lustre/lmv/lmv_intent.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Copyright (C) 2002, 2003 Cluster File Systems, Inc.
   5  *
   6  *   This file is part of Lustre, http://www.lustre.org.
   7  *
   8  *   Lustre is free software; you can redistribute it and/or
   9  *   modify it under the terms of version 2 of the GNU General Public
  10  *   License as published by the Free Software Foundation.
  11  *
  12  *   Lustre is distributed in the hope that it will be useful,
  13  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *   GNU General Public License for more details.
  16  *
  17  *   You should have received a copy of the GNU General Public License
  18  *   along with Lustre; if not, write to the Free Software
  19  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20  */
  21
  22 #ifndef EXPORT_SYMTAB
  23 # define EXPORT_SYMTAB
  24 #endif
  25 #define DEBUG_SUBSYSTEM S_LMV
  26 #ifdef __KERNEL__
  27 #include <linux/slab.h>
  28 #include <linux/module.h>
  29 #include <linux/init.h>
  30 #include <linux/slab.h>
  31 #include <linux/pagemap.h>
  32 #include <asm/div64.h>
  33 #else
  34 #include <liblustre.h>
  35 #endif
  36
  37 #include <linux/obd_support.h>
  38 #include <linux/lustre_lib.h>
  39 #include <linux/lustre_net.h>
  40 #include <linux/lustre_idl.h>
  41 #include <linux/lustre_dlm.h>
  42 #include <linux/lustre_mds.h>
  43 #include <linux/obd_class.h>
  44 #include <linux/obd_ost.h>
  45 #include <linux/seq_file.h>
  46 #include <linux/lprocfs_status.h>
  47 #include <linux/lustre_fsfilt.h>
  48 #include <linux/obd_lmv.h>
  49 #include "lmv_internal.h"
  50
  51
  52 int lmv_handle_remote_inode(struct obd_export *exp, struct ll_uctxt *uctxt,
  53                             void *lmm, int lmmsize,
  54                             struct lookup_intent *it, int flags,
  55                             struct ptlrpc_request **reqp,
  56                             ldlm_blocking_callback cb_blocking)
  57 {
  58         struct obd_device *obd = exp->exp_obd;
  59         struct lmv_obd *lmv = &obd->u.lmv;
  60         struct mds_body *body = NULL;
  61         int rc = 0;
  62         ENTRY;
  63
  64         body = lustre_msg_buf((*reqp)->rq_repmsg, 1, sizeof(*body));
  65         LASSERT(body != NULL);
  66
  67         if (body->valid & OBD_MD_MDS) {
  68                 /* oh, MDS reports that this is remote inode case
  69                  * i.e. we have to ask for real attrs on another MDS */
  70                 struct ptlrpc_request *req;
  71                 struct ll_fid nfid;
  72                 struct lustre_handle plock;
  73                 int pmode;
  74
  75                 if (it->it_op == IT_LOOKUP) {
  76                         /* unfortunately, we have to lie to MDC/MDS to
  77                          * retrieve attributes llite needs */
  78                         it->it_op = IT_GETATTR;
  79                 }
  80
  81                 /* we got LOOKUP lock, but we really need attrs */
  82                 pmode = it->d.lustre.it_lock_mode;
  83                 if (pmode) {
  84                         memcpy(&plock, &it->d.lustre.it_lock_handle,
  85                                         sizeof(plock));
  86                         it->d.lustre.it_lock_mode = 0;
  87                 }
  88
  89                 nfid = body->fid1;
  90                 it->d.lustre.it_disposition &= ~DISP_ENQ_COMPLETE;
  91                 rc = md_intent_lock(lmv->tgts[nfid.mds].exp, uctxt, &nfid,
  92                                     NULL, 0, lmm, lmmsize, NULL, it, flags,
  93                                     &req, cb_blocking);
  94
  95                 /* llite needs LOOKUP lock to track dentry revocation in
  96                  * order to maintain dcache consistency. thus drop UPDATE
  97                  * lock here and put LOOKUP in request */
  98                 if (rc == 0) {
  99                         LASSERT(it->d.lustre.it_lock_mode != 0);
 100                         ldlm_lock_decref((void *)&it->d.lustre.it_lock_handle,
 101                                          it->d.lustre.it_lock_mode);
 102                         memcpy(&it->d.lustre.it_lock_handle, &plock,
