lustre/llite/dir.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  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 <asm/uaccess.h>
  46 #include <linux/buffer_head.h>   // for wait_on_buffer
  47 #include <linux/pagevec.h>
  48
  49 #define DEBUG_SUBSYSTEM S_LLITE
  50
  51 #include <lustre/lustre_idl.h>
  52 #include <obd_support.h>
  53 #include <obd_class.h>
  54 #include <lustre_lib.h>
  55 #include <lustre/lustre_idl.h>
  56 #include <lustre_lite.h>
  57 #include <lustre_dlm.h>
  58 #include <lustre_fid.h>
  59 #include "llite_internal.h"
  60
  61 /*
  62  * (new) readdir implementation overview.
  63  *
  64  * Original lustre readdir implementation cached exact copy of raw directory
  65  * pages on the client. These pages were indexed in client page cache by
  66  * logical offset in the directory file. This design, while very simple and
  67  * intuitive had some inherent problems:
  68  *
  69  *     . it implies that byte offset to the directory entry serves as a
  70  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  71  *     ext3/htree directory entries may move due to splits, and more
  72  *     importantly,
  73  *
  74  *     . it is incompatible with the design of split directories for cmd3,
  75  *     that assumes that names are distributed across nodes based on their
  76  *     hash, and so readdir should be done in hash order.
  77  *
  78  * New readdir implementation does readdir in hash order, and uses hash of a
  79  * file name as a telldir/seekdir cookie. This led to number of complications:
  80  *
  81  *     . hash is not unique, so it cannot be used to index cached directory
  82  *     pages on the client (note, that it requires a whole pageful of hash
  83  *     collided entries to cause two pages to have identical hashes);
  84  *
  85  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  86  *     entry cookie. ext3/htree has the same problem and lustre implementation
  87  *     mimics their solution: seekdir(hash) positions directory at the first
  88  *     entry with the given hash.
  89  *
  90  * Client side.
  91  *
  92  * 0. caching
  93  *
  94  * Client caches directory pages using hash of the first entry as an index. As
  95  * noted above hash is not unique, so this solution doesn't work as is:
  96  * special processing is needed for "page hash chains" (i.e., sequences of
  97  * pages filled with entries all having the same hash value).
  98  *
  99  * First, such chains have to be detected. To this end, server returns to the
 100  * client the hash of the first entry on the page next to one returned. When
 101  * client detects that this hash is the same as hash of the first entry on the
 102  * returned page, page hash collision has to be handled. Pages in the
 103  * hash chain, except first one, are termed "overflow pages".
 104  *
 105  * Solution to index uniqueness problem is to not cache overflow
 106  * pages. Instead, when page hash collision is detected, all overflow pages
 107  * from emerging chain are immediately requested from the server and placed in
 108  * a special data structure (struct ll_dir_chain). This data structure is used
 109  * by ll_readdir() to process entries from overflow pages. When readdir
 110  * invocation finishes, overflow pages are discarded. If page hash collision
 111  * chain weren't completely processed, next call to readdir will again detect
 112  * page hash collision, again read overflow pages in, process next portion of
 113  * entries and again discard the pages. This is not as wasteful as it looks,
 114  * because, given reasonable hash, page hash collisions are extremely rare.
 115  *
 116  * 1. directory positioning
 117  *
 118  * When seekdir(hash) is called, original
 119  *
 120  *
 121  *
 122  *
 123  *
 124  *
 125  *
 126  *
 127  * Server.
 128  *
 129  * identification of and access to overflow pages
 130  *
 131  * page format
 132  *
 133  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 134  * a header lu_dirpage which describes the start/end hash, and whether this
 135  * page is empty (contains no dir entry) or hash collide with next page.
 136  * After client receives reply, several pages will be integrated into dir page
 137  * in CFS_PAGE_SIZE (if CFS_PAGE_SIZE greater than LU_PAGE_SIZE), and the
 138  * lu_dirpage for this integrated page will be adjusted.
 139  *
 140  */
 141
 142 /* returns the page unlocked, but with a reference */
 143 static int ll_dir_filler(void *_hash, struct page *page0)
 144 {
 145         struct inode *inode = page0->mapping->host;
 146         int hash64 = ll_i2sbi(inode)->ll_flags & LL_SBI_64BIT_HASH;
 147         struct obd_export *exp = ll_i2sbi(inode)->ll_md_exp;
 148         struct ptlrpc_request *request;
 149         struct mdt_body *body;
 150         struct md_op_data *op_data;
 151         __u64 hash = *((__u64 *)_hash);
 152         struct page **page_pool;
 153         struct page *page;
 154 #ifndef HAVE_ADD_TO_PAGE_CACHE_LRU
 155         struct pagevec lru_pvec;
 156 #endif
 157         struct lu_dirpage *dp;
 158         int max_pages = ll_i2sbi(inode)->ll_md_brw_size >> CFS_PAGE_SHIFT;
 159         int nrdpgs = 0; /* number of pages read actually */
 160         int npages;
 161         int i;
 162         int rc;
 163         ENTRY;
 164
 165         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) hash "LPU64"\n",
 166                inode->i_ino, inode->i_generation, inode, hash);
 167
 168         LASSERT(max_pages > 0 && max_pages <= MD_MAX_BRW_PAGES);
 169
 170         OBD_ALLOC(page_pool, sizeof(page) * max_pages);
 171         if (page_pool != NULL) {
 172                 page_pool[0] = page0;
 173         } else {
 174                 page_pool = &page0;
 175                 max_pages = 1;
 176         }
 177         for (npages = 1; npages < max_pages; npages++) {
 178                 page = page_cache_alloc_cold(inode->i_mapping);
 179                 if (!page)
 180                         break;
 181                 page_pool[npages] = page;
 182         }
 183
 184         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 185                                      LUSTRE_OPC_ANY, NULL);
 186         op_data->op_npages = npages;
 187         op_data->op_offset = hash;
 188         rc = md_readpage(exp, op_data, page_pool, &request);
 189         ll_finish_md_op_data(op_data);
 190         if (rc == 0) {
 191                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 192                 /* Checked by mdc_readpage() */
 193                 LASSERT(body != NULL);
 194
 195                 if (body->valid & OBD_MD_FLSIZE)
 196                         cl_isize_write(inode, body->size);
 197
 198                 nrdpgs = (request->rq_bulk->bd_nob_transferred+CFS_PAGE_SIZE-1)
 199                          >> CFS_PAGE_SHIFT;
 200                 SetPageUptodate(page0);
 201         }
 202         unlock_page(page0);
 203         ptlrpc_req_finished(request);
 204
 205         CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages);
 206
 207         ll_pagevec_init(&lru_pvec, 0);
 208         for (i = 1; i < npages; i++) {
 209                 unsigned long offset;
 210                 int ret;
 211
 212                 page = page_pool[i];
 213
 214                 if (rc < 0 || i >= nrdpgs) {
 215                         page_cache_release(page);
 216                         continue;
 217                 }
 218
 219                 SetPageUptodate(page);
 220
 221                 dp = cfs_kmap(page);
 222                 hash = le64_to_cpu(dp->ldp_hash_start);
 223                 cfs_kunmap(page);
 224
 225                 offset = hash_x_index(hash, hash64);
 226
 227                 prefetchw(&page->flags);
 228                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
 229                                             GFP_KERNEL);
 230                 if (ret == 0) {
 231                         unlock_page(page);
 232                         if (ll_pagevec_add(&lru_pvec, page) == 0)
 233                                 ll_pagevec_lru_add_file(&lru_pvec);
 234                 } else {
 235                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
 236                                " %d\n", offset, ret);
 237                 }
 238                 page_cache_release(page);
 239         }
 240         ll_pagevec_lru_add_file(&lru_pvec);
 241
 242         if (page_pool != &page0)
 243                 OBD_FREE(page_pool, sizeof(struct page *) * max_pages);
 244         EXIT;
 245         return rc;
 246 }
 247
 248 static void ll_check_page(struct inode *dir, struct page *page)
 249 {
 250         /* XXX: check page format later */
 251         SetPageChecked(page);
 252 }
 253
 254 void ll_release_page(struct page *page, int remove)
 255 {
 256         kunmap(page);
 257         if (remove) {
 258                 lock_page(page);
 259                 if (likely(page->mapping != NULL))
 260                         truncate_complete_page(page->mapping, page);
 261                 unlock_page(page);
 262         }
 263         page_cache_release(page);
 264 }
 265
 266 /*
 267  * Find, kmap and return page that contains given hash.
