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((__u32)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, dir);
 311         }
 312         ldlm_lock_dump_handle(D_OTHER, &lockh);
 313
 314         page = ll_dir_page_locate(dir, hash, &start, &end);
 315         if (IS_ERR(page))
 316                 GOTO(out_unlock, page);
 317
 318         if (page != NULL) {
 319                 /*
 320                  * XXX nikita: not entirely correct handling of a corner case:
 321                  * suppose hash chain of entries with hash value HASH crosses
 322                  * border between pages P0 and P1. First both P0 and P1 are
 323                  * cached, seekdir() is called for some entry from the P0 part
 324                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 325                  * happens and finds P1, as it starts with matching hash
 326                  * value. Remaining entries from P0 part of the chain are
 327                  * skipped. (Is that really a bug?)
 328                  *
 329                  * Possible solutions: 0. don't cache P1 is such case, handle
 330                  * it as an "overflow" page. 1. invalidate all pages at
 331                  * once. 2. use HASH|1 as an index for P1.
 332                  */
 333                 if (exact && hash != start) {
 334                         /*
 335                          * readdir asked for a page starting _exactly_ from
 336                          * given hash, but cache contains stale page, with
 337                          * entries with smaller hash values. Stale page should
 338                          * be invalidated, and new one fetched.
 339                          */
 340                         CWARN("Stale readpage page %p: "LPX64" != "LPX64"\n",
 341                               page, hash, start);
 342                         ll_release_page(page, hash, start, end);
 343                 } else {
 344                         GOTO(hash_collision, page);
 345                 }
 346         }
 347
 348         page = read_cache_page(mapping, hash_x_index((__u32)hash),
 349                                (filler_t*)mapping->a_ops->readpage, NULL);
 350         if (IS_ERR(page))
 351                 GOTO(out_unlock, page);
 352
 353         wait_on_page(page);
 354         (void)kmap(page);
 355         if (!PageUptodate(page))
 356                 goto fail;
 357         if (!PageChecked(page))
 358                 ll_check_page(dir, page);
 359         if (PageError(page))
 360                 goto fail;
 361 hash_collision:
 362         dp = page_address(page);
 363
 364         start = le64_to_cpu(dp->ldp_hash_start);
 365         end   = le64_to_cpu(dp->ldp_hash_end);
 366         if (end == start) {
 367                 LASSERT(start == hash);
 368                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
 369                 /*
 370                  * Fetch whole overflow chain...
 371                  *
 372                  * XXX not yet.
 373                  */
 374                 goto fail;
 375         }
 376 out_unlock:
 377         ldlm_lock_decref(&lockh, mode);
 378         return page;
 379
 380 fail:
 381         ll_put_page(page);
 382         page = ERR_PTR(-EIO);
 383         goto out_unlock;
 384 }
 385
 386 int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 387 {
 388         struct inode         *inode = filp->f_dentry->d_inode;
 389         struct ll_inode_info *info  = ll_i2info(inode);
 390         struct ll_sb_info    *sbi   = ll_i2sbi(inode);
 391         __u64                 pos   = filp->f_pos;
 392         struct page          *page;
 393         struct ll_dir_chain   chain;
 394         int rc;
 395         int done;
 396         int shift;
 397         ENTRY;
 398
 399         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu\n",
 400                inode->i_ino, inode->i_generation, inode,
 401                (unsigned long)pos, i_size_read(inode));
 402
 403         if (pos == DIR_END_OFF)
 404                 /*
 405                  * end-of-file.
 406                  */
 407                 RETURN(0);
 408
 409         rc    = 0;
 410         done  = 0;
 411         shift = 0;
 412         ll_dir_chain_init(&chain);
 413
 414         page = ll_get_dir_page(inode, pos, 0, &chain);
 415
 416         while (rc == 0 && !done) {
 417                 struct lu_dirpage *dp;
 418                 struct lu_dirent  *ent;
 419
 420                 if (!IS_ERR(page)) {
 421                         /*
 422                          * If page is empty (end of directory is reached),
 423                          * use this value.
