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                         cl_isize_write(inode, body->size);
 173                 SetPageUptodate(page);
 174         }
 175         ptlrpc_req_finished(request);
 176
 177         unlock_page(page);
 178         EXIT;
 179         return rc;
 180 }
 181
 182 struct address_space_operations ll_dir_aops = {
 183         .readpage  = ll_dir_readpage,
 184 };
 185
 186 static void ll_check_page(struct inode *dir, struct page *page)
 187 {
 188         /* XXX: check page format later */
 189         SetPageChecked(page);
 190 }
 191
 192 static void ll_release_page(struct page *page, __u64 hash,
 193                             __u64 start, __u64 end)
 194 {
 195         kunmap(page);
 196         lock_page(page);
 197         if (likely(page->mapping != NULL)) {
 198                 ll_truncate_complete_page(page);
 199                 unlock_page(page);
 200         } else {
 201                 unlock_page(page);
 202                 CWARN("NULL mapping page %p, truncated by others: "
 203                       "hash("LPX64") | start("LPX64") | end("LPX64")\n",
 204                       page, hash, start, end);
 205         }
 206         page_cache_release(page);
 207 }
 208
 209 /*
 210  * Find, kmap and return page that contains given hash.
 211  */
 212 static struct page *ll_dir_page_locate(struct inode *dir, __u64 hash,
 213                                        __u64 *start, __u64 *end)
 214 {
 215         struct address_space *mapping = dir->i_mapping;
 216         /*
 217          * Complement of hash is used as an index so that
 218          * radix_tree_gang_lookup() can be used to find a page with starting
 219          * hash _smaller_ than one we are looking for.
 220          */
 221         unsigned long offset = hash_x_index((unsigned long)hash);
 222         struct page *page;
 223         int found;
 224
 225         TREE_READ_LOCK_IRQ(mapping);
 226         found = radix_tree_gang_lookup(&mapping->page_tree,
 227                                        (void **)&page, offset, 1);
 228         if (found > 0) {
 229                 struct lu_dirpage *dp;
 230
 231                 page_cache_get(page);
 232                 TREE_READ_UNLOCK_IRQ(mapping);
 233                 /*
 234                  * In contrast to find_lock_page() we are sure that directory
 235                  * page cannot be truncated (while DLM lock is held) and,
 236                  * hence, can avoid restart.
 237                  *
 238                  * In fact, page cannot be locked here at all, because
 239                  * ll_dir_readpage() does synchronous io.
 240                  */
 241                 wait_on_page(page);
 242                 if (PageUptodate(page)) {
 243                         dp = kmap(page);
 244                         *start = le64_to_cpu(dp->ldp_hash_start);
 245                         *end   = le64_to_cpu(dp->ldp_hash_end);
 246                         LASSERT(*start <= hash);
 247                         if (hash > *end || (*end != *start && hash == *end)) {
 248                                 ll_release_page(page, hash, *start, *end);
 249                                 page = NULL;
 250                         }
 251                 } else {
 252                         page_cache_release(page);
 253                         page = ERR_PTR(-EIO);
 254                 }
 255
 256         } else {
 257                 TREE_READ_UNLOCK_IRQ(mapping);
 258                 page = NULL;
 259         }
 260         return page;
 261 }
 262
 263 struct page *ll_get_dir_page(struct inode *dir, __u64 hash, int exact,
 264                              struct ll_dir_chain *chain)
 265 {
 266         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 267         struct address_space *mapping = dir->i_mapping;
 268         struct lustre_handle lockh;
 269         struct lu_dirpage *dp;
 270         struct page *page;
 271         ldlm_mode_t mode;
 272         int rc;
 273         __u64 start = 0;
 274         __u64 end = 0;
 275
 276         mode = LCK_PR;
 277         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 278                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
 279         if (!rc) {
 280                 struct ldlm_enqueue_info einfo = { LDLM_IBITS, mode,
 281                        ll_md_blocking_ast, ldlm_completion_ast,
 282                        NULL, NULL, dir };
 283                 struct lookup_intent it = { .it_op = IT_READDIR };
 284                 struct ptlrpc_request *request;
 285                 struct md_op_data *op_data;
 286
 287                 op_data = ll_prep_md_op_data(NULL, dir, NULL, NULL, 0, 0,
 288                                              LUSTRE_OPC_ANY, NULL);
 289                 if (IS_ERR(op_data))
 290                         return (void *)op_data;
 291
 292                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
 293                                 op_data, &lockh, NULL, 0, NULL, 0);
 294
 295                 ll_finish_md_op_data(op_data);
 296
 297                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 298                 if (request)
 299                         ptlrpc_req_finished(request);
 300                 if (rc < 0) {
 301                         CERROR("lock enqueue: rc: %d\n", rc);
 302                         return ERR_PTR(rc);
 303                 }
 304         } else {
 305                 /* for cross-ref object, l_ast_data of the lock may not be set,
 306                  * we reset it here */
 307                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
 308                                  dir, NULL);
 309         }
 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                 GOTO(out_unlock, page);
 315
 316         if (page != NULL) {
 317                 /*
 318                  * XXX nikita: not entirely correct handling of a corner case:
 319                  * suppose hash chain of entries with hash value HASH crosses
 320                  * border between pages P0 and P1. First both P0 and P1 are
 321                  * cached, seekdir() is called for some entry from the P0 part
 322                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 323                  * happens and finds P1, as it starts with matching hash
 324                  * value. Remaining entries from P0 part of the chain are
 325                  * skipped. (Is that really a bug?)
