lustre/llite/dir.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/dir.c
  37  *
  38  * Directory code for lustre client.
  39  */
  40
  41 #include <linux/fs.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/mm.h>
  44 #include <linux/version.h>
  45 #include <linux/smp_lock.h>
  46 #include <asm/uaccess.h>
  47 #include <linux/buffer_head.h>   // for wait_on_buffer
  48
  49 #define DEBUG_SUBSYSTEM S_LLITE
  50
  51 #include <obd_support.h>
  52 #include <obd_class.h>
  53 #include <lustre_lib.h>
  54 #include <lustre/lustre_idl.h>
  55 #include <lustre_lite.h>
  56 #include <lustre_dlm.h>
  57 #include <lustre_fid.h>
  58 #include "llite_internal.h"
  59
  60 #ifndef HAVE_PAGE_CHECKED
  61 #ifdef HAVE_PG_FS_MISC
  62 #define PageChecked(page)        test_bit(PG_fs_misc, &(page)->flags)
  63 #define SetPageChecked(page)     set_bit(PG_fs_misc, &(page)->flags)
  64 #else
  65 #error PageChecked or PageFsMisc not defined in kernel
  66 #endif
  67 #endif
  68
  69 /*
  70  * (new) readdir implementation overview.
  71  *
  72  * Original lustre readdir implementation cached exact copy of raw directory
  73  * pages on the client. These pages were indexed in client page cache by
  74  * logical offset in the directory file. This design, while very simple and
  75  * intuitive had some inherent problems:
  76  *
  77  *     . it implies that byte offset to the directory entry serves as a
  78  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  79  *     ext3/htree directory entries may move due to splits, and more
  80  *     importantly,
  81  *
  82  *     . it is incompatible with the design of split directories for cmd3,
  83  *     that assumes that names are distributed across nodes based on their
  84  *     hash, and so readdir should be done in hash order.
  85  *
  86  * New readdir implementation does readdir in hash order, and uses hash of a
  87  * file name as a telldir/seekdir cookie. This led to number of complications:
  88  *
  89  *     . hash is not unique, so it cannot be used to index cached directory
  90  *     pages on the client (note, that it requires a whole pageful of hash
  91  *     collided entries to cause two pages to have identical hashes);
  92  *
  93  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  94  *     entry cookie. ext3/htree has the same problem and lustre implementation
  95  *     mimics their solution: seekdir(hash) positions directory at the first
  96  *     entry with the given hash.
  97  *
  98  * Client side.
  99  *
 100  * 0. caching
 101  *
 102  * Client caches directory pages using hash of the first entry as an index. As
 103  * noted above hash is not unique, so this solution doesn't work as is:
 104  * special processing is needed for "page hash chains" (i.e., sequences of
 105  * pages filled with entries all having the same hash value).
 106  *
 107  * First, such chains have to be detected. To this end, server returns to the
 108  * client the hash of the first entry on the page next to one returned. When
 109  * client detects that this hash is the same as hash of the first entry on the
 110  * returned page, page hash collision has to be handled. Pages in the
 111  * hash chain, except first one, are termed "overflow pages".
 112  *
 113  * Solution to index uniqueness problem is to not cache overflow
 114  * pages. Instead, when page hash collision is detected, all overflow pages
 115  * from emerging chain are immediately requested from the server and placed in
 116  * a special data structure (struct ll_dir_chain). This data structure is used
 117  * by ll_readdir() to process entries from overflow pages. When readdir
 118  * invocation finishes, overflow pages are discarded. If page hash collision
 119  * chain weren't completely processed, next call to readdir will again detect
 120  * page hash collision, again read overflow pages in, process next portion of
 121  * entries and again discard the pages. This is not as wasteful as it looks,
 122  * because, given reasonable hash, page hash collisions are extremely rare.
 123  *
 124  * 1. directory positioning
 125  *
 126  * When seekdir(hash) is called, original
 127  *
 128  *
 129  *
 130  *
 131  *
 132  *
 133  *
 134  *
 135  * Server.
 136  *
 137  * identification of and access to overflow pages
 138  *
 139  * page format
 140  *
 141  *
 142  *
 143  *
 144  *
 145  */
 146
 147 /* returns the page unlocked, but with a reference */
 148 static int ll_dir_readpage(struct file *file, struct page *page)
 149 {
 150         struct inode *inode = page->mapping->host;
 151         struct ptlrpc_request *request;
 152         struct mdt_body *body;
 153         struct obd_capa *oc;
 154         __u64 hash;
 155         int rc;
 156         ENTRY;
 157
 158         hash = (__u64)hash_x_index(page->index);
 159         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) off %lu\n",
 160                inode->i_ino, inode->i_generation, inode, (unsigned long)hash);
 161
 162         oc = ll_mdscapa_get(inode);
 163         rc = md_readpage(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode),
 164                          oc, hash, page, &request);
 165         capa_put(oc);
 166         if (!rc) {
 167                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 168                 /* Checked by mdc_readpage() */
 169                 LASSERT(body != NULL);
 170
 171                 if (body->valid & OBD_MD_FLSIZE) {
 172                         ll_inode_size_lock(inode, 0);
 173                         i_size_write(inode, body->size);
 174                         ll_inode_size_unlock(inode, 0);
 175                 }
 176                 SetPageUptodate(page);
 177         }
 178         ptlrpc_req_finished(request);
 179
 180         unlock_page(page);
 181         EXIT;
 182         return rc;
 183 }
 184
 185 struct address_space_operations ll_dir_aops = {
 186         .readpage  = ll_dir_readpage,
 187 };
 188
 189 static void ll_check_page(struct inode *dir, struct page *page)
 190 {
 191         /* XXX: check page format later */
 192         SetPageChecked(page);
 193 }
 194
 195 static void ll_release_page(struct page *page, __u64 hash,
 196                             __u64 start, __u64 end)
 197 {
 198         kunmap(page);
 199         lock_page(page);
 200         if (likely(page->mapping != NULL)) {
 201                 ll_truncate_complete_page(page);
 202                 unlock_page(page);
 203         } else {
 204                 unlock_page(page);
 205                 CWARN("NULL mapping page %p, truncated by others: "
 206                       "hash("LPX64") | start("LPX64") | end("LPX64")\n",
 207                       page, hash, start, end);
 208         }
 209         page_cache_release(page);
 210 }
 211
 212 /*
 213  * Find, kmap and return page that contains given hash.