 424                          */
 425                         __u64 hash = DIR_END_OFF;
 426                         __u64 next;
 427
 428                         dp = page_address(page);
 429                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 430                              ent = lu_dirent_next(ent)) {
 431                                 char          *name;
 432                                 int            namelen;
 433                                 struct lu_fid  fid;
 434                                 ino_t          ino;
 435
 436                                 /*
 437                                  * XXX: implement correct swabbing here.
 438                                  */
 439
 440                                 hash    = le64_to_cpu(ent->lde_hash);
 441                                 namelen = le16_to_cpu(ent->lde_namelen);
 442
 443                                 if (hash < pos)
 444                                         /*
 445                                          * Skip until we find target hash
 446                                          * value.
 447                                          */
 448                                         continue;
 449
 450                                 if (namelen == 0)
 451                                         /*
 452                                          * Skip dummy record.
 453                                          */
 454                                         continue;
 455
 456                                 fid  = ent->lde_fid;
 457                                 name = ent->lde_name;
 458                                 fid_le_to_cpu(&fid, &fid);
 459                                 ino  = ll_fid_build_ino(sbi, &fid);
 460
 461                                 done = filldir(cookie, name, namelen,
 462                                                (loff_t)hash, ino, DT_UNKNOWN);
 463                         }
 464                         next = le64_to_cpu(dp->ldp_hash_end);
 465                         ll_put_page(page);
 466                         if (!done) {
 467                                 pos = next;
 468                                 if (pos == DIR_END_OFF)
 469                                         /*
 470                                          * End of directory reached.
 471                                          */
 472                                         done = 1;
 473                                 else if (1 /* chain is exhausted*/)
 474                                         /*
 475                                          * Normal case: continue to the next
 476                                          * page.
 477                                          */
 478                                         page = ll_get_dir_page(inode, pos, 1,
 479                                                                &chain);
 480                                 else {
 481                                         /*
 482                                          * go into overflow page.
 483                                          */
 484                                 }
 485                         } else
 486                                 pos = hash;
 487                 } else {
 488                         rc = PTR_ERR(page);
 489                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 490                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 491                 }
 492         }
 493
 494         filp->f_pos = (loff_t)(__s32)pos;
 495         filp->f_version = inode->i_version;
 496         touch_atime(filp->f_vfsmnt, filp->f_dentry);
 497
 498         ll_dir_chain_fini(&chain);
 499
 500         RETURN(rc);
 501 }
 502
 503 int ll_send_mgc_param(struct obd_export *mgc, char *string)
 504 {
 505         struct mgs_send_param *msp;
 506         int rc = 0;
 507
 508         OBD_ALLOC_PTR(msp);
 509         if (!msp)
 510                 return -ENOMEM;
 511
 512         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 513         rc = obd_set_info_async(mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 514                                 sizeof(struct mgs_send_param), msp, NULL);
 515         if (rc)
 516                 CERROR("Failed to set parameter: %d\n", rc);
 517         OBD_FREE_PTR(msp);
 518
 519         return rc;
 520 }
 521
 522 char *ll_get_fsname(struct inode *inode)
 523 {
 524         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 525         char *ptr, *fsname;
 526         int len;
 527
 528         OBD_ALLOC(fsname, MGS_PARAM_MAXLEN);
 529         len = strlen(lsi->lsi_lmd->lmd_profile);
 530         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
 531         if (ptr && (strcmp(ptr, "-client") == 0))
 532                 len -= 7;
 533         strncpy(fsname, lsi->lsi_lmd->lmd_profile, len);
 534         fsname[len] = '\0';
 535
 536         return fsname;
 537 }
 538
 539 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 540                      int set_default)
 541 {
 542         struct ll_sb_info *sbi = ll_i2sbi(inode);
 543         struct md_op_data *op_data;
 544         struct ptlrpc_request *req = NULL;
 545         int rc = 0;
 546         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 547         struct obd_device *mgc = lsi->lsi_mgc;
 548         char *fsname = NULL, *param = NULL;
 549         int lum_size;
 550
 551         /*
 552          * This is coming from userspace, so should be in
 553          * local endian.  But the MDS would like it in little
 554          * endian, so we swab it before we send it.
