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: "DFID" at "LPU64": rc %d\n",
 305                                PFID(ll_inode2fid(dir)), hash, rc);
 306                         return ERR_PTR(rc);
 307                 }
 308         }
 309         md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie, dir, NULL);
 310         ldlm_lock_dump_handle(D_OTHER, &lockh);
 311
 312         page = ll_dir_page_locate(dir, hash, &start, &end);
 313         if (IS_ERR(page)) {
 314                 CERROR("dir page locate: "DFID" at "LPU64": rc %ld\n",
 315                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 316                 GOTO(out_unlock, page);
 317         }
 318
 319         if (page != NULL) {
 320                 /*
 321                  * XXX nikita: not entirely correct handling of a corner case:
 322                  * suppose hash chain of entries with hash value HASH crosses
 323                  * border between pages P0 and P1. First both P0 and P1 are
 324                  * cached, seekdir() is called for some entry from the P0 part
 325                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 326                  * happens and finds P1, as it starts with matching hash
 327                  * value. Remaining entries from P0 part of the chain are
 328                  * skipped. (Is that really a bug?)
 329                  *
 330                  * Possible solutions: 0. don't cache P1 is such case, handle
 331                  * it as an "overflow" page. 1. invalidate all pages at
 332                  * once. 2. use HASH|1 as an index for P1.
 333                  */
 334                 if (exact && hash != start) {
 335                         /*
 336                          * readdir asked for a page starting _exactly_ from
 337                          * given hash, but cache contains stale page, with
 338                          * entries with smaller hash values. Stale page should
 339                          * be invalidated, and new one fetched.
 340                          */
 341                         CDEBUG(D_OTHER, "Stale readpage page %p: "LPX64" != "LPX64"\n",
 342                                page, hash, start);
 343                         ll_release_page(page, hash, start, end);
 344                 } else {
 345                         GOTO(hash_collision, page);
 346                 }
 347         }
 348
 349         page = read_cache_page(mapping, hash_x_index((unsigned long)hash),
 350                                (filler_t*)mapping->a_ops->readpage, NULL);
 351         if (IS_ERR(page)) {
 352                 CERROR("read cache page: "DFID" at "LPU64": rc %ld\n",
 353                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 354                 GOTO(out_unlock, page);
 355         }
 356
 357         wait_on_page(page);
 358         (void)kmap(page);
 359         if (!PageUptodate(page)) {
 360                 CERROR("page not updated: "DFID" at "LPU64": rc %d\n",
 361                        PFID(ll_inode2fid(dir)), hash, -5);
 362                 goto fail;
 363         }
 364         if (!PageChecked(page))
 365                 ll_check_page(dir, page);
 366         if (PageError(page)) {
 367                 CERROR("page error: "DFID" at "LPU64": rc %d\n",
 368                        PFID(ll_inode2fid(dir)), hash, -5);
 369                 goto fail;
 370         }
 371 hash_collision:
 372         dp = page_address(page);
 373
 374         start = le64_to_cpu(dp->ldp_hash_start);
 375         end   = le64_to_cpu(dp->ldp_hash_end);
 376         if (end == start) {
 377                 LASSERT(start == hash);
 378                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
 379                 /*
 380                  * Fetch whole overflow chain...
 381                  *
 382                  * XXX not yet.
 383                  */
 384                 goto fail;
 385         }
 386 out_unlock:
 387         ldlm_lock_decref(&lockh, mode);
 388         return page;
 389
 390 fail:
 391         ll_put_page(page);
 392         page = ERR_PTR(-EIO);
 393         goto out_unlock;
 394 }
 395
 396 int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 397 {
 398         struct inode         *inode = filp->f_dentry->d_inode;
 399         struct ll_inode_info *info  = ll_i2info(inode);
 400         __u64                 pos   = filp->f_pos;
 401         struct page          *page;
 402         struct ll_dir_chain   chain;
 403         int rc;
 404         int done;
 405         int shift;
 406         __u16 type;
 407         ENTRY;
 408
 409         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu\n",
 410                inode->i_ino, inode->i_generation, inode,
 411                (unsigned long)pos, i_size_read(inode));
 412
 413         if (pos == DIR_END_OFF)
 414                 /*
 415                  * end-of-file.
