X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=blobdiff_plain;f=lustre%2Fmdt%2Fmdt_open.c;h=e70858467ca2cd1c174b309707cb154e464de0d7;hp=884eddb1e688932b963e92265b0f77aa9f183891;hb=f2a9374170e4522b9d2ac3b7096cf2912339d480;hpb=bc90073ea5bf2f354a96dfabf0444d36bdb2fa33 diff --git a/lustre/mdt/mdt_open.c b/lustre/mdt/mdt_open.c index 884eddb..e708584 100644 --- a/lustre/mdt/mdt_open.c +++ b/lustre/mdt/mdt_open.c @@ -26,8 +26,11 @@ * GPL HEADER END */ /* - * Copyright 2008 Sun Microsystems, Inc. All rights reserved + * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. + * + * Copyright (c) 2011 Whamcloud, Inc. + * */ /* * This file is part of Lustre, http://www.lustre.org/ @@ -85,10 +88,9 @@ struct mdt_file_data *mdt_handle2mfd(struct mdt_thread_info *info, LASSERT(handle != NULL); mfd = class_handle2object(handle->cookie); /* during dw/setattr replay the mfd can be found by old handle */ - if (mfd == NULL && - lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) { + if (mfd == NULL && req_is_replay(req)) { struct mdt_export_data *med = &req->rq_export->exp_mdt_data; - list_for_each_entry(mfd, &med->med_open_head, mfd_list) { + cfs_list_for_each_entry(mfd, &med->med_open_head, mfd_list) { if (mfd->mfd_old_handle.cookie == handle->cookie) RETURN (mfd); } @@ -100,7 +102,7 @@ struct mdt_file_data *mdt_handle2mfd(struct mdt_thread_info *info, /* free mfd */ void mdt_mfd_free(struct mdt_file_data *mfd) { - LASSERT(list_empty(&mfd->mfd_list)); + LASSERT(cfs_list_empty(&mfd->mfd_list)); OBD_FREE_RCU(mfd, sizeof *mfd, &mfd->mfd_handle); } @@ -108,8 +110,8 @@ static int mdt_create_data(struct mdt_thread_info *info, struct mdt_object *p, struct mdt_object *o) { struct md_op_spec *spec = &info->mti_spec; - struct md_attr *ma = &info->mti_attr; - int rc; + struct md_attr *ma = &info->mti_attr; + int rc = 0; ENTRY; if (!md_should_create(spec->sp_cr_flags)) @@ -117,9 +119,15 @@ static int mdt_create_data(struct mdt_thread_info *info, ma->ma_need = MA_INODE | MA_LOV; ma->ma_valid = 0; - rc = mdo_create_data(info->mti_env, - p ? mdt_object_child(p) : NULL, - mdt_object_child(o), spec, ma); + cfs_down(&o->mot_lov_sem); + if (!(o->mot_flags & MOF_LOV_CREATED)) { + rc = mdo_create_data(info->mti_env, + p ? mdt_object_child(p) : NULL, + mdt_object_child(o), spec, ma); + if (rc == 0 && ma->ma_valid & MA_LOV) + o->mot_flags |= MOF_LOV_CREATED; + } + cfs_up(&o->mot_lov_sem); RETURN(rc); } @@ -162,14 +170,14 @@ int mdt_ioepoch_open(struct mdt_thread_info *info, struct mdt_object *o, !S_ISREG(lu_object_attr(&o->mot_obj.mo_lu))) RETURN(0); - down(&o->mot_ioepoch_sem); + cfs_down(&o->mot_ioepoch_sem); if (mdt_ioepoch_opened(o)) { /* Epoch continues even if there is no writers yet. */ CDEBUG(D_INODE, "continue epoch "LPU64" for "DFID"\n", o->mot_ioepoch, PFID(mdt_object_fid(o))); } else { /* XXX: ->mdt_ioepoch is not initialized at the mount */ - spin_lock(&mdt->mdt_ioepoch_lock); + cfs_spin_lock(&mdt->mdt_ioepoch_lock); if (mdt->mdt_ioepoch < info->mti_replayepoch) mdt->mdt_ioepoch = info->mti_replayepoch; @@ -180,7 +188,7 @@ int mdt_ioepoch_open(struct mdt_thread_info *info, struct mdt_object *o, else o->mot_ioepoch = mdt->mdt_ioepoch; - spin_unlock(&mdt->mdt_ioepoch_lock); + cfs_spin_unlock(&mdt->mdt_ioepoch_lock); CDEBUG(D_INODE, "starting epoch "LPU64" for "DFID"\n", o->mot_ioepoch, PFID(mdt_object_fid(o))); @@ -189,7 +197,7 @@ int mdt_ioepoch_open(struct mdt_thread_info *info, struct mdt_object *o, cancel = 1; } o->mot_ioepoch_count++; - up(&o->mot_ioepoch_sem); + cfs_up(&o->mot_ioepoch_sem); /* Cancel Size-on-MDS attributes cached on clients for the open case. * In the truncate case, see mdt_reint_setattr(). */ @@ -231,7 +239,7 @@ static int mdt_som_attr_set(struct mdt_thread_info *info, ma->ma_som->msd_size = la->la_valid & LA_SIZE ? la->la_size : 0; ma->ma_som->msd_blocks = la->la_valid & LA_BLOCKS ? la->la_blocks : 0; - ma->ma_som->msd_mountid = mdt->mdt_mount_count; + ma->ma_som->msd_mountid = mdt->mdt_lut.lut_obd->u.obt.obt_mount_count; ma->ma_attr.la_valid &= LA_ATIME | LA_MTIME | LA_CTIME; } else { ma->ma_som->msd_ioepoch = IOEPOCH_INVAL; @@ -255,7 +263,7 @@ static inline int mdt_ioepoch_close_on_eviction(struct mdt_thread_info *info, { int rc = 0; - down(&o->mot_ioepoch_sem); + cfs_down(&o->mot_ioepoch_sem); CDEBUG(D_INODE, "Eviction. Closing IOepoch "LPU64" on "DFID". " "Count %d\n", o->mot_ioepoch, PFID(mdt_object_fid(o)), o->mot_ioepoch_count); @@ -268,7 +276,7 @@ static inline int mdt_ioepoch_close_on_eviction(struct mdt_thread_info *info, rc = mdt_som_attr_set(info, o, o->mot_ioepoch, MDT_SOM_DISABLE); mdt_object_som_enable(o, o->mot_ioepoch); } - up(&o->mot_ioepoch_sem); + cfs_up(&o->mot_ioepoch_sem); RETURN(rc); } @@ -283,7 +291,7 @@ static inline int mdt_ioepoch_close_on_replay(struct mdt_thread_info *info, int rc = MDT_IOEPOCH_CLOSED; ENTRY; - down(&o->mot_ioepoch_sem); + cfs_down(&o->mot_ioepoch_sem); CDEBUG(D_INODE, "Replay. Closing epoch "LPU64" on "DFID". Count %d\n", o->mot_ioepoch, PFID(mdt_object_fid(o)), o->mot_ioepoch_count); o->mot_ioepoch_count--; @@ -295,7 +303,7 @@ static inline int mdt_ioepoch_close_on_replay(struct mdt_thread_info *info, if (!mdt_ioepoch_opened(o)) mdt_object_som_enable(o, info->mti_ioepoch->ioepoch); - up(&o->mot_ioepoch_sem); + cfs_up(&o->mot_ioepoch_sem); RETURN(rc); } @@ -324,7 +332,7 @@ static inline int mdt_ioepoch_close_reg(struct mdt_thread_info *info, la = &info->mti_attr.ma_attr; achange = (info->mti_ioepoch->flags & MF_SOM_CHANGE); - down(&o->mot_ioepoch_sem); + cfs_down(&o->mot_ioepoch_sem); o->mot_ioepoch_count--; tmp_ma = &info->mti_u.som.attr; @@ -391,7 +399,7 @@ static inline int mdt_ioepoch_close_reg(struct mdt_thread_info *info, mdt_object_som_enable(o, o->mot_ioepoch); } - up(&o->mot_ioepoch_sem); + cfs_up(&o->mot_ioepoch_sem); /* If recovery is needed, tell the client to perform GETATTR under * the lock. */ if (ret == MDT_IOEPOCH_GETATTR && recovery) { @@ -403,7 +411,7 @@ static inline int mdt_ioepoch_close_reg(struct mdt_thread_info *info, RETURN(rc ? : ret); error_up: - up(&o->mot_ioepoch_sem); + cfs_up(&o->mot_ioepoch_sem); return rc; } @@ -472,7 +480,7 @@ int mdt_som_au_close(struct mdt_thread_info *info, struct mdt_object *o) !