 555          */
 556         switch (lump->lmm_magic) {
 557         case LOV_USER_MAGIC_V1: {
 558                 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 559                         lustre_swab_lov_user_md_v1(lump);
 560                 lum_size = sizeof(struct lov_user_md_v1);
 561                 break;
 562                 }
 563         case LOV_USER_MAGIC_V3: {
 564                 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 565                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lump);
 566                 lum_size = sizeof(struct lov_user_md_v3);
 567                 break;
 568                 }
 569         default: {
 570                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 571                                 " %#08x != %#08x nor %#08x\n",
 572                                 lump->lmm_magic, LOV_USER_MAGIC_V1,
 573                                 LOV_USER_MAGIC_V3);
 574                 RETURN(-EINVAL);
 575                 }
 576         }
 577
 578         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 579                                      LUSTRE_OPC_ANY, NULL);
 580         if (IS_ERR(op_data))
 581                 RETURN(PTR_ERR(op_data));
 582
 583         /* swabbing is done in lov_setstripe() on server side */
 584         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 585                         NULL, 0, &req, NULL);
 586         ll_finish_md_op_data(op_data);
 587         ptlrpc_req_finished(req);
 588         if (rc) {
 589                 if (rc != -EPERM && rc != -EACCES)
 590                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 591         }
 592
 593         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 594          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 595          need the make the distiction between the 2 versions */
 596         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 597                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 598
 599                 /* Get fsname and assume devname to be -MDT0000. */
 600                 fsname = ll_get_fsname(inode);
 601                 /* Set root stripesize */
 602                 sprintf(param, "%s-MDT0000.lov.stripesize=%u", fsname,
 603                         lump->lmm_stripe_size);
 604                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 605                 if (rc)
 606                         goto end;
 607
 608                 /* Set root stripecount */
 609                 sprintf(param, "%s-MDT0000.lov.stripecount=%hd", fsname,
 610                         lump->lmm_stripe_count);
 611                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 612                 if (rc)
 613                         goto end;
 614
 615                 /* Set root stripeoffset */
 616                 sprintf(param, "%s-MDT0000.lov.stripeoffset=%hd", fsname,
 617                         lump->lmm_stripe_offset);
 618                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 619                 if (rc)
 620                         goto end;
 621 end:
 622                 if (fsname)
 623                         OBD_FREE(fsname, MGS_PARAM_MAXLEN);
 624                 if (param)
 625                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 626         }
 627         return rc;
 628 }
 629
 630 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 631                      int *lmm_size, struct ptlrpc_request **request)
 632 {
 633         struct ll_sb_info *sbi = ll_i2sbi(inode);
 634         struct mdt_body   *body;
 635         struct lov_mds_md *lmm = NULL;
 636         struct ptlrpc_request *req = NULL;
 637         int rc, lmmsize;
 638         struct obd_capa *oc;
 639
 640         rc = ll_get_max_mdsize(sbi, &lmmsize);
 641         if (rc)
 642                 RETURN(rc);
 643
 644         oc = ll_mdscapa_get(inode);
 645         rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode),
 646                         oc, OBD_MD_FLEASIZE | OBD_MD_FLDIREA,
 647                         lmmsize, &req);
 648         capa_put(oc);
 649         if (rc < 0) {
 650                 CDEBUG(D_INFO, "md_getattr failed on inode "
 651                        "%lu/%u: rc %d\n", inode->i_ino,
 652                        inode->i_generation, rc);
 653                 GOTO(out, rc);
 654         }
 655
 656         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 657         LASSERT(body != NULL);
 658
 659         lmmsize = body->eadatasize;
 660
 661         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 662             lmmsize == 0) {
 663                 GOTO(out, rc = -ENODATA);
 664         }
 665
 666         lmm = req_capsule_server_sized_get(&req->rq_pill,
 667                                            &RMF_MDT_MD, lmmsize);
 668         LASSERT(lmm != NULL);
 669
 670         /*
 671          * This is coming from the MDS, so is probably in
 672          * little endian.  We convert it to host endian before
 673          * passing it to userspace.
