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 <portals/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 /* If we are unlinking an open file/dir (i.e. creating an orphan) then
  45  * we instead link the inode into the PENDING directory until it is
  46  * finally released.  We can't simply call mds_reint_rename() or some
  47  * part thereof, because we don't have the inode to check for link
  48  * count/open status until after it is locked.
  49  *
  50  * For lock ordering, we always get the PENDING, then pending_child lock
  51  * last to avoid deadlocks.
  52  */
  53
  54 int mds_open_unlink_rename(struct mds_update_record *rec,
  55                            struct obd_device *obd, struct dentry *dparent,
  56                            struct dentry *dchild, void **handle)
  57 {
  58         struct mds_obd *mds = &obd->u.mds;
  59         struct inode *pending_dir = mds->mds_pending_dir->d_inode;
  60         struct dentry *pending_child;
  61         char fidname[LL_FID_NAMELEN];
  62         int fidlen = 0, rc;
  63         unsigned mode;
  64         ENTRY;
  65
  66         LASSERT(!mds_inode_is_orphan(dchild->d_inode));
  67
  68         down(&pending_dir->i_sem);
  69         fidlen = ll_fid2str(fidname, dchild->d_inode->i_ino,
  70                             dchild->d_inode->i_generation);
  71
  72         CDEBUG(D_HA, "pending destroy of %dx open file %s = %s\n",
  73                mds_open_orphan_count(dchild->d_inode), rec->ur_name, fidname);
  74
  75         pending_child = lookup_one_len(fidname, mds->mds_pending_dir, fidlen);
  76         if (IS_ERR(pending_child))
  77                 GOTO(out_lock, rc = PTR_ERR(pending_child));
  78
  79         if (pending_child->d_inode != NULL) {
  80                 CERROR("re-destroying orphan file %s?\n", rec->ur_name);
  81                 LASSERT(pending_child->d_inode == dchild->d_inode);
  82                 GOTO(out_dput, rc = 0);
  83         }
  84
  85         /* link() is semanticaly-wrong for S_IFDIR, so we set S_IFREG
  86          * for linking and return real mode back then -bzzz */
  87         mode = dchild->d_inode->i_mode;
  88         dchild->d_inode->i_mode = S_IFREG;
  89         rc = vfs_link(dchild, pending_dir, pending_child);
  90         if (rc)
  91                 CERROR("error linking orphan %s to PENDING: rc = %d\n",
  92                        rec->ur_name, rc);
  93         else
  94                 mds_inode_set_orphan(dchild->d_inode);
  95
  96         /* return mode and correct i_nlink if inode is directory */
  97         LASSERT(dchild->d_inode->i_nlink == 1);
  98         dchild->d_inode->i_mode = mode;
  99         if ((mode & S_IFMT) == S_IFDIR) {
 100                 dchild->d_inode->i_nlink++;
 101                 pending_dir->i_nlink++;
 102         }
 103         mark_inode_dirty(dchild->d_inode);
 104
 105 out_dput:
 106         dput(pending_child);
 107 out_lock:
 108         up(&pending_dir->i_sem);
 109         RETURN(rc);
 110 }
 111
 112 static int mds_osc_destroy_orphan(struct mds_obd *mds,
 113                                   struct inode *inode,
 114                                   struct lov_mds_md *lmm,
 115                                   int lmm_size,
 116                                   struct llog_cookie *logcookies,
 117                                   int log_unlink)
 118 {
 119         struct lov_stripe_md *lsm = NULL;
 120         struct obd_trans_info oti = { 0 };
 121         struct obdo *oa;
 122         int rc;
 123         ENTRY;
 124
 125         if (lmm_size == 0)
 126                 RETURN(0);
 127
 128         rc = obd_unpackmd(mds->mds_osc_exp, &lsm, lmm, lmm_size);
 129         if (rc < 0) {
 130                 CERROR("Error unpack md %p\n", lmm);
 131                 RETURN(rc);
 132         } else {
 133                 LASSERT(rc >= sizeof(*lsm));
 134                 rc = 0;
 135         }
 136
 137         oa = obdo_alloc();
 138         if (oa == NULL)
 139                 GOTO(out_free_memmd, rc = -ENOMEM);
 140         oa->o_id = lsm->lsm_object_id;
