mds_reint.c

lustre 1.6.5 source code

/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
 * vim:expandtab:shiftwidth=8:tabstop=8:
 *
 *  linux/mds/mds_reint.c
 *  Lustre Metadata Server (mds) reintegration routines
 *
 *  Copyright (C) 2002-2005 Cluster File Systems, Inc.
 *   Author: Peter Braam <[e-mail]>
 *   Author: Andreas Dilger <[e-mail]>
 *   Author: Phil Schwan <[e-mail]>
 *
 *   This file is part of the Lustre file system, http://www.lustre.org
 *   Lustre is a trademark of Cluster File Systems, Inc.
 *
 *   You may have signed or agreed to another license before downloading
 *   this software.  If so, you are bound by the terms and conditions
 *   of that agreement, and the following does not apply to you.  See the
 *   LICENSE file included with this distribution for more information.
 *
 *   If you did not agree to a different license, then this copy of Lustre
 *   is open source 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.
 *
 *   In either case, 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
 *   license text for more details.
 */

#ifndef EXPORT_SYMTAB
# define EXPORT_SYMTAB
#endif
#define DEBUG_SUBSYSTEM S_MDS

#include <linux/fs.h>
#include <obd_support.h>
#include <obd_class.h>
#include <obd.h>
#include <lustre_lib.h>
#include <lustre/lustre_idl.h>
#include <lustre_mds.h>
#include <lustre_dlm.h>
#include <lustre_fsfilt.h>
#include <lustre_ucache.h>

#include "mds_internal.h"

void mds_commit_cb(struct obd_device *obd, __u64 transno, void *data,
                   int error)
{
        obd_transno_commit_cb(obd, transno, error);
}

struct mds_logcancel_data {
        struct lov_mds_md      *mlcd_lmm;
        int                     mlcd_size;
        int                     mlcd_cookielen;
        int                     mlcd_eadatalen;
        struct llog_cookie      mlcd_cookies[0];
};


static void mds_cancel_cookies_cb(struct obd_device *obd, __u64 transno,
                                  void *cb_data, int error)
{
        struct mds_logcancel_data *mlcd = cb_data;
        struct lov_stripe_md *lsm = NULL;
        struct llog_ctxt *ctxt;
        int rc;

        obd_transno_commit_cb(obd, transno, error);

        CDEBUG(D_RPCTRACE, "cancelling %d cookies\n",
               (int)(mlcd->mlcd_cookielen / sizeof(*mlcd->mlcd_cookies)));

        rc = obd_unpackmd(obd->u.mds.mds_osc_exp, &lsm, mlcd->mlcd_lmm,
                          mlcd->mlcd_eadatalen);
        if (rc < 0) {
                CERROR("bad LSM cancelling %d log cookies: rc %d\n",
                       (int)(mlcd->mlcd_cookielen/sizeof(*mlcd->mlcd_cookies)),
                       rc);
        } else {
                ///* XXX 0 normally, SENDNOW for debug */);
                rc = obd_checkmd(obd->u.mds.mds_osc_exp, obd->obd_self_export,
                                 lsm);
                if (rc)
                        CERROR("Can not revalidate lsm %p \n", lsm);

                ctxt = llog_get_context(obd,mlcd->mlcd_cookies[0].lgc_subsys+1);
                rc = llog_cancel(ctxt, lsm, mlcd->mlcd_cookielen /
                                                sizeof(*mlcd->mlcd_cookies),
                                 mlcd->mlcd_cookies, OBD_LLOG_FL_SENDNOW);
                llog_ctxt_put(ctxt);

                if (rc)
                        CERROR("error cancelling %d log cookies: rc %d\n",
                               (int)(mlcd->mlcd_cookielen /
                                     sizeof(*mlcd->mlcd_cookies)), rc);
        }

        OBD_FREE(mlcd, mlcd->mlcd_size);
}

/* Assumes caller has already pushed us into the kernel context. */
int mds_finish_transno(struct mds_obd *mds, struct inode *inode, void *handle,
                       struct ptlrpc_request *req, int rc, __u32 op_data, 
                       int force_sync)
{
        struct mds_export_data *med = &req->rq_export->exp_mds_data;
        struct mds_client_data *mcd = med->med_mcd;
        struct obd_device *obd = req->rq_export->exp_obd;
        __u64 transno, prev_transno;
        int err;
        loff_t off;
        int log_pri = D_RPCTRACE;
        ENTRY;

        if (IS_ERR(handle)) {
                LASSERT(rc != 0);
                RETURN(rc);
        }

