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: "DFID" at "LPU64": rc %d\n",
 302                                PFID(ll_inode2fid(dir)), hash, rc);
 303                         return ERR_PTR(rc);
 304                 }
 305         } else {
 306                 /* for cross-ref object, l_ast_data of the lock may not be set,
 307                  * we reset it here */
 308                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
 309                                  dir, NULL);
 310         }
 311         ldlm_lock_dump_handle(D_OTHER, &lockh);
 312
 313         page = ll_dir_page_locate(dir, hash, &start, &end);
 314         if (IS_ERR(page)) {
 315                 CERROR("dir page locate: "DFID" at "LPU64": rc %ld\n",
 316                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 317                 GOTO(out_unlock, page);
 318         }
 319
 320         if (page != NULL) {
 321                 /*
 322                  * XXX nikita: not entirely correct handling of a corner case:
 323                  * suppose hash chain of entries with hash value HASH crosses
 324                  * border between pages P0 and P1. First both P0 and P1 are
 325                  * cached, seekdir() is called for some entry from the P0 part
 326                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 327                  * happens and finds P1, as it starts with matching hash
 328                  * value. Remaining entries from P0 part of the chain are
 329                  * skipped. (Is that really a bug?)
 330                  *
 331                  * Possible solutions: 0. don't cache P1 is such case, handle
 332                  * it as an "overflow" page. 1. invalidate all pages at
 333                  * once. 2. use HASH|1 as an index for P1.
 334                  */
 335                 if (exact && hash != start) {
 336                         /*
 337                          * readdir asked for a page starting _exactly_ from
 338                          * given hash, but cache contains stale page, with
 339                          * entries with smaller hash values. Stale page should
 340                          * be invalidated, and new one fetched.
 341                          */
 342                         CDEBUG(D_OTHER, "Stale readpage page %p: "LPX64" != "LPX64"\n",
 343                                page, hash, start);
 344                         ll_release_page(page, hash, start, end);
 345                 } else {
 346                         GOTO(hash_collision, page);
 347                 }
 348         }
 349
 350         page = read_cache_page(mapping, hash_x_index((unsigned long)hash),
 351                                (filler_t*)mapping->a_ops->readpage, NULL);
 352         if (IS_ERR(page)) {
 353                 CERROR("read cache page: "DFID" at "LPU64": rc %ld\n",
 354                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 355                 GOTO(out_unlock, page);
 356         }
 357
 358         wait_on_page(page);
 359         (void)kmap(page);
 360         if (!PageUptodate(page)) {
 361                 CERROR("page not updated: "DFID" at "LPU64": rc %d\n",
 362                        PFID(ll_inode2fid(dir)), hash, -5);
 363                 goto fail;
 364         }
 365         if (!PageChecked(page))
 366                 ll_check_page(dir, page);
 367         if (PageError(page)) {
 368                 CERROR("page error: "DFID" at "LPU64": rc %d\n",
 369                        PFID(ll_inode2fid(dir)), hash, -5);
 370                 goto fail;
 371         }
 372 hash_collision:
 373         dp = page_address(page);
 374
 375         start = le64_to_cpu(dp->ldp_hash_start);
 376         end   = le64_to_cpu(dp->ldp_hash_end);
 377         if (end == start) {
 378                 LASSERT(start == hash);
 379                 CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
 380                 /*
 381                  * Fetch whole overflow chain...
 382                  *
 383                  * XXX not yet.
 384                  */
 385                 goto fail;
 386         }
 387 out_unlock:
 388         ldlm_lock_decref(&lockh, mode);
 389         return page;
 390
 391 fail:
 392         ll_put_page(page);
 393         page = ERR_PTR(-EIO);
 394         goto out_unlock;
 395 }
 396
 397 int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 398 {
 399         struct inode         *inode = filp->f_dentry->d_inode;
 400         struct ll_inode_info *info  = ll_i2info(inode);
 401         __u64                 pos   = filp->f_pos;
 402         struct page          *page;
 403         struct ll_dir_chain   chain;
 404         int rc;
 405         int done;
 406         int shift;
 407         __u16 type;
 408         ENTRY;
 409
 410         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu\n",
 411                inode->i_ino, inode->i_generation, inode,
 412                (unsigned long)pos, i_size_read(inode));
 413
 414         if (pos == DIR_END_OFF)
 415                 /*
 416                  * end-of-file.
