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         if (file) {
 159                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
 160
 161                 hash = fd->fd_dir.lfd_next;
 162         } else {
 163                 struct ll_inode_info *lli = ll_i2info(inode);
 164
 165                 cfs_spin_lock(&lli->lli_sa_lock);
 166                 if (lli->lli_sai)
 167                         LASSERT(lli->lli_sai->sai_pid == cfs_curproc_pid());
 168                 else
 169                         LASSERT(lli->lli_opendir_pid == cfs_curproc_pid());
 170                 hash = lli->lli_sa_pos;
 171                 cfs_spin_unlock(&lli->lli_sa_lock);
 172         }
 173         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) off %lu\n",
 174                inode->i_ino, inode->i_generation, inode, (unsigned long)hash);
 175
 176         oc = ll_mdscapa_get(inode);
 177         rc = md_readpage(ll_i2sbi(inode)->ll_md_exp, ll_inode2fid(inode),
 178                          oc, hash, page, &request);
 179         capa_put(oc);
 180         if (!rc) {
 181                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 182                 /* Checked by mdc_readpage() */
 183                 LASSERT(body != NULL);
 184
 185                 if (body->valid & OBD_MD_FLSIZE)
 186                         cl_isize_write(inode, body->size);
 187                 SetPageUptodate(page);
 188         }
 189         ptlrpc_req_finished(request);
 190
 191         unlock_page(page);
 192         EXIT;
 193         return rc;
 194 }
 195
 196 #ifndef MS_HAS_NEW_AOPS
 197 struct address_space_operations ll_dir_aops = {
 198         .readpage  = ll_dir_readpage,
 199 };
 200 #else
 201 struct address_space_operations_ext ll_dir_aops = {
 202         .orig_aops.readpage  = ll_dir_readpage,
 203 };
 204 #endif
 205
 206 static void ll_check_page(struct inode *dir, struct page *page)
 207 {
 208         /* XXX: check page format later */
 209         SetPageChecked(page);
 210 }
 211
 212 static void ll_release_page(struct page *page, __u64 hash,
 213                             __u64 start, __u64 end)
 214 {
 215         kunmap(page);
 216         lock_page(page);
 217         if (likely(page->mapping != NULL)) {
 218                 ll_truncate_complete_page(page);
 219                 unlock_page(page);
 220         } else {
 221                 unlock_page(page);
 222                 CWARN("NULL mapping page %p, truncated by others: "
 223                       "hash("LPX64") | start("LPX64") | end("LPX64")\n",
 224                       page, hash, start, end);
 225         }
 226         page_cache_release(page);
 227 }
 228
 229 /*
 230  * Find, kmap and return page that contains given hash.
 231  */
 232 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
 233                                        __u64 *start, __u64 *end)
 234 {
 235         struct address_space *mapping = dir->i_mapping;
 236         /*
 237          * Complement of hash is used as an index so that
 238          * radix_tree_gang_lookup() can be used to find a page with starting
 239          * hash _smaller_ than one we are looking for.
 240          */
 241         unsigned long offset = hash_x_index(*hash);
 242         struct page *page;
 243         int found;
 244
 245         TREE_READ_LOCK_IRQ(mapping);
 246         found = radix_tree_gang_lookup(&mapping->page_tree,
 247                                        (void **)&page, offset, 1);
 248         if (found > 0) {
 249                 struct lu_dirpage *dp;
 250
 251                 page_cache_get(page);
 252                 TREE_READ_UNLOCK_IRQ(mapping);
 253                 /*
 254                  * In contrast to find_lock_page() we are sure that directory
 255                  * page cannot be truncated (while DLM lock is held) and,
 256                  * hence, can avoid restart.
 257                  *
 258                  * In fact, page cannot be locked here at all, because
 259                  * ll_dir_readpage() does synchronous io.
 260                  */
 261                 wait_on_page(page);
 262                 if (PageUptodate(page)) {
 263                         dp = kmap(page);
 264 #if BITS_PER_LONG == 32
 265                         *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 266                         *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 267                         *hash  = *hash >> 32;
 268 #else
 269                         *start = le64_to_cpu(dp->ldp_hash_start);
 270                         *end   = le64_to_cpu(dp->ldp_hash_end);
 271 #endif
 272                         LASSERTF(*start <= *hash, "start = "LPX64",end = "
 273                                  LPX64",hash = "LPX64"\n", *start, *end, *hash);
 274                         if (*hash > *end || (*end != *start && *hash == *end)) {
 275                                 ll_release_page(page, *hash, *start, *end);
 276                                 page = NULL;
 277                         }
 278                 } else {
 279                         page_cache_release(page);
 280                         page = ERR_PTR(-EIO);
 281                 }
 282
 283         } else {
 284                 TREE_READ_UNLOCK_IRQ(mapping);
 285                 page = NULL;
 286         }
 287         return page;
 288 }
 289
 290 struct page *ll_get_dir_page(struct file *filp, struct inode *dir, __u64 hash,
 291                              int exact, struct ll_dir_chain *chain)
 292 {
 293         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 294         struct address_space *mapping = dir->i_mapping;
 295         struct lustre_handle lockh;
 296         struct lu_dirpage *dp;
 297         struct page *page;
 298         ldlm_mode_t mode;
 299         int rc;
 300         __u64 start = 0;
 301         __u64 end = 0;
 302         __u64 lhash = hash;
 303         struct ll_inode_info *lli = ll_i2info(dir);
 304
 305         mode = LCK_PR;
 306         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 307                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
 308         if (!rc) {
 309                 struct ldlm_enqueue_info einfo = { LDLM_IBITS, mode,
 310                        ll_md_blocking_ast, ldlm_completion_ast,
 311                        NULL, NULL, dir };
 312                 struct lookup_intent it = { .it_op = IT_READDIR };
 313                 struct ptlrpc_request *request;
 314                 struct md_op_data *op_data;
 315
 316                 op_data = ll_prep_md_op_data(NULL, dir, NULL, NULL, 0, 0,