 326                  *
 327                  * Possible solutions: 0. don't cache P1 is such case, handle
 328                  * it as an "overflow" page. 1. invalidate all pages at
 329                  * once. 2. use HASH|1 as an index for P1.
 330                  */
 331                 if (exact && hash != start) {
 332                         /*
 333                          * readdir asked for a page starting _exactly_ from
 334                          * given hash, but cache contains stale page, with
 335                          * entries with smaller hash values. Stale page should
 336                          * be invalidated, and new one fetched.
 337                          */
 338                         CDEBUG(D_OTHER, "Stale readpage page %p: "LPX64" != "LPX64"\n",
 339                                page, hash, start);
 340                         ll_release_page(page, hash, start, end);
 341                 } else {
 342                         GOTO(hash_collision, page);
 343                 }
 344         }
 345
 346         page = read_cache_page(mapping, hash_x_index((unsigned long)hash),
 347                                (filler_t*)mapping->a_ops->readpage, NULL);
 348         if (IS_ERR(page))
 349                 GOTO(out_unlock, page);
 350
 351         wait_on_page(page);
 352         (void)kmap(page);
 353         if (!PageUptodate(page))
 354                 goto fail;
 355         if (!PageChecked(page))
 356                 ll_check_page(dir, page);
 357         if (PageError(page))
 358                 goto fail;
 359 hash_collision:
 360         dp = page_address(page);
 361
 362         start = le64_to_cpu(dp->ldp_hash_start);
 363         end   = le64_to_cpu(dp->ldp_hash_end);
 364         if (end == start) {
 365                 LASSERT(start == hash);
 366                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
 367                 /*
 368                  * Fetch whole overflow chain...
 369                  *
 370                  * XXX not yet.
 371                  */
 372                 goto fail;
 373         }
 374 out_unlock:
 375         ldlm_lock_decref(&lockh, mode);
 376         return page;
 377
 378 fail:
 379         ll_put_page(page);
 380         page = ERR_PTR(-EIO);
 381         goto out_unlock;
 382 }
 383
 384 int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 385 {
 386         struct inode         *inode = filp->f_dentry->d_inode;
 387         struct ll_inode_info *info  = ll_i2info(inode);
 388         __u64                 pos   = filp->f_pos;
 389         struct page          *page;
 390         struct ll_dir_chain   chain;
 391         int rc;
 392         int done;
 393         int shift;
 394         __u16 type;
 395         ENTRY;
 396
 397         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu\n",
 398                inode->i_ino, inode->i_generation, inode,
 399                (unsigned long)pos, i_size_read(inode));
 400
 401         if (pos == DIR_END_OFF)
 402                 /*
 403                  * end-of-file.
 404                  */
 405                 RETURN(0);
 406
 407         rc    = 0;
 408         done  = 0;
 409         shift = 0;
 410         ll_dir_chain_init(&chain);
 411
 412         page = ll_get_dir_page(inode, pos, 0, &chain);
 413
 414         while (rc == 0 && !done) {
 415                 struct lu_dirpage *dp;
 416                 struct lu_dirent  *ent;
 417
 418                 if (!IS_ERR(page)) {
 419                         /*
 420                          * If page is empty (end of directory is reached),
 421                          * use this value.
 422                          */
 423                         __u64 hash = DIR_END_OFF;
 424                         __u64 next;
 425
 426                         dp = page_address(page);
 427                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 428                              ent = lu_dirent_next(ent)) {
 429                                 char          *name;
 430                                 int            namelen;
 431                                 struct lu_fid  fid;
 432                                 ino_t          ino;
 433
 434                                 /*
 435                                  * XXX: implement correct swabbing here.
 436                                  */
 437
 438                                 hash    = le64_to_cpu(ent->lde_hash);
 439                                 namelen = le16_to_cpu(ent->lde_namelen);
 440
 441                                 if (hash < pos)
 442                                         /*
 443                                          * Skip until we find target hash
 444                                          * value.
 445                                          */
 446                                         continue;
 447
 448                                 if (namelen == 0)
 449                                         /*
 450                                          * Skip dummy record.