 417                  */
 418                 RETURN(0);
 419
 420         rc    = 0;
 421         done  = 0;
 422         shift = 0;
 423         ll_dir_chain_init(&chain);
 424
 425         page = ll_get_dir_page(inode, pos, 0, &chain);
 426
 427         while (rc == 0 && !done) {
 428                 struct lu_dirpage *dp;
 429                 struct lu_dirent  *ent;
 430
 431                 if (!IS_ERR(page)) {
 432                         /*
 433                          * If page is empty (end of directory is reached),
 434                          * use this value.
 435                          */
 436                         __u64 hash = DIR_END_OFF;
 437                         __u64 next;
 438
 439                         dp = page_address(page);
 440                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 441                              ent = lu_dirent_next(ent)) {
 442                                 char          *name;
 443                                 int            namelen;
 444                                 struct lu_fid  fid;
 445                                 ino_t          ino;
 446
 447                                 /*
 448                                  * XXX: implement correct swabbing here.
 449                                  */
 450
 451                                 hash    = le64_to_cpu(ent->lde_hash);
 452                                 namelen = le16_to_cpu(ent->lde_namelen);
 453
 454                                 if (hash < pos)
 455                                         /*
 456                                          * Skip until we find target hash
 457                                          * value.
 458                                          */
 459                                         continue;
 460
 461                                 if (namelen == 0)
 462                                         /*
 463                                          * Skip dummy record.
 464                                          */
 465                                         continue;
 466
 467                                 fid  = ent->lde_fid;
 468                                 name = ent->lde_name;
 469                                 fid_le_to_cpu(&fid, &fid);
 470                                 ino  = cl_fid_build_ino(&fid);
 471                                 type = ll_dirent_type_get(ent);
 472                                 done = filldir(cookie, name, namelen,
 473                                                (loff_t)hash, ino, type);
 474                         }
 475                         next = le64_to_cpu(dp->ldp_hash_end);
 476                         ll_put_page(page);
 477                         if (!done) {
 478                                 pos = next;
 479                                 if (pos == DIR_END_OFF)
 480                                         /*
 481                                          * End of directory reached.
 482                                          */
 483                                         done = 1;
 484                                 else if (1 /* chain is exhausted*/)
 485                                         /*
 486                                          * Normal case: continue to the next
 487                                          * page.
 488                                          */
 489                                         page = ll_get_dir_page(inode, pos, 1,
 490                                                                &chain);
 491                                 else {
 492                                         /*
 493                                          * go into overflow page.
 494                                          */
 495                                 }
 496                         } else
 497                                 pos = hash;
 498                 } else {
 499                         rc = PTR_ERR(page);
 500                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 501                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 502                 }
 503         }
 504
 505         filp->f_pos = (loff_t)pos;
 506         filp->f_version = inode->i_version;
 507         touch_atime(filp->f_vfsmnt, filp->f_dentry);
 508
 509         ll_dir_chain_fini(&chain);
 510
 511         RETURN(rc);
 512 }
 513
 514 int ll_send_mgc_param(struct obd_export *mgc, char *string)
 515 {
 516         struct mgs_send_param *msp;
 517         int rc = 0;
 518
 519         OBD_ALLOC_PTR(msp);
 520         if (!msp)
 521                 return -ENOMEM;
 522
 523         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 524         rc = obd_set_info_async(mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 525                                 sizeof(struct mgs_send_param), msp, NULL);
 526         if (rc)
 527                 CERROR("Failed to set parameter: %d\n", rc);
 528         OBD_FREE_PTR(msp);
 529
 530         return rc;
 531 }
 532
 533 char *ll_get_fsname(struct inode *inode)
 534 {
 535         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 536         char *ptr, *fsname;
 537         int len;
 538
 539         OBD_ALLOC(fsname, MGS_PARAM_MAXLEN);
 540         len = strlen(lsi->lsi_lmd->lmd_profile);
 541         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
 542         if (ptr && (strcmp(ptr, "-client") == 0))
 543                 len -= 7;
 544         strncpy(fsname, lsi->lsi_lmd->lmd_profile, len);
 545         fsname[len] = '\0';
 546
 547         return fsname;
 548 }
 549
 550 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 551                      int set_default)
 552 {
 553         struct ll_sb_info *sbi = ll_i2sbi(inode);
 554         struct md_op_data *op_data;
 555         struct ptlrpc_request *req = NULL;
 556         int rc = 0;
 557         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 558         struct obd_device *mgc = lsi->lsi_mgc;
 559         char *fsname = NULL, *param = NULL;
 560         int lum_size;
 561
 562         /*
 563          * This is coming from userspace, so should be in
 564          * local endian.  But the MDS would like it in little
 565          * endian, so we swab it before we send it.