 214  */
 215 static struct page *ll_dir_page_locate(struct inode *dir, __u64 hash,
 216                                        __u64 *start, __u64 *end)
 217 {
 218         struct address_space *mapping = dir->i_mapping;
 219         /*
 220          * Complement of hash is used as an index so that
 221          * radix_tree_gang_lookup() can be used to find a page with starting
 222          * hash _smaller_ than one we are looking for.
 223          */
 224         unsigned long offset = hash_x_index((unsigned long)hash);
 225         struct page *page;
 226         int found;
 227
 228         TREE_READ_LOCK_IRQ(mapping);
 229         found = radix_tree_gang_lookup(&mapping->page_tree,
 230                                        (void **)&page, offset, 1);
 231         if (found > 0) {
 232                 struct lu_dirpage *dp;
 233
 234                 page_cache_get(page);
 235                 TREE_READ_UNLOCK_IRQ(mapping);
 236                 /*
 237                  * In contrast to find_lock_page() we are sure that directory
 238                  * page cannot be truncated (while DLM lock is held) and,
 239                  * hence, can avoid restart.
 240                  *
 241                  * In fact, page cannot be locked here at all, because
 242                  * ll_dir_readpage() does synchronous io.
 243                  */
 244                 wait_on_page(page);
 245                 if (PageUptodate(page)) {
 246                         dp = kmap(page);
 247                         *start = le64_to_cpu(dp->ldp_hash_start);
 248                         *end   = le64_to_cpu(dp->ldp_hash_end);
 249                         LASSERT(*start <= hash);
 250                         if (hash > *end || (*end != *start && hash == *end)) {
 251                                 ll_release_page(page, hash, *start, *end);
 252                                 page = NULL;
 253                         }
 254                 } else {
 255                         page_cache_release(page);
 256                         page = ERR_PTR(-EIO);
 257                 }
 258
 259         } else {
 260                 TREE_READ_UNLOCK_IRQ(mapping);
 261                 page = NULL;
 262         }
 263         return page;
 264 }
 265
 266 struct page *ll_get_dir_page(struct inode *dir, __u64 hash, int exact,
 267                              struct ll_dir_chain *chain)
 268 {
 269         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 270         struct address_space *mapping = dir->i_mapping;
 271         struct lustre_handle lockh;
 272         struct lu_dirpage *dp;
 273         struct page *page;
 274         ldlm_mode_t mode;
 275         int rc;
 276         __u64 start = 0;
 277         __u64 end = 0;
 278
 279         mode = LCK_PR;
 280         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 281                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
 282         if (!rc) {
 283                 struct ldlm_enqueue_info einfo = { LDLM_IBITS, mode,
 284                        ll_md_blocking_ast, ldlm_completion_ast,
 285                        NULL, NULL, dir };
 286                 struct lookup_intent it = { .it_op = IT_READDIR };
 287                 struct ptlrpc_request *request;
 288                 struct md_op_data *op_data;
 289
 290                 op_data = ll_prep_md_op_data(NULL, dir, NULL, NULL, 0, 0,
 291                                              LUSTRE_OPC_ANY, NULL);
 292                 if (IS_ERR(op_data))
 293                         return (void *)op_data;
 294
 295                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
 296                                 op_data, &lockh, NULL, 0, NULL, 0);
 297
 298                 ll_finish_md_op_data(op_data);
 299
 300                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 301                 if (request)
 302                         ptlrpc_req_finished(request);
 303                 if (rc < 0) {
 304                         CERROR("lock enqueue: rc: %d\n", rc);
 305                         return ERR_PTR(rc);
 306                 }
 307         } else {
 308                 /* for cross-ref object, l_ast_data of the lock may not be set,
 309                  * we reset it here */
 310                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
 311                                  dir, NULL);
 312         }
 313         ldlm_lock_dump_handle(D_OTHER, &lockh);
 314
 315         page = ll_dir_page_locate(dir, hash, &start, &end);
 316         if (IS_ERR(page))
 317                 GOTO(out_unlock, page);
 318
 319         if (page != NULL) {
 320                 /*
 321                  * XXX nikita: not entirely correct handling of a corner case:
 322                  * suppose hash chain of entries with hash value HASH crosses
 323                  * border between pages P0 and P1. First both P0 and P1 are
 324                  * cached, seekdir() is called for some entry from the P0 part
 325                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 326                  * happens and finds P1, as it starts with matching hash
 327                  * value. Remaining entries from P0 part of the chain are
 328                  * skipped. (Is that really a bug?)