 317                                              LUSTRE_OPC_ANY, NULL);
 318                 if (IS_ERR(op_data))
 319                         return (void *)op_data;
 320
 321                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
 322                                 op_data, &lockh, NULL, 0, NULL, 0);
 323
 324                 ll_finish_md_op_data(op_data);
 325
 326                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 327                 if (request)
 328                         ptlrpc_req_finished(request);
 329                 if (rc < 0) {
 330                         CERROR("lock enqueue: "DFID" at "LPU64": rc %d\n",
 331                                PFID(ll_inode2fid(dir)), hash, rc);
 332                         return ERR_PTR(rc);
 333                 }
 334         } else {
 335                 /* for cross-ref object, l_ast_data of the lock may not be set,
 336                  * we reset it here */
 337                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
 338                                  dir, NULL);
 339         }
 340         ldlm_lock_dump_handle(D_OTHER, &lockh);
 341
 342         cfs_down(&lli->lli_readdir_sem);
 343         page = ll_dir_page_locate(dir, &lhash, &start, &end);
 344         if (IS_ERR(page)) {
 345                 CERROR("dir page locate: "DFID" at "LPU64": rc %ld\n",
 346                        PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
 347                 GOTO(out_unlock, page);
 348         }
 349
 350         if (page != NULL) {
 351                 /*
 352                  * XXX nikita: not entirely correct handling of a corner case:
 353                  * suppose hash chain of entries with hash value HASH crosses
 354                  * border between pages P0 and P1. First both P0 and P1 are
 355                  * cached, seekdir() is called for some entry from the P0 part
 356                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 357                  * happens and finds P1, as it starts with matching hash
 358                  * value. Remaining entries from P0 part of the chain are
 359                  * skipped. (Is that really a bug?)
 360                  *
 361                  * Possible solutions: 0. don't cache P1 is such case, handle
 362                  * it as an "overflow" page. 1. invalidate all pages at
 363                  * once. 2. use HASH|1 as an index for P1.
 364                  */
 365                 if (exact && lhash != start) {
 366                         /*
 367                          * readdir asked for a page starting _exactly_ from
 368                          * given hash, but cache contains stale page, with
 369                          * entries with smaller hash values. Stale page should
 370                          * be invalidated, and new one fetched.
 371                          */
 372                         CDEBUG(D_OTHER, "Stale readpage page %p: "
 373                                "start = "LPX64",end = "LPX64"hash ="LPX64"\n",
 374                                page, start, end, lhash);
 375                         ll_release_page(page, lhash, start, end);
 376                 } else {
 377                         GOTO(hash_collision, page);
 378                 }
 379         }
 380
 381         page = read_cache_page(mapping, hash_x_index(hash),
 382                                (filler_t*)mapping->a_ops->readpage, filp);
 383         if (IS_ERR(page)) {
 384                 CERROR("read cache page: "DFID" at "LPU64": rc %ld\n",
 385                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 386                 GOTO(out_unlock, page);
 387         }
 388
 389         wait_on_page(page);
 390         (void)kmap(page);
 391         if (!PageUptodate(page)) {
 392                 CERROR("page not updated: "DFID" at "LPU64": rc %d\n",
 393                        PFID(ll_inode2fid(dir)), hash, -5);
 394                 goto fail;
 395         }
 396         if (!PageChecked(page))
 397                 ll_check_page(dir, page);
 398         if (PageError(page)) {
 399                 CERROR("page error: "DFID" at "LPU64": rc %d\n",
 400                        PFID(ll_inode2fid(dir)), hash, -5);
 401                 goto fail;
 402         }
 403 hash_collision:
 404         dp = page_address(page);
 405 #if BITS_PER_LONG == 32
 406         start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 407         end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 408         lhash = hash >> 32;
 409 #else
 410         start = le64_to_cpu(dp->ldp_hash_start);
 411         end   = le64_to_cpu(dp->ldp_hash_end);
 412         lhash = hash;
 413 #endif
 414         if (end == start) {
 415                 LASSERT(start == lhash);
 416                 CWARN("Page-wide hash collision: "LPU64"\n", end);
 417 #if BITS_PER_LONG == 32
 418                 CWARN("Real page-wide hash collision at ["LPU64" "LPU64"] with "
 419                       "hash "LPU64"\n", le64_to_cpu(dp->ldp_hash_start),
 420                       le64_to_cpu(dp->ldp_hash_end), hash);
 421 #endif
 422                 /*
 423                  * Fetch whole overflow chain...
 424                  *
 425                  * XXX not yet.
 426                  */
 427                 goto fail;
 428         }
 429 out_unlock:
 430         cfs_up(&lli->lli_readdir_sem);
 431         ldlm_lock_decref(&lockh, mode);
 432         return page;
 433
 434 fail:
 435         ll_put_page(page);
 436         page = ERR_PTR(-EIO);
 437         goto out_unlock;
 438 }
 439
 440 int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 441 {
 442         struct inode         *inode      = filp->f_dentry->d_inode;
 443         struct ll_inode_info *info       = ll_i2info(inode);
 444         struct ll_sb_info    *sbi        = ll_i2sbi(inode);
 445         struct ll_file_data  *fd         = LUSTRE_FPRIVATE(filp);
 446         __u64                 pos        = fd->fd_dir.lfd_pos;
 447         int                   need_32bit = ll_need_32bit_api(sbi);
 448         struct page          *page;
 449         struct ll_dir_chain   chain;
 450         int                   done;
 451         int                   shift;
 452         int                   rc;
 453         ENTRY;
 454
 455         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu 32bit_api %d\n",
 456                inode->i_ino, inode->i_generation, inode,
 457                (unsigned long)pos, i_size_read(inode), need_32bit);
 458
 459         if (pos == DIR_END_OFF)
 460                 /*
 461                  * end-of-file.