 566          */
 567         switch (lump->lmm_magic) {
 568         case LOV_USER_MAGIC_V1: {
 569                 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 570                         lustre_swab_lov_user_md_v1(lump);
 571                 lum_size = sizeof(struct lov_user_md_v1);
 572                 break;
 573                 }
 574         case LOV_USER_MAGIC_V3: {
 575                 if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 576                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lump);
 577                 lum_size = sizeof(struct lov_user_md_v3);
 578                 break;
 579                 }
 580         default: {
 581                 CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 582                                 " %#08x != %#08x nor %#08x\n",
 583                                 lump->lmm_magic, LOV_USER_MAGIC_V1,
 584                                 LOV_USER_MAGIC_V3);
 585                 RETURN(-EINVAL);
 586                 }
 587         }
 588
 589         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 590                                      LUSTRE_OPC_ANY, NULL);
 591         if (IS_ERR(op_data))
 592                 RETURN(PTR_ERR(op_data));
 593
 594         /* swabbing is done in lov_setstripe() on server side */
 595         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 596                         NULL, 0, &req, NULL);
 597         ll_finish_md_op_data(op_data);
 598         ptlrpc_req_finished(req);
 599         if (rc) {
 600                 if (rc != -EPERM && rc != -EACCES)
 601                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 602         }
 603
 604         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 605          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 606          need the make the distiction between the 2 versions */
 607         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 608                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 609
 610                 /* Get fsname and assume devname to be -MDT0000. */
 611                 fsname = ll_get_fsname(inode);
 612                 /* Set root stripesize */
 613                 sprintf(param, "%s-MDT0000.lov.stripesize=%u", fsname,
 614                         lump->lmm_stripe_size);
 615                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 616                 if (rc)
 617                         goto end;
 618
 619                 /* Set root stripecount */
 620                 sprintf(param, "%s-MDT0000.lov.stripecount=%hd", fsname,
 621                         lump->lmm_stripe_count);
 622                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 623                 if (rc)
 624                         goto end;
 625
 626                 /* Set root stripeoffset */
 627                 sprintf(param, "%s-MDT0000.lov.stripeoffset=%hd", fsname,
 628                         lump->lmm_stripe_offset);
 629                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 630                 if (rc)
 631                         goto end;
 632 end:
 633                 if (fsname)
 634                         OBD_FREE(fsname, MGS_PARAM_MAXLEN);
 635                 if (param)
 636                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 637         }
 638         return rc;
 639 }
 640
 641 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 642                      int *lmm_size, struct ptlrpc_request **request)
 643 {
 644         struct ll_sb_info *sbi = ll_i2sbi(inode);
 645         struct mdt_body   *body;
 646         struct lov_mds_md *lmm = NULL;
 647         struct ptlrpc_request *req = NULL;
 648         int rc, lmmsize;
 649         struct md_op_data *op_data;
 650
 651         rc = ll_get_max_mdsize(sbi, &lmmsize);
 652         if (rc)
 653                 RETURN(rc);
 654
 655         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 656                                      0, lmmsize, LUSTRE_OPC_ANY,
 657                                      NULL);
 658         if (op_data == NULL)
 659                 RETURN(-ENOMEM);
 660
 661         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 662         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 663         ll_finish_md_op_data(op_data);
 664         if (rc < 0) {
 665                 CDEBUG(D_INFO, "md_getattr failed on inode "
 666                        "%lu/%u: rc %d\n", inode->i_ino,
 667                        inode->i_generation, rc);
 668                 GOTO(out, rc);
 669         }
 670
 671         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 672         LASSERT(body != NULL);
 673
 674         lmmsize = body->eadatasize;
 675
 676         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 677             lmmsize == 0) {
 678                 GOTO(out, rc = -ENODATA);
 679         }
 680
 681         lmm = req_capsule_server_sized_get(&req->rq_pill,
 682                                            &RMF_MDT_MD, lmmsize);
 683         LASSERT(lmm != NULL);
 684
 685         /*
 686          * This is coming from the MDS, so is probably in
 687          * little endian.  We convert it to host endian before
 688          * passing it to userspace.