 329                  *
 330                  * Possible solutions: 0. don't cache P1 is such case, handle
 331                  * it as an "overflow" page. 1. invalidate all pages at
 332                  * once. 2. use HASH|1 as an index for P1.
 333                  */
 334                 if (exact && hash != start) {
 335                         /*
 336                          * readdir asked for a page starting _exactly_ from
 337                          * given hash, but cache contains stale page, with
 338                          * entries with smaller hash values. Stale page should
 339                          * be invalidated, and new one fetched.
 340                          */
 341                         CWARN("Stale readpage page %p: "LPX64" != "LPX64"\n",
 342                               page, hash, start);
 343                         ll_release_page(page, hash, start, end);
 344                 } else {
 345                         GOTO(hash_collision, page);
 346                 }
 347         }
 348
 349         page = read_cache_page(mapping, hash_x_index((unsigned long)hash),
 350                                (filler_t*)mapping->a_ops->readpage, NULL);
 351         if (IS_ERR(page))
 352                 GOTO(out_unlock, page);
 353
 354         wait_on_page(page);
 355         (void)kmap(page);
 356         if (!PageUptodate(page))
 357                 goto fail;
 358         if (!PageChecked(page))
 359                 ll_check_page(dir, page);
 360         if (PageError(page))
 361                 goto fail;
 362 hash_collision:
 363         dp = page_address(page);
 364
 365         start = le64_to_cpu(dp->ldp_hash_start);
 366         end   = le64_to_cpu(dp->ldp_hash_end);
 367         if (end == start) {
 368                 LASSERT(start == hash);
 369                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
 370                 /*
 371                  * Fetch whole overflow chain...
 372                  *
 373                  * XXX not yet.
 374                  */
 375                 goto fail;
 376         }
 377 out_unlock:
 378         ldlm_lock_decref(&lockh, mode);
 379         return page;
 380
 381 fail:
 382         ll_put_page(page);
 383         page = ERR_PTR(-EIO);
 384         goto out_unlock;
 385 }
 386
 387 int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 388 {
 389         struct inode         *inode = filp->f_dentry->d_inode;
 390         struct ll_inode_info *info  = ll_i2info(inode);
 391         struct ll_sb_info    *sbi   = ll_i2sbi(inode);
 392         __u64                 pos   = filp->f_pos;
 393         struct page          *page;
 394         struct ll_dir_chain   chain;
 395         int rc;
 396         int done;
 397         int shift;
 398         __u16 type;
 399         ENTRY;
 400
 401         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu\n",
 402                inode->i_ino, inode->i_generation, inode,
 403                (unsigned long)pos, i_size_read(inode));
 404
 405         if (pos == DIR_END_OFF)
 406                 /*
 407                  * end-of-file.
 408                  */
 409                 RETURN(0);
 410
 411         rc    = 0;
 412         done  = 0;
 413         shift = 0;
 414         ll_dir_chain_init(&chain);
 415
 416         page = ll_get_dir_page(inode, pos, 0, &chain);
 417
 418         while (rc == 0 && !done) {
 419                 struct lu_dirpage *dp;
 420                 struct lu_dirent  *ent;
 421
 422                 if (!IS_ERR(page)) {
 423                         /*
 424                          * If page is empty (end of directory is reached),
 425                          * use this value.
 426                          */
 427                         __u64 hash = DIR_END_OFF;
 428                         __u64 next;
 429
 430                         dp = page_address(page);
 431                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 432                              ent = lu_dirent_next(ent)) {
 433                                 char          *name;
 434                                 int            namelen;
 435                                 struct lu_fid  fid;
 436                                 ino_t          ino;
 437
 438                                 /*
 439                                  * XXX: implement correct swabbing here.
 440                                  */
 441
 442                                 hash    = le64_to_cpu(ent->lde_hash);
 443                                 namelen = le16_to_cpu(ent->lde_namelen);
 444
 445                                 if (hash < pos)
 446                                         /*
 447                                          * Skip until we find target hash
 448                                          * value.
 449                                          */
 450                                         continue;
 451
 452                                 if (namelen == 0)
 453                                         /*
 454                                          * Skip dummy record.
 455                                          */
 456                                         continue;
 457
 458                                 fid  = ent->lde_fid;
 459                                 name = ent->lde_name;
 460                                 fid_le_to_cpu(&fid, &fid);
 461                                 ino  = ll_fid_build_ino(sbi, &fid);
 462                                 type = ll_dirent_type_get(ent);
 463                                 done = filldir(cookie, name, namelen,
 464                                                (loff_t)hash, ino, type);
 465                         }
 466                         next = le64_to_cpu(dp->ldp_hash_end);
 467                         ll_put_page(page);
 468                         if (!done) {
 469                                 pos = next;
 470                                 if (pos == DIR_END_OFF)
 471                                         /*
 472                                          * End of directory reached.
 473                                          */
 474                                         done = 1;
 475                                 else if (1 /* chain is exhausted*/)
 476                                         /*
 477                                          * Normal case: continue to the next
 478                                          * page.
 479                                          */
 480                                         page = ll_get_dir_page(inode, pos, 1,
 481                                                                &chain);
 482                                 else {
 483                                         /*
 484                                          * go into overflow page.
