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 (c) 2002, 2010, Oracle and/or its affiliates. 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                                 __u64          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                                 if (cfs_curproc_is_32bit())
 471                                         ino = cl_fid_build_ino32(&fid);
 472                                 else
 473                                         ino = cl_fid_build_ino(&fid);
 474                                 type = ll_dirent_type_get(ent);
 475                                 done = filldir(cookie, name, namelen,
 476                                                (loff_t)hash, ino, type);
 477                         }
 478                         next = le64_to_cpu(dp->ldp_hash_end);
 479                         ll_put_page(page);
 480                         if (!done) {
 481                                 pos = next;
 482                                 if (pos == DIR_END_OFF)
 483                                         /*
 484                                          * End of directory reached.
 485                                          */
 486                                         done = 1;
 487                                 else if (1 /* chain is exhausted*/)
 488                                         /*
 489                                          * Normal case: continue to the next
 490                                          * page.
 491                                          */
 492                                         page = ll_get_dir_page(inode, pos, 1,
 493                                                                &chain);
 494                                 else {
 495                                         /*
 496                                          * go into overflow page.
 497                                          */
 498                                 }
 499                         } else
 500                                 pos = hash;
 501                 } else {
 502                         rc = PTR_ERR(page);
 503                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 504                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 505                 }
 506         }
 507
 508         filp->f_pos = (loff_t)pos;
 509         filp->f_version = inode->i_version;
 510         touch_atime(filp->f_vfsmnt, filp->f_dentry);
 511
 512         ll_dir_chain_fini(&chain);
 513
 514         RETURN(rc);
 515 }
 516
 517 int ll_send_mgc_param(struct obd_export *mgc, char *string)
 518 {
 519         struct mgs_send_param *msp;
 520         int rc = 0;
 521
 522         OBD_ALLOC_PTR(msp);
 523         if (!msp)
 524                 return -ENOMEM;
 525
 526         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 527         rc = obd_set_info_async(mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 528                                 sizeof(struct mgs_send_param), msp, NULL);
 529         if (rc)
 530                 CERROR("Failed to set parameter: %d\n", rc);
 531         OBD_FREE_PTR(msp);
 532
 533         return rc;
 534 }
 535
 536 char *ll_get_fsname(struct inode *inode)
 537 {
 538         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 539         char *ptr, *fsname;
 540         int len;
 541
 542         OBD_ALLOC(fsname, MGS_PARAM_MAXLEN);
 543         len = strlen(lsi->lsi_lmd->lmd_profile);
 544         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
 545         if (ptr && (strcmp(ptr, "-client") == 0))
 546                 len -= 7;
 547         strncpy(fsname, lsi->lsi_lmd->lmd_profile, len);
 548         fsname[len] = '\0';
 549
 550         return fsname;
 551 }
 552
 553 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 554                      int set_default)
 555 {
 556         struct ll_sb_info *sbi = ll_i2sbi(inode);
 557         struct md_op_data *op_data;
 558         struct ptlrpc_request *req = NULL;
 559         int rc = 0;
 560         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 561         struct obd_device *mgc = lsi->lsi_mgc;
 562         char *fsname = NULL, *param = NULL;
 563         int lum_size;
 564
 565         if (lump != NULL) {
 566                 /*
 567                  * This is coming from userspace, so should be in
 568                  * local endian.  But the MDS would like it in little
 569                  * endian, so we swab it before we send it.