 689          */
 690         /* We don't swab objects for directories */
 691         switch (le32_to_cpu(lmm->lmm_magic)) {
 692         case LOV_MAGIC_V1:
 693                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 694                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 695                 break;
 696         case LOV_MAGIC_V3:
 697                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 698                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 699                 break;
 700         default:
 701                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 702                 rc = -EPROTO;
 703         }
 704 out:
 705         *lmmp = lmm;
 706         *lmm_size = lmmsize;
 707         *request = req;
 708         return rc;
 709 }
 710
 711 /*
 712  *  Get MDT index for the inode.
 713  */
 714 int ll_get_mdt_idx(struct inode *inode)
 715 {
 716         struct ll_sb_info *sbi = ll_i2sbi(inode);
 717         struct md_op_data *op_data;
 718         int rc, mdtidx;
 719         ENTRY;
 720
 721         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 722                                      0, LUSTRE_OPC_ANY, NULL);
 723         if (op_data == NULL)
 724                 RETURN(-ENOMEM);
 725
 726         op_data->op_valid |= OBD_MD_MDTIDX;
 727         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 728         mdtidx = op_data->op_mds;
 729         ll_finish_md_op_data(op_data);
 730         if (rc < 0) {
 731                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 732                 RETURN(rc);
 733         }
 734         return mdtidx;
 735 }
 736
 737 static int copy_and_ioctl(int cmd, struct obd_export *exp, void *data, int len)
 738 {
 739         void *ptr;
 740         int rc;
 741
 742         OBD_ALLOC(ptr, len);
 743         if (ptr == NULL)
 744                 return -ENOMEM;
 745         if (cfs_copy_from_user(ptr, data, len)) {
 746                 OBD_FREE(ptr, len);
 747                 return -EFAULT;
 748         }
 749         rc = obd_iocontrol(cmd, exp, len, data, NULL);
 750         OBD_FREE(ptr, len);
 751         return rc;
 752 }
 753
 754 static int ll_dir_ioctl(struct inode *inode, struct file *file,
 755                         unsigned int cmd, unsigned long arg)
 756 {
 757         struct ll_sb_info *sbi = ll_i2sbi(inode);
 758         struct obd_ioctl_data *data;
 759         int rc = 0;
 760         ENTRY;
 761
 762         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
 763                inode->i_ino, inode->i_generation, inode, cmd);
 764
 765         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
 766         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
 767                 return -ENOTTY;
 768
 769         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
 770         switch(cmd) {
 771         case FSFILT_IOC_GETFLAGS:
 772         case FSFILT_IOC_SETFLAGS:
 773                 RETURN(ll_iocontrol(inode, file, cmd, arg));
 774         case FSFILT_IOC_GETVERSION_OLD:
 775         case FSFILT_IOC_GETVERSION:
 776                 RETURN(put_user(inode->i_generation, (int *)arg));
 777         /* We need to special case any other ioctls we want to handle,
 778          * to send them to the MDS/OST as appropriate and to properly
 779          * network encode the arg field.
 780         case FSFILT_IOC_SETVERSION_OLD:
 781         case FSFILT_IOC_SETVERSION:
 782         */
 783         case LL_IOC_GET_MDTIDX: {
 784                 int mdtidx;
 785
 786                 mdtidx = ll_get_mdt_idx(inode);
 787                 if (mdtidx < 0)
 788                         RETURN(mdtidx);
 789
 790                 if (put_user((int)mdtidx, (int*)arg))
 791                         RETURN(-EFAULT);
 792
 793                 return 0;
 794         }
 795         case IOC_MDC_LOOKUP: {
 796                 struct ptlrpc_request *request = NULL;
 797                 int namelen, len = 0;
 798                 char *buf = NULL;
 799                 char *filename;
 800                 struct md_op_data *op_data;
 801
 802                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 803                 if (rc)
 804                         RETURN(rc);
 805                 data = (void *)buf;
 806
 807                 filename = data->ioc_inlbuf1;
 808                 namelen = strlen(filename);
 809
 810                 if (namelen < 1) {
 811                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
 812                         GOTO(out_free, rc = -EINVAL);
 813                 }
 814
 815                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
 816                                              0, LUSTRE_OPC_ANY, NULL);
 817                 if (op_data == NULL)
 818                         GOTO(out_free, rc = -ENOMEM);
 819
 820                 op_data->op_valid = OBD_MD_FLID;
 821                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
 822                 ll_finish_md_op_data(op_data);
 823                 if (rc < 0) {
 824                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 825                         GOTO(out_free, rc);
 826                 }
 827                 ptlrpc_req_finished(request);
 828                 EXIT;
 829 out_free:
 830                 obd_ioctl_freedata(buf, len);
 831                 return rc;
 832         }
 833         case LL_IOC_LOV_SETSTRIPE: {
 834                 struct lov_user_md_v3 lumv3;
