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, 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((unsigned long)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         __u16 type;
 398         ENTRY;
 399
 400         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu\n",
 401                inode->i_ino, inode->i_generation, inode,
 402                (unsigned long)pos, i_size_read(inode));
 403
 404         if (pos == DIR_END_OFF)
 405                 /*
 406                  * end-of-file.
 407                  */
 408                 RETURN(0);
 409
 410         rc    = 0;
 411         done  = 0;
 412         shift = 0;
 413         ll_dir_chain_init(&chain);
 414
 415         page = ll_get_dir_page(inode, pos, 0, &chain);
 416
 417         while (rc == 0 && !done) {
 418                 struct lu_dirpage *dp;
 419                 struct lu_dirent  *ent;
 420
 421                 if (!IS_ERR(page)) {
 422                         /*
 423                          * If page is empty (end of directory is reached),
 424                          * use this value.
 425                          */
 426                         __u64 hash = DIR_END_OFF;
 427                         __u64 next;
 428
 429                         dp = page_address(page);
 430                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 431                              ent = lu_dirent_next(ent)) {
 432                                 char          *name;
 433                                 int            namelen;
 434                                 struct lu_fid  fid;
 435                                 ino_t          ino;
 436
 437                                 /*
 438                                  * XXX: implement correct swabbing here.
 439                                  */
 440
 441                                 hash    = le64_to_cpu(ent->lde_hash);
 442                                 namelen = le16_to_cpu(ent->lde_namelen);
 443
 444                                 if (hash < pos)
 445                                         /*
 446                                          * Skip until we find target hash
 447                                          * value.
 448                                          */
 449                                         continue;
 450
 451                                 if (namelen == 0)
 452                                         /*
 453                                          * Skip dummy record.
 454                                          */
 455                                         continue;
 456
 457                                 fid  = ent->lde_fid;
 458                                 name = ent->lde_name;
 459                                 fid_le_to_cpu(&fid, &fid);
 460                                 ino  = ll_fid_build_ino(sbi, &fid);
 461                                 type = ll_dirent_type_get(ent);
 462                                 done = filldir(cookie, name, namelen,
 463                                                (loff_t)hash, ino, type);
 464                         }
 465                         next = le64_to_cpu(dp->ldp_hash_end);
 466                         ll_put_page(page);
 467                         if (!done) {
 468                                 pos = next;
 469                                 if (pos == DIR_END_OFF)
 470                                         /*
 471                                          * End of directory reached.
 472                                          */
 473                                         done = 1;
 474                                 else if (1 /* chain is exhausted*/)
 475                                         /*
 476                                          * Normal case: continue to the next
 477                                          * page.
 478                                          */
 479                                         page = ll_get_dir_page(inode, pos, 1,
 480                                                                &chain);
 481                                 else {
 482                                         /*
 483                                          * go into overflow page.
 484                                          */
 485                                 }
 486                         } else
 487                                 pos = hash;
 488                 } else {
 489                         rc = PTR_ERR(page);
 490                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 491                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 492                 }
 493         }
 494
 495         filp->f_pos = (loff_t)pos;
 496         filp->f_version = inode->i_version;
 497         touch_atime(filp->f_vfsmnt, filp->f_dentry);
 498
 499         ll_dir_chain_fini(&chain);
 500
 501         RETURN(rc);
 502 }
 503
 504 int ll_send_mgc_param(struct obd_export *mgc, char *string)
 505 {
 506         struct mgs_send_param *msp;
 507         int rc = 0;
 508
 509         OBD_ALLOC_PTR(msp);
 510         if (!msp)
 511                 return -ENOMEM;
 512
 513         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 514         rc = obd_set_info_async(mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 515                                 sizeof(struct mgs_send_param), msp, NULL);
 516         if (rc)
 517                 CERROR("Failed to set parameter: %d\n", rc);
 518         OBD_FREE_PTR(msp);
 519
 520         return rc;
 521 }
 522
 523 char *ll_get_fsname(struct inode *inode)
 524 {
 525         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 526         char *ptr, *fsname;
 527         int len;
 528
 529         OBD_ALLOC(fsname, MGS_PARAM_MAXLEN);
 530         len = strlen(lsi->lsi_lmd->lmd_profile);
 531         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
 532         if (ptr && (strcmp(ptr, "-client") == 0))
 533                 len -= 7;
 534         strncpy(fsname, lsi->lsi_lmd->lmd_profile, len);
 535         fsname[len] = '\0';
 536
 537         return fsname;
 538 }
 539
 540 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 541                      int set_default)
 542 {
 543         struct ll_sb_info *sbi = ll_i2sbi(inode);
 544         struct md_op_data *op_data;
 545         struct ptlrpc_request *req = NULL;
 546         int rc = 0;
 547         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 548         struct obd_device *mgc = lsi->lsi_mgc;
 549         char *fsname = NULL, *param = NULL;
 550         int lum_size;
 551
 552         /*
 553          * This is coming from userspace, so should be in
 554          * local endian.  But the MDS would like it in little
 555          * endian, so we swab it before we send it.