 485                                          */
 486                                 }
 487                         } else
 488                                 pos = hash;
 489                 } else {
 490                         rc = PTR_ERR(page);
 491                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 492                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 493                 }
 494         }
 495
 496         filp->f_pos = (loff_t)pos;
 497         filp->f_version = inode->i_version;
 498         touch_atime(filp->f_vfsmnt, filp->f_dentry);
 499
 500         ll_dir_chain_fini(&chain);
 501
 502         RETURN(rc);
 503 }
 504
 505 int ll_send_mgc_param(struct obd_export *mgc, char *string)
 506 {
 507         struct mgs_send_param *msp;
 508         int rc = 0;
 509
 510         OBD_ALLOC_PTR(msp);
 511         if (!msp)
 512                 return -ENOMEM;
 513
 514         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 515         rc = obd_set_info_async(mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 516                                 sizeof(struct mgs_send_param), msp, NULL);
 517         if (rc)
 518                 CERROR("Failed to set parameter: %d\n", rc);
 519         OBD_FREE_PTR(msp);
 520
 521         return rc;
 522 }
 523
 524 char *ll_get_fsname(struct inode *inode)
 525 {
 526         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 527         char *ptr, *fsname;
 528         int len;
 529
 530         OBD_ALLOC(fsname, MGS_PARAM_MAXLEN);
 531         len = strlen(lsi->lsi_lmd->lmd_profile);
 532         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
 533         if (ptr && (strcmp(ptr, "-client") == 0))
 534                 len -= 7;
 535         strncpy(fsname, lsi->lsi_lmd->lmd_profile, len);
 536         fsname[len] = '\0';
 537
 538         return fsname;
 539 }
 540
 541 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 542                      int set_default)
 543 {
 544         struct ll_sb_info *sbi = ll_i2sbi(inode);
 545         struct md_op_data *op_data;
 546         struct ptlrpc_request *req = NULL;
 547         int rc = 0;
 548         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 549         struct obd_device *mgc = lsi->lsi_mgc;
 550         char *fsname = NULL, *param = NULL;
 551         int lum_size;
 552
 553         /*
 554          * This is coming from userspace, so should be in
 555          * local endian.  But the MDS would like it in little
 556          * endian, so we swab it before we send it.
 557          */
 558         switch (lump->lmm_magic) {
 559         case LOV_USER_MAGIC_V1: {
 560                 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 561                         lustre_swab_lov_user_md_v1(lump);
 562                 lum_size = sizeof(struct lov_user_md_v1);
 563                 break;
 564                 }
 565         case LOV_USER_MAGIC_V3: {
 566                 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 567                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lump);
 568                 lum_size = sizeof(struct lov_user_md_v3);
 569                 break;
 570                 }
 571         default: {
 572                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 573                                 " %#08x != %#08x nor %#08x\n",
 574                                 lump->lmm_magic, LOV_USER_MAGIC_V1,
 575                                 LOV_USER_MAGIC_V3);
 576                 RETURN(-EINVAL);
 577                 }
 578         }
 579
 580         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 581                                      LUSTRE_OPC_ANY, NULL);
 582         if (IS_ERR(op_data))
 583                 RETURN(PTR_ERR(op_data));
 584
 585         /* swabbing is done in lov_setstripe() on server side */
 586         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 587                         NULL, 0, &req, NULL);
 588         ll_finish_md_op_data(op_data);
 589         ptlrpc_req_finished(req);
 590         if (rc) {
 591                 if (rc != -EPERM && rc != -EACCES)
 592                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 593         }
 594
 595         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 596          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 597          need the make the distiction between the 2 versions */
 598         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 599                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 600
 601                 /* Get fsname and assume devname to be -MDT0000. */
 602                 fsname = ll_get_fsname(inode);
 603                 /* Set root stripesize */
 604                 sprintf(param, "%s-MDT0000.lov.stripesize=%u", fsname,
 605                         lump->lmm_stripe_size);
 606                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 607                 if (rc)
 608                         goto end;
 609
 610                 /* Set root stripecount */
 611                 sprintf(param, "%s-MDT0000.lov.stripecount=%hd", fsname,
 612                         lump->lmm_stripe_count);
 613                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 614                 if (rc)
 615                         goto end;
 616
 617                 /* Set root stripeoffset */
 618                 sprintf(param, "%s-MDT0000.lov.stripeoffset=%hd", fsname,
 619                         lump->lmm_stripe_offset);
 620                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 621                 if (rc)
 622                         goto end;
 623 end:
 624                 if (fsname)
 625                         OBD_FREE(fsname, MGS_PARAM_MAXLEN);
 626                 if (param)
 627                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 628         }
 629         return rc;
 630 }
 631
 632 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 633                      int *lmm_size, struct ptlrpc_request **request)
 634 {
 635         struct ll_sb_info *sbi = ll_i2sbi(inode);
 636         struct mdt_body   *body;
 637         struct lov_mds_md *lmm = NULL;
 638         struct ptlrpc_request *req = NULL;
 639         int rc, lmmsize;
 640         struct obd_capa *oc;
 641
 642         rc = ll_get_max_mdsize(sbi, &lmmsize);
 643         if (rc)
 644                 RETURN(rc);
 645
 646         oc = ll_mdscapa_get(inode);
 647         rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode),
 648                         oc, OBD_MD_FLEASIZE | OBD_MD_FLDIREA,
 649                         lmmsize, &req);
 650         capa_put(oc);
 651         if (rc < 0) {
 652                 CDEBUG(D_INFO, "md_getattr failed on inode "
 653                        "%lu/%u: rc %d\n", inode->i_ino,
 654                        inode->i_generation, rc);
 655                 GOTO(out, rc);
 656         }
 657
 658         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 659         LASSERT(body != NULL);
 660
 661         lmmsize = body->eadatasize;
 662
 663         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 664             lmmsize == 0) {
 665                 GOTO(out, rc = -ENODATA);
 666         }
 667
 668         lmm = req_capsule_server_sized_get(&req->rq_pill,
 669                                            &RMF_MDT_MD, lmmsize);
 670         LASSERT(lmm != NULL);
 671
 672         /*
 673          * This is coming from the MDS, so is probably in
 674          * little endian.  We convert it to host endian before
 675          * passing it to userspace.