 674          */
 675         /* We don't swab objects for directories */
 676         switch (le32_to_cpu(lmm->lmm_magic)) {
 677         case LOV_MAGIC_V1:
 678                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 679                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 680                 break;
 681         case LOV_MAGIC_V3:
 682                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 683                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 684                 break;
 685         default:
 686                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 687                 rc = -EPROTO;
 688         }
 689 out:
 690         *lmmp = lmm;
 691         *lmm_size = lmmsize;
 692         *request = req;
 693         return rc;
 694 }
 695
 696 static int ll_dir_ioctl(struct inode *inode, struct file *file,
 697                         unsigned int cmd, unsigned long arg)
 698 {
 699         struct ll_sb_info *sbi = ll_i2sbi(inode);
 700         struct obd_ioctl_data *data;
 701         ENTRY;
 702
 703         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
 704                inode->i_ino, inode->i_generation, inode, cmd);
 705
 706         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
 707         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
 708                 return -ENOTTY;
 709
 710         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
 711         switch(cmd) {
 712         case EXT3_IOC_GETFLAGS:
 713         case EXT3_IOC_SETFLAGS:
 714                 RETURN(ll_iocontrol(inode, file, cmd, arg));
 715         case EXT3_IOC_GETVERSION_OLD:
 716         case EXT3_IOC_GETVERSION:
 717                 RETURN(put_user(inode->i_generation, (int *)arg));
 718         /* We need to special case any other ioctls we want to handle,
 719          * to send them to the MDS/OST as appropriate and to properly
 720          * network encode the arg field.
 721         case EXT3_IOC_SETVERSION_OLD:
 722         case EXT3_IOC_SETVERSION:
 723         */
 724         case IOC_MDC_LOOKUP: {
 725                 struct ptlrpc_request *request = NULL;
 726                 int namelen, rc, len = 0;
 727                 char *buf = NULL;
 728                 char *filename;
 729                 struct obd_capa *oc;
 730
 731                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 732                 if (rc)
 733                         RETURN(rc);
 734                 data = (void *)buf;
 735
 736                 filename = data->ioc_inlbuf1;
 737                 namelen = data->ioc_inllen1;
 738
 739                 if (namelen < 1) {
 740                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
 741                         GOTO(out, rc = -EINVAL);
 742                 }
 743
 744                 oc = ll_mdscapa_get(inode);
 745                 rc = md_getattr_name(sbi->ll_md_exp, ll_inode2fid(inode), oc,
 746                                      filename, namelen, OBD_MD_FLID, 0,
 747                                      ll_i2suppgid(inode), &request);
 748                 capa_put(oc);
 749                 if (rc < 0) {
 750                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 751                         GOTO(out, rc);
 752                 }
 753
 754                 ptlrpc_req_finished(request);
 755
 756                 EXIT;
 757         out:
 758                 obd_ioctl_freedata(buf, len);
 759                 return rc;
 760         }
 761         case LL_IOC_LOV_SETSTRIPE: {
 762                 struct lov_user_md_v3 lumv3;
 763                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
 764                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
 765                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
 766
 767                 int rc = 0;
 768                 int set_default = 0;
 769
 770                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
 771                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
 772                         sizeof(lumv3p->lmm_objects[0]));
 773                 /* first try with v1 which is smaller than v3 */
 774                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
 775                         RETURN(-EFAULT);
 776
 777                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
 778                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
 779                                 RETURN(-EFAULT);
 780                 }
 781
 782                 if (inode->i_sb->s_root == file->f_dentry)
 783                         set_default = 1;
 784
 785                 /* in v1 and v3 cases lumv1 points to data */
 786                 rc = ll_dir_setstripe(inode, lumv1, set_default);
 787
 788                 RETURN(rc);
 789         }
 790         case LL_IOC_OBD_STATFS:
 791                 RETURN(ll_obd_statfs(inode, (void *)arg));
 792         case LL_IOC_LOV_GETSTRIPE:
 793         case LL_IOC_MDC_GETINFO:
 794         case IOC_MDC_GETFILEINFO:
 795         case IOC_MDC_GETFILESTRIPE: {
 796                 struct ptlrpc_request *request = NULL;
 797                 struct lov_user_md *lump;
 798                 struct lov_mds_md *lmm = NULL;
 799                 struct mdt_body *body;
 800                 char *filename = NULL;
 801                 int rc, lmmsize;
 802
 803                 if (cmd == IOC_MDC_GETFILEINFO ||
 804                     cmd == IOC_MDC_GETFILESTRIPE) {
 805                         filename = getname((const char *)arg);
 806                         if (IS_ERR(filename))
 807                                 RETURN(PTR_ERR(filename));
 808
 809                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
 810                                                       &lmmsize, &request);
 811                 } else {
 812                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
 813                 }
 814
 815                 if (request) {
 816                         body = req_capsule_server_get(&request->rq_pill,
 817                                                       &RMF_MDT_BODY);
 818                         LASSERT(body != NULL);
 819                 } else {
 820                         GOTO(out_req, rc);
 821                 }
 822
 823                 if (rc < 0) {
 824                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
 825                                                cmd == LL_IOC_MDC_GETINFO))
 826                                 GOTO(skip_lmm, rc = 0);
 827                         else
 828                                 GOTO(out_req, rc);
 829                 }
 830
 831                 if (cmd == IOC_MDC_GETFILESTRIPE ||
 832                     cmd == LL_IOC_LOV_GETSTRIPE) {
 833                         lump = (struct lov_user_md *)arg;
 834                 } else {
 835                         struct lov_user_mds_data *lmdp;
 836                         lmdp = (struct lov_user_mds_data *)arg;
 837                         lump = &lmdp->lmd_lmm;
 838                 }
 839                 if (copy_to_user(lump, lmm, lmmsize))
 840                         GOTO(out_lmm, rc = -EFAULT);
 841         skip_lmm:
 842                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
 843                         struct lov_user_mds_data *lmdp;
 844                         lstat_t st = { 0 };
 845
 846                         st.st_dev     = inode->i_sb->s_dev;
 847                         st.st_mode    = body->mode;
 848                         st.st_nlink   = body->nlink;
 849                         st.st_uid     = body->uid;
 850                         st.st_gid     = body->gid;
 851                         st.st_rdev    = body->rdev;
 852                         st.st_size    = body->size;
 853                         st.st_blksize = CFS_PAGE_SIZE;
 854                         st.st_blocks  = body->blocks;
 855                         st.st_atime   = body->atime;
 856                         st.st_mtime   = body->mtime;
 857                         st.st_ctime   = body->ctime;
 858                         st.st_ino     = inode->i_ino;
 859
 860                         lmdp = (struct lov_user_mds_data *)arg;
 861                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
 862                                 GOTO(out_lmm, rc = -EFAULT);
 863                 }
 864
 865                 EXIT;
 866         out_lmm:
 867                 if (lmm && lmm->lmm_magic == LOV_MAGIC_JOIN)
 868                         OBD_FREE(lmm, lmmsize);
 869         out_req:
 870                 ptlrpc_req_finished(request);
 871                 if (filename)
 872                         putname(filename);
 873                 return rc;
 874         }
 875         case IOC_LOV_GETINFO: {
 876                 struct lov_user_mds_data *lumd;
 877                 struct lov_stripe_md *lsm;
 878                 struct lov_user_md *lum;
 879                 struct lov_mds_md *lmm;
 880                 int lmmsize;
 881                 lstat_t st;
 882                 int rc;
 883
 884                 lumd = (struct lov_user_mds_data *)arg;
 885                 lum = &lumd->lmd_lmm;
 886
 887                 rc = ll_get_max_mdsize(sbi, &lmmsize);
 888                 if (rc)
 889                         RETURN(rc);
 890
 891                 OBD_ALLOC(lmm, lmmsize);
 892                 if (copy_from_user(lmm, lum, lmmsize))
 893                         GOTO(free_lmm, rc = -EFAULT);
 894
 895                 switch (lmm->lmm_magic) {
 896                 case LOV_USER_MAGIC_V1:
 897                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
 898                                 break;
 899                         /* swab objects first so that stripes num will be sane */
 900                         lustre_swab_lov_user_md_objects(
 901                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
