Whamcloud - gitweb
LU-9728 osd: use GFP_HIGHUSER for non-local IO
[fs/lustre-release.git] / lustre / ofd / ofd_io.c
index bd2da51..5313655 100644 (file)
 
 #include <linux/kthread.h>
 #include "ofd_internal.h"
+#include <lustre_nodemap.h>
 
 struct ofd_inconsistency_item {
        struct list_head         oii_list;
        struct ofd_object       *oii_obj;
-       struct lu_fid            oii_pfid;
+       struct filter_fid        oii_ff;
 };
 
 /**
@@ -60,27 +61,28 @@ struct ofd_inconsistency_item {
  * \param[in] env      execution environment
  * \param[in] ofd      OFD device
  * \param[in] oii      object-related local data
- * \param[in] lr       LFSCK request data
+ * \param[in] lrl      LFSCK request data
  */
 static void ofd_inconsistency_verify_one(const struct lu_env *env,
                                         struct ofd_device *ofd,
                                         struct ofd_inconsistency_item *oii,
-                                        struct lfsck_request *lr)
+                                        struct lfsck_req_local *lrl)
 {
-       struct ofd_object       *fo     = oii->oii_obj;
-       struct lu_fid           *pfid   = &fo->ofo_pfid;
-       int                      rc;
+       struct ofd_object *fo = oii->oii_obj;
+       struct filter_fid *client_ff = &oii->oii_ff;
+       struct filter_fid *local_ff = &fo->ofo_ff;
+       int rc;
 
        LASSERT(fo->ofo_pfid_checking);
        LASSERT(!fo->ofo_pfid_verified);
 
-       lr->lr_fid = fo->ofo_header.loh_fid; /* OST-object itself FID. */
-       lr->lr_fid2 = oii->oii_pfid; /* client given PFID. */
-       lr->lr_fid3 = *pfid; /* OST local stored PFID. */
+       lrl->lrl_fid = fo->ofo_header.loh_fid; /* OST-object itself FID. */
+       lrl->lrl_ff_client = *client_ff; /* client given PFID. */
+       lrl->lrl_ff_local = *local_ff; /* OST local stored PFID. */
 
-       rc = lfsck_in_notify(env, ofd->ofd_osd, lr, NULL);
+       rc = lfsck_in_notify_local(env, ofd->ofd_osd, lrl, NULL);
        ofd_write_lock(env, fo);
-       switch (lr->lr_status) {
+       switch (lrl->lrl_status) {
        case LPVS_INIT:
                LASSERT(rc <= 0);
 
@@ -89,7 +91,8 @@ static void ofd_inconsistency_verify_one(const struct lu_env *env,
                               "PFID xattr for "DFID", the client given PFID "
                               DFID", OST local stored PFID "DFID": rc = %d\n",
                               ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
-                              PFID(&oii->oii_pfid), PFID(pfid), rc);
+                              PFID(&client_ff->ff_parent),
+                              PFID(&local_ff->ff_parent), rc);
                else
                        fo->ofo_pfid_verified = 1;
                break;
@@ -102,14 +105,16 @@ static void ofd_inconsistency_verify_one(const struct lu_env *env,
                               "PFID for "DFID", the client given PFID "DFID
                               ", local stored PFID "DFID": rc = %d\n",
                               ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
-                              PFID(&oii->oii_pfid), PFID(pfid), rc);
+                              PFID(&client_ff->ff_parent),
+                              PFID(&local_ff->ff_parent), rc);
                else
                        CDEBUG(D_LFSCK, "%s: both the client given PFID and "
                               "the OST local stored PFID are stale for the "
                               "OST-object "DFID", client given PFID is "DFID
                               ", local stored PFID is "DFID"\n",
                               ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
-                              PFID(&oii->oii_pfid), PFID(pfid));
+                              PFID(&client_ff->ff_parent),
+                              PFID(&local_ff->ff_parent));
                break;
        case LPVS_INCONSISTENT_TOFIX:
                ofd->ofd_inconsistency_self_detected++;
@@ -119,15 +124,17 @@ static void ofd_inconsistency_verify_one(const struct lu_env *env,
                               "for "DFID", with the client given PFID "DFID
                               ", the old stored PFID "DFID"\n",
                               ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
-                              PFID(&oii->oii_pfid), PFID(pfid));
+                              PFID(&client_ff->ff_parent),
+                              PFID(&local_ff->ff_parent));
                } else if (rc < 0) {
                        CDEBUG(D_LFSCK, "%s: fail to fix the OST PFID xattr "
                               "for "DFID", client given PFID "DFID", local "
                               "stored PFID "DFID": rc = %d\n",
                               ofd_name(ofd), PFID(&fo->ofo_header.loh_fid),
-                              PFID(&oii->oii_pfid), PFID(pfid), rc);
+                              PFID(&client_ff->ff_parent),
+                              PFID(&local_ff->ff_parent), rc);
                }
-               *pfid = oii->oii_pfid;
+               local_ff->ff_parent = client_ff->ff_parent;
                fo->ofo_pfid_verified = 1;
                break;
        default:
@@ -136,7 +143,7 @@ static void ofd_inconsistency_verify_one(const struct lu_env *env,
        fo->ofo_pfid_checking = 0;
        ofd_write_unlock(env, fo);
 