 268  */
 269 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
 270                                        __u64 *start, __u64 *end)
 271 {
 272         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 273         struct address_space *mapping = dir->i_mapping;
 274         /*
 275          * Complement of hash is used as an index so that
 276          * radix_tree_gang_lookup() can be used to find a page with starting
 277          * hash _smaller_ than one we are looking for.
 278          */
 279         unsigned long offset = hash_x_index(*hash, hash64);
 280         struct page *page;
 281         int found;
 282
 283         TREE_READ_LOCK_IRQ(mapping);
 284         found = radix_tree_gang_lookup(&mapping->page_tree,
 285                                        (void **)&page, offset, 1);
 286         if (found > 0) {
 287                 struct lu_dirpage *dp;
 288
 289                 page_cache_get(page);
 290                 TREE_READ_UNLOCK_IRQ(mapping);
 291                 /*
 292                  * In contrast to find_lock_page() we are sure that directory
 293                  * page cannot be truncated (while DLM lock is held) and,
 294                  * hence, can avoid restart.
 295                  *
 296                  * In fact, page cannot be locked here at all, because
 297                  * ll_dir_filler() does synchronous io.
 298                  */
 299                 wait_on_page(page);
 300                 if (PageUptodate(page)) {
 301                         dp = cfs_kmap(page);
 302                         if (BITS_PER_LONG == 32 && hash64) {
 303                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 304                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 305                                 *hash  = *hash >> 32;
 306                         } else {
 307                                 *start = le64_to_cpu(dp->ldp_hash_start);
 308                                 *end   = le64_to_cpu(dp->ldp_hash_end);
 309                         }
 310                         LASSERTF(*start <= *hash, "start = "LPX64",end = "
 311                                  LPX64",hash = "LPX64"\n", *start, *end, *hash);
 312                         CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash "LPU64"\n",
 313                                offset, *start, *end, *hash);
 314                         if (*hash > *end) {
 315                                 ll_release_page(page, 0);
 316                                 page = NULL;
 317                         } else if (*end != *start && *hash == *end) {
 318                                 /*
 319                                  * upon hash collision, remove this page,
 320                                  * otherwise put page reference, and
 321                                  * ll_get_dir_page() will issue RPC to fetch
 322                                  * the page we want.
 323                                  */
 324                                 ll_release_page(page,
 325                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
 326                                 page = NULL;
 327                         }
 328                 } else {
 329                         page_cache_release(page);
 330                         page = ERR_PTR(-EIO);
 331                 }
 332
 333         } else {
 334                 TREE_READ_UNLOCK_IRQ(mapping);
 335                 page = NULL;
 336         }
 337         return page;
 338 }
 339
 340 struct page *ll_get_dir_page(struct inode *dir, __u64 hash,
 341                              struct ll_dir_chain *chain)
 342 {
 343         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 344         struct address_space *mapping = dir->i_mapping;
 345         struct lustre_handle lockh;
 346         struct lu_dirpage *dp;
 347         struct page *page;
 348         ldlm_mode_t mode;
 349         int rc;
 350         __u64 start = 0;
 351         __u64 end = 0;
 352         __u64 lhash = hash;
 353         struct ll_inode_info *lli = ll_i2info(dir);
 354         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 355
 356         mode = LCK_PR;
 357         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 358                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
 359         if (!rc) {
 360                 struct ldlm_enqueue_info einfo = { LDLM_IBITS, mode,
 361                        ll_md_blocking_ast, ldlm_completion_ast,
 362                        NULL, NULL, dir };
 363                 struct lookup_intent it = { .it_op = IT_READDIR };
 364                 struct ptlrpc_request *request;
 365                 struct md_op_data *op_data;
 366
 367                 op_data = ll_prep_md_op_data(NULL, dir, NULL, NULL, 0, 0,
 368                                              LUSTRE_OPC_ANY, NULL);
 369                 if (IS_ERR(op_data))
 370                         return (void *)op_data;
 371
 372                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
 373                                 op_data, &lockh, NULL, 0, NULL, 0);
 374
 375                 ll_finish_md_op_data(op_data);
 376
 377                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 378                 if (request)
 379                         ptlrpc_req_finished(request);
 380                 if (rc < 0) {
 381                         CERROR("lock enqueue: "DFID" at "LPU64": rc %d\n",
 382                                PFID(ll_inode2fid(dir)), hash, rc);
 383                         return ERR_PTR(rc);
 384                 }
 385         } else {
 386                 /* for cross-ref object, l_ast_data of the lock may not be set,
 387                  * we reset it here */
 388                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
 389                                  dir, NULL);
 390         }
 391         ldlm_lock_dump_handle(D_OTHER, &lockh);
 392
 393         mutex_lock(&lli->lli_readdir_mutex);
 394         page = ll_dir_page_locate(dir, &lhash, &start, &end);
 395         if (IS_ERR(page)) {
 396                 CERROR("dir page locate: "DFID" at "LPU64": rc %ld\n",
 397                        PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
 398                 GOTO(out_unlock, page);
 399         } else if (page != NULL) {
 400                 /*
 401                  * XXX nikita: not entirely correct handling of a corner case:
 402                  * suppose hash chain of entries with hash value HASH crosses
 403                  * border between pages P0 and P1. First both P0 and P1 are
 404                  * cached, seekdir() is called for some entry from the P0 part
 405                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 406                  * happens and finds P1, as it starts with matching hash
 407                  * value. Remaining entries from P0 part of the chain are
 408                  * skipped. (Is that really a bug?)
 409                  *
 410                  * Possible solutions: 0. don't cache P1 is such case, handle
 411                  * it as an "overflow" page. 1. invalidate all pages at
 412                  * once. 2. use HASH|1 as an index for P1.
 413                  */
 414                 GOTO(hash_collision, page);
 415         }
 416
 417         page = read_cache_page(mapping, hash_x_index(hash, hash64),
 418                                ll_dir_filler, &lhash);
 419         if (IS_ERR(page)) {
 420                 CERROR("read cache page: "DFID" at "LPU64": rc %ld\n",
 421                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 422                 GOTO(out_unlock, page);
 423         }
 424
 425         wait_on_page(page);
 426         (void)kmap(page);
 427         if (!PageUptodate(page)) {
 428                 CERROR("page not updated: "DFID" at "LPU64": rc %d\n",
 429                        PFID(ll_inode2fid(dir)), hash, -5);
 430                 goto fail;
 431         }
 432         if (!PageChecked(page))
 433                 ll_check_page(dir, page);
 434         if (PageError(page)) {
 435                 CERROR("page error: "DFID" at "LPU64": rc %d\n",
 436                        PFID(ll_inode2fid(dir)), hash, -5);
 437                 goto fail;
 438         }
 439 hash_collision:
 440         dp = page_address(page);
 441         if (BITS_PER_LONG == 32 && hash64) {
 442                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 443                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 444                 lhash = hash >> 32;
 445         } else {
 446                 start = le64_to_cpu(dp->ldp_hash_start);
 447                 end   = le64_to_cpu(dp->ldp_hash_end);
 448                 lhash = hash;
 449         }
 450         if (end == start) {
 451                 LASSERT(start == lhash);
 452                 CWARN("Page-wide hash collision: "LPU64"\n", end);
 453                 if (BITS_PER_LONG == 32 && hash64)
 454                         CWARN("Real page-wide hash collision at ["LPU64" "LPU64
 455                               "] with hash "LPU64"\n",
 456                               le64_to_cpu(dp->ldp_hash_start),
 457                               le64_to_cpu(dp->ldp_hash_end), hash);
 458                 /*
 459                  * Fetch whole overflow chain...