 570                  */
 571                 switch (lump->lmm_magic) {
 572                 case LOV_USER_MAGIC_V1: {
 573                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 574                                 lustre_swab_lov_user_md_v1(lump);
 575                         lum_size = sizeof(struct lov_user_md_v1);
 576                         break;
 577                         }
 578                 case LOV_USER_MAGIC_V3: {
 579                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 580                                 lustre_swab_lov_user_md_v3(
 581                                         (struct lov_user_md_v3 *)lump);
 582                         lum_size = sizeof(struct lov_user_md_v3);
 583                         break;
 584                         }
 585                 default: {
 586                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 587                                         " %#08x != %#08x nor %#08x\n",
 588                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 589                                         LOV_USER_MAGIC_V3);
 590                         RETURN(-EINVAL);
 591                         }
 592                }
 593         } else {
 594                 lum_size = sizeof(struct lov_user_md_v1);
 595         }
 596
 597         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 598                                      LUSTRE_OPC_ANY, NULL);
 599         if (IS_ERR(op_data))
 600                 RETURN(PTR_ERR(op_data));
 601
 602         /* swabbing is done in lov_setstripe() on server side */
 603         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 604                         NULL, 0, &req, NULL);
 605         ll_finish_md_op_data(op_data);
 606         ptlrpc_req_finished(req);
 607         if (rc) {
 608                 if (rc != -EPERM && rc != -EACCES)
 609                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 610         }
 611
 612         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 613          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 614          need the make the distiction between the 2 versions */
 615         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 616                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 617
 618                 /* Get fsname and assume devname to be -MDT0000. */
 619                 fsname = ll_get_fsname(inode);
 620                 /* Set root stripesize */
 621                 sprintf(param, "%s-MDT0000.lov.stripesize=%u", fsname,
 622                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 623                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 624                 if (rc)
 625                         goto end;
 626
 627                 /* Set root stripecount */
 628                 sprintf(param, "%s-MDT0000.lov.stripecount=%hd", fsname,
 629                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 630                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 631                 if (rc)
 632                         goto end;
 633
 634                 /* Set root stripeoffset */
 635                 sprintf(param, "%s-MDT0000.lov.stripeoffset=%hd", fsname,
 636                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 637                         (typeof(lump->lmm_stripe_offset))(-1));
 638                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 639                 if (rc)
 640                         goto end;
 641 end:
 642                 if (fsname)
 643                         OBD_FREE(fsname, MGS_PARAM_MAXLEN);
 644                 if (param)
 645                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 646         }
 647         return rc;
 648 }
 649
 650 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 651                      int *lmm_size, struct ptlrpc_request **request)
 652 {
 653         struct ll_sb_info *sbi = ll_i2sbi(inode);
 654         struct mdt_body   *body;
 655         struct lov_mds_md *lmm = NULL;
 656         struct ptlrpc_request *req = NULL;
 657         int rc, lmmsize;
 658         struct md_op_data *op_data;
 659
 660         rc = ll_get_max_mdsize(sbi, &lmmsize);
 661         if (rc)
 662                 RETURN(rc);
 663
 664         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 665                                      0, lmmsize, LUSTRE_OPC_ANY,
 666                                      NULL);
 667         if (op_data == NULL)
 668                 RETURN(-ENOMEM);
 669
 670         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 671         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 672         ll_finish_md_op_data(op_data);
 673         if (rc < 0) {
 674                 CDEBUG(D_INFO, "md_getattr failed on inode "
 675                        "%lu/%u: rc %d\n", inode->i_ino,
 676                        inode->i_generation, rc);
 677                 GOTO(out, rc);
 678         }
 679
 680         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 681         LASSERT(body != NULL);
 682
 683         lmmsize = body->eadatasize;
 684
 685         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 686             lmmsize == 0) {
 687                 GOTO(out, rc = -ENODATA);
 688         }
 689
 690         lmm = req_capsule_server_sized_get(&req->rq_pill,
 691                                            &RMF_MDT_MD, lmmsize);
 692         LASSERT(lmm != NULL);
 693
 694         /*
 695          * This is coming from the MDS, so is probably in
 696          * little endian.  We convert it to host endian before
 697          * passing it to userspace.
 698          */
 699         /* We don't swab objects for directories */
 700         switch (le32_to_cpu(lmm->lmm_magic)) {
 701         case LOV_MAGIC_V1:
 702                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 703                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 704                 break;
 705         case LOV_MAGIC_V3:
 706                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 707                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 708                 break;
 709         default:
 710                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 711                 rc = -EPROTO;
 712         }
 713 out:
 714         *lmmp = lmm;
 715         *lmm_size = lmmsize;
 716         *request = req;
 717         return rc;
 718 }
 719
 720 /*
 721  *  Get MDT index for the inode.