 451                                          */
 452                                         continue;
 453
 454                                 fid  = ent->lde_fid;
 455                                 name = ent->lde_name;
 456                                 fid_le_to_cpu(&fid, &fid);
 457                                 ino  = cl_fid_build_ino(&fid);
 458                                 type = ll_dirent_type_get(ent);
 459                                 done = filldir(cookie, name, namelen,
 460                                                (loff_t)hash, ino, type);
 461                         }
 462                         next = le64_to_cpu(dp->ldp_hash_end);
 463                         ll_put_page(page);
 464                         if (!done) {
 465                                 pos = next;
 466                                 if (pos == DIR_END_OFF)
 467                                         /*
 468                                          * End of directory reached.
 469                                          */
 470                                         done = 1;
 471                                 else if (1 /* chain is exhausted*/)
 472                                         /*
 473                                          * Normal case: continue to the next
 474                                          * page.
 475                                          */
 476                                         page = ll_get_dir_page(inode, pos, 1,
 477                                                                &chain);
 478                                 else {
 479                                         /*
 480                                          * go into overflow page.
 481                                          */
 482                                 }
 483                         } else
 484                                 pos = hash;
 485                 } else {
 486                         rc = PTR_ERR(page);
 487                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 488                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 489                 }
 490         }
 491
 492         filp->f_pos = (loff_t)pos;
 493         filp->f_version = inode->i_version;
 494         touch_atime(filp->f_vfsmnt, filp->f_dentry);
 495
 496         ll_dir_chain_fini(&chain);
 497
 498         RETURN(rc);
 499 }
 500
 501 int ll_send_mgc_param(struct obd_export *mgc, char *string)
 502 {
 503         struct mgs_send_param *msp;
 504         int rc = 0;
 505
 506         OBD_ALLOC_PTR(msp);
 507         if (!msp)
 508                 return -ENOMEM;
 509
 510         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 511         rc = obd_set_info_async(mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 512                                 sizeof(struct mgs_send_param), msp, NULL);
 513         if (rc)
 514                 CERROR("Failed to set parameter: %d\n", rc);
 515         OBD_FREE_PTR(msp);
 516
 517         return rc;
 518 }
 519
 520 char *ll_get_fsname(struct inode *inode)
 521 {
 522         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 523         char *ptr, *fsname;
 524         int len;
 525
 526         OBD_ALLOC(fsname, MGS_PARAM_MAXLEN);
 527         len = strlen(lsi->lsi_lmd->lmd_profile);
 528         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
 529         if (ptr && (strcmp(ptr, "-client") == 0))
 530                 len -= 7;
 531         strncpy(fsname, lsi->lsi_lmd->lmd_profile, len);
 532         fsname[len] = '\0';
 533
 534         return fsname;
 535 }
 536
 537 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 538                      int set_default)
 539 {
 540         struct ll_sb_info *sbi = ll_i2sbi(inode);
 541         struct md_op_data *op_data;
 542         struct ptlrpc_request *req = NULL;
 543         int rc = 0;
 544         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 545         struct obd_device *mgc = lsi->lsi_mgc;
 546         char *fsname = NULL, *param = NULL;
 547         int lum_size;
 548
 549         /*
 550          * This is coming from userspace, so should be in
 551          * local endian.  But the MDS would like it in little
 552          * endian, so we swab it before we send it.
 553          */
 554         switch (lump->lmm_magic) {
 555         case LOV_USER_MAGIC_V1: {
 556                 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 557                         lustre_swab_lov_user_md_v1(lump);
 558                 lum_size = sizeof(struct lov_user_md_v1);
 559                 break;
 560                 }
 561         case LOV_USER_MAGIC_V3: {
 562                 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 563                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lump);
 564                 lum_size = sizeof(struct lov_user_md_v3);
 565                 break;
 566                 }
 567         default: {
 568                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 569                                 " %#08x != %#08x nor %#08x\n",
 570                                 lump->lmm_magic, LOV_USER_MAGIC_V1,
 571                                 LOV_USER_MAGIC_V3);
 572                 RETURN(-EINVAL);
 573                 }
 574         }
 575
 576         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 577                                      LUSTRE_OPC_ANY, NULL);
 578         if (IS_ERR(op_data))
 579                 RETURN(PTR_ERR(op_data));
 580
 581         /* swabbing is done in lov_setstripe() on server side */
 582         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 583                         NULL, 0, &req, NULL);
 584         ll_finish_md_op_data(op_data);
 585         ptlrpc_req_finished(req);
 586         if (rc) {
 587                 if (rc != -EPERM && rc != -EACCES)
 588                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 589         }
 590
 591         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 592          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 593          need the make the distiction between the 2 versions */
