/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
* vim:expandtab:shiftwidth=8:tabstop=8:
*
- * Lustre Light file operations
+ * 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.
*
- * Copyright (c) 2002-2004 Cluster File Systems, Inc.
+ * 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).
*
- * This file is part of Lustre, http://www.lustre.org.
+ * You should have received a copy of the GNU General Public License
+ * version 2 along with this program; If not, see
+ * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
*
- * Lustre is free software; you can redistribute it and/or
- * modify it under the terms of version 2 of the GNU General Public
- * License as published by the Free Software Foundation.
+ * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
+ * CA 95054 USA or visit www.sun.com if you need additional information or
+ * have any questions.
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Use is subject to license terms.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
*
- * Lustre 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 for more details.
+ * lustre/liblustre/file.c
*
- * You should have received a copy of the GNU General Public License
- * along with Lustre; if not, write to the Free Software
- * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ * Lustre Light file operations
*/
#define DEBUG_SUBSYSTEM S_LLITE
#include <sys/queue.h>
#include <fcntl.h>
-#ifdef HAVE_XTIO_H
-#include <xtio.h>
-#endif
-#include <sysio.h>
-#include <fs.h>
-#include <mount.h>
-#include <inode.h>
-#ifdef HAVE_FILE_H
-#include <file.h>
-#endif
-
-#undef LIST_HEAD
-
#include "llite_lib.h"
/* Pack the required supplementary groups into the supplied groups array.
LASSERT(i1 != NULL);
LASSERT(suppgids != NULL);
- if (in_group_p(i1->i_stbuf.st_gid))
+ if (cfs_curproc_is_in_groups(i1->i_stbuf.st_gid))
suppgids[0] = i1->i_stbuf.st_gid;
else
suppgids[0] = -1;
if (i2) {
- if (in_group_p(i2->i_stbuf.st_gid))
+ if (cfs_curproc_is_in_groups(i2->i_stbuf.st_gid))
suppgids[1] = i2->i_stbuf.st_gid;
else
suppgids[1] = -1;
{
LASSERT(i1 != NULL || i2 != NULL);
LASSERT(op_data);
- memset(op_data, 0, sizeof(*op_data));
if (i1) {
ll_i2gids(op_data->op_suppgids, i1, i2);
op_data->op_name = name;
op_data->op_mode = mode;
op_data->op_namelen = namelen;
- op_data->op_mod_time = CURRENT_TIME;
+ op_data->op_mod_time = CFS_CURRENT_TIME;
+ op_data->op_data = NULL;
}
void llu_finish_md_op_data(struct md_op_data *op_data)
valid &= src->o_valid;
if (valid & (OBD_MD_FLCTIME | OBD_MD_FLMTIME))
- CDEBUG(D_INODE,"valid "LPX64", cur time %lu/%lu, new %lu/%lu\n",
+ CDEBUG(D_INODE,"valid "LPX64", cur time "CFS_TIME_T"/"CFS_TIME_T
+ ", new %lu/%lu\n",
src->o_valid, LTIME_S(st->st_mtime),
LTIME_S(st->st_ctime),
(long)src->o_mtime, (long)src->o_ctime);
if (valid & OBD_MD_FLATIME && src->o_atime > LTIME_S(st->st_atime))
LTIME_S(st->st_atime) = src->o_atime;
-
- /* mtime is always updated with ctime, but can be set in past.
