lustre/mds/mds_reint.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  linux/mds/mds_reint.c
   5  *  Lustre Metadata Server (mds) reintegration routines
   6  *
   7  *  Copyright (C) 2002, 2003 Cluster File Systems, Inc.
   8  *   Author: Peter Braam <braam@clusterfs.com>
   9  *   Author: Andreas Dilger <adilger@clusterfs.com>
  10  *   Author: Phil Schwan <phil@clusterfs.com>
  11  *
  12  *   This file is part of Lustre, http://www.lustre.org.
  13  *
  14  *   Lustre is free software; you can redistribute it and/or
  15  *   modify it under the terms of version 2 of the GNU General Public
  16  *   License as published by the Free Software Foundation.
  17  *
  18  *   Lustre is distributed in the hope that it will be useful,
  19  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  20  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  21  *   GNU General Public License for more details.
  22  *
  23  *   You should have received a copy of the GNU General Public License
  24  *   along with Lustre; if not, write to the Free Software
  25  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  26  */
  27
  28 #ifndef EXPORT_SYMTAB
  29 # define EXPORT_SYMTAB
  30 #endif
  31 #define DEBUG_SUBSYSTEM S_MDS
  32
  33 #include <linux/fs.h>
  34 #include <linux/obd_support.h>
  35 #include <linux/obd_class.h>
  36 #include <linux/obd.h>
  37 #include <linux/lustre_lib.h>
  38 #include <linux/lustre_idl.h>
  39 #include <linux/lustre_mds.h>
  40 #include <linux/lustre_dlm.h>
  41 #include <linux/lustre_fsfilt.h>
  42
  43 #include "mds_internal.h"
  44
  45 void mds_commit_cb(struct obd_device *obd, __u64 transno, void *data,
  46                    int error)
  47 {
  48         obd_transno_commit_cb(obd, transno, error);
  49 }
  50
  51 struct mds_logcancel_data {
  52         struct lov_mds_md      *mlcd_lmm;
  53         int                     mlcd_size;
  54         int                     mlcd_cookielen;
  55         int                     mlcd_eadatalen;
  56         struct llog_cookie      mlcd_cookies[0];
  57 };
  58
  59
  60 static void mds_cancel_cookies_cb(struct obd_device *obd, __u64 transno,
  61                                   void *cb_data, int error)
  62 {
  63         struct mds_logcancel_data *mlcd = cb_data;
  64         struct lov_stripe_md *lsm = NULL;
  65 #ifdef ENABLE_ORPHANS
  66         struct llog_ctxt *ctxt;
  67 #endif
  68         int rc;
  69
  70         obd_transno_commit_cb(obd, transno, error);
  71
  72         CDEBUG(D_HA, "cancelling %d cookies\n",
  73                (int)(mlcd->mlcd_cookielen / sizeof(*mlcd->mlcd_cookies)));
  74
  75         rc = obd_unpackmd(obd->u.mds.mds_osc_exp, &lsm, mlcd->mlcd_lmm,
  76                           mlcd->mlcd_eadatalen);
  77         if (rc < 0) {
  78                 CERROR("bad LSM cancelling %d log cookies: rc %d\n",
  79                        (int)(mlcd->mlcd_cookielen/sizeof(*mlcd->mlcd_cookies)),
  80                        rc);
  81         } else {
  82 #ifdef ENABLE_ORPHANS
  83                 ///* XXX 0 normally, SENDNOW for debug */);
  84                 ctxt = llog_get_context(obd, mlcd->mlcd_cookies[0].lgc_subsys + 1);
  85                 rc = llog_cancel(ctxt, lsm,
  86                                          mlcd->mlcd_cookielen /
  87                                          sizeof(*mlcd->mlcd_cookies),
  88                                          mlcd->mlcd_cookies, OBD_LLOG_FL_SENDNOW);
  89                 if (rc)
  90                         CERROR("error cancelling %d log cookies: rc %d\n",
  91                                (int)(mlcd->mlcd_cookielen /
  92                                      sizeof(*mlcd->mlcd_cookies)), rc);
  93 #endif
  94         }
  95
  96         OBD_FREE(mlcd, mlcd->mlcd_size);
  97 }
  98
  99 /* Assumes caller has already pushed us into the kernel context. */
 100 int mds_finish_transno(struct mds_obd *mds, struct inode *inode, void *handle,
 101                        struct ptlrpc_request *req, int rc, __u32 op_data)
 102 {
 103         struct mds_export_data *med = &req->rq_export->exp_mds_data;
 104         struct mds_client_data *mcd = med->med_mcd;
 105         struct obd_device *obd = req->rq_export->exp_obd;
 106         int err;
 107         __u64 transno;
 108         loff_t off;
 109         int log_pri = D_HA;
 110         ENTRY;
 111
 112         /* if the export has already been failed, we have no last_rcvd slot */
 113         if (req->rq_export->exp_failed) {
 114                 CERROR("committing transaction for disconnected client\n");
 115                 if (handle)
 116                         GOTO(commit, rc);
 117                 RETURN(rc);
 118         }
 119
 120         if (IS_ERR(handle))
 121                 RETURN(rc);
 122
 123         if (handle == NULL) {
 124                 /* if we're starting our own xaction, use our own inode */
 125                 inode = mds->mds_rcvd_filp->f_dentry->d_inode;
 126                 handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR, NULL);
 127                 if (IS_ERR(handle)) {
 128                         CERROR("fsfilt_start: %ld\n", PTR_ERR(handle));
 129                         RETURN(PTR_ERR(handle));
 130                 }
 131         }
 132
 133         off = med->med_off;
 134
 135         transno = req->rq_reqmsg->transno;
 136         if (transno == 0) {
 137                 spin_lock(&mds->mds_transno_lock);
 138                 transno = ++mds->mds_last_transno;
 139                 spin_unlock(&mds->mds_transno_lock);
 140         } else {
 141                 spin_lock(&mds->mds_transno_lock);
 142                 if (transno > mds->mds_last_transno)
 143                         mds->mds_last_transno = transno;
 144                 spin_unlock(&mds->mds_transno_lock);
 145         }
 146         req->rq_repmsg->transno = req->rq_transno = transno;
 147         mcd->mcd_last_transno = cpu_to_le64(transno);
 148         mcd->mcd_mount_count = cpu_to_le64(mds->mds_mount_count);
 149         mcd->mcd_last_xid = cpu_to_le64(req->rq_xid);
 150         mcd->mcd_last_result = cpu_to_le32(rc);
 151         mcd->mcd_last_data = cpu_to_le32(op_data);
 152
 153         fsfilt_add_journal_cb(req->rq_export->exp_obd, transno, handle,
 154                               mds_commit_cb, NULL);
 155         err = fsfilt_write_record(obd, mds->mds_rcvd_filp, mcd, sizeof(*mcd),
 156                                   &off, 0);
 157
 158         if (err) {
 159                 log_pri = D_ERROR;
 160                 if (rc == 0)
 161                         rc = err;
 162         }
 163
 164         DEBUG_REQ(log_pri, req,
 165                   "wrote trans #"LPU64" client %s at idx %u: err = %d\n",
 166                   transno, mcd->mcd_uuid, med->med_idx, err);
 167
 168         err = mds_lov_write_objids(obd);
 169         if (err) {
 170                 log_pri = D_ERROR;
 171                 if (rc == 0)
 172                         rc = err;
 173         }
 174         CDEBUG(log_pri, "wrote objids: err = %d\n", err);
 175
 176 commit:
 177         err = fsfilt_commit(obd, inode, handle, 0);
 178         if (err) {
 179                 CERROR("error committing transaction: %d\n", err);
 180                 if (!rc)
 181                         rc = err;
 182         }
 183
 184         RETURN(rc);
 185 }
 186
 187 /* this gives the same functionality as the code between
 188  * sys_chmod and inode_setattr
 189  * chown_common and inode_setattr
 190  * utimes and inode_setattr
 191  */
 192 int mds_fix_attr(struct inode *inode, struct mds_update_record *rec)
 193 {
 194         time_t now = LTIME_S(CURRENT_TIME);
 195         struct iattr *attr = &rec->ur_iattr;
 196         unsigned int ia_valid = attr->ia_valid;
 197         int error;
 198         ENTRY;
 199
 200         /* only fix up attrs if the client VFS didn't already */
 201         if (!(ia_valid & ATTR_RAW))
 202                 RETURN(0);
 203
 204         if (!(ia_valid & ATTR_CTIME_SET))
 205                 LTIME_S(attr->ia_ctime) = now;
 206         if (!(ia_valid & ATTR_ATIME_SET))
 207                 LTIME_S(attr->ia_atime) = now;
 208         if (!(ia_valid & ATTR_MTIME_SET))
 209                 LTIME_S(attr->ia_mtime) = now;
 210
 211         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 212                 RETURN(-EPERM);
 213
 214         /* times */
 215         if ((ia_valid & (ATTR_MTIME|ATTR_ATIME)) == (ATTR_MTIME|ATTR_ATIME)) {
 216                 if (rec->ur_fsuid != inode->i_uid &&
 217                     (error = ll_permission(inode, MAY_WRITE, NULL)) != 0)
 218                         RETURN(error);
 219         }
 220
 221         if (ia_valid & ATTR_SIZE) {
 222                 if ((error = ll_permission(inode, MAY_WRITE, NULL)) != 0)
 223                         RETURN(error);
 224         }
 225
 226         if (ia_valid & ATTR_UID) {
 227                 /* chown */
 228                 error = -EPERM;
 229                 if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
 230                         RETURN(-EPERM);
 231                 if (attr->ia_uid == (uid_t) -1)
 232                         attr->ia_uid = inode->i_uid;
 233                 if (attr->ia_gid == (gid_t) -1)
 234                         attr->ia_gid = inode->i_gid;
 235                 attr->ia_mode = inode->i_mode;
 236                 /*
 237                  * If the user or group of a non-directory has been
 238                  * changed by a non-root user, remove the setuid bit.
 239                  * 19981026 David C Niemi <niemi@tux.org>
 240                  *
 241                  * Changed this to apply to all users, including root,
 242                  * to avoid some races. This is the behavior we had in
 243                  * 2.0. The check for non-root was definitely wrong
 244                  * for 2.2 anyway, as it should have been using
 245                  * CAP_FSETID rather than fsuid -- 19990830 SD.
