lustre/mds/mds_unlink_open.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  lustre/mds/mds_orphan.c
   5  *
   6  *  Copyright (c) 2001-2003 Cluster File Systems, Inc.
   7  *   Author: Peter Braam <braam@clusterfs.com>
   8  *   Author: Andreas Dilger <adilger@clusterfs.com>
   9  *   Author: Phil Schwan <phil@clusterfs.com>
  10  *
  11  *   This file is part of Lustre, http://www.lustre.org.
  12  *
  13  *   Lustre is free software; you can redistribute it and/or
  14  *   modify it under the terms of version 2 of the GNU General Public
  15  *   License as published by the Free Software Foundation.
  16  *
  17  *   Lustre is distributed in the hope that it will be useful,
  18  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  19  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  20  *   GNU General Public License for more details.
  21  *
  22  *   You should have received a copy of the GNU General Public License
  23  *   along with Lustre; if not, write to the Free Software
  24  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  25  */
  26
  27 /* code for handling open unlinked files */
  28
  29 #define DEBUG_SUBSYSTEM S_MDS
  30
  31 #include <linux/config.h>
  32 #include <linux/module.h>
  33 #include <linux/version.h>
  34
  35 #include <libcfs/list.h>
  36 #include <linux/obd_class.h>
  37 #include <linux/lustre_fsfilt.h>
  38 #include <linux/lustre_commit_confd.h>
  39 #include <linux/lvfs.h>
  40
  41 #include "mds_internal.h"
  42
  43 /*
  44  * used when destroying orphanes and from mds_reint_unlink() when MDS wants to
  45  * destroy objects on OSS.
  46  */
  47 int
  48 mds_unlink_object(struct mds_obd *mds, struct inode *inode,
  49                   struct lov_mds_md *lmm, int lmm_size,
  50                   struct llog_cookie *logcookies,
  51                   int log_unlink, int async)
  52 {
  53         struct lov_stripe_md *lsm = NULL;
  54         struct obd_trans_info oti = { 0 };
  55         struct obdo *oa;
  56         int rc;
  57         ENTRY;
  58
  59         if (lmm_size == 0)
  60                 RETURN(0);
  61
  62         rc = obd_unpackmd(mds->mds_dt_exp, &lsm, lmm, lmm_size);
  63         if (rc < 0) {
  64                 CERROR("Error unpack md %p\n", lmm);
  65                 RETURN(rc);
  66         } else {
  67                 LASSERT(rc >= sizeof(*lsm));
  68                 rc = 0;
  69         }
  70
  71         oa = obdo_alloc();
  72         if (oa == NULL)
  73                 GOTO(out_free_memmd, rc = -ENOMEM);
  74         oa->o_id = lsm->lsm_object_id;
  75         oa->o_gr = FILTER_GROUP_FIRST_MDS + mds->mds_num;
  76         oa->o_mode = inode->i_mode & S_IFMT;
  77         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
  78
  79         if (log_unlink && logcookies) {
  80                 oa->o_valid |= OBD_MD_FLCOOKIE;
  81                 oti.oti_logcookies = logcookies;
  82         }
  83
  84         CDEBUG(D_INODE, "destroy OSS object %d/%d\n",
  85                (int)oa->o_id, (int)oa->o_gr);
  86
  87         oti.oti_async = async;
  88         rc = obd_destroy(mds->mds_dt_exp, oa, lsm, &oti);
  89         obdo_free(oa);
  90 out_free_memmd:
  91         obd_free_memmd(mds->mds_dt_exp, &lsm);
  92         RETURN(rc);
  93 }
  94
  95 static int mds_unlink_orphan(struct obd_device *obd, struct dentry *dchild,
  96                              struct inode *inode, struct inode *pending_dir)
  97 {
  98         struct mds_obd *mds = &obd->u.mds;
  99         struct lov_mds_md *lmm = NULL;
 100         struct llog_cookie *logcookies = NULL;
 101         int lmm_size, log_unlink = 0;
 102         void *handle = NULL;
 103         int rc, err;
 104         ENTRY;