- As write and utime(2) may happen within 1 second, and utime's
- mtime has a priority over write's one, leave mtime from mds
- for the same ctimes. */
- if (valid & OBD_MD_FLCTIME && src->o_ctime > LTIME_S(st->st_ctime)) {
+ if (valid & OBD_MD_FLMTIME && src->o_mtime > LTIME_S(st->st_mtime))
+ LTIME_S(st->st_mtime) = src->o_mtime;
+ if (valid & OBD_MD_FLCTIME && src->o_ctime > LTIME_S(st->st_ctime))
LTIME_S(st->st_ctime) = src->o_ctime;
- if (valid & OBD_MD_FLMTIME)
- LTIME_S(st->st_mtime) = src->o_mtime;
- }
if (valid & OBD_MD_FLSIZE && src->o_size > st->st_size)
st->st_size = src->o_size;
/* optimum IO size */
st->st_blocks = src->o_blocks;
}
+/**
+ * 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 llu_ioepoch_open(struct llu_inode_info *lli, __u64 ioepoch)
+{
+ if (ioepoch && lli->lli_ioepoch != ioepoch) {
+ lli->lli_ioepoch = ioepoch;
+ CDEBUG(D_INODE, "Epoch "LPU64" opened on "DFID" for truncate\n",
+ ioepoch, PFID(&lli->lli_fid));
+ }
+}
+
int llu_local_open(struct llu_inode_info *lli, struct lookup_intent *it)
{
struct ptlrpc_request *req = it->d.lustre.it_data;
struct mdt_body *body;
ENTRY;
- body = lustre_msg_buf(req->rq_repmsg, DLM_REPLY_REC_OFF, sizeof(*body));
- LASSERT(body != NULL); /* reply already checked out */
- LASSERT_REPSWABBED(req, DLM_REPLY_REC_OFF); /* and swabbed down */
+ body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
+ LASSERT(body != NULL);
/* already opened? */
if (lli->lli_open_count++)
fd->fd_mds_och.och_magic = OBD_CLIENT_HANDLE_MAGIC;
fd->fd_mds_och.och_fid = lli->lli_fid;
lli->lli_file_data = fd;
-
+ llu_ioepoch_open(lli, body->ioepoch);
md_set_open_replay_data(lli->lli_sbi->ll_md_exp,
&fd->fd_mds_och, it->d.lustre.it_data);
fd = lli->lli_file_data;
lsm = lli->lli_smd;
- if (lsm == NULL) {
- if (fd->fd_flags & O_LOV_DELAY_CREATE) {
- CDEBUG(D_INODE, "object creation was delayed\n");
- GOTO(out_release, rc);
- }
- }
- fd->fd_flags &= ~O_LOV_DELAY_CREATE;
-
+ if (lsm)
+ flags &= ~O_LOV_DELAY_CREATE;
+ /*XXX: open_flags are overwritten and the previous ones are lost */
lli->lli_open_flags = flags & ~(O_CREAT | O_EXCL | O_TRUNC);
out_release:
RETURN(rc);
}
-int llu_objects_destroy(struct ptlrpc_request *request, struct inode *dir)
+int llu_objects_destroy(struct ptlrpc_request *req, struct inode *dir)
{
struct mdt_body *body;
struct lov_mds_md *eadata;
int rc;
ENTRY;
- /* req is swabbed so this is safe */
- body = lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF, sizeof(*body));
+ body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
if (!(body->valid & OBD_MD_FLEASIZE))
RETURN(0);
* to this file. Use this EA to unlink the objects on the OST.
* It's opaque so we don't swab here; we leave it to obd_unpackmd() to
* check it is complete and sensible. */
- eadata = lustre_swab_repbuf(request, REPLY_REC_OFF+1, body->eadatasize,
- NULL);
+ eadata = req_capsule_server_sized_get(&req->rq_pill, &RMF_MDT_MD,
+ body->eadatasize);
+
LASSERT(eadata != NULL);
- if (eadata == NULL) {
- CERROR("Can't unpack MDS EA data\n");
- GOTO(out, rc = -EPROTO);
- }
rc = obd_unpackmd(llu_i2obdexp(dir), &lsm, eadata,body->eadatasize);