 835                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
 836                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
 837                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
 838
 839                 int set_default = 0;
 840
 841                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
 842                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
 843                         sizeof(lumv3p->lmm_objects[0]));
 844                 /* first try with v1 which is smaller than v3 */
 845                 if (cfs_copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
 846                         RETURN(-EFAULT);
 847
 848                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
 849                         if (cfs_copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
 850                                 RETURN(-EFAULT);
 851                 }
 852
 853                 if (inode->i_sb->s_root == file->f_dentry)
 854                         set_default = 1;
 855
 856                 /* in v1 and v3 cases lumv1 points to data */
 857                 rc = ll_dir_setstripe(inode, lumv1, set_default);
 858
 859                 RETURN(rc);
 860         }
 861         case LL_IOC_OBD_STATFS:
 862                 RETURN(ll_obd_statfs(inode, (void *)arg));
 863         case LL_IOC_LOV_GETSTRIPE:
 864         case LL_IOC_MDC_GETINFO:
 865         case IOC_MDC_GETFILEINFO:
 866         case IOC_MDC_GETFILESTRIPE: {
 867                 struct ptlrpc_request *request = NULL;
 868                 struct lov_user_md *lump;
 869                 struct lov_mds_md *lmm = NULL;
 870                 struct mdt_body *body;
 871                 char *filename = NULL;
 872                 int lmmsize;
 873
 874                 if (cmd == IOC_MDC_GETFILEINFO ||
 875                     cmd == IOC_MDC_GETFILESTRIPE) {
 876                         filename = getname((const char *)arg);
 877                         if (IS_ERR(filename))
 878                                 RETURN(PTR_ERR(filename));
 879
 880                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
 881                                                       &lmmsize, &request);
 882                 } else {
 883                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
 884                 }
 885
 886                 if (request) {
 887                         body = req_capsule_server_get(&request->rq_pill,
 888                                                       &RMF_MDT_BODY);
 889                         LASSERT(body != NULL);
 890                 } else {
 891                         GOTO(out_req, rc);
 892                 }
 893
 894                 if (rc < 0) {
 895                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
 896                                                cmd == LL_IOC_MDC_GETINFO))
 897                                 GOTO(skip_lmm, rc = 0);
 898                         else
 899                                 GOTO(out_req, rc);
 900                 }
 901
 902                 if (cmd == IOC_MDC_GETFILESTRIPE ||
 903                     cmd == LL_IOC_LOV_GETSTRIPE) {
 904                         lump = (struct lov_user_md *)arg;
 905                 } else {
 906                         struct lov_user_mds_data *lmdp;
 907                         lmdp = (struct lov_user_mds_data *)arg;
 908                         lump = &lmdp->lmd_lmm;
 909                 }
 910                 if (cfs_copy_to_user(lump, lmm, lmmsize))
 911                         GOTO(out_req, rc = -EFAULT);
 912         skip_lmm:
 913                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
 914                         struct lov_user_mds_data *lmdp;
 915                         lstat_t st = { 0 };
 916
 917                         st.st_dev     = inode->i_sb->s_dev;
 918                         st.st_mode    = body->mode;
 919                         st.st_nlink   = body->nlink;
 920                         st.st_uid     = body->uid;
 921                         st.st_gid     = body->gid;
 922                         st.st_rdev    = body->rdev;
 923                         st.st_size    = body->size;
 924                         st.st_blksize = CFS_PAGE_SIZE;
 925                         st.st_blocks  = body->blocks;
 926                         st.st_atime   = body->atime;
 927                         st.st_mtime   = body->mtime;
 928                         st.st_ctime   = body->ctime;
 929                         st.st_ino     = inode->i_ino;
 930
 931                         lmdp = (struct lov_user_mds_data *)arg;
 932                         if (cfs_copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
 933                                 GOTO(out_req, rc = -EFAULT);
 934                 }
 935
 936                 EXIT;
 937         out_req:
 938                 ptlrpc_req_finished(request);
 939                 if (filename)
 940                         putname(filename);
 941                 return rc;
 942         }
 943         case IOC_LOV_GETINFO: {
 944                 struct lov_user_mds_data *lumd;
 945                 struct lov_stripe_md *lsm;
 946                 struct lov_user_md *lum;
 947                 struct lov_mds_md *lmm;
 948                 int lmmsize;
 949                 lstat_t st;
 950
 951                 lumd = (struct lov_user_mds_data *)arg;
 952                 lum = &lumd->lmd_lmm;
 953
 954                 rc = ll_get_max_mdsize(sbi, &lmmsize);