 416                  */
 417                 RETURN(0);
 418
 419         rc    = 0;
 420         done  = 0;
 421         shift = 0;
 422         ll_dir_chain_init(&chain);
 423
 424         page = ll_get_dir_page(inode, pos, 0, &chain);
 425
 426         while (rc == 0 && !done) {
 427                 struct lu_dirpage *dp;
 428                 struct lu_dirent  *ent;
 429
 430                 if (!IS_ERR(page)) {
 431                         /*
 432                          * If page is empty (end of directory is reached),
 433                          * use this value.
 434                          */
 435                         __u64 hash = DIR_END_OFF;
 436                         __u64 next;
 437
 438                         dp = page_address(page);
 439                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 440                              ent = lu_dirent_next(ent)) {
 441                                 char          *name;
 442                                 int            namelen;
 443                                 struct lu_fid  fid;
 444                                 ino_t          ino;
 445
 446                                 /*
 447                                  * XXX: implement correct swabbing here.
 448                                  */
 449
 450                                 hash    = le64_to_cpu(ent->lde_hash);
 451                                 namelen = le16_to_cpu(ent->lde_namelen);
 452
 453                                 if (hash < pos)
 454                                         /*
 455                                          * Skip until we find target hash
 456                                          * value.
 457                                          */
 458                                         continue;
 459
 460                                 if (namelen == 0)
 461                                         /*
 462                                          * Skip dummy record.
 463                                          */
 464                                         continue;
 465
 466                                 fid  = ent->lde_fid;
 467                                 name = ent->lde_name;
 468                                 fid_le_to_cpu(&fid, &fid);
 469                                 ino  = cl_fid_build_ino(&fid);
 470                                 type = ll_dirent_type_get(ent);
 471                                 done = filldir(cookie, name, namelen,
 472                                                (loff_t)hash, ino, type);
 473                         }
 474                         next = le64_to_cpu(dp->ldp_hash_end);
 475                         ll_put_page(page);
 476                         if (!done) {
 477                                 pos = next;
 478                                 if (pos == DIR_END_OFF)
 479                                         /*
 480                                          * End of directory reached.
 481                                          */
 482                                         done = 1;
 483                                 else if (1 /* chain is exhausted*/)
 484                                         /*
 485                                          * Normal case: continue to the next
 486                                          * page.
 487                                          */
 488                                         page = ll_get_dir_page(inode, pos, 1,
 489                                                                &chain);
 490                                 else {
 491                                         /*
 492                                          * go into overflow page.
 493                                          */
 494                                 }
 495                         } else
 496                                 pos = hash;
 497                 } else {
 498                         rc = PTR_ERR(page);
 499                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 500                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 501                 }
 502         }
 503
 504         filp->f_pos = (loff_t)pos;
 505         filp->f_version = inode->i_version;
 506         touch_atime(filp->f_vfsmnt, filp->f_dentry);
 507
 508         ll_dir_chain_fini(&chain);
 509
 510         RETURN(rc);
 511 }
 512
 513 int ll_send_mgc_param(struct obd_export *mgc, char *string)
 514 {
 515         struct mgs_send_param *msp;
 516         int rc = 0;
 517
 518         OBD_ALLOC_PTR(msp);
 519         if (!msp)
 520                 return -ENOMEM;
 521
 522         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 523         rc = obd_set_info_async(mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 524                                 sizeof(struct mgs_send_param), msp, NULL);
 525         if (rc)
 526                 CERROR("Failed to set parameter: %d\n", rc);
 527         OBD_FREE_PTR(msp);
 528
 529         return rc;
 530 }
 531
 532 char *ll_get_fsname(struct inode *inode)
 533 {
 534         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 535         char *ptr, *fsname;
 536         int len;
 537
 538         OBD_ALLOC(fsname, MGS_PARAM_MAXLEN);
 539         len = strlen(lsi->lsi_lmd->lmd_profile);
 540         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
 541         if (ptr && (strcmp(ptr, "-client") == 0))
 542                 len -= 7;
 543         strncpy(fsname, lsi->lsi_lmd->lmd_profile, len);
 544         fsname[len] = '\0';
 545
 546         return fsname;
 547 }
 548
 549 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 550                      int set_default)
 551 {
 552         struct ll_sb_info *sbi = ll_i2sbi(inode);
 553         struct md_op_data *op_data;
 554         struct ptlrpc_request *req = NULL;
 555         int rc = 0;
 556         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 557         struct obd_device *mgc = lsi->lsi_mgc;
 558         char *fsname = NULL, *param = NULL;
 559         int lum_size;
 560
 561         /*
 562          * This is coming from userspace, so should be in
 563          * local endian.  But the MDS would like it in little
 564          * endian, so we swab it before we send it.