 556          */
 557         switch (lump->lmm_magic) {
 558         case LOV_USER_MAGIC_V1: {
 559                 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 560                         lustre_swab_lov_user_md_v1(lump);
 561                 lum_size = sizeof(struct lov_user_md_v1);
 562                 break;
 563                 }
 564         case LOV_USER_MAGIC_V3: {
 565                 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 566                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lump);
 567                 lum_size = sizeof(struct lov_user_md_v3);
 568                 break;
 569                 }
 570         default: {
 571                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 572                                 " %#08x != %#08x nor %#08x\n",
 573                                 lump->lmm_magic, LOV_USER_MAGIC_V1,
 574                                 LOV_USER_MAGIC_V3);
 575                 RETURN(-EINVAL);
 576                 }
 577         }
 578
 579         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 580                                      LUSTRE_OPC_ANY, NULL);
 581         if (IS_ERR(op_data))
 582                 RETURN(PTR_ERR(op_data));
 583
 584         /* swabbing is done in lov_setstripe() on server side */
 585         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 586                         NULL, 0, &req, NULL);
 587         ll_finish_md_op_data(op_data);
 588         ptlrpc_req_finished(req);
 589         if (rc) {
 590                 if (rc != -EPERM && rc != -EACCES)
 591                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 592         }
 593
 594         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 595          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 596          need the make the distiction between the 2 versions */
 597         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 598                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 599
 600                 /* Get fsname and assume devname to be -MDT0000. */
 601                 fsname = ll_get_fsname(inode);
 602                 /* Set root stripesize */
 603                 sprintf(param, "%s-MDT0000.lov.stripesize=%u", fsname,
 604                         lump->lmm_stripe_size);
 605                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 606                 if (rc)
 607                         goto end;
 608
 609                 /* Set root stripecount */
 610                 sprintf(param, "%s-MDT0000.lov.stripecount=%hd", fsname,
 611                         lump->lmm_stripe_count);
 612                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 613                 if (rc)
 614                         goto end;
 615
 616                 /* Set root stripeoffset */
 617                 sprintf(param, "%s-MDT0000.lov.stripeoffset=%hd", fsname,
 618                         lump->lmm_stripe_offset);
 619                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 620                 if (rc)
 621                         goto end;
 622 end:
 623                 if (fsname)
 624                         OBD_FREE(fsname, MGS_PARAM_MAXLEN);
 625                 if (param)
 626                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 627         }
 628         return rc;
 629 }
 630
 631 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 632                      int *lmm_size, struct ptlrpc_request **request)
 633 {
 634         struct ll_sb_info *sbi = ll_i2sbi(inode);
 635         struct mdt_body   *body;
 636         struct lov_mds_md *lmm = NULL;
 637         struct ptlrpc_request *req = NULL;
 638         int rc, lmmsize;
 639         struct obd_capa *oc;
 640
 641         rc = ll_get_max_mdsize(sbi, &lmmsize);
 642         if (rc)
 643                 RETURN(rc);
 644
 645         oc = ll_mdscapa_get(inode);
 646         rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode),
 647                         oc, OBD_MD_FLEASIZE | OBD_MD_FLDIREA,
 648                         lmmsize, &req);
 649         capa_put(oc);
 650         if (rc < 0) {
 651                 CDEBUG(D_INFO, "md_getattr failed on inode "
 652                        "%lu/%u: rc %d\n", inode->i_ino,
 653                        inode->i_generation, rc);
 654                 GOTO(out, rc);
 655         }
 656
 657         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 658         LASSERT(body != NULL);
 659
 660         lmmsize = body->eadatasize;
 661
 662         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 663             lmmsize == 0) {
 664                 GOTO(out, rc = -ENODATA);
 665         }
 666
 667         lmm = req_capsule_server_sized_get(&req->rq_pill,
 668                                            &RMF_MDT_MD, lmmsize);
 669         LASSERT(lmm != NULL);
 670
 671         /*
 672          * This is coming from the MDS, so is probably in
 673          * little endian.  We convert it to host endian before
 674          * passing it to userspace.