 460                  *
 461                  * XXX not yet.
 462                  */
 463                 goto fail;
 464         }
 465 out_unlock:
 466         mutex_unlock(&lli->lli_readdir_mutex);
 467         ldlm_lock_decref(&lockh, mode);
 468         return page;
 469
 470 fail:
 471         ll_release_page(page, 1);
 472         page = ERR_PTR(-EIO);
 473         goto out_unlock;
 474 }
 475
 476 int ll_dir_read(struct inode *inode, __u64 *_pos, void *cookie,
 477                 filldir_t filldir)
 478 {
 479         struct ll_inode_info *info       = ll_i2info(inode);
 480         struct ll_sb_info    *sbi        = ll_i2sbi(inode);
 481         __u64                 pos        = *_pos;
 482         int                   api32      = ll_need_32bit_api(sbi);
 483         int                   hash64     = sbi->ll_flags & LL_SBI_64BIT_HASH;
 484         struct page          *page;
 485         struct ll_dir_chain   chain;
 486         int                   done = 0;
 487         int                   rc = 0;
 488         ENTRY;
 489
 490         ll_dir_chain_init(&chain);
 491
 492         page = ll_get_dir_page(inode, pos, &chain);
 493
 494         while (rc == 0 && !done) {
 495                 struct lu_dirpage *dp;
 496                 struct lu_dirent  *ent;
 497
 498                 if (!IS_ERR(page)) {
 499                         /*
 500                          * If page is empty (end of directory is reached),
 501                          * use this value.
 502                          */
 503                         __u64 hash = MDS_DIR_END_OFF;
 504                         __u64 next;
 505
 506                         dp = page_address(page);
 507                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 508                              ent = lu_dirent_next(ent)) {
 509                                 __u16          type;
 510                                 int            namelen;
 511                                 struct lu_fid  fid;
 512                                 __u64          lhash;
 513                                 __u64          ino;
 514
 515                                 /*
 516                                  * XXX: implement correct swabbing here.
 517                                  */
 518
 519                                 hash = le64_to_cpu(ent->lde_hash);
 520                                 if (hash < pos)
 521                                         /*
 522                                          * Skip until we find target hash
 523                                          * value.
 524                                          */
 525                                         continue;
 526
 527                                 namelen = le16_to_cpu(ent->lde_namelen);
 528                                 if (namelen == 0)
 529                                         /*
 530                                          * Skip dummy record.
 531                                          */
 532                                         continue;
 533
 534                                 if (api32 && hash64)
 535                                         lhash = hash >> 32;
 536                                 else
 537                                         lhash = hash;
 538                                 fid_le_to_cpu(&fid, &ent->lde_fid);
 539                                 ino = cl_fid_build_ino(&fid, api32);
 540                                 type = ll_dirent_type_get(ent);
 541                                 /* For 'll_nfs_get_name_filldir()', it will try
 542                                  * to access the 'ent' through its 'lde_name',
 543                                  * so the parameter 'name' for 'filldir()' must
 544                                  * be part of the 'ent'. */
 545                                 done = filldir(cookie, ent->lde_name, namelen,
 546                                                lhash, ino, type);
 547                         }
 548                         next = le64_to_cpu(dp->ldp_hash_end);
 549                         if (!done) {
 550                                 pos = next;
 551                                 if (pos == MDS_DIR_END_OFF) {
 552                                         /*
 553                                          * End of directory reached.
 554                                          */
 555                                         done = 1;
 556                                         ll_release_page(page, 0);
 557                                 } else if (1 /* chain is exhausted*/) {
 558                                         /*
 559                                          * Normal case: continue to the next
 560                                          * page.
 561                                          */
 562                                         ll_release_page(page,
 563                                             le32_to_cpu(dp->ldp_flags) &
 564                                                         LDF_COLLIDE);
 565                                         next = pos;
 566                                         page = ll_get_dir_page(inode, pos,
 567                                                                &chain);
 568                                 } else {
 569                                         /*
 570                                          * go into overflow page.
 571                                          */
 572                                         LASSERT(le32_to_cpu(dp->ldp_flags) &
 573                                                 LDF_COLLIDE);
 574                                         ll_release_page(page, 1);
 575                                 }
 576                         } else {
 577                                 pos = hash;
 578                                 ll_release_page(page, 0);
 579                         }
 580                 } else {
 581                         rc = PTR_ERR(page);
 582                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 583                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 584                 }
 585         }
 586
 587         *_pos = pos;
 588         ll_dir_chain_fini(&chain);
 589         RETURN(rc);
 590 }
 591
 592 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 593 {
 594         struct inode            *inode  = filp->f_dentry->d_inode;
 595         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 596         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 597         __u64                   pos     = lfd->lfd_pos;
 598         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 599         int                     api32   = ll_need_32bit_api(sbi);
 600         int                     rc;
 601 #ifdef HAVE_TOUCH_ATIME_1ARG
 602         struct path             path;
 603 #endif
 604         ENTRY;
 605
 606         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu "
 607                " 32bit_api %d\n", inode->i_ino, inode->i_generation,
 608                inode, (unsigned long)pos, i_size_read(inode), api32);
 609
 610         if (pos == MDS_DIR_END_OFF)
 611                 /*
 612                  * end-of-file.
 613                  */
 614                 GOTO(out, rc = 0);
 615
 616         rc = ll_dir_read(inode, &pos, cookie, filldir);
 617         lfd->lfd_pos = pos;
 618         if (pos == MDS_DIR_END_OFF) {
 619                 if (api32)
 620                         filp->f_pos = LL_DIR_END_OFF_32BIT;
 621                 else
 622                         filp->f_pos = LL_DIR_END_OFF;
 623         } else {
 624                 if (api32 && hash64)
 625                         filp->f_pos = pos >> 32;
 626                 else
 627                         filp->f_pos = pos;
 628         }
 629         filp->f_version = inode->i_version;
 630 #ifdef HAVE_TOUCH_ATIME_1ARG
 631         path.mnt = filp->f_vfsmnt;
 632         path.dentry = filp->f_dentry;
 633         touch_atime(&path);
 634 #else
 635         touch_atime(filp->f_vfsmnt, filp->f_dentry);
 636 #endif
 637
 638 out:
 639         if (!rc)
 640                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 641
 642         RETURN(rc);
 643 }
 644
 645 int ll_send_mgc_param(struct obd_export *mgc, char *string)
 646 {
 647         struct mgs_send_param *msp;
 648         int rc = 0;
 649
 650         OBD_ALLOC_PTR(msp);
 651         if (!msp)
 652                 return -ENOMEM;
 653
 654         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 655         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 656                                 sizeof(struct mgs_send_param), msp, NULL);
 657         if (rc)
 658                 CERROR("Failed to set parameter: %d\n", rc);
 659         OBD_FREE_PTR(msp);
 660
 661         return rc;
 662 }
 663
 664 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 665                      int set_default)
 666 {
 667         struct ll_sb_info *sbi = ll_i2sbi(inode);
 668         struct md_op_data *op_data;
 669         struct ptlrpc_request *req = NULL;
 670         int rc = 0;
 671         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 672         struct obd_device *mgc = lsi->lsi_mgc;
 673         int lum_size;
 674         ENTRY;
 675
 676         if (lump != NULL) {
 677                 /*
 678                  * This is coming from userspace, so should be in
 679                  * local endian.  But the MDS would like it in little
 680                  * endian, so we swab it before we send it.