 565          */
 566         switch (lump->lmm_magic) {
 567         case LOV_USER_MAGIC_V1: {
 568                 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 569                         lustre_swab_lov_user_md_v1(lump);
 570                 lum_size = sizeof(struct lov_user_md_v1);
 571                 break;
 572                 }
 573         case LOV_USER_MAGIC_V3: {
 574                 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 575                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lump);
 576                 lum_size = sizeof(struct lov_user_md_v3);
 577                 break;
 578                 }
 579         default: {
 580                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 581                                 " %#08x != %#08x nor %#08x\n",
 582                                 lump->lmm_magic, LOV_USER_MAGIC_V1,
 583                                 LOV_USER_MAGIC_V3);
 584                 RETURN(-EINVAL);
 585                 }
 586         }
 587
 588         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 589                                      LUSTRE_OPC_ANY, NULL);
 590         if (IS_ERR(op_data))
 591                 RETURN(PTR_ERR(op_data));
 592
 593         /* swabbing is done in lov_setstripe() on server side */
 594         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 595                         NULL, 0, &req, NULL);
 596         ll_finish_md_op_data(op_data);
 597         ptlrpc_req_finished(req);
 598         if (rc) {
 599                 if (rc != -EPERM && rc != -EACCES)
 600                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 601         }
 602
 603         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 604          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 605          need the make the distiction between the 2 versions */
 606         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 607                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 608
 609                 /* Get fsname and assume devname to be -MDT0000. */
 610                 fsname = ll_get_fsname(inode);
 611                 /* Set root stripesize */
 612                 sprintf(param, "%s-MDT0000.lov.stripesize=%u", fsname,
 613                         lump->lmm_stripe_size);
 614                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 615                 if (rc)
 616                         goto end;
 617
 618                 /* Set root stripecount */
 619                 sprintf(param, "%s-MDT0000.lov.stripecount=%hd", fsname,
 620                         lump->lmm_stripe_count);
 621                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 622                 if (rc)
 623                         goto end;
 624
 625                 /* Set root stripeoffset */
 626                 sprintf(param, "%s-MDT0000.lov.stripeoffset=%hd", fsname,
 627                         lump->lmm_stripe_offset);
 628                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 629                 if (rc)
 630                         goto end;
 631 end:
 632                 if (fsname)
 633                         OBD_FREE(fsname, MGS_PARAM_MAXLEN);
 634                 if (param)
 635                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 636         }
 637         return rc;
 638 }
 639
 640 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 641                      int *lmm_size, struct ptlrpc_request **request)
 642 {
 643         struct ll_sb_info *sbi = ll_i2sbi(inode);
 644         struct mdt_body   *body;
 645         struct lov_mds_md *lmm = NULL;
 646         struct ptlrpc_request *req = NULL;
 647         int rc, lmmsize;
 648         struct obd_capa *oc;
 649
 650         rc = ll_get_max_mdsize(sbi, &lmmsize);
 651         if (rc)
 652                 RETURN(rc);
 653
 654         oc = ll_mdscapa_get(inode);
 655         rc = md_getattr(sbi->ll_md_exp, ll_inode2fid(inode),
 656                         oc, OBD_MD_FLEASIZE | OBD_MD_FLDIREA,
 657                         lmmsize, &req);
 658         capa_put(oc);
 659         if (rc < 0) {
 660                 CDEBUG(D_INFO, "md_getattr failed on inode "
 661                        "%lu/%u: rc %d\n", inode->i_ino,
 662                        inode->i_generation, rc);
 663                 GOTO(out, rc);
 664         }
 665
 666         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 667         LASSERT(body != NULL);
 668
 669         lmmsize = body->eadatasize;
 670
 671         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 672             lmmsize == 0) {
 673                 GOTO(out, rc = -ENODATA);
 674         }
 675
 676         lmm = req_capsule_server_sized_get(&req->rq_pill,
 677                                            &RMF_MDT_MD, lmmsize);
 678         LASSERT(lmm != NULL);
 679
 680         /*
 681          * This is coming from the MDS, so is probably in
 682          * little endian.  We convert it to host endian before
 683          * passing it to userspace.