 676          */
 677         /* We don't swab objects for directories */
 678         switch (le32_to_cpu(lmm->lmm_magic)) {
 679         case LOV_MAGIC_V1:
 680                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 681                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 682                 break;
 683         case LOV_MAGIC_V3:
 684                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 685                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 686                 break;
 687         default:
 688                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 689                 rc = -EPROTO;
 690         }
 691 out:
 692         *lmmp = lmm;
 693         *lmm_size = lmmsize;
 694         *request = req;
 695         return rc;
 696 }
 697
 698 static int ll_dir_ioctl(struct inode *inode, struct file *file,
 699                         unsigned int cmd, unsigned long arg)
 700 {
 701         struct ll_sb_info *sbi = ll_i2sbi(inode);
 702         struct obd_ioctl_data *data;
 703         ENTRY;
 704
 705         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
 706                inode->i_ino, inode->i_generation, inode, cmd);
 707
 708         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
 709         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
 710                 return -ENOTTY;
 711
 712         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
 713         switch(cmd) {
 714         case EXT3_IOC_GETFLAGS:
 715         case EXT3_IOC_SETFLAGS:
 716                 RETURN(ll_iocontrol(inode, file, cmd, arg));
 717         case EXT3_IOC_GETVERSION_OLD:
 718         case EXT3_IOC_GETVERSION:
 719                 RETURN(put_user(inode->i_generation, (int *)arg));
 720         /* We need to special case any other ioctls we want to handle,
 721          * to send them to the MDS/OST as appropriate and to properly
 722          * network encode the arg field.
 723         case EXT3_IOC_SETVERSION_OLD:
 724         case EXT3_IOC_SETVERSION:
 725         */
 726         case IOC_MDC_LOOKUP: {
 727                 struct ptlrpc_request *request = NULL;
 728                 int namelen, rc, len = 0;
 729                 char *buf = NULL;
 730                 char *filename;
 731                 struct obd_capa *oc;
 732
 733                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 734                 if (rc)
 735                         RETURN(rc);
 736                 data = (void *)buf;
 737
 738                 filename = data->ioc_inlbuf1;
 739                 namelen = data->ioc_inllen1;
 740
 741                 if (namelen < 1) {
 742                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
 743                         GOTO(out, rc = -EINVAL);
 744                 }
 745
 746                 oc = ll_mdscapa_get(inode);
 747                 rc = md_getattr_name(sbi->ll_md_exp, ll_inode2fid(inode), oc,
 748                                      filename, namelen, OBD_MD_FLID, 0,
 749                                      ll_i2suppgid(inode), &request);
 750                 capa_put(oc);
 751                 if (rc < 0) {
 752                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 753                         GOTO(out, rc);
 754                 }
 755
 756                 ptlrpc_req_finished(request);
 757
 758                 EXIT;
 759         out:
 760                 obd_ioctl_freedata(buf, len);
 761                 return rc;
 762         }
 763         case LL_IOC_LOV_SETSTRIPE: {
 764                 struct lov_user_md_v3 lumv3;
 765                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
 766                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
 767                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
 768
 769                 int rc = 0;
 770                 int set_default = 0;
 771
 772                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
 773                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
 774                         sizeof(lumv3p->lmm_objects[0]));
 775                 /* first try with v1 which is smaller than v3 */
 776                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
 777                         RETURN(-EFAULT);
 778
 779                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
 780                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
 781                                 RETURN(-EFAULT);
 782                 }
 783
 784                 if (inode->i_sb->s_root == file->f_dentry)
 785                         set_default = 1;
 786
 787                 /* in v1 and v3 cases lumv1 points to data */
 788                 rc = ll_dir_setstripe(inode, lumv1, set_default);
 789
 790                 RETURN(rc);
 791         }
 792         case LL_IOC_OBD_STATFS:
 793                 RETURN(ll_obd_statfs(inode, (void *)arg));
 794         case LL_IOC_LOV_GETSTRIPE:
 795         case LL_IOC_MDC_GETINFO:
 796         case IOC_MDC_GETFILEINFO:
 797         case IOC_MDC_GETFILESTRIPE: {
 798                 struct ptlrpc_request *request = NULL;
 799                 struct lov_user_md *lump;
 800                 struct lov_mds_md *lmm = NULL;
 801                 struct mdt_body *body;
 802                 char *filename = NULL;
 803                 int rc, lmmsize;
 804
 805                 if (cmd == IOC_MDC_GETFILEINFO ||
 806                     cmd == IOC_MDC_GETFILESTRIPE) {
 807                         filename = getname((const char *)arg);
 808                         if (IS_ERR(filename))
 809                                 RETURN(PTR_ERR(filename));
 810
 811                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
 812                                                       &lmmsize, &request);
 813                 } else {
 814                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
 815                 }
 816
 817                 if (request) {
 818                         body = req_capsule_server_get(&request->rq_pill,
 819                                                       &RMF_MDT_BODY);
 820                         LASSERT(body != NULL);
 821                 } else {
 822                         GOTO(out_req, rc);
 823                 }
 824
 825                 if (rc < 0) {
 826                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
 827                                                cmd == LL_IOC_MDC_GETINFO))