 675          */
 676         /* We don't swab objects for directories */
 677         switch (le32_to_cpu(lmm->lmm_magic)) {
 678         case LOV_MAGIC_V1:
 679                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 680                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 681                 break;
 682         case LOV_MAGIC_V3:
 683                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 684                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 685                 break;
 686         default:
 687                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 688                 rc = -EPROTO;
 689         }
 690 out:
 691         *lmmp = lmm;
 692         *lmm_size = lmmsize;
 693         *request = req;
 694         return rc;
 695 }
 696
 697 static int ll_dir_ioctl(struct inode *inode, struct file *file,
 698                         unsigned int cmd, unsigned long arg)
 699 {
 700         struct ll_sb_info *sbi = ll_i2sbi(inode);
 701         struct obd_ioctl_data *data;
 702         ENTRY;
 703
 704         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
 705                inode->i_ino, inode->i_generation, inode, cmd);
 706
 707         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
 708         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
 709                 return -ENOTTY;
 710
 711         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
 712         switch(cmd) {
 713         case EXT3_IOC_GETFLAGS:
 714         case EXT3_IOC_SETFLAGS:
 715                 RETURN(ll_iocontrol(inode, file, cmd, arg));
 716         case EXT3_IOC_GETVERSION_OLD:
 717         case EXT3_IOC_GETVERSION:
 718                 RETURN(put_user(inode->i_generation, (int *)arg));
 719         /* We need to special case any other ioctls we want to handle,
 720          * to send them to the MDS/OST as appropriate and to properly
 721          * network encode the arg field.
 722         case EXT3_IOC_SETVERSION_OLD:
 723         case EXT3_IOC_SETVERSION:
 724         */
 725         case IOC_MDC_LOOKUP: {
 726                 struct ptlrpc_request *request = NULL;
 727                 int namelen, rc, len = 0;
 728                 char *buf = NULL;
 729                 char *filename;
 730                 struct obd_capa *oc;
 731
 732                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 733                 if (rc)
 734                         RETURN(rc);
 735                 data = (void *)buf;
 736
 737                 filename = data->ioc_inlbuf1;
 738                 namelen = data->ioc_inllen1;
 739
 740                 if (namelen < 1) {
 741                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
 742                         GOTO(out, rc = -EINVAL);
 743                 }
 744
 745                 oc = ll_mdscapa_get(inode);
 746                 rc = md_getattr_name(sbi->ll_md_exp, ll_inode2fid(inode), oc,
 747                                      filename, namelen, OBD_MD_FLID, 0,
 748                                      ll_i2suppgid(inode), &request);
 749                 capa_put(oc);
 750                 if (rc < 0) {
 751                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 752                         GOTO(out, rc);
 753                 }
 754
 755                 ptlrpc_req_finished(request);
 756
 757                 EXIT;
 758         out:
 759                 obd_ioctl_freedata(buf, len);
 760                 return rc;
 761         }
 762         case LL_IOC_LOV_SETSTRIPE: {
 763                 struct lov_user_md_v3 lumv3;
 764                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
 765                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
 766                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
 767
 768                 int rc = 0;
 769                 int set_default = 0;
 770
 771                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
 772                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
 773                         sizeof(lumv3p->lmm_objects[0]));
 774                 /* first try with v1 which is smaller than v3 */
 775                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
 776                         RETURN(-EFAULT);
 777
 778                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
 779                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
 780                                 RETURN(-EFAULT);
 781                 }
 782
 783                 if (inode->i_sb->s_root == file->f_dentry)
 784                         set_default = 1;
 785
 786                 /* in v1 and v3 cases lumv1 points to data */
 787                 rc = ll_dir_setstripe(inode, lumv1, set_default);
 788
 789                 RETURN(rc);
 790         }
 791         case LL_IOC_OBD_STATFS:
 792                 RETURN(ll_obd_statfs(inode, (void *)arg));
 793         case LL_IOC_LOV_GETSTRIPE:
 794         case LL_IOC_MDC_GETINFO:
 795         case IOC_MDC_GETFILEINFO:
 796         case IOC_MDC_GETFILESTRIPE: {
 797                 struct ptlrpc_request *request = NULL;