 103                                         sizeof(plock));
 104                         it->d.lustre.it_lock_mode = pmode;
 105
 106                 } else if (pmode)
 107                         ldlm_lock_decref(&plock, pmode);
 108
 109                 ptlrpc_req_finished(*reqp);
 110                 *reqp = req;
 111         }
 112         RETURN(rc);
 113 }
 114
 115 int lmv_intent_open(struct obd_export *exp, struct ll_uctxt *uctxt,
 116                     struct ll_fid *pfid, const char *name, int len,
 117                     void *lmm, int lmmsize, struct ll_fid *cfid,
 118                     struct lookup_intent *it, int flags,
 119                     struct ptlrpc_request **reqp,
 120                     ldlm_blocking_callback cb_blocking)
 121 {
 122         struct obd_device *obd = exp->exp_obd;
 123         struct lmv_obd *lmv = &obd->u.lmv;
 124         struct mds_body *body = NULL;
 125         struct ll_fid rpfid = *pfid;
 126         struct lmv_obj *obj;
 127         struct mea *mea;
 128         int rc, mds;
 129         ENTRY;
 130
 131         /* IT_OPEN is intended to open (and create, possible) an object.
 132          * parent (pfid) may be splitted dir */
 133
 134         mds = pfid->mds;
 135         obj = lmv_grab_obj(obd, pfid, 0);
 136         if (obj) {
 137                 /* directory is already splitted, so we have to forward
 138                  * request to the right MDS */
 139                 mds = raw_name2idx(obj->objcount, name, len);
 140                 rpfid = obj->objs[mds].fid;
 141                 CDEBUG(D_OTHER, "forward to MDS #%u\n", mds);
 142         }
 143
 144         rc = md_intent_lock(lmv->tgts[mds].exp, uctxt, &rpfid, name, len,
 145                             lmm, lmmsize, cfid, it, flags, reqp, cb_blocking);
 146
 147         lmv_put_obj(obj);
 148         if (rc != 0)
 149                 RETURN(rc);
 150
 151         /* okay, MDS has returned success. probably name has been
 152          * resolved in remote inode */
 153         rc = lmv_handle_remote_inode(exp, uctxt, lmm, lmmsize, it, flags,
 154                                      reqp, cb_blocking);
 155         if (rc != 0) {
 156                 LASSERT(rc < 0);
 157                 RETURN(rc);
 158         }
 159
 160         /* caller may use attrs MDS returns on IT_OPEN lock request
 161          * so, we have to update them for splitted dir */
 162         body = lustre_msg_buf((*reqp)->rq_repmsg, 1, sizeof(*body));
 163         LASSERT(body != NULL);
 164         cfid = &body->fid1;
 165         obj = lmv_grab_obj(obd, cfid, 0);
 166         if (rc == 0 && !obj && (mea = is_body_of_splitted_dir(*reqp, 1))) {
 167                 /* wow! this is splitted dir, we'd like to handle it */
 168                 rc = lmv_create_obj_from_attrs(exp, &body->fid1, mea);
 169         }
 170         obj = lmv_grab_obj(obd, cfid, 0);
 171         if (obj) {
 172                 /* this is splitted dir and we'd want to get attrs */
 173                 CDEBUG(D_OTHER, "attrs from slaves for %lu/%lu/%lu\n",
 174                                 (unsigned long) cfid->mds,
 175                                 (unsigned long) cfid->id,
 176                                 (unsigned long) cfid->generation);
 177                 rc = lmv_revalidate_slaves(exp, reqp, cfid,
 178                                 it, 1, cb_blocking);
 179         } else if (S_ISDIR(body->mode)) {
 180                 /*CWARN("hmmm, %lu/%lu/%lu has not lmv obj?!\n",
 181                                 (unsigned long) cfid->mds,
 182                                 (unsigned long) cfid->id,
 183                                 (unsigned long) cfid->generation);*/
 184         }
 185         lmv_put_obj(obj);
 186         RETURN(rc);
 187 }
 188
 189 int lmv_intent_getattr(struct obd_export *exp, struct ll_uctxt *uctxt,
 190                        struct ll_fid *pfid, const char *name, int len,
 191                        void *lmm, int lmmsize, struct ll_fid *cfid,
 192                        struct lookup_intent *it, int flags,
 193                        struct ptlrpc_request **reqp,
 194                        ldlm_blocking_callback cb_blocking)
 195 {