 594         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 595                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 596
 597                 /* Get fsname and assume devname to be -MDT0000. */
 598                 fsname = ll_get_fsname(inode);
 599                 /* Set root stripesize */
 600                 sprintf(param, "%s-MDT0000.lov.stripesize=%u", fsname,
 601                         lump->lmm_stripe_size);
 602                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 603                 if (rc)
 604                         goto end;
 605
 606                 /* Set root stripecount */
 607                 sprintf(param, "%s-MDT0000.lov.stripecount=%hd", fsname,
 608                         lump->lmm_stripe_count);
 609                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 610                 if (rc)
 611                         goto end;
 612
 613                 /* Set root stripeoffset */
 614                 sprintf(param, "%s-MDT0000.lov.stripeoffset=%hd", fsname,
 615                         lump->lmm_stripe_offset);
 616                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 617                 if (rc)
 618                         goto end;
 619 end:
 620                 if (fsname)
 621                         OBD_FREE(fsname, MGS_PARAM_MAXLEN);
 622                 if (param)
 623                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 624         }
 625         return rc;
 626 }
 627
 628 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 629                      int *lmm_size, struct ptlrpc_request **request)
 630 {
 631         struct ll_sb_info *sbi = ll_i2sbi(inode);
 632         struct mdt_body   *body;
 633         struct lov_mds_md *lmm = NULL;
 634         struct ptlrpc_request *req = NULL;
 635         int rc, lmmsize;
 636         struct md_op_data *op_data;
 637
 638         rc = ll_get_max_mdsize(sbi, &lmmsize);
 639         if (rc)
 640                 RETURN(rc);
 641
 642         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 643                                      0, lmmsize, LUSTRE_OPC_ANY,
 644                                      NULL);
 645         if (op_data == NULL)
 646                 RETURN(-ENOMEM);
 647
 648         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 649         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 650         ll_finish_md_op_data(op_data);
 651         if (rc < 0) {
 652                 CDEBUG(D_INFO, "md_getattr failed on inode "
 653                        "%lu/%u: rc %d\n", inode->i_ino,
 654                        inode->i_generation, rc);
 655                 GOTO(out, rc);
 656         }
 657
 658         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 659         LASSERT(body != NULL);
 660
 661         lmmsize = body->eadatasize;
 662
 663         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 664             lmmsize == 0) {
 665                 GOTO(out, rc = -ENODATA);
 666         }
 667
 668         lmm = req_capsule_server_sized_get(&req->rq_pill,
 669                                            &RMF_MDT_MD, lmmsize);
 670         LASSERT(lmm != NULL);
 671
 672         /*
 673          * This is coming from the MDS, so is probably in
 674          * little endian.  We convert it to host endian before
 675          * passing it to userspace.
 676          */
 677         /* We don't swab objects for directories */
 678         switch (le32_to_cpu(lmm->lmm_magic)) {
 679         case LOV_MAGIC_V1:
 680                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 681                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 682                 break;
 683         case LOV_MAGIC_V3:
 684                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 685                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 686                 break;
 687         default:
 688                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 689                 rc = -EPROTO;
 690         }
 691 out:
 692         *lmmp = lmm;
 693         *lmm_size = lmmsize;
 694         *request = req;
 695         return rc;
 696 }
 697
 698 /*
 699  *  Get MDT index for the inode.
 700  */
 701 int ll_get_mdt_idx(struct inode *inode)
 702 {
 703         struct ll_sb_info *sbi = ll_i2sbi(inode);
 704         struct md_op_data *op_data;
 705         int rc, mdtidx;
 706         ENTRY;
 707
 708         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 709                                      0, LUSTRE_OPC_ANY, NULL);
 710         if (op_data == NULL)
 711                 RETURN(-ENOMEM);
 712
 713         op_data->op_valid |= OBD_MD_MDTIDX;
 714         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 715         mdtidx = op_data->op_mds;
 716         ll_finish_md_op_data(op_data);
 717         if (rc < 0) {
 718                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 719                 RETURN(rc);
 720         }
 721         return mdtidx;
 722 }
 723
 724 static int ll_dir_ioctl(struct inode *inode, struct file *file,
 725                         unsigned int cmd, unsigned long arg)
 726 {
 727         struct ll_sb_info *sbi = ll_i2sbi(inode);
 728         struct obd_ioctl_data *data;
 729         ENTRY;
 730
 731         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
 732                inode->i_ino, inode->i_generation, inode, cmd);
 733
 734         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
 735         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
 736                 return -ENOTTY;
 737
 738         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
 739         switch(cmd) {
 740         case FSFILT_IOC_GETFLAGS:
 741         case FSFILT_IOC_SETFLAGS:
 742                 RETURN(ll_iocontrol(inode, file, cmd, arg));
 743         case FSFILT_IOC_GETVERSION_OLD:
 744         case FSFILT_IOC_GETVERSION:
 745                 RETURN(put_user(inode->i_generation, (int *)arg));
 746         /* We need to special case any other ioctls we want to handle,
 747          * to send them to the MDS/OST as appropriate and to properly
 748          * network encode the arg field.