 246                  */
 247                 if ((inode->i_mode & S_ISUID) == S_ISUID &&
 248                     !S_ISDIR(inode->i_mode)) {
 249                         attr->ia_mode &= ~S_ISUID;
 250                         attr->ia_valid |= ATTR_MODE;
 251                 }
 252                 /*
 253                  * Likewise, if the user or group of a non-directory
 254                  * has been changed by a non-root user, remove the
 255                  * setgid bit UNLESS there is no group execute bit
 256                  * (this would be a file marked for mandatory
 257                  * locking).  19981026 David C Niemi <niemi@tux.org>
 258                  *
 259                  * Removed the fsuid check (see the comment above) --
 260                  * 19990830 SD.
 261                  */
 262                 if (((inode->i_mode & (S_ISGID | S_IXGRP)) ==
 263                      (S_ISGID | S_IXGRP)) && !S_ISDIR(inode->i_mode)) {
 264                         attr->ia_mode &= ~S_ISGID;
 265                         attr->ia_valid |= ATTR_MODE;
 266                 }
 267         } else if (ia_valid & ATTR_MODE) {
 268                 int mode = attr->ia_mode;
 269                 /* chmod */
 270                 if (attr->ia_mode == (mode_t) -1)
 271                         attr->ia_mode = inode->i_mode;
 272                 attr->ia_mode =
 273                         (mode & S_IALLUGO) | (inode->i_mode & ~S_IALLUGO);
 274         }
 275         RETURN(0);
 276 }
 277
 278 void mds_steal_ack_locks(struct obd_export *exp, struct ptlrpc_request *req)
 279 {
 280         unsigned long flags;
 281         struct ptlrpc_request *oldrep = exp->exp_outstanding_reply;
 282
 283         if (oldrep == NULL)
 284                 return;
 285         memcpy(req->rq_ack_locks, oldrep->rq_ack_locks,
 286                sizeof req->rq_ack_locks);
 287         spin_lock_irqsave(&req->rq_lock, flags);
 288         oldrep->rq_resent = 1;
 289         wake_up(&oldrep->rq_reply_waitq);
 290         spin_unlock_irqrestore(&req->rq_lock, flags);
 291         DEBUG_REQ(D_HA, oldrep, "stole locks from");
 292         DEBUG_REQ(D_HA, req, "stole locks for");
 293 }
 294
 295 void mds_req_from_mcd(struct ptlrpc_request *req, struct mds_client_data *mcd)
 296 {
 297         DEBUG_REQ(D_HA, req, "restoring transno "LPD64"/status %d",
 298                   mcd->mcd_last_transno, mcd->mcd_last_result);
 299         req->rq_repmsg->transno = req->rq_transno = mcd->mcd_last_transno;
 300         req->rq_repmsg->status = req->rq_status = mcd->mcd_last_result;
 301
 302         if (req->rq_export->exp_outstanding_reply)
 303                 mds_steal_ack_locks(req->rq_export, req);
 304 }
 305
 306 static void reconstruct_reint_setattr(struct mds_update_record *rec,
 307                                       int offset, struct ptlrpc_request *req)
 308 {
 309         struct mds_export_data *med = &req->rq_export->exp_mds_data;
 310         struct mds_obd *obd = &req->rq_export->exp_obd->u.mds;
 311         struct dentry *de;
 312         struct mds_body *body;
 313
 314         mds_req_from_mcd(req, med->med_mcd);
 315
 316         de = mds_fid2dentry(obd, rec->ur_fid1, NULL);
 317         if (IS_ERR(de)) {
 318                 LASSERT(PTR_ERR(de) == req->rq_status);
 319                 return;
 320         }
 321
 322         body = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*body));
 323         mds_pack_inode2fid(&body->fid1, de->d_inode);
 324         mds_pack_inode2body(body, de->d_inode);
 325
 326         /* Don't return OST-specific attributes if we didn't just set them */
 327         if (rec->ur_iattr.ia_valid & ATTR_SIZE)
 328                 body->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 329         if (rec->ur_iattr.ia_valid & (ATTR_MTIME | ATTR_MTIME_SET))
 330                 body->valid |= OBD_MD_FLMTIME;
 331         if (rec->ur_iattr.ia_valid & (ATTR_ATIME | ATTR_ATIME_SET))
 332                 body->valid |= OBD_MD_FLATIME;
 333
 334         l_dput(de);
 335 }
 336
 337 /* In the raw-setattr case, we lock the child inode.
 338  * In the write-back case or if being called from open, the client holds a lock
 339  * already.
 340  *
 341  * We use the ATTR_FROM_OPEN flag to tell these cases apart. */
 342 static int mds_reint_setattr(struct mds_update_record *rec, int offset,
 343                              struct ptlrpc_request *req,
 344                              struct lustre_handle *lh)
 345 {
 346         struct mds_obd *mds = mds_req2mds(req);
 347         struct obd_device *obd = req->rq_export->exp_obd;
 348         struct mds_body *body;
 349         struct dentry *de;
 350         struct inode *inode = NULL;
 351         struct lustre_handle lockh;
 352         void *handle = NULL;
 353         struct mds_logcancel_data *mlcd = NULL;
 354         int rc = 0, cleanup_phase = 0, err, locked = 0;
 355         ENTRY;
 356
 357         LASSERT(offset == 0);
 358
 359         DEBUG_REQ(D_INODE, req, "setattr "LPU64"/%u %x\n", rec->ur_fid1->id,
 360                   rec->ur_fid1->generation, rec->ur_iattr.ia_valid);
 361
 362         MDS_CHECK_RESENT(req, reconstruct_reint_setattr(rec, offset, req));
 363
 364         if (rec->ur_iattr.ia_valid & ATTR_FROM_OPEN) {
 365                 de = mds_fid2dentry(mds, rec->ur_fid1, NULL);
 366                 if (IS_ERR(de))
 367                         GOTO(cleanup, rc = PTR_ERR(de));
 368         } else {
 369                 de = mds_fid2locked_dentry(obd, rec->ur_fid1, NULL, LCK_PW,
 370                                            &lockh, NULL, 0);
 371                 if (IS_ERR(de))
 372                         GOTO(cleanup, rc = PTR_ERR(de));
 373                 locked = 1;
 374         }
 375
 376         cleanup_phase = 1;
 377         inode = de->d_inode;
 378         LASSERT(inode);
 379         if (S_ISREG(inode->i_mode) && rec->ur_eadata != NULL)
 380                 down(&inode->i_sem);
 381
 382         OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_SETATTR_WRITE, inode->i_sb);
 383
 384         handle = fsfilt_start(obd, inode, FSFILT_OP_SETATTR, NULL);
 385         if (IS_ERR(handle))
 386                 GOTO(cleanup, rc = PTR_ERR(handle));
 387
 388         if (rec->ur_iattr.ia_valid & (ATTR_MTIME | ATTR_CTIME))
 389                 CDEBUG(D_INODE, "setting mtime %lu, ctime %lu\n",
 390                        LTIME_S(rec->ur_iattr.ia_mtime),
 391                        LTIME_S(rec->ur_iattr.ia_ctime));
 392         rc = mds_fix_attr(inode, rec);
 393         if (rc)
 394                 GOTO(cleanup, rc);
 395
 396         if (rec->ur_iattr.ia_valid & ATTR_ATTR_FLAG)    /* ioctl */
 397                 rc = fsfilt_iocontrol(obd, inode, NULL, EXT3_IOC_SETFLAGS,
 398                                       (long)&rec->ur_iattr.ia_attr_flags);
 399         else                                            /* setattr */
 400                 rc = fsfilt_setattr(obd, de, handle, &rec->ur_iattr, 0);
 401
 402         if (rc == 0 && S_ISREG(inode->i_mode) && rec->ur_eadata != NULL) {
 403                 rc = fsfilt_set_md(obd, inode, handle,
 404                                    rec->ur_eadata, rec->ur_eadatalen);
 405         }
 406
 407         body = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*body));
 408         mds_pack_inode2fid(&body->fid1, inode);
 409         mds_pack_inode2body(body, inode);
 410
 411         /* Don't return OST-specific attributes if we didn't just set them */
 412         if (rec->ur_iattr.ia_valid & ATTR_SIZE)
 413                 body->valid |= OBD_MD_FLSIZE | OBD_MD_FLBLOCKS;
 414         if (rec->ur_iattr.ia_valid & (ATTR_MTIME | ATTR_MTIME_SET))
 415                 body->valid |= OBD_MD_FLMTIME;
 416         if (rec->ur_iattr.ia_valid & (ATTR_ATIME | ATTR_ATIME_SET))
 417                 body->valid |= OBD_MD_FLATIME;
 418
 419         if (rc == 0 && rec->ur_cookielen && !IS_ERR(mds->mds_osc_obd)) {
 420                 OBD_ALLOC(mlcd, sizeof(*mlcd) + rec->ur_cookielen +
 421                           rec->ur_eadatalen);
 422                 if (mlcd) {
 423                         mlcd->mlcd_size = sizeof(*mlcd) + rec->ur_cookielen +
 424                                 rec->ur_eadatalen;
 425                         mlcd->mlcd_eadatalen = rec->ur_eadatalen;
 426                         mlcd->mlcd_cookielen = rec->ur_cookielen;
 427                         mlcd->mlcd_lmm = (void *)&mlcd->mlcd_cookies +
 428                                 mlcd->mlcd_cookielen;
 429                         memcpy(&mlcd->mlcd_cookies, rec->ur_logcookies,
 430                                mlcd->mlcd_cookielen);
 431                         memcpy(mlcd->mlcd_lmm, rec->ur_eadata,
 432                                mlcd->mlcd_eadatalen);
 433                 } else {
 434                         CERROR("unable to allocate log cancel data\n");
 435                 }
 436         }
 437         EXIT;
 438  cleanup:
 439         if (mlcd != NULL)
 440                 fsfilt_add_journal_cb(req->rq_export->exp_obd, 0, handle,
 441                                       mds_cancel_cookies_cb, mlcd);
 442         err = mds_finish_transno(mds, inode, handle, req, rc, 0);
 443         switch (cleanup_phase) {
 444         case 1:
 445                 if (S_ISREG(inode->i_mode) && rec->ur_eadata != NULL)
 446                         down(&inode->i_sem);
 447                 l_dput(de);
 448                 if (locked) {
 449                         if (rc) {
 450                                 ldlm_lock_decref(&lockh, LCK_PW);
 451                         } else {
 452                                 ldlm_put_lock_into_req(req, &lockh, LCK_PW);
 453                         }
 454                 }
 455         case 0:
 456                 break;
 457         default:
 458                 LBUG();
 459         }
 460         if (err && !rc)