if (rc < 0) {
GOTO(out_free_memmd, rc = -ENOMEM);
oa->o_id = lsm->lsm_object_id;
- oa->o_gr = lsm->lsm_object_gr;
+ oa->o_seq = lsm->lsm_object_seq;
oa->o_mode = body->mode & S_IFMT;
oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
+ obdo_from_inode(oa, NULL, &llu_i2info(dir)->lli_fid, 0);
if (body->valid & OBD_MD_FLCOOKIE) {
oa->o_valid |= OBD_MD_FLCOOKIE;
oti.oti_logcookies =
- lustre_msg_buf(request->rq_repmsg, REPLY_REC_OFF + 2,
- sizeof(struct llog_cookie) *
- lsm->lsm_stripe_count);
+ req_capsule_server_sized_get(&req->rq_pill,
+ &RMF_LOGCOOKIES,
+ sizeof(struct llog_cookie) *
+ lsm->lsm_stripe_count);
if (oti.oti_logcookies == NULL) {
oa->o_valid &= ~OBD_MD_FLCOOKIE;
body->valid &= ~OBD_MD_FLCOOKIE;
}
}
- rc = obd_destroy(llu_i2obdexp(dir), oa, lsm, &oti, NULL);
+ rc = obd_destroy(llu_i2obdexp(dir), oa, lsm, &oti, NULL, NULL);
OBDO_FREE(oa);
if (rc)
CERROR("obd destroy objid 0x"LPX64" error %d\n",
return rc;
}
-int llu_sizeonmds_update(struct inode *inode, struct lustre_handle *fh,
- __u64 ioepoch)
+/** Cliens updates SOM attributes on MDS: obd_getattr and md_setattr. */
+int llu_som_update(struct inode *inode, struct md_op_data *op_data)
{
struct llu_inode_info *lli = llu_i2info(inode);
struct llu_sb_info *sbi = llu_i2sbi(inode);
- struct md_op_data op_data;
- struct obdo oa;
+ struct obdo oa = { 0 };
+ __u32 old_flags;
int rc;
ENTRY;
-
+
LASSERT(!(lli->lli_flags & LLIF_MDS_SIZE_LOCK));
LASSERT(sbi->ll_lco.lco_flags & OBD_CONNECT_SOM);
-
- rc = llu_inode_getattr(inode, &oa);
- if (rc) {
- CERROR("inode_getattr failed (%d): unable to send a "
- "Size-on-MDS attribute update for inode %llu/%lu\n",
- rc, (long long)llu_i2stat(inode)->st_ino,
- lli->lli_st_generation);
- RETURN(rc);
+
+ 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 = llu_inode_getattr(inode, &oa, op_data->op_ioepoch,
+ old_flags & MF_GETATTR_LOCK);
+ if (rc) {
+ oa.o_valid = 0;
+ if (rc == -ENOENT)
+ CDEBUG(D_INODE, "objid "LPX64" is destroyed\n",
+ lli->lli_smd->lsm_object_id);
+ else
+ CERROR("inode_getattr failed (%d): unable to "
+ "send a Size-on-MDS attribute update "
+ "for inode %llu/%lu\n", rc,
+ (long long)llu_i2stat(inode)->st_ino,
+ lli->lli_st_generation);
+ } 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);
}
-
- md_from_obdo(&op_data, &oa, oa.o_valid);
- memcpy(&op_data.op_handle, fh, sizeof(*fh));
- op_data.op_ioepoch = ioepoch;
- op_data.op_flags |= MF_SOM_CHANGE;
- rc = llu_md_setattr(inode, &op_data);
+ rc = llu_md_setattr(inode, op_data, NULL);
RETURN(rc);
}
-int llu_md_close(struct obd_export *md_exp, struct inode *inode)
+void llu_pack_inode2opdata(struct inode *inode, struct md_op_data *op_data,
+ struct lustre_handle *fh)
{
struct llu_inode_info *lli = llu_i2info(inode);
- struct ll_file_data *fd = lli->lli_file_data;
- struct ptlrpc_request *req = NULL;
- struct obd_client_handle *och = &fd->fd_mds_och;
struct intnl_stat *st = llu_i2stat(inode);
- struct md_op_data op_data = { { 0 } };
- int rc;
ENTRY;
- /* clear group lock, if present */
- if (fd->fd_flags & LL_FILE_GROUP_LOCKED) {
- struct lov_stripe_md *lsm = llu_i2info(inode)->lli_smd;