 749         case FSFILT_IOC_SETVERSION_OLD:
 750         case FSFILT_IOC_SETVERSION:
 751         */
 752         case LL_IOC_GET_MDTIDX: {
 753                 int mdtidx;
 754
 755                 mdtidx = ll_get_mdt_idx(inode);
 756                 if (mdtidx < 0)
 757                         RETURN(mdtidx);
 758
 759                 if (put_user((int)mdtidx, (int*)arg))
 760                         RETURN(-EFAULT);
 761
 762                 return 0;
 763         }
 764         case IOC_MDC_LOOKUP: {
 765                 struct ptlrpc_request *request = NULL;
 766                 int namelen, rc, len = 0;
 767                 char *buf = NULL;
 768                 char *filename;
 769                 struct md_op_data *op_data;
 770
 771                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 772                 if (rc)
 773                         RETURN(rc);
 774                 data = (void *)buf;
 775
 776                 filename = data->ioc_inlbuf1;
 777                 namelen = strlen(filename);
 778
 779                 if (namelen < 1) {
 780                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
 781                         GOTO(out_free, rc = -EINVAL);
 782                 }
 783
 784                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
 785                                              0, LUSTRE_OPC_ANY, NULL);
 786                 if (op_data == NULL)
 787                         GOTO(out_free, rc = -ENOMEM);
 788
 789                 op_data->op_valid = OBD_MD_FLID;
 790                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
 791                 ll_finish_md_op_data(op_data);
 792                 if (rc < 0) {
 793                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 794                         GOTO(out_free, rc);
 795                 }
 796                 ptlrpc_req_finished(request);
 797                 EXIT;
 798 out_free:
 799                 obd_ioctl_freedata(buf, len);
 800                 return rc;
 801         }
 802         case LL_IOC_LOV_SETSTRIPE: {
 803                 struct lov_user_md_v3 lumv3;
 804                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
 805                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
 806                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
 807
 808                 int rc = 0;
 809                 int set_default = 0;
 810
 811                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
 812                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
 813                         sizeof(lumv3p->lmm_objects[0]));
 814                 /* first try with v1 which is smaller than v3 */
 815                 if (cfs_copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
 816                         RETURN(-EFAULT);
 817
 818                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
 819                         if (cfs_copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
 820                                 RETURN(-EFAULT);
 821                 }
 822
 823                 if (inode->i_sb->s_root == file->f_dentry)
 824                         set_default = 1;
 825
 826                 /* in v1 and v3 cases lumv1 points to data */
 827                 rc = ll_dir_setstripe(inode, lumv1, set_default);
 828
 829                 RETURN(rc);
 830         }
 831         case LL_IOC_OBD_STATFS:
 832                 RETURN(ll_obd_statfs(inode, (void *)arg));
 833         case LL_IOC_LOV_GETSTRIPE:
 834         case LL_IOC_MDC_GETINFO:
 835         case IOC_MDC_GETFILEINFO:
 836         case IOC_MDC_GETFILESTRIPE: {
 837                 struct ptlrpc_request *request = NULL;
 838                 struct lov_user_md *lump;
 839                 struct lov_mds_md *lmm = NULL;
 840                 struct mdt_body *body;
 841                 char *filename = NULL;
 842                 int rc, lmmsize;
 843
 844                 if (cmd == IOC_MDC_GETFILEINFO ||
 845                     cmd == IOC_MDC_GETFILESTRIPE) {
 846                         filename = getname((const char *)arg);
 847                         if (IS_ERR(filename))
 848                                 RETURN(PTR_ERR(filename));
 849
 850                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
 851                                                       &lmmsize, &request);
 852                 } else {
 853                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
 854                 }
 855
 856                 if (request) {
 857                         body = req_capsule_server_get(&request->rq_pill,
 858                                                       &RMF_MDT_BODY);
 859                         LASSERT(body != NULL);
 860                 } else {
 861                         GOTO(out_req, rc);
 862                 }
 863
 864                 if (rc < 0) {
 865                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
 866                                                cmd == LL_IOC_MDC_GETINFO))
 867                                 GOTO(skip_lmm, rc = 0);
 868                         else
 869                                 GOTO(out_req, rc);
 870                 }
 871
 872                 if (cmd == IOC_MDC_GETFILESTRIPE ||
 873                     cmd == LL_IOC_LOV_GETSTRIPE) {
 874                         lump = (struct lov_user_md *)arg;
 875                 } else {
 876                         struct lov_user_mds_data *lmdp;
 877                         lmdp = (struct lov_user_mds_data *)arg;
 878                         lump = &lmdp->lmd_lmm;
 879                 }
 880                 if (cfs_copy_to_user(lump, lmm, lmmsize))
 881                         GOTO(out_req, rc = -EFAULT);
 882         skip_lmm:
 883                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
 884                         struct lov_user_mds_data *lmdp;
 885                         lstat_t st = { 0 };
 886
 887                         st.st_dev     = inode->i_sb->s_dev;
 888                         st.st_mode    = body->mode;
