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_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
 665                         char *filename)
 666 {
 667         struct ptlrpc_request *request = NULL;
 668         struct md_op_data *op_data;
 669         struct ll_sb_info *sbi = ll_i2sbi(dir);
 670         int mode;
 671         int err;
 672
 673         ENTRY;
 674
 675         mode = (0755 & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
 676         op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
 677                                      strlen(filename), mode, LUSTRE_OPC_MKDIR,
 678                                      lump);
 679         if (IS_ERR(op_data))
 680                 GOTO(err_exit, err = PTR_ERR(op_data));
 681
 682         op_data->op_cli_flags |= CLI_SET_MEA;
 683         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 684                         cfs_curproc_fsuid(), cfs_curproc_fsgid(),
 685                         cfs_curproc_cap_pack(), 0, &request);
 686         ll_finish_md_op_data(op_data);
 687         if (err)
 688                 GOTO(err_exit, err);
 689 err_exit:
 690         ptlrpc_req_finished(request);
 691         return err;
 692 }
 693
 694 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 695                      int set_default)
 696 {
 697         struct ll_sb_info *sbi = ll_i2sbi(inode);
 698         struct md_op_data *op_data;
 699         struct ptlrpc_request *req = NULL;
 700         int rc = 0;
 701         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 702         struct obd_device *mgc = lsi->lsi_mgc;
 703         int lum_size;
 704         ENTRY;
 705
 706         if (lump != NULL) {
 707                 /*
 708                  * This is coming from userspace, so should be in
 709                  * local endian.  But the MDS would like it in little
 710                  * endian, so we swab it before we send it.
 711                  */
 712                 switch (lump->lmm_magic) {
 713                 case LOV_USER_MAGIC_V1: {
 714                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 715                                 lustre_swab_lov_user_md_v1(lump);
 716                         lum_size = sizeof(struct lov_user_md_v1);
 717                         break;
 718                 }
 719                 case LOV_USER_MAGIC_V3: {
 720                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 721                                 lustre_swab_lov_user_md_v3(
 722                                         (struct lov_user_md_v3 *)lump);
 723                         lum_size = sizeof(struct lov_user_md_v3);
 724                         break;
 725                 }
 726                 default: {
 727                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 728                                         " %#08x != %#08x nor %#08x\n",
 729                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 730                                         LOV_USER_MAGIC_V3);
 731                         RETURN(-EINVAL);
 732                 }
 733                 }
 734         } else {
 735                 lum_size = sizeof(struct lov_user_md_v1);
 736         }
 737
 738         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 739                                      LUSTRE_OPC_ANY, NULL);
 740         if (IS_ERR(op_data))
 741                 RETURN(PTR_ERR(op_data));
 742
 743         if (lump != NULL && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC))
 744                 op_data->op_cli_flags |= CLI_SET_MEA;
 745
 746         /* swabbing is done in lov_setstripe() on server side */
 747         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 748                         NULL, 0, &req, NULL);
 749         ll_finish_md_op_data(op_data);
 750         ptlrpc_req_finished(req);
 751         if (rc) {
 752                 if (rc != -EPERM && rc != -EACCES)
 753                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 754         }
 755
 756         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 757          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 758          need the make the distiction between the 2 versions */
 759         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 760                 char *param = NULL;
 761                 char *buf;
 762
 763                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 764                 if (param == NULL)
 765                         GOTO(end, rc = -ENOMEM);
 766
 767                 buf = param;
 768                 /* Get fsname and assume devname to be -MDT0000. */
 769                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 770                 strcat(buf, "-MDT0000.lov");
 771                 buf += strlen(buf);
 772
 773                 /* Set root stripesize */
 774                 sprintf(buf, ".stripesize=%u",
 775                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 776                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 777                 if (rc)
 778                         GOTO(end, rc);
 779
 780                 /* Set root stripecount */
 781                 sprintf(buf, ".stripecount=%hd",
 782                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 783                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 784                 if (rc)
 785                         GOTO(end, rc);
 786
 787                 /* Set root stripeoffset */
 788                 sprintf(buf, ".stripeoffset=%hd",
 789                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 790                         (typeof(lump->lmm_stripe_offset))(-1));
 791                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 792
 793 end:
 794                 if (param != NULL)
 795                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 796         }
 797         RETURN(rc);
 798 }
 799
 800 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 801                      int *lmm_size, struct ptlrpc_request **request)
 802 {
 803         struct ll_sb_info *sbi = ll_i2sbi(inode);
 804         struct mdt_body   *body;
 805         struct lov_mds_md *lmm = NULL;
 806         struct ptlrpc_request *req = NULL;
 807         int rc, lmmsize;
 808         struct md_op_data *op_data;
 809
 810         rc = ll_get_max_mdsize(sbi, &lmmsize);
 811         if (rc)
 812                 RETURN(rc);
 813
 814         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 815                                      0, lmmsize, LUSTRE_OPC_ANY,
 816                                      NULL);
 817         if (IS_ERR(op_data))
 818                 RETURN(PTR_ERR(op_data));
 819
 820         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 821         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 822         ll_finish_md_op_data(op_data);
 823         if (rc < 0) {
 824                 CDEBUG(D_INFO, "md_getattr failed on inode "
 825                        "%lu/%u: rc %d\n", inode->i_ino,
 826                        inode->i_generation, rc);
 827                 GOTO(out, rc);
 828         }
 829
 830         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 831         LASSERT(body != NULL);
 832
 833         lmmsize = body->eadatasize;
 834
 835         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 836             lmmsize == 0) {
 837                 GOTO(out, rc = -ENODATA);
 838         }
 839
 840         lmm = req_capsule_server_sized_get(&req->rq_pill,
 841                                            &RMF_MDT_MD, lmmsize);
 842         LASSERT(lmm != NULL);
 843
 844         /*
 845          * This is coming from the MDS, so is probably in
 846          * little endian.  We convert it to host endian before
 847          * passing it to userspace.