 828                                 GOTO(skip_lmm, rc = 0);
 829                         else
 830                                 GOTO(out_req, rc);
 831                 }
 832
 833                 if (cmd == IOC_MDC_GETFILESTRIPE ||
 834                     cmd == LL_IOC_LOV_GETSTRIPE) {
 835                         lump = (struct lov_user_md *)arg;
 836                 } else {
 837                         struct lov_user_mds_data *lmdp;
 838                         lmdp = (struct lov_user_mds_data *)arg;
 839                         lump = &lmdp->lmd_lmm;
 840                 }
 841                 if (copy_to_user(lump, lmm, lmmsize))
 842                         GOTO(out_lmm, rc = -EFAULT);
 843         skip_lmm:
 844                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
 845                         struct lov_user_mds_data *lmdp;
 846                         lstat_t st = { 0 };
 847
 848                         st.st_dev     = inode->i_sb->s_dev;
 849                         st.st_mode    = body->mode;
 850                         st.st_nlink   = body->nlink;
 851                         st.st_uid     = body->uid;
 852                         st.st_gid     = body->gid;
 853                         st.st_rdev    = body->rdev;
 854                         st.st_size    = body->size;
 855                         st.st_blksize = CFS_PAGE_SIZE;
 856                         st.st_blocks  = body->blocks;
 857                         st.st_atime   = body->atime;
 858                         st.st_mtime   = body->mtime;
 859                         st.st_ctime   = body->ctime;
 860                         st.st_ino     = inode->i_ino;
 861
 862                         lmdp = (struct lov_user_mds_data *)arg;
 863                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
 864                                 GOTO(out_lmm, rc = -EFAULT);
 865                 }
 866
 867                 EXIT;
 868         out_lmm:
 869                 if (lmm && lmm->lmm_magic == LOV_MAGIC_JOIN)
 870                         OBD_FREE(lmm, lmmsize);
 871         out_req:
 872                 ptlrpc_req_finished(request);
 873                 if (filename)
 874                         putname(filename);
 875                 return rc;
 876         }
 877         case IOC_LOV_GETINFO: {
 878                 struct lov_user_mds_data *lumd;
 879                 struct lov_stripe_md *lsm;
 880                 struct lov_user_md *lum;
 881                 struct lov_mds_md *lmm;
 882                 int lmmsize;
 883                 lstat_t st;
 884                 int rc;
 885
 886                 lumd = (struct lov_user_mds_data *)arg;
 887                 lum = &lumd->lmd_lmm;
 888
 889                 rc = ll_get_max_mdsize(sbi, &lmmsize);
 890                 if (rc)
 891                         RETURN(rc);
 892
 893                 OBD_ALLOC(lmm, lmmsize);
 894                 if (copy_from_user(lmm, lum, lmmsize))
 895                         GOTO(free_lmm, rc = -EFAULT);
 896
 897                 switch (lmm->lmm_magic) {
 898                 case LOV_USER_MAGIC_V1:
 899                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
 900                                 break;
 901                         /* swab objects first so that stripes num will be sane */
 902                         lustre_swab_lov_user_md_objects(
 903                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
 904                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
 905                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 906                         break;
 907                 case LOV_USER_MAGIC_V3:
 908                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
 909                                 break;
 910                         /* swab objects first so that stripes num will be sane */
 911                         lustre_swab_lov_user_md_objects(
 912                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
 913                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
 914                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 915                         break;
 916                 default:
 917                         GOTO(free_lmm, rc = -EINVAL);
 918                 }
 919
 920                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
 921                 if (rc < 0)
 922                         GOTO(free_lmm, rc = -ENOMEM);
 923
 924                 rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp, lsm);
 925                 if (rc)
 926                         GOTO(free_lsm, rc);
 927
 928                 /* Perform glimpse_size operation. */
 929                 memset(&st, 0, sizeof(st));
 930
 931                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
 932                 if (rc)
 933                         GOTO(free_lsm, rc);
 934
 935                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
 936                         GOTO(free_lsm, rc = -EFAULT);
 937
 938                 EXIT;
 939         free_lsm:
 940                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
 941         free_lmm:
 942                 OBD_FREE(lmm, lmmsize);
 943                 return rc;
 944         }
 945         case OBD_IOC_LLOG_CATINFO: {
 946                 struct ptlrpc_request *req = NULL;
 947                 char                  *buf = NULL;
 948                 char                  *str;
 949                 int                    len = 0;
 950                 int                    rc;
 951
 952                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 953                 if (rc)
 954                         RETURN(rc);
 955                 data = (void *)buf;
 956
 957                 if (!data->ioc_inlbuf1) {
 958                         obd_ioctl_freedata(buf, len);
 959                         RETURN(-EINVAL);
 960                 }
 961
 962                 req = ptlrpc_request_alloc(sbi2mdc(sbi)->cl_import,
 963                                            &RQF_LLOG_CATINFO);
 964                 if (req == NULL)
 965                         GOTO(out_catinfo, rc = -ENOMEM);
 966
 967                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 968                                      data->ioc_inllen1);
 969                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_CLIENT,
 970                                      data->ioc_inllen2);