 722  */
 723 int ll_get_mdt_idx(struct inode *inode)
 724 {
 725         struct ll_sb_info *sbi = ll_i2sbi(inode);
 726         struct md_op_data *op_data;
 727         int rc, mdtidx;
 728         ENTRY;
 729
 730         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 731                                      0, LUSTRE_OPC_ANY, NULL);
 732         if (op_data == NULL)
 733                 RETURN(-ENOMEM);
 734
 735         op_data->op_valid |= OBD_MD_MDTIDX;
 736         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 737         mdtidx = op_data->op_mds;
 738         ll_finish_md_op_data(op_data);
 739         if (rc < 0) {
 740                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 741                 RETURN(rc);
 742         }
 743         return mdtidx;
 744 }
 745
 746 static int copy_and_ioctl(int cmd, struct obd_export *exp, void *data, int len)
 747 {
 748         void *ptr;
 749         int rc;
 750
 751         OBD_ALLOC(ptr, len);
 752         if (ptr == NULL)
 753                 return -ENOMEM;
 754         if (cfs_copy_from_user(ptr, data, len)) {
 755                 OBD_FREE(ptr, len);
 756                 return -EFAULT;
 757         }
 758         rc = obd_iocontrol(cmd, exp, len, data, NULL);
 759         OBD_FREE(ptr, len);
 760         return rc;
 761 }
 762
 763 static int ll_dir_ioctl(struct inode *inode, struct file *file,
 764                         unsigned int cmd, unsigned long arg)
 765 {
 766         struct ll_sb_info *sbi = ll_i2sbi(inode);
 767         struct obd_ioctl_data *data;
 768         int rc = 0;
 769         ENTRY;
 770
 771         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
 772                inode->i_ino, inode->i_generation, inode, cmd);
 773
 774         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
 775         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
 776                 return -ENOTTY;
 777
 778         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
 779         switch(cmd) {
 780         case FSFILT_IOC_GETFLAGS:
 781         case FSFILT_IOC_SETFLAGS:
 782                 RETURN(ll_iocontrol(inode, file, cmd, arg));
 783         case FSFILT_IOC_GETVERSION_OLD:
 784         case FSFILT_IOC_GETVERSION:
 785                 RETURN(put_user(inode->i_generation, (int *)arg));
 786         /* We need to special case any other ioctls we want to handle,
 787          * to send them to the MDS/OST as appropriate and to properly
 788          * network encode the arg field.
 789         case FSFILT_IOC_SETVERSION_OLD:
 790         case FSFILT_IOC_SETVERSION:
 791         */
 792         case LL_IOC_GET_MDTIDX: {
 793                 int mdtidx;
 794
 795                 mdtidx = ll_get_mdt_idx(inode);
 796                 if (mdtidx < 0)
 797                         RETURN(mdtidx);
 798
 799                 if (put_user((int)mdtidx, (int*)arg))
 800                         RETURN(-EFAULT);
 801
 802                 return 0;
 803         }
 804         case IOC_MDC_LOOKUP: {
 805                 struct ptlrpc_request *request = NULL;
 806                 int namelen, len = 0;
 807                 char *buf = NULL;
 808                 char *filename;
 809                 struct md_op_data *op_data;
 810
 811                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 812                 if (rc)
 813                         RETURN(rc);
 814                 data = (void *)buf;
 815
 816                 filename = data->ioc_inlbuf1;
 817                 namelen = strlen(filename);
 818
 819                 if (namelen < 1) {
 820                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
 821                         GOTO(out_free, rc = -EINVAL);
 822                 }
 823
 824                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
 825                                              0, LUSTRE_OPC_ANY, NULL);
 826                 if (op_data == NULL)
 827                         GOTO(out_free, rc = -ENOMEM);
 828
 829                 op_data->op_valid = OBD_MD_FLID;
 830                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
 831                 ll_finish_md_op_data(op_data);
 832                 if (rc < 0) {
 833                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 834                         GOTO(out_free, rc);
 835                 }
 836                 ptlrpc_req_finished(request);
 837                 EXIT;
 838 out_free:
 839                 obd_ioctl_freedata(buf, len);
 840                 return rc;
 841         }
 842         case LL_IOC_LOV_SETSTRIPE: {
 843                 struct lov_user_md_v3 lumv3;
 844                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
 845                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
 846                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
 847
 848                 int set_default = 0;
 849
 850                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
 851                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
 852                         sizeof(lumv3p->lmm_objects[0]));
 853                 /* first try with v1 which is smaller than v3 */
 854                 if (cfs_copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
 855                         RETURN(-EFAULT);