 902                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
 903                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 904                         break;
 905                 case LOV_USER_MAGIC_V3:
 906                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
 907                                 break;
 908                         /* swab objects first so that stripes num will be sane */
 909                         lustre_swab_lov_user_md_objects(
 910                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
 911                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
 912                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 913                         break;
 914                 default:
 915                         GOTO(free_lmm, rc = -EINVAL);
 916                 }
 917
 918                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
 919                 if (rc < 0)
 920                         GOTO(free_lmm, rc = -ENOMEM);
 921
 922                 rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp, lsm);
 923                 if (rc)
 924                         GOTO(free_lsm, rc);
 925
 926                 /* Perform glimpse_size operation. */
 927                 memset(&st, 0, sizeof(st));
 928
 929                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
 930                 if (rc)
 931                         GOTO(free_lsm, rc);
 932
 933                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
 934                         GOTO(free_lsm, rc = -EFAULT);
 935
 936                 EXIT;
 937         free_lsm:
 938                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
 939         free_lmm:
 940                 OBD_FREE(lmm, lmmsize);
 941                 return rc;
 942         }
 943         case OBD_IOC_LLOG_CATINFO: {
 944                 struct ptlrpc_request *req = NULL;
 945                 char                  *buf = NULL;
 946                 char                  *str;
 947                 int                    len = 0;
 948                 int                    rc;
 949
 950                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 951                 if (rc)
 952                         RETURN(rc);
 953                 data = (void *)buf;
 954
 955                 if (!data->ioc_inlbuf1) {
 956                         obd_ioctl_freedata(buf, len);
 957                         RETURN(-EINVAL);
 958                 }
 959
 960                 req = ptlrpc_request_alloc(sbi2mdc(sbi)->cl_import,
 961                                            &RQF_LLOG_CATINFO);
 962                 if (req == NULL)
 963                         GOTO(out_catinfo, rc = -ENOMEM);
 964
 965                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 966                                      data->ioc_inllen1);
 967                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_CLIENT,
 968                                      data->ioc_inllen2);
 969
 970                 rc = ptlrpc_request_pack(req, LUSTRE_LOG_VERSION, LLOG_CATINFO);
 971                 if (rc) {
 972                         ptlrpc_request_free(req);
 973                         GOTO(out_catinfo, rc);
 974                 }
 975
 976                 str = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 977                 memcpy(str, data->ioc_inlbuf1, data->ioc_inllen1);
 978                 if (data->ioc_inllen2) {
 979                         str = req_capsule_client_get(&req->rq_pill,
 980                                                      &RMF_STRING);
 981                         memcpy(str, data->ioc_inlbuf2, data->ioc_inllen2);
 982                 }
 983
 984                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_SERVER,
 985                                      data->ioc_plen1);
 986                 ptlrpc_request_set_replen(req);
 987
 988                 rc = ptlrpc_queue_wait(req);
 989                 if (!rc) {
 990                         str = req_capsule_server_get(&req->rq_pill,
 991                                                      &RMF_STRING);
 992                         if (copy_to_user(data->ioc_pbuf1, str, data->ioc_plen1))
 993                                 rc = -EFAULT;
 994                 }
 995                 ptlrpc_req_finished(req);
 996         out_catinfo:
 997                 obd_ioctl_freedata(buf, len);
 998                 RETURN(rc);
 999         }
1000         case OBD_IOC_QUOTACHECK: {
1001                 struct obd_quotactl *oqctl;
1002                 int rc, error = 0;
1003
1004                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1005                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1006                         RETURN(-EPERM);
1007
1008                 OBD_ALLOC_PTR(oqctl);
1009                 if (!oqctl)
1010                         RETURN(-ENOMEM);
1011                 oqctl->qc_type = arg;