 971
 972                 rc = ptlrpc_request_pack(req, LUSTRE_LOG_VERSION, LLOG_CATINFO);
 973                 if (rc) {
 974                         ptlrpc_request_free(req);
 975                         GOTO(out_catinfo, rc);
 976                 }
 977
 978                 str = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 979                 memcpy(str, data->ioc_inlbuf1, data->ioc_inllen1);
 980                 if (data->ioc_inllen2) {
 981                         str = req_capsule_client_get(&req->rq_pill,
 982                                                      &RMF_STRING);
 983                         memcpy(str, data->ioc_inlbuf2, data->ioc_inllen2);
 984                 }
 985
 986                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_SERVER,
 987                                      data->ioc_plen1);
 988                 ptlrpc_request_set_replen(req);
 989
 990                 rc = ptlrpc_queue_wait(req);
 991                 if (!rc) {
 992                         str = req_capsule_server_get(&req->rq_pill,
 993                                                      &RMF_STRING);
 994                         if (copy_to_user(data->ioc_pbuf1, str, data->ioc_plen1))
 995                                 rc = -EFAULT;
 996                 }
 997                 ptlrpc_req_finished(req);
 998         out_catinfo:
 999                 obd_ioctl_freedata(buf, len);
1000                 RETURN(rc);
1001         }
1002         case OBD_IOC_QUOTACHECK: {
1003                 struct obd_quotactl *oqctl;
1004                 int rc, error = 0;
1005
1006                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1007                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1008                         RETURN(-EPERM);
1009
1010                 OBD_ALLOC_PTR(oqctl);
1011                 if (!oqctl)
1012                         RETURN(-ENOMEM);
1013                 oqctl->qc_type = arg;
1014                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1015                 if (rc < 0) {
1016                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1017                         error = rc;
1018                 }
1019
1020                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1021                 if (rc < 0)
1022                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1023
1024                 OBD_FREE_PTR(oqctl);
1025                 return error ?: rc;
1026         }
1027         case OBD_IOC_POLL_QUOTACHECK: {
1028                 struct if_quotacheck *check;
1029                 int rc;
1030
1031                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1032                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1033                         RETURN(-EPERM);
1034
1035                 OBD_ALLOC_PTR(check);
1036                 if (!check)
1037                         RETURN(-ENOMEM);
1038
1039                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1040                                    NULL);
1041                 if (rc) {
1042                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1043                         if (copy_to_user((void *)arg, check, sizeof(*check)))
1044                                 rc = -EFAULT;
1045                         GOTO(out_poll, rc);
1046                 }
1047
1048                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1049                                    NULL);
1050                 if (rc) {
1051                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1052                         if (copy_to_user((void *)arg, check, sizeof(*check)))
1053                                 rc = -EFAULT;
1054                         GOTO(out_poll, rc);
1055                 }
1056         out_poll:
1057                 OBD_FREE_PTR(check);
1058                 RETURN(rc);
1059         }
1060         case OBD_IOC_QUOTACTL: {
1061                 struct if_quotactl *qctl;
1062                 int cmd, type, id, valid, rc = 0;
1063
1064                 OBD_ALLOC_PTR(qctl);
1065                 if (!qctl)
1066                         RETURN(-ENOMEM);
1067
1068                 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1069                         GOTO(out_quotactl, rc = -EFAULT);
1070
1071                 cmd = qctl->qc_cmd;
1072                 type = qctl->qc_type;
1073                 id = qctl->qc_id;
1074                 valid = qctl->qc_valid;
1075
1076                 switch (cmd) {
1077                 case LUSTRE_Q_INVALIDATE:
1078                 case LUSTRE_Q_FINVALIDATE:
1079                 case Q_QUOTAON:
1080                 case Q_QUOTAOFF:
1081                 case Q_SETQUOTA:
1082                 case Q_SETINFO:
1083                         if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1084                             sbi->ll_flags & LL_SBI_RMT_CLIENT)
1085                                 GOTO(out_quotactl, rc = -EPERM);
1086                         break;
1087                 case Q_GETQUOTA:
1088                         if (((type == USRQUOTA && current->euid != id) ||
1089                              (type == GRPQUOTA && !in_egroup_p(id))) &&
1090                             (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1091                              sbi->ll_flags & LL_SBI_RMT_CLIENT))
1092                                 GOTO(out_quotactl, rc = -EPERM);
1093                         break;
1094                 case Q_GETINFO:
1095                         break;
1096                 default:
1097                         CERROR("unsupported quotactl op: %#x\n", cmd);
1098                         GOTO(out_quotactl, rc = -ENOTTY);
1099                 }
1100
1101                 if (valid != QC_GENERAL) {
1102                         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1103                                 GOTO(out_quotactl, rc = -EOPNOTSUPP);
1104
1105                         if (cmd == Q_GETINFO)
1106                                 qctl->qc_cmd = Q_GETOINFO;
1107                         else if (cmd == Q_GETQUOTA)
1108                                 qctl->qc_cmd = Q_GETOQUOTA;
1109                         else
1110                                 GOTO(out_quotactl, rc = -EINVAL);
1111
1112                         switch (valid) {
1113                         case QC_MDTIDX:
1114                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1115                                                    sbi->ll_md_exp,