 798                 struct lov_user_md *lump;
 799                 struct lov_mds_md *lmm = NULL;
 800                 struct mdt_body *body;
 801                 char *filename = NULL;
 802                 int rc, lmmsize;
 803
 804                 if (cmd == IOC_MDC_GETFILEINFO ||
 805                     cmd == IOC_MDC_GETFILESTRIPE) {
 806                         filename = getname((const char *)arg);
 807                         if (IS_ERR(filename))
 808                                 RETURN(PTR_ERR(filename));
 809
 810                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
 811                                                       &lmmsize, &request);
 812                 } else {
 813                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
 814                 }
 815
 816                 if (request) {
 817                         body = req_capsule_server_get(&request->rq_pill,
 818                                                       &RMF_MDT_BODY);
 819                         LASSERT(body != NULL);
 820                 } else {
 821                         GOTO(out_req, rc);
 822                 }
 823
 824                 if (rc < 0) {
 825                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
 826                                                cmd == LL_IOC_MDC_GETINFO))
 827                                 GOTO(skip_lmm, rc = 0);
 828                         else
 829                                 GOTO(out_req, rc);
 830                 }
 831
 832                 if (cmd == IOC_MDC_GETFILESTRIPE ||
 833                     cmd == LL_IOC_LOV_GETSTRIPE) {
 834                         lump = (struct lov_user_md *)arg;
 835                 } else {
 836                         struct lov_user_mds_data *lmdp;
 837                         lmdp = (struct lov_user_mds_data *)arg;
 838                         lump = &lmdp->lmd_lmm;
 839                 }
 840                 if (copy_to_user(lump, lmm, lmmsize))
 841                         GOTO(out_lmm, rc = -EFAULT);
 842         skip_lmm:
 843                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
 844                         struct lov_user_mds_data *lmdp;
 845                         lstat_t st = { 0 };
 846
 847                         st.st_dev     = inode->i_sb->s_dev;
 848                         st.st_mode    = body->mode;
 849                         st.st_nlink   = body->nlink;
 850                         st.st_uid     = body->uid;
 851                         st.st_gid     = body->gid;
 852                         st.st_rdev    = body->rdev;
 853                         st.st_size    = body->size;
 854                         st.st_blksize = CFS_PAGE_SIZE;
 855                         st.st_blocks  = body->blocks;
 856                         st.st_atime   = body->atime;
 857                         st.st_mtime   = body->mtime;
 858                         st.st_ctime   = body->ctime;
 859                         st.st_ino     = inode->i_ino;
 860
 861                         lmdp = (struct lov_user_mds_data *)arg;
 862                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
 863                                 GOTO(out_lmm, rc = -EFAULT);
 864                 }
 865
 866                 EXIT;
 867         out_lmm:
 868                 if (lmm && lmm->lmm_magic == LOV_MAGIC_JOIN)
 869                         OBD_FREE(lmm, lmmsize);
 870         out_req:
 871                 ptlrpc_req_finished(request);
 872                 if (filename)
 873                         putname(filename);
 874                 return rc;
 875         }
 876         case IOC_LOV_GETINFO: {
 877                 struct lov_user_mds_data *lumd;
 878                 struct lov_stripe_md *lsm;
 879                 struct lov_user_md *lum;
 880                 struct lov_mds_md *lmm;
 881                 int lmmsize;
 882                 lstat_t st;
 883                 int rc;
 884
 885                 lumd = (struct lov_user_mds_data *)arg;
 886                 lum = &lumd->lmd_lmm;
 887
 888                 rc = ll_get_max_mdsize(sbi, &lmmsize);
 889                 if (rc)
 890                         RETURN(rc);
 891
 892                 OBD_ALLOC(lmm, lmmsize);
 893                 if (copy_from_user(lmm, lum, lmmsize))
 894                         GOTO(free_lmm, rc = -EFAULT);
 895
 896                 switch (lmm->lmm_magic) {
 897                 case LOV_USER_MAGIC_V1:
 898                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
 899                                 break;
 900                         /* swab objects first so that stripes num will be sane */
 901                         lustre_swab_lov_user_md_objects(
 902                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
 903                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
 904                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 905                         break;
 906                 case LOV_USER_MAGIC_V3:
 907                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
 908                                 break;
 909                         /* swab objects first so that stripes num will be sane */