-       lu_object_put(env, &fo->ofo_obj.do_lu);
+       ofd_object_put(env, fo);
        OBD_FREE_PTR(oii);
 }
 
@@ -153,29 +160,29 @@ static void ofd_inconsistency_verify_one(const struct lu_env *env,
  */
 static int ofd_inconsistency_verification_main(void *args)
 {
-       struct lu_env                  env;
-       struct ofd_device             *ofd    = args;
-       struct ptlrpc_thread          *thread = &ofd->ofd_inconsistency_thread;
+       struct lu_env env;
+       struct ofd_device *ofd = args;
+       struct ptlrpc_thread *thread = &ofd->ofd_inconsistency_thread;
        struct ofd_inconsistency_item *oii;
-       struct lfsck_request          *lr     = NULL;
-       struct l_wait_info             lwi    = { 0 };
-       int                            rc;
+       struct lfsck_req_local *lrl = NULL;
+       struct l_wait_info lwi = { 0 };
+       int rc;
        ENTRY;
 
        rc = lu_env_init(&env, LCT_DT_THREAD);
        spin_lock(&ofd->ofd_inconsistency_lock);
-       thread_set_flags(thread, rc != 0 ? SVC_STOPPED : SVC_RUNNING);
+       thread_set_flags(thread, rc ? SVC_STOPPED : SVC_RUNNING);
        wake_up_all(&thread->t_ctl_waitq);
        spin_unlock(&ofd->ofd_inconsistency_lock);
-       if (rc != 0)
+       if (rc)
                RETURN(rc);
 
-       OBD_ALLOC_PTR(lr);
-       if (unlikely(lr == NULL))
+       OBD_ALLOC_PTR(lrl);
+       if (unlikely(!lrl))
                GOTO(out_unlocked, rc = -ENOMEM);
 
-       lr->lr_event = LE_PAIRS_VERIFY;
-       lr->lr_active = LFSCK_TYPE_LAYOUT;
+       lrl->lrl_event = LEL_PAIRS_VERIFY_LOCAL;
+       lrl->lrl_active = LFSCK_TYPE_LAYOUT;
 
        spin_lock(&ofd->ofd_inconsistency_lock);
        while (1) {
@@ -188,7 +195,7 @@ static int ofd_inconsistency_verification_main(void *args)
                                         oii_list);
                        list_del_init(&oii->oii_list);
                        spin_unlock(&ofd->ofd_inconsistency_lock);
-                       ofd_inconsistency_verify_one(&env, ofd, oii, lr);
+                       ofd_inconsistency_verify_one(&env, ofd, oii, lrl);
                        spin_lock(&ofd->ofd_inconsistency_lock);
                }
 
@@ -214,12 +221,12 @@ static int ofd_inconsistency_verification_main(void *args)
                fo->ofo_pfid_checking = 0;
                ofd_write_unlock(&env, fo);
 
-               lu_object_put(&env, &fo->ofo_obj.do_lu);
+               ofd_object_put(&env, fo);
                OBD_FREE_PTR(oii);
                spin_lock(&ofd->ofd_inconsistency_lock);
        }
 
-       OBD_FREE_PTR(lr);
+       OBD_FREE_PTR(lrl);
 
        GOTO(out, rc = 0);
 