 196         struct obd_device *obd = exp->exp_obd;
 197         struct lmv_obd *lmv = &obd->u.lmv;
 198         struct mds_body *body = NULL;
 199         struct ll_fid rpfid = *pfid;
 200         struct lmv_obj *obj, *obj2;
 201         struct mea *mea;
 202         int rc = 0, mds;
 203         ENTRY;
 204
 205         if (cfid) {
 206                 /* caller wants to revalidate attrs of obj
 207                  * we have to revalidate slaves if requested
 208                  * object is splitted directory */
 209                 CDEBUG(D_OTHER, "revalidate attrs for %lu/%lu/%lu\n",
 210                        (unsigned long) cfid->mds,
 211                        (unsigned long) cfid->id,
 212                        (unsigned long) cfid->generation);
 213                 mds = cfid->mds;
 214                 obj = lmv_grab_obj(obd, cfid, 0);
 215                 if (obj) {
 216                         /* in fact, we need not this with current
 217                          * _intent_lock(), but it may change some day */
 218                         rpfid = obj->objs[mds].fid;
 219                 }
 220                 rc = md_intent_lock(lmv->tgts[mds].exp, uctxt, &rpfid, name,
 221                                     len, lmm, lmmsize, cfid, it, flags, reqp,
 222                                     cb_blocking);
 223                 if (obj && rc >= 0) {
 224                         /* this is splitted dir. in order to optimize things
 225                          * a bit, we consider obj valid updating missing
 226                          * parts. FIXME: do we need to return any lock here?
 227                          * it would be fine if we don't. this means that
 228                          * nobody should use UPDATE lock to notify about
 229                          * object removal */
 230                         CDEBUG(D_OTHER,
 231                                "revalidate slaves for %lu/%lu/%lu, rc %d\n",
 232                                (unsigned long) cfid->mds,
 233                                (unsigned long) cfid->id,
 234                                (unsigned long) cfid->generation, rc);
 235                         rc = lmv_revalidate_slaves(exp, reqp, cfid, it, rc,
 236                                                    cb_blocking);
 237                 }
 238                 RETURN(rc);
 239         }
 240
 241         CDEBUG(D_OTHER, "INTENT getattr for %*s on %lu/%lu/%lu\n",
 242                len, name, (unsigned long) pfid->mds,
 243                (unsigned long) pfid->id,
 244                (unsigned long) pfid->generation);
 245
 246         mds = pfid->mds;
 247         obj = lmv_grab_obj(obd, pfid, 0);
 248         if (obj && len) {
 249                 /* directory is already splitted. calculate mds */
 250                 mds = raw_name2idx(obj->objcount, (char *) name, len);
 251                 rpfid = obj->objs[mds].fid;
 252                 CDEBUG(D_OTHER, "forward to MDS #%u (slave %lu/%lu/%lu)\n",
 253                        mds, (unsigned long) rpfid.mds,
 254                        (unsigned long) rpfid.id,
 255                        (unsigned long) rpfid.generation);
 256         }
 257         rc = md_intent_lock(lmv->tgts[mds].exp, uctxt, &rpfid, name,
 258                             len, lmm, lmmsize, NULL, it, flags, reqp,
 259                             cb_blocking);
 260         if (rc < 0)
 261                 RETURN(rc);
 262         LASSERT(rc == 0);
 263
 264         /* okay, MDS has returned success. probably name has been
 265          * resolved in remote inode */
 266         rc = lmv_handle_remote_inode(exp, uctxt, lmm, lmmsize, it, flags,
 267                                      reqp, cb_blocking);
 268         if (rc < 0)
 269                 RETURN(rc);
 270
 271         body = lustre_msg_buf((*reqp)->rq_repmsg, 1, sizeof(*body));
 272         LASSERT(body != NULL);
 273         cfid = &body->fid1;
 274         obj2 = lmv_grab_obj(obd, cfid, 0);
 275
 276         if (rc == 0 && !obj2 && (mea = is_body_of_splitted_dir(*reqp, 1))) {
 277                 /* wow! this is splitted dir, we'd like to handle it */
 278                 body = lustre_msg_buf((*reqp)->rq_repmsg, 1, sizeof(*body));