 910                         lustre_swab_lov_user_md_objects(
 911                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
 912                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
 913                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 914                         break;
 915                 default:
 916                         GOTO(free_lmm, rc = -EINVAL);
 917                 }
 918
 919                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
 920                 if (rc < 0)
 921                         GOTO(free_lmm, rc = -ENOMEM);
 922
 923                 rc = obd_checkmd(sbi->ll_dt_exp, sbi->ll_md_exp, lsm);
 924                 if (rc)
 925                         GOTO(free_lsm, rc);
 926
 927                 /* Perform glimpse_size operation. */
 928                 memset(&st, 0, sizeof(st));
 929
 930                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
 931                 if (rc)
 932                         GOTO(free_lsm, rc);
 933
 934                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
 935                         GOTO(free_lsm, rc = -EFAULT);
 936
 937                 EXIT;
 938         free_lsm:
 939                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
 940         free_lmm:
 941                 OBD_FREE(lmm, lmmsize);
 942                 return rc;
 943         }
 944         case OBD_IOC_LLOG_CATINFO: {
 945                 struct ptlrpc_request *req = NULL;
 946                 char                  *buf = NULL;
 947                 char                  *str;
 948                 int                    len = 0;
 949                 int                    rc;
 950
 951                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 952                 if (rc)
 953                         RETURN(rc);
 954                 data = (void *)buf;
 955
 956                 if (!data->ioc_inlbuf1) {
 957                         obd_ioctl_freedata(buf, len);
 958                         RETURN(-EINVAL);
 959                 }
 960
 961                 req = ptlrpc_request_alloc(sbi2mdc(sbi)->cl_import,
 962                                            &RQF_LLOG_CATINFO);
 963                 if (req == NULL)
 964                         GOTO(out_catinfo, rc = -ENOMEM);
 965
 966                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 967                                      data->ioc_inllen1);
 968                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_CLIENT,
 969                                      data->ioc_inllen2);
 970
 971                 rc = ptlrpc_request_pack(req, LUSTRE_LOG_VERSION, LLOG_CATINFO);
 972                 if (rc) {
 973                         ptlrpc_request_free(req);
 974                         GOTO(out_catinfo, rc);
 975                 }
 976
 977                 str = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 978                 memcpy(str, data->ioc_inlbuf1, data->ioc_inllen1);
 979                 if (data->ioc_inllen2) {
 980                         str = req_capsule_client_get(&req->rq_pill,
 981                                                      &RMF_STRING);
 982                         memcpy(str, data->ioc_inlbuf2, data->ioc_inllen2);
 983                 }
 984
 985                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_SERVER,
 986                                      data->ioc_plen1);
 987                 ptlrpc_request_set_replen(req);
 988
 989                 rc = ptlrpc_queue_wait(req);
 990                 if (!rc) {
 991                         str = req_capsule_server_get(&req->rq_pill,
 992                                                      &RMF_STRING);
 993                         if (copy_to_user(data->ioc_pbuf1, str, data->ioc_plen1))
 994                                 rc = -EFAULT;
 995                 }
 996                 ptlrpc_req_finished(req);
 997         out_catinfo:
 998                 obd_ioctl_freedata(buf, len);
 999                 RETURN(rc);
1000         }
1001         case OBD_IOC_QUOTACHECK: {
1002                 struct obd_quotactl *oqctl;
1003                 int rc, error = 0;
1004
1005                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1006                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1007                         RETURN(-EPERM);
1008
1009                 OBD_ALLOC_PTR(oqctl);
1010                 if (!oqctl)
1011                         RETURN(-ENOMEM);
1012                 oqctl->qc_type = arg;
1013                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1014                 if (rc < 0) {
1015                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1016                         error = rc;
1017                 }
1018
1019                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1020                 if (rc < 0)
1021                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1022
1023                 OBD_FREE_PTR(oqctl);
1024                 return error ?: rc;
1025         }
1026         case OBD_IOC_POLL_QUOTACHECK: {
1027                 struct if_quotacheck *check;