 684          */
 685         /* We don't swab objects for directories */
 686         switch (le32_to_cpu(lmm->lmm_magic)) {
 687         case LOV_MAGIC_V1:
 688                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 689                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 690                 break;
 691         case LOV_MAGIC_V3:
 692                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 693                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 694                 break;
 695         default:
 696                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 697                 rc = -EPROTO;
 698         }
 699 out:
 700         *lmmp = lmm;
 701         *lmm_size = lmmsize;
 702         *request = req;
 703         return rc;
 704 }
 705
 706 static int ll_dir_ioctl(struct inode *inode, struct file *file,
 707                         unsigned int cmd, unsigned long arg)
 708 {
 709         struct ll_sb_info *sbi = ll_i2sbi(inode);
 710         struct obd_ioctl_data *data;
 711         ENTRY;
 712
 713         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
 714                inode->i_ino, inode->i_generation, inode, cmd);
 715
 716         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
 717         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
 718                 return -ENOTTY;
 719
 720         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
 721         switch(cmd) {
 722         case FSFILT_IOC_GETFLAGS:
 723         case FSFILT_IOC_SETFLAGS:
 724                 RETURN(ll_iocontrol(inode, file, cmd, arg));
 725         case FSFILT_IOC_GETVERSION_OLD:
 726         case FSFILT_IOC_GETVERSION:
 727                 RETURN(put_user(inode->i_generation, (int *)arg));
 728         /* We need to special case any other ioctls we want to handle,
 729          * to send them to the MDS/OST as appropriate and to properly
 730          * network encode the arg field.
 731         case FSFILT_IOC_SETVERSION_OLD:
 732         case FSFILT_IOC_SETVERSION:
 733         */
 734         case IOC_MDC_LOOKUP: {
 735                 struct ptlrpc_request *request = NULL;
 736                 int namelen, rc, len = 0;
 737                 char *buf = NULL;
 738                 char *filename;
 739                 struct obd_capa *oc;
 740
 741                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 742                 if (rc)
 743                         RETURN(rc);
 744                 data = (void *)buf;
 745
 746                 filename = data->ioc_inlbuf1;
 747                 namelen = data->ioc_inllen1;
 748
 749                 if (namelen < 1) {
 750                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
 751                         GOTO(out, rc = -EINVAL);
 752                 }
 753
 754                 oc = ll_mdscapa_get(inode);
 755                 rc = md_getattr_name(sbi->ll_md_exp, ll_inode2fid(inode), oc,
 756                                      filename, namelen, OBD_MD_FLID, 0,
 757                                      ll_i2suppgid(inode), &request);
 758                 capa_put(oc);
 759                 if (rc < 0) {
 760                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 761                         GOTO(out, rc);
 762                 }
 763
 764                 ptlrpc_req_finished(request);
 765
 766                 EXIT;
 767         out:
 768                 obd_ioctl_freedata(buf, len);
 769                 return rc;
 770         }
 771         case LL_IOC_LOV_SETSTRIPE: {
 772                 struct lov_user_md_v3 lumv3;
 773                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
 774                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
 775                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
 776
 777                 int rc = 0;
 778                 int set_default = 0;
 779
 780                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
 781                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
 782                         sizeof(lumv3p->lmm_objects[0]));
 783                 /* first try with v1 which is smaller than v3 */
 784                 if (cfs_copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
 785                         RETURN(-EFAULT);
 786
 787                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
 788                         if (cfs_copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
 789                                 RETURN(-EFAULT);
 790                 }
 791
 792                 if (inode->i_sb->s_root == file->f_dentry)
 793                         set_default = 1;
 794
 795                 /* in v1 and v3 cases lumv1 points to data */
 796                 rc = ll_dir_setstripe(inode, lumv1, set_default);
 797
 798                 RETURN(rc);
 799         }
 800         case LL_IOC_OBD_STATFS:
 801                 RETURN(ll_obd_statfs(inode, (void *)arg));
 802         case LL_IOC_LOV_GETSTRIPE:
 803         case LL_IOC_MDC_GETINFO:
 804         case IOC_MDC_GETFILEINFO:
 805         case IOC_MDC_GETFILESTRIPE: {
 806                 struct ptlrpc_request *request = NULL;