 461                 rc = err;
 462
 463         req->rq_status = rc;
 464         return 0;
 465 }
 466
 467 static void reconstruct_reint_create(struct mds_update_record *rec, int offset,
 468                                      struct ptlrpc_request *req)
 469 {
 470         struct mds_export_data *med = &req->rq_export->exp_mds_data;
 471         struct mds_obd *obd = &req->rq_export->exp_obd->u.mds;
 472         struct dentry *parent, *child;
 473         struct mds_body *body;
 474
 475         mds_req_from_mcd(req, med->med_mcd);
 476
 477         if (req->rq_status)
 478                 return;
 479
 480         parent = mds_fid2dentry(obd, rec->ur_fid1, NULL);
 481         LASSERT(!IS_ERR(parent));
 482         child = ll_lookup_one_len(rec->ur_name, parent, rec->ur_namelen - 1);
 483         LASSERT(!IS_ERR(child));
 484         body = lustre_msg_buf(req->rq_repmsg, offset, sizeof (*body));
 485         mds_pack_inode2fid(&body->fid1, child->d_inode);
 486         mds_pack_inode2body(body, child->d_inode);
 487         l_dput(parent);
 488         l_dput(child);
 489 }
 490
 491 static int mds_reint_create(struct mds_update_record *rec, int offset,
 492                             struct ptlrpc_request *req,
 493                             struct lustre_handle *lh)
 494 {
 495         struct dentry *dparent = NULL;
 496         struct mds_obd *mds = mds_req2mds(req);
 497         struct obd_device *obd = req->rq_export->exp_obd;
 498         struct dentry *dchild = NULL;
 499         struct inode *dir = NULL;
 500         void *handle = NULL;
 501         struct lustre_handle lockh;
 502         int rc = 0, err, type = rec->ur_mode & S_IFMT, cleanup_phase = 0;
 503         int created = 0;
 504         struct dentry_params dp;
 505         ENTRY;
 506
 507         LASSERT(offset == 0);
 508         LASSERT(!strcmp(req->rq_export->exp_obd->obd_type->typ_name, "mds"));
 509
 510         DEBUG_REQ(D_INODE, req, "parent "LPU64"/%u name %s mode %o\n",
 511                   rec->ur_fid1->id, rec->ur_fid1->generation,
 512                   rec->ur_name, rec->ur_mode);
 513
 514         MDS_CHECK_RESENT(req, reconstruct_reint_create(rec, offset, req));
 515
 516         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_CREATE))
 517                 GOTO(cleanup, rc = -ESTALE);
 518
 519         dparent = mds_fid2locked_dentry(obd, rec->ur_fid1, NULL, LCK_PW, &lockh,
 520                                         rec->ur_name, rec->ur_namelen - 1);
 521         if (IS_ERR(dparent)) {
 522                 rc = PTR_ERR(dparent);
 523                 CERROR("parent lookup error %d\n", rc);
 524                 GOTO(cleanup, rc);
 525         }
 526         cleanup_phase = 1; /* locked parent dentry */
 527         dir = dparent->d_inode;
 528         LASSERT(dir);
 529
 530         ldlm_lock_dump_handle(D_OTHER, &lockh);
 531
 532         dchild = ll_lookup_one_len(rec->ur_name, dparent, rec->ur_namelen - 1);
 533         if (IS_ERR(dchild)) {
 534                 rc = PTR_ERR(dchild);
 535                 CERROR("child lookup error %d\n", rc);
 536                 GOTO(cleanup, rc);
 537         }
 538
 539         cleanup_phase = 2; /* child dentry */
 540
 541         OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_CREATE_WRITE, dir->i_sb);
 542
 543         if (dir->i_mode & S_ISGID) {
 544                 if (S_ISDIR(rec->ur_mode))
 545                         rec->ur_mode |= S_ISGID;
 546         }
 547
 548         dchild->d_fsdata = (void *)&dp;
 549         dp.p_inum = (unsigned long)rec->ur_fid2->id;
 550         dp.p_ptr = req;
 551
 552         switch (type) {
 553         case S_IFREG:{
 554                 handle = fsfilt_start(obd, dir, FSFILT_OP_CREATE, NULL);
 555                 if (IS_ERR(handle))
 556                         GOTO(cleanup, rc = PTR_ERR(handle));
 557                 rc = ll_vfs_create(dir, dchild, rec->ur_mode, NULL);
 558                 EXIT;
 559                 break;
 560         }
 561         case S_IFDIR:{
 562                 handle = fsfilt_start(obd, dir, FSFILT_OP_MKDIR, NULL);
 563                 if (IS_ERR(handle))
 564                         GOTO(cleanup, rc = PTR_ERR(handle));
 565                 rc = vfs_mkdir(dir, dchild, rec->ur_mode);
 566                 EXIT;
 567                 break;
 568         }
 569         case S_IFLNK:{
 570                 handle = fsfilt_start(obd, dir, FSFILT_OP_SYMLINK, NULL);
 571                 if (IS_ERR(handle))
 572                         GOTO(cleanup, rc = PTR_ERR(handle));
 573                 if (rec->ur_tgt == NULL)        /* no target supplied */
 574                         rc = -EINVAL;           /* -EPROTO? */
 575                 else
 576                         rc = vfs_symlink(dir, dchild, rec->ur_tgt);
 577                 EXIT;
 578                 break;
 579         }
 580         case S_IFCHR:
 581         case S_IFBLK:
 582         case S_IFIFO:
 583         case S_IFSOCK:{
 584                 int rdev = rec->ur_rdev;
 585                 handle = fsfilt_start(obd, dir, FSFILT_OP_MKNOD, NULL);
 586                 if (IS_ERR(handle))
 587                         GOTO(cleanup, (handle = NULL, rc = PTR_ERR(handle)));
 588                 rc = vfs_mknod(dir, dchild, rec->ur_mode, rdev);
 589                 EXIT;
 590                 break;
 591         }
 592         default:
 593                 CERROR("bad file type %o creating %s\n", type, rec->ur_name);
 594                 dchild->d_fsdata = NULL;
 595                 GOTO(cleanup, rc = -EINVAL);
 596         }
 597
 598         /* In case we stored the desired inum in here, we want to clean up. */
 599         if (dchild->d_fsdata == (void *)(unsigned long)rec->ur_fid2->id)
 600                 dchild->d_fsdata = NULL;
 601
 602         if (rc) {
 603                 CDEBUG(D_INODE, "error during create: %d\n", rc);
 604                 GOTO(cleanup, rc);
 605         } else {
 606                 struct iattr iattr;
 607                 struct inode *inode = dchild->d_inode;
 608                 struct mds_body *body;
 609
 610                 created = 1;
 611                 LTIME_S(iattr.ia_atime) = rec->ur_time;
 612                 LTIME_S(iattr.ia_ctime) = rec->ur_time;
 613                 LTIME_S(iattr.ia_mtime) = rec->ur_time;
 614                 iattr.ia_uid = rec->ur_fsuid;
 615                 if (dir->i_mode & S_ISGID)
 616                         iattr.ia_gid = dir->i_gid;
 617                 else
 618                         iattr.ia_gid = rec->ur_fsgid;
 619                 iattr.ia_valid = ATTR_UID | ATTR_GID | ATTR_ATIME |
 620                         ATTR_MTIME | ATTR_CTIME;
 621
 622                 if (rec->ur_fid2->id) {
 623                         LASSERT(rec->ur_fid2->id == inode->i_ino);
 624                         inode->i_generation = rec->ur_fid2->generation;
 625                         /* Dirtied and committed by the upcoming setattr. */
 626                         CDEBUG(D_INODE, "recreated ino %lu with gen %u\n",
 627                                inode->i_ino, inode->i_generation);
 628                 } else {
 629                         struct lustre_handle child_ino_lockh;
 630                         struct ldlm_res_id child_res_id =
 631                              { .name = { inode->i_ino, 0 } };
 632                         int lock_flags = 0;
 633
 634                         CDEBUG(D_INODE, "created ino %lu with gen %x\n",
 635                                inode->i_ino, inode->i_generation);
 636
 637                         /* The inode we were allocated may have just been freed
 638                          * by an unlink operation.  We take this lock to
 639                          * synchronize against the matching reply-ack-lock taken
 640                          * in unlink, to avoid replay problems if this reply
 641                          * makes it out to the client but the unlink's does not.
 642                          * See bug 2029 for more detail.*/
 643                         rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace,
 644                                               NULL, child_res_id, LDLM_PLAIN,
 645                                               NULL, 0, LCK_EX, &lock_flags,
 646                                               ldlm_completion_ast,
 647                                               mds_blocking_ast, NULL,
 648                                               &child_ino_lockh);
 649                         if (rc != ELDLM_OK) {
 650                                 CERROR("error locking for unlink/create sync: "
 651                                        "%d\n", rc);
 652                         } else {
 653                                 ldlm_lock_decref(&child_ino_lockh, LCK_EX);
 654                         }
 655                 }
 656
 657                 rc = fsfilt_setattr(obd, dchild, handle, &iattr, 0);
 658                 if (rc)
 659                         CERROR("error on child setattr: rc = %d\n", rc);
 660
 661                 iattr.ia_valid = ATTR_MTIME | ATTR_CTIME;
 662                 rc = fsfilt_setattr(obd, dparent, handle, &iattr, 0);
 663                 if (rc)
 664                         CERROR("error on parent setattr: rc = %d\n", rc);
 665
 666                 body = lustre_msg_buf(req->rq_repmsg, offset, sizeof (*body));
 667                 mds_pack_inode2fid(&body->fid1, inode);
 668                 mds_pack_inode2body(body, inode);
 669         }
 670         EXIT;
 671
 672 cleanup:
 673         err = mds_finish_transno(mds, dir, handle, req, rc, 0);
 674
 675         if (rc && created) {
 676                 /* Destroy the file we just created.  This should not need
 677                  * extra journal credits, as we have already modified all of
 678                  * the blocks needed in order to create the file in the first
 679                  * place.
