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         ENTRY;
  64
  65         LASSERT(!mds_inode_is_orphan(dchild->d_inode));
  66
  67         down(&pending_dir->i_sem);
  68         fidlen = ll_fid2str(fidname, dchild->d_inode->i_ino,
  69                             dchild->d_inode->i_generation);
  70
  71         CWARN("pending destroy of %dx open file %s = %s\n",
  72               mds_open_orphan_count(dchild->d_inode),
  73               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         *handle = fsfilt_start(obd, pending_dir, FSFILT_OP_RENAME, NULL);
  86         if (IS_ERR(*handle))
  87                 GOTO(out_dput, rc = PTR_ERR(*handle));
  88
  89         lock_kernel();
  90         rc = vfs_rename(dparent->d_inode, dchild, pending_dir, pending_child);
  91         unlock_kernel();
  92         if (rc)
  93                 CERROR("error renaming orphan %lu/%s to PENDING: rc = %d\n",
  94                        dparent->d_inode->i_ino, rec->ur_name, rc);
  95         else
  96                 mds_inode_set_orphan(dchild->d_inode);
  97 out_dput:
  98         dput(pending_child);
  99 out_lock:
 100         up(&pending_dir->i_sem);
 101         RETURN(rc);
 102 }
 103
 104 static int mds_osc_destroy_orphan(struct mds_obd *mds,
 105                                   struct ptlrpc_request *request)
 106 {
 107         struct mds_body *body;
 108         struct lov_mds_md *lmm = NULL;
 109         struct lov_stripe_md *lsm = NULL;
 110         struct obd_trans_info oti = { 0 };
 111         struct obdo *oa;
 112         int rc;
 113         ENTRY;
 114
 115         body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
 116         if (!(body->valid & OBD_MD_FLEASIZE))
 117                 RETURN(0);
 118         if (body->eadatasize == 0) {
 119                 CERROR("OBD_MD_FLEASIZE set but eadatasize zero\n");
 120                 RETURN(rc = -EPROTO);
 121         }
 122
 123         lmm = lustre_msg_buf(request->rq_repmsg, 1, body->eadatasize);
 124         LASSERT(lmm != NULL);
 125
 126         rc = obd_unpackmd(mds->mds_osc_exp, &lsm, lmm, body->eadatasize);
 127         if (rc < 0) {
 128                 CERROR("Error unpack md %p\n", lmm);
 129                 RETURN(rc);
 130         } else {
 131                 LASSERT(rc >= sizeof(*lsm));
 132                 rc = 0;
 133         }
 134
 135         oa = obdo_alloc();
 136         if (oa == NULL)
 137                 GOTO(out_free_memmd, rc = -ENOMEM);
 138         oa->o_id = lsm->lsm_object_id;
 139         oa->o_mode = body->mode & S_IFMT;
 140         oa->o_valid = OBD_MD_FLID | OBD_MD_FLTYPE;
 141
 142         if (body->valid & OBD_MD_FLCOOKIE) {
 143                 oa->o_valid |= OBD_MD_FLCOOKIE;
 144                 oti.oti_logcookies =
 145                         lustre_msg_buf(request->rq_repmsg, 2,
 146                                        sizeof(struct llog_cookie) *
 147                                        lsm->lsm_stripe_count);
 148                 if (oti.oti_logcookies == NULL)
 149                         oa->o_valid &= ~OBD_MD_FLCOOKIE;
 150                         body->valid &= ~OBD_MD_FLCOOKIE;
 151         }
 152
 153         rc = obd_destroy(mds->mds_osc_exp, oa, lsm, &oti);
 154         obdo_free(oa);
 155         if (rc)
 156                 CERROR("destroy orphan objid 0x"LPX64" on ost error "
 157                        "%d\n", lsm->lsm_object_id, rc);
 158 out_free_memmd:
 159         obd_free_memmd(mds->mds_osc_exp, &lsm);
 160         RETURN(rc);
 161 }
 162
 163 static int mds_unlink_orphan(struct obd_device *obd, struct dentry *dchild,
 164                              struct inode *inode, struct inode *pending_dir)
 165 {
 166         struct mds_obd *mds = &obd->u.mds;
 167         struct mds_body *body;
 168         void *handle = NULL;
 169         struct ptlrpc_request *req;
 170         int lengths[3] = {sizeof(struct mds_body),
 171                           mds->mds_max_mdsize,
 172                           mds->mds_max_cookiesize};
 173         int rc;
 174         ENTRY;
 175
 176         LASSERT(mds->mds_osc_obd != NULL);
 177         OBD_ALLOC(req, sizeof(*req));
 178         if (!req) {
 179                 CERROR("request allocation out of memory\n");
 180                 GOTO(err_alloc_req, rc = -ENOMEM);
 181         }
 182         rc = lustre_pack_reply(req, 3, lengths, NULL);
 183         if (rc) {
 184                 CERROR("cannot pack request %d\n", rc);
 185                 GOTO(out_free_req, rc);
 186         }
 187         body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
 188         LASSERT(body != NULL);
 189
 190         mds_pack_inode2body(body, inode);
 191         mds_pack_md(obd, req->rq_repmsg, 1, body, inode, 1);
 192
 193         handle = fsfilt_start(obd, pending_dir, FSFILT_OP_UNLINK_LOG, NULL);
 194         if (IS_ERR(handle)) {
 195                 rc = PTR_ERR(handle);
 196                 CERROR("error fsfilt_start: %d\n", rc);
 197                 handle = NULL;
 198                 GOTO(out_free_msg, rc);
 199         }
 200
 201         if (S_ISDIR(inode->i_mode)) {