 955                 if (rc)
 956                         RETURN(rc);
 957
 958                 OBD_ALLOC(lmm, lmmsize);
 959                 if (cfs_copy_from_user(lmm, lum, lmmsize))
 960                         GOTO(free_lmm, rc = -EFAULT);
 961
 962                 switch (lmm->lmm_magic) {
 963                 case LOV_USER_MAGIC_V1:
 964                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
 965                                 break;
 966                         /* swab objects first so that stripes num will be sane */
 967                         lustre_swab_lov_user_md_objects(
 968                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
 969                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
 970                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 971                         break;
 972                 case LOV_USER_MAGIC_V3:
 973                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
 974                                 break;
 975                         /* swab objects first so that stripes num will be sane */
 976                         lustre_swab_lov_user_md_objects(
 977                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
 978                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
 979                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 980                         break;
 981                 default:
 982                         GOTO(free_lmm, rc = -EINVAL);
 983                 }
 984
 985                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
 986                 if (rc < 0)
 987                         GOTO(free_lmm, rc = -ENOMEM);
 988
 989                 /* Perform glimpse_size operation. */
 990                 memset(&st, 0, sizeof(st));
 991
 992                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
 993                 if (rc)
 994                         GOTO(free_lsm, rc);
 995
 996                 if (cfs_copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
 997                         GOTO(free_lsm, rc = -EFAULT);
 998
 999                 EXIT;
1000         free_lsm:
1001                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1002         free_lmm:
1003                 OBD_FREE(lmm, lmmsize);
1004                 return rc;
1005         }
1006         case OBD_IOC_LLOG_CATINFO: {
1007                 struct ptlrpc_request *req = NULL;
1008                 char                  *buf = NULL;
1009                 char                  *str;
1010                 int                    len = 0;
1011
1012                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1013                 if (rc)
1014                         RETURN(rc);
1015                 data = (void *)buf;
1016
1017                 if (!data->ioc_inlbuf1) {
1018                         obd_ioctl_freedata(buf, len);
1019                         RETURN(-EINVAL);
1020                 }
1021
1022                 req = ptlrpc_request_alloc(sbi2mdc(sbi)->cl_import,
1023                                            &RQF_LLOG_CATINFO);
1024                 if (req == NULL)
1025                         GOTO(out_catinfo, rc = -ENOMEM);
1026
1027                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
1028                                      data->ioc_inllen1);
1029                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_CLIENT,
1030                                      data->ioc_inllen2);
1031
1032                 rc = ptlrpc_request_pack(req, LUSTRE_LOG_VERSION, LLOG_CATINFO);
1033                 if (rc) {
1034                         ptlrpc_request_free(req);
1035                         GOTO(out_catinfo, rc);
1036                 }
1037
1038                 str = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
1039                 memcpy(str, data->ioc_inlbuf1, data->ioc_inllen1);
1040                 if (data->ioc_inllen2) {
1041                         str = req_capsule_client_get(&req->rq_pill,
1042                                                      &RMF_STRING);
1043                         memcpy(str, data->ioc_inlbuf2, data->ioc_inllen2);
1044                 }
1045
1046                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_SERVER,
1047                                      data->ioc_plen1);
1048                 ptlrpc_request_set_replen(req);
1049
1050                 rc = ptlrpc_queue_wait(req);
1051                 if (!rc) {
1052                         str = req_capsule_server_get(&req->rq_pill,
1053                                                      &RMF_STRING);
1054                         if (cfs_copy_to_user(data->ioc_pbuf1, str,
1055                                              data->ioc_plen1))
1056                                 rc = -EFAULT;
1057                 }
1058                 ptlrpc_req_finished(req);
1059         out_catinfo:
1060                 obd_ioctl_freedata(buf, len);
1061                 RETURN(rc);
1062         }
1063         case OBD_IOC_QUOTACHECK: {
1064                 struct obd_quotactl *oqctl;
1065                 int error = 0;
1066
1067                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1068                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1069                         RETURN(-EPERM);
1070
1071                 OBD_ALLOC_PTR(oqctl);
1072                 if (!oqctl)
1073                         RETURN(-ENOMEM);