 680                  */
 681                 switch (type) {
 682                 case S_IFDIR:
 683                         err = vfs_rmdir(dir, dchild);
 684                         if (err)
 685                                 CERROR("rmdir in error path: %d\n", err);
 686                         break;
 687                 default:
 688                         err = vfs_unlink(dir, dchild);
 689                         if (err)
 690                                 CERROR("unlink in error path: %d\n", err);
 691                         break;
 692                 }
 693         } else {
 694                 rc = err;
 695         }
 696         switch (cleanup_phase) {
 697         case 2: /* child dentry */
 698                 l_dput(dchild);
 699         case 1: /* locked parent dentry */
 700                 if (rc) {
 701                         ldlm_lock_decref(&lockh, LCK_PW);
 702                 } else {
 703                         ldlm_put_lock_into_req(req, &lockh, LCK_PW);
 704                 }
 705                 l_dput(dparent);
 706         case 0:
 707                 break;
 708         default:
 709                 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
 710                 LBUG();
 711         }
 712         req->rq_status = rc;
 713         return 0;
 714 }
 715
 716 int res_gt(struct ldlm_res_id *res1, struct ldlm_res_id *res2)
 717 {
 718         int i;
 719
 720         for (i = 0; i < RES_NAME_SIZE; i++) {
 721                 /* return 1 here, because enqueue_ordered will skip resources
 722                  * of all zeroes if they're sorted to the end of the list. */
 723                 if (res1->name[i] == 0 && res2->name[i] != 0)
 724                         return 1;
 725                 if (res2->name[i] == 0 && res1->name[i] != 0)
 726                         return 0;
 727
 728                 if (res1->name[i] > res2->name[i])
 729                         return 1;
 730                 if (res1->name[i] < res2->name[i])
 731                         return 0;
 732         }
 733         return 0;
 734 }
 735
 736 /* This function doesn't use ldlm_match_or_enqueue because we're always called
 737  * with EX or PW locks, and the MDS is no longer allowed to match write locks,
 738  * because they take the place of local semaphores.
 739  *
 740  * One or two locks are taken in numerical order.  A res_id->name[0] of 0 means
 741  * no lock is taken for that res_id.  Must be at least one non-zero res_id. */
 742 int enqueue_ordered_locks(struct obd_device *obd, struct ldlm_res_id *p1_res_id,
 743                           struct lustre_handle *p1_lockh, int p1_lock_mode,
 744                           struct ldlm_res_id *p2_res_id,
 745                           struct lustre_handle *p2_lockh, int p2_lock_mode)
 746 {
 747         struct ldlm_res_id *res_id[2] = { p1_res_id, p2_res_id };
 748         struct lustre_handle *handles[2] = { p1_lockh, p2_lockh };
 749         int lock_modes[2] = { p1_lock_mode, p2_lock_mode };
 750         int rc, flags;
 751         ENTRY;
 752
 753         LASSERT(p1_res_id != NULL && p2_res_id != NULL);
 754
 755         CDEBUG(D_INFO, "locks before: "LPU64"/"LPU64"\n",
 756                res_id[0]->name[0], res_id[1]->name[0]);
 757
 758         if (res_gt(p1_res_id, p2_res_id)) {
 759                 handles[1] = p1_lockh;
 760                 handles[0] = p2_lockh;
 761                 res_id[1] = p1_res_id;
 762                 res_id[0] = p2_res_id;
 763                 lock_modes[1] = p1_lock_mode;
 764                 lock_modes[0] = p2_lock_mode;
 765         }
 766
 767         CDEBUG(D_DLMTRACE, "lock order: "LPU64"/"LPU64"\n",
 768                res_id[0]->name[0], res_id[1]->name[0]);
 769
 770         flags = LDLM_FL_LOCAL_ONLY;
 771         rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, NULL, *res_id[0],
 772                               LDLM_PLAIN, NULL, 0, lock_modes[0], &flags,
 773                               ldlm_completion_ast, mds_blocking_ast, NULL,
 774                               handles[0]);
 775         if (rc != ELDLM_OK)
 776                 RETURN(-EIO);
 777         ldlm_lock_dump_handle(D_OTHER, handles[0]);
 778
 779         if (memcmp(res_id[0], res_id[1], sizeof(*res_id[0])) == 0) {
 780                 memcpy(handles[1], handles[0], sizeof(*(handles[1])));
 781                 ldlm_lock_addref(handles[1], lock_modes[1]);
 782         } else if (res_id[1]->name[0] != 0) {
 783                 flags = LDLM_FL_LOCAL_ONLY;
 784                 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace, NULL,
 785                                       *res_id[1], LDLM_PLAIN, NULL, 0,
 786                                       lock_modes[1], &flags,ldlm_completion_ast,
 787                                       mds_blocking_ast, NULL, handles[1]);
 788                 if (rc != ELDLM_OK) {
 789                         ldlm_lock_decref(handles[0], lock_modes[0]);
 790                         RETURN(-EIO);
 791                 }
 792                 ldlm_lock_dump_handle(D_OTHER, handles[1]);
 793         }
 794
 795         RETURN(0);
 796 }
 797
 798 int enqueue_4ordered_locks(struct obd_device *obd,struct ldlm_res_id *p1_res_id,
 799                            struct lustre_handle *p1_lockh, int p1_lock_mode,
 800                            struct ldlm_res_id *p2_res_id,
 801                            struct lustre_handle *p2_lockh, int p2_lock_mode,
 802                            struct ldlm_res_id *c1_res_id,
 803                            struct lustre_handle *c1_lockh, int c1_lock_mode,
 804                            struct ldlm_res_id *c2_res_id,
 805                            struct lustre_handle *c2_lockh, int c2_lock_mode)
 806 {
 807         struct ldlm_res_id *res_id[5] = { p1_res_id, p2_res_id,
 808                                           c1_res_id, c2_res_id };
 809         struct lustre_handle *dlm_handles[5] = { p1_lockh, p2_lockh,
 810                                                  c1_lockh, c2_lockh };
 811         int lock_modes[5] = { p1_lock_mode, p2_lock_mode,
 812                               c1_lock_mode, c2_lock_mode };
 813         int rc, i, j, sorted, flags;
 814         ENTRY;
 815
 816         CDEBUG(D_DLMTRACE, "locks before: "LPU64"/"LPU64"/"LPU64"/"LPU64"\n",
 817                res_id[0]->name[0], res_id[1]->name[0], res_id[2]->name[0],
 818                res_id[3]->name[0]);
 819
 820         /* simple insertion sort - we have at most 4 elements */
 821         for (i = 1; i < 4; i++) {
 822                 j = i - 1;
 823                 dlm_handles[4] = dlm_handles[i];
 824                 res_id[4] = res_id[i];
 825                 lock_modes[4] = lock_modes[i];
 826
 827                 sorted = 0;
 828                 do {
 829                         if (res_gt(res_id[j], res_id[4])) {
 830                                 dlm_handles[j + 1] = dlm_handles[j];
 831                                 res_id[j + 1] = res_id[j];
 832                                 lock_modes[j + 1] = lock_modes[j];
 833                                 j--;
 834                         } else {
 835                                 sorted = 1;
 836                         }
 837                 } while (j >= 0 && !sorted);
 838
 839                 dlm_handles[j + 1] = dlm_handles[4];
 840                 res_id[j + 1] = res_id[4];
 841                 lock_modes[j + 1] = lock_modes[4];
 842         }
 843
 844         CDEBUG(D_DLMTRACE, "lock order: "LPU64"/"LPU64"/"LPU64"/"LPU64"\n",
 845                res_id[0]->name[0], res_id[1]->name[0], res_id[2]->name[0],
 846                res_id[3]->name[0]);
 847
 848         /* XXX we could send ASTs on all these locks first before blocking? */
 849         for (i = 0; i < 4; i++) {
 850                 flags = 0;
 851                 if (res_id[i]->name[0] == 0)
 852                         break;
 853                 if (i != 0 &&
 854                     memcmp(res_id[i], res_id[i-1], sizeof(*res_id[i])) == 0) {
 855                         memcpy(dlm_handles[i], dlm_handles[i-1],
 856                                sizeof(*(dlm_handles[i])));
 857                         ldlm_lock_addref(dlm_handles[i], lock_modes[i]);
 858                 } else {
 859                         rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace,
 860                                               NULL, *res_id[i], LDLM_PLAIN,
 861                                               NULL, 0, lock_modes[i], &flags,
 862                                               ldlm_completion_ast,
 863                                               mds_blocking_ast, NULL,
 864                                               dlm_handles[i]);
 865                         if (rc != ELDLM_OK)
 866                                 GOTO(out_err, rc = -EIO);
 867                         ldlm_lock_dump_handle(D_OTHER, dlm_handles[i]);
 868                 }
 869         }
 870
 871         RETURN(0);
 872 out_err:
 873         while (i-- > 0)
 874                 ldlm_lock_decref(dlm_handles[i], lock_modes[i]);
 875
 876         return rc;
 877 }
 878
 879 /* In the unlikely case that the child changed while we were waiting
 880  * on the lock, we need to drop the lock on the old child and either:
 881  * - if the child has a lower resource name, then we have to also
 882  *   drop the parent lock and regain the locks in the right order
 883  * - in the rename case, if the child has a lower resource name than one of
 884  *   the other parent/child resources (maxres) we also need to reget the locks
 885  * - if the child has a higher resource name (this is the common case)
 886  *   we can just get the lock on the new child (still in lock order)
 887  *
 888  * Returns 0 if the child did not change or if it changed but could be locked.
 889  * Returns 1 if the child changed and we need to re-lock (no locks held).