(info->mti_attr.ma_attr.la_valid & LA_SIZE))) act = MDT_SOM_DISABLE; - down(&o->mot_ioepoch_sem); + cfs_down(&o->mot_ioepoch_sem); /* Mark the object it is the recovery state if we failed to obtain * SOM attributes. */ if (act == MDT_SOM_DISABLE) @@ -486,7 +494,7 @@ int mdt_som_au_close(struct mdt_thread_info *info, struct mdt_object *o) rc = mdt_som_attr_set(info, o, ioepoch, act); mdt_object_som_enable(o, ioepoch); } - up(&o->mot_ioepoch_sem); + cfs_up(&o->mot_ioepoch_sem); RETURN(rc); } @@ -494,9 +502,9 @@ int mdt_write_read(struct mdt_object *o) { int rc = 0; ENTRY; - down(&o->mot_ioepoch_sem); + cfs_down(&o->mot_ioepoch_sem); rc = o->mot_writecount; - up(&o->mot_ioepoch_sem); + cfs_up(&o->mot_ioepoch_sem); RETURN(rc); } @@ -504,21 +512,21 @@ int mdt_write_get(struct mdt_object *o) { int rc = 0; ENTRY; - down(&o->mot_ioepoch_sem); + cfs_down(&o->mot_ioepoch_sem); if (o->mot_writecount < 0) rc = -ETXTBSY; else o->mot_writecount++; - up(&o->mot_ioepoch_sem); + cfs_up(&o->mot_ioepoch_sem); RETURN(rc); } void mdt_write_put(struct mdt_object *o) { ENTRY; - down(&o->mot_ioepoch_sem); + cfs_down(&o->mot_ioepoch_sem); o->mot_writecount--; - up(&o->mot_ioepoch_sem); + cfs_up(&o->mot_ioepoch_sem); EXIT; } @@ -526,21 +534,21 @@ static int mdt_write_deny(struct mdt_object *o) { int rc = 0; ENTRY; - down(&o->mot_ioepoch_sem); + cfs_down(&o->mot_ioepoch_sem); if (o->mot_writecount > 0) rc = -ETXTBSY; else o->mot_writecount--; - up(&o->mot_ioepoch_sem); + cfs_up(&o->mot_ioepoch_sem); RETURN(rc); } static void mdt_write_allow(struct mdt_object *o) { ENTRY; - down(&o->mot_ioepoch_sem); + cfs_down(&o->mot_ioepoch_sem); o->mot_writecount++; - up(&o->mot_ioepoch_sem); + cfs_up(&o->mot_ioepoch_sem); EXIT; } @@ -552,22 +560,20 @@ static void mdt_empty_transno(struct mdt_thread_info* info) ENTRY; /* transaction has occurred already */ - if (lustre_msg_get_transno(req->rq_repmsg) != 0) { - EXIT; - return; - } + if (lustre_msg_get_transno(req->rq_repmsg) != 0) + RETURN_EXIT; - spin_lock(&mdt->mdt_transno_lock); + cfs_spin_lock(&mdt->mdt_lut.lut_translock); if (info->mti_transno == 0) { - info->mti_transno = ++ mdt->mdt_last_transno; + info->mti_transno = ++ mdt->mdt_lut.lut_last_transno; } else { /* should be replay */ - if (info->mti_transno > mdt->mdt_last_transno) - mdt->mdt_last_transno = info->mti_transno; + if (info->mti_transno > mdt->mdt_lut.lut_last_transno) + mdt->mdt_lut.lut_last_transno = info->mti_transno; } - spin_unlock(&mdt->mdt_transno_lock); + cfs_spin_unlock(&mdt->mdt_lut.lut_translock); - CDEBUG(D_INODE, "transno = %llu, last_committed = %llu\n", + CDEBUG(D_INODE, "transno = "LPU64", last_committed = "LPU64"\n", info->mti_transno, req->rq_export->exp_obd->obd_last_committed); @@ -587,7 +593,7 @@ void mdt_mfd_set_mode(struct mdt_file_data *mfd, int mode) } static int mdt_mfd_open(struct mdt_thread_info *info, struct mdt_object *p, - struct mdt_object *o, int flags, int created) + struct mdt_object *o, __u64 flags, int created) { struct ptlrpc_request *req = mdt_info_req(info); struct mdt_export_data *med = &req->rq_export->exp_mdt_data; @@ -667,7 +673,7 @@ static int mdt_mfd_open(struct mdt_thread_info *info, struct mdt_object *p, mfd->mfd_xid = req->rq_xid; /* replay handle */ - if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) { + if (req_is_replay(req)) { struct mdt_file_data *old_mfd; /* Check wheather old cookie already exist in * the list, becasue when do recovery, client @@ -678,14 +684,15 @@ static int mdt_mfd_open(struct mdt_thread_info *info, struct mdt_object *p, old_mfd = mdt_handle2mfd(info, info->mti_rr.rr_handle); if (old_mfd) { CDEBUG(D_HA, "del orph mfd %p fid=("DFID") " - "cookie=" LPX64"\n", - mfd, + "cookie=" LPX64"\n", mfd, PFID(mdt_object_fid(mfd->mfd_object)), info->mti_rr.rr_handle->cookie); - spin_lock(&med->med_open_lock); + cfs_spin_lock(&med->med_open_lock); class_handle_unhash(&old_mfd->mfd_handle); - list_del_init(&old_mfd->mfd_list); - spin_unlock(&med->med_open_lock); + cfs_list_del_init(&old_mfd->mfd_list); + cfs_spin_unlock(&med->med_open_lock); + /* no attr update for that close */ + la->la_valid = 0; mdt_mfd_close(info, old_mfd); } CDEBUG(D_HA, "Store old cookie "LPX64" in new mfd\n", @@ -696,15 +703,15 @@ static int mdt_mfd_open(struct mdt_thread_info *info, struct mdt_object *p, repbody->handle.cookie = mfd->mfd_handle.h_cookie; if (req->rq_export->exp_disconnected) { - spin_lock(&med->med_open_lock); + cfs_spin_lock(&med->med_open_lock); class_handle_unhash(&mfd->mfd_handle); - list_del_init(&mfd->mfd_list); - spin_unlock(&med->med_open_lock); + cfs_list_del_init(&mfd->mfd_list); + cfs_spin_unlock(&med->med_open_lock); mdt_mfd_close(info, mfd); } else { - spin_lock(&med->med_open_lock); - list_add(&mfd->mfd_list, &med->med_open_head); - spin_unlock(&med->med_open_lock); + cfs_spin_lock(&med->med_open_lock); + cfs_list_add(&mfd->mfd_list, &med->med_open_head); + cfs_spin_unlock(&med->med_open_lock); } mdt_empty_transno(info); @@ -717,7 +724,7 @@ static int mdt_mfd_open(struct mdt_thread_info *info, struct mdt_object *p, static int mdt_finish_open(struct mdt_thread_info *info, struct mdt_object *p, struct mdt_object *o, - int flags, int created, struct ldlm_reply *rep) + __u64 flags, int created, struct ldlm_reply *rep) { struct ptlrpc_request *req = mdt_info_req(info); struct obd_export *exp = req->rq_export; @@ -728,7 +735,7 @@ static int mdt_finish_open(struct mdt_thread_info *info, struct mdt_body *repbody; int rc = 0; int isreg, isdir, islnk; - struct list_head *t; + cfs_list_t *t; ENTRY; LASSERT(ma->ma_valid & MA_INODE); @@ -844,15 +851,15 @@ static int mdt_finish_open(struct mdt_thread_info *info, mfd = NULL; if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) { - spin_lock(&med->med_open_lock); - list_for_each(t, &med->med_open_head) { - mfd = list_entry(t, struct mdt_file_data, mfd_list); + cfs_spin_lock(&med->med_open_lock); + cfs_list_for_each(t, &med->med_open_head) { + mfd = cfs_list_entry(t, struct mdt_file_data, mfd_list); if (mfd->mfd_xid == req->rq_xid) { break; } mfd = NULL; } - spin_unlock(&med->med_open_lock); + cfs_spin_unlock(&med->med_open_lock); if (mfd != NULL) { repbody->handle.cookie = mfd->mfd_handle.h_cookie; @@ -883,8 +890,8 @@ void mdt_reconstruct_open(struct mdt_thread_info *info, struct mdt_device *mdt = info->mti_mdt; struct req_capsule *pill = info->mti_pill; struct ptlrpc_request *req = mdt_info_req(info); - struct mdt_export_data *med = &req->rq_export->exp_mdt_data; - struct lsd_client_data *lcd = med->med_lcd; + struct tg_export_data *ted = &req->rq_export->exp_target_data; + struct lsd_client_data *lcd = ted->ted_lcd; struct md_attr *ma = &info->mti_attr; struct mdt_reint_record *rr = &info->mti_rr; __u32 flags = info->mti_spec.sp_cr_flags; @@ -908,11 +915,11 @@ void mdt_reconstruct_open(struct mdt_thread_info *info, ma->ma_valid = 0; - mdt_req_from_lcd(req, med->med_lcd); + mdt_req_from_lcd(req, lcd); mdt_set_disposition(info, ldlm_rep, lcd->lcd_last_data); - CERROR("This is reconstruct open: disp="LPX64", result=%d\n", - ldlm_rep->lock_policy_res1, req->rq_status); + CDEBUG(D_INODE, "This is reconstruct open: disp="LPX64", result=%d\n", + ldlm_rep->lock_policy_res1, req->rq_status); if (mdt_get_disposition(ldlm_rep, DISP_OPEN_CREATE) && req->rq_status != 0) @@ -930,12 +937,11 @@ void mdt_reconstruct_open(struct mdt_thread_info *info, rc = PTR_ERR(parent); LCONSOLE_WARN("Parent "DFID" lookup error %d." " Evicting client %s with export %s.\n", - PFID(mdt_object_fid(parent)), rc, + PFID(rr->rr_fid1), rc, obd_uuid2str(&exp->exp_client_uuid), obd_export_nid2str(exp)); mdt_export_evict(exp); - EXIT; - return; + RETURN_EXIT; } child = mdt_object_find(env, mdt, rr->rr_fid2); if (IS_ERR(child)) { @@ -947,8 +953,7 @@ void mdt_reconstruct_open(struct mdt_thread_info *info, obd_export_nid2str(exp)); mdt_object_put(env, parent); mdt_export_evict(exp); - EXIT; - return; + RETURN_EXIT; } rc = mdt_object_exists(child); if (rc > 0) { @@ -1026,19 +1031,37 @@ static int mdt_open_by_fid(struct mdt_thread_info* info, RETURN(rc); } -static int mdt_open_anon_by_fid(struct mdt_thread_info* info, +static int mdt_open_anon_by_fid(struct mdt_thread_info *info, struct ldlm_reply *rep, struct mdt_lock_handle *lhc) { + const struct lu_env *env = info->mti_env; + struct mdt_device *mdt = info->mti_mdt; __u32 flags = info->mti_spec.sp_cr_flags; - struct mdt_reint_record *rr = &info->mti_rr; - struct md_attr *ma = &info->mti_attr; + struct mdt_reint_record *rr = &info->mti_rr; + struct md_attr *ma = &info->mti_attr; + struct mdt_object *parent= NULL; struct mdt_object *o; int rc; ldlm_mode_t lm; ENTRY; - o = mdt_object_find(info->mti_env, info->mti_mdt, rr->rr_fid2); + if (md_should_create(flags)) { + if (!lu_fid_eq(rr->rr_fid1, rr->rr_fid2)) { + parent = mdt_object_find(env, mdt, rr->rr_fid1); + if (IS_ERR(parent)) { + CDEBUG(D_INODE, "Fail to find parent "DFID + " for anonymous created %ld, try to" + " use server-side parent.