 681                  */
 682                 switch (lump->lmm_magic) {
 683                 case LOV_USER_MAGIC_V1: {
 684                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 685                                 lustre_swab_lov_user_md_v1(lump);
 686                         lum_size = sizeof(struct lov_user_md_v1);
 687                         break;
 688                         }
 689                 case LOV_USER_MAGIC_V3: {
 690                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 691                                 lustre_swab_lov_user_md_v3(
 692                                         (struct lov_user_md_v3 *)lump);
 693                         lum_size = sizeof(struct lov_user_md_v3);
 694                         break;
 695                         }
 696                 default: {
 697                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 698                                         " %#08x != %#08x nor %#08x\n",
 699                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 700                                         LOV_USER_MAGIC_V3);
 701                         RETURN(-EINVAL);
 702                         }
 703                }
 704         } else {
 705                 lum_size = sizeof(struct lov_user_md_v1);
 706         }
 707
 708         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 709                                      LUSTRE_OPC_ANY, NULL);
 710         if (IS_ERR(op_data))
 711                 RETURN(PTR_ERR(op_data));
 712
 713         /* swabbing is done in lov_setstripe() on server side */
 714         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 715                         NULL, 0, &req, NULL);
 716         ll_finish_md_op_data(op_data);
 717         ptlrpc_req_finished(req);
 718         if (rc) {
 719                 if (rc != -EPERM && rc != -EACCES)
 720                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 721         }
 722
 723         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 724          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 725          need the make the distiction between the 2 versions */
 726         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 727                 char *param = NULL;
 728                 char *buf;
 729
 730                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 731                 if (param == NULL)
 732                         GOTO(end, rc = -ENOMEM);
 733
 734                 buf = param;
 735                 /* Get fsname and assume devname to be -MDT0000. */
 736                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 737                 strcat(buf, "-MDT0000.lov");
 738                 buf += strlen(buf);
 739
 740                 /* Set root stripesize */
 741                 sprintf(buf, ".stripesize=%u",
 742                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 743                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 744                 if (rc)
 745                         GOTO(end, rc);
 746
 747                 /* Set root stripecount */
 748                 sprintf(buf, ".stripecount=%hd",
 749                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 750                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 751                 if (rc)
 752                         GOTO(end, rc);
 753
 754                 /* Set root stripeoffset */
 755                 sprintf(buf, ".stripeoffset=%hd",
 756                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 757                         (typeof(lump->lmm_stripe_offset))(-1));
 758                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 759
 760 end:
 761                 if (param != NULL)
 762                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 763         }
 764         RETURN(rc);
 765 }
 766
 767 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 768                      int *lmm_size, struct ptlrpc_request **request)
 769 {
 770         struct ll_sb_info *sbi = ll_i2sbi(inode);
 771         struct mdt_body   *body;
 772         struct lov_mds_md *lmm = NULL;
 773         struct ptlrpc_request *req = NULL;
 774         int rc, lmmsize;
 775         struct md_op_data *op_data;
 776
 777         rc = ll_get_max_mdsize(sbi, &lmmsize);
 778         if (rc)
 779                 RETURN(rc);
 780
 781         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 782                                      0, lmmsize, LUSTRE_OPC_ANY,
 783                                      NULL);
 784         if (IS_ERR(op_data))
 785                 RETURN(PTR_ERR(op_data));
 786
 787         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 788         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 789         ll_finish_md_op_data(op_data);
 790         if (rc < 0) {
 791                 CDEBUG(D_INFO, "md_getattr failed on inode "
 792                        "%lu/%u: rc %d\n", inode->i_ino,
 793                        inode->i_generation, rc);
 794                 GOTO(out, rc);
 795         }
 796
 797         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 798         LASSERT(body != NULL);
 799
 800         lmmsize = body->eadatasize;
 801
 802         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 803             lmmsize == 0) {
 804                 GOTO(out, rc = -ENODATA);
 805         }
 806
 807         lmm = req_capsule_server_sized_get(&req->rq_pill,
 808                                            &RMF_MDT_MD, lmmsize);
 809         LASSERT(lmm != NULL);
 810
 811         /*
 812          * This is coming from the MDS, so is probably in
 813          * little endian.  We convert it to host endian before
 814          * passing it to userspace.
 815          */
 816         /* We don't swab objects for directories */
 817         switch (le32_to_cpu(lmm->lmm_magic)) {
 818         case LOV_MAGIC_V1:
 819                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 820                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 821                 break;
 822         case LOV_MAGIC_V3:
 823                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 824                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 825                 break;
 826         default:
 827                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 828                 rc = -EPROTO;
 829         }
 830 out:
 831         *lmmp = lmm;
 832         *lmm_size = lmmsize;
 833         *request = req;
 834         return rc;
 835 }
 836
 837 /*
 838  *  Get MDT index for the inode.
 839  */
 840 int ll_get_mdt_idx(struct inode *inode)
 841 {
 842         struct ll_sb_info *sbi = ll_i2sbi(inode);
 843         struct md_op_data *op_data;
 844         int rc, mdtidx;
 845         ENTRY;
 846
 847         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 848                                      0, LUSTRE_OPC_ANY, NULL);
 849         if (IS_ERR(op_data))
 850                 RETURN(PTR_ERR(op_data));
 851
 852         op_data->op_flags |= MF_GET_MDT_IDX;
 853         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 854         mdtidx = op_data->op_mds;
 855         ll_finish_md_op_data(op_data);
 856         if (rc < 0) {
 857                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 858                 RETURN(rc);
 859         }
 860         return mdtidx;
 861 }
 862
 863 /**
 864  * Generic handler to do any pre-copy work.
 865  *
 866  * It send a first hsm_progress (with extent length == 0) to coordinator as a
 867  * first information for it that real work has started.
 868  *
 869  * Moreover, for a ARCHIVE request, it will sample the file data version and
 870  * store it in \a copy.
 871  *
 872  * \return 0 on success.
 873  */
 874 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 875 {
 876         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 877         struct hsm_progress_kernel       hpk;
 878         int                              rc;
 879         ENTRY;
 880
 881         /* Forge a hsm_progress based on data from copy. */
 882         hpk.hpk_fid = copy->hc_hai.hai_fid;
 883         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 884         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 885         hpk.hpk_extent.length = 0;
 886         hpk.hpk_flags = 0;
 887         hpk.hpk_errval = 0;
 888         hpk.hpk_data_version = 0;
 889
 890
 891         /* For archive request, we need to read the current file version. */
 892         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 893                 struct inode    *inode;
 894                 __u64            data_version = 0;
 895
 896                 /* Get inode for this fid */
 897                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 898                 if (IS_ERR(inode)) {
 899                         hpk.hpk_flags |= HP_FLAG_RETRY;
 900                         /* hpk_errval is >= 0 */
 901                         hpk.hpk_errval = -PTR_ERR(inode);
 902                         GOTO(progress, rc = PTR_ERR(inode));
 903                 }
 904
 905                 /* Read current file data version */
 906                 rc = ll_data_version(inode, &data_version, 1);
 907                 iput(inode);
 908                 if (rc != 0) {
 909                         CDEBUG(D_HSM, "Could not read file data version of "
 910                                       DFID" (rc = %d). Archive request ("
 911                                       LPX64") could not be done.\n",
 912                                       PFID(&copy->hc_hai.hai_fid), rc,
 913                                       copy->hc_hai.hai_cookie);
 914                         hpk.hpk_flags |= HP_FLAG_RETRY;
 915                         /* hpk_errval must be >= 0 */
 916                         hpk.hpk_errval = -rc;
 917                         GOTO(progress, rc);
 918                 }
 919
 920                 /* Store it the hsm_copy for later copytool use.