 856
 857                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
 858                         if (cfs_copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
 859                                 RETURN(-EFAULT);
 860                 }
 861
 862                 if (inode->i_sb->s_root == file->f_dentry)
 863                         set_default = 1;
 864
 865                 /* in v1 and v3 cases lumv1 points to data */
 866                 rc = ll_dir_setstripe(inode, lumv1, set_default);
 867
 868                 RETURN(rc);
 869         }
 870         case LL_IOC_OBD_STATFS:
 871                 RETURN(ll_obd_statfs(inode, (void *)arg));
 872         case LL_IOC_LOV_GETSTRIPE:
 873         case LL_IOC_MDC_GETINFO:
 874         case IOC_MDC_GETFILEINFO:
 875         case IOC_MDC_GETFILESTRIPE: {
 876                 struct ptlrpc_request *request = NULL;
 877                 struct lov_user_md *lump;
 878                 struct lov_mds_md *lmm = NULL;
 879                 struct mdt_body *body;
 880                 char *filename = NULL;
 881                 int lmmsize;
 882
 883                 if (cmd == IOC_MDC_GETFILEINFO ||
 884                     cmd == IOC_MDC_GETFILESTRIPE) {
 885                         filename = getname((const char *)arg);
 886                         if (IS_ERR(filename))
 887                                 RETURN(PTR_ERR(filename));
 888
 889                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
 890                                                       &lmmsize, &request);
 891                 } else {
 892                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
 893                 }
 894
 895                 if (request) {
 896                         body = req_capsule_server_get(&request->rq_pill,
 897                                                       &RMF_MDT_BODY);
 898                         LASSERT(body != NULL);
 899                 } else {
 900                         GOTO(out_req, rc);
 901                 }
 902
 903                 if (rc < 0) {
 904                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
 905                                                cmd == LL_IOC_MDC_GETINFO))
 906                                 GOTO(skip_lmm, rc = 0);
 907                         else
 908                                 GOTO(out_req, rc);
 909                 }
 910
 911                 if (cmd == IOC_MDC_GETFILESTRIPE ||
 912                     cmd == LL_IOC_LOV_GETSTRIPE) {
 913                         lump = (struct lov_user_md *)arg;
 914                 } else {
 915                         struct lov_user_mds_data *lmdp;
 916                         lmdp = (struct lov_user_mds_data *)arg;
 917                         lump = &lmdp->lmd_lmm;
 918                 }
 919                 if (cfs_copy_to_user(lump, lmm, lmmsize))
 920                         GOTO(out_req, rc = -EFAULT);
 921         skip_lmm:
 922                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
 923                         struct lov_user_mds_data *lmdp;
 924                         lstat_t st = { 0 };
 925
 926                         st.st_dev     = inode->i_sb->s_dev;
 927                         st.st_mode    = body->mode;
 928                         st.st_nlink   = body->nlink;
 929                         st.st_uid     = body->uid;
 930                         st.st_gid     = body->gid;
 931                         st.st_rdev    = body->rdev;
 932                         st.st_size    = body->size;
 933                         st.st_blksize = CFS_PAGE_SIZE;
 934                         st.st_blocks  = body->blocks;
 935                         st.st_atime   = body->atime;
 936                         st.st_mtime   = body->mtime;
 937                         st.st_ctime   = body->ctime;
 938                         st.st_ino     = inode->i_ino;
 939
 940                         lmdp = (struct lov_user_mds_data *)arg;
 941                         if (cfs_copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
 942                                 GOTO(out_req, rc = -EFAULT);
 943                 }
 944
 945                 EXIT;
 946         out_req:
 947                 ptlrpc_req_finished(request);
 948                 if (filename)
 949                         putname(filename);
 950                 return rc;
 951         }
 952         case IOC_LOV_GETINFO: {
 953                 struct lov_user_mds_data *lumd;
 954                 struct lov_stripe_md *lsm;
 955                 struct lov_user_md *lum;
 956                 struct lov_mds_md *lmm;
 957                 int lmmsize;
 958                 lstat_t st;
 959
 960                 lumd = (struct lov_user_mds_data *)arg;
 961                 lum = &lumd->lmd_lmm;
 962
 963                 rc = ll_get_max_mdsize(sbi, &lmmsize);
 964                 if (rc)
 965                         RETURN(rc);
 966
 967                 OBD_ALLOC(lmm, lmmsize);
 968                 if (cfs_copy_from_user(lmm, lum, lmmsize))
 969                         GOTO(free_lmm, rc = -EFAULT);
 970
 971                 switch (lmm->lmm_magic) {
 972                 case LOV_USER_MAGIC_V1:
 973                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
 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_v1 *)lmm)->lmm_objects,