 279                 LASSERT(body != NULL);
 280                 rc = lmv_create_obj_from_attrs(exp, &body->fid1, mea);
 281                 obj2 = lmv_grab_obj(obd, cfid, 0);
 282         }
 283
 284         if (obj2) {
 285                 /* this is splitted dir and we'd want to get attrs */
 286                 CDEBUG(D_OTHER,
 287                                 "attrs from slaves for %lu/%lu/%lu, rc %d\n",
 288                                 (unsigned long) cfid->mds,
 289                                 (unsigned long) cfid->id,
 290                                 (unsigned long) cfid->generation, rc);
 291                 rc = lmv_revalidate_slaves(exp, reqp, cfid,
 292                                 it, 1, cb_blocking);
 293         }
 294         RETURN(rc);
 295 }
 296
 297 void lmv_update_body_from_obj(struct mds_body *body, struct lmv_inode *obj)
 298 {
 299         /* update size */
 300         body->size += obj->size;
 301
 302         /* update atime */
 303         /* update ctime */
 304         /* update mtime */
 305         /* update nlink */
 306 }
 307
 308 int lmv_lookup_slaves(struct obd_export *exp, struct ptlrpc_request **reqp)
 309 {
 310         struct obd_device *obd = exp->exp_obd;
 311         struct lmv_obd *lmv = &obd->u.lmv;
 312         struct mds_body *body = NULL;
 313         struct lustre_handle *lockh;
 314         struct ldlm_lock *lock;
 315         struct mds_body *body2;
 316         struct ll_uctxt uctxt;
 317         struct lmv_obj *obj;
 318         int i, rc = 0;
 319         ENTRY;
 320
 321         LASSERT(reqp);
 322         LASSERT(*reqp);
 323
 324         /* master is locked. we'd like to take locks on slaves
 325          * and update attributes to be returned from the slaves
 326          * it's important that lookup is called in two cases:
 327          *  - for first time (dcache has no such a resolving yet
 328          *  - ->d_revalidate() returned false
 329          * last case possible only if all the objs (master and
 330          * all slaves aren't valid */
 331
 332         body = lustre_msg_buf((*reqp)->rq_repmsg, 1, sizeof(*body));
 333         LASSERT(body != NULL);
 334
 335         obj = lmv_grab_obj(obd, &body->fid1, 0);
 336         LASSERT(obj);
 337
 338         CDEBUG(D_OTHER, "lookup slaves for %lu/%lu/%lu\n",
 339                (unsigned long) body->fid1.mds,
 340                (unsigned long) body->fid1.id,
 341                (unsigned long) body->fid1.generation);
 342
 343         uctxt.gid1 = 0;
 344         uctxt.gid2 = 0;
 345         for (i = 0; i < obj->objcount; i++) {
 346                 struct ll_fid fid = obj->objs[i].fid;
 347                 struct ptlrpc_request *req = NULL;
 348                 struct lookup_intent it;
 349
 350                 if (fid_equal(&fid, &obj->fid)) {
 351                         /* skip master obj */
 352                         continue;
 353                 }
 354
 355                 CDEBUG(D_OTHER, "lookup slave %lu/%lu/%lu\n",
 356                        (unsigned long) fid.mds,
 357                        (unsigned long) fid.id,
 358                        (unsigned long) fid.generation);
 359
 360                 /* is obj valid? */
 361                 memset(&it, 0, sizeof(it));
 362                 it.it_op = IT_GETATTR;
 363                 rc = md_intent_lock(lmv->tgts[fid.mds].exp, &uctxt, &fid,
 364                                     NULL, 0, NULL, 0, &fid, &it, 0, &req,
 365                                     lmv_dirobj_blocking_ast);
 366                 lockh = (struct lustre_handle *) &it.d.lustre.it_lock_handle;
 367                 if (rc > 0) {
 368                         /* nice, this slave is valid */
 369                         LASSERT(req == NULL);
 370                         CDEBUG(D_OTHER, "cached\n");
 371                         goto release_lock;
 372                 }
 373
 374                 if (rc < 0) {
 375                         /* error during revalidation */
 376                         GOTO(cleanup, rc);
 377                 }
 378
 379                 /* rc == 0, this means we have no such a lock and can't