 921                  * Always modified even if no lsm. */
 922                 copy->hc_data_version = data_version;
 923         }
 924
 925 progress:
 926         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 927                            &hpk, NULL);
 928
 929         RETURN(rc);
 930 }
 931
 932 /**
 933  * Generic handler to do any post-copy work.
 934  *
 935  * It will send the last hsm_progress update to coordinator to inform it
 936  * that copy is finished and whether it was successful or not.
 937  *
 938  * Moreover,
 939  * - for ARCHIVE request, it will sample the file data version and compare it
 940  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 941  *   will be considered as failed.
 942  * - for RESTORE request, it will sample the file data version and send it to
 943  *   coordinator which is useful if the file was imported as 'released'.
 944  *
 945  * \return 0 on success.
 946  */
 947 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 948 {
 949         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 950         struct hsm_progress_kernel       hpk;
 951         int                              rc;
 952         ENTRY;
 953
 954         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 955         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 956          * initialized if copy_end was called with copy == NULL.
 957          */
 958
 959         /* Forge a hsm_progress based on data from copy. */
 960         hpk.hpk_fid = copy->hc_hai.hai_fid;
 961         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 962         hpk.hpk_extent = copy->hc_hai.hai_extent;
 963         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 964         hpk.hpk_errval = copy->hc_errval;
 965         hpk.hpk_data_version = 0;
 966
 967         /* For archive request, we need to check the file data was not changed.
 968          *
 969          * For restore request, we need to send the file data version, this is
 970          * useful when the file was created using hsm_import.
 971          */
 972         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 973              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 974             (copy->hc_errval == 0)) {
 975                 struct inode    *inode;
 976                 __u64            data_version = 0;
 977
 978                 /* Get lsm for this fid */
 979                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 980                 if (IS_ERR(inode)) {
 981                         hpk.hpk_flags |= HP_FLAG_RETRY;
 982                         /* hpk_errval must be >= 0 */
 983                         hpk.hpk_errval = -PTR_ERR(inode);
 984                         GOTO(progress, rc = PTR_ERR(inode));
 985                 }
 986
 987                 rc = ll_data_version(inode, &data_version,
 988                                      copy->hc_hai.hai_action == HSMA_ARCHIVE);
 989                 iput(inode);
 990                 if (rc) {
 991                         CDEBUG(D_HSM, "Could not read file data version. "
 992                                       "Request could not be confirmed.\n");
 993                         if (hpk.hpk_errval == 0)
 994                                 hpk.hpk_errval = rc;
 995                         GOTO(progress, rc);
 996                 }
 997
 998                 /* Store it the hsm_copy for later copytool use.
 999                  * Always modified even if no lsm. */
1000                 hpk.hpk_data_version = data_version;
1001
1002                 /* File could have been stripped during archiving, so we need
1003                  * to check anyway. */
1004                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1005                     (copy->hc_data_version != data_version)) {
1006                         CDEBUG(D_HSM, "File data version mismatched. "
1007                               "File content was changed during archiving. "
1008                                DFID", start:"LPX64" current:"LPX64"\n",
1009                                PFID(&copy->hc_hai.hai_fid),
1010                                copy->hc_data_version, data_version);
1011                         /* File was changed, send error to cdt. Do not ask for
1012                          * retry because if a file is modified frequently,
1013                          * the cdt will loop on retried archive requests.
1014                          * The policy engine will ask for a new archive later
1015                          * when the file will not be modified for some tunable
1016                          * time */
1017                         /* we do not notify caller */
1018                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
1019                         /* hpk_errval must be >= 0 */
1020                         hpk.hpk_errval = EBUSY;
1021                 }
1022
1023         }
1024
1025 progress:
1026         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1027                            &hpk, NULL);
1028
1029         RETURN(rc);
1030 }
1031
1032
1033 static int copy_and_ioctl(int cmd, struct obd_export *exp, void *data, int len)
1034 {
1035         void *ptr;
1036         int rc;
1037
1038         OBD_ALLOC(ptr, len);
1039         if (ptr == NULL)
1040                 return -ENOMEM;
1041         if (cfs_copy_from_user(ptr, data, len)) {
1042                 OBD_FREE(ptr, len);
1043                 return -EFAULT;
1044         }
1045         rc = obd_iocontrol(cmd, exp, len, data, NULL);
1046         OBD_FREE(ptr, len);
1047         return rc;
1048 }
1049
1050 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1051 {
1052         int cmd = qctl->qc_cmd;
1053         int type = qctl->qc_type;
1054         int id = qctl->qc_id;
1055         int valid = qctl->qc_valid;
1056         int rc = 0;
1057         ENTRY;
1058
1059         switch (cmd) {
1060         case LUSTRE_Q_INVALIDATE:
1061         case LUSTRE_Q_FINVALIDATE:
1062         case Q_QUOTAON:
1063         case Q_QUOTAOFF:
1064         case Q_SETQUOTA:
1065         case Q_SETINFO:
1066                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1067                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1068                         RETURN(-EPERM);
1069                 break;
1070         case Q_GETQUOTA:
1071                 if (((type == USRQUOTA && cfs_curproc_euid() != id) ||
1072                      (type == GRPQUOTA && !in_egroup_p(id))) &&
1073                     (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1074                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
1075                         RETURN(-EPERM);
1076                 break;
1077         case Q_GETINFO:
1078                 break;
1079         default:
1080                 CERROR("unsupported quotactl op: %#x\n", cmd);
1081                 RETURN(-ENOTTY);
1082         }
1083
1084         if (valid != QC_GENERAL) {
1085                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1086                         RETURN(-EOPNOTSUPP);
1087
1088                 if (cmd == Q_GETINFO)
1089                         qctl->qc_cmd = Q_GETOINFO;
1090                 else if (cmd == Q_GETQUOTA)
1091                         qctl->qc_cmd = Q_GETOQUOTA;
1092                 else
1093                         RETURN(-EINVAL);
1094
1095                 switch (valid) {
1096                 case QC_MDTIDX:
1097                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1098                                            sizeof(*qctl), qctl, NULL);
1099                         break;
1100                 case QC_OSTIDX:
1101                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1102                                            sizeof(*qctl), qctl, NULL);
1103                         break;
1104                 case QC_UUID:
1105                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1106                                            sizeof(*qctl), qctl, NULL);
1107                         if (rc == -EAGAIN)
1108                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1109                                                    sbi->ll_dt_exp,
1110                                                    sizeof(*qctl), qctl, NULL);
1111                         break;
1112                 default:
1113                         rc = -EINVAL;
1114                         break;
1115                 }
1116
1117                 if (rc)
1118                         RETURN(rc);
1119
1120                 qctl->qc_cmd = cmd;
1121         } else {
1122                 struct obd_quotactl *oqctl;
1123
1124                 OBD_ALLOC_PTR(oqctl);
1125                 if (oqctl == NULL)
1126                         RETURN(-ENOMEM);
1127
1128                 QCTL_COPY(oqctl, qctl);
1129                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1130                 if (rc) {
1131                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
1132                                 oqctl->qc_cmd = Q_QUOTAOFF;
1133                                 obd_quotactl(sbi->ll_md_exp, oqctl);
1134                         }
1135                         OBD_FREE_PTR(oqctl);
1136                         RETURN(rc);
1137                 }
1138                 /* If QIF_SPACE is not set, client should collect the
1139                  * space usage from OSSs by itself */
1140                 if (cmd == Q_GETQUOTA &&