1074                 oqctl->qc_type = arg;
1075                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1076                 if (rc < 0) {
1077                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1078                         error = rc;
1079                 }
1080
1081                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1082                 if (rc < 0)
1083                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1084
1085                 OBD_FREE_PTR(oqctl);
1086                 return error ?: rc;
1087         }
1088         case OBD_IOC_POLL_QUOTACHECK: {
1089                 struct if_quotacheck *check;
1090
1091                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1092                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1093                         RETURN(-EPERM);
1094
1095                 OBD_ALLOC_PTR(check);
1096                 if (!check)
1097                         RETURN(-ENOMEM);
1098
1099                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1100                                    NULL);
1101                 if (rc) {
1102                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1103                         if (cfs_copy_to_user((void *)arg, check,
1104                                              sizeof(*check)))
1105                                 rc = -EFAULT;
1106                         GOTO(out_poll, rc);
1107                 }
1108
1109                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1110                                    NULL);
1111                 if (rc) {
1112                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1113                         if (cfs_copy_to_user((void *)arg, check,
1114                                              sizeof(*check)))
1115                                 rc = -EFAULT;
1116                         GOTO(out_poll, rc);
1117                 }
1118         out_poll:
1119                 OBD_FREE_PTR(check);
1120                 RETURN(rc);
1121         }
1122         case OBD_IOC_QUOTACTL: {
1123                 struct if_quotactl *qctl;
1124                 int cmd, type, id, valid;
1125
1126                 OBD_ALLOC_PTR(qctl);
1127                 if (!qctl)
1128                         RETURN(-ENOMEM);
1129
1130                 if (cfs_copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1131                         GOTO(out_quotactl, rc = -EFAULT);
1132
1133                 cmd = qctl->qc_cmd;
1134                 type = qctl->qc_type;
1135                 id = qctl->qc_id;
1136                 valid = qctl->qc_valid;
1137
1138                 switch (cmd) {
1139                 case LUSTRE_Q_INVALIDATE:
1140                 case LUSTRE_Q_FINVALIDATE:
1141                 case Q_QUOTAON:
1142                 case Q_QUOTAOFF:
1143                 case Q_SETQUOTA:
1144                 case Q_SETINFO:
1145                         if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1146                             sbi->ll_flags & LL_SBI_RMT_CLIENT)
1147                                 GOTO(out_quotactl, rc = -EPERM);
1148                         break;
1149                 case Q_GETQUOTA:
1150                         if (((type == USRQUOTA && cfs_curproc_euid() != id) ||
1151                              (type == GRPQUOTA && !in_egroup_p(id))) &&
1152                             (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1153                              sbi->ll_flags & LL_SBI_RMT_CLIENT))
1154                                 GOTO(out_quotactl, rc = -EPERM);
1155                         break;
1156                 case Q_GETINFO:
1157                         break;
1158                 default:
1159                         CERROR("unsupported quotactl op: %#x\n", cmd);
1160                         GOTO(out_quotactl, rc = -ENOTTY);
1161                 }
1162
1163                 if (valid != QC_GENERAL) {
1164                         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1165                                 GOTO(out_quotactl, rc = -EOPNOTSUPP);
1166
1167                         if (cmd == Q_GETINFO)
1168                                 qctl->qc_cmd = Q_GETOINFO;
1169                         else if (cmd == Q_GETQUOTA)
1170                                 qctl->qc_cmd = Q_GETOQUOTA;
1171                         else
1172                                 GOTO(out_quotactl, rc = -EINVAL);
1173
1174                         switch (valid) {
1175                         case QC_MDTIDX:
1176                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1177                                                    sbi->ll_md_exp,
1178                                                    sizeof(*qctl), qctl, NULL);
1179                                 break;
1180                         case QC_OSTIDX:
1181                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1182                                                    sbi->ll_dt_exp,
1183                                                    sizeof(*qctl), qctl, NULL);
1184                                 break;
1185                         case QC_UUID:
1186                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1187                                                    sbi->ll_md_exp,
1188                                                    sizeof(*qctl), qctl, NULL);
1189                                 if (rc == -EAGAIN)