@@ -320,9 +327,10 @@ int ofd_stop_inconsistency_verification_thread(struct ofd_device *ofd)
 static void ofd_add_inconsistency_item(const struct lu_env *env,
                                       struct ofd_object *fo, struct obdo *oa)
 {
-       struct ofd_device               *ofd    = ofd_obj2dev(fo);
-       struct ofd_inconsistency_item   *oii;
-       bool                             wakeup = false;
+       struct ofd_device *ofd = ofd_obj2dev(fo);
+       struct ofd_inconsistency_item *oii;
+       struct filter_fid *ff;
+       bool wakeup = false;
 
        OBD_ALLOC_PTR(oii);
        if (oii == NULL)
@@ -331,9 +339,11 @@ static void ofd_add_inconsistency_item(const struct lu_env *env,
        INIT_LIST_HEAD(&oii->oii_list);
        lu_object_get(&fo->ofo_obj.do_lu);
        oii->oii_obj = fo;
-       oii->oii_pfid.f_seq = oa->o_parent_seq;
-       oii->oii_pfid.f_oid = oa->o_parent_oid;
-       oii->oii_pfid.f_stripe_idx = oa->o_stripe_idx;
+       ff = &oii->oii_ff;
+       ff->ff_parent.f_seq = oa->o_parent_seq;
+       ff->ff_parent.f_oid = oa->o_parent_oid;
+       ff->ff_parent.f_stripe_idx = oa->o_stripe_idx;
+       ff->ff_layout = oa->o_layout;
 
        spin_lock(&ofd->ofd_inconsistency_lock);
        if (fo->ofo_pfid_checking || fo->ofo_pfid_verified) {
@@ -378,8 +388,8 @@ static void ofd_add_inconsistency_item(const struct lu_env *env,
 int ofd_verify_ff(const struct lu_env *env, struct ofd_object *fo,
                  struct obdo *oa)
 {
-       struct lu_fid   *pfid   = &fo->ofo_pfid;
-       int              rc     = 0;
+       struct lu_fid *pfid = &fo->ofo_ff.ff_parent;
+       int rc = 0;
        ENTRY;
 
        if (fid_is_sane(pfid)) {
@@ -443,8 +453,9 @@ static int ofd_preprw_read(const struct lu_env *env, struct obd_export *exp,
                           struct niobuf_remote *rnb, int *nr_local,
                           struct niobuf_local *lnb, char *jobid)
 {
-       struct ofd_object       *fo;
-       int                      i, j, rc, tot_bytes = 0;
+       struct ofd_object *fo;
+       int i, j, rc, tot_bytes = 0;
+       enum dt_bufs_type dbt = DT_BUFS_TYPE_READ;
 
        ENTRY;
        LASSERT(env != NULL);
@@ -464,10 +475,12 @@ static int ofd_preprw_read(const struct lu_env *env, struct obd_export *exp,
                        GOTO(unlock, rc);
        }
 
-       *nr_local = 0;
-       for (i = 0, j = 0; i < niocount; i++) {
+       if (ptlrpc_connection_is_local(exp->exp_connection))
+               dbt |= DT_BUFS_TYPE_LOCAL;
+
+       for (*nr_local = 0, i = 0, j = 0; i < niocount; i++) {
                rc = dt_bufs_get(env, ofd_object_child(fo), rnb + i,
-                                lnb + j, 0);
+                                lnb + j, dbt);
                if (unlikely(rc < 0))
                        GOTO(buf_put, rc);
                LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
@@ -528,8 +541,9 @@ static int ofd_preprw_write(const struct lu_env *env, struct obd_export *exp,
                            struct niobuf_remote *rnb, int *nr_local,
                            struct niobuf_local *lnb, char *jobid)
 {
-       struct ofd_object       *fo;
-       int                      i, j, k, rc = 0, tot_bytes = 0;
+       struct ofd_object *fo;
+       int i, j, k, rc = 0, tot_bytes = 0;
+       enum dt_bufs_type dbt = DT_BUFS_TYPE_WRITE;
 
        ENTRY;
        LASSERT(env != NULL);
@@ -616,13 +630,15 @@ static int ofd_preprw_write(const struct lu_env *env, struct obd_export *exp,
 