1141                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1142                     !oqctl->qc_dqblk.dqb_curspace) {
1143                         struct obd_quotactl *oqctl_tmp;
1144
1145                         OBD_ALLOC_PTR(oqctl_tmp);
1146                         if (oqctl_tmp == NULL)
1147                                 GOTO(out, rc = -ENOMEM);
1148
1149                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1150                         oqctl_tmp->qc_id = oqctl->qc_id;
1151                         oqctl_tmp->qc_type = oqctl->qc_type;
1152
1153                         /* collect space usage from OSTs */
1154                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1155                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1156                         if (!rc || rc == -EREMOTEIO) {
1157                                 oqctl->qc_dqblk.dqb_curspace =
1158                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1159                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1160                         }
1161
1162                         /* collect space & inode usage from MDTs */
1163                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1164                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1165                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1166                         if (!rc || rc == -EREMOTEIO) {
1167                                 oqctl->qc_dqblk.dqb_curspace +=
1168                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1169                                 oqctl->qc_dqblk.dqb_curinodes =
1170                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1171                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1172                         } else {
1173                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1174                         }
1175
1176                         OBD_FREE_PTR(oqctl_tmp);
1177                 }
1178 out:
1179                 QCTL_COPY(qctl, oqctl);
1180                 OBD_FREE_PTR(oqctl);
1181         }
1182
1183         RETURN(rc);
1184 }
1185
1186 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1187 {
1188         struct inode *inode = file->f_dentry->d_inode;
1189         struct ll_sb_info *sbi = ll_i2sbi(inode);
1190         struct obd_ioctl_data *data;
1191         int rc = 0;
1192         ENTRY;
1193
1194         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1195                inode->i_ino, inode->i_generation, inode, cmd);
1196
1197         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1198         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1199                 return -ENOTTY;
1200
1201         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1202         switch(cmd) {
1203         case FSFILT_IOC_GETFLAGS:
1204         case FSFILT_IOC_SETFLAGS:
1205                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1206         case FSFILT_IOC_GETVERSION_OLD:
1207         case FSFILT_IOC_GETVERSION:
1208                 RETURN(put_user(inode->i_generation, (int *)arg));
1209         /* We need to special case any other ioctls we want to handle,
1210          * to send them to the MDS/OST as appropriate and to properly
1211          * network encode the arg field.
1212         case FSFILT_IOC_SETVERSION_OLD:
1213         case FSFILT_IOC_SETVERSION:
1214         */
1215         case LL_IOC_GET_MDTIDX: {
1216                 int mdtidx;
1217
1218                 mdtidx = ll_get_mdt_idx(inode);
1219                 if (mdtidx < 0)
1220                         RETURN(mdtidx);
1221
1222                 if (put_user((int)mdtidx, (int*)arg))
1223                         RETURN(-EFAULT);
1224
1225                 return 0;
1226         }
1227         case IOC_MDC_LOOKUP: {
1228                 struct ptlrpc_request *request = NULL;
1229                 int namelen, len = 0;
1230                 char *buf = NULL;
1231                 char *filename;
1232                 struct md_op_data *op_data;
1233
1234                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1235                 if (rc)
1236                         RETURN(rc);
1237                 data = (void *)buf;
1238
1239                 filename = data->ioc_inlbuf1;
1240                 namelen = strlen(filename);
1241
1242                 if (namelen < 1) {
1243                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1244                         GOTO(out_free, rc = -EINVAL);
1245                 }
1246
1247                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1248                                              0, LUSTRE_OPC_ANY, NULL);
1249                 if (IS_ERR(op_data))
1250                         GOTO(out_free, rc = PTR_ERR(op_data));
1251
1252                 op_data->op_valid = OBD_MD_FLID;
1253                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1254                 ll_finish_md_op_data(op_data);
1255                 if (rc < 0) {
1256                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1257                         GOTO(out_free, rc);
1258                 }
1259                 ptlrpc_req_finished(request);
1260                 EXIT;
1261 out_free:
1262                 obd_ioctl_freedata(buf, len);
1263                 return rc;
1264         }
1265         case LL_IOC_LOV_SETSTRIPE: {
1266                 struct lov_user_md_v3 lumv3;
1267                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1268                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1269                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1270
1271                 int set_default = 0;
1272
1273                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1274                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1275                         sizeof(lumv3p->lmm_objects[0]));
1276                 /* first try with v1 which is smaller than v3 */
1277                 if (cfs_copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1278                         RETURN(-EFAULT);
1279
1280                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1281                         if (cfs_copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1282                                 RETURN(-EFAULT);
1283                 }
1284
1285                 if (inode->i_sb->s_root == file->f_dentry)
1286                         set_default = 1;
1287
1288                 /* in v1 and v3 cases lumv1 points to data */
1289                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1290
1291                 RETURN(rc);
1292         }
1293         case LL_IOC_OBD_STATFS:
1294                 RETURN(ll_obd_statfs(inode, (void *)arg));
1295         case LL_IOC_LOV_GETSTRIPE:
1296         case LL_IOC_MDC_GETINFO:
1297         case IOC_MDC_GETFILEINFO:
1298         case IOC_MDC_GETFILESTRIPE: {
1299                 struct ptlrpc_request *request = NULL;
1300                 struct lov_user_md *lump;
1301                 struct lov_mds_md *lmm = NULL;
1302                 struct mdt_body *body;
1303                 char *filename = NULL;
1304                 int lmmsize;
1305
1306                 if (cmd == IOC_MDC_GETFILEINFO ||
1307                     cmd == IOC_MDC_GETFILESTRIPE) {
1308                         filename = getname((const char *)arg);
1309                         if (IS_ERR(filename))
1310                                 RETURN(PTR_ERR(filename));
1311
1312                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1313                                                       &lmmsize, &request);
1314                 } else {
1315                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1316                 }
1317
1318                 if (request) {
1319                         body = req_capsule_server_get(&request->rq_pill,
1320                                                       &RMF_MDT_BODY);
1321                         LASSERT(body != NULL);
1322                 } else {
1323                         GOTO(out_req, rc);
1324                 }
1325
1326                 if (rc < 0) {
1327                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1328                                                cmd == LL_IOC_MDC_GETINFO))
1329                                 GOTO(skip_lmm, rc = 0);
1330                         else
1331                                 GOTO(out_req, rc);
1332                 }
1333
1334                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1335                     cmd == LL_IOC_LOV_GETSTRIPE) {
1336                         lump = (struct lov_user_md *)arg;
1337                 } else {
1338                         struct lov_user_mds_data *lmdp;
1339                         lmdp = (struct lov_user_mds_data *)arg;
1340                         lump = &lmdp->lmd_lmm;
1341                 }
1342                 if (cfs_copy_to_user(lump, lmm, lmmsize)) {
1343                         if (cfs_copy_to_user(lump, lmm, sizeof(*lump)))
1344                                 GOTO(out_req, rc = -EFAULT);
1345                         rc = -EOVERFLOW;
1346                 }
1347         skip_lmm:
1348                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1349                         struct lov_user_mds_data *lmdp;
1350                         lstat_t st = { 0 };
1351
1352                         st.st_dev     = inode->i_sb->s_dev;
