obj-$(CONFIG_LUSTRE_FS) += lustre.o
-lustre-y := dcache.o dir.o file.o llite_close.o llite_lib.o llite_nfs.o \
+lustre-y := dcache.o dir.o file.o llite_lib.o llite_nfs.o \
rw.o namei.o symlink.o llite_mmap.o range_lock.o \
xattr.o xattr_cache.o rw26.o super25.o statahead.o \
glimpse.o lcommon_cl.o lcommon_misc.o \
op_data->op_attr.ia_size = i_size_read(inode);
op_data->op_attr_blocks = inode->i_blocks;
op_data->op_attr_flags = ll_inode_to_ext_flags(inode->i_flags);
- op_data->op_ioepoch = ll_i2info(inode)->lli_ioepoch;
if (fh)
op_data->op_handle = *fh;
}
/**
- * Closes the IO epoch and packs all the attributes into @op_data for
- * the CLOSE rpc.
+ * Packs all the attributes into @op_data for the CLOSE rpc.
*/
static void ll_prepare_close(struct inode *inode, struct md_op_data *op_data,
struct obd_client_handle *och)
if (!(och->och_flags & FMODE_WRITE))
goto out;
- if (!exp_connect_som(ll_i2mdexp(inode)) || !S_ISREG(inode->i_mode))
- op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
- else
- ll_ioepoch_close(inode, op_data, &och, 0);
-
+ op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
out:
ll_pack_inode2opdata(inode, op_data, &och->och_fh);
ll_prep_md_op_data(op_data, inode, NULL, NULL,
struct md_op_data *op_data;
struct ptlrpc_request *req = NULL;
struct obd_device *obd = class_exp2obd(exp);
- int epoch_close = 1;
int rc;
if (!obd) {
op_data->op_lease_handle = och->och_lease_handle;
op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
}
- epoch_close = op_data->op_flags & MF_EPOCH_CLOSE;
+
rc = md_close(md_exp, op_data, och->och_mod, &req);
- if (rc == -EAGAIN) {
- /* This close must have the epoch closed. */
- LASSERT(epoch_close);
- /* MDS has instructed us to obtain Size-on-MDS attribute from
- * OSTs and send setattr to back to MDS.
- */
- rc = ll_som_update(inode, op_data);
- if (rc) {
- CERROR("%s: inode "DFID" mdc Size-on-MDS update failed: rc = %d\n",
- ll_i2mdexp(inode)->exp_obd->obd_name,
- PFID(ll_inode2fid(inode)), rc);
- rc = 0;
- }
- } else if (rc) {
+ if (rc) {
CERROR("%s: inode "DFID" mdc close failed: rc = %d\n",
ll_i2mdexp(inode)->exp_obd->obd_name,
PFID(ll_inode2fid(inode)), rc);
ll_finish_md_op_data(op_data);
out:
- if (exp_connect_som(exp) && !epoch_close &&
- S_ISREG(inode->i_mode) && (och->och_flags & FMODE_WRITE)) {
- ll_queue_done_writing(inode, LLIF_DONE_WRITING);
- } else {
- md_clear_open_replay_data(md_exp, och);
- /* Free @och if it is not waiting for DONE_WRITING. */
- och->och_fh.cookie = DEAD_HANDLE_MAGIC;
- kfree(och);
- }
+ md_clear_open_replay_data(md_exp, och);
+ och->och_fh.cookie = DEAD_HANDLE_MAGIC;
+ kfree(och);
+
if (req) /* This is close request */
ptlrpc_req_finished(req);
return rc;
return rc;
}
-/**
- * Assign an obtained @ioepoch to client's inode. No lock is needed, MDS does
- * not believe attributes if a few ioepoch holders exist. Attributes for
- * previous ioepoch if new one is opened are also skipped by MDS.
- */
-void ll_ioepoch_open(struct ll_inode_info *lli, __u64 ioepoch)
-{
- if (ioepoch && lli->lli_ioepoch != ioepoch) {
- lli->lli_ioepoch = ioepoch;
- CDEBUG(D_INODE, "Epoch %llu opened on "DFID"\n",
- ioepoch, PFID(&lli->lli_fid));
- }
-}
-
static int ll_och_fill(struct obd_export *md_exp, struct lookup_intent *it,
struct obd_client_handle *och)
{
struct ll_file_data *fd, struct obd_client_handle *och)
{
struct inode *inode = file_inode(file);
- struct ll_inode_info *lli = ll_i2info(inode);
LASSERT(!LUSTRE_FPRIVATE(file));
LASSERT(fd);
if (och) {
- struct mdt_body *body;
int rc;
rc = ll_och_fill(ll_i2sbi(inode)->ll_md_exp, it, och);
if (rc != 0)
return rc;
-
- body = req_capsule_server_get(&it->it_request->rq_pill,
- &RMF_MDT_BODY);
- ll_ioepoch_open(lli, body->mbo_ioepoch);
}
LUSTRE_FPRIVATE(file) = fd;
/* Fills the obdo with the attributes for the lsm */
static int ll_lsm_getattr(struct lov_stripe_md *lsm, struct obd_export *exp,
- struct obdo *obdo, __u64 ioepoch, int dv_flags)
+ struct obdo *obdo, int dv_flags)
{
struct ptlrpc_request_set *set;
struct obd_info oinfo = { };
oinfo.oi_oa = obdo;
oinfo.oi_oa->o_oi = lsm->lsm_oi;
oinfo.oi_oa->o_mode = S_IFREG;
- oinfo.oi_oa->o_ioepoch = ioepoch;
oinfo.oi_oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE |
OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
OBD_MD_FLBLKSZ | OBD_MD_FLATIME |
OBD_MD_FLMTIME | OBD_MD_FLCTIME |
- OBD_MD_FLGROUP | OBD_MD_FLEPOCH |
- OBD_MD_FLDATAVERSION;
+ OBD_MD_FLGROUP | OBD_MD_FLDATAVERSION;
if (dv_flags & (LL_DV_WR_FLUSH | LL_DV_RD_FLUSH)) {
oinfo.oi_oa->o_valid |= OBD_MD_FLFLAGS;
oinfo.oi_oa->o_flags |= OBD_FL_SRVLOCK;
return rc;
}
-/**
- * Performs the getattr on the inode and updates its fields.