 462                  */
 463                 RETURN(0);
 464
 465         rc    = 0;
 466         done  = 0;
 467         shift = 0;
 468         ll_dir_chain_init(&chain);
 469
 470         fd->fd_dir.lfd_next = pos;
 471         page = ll_get_dir_page(filp, inode, pos, 0, &chain);
 472
 473         while (rc == 0 && !done) {
 474                 struct lu_dirpage *dp;
 475                 struct lu_dirent  *ent;
 476
 477                 if (!IS_ERR(page)) {
 478                         /*
 479                          * If page is empty (end of directory is reached),
 480                          * use this value.
 481                          */
 482                         __u64 hash = DIR_END_OFF;
 483                         __u64 next;
 484
 485                         dp = page_address(page);
 486                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 487                              ent = lu_dirent_next(ent)) {
 488                                 __u16          type;
 489                                 int            namelen;
 490                                 struct lu_fid  fid;
 491                                 __u64          lhash;
 492                                 __u64          ino;
 493
 494                                 /*
 495                                  * XXX: implement correct swabbing here.
 496                                  */
 497
 498                                 hash = le64_to_cpu(ent->lde_hash);
 499                                 if (hash < pos)
 500                                         /*
 501                                          * Skip until we find target hash
 502                                          * value.
 503                                          */
 504                                         continue;
 505
 506                                 namelen = le16_to_cpu(ent->lde_namelen);
 507                                 if (namelen == 0)
 508                                         /*
 509                                          * Skip dummy record.
 510                                          */
 511                                         continue;
 512
 513                                 if (need_32bit)
 514                                         lhash = hash >> 32;
 515                                 else
 516                                         lhash = hash;
 517                                 fid_le_to_cpu(&fid, &ent->lde_fid);
 518                                 ino = cl_fid_build_ino(&fid,need_32bit);
 519                                 type = ll_dirent_type_get(ent);
 520                                 /* For 'll_nfs_get_name_filldir()', it will try
 521                                  * to access the 'ent' through its 'lde_name',
 522                                  * so the parameter 'name' for 'filldir()' must
 523                                  * be part of the 'ent'. */
 524                                 done = filldir(cookie, ent->lde_name, namelen,
 525                                                lhash, ino, type);
 526                         }
 527                         next = le64_to_cpu(dp->ldp_hash_end);
 528                         ll_put_page(page);
 529                         if (!done) {
 530                                 pos = next;
 531                                 if (pos == DIR_END_OFF) {
 532                                         /*
 533                                          * End of directory reached.
 534                                          */
 535                                         done = 1;
 536                                 } else if (1 /* chain is exhausted*/) {
 537                                         /*
 538                                          * Normal case: continue to the next
 539                                          * page.
 540                                          */
 541                                         fd->fd_dir.lfd_next = pos;
 542                                         page = ll_get_dir_page(filp, inode, pos,
 543                                                                1, &chain);
 544                                 } else {
 545                                         /*
 546                                          * go into overflow page.
 547                                          */
 548                                 }
 549                         } else {
 550                                 pos = hash;
 551                         }
 552                 } else {
 553                         rc = PTR_ERR(page);
 554                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 555                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 556                 }
 557         }
 558
 559         fd->fd_dir.lfd_pos = pos;
 560         if (need_32bit) {
 561                 if (pos == DIR_END_OFF)
 562                         filp->f_pos = DIR_END_OFF_32BIT;
 563                 else
 564                         filp->f_pos = pos >> 32;
 565         } else {
 566                 filp->f_pos = pos;
 567         }
 568         filp->f_version = inode->i_version;
 569         touch_atime(filp->f_vfsmnt, filp->f_dentry);
 570
 571         ll_dir_chain_fini(&chain);
 572
 573         RETURN(rc);
 574 }
 575
 576 int ll_send_mgc_param(struct obd_export *mgc, char *string)
 577 {
 578         struct mgs_send_param *msp;
 579         int rc = 0;
 580
 581         OBD_ALLOC_PTR(msp);
 582         if (!msp)
 583                 return -ENOMEM;
 584
 585         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 586         rc = obd_set_info_async(mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 587                                 sizeof(struct mgs_send_param), msp, NULL);
 588         if (rc)
 589                 CERROR("Failed to set parameter: %d\n", rc);
 590         OBD_FREE_PTR(msp);
 591
 592         return rc;
 593 }
 594
 595 char *ll_get_fsname(struct inode *inode)
 596 {
 597         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 598         char *ptr, *fsname;
 599         int len;
 600
 601         OBD_ALLOC(fsname, MGS_PARAM_MAXLEN);
 602         len = strlen(lsi->lsi_lmd->lmd_profile);
 603         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
 604         if (ptr && (strcmp(ptr, "-client") == 0))
 605                 len -= 7;
 606         strncpy(fsname, lsi->lsi_lmd->lmd_profile, len);
 607         fsname[len] = '\0';
 608
 609         return fsname;
 610 }
 611
 612 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 613                      int set_default)
 614 {
 615         struct ll_sb_info *sbi = ll_i2sbi(inode);
 616         struct md_op_data *op_data;
 617         struct ptlrpc_request *req = NULL;
 618         int rc = 0;
 619         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 620         struct obd_device *mgc = lsi->lsi_mgc;
 621         char *fsname = NULL, *param = NULL;
 622         int lum_size;
 623
 624         if (lump != NULL) {
 625                 /*
 626                  * This is coming from userspace, so should be in
 627                  * local endian.  But the MDS would like it in little
 628                  * endian, so we swab it before we send it.