- fd->fd_flags &= ~(LL_FILE_GROUP_LOCKED|LL_FILE_IGNORE_LOCK);
- rc = llu_extent_unlock(fd, inode, lsm, LCK_GROUP,
- &fd->fd_cwlockh);
- }
+ op_data->op_fid1 = lli->lli_fid;
+ op_data->op_attr.ia_atime = st->st_atime;
+ op_data->op_attr.ia_mtime = st->st_mtime;
+ op_data->op_attr.ia_ctime = st->st_ctime;
+ op_data->op_attr.ia_size = st->st_size;
+ op_data->op_attr_blocks = st->st_blocks;
+ op_data->op_attr.ia_attr_flags = lli->lli_st_flags;
+ op_data->op_ioepoch = lli->lli_ioepoch;
+ if (fh)
+ op_data->op_handle = *fh;
+ EXIT;
+}
+
+/** Pack SOM attributes info @opdata for CLOSE, DONE_WRITING rpc. */
+void llu_done_writing_attr(struct inode *inode, struct md_op_data *op_data)
+{
+ struct llu_inode_info *lli = llu_i2info(inode);
+ ENTRY;
+
+ op_data->op_flags |= MF_SOM_CHANGE;
+
+ /* Pack Size-on-MDS attributes if we are in IO
+ * epoch and attributes are valid. */
+ LASSERT(!(lli->lli_flags & LLIF_MDS_SIZE_LOCK));
+ if (!cl_local_size(inode))
+ op_data->op_attr.ia_valid |= ATTR_MTIME_SET | ATTR_CTIME_SET |
+ ATTR_ATIME_SET | ATTR_SIZE | ATTR_BLOCKS;
+
+ EXIT;
+}
+
+static void llu_prepare_close(struct inode *inode, struct md_op_data *op_data,
+ struct ll_file_data *fd)
+{
+ struct obd_client_handle *och = &fd->fd_mds_och;
+
+ op_data->op_attr.ia_valid = ATTR_MODE | ATTR_ATIME_SET |
+ ATTR_MTIME_SET | ATTR_CTIME_SET;
- op_data.op_attr.ia_valid = ATTR_MODE | ATTR_ATIME_SET |
- ATTR_MTIME_SET | ATTR_CTIME_SET;
-
if (fd->fd_flags & FMODE_WRITE) {
struct llu_sb_info *sbi = llu_i2sbi(inode);
if (!(sbi->ll_lco.lco_flags & OBD_CONNECT_SOM) ||
!S_ISREG(llu_i2stat(inode)->st_mode)) {
- op_data.op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
+ op_data->op_attr.ia_valid |= ATTR_SIZE | ATTR_BLOCKS;
} else {
/* Inode cannot be dirty. Close the epoch. */
- op_data.op_flags |= MF_EPOCH_CLOSE;
- /* XXX: Send CHANGE flag only if Size-on-MDS inode attributes
- * are really changed. */
- op_data.op_flags |= MF_SOM_CHANGE;
-
- /* Pack Size-on-MDS attributes if we are in IO epoch and
- * attributes are valid. */
- LASSERT(!(lli->lli_flags & LLIF_MDS_SIZE_LOCK));
- if (!llu_local_size(inode))
- op_data.op_attr.ia_valid |=
- OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
+ op_data->op_flags |= MF_EPOCH_CLOSE;
+ /* XXX: Send SOM attributes only if they are really
+ * changed. */
+ llu_done_writing_attr(inode, op_data);
}
}
- op_data.op_fid1 = lli->lli_fid;
- op_data.op_attr.ia_atime = st->st_atime;
- op_data.op_attr.ia_mtime = st->st_mtime;
- op_data.op_attr.ia_ctime = st->st_ctime;
- op_data.op_attr.ia_size = st->st_size;
- op_data.op_attr_blocks = st->st_blocks;
- op_data.op_attr.ia_attr_flags = lli->lli_st_flags;
- op_data.op_ioepoch = lli->lli_ioepoch;
- memcpy(&op_data.op_handle, &och->och_fh, sizeof(op_data.op_handle));
-
- rc = md_close(md_exp, &op_data, och, &req);
+ llu_pack_inode2opdata(inode, op_data, &och->och_fh);
+ llu_prep_md_op_data(op_data, inode, NULL, NULL,
+ 0, 0, LUSTRE_OPC_ANY);
+}
+
+int llu_md_close(struct obd_export *md_exp, struct inode *inode)
+{
+ struct llu_inode_info *lli = llu_i2info(inode);
+ struct ll_file_data *fd = lli->lli_file_data;