- * If @sync != 0, perform the getattr under the server-side lock.
- */
-int ll_inode_getattr(struct inode *inode, struct obdo *obdo,
- __u64 ioepoch, int sync)
-{
- struct lov_stripe_md *lsm;
- int rc;
-
- lsm = ccc_inode_lsm_get(inode);
- rc = ll_lsm_getattr(lsm, ll_i2dtexp(inode),
- obdo, ioepoch, sync ? LL_DV_RD_FLUSH : 0);
- if (rc == 0) {
- struct ost_id *oi = lsm ? &lsm->lsm_oi : &obdo->o_oi;
-
- obdo_refresh_inode(inode, obdo, obdo->o_valid);
- CDEBUG(D_INODE, "objid " DOSTID " size %llu, blocks %llu, blksize %lu\n",
- POSTID(oi), i_size_read(inode),
- (unsigned long long)inode->i_blocks,
- 1UL << inode->i_blkbits);
- }
- ccc_inode_lsm_put(inode, lsm);
- return rc;
-}
-
int ll_merge_attr(const struct lu_env *env, struct inode *inode)
{
struct ll_inode_info *lli = ll_i2info(inode);
struct obdo obdo = { 0 };
int rc;
- rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, &obdo, 0, 0);
+ rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, &obdo, 0);
if (rc == 0) {
st->st_size = obdo.o_size;
st->st_blocks = obdo.o_blocks;
goto out;
}
- rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, obdo, 0, flags);
+ rc = ll_lsm_getattr(lsm, sbi->ll_dt_exp, obdo, flags);
if (rc == 0) {
if (!(obdo->o_valid & OBD_MD_FLDATAVERSION))
rc = -EOPNOTSUPP;
{
struct ll_inode_info *lli = ll_i2info(inode);
const struct lu_fid *fid = lu_object_fid(&clob->co_lu);
- int result;
+ int result = 0;
- result = 0;
- if (!(lli->lli_flags & LLIF_MDS_SIZE_LOCK)) {
- CDEBUG(D_DLMTRACE, "Glimpsing inode " DFID "\n", PFID(fid));
- if (lli->lli_has_smd) {
- struct cl_lock *lock = vvp_env_lock(env);
- struct cl_lock_descr *descr = &lock->cll_descr;
-
- /* NOTE: this looks like DLM lock request, but it may
- * not be one. Due to CEF_ASYNC flag (translated
- * to LDLM_FL_HAS_INTENT by osc), this is
- * glimpse request, that won't revoke any
- * conflicting DLM locks held. Instead,
- * ll_glimpse_callback() will be called on each
- * client holding a DLM lock against this file,
- * and resulting size will be returned for each
- * stripe. DLM lock on [0, EOF] is acquired only
- * if there were no conflicting locks. If there
- * were conflicting locks, enqueuing or waiting
- * fails with -ENAVAIL, but valid inode
- * attributes are returned anyway.
- */
- *descr = whole_file;
- descr->cld_obj = clob;
- descr->cld_mode = CLM_READ;
- descr->cld_enq_flags = CEF_ASYNC | CEF_MUST;
- if (agl)
- descr->cld_enq_flags |= CEF_AGL;
- /*
- * CEF_ASYNC is used because glimpse sub-locks cannot
- * deadlock (because they never conflict with other
- * locks) and, hence, can be enqueued out-of-order.
- *
- * CEF_MUST protects glimpse lock from conversion into
- * a lockless mode.
- */
- result = cl_lock_request(env, io, lock);
- if (result < 0)
- return result;
-
- if (!agl) {
- ll_merge_attr(env, inode);
- if (i_size_read(inode) > 0 &&
- inode->i_blocks == 0) {
- /*
- * LU-417: Add dirty pages block count
- * lest i_blocks reports 0, some "cp" or
- * "tar" may think it's a completely
- * sparse file and skip it.
- */
- inode->i_blocks = dirty_cnt(inode);
- }
- }
- cl_lock_release(env, lock);
- } else {
- CDEBUG(D_DLMTRACE, "No objects for inode\n");
+ CDEBUG(D_DLMTRACE, "Glimpsing inode " DFID "\n", PFID(fid));
+ if (lli->lli_has_smd) {
+ struct cl_lock *lock = vvp_env_lock(env);
+ struct cl_lock_descr *descr = &lock->cll_descr;
+
+ /* NOTE: this looks like DLM lock request, but it may
+ * not be one. Due to CEF_ASYNC flag (translated
+ * to LDLM_FL_HAS_INTENT by osc), this is
+ * glimpse request, that won't revoke any
+ * conflicting DLM locks held. Instead,
+ * ll_glimpse_callback() will be called on each
+ * client holding a DLM lock against this file,
+ * and resulting size will be returned for each
+ * stripe. DLM lock on [0, EOF] is acquired only
+ * if there were no conflicting locks. If there
+ * were conflicting locks, enqueuing or waiting
+ * fails with -ENAVAIL, but valid inode
+ * attributes are returned anyway.
+ */
+ *descr = whole_file;
+ descr->cld_obj = clob;
+ descr->cld_mode = CLM_READ;
+ descr->cld_enq_flags = CEF_ASYNC | CEF_MUST;
+ if (agl)
+ descr->cld_enq_flags |= CEF_AGL;
+ /*
+ * CEF_ASYNC is used because glimpse sub-locks cannot
+ * deadlock (because they never conflict with other
+ * locks) and, hence, can be enqueued out-of-order.
+ *
+ * CEF_MUST protects glimpse lock from conversion into
+ * a lockless mode.
+ */
+ result = cl_lock_request(env, io, lock);
+ if (result < 0)
+ return result;
+
+ if (!agl) {
ll_merge_attr(env, inode);
+ if (i_size_read(inode) > 0 && !inode->i_blocks) {
+ /*
+ * LU-417: Add dirty pages block count
+ * lest i_blocks reports 0, some "cp" or
+ * "tar" may think it's a completely
+ * sparse file and skip it.