\n", + PFID(rr->rr_fid1), PTR_ERR(parent)); + parent = NULL; + } + } + if (parent == NULL) + ma->ma_need |= MA_PFID; + } + + o = mdt_object_find(env, mdt, rr->rr_fid2); if (IS_ERR(o)) RETURN(rc = PTR_ERR(o)); @@ -1051,7 +1074,6 @@ static int mdt_open_anon_by_fid(struct mdt_thread_info* info, CERROR("NFS remote open shouldn't happen.\n"); GOTO(out, rc); } - mdt_set_disposition(info, rep, (DISP_IT_EXECD | DISP_LOOKUP_EXECD | DISP_LOOKUP_POS)); @@ -1071,20 +1093,33 @@ static int mdt_open_anon_by_fid(struct mdt_thread_info* info, if (rc) GOTO(out, rc); - rc = mo_attr_get(info->mti_env, mdt_object_child(o), ma); + rc = mo_attr_get(env, mdt_object_child(o), ma); if (rc) GOTO(out, rc); + if (ma->ma_valid & MA_PFID) { + parent = mdt_object_find(env, mdt, &ma->ma_pfid); + if (IS_ERR(parent)) { + CDEBUG(D_INODE, "Fail to find parent "DFID + " for anonymous created %ld, try to" + " use system default.\n", + PFID(&ma->ma_pfid), PTR_ERR(parent)); + parent = NULL; + } + } + if (flags & MDS_OPEN_LOCK) mdt_set_disposition(info, rep, DISP_OPEN_LOCK); - rc = mdt_finish_open(info, NULL, o, flags, 0, rep); + rc = mdt_finish_open(info, parent, o, flags, 0, rep); - if (!(flags & MDS_OPEN_LOCK)) + if (!(flags & MDS_OPEN_LOCK) || rc) mdt_object_unlock(info, o, lhc, 1); GOTO(out, rc); out: - mdt_object_put(info->mti_env, o); + mdt_object_put(env, o); + if (parent != NULL) + mdt_object_put(env, parent); return rc; } @@ -1153,7 +1188,7 @@ int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc) struct mdt_body *repbody; struct lu_fid *child_fid = &info->mti_tmp_fid1; struct md_attr *ma = &info->mti_attr; - __u32 create_flags = info->mti_spec.sp_cr_flags; + __u64 create_flags = info->mti_spec.sp_cr_flags; struct mdt_reint_record *rr = &info->mti_rr; struct lu_name *lname; int result, rc; @@ -1164,6 +1199,7 @@ int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc) OBD_FAIL_TIMEOUT_ORSET(OBD_FAIL_MDS_PAUSE_OPEN, OBD_FAIL_ONCE, (obd_timeout + 1) / 4); + mdt_counter_incr(req->rq_export, LPROC_MDT_OPEN); repbody = req_capsule_server_get(info->mti_pill, &RMF_MDT_BODY); ma->ma_lmm = req_capsule_server_get(info->mti_pill, &RMF_MDT_MD); @@ -1189,12 +1225,12 @@ int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc) GOTO(out, result = err_serious(-EINVAL)); CDEBUG(D_INODE, "I am going to open "DFID"/(%s->"DFID") " - "cr_flag=0%o mode=0%06o msg_flag=0x%x\n", + "cr_flag="LPO64" mode=0%06o msg_flag=0x%x\n", PFID(rr->rr_fid1), rr->rr_name, PFID(rr->rr_fid2), create_flags, ma->ma_attr.la_mode, msg_flags); - if (msg_flags & MSG_REPLAY || + if (req_is_replay(req) || (req->rq_export->exp_libclient && create_flags&MDS_OPEN_HAS_EA)) { /* This is a replay request or from liblustre with ea. */ result = mdt_open_by_fid(info, ldlm_rep); @@ -1210,15 +1246,16 @@ int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc) * via a regular replay. */ if (!