 380                  * think obj is still valid. lookup it again */
 381                 LASSERT(req == NULL);
 382                 req = NULL;
 383                 memset(&it, 0, sizeof(it));
 384                 it.it_op = IT_GETATTR;
 385                 rc = md_intent_lock(lmv->tgts[fid.mds].exp, &uctxt, &fid,
 386                                     NULL, 0, NULL, 0, NULL, &it, 0, &req,
 387                                     lmv_dirobj_blocking_ast);
 388                 lockh = (struct lustre_handle *) &it.d.lustre.it_lock_handle;
 389                 LASSERT(rc <= 0);
 390                 if (rc < 0) {
 391                         /* error during lookup */
 392                         GOTO(cleanup, rc);
 393                 }
 394
 395                 lock = ldlm_handle2lock(lockh);
 396                 LASSERT(lock);
 397                 lock->l_ast_data = obj;
 398                 atomic_inc(&obj->count);
 399
 400                 body2 = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*body2));
 401                 LASSERT(body2);
 402
 403                 obj->objs[i].size = body2->size;
 404                 CDEBUG(D_OTHER, "fresh: %lu\n",
 405                                 (unsigned long) obj->objs[i].size);
 406
 407                 LDLM_LOCK_PUT(lock);
 408
 409                 if (req)
 410                         ptlrpc_req_finished(req);
 411 release_lock:
 412                 lmv_update_body_from_obj(body, obj->objs + i);
 413                 if (it.d.lustre.it_lock_mode)
 414                         ldlm_lock_decref(lockh, it.d.lustre.it_lock_mode);
 415         }
 416 cleanup:
 417         RETURN(rc);
 418 }
 419
 420 int lmv_intent_lookup(struct obd_export *exp, struct ll_uctxt *uctxt,
 421                 struct ll_fid *pfid, const char *name, int len,
 422                 void *lmm, int lmmsize, struct ll_fid *cfid,
 423                 struct lookup_intent *it, int flags,
 424                 struct ptlrpc_request **reqp,
 425                 ldlm_blocking_callback cb_blocking)
 426 {
 427         struct obd_device *obd = exp->exp_obd;
 428         struct lmv_obd *lmv = &obd->u.lmv;
 429         struct mds_body *body = NULL;
 430         struct ll_fid rpfid = *pfid;
 431         struct lmv_obj *obj;
 432         struct mea *mea;
 433         int rc, mds;
 434         ENTRY;
 435
 436         /* IT_LOOKUP is intended to produce name -> fid resolving
 437          * (let's call this lookup below) or to confirm requested
 438          * resolving is still valid (let's call this revalidation)
 439          * cfid != NULL specifies revalidation */
 440
 441         if (cfid) {
 442                 /* this is revalidation: we have to check is LOOKUP
 443                  * lock still valid for given fid. very important
 444                  * part is that we have to choose right mds because
 445                  * namespace is per mds */
 446                 rpfid = *pfid;
 447                 obj = lmv_grab_obj(obd, pfid, 0);
 448                 if (obj) {
 449                         mds = raw_name2idx(obj->objcount, (char *) name, len);
 450                         rpfid = obj->objs[mds].fid;
 451                         lmv_put_obj(obj);
 452                 }
 453                 mds = rpfid.mds;
 454                 CDEBUG(D_OTHER, "revalidate lookup for %lu/%lu/%lu to %d MDS\n",
 455                        (unsigned long) cfid->mds,
 456                        (unsigned long) cfid->id,
 457                        (unsigned long) cfid->generation, mds);
 458                 rc = md_intent_lock(lmv->tgts[mds].exp, uctxt, pfid, name,
 459                                     len, lmm, lmmsize, cfid, it, flags,
 460                                     reqp, cb_blocking);
 461                 RETURN(rc);
 462         }
 463
 464         mds = pfid->mds;
 465 repeat:
 466         /* this is lookup. during lookup we have to update all the
 467          * attributes, because returned values will be put in struct
 468          * inode */
 469
 470         obj = lmv_grab_obj(obd, pfid, 0);
 471         if (obj && len) {
 472                 /* directory is already splitted. calculate mds */
 473                 mds = raw_name2idx(obj->objcount, (char *) name, len);