1012                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1013                 if (rc < 0) {
1014                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1015                         error = rc;
1016                 }
1017
1018                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1019                 if (rc < 0)
1020                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1021
1022                 OBD_FREE_PTR(oqctl);
1023                 return error ?: rc;
1024         }
1025         case OBD_IOC_POLL_QUOTACHECK: {
1026                 struct if_quotacheck *check;
1027                 int rc;
1028
1029                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1030                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1031                         RETURN(-EPERM);
1032
1033                 OBD_ALLOC_PTR(check);
1034                 if (!check)
1035                         RETURN(-ENOMEM);
1036
1037                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1038                                    NULL);
1039                 if (rc) {
1040                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1041                         if (copy_to_user((void *)arg, check, sizeof(*check)))
1042                                 rc = -EFAULT;
1043                         GOTO(out_poll, rc);
1044                 }
1045
1046                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1047                                    NULL);
1048                 if (rc) {
1049                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1050                         if (copy_to_user((void *)arg, check, sizeof(*check)))
1051                                 rc = -EFAULT;
1052                         GOTO(out_poll, rc);
1053                 }
1054         out_poll:
1055                 OBD_FREE_PTR(check);
1056                 RETURN(rc);
1057         }
1058         case OBD_IOC_QUOTACTL: {
1059                 struct if_quotactl *qctl;
1060                 int cmd, type, id, valid, rc = 0;
1061
1062                 OBD_ALLOC_PTR(qctl);
1063                 if (!qctl)
1064                         RETURN(-ENOMEM);
1065
1066                 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1067                         GOTO(out_quotactl, rc = -EFAULT);
1068
1069                 cmd = qctl->qc_cmd;
1070                 type = qctl->qc_type;
1071                 id = qctl->qc_id;
1072                 valid = qctl->qc_valid;
1073
1074                 switch (cmd) {
1075                 case LUSTRE_Q_INVALIDATE:
1076                 case LUSTRE_Q_FINVALIDATE:
1077                 case Q_QUOTAON:
1078                 case Q_QUOTAOFF:
1079                 case Q_SETQUOTA:
1080                 case Q_SETINFO:
1081                         if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1082                             sbi->ll_flags & LL_SBI_RMT_CLIENT)
1083                                 GOTO(out_quotactl, rc = -EPERM);
1084                         break;
1085                 case Q_GETQUOTA:
1086                         if (((type == USRQUOTA && current->euid != id) ||
1087                              (type == GRPQUOTA && !in_egroup_p(id))) &&
1088                             (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1089                              sbi->ll_flags & LL_SBI_RMT_CLIENT))
1090                                 GOTO(out_quotactl, rc = -EPERM);
1091                         break;
1092                 case Q_GETINFO:
1093                         break;
1094                 default:
1095                         CERROR("unsupported quotactl op: %#x\n", cmd);
1096                         GOTO(out_quotactl, rc = -ENOTTY);
1097                 }
1098
1099                 if (valid != QC_GENERAL) {
1100                         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1101                                 GOTO(out_quotactl, rc = -EOPNOTSUPP);
1102
1103                         if (cmd == Q_GETINFO)
1104                                 qctl->qc_cmd = Q_GETOINFO;
1105                         else if (cmd == Q_GETQUOTA)
1106                                 qctl->qc_cmd = Q_GETOQUOTA;
1107                         else
1108                                 GOTO(out_quotactl, rc = -EINVAL);
1109
1110                         switch (valid) {
1111                         case QC_MDTIDX:
1112                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1113                                                    sbi->ll_md_exp,
1114                                                    sizeof(*qctl), qctl, NULL);
1115                                 break;
1116                         case QC_OSTIDX:
1117                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1118                                                    sbi->ll_dt_exp,
1119                                                    sizeof(*qctl), qctl, NULL);
1120                                 break;