 807                 struct lov_user_md *lump;
 808                 struct lov_mds_md *lmm = NULL;
 809                 struct mdt_body *body;
 810                 char *filename = NULL;
 811                 int rc, lmmsize;
 812
 813                 if (cmd == IOC_MDC_GETFILEINFO ||
 814                     cmd == IOC_MDC_GETFILESTRIPE) {
 815                         filename = getname((const char *)arg);
 816                         if (IS_ERR(filename))
 817                                 RETURN(PTR_ERR(filename));
 818
 819                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
 820                                                       &lmmsize, &request);
 821                 } else {
 822                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
 823                 }
 824
 825                 if (request) {
 826                         body = req_capsule_server_get(&request->rq_pill,
 827                                                       &RMF_MDT_BODY);
 828                         LASSERT(body != NULL);
 829                 } else {
 830                         GOTO(out_req, rc);
 831                 }
 832
 833                 if (rc < 0) {
 834                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
 835                                                cmd == LL_IOC_MDC_GETINFO))
 836                                 GOTO(skip_lmm, rc = 0);
 837                         else
 838                                 GOTO(out_req, rc);
 839                 }
 840
 841                 if (cmd == IOC_MDC_GETFILESTRIPE ||
 842                     cmd == LL_IOC_LOV_GETSTRIPE) {
 843                         lump = (struct lov_user_md *)arg;
 844                 } else {
 845                         struct lov_user_mds_data *lmdp;
 846                         lmdp = (struct lov_user_mds_data *)arg;
 847                         lump = &lmdp->lmd_lmm;
 848                 }
 849                 if (cfs_copy_to_user(lump, lmm, lmmsize))
 850                         GOTO(out_req, rc = -EFAULT);
 851         skip_lmm:
 852                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
 853                         struct lov_user_mds_data *lmdp;
 854                         lstat_t st = { 0 };
 855
 856                         st.st_dev     = inode->i_sb->s_dev;
 857                         st.st_mode    = body->mode;
 858                         st.st_nlink   = body->nlink;
 859                         st.st_uid     = body->uid;
 860                         st.st_gid     = body->gid;
 861                         st.st_rdev    = body->rdev;
 862                         st.st_size    = body->size;
 863                         st.st_blksize = CFS_PAGE_SIZE;
 864                         st.st_blocks  = body->blocks;
 865                         st.st_atime   = body->atime;
 866                         st.st_mtime   = body->mtime;
 867                         st.st_ctime   = body->ctime;
 868                         st.st_ino     = inode->i_ino;
 869
 870                         lmdp = (struct lov_user_mds_data *)arg;
 871                         if (cfs_copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
 872                                 GOTO(out_req, rc = -EFAULT);
 873                 }
 874
 875                 EXIT;
 876         out_req:
 877                 ptlrpc_req_finished(request);
 878                 if (filename)
 879                         putname(filename);
 880                 return rc;
 881         }
 882         case IOC_LOV_GETINFO: {
 883                 struct lov_user_mds_data *lumd;
 884                 struct lov_stripe_md *lsm;
 885                 struct lov_user_md *lum;
 886                 struct lov_mds_md *lmm;
 887                 int lmmsize;
 888                 lstat_t st;
 889                 int rc;
 890
 891                 lumd = (struct lov_user_mds_data *)arg;
 892                 lum = &lumd->lmd_lmm;
 893
 894                 rc = ll_get_max_mdsize(sbi, &lmmsize);
 895                 if (rc)
 896                         RETURN(rc);
 897
 898                 OBD_ALLOC(lmm, lmmsize);
 899                 if (cfs_copy_from_user(lmm, lum, lmmsize))
 900                         GOTO(free_lmm, rc = -EFAULT);
 901
 902                 switch (lmm->lmm_magic) {
 903                 case LOV_USER_MAGIC_V1:
 904                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
 905                                 break;
 906                         /* swab objects first so that stripes num will be sane */
 907                         lustre_swab_lov_user_md_objects(
 908                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
 909                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
 910                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 911                         break;
 912                 case LOV_USER_MAGIC_V3:
 913                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
 914                                 break;
 915                         /* swab objects first so that stripes num will be sane */
 916                         lustre_swab_lov_user_md_objects(