 629                  */
 630                 switch (lump->lmm_magic) {
 631                 case LOV_USER_MAGIC_V1: {
 632                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 633                                 lustre_swab_lov_user_md_v1(lump);
 634                         lum_size = sizeof(struct lov_user_md_v1);
 635                         break;
 636                         }
 637                 case LOV_USER_MAGIC_V3: {
 638                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 639                                 lustre_swab_lov_user_md_v3(
 640                                         (struct lov_user_md_v3 *)lump);
 641                         lum_size = sizeof(struct lov_user_md_v3);
 642                         break;
 643                         }
 644                 default: {
 645                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 646                                         " %#08x != %#08x nor %#08x\n",
 647                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 648                                         LOV_USER_MAGIC_V3);
 649                         RETURN(-EINVAL);
 650                         }
 651                }
 652         } else {
 653                 lum_size = sizeof(struct lov_user_md_v1);
 654         }
 655
 656         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 657                                      LUSTRE_OPC_ANY, NULL);
 658         if (IS_ERR(op_data))
 659                 RETURN(PTR_ERR(op_data));
 660
 661         /* swabbing is done in lov_setstripe() on server side */
 662         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 663                         NULL, 0, &req, NULL);
 664         ll_finish_md_op_data(op_data);
 665         ptlrpc_req_finished(req);
 666         if (rc) {
 667                 if (rc != -EPERM && rc != -EACCES)
 668                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 669         }
 670
 671         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 672          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 673          need the make the distiction between the 2 versions */
 674         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 675                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 676
 677                 /* Get fsname and assume devname to be -MDT0000. */
 678                 fsname = ll_get_fsname(inode);
 679                 /* Set root stripesize */
 680                 sprintf(param, "%s-MDT0000.lov.stripesize=%u", fsname,
 681                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 682                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 683                 if (rc)
 684                         goto end;
 685
 686                 /* Set root stripecount */
 687                 sprintf(param, "%s-MDT0000.lov.stripecount=%hd", fsname,
 688                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 689                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 690                 if (rc)
 691                         goto end;
 692
 693                 /* Set root stripeoffset */
 694                 sprintf(param, "%s-MDT0000.lov.stripeoffset=%hd", fsname,
 695                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 696                         (typeof(lump->lmm_stripe_offset))(-1));
 697                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 698                 if (rc)
 699                         goto end;
 700 end:
 701                 if (fsname)
 702                         OBD_FREE(fsname, MGS_PARAM_MAXLEN);
 703                 if (param)
 704                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 705         }
 706         return rc;
 707 }
 708
 709 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 710                      int *lmm_size, struct ptlrpc_request **request)
 711 {
 712         struct ll_sb_info *sbi = ll_i2sbi(inode);
 713         struct mdt_body   *body;
 714         struct lov_mds_md *lmm = NULL;
 715         struct ptlrpc_request *req = NULL;
 716         int rc, lmmsize;
 717         struct md_op_data *op_data;
 718
 719         rc = ll_get_max_mdsize(sbi, &lmmsize);
 720         if (rc)
 721                 RETURN(rc);
 722
 723         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 724                                      0, lmmsize, LUSTRE_OPC_ANY,
 725                                      NULL);
 726         if (op_data == NULL)
 727                 RETURN(-ENOMEM);
 728
 729         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 730         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 731         ll_finish_md_op_data(op_data);
 732         if (rc < 0) {
 733                 CDEBUG(D_INFO, "md_getattr failed on inode "
 734                        "%lu/%u: rc %d\n", inode->i_ino,
 735                        inode->i_generation, rc);
 736                 GOTO(out, rc);
 737         }
 738
 739         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 740         LASSERT(body != NULL);
 741
 742         lmmsize = body->eadatasize;
 743
 744         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 745             lmmsize == 0) {
 746                 GOTO(out, rc = -ENODATA);
 747         }
 748
 749         lmm = req_capsule_server_sized_get(&req->rq_pill,
 750                                            &RMF_MDT_MD, lmmsize);
 751         LASSERT(lmm != NULL);
 752
 753         /*
 754          * This is coming from the MDS, so is probably in
 755          * little endian.  We convert it to host endian before
 756          * passing it to userspace.
 757          */
 758         /* We don't swab objects for directories */
 759         switch (le32_to_cpu(lmm->lmm_magic)) {
 760         case LOV_MAGIC_V1:
 761                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 762                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 763                 break;
 764         case LOV_MAGIC_V3:
 765                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 766                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 767                 break;
 768         default:
 769                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 770                 rc = -EPROTO;
 771         }
 772 out:
 773         *lmmp = lmm;
 774         *lmm_size = lmmsize;
 775         *request = req;
 776         return rc;
 777 }
 778
 779 /*
 780  *  Get MDT index for the inode.