1028                 int rc;
1029
1030                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1031                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1032                         RETURN(-EPERM);
1033
1034                 OBD_ALLOC_PTR(check);
1035                 if (!check)
1036                         RETURN(-ENOMEM);
1037
1038                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1039                                    NULL);
1040                 if (rc) {
1041                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1042                         if (copy_to_user((void *)arg, check, sizeof(*check)))
1043                                 rc = -EFAULT;
1044                         GOTO(out_poll, rc);
1045                 }
1046
1047                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1048                                    NULL);
1049                 if (rc) {
1050                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1051                         if (copy_to_user((void *)arg, check, sizeof(*check)))
1052                                 rc = -EFAULT;
1053                         GOTO(out_poll, rc);
1054                 }
1055         out_poll:
1056                 OBD_FREE_PTR(check);
1057                 RETURN(rc);
1058         }
1059         case OBD_IOC_QUOTACTL: {
1060                 struct if_quotactl *qctl;
1061                 int cmd, type, id, valid, rc = 0;
1062
1063                 OBD_ALLOC_PTR(qctl);
1064                 if (!qctl)
1065                         RETURN(-ENOMEM);
1066
1067                 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1068                         GOTO(out_quotactl, rc = -EFAULT);
1069
1070                 cmd = qctl->qc_cmd;
1071                 type = qctl->qc_type;
1072                 id = qctl->qc_id;
1073                 valid = qctl->qc_valid;
1074
1075                 switch (cmd) {
1076                 case LUSTRE_Q_INVALIDATE:
1077                 case LUSTRE_Q_FINVALIDATE:
1078                 case Q_QUOTAON:
1079                 case Q_QUOTAOFF:
1080                 case Q_SETQUOTA:
1081                 case Q_SETINFO:
1082                         if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1083                             sbi->ll_flags & LL_SBI_RMT_CLIENT)
1084                                 GOTO(out_quotactl, rc = -EPERM);
1085                         break;
1086                 case Q_GETQUOTA:
1087                         if (((type == USRQUOTA && current->euid != id) ||
1088                              (type == GRPQUOTA && !in_egroup_p(id))) &&
1089                             (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1090                              sbi->ll_flags & LL_SBI_RMT_CLIENT))
1091                                 GOTO(out_quotactl, rc = -EPERM);
1092                         break;
1093                 case Q_GETINFO:
1094                         break;
1095                 default:
1096                         CERROR("unsupported quotactl op: %#x\n", cmd);
1097                         GOTO(out_quotactl, rc = -ENOTTY);
1098                 }
1099
1100                 if (valid != QC_GENERAL) {
1101                         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1102                                 GOTO(out_quotactl, rc = -EOPNOTSUPP);
1103
1104                         if (cmd == Q_GETINFO)
1105                                 qctl->qc_cmd = Q_GETOINFO;
1106                         else if (cmd == Q_GETQUOTA)
1107                                 qctl->qc_cmd = Q_GETOQUOTA;
1108                         else
1109                                 GOTO(out_quotactl, rc = -EINVAL);
1110
1111                         switch (valid) {
1112                         case QC_MDTIDX:
1113                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1114                                                    sbi->ll_md_exp,
1115                                                    sizeof(*qctl), qctl, NULL);
1116                                 break;
1117                         case QC_OSTIDX:
1118                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1119                                                    sbi->ll_dt_exp,
1120                                                    sizeof(*qctl), qctl, NULL);
1121                                 break;
1122                         case QC_UUID:
1123                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1124                                                    sbi->ll_md_exp,
1125                                                    sizeof(*qctl), qctl, NULL);
1126                                 if (rc == -EAGAIN)
1127                                         rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1128                                                            sbi->ll_dt_exp,
1129                                                            sizeof(*qctl), qctl,
1130                                                            NULL);
1131                                 break;
1132                         default:
1133                                 rc = -EINVAL;