+ */
+ inode->i_blocks = dirty_cnt(inode);
+ }
}
+
+ cl_lock_release(env, lock);
+ } else {
+ CDEBUG(D_DLMTRACE, "No objects for inode\n");
+ ll_merge_attr(env, inode);
}
return result;
+++ /dev/null
-/*
- * GPL HEADER START
- *
- * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
- *
- * This program is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License version 2 only,
- * as published by the Free Software Foundation.
- *
- * This program is distributed in the hope that it will be useful, but
- * WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
- * General Public License version 2 for more details (a copy is included
- * in the LICENSE file that accompanied this code).
- *
- * You should have received a copy of the GNU General Public License
- * version 2 along with this program; If not, see
- * http://www.gnu.org/licenses/gpl-2.0.html
- *
- * GPL HEADER END
- */
-/*
- * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
- * Use is subject to license terms.
- *
- * Copyright (c) 2011, 2012, Intel Corporation.
- */
-/*
- * This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
- *
- * lustre/llite/llite_close.c
- *
- * Lustre Lite routines to issue a secondary close after writeback
- */
-
-#include <linux/module.h>
-
-#define DEBUG_SUBSYSTEM S_LLITE
-
-#include "llite_internal.h"
-
-/** records that a write is in flight */
-void vvp_write_pending(struct vvp_object *club, struct vvp_page *page)
-{
- struct ll_inode_info *lli = ll_i2info(club->vob_inode);
-
- spin_lock(&lli->lli_lock);
- lli->lli_flags |= LLIF_SOM_DIRTY;
- if (page && list_empty(&page->vpg_pending_linkage))
- list_add(&page->vpg_pending_linkage, &club->vob_pending_list);
- spin_unlock(&lli->lli_lock);
-}
-
-/** records that a write has completed */
-void vvp_write_complete(struct vvp_object *club, struct vvp_page *page)
-{
- struct ll_inode_info *lli = ll_i2info(club->vob_inode);
- int rc = 0;
-
- spin_lock(&lli->lli_lock);
- if (page && !list_empty(&page->vpg_pending_linkage)) {
- list_del_init(&page->vpg_pending_linkage);
- rc = 1;
- }
- spin_unlock(&lli->lli_lock);
- if (rc)
- ll_queue_done_writing(club->vob_inode, 0);
-}
-
-/** Queues DONE_WRITING if
- * - done writing is allowed;
- * - inode has no no dirty pages;
- */
-void ll_queue_done_writing(struct inode *inode, unsigned long flags)
-{
- struct ll_inode_info *lli = ll_i2info(inode);
- struct vvp_object *club = cl2vvp(ll_i2info(inode)->lli_clob);
-
- spin_lock(&lli->lli_lock);
- lli->lli_flags |= flags;
-
- if ((lli->lli_flags & LLIF_DONE_WRITING) &&
- list_empty(&club->vob_pending_list)) {
- struct ll_close_queue *lcq = ll_i2sbi(inode)->ll_lcq;
-
- if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
- CWARN("%s: file "DFID"(flags %u) Size-on-MDS valid, done writing allowed and no diry pages\n",
- ll_get_fsname(inode->i_sb, NULL, 0),
- PFID(ll_inode2fid(inode)), lli->lli_flags);
- /* DONE_WRITING is allowed and inode has no dirty page. */
- spin_lock(&lcq->lcq_lock);
-
- LASSERT(list_empty(&lli->lli_close_list));
- CDEBUG(D_INODE, "adding inode "DFID" to close list\n",
- PFID(ll_inode2fid(inode)));
- list_add_tail(&lli->lli_close_list, &lcq->lcq_head);
-
- /* Avoid a concurrent insertion into the close thread queue:
- * an inode is already in the close thread, open(), write(),
- * close() happen, epoch is closed as the inode is marked as
- * LLIF_EPOCH_PENDING. When pages are written inode should not
- * be inserted into the queue again, clear this flag to avoid
- * it.
- */
- lli->lli_flags &= ~LLIF_DONE_WRITING;
-
- wake_up(&lcq->lcq_waitq);
- spin_unlock(&lcq->lcq_lock);
- }
- spin_unlock(&lli->lli_lock);
-}
-
-/** Pack SOM attributes info @opdata for CLOSE, DONE_WRITING rpc. */
-void ll_done_writing_attr(struct inode *inode, struct md_op_data *op_data)
-{
- struct ll_inode_info *lli = ll_i2info(inode);
-
- op_data->op_flags |= MF_SOM_CHANGE;
- /* Check if Size-on-MDS attributes are valid. */
- if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
- CERROR("%s: inode "DFID"(flags %u) MDS holds lock on Size-on-MDS attributes\n",
- ll_get_fsname(inode->i_sb, NULL, 0),
- PFID(ll_inode2fid(inode)), lli->lli_flags);
-
- if (!cl_local_size(inode)) {
- /* Send Size-on-MDS Attributes if valid. */
- op_data->op_attr.ia_valid |= ATTR_MTIME_SET | ATTR_CTIME_SET |
- ATTR_ATIME_SET | ATTR_SIZE | ATTR_BLOCKS;
- }
-}
-
-/** Closes ioepoch and packs Size-on-MDS attribute if needed into @op_data. */
-void ll_ioepoch_close(struct inode *inode, struct md_op_data *op_data,
- struct obd_client_handle **och, unsigned long flags)
-{
- struct ll_inode_info *lli = ll_i2info(inode);
- struct vvp_object *club = cl2vvp(ll_i2info(inode)->lli_clob);
-
- spin_lock(&lli->lli_lock);
- if (!(list_empty(&club->vob_pending_list))) {
- if (!(lli->lli_flags & LLIF_EPOCH_PENDING)) {
- LASSERT(*och);
- LASSERT(!lli->lli_pending_och);
- /* Inode is dirty and there is no pending write done
- * request yet, DONE_WRITE is to be sent later.