 781  */
 782 int ll_get_mdt_idx(struct inode *inode)
 783 {
 784         struct ll_sb_info *sbi = ll_i2sbi(inode);
 785         struct md_op_data *op_data;
 786         int rc, mdtidx;
 787         ENTRY;
 788
 789         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 790                                      0, LUSTRE_OPC_ANY, NULL);
 791         if (op_data == NULL)
 792                 RETURN(-ENOMEM);
 793
 794         op_data->op_valid |= OBD_MD_MDTIDX;
 795         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 796         mdtidx = op_data->op_mds;
 797         ll_finish_md_op_data(op_data);
 798         if (rc < 0) {
 799                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 800                 RETURN(rc);
 801         }
 802         return mdtidx;
 803 }
 804
 805 static int copy_and_ioctl(int cmd, struct obd_export *exp, void *data, int len)
 806 {
 807         void *ptr;
 808         int rc;
 809
 810         OBD_ALLOC(ptr, len);
 811         if (ptr == NULL)
 812                 return -ENOMEM;
 813         if (cfs_copy_from_user(ptr, data, len)) {
 814                 OBD_FREE(ptr, len);
 815                 return -EFAULT;
 816         }
 817         rc = obd_iocontrol(cmd, exp, len, data, NULL);
 818         OBD_FREE(ptr, len);
 819         return rc;
 820 }
 821
 822 static int ll_dir_ioctl(struct inode *inode, struct file *file,
 823                         unsigned int cmd, unsigned long arg)
 824 {
 825         struct ll_sb_info *sbi = ll_i2sbi(inode);
 826         struct obd_ioctl_data *data;
 827         int rc = 0;
 828         ENTRY;
 829
 830         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
 831                inode->i_ino, inode->i_generation, inode, cmd);
 832
 833         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
 834         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
 835                 return -ENOTTY;
 836
 837         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
 838         switch(cmd) {
 839         case FSFILT_IOC_GETFLAGS:
 840         case FSFILT_IOC_SETFLAGS:
 841                 RETURN(ll_iocontrol(inode, file, cmd, arg));
 842         case FSFILT_IOC_GETVERSION_OLD:
 843         case FSFILT_IOC_GETVERSION:
 844                 RETURN(put_user(inode->i_generation, (int *)arg));
 845         /* We need to special case any other ioctls we want to handle,
 846          * to send them to the MDS/OST as appropriate and to properly
 847          * network encode the arg field.
 848         case FSFILT_IOC_SETVERSION_OLD:
 849         case FSFILT_IOC_SETVERSION:
 850         */
 851         case LL_IOC_GET_MDTIDX: {
 852                 int mdtidx;
 853
 854                 mdtidx = ll_get_mdt_idx(inode);
 855                 if (mdtidx < 0)
 856                         RETURN(mdtidx);
 857
 858                 if (put_user((int)mdtidx, (int*)arg))
 859                         RETURN(-EFAULT);
 860
 861                 return 0;
 862         }
 863         case IOC_MDC_LOOKUP: {
 864                 struct ptlrpc_request *request = NULL;
 865                 int namelen, len = 0;
 866                 char *buf = NULL;
 867                 char *filename;
 868                 struct md_op_data *op_data;
 869
 870                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 871                 if (rc)
 872                         RETURN(rc);
 873                 data = (void *)buf;
 874
 875                 filename = data->ioc_inlbuf1;
 876                 namelen = strlen(filename);
 877
 878                 if (namelen < 1) {
 879                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
 880                         GOTO(out_free, rc = -EINVAL);
 881                 }
 882
 883                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
 884                                              0, LUSTRE_OPC_ANY, NULL);
 885                 if (op_data == NULL)
 886                         GOTO(out_free, rc = -ENOMEM);
 887
 888                 op_data->op_valid = OBD_MD_FLID;
 889                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
 890                 ll_finish_md_op_data(op_data);
 891                 if (rc < 0) {
 892                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 893                         GOTO(out_free, rc);
 894                 }
 895                 ptlrpc_req_finished(request);
 896                 EXIT;
 897 out_free:
 898                 obd_ioctl_freedata(buf, len);
 899                 return rc;
 900         }
 901         case LL_IOC_LOV_SETSTRIPE: {
 902                 struct lov_user_md_v3 lumv3;
 903                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
 904                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
 905                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
 906
 907                 int set_default = 0;
 908
 909                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
 910                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
 911                         sizeof(lumv3p->lmm_objects[0]));
 912                 /* first try with v1 which is smaller than v3 */
 913                 if (cfs_copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
 914                         RETURN(-EFAULT);
 915
 916                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
 917                         if (cfs_copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
 918                                 RETURN(-EFAULT);
 919                 }
 920
 921                 if (inode->i_sb->s_root == file->f_dentry)
 922                         set_default = 1;
 923
 924                 /* in v1 and v3 cases lumv1 points to data */
 925                 rc = ll_dir_setstripe(inode, lumv1, set_default);
 926
 927                 RETURN(rc);
 928         }
 929         case LL_IOC_OBD_STATFS:
 930                 RETURN(ll_obd_statfs(inode, (void *)arg));
 931         case LL_IOC_LOV_GETSTRIPE:
 932         case LL_IOC_MDC_GETINFO:
 933         case IOC_MDC_GETFILEINFO:
 934         case IOC_MDC_GETFILESTRIPE: {
 935                 struct ptlrpc_request *request = NULL;
 936                 struct lov_user_md *lump;
 937                 struct lov_mds_md *lmm = NULL;
 938                 struct mdt_body *body;
 939                 char *filename = NULL;
 940                 int lmmsize;
 941
 942                 if (cmd == IOC_MDC_GETFILEINFO ||
 943                     cmd == IOC_MDC_GETFILESTRIPE) {
 944                         filename = getname((const char *)arg);
 945                         if (IS_ERR(filename))
 946                                 RETURN(PTR_ERR(filename));
 947
 948                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
 949                                                       &lmmsize, &request);
 950                 } else {
 951                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
 952                 }
 953
 954                 if (request) {
 955                         body = req_capsule_server_get(&request->rq_pill,
 956                                                       &RMF_MDT_BODY);
 957                         LASSERT(body != NULL);