(create_flags & MDS_OPEN_CREAT)) { - DEBUG_REQ(D_ERROR, req,"OPEN & CREAT not in open replay."); + DEBUG_REQ(D_ERROR, req, + "OPEN & CREAT not in open replay."); GOTO(out, result = -EFAULT); } CDEBUG(D_INFO, "Open replay did find object, continue as " "regular open\n"); - } else if (rr->rr_namelen == 0 && !info->mti_cross_ref) { + } else if (rr->rr_namelen == 0 && !info->mti_cross_ref && + create_flags & MDS_OPEN_LOCK) { result = mdt_open_anon_by_fid(info, ldlm_rep, lhc); - if (result != -ENOENT) - GOTO(out, result); + GOTO(out, result); } if (OBD_FAIL_CHECK(OBD_FAIL_MDS_OPEN_PACK)) @@ -1244,10 +1281,14 @@ int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc) if (IS_ERR(parent)) GOTO(out, result = PTR_ERR(parent)); + /* get and check version of parent */ + result = mdt_version_get_check(info, parent, 0); + if (result) + GOTO(out_parent, result); + fid_zero(child_fid); lname = mdt_name(info->mti_env, (char *)rr->rr_name, rr->rr_namelen); - result = mdo_lookup(info->mti_env, mdt_object_child(parent), lname, child_fid, &info->mti_spec); LASSERTF(ergo(result == 0, fid_is_sane(child_fid)), @@ -1284,17 +1325,23 @@ int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc) if (IS_ERR(child)) GOTO(out_parent, result = PTR_ERR(child)); + /** check version of child */ + rc = mdt_version_get_check(info, child, 1); + if (rc) + GOTO(out_child, result = rc); + mdt_set_capainfo(info, 1, child_fid, BYPASS_CAPA); if (result == -ENOENT) { + /* save versions in reply */ + mdt_version_get_save(info, parent, 0); + mdt_version_get_save(info, child, 1); + + /* version of child will be changed */ + info->mti_mos = child; + /* Not found and with MDS_OPEN_CREAT: let's create it. */ mdt_set_disposition(info, ldlm_rep, DISP_OPEN_CREATE); - info->mti_mos[0] = parent; - info->mti_mos[1] = child; - result = mdt_version_get_check(info, 0); - if (result) - GOTO(out_child, result); - /* Let lower layers know what is lock mode on directory. */ info->mti_spec.sp_cr_mode = mdt_dlm_mode2mdl_mode(lh->mlh_pdo_mode); @@ -1369,10 +1416,9 @@ int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc) LASSERT(!lustre_handle_is_used(&lhc->mlh_reg_lh)); /* get openlock if this is not replay and if a client requested it */ - if (!(msg_flags & MSG_REPLAY) && create_flags & MDS_OPEN_LOCK) { + if (!req_is_replay(req) && create_flags & MDS_OPEN_LOCK) { ldlm_mode_t lm; - LASSERT(!created); if (create_flags & FMODE_WRITE) lm = LCK_CW; else if (create_flags & MDS_FMODE_EXEC) @@ -1420,7 +1466,7 @@ int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc) out_child: mdt_object_put(info->mti_env, child); out_parent: - mdt_object_unlock_put(info, parent, lh, result); + mdt_object_unlock_put(info, parent, lh, result || !created); out: if (result && result != -EREMOTE) lustre_msg_set_transno(req->rq_repmsg, 0); @@ -1462,12 +1508,13 @@ int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd) if ((mode & MDS_FMODE_EXEC || mode & FMODE_READ || mode & FMODE_WRITE) && (ma->ma_valid & MA_INODE) && (ma->ma_attr.la_valid & LA_ATIME)) { /* Set the atime only. */ + ma->ma_valid = MA_INODE; ma->ma_attr.la_valid = LA_ATIME; rc = mo_attr_set(info->mti_env, next, ma); } ma->ma_need |= MA_INODE; - ma->ma_valid = 0; + ma->ma_valid &= ~MA_INODE; if (!MFD_CLOSED(mode)) rc = mo_close(info->mti_env, next, ma); @@ -1483,10 +1530,10 @@ int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd) LASSERT(mdt_info_req(info)); med = &mdt_info_req(info)->rq_export->exp_mdt_data; - spin_lock(&med->med_open_lock); - list_add(&mfd->mfd_list, &med->med_open_head); + cfs_spin_lock(&med->med_open_lock); + cfs_list_add(&mfd->mfd_list, &med->med_open_head); class_handle_hash_back(&mfd->mfd_handle); - spin_unlock(&med->med_open_lock); + cfs_spin_unlock(&med->med_open_lock); if (ret == MDT_IOEPOCH_OPENED) { ret = 0; @@ -1514,6 +1561,7 @@ int mdt_close(struct mdt_thread_info *info) int rc, ret = 0; ENTRY; + mdt_counter_incr(req->rq_export, LPROC_MDT_CLOSE); /* Close may come with the Size-on-MDS update. Unpack it. */ rc = mdt_close_unpack(info); if (rc) @@ -1554,10 +1602,10 @@ int mdt_close(struct mdt_thread_info *info) } med = &req->rq_export->exp_mdt_data; - spin_lock(&med->med_open_lock); + cfs_spin_lock(&med->med_open_lock); mfd = mdt_handle2mfd(info, &info->mti_ioepoch->handle); if (mdt_mfd_closed(mfd)) { - spin_unlock(&med->med_open_lock); + cfs_spin_unlock(&med->med_open_lock); CDEBUG(D_INODE, "no handle for file close: fid = "DFID ": cookie = "LPX64"\n", PFID(info->mti_rr.rr_fid1), info->mti_ioepoch->handle.cookie); @@ -1565,8 +1613,8 @@ int mdt_close(struct mdt_thread_info *info) rc = -ESTALE; } else { class_handle_unhash(&mfd->mfd_handle); - list_del_init(&mfd->mfd_list); - spin_unlock(&med->med_open_lock); + cfs_list_del_init(&mfd->mfd_list); + cfs_spin_unlock(&med->med_open_lock); /* Do not lose object before last unlink. */ o = mfd->mfd_object; @@ -1624,10 +1672,10 @@ int mdt_done_writing(struct mdt_thread_info *info) RETURN(lustre_msg_get_status(req->rq_repmsg)); med = &info->mti_exp->exp_mdt_data; - spin_lock(&med->med_open_lock); + cfs_spin_lock(&med->med_open_lock); mfd = mdt_handle2mfd(info, &info->mti_ioepoch->handle); if (mfd == NULL) { - spin_unlock(&med->med_open_lock); + cfs_spin_unlock(&med->med_open_lock); CDEBUG(D_INODE, "no handle for done write: fid = "DFID ": cookie = "LPX64" ioepoch = "LPU64"\n", PFID(info->mti_rr.rr_fid1), @@ -1645,21 +1693,21 @@ int mdt_done_writing(struct mdt_thread_info *info) LASSERT(mfd->mfd_mode == FMODE_EPOCH || mfd->mfd_mode == FMODE_TRUNC); class_handle_unhash(&mfd->mfd_handle); - list_del_init(&mfd->mfd_list); - spin_unlock(&med->med_open_lock); + cfs_list_del_init(&mfd->mfd_list); + cfs_spin_unlock(&med->med_open_lock); /* Set EPOCH CLOSE flag if not set by client. */ info->mti_ioepoch->flags |= MF_EPOCH_CLOSE; info->mti_attr.ma_valid = 0; info->mti_attr.ma_lmm_size = info->mti_mdt->mdt_max_mdsize; - OBD_ALLOC(info->mti_attr.ma_lmm, info->mti_attr.ma_lmm_size); + OBD_ALLOC_LARGE(info->mti_attr.ma_lmm, info->mti_mdt->mdt_max_mdsize); if (info->mti_attr.ma_lmm == NULL) RETURN(-ENOMEM); rc = mdt_mfd_close(info, mfd); - OBD_FREE(info->mti_attr.ma_lmm, info->mti_attr.ma_lmm_size); + OBD_FREE_LARGE(info->mti_attr.ma_lmm, info->mti_mdt->mdt_max_mdsize); mdt_empty_transno(info); RETURN(rc); }