- */
- lli->lli_flags |= LLIF_EPOCH_PENDING;
- lli->lli_pending_och = *och;
- spin_unlock(&lli->lli_lock);
-
- inode = igrab(inode);
- LASSERT(inode);
- goto out;
- }
- if (flags & LLIF_DONE_WRITING) {
- /* Some pages are still dirty, it is early to send
- * DONE_WRITE. Wait until all pages will be flushed
- * and try DONE_WRITE again later.
- */
- LASSERT(!(lli->lli_flags & LLIF_DONE_WRITING));
- lli->lli_flags |= LLIF_DONE_WRITING;
- spin_unlock(&lli->lli_lock);
-
- inode = igrab(inode);
- LASSERT(inode);
- goto out;
- }
- }
- CDEBUG(D_INODE, "Epoch %llu closed on "DFID"\n",
- ll_i2info(inode)->lli_ioepoch, PFID(&lli->lli_fid));
- op_data->op_flags |= MF_EPOCH_CLOSE;
-
- if (flags & LLIF_DONE_WRITING) {
- LASSERT(lli->lli_flags & LLIF_SOM_DIRTY);
- LASSERT(!(lli->lli_flags & LLIF_DONE_WRITING));
- *och = lli->lli_pending_och;
- lli->lli_pending_och = NULL;
- lli->lli_flags &= ~LLIF_EPOCH_PENDING;
- } else {
- /* Pack Size-on-MDS inode attributes only if they has changed */
- if (!(lli->lli_flags & LLIF_SOM_DIRTY)) {
- spin_unlock(&lli->lli_lock);
- goto out;
- }
-
- /* There is a pending DONE_WRITE -- close epoch with no
- * attribute change.
- */
- if (lli->lli_flags & LLIF_EPOCH_PENDING) {
- spin_unlock(&lli->lli_lock);
- goto out;
- }
- }
-
- LASSERT(list_empty(&club->vob_pending_list));
- lli->lli_flags &= ~LLIF_SOM_DIRTY;
- spin_unlock(&lli->lli_lock);
- ll_done_writing_attr(inode, op_data);
-
-out:
- return;
-}
-
-/**
- * Cliens updates SOM attributes on MDS (including llog cookies):
- * obd_getattr with no lock and md_setattr.
- */
-int ll_som_update(struct inode *inode, struct md_op_data *op_data)
-{
- struct ll_inode_info *lli = ll_i2info(inode);
- struct ptlrpc_request *request = NULL;
- __u32 old_flags;
- struct obdo *oa;
- int rc;
-
- LASSERT(op_data);
- if (lli->lli_flags & LLIF_MDS_SIZE_LOCK)
- CERROR("%s: inode "DFID"(flags %u) MDS holds lock on Size-on-MDS attributes\n",
- ll_get_fsname(inode->i_sb, NULL, 0),
- PFID(ll_inode2fid(inode)), lli->lli_flags);
-
- oa = kmem_cache_zalloc(obdo_cachep, GFP_NOFS);
- if (!oa) {
- CERROR("can't allocate memory for Size-on-MDS update.\n");
- return -ENOMEM;
- }
-
- old_flags = op_data->op_flags;
- op_data->op_flags = MF_SOM_CHANGE;
-
- /* If inode is already in another epoch, skip getattr from OSTs. */
- if (lli->lli_ioepoch == op_data->op_ioepoch) {
- rc = ll_inode_getattr(inode, oa, op_data->op_ioepoch,
- old_flags & MF_GETATTR_LOCK);
- if (rc) {
- oa->o_valid = 0;
- if (rc != -ENOENT)
- CERROR("%s: inode_getattr failed - unable to send a Size-on-MDS attribute update for inode "DFID": rc = %d\n",
- ll_get_fsname(inode->i_sb, NULL, 0),
- PFID(ll_inode2fid(inode)), rc);
- } else {
- CDEBUG(D_INODE, "Size-on-MDS update on "DFID"\n",
- PFID(&lli->lli_fid));
- }
- /* Install attributes into op_data. */
- md_from_obdo(op_data, oa, oa->o_valid);
- }
-
- rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data,
- NULL, 0, NULL, 0, &request, NULL);
- ptlrpc_req_finished(request);
-
- kmem_cache_free(obdo_cachep, oa);
- return rc;
-}
-
-/**
- * Closes the ioepoch and packs all the attributes into @op_data for
- * DONE_WRITING rpc.
- */
-static void ll_prepare_done_writing(struct inode *inode,
- struct md_op_data *op_data,
- struct obd_client_handle **och)
-{
- ll_ioepoch_close(inode, op_data, och, LLIF_DONE_WRITING);
- /* If there is no @och, we do not do D_W yet. */
- if (!*och)
- return;
-
- ll_pack_inode2opdata(inode, op_data, &(*och)->och_fh);
- ll_prep_md_op_data(op_data, inode, NULL, NULL,
- 0, 0, LUSTRE_OPC_ANY, NULL);
-}
-
-/** Send a DONE_WRITING rpc. */
-static void ll_done_writing(struct inode *inode)
-{
- struct obd_client_handle *och = NULL;
- struct md_op_data *op_data;
- int rc;
-
- LASSERT(exp_connect_som(ll_i2mdexp(inode)));
-
- op_data = kzalloc(sizeof(*op_data), GFP_NOFS);
- if (!op_data)
- return;
-
- ll_prepare_done_writing(inode, op_data, &och);
- /* If there is no @och, we do not do D_W yet. */
- if (!och)
- goto out;
-
- rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, NULL);
- if (rc == -EAGAIN)
- /* MDS has instructed us to obtain Size-on-MDS attribute from
- * OSTs and send setattr to back to MDS.