 890  * Returns -ve error with a valid dchild (no locks held). */
 891 static int mds_verify_child(struct obd_device *obd,
 892                             struct ldlm_res_id *parent_res_id,
 893                             struct lustre_handle *parent_lockh,
 894                             struct dentry *dparent, int parent_mode,
 895                             struct ldlm_res_id *child_res_id,
 896                             struct lustre_handle *child_lockh,
 897                             struct dentry **dchildp, int child_mode,
 898                             const char *name, int namelen,
 899                             struct ldlm_res_id *maxres)
 900 {
 901         struct dentry *vchild, *dchild = *dchildp;
 902         int rc = 0, cleanup_phase = 2; /* parent, child locks */
 903         ENTRY;
 904
 905         vchild = ll_lookup_one_len(name, dparent, namelen - 1);
 906         if (IS_ERR(vchild))
 907                 GOTO(cleanup, rc = PTR_ERR(vchild));
 908
 909         if (likely((vchild->d_inode == NULL && child_res_id->name[0] == 0) ||
 910                    (vchild->d_inode != NULL &&
 911                     child_res_id->name[0] == vchild->d_inode->i_ino &&
 912                     child_res_id->name[1] == vchild->d_inode->i_generation))) {
 913                 if (dchild != NULL)
 914                         l_dput(dchild);
 915                 *dchildp = vchild;
 916
 917                 RETURN(0);
 918         }
 919
 920         CDEBUG(D_DLMTRACE, "child inode changed: %p != %p (%lu != "LPU64")\n",
 921                vchild->d_inode, dchild ? dchild->d_inode : 0,
 922                vchild->d_inode ? vchild->d_inode->i_ino : 0,
 923                child_res_id->name[0]);
 924         if (child_res_id->name[0] != 0)
 925                 ldlm_lock_decref(child_lockh, child_mode);
 926         if (dchild)
 927                 l_dput(dchild);
 928
 929         cleanup_phase = 1; /* parent lock only */
 930         *dchildp = dchild = vchild;
 931
 932         if (dchild->d_inode) {
 933                 int flags = 0;
 934                 child_res_id->name[0] = dchild->d_inode->i_ino;
 935                 child_res_id->name[1] = dchild->d_inode->i_generation;
 936
 937                 if (res_gt(parent_res_id, child_res_id) ||
 938                     res_gt(maxres, child_res_id)) {
 939                         CDEBUG(D_DLMTRACE, "relock "LPU64"<("LPU64"|"LPU64")\n",
 940                                child_res_id->name[0], parent_res_id->name[0],
 941                                maxres->name[0]);
 942                         GOTO(cleanup, rc = 1);
 943                 }
 944
 945                 rc = ldlm_cli_enqueue(NULL, NULL, obd->obd_namespace,
 946                                       NULL, *child_res_id, LDLM_PLAIN,
 947                                       NULL, 0, child_mode, &flags,
 948                                       ldlm_completion_ast, mds_blocking_ast,
 949                                       NULL, child_lockh);
 950                 if (rc != ELDLM_OK)
 951                         GOTO(cleanup, rc = -EIO);
 952         } else {
 953                 memset(child_res_id, 0, sizeof(*child_res_id));
 954         }
 955
 956         EXIT;
 957 cleanup:
 958         if (rc) {
 959                 switch(cleanup_phase) {
 960                 case 2:
 961                         if (child_res_id->name[0] != 0)
 962                                 ldlm_lock_decref(child_lockh, child_mode);
 963                 case 1:
 964                         ldlm_lock_decref(parent_lockh, parent_mode);
 965                 }
 966         }
 967         return rc;
 968 }
 969
 970 int mds_get_parent_child_locked(struct obd_device *obd, struct mds_obd *mds,
 971                                 struct ll_fid *fid,
 972                                 struct lustre_handle *parent_lockh,
 973                                 struct dentry **dparentp, int parent_mode,
 974                                 char *name, int namelen,
 975                                 struct lustre_handle *child_lockh,
 976                                 struct dentry **dchildp, int child_mode)
 977 {
 978         struct ldlm_res_id child_res_id = { .name = {0} };
 979         struct ldlm_res_id parent_res_id = { .name = {0} };
 980         struct inode *inode;
 981         int rc = 0, cleanup_phase = 0;
 982         ENTRY;
 983
 984         /* Step 1: Lookup parent */
 985         *dparentp = mds_fid2dentry(mds, fid, NULL);
 986         if (IS_ERR(*dparentp))
 987                 RETURN(rc = PTR_ERR(*dparentp));
 988         LASSERT((*dparentp)->d_inode);
 989
 990         CDEBUG(D_INODE, "parent ino %lu, name %s\n",
 991                (*dparentp)->d_inode->i_ino, name);
 992
 993         parent_res_id.name[0] = (*dparentp)->d_inode->i_ino;
 994         parent_res_id.name[1] = (*dparentp)->d_inode->i_generation;
 995
 996         cleanup_phase = 1; /* parent dentry */
 997
 998         /* Step 2: Lookup child (without lock, to get resource name) */
 999         *dchildp = ll_lookup_one_len(name, *dparentp, namelen - 1);
1000         if (IS_ERR(*dchildp)) {
1001                 rc = PTR_ERR(*dchildp);
1002                 CDEBUG(D_INODE, "child lookup error %d\n", rc);
1003                 GOTO(cleanup, rc);
1004         }
1005
1006         inode = (*dchildp)->d_inode;
1007         if (inode != NULL)
1008                 inode = igrab(inode);
1009         if (inode == NULL)
1010                 goto retry_locks;
1011
1012         child_res_id.name[0] = inode->i_ino;
1013         child_res_id.name[1] = inode->i_generation;
1014         iput(inode);
1015
1016 retry_locks:
1017         cleanup_phase = 2; /* child dentry */
1018
1019         /* Step 3: Lock parent and child in resource order.  If child doesn't
1020          *         exist, we still have to lock the parent and re-lookup. */
1021         rc = enqueue_ordered_locks(obd, &parent_res_id,parent_lockh,parent_mode,
1022                                    &child_res_id, child_lockh, child_mode);
1023         if (rc)
1024                 GOTO(cleanup, rc);
1025
1026         if (!(*dchildp)->d_inode)
1027                 cleanup_phase = 3; /* parent lock */
1028         else
1029                 cleanup_phase = 4; /* child lock */
1030
1031         /* Step 4: Re-lookup child to verify it hasn't changed since locking */
1032         rc = mds_verify_child(obd, &parent_res_id, parent_lockh, *dparentp,
1033                               parent_mode, &child_res_id, child_lockh, dchildp,
1034                               child_mode, name, namelen, &parent_res_id);
1035         if (rc > 0)
1036                 goto retry_locks;
1037         if (rc < 0) {
1038                 cleanup_phase = 3;
1039                 GOTO(cleanup, rc);
1040         }
1041
1042 cleanup:
1043         if (rc) {
1044                 switch (cleanup_phase) {
1045                 case 4:
1046                         ldlm_lock_decref(child_lockh, child_mode);
1047                 case 3:
1048                         ldlm_lock_decref(parent_lockh, parent_mode);
1049                 case 2:
1050                         l_dput(*dchildp);
1051                 case 1:
1052                         l_dput(*dparentp);
1053                 default: ;
1054                 }
1055         }
1056         return rc;
1057 }
1058
1059 void mds_reconstruct_generic(struct ptlrpc_request *req)
1060 {
1061         struct mds_export_data *med = &req->rq_export->exp_mds_data;
1062
1063         mds_req_from_mcd(req, med->med_mcd);
1064 }
1065
1066 static int mds_reint_unlink(struct mds_update_record *rec, int offset,
1067                             struct ptlrpc_request *req,
1068                             struct lustre_handle *lh)
1069 {
1070         struct dentry *dparent, *dchild;
1071         struct mds_obd *mds = mds_req2mds(req);
1072         struct obd_device *obd = req->rq_export->exp_obd;
1073         struct mds_body *body = NULL;
1074         struct inode *child_inode;
1075         struct lustre_handle parent_lockh, child_lockh, child_reuse_lockh;
1076         void *handle = NULL;
1077         int rc = 0, log_unlink = 0, cleanup_phase = 0;
1078         ENTRY;
1079
1080         LASSERT(offset == 0 || offset == 2);
1081
1082         DEBUG_REQ(D_INODE, req, "parent ino "LPU64"/%u, child %s",
1083                   rec->ur_fid1->id, rec->ur_fid1->generation, rec->ur_name);
1084
1085         MDS_CHECK_RESENT(req, mds_reconstruct_generic(req));
1086
1087         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_UNLINK))
1088                 GOTO(cleanup, rc = -ENOENT);
1089
1090         rc = mds_get_parent_child_locked(obd, mds, rec->ur_fid1,
1091                                          &parent_lockh, &dparent, LCK_PW,
1092                                          rec->ur_name, rec->ur_namelen,
1093                                          &child_lockh, &dchild, LCK_EX);
1094         if (rc)
1095                 GOTO(cleanup, rc);
1096
1097         cleanup_phase = 1; /* dchild, dparent, locks */
1098
1099         child_inode = dchild->d_inode;
1100         if (child_inode == NULL) {
1101                 CDEBUG(D_INODE, "child doesn't exist (dir %lu, name %s)\n",
1102                        dparent->d_inode->i_ino, rec->ur_name);
1103                 GOTO(cleanup, rc = -ENOENT);
1104         }
1105
1106         cleanup_phase = 2; /* dchild has a lock */
1107
1108         /* Step 4: Get a lock on the ino to sync with creation WRT inode
1109          * reuse (see bug 2029). */
1110         rc = mds_lock_new_child(obd, child_inode, &child_reuse_lockh);
1111         if (rc != ELDLM_OK)
1112                 GOTO(cleanup, rc);
1113
1114         cleanup_phase = 3; /* child inum lock */
1115
1116         OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_UNLINK_WRITE, dparent->d_inode->i_sb);
1117
1118         /* ldlm_reply in buf[0] if called via intent */
1119         if (offset)
1120                 offset = 1;
1121
1122         body = lustre_msg_buf(req->rq_repmsg, offset, sizeof (*body));
1123         LASSERT(body != NULL);
1124
1125         /* If this is the last reference to this inode, get the OBD EA
1126          * data first so the client can destroy OST objects.