 978                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
 979                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 980                         break;
 981                 case LOV_USER_MAGIC_V3:
 982                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
 983                                 break;
 984                         /* swab objects first so that stripes num will be sane */
 985                         lustre_swab_lov_user_md_objects(
 986                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
 987                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
 988                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 989                         break;
 990                 default:
 991                         GOTO(free_lmm, rc = -EINVAL);
 992                 }
 993
 994                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
 995                 if (rc < 0)
 996                         GOTO(free_lmm, rc = -ENOMEM);
 997
 998                 /* Perform glimpse_size operation. */
 999                 memset(&st, 0, sizeof(st));
1000
1001                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1002                 if (rc)
1003                         GOTO(free_lsm, rc);
1004
1005                 if (cfs_copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1006                         GOTO(free_lsm, rc = -EFAULT);
1007
1008                 EXIT;
1009         free_lsm:
1010                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1011         free_lmm:
1012                 OBD_FREE(lmm, lmmsize);
1013                 return rc;
1014         }
1015         case OBD_IOC_LLOG_CATINFO: {
1016                 struct ptlrpc_request *req = NULL;
1017                 char                  *buf = NULL;
1018                 char                  *str;
1019                 int                    len = 0;
1020
1021                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1022                 if (rc)
1023                         RETURN(rc);
1024                 data = (void *)buf;
1025
1026                 if (!data->ioc_inlbuf1) {
1027                         obd_ioctl_freedata(buf, len);
1028                         RETURN(-EINVAL);
1029                 }
1030
1031                 req = ptlrpc_request_alloc(sbi2mdc(sbi)->cl_import,
1032                                            &RQF_LLOG_CATINFO);
1033                 if (req == NULL)
1034                         GOTO(out_catinfo, rc = -ENOMEM);
1035
1036                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
1037                                      data->ioc_inllen1);
1038                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_CLIENT,
1039                                      data->ioc_inllen2);
1040
1041                 rc = ptlrpc_request_pack(req, LUSTRE_LOG_VERSION, LLOG_CATINFO);
1042                 if (rc) {
1043                         ptlrpc_request_free(req);
1044                         GOTO(out_catinfo, rc);
1045                 }
1046
1047                 str = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
1048                 memcpy(str, data->ioc_inlbuf1, data->ioc_inllen1);
1049                 if (data->ioc_inllen2) {
1050                         str = req_capsule_client_get(&req->rq_pill,
1051                                                      &RMF_STRING);
1052                         memcpy(str, data->ioc_inlbuf2, data->ioc_inllen2);
1053                 }
1054
1055                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_SERVER,
1056                                      data->ioc_plen1);
1057                 ptlrpc_request_set_replen(req);
1058
1059                 rc = ptlrpc_queue_wait(req);
1060                 if (!rc) {
1061                         str = req_capsule_server_get(&req->rq_pill,
1062                                                      &RMF_STRING);
1063                         if (cfs_copy_to_user(data->ioc_pbuf1, str,
1064                                              data->ioc_plen1))
1065                                 rc = -EFAULT;
1066                 }
1067                 ptlrpc_req_finished(req);
1068         out_catinfo:
1069                 obd_ioctl_freedata(buf, len);
1070                 RETURN(rc);
1071         }
1072         case OBD_IOC_QUOTACHECK: {
1073                 struct obd_quotactl *oqctl;
1074                 int error = 0;
1075
1076                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1077                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1078                         RETURN(-EPERM);
1079
1080                 OBD_ALLOC_PTR(oqctl);
1081                 if (!oqctl)
1082                         RETURN(-ENOMEM);
1083                 oqctl->qc_type = arg;
1084                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1085                 if (rc < 0) {
1086                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1087                         error = rc;
1088                 }
1089
1090                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1091                 if (rc < 0)
1092                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1093
1094                 OBD_FREE_PTR(oqctl);