 474                 rpfid = obj->objs[mds].fid;
 475                 lmv_put_obj(obj);
 476         }
 477
 478         rc = md_intent_lock(lmv->tgts[mds].exp, uctxt, &rpfid, name,
 479                             len, lmm, lmmsize, NULL, it, flags, reqp,
 480                             cb_blocking);
 481         if (rc > 0) {
 482                 /* very interesting. it seems object is still valid
 483                  * but for some reason llite calls lookup, not revalidate */
 484                 CWARN("lookup for %lu/%lu/%lu and data should be uptodate\n",
 485                        (unsigned long) rpfid.mds,
 486                        (unsigned long) rpfid.id,
 487                        (unsigned long) rpfid.generation);
 488                 LASSERT(*reqp == NULL);
 489                 RETURN(rc);
 490         }
 491
 492         if (rc == 0 && *reqp == NULL) {
 493                 /* once again, we're asked for lookup, not revalidate */
 494                 CWARN("lookup for %lu/%lu/%lu and data should be uptodate\n",
 495                        (unsigned long) rpfid.mds,
 496                        (unsigned long) rpfid.id,
 497                        (unsigned long) rpfid.generation);
 498                 RETURN(rc);
 499         }
 500
 501         if (rc == -ESTALE) {
 502                 /* directory got splitted since last update. this shouldn't
 503                  * be becasue splitting causes lock revocation, so revalidate
 504                  * had to fail and lookup on dir had to return mea */
 505                 CWARN("we haven't knew about directory splitting!\n");
 506                 LASSERT(obj == NULL);
 507                 rc = lmv_create_obj_from_attrs(exp, &rpfid, NULL);
 508                 if (rc)
 509                         RETURN(rc);
 510                 goto repeat;
 511         }
 512
 513         if (rc < 0)
 514                 RETURN(rc);
 515
 516         /* okay, MDS has returned success. probably name has been
 517          * resolved in remote inode */
 518         rc = lmv_handle_remote_inode(exp, uctxt, lmm, lmmsize, it, flags,
 519                                      reqp, cb_blocking);
 520
 521         if (rc == 0 && (mea = is_body_of_splitted_dir(*reqp, 1))) {
 522                 /* wow! this is splitted dir, we'd like to handle it */
 523                 body = lustre_msg_buf((*reqp)->rq_repmsg, 1, sizeof(*body));
 524                 LASSERT(body != NULL);
 525                 obj = lmv_grab_obj(obd, &body->fid1, 0);
 526                 if (!obj)
 527                         rc = lmv_create_obj_from_attrs(exp, &body->fid1, mea);
 528                 lmv_put_obj(obj);
 529         }
 530
 531         RETURN(rc);
 532 }
 533
 534 int lmv_intent_lock(struct obd_export *exp, struct ll_uctxt *uctxt,
 535                 struct ll_fid *pfid, const char *name, int len,
 536                 void *lmm, int lmmsize, struct ll_fid *cfid,
 537                 struct lookup_intent *it, int flags,
 538                 struct ptlrpc_request **reqp,
 539                 ldlm_blocking_callback cb_blocking)
 540 {
 541         struct obd_device *obd = exp->exp_obd;
 542         int rc = 0;
 543         ENTRY;
 544
 545         LASSERT(it);
 546         LASSERT(pfid);
 547
 548         CDEBUG(D_OTHER, "INTENT LOCK '%s' for '%*s' on %lu/%lu -> %u\n",
 549                         LL_IT2STR(it), len, name, (unsigned long) pfid->id,
 550                         (unsigned long) pfid->generation, pfid->mds);
 551
 552         lmv_connect(obd);
 553         if (it->it_op == IT_LOOKUP)
 554                 rc = lmv_intent_lookup(exp, uctxt, pfid, name, len, lmm,
 555                                        lmmsize, cfid, it, flags, reqp,
 556                                        cb_blocking);
 557         else if (it->it_op & IT_OPEN)
 558                 rc = lmv_intent_open(exp, uctxt, pfid, name, len, lmm,
 559                                      lmmsize, cfid, it, flags, reqp,
 560                                      cb_blocking);
 561         else if (it->it_op == IT_GETATTR)
 562                 rc = lmv_intent_getattr(exp, uctxt, pfid, name, len, lmm,