1190                                         rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1191                                                            sbi->ll_dt_exp,
1192                                                            sizeof(*qctl), qctl,
1193                                                            NULL);
1194                                 break;
1195                         default:
1196                                 rc = -EINVAL;
1197                                 break;
1198                         }
1199
1200                         if (rc)
1201                                 GOTO(out_quotactl, rc);
1202                         else
1203                                 qctl->qc_cmd = cmd;
1204                 } else {
1205                         struct obd_quotactl *oqctl;
1206
1207                         OBD_ALLOC_PTR(oqctl);
1208                         if (!oqctl)
1209                                 GOTO(out_quotactl, rc = -ENOMEM);
1210
1211                         QCTL_COPY(oqctl, qctl);
1212                         rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1213                         if (rc) {
1214                                 if (rc != -EALREADY && cmd == Q_QUOTAON) {
1215                                         oqctl->qc_cmd = Q_QUOTAOFF;
1216                                         obd_quotactl(sbi->ll_md_exp, oqctl);
1217                                 }
1218                                 OBD_FREE_PTR(oqctl);
1219                                 GOTO(out_quotactl, rc);
1220                         } else {
1221                                 QCTL_COPY(qctl, oqctl);
1222                                 OBD_FREE_PTR(oqctl);
1223                         }
1224                 }
1225
1226                 if (cfs_copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1227                         rc = -EFAULT;
1228
1229         out_quotactl:
1230                 OBD_FREE_PTR(qctl);
1231                 RETURN(rc);
1232         }
1233         case OBD_IOC_GETNAME: {
1234                 struct obd_device *obd = class_exp2obd(sbi->ll_dt_exp);
1235                 if (!obd)
1236                         RETURN(-EFAULT);
1237                 if (cfs_copy_to_user((void *)arg, obd->obd_name,
1238                                      strlen(obd->obd_name) + 1))
1239                         RETURN (-EFAULT);
1240                 RETURN(0);
1241         }
1242         case LL_IOC_FLUSHCTX:
1243                 RETURN(ll_flush_ctx(inode));
1244 #ifdef CONFIG_FS_POSIX_ACL
1245         case LL_IOC_RMTACL: {
1246             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1247                 inode == inode->i_sb->s_root->d_inode) {
1248                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1249
1250                 LASSERT(fd != NULL);
1251                 rc = rct_add(&sbi->ll_rct, cfs_curproc_pid(), arg);
1252                 if (!rc)
1253                         fd->fd_flags |= LL_FILE_RMTACL;
1254                 RETURN(rc);
1255             } else
1256                 RETURN(0);
1257         }
1258 #endif
1259         case LL_IOC_GETOBDCOUNT: {
1260                 int count;
1261
1262                 if (cfs_copy_from_user(&count, (int *)arg, sizeof(int)))
1263                         RETURN(-EFAULT);
1264
1265                 if (!count) {
1266                         /* get ost count */
1267                         struct lov_obd *lov = &sbi->ll_dt_exp->exp_obd->u.lov;
1268                         count = lov->desc.ld_tgt_count;
1269                 } else {
1270                         /* get mdt count */
1271                         struct lmv_obd *lmv = &sbi->ll_md_exp->exp_obd->u.lmv;
1272                         count = lmv->desc.ld_tgt_count;
1273                 }
1274
1275                 if (cfs_copy_to_user((int *)arg, &count, sizeof(int)))
1276                         RETURN(-EFAULT);
1277
1278                 RETURN(0);
1279         }
1280         case LL_IOC_PATH2FID:
1281                 if (cfs_copy_to_user((void *)arg, ll_inode2fid(inode),
1282                                      sizeof(struct lu_fid)))
1283                         RETURN(-EFAULT);
1284                 RETURN(0);
1285         case OBD_IOC_CHANGELOG_SEND:
1286         case OBD_IOC_CHANGELOG_CLEAR:
1287                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1288                                     sizeof(struct ioc_changelog));
1289                 RETURN(rc);
1290         case OBD_IOC_FID2PATH:
1291                 RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg));
1292         case LL_IOC_HSM_CT_START:
1293                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1294                                     sizeof(struct lustre_kernelcomm));
1295                 RETURN(rc);
1296
1297         default:
1298                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp,0,NULL,(void *)arg));
1299         }
1300 }
1301
1302 int ll_dir_open(struct inode *inode, struct file *file)
1303 {
1304         ENTRY;
1305         RETURN(ll_file_open(inode, file));
1306 }
1307
1308 int ll_dir_release(struct inode *inode, struct file *file)
1309 {
1310         ENTRY;
1311         RETURN(ll_file_release(inode, file));
1312 }
1313
1314 struct file_operations ll_dir_operations = {
1315         .open     = ll_dir_open,
1316         .release  = ll_dir_release,
1317         .read     = generic_read_dir,
1318         .readdir  = ll_readdir,
1319         .ioctl    = ll_dir_ioctl
1320 };