- */
- rc = ll_som_update(inode, op_data);
- else if (rc) {
- CERROR("%s: inode "DFID" mdc done_writing failed: rc = %d\n",
- ll_get_fsname(inode->i_sb, NULL, 0),
- PFID(ll_inode2fid(inode)), rc);
- }
-out:
- ll_finish_md_op_data(op_data);
- if (och) {
- md_clear_open_replay_data(ll_i2sbi(inode)->ll_md_exp, och);
- kfree(och);
- }
-}
-
-static struct ll_inode_info *ll_close_next_lli(struct ll_close_queue *lcq)
-{
- struct ll_inode_info *lli = NULL;
-
- spin_lock(&lcq->lcq_lock);
-
- if (!list_empty(&lcq->lcq_head)) {
- lli = list_entry(lcq->lcq_head.next, struct ll_inode_info,
- lli_close_list);
- list_del_init(&lli->lli_close_list);
- } else if (atomic_read(&lcq->lcq_stop)) {
- lli = ERR_PTR(-EALREADY);
- }
-
- spin_unlock(&lcq->lcq_lock);
- return lli;
-}
-
-static int ll_close_thread(void *arg)
-{
- struct ll_close_queue *lcq = arg;
-
- complete(&lcq->lcq_comp);
-
- while (1) {
- struct l_wait_info lwi = { 0 };
- struct ll_inode_info *lli;
- struct inode *inode;
-
- l_wait_event_exclusive(lcq->lcq_waitq,
- (lli = ll_close_next_lli(lcq)) != NULL,
- &lwi);
- if (IS_ERR(lli))
- break;
-
- inode = ll_info2i(lli);
- CDEBUG(D_INFO, "done_writing for inode "DFID"\n",
- PFID(ll_inode2fid(inode)));
- ll_done_writing(inode);
- iput(inode);
- }
-
- CDEBUG(D_INFO, "ll_close exiting\n");
- complete(&lcq->lcq_comp);
- return 0;
-}
-
-int ll_close_thread_start(struct ll_close_queue **lcq_ret)
-{
- struct ll_close_queue *lcq;
- struct task_struct *task;
-
- if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_CLOSE_THREAD))
- return -EINTR;
-
- lcq = kzalloc(sizeof(*lcq), GFP_NOFS);
- if (!lcq)
- return -ENOMEM;
-
- spin_lock_init(&lcq->lcq_lock);
- INIT_LIST_HEAD(&lcq->lcq_head);
- init_waitqueue_head(&lcq->lcq_waitq);
- init_completion(&lcq->lcq_comp);
-
- task = kthread_run(ll_close_thread, lcq, "ll_close");
- if (IS_ERR(task)) {
- kfree(lcq);
- return PTR_ERR(task);
- }
-
- wait_for_completion(&lcq->lcq_comp);
- *lcq_ret = lcq;
- return 0;
-}
-
-void ll_close_thread_shutdown(struct ll_close_queue *lcq)
-{
- init_completion(&lcq->lcq_comp);
- atomic_inc(&lcq->lcq_stop);
- wake_up(&lcq->lcq_waitq);
- wait_for_completion(&lcq->lcq_comp);
- kfree(lcq);
-}
};
enum lli_flags {
- /* MDS has an authority for the Size-on-MDS attributes. */
- LLIF_MDS_SIZE_LOCK = (1 << 0),
- /* Epoch close is postponed. */
- LLIF_EPOCH_PENDING = (1 << 1),
- /* DONE WRITING is allowed. */
- LLIF_DONE_WRITING = (1 << 2),
- /* Sizeon-on-MDS attributes are changed. An attribute update needs to
- * be sent to MDS.
- */
- LLIF_SOM_DIRTY = (1 << 3),
/* File data is modified. */
- LLIF_DATA_MODIFIED = (1 << 4),
+ LLIF_DATA_MODIFIED = BIT(0),
/* File is being restored */
- LLIF_FILE_RESTORING = (1 << 5),
+ LLIF_FILE_RESTORING = BIT(1),
/* Xattr cache is attached to the file */
- LLIF_XATTR_CACHE = (1 << 6),
+ LLIF_XATTR_CACHE = BIT(2),
};
struct ll_inode_info {
__u32 lli_inode_magic;
__u32 lli_flags;
- __u64 lli_ioepoch;
spinlock_t lli_lock;
struct posix_acl *lli_posix_acl;
/* master inode fid for stripe directory */
struct lu_fid lli_pfid;
- struct list_head lli_close_list;
-
- /* handle is to be sent to MDS later on done_writing and setattr.
- * Open handle data are needed for the recovery to reconstruct
- * the inode state on the MDS. XXX: recovery is not ready yet.
- */
- struct obd_client_handle *lli_pending_och;
-
/* We need all three because every inode may be opened in different
* modes
*/
#define LL_SBI_LOCALFLOCK 0x200 /* Local flocks support by kernel */
#define LL_SBI_LRU_RESIZE 0x400 /* lru resize support */
#define LL_SBI_LAZYSTATFS 0x800 /* lazystatfs mount option */
-#define LL_SBI_SOM_PREVIEW 0x1000 /* SOM preview mount option */
+/* LL_SBI_SOM_PREVIEW 0x1000 SOM preview mount option, obsolete */
#define LL_SBI_32BIT_API 0x2000 /* generate 32 bit inodes. */
#define LL_SBI_64BIT_HASH 0x4000 /* support 64-bits dir hash/offset */
#define LL_SBI_AGL_ENABLED 0x8000 /* enable agl */
int ll_flags;
unsigned int ll_umounting:1,
- ll_xattr_cache_enabled:1;
- struct lustre_client_ocd ll_lco;
+ ll_xattr_cache_enabled:1,
+ ll_client_common_fill_super_succeeded:1;
- struct ll_close_queue *ll_lcq;
+ struct lustre_client_ocd ll_lco;
struct lprocfs_stats *ll_stats; /* lprocfs stats counter */
int ll_file_release(struct inode *inode, struct file *file);
int ll_glimpse_ioctl(struct ll_sb_info *sbi,
struct lov_stripe_md *lsm, lstat_t *st);
-void ll_ioepoch_open(struct ll_inode_info *lli, __u64 ioepoch);
int ll_release_openhandle(struct inode *, struct lookup_intent *);
int ll_md_real_close(struct inode *inode, fmode_t fmode);
-void ll_ioepoch_close(struct inode *inode, struct md_op_data *op_data,
- struct obd_client_handle **och, unsigned long flags);
-void ll_done_writing_attr(struct inode *inode, struct md_op_data *op_data);
-int ll_som_update(struct inode *inode, struct md_op_data *op_data);
-int ll_inode_getattr(struct inode *inode, struct obdo *obdo,
- __u64 ioepoch, int sync);
void ll_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data,
struct lustre_handle *fh);
int ll_getattr(struct vfsmount *mnt, struct dentry *de, struct kstat *stat);
/* llite/symlink.c */
extern const struct inode_operations ll_fast_symlink_inode_operations;
-/* llite/llite_close.c */
-struct ll_close_queue {
- spinlock_t lcq_lock;
- struct list_head lcq_head;
- wait_queue_head_t lcq_waitq;
- struct completion lcq_comp;
- atomic_t lcq_stop;
-};
-
-void vvp_write_pending(struct vvp_object *club, struct vvp_page *page);
-void vvp_write_complete(struct vvp_object *club, struct vvp_page *page);
-
/**
* IO arguments for various VFS I/O interfaces.