 958                 } else {
 959                         GOTO(out_req, rc);
 960                 }
 961
 962                 if (rc < 0) {
 963                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
 964                                                cmd == LL_IOC_MDC_GETINFO))
 965                                 GOTO(skip_lmm, rc = 0);
 966                         else
 967                                 GOTO(out_req, rc);
 968                 }
 969
 970                 if (cmd == IOC_MDC_GETFILESTRIPE ||
 971                     cmd == LL_IOC_LOV_GETSTRIPE) {
 972                         lump = (struct lov_user_md *)arg;
 973                 } else {
 974                         struct lov_user_mds_data *lmdp;
 975                         lmdp = (struct lov_user_mds_data *)arg;
 976                         lump = &lmdp->lmd_lmm;
 977                 }
 978                 if (cfs_copy_to_user(lump, lmm, lmmsize)) {
 979                         if (cfs_copy_to_user(lump, lmm, sizeof(*lump)))
 980                                 GOTO(out_req, rc = -EFAULT);
 981                         rc = -EOVERFLOW;
 982                 }
 983         skip_lmm:
 984                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
 985                         struct lov_user_mds_data *lmdp;
 986                         lstat_t st = { 0 };
 987
 988                         st.st_dev     = inode->i_sb->s_dev;
 989                         st.st_mode    = body->mode;
 990                         st.st_nlink   = body->nlink;
 991                         st.st_uid     = body->uid;
 992                         st.st_gid     = body->gid;
 993                         st.st_rdev    = body->rdev;
 994                         st.st_size    = body->size;
 995                         st.st_blksize = CFS_PAGE_SIZE;
 996                         st.st_blocks  = body->blocks;
 997                         st.st_atime   = body->atime;
 998                         st.st_mtime   = body->mtime;
 999                         st.st_ctime   = body->ctime;
1000                         st.st_ino     = inode->i_ino;
1001
1002                         lmdp = (struct lov_user_mds_data *)arg;
1003                         if (cfs_copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1004                                 GOTO(out_req, rc = -EFAULT);
1005                 }
1006
1007                 EXIT;
1008         out_req:
1009                 ptlrpc_req_finished(request);
1010                 if (filename)
1011                         putname(filename);
1012                 return rc;
1013         }
1014         case IOC_LOV_GETINFO: {
1015                 struct lov_user_mds_data *lumd;
1016                 struct lov_stripe_md *lsm;
1017                 struct lov_user_md *lum;
1018                 struct lov_mds_md *lmm;
1019                 int lmmsize;
1020                 lstat_t st;
1021
1022                 lumd = (struct lov_user_mds_data *)arg;
1023                 lum = &lumd->lmd_lmm;
1024
1025                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1026                 if (rc)
1027                         RETURN(rc);
1028
1029                 OBD_ALLOC_LARGE(lmm, lmmsize);
1030                 if (cfs_copy_from_user(lmm, lum, lmmsize))
1031                         GOTO(free_lmm, rc = -EFAULT);
1032
1033                 switch (lmm->lmm_magic) {
1034                 case LOV_USER_MAGIC_V1:
1035                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1036                                 break;
1037                         /* swab objects first so that stripes num will be sane */
1038                         lustre_swab_lov_user_md_objects(
1039                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1040                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1041                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1042                         break;
1043                 case LOV_USER_MAGIC_V3:
1044                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1045                                 break;
1046                         /* swab objects first so that stripes num will be sane */
1047                         lustre_swab_lov_user_md_objects(
1048                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1049                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1050                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1051                         break;
1052                 default:
1053                         GOTO(free_lmm, rc = -EINVAL);
1054                 }
1055
1056                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1057                 if (rc < 0)
1058                         GOTO(free_lmm, rc = -ENOMEM);
1059
1060                 /* Perform glimpse_size operation. */
1061                 memset(&st, 0, sizeof(st));
1062
1063                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1064                 if (rc)
1065                         GOTO(free_lsm, rc);
1066
1067                 if (cfs_copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1068                         GOTO(free_lsm, rc = -EFAULT);
1069
1070                 EXIT;
1071         free_lsm:
1072                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1073         free_lmm:
1074                 OBD_FREE_LARGE(lmm, lmmsize);
1075                 return rc;
1076         }
1077         case OBD_IOC_LLOG_CATINFO: {
1078                 struct ptlrpc_request *req = NULL;
1079                 char                  *buf = NULL;
1080                 char                  *str;
1081                 int                    len = 0;
1082
1083                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1084                 if (rc)
1085                         RETURN(rc);
1086                 data = (void *)buf;
1087
1088                 if (!data->ioc_inlbuf1) {
1089                         obd_ioctl_freedata(buf, len);
1090                         RETURN(-EINVAL);
1091                 }
1092
1093                 req = ptlrpc_request_alloc(sbi2mdc(sbi)->cl_import,
1094                                            &RQF_LLOG_CATINFO);
1095                 if (req == NULL)
1096                         GOTO(out_catinfo, rc = -ENOMEM);
1097
1098                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
1099                                      data->ioc_inllen1);
1100                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_CLIENT,
1101                                      data->ioc_inllen2);
1102
1103                 rc = ptlrpc_request_pack(req, LUSTRE_LOG_VERSION, LLOG_CATINFO);
1104                 if (rc) {
1105                         ptlrpc_request_free(req);
1106                         GOTO(out_catinfo, rc);
1107                 }
1108
1109                 str = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
1110                 memcpy(str, data->ioc_inlbuf1, data->ioc_inllen1);
1111                 if (data->ioc_inllen2) {
1112                         str = req_capsule_client_get(&req->rq_pill,
1113                                                      &RMF_STRING);
1114                         memcpy(str, data->ioc_inlbuf2, data->ioc_inllen2);