 917                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
 918                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
 919                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 920                         break;
 921                 default:
 922                         GOTO(free_lmm, rc = -EINVAL);
 923                 }
 924
 925                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
 926                 if (rc < 0)
 927                         GOTO(free_lmm, rc = -ENOMEM);
 928
 929                 /* Perform glimpse_size operation. */
 930                 memset(&st, 0, sizeof(st));
 931
 932                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
 933                 if (rc)
 934                         GOTO(free_lsm, rc);
 935
 936                 if (cfs_copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
 937                         GOTO(free_lsm, rc = -EFAULT);
 938
 939                 EXIT;
 940         free_lsm:
 941                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
 942         free_lmm:
 943                 OBD_FREE(lmm, lmmsize);
 944                 return rc;
 945         }
 946         case OBD_IOC_LLOG_CATINFO: {
 947                 struct ptlrpc_request *req = NULL;
 948                 char                  *buf = NULL;
 949                 char                  *str;
 950                 int                    len = 0;
 951                 int                    rc;
 952
 953                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 954                 if (rc)
 955                         RETURN(rc);
 956                 data = (void *)buf;
 957
 958                 if (!data->ioc_inlbuf1) {
 959                         obd_ioctl_freedata(buf, len);
 960                         RETURN(-EINVAL);
 961                 }
 962
 963                 req = ptlrpc_request_alloc(sbi2mdc(sbi)->cl_import,
 964                                            &RQF_LLOG_CATINFO);
 965                 if (req == NULL)
 966                         GOTO(out_catinfo, rc = -ENOMEM);
 967
 968                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
 969                                      data->ioc_inllen1);
 970                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_CLIENT,
 971                                      data->ioc_inllen2);
 972
 973                 rc = ptlrpc_request_pack(req, LUSTRE_LOG_VERSION, LLOG_CATINFO);
 974                 if (rc) {
 975                         ptlrpc_request_free(req);
 976                         GOTO(out_catinfo, rc);
 977                 }
 978
 979                 str = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 980                 memcpy(str, data->ioc_inlbuf1, data->ioc_inllen1);
 981                 if (data->ioc_inllen2) {
 982                         str = req_capsule_client_get(&req->rq_pill,
 983                                                      &RMF_STRING);
 984                         memcpy(str, data->ioc_inlbuf2, data->ioc_inllen2);
 985                 }
 986
 987                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_SERVER,
 988                                      data->ioc_plen1);
 989                 ptlrpc_request_set_replen(req);
 990
 991                 rc = ptlrpc_queue_wait(req);
 992                 if (!rc) {
 993                         str = req_capsule_server_get(&req->rq_pill,
 994                                                      &RMF_STRING);
 995                         if (cfs_copy_to_user(data->ioc_pbuf1, str,
 996                                              data->ioc_plen1))
 997                                 rc = -EFAULT;
 998                 }
 999                 ptlrpc_req_finished(req);
1000         out_catinfo:
1001                 obd_ioctl_freedata(buf, len);
1002                 RETURN(rc);
1003         }
1004         case OBD_IOC_QUOTACHECK: {
1005                 struct obd_quotactl *oqctl;
1006                 int rc, error = 0;
1007
1008                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1009                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1010                         RETURN(-EPERM);
1011
1012                 OBD_ALLOC_PTR(oqctl);
1013                 if (!oqctl)
1014                         RETURN(-ENOMEM);
1015                 oqctl->qc_type = arg;
1016                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1017                 if (rc < 0) {
1018                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1019                         error = rc;
1020                 }
1021
1022                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1023                 if (rc < 0)
1024                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1025
1026                 OBD_FREE_PTR(oqctl);
1027                 return error ?: rc;
1028         }
1029         case OBD_IOC_POLL_QUOTACHECK: {
1030                 struct if_quotacheck *check;
1031                 int rc;
1032
1033                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1034                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1035                         RETURN(-EPERM);