*/
return &ll_env_info(env)->lti_args;
}
-void ll_queue_done_writing(struct inode *inode, unsigned long flags);
-void ll_close_thread_shutdown(struct ll_close_queue *lcq);
-int ll_close_thread_start(struct ll_close_queue **lcq_ret);
-
/* llite/llite_mmap.c */
int ll_teardown_mmaps(struct address_space *mapping, __u64 first, __u64 last);
OBD_CONNECT_OPEN_BY_FID |
OBD_CONNECT_DIR_STRIPE;
- if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
- data->ocd_connect_flags |= OBD_CONNECT_SOM;
-
if (sbi->ll_flags & LL_SBI_LRU_RESIZE)
data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
#ifdef CONFIG_FS_POSIX_ACL
OBD_CONNECT_JOBSTATS | OBD_CONNECT_LVB_TYPE |
OBD_CONNECT_LAYOUTLOCK | OBD_CONNECT_PINGLESS;
- if (sbi->ll_flags & LL_SBI_SOM_PREVIEW)
- data->ocd_connect_flags |= OBD_CONNECT_SOM;
-
if (!OBD_FAIL_CHECK(OBD_FAIL_OSC_CONNECT_CKSUM)) {
/* OBD_CONNECT_CKSUM should always be set, even if checksums are
* disabled by default, because it can still be enabled on the
goto out_root;
}
- err = ll_close_thread_start(&sbi->ll_lcq);
- if (err) {
- CERROR("cannot start close thread: rc %d\n", err);
- goto out_root;
- }
-
checksum = sbi->ll_flags & LL_SBI_CHECKSUM;
err = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
KEY_CHECKSUM, sizeof(checksum), &checksum,
{
struct ll_sb_info *sbi = ll_s2sbi(sb);
- ll_close_thread_shutdown(sbi->ll_lcq);
-
cl_sb_fini(sb);
obd_fid_fini(sbi->ll_dt_exp->exp_obd);
*flags &= ~tmp;
goto next;
}
- tmp = ll_set_opt("som_preview", s1, LL_SBI_SOM_PREVIEW);
- if (tmp) {
- *flags |= tmp;
- goto next;
- }
tmp = ll_set_opt("32bitapi", s1, LL_SBI_32BIT_API);
if (tmp) {
*flags |= tmp;
{
lli->lli_inode_magic = LLI_INODE_MAGIC;
lli->lli_flags = 0;
- lli->lli_ioepoch = 0;
lli->lli_maxbytes = MAX_LFS_FILESIZE;
spin_lock_init(&lli->lli_lock);
lli->lli_posix_acl = NULL;
/* Do not set lli_fid, it has been initialized already. */
fid_zero(&lli->lli_pfid);
- INIT_LIST_HEAD(&lli->lli_close_list);
- lli->lli_pending_och = NULL;
lli->lli_mds_read_och = NULL;
lli->lli_mds_write_och = NULL;
lli->lli_mds_exec_och = NULL;
/* connections, registrations, sb setup */
err = client_common_fill_super(sb, md, dt, mnt);
+ if (!err)
+ sbi->ll_client_common_fill_super_succeeded = 1;
out_free:
kfree(md);
}
}
- if (sbi->ll_lcq) {
+ if (sbi->ll_client_common_fill_super_succeeded) {
/* Only if client_common_fill_super succeeded */
client_common_put_super(sb);
}
LASSERT(lli->lli_opendir_pid == 0);
}
- spin_lock(&lli->lli_lock);
- ll_i2info(inode)->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
- spin_unlock(&lli->lli_lock);
md_null_inode(sbi->ll_md_exp, ll_inode2fid(inode));
LASSERT(!lli->lli_open_fd_write_count);
rc = simple_setattr(dentry, &op_data->op_attr);
op_data->op_attr.ia_valid = ia_valid;
- /* Extract epoch data if obtained. */
- op_data->op_handle = md.body->mbo_handle;
- op_data->op_ioepoch = md.body->mbo_ioepoch;
-
rc = ll_update_inode(inode, &md);
ptlrpc_req_finished(request);
return rc;
}
-/* Close IO epoch and send Size-on-MDS attribute update. */
-static int ll_setattr_done_writing(struct inode *inode,
- struct md_op_data *op_data,
- struct md_open_data *mod)
-{
- struct ll_inode_info *lli = ll_i2info(inode);
- int rc = 0;
-
- if (!S_ISREG(inode->i_mode))
- return 0;
-
- CDEBUG(D_INODE, "Epoch %llu closed on "DFID" for truncate\n",
- op_data->op_ioepoch, PFID(&lli->lli_fid));
-
- op_data->op_flags = MF_EPOCH_CLOSE;
- ll_done_writing_attr(inode, op_data);
- ll_pack_inode2opdata(inode, op_data, NULL);
-
- rc = md_done_writing(ll_i2sbi(inode)->ll_md_exp, op_data, mod);
- if (rc == -EAGAIN)
- /* MDS has instructed us to obtain Size-on-MDS attribute
- * from OSTs and send setattr to back to MDS.