        /* Process incoming grant info, set OBD_BRW_GRANTED flag and grant some
         * space back if possible */
-       ofd_grant_prepare_write(env, exp, oa, rnb, obj->ioo_bufcnt);
+       tgt_grant_prepare_write(env, exp, oa, rnb, obj->ioo_bufcnt);
+
+       if (ptlrpc_connection_is_local(exp->exp_connection))
+               dbt |= DT_BUFS_TYPE_LOCAL;
 
        /* parse remote buffers to local buffers and prepare the latter */
-       *nr_local = 0;
-       for (i = 0, j = 0; i < obj->ioo_bufcnt; i++) {
+       for (*nr_local = 0, i = 0, j = 0; i < obj->ioo_bufcnt; i++) {
                rc = dt_bufs_get(env, ofd_object_child(fo),
-                                rnb + i, lnb + j, 1);
+                                rnb + i, lnb + j, dbt);
                if (unlikely(rc < 0))
                        GOTO(err, rc);
                LASSERT(rc <= PTLRPC_MAX_BRW_PAGES);
@@ -649,13 +665,13 @@ err:
        dt_bufs_put(env, ofd_object_child(fo), lnb, *nr_local);
        ofd_read_unlock(env, fo);
        ofd_object_put(env, fo);
-       /* ofd_grant_prepare_write() was called, so we must commit */
-       ofd_grant_commit(exp, oa->o_grant_used, rc);
+       /* tgt_grant_prepare_write() was called, so we must commit */
+       tgt_grant_commit(exp, oa->o_grant_used, rc);
 out:
        /* let's still process incoming grant information packed in the oa,
         * but without enforcing grant since we won't proceed with the write.
         * Just like a read request actually. */
-       ofd_grant_prepare_read(env, exp, oa);
+       tgt_grant_prepare_read(env, exp, oa);
        return rc;
 }
 
@@ -736,7 +752,7 @@ int ofd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
                rc = ofd_preprw_write(env, exp, ofd, fid, &info->fti_attr, oa,
                                      objcount, obj, rnb, nr_local, lnb, jobid);
        } else if (cmd == OBD_BRW_READ) {
-               ofd_grant_prepare_read(env, exp, oa);
+               tgt_grant_prepare_read(env, exp, oa);
                rc = ofd_preprw_read(env, exp, ofd, fid, &info->fti_attr, oa,
                                     obj->ioo_bufcnt, rnb, nr_local, lnb,
                                     jobid);
@@ -824,9 +840,9 @@ ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd,
        dt_obj = ofd_object_child(ofd_obj);
        LASSERT(dt_obj != NULL);
 
-       la->la_valid &= LA_UID | LA_GID;
+       la->la_valid &= LA_UID | LA_GID | LA_PROJID;
 
-       rc = ofd_attr_handle_ugid(env, ofd_obj, la, 0 /* !is_setattr */);
+       rc = ofd_attr_handle_id(env, ofd_obj, la, 0 /* !is_setattr */);
        if (rc != 0)
                GOTO(out, rc);
 
@@ -855,9 +871,8 @@ ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd,
        if (ff_needed) {
                if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR1))
                        ff->ff_parent.f_oid = cpu_to_le32(1UL << 31);
-               if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
-                       ff->ff_parent.f_oid =
-                       cpu_to_le32(le32_to_cpu(ff->ff_parent.f_oid) - 1);
+               else if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_UNMATCHED_PAIR2))
+                       le32_add_cpu(&ff->ff_parent.f_oid, -1);
 
                info->fti_buf.lb_buf = ff;
                info->fti_buf.lb_len = sizeof(*ff);
@@ -873,7 +888,7 @@ ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd,
        if (rc)
                GOTO(out_tx, rc);
 
-       /* set uid/gid */
+       /* set uid/gid/projid */
        if (la->la_valid) {
                rc = dt_attr_set(env, dt_obj, la, th);
                if (rc)
@@ -887,16 +902,8 @@ ofd_write_attr_set(const struct lu_env *env, struct ofd_device *ofd,
 
                rc = dt_xattr_set(env, dt_obj, &info->fti_buf, XATTR_NAME_FID,
                                  0, th);
-               if (rc == 0) {
-                       ofd_obj->ofo_pfid.f_seq = le64_to_cpu(ff->ff_parent.f_seq);
-                       ofd_obj->ofo_pfid.f_oid = le32_to_cpu(ff->ff_parent.f_oid);
-                       /* Currently, the filter_fid::ff_parent::f_ver is not
-                        * the real parent MDT-object's FID::f_ver, instead it
-                        * is the OST-object index in its parent MDT-object's
-                        * layout EA. */
-                       ofd_obj->ofo_pfid.f_stripe_idx =
-                                       le32_to_cpu(ff->ff_parent.f_stripe_idx);
-               }
+               if (!rc)
+                       filter_fid_le_to_cpu(&ofd_obj->ofo_ff, ff, sizeof(*ff));
        }
 