1116                                                    sizeof(*qctl), qctl, NULL);
1117                                 break;
1118                         case QC_OSTIDX:
1119                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1120                                                    sbi->ll_dt_exp,
1121                                                    sizeof(*qctl), qctl, NULL);
1122                                 break;
1123                         case QC_UUID:
1124                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1125                                                    sbi->ll_md_exp,
1126                                                    sizeof(*qctl), qctl, NULL);
1127                                 if (rc == -EAGAIN)
1128                                         rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1129                                                            sbi->ll_dt_exp,
1130                                                            sizeof(*qctl), qctl,
1131                                                            NULL);
1132                                 break;
1133                         default:
1134                                 rc = -EINVAL;
1135                                 break;
1136                         }
1137
1138                         if (rc)
1139                                 GOTO(out_quotactl, rc);
1140                         else
1141                                 qctl->qc_cmd = cmd;
1142                 } else {
1143                         struct obd_quotactl *oqctl;
1144
1145                         OBD_ALLOC_PTR(oqctl);
1146                         if (!oqctl)
1147                                 GOTO(out_quotactl, rc = -ENOMEM);
1148
1149                         QCTL_COPY(oqctl, qctl);
1150                         rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1151                         if (rc) {
1152                                 if (rc != -EBUSY && cmd == Q_QUOTAON) {
1153                                         oqctl->qc_cmd = Q_QUOTAOFF;
1154                                         obd_quotactl(sbi->ll_md_exp, oqctl);
1155                                 }
1156                                 OBD_FREE_PTR(oqctl);
1157                                 GOTO(out_quotactl, rc);
1158                         } else {
1159                                 QCTL_COPY(qctl, oqctl);
1160                                 OBD_FREE_PTR(oqctl);
1161                         }
1162                 }
1163
1164                 if (copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1165                         rc = -EFAULT;
1166
1167         out_quotactl:
1168                 OBD_FREE_PTR(qctl);
1169                 RETURN(rc);
1170         }
1171         case OBD_IOC_GETNAME: {
1172                 struct obd_device *obd = class_exp2obd(sbi->ll_dt_exp);
1173                 if (!obd)
1174                         RETURN(-EFAULT);
1175                 if (copy_to_user((void *)arg, obd->obd_name,
1176                                 strlen(obd->obd_name) + 1))
1177                         RETURN (-EFAULT);
1178                 RETURN(0);
1179         }
1180         case LL_IOC_FLUSHCTX:
1181                 RETURN(ll_flush_ctx(inode));
1182 #ifdef CONFIG_FS_POSIX_ACL
1183         case LL_IOC_RMTACL: {
1184             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1185                 inode == inode->i_sb->s_root->d_inode) {
1186                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1187                 int rc;
1188
1189                 LASSERT(fd != NULL);
1190                 rc = rct_add(&sbi->ll_rct, cfs_curproc_pid(), arg);
1191                 if (!rc)
1192                         fd->fd_flags |= LL_FILE_RMTACL;
1193                 RETURN(rc);
1194             } else
1195                 RETURN(0);
1196         }
1197 #endif
1198         case LL_IOC_GETOBDCOUNT: {
1199                 int count;
1200
1201                 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1202                         RETURN(-EFAULT);
1203
1204                 if (!count) {
1205                         /* get ost count */
1206                         struct lov_obd *lov = &sbi->ll_dt_exp->exp_obd->u.lov;
1207                         count = lov->desc.ld_tgt_count;
1208                 } else {
1209                         /* get mdt count */
1210                         struct lmv_obd *lmv = &sbi->ll_md_exp->exp_obd->u.lmv;
1211                         count = lmv->desc.ld_tgt_count;
1212                 }
1213
1214                 if (copy_to_user((int *)arg, &count, sizeof(int)))
1215                         RETURN(-EFAULT);
1216
1217                 RETURN(0);
1218         }
1219         case LL_IOC_PATH2FID:
1220                 if (copy_to_user((void *)arg, &ll_i2info(inode)->lli_fid,
1221                                  sizeof(struct lu_fid)))
1222                         RETURN(-EFAULT);
1223                 RETURN(0);
1224         case OBD_IOC_CHANGELOG_CLEAR: {
1225                 struct ioc_changelog_clear *icc;
1226                 int rc;
1227
1228                 OBD_ALLOC_PTR(icc);
1229                 if (icc == NULL)
1230                         RETURN(-ENOMEM);
1231                 if (copy_from_user(icc, (void *)arg, sizeof(*icc)))
1232                         GOTO(icc_free, rc = -EFAULT);
1233
1234                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(*icc), icc,NULL);
1235
1236 icc_free:
1237                 OBD_FREE_PTR(icc);
1238                 RETURN(rc);
1239         }
1240         case OBD_IOC_FID2PATH:
1241                 RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg));
1242
1243         default:
1244                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp,0,NULL,(void *)arg));
1245         }
1246 }
1247
1248 int ll_dir_open(struct inode *inode, struct file *file)
1249 {
1250         ENTRY;
1251         RETURN(ll_file_open(inode, file));
1252 }
1253
1254 int ll_dir_release(struct inode *inode, struct file *file)
1255 {
1256         ENTRY;
1257         RETURN(ll_file_release(inode, file));
1258 }
1259
1260 struct file_operations ll_dir_operations = {
1261         .open     = ll_dir_open,
1262         .release  = ll_dir_release,
1263         .read     = generic_read_dir,
1264         .readdir  = ll_readdir,
1265         .ioctl    = ll_dir_ioctl
1266 };