- */
- rc = ll_som_update(inode, op_data);
- else if (rc) {
- CERROR("%s: inode "DFID" mdc truncate failed: rc = %d\n",
- ll_i2sbi(inode)->ll_md_exp->exp_obd->obd_name,
- PFID(ll_inode2fid(inode)), rc);
- }
- return rc;
-}
-
/* If this inode has objects allocated to it (lsm != NULL), then the OST
* object(s) determine the file size and mtime. Otherwise, the MDS will
* keep these values until such a time that objects are allocated for it.
struct md_op_data *op_data = NULL;
struct md_open_data *mod = NULL;
bool file_is_released = false;
- int rc = 0, rc1 = 0;
+ int rc = 0;
CDEBUG(D_VFSTRACE, "%s: setattr inode "DFID"(%p) from %llu to %llu, valid %x, hsm_import %d\n",
ll_get_fsname(inode->i_sb, NULL, 0), PFID(&lli->lli_fid), inode,
memcpy(&op_data->op_attr, attr, sizeof(*attr));
- /* Open epoch for truncate. */
- if (exp_connect_som(ll_i2mdexp(inode)) && !hsm_import &&
- (attr->ia_valid & (ATTR_SIZE | ATTR_MTIME | ATTR_MTIME_SET)))
- op_data->op_flags = MF_EPOCH_OPEN;
-
rc = ll_md_setattr(dentry, op_data, &mod);
if (rc)
goto out;
spin_unlock(&lli->lli_lock);
}
- ll_ioepoch_open(lli, op_data->op_ioepoch);
if (!S_ISREG(inode->i_mode) || file_is_released) {
rc = 0;
goto out;
up_write(&lli->lli_trunc_sem);
}
out:
- if (op_data->op_ioepoch) {
- rc1 = ll_setattr_done_writing(inode, op_data, mod);
- if (!rc)
- rc = rc1;
- }
- ll_finish_md_op_data(op_data);
+ if (op_data)
+ ll_finish_md_op_data(op_data);
if (!S_ISDIR(inode->i_mode)) {
inode_lock(inode);
LASSERT(fid_seq(&lli->lli_fid) != 0);
if (body->mbo_valid & OBD_MD_FLSIZE) {
- if (exp_connect_som(ll_i2mdexp(inode)) &&
- S_ISREG(inode->i_mode)) {
- struct lustre_handle lockh;
- enum ldlm_mode mode;
-
- /* As it is possible a blocking ast has been processed
- * by this time, we need to check there is an UPDATE
- * lock on the client and set LLIF_MDS_SIZE_LOCK holding
- * it.
- */
- mode = ll_take_md_lock(inode, MDS_INODELOCK_UPDATE,
- &lockh, LDLM_FL_CBPENDING,
- LCK_CR | LCK_CW |
- LCK_PR | LCK_PW);
- if (mode) {
- if (lli->lli_flags & (LLIF_DONE_WRITING |
- LLIF_EPOCH_PENDING |
- LLIF_SOM_DIRTY)) {
- CERROR("%s: inode "DFID" flags %u still has size authority! do not trust the size got from MDS\n",
- sbi->ll_md_exp->exp_obd->obd_name,
- PFID(ll_inode2fid(inode)),
- lli->lli_flags);
- } else {
- /* Use old size assignment to avoid
- * deadlock bz14138 & bz14326
- */
- i_size_write(inode, body->mbo_size);
- spin_lock(&lli->lli_lock);
- lli->lli_flags |= LLIF_MDS_SIZE_LOCK;
- spin_unlock(&lli->lli_lock);
- }
- ldlm_lock_decref(&lockh, mode);
- }
- } else {
- /* Use old size assignment to avoid
- * deadlock bz14138 & bz14326
- */
- i_size_write(inode, body->mbo_size);
+ i_size_write(inode, body->mbo_size);
- CDEBUG(D_VFSTRACE, "inode=%lu, updating i_size %llu\n",
- inode->i_ino, (unsigned long long)body->mbo_size);
- }
+ CDEBUG(D_VFSTRACE, "inode=" DFID ", updating i_size %llu\n",
+ PFID(ll_inode2fid(inode)),
+ (unsigned long long)body->mbo_size);
if (body->mbo_valid & OBD_MD_FLBLOCKS)
inode->i_blocks = body->mbo_blocks;
return;
op_data->op_fid1 = body->mbo_fid1;
- op_data->op_ioepoch = body->mbo_ioepoch;
op_data->op_handle = body->mbo_handle;
op_data->op_mod_time = get_seconds();
md_close(exp, op_data, NULL, &close_req);
PFID(ll_inode2fid(inode)), rc);
}
- if (bits & MDS_INODELOCK_UPDATE) {
- struct ll_inode_info *lli = ll_i2info(inode);
-
- spin_lock(&lli->lli_lock);
- lli->lli_flags &= ~LLIF_MDS_SIZE_LOCK;
- spin_unlock(&lli->lli_lock);
- }
-
if ((bits & MDS_INODELOCK_UPDATE) && S_ISDIR(inode->i_mode)) {
struct ll_inode_info *lli = ll_i2info(inode);
vpg = cl2vvp_page(cl_page_at(page, &vvp_device_type));
vmpage = vpg->vpg_page;
- seq_printf(seq, " %5i | %p %p %s %s %s %s | %p "DFID"(%p) %lu %u [",
+ seq_printf(seq, " %5i | %p %p %s %s %s | %p " DFID "(%p) %lu %u [",
0 /* gen */,
vpg, page,
"none",
- vpg->vpg_write_queued ? "wq" : "- ",
vpg->vpg_defer_uptodate ? "du" : "- ",
PageWriteback(vmpage) ? "wb" : "-",
vmpage, PFID(ll_inode2fid(vmpage->mapping->host)),
struct cl_object vob_cl;
struct inode *vob_inode;
- /**
- * A list of dirty pages pending IO in the cache. Used by
- * SOM. Protected by ll_inode_info::lli_lock.
- *
- * \see vvp_page::vpg_pending_linkage
- */
- struct list_head vob_pending_list;
-
/**
* Number of transient pages. This is no longer protected by i_sem,
* and needs to be atomic. This is not actually used for anything,
struct vvp_page {
struct cl_page_slice vpg_cl;
unsigned int vpg_defer_uptodate:1,
- vpg_ra_used:1,
- vpg_write_queued:1;
- /**
- * Non-empty iff this page is already counted in
- * vvp_object::vob_pending_list. This list is only used as a flag,
- * that is, never iterated through, only checked for list_empty(), but
- * having a list is useful for debugging.