 889                         st.st_nlink   = body->nlink;
 890                         st.st_uid     = body->uid;
 891                         st.st_gid     = body->gid;
 892                         st.st_rdev    = body->rdev;
 893                         st.st_size    = body->size;
 894                         st.st_blksize = CFS_PAGE_SIZE;
 895                         st.st_blocks  = body->blocks;
 896                         st.st_atime   = body->atime;
 897                         st.st_mtime   = body->mtime;
 898                         st.st_ctime   = body->ctime;
 899                         st.st_ino     = inode->i_ino;
 900
 901                         lmdp = (struct lov_user_mds_data *)arg;
 902                         if (cfs_copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
 903                                 GOTO(out_req, rc = -EFAULT);
 904                 }
 905
 906                 EXIT;
 907         out_req:
 908                 ptlrpc_req_finished(request);
 909                 if (filename)
 910                         putname(filename);
 911                 return rc;
 912         }
 913         case IOC_LOV_GETINFO: {
 914                 struct lov_user_mds_data *lumd;
 915                 struct lov_stripe_md *lsm;
 916                 struct lov_user_md *lum;
 917                 struct lov_mds_md *lmm;
 918                 int lmmsize;
 919                 lstat_t st;
 920                 int rc;
 921
 922                 lumd = (struct lov_user_mds_data *)arg;
 923                 lum = &lumd->lmd_lmm;
 924
 925                 rc = ll_get_max_mdsize(sbi, &lmmsize);
 926                 if (rc)
 927                         RETURN(rc);
 928
 929                 OBD_ALLOC(lmm, lmmsize);
 930                 if (cfs_copy_from_user(lmm, lum, lmmsize))
 931                         GOTO(free_lmm, rc = -EFAULT);
 932
 933                 switch (lmm->lmm_magic) {
 934                 case LOV_USER_MAGIC_V1:
 935                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
 936                                 break;
 937                         /* swab objects first so that stripes num will be sane */
 938                         lustre_swab_lov_user_md_objects(
 939                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
 940                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
 941                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 942                         break;
 943                 case LOV_USER_MAGIC_V3:
 944                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
 945                                 break;
 946                         /* swab objects first so that stripes num will be sane */
 947                         lustre_swab_lov_user_md_objects(
 948                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
 949                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
 950                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 951                         break;
 952                 default:
 953                         GOTO(free_lmm, rc = -EINVAL);
 954                 }
 955
 956                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
 957                 if (rc < 0)
 958                         GOTO(free_lmm, rc = -ENOMEM);
 959
 960                 /* Perform glimpse_size operation. */
 961                 memset(&st, 0, sizeof(st));
 962
 963                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
 964                 if (rc)
 965                         GOTO(free_lsm, rc);
 966
 967                 if (cfs_copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
 968                         GOTO(free_lsm, rc = -EFAULT);
 969
 970                 EXIT;
 971         free_lsm:
 972                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
 973         free_lmm:
 974                 OBD_FREE(lmm, lmmsize);
 975                 return rc;
 976         }
 977         case OBD_IOC_LLOG_CATINFO: {
 978                 struct ptlrpc_request *req = NULL;
 979                 char                  *buf = NULL;
 980                 char                  *str;
 981                 int                    len = 0;
 982                 int                    rc;
 983
 984                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 985                 if (rc)
 986                         RETURN(rc);
 987                 data = (void *)buf;
 988
 989                 if (!data->ioc_inlbuf1) {
 990                         obd_ioctl_freedata(buf, len);
 991                         RETURN(-EINVAL);
 992                 }
 993
 994                 req = ptlrpc_request_alloc(sbi2mdc(sbi)->cl_import,
 995                                            &RQF_LLOG_CATINFO);
 996                 if (req == NULL)
 997                         GOTO(out_catinfo, rc = -ENOMEM);
 998
 999                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
1000                                      data->ioc_inllen1);
1001                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_CLIENT,
1002                                      data->ioc_inllen2);
1003
1004                 rc = ptlrpc_request_pack(req, LUSTRE_LOG_VERSION, LLOG_CATINFO);
1005                 if (rc) {
1006                         ptlrpc_request_free(req);
1007                         GOTO(out_catinfo, rc);
1008                 }
1009
1010                 str = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
1011                 memcpy(str, data->ioc_inlbuf1, data->ioc_inllen1);
1012                 if (data->ioc_inllen2) {
1013                         str = req_capsule_client_get(&req->rq_pill,
1014                                                      &RMF_STRING);
1015                         memcpy(str, data->ioc_inlbuf2, data->ioc_inllen2);
1016                 }
1017
1018                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_SERVER,
1019                                      data->ioc_plen1);
1020                 ptlrpc_request_set_replen(req);
1021
1022                 rc = ptlrpc_queue_wait(req);