1127          * we only do the object removal if no open files remain.
1128          * Nobody can get at this name anymore because of the locks so
1129          * we make decisions here as to whether to remove the inode */
1130         if (S_ISREG(child_inode->i_mode) && child_inode->i_nlink == 1 &&
1131             mds_open_orphan_count(child_inode) == 0) {
1132                 mds_pack_inode2fid(&body->fid1, child_inode);
1133                 mds_pack_inode2body(body, child_inode);
1134                 mds_pack_md(obd, req->rq_repmsg, offset + 1, body,
1135                             child_inode, 1);
1136                 if (!(body->valid & OBD_MD_FLEASIZE)) {
1137                         body->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
1138                                         OBD_MD_FLATIME | OBD_MD_FLMTIME);
1139                 } else {
1140                         log_unlink = 1;
1141                 }
1142         }
1143
1144         /* We have to do these checks ourselves, in case we are making an
1145          * orphan.  The client tells us whether rmdir() or unlink() was called,
1146          * so we need to return appropriate errors (bug 72).
1147          *
1148          * We don't have to check permissions, because vfs_rename (called from
1149          * mds_open_unlink_rename) also calls may_delete. */
1150         if ((rec->ur_mode & S_IFMT) == S_IFDIR) {
1151                 if (!S_ISDIR(child_inode->i_mode))
1152                         GOTO(cleanup, rc = -ENOTDIR);
1153         } else {
1154                 if (S_ISDIR(child_inode->i_mode))
1155                         GOTO(cleanup, rc = -EISDIR);
1156         }
1157
1158         if (child_inode->i_nlink == (S_ISDIR(child_inode->i_mode) ? 2 : 1) &&
1159             mds_open_orphan_count(child_inode) > 0) {
1160                 rc = mds_open_unlink_rename(rec, obd, dparent, dchild, &handle);
1161                 cleanup_phase = 4; /* transaction */
1162                 GOTO(cleanup, rc);
1163         }
1164
1165         /* Step 4: Do the unlink: we already verified ur_mode above (bug 72) */
1166         switch (child_inode->i_mode & S_IFMT) {
1167         case S_IFDIR:
1168                 /* Drop any lingering child directories before we start our
1169                  * transaction, to avoid doing multiple inode dirty/delete
1170                  * in our compound transaction (bug 1321). */
1171                 shrink_dcache_parent(dchild);
1172                 handle = fsfilt_start(obd, dparent->d_inode, FSFILT_OP_RMDIR,
1173                                       NULL);
1174                 if (IS_ERR(handle))
1175                         GOTO(cleanup, rc = PTR_ERR(handle));
1176                 cleanup_phase = 4; /* transaction */
1177                 rc = vfs_rmdir(dparent->d_inode, dchild);
1178                 break;
1179         case S_IFREG: {
1180                 handle = fsfilt_start(obd, dparent->d_inode,
1181                                       FSFILT_OP_UNLINK_LOG, NULL);
1182                 if (IS_ERR(handle))
1183                         GOTO(cleanup, rc = PTR_ERR(handle));
1184
1185                 cleanup_phase = 4; /* transaction */
1186                 rc = vfs_unlink(dparent->d_inode, dchild);
1187
1188 #ifdef ENABLE_ORPHANS
1189
1190                 if (!rc && log_unlink)
1191                         if (mds_log_op_unlink(obd, child_inode, req->rq_repmsg,
1192                                               offset + 1) > 0)
1193                                 body->valid |= OBD_MD_FLCOOKIE;
1194 #endif
1195                 break;
1196         }
1197         case S_IFLNK:
1198         case S_IFCHR:
1199         case S_IFBLK:
1200         case S_IFIFO:
1201         case S_IFSOCK:
1202                 handle = fsfilt_start(obd, dparent->d_inode, FSFILT_OP_UNLINK,
1203                                       NULL);
1204                 if (IS_ERR(handle))
1205                         GOTO(cleanup, rc = PTR_ERR(handle));
1206                 cleanup_phase = 4; /* transaction */
1207                 rc = vfs_unlink(dparent->d_inode, dchild);
1208                 break;
1209         default:
1210                 CERROR("bad file type %o unlinking %s\n", rec->ur_mode,
1211                        rec->ur_name);
1212                 LBUG();
1213                 GOTO(cleanup, rc = -EINVAL);
1214         }
1215
1216  cleanup:
1217         if (rc == 0) {
1218                 struct iattr iattr;
1219                 int err;
1220
1221                 iattr.ia_valid = ATTR_MTIME | ATTR_CTIME;
1222                 LTIME_S(iattr.ia_mtime) = rec->ur_time;
1223                 LTIME_S(iattr.ia_ctime) = rec->ur_time;
1224
1225                 err = fsfilt_setattr(obd, dparent, handle, &iattr, 0);
1226                 if (err)
1227                         CERROR("error on parent setattr: rc = %d\n", err);
1228         }
1229
1230         switch(cleanup_phase) {
1231         case 4:
1232                 rc = mds_finish_transno(mds, dparent->d_inode, handle, req,
1233                                         rc, 0);
1234                 if (!rc)
1235                         (void)obd_set_info(mds->mds_osc_exp, strlen("unlinked"),
1236                                            "unlinked", 0, NULL);
1237         case 3: /* child ino-reuse lock */
1238                 if (rc && body != NULL) {
1239                         // Don't unlink the OST objects if the MDS unlink failed
1240                         body->valid = 0;
1241                 }
1242                 if (rc)
1243                         ldlm_lock_decref(&child_reuse_lockh, LCK_EX);
1244                 else
1245                         ldlm_put_lock_into_req(req, &child_reuse_lockh, LCK_EX);
1246         case 2: /* child lock */
1247                 ldlm_lock_decref(&child_lockh, LCK_EX);
1248         case 1: /* child and parent dentry, parent lock */
1249                 if (rc)
1250                         ldlm_lock_decref(&parent_lockh, LCK_PW);
1251                 else
1252                         ldlm_put_lock_into_req(req, &parent_lockh, LCK_PW);
1253                 l_dput(dchild);
1254                 l_dput(dparent);
1255         case 0:
1256                 break;
1257         default:
1258                 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
1259                 LBUG();
1260         }
1261         req->rq_status = rc;
1262         return 0;
1263 }
1264
1265 static int mds_reint_link(struct mds_update_record *rec, int offset,
1266                           struct ptlrpc_request *req,
1267                           struct lustre_handle *lh)
1268 {
1269         struct obd_device *obd = req->rq_export->exp_obd;
1270         struct dentry *de_src = NULL;
1271         struct dentry *de_tgt_dir = NULL;
1272         struct dentry *dchild = NULL;
1273         struct mds_obd *mds = mds_req2mds(req);
1274         struct lustre_handle *handle = NULL, tgt_dir_lockh, src_lockh;
1275         struct ldlm_res_id src_res_id = { .name = {0} };
1276         struct ldlm_res_id tgt_dir_res_id = { .name = {0} };
1277         int rc = 0, cleanup_phase = 0;
1278         ENTRY;
1279
1280         LASSERT(offset == 0);
1281
1282         DEBUG_REQ(D_INODE, req, "original "LPU64"/%u to "LPU64"/%u %s",
1283                   rec->ur_fid1->id, rec->ur_fid1->generation,
1284                   rec->ur_fid2->id, rec->ur_fid2->generation, rec->ur_name);
1285
1286         MDS_CHECK_RESENT(req, mds_reconstruct_generic(req));
1287
1288         if (OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_LINK))
1289                 GOTO(cleanup, rc = -ENOENT);
1290
1291         /* Step 1: Lookup the source inode and target directory by FID */
1292         de_src = mds_fid2dentry(mds, rec->ur_fid1, NULL);
1293         if (IS_ERR(de_src))
1294                 GOTO(cleanup, rc = PTR_ERR(de_src));
1295
1296         cleanup_phase = 1; /* source dentry */
1297
1298         de_tgt_dir = mds_fid2dentry(mds, rec->ur_fid2, NULL);
1299         if (IS_ERR(de_tgt_dir))
1300                 GOTO(cleanup, rc = PTR_ERR(de_tgt_dir));
1301
1302         cleanup_phase = 2; /* target directory dentry */
1303
1304         CDEBUG(D_INODE, "linking %*s/%s to inode %lu\n",
1305                de_tgt_dir->d_name.len, de_tgt_dir->d_name.name, rec->ur_name,
1306                de_src->d_inode->i_ino);
1307
1308         /* Step 2: Take the two locks */
1309         src_res_id.name[0] = de_src->d_inode->i_ino;
1310         src_res_id.name[1] = de_src->d_inode->i_generation;
1311         tgt_dir_res_id.name[0] = de_tgt_dir->d_inode->i_ino;
1312         tgt_dir_res_id.name[1] = de_tgt_dir->d_inode->i_generation;
1313
1314         rc = enqueue_ordered_locks(obd, &src_res_id, &src_lockh, LCK_EX,
1315                                    &tgt_dir_res_id, &tgt_dir_lockh, LCK_EX);
1316         if (rc)
1317                 GOTO(cleanup, rc);
1318
1319         cleanup_phase = 3; /* locks */
1320
1321         /* Step 3: Lookup the child */
1322         dchild = ll_lookup_one_len(rec->ur_name, de_tgt_dir, rec->ur_namelen-1);
1323         if (IS_ERR(dchild)) {
1324                 rc = PTR_ERR(dchild);
1325                 if (rc != -EPERM && rc != -EACCES)
1326                         CERROR("child lookup error %d\n", rc);
1327                 GOTO(cleanup, rc);
1328         }
1329
1330         cleanup_phase = 4; /* child dentry */
1331
1332         if (dchild->d_inode) {
1333                 CDEBUG(D_INODE, "child exists (dir %lu, name %s)\n",
1334                        de_tgt_dir->d_inode->i_ino, rec->ur_name);
1335                 rc = -EEXIST;
1336                 GOTO(cleanup, rc);
1337         }
1338
1339         /* Step 4: Do it. */
1340         OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_LINK_WRITE, de_src->d_inode->i_sb);