1121                         case QC_UUID:
1122                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1123                                                    sbi->ll_md_exp,
1124                                                    sizeof(*qctl), qctl, NULL);
1125                                 if (rc == -EAGAIN)
1126                                         rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1127                                                            sbi->ll_dt_exp,
1128                                                            sizeof(*qctl), qctl,
1129                                                            NULL);
1130                                 break;
1131                         default:
1132                                 rc = -EINVAL;
1133                                 break;
1134                         }
1135
1136                         if (rc)
1137                                 GOTO(out_quotactl, rc);
1138                         else
1139                                 qctl->qc_cmd = cmd;
1140                 } else {
1141                         struct obd_quotactl *oqctl;
1142
1143                         OBD_ALLOC_PTR(oqctl);
1144                         if (!oqctl)
1145                                 GOTO(out_quotactl, rc = -ENOMEM);
1146
1147                         QCTL_COPY(oqctl, qctl);
1148                         rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1149                         if (rc) {
1150                                 if (rc != -EBUSY && cmd == Q_QUOTAON) {
1151                                         oqctl->qc_cmd = Q_QUOTAOFF;
1152                                         obd_quotactl(sbi->ll_md_exp, oqctl);
1153                                 }
1154                                 OBD_FREE_PTR(oqctl);
1155                                 GOTO(out_quotactl, rc);
1156                         } else {
1157                                 QCTL_COPY(qctl, oqctl);
1158                                 OBD_FREE_PTR(oqctl);
1159                         }
1160                 }
1161
1162                 if (copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1163                         rc = -EFAULT;
1164
1165         out_quotactl:
1166                 OBD_FREE_PTR(qctl);
1167                 RETURN(rc);
1168         }
1169         case OBD_IOC_GETNAME: {
1170                 struct obd_device *obd = class_exp2obd(sbi->ll_dt_exp);
1171                 if (!obd)
1172                         RETURN(-EFAULT);
1173                 if (copy_to_user((void *)arg, obd->obd_name,
1174                                 strlen(obd->obd_name) + 1))
1175                         RETURN (-EFAULT);
1176                 RETURN(0);
1177         }
1178         case LL_IOC_FLUSHCTX:
1179                 RETURN(ll_flush_ctx(inode));
1180 #ifdef CONFIG_FS_POSIX_ACL
1181         case LL_IOC_RMTACL: {
1182             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1183                 inode == inode->i_sb->s_root->d_inode) {
1184                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1185                 int rc;
1186
1187                 LASSERT(fd != NULL);
1188                 rc = rct_add(&sbi->ll_rct, cfs_curproc_pid(), arg);
1189                 if (!rc)
1190                         fd->fd_flags |= LL_FILE_RMTACL;
1191                 RETURN(rc);
1192             } else
1193                 RETURN(0);
1194         }
1195 #endif
1196         case LL_IOC_GETOBDCOUNT: {
1197                 int count;
1198
1199                 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1200                         RETURN(-EFAULT);
1201
1202                 if (!count) {
1203                         /* get ost count */
1204                         struct lov_obd *lov = &sbi->ll_dt_exp->exp_obd->u.lov;
1205                         count = lov->desc.ld_tgt_count;
1206                 } else {
1207                         /* get mdt count */
1208                         struct lmv_obd *lmv = &sbi->ll_md_exp->exp_obd->u.lmv;
1209                         count = lmv->desc.ld_tgt_count;
1210                 }
1211
1212                 if (copy_to_user((int *)arg, &count, sizeof(int)))
1213                         RETURN(-EFAULT);
1214
1215                 RETURN(0);
1216         }
1217         default:
1218                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp,0,NULL,(void *)arg));
1219         }
1220 }
1221
1222 int ll_dir_open(struct inode *inode, struct file *file)
1223 {
1224         ENTRY;
1225         RETURN(ll_file_open(inode, file));
1226 }
1227
1228 int ll_dir_release(struct inode *inode, struct file *file)
1229 {
1230         ENTRY;
1231         RETURN(ll_file_release(inode, file));
1232 }
1233
1234 struct file_operations ll_dir_operations = {
1235         .open     = ll_dir_open,
1236         .release  = ll_dir_release,
1237         .read     = generic_read_dir,
1238         .readdir  = ll_readdir,
1239         .ioctl    = ll_dir_ioctl
1240 };