 202                 rc = vfs_rmdir(pending_dir, dchild);
 203         } else {
 204                 rc = vfs_unlink(pending_dir, dchild);
 205         }
 206         if (rc)
 207                 CERROR("error %d unlinking orphan %*s from PENDING directory\n",
 208                        rc, dchild->d_name.len, dchild->d_name.name);
 209
 210         if ((body->valid & OBD_MD_FLEASIZE)) {
 211                 if (mds_log_op_unlink(obd, inode, req->rq_repmsg, 1) > 0)
 212                         body->valid |= OBD_MD_FLCOOKIE;
 213         }
 214
 215         if (handle) {
 216                 int err = fsfilt_commit(obd, pending_dir, handle, 0);
 217                 if (err) {
 218                         CERROR("error committing orphan unlink: %d\n", err);
 219                         rc = err;
 220                         GOTO(out_free_msg, rc);
 221                 }
 222         }
 223         rc = mds_osc_destroy_orphan(mds, req);
 224 out_free_msg:
 225         OBD_FREE(req->rq_repmsg, req->rq_replen);
 226         req->rq_repmsg = NULL;
 227 out_free_req:
 228         OBD_FREE(req, sizeof(*req));
 229 err_alloc_req:
 230         RETURN(rc);
 231 }
 232
 233 int mds_cleanup_orphans(struct obd_device *obd)
 234 {
 235         struct mds_obd *mds = &obd->u.mds;
 236         struct obd_run_ctxt saved;
 237         struct file *file;
 238         struct dentry *dchild, *dentry;
 239         struct vfsmount *mnt;
 240         struct inode *child_inode, *pending_dir = mds->mds_pending_dir->d_inode;
 241         struct l_linux_dirent *dirent, *n;
 242         struct list_head dentry_list;
 243         char d_name[LL_FID_NAMELEN];
 244         __u64 i = 0;
 245         int rc = 0, item = 0, namlen;
 246         ENTRY;
 247
 248         push_ctxt(&saved, &obd->obd_ctxt, NULL);
 249         dentry = dget(mds->mds_pending_dir);
 250         if (IS_ERR(dentry))
 251                 GOTO(err_pop, rc = PTR_ERR(dentry));
 252         mnt = mntget(mds->mds_vfsmnt);
 253         if (IS_ERR(mnt))
 254                 GOTO(err_mntget, rc = PTR_ERR(mnt));
 255
 256         file = dentry_open(mds->mds_pending_dir, mds->mds_vfsmnt,
 257                            O_RDONLY | O_LARGEFILE);
 258         if (IS_ERR(file))
 259                 GOTO(err_pop, rc = PTR_ERR(file));
 260
 261         INIT_LIST_HEAD(&dentry_list);
 262         rc = l_readdir(file, &dentry_list);
 263         filp_close(file, 0);
 264         if (rc < 0)
 265                 GOTO(err_out, rc);
 266
 267         list_for_each_entry_safe(dirent, n, &dentry_list, lld_list) {
 268                 i ++;
 269                 list_del(&dirent->lld_list);
 270
 271                 namlen = strlen(dirent->lld_name);
 272                 LASSERT(sizeof(d_name) >= namlen + 1);
 273                 strcpy(d_name, dirent->lld_name);
 274                 OBD_FREE(dirent, sizeof(*dirent));
 275
 276                 CDEBUG(D_INODE, "entry "LPU64" of PENDING DIR: %s\n",
 277                        i, d_name);
 278
 279                 if (((namlen == 1) && !strcmp(d_name, ".")) ||
 280                     ((namlen == 2) && !strcmp(d_name, ".."))) {
 281                         continue;
 282                 }
 283
 284                 down(&pending_dir->i_sem);
 285                 dchild = lookup_one_len(d_name, mds->mds_pending_dir, namlen);
 286                 if (IS_ERR(dchild)) {
 287                         up(&pending_dir->i_sem);
 288                         GOTO(err_out, rc = PTR_ERR(dchild));
 289                 }
 290                 if (!dchild->d_inode) {
 291                         CERROR("orphan %s has been removed\n", d_name);
 292                         GOTO(next, rc = 0);
 293                 }
 294
 295                 child_inode = dchild->d_inode;
 296                 if (mds_inode_is_orphan(child_inode) &&
 297                     mds_open_orphan_count(child_inode)) {
 298                         CWARN("orphan %s was re-opened during recovery\n", d_name);
 299                         GOTO(next, rc = 0);
 300                 }
 301
 302                 rc = mds_unlink_orphan(obd, dchild, child_inode, pending_dir);
 303                 if (rc == 0) {
 304                         item ++;
 305                         CWARN("removed orphan %s from MDS and OST\n", d_name);
 306                 } else {
 307                         CERROR("removed orphan %s from MDS and OST failed,"
 308                                " rc = %d\n", d_name, rc);
 309                         rc = 0;
 310                 }
 311 next:
 312                 l_dput(dchild);
 313                 up(&pending_dir->i_sem);
 314         }
 315 err_out:
 316         list_for_each_entry_safe(dirent, n, &dentry_list, lld_list) {
 317                 list_del(&dirent->lld_list);
 318                 OBD_FREE(dirent, sizeof(*dirent));
 319         }
 320 err_pop:
 321         pop_ctxt(&saved, &obd->obd_ctxt, NULL);
 322         if (rc == 0)
 323                 rc = item;
 324         RETURN(rc);
 325
 326 err_mntget:
 327         l_dput(mds->mds_pending_dir);
 328         goto err_pop;
 329 }