 105
 106         LASSERT(mds->mds_dt_obd != NULL);
 107         LASSERT(obd->obd_recovering == 0);
 108
 109         /* We don't need to do any of these other things for orhpan dirs,
 110          * especially not mds_get_md (may get a default LOV EA, bug 4554) */
 111         if (S_ISDIR(inode->i_mode)) {
 112                 rc = vfs_rmdir(pending_dir, dchild);
 113                 if (rc)
 114                         CERROR("error %d unlinking dir %*s from PENDING\n",
 115                                rc, dchild->d_name.len, dchild->d_name.name);
 116                 RETURN(rc);
 117         }
 118
 119         lmm_size = mds->mds_max_mdsize;
 120         OBD_ALLOC(lmm, lmm_size);
 121         if (lmm == NULL)
 122                 RETURN(-ENOMEM);
 123
 124         rc = mds_get_md(obd, inode, lmm, &lmm_size, 1, 0);
 125         if (rc < 0)
 126                 GOTO(out_free_lmm, rc);
 127
 128         handle = fsfilt_start_log(obd, pending_dir, FSFILT_OP_UNLINK, NULL,
 129                                   le32_to_cpu(lmm->lmm_stripe_count));
 130         if (IS_ERR(handle)) {
 131                 rc = PTR_ERR(handle);
 132                 CERROR("error fsfilt_start: %d\n", rc);
 133                 handle = NULL;
 134                 GOTO(out_free_lmm, rc);
 135         }
 136
 137         rc = vfs_unlink(pending_dir, dchild);
 138         if (rc) {
 139                 CERROR("error %d unlinking orphan %.*s from PENDING\n",
 140                        rc, dchild->d_name.len, dchild->d_name.name);
 141         } else if (lmm_size) {
 142                 OBD_ALLOC(logcookies, mds->mds_max_cookiesize);
 143                 if (logcookies == NULL)
 144                         rc = -ENOMEM;
 145                 else if (mds_log_op_unlink(obd, inode, lmm,lmm_size,logcookies,
 146                                            mds->mds_max_cookiesize, NULL) > 0)
 147                         log_unlink = 1;
 148         }
 149         err = fsfilt_commit(obd, mds->mds_sb, pending_dir, handle, 0);
 150         if (err) {
 151                 CERROR("error committing orphan unlink: %d\n", err);
 152                 if (!rc)
 153                         rc = err;
 154         } else if (!rc) {
 155                 rc = mds_unlink_object(mds, inode, lmm, lmm_size,
 156                                        logcookies, log_unlink, 0);
 157         }
 158
 159         if (logcookies != NULL)
 160                 OBD_FREE(logcookies, mds->mds_max_cookiesize);
 161 out_free_lmm:
 162         OBD_FREE(lmm, mds->mds_max_mdsize);
 163         RETURN(rc);
 164 }
 165
 166 int mds_cleanup_orphans(struct obd_device *obd)
 167 {
 168         struct mds_obd *mds = &obd->u.mds;
 169         struct lvfs_run_ctxt saved;
 170         struct file *file;
 171         struct dentry *dchild, *dentry;
 172         struct vfsmount *mnt;
 173         struct inode *child_inode, *pending_dir = mds->mds_pending_dir->d_inode;
 174         struct l_linux_dirent *dirent, *n;
 175         struct list_head dentry_list;
 176         char d_name[LL_ID_NAMELEN];
 177         unsigned long inum;
 178         __u64 i = 0;
 179         int rc = 0, item = 0, namlen;
 180         ENTRY;
 181
 182         LASSERT(obd->obd_recovering == 0);
 183         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 184         dentry = dget(mds->mds_pending_dir);
 185         if (IS_ERR(dentry))
 186                 GOTO(err_pop, rc = PTR_ERR(dentry));
 187         mnt = mntget(mds->mds_vfsmnt);
 188         if (IS_ERR(mnt))
 189                 GOTO(err_mntget, rc = PTR_ERR(mnt));
 190
 191         file = dentry_open(mds->mds_pending_dir, mds->mds_vfsmnt,
 192                            O_RDONLY | O_LARGEFILE);