1023                 if (!rc) {
1024                         str = req_capsule_server_get(&req->rq_pill,
1025                                                      &RMF_STRING);
1026                         if (cfs_copy_to_user(data->ioc_pbuf1, str,
1027                                              data->ioc_plen1))
1028                                 rc = -EFAULT;
1029                 }
1030                 ptlrpc_req_finished(req);
1031         out_catinfo:
1032                 obd_ioctl_freedata(buf, len);
1033                 RETURN(rc);
1034         }
1035         case OBD_IOC_QUOTACHECK: {
1036                 struct obd_quotactl *oqctl;
1037                 int rc, error = 0;
1038
1039                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1040                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1041                         RETURN(-EPERM);
1042
1043                 OBD_ALLOC_PTR(oqctl);
1044                 if (!oqctl)
1045                         RETURN(-ENOMEM);
1046                 oqctl->qc_type = arg;
1047                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1048                 if (rc < 0) {
1049                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1050                         error = rc;
1051                 }
1052
1053                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1054                 if (rc < 0)
1055                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1056
1057                 OBD_FREE_PTR(oqctl);
1058                 return error ?: rc;
1059         }
1060         case OBD_IOC_POLL_QUOTACHECK: {
1061                 struct if_quotacheck *check;
1062                 int rc;
1063
1064                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1065                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1066                         RETURN(-EPERM);
1067
1068                 OBD_ALLOC_PTR(check);
1069                 if (!check)
1070                         RETURN(-ENOMEM);
1071
1072                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1073                                    NULL);
1074                 if (rc) {
1075                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1076                         if (cfs_copy_to_user((void *)arg, check,
1077                                              sizeof(*check)))
1078                                 rc = -EFAULT;
1079                         GOTO(out_poll, rc);
1080                 }
1081
1082                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1083                                    NULL);
1084                 if (rc) {
1085                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1086                         if (cfs_copy_to_user((void *)arg, check,
1087                                              sizeof(*check)))
1088                                 rc = -EFAULT;
1089                         GOTO(out_poll, rc);
1090                 }
1091         out_poll:
1092                 OBD_FREE_PTR(check);
1093                 RETURN(rc);
1094         }
1095         case OBD_IOC_QUOTACTL: {
1096                 struct if_quotactl *qctl;
1097                 int cmd, type, id, valid, rc = 0;
1098
1099                 OBD_ALLOC_PTR(qctl);
1100                 if (!qctl)
1101                         RETURN(-ENOMEM);
1102
1103                 if (cfs_copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1104                         GOTO(out_quotactl, rc = -EFAULT);
1105
1106                 cmd = qctl->qc_cmd;
1107                 type = qctl->qc_type;
1108                 id = qctl->qc_id;
1109                 valid = qctl->qc_valid;
1110
1111                 switch (cmd) {
1112                 case LUSTRE_Q_INVALIDATE:
1113                 case LUSTRE_Q_FINVALIDATE:
1114                 case Q_QUOTAON:
1115                 case Q_QUOTAOFF:
1116                 case Q_SETQUOTA:
1117                 case Q_SETINFO:
1118                         if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1119                             sbi->ll_flags & LL_SBI_RMT_CLIENT)
1120                                 GOTO(out_quotactl, rc = -EPERM);
1121                         break;
1122                 case Q_GETQUOTA:
1123                         if (((type == USRQUOTA && cfs_curproc_euid() != id) ||
1124                              (type == GRPQUOTA && !in_egroup_p(id))) &&
1125                             (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1126                              sbi->ll_flags & LL_SBI_RMT_CLIENT))
1127                                 GOTO(out_quotactl, rc = -EPERM);
1128                         break;
1129                 case Q_GETINFO:
1130                         break;
1131                 default:
1132                         CERROR("unsupported quotactl op: %#x\n", cmd);
1133                         GOTO(out_quotactl, rc = -ENOTTY);
1134                 }
1135
1136                 if (valid != QC_GENERAL) {
1137                         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1138                                 GOTO(out_quotactl, rc = -EOPNOTSUPP);
1139
1140                         if (cmd == Q_GETINFO)
1141                                 qctl->qc_cmd = Q_GETOINFO;
1142                         else if (cmd == Q_GETQUOTA)
1143                                 qctl->qc_cmd = Q_GETOQUOTA;
1144                         else
1145                                 GOTO(out_quotactl, rc = -EINVAL);
1146
1147                         switch (valid) {
1148                         case QC_MDTIDX:
1149                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1150                                                    sbi->ll_md_exp,
1151                                                    sizeof(*qctl), qctl, NULL);
1152                                 break;
1153                         case QC_OSTIDX:
1154                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1155                                                    sbi->ll_dt_exp,