1134                                 break;
1135                         }
1136
1137                         if (rc)
1138                                 GOTO(out_quotactl, rc);
1139                         else
1140                                 qctl->qc_cmd = cmd;
1141                 } else {
1142                         struct obd_quotactl *oqctl;
1143
1144                         OBD_ALLOC_PTR(oqctl);
1145                         if (!oqctl)
1146                                 GOTO(out_quotactl, rc = -ENOMEM);
1147
1148                         QCTL_COPY(oqctl, qctl);
1149                         rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1150                         if (rc) {
1151                                 if (rc != -EBUSY && cmd == Q_QUOTAON) {
1152                                         oqctl->qc_cmd = Q_QUOTAOFF;
1153                                         obd_quotactl(sbi->ll_md_exp, oqctl);
1154                                 }
1155                                 OBD_FREE_PTR(oqctl);
1156                                 GOTO(out_quotactl, rc);
1157                         } else {
1158                                 QCTL_COPY(qctl, oqctl);
1159                                 OBD_FREE_PTR(oqctl);
1160                         }
1161                 }
1162
1163                 if (copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1164                         rc = -EFAULT;
1165
1166         out_quotactl:
1167                 OBD_FREE_PTR(qctl);
1168                 RETURN(rc);
1169         }
1170         case OBD_IOC_GETNAME: {
1171                 struct obd_device *obd = class_exp2obd(sbi->ll_dt_exp);
1172                 if (!obd)
1173                         RETURN(-EFAULT);
1174                 if (copy_to_user((void *)arg, obd->obd_name,
1175                                 strlen(obd->obd_name) + 1))
1176                         RETURN (-EFAULT);
1177                 RETURN(0);
1178         }
1179         case LL_IOC_FLUSHCTX:
1180                 RETURN(ll_flush_ctx(inode));
1181 #ifdef CONFIG_FS_POSIX_ACL
1182         case LL_IOC_RMTACL: {
1183             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1184                 inode == inode->i_sb->s_root->d_inode) {
1185                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1186                 int rc;
1187
1188                 LASSERT(fd != NULL);
1189                 rc = rct_add(&sbi->ll_rct, cfs_curproc_pid(), arg);
1190                 if (!rc)
1191                         fd->fd_flags |= LL_FILE_RMTACL;
1192                 RETURN(rc);
1193             } else
1194                 RETURN(0);
1195         }
1196 #endif
1197         case LL_IOC_GETOBDCOUNT: {
1198                 int count;
1199
1200                 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1201                         RETURN(-EFAULT);
1202
1203                 if (!count) {
1204                         /* get ost count */
1205                         struct lov_obd *lov = &sbi->ll_dt_exp->exp_obd->u.lov;
1206                         count = lov->desc.ld_tgt_count;
1207                 } else {
1208                         /* get mdt count */
1209                         struct lmv_obd *lmv = &sbi->ll_md_exp->exp_obd->u.lmv;
1210                         count = lmv->desc.ld_tgt_count;
1211                 }
1212
1213                 if (copy_to_user((int *)arg, &count, sizeof(int)))
1214                         RETURN(-EFAULT);
1215
1216                 RETURN(0);
1217         }
1218         case LL_IOC_PATH2FID:
1219                 if (copy_to_user((void *)arg, &ll_i2info(inode)->lli_fid,
1220                                  sizeof(struct lu_fid)))
1221                         RETURN(-EFAULT);
1222                 RETURN(0);
1223         case OBD_IOC_CHANGELOG_CLEAR: {
1224                 struct ioc_changelog_clear *icc;
1225                 int rc;
1226
1227                 OBD_ALLOC_PTR(icc);
1228                 if (icc == NULL)
1229                         RETURN(-ENOMEM);
1230                 if (copy_from_user(icc, (void *)arg, sizeof(*icc)))
1231                         GOTO(icc_free, rc = -EFAULT);
1232
1233                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(*icc), icc,NULL);
1234
1235 icc_free:
1236                 OBD_FREE_PTR(icc);
1237                 RETURN(rc);
1238         }
1239         case OBD_IOC_FID2PATH:
1240                 RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg));
1241
1242         default:
1243                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp,0,NULL,(void *)arg));
1244         }
1245 }
1246
1247 int ll_dir_open(struct inode *inode, struct file *file)
1248 {
1249         ENTRY;
1250         RETURN(ll_file_open(inode, file));
1251 }
1252
1253 int ll_dir_release(struct inode *inode, struct file *file)
1254 {
1255         ENTRY;
1256         RETURN(ll_file_release(inode, file));
1257 }
1258
1259 struct file_operations ll_dir_operations = {
1260         .open     = ll_dir_open,
1261         .release  = ll_dir_release,
1262         .read     = generic_read_dir,
1263         .readdir  = ll_readdir,
1264         .ioctl    = ll_dir_ioctl
1265 };