- */
- struct list_head vpg_pending_linkage;
+ vpg_ra_used:1;
/** VM page */
struct page *vpg_page;
};
static void write_commit_callback(const struct lu_env *env, struct cl_io *io,
struct cl_page *page)
{
- struct vvp_page *vpg;
struct page *vmpage = page->cp_vmpage;
- struct cl_object *clob = cl_io_top(io)->ci_obj;
SetPageUptodate(vmpage);
set_page_dirty(vmpage);
- vpg = cl2vvp_page(cl_object_page_slice(clob, page));
- vvp_write_pending(cl2vvp(clob), vpg);
-
cl_page_disown(env, io, page);
/* held in ll_cl_init() */
static void mkwrite_commit_callback(const struct lu_env *env, struct cl_io *io,
struct cl_page *page)
{
- struct vvp_page *vpg;
- struct cl_object *clob = cl_io_top(io)->ci_obj;
-
set_page_dirty(page->cp_vmpage);
-
- vpg = cl2vvp_page(cl_object_page_slice(clob, page));
- vvp_write_pending(cl2vvp(clob), vpg);
}
static int vvp_io_fault_start(const struct lu_env *env,
struct inode *inode = obj->vob_inode;
struct ll_inode_info *lli;
- (*p)(env, cookie, "(%s %d %d) inode: %p ",
- list_empty(&obj->vob_pending_list) ? "-" : "+",
+ (*p)(env, cookie, "(%d %d) inode: %p ",
atomic_read(&obj->vob_transient_pages),
atomic_read(&obj->vob_mmap_cnt), inode);
if (inode) {
const struct cl_object_conf *cconf;
cconf = lu2cl_conf(conf);
- INIT_LIST_HEAD(&vob->vob_pending_list);
lu_object_add(obj, below);
result = vvp_object_init0(env, vob, cconf);
} else {
LASSERT((struct cl_page *)vmpage->private == page);
LASSERT(inode == vvp_object_inode(obj));
- vvp_write_complete(cl2vvp(obj), cl2vvp_page(slice));
-
/* Drop the reference count held in vvp_page_init */
refc = atomic_dec_return(&page->cp_ref);
LASSERTF(refc >= 1, "page = %p, refc = %d\n", page, refc);
if (!pg->cp_sync_io)
set_page_writeback(vmpage);
- vvp_write_pending(cl2vvp(slice->cpl_obj), cl2vvp_page(slice));
-
return 0;
}
CL_PAGE_HEADER(D_PAGE, env, pg, "completing WRITE with %d\n", ioret);
- /*
- * TODO: Actually it makes sense to add the page into oap pending
- * list again and so that we don't need to take the page out from
- * SoM write pending list, if we just meet a recoverable error,
- * -ENOMEM, etc.
- * To implement this, we just need to return a non zero value in
- * ->cpo_completion method. The underlying transfer should be notified
- * and then re-add the page into pending transfer queue. -jay
- */
-
- vpg->vpg_write_queued = 0;
- vvp_write_complete(cl2vvp(slice->cpl_obj), vpg);
-
if (pg->cp_sync_io) {
LASSERT(PageLocked(vmpage));
LASSERT(!PageWriteback(vmpage));
LASSERT(pg->cp_state == CPS_CACHED);
/* This actually clears the dirty bit in the radix tree. */
set_page_writeback(vmpage);
- vvp_write_pending(cl2vvp(slice->cpl_obj), cl2vvp_page(slice));
CL_PAGE_HEADER(D_PAGE, env, pg, "readied\n");
} else if (pg->cp_state == CPS_PAGEOUT) {
/* is it possible for osc_flush_async_page() to already
struct vvp_page *vpg = cl2vvp_page(slice);
struct page *vmpage = vpg->vpg_page;
- (*printer)(env, cookie, LUSTRE_VVP_NAME "-page@%p(%d:%d:%d) vm@%p ",
- vpg, vpg->vpg_defer_uptodate, vpg->vpg_ra_used,
- vpg->vpg_write_queued, vmpage);
+ (*printer)(env, cookie, LUSTRE_VVP_NAME "-page@%p(%d:%d) vm@%p ",
+ vpg, vpg->vpg_defer_uptodate, vpg->vpg_ra_used, vmpage);
if (vmpage) {
(*printer)(env, cookie, "%lx %d:%d %lx %lu %slru",
(long)vmpage->flags, page_count(vmpage),
vpg->vpg_page = vmpage;
get_page(vmpage);
- INIT_LIST_HEAD(&vpg->vpg_pending_linkage);
if (page->cp_type == CPT_CACHEABLE) {
/* in cache, decref in vvp_page_delete */
atomic_inc(&page->cp_ref);
*
* - o_parent_ver
*
- * - o_ioepoch,
- *
*/
static void vvp_req_attr_set(const struct lu_env *env,
const struct cl_req_slice *slice,
inode = vvp_object_inode(obj);
valid_flags = OBD_MD_FLTYPE;
- if (slice->crs_req->crq_type == CRT_WRITE) {
- if (flags & OBD_MD_FLEPOCH) {
- oa->o_valid |= OBD_MD_FLEPOCH;
- oa->o_ioepoch = ll_i2info(inode)->lli_ioepoch;
- valid_flags |= OBD_MD_FLMTIME | OBD_MD_FLCTIME |
- OBD_MD_FLUID | OBD_MD_FLGID;
- }
- }
+ if (slice->crs_req->crq_type == CRT_WRITE)
+ valid_flags |= OBD_MD_FLMTIME | OBD_MD_FLCTIME |
+ OBD_MD_FLUID | OBD_MD_FLGID;
obdo_from_inode(oa, inode, valid_flags & flags);
obdo_set_parent_fid(oa, &ll_i2info(inode)->lli_fid);
if (OBD_FAIL_CHECK(OBD_FAIL_LFSCK_INVALID_PFID))
projects / GitHub / moto-9609 / android_kernel_motorola_exynos9610.git / commitdiff