 563                                         lmmsize, cfid, it, flags, reqp,
 564                                         cb_blocking);
 565         else
 566                 LBUG();
 567         RETURN(rc);
 568 }
 569
 570 int lmv_revalidate_slaves(struct obd_export *exp, struct ptlrpc_request **reqp,
 571                           struct ll_fid *mfid, struct lookup_intent *oit,
 572                           int master_valid, ldlm_blocking_callback cb_blocking)
 573 {
 574         struct obd_device *obd = exp->exp_obd;
 575         struct ptlrpc_request *mreq = *reqp;
 576         struct lmv_obd *lmv = &obd->u.lmv;
 577         struct lustre_handle master_lockh;
 578         unsigned long size = 0;
 579         struct ldlm_lock *lock;
 580         struct mds_body *body;
 581         struct ll_uctxt uctxt;
 582         struct lmv_obj *obj;
 583         int master_lock_mode;
 584         int i, rc = 0;
 585         ENTRY;
 586
 587         /* we have to loop over the subobjects, check validity and update
 588          * them from MDSs if needed. it's very useful that we need not to
 589          * update all the fields. say, common fields (that are equal on
 590          * all the subojects need not to be update, another fields (i_size,
 591          * for example) are cached all the time */
 592         obj = lmv_grab_obj(obd, mfid, 0);
 593         LASSERT(obj);
 594
 595         master_lock_mode = 0;
 596         uctxt.gid1 = 0;
 597         uctxt.gid2 = 0;
 598         for (i = 0; i < obj->objcount; i++) {
 599                 struct ll_fid fid = obj->objs[i].fid;
 600                 struct lustre_handle *lockh = NULL;
 601                 struct ptlrpc_request *req = NULL;
 602                 ldlm_blocking_callback cb;
 603                 struct lookup_intent it;
 604                 int master = 0;
 605
 606                 CDEBUG(D_OTHER, "revalidate subobj %lu/%lu/%lu\n",
 607                        (unsigned long) fid.mds,
 608                        (unsigned long) fid.id,
 609                        (unsigned long) fid.generation);
 610
 611                 memset(&it, 0, sizeof(it));
 612                 it.it_op = IT_GETATTR;
 613                 cb = lmv_dirobj_blocking_ast;
 614
 615                 if (fid_equal(&fid, &obj->fid)) {
 616                         if (master_valid) {
 617                                 /* lmv_intent_getattr() already checked
 618                                  * validness and took the lock */
 619                                 if (mreq) {
 620                                         /* it even got the reply
 621                                          * refresh attrs from that reply */
 622                                         body = lustre_msg_buf(mreq->rq_repmsg,
 623                                                               1,sizeof(*body));
 624                                         LASSERT(body != NULL);
 625                                         goto update;
 626                                 }
 627                                 /* take already cached attrs into account */
 628                                 CDEBUG(D_OTHER,
 629                                        "master is locked and cached\n");
 630                                 goto release_lock;
 631                         }
 632                         master = 1;
 633                         cb = cb_blocking;
 634                 }
 635
 636                 /* is obj valid? */
 637                 rc = md_intent_lock(lmv->tgts[fid.mds].exp, &uctxt, &fid,
 638                                     NULL, 0, NULL, 0, &fid, &it, 0, &req, cb);
 639                 lockh = (struct lustre_handle *) &it.d.lustre.it_lock_handle;
 640                 if (rc > 0) {
 641                         /* nice, this slave is valid */
 642                         LASSERT(req == NULL);
 643                         CDEBUG(D_OTHER, "cached\n");
 644                         goto release_lock;
 645                 }
 646
 647                 if (rc < 0) {
 648                         /* error during revalidation */
 649                         GOTO(cleanup, rc);
 650                 }
 651
 652                 /* rc == 0, this means we have no such a lock and can't