        /* if the export has already been failed, we have no last_rcvd slot */
        if (req->rq_export->exp_failed || obd->obd_fail) {
                CWARN("commit transaction for disconnected client %s: rc %d\n",
                      req->rq_export->exp_client_uuid.uuid, rc);
                if (rc == 0)
                        rc = -ENOTCONN;
                if (handle)
                        GOTO(commit, rc);
                RETURN(rc);
        }

        if (handle == NULL) {
                /* if we're starting our own xaction, use our own inode */
                inode = mds->mds_rcvd_filp->f_dentry->d_inode;
                handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR, NULL);
                if (IS_ERR(handle)) {
                        CERROR("fsfilt_start: %ld\n", PTR_ERR(handle));
                        RETURN(PTR_ERR(handle));
                }
        }

        off = med->med_lr_off;

        transno = lustre_msg_get_transno(req->rq_reqmsg);
        if (rc != 0) {
                if (transno != 0) {
                        CERROR("%s: replay %s transno "LPU64" failed: rc %d\n",
                               obd->obd_name,
                               libcfs_nid2str(req->rq_export->exp_connection->c_peer.nid),
                               transno, rc);
                        transno = 0;
                }
        } else if (transno == 0) {
                spin_lock(&mds->mds_transno_lock);
                transno = ++mds->mds_last_transno;
                spin_unlock(&mds->mds_transno_lock);
        } else {
                spin_lock(&mds->mds_transno_lock);
                if (transno > mds->mds_last_transno)
                        mds->mds_last_transno = transno;
                spin_unlock(&mds->mds_transno_lock);
        }

        req->rq_transno = transno;
        lustre_msg_set_transno(req->rq_repmsg, transno);
        if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
                prev_transno = le64_to_cpu(mcd->mcd_last_close_transno);
                mcd->mcd_last_close_transno = cpu_to_le64(transno);
                mcd->mcd_last_close_xid = cpu_to_le64(req->rq_xid);
                mcd->mcd_last_close_result = cpu_to_le32(rc);
                mcd->mcd_last_close_data = cpu_to_le32(op_data);
        } else {
                prev_transno = le64_to_cpu(mcd->mcd_last_transno);
                if (((lustre_msg_get_flags(req->rq_reqmsg) &
                      (MSG_RESENT | MSG_REPLAY)) == 0) ||
                    (transno > prev_transno)) {
                        mcd->mcd_last_transno = cpu_to_le64(transno);
                        mcd->mcd_last_xid     = cpu_to_le64(req->rq_xid);
                        mcd->mcd_last_result  = cpu_to_le32(rc);
                        mcd->mcd_last_data    = cpu_to_le32(op_data);
                }
        }
        /* update the server data to not lose the greatest transno. Bug 11125 */
        if ((transno == 0) && (prev_transno == mds->mds_last_transno))
                mds_update_server_data(obd, 0);

        if (off <= 0) {
                CERROR("client idx %d has offset %lld\n", med->med_lr_idx, off);
                err = -EINVAL;
        } else {
                struct obd_export *exp = req->rq_export;

                if (!force_sync)
                        force_sync = fsfilt_add_journal_cb(exp->exp_obd,transno, 
                                                          handle, mds_commit_cb,
                                                          NULL);

                err = fsfilt_write_record(obd, mds->mds_rcvd_filp, mcd,
                                          sizeof(*mcd), &off, 
                                          force_sync | exp->exp_need_sync);
                if (force_sync)
                        mds_commit_cb(obd, transno, NULL, err);
        }

        if (err) {
                log_pri = D_ERROR;
                if (rc == 0)
                        rc = err;
        }