1036
1037                 OBD_ALLOC_PTR(check);
1038                 if (!check)
1039                         RETURN(-ENOMEM);
1040
1041                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1042                                    NULL);
1043                 if (rc) {
1044                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1045                         if (cfs_copy_to_user((void *)arg, check,
1046                                              sizeof(*check)))
1047                                 rc = -EFAULT;
1048                         GOTO(out_poll, rc);
1049                 }
1050
1051                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1052                                    NULL);
1053                 if (rc) {
1054                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1055                         if (cfs_copy_to_user((void *)arg, check,
1056                                              sizeof(*check)))
1057                                 rc = -EFAULT;
1058                         GOTO(out_poll, rc);
1059                 }
1060         out_poll:
1061                 OBD_FREE_PTR(check);
1062                 RETURN(rc);
1063         }
1064         case OBD_IOC_QUOTACTL: {
1065                 struct if_quotactl *qctl;
1066                 int cmd, type, id, valid, rc = 0;
1067
1068                 OBD_ALLOC_PTR(qctl);
1069                 if (!qctl)
1070                         RETURN(-ENOMEM);
1071
1072                 if (cfs_copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1073                         GOTO(out_quotactl, rc = -EFAULT);
1074
1075                 cmd = qctl->qc_cmd;
1076                 type = qctl->qc_type;
1077                 id = qctl->qc_id;
1078                 valid = qctl->qc_valid;
1079
1080                 switch (cmd) {
1081                 case LUSTRE_Q_INVALIDATE:
1082                 case LUSTRE_Q_FINVALIDATE:
1083                 case Q_QUOTAON:
1084                 case Q_QUOTAOFF:
1085                 case Q_SETQUOTA:
1086                 case Q_SETINFO:
1087                         if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1088                             sbi->ll_flags & LL_SBI_RMT_CLIENT)
1089                                 GOTO(out_quotactl, rc = -EPERM);
1090                         break;
1091                 case Q_GETQUOTA:
1092                         if (((type == USRQUOTA && current->euid != id) ||
1093                              (type == GRPQUOTA && !in_egroup_p(id))) &&
1094                             (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1095                              sbi->ll_flags & LL_SBI_RMT_CLIENT))
1096                                 GOTO(out_quotactl, rc = -EPERM);
1097                         break;
1098                 case Q_GETINFO:
1099                         break;
1100                 default:
1101                         CERROR("unsupported quotactl op: %#x\n", cmd);
1102                         GOTO(out_quotactl, rc = -ENOTTY);
1103                 }
1104
1105                 if (valid != QC_GENERAL) {
1106                         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1107                                 GOTO(out_quotactl, rc = -EOPNOTSUPP);
1108
1109                         if (cmd == Q_GETINFO)
1110                                 qctl->qc_cmd = Q_GETOINFO;
1111                         else if (cmd == Q_GETQUOTA)
1112                                 qctl->qc_cmd = Q_GETOQUOTA;
1113                         else
1114                                 GOTO(out_quotactl, rc = -EINVAL);
1115
1116                         switch (valid) {
1117                         case QC_MDTIDX:
1118                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1119                                                    sbi->ll_md_exp,
1120                                                    sizeof(*qctl), qctl, NULL);
1121                                 break;
1122                         case QC_OSTIDX:
1123                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1124                                                    sbi->ll_dt_exp,
1125                                                    sizeof(*qctl), qctl, NULL);
1126                                 break;
1127                         case QC_UUID:
1128                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1129                                                    sbi->ll_md_exp,
1130                                                    sizeof(*qctl), qctl, NULL);
1131                                 if (rc == -EAGAIN)
1132                                         rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1133                                                            sbi->ll_dt_exp,
1134                                                            sizeof(*qctl), qctl,
1135                                                            NULL);
1136                                 break;
1137                         default:
1138                                 rc = -EINVAL;
1139                                 break;