        GOTO(out_tx, rc);
@@ -1133,7 +1140,7 @@ out_stop:
        }
 
        if (rc == 0 && granted > 0) {
-               if (ofd_grant_commit_cb_add(th, exp, granted) == 0)
+               if (tgt_grant_commit_cb_add(th, exp, granted) == 0)
                        granted = 0;
        }
 
@@ -1160,7 +1167,7 @@ out:
        /* second put is pair to object_get in ofd_preprw_write */
        ofd_object_put(env, fo);
        if (granted > 0)
-               ofd_grant_commit(exp, granted, old_rc);
+               tgt_grant_commit(exp, granted, old_rc);
        RETURN(rc);
 }
 
@@ -1202,6 +1209,8 @@ int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
        LASSERT(npages > 0);
 
        if (cmd == OBD_BRW_WRITE) {
+               struct lu_nodemap *nodemap;
+
                /* Don't update timestamps if this write is older than a
                 * setattr which modifies the timestamps. b=10150 */
 
@@ -1209,7 +1218,7 @@ int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
                 * to be changed to ofd_fmd_get() to create the fmd if it
                 * doesn't already exist so we can store the reservation handle
                 * there. */
-               valid = OBD_MD_FLUID | OBD_MD_FLGID;
+               valid = OBD_MD_FLUID | OBD_MD_FLGID | OBD_MD_FLPROJID;
                fmd = ofd_fmd_find(exp, fid);
                if (!fmd || fmd->fmd_mactime_xid < info->fti_xid)
                        valid |= OBD_MD_FLATIME | OBD_MD_FLMTIME |
@@ -1227,9 +1236,11 @@ int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
                                        oa->o_grant_used, old_rc);
                if (rc == 0)
                        obdo_from_la(oa, &info->fti_attr,
-                                    OFD_VALID_FLAGS | LA_GID | LA_UID);
+                                    OFD_VALID_FLAGS | LA_GID | LA_UID |
+                                    LA_PROJID);
                else
-                       obdo_from_la(oa, &info->fti_attr, LA_GID | LA_UID);
+                       obdo_from_la(oa, &info->fti_attr, LA_GID | LA_UID |
+                                    LA_PROJID);
 
                /* don't report overquota flag if we failed before reaching
                 * commit */
@@ -1248,9 +1259,29 @@ int ofd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
                                else
                                        oa->o_flags = OBD_FL_NO_GRPQUOTA;
                        }
+                       if (lnb[0].lnb_flags & OBD_BRW_OVER_PRJQUOTA) {
+                               if (oa->o_valid & OBD_MD_FLFLAGS)
+                                       oa->o_flags |= OBD_FL_NO_PRJQUOTA;
+                               else
+                                       oa->o_flags = OBD_FL_NO_PRJQUOTA;
+                       }
 
                        oa->o_valid |= OBD_MD_FLFLAGS;
-                       oa->o_valid |= OBD_MD_FLUSRQUOTA | OBD_MD_FLGRPQUOTA;
+                       oa->o_valid |= OBD_MD_FLALLQUOTA;
+               }
+
+               /* Convert back to client IDs. LU-9671.
+                * nodemap_get_from_exp() may fail due to nodemap deactivated,
+                * server ID will be returned back to client in that case. */
+               nodemap = nodemap_get_from_exp(exp);
+               if (nodemap != NULL && !IS_ERR(nodemap)) {
+                       oa->o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
+                                                  NODEMAP_FS_TO_CLIENT,
+                                                  oa->o_uid);
+                       oa->o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
+                                                  NODEMAP_FS_TO_CLIENT,
+                                                  oa->o_gid);
+                       nodemap_putref(nodemap);
                }
        } else if (cmd == OBD_BRW_READ) {
                struct ldlm_namespace *ns = ofd->ofd_namespace;