1115                 }
1116
1117                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_SERVER,
1118                                      data->ioc_plen1);
1119                 ptlrpc_request_set_replen(req);
1120
1121                 rc = ptlrpc_queue_wait(req);
1122                 if (!rc) {
1123                         str = req_capsule_server_get(&req->rq_pill,
1124                                                      &RMF_STRING);
1125                         if (cfs_copy_to_user(data->ioc_pbuf1, str,
1126                                              data->ioc_plen1))
1127                                 rc = -EFAULT;
1128                 }
1129                 ptlrpc_req_finished(req);
1130         out_catinfo:
1131                 obd_ioctl_freedata(buf, len);
1132                 RETURN(rc);
1133         }
1134         case OBD_IOC_QUOTACHECK: {
1135                 struct obd_quotactl *oqctl;
1136                 int error = 0;
1137
1138                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1139                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1140                         RETURN(-EPERM);
1141
1142                 OBD_ALLOC_PTR(oqctl);
1143                 if (!oqctl)
1144                         RETURN(-ENOMEM);
1145                 oqctl->qc_type = arg;
1146                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1147                 if (rc < 0) {
1148                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1149                         error = rc;
1150                 }
1151
1152                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1153                 if (rc < 0)
1154                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1155
1156                 OBD_FREE_PTR(oqctl);
1157                 return error ?: rc;
1158         }
1159         case OBD_IOC_POLL_QUOTACHECK: {
1160                 struct if_quotacheck *check;
1161
1162                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1163                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1164                         RETURN(-EPERM);
1165
1166                 OBD_ALLOC_PTR(check);
1167                 if (!check)
1168                         RETURN(-ENOMEM);
1169
1170                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1171                                    NULL);
1172                 if (rc) {
1173                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1174                         if (cfs_copy_to_user((void *)arg, check,
1175                                              sizeof(*check)))
1176                                 CDEBUG(D_QUOTA, "cfs_copy_to_user failed\n");
1177                         GOTO(out_poll, rc);
1178                 }
1179
1180                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1181                                    NULL);
1182                 if (rc) {
1183                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1184                         if (cfs_copy_to_user((void *)arg, check,
1185                                              sizeof(*check)))
1186                                 CDEBUG(D_QUOTA, "cfs_copy_to_user failed\n");
1187                         GOTO(out_poll, rc);
1188                 }
1189         out_poll:
1190                 OBD_FREE_PTR(check);
1191                 RETURN(rc);
1192         }
1193         case OBD_IOC_QUOTACTL: {
1194                 struct if_quotactl *qctl;
1195                 int cmd, type, id, valid;
1196
1197                 OBD_ALLOC_PTR(qctl);
1198                 if (!qctl)
1199                         RETURN(-ENOMEM);
1200
1201                 if (cfs_copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1202                         GOTO(out_quotactl, rc = -EFAULT);
1203
1204                 cmd = qctl->qc_cmd;
1205                 type = qctl->qc_type;
1206                 id = qctl->qc_id;
1207                 valid = qctl->qc_valid;
1208
1209                 switch (cmd) {
1210                 case LUSTRE_Q_INVALIDATE:
1211                 case LUSTRE_Q_FINVALIDATE:
1212                 case Q_QUOTAON:
1213                 case Q_QUOTAOFF:
1214                 case Q_SETQUOTA:
1215                 case Q_SETINFO:
1216                         if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1217                             sbi->ll_flags & LL_SBI_RMT_CLIENT)
1218                                 GOTO(out_quotactl, rc = -EPERM);
1219                         break;
1220                 case Q_GETQUOTA:
1221                         if (((type == USRQUOTA && cfs_curproc_euid() != id) ||
1222                              (type == GRPQUOTA && !in_egroup_p(id))) &&
1223                             (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1224                              sbi->ll_flags & LL_SBI_RMT_CLIENT))
1225                                 GOTO(out_quotactl, rc = -EPERM);
1226                         break;
1227                 case Q_GETINFO:
1228                         break;
1229                 default:
1230                         CERROR("unsupported quotactl op: %#x\n", cmd);
1231                         GOTO(out_quotactl, rc = -ENOTTY);
1232                 }
1233
1234                 if (valid != QC_GENERAL) {
1235                         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1236                                 GOTO(out_quotactl, rc = -EOPNOTSUPP);
1237
1238                         if (cmd == Q_GETINFO)
1239                                 qctl->qc_cmd = Q_GETOINFO;
1240                         else if (cmd == Q_GETQUOTA)
1241                                 qctl->qc_cmd = Q_GETOQUOTA;
1242                         else
1243                                 GOTO(out_quotactl, rc = -EINVAL);
1244
1245                         switch (valid) {
1246                         case QC_MDTIDX:
1247                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1248                                                    sbi->ll_md_exp,
1249                                                    sizeof(*qctl), qctl, NULL);
1250                                 break;
1251                         case QC_OSTIDX:
1252                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1253                                                    sbi->ll_dt_exp,
1254                                                    sizeof(*qctl), qctl, NULL);
1255                                 break;
1256                         case QC_UUID:
1257                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1258                                                    sbi->ll_md_exp,
1259                                                    sizeof(*qctl), qctl, NULL);
1260                                 if (rc == -EAGAIN)
1261                                         rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1262                                                            sbi->ll_dt_exp,
1263                                                            sizeof(*qctl), qctl,
1264                                                            NULL);
1265                                 break;
1266                         default:
1267                                 rc = -EINVAL;
1268                                 break;
1269                         }
1270
1271                         if (rc)
1272                                 GOTO(out_quotactl, rc);
1273                         else
1274                                 qctl->qc_cmd = cmd;
1275                 } else {
1276                         struct obd_quotactl *oqctl;
1277
1278                         OBD_ALLOC_PTR(oqctl);
1279                         if (!oqctl)
1280                                 GOTO(out_quotactl, rc = -ENOMEM);
1281
1282                         QCTL_COPY(oqctl, qctl);
1283                         rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1284                         if (rc) {
1285                                 if (rc != -EALREADY && cmd == Q_QUOTAON) {
1286                                         oqctl->qc_cmd = Q_QUOTAOFF;
1287                                         obd_quotactl(sbi->ll_md_exp, oqctl);
1288                                 }
1289                                 OBD_FREE_PTR(oqctl);
1290                                 GOTO(out_quotactl, rc);
1291                         } else {
1292                                 QCTL_COPY(qctl, oqctl);
1293                                 OBD_FREE_PTR(oqctl);
1294                         }
1295                 }
1296
1297                 if (cfs_copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1298                         rc = -EFAULT;
1299
1300         out_quotactl:
1301                 OBD_FREE_PTR(qctl);
1302                 RETURN(rc);
1303         }
1304         case OBD_IOC_GETNAME: {
1305                 struct obd_device *obd = class_exp2obd(sbi->ll_dt_exp);
1306                 if (!obd)
1307                         RETURN(-EFAULT);
1308                 if (cfs_copy_to_user((void *)arg, obd->obd_name,
1309                                      strlen(obd->obd_name) + 1))
1310                         RETURN (-EFAULT);
1311                 RETURN(0);
1312         }
1313         case LL_IOC_FLUSHCTX:
1314                 RETURN(ll_flush_ctx(inode));
1315 #ifdef CONFIG_FS_POSIX_ACL
1316         case LL_IOC_RMTACL: {
1317             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1318                 inode == inode->i_sb->s_root->d_inode) {
1319                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1320
1321                 LASSERT(fd != NULL);
1322                 rc = rct_add(&sbi->ll_rct, cfs_curproc_pid(), arg);
1323                 if (!rc)
1324                         fd->fd_flags |= LL_FILE_RMTACL;
1325                 RETURN(rc);
1326             } else
1327                 RETURN(0);
1328         }
1329 #endif
1330         case LL_IOC_GETOBDCOUNT: {
1331                 int count;
1332
1333                 if (cfs_copy_from_user(&count, (int *)arg, sizeof(int)))
1334                         RETURN(-EFAULT);
1335
1336                 if (!count) {
1337                         /* get ost count */
1338                         struct lov_obd *lov = &sbi->ll_dt_exp->exp_obd->u.lov;
1339                         count = lov->desc.ld_tgt_count;
1340                 } else {
1341                         /* get mdt count */
1342                         struct lmv_obd *lmv = &sbi->ll_md_exp->exp_obd->u.lmv;
1343                         count = lmv->desc.ld_tgt_count;
1344                 }
1345
1346                 if (cfs_copy_to_user((int *)arg, &count, sizeof(int)))
1347                         RETURN(-EFAULT);
1348
1349                 RETURN(0);
1350         }
1351         case LL_IOC_PATH2FID:
1352                 if (cfs_copy_to_user((void *)arg, ll_inode2fid(inode),
1353                                      sizeof(struct lu_fid)))
1354                         RETURN(-EFAULT);
1355                 RETURN(0);
1356         case LL_IOC_GET_CONNECT_FLAGS: {
1357                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1358         }
1359         case OBD_IOC_CHANGELOG_SEND:
1360         case OBD_IOC_CHANGELOG_CLEAR:
1361                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1362                                     sizeof(struct ioc_changelog));
1363                 RETURN(rc);
1364         case OBD_IOC_FID2PATH:
1365                 RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg));
1366         case LL_IOC_HSM_CT_START:
1367                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1368                                     sizeof(struct lustre_kernelcomm));
1369                 RETURN(rc);
1370
1371         default:
1372                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp,0,NULL,(void *)arg));
1373         }
1374 }
1375
1376 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1377 {
1378         struct inode *inode = file->f_mapping->host;
1379         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1380         int need_32bit = ll_need_32bit_api(ll_i2sbi(inode));
1381         loff_t ret = -EINVAL;
1382         ENTRY;
1383
1384         cfs_mutex_lock(&inode->i_mutex);
1385         switch (origin) {
1386                 case SEEK_SET:
1387                         break;
1388                 case SEEK_CUR:
1389                         offset += file->f_pos;
1390                         break;
1391                 case SEEK_END:
1392                         if (offset > 0)
1393                                 GOTO(out, ret);
1394                         if (need_32bit)
1395                                 offset += DIR_END_OFF_32BIT;
1396                         else
1397                                 offset += DIR_END_OFF;
1398                         break;
1399                 default:
1400                         GOTO(out, ret);
1401         }
1402
1403         if (offset >= 0 &&
1404             ((need_32bit && offset <= DIR_END_OFF_32BIT) || !need_32bit)) {
1405                 if (offset != file->f_pos) {
1406                         if (need_32bit) {
1407                                 if (offset == DIR_END_OFF_32BIT)
1408                                         fd->fd_dir.lfd_pos = DIR_END_OFF;
1409                                 else
1410                                         fd->fd_dir.lfd_pos = offset << 32;
1411                         } else {
1412                                 fd->fd_dir.lfd_pos = offset;
1413                         }
1414                         file->f_pos = offset;
1415                         file->f_version = 0;
1416                 }
1417                 ret = offset;
1418         }
1419         EXIT;
1420
1421 out:
1422         cfs_mutex_unlock(&inode->i_mutex);
1423         return ret;
1424 }
1425
1426 int ll_dir_open(struct inode *inode, struct file *file)
1427 {
1428         ENTRY;
1429         RETURN(ll_file_open(inode, file));
1430 }
1431
1432 int ll_dir_release(struct inode *inode, struct file *file)
1433 {
1434         ENTRY;
1435         RETURN(ll_file_release(inode, file));
1436 }
1437
1438 struct file_operations ll_dir_operations = {
1439         .llseek   = ll_dir_seek,
1440         .open     = ll_dir_open,
1441         .release  = ll_dir_release,
1442         .read     = generic_read_dir,
1443         .readdir  = ll_readdir,
1444         .ioctl    = ll_dir_ioctl
1445 };