1341
1342         handle = fsfilt_start(obd, de_tgt_dir->d_inode, FSFILT_OP_LINK, NULL);
1343         if (IS_ERR(handle)) {
1344                 rc = PTR_ERR(handle);
1345                 GOTO(cleanup, rc);
1346         }
1347
1348         rc = vfs_link(de_src, de_tgt_dir->d_inode, dchild);
1349         if (rc && rc != -EPERM && rc != -EACCES)
1350                 CERROR("vfs_link error %d\n", rc);
1351 cleanup:
1352         rc = mds_finish_transno(mds, de_tgt_dir ? de_tgt_dir->d_inode : NULL,
1353                                 handle, req, rc, 0);
1354         EXIT;
1355
1356         switch (cleanup_phase) {
1357         case 4: /* child dentry */
1358                 l_dput(dchild);
1359         case 3: /* locks */
1360                 if (rc) {
1361                         ldlm_lock_decref(&src_lockh, LCK_EX);
1362                         ldlm_lock_decref(&tgt_dir_lockh, LCK_EX);
1363                 } else {
1364                         ldlm_put_lock_into_req(req, &src_lockh, LCK_EX);
1365                         ldlm_put_lock_into_req(req, &tgt_dir_lockh, LCK_EX);
1366                 }
1367         case 2: /* target dentry */
1368                 l_dput(de_tgt_dir);
1369         case 1: /* source dentry */
1370                 l_dput(de_src);
1371         case 0:
1372                 break;
1373         default:
1374                 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
1375                 LBUG();
1376         }
1377         req->rq_status = rc;
1378         return 0;
1379 }
1380
1381 /*
1382  * add a hard link in the PENDING directory, only used by rename()
1383  */
1384 static int mds_add_link_orphan(struct mds_update_record *rec,
1385                                struct obd_device *obd,
1386                                struct dentry *dentry)
1387 {
1388         struct mds_obd *mds = &obd->u.mds;
1389         struct inode *pending_dir = mds->mds_pending_dir->d_inode;
1390         struct dentry *pending_child;
1391         char fidname[LL_FID_NAMELEN];
1392         int fidlen = 0, rc;
1393         ENTRY;
1394
1395         LASSERT(dentry->d_inode);
1396         LASSERT(!mds_inode_is_orphan(dentry->d_inode));
1397
1398         down(&pending_dir->i_sem);
1399         fidlen = ll_fid2str(fidname, dentry->d_inode->i_ino,
1400                             dentry->d_inode->i_generation);
1401
1402         CDEBUG(D_ERROR, "pending destroy of %dx open file %s = %s\n",
1403                mds_open_orphan_count(dentry->d_inode),
1404                rec->ur_name, fidname);
1405
1406         pending_child = lookup_one_len(fidname, mds->mds_pending_dir, fidlen);
1407         if (IS_ERR(pending_child))
1408                 GOTO(out_lock, rc = PTR_ERR(pending_child));
1409
1410         if (pending_child->d_inode != NULL) {
1411                 CERROR("re-destroying orphan file %s?\n", rec->ur_name);
1412                 LASSERT(pending_child->d_inode == dentry->d_inode);
1413                 GOTO(out_dput, rc = 0);
1414         }
1415
1416         lock_kernel();
1417         rc = vfs_link(dentry, pending_dir, pending_child);
1418         unlock_kernel();
1419         if (rc)
1420                 CERROR("error addlink orphan %s to PENDING: rc = %d\n",
1421                        rec->ur_name, rc);
1422         else
1423                 mds_inode_set_orphan(dentry->d_inode);
1424 out_dput:
1425         l_dput(pending_child);
1426 out_lock:
1427         up(&pending_dir->i_sem);
1428         RETURN(rc);
1429 }
1430
1431 /* The idea here is that we need to get four locks in the end:
1432  * one on each parent directory, one on each child.  We need to take
1433  * these locks in some kind of order (to avoid deadlocks), and the order
1434  * I selected is "increasing resource number" order.  We need to look up
1435  * the children, however, before we know what the resource number(s) are.
1436  * Thus the following plan:
1437  *
1438  * 1,2. Look up the parents
1439  * 3,4. Look up the children
1440  * 5. Take locks on the parents and children, in order
1441  * 6. Verify that the children haven't changed since they were looked up
1442  *
1443  * If there was a race and the children changed since they were first looked
1444  * up, it is possible that mds_verify_child() will be able to just grab the
1445  * lock on the new child resource (if it has a higher resource than any other)
1446  * but we need to compare against not only its parent, but also against the
1447  * parent and child of the "other half" of the rename, hence maxres_{src,tgt}.
1448  *
1449  * We need the fancy igrab() on the child inodes because we aren't holding a
1450  * lock on the parent after the lookup is done, so dentry->d_inode may change
1451  * at any time, and igrab() itself doesn't like getting passed a NULL argument.
1452  */
1453 static int mds_get_parents_children_locked(struct obd_device *obd,
1454                                            struct mds_obd *mds,
1455                                            struct ll_fid *p1_fid,
1456                                            struct dentry **de_srcdirp,
1457                                            struct ll_fid *p2_fid,
1458                                            struct dentry **de_tgtdirp,
1459                                            int parent_mode,
1460                                            const char *old_name, int old_len,
1461                                            struct dentry **de_oldp,
1462                                            const char *new_name, int new_len,
1463                                            struct dentry **de_newp,
1464                                            struct lustre_handle *dlm_handles,
1465                                            int child_mode)
1466 {
1467         struct ldlm_res_id p1_res_id = { .name = {0} };
1468         struct ldlm_res_id p2_res_id = { .name = {0} };
1469         struct ldlm_res_id c1_res_id = { .name = {0} };
1470         struct ldlm_res_id c2_res_id = { .name = {0} };
1471         struct ldlm_res_id *maxres_src, *maxres_tgt;
1472         struct inode *inode;
1473         int rc = 0, cleanup_phase = 0;
1474         ENTRY;
1475
1476         /* Step 1: Lookup the source directory */
1477         *de_srcdirp = mds_fid2dentry(mds, p1_fid, NULL);
1478         if (IS_ERR(*de_srcdirp))
1479                 GOTO(cleanup, rc = PTR_ERR(*de_srcdirp));
1480
1481         cleanup_phase = 1; /* source directory dentry */
1482
1483         p1_res_id.name[0] = (*de_srcdirp)->d_inode->i_ino;
1484         p1_res_id.name[1] = (*de_srcdirp)->d_inode->i_generation;
1485
1486         /* Step 2: Lookup the target directory */
1487         if (memcmp(p1_fid, p2_fid, sizeof(*p1_fid)) == 0) {
1488                 *de_tgtdirp = dget(*de_srcdirp);
1489         } else {
1490                 *de_tgtdirp = mds_fid2dentry(mds, p2_fid, NULL);
1491                 if (IS_ERR(*de_tgtdirp))
1492                         GOTO(cleanup, rc = PTR_ERR(*de_tgtdirp));
1493         }
1494
1495         cleanup_phase = 2; /* target directory dentry */
1496
1497         p2_res_id.name[0] = (*de_tgtdirp)->d_inode->i_ino;
1498         p2_res_id.name[1] = (*de_tgtdirp)->d_inode->i_generation;
1499
1500         /* Step 3: Lookup the source child entry */
1501         *de_oldp = ll_lookup_one_len(old_name, *de_srcdirp, old_len - 1);
1502         if (IS_ERR(*de_oldp)) {
1503                 rc = PTR_ERR(*de_oldp);
1504                 CERROR("old child lookup error (%*s): %d\n",
1505                        old_len - 1, old_name, rc);
1506                 GOTO(cleanup, rc);
1507         }
1508
1509         cleanup_phase = 3; /* original name dentry */
1510
1511         inode = (*de_oldp)->d_inode;
1512         if (inode != NULL)
1513                 inode = igrab(inode);
1514         if (inode == NULL)
1515                 GOTO(cleanup, rc = -ENOENT);
1516
1517         c1_res_id.name[0] = inode->i_ino;
1518         c1_res_id.name[1] = inode->i_generation;
1519         iput(inode);
1520
1521         /* Step 4: Lookup the target child entry */
1522         *de_newp = ll_lookup_one_len(new_name, *de_tgtdirp, new_len - 1);
1523         if (IS_ERR(*de_newp)) {
1524                 rc = PTR_ERR(*de_newp);
1525                 CERROR("new child lookup error (%*s): %d\n",
1526                        old_len - 1, old_name, rc);
1527                 GOTO(cleanup, rc);
1528         }
1529
1530         cleanup_phase = 4; /* target dentry */
1531
1532         inode = (*de_newp)->d_inode;
1533         if (inode != NULL)
1534                 inode = igrab(inode);
1535         if (inode == NULL)
1536                 goto retry_locks;
1537
1538         c2_res_id.name[0] = inode->i_ino;
1539         c2_res_id.name[1] = inode->i_generation;
1540         iput(inode);
1541
1542 retry_locks:
1543         /* Step 5: Take locks on the parents and child(ren) */
1544         maxres_src = &p1_res_id;
1545         maxres_tgt = &p2_res_id;
1546         cleanup_phase = 4; /* target dentry */
1547
1548         if (c1_res_id.name[0] != 0 && res_gt(&c1_res_id, &p1_res_id))
1549                 maxres_src = &c1_res_id;
1550         if (c2_res_id.name[0] != 0 && res_gt(&c2_res_id, &p2_res_id))
1551                 maxres_tgt = &c2_res_id;
1552
1553         rc = enqueue_4ordered_locks(obd, &p1_res_id,&dlm_handles[0],parent_mode,
1554                                     &p2_res_id, &dlm_handles[1], parent_mode,
1555                                     &c1_res_id, &dlm_handles[2], child_mode,
1556                                     &c2_res_id, &dlm_handles[3], child_mode);
1557         if (rc)
1558                 GOTO(cleanup, rc);
1559
1560         cleanup_phase = 6; /* parent and child(ren) locks */
1561