 653                  * think obj is still valid. lookup it again */
 654                 LASSERT(req == NULL);
 655                 req = NULL;
 656                 memset(&it, 0, sizeof(it));
 657                 it.it_op = IT_GETATTR;
 658                 rc = md_intent_lock(lmv->tgts[fid.mds].exp, &uctxt, &fid,
 659                                     NULL, 0, NULL, 0, NULL, &it, 0, &req, cb);
 660                 lockh = (struct lustre_handle *) &it.d.lustre.it_lock_handle;
 661                 LASSERT(rc <= 0);
 662                 if (rc < 0) {
 663                         /* error during lookup */
 664                         GOTO(cleanup, rc);
 665                 }
 666
 667                 if (master) {
 668                         LASSERT(master_valid == 0);
 669                         /* save lock on master to be returned to the caller */
 670                         CDEBUG(D_OTHER, "no lock on master yet\n");
 671                         memcpy(&master_lockh, lockh, sizeof(master_lockh));
 672                         master_lock_mode = it.d.lustre.it_lock_mode;
 673                         it.d.lustre.it_lock_mode = 0;
 674                 } else {
 675                         /* this is slave. we want to control it */
 676                         lock = ldlm_handle2lock(lockh);
 677                         LASSERT(lock);
 678                         lock->l_ast_data = obj;
 679                         atomic_inc(&obj->count);
 680                         LDLM_LOCK_PUT(lock);
 681                 }
 682
 683                 if (*reqp == NULL) {
 684                         /* this is first reply, we'll use it to return
 685                          * updated data back to the caller */
 686                         LASSERT(req);
 687                         ptlrpc_request_addref(req);
 688                         *reqp = req;
 689
 690                 }
 691
 692                 body = lustre_msg_buf(req->rq_repmsg, 1, sizeof(*body));
 693                 LASSERT(body);
 694
 695 update:
 696                 obj->objs[i].size = body->size;
 697                 CDEBUG(D_OTHER, "fresh: %lu\n",
 698                        (unsigned long) obj->objs[i].size);
 699
 700                 if (req)
 701                         ptlrpc_req_finished(req);
 702 release_lock:
 703                 size += obj->objs[i].size;
 704                 if (it.d.lustre.it_lock_mode)
 705                         ldlm_lock_decref(lockh, it.d.lustre.it_lock_mode);
 706         }
 707
 708         if (*reqp) {
 709                 /* some attrs got refreshed, we have reply and it's time
 710                  * to put fresh attrs to it */
 711                 CDEBUG(D_OTHER, "return refreshed attrs: size = %lu\n",
 712                        (unsigned long) size);
 713                 body = lustre_msg_buf((*reqp)->rq_repmsg, 1, sizeof(*body));
 714                 LASSERT(body);
 715                 /* FIXME: what about another attributes? */
 716                 body->size = size;
 717                 if (mreq == NULL) {
 718                         /* very important to maintain lli->mds the same
 719                          * because of revalidation. mreq == NULL means
 720                          * that caller has no reply and the only attr
 721                          * we can return is size */
 722                         body->valid = OBD_MD_FLSIZE;
 723                         body->mds = obj->fid.mds;
 724                 }
 725                 if (master_valid == 0) {
 726                         memcpy(&oit->d.lustre.it_lock_handle,
 727                                &master_lockh, sizeof(master_lockh));
 728                         oit->d.lustre.it_lock_mode = master_lock_mode;
 729                 }
 730                 rc = 0;
 731         } else {
 732                 /* it seems all the attrs are fresh and we did no request */
 733                 CDEBUG(D_OTHER, "all the attrs were fresh\n");
 734                 if (master_valid == 0)
 735                         oit->d.lustre.it_lock_mode = master_lock_mode;
 736                 rc = 1;
 737         }
 738 cleanup:
 739         RETURN(rc);
 740 }
 741