1095                 return error ?: rc;
1096         }
1097         case OBD_IOC_POLL_QUOTACHECK: {
1098                 struct if_quotacheck *check;
1099
1100                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1101                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1102                         RETURN(-EPERM);
1103
1104                 OBD_ALLOC_PTR(check);
1105                 if (!check)
1106                         RETURN(-ENOMEM);
1107
1108                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1109                                    NULL);
1110                 if (rc) {
1111                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1112                         if (cfs_copy_to_user((void *)arg, check,
1113                                              sizeof(*check)))
1114                                 rc = -EFAULT;
1115                         GOTO(out_poll, rc);
1116                 }
1117
1118                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1119                                    NULL);
1120                 if (rc) {
1121                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1122                         if (cfs_copy_to_user((void *)arg, check,
1123                                              sizeof(*check)))
1124                                 rc = -EFAULT;
1125                         GOTO(out_poll, rc);
1126                 }
1127         out_poll:
1128                 OBD_FREE_PTR(check);
1129                 RETURN(rc);
1130         }
1131         case OBD_IOC_QUOTACTL: {
1132                 struct if_quotactl *qctl;
1133                 int cmd, type, id, valid;
1134
1135                 OBD_ALLOC_PTR(qctl);
1136                 if (!qctl)
1137                         RETURN(-ENOMEM);
1138
1139                 if (cfs_copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1140                         GOTO(out_quotactl, rc = -EFAULT);
1141
1142                 cmd = qctl->qc_cmd;
1143                 type = qctl->qc_type;
1144                 id = qctl->qc_id;
1145                 valid = qctl->qc_valid;
1146
1147                 switch (cmd) {
1148                 case LUSTRE_Q_INVALIDATE:
1149                 case LUSTRE_Q_FINVALIDATE:
1150                 case Q_QUOTAON:
1151                 case Q_QUOTAOFF:
1152                 case Q_SETQUOTA:
1153                 case Q_SETINFO:
1154                         if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1155                             sbi->ll_flags & LL_SBI_RMT_CLIENT)
1156                                 GOTO(out_quotactl, rc = -EPERM);
1157                         break;
1158                 case Q_GETQUOTA:
1159                         if (((type == USRQUOTA && cfs_curproc_euid() != id) ||
1160                              (type == GRPQUOTA && !in_egroup_p(id))) &&
1161                             (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1162                              sbi->ll_flags & LL_SBI_RMT_CLIENT))
1163                                 GOTO(out_quotactl, rc = -EPERM);
1164                         break;
1165                 case Q_GETINFO:
1166                         break;
1167                 default:
1168                         CERROR("unsupported quotactl op: %#x\n", cmd);
1169                         GOTO(out_quotactl, rc = -ENOTTY);
1170                 }
1171
1172                 if (valid != QC_GENERAL) {
1173                         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1174                                 GOTO(out_quotactl, rc = -EOPNOTSUPP);
1175
1176                         if (cmd == Q_GETINFO)
1177                                 qctl->qc_cmd = Q_GETOINFO;
1178                         else if (cmd == Q_GETQUOTA)
1179                                 qctl->qc_cmd = Q_GETOQUOTA;
1180                         else
1181                                 GOTO(out_quotactl, rc = -EINVAL);
1182
1183                         switch (valid) {
1184                         case QC_MDTIDX:
1185                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1186                                                    sbi->ll_md_exp,
1187                                                    sizeof(*qctl), qctl, NULL);
1188                                 break;
1189                         case QC_OSTIDX:
1190                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1191                                                    sbi->ll_dt_exp,
1192                                                    sizeof(*qctl), qctl, NULL);
1193                                 break;
1194                         case QC_UUID:
1195                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1196                                                    sbi->ll_md_exp,
1197                                                    sizeof(*qctl), qctl, NULL);
1198                                 if (rc == -EAGAIN)
1199                                         rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1200                                                            sbi->ll_dt_exp,
1201                                                            sizeof(*qctl), qctl,
1202                                                            NULL);