 193         if (IS_ERR(file))
 194                 GOTO(err_pop, rc = PTR_ERR(file));
 195
 196         INIT_LIST_HEAD(&dentry_list);
 197         rc = l_readdir(file, &dentry_list);
 198         filp_close(file, 0);
 199         if (rc < 0)
 200                 GOTO(err_out, rc);
 201
 202         list_for_each_entry_safe(dirent, n, &dentry_list, lld_list) {
 203                 i++;
 204                 list_del(&dirent->lld_list);
 205
 206                 namlen = strlen(dirent->lld_name);
 207                 LASSERT(sizeof(d_name) >= namlen + 1);
 208                 strcpy(d_name, dirent->lld_name);
 209                 inum = dirent->lld_ino;
 210                 OBD_FREE(dirent, sizeof(*dirent));
 211
 212                 CDEBUG(D_INODE, "entry "LPU64" of PENDING DIR: %s\n",
 213                        i, d_name);
 214
 215                 if (((namlen == 1) && !strcmp(d_name, ".")) ||
 216                     ((namlen == 2) && !strcmp(d_name, "..")) || inum == 0)
 217                         continue;
 218
 219                 down(&pending_dir->i_sem);
 220                 dchild = lookup_one_len(d_name, mds->mds_pending_dir, namlen);
 221                 if (IS_ERR(dchild)) {
 222                         up(&pending_dir->i_sem);
 223                         GOTO(err_out, rc = PTR_ERR(dchild));
 224                 }
 225                 if (!dchild->d_inode) {
 226                         CERROR("orphan %s has been removed\n", d_name);
 227                         GOTO(next, rc = 0);
 228                 }
 229
 230                 if (is_bad_inode(dchild->d_inode)) {
 231                         CERROR("bad orphan inode found %lu/%u\n",
 232                                dchild->d_inode->i_ino,
 233                                dchild->d_inode->i_generation);
 234                         GOTO(next, rc = -ENOENT);
 235                 }
 236
 237                 child_inode = dchild->d_inode;
 238                 DOWN_READ_I_ALLOC_SEM(child_inode);
 239                 if (mds_orphan_open_count(child_inode)) {
 240                         UP_READ_I_ALLOC_SEM(child_inode);
 241                         CWARN("orphan %s re-opened during recovery\n", d_name);
 242                         GOTO(next, rc = 0);
 243                 }
 244                 if (!mds_inode_is_orphan(child_inode)) {
 245                         UP_READ_I_ALLOC_SEM(child_inode);
 246                         CWARN("orphan %s has been removed by CLOSE\n", d_name);
 247                         GOTO(next, rc = 0);
 248                 }
 249                 mds_inode_unset_orphan(child_inode);
 250                 UP_READ_I_ALLOC_SEM(child_inode);
 251                 rc = mds_unlink_orphan(obd, dchild, child_inode, pending_dir);
 252                 if (rc == 0) {
 253                         item ++;
 254                         CWARN("removed orphan %s from MDS and OST\n", d_name);
 255                 } else {
 256                         CDEBUG(D_INODE, "removed orphan %s from MDS/OST failed,"
 257                                " rc = %d\n", d_name, rc);
 258                         rc = 0;
 259                 }
 260 next:
 261                 l_dput(dchild);
 262                 up(&pending_dir->i_sem);
 263         }
 264 err_out:
 265         list_for_each_entry_safe(dirent, n, &dentry_list, lld_list) {
 266                 list_del(&dirent->lld_list);
 267                 OBD_FREE(dirent, sizeof(*dirent));
 268         }
 269 err_pop:
 270         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 271         if (rc == 0)
 272                 rc = item;
 273         RETURN(rc);
 274
 275 err_mntget:
 276         l_dput(mds->mds_pending_dir);
 277         goto err_pop;
 278 }