1140                         }
1141
1142                         if (rc)
1143                                 GOTO(out_quotactl, rc);
1144                         else
1145                                 qctl->qc_cmd = cmd;
1146                 } else {
1147                         struct obd_quotactl *oqctl;
1148
1149                         OBD_ALLOC_PTR(oqctl);
1150                         if (!oqctl)
1151                                 GOTO(out_quotactl, rc = -ENOMEM);
1152
1153                         QCTL_COPY(oqctl, qctl);
1154                         rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1155                         if (rc) {
1156                                 if (rc != -EALREADY && cmd == Q_QUOTAON) {
1157                                         oqctl->qc_cmd = Q_QUOTAOFF;
1158                                         obd_quotactl(sbi->ll_md_exp, oqctl);
1159                                 }
1160                                 OBD_FREE_PTR(oqctl);
1161                                 GOTO(out_quotactl, rc);
1162                         } else {
1163                                 QCTL_COPY(qctl, oqctl);
1164                                 OBD_FREE_PTR(oqctl);
1165                         }
1166                 }
1167
1168                 if (cfs_copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1169                         rc = -EFAULT;
1170
1171         out_quotactl:
1172                 OBD_FREE_PTR(qctl);
1173                 RETURN(rc);
1174         }
1175         case OBD_IOC_GETNAME: {
1176                 struct obd_device *obd = class_exp2obd(sbi->ll_dt_exp);
1177                 if (!obd)
1178                         RETURN(-EFAULT);
1179                 if (cfs_copy_to_user((void *)arg, obd->obd_name,
1180                                      strlen(obd->obd_name) + 1))
1181                         RETURN (-EFAULT);
1182                 RETURN(0);
1183         }
1184         case LL_IOC_FLUSHCTX:
1185                 RETURN(ll_flush_ctx(inode));
1186 #ifdef CONFIG_FS_POSIX_ACL
1187         case LL_IOC_RMTACL: {
1188             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1189                 inode == inode->i_sb->s_root->d_inode) {
1190                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1191                 int rc;
1192
1193                 LASSERT(fd != NULL);
1194                 rc = rct_add(&sbi->ll_rct, cfs_curproc_pid(), arg);
1195                 if (!rc)
1196                         fd->fd_flags |= LL_FILE_RMTACL;
1197                 RETURN(rc);
1198             } else
1199                 RETURN(0);
1200         }
1201 #endif
1202         case LL_IOC_GETOBDCOUNT: {
1203                 int count;
1204
1205                 if (cfs_copy_from_user(&count, (int *)arg, sizeof(int)))
1206                         RETURN(-EFAULT);
1207
1208                 if (!count) {
1209                         /* get ost count */
1210                         struct lov_obd *lov = &sbi->ll_dt_exp->exp_obd->u.lov;
1211                         count = lov->desc.ld_tgt_count;
1212                 } else {
1213                         /* get mdt count */
1214                         struct lmv_obd *lmv = &sbi->ll_md_exp->exp_obd->u.lmv;
1215                         count = lmv->desc.ld_tgt_count;
1216                 }
1217
1218                 if (cfs_copy_to_user((int *)arg, &count, sizeof(int)))
1219                         RETURN(-EFAULT);
1220
1221                 RETURN(0);
1222         }
1223         case LL_IOC_PATH2FID:
1224                 if (cfs_copy_to_user((void *)arg, ll_inode2fid(inode),
1225                                      sizeof(struct lu_fid)))
1226                         RETURN(-EFAULT);
1227                 RETURN(0);
1228         case OBD_IOC_CHANGELOG_CLEAR: {
1229                 struct ioc_changelog_clear *icc;
1230                 int rc;
1231
1232                 OBD_ALLOC_PTR(icc);
1233                 if (icc == NULL)
1234                         RETURN(-ENOMEM);
1235                 if (cfs_copy_from_user(icc, (void *)arg, sizeof(*icc)))
1236                         GOTO(icc_free, rc = -EFAULT);
1237
1238                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(*icc), icc,NULL);
1239
1240 icc_free:
1241                 OBD_FREE_PTR(icc);
1242                 RETURN(rc);
1243         }
1244         case OBD_IOC_FID2PATH:
1245                 RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg));
1246
1247         default:
1248                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp,0,NULL,(void *)arg));
1249         }
1250 }
1251
1252 int ll_dir_open(struct inode *inode, struct file *file)
1253 {
1254         ENTRY;
1255         RETURN(ll_file_open(inode, file));
1256 }
1257
1258 int ll_dir_release(struct inode *inode, struct file *file)
1259 {
1260         ENTRY;
1261         RETURN(ll_file_release(inode, file));
1262 }
1263
1264 struct file_operations ll_dir_operations = {
1265         .open     = ll_dir_open,
1266         .release  = ll_dir_release,
1267         .read     = generic_read_dir,
1268         .readdir  = ll_readdir,
1269         .ioctl    = ll_dir_ioctl
1270 };