ((int)lockrep->lock_policy_res2 == -ERANGE &&
            it->it_op & (IT_OPEN | IT_GETATTR | IT_LOOKUP) &&
-           acl_bufsize != imp->imp_connect_data.ocd_max_easize) {
+           acl_bufsize == LUSTRE_POSIX_ACL_MAX_SIZE_OLD) {
                mdc_clear_replay_flag(req, -ERANGE);
                ptlrpc_req_finished(req);
-               acl_bufsize = imp->imp_connect_data.ocd_max_easize;
+               acl_bufsize = MIN(imp->imp_connect_data.ocd_max_easize,
+                                 XATTR_SIZE_MAX);
                goto resend;
        }
 
@@ -1218,18 +1327,18 @@ int mdc_intent_lock(struct obd_export *exp, struct md_op_data *op_data,
 }
 
 static int mdc_intent_getattr_async_interpret(const struct lu_env *env,
-                                              struct ptlrpc_request *req,
-                                              void *args, int rc)
+                                             struct ptlrpc_request *req,
+                                             void *args, int rc)
 {
        struct mdc_getattr_args  *ga = args;
-       struct obd_export        *exp = ga->ga_exp;
-       struct md_enqueue_info   *minfo = ga->ga_minfo;
+       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 obd_device *obddev;
+       struct ldlm_reply *lockrep;
+       __u64 flags = LDLM_FL_HAS_INTENT;
        ENTRY;
 
         it    = &minfo->mi_it;