1203                                 break;
1204                         default:
1205                                 rc = -EINVAL;
1206                                 break;
1207                         }
1208
1209                         if (rc)
1210                                 GOTO(out_quotactl, rc);
1211                         else
1212                                 qctl->qc_cmd = cmd;
1213                 } else {
1214                         struct obd_quotactl *oqctl;
1215
1216                         OBD_ALLOC_PTR(oqctl);
1217                         if (!oqctl)
1218                                 GOTO(out_quotactl, rc = -ENOMEM);
1219
1220                         QCTL_COPY(oqctl, qctl);
1221                         rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1222                         if (rc) {
1223                                 if (rc != -EALREADY && cmd == Q_QUOTAON) {
1224                                         oqctl->qc_cmd = Q_QUOTAOFF;
1225                                         obd_quotactl(sbi->ll_md_exp, oqctl);
1226                                 }
1227                                 OBD_FREE_PTR(oqctl);
1228                                 GOTO(out_quotactl, rc);
1229                         } else {
1230                                 QCTL_COPY(qctl, oqctl);
1231                                 OBD_FREE_PTR(oqctl);
1232                         }
1233                 }
1234
1235                 if (cfs_copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1236                         rc = -EFAULT;
1237
1238         out_quotactl:
1239                 OBD_FREE_PTR(qctl);
1240                 RETURN(rc);
1241         }
1242         case OBD_IOC_GETNAME: {
1243                 struct obd_device *obd = class_exp2obd(sbi->ll_dt_exp);
1244                 if (!obd)
1245                         RETURN(-EFAULT);
1246                 if (cfs_copy_to_user((void *)arg, obd->obd_name,
1247                                      strlen(obd->obd_name) + 1))
1248                         RETURN (-EFAULT);
1249                 RETURN(0);
1250         }
1251         case LL_IOC_FLUSHCTX:
1252                 RETURN(ll_flush_ctx(inode));
1253 #ifdef CONFIG_FS_POSIX_ACL
1254         case LL_IOC_RMTACL: {
1255             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1256                 inode == inode->i_sb->s_root->d_inode) {
1257                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1258
1259                 LASSERT(fd != NULL);
1260                 rc = rct_add(&sbi->ll_rct, cfs_curproc_pid(), arg);
1261                 if (!rc)
1262                         fd->fd_flags |= LL_FILE_RMTACL;
1263                 RETURN(rc);
1264             } else
1265                 RETURN(0);
1266         }
1267 #endif
1268         case LL_IOC_GETOBDCOUNT: {
1269                 int count;
1270
1271                 if (cfs_copy_from_user(&count, (int *)arg, sizeof(int)))
1272                         RETURN(-EFAULT);
1273
1274                 if (!count) {
1275                         /* get ost count */
1276                         struct lov_obd *lov = &sbi->ll_dt_exp->exp_obd->u.lov;
1277                         count = lov->desc.ld_tgt_count;
1278                 } else {
1279                         /* get mdt count */
1280                         struct lmv_obd *lmv = &sbi->ll_md_exp->exp_obd->u.lmv;
1281                         count = lmv->desc.ld_tgt_count;
1282                 }
1283
1284                 if (cfs_copy_to_user((int *)arg, &count, sizeof(int)))
1285                         RETURN(-EFAULT);
1286
1287                 RETURN(0);
1288         }
1289         case LL_IOC_PATH2FID:
1290                 if (cfs_copy_to_user((void *)arg, ll_inode2fid(inode),
1291                                      sizeof(struct lu_fid)))
1292                         RETURN(-EFAULT);
1293                 RETURN(0);
1294         case LL_IOC_GET_CONNECT_FLAGS: {
1295                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1296         }
1297         case OBD_IOC_CHANGELOG_SEND:
1298         case OBD_IOC_CHANGELOG_CLEAR:
1299                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1300                                     sizeof(struct ioc_changelog));
1301                 RETURN(rc);
1302         case OBD_IOC_FID2PATH:
1303                 RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg));
1304         case LL_IOC_HSM_CT_START:
1305                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1306                                     sizeof(struct lustre_kernelcomm));
1307                 RETURN(rc);
1308
1309         default:
1310                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp,0,NULL,(void *)arg));
1311         }
1312 }
1313
1314 int ll_dir_open(struct inode *inode, struct file *file)
1315 {
1316         ENTRY;
1317         RETURN(ll_file_open(inode, file));
1318 }
1319
1320 int ll_dir_release(struct inode *inode, struct file *file)
1321 {
1322         ENTRY;
1323         RETURN(ll_file_release(inode, file));
1324 }
1325
1326 struct file_operations ll_dir_operations = {
1327         .open     = ll_dir_open,
1328         .release  = ll_dir_release,
1329         .read     = generic_read_dir,
1330         .readdir  = ll_readdir,
1331         .ioctl    = ll_dir_ioctl
1332 };