 848          */
 849         /* We don't swab objects for directories */
 850         switch (le32_to_cpu(lmm->lmm_magic)) {
 851         case LOV_MAGIC_V1:
 852                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 853                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 854                 break;
 855         case LOV_MAGIC_V3:
 856                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 857                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 858                 break;
 859         default:
 860                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 861                 rc = -EPROTO;
 862         }
 863 out:
 864         *lmmp = lmm;
 865         *lmm_size = lmmsize;
 866         *request = req;
 867         return rc;
 868 }
 869
 870 /*
 871  *  Get MDT index for the inode.
 872  */
 873 int ll_get_mdt_idx(struct inode *inode)
 874 {
 875         struct ll_sb_info *sbi = ll_i2sbi(inode);
 876         struct md_op_data *op_data;
 877         int rc, mdtidx;
 878         ENTRY;
 879
 880         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 881                                      0, LUSTRE_OPC_ANY, NULL);
 882         if (IS_ERR(op_data))
 883                 RETURN(PTR_ERR(op_data));
 884
 885         op_data->op_flags |= MF_GET_MDT_IDX;
 886         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 887         mdtidx = op_data->op_mds;
 888         ll_finish_md_op_data(op_data);
 889         if (rc < 0) {
 890                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 891                 RETURN(rc);
 892         }
 893         return mdtidx;
 894 }
 895
 896 /**
 897  * Generic handler to do any pre-copy work.
 898  *
 899  * It send a first hsm_progress (with extent length == 0) to coordinator as a
 900  * first information for it that real work has started.
 901  *
 902  * Moreover, for a ARCHIVE request, it will sample the file data version and
 903  * store it in \a copy.
 904  *
 905  * \return 0 on success.
 906  */
 907 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 908 {
 909         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 910         struct hsm_progress_kernel       hpk;
 911         int                              rc;
 912         ENTRY;
 913
 914         /* Forge a hsm_progress based on data from copy. */
 915         hpk.hpk_fid = copy->hc_hai.hai_fid;
 916         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 917         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 918         hpk.hpk_extent.length = 0;
 919         hpk.hpk_flags = 0;
 920         hpk.hpk_errval = 0;
 921         hpk.hpk_data_version = 0;
 922
 923
 924         /* For archive request, we need to read the current file version. */
 925         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 926                 struct inode    *inode;
 927                 __u64            data_version = 0;
 928
 929                 /* Get inode for this fid */
 930                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 931                 if (IS_ERR(inode)) {
 932                         hpk.hpk_flags |= HP_FLAG_RETRY;
 933                         /* hpk_errval is >= 0 */
 934                         hpk.hpk_errval = -PTR_ERR(inode);
 935                         GOTO(progress, rc = PTR_ERR(inode));
 936                 }
 937
 938                 /* Read current file data version */
 939                 rc = ll_data_version(inode, &data_version, 1);
 940                 iput(inode);
 941                 if (rc != 0) {
 942                         CDEBUG(D_HSM, "Could not read file data version of "
 943                                       DFID" (rc = %d). Archive request ("
 944                                       LPX64") could not be done.\n",
 945                                       PFID(&copy->hc_hai.hai_fid), rc,
 946                                       copy->hc_hai.hai_cookie);
 947                         hpk.hpk_flags |= HP_FLAG_RETRY;
 948                         /* hpk_errval must be >= 0 */
 949                         hpk.hpk_errval = -rc;
 950                         GOTO(progress, rc);
 951                 }
 952
 953                 /* Store it the hsm_copy for later copytool use.
 954                  * Always modified even if no lsm. */
 955                 copy->hc_data_version = data_version;
 956         }
 957
 958 progress:
 959         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 960                            &hpk, NULL);
 961
 962         RETURN(rc);
 963 }
 964
 965 /**
 966  * Generic handler to do any post-copy work.
 967  *
 968  * It will send the last hsm_progress update to coordinator to inform it
 969  * that copy is finished and whether it was successful or not.
 970  *
 971  * Moreover,
 972  * - for ARCHIVE request, it will sample the file data version and compare it
 973  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 974  *   will be considered as failed.
 975  * - for RESTORE request, it will sample the file data version and send it to
 976  *   coordinator which is useful if the file was imported as 'released'.
 977  *
 978  * \return 0 on success.