 141         oa->o_gr = FILTER_GROUP_FIRST_MDS + mds->mds_num;
 142         oa->o_mode = inode->i_mode & S_IFMT;
 143         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE | OBD_MD_FLGROUP;
 144
 145         if (log_unlink && logcookies) {
 146                 oa->o_valid |= OBD_MD_FLCOOKIE;
 147                 oti.oti_logcookies = logcookies;
 148         }
 149
 150         rc = obd_destroy(mds->mds_osc_exp, oa, lsm, &oti);
 151         obdo_free(oa);
 152         if (rc)
 153                 CDEBUG(D_INODE, "destroy orphan objid 0x"LPX64" on ost error "
 154                        "%d\n", lsm->lsm_object_id, rc);
 155 out_free_memmd:
 156         obd_free_memmd(mds->mds_osc_exp, &lsm);
 157         RETURN(rc);
 158 }
 159
 160 static int mds_unlink_orphan(struct obd_device *obd, struct dentry *dchild,
 161                              struct inode *inode, struct inode *pending_dir)
 162 {
 163         struct mds_obd *mds = &obd->u.mds;
 164         struct lov_mds_md *lmm = NULL;
 165         struct llog_cookie *logcookies = NULL;
 166         int lmm_size = 0, log_unlink = 0;
 167         void *handle = NULL;
 168         int rc, err;
 169         ENTRY;
 170
 171         LASSERT(mds->mds_osc_obd != NULL);
 172
 173         OBD_ALLOC(lmm, mds->mds_max_mdsize);
 174         if (lmm == NULL)
 175                 RETURN(-ENOMEM);
 176
 177         down(&inode->i_sem);
 178         rc = fsfilt_get_md(obd, inode, lmm, mds->mds_max_mdsize);
 179         up(&inode->i_sem);
 180
 181         if (rc < 0) {
 182                 CERROR("Error %d reading eadata for ino %lu\n",
 183                        rc, inode->i_ino);
 184                 GOTO(out_free_lmm, rc);
 185         } else if (rc > 0) {
 186                 lmm_size = rc;
 187                 rc = mds_convert_lov_ea(obd, inode, lmm, lmm_size);
 188                 if (rc > 0)
 189                         lmm_size = rc;
 190                 rc = 0;
 191         }
 192
 193         handle = fsfilt_start_log(obd, pending_dir, FSFILT_OP_UNLINK, NULL,
 194                                   le32_to_cpu(lmm->lmm_stripe_count));
 195         if (IS_ERR(handle)) {
 196                 rc = PTR_ERR(handle);
 197                 CERROR("error fsfilt_start: %d\n", rc);
 198                 handle = NULL;
 199                 GOTO(out_free_lmm, rc);
 200         }
 201
 202         down(&inode->i_sem);
 203         rc = fsfilt_get_md(obd, inode, lmm, mds->mds_max_mdsize);
 204         up(&inode->i_sem);
 205
 206         if (rc < 0) {
 207                 CERROR("Error %d reading eadata for ino %lu\n",
 208                        rc, inode->i_ino);
 209                 GOTO(out_free_lmm, rc);
 210         } else if (rc > 0) {
 211                 lmm_size = rc;
 212                 rc = 0;
 213         }
 214
 215         if (S_ISDIR(inode->i_mode))
 216                 rc = vfs_rmdir(pending_dir, dchild);
 217         else
 218                 rc = vfs_unlink(pending_dir, dchild);
 219
 220         if (rc)
 221                 CERROR("error %d unlinking orphan %*s from PENDING directory\n",
 222                        rc, dchild->d_name.len, dchild->d_name.name);
 223
 224         if (!rc && lmm_size) {
 225                 OBD_ALLOC(logcookies, mds->mds_max_cookiesize);
 226                 if (logcookies == NULL)
 227                         rc = -ENOMEM;
 228                 else if (mds_log_op_unlink(obd, inode, lmm,lmm_size,logcookies,
 229                                            mds->mds_max_cookiesize) > 0)
 230                         log_unlink = 1;
 231         }
 232         err = fsfilt_commit(obd, mds->mds_sb, pending_dir, handle, 0);
 233         if (err) {
 234                 CERROR("error committing orphan unlink: %d\n", err);
 235                 if (!rc)
 236                         rc = err;
 237         }
 238         if (!rc) {
 239                 rc = mds_osc_destroy_orphan(mds, inode, lmm, lmm_size,
 240                                             logcookies, log_unlink);
 241         }
 242
 243         if (logcookies != NULL)