1562         /* Step 6a: Re-lookup source child to verify it hasn't changed */
1563         rc = mds_verify_child(obd, &p1_res_id, &dlm_handles[0], *de_srcdirp,
1564                               parent_mode, &c1_res_id, &dlm_handles[2], de_oldp,
1565                               child_mode, old_name, old_len, maxres_tgt);
1566         if (rc) {
1567                 if (c2_res_id.name[0] != 0)
1568                         ldlm_lock_decref(&dlm_handles[3], child_mode);
1569                 ldlm_lock_decref(&dlm_handles[1], parent_mode);
1570                 cleanup_phase = 4;
1571                 if (rc > 0)
1572                         goto retry_locks;
1573                 GOTO(cleanup, rc);
1574         }
1575
1576         if ((*de_oldp)->d_inode == NULL)
1577                 GOTO(cleanup, rc = -ENOENT);
1578
1579         /* Step 6b: Re-lookup target child to verify it hasn't changed */
1580         rc = mds_verify_child(obd, &p2_res_id, &dlm_handles[1], *de_tgtdirp,
1581                               parent_mode, &c2_res_id, &dlm_handles[3], de_newp,
1582                               child_mode, new_name, new_len, maxres_src);
1583         if (rc) {
1584                 ldlm_lock_decref(&dlm_handles[2], child_mode);
1585                 ldlm_lock_decref(&dlm_handles[0], parent_mode);
1586                 cleanup_phase = 4;
1587                 if (rc > 0)
1588                         goto retry_locks;
1589                 GOTO(cleanup, rc);
1590         }
1591
1592         EXIT;
1593 cleanup:
1594         if (rc) {
1595                 switch (cleanup_phase) {
1596                 case 6: /* child lock(s) */
1597                         if (c2_res_id.name[0] != 0)
1598                                 ldlm_lock_decref(&dlm_handles[3], child_mode);
1599                         if (c1_res_id.name[0] != 0)
1600                                 ldlm_lock_decref(&dlm_handles[2], child_mode);
1601                 case 5: /* parent locks */
1602                         ldlm_lock_decref(&dlm_handles[1], parent_mode);
1603                         ldlm_lock_decref(&dlm_handles[0], parent_mode);
1604                 case 4: /* target dentry */
1605                         l_dput(*de_newp);
1606                 case 3: /* source dentry */
1607                         l_dput(*de_oldp);
1608                 case 2: /* target directory dentry */
1609                         l_dput(*de_tgtdirp);
1610                 case 1: /* source directry dentry */
1611                         l_dput(*de_srcdirp);
1612                 }
1613         }
1614
1615         return rc;
1616 }
1617
1618 static int mds_reint_rename(struct mds_update_record *rec, int offset,
1619                             struct ptlrpc_request *req,
1620                             struct lustre_handle *lockh)
1621 {
1622         struct obd_device *obd = req->rq_export->exp_obd;
1623         struct dentry *de_srcdir = NULL;
1624         struct dentry *de_tgtdir = NULL;
1625         struct dentry *de_old = NULL;
1626         struct dentry *de_new = NULL;
1627         struct mds_obd *mds = mds_req2mds(req);
1628         struct lustre_handle dlm_handles[4];
1629         struct mds_body *body = NULL;
1630         int rc = 0, lock_count = 3;
1631         int cleanup_phase = 0;
1632         void *handle = NULL;
1633         ENTRY;
1634
1635         LASSERT(offset == 0);
1636
1637         DEBUG_REQ(D_INODE, req, "parent "LPU64"/%u %s to "LPU64"/%u %s",
1638                   rec->ur_fid1->id, rec->ur_fid1->generation, rec->ur_name,
1639                   rec->ur_fid2->id, rec->ur_fid2->generation, rec->ur_tgt);
1640
1641         MDS_CHECK_RESENT(req, mds_reconstruct_generic(req));
1642
1643         rc = mds_get_parents_children_locked(obd, mds, rec->ur_fid1, &de_srcdir,
1644                                              rec->ur_fid2, &de_tgtdir, LCK_PW,
1645                                              rec->ur_name, rec->ur_namelen,
1646                                              &de_old, rec->ur_tgt,
1647                                              rec->ur_tgtlen, &de_new,
1648                                              dlm_handles, LCK_EX);
1649         if (rc)
1650                 GOTO(cleanup, rc);
1651
1652         cleanup_phase = 1; /* parent(s), children, locks */
1653
1654         if (de_new->d_inode)
1655                 lock_count = 4;
1656
1657         /* sanity check for src inode */
1658         if (de_old->d_inode->i_ino == de_srcdir->d_inode->i_ino ||
1659             de_old->d_inode->i_ino == de_tgtdir->d_inode->i_ino)
1660                 GOTO(cleanup, rc = -EINVAL);
1661
1662         /* sanity check for dest inode */
1663         if (de_new->d_inode &&
1664             (de_new->d_inode->i_ino == de_srcdir->d_inode->i_ino ||
1665              de_new->d_inode->i_ino == de_tgtdir->d_inode->i_ino))
1666                 GOTO(cleanup, rc = -EINVAL);
1667
1668         /* if we are about to remove the target at first, pass the EA of
1669          * that inode to client to perform and cleanup on OST */
1670         body = lustre_msg_buf(req->rq_repmsg, 0, sizeof (*body));
1671         LASSERT(body != NULL);
1672
1673         if (de_new->d_inode &&
1674             S_ISREG(de_new->d_inode->i_mode) &&
1675             de_new->d_inode->i_nlink == 1 &&
1676             mds_open_orphan_count(de_new->d_inode) == 0) {
1677                 mds_pack_inode2fid(&body->fid1, de_new->d_inode);
1678                 mds_pack_inode2body(body, de_new->d_inode);
1679                 mds_pack_md(obd, req->rq_repmsg, 1, body, de_new->d_inode, 1);
1680                 if (!(body->valid & OBD_MD_FLEASIZE)) {
1681                         body->valid |= (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS |
1682                         OBD_MD_FLATIME | OBD_MD_FLMTIME);
1683                 } else {
1684                         /* XXX need log unlink? */
1685                 }
1686         }
1687
1688         OBD_FAIL_WRITE(OBD_FAIL_MDS_REINT_RENAME_WRITE,
1689                        de_srcdir->d_inode->i_sb);
1690
1691         handle = fsfilt_start(obd, de_tgtdir->d_inode, FSFILT_OP_RENAME, NULL);
1692         if (IS_ERR(handle))
1693                 GOTO(cleanup, rc = PTR_ERR(handle));
1694
1695         /* FIXME need adjust the journal block count? */
1696         /* if the target should be moved to PENDING, we at first increase the
1697          * link and later vfs_rename() will decrease the link count again */
1698         if (de_new->d_inode &&
1699             S_ISREG(de_new->d_inode->i_mode) &&
1700             de_new->d_inode->i_nlink == 1 &&
1701             mds_open_orphan_count(de_new->d_inode) > 0) {
1702                 rc = mds_add_link_orphan(rec, obd, de_new);
1703                 if (rc)
1704                         GOTO(cleanup, rc);
1705         }
1706
1707         lock_kernel();
1708         de_old->d_fsdata = req;
1709         de_new->d_fsdata = req;
1710         rc = vfs_rename(de_srcdir->d_inode, de_old, de_tgtdir->d_inode, de_new);
1711         unlock_kernel();
1712
1713         GOTO(cleanup, rc);
1714 cleanup:
1715         rc = mds_finish_transno(mds, de_tgtdir ? de_tgtdir->d_inode : NULL,
1716                                 handle, req, rc, 0);
1717         switch (cleanup_phase) {
1718         case 1:
1719                 if (rc) {
1720                         if (lock_count == 4)
1721                                 ldlm_lock_decref(&(dlm_handles[3]), LCK_EX);
1722                         ldlm_lock_decref(&(dlm_handles[2]), LCK_EX);
1723                         ldlm_lock_decref(&(dlm_handles[1]), LCK_PW);
1724                         ldlm_lock_decref(&(dlm_handles[0]), LCK_PW);
1725                 } else {
1726                         if (lock_count == 4)
1727                                 ldlm_put_lock_into_req(req,
1728                                                     &(dlm_handles[3]), LCK_EX);
1729                         ldlm_put_lock_into_req(req, &(dlm_handles[2]), LCK_EX);
1730                         ldlm_put_lock_into_req(req, &(dlm_handles[1]), LCK_PW);
1731                         ldlm_put_lock_into_req(req, &(dlm_handles[0]), LCK_PW);
1732                 }
1733                 l_dput(de_new);
1734                 l_dput(de_old);
1735                 l_dput(de_tgtdir);
1736                 l_dput(de_srcdir);
1737         case 0:
1738                 break;
1739         default:
1740                 CERROR("invalid cleanup_phase %d\n", cleanup_phase);
1741                 LBUG();
1742         }
1743         req->rq_status = rc;
1744         return 0;
1745 }
1746
1747 typedef int (*mds_reinter)(struct mds_update_record *, int offset,
1748                            struct ptlrpc_request *, struct lustre_handle *);
1749
1750 static mds_reinter reinters[REINT_MAX + 1] = {
1751         [REINT_SETATTR] mds_reint_setattr,
1752         [REINT_CREATE] mds_reint_create,
1753         [REINT_UNLINK] mds_reint_unlink,
1754         [REINT_LINK] mds_reint_link,
1755         [REINT_RENAME] mds_reint_rename,
1756         [REINT_OPEN] mds_open
1757 };
1758
1759 int mds_reint_rec(struct mds_update_record *rec, int offset,
1760                   struct ptlrpc_request *req, struct lustre_handle *lockh)
1761 {
1762         struct obd_device *obd = req->rq_export->exp_obd;
1763         struct obd_run_ctxt saved;
1764         int rc;
1765         ENTRY;
1766
1767         /* checked by unpacker */
1768         LASSERT(rec->ur_opcode <= REINT_MAX &&
1769                 reinters[rec->ur_opcode] != NULL);
1770
1771         push_ctxt(&saved, &obd->obd_ctxt, &rec->ur_uc);
1772         rc = reinters[rec->ur_opcode] (rec, offset, req, lockh);
1773         pop_ctxt(&saved, &obd->obd_ctxt, &rec->ur_uc);
1774
1775         RETURN(rc);
1776 }