+ struct ptlrpc_request *req = NULL;
+ struct obd_client_handle *och = &fd->fd_mds_och;
+ struct intnl_stat *st = llu_i2stat(inode);
+ struct md_op_data op_data = { { 0 } };
+ int rc;
+ ENTRY;
+
+ /* clear group lock, if present */
+ if (fd->fd_flags & LL_FILE_GROUP_LOCKED)
+ llu_put_grouplock(inode, fd->fd_grouplock.cg_gid);
+
+ llu_prepare_close(inode, &op_data, fd);
+ rc = md_close(md_exp, &op_data, och->och_mod, &req);
if (rc == -EAGAIN) {
/* We are the last writer, so the MDS has instructed us to get
* the file size and any write cookies, then close again. */
- LASSERT(fd->fd_flags & FMODE_WRITE);
- rc = llu_sizeonmds_update(inode, &och->och_fh,
- op_data.op_ioepoch);
+ LASSERT(lli->lli_open_flags & FMODE_WRITE);
+ rc = llu_som_update(inode, &op_data);
if (rc) {
CERROR("inode %llu mdc Size-on-MDS update failed: "
"rc = %d\n", (long long)st->st_ino, rc);
}
/* if open count == 0 && stale_flag is set, should we
* remove the inode immediately? */
- liblustre_wait_event(0);
+ liblustre_wait_idle();
return 0;
}
RETURN(off);
}
-
-/* this isn't where truncate starts. roughly:
- * llu_iop_{open,setattr}->llu_setattr_raw->llu_vmtruncate->llu_truncate
- * we grab the lock back in setattr_raw to avoid races. */
-static void llu_truncate(struct inode *inode, obd_flag flags)
-{
- struct llu_inode_info *lli = llu_i2info(inode);
- struct intnl_stat *st = llu_i2stat(inode);
- struct obd_info oinfo = { { { 0 } } };
- struct obdo oa = { 0 };
- int rc;
- ENTRY;
- CDEBUG(D_VFSTRACE, "VFS Op:inode=%llu/%lu(%p) to %llu\n",
- (long long)st->st_ino, lli->lli_st_generation, inode,
- (long long)st->st_size);
-
- if (!lli->lli_smd) {
- CDEBUG(D_INODE, "truncate on inode %llu with no objects\n",
- (long long)st->st_ino);
- EXIT;
- return;
- }
-
- oinfo.oi_md = lli->lli_smd;
- oinfo.oi_policy.l_extent.start = st->st_size;
- oinfo.oi_policy.l_extent.end = OBD_OBJECT_EOF;
- oinfo.oi_oa = &oa;
- oa.o_id = lli->lli_smd->lsm_object_id;
- oa.o_valid = OBD_MD_FLID | OBD_MD_FLFLAGS;
- oa.o_flags = flags; /* We don't actually want to copy inode flags */
-
- obdo_from_inode(&oa, inode,
- OBD_MD_FLTYPE | OBD_MD_FLMODE | OBD_MD_FLATIME |
- OBD_MD_FLMTIME | OBD_MD_FLCTIME);
-
- obd_adjust_kms(llu_i2obdexp(inode), lli->lli_smd, st->st_size, 1);
-
- CDEBUG(D_INFO, "calling punch for "LPX64" (all bytes after %Lu)\n",
- oa.o_id, (long long)st->st_size);
-
- /* truncate == punch from new size to absolute end of file */
- rc = obd_punch_rqset(llu_i2obdexp(inode), &oinfo, NULL);
- if (rc)
- CERROR("obd_truncate fails (%d) ino %llu\n",
- rc, (long long)st->st_ino);
- else
- obdo_to_inode(inode, &oa, OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
- OBD_MD_FLATIME | OBD_MD_FLMTIME |
- OBD_MD_FLCTIME);
-
- EXIT;
- return;
-} /* llu_truncate */
-
-int llu_vmtruncate(struct inode * inode, loff_t offset, obd_flag flags)
-{
- llu_i2stat(inode)->st_size = offset;
-
- /*
- * llu_truncate() is only called from this
- * point. llu_vmtruncate/llu_truncate split exists to mimic the
- * structure of Linux VFS truncate code path.
- */
-
- llu_truncate(inode, flags);
-
- return 0;
-}