1156                                                    sizeof(*qctl), qctl, NULL);
1157                                 break;
1158                         case QC_UUID:
1159                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1160                                                    sbi->ll_md_exp,
1161                                                    sizeof(*qctl), qctl, NULL);
1162                                 if (rc == -EAGAIN)
1163                                         rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1164                                                            sbi->ll_dt_exp,
1165                                                            sizeof(*qctl), qctl,
1166                                                            NULL);
1167                                 break;
1168                         default:
1169                                 rc = -EINVAL;
1170                                 break;
1171                         }
1172
1173                         if (rc)
1174                                 GOTO(out_quotactl, rc);
1175                         else
1176                                 qctl->qc_cmd = cmd;
1177                 } else {
1178                         struct obd_quotactl *oqctl;
1179
1180                         OBD_ALLOC_PTR(oqctl);
1181                         if (!oqctl)
1182                                 GOTO(out_quotactl, rc = -ENOMEM);
1183
1184                         QCTL_COPY(oqctl, qctl);
1185                         rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1186                         if (rc) {
1187                                 if (rc != -EALREADY && cmd == Q_QUOTAON) {
1188                                         oqctl->qc_cmd = Q_QUOTAOFF;
1189                                         obd_quotactl(sbi->ll_md_exp, oqctl);
1190                                 }
1191                                 OBD_FREE_PTR(oqctl);
1192                                 GOTO(out_quotactl, rc);
1193                         } else {
1194                                 QCTL_COPY(qctl, oqctl);
1195                                 OBD_FREE_PTR(oqctl);
1196                         }
1197                 }
1198
1199                 if (cfs_copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1200                         rc = -EFAULT;
1201
1202         out_quotactl:
1203                 OBD_FREE_PTR(qctl);
1204                 RETURN(rc);
1205         }
1206         case OBD_IOC_GETNAME: {
1207                 struct obd_device *obd = class_exp2obd(sbi->ll_dt_exp);
1208                 if (!obd)
1209                         RETURN(-EFAULT);
1210                 if (cfs_copy_to_user((void *)arg, obd->obd_name,
1211                                      strlen(obd->obd_name) + 1))
1212                         RETURN (-EFAULT);
1213                 RETURN(0);
1214         }
1215         case LL_IOC_FLUSHCTX:
1216                 RETURN(ll_flush_ctx(inode));
1217 #ifdef CONFIG_FS_POSIX_ACL
1218         case LL_IOC_RMTACL: {
1219             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1220                 inode == inode->i_sb->s_root->d_inode) {
1221                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1222                 int rc;
1223
1224                 LASSERT(fd != NULL);
1225                 rc = rct_add(&sbi->ll_rct, cfs_curproc_pid(), arg);
1226                 if (!rc)
1227                         fd->fd_flags |= LL_FILE_RMTACL;
1228                 RETURN(rc);
1229             } else
1230                 RETURN(0);
1231         }
1232 #endif
1233         case LL_IOC_GETOBDCOUNT: {
1234                 int count;
1235
1236                 if (cfs_copy_from_user(&count, (int *)arg, sizeof(int)))
1237                         RETURN(-EFAULT);
1238
1239                 if (!count) {
1240                         /* get ost count */
1241                         struct lov_obd *lov = &sbi->ll_dt_exp->exp_obd->u.lov;
1242                         count = lov->desc.ld_tgt_count;
1243                 } else {
1244                         /* get mdt count */
1245                         struct lmv_obd *lmv = &sbi->ll_md_exp->exp_obd->u.lmv;
1246                         count = lmv->desc.ld_tgt_count;
1247                 }
1248
1249                 if (cfs_copy_to_user((int *)arg, &count, sizeof(int)))
1250                         RETURN(-EFAULT);
1251
1252                 RETURN(0);
1253         }
1254         case LL_IOC_PATH2FID:
1255                 if (cfs_copy_to_user((void *)arg, ll_inode2fid(inode),
1256                                      sizeof(struct lu_fid)))
1257                         RETURN(-EFAULT);
1258                 RETURN(0);
1259         case OBD_IOC_CHANGELOG_CLEAR: {
1260                 struct ioc_changelog_clear *icc;
1261                 int rc;
1262
1263                 OBD_ALLOC_PTR(icc);
1264                 if (icc == NULL)
1265                         RETURN(-ENOMEM);
1266                 if (cfs_copy_from_user(icc, (void *)arg, sizeof(*icc)))
1267                         GOTO(icc_free, rc = -EFAULT);
1268
1269                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(*icc), icc,NULL);
1270
1271 icc_free:
1272                 OBD_FREE_PTR(icc);
1273                 RETURN(rc);
1274         }
1275         case OBD_IOC_FID2PATH:
1276                 RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg));
1277
1278         default:
1279                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp,0,NULL,(void *)arg));
1280         }
1281 }
1282
1283 int ll_dir_open(struct inode *inode, struct file *file)
1284 {
1285         ENTRY;
1286         RETURN(ll_file_open(inode, file));
1287 }
1288
1289 int ll_dir_release(struct inode *inode, struct file *file)
1290 {
1291         ENTRY;
1292         RETURN(ll_file_release(inode, file));
1293 }
1294
1295 struct file_operations ll_dir_operations = {
1296         .open     = ll_dir_open,
1297         .release  = ll_dir_release,
1298         .read     = generic_read_dir,
1299         .readdir  = ll_readdir,
1300         .ioctl    = ll_dir_ioctl
1301 };