1353                         st.st_mode    = body->mode;
1354                         st.st_nlink   = body->nlink;
1355                         st.st_uid     = body->uid;
1356                         st.st_gid     = body->gid;
1357                         st.st_rdev    = body->rdev;
1358                         st.st_size    = body->size;
1359                         st.st_blksize = CFS_PAGE_SIZE;
1360                         st.st_blocks  = body->blocks;
1361                         st.st_atime   = body->atime;
1362                         st.st_mtime   = body->mtime;
1363                         st.st_ctime   = body->ctime;
1364                         st.st_ino     = inode->i_ino;
1365
1366                         lmdp = (struct lov_user_mds_data *)arg;
1367                         if (cfs_copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1368                                 GOTO(out_req, rc = -EFAULT);
1369                 }
1370
1371                 EXIT;
1372         out_req:
1373                 ptlrpc_req_finished(request);
1374                 if (filename)
1375                         putname(filename);
1376                 return rc;
1377         }
1378         case IOC_LOV_GETINFO: {
1379                 struct lov_user_mds_data *lumd;
1380                 struct lov_stripe_md *lsm;
1381                 struct lov_user_md *lum;
1382                 struct lov_mds_md *lmm;
1383                 int lmmsize;
1384                 lstat_t st;
1385
1386                 lumd = (struct lov_user_mds_data *)arg;
1387                 lum = &lumd->lmd_lmm;
1388
1389                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1390                 if (rc)
1391                         RETURN(rc);
1392
1393                 OBD_ALLOC_LARGE(lmm, lmmsize);
1394                 if (cfs_copy_from_user(lmm, lum, lmmsize))
1395                         GOTO(free_lmm, rc = -EFAULT);
1396
1397                 switch (lmm->lmm_magic) {
1398                 case LOV_USER_MAGIC_V1:
1399                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1400                                 break;
1401                         /* swab objects first so that stripes num will be sane */
1402                         lustre_swab_lov_user_md_objects(
1403                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1404                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1405                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1406                         break;
1407                 case LOV_USER_MAGIC_V3:
1408                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1409                                 break;
1410                         /* swab objects first so that stripes num will be sane */
1411                         lustre_swab_lov_user_md_objects(
1412                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1413                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1414                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1415                         break;
1416                 default:
1417                         GOTO(free_lmm, rc = -EINVAL);
1418                 }
1419
1420                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1421                 if (rc < 0)
1422                         GOTO(free_lmm, rc = -ENOMEM);
1423
1424                 /* Perform glimpse_size operation. */
1425                 memset(&st, 0, sizeof(st));
1426
1427                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1428                 if (rc)
1429                         GOTO(free_lsm, rc);
1430
1431                 if (cfs_copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1432                         GOTO(free_lsm, rc = -EFAULT);
1433
1434                 EXIT;
1435         free_lsm:
1436                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1437         free_lmm:
1438                 OBD_FREE_LARGE(lmm, lmmsize);
1439                 return rc;
1440         }
1441         case OBD_IOC_LLOG_CATINFO: {
1442                 RETURN(-EOPNOTSUPP);
1443         }
1444         case OBD_IOC_QUOTACHECK: {
1445                 struct obd_quotactl *oqctl;
1446                 int error = 0;
1447
1448                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1449                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1450                         RETURN(-EPERM);
1451
1452                 OBD_ALLOC_PTR(oqctl);
1453                 if (!oqctl)
1454                         RETURN(-ENOMEM);
1455                 oqctl->qc_type = arg;
1456                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1457                 if (rc < 0) {
1458                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1459                         error = rc;
1460                 }
1461
1462                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1463                 if (rc < 0)
1464                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1465
1466                 OBD_FREE_PTR(oqctl);
1467                 return error ?: rc;
1468         }
1469         case OBD_IOC_POLL_QUOTACHECK: {
1470                 struct if_quotacheck *check;
1471
1472                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1473                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1474                         RETURN(-EPERM);
1475
1476                 OBD_ALLOC_PTR(check);
1477                 if (!check)
1478                         RETURN(-ENOMEM);
1479
1480                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1481                                    NULL);
1482                 if (rc) {
1483                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1484                         if (cfs_copy_to_user((void *)arg, check,
1485                                              sizeof(*check)))
1486                                 CDEBUG(D_QUOTA, "cfs_copy_to_user failed\n");
1487                         GOTO(out_poll, rc);
1488                 }
1489
1490                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1491                                    NULL);
1492                 if (rc) {
1493                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1494                         if (cfs_copy_to_user((void *)arg, check,
1495                                              sizeof(*check)))
1496                                 CDEBUG(D_QUOTA, "cfs_copy_to_user failed\n");
1497                         GOTO(out_poll, rc);
1498                 }
1499         out_poll:
1500                 OBD_FREE_PTR(check);
1501                 RETURN(rc);
1502         }
1503 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
1504         case LL_IOC_QUOTACTL_18: {
1505                 /* copy the old 1.x quota struct for internal use, then copy
1506                  * back into old format struct.  For 1.8 compatibility. */
1507                 struct if_quotactl_18 *qctl_18;
1508                 struct if_quotactl *qctl_20;
1509
1510                 OBD_ALLOC_PTR(qctl_18);
1511                 if (!qctl_18)
1512                         RETURN(-ENOMEM);
1513
1514                 OBD_ALLOC_PTR(qctl_20);
1515                 if (!qctl_20)
1516                         GOTO(out_quotactl_18, rc = -ENOMEM);
1517
1518                 if (cfs_copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1519                         GOTO(out_quotactl_20, rc = -ENOMEM);
1520
1521                 QCTL_COPY(qctl_20, qctl_18);
1522                 qctl_20->qc_idx = 0;
1523
1524                 /* XXX: dqb_valid was borrowed as a flag to mark that
1525                  *      only mds quota is wanted */
1526                 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1527                     qctl_18->qc_dqblk.dqb_valid) {
1528                         qctl_20->qc_valid = QC_MDTIDX;
1529                         qctl_20->qc_dqblk.dqb_valid = 0;
1530                 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1531                         qctl_20->qc_valid = QC_UUID;
1532                         qctl_20->obd_uuid = qctl_18->obd_uuid;
1533                 } else {
1534                         qctl_20->qc_valid = QC_GENERAL;
1535                 }
1536
1537                 rc = quotactl_ioctl(sbi, qctl_20);
1538
1539                 if (rc == 0) {
1540                         QCTL_COPY(qctl_18, qctl_20);
1541                         qctl_18->obd_uuid = qctl_20->obd_uuid;
1542
1543                         if (cfs_copy_to_user((void *)arg, qctl_18,
1544                                              sizeof(*qctl_18)))
1545                                 rc = -EFAULT;
1546                 }
1547
1548         out_quotactl_20:
1549                 OBD_FREE_PTR(qctl_20);
1550         out_quotactl_18:
1551                 OBD_FREE_PTR(qctl_18);
1552                 RETURN(rc);
1553         }
1554 #else
1555 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1556 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
1557         case LL_IOC_QUOTACTL: {
1558                 struct if_quotactl *qctl;
1559
1560                 OBD_ALLOC_PTR(qctl);
1561                 if (!qctl)
1562                         RETURN(-ENOMEM);
1563
1564                 if (cfs_copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1565                         GOTO(out_quotactl, rc = -EFAULT);
1566
1567                 rc = quotactl_ioctl(sbi, qctl);
1568
1569                 if (rc == 0 && cfs_copy_to_user((void *)arg,qctl,sizeof(*qctl)))