 244                 OBD_FREE(logcookies, mds->mds_max_cookiesize);
 245 out_free_lmm:
 246         OBD_FREE(lmm, mds->mds_max_mdsize);
 247         RETURN(rc);
 248 }
 249
 250 int mds_cleanup_orphans(struct obd_device *obd)
 251 {
 252         struct mds_obd *mds = &obd->u.mds;
 253         struct lvfs_run_ctxt saved;
 254         struct file *file;
 255         struct dentry *dchild, *dentry;
 256         struct vfsmount *mnt;
 257         struct inode *child_inode, *pending_dir = mds->mds_pending_dir->d_inode;
 258         struct l_linux_dirent *dirent, *n;
 259         struct list_head dentry_list;
 260         char d_name[LL_FID_NAMELEN];
 261         __u64 i = 0;
 262         int rc = 0, item = 0, namlen;
 263         ENTRY;
 264
 265         push_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 266         dentry = dget(mds->mds_pending_dir);
 267         if (IS_ERR(dentry))
 268                 GOTO(err_pop, rc = PTR_ERR(dentry));
 269         mnt = mntget(mds->mds_vfsmnt);
 270         if (IS_ERR(mnt))
 271                 GOTO(err_mntget, rc = PTR_ERR(mnt));
 272
 273         file = dentry_open(mds->mds_pending_dir, mds->mds_vfsmnt,
 274                            O_RDONLY | O_LARGEFILE);
 275         if (IS_ERR(file))
 276                 GOTO(err_pop, rc = PTR_ERR(file));
 277
 278         INIT_LIST_HEAD(&dentry_list);
 279         rc = l_readdir(file, &dentry_list);
 280         filp_close(file, 0);
 281         if (rc < 0)
 282                 GOTO(err_out, rc);
 283
 284         list_for_each_entry_safe(dirent, n, &dentry_list, lld_list) {
 285                 i ++;
 286                 list_del(&dirent->lld_list);
 287
 288                 namlen = strlen(dirent->lld_name);
 289                 LASSERT(sizeof(d_name) >= namlen + 1);
 290                 strcpy(d_name, dirent->lld_name);
 291                 OBD_FREE(dirent, sizeof(*dirent));
 292
 293                 CDEBUG(D_INODE, "entry "LPU64" of PENDING DIR: %s\n",
 294                        i, d_name);
 295
 296                 if (((namlen == 1) && !strcmp(d_name, ".")) ||
 297                     ((namlen == 2) && !strcmp(d_name, ".."))) {
 298                         continue;
 299                 }
 300
 301                 down(&pending_dir->i_sem);
 302                 dchild = lookup_one_len(d_name, mds->mds_pending_dir, namlen);
 303                 if (IS_ERR(dchild)) {
 304                         up(&pending_dir->i_sem);
 305                         GOTO(err_out, rc = PTR_ERR(dchild));
 306                 }
 307                 if (!dchild->d_inode) {
 308                         CERROR("orphan %s has been removed\n", d_name);
 309                         GOTO(next, rc = 0);
 310                 }
 311
 312                 child_inode = dchild->d_inode;
 313                 if (mds_inode_is_orphan(child_inode) &&
 314                     mds_open_orphan_count(child_inode)) {
 315                         CWARN("orphan %s was re-opened during recovery\n", d_name);
 316                         GOTO(next, rc = 0);
 317                 }
 318
 319                 rc = mds_unlink_orphan(obd, dchild, child_inode, pending_dir);
 320                 if (rc == 0) {
 321                         item ++;
 322                         CWARN("removed orphan %s from MDS and OST\n", d_name);
 323                 } else {
 324                         CDEBUG(D_INODE, "removed orphan %s from MDS/OST failed,"
 325                                " rc = %d\n", d_name, rc);
 326                         rc = 0;
 327                 }
 328 next:
 329                 l_dput(dchild);
 330                 up(&pending_dir->i_sem);
 331         }
 332 err_out:
 333         list_for_each_entry_safe(dirent, n, &dentry_list, lld_list) {
 334                 list_del(&dirent->lld_list);
 335                 OBD_FREE(dirent, sizeof(*dirent));
 336         }
 337 err_pop:
 338         pop_ctxt(&saved, &obd->obd_lvfs_ctxt, NULL);
 339         if (rc == 0)
 340                 rc = item;
 341         RETURN(rc);
 342
 343 err_mntget:
 344         l_dput(mds->mds_pending_dir);
 345         goto err_pop;
 346 }