        DEBUG_REQ(log_pri, req,
                  "wrote trans #"LPU64" rc %d client %s at idx %u: err = %d",
                  transno, rc, mcd->mcd_uuid, med->med_lr_idx, err);

        err = mds_lov_write_objids(obd);
        if (err) {
                log_pri = D_ERROR;
                if (rc == 0)
                        rc = err;
        }
        CDEBUG(log_pri, "wrote objids: err = %d\n", err);

commit:
        err = fsfilt_commit(obd, inode, handle, 0);
        if (err) {
                CERROR("error committing transaction: %d\n", err);
                if (!rc)
                        rc = err;
        }

        RETURN(rc);
}

/* this gives the same functionality as the code between
 * sys_chmod and inode_setattr
 * chown_common and inode_setattr
 * utimes and inode_setattr
 */
int mds_fix_attr(struct inode *inode, struct mds_update_record *rec)
{
        time_t now = CURRENT_SECONDS;
        struct iattr *attr = &rec->ur_iattr;
        unsigned int ia_valid = attr->ia_valid;
        int error;
        ENTRY;

        if (ia_valid & ATTR_RAW)
                attr->ia_valid &= ~ATTR_RAW;

        if (!(ia_valid & ATTR_CTIME_SET))
                LTIME_S(attr->ia_ctime) = now;
        else
                attr->ia_valid &= ~ATTR_CTIME_SET;
        if (!(ia_valid & ATTR_ATIME_SET))
                LTIME_S(attr->ia_atime) = now;
        if (!(ia_valid & ATTR_MTIME_SET))
                LTIME_S(attr->ia_mtime) = now;

        if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
                RETURN((attr->ia_valid & ~ATTR_ATTR_FLAG) ? -EPERM : 0);

        /* times */
        if ((ia_valid & (ATTR_MTIME|ATTR_ATIME)) == (ATTR_MTIME|ATTR_ATIME)) {
                if (current->fsuid != inode->i_uid &&
                    (error = ll_permission(inode, MAY_WRITE, NULL)) != 0)
                        RETURN(error);
        }

        if (ia_valid & ATTR_SIZE &&
            /* NFSD hack for open(O_CREAT|O_TRUNC)=mknod+truncate (bug 5781) */
            !(rec->ur_uc.luc_fsuid == inode->i_uid &&
              ia_valid & MDS_OPEN_OWNEROVERRIDE)) {
                if ((error = ll_permission(inode, MAY_WRITE, NULL)) != 0)
                        RETURN(error);
        }

        if (ia_valid & (ATTR_UID | ATTR_GID)) {
                /* chown */
                error = -EPERM;
                if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
                        RETURN(-EPERM);
                if (attr->ia_uid == (uid_t) -1)
                        attr->ia_uid = inode->i_uid;
                if (attr->ia_gid == (gid_t) -1)
                        attr->ia_gid = inode->i_gid;
                if (!(ia_valid & ATTR_MODE))
                        attr->ia_mode = inode->i_mode;
                /*
                 * If the user or group of a non-directory has been
                 * changed by a non-root user, remove the setuid bit.
                 * 19981026 David C Niemi <[e-mail]>
                 *
                 * Changed this to apply to all users, including root,
                 * to avoid some races. This is the behavior we had in
                 * 2.0. The check for non-root was definitely wrong
                 * for 2.2 anyway, as it should have been using
                 * CAP_FSETID rather than fsuid -- 19990830 SD.
                 */
                if ((inode->i_mode & S_ISUID) == S_ISUID &&
                    !S_ISDIR(inode->i_mode)) {
                        attr->ia_mode &= ~S_ISUID;
                        attr->ia_valid |= ATTR_MODE;
                }
                /*
                 * Likewise, if the user or group of a non-directory
                 * has been changed by a non-root user, remove the
                 * setgid bit UNLESS there is no group execute bit
                 * (this would be a file marked for mandatory
                 * locking).  19981026 David C Niemi <[e-mail]>
                 *
                 * Removed the fsuid check (see the comment above) --
                 * 19990830 SD.
                 */
                if (((inode->i_mode & (S_ISGID | S_IXGRP)) ==
                     (S_ISGID | S_IXGRP)) && !S_ISDIR(inode->i_mode)) {
                        attr->ia_mode &= ~S_ISGID;
                        attr->ia_valid |= ATTR_MODE;
                }