1570                         rc = -EFAULT;
1571
1572         out_quotactl:
1573                 OBD_FREE_PTR(qctl);
1574                 RETURN(rc);
1575         }
1576         case OBD_IOC_GETDTNAME:
1577         case OBD_IOC_GETMDNAME:
1578                 RETURN(ll_get_obd_name(inode, cmd, arg));
1579         case LL_IOC_FLUSHCTX:
1580                 RETURN(ll_flush_ctx(inode));
1581 #ifdef CONFIG_FS_POSIX_ACL
1582         case LL_IOC_RMTACL: {
1583             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1584                 inode == inode->i_sb->s_root->d_inode) {
1585                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1586
1587                 LASSERT(fd != NULL);
1588                 rc = rct_add(&sbi->ll_rct, cfs_curproc_pid(), arg);
1589                 if (!rc)
1590                         fd->fd_flags |= LL_FILE_RMTACL;
1591                 RETURN(rc);
1592             } else
1593                 RETURN(0);
1594         }
1595 #endif
1596         case LL_IOC_GETOBDCOUNT: {
1597                 int count, vallen;
1598                 struct obd_export *exp;
1599
1600                 if (cfs_copy_from_user(&count, (int *)arg, sizeof(int)))
1601                         RETURN(-EFAULT);
1602
1603                 /* get ost count when count is zero, get mdt count otherwise */
1604                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1605                 vallen = sizeof(count);
1606                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1607                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1608                 if (rc) {
1609                         CERROR("get target count failed: %d\n", rc);
1610                         RETURN(rc);
1611                 }
1612
1613                 if (cfs_copy_to_user((int *)arg, &count, sizeof(int)))
1614                         RETURN(-EFAULT);
1615
1616                 RETURN(0);
1617         }
1618         case LL_IOC_PATH2FID:
1619                 if (cfs_copy_to_user((void *)arg, ll_inode2fid(inode),
1620                                      sizeof(struct lu_fid)))
1621                         RETURN(-EFAULT);
1622                 RETURN(0);
1623         case LL_IOC_GET_CONNECT_FLAGS: {
1624                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1625         }
1626         case OBD_IOC_CHANGELOG_SEND:
1627         case OBD_IOC_CHANGELOG_CLEAR:
1628                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1629                                     sizeof(struct ioc_changelog));
1630                 RETURN(rc);
1631         case OBD_IOC_FID2PATH:
1632                 RETURN(ll_fid2path(inode, (void *)arg));
1633         case LL_IOC_HSM_REQUEST: {
1634                 struct hsm_user_request *hur;
1635                 int                      totalsize;
1636
1637                 OBD_ALLOC_PTR(hur);
1638                 if (hur == NULL)
1639                         RETURN(-ENOMEM);
1640
1641                 /* We don't know the true size yet; copy the fixed-size part */
1642                 if (cfs_copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1643                         OBD_FREE_PTR(hur);
1644                         RETURN(-EFAULT);
1645                 }
1646
1647                 /* Compute the whole struct size */
1648                 totalsize = hur_len(hur);
1649                 OBD_FREE_PTR(hur);
1650                 OBD_ALLOC_LARGE(hur, totalsize);
1651                 if (hur == NULL)
1652                         RETURN(-ENOMEM);
1653
1654                 /* Copy the whole struct */
1655                 if (cfs_copy_from_user(hur, (void *)arg, totalsize)) {
1656                         OBD_FREE_LARGE(hur, totalsize);
1657                         RETURN(-EFAULT);
1658                 }
1659
1660                 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1661                                    hur, NULL);
1662
1663                 OBD_FREE_LARGE(hur, totalsize);
1664
1665                 RETURN(rc);
1666         }
1667         case LL_IOC_HSM_PROGRESS: {
1668                 struct hsm_progress_kernel      hpk;
1669                 struct hsm_progress             hp;
1670
1671                 if (cfs_copy_from_user(&hp, (void *)arg, sizeof(hp)))
1672                         RETURN(-EFAULT);
1673
1674                 hpk.hpk_fid = hp.hp_fid;
1675                 hpk.hpk_cookie = hp.hp_cookie;
1676                 hpk.hpk_extent = hp.hp_extent;
1677                 hpk.hpk_flags = hp.hp_flags;
1678                 hpk.hpk_errval = hp.hp_errval;
1679                 hpk.hpk_data_version = 0;
1680
1681                 /* File may not exist in Lustre; all progress
1682                  * reported to Lustre root */
1683                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1684                                    NULL);
1685                 RETURN(rc);
1686         }
1687         case LL_IOC_HSM_CT_START:
1688                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1689                                     sizeof(struct lustre_kernelcomm));
1690                 RETURN(rc);
1691
1692         case LL_IOC_HSM_COPY_START: {
1693                 struct hsm_copy *copy;
1694                 int              rc;
1695
1696                 OBD_ALLOC_PTR(copy);
1697                 if (copy == NULL)
1698                         RETURN(-ENOMEM);
1699                 if (cfs_copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1700                         OBD_FREE_PTR(copy);
1701                         RETURN(-EFAULT);
1702                 }
1703
1704                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1705                 if (cfs_copy_to_user((char *)arg, copy, sizeof(*copy)))
1706                         rc = -EFAULT;
1707
1708                 OBD_FREE_PTR(copy);
1709                 RETURN(rc);
1710         }
1711         case LL_IOC_HSM_COPY_END: {
1712                 struct hsm_copy *copy;
1713                 int              rc;
1714
1715                 OBD_ALLOC_PTR(copy);
1716                 if (copy == NULL)
1717                         RETURN(-ENOMEM);
1718                 if (cfs_copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1719                         OBD_FREE_PTR(copy);
1720                         RETURN(-EFAULT);
1721                 }
1722
1723                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1724                 if (cfs_copy_to_user((char *)arg, copy, sizeof(*copy)))
1725                         rc = -EFAULT;
1726
1727                 OBD_FREE_PTR(copy);
1728                 RETURN(rc);
1729         }
1730
1731         default:
1732                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1733                                      (void *)arg));
1734         }
1735 }
1736
1737 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1738 {
1739         struct inode *inode = file->f_mapping->host;
1740         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1741         struct ll_sb_info *sbi = ll_i2sbi(inode);
1742         int api32 = ll_need_32bit_api(sbi);
1743         loff_t ret = -EINVAL;
1744         ENTRY;
1745
1746         mutex_lock(&inode->i_mutex);
1747         switch (origin) {
1748                 case SEEK_SET:
1749                         break;
1750                 case SEEK_CUR:
1751                         offset += file->f_pos;
1752                         break;
1753                 case SEEK_END:
1754                         if (offset > 0)
1755                                 GOTO(out, ret);
1756                         if (api32)
1757                                 offset += LL_DIR_END_OFF_32BIT;
1758                         else
1759                                 offset += LL_DIR_END_OFF;
1760                         break;
1761                 default:
1762                         GOTO(out, ret);
1763         }
1764
1765         if (offset >= 0 &&
1766             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1767              (!api32 && offset <= LL_DIR_END_OFF))) {
1768                 if (offset != file->f_pos) {
1769                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1770                             (!api32 && offset == LL_DIR_END_OFF))
1771                                 fd->lfd_pos = MDS_DIR_END_OFF;
1772                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1773                                 fd->lfd_pos = offset << 32;
1774                         else
1775                                 fd->lfd_pos = offset;
1776                         file->f_pos = offset;
1777                         file->f_version = 0;
1778                 }
1779                 ret = offset;
1780         }
1781         GOTO(out, ret);
1782
1783 out:
1784         mutex_unlock(&inode->i_mutex);
1785         return ret;
1786 }
1787
1788 int ll_dir_open(struct inode *inode, struct file *file)
1789 {
1790         ENTRY;
1791         RETURN(ll_file_open(inode, file));
1792 }
1793
1794 int ll_dir_release(struct inode *inode, struct file *file)
1795 {
1796         ENTRY;
1797         RETURN(ll_file_release(inode, file));
1798 }
1799
1800 struct file_operations ll_dir_operations = {
1801         .llseek   = ll_dir_seek,
1802         .open     = ll_dir_open,
1803         .release  = ll_dir_release,
1804         .read     = generic_read_dir,
1805         .readdir  = ll_readdir,
1806         .unlocked_ioctl   = ll_dir_ioctl,
1807         .fsync    = ll_fsync,
1808 };