 979  */
 980 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 981 {
 982         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 983         struct hsm_progress_kernel       hpk;
 984         int                              rc;
 985         ENTRY;
 986
 987         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 988         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 989          * initialized if copy_end was called with copy == NULL.
 990          */
 991
 992         /* Forge a hsm_progress based on data from copy. */
 993         hpk.hpk_fid = copy->hc_hai.hai_fid;
 994         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 995         hpk.hpk_extent = copy->hc_hai.hai_extent;
 996         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 997         hpk.hpk_errval = copy->hc_errval;
 998         hpk.hpk_data_version = 0;
 999
1000         /* For archive request, we need to check the file data was not changed.
1001          *
1002          * For restore request, we need to send the file data version, this is
1003          * useful when the file was created using hsm_import.
1004          */
1005         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
1006              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
1007             (copy->hc_errval == 0)) {
1008                 struct inode    *inode;
1009                 __u64            data_version = 0;
1010
1011                 /* Get lsm for this fid */
1012                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
1013                 if (IS_ERR(inode)) {
1014                         hpk.hpk_flags |= HP_FLAG_RETRY;
1015                         /* hpk_errval must be >= 0 */
1016                         hpk.hpk_errval = -PTR_ERR(inode);
1017                         GOTO(progress, rc = PTR_ERR(inode));
1018                 }
1019
1020                 rc = ll_data_version(inode, &data_version,
1021                                      copy->hc_hai.hai_action == HSMA_ARCHIVE);
1022                 iput(inode);
1023                 if (rc) {
1024                         CDEBUG(D_HSM, "Could not read file data version. "
1025                                       "Request could not be confirmed.\n");
1026                         if (hpk.hpk_errval == 0)
1027                                 hpk.hpk_errval = rc;
1028                         GOTO(progress, rc);
1029                 }
1030
1031                 /* Store it the hsm_copy for later copytool use.
1032                  * Always modified even if no lsm. */
1033                 hpk.hpk_data_version = data_version;
1034
1035                 /* File could have been stripped during archiving, so we need
1036                  * to check anyway. */
1037                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1038                     (copy->hc_data_version != data_version)) {
1039                         CDEBUG(D_HSM, "File data version mismatched. "
1040                               "File content was changed during archiving. "
1041                                DFID", start:"LPX64" current:"LPX64"\n",
1042                                PFID(&copy->hc_hai.hai_fid),
1043                                copy->hc_data_version, data_version);
1044                         /* File was changed, send error to cdt. Do not ask for
1045                          * retry because if a file is modified frequently,
1046                          * the cdt will loop on retried archive requests.
1047                          * The policy engine will ask for a new archive later
1048                          * when the file will not be modified for some tunable
1049                          * time */
1050                         /* we do not notify caller */
1051                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
1052                         /* hpk_errval must be >= 0 */
1053                         hpk.hpk_errval = EBUSY;
1054                 }
1055
1056         }
1057
1058 progress:
1059         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1060                            &hpk, NULL);
1061
1062         RETURN(rc);
1063 }
1064
1065
1066 static int copy_and_ioctl(int cmd, struct obd_export *exp, void *data, int len)
1067 {
1068         void *ptr;
1069         int rc;
1070
1071         OBD_ALLOC(ptr, len);
1072         if (ptr == NULL)
1073                 return -ENOMEM;
1074         if (cfs_copy_from_user(ptr, data, len)) {
1075                 OBD_FREE(ptr, len);
1076                 return -EFAULT;
1077         }
1078         rc = obd_iocontrol(cmd, exp, len, data, NULL);
1079         OBD_FREE(ptr, len);
1080         return rc;
1081 }
1082
1083 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1084 {
1085         int cmd = qctl->qc_cmd;
1086         int type = qctl->qc_type;
1087         int id = qctl->qc_id;
1088         int valid = qctl->qc_valid;
1089         int rc = 0;
1090         ENTRY;
1091
1092         switch (cmd) {
1093         case LUSTRE_Q_INVALIDATE:
1094         case LUSTRE_Q_FINVALIDATE:
1095         case Q_QUOTAON:
1096         case Q_QUOTAOFF:
1097         case Q_SETQUOTA:
1098         case Q_SETINFO:
1099                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1100                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1101                         RETURN(-EPERM);
1102                 break;
1103         case Q_GETQUOTA:
1104                 if (((type == USRQUOTA && cfs_curproc_euid() != id) ||
1105                      (type == GRPQUOTA && !in_egroup_p(id))) &&
1106                     (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1107                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
1108                         RETURN(-EPERM);
1109                 break;
1110         case Q_GETINFO:
1111                 break;
1112         default:
1113                 CERROR("unsupported quotactl op: %#x\n", cmd);
1114                 RETURN(-ENOTTY);
1115         }
1116
1117         if (valid != QC_GENERAL) {
1118                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1119                         RETURN(-EOPNOTSUPP);
1120
1121                 if (cmd == Q_GETINFO)
1122                         qctl->qc_cmd = Q_GETOINFO;
1123                 else if (cmd == Q_GETQUOTA)
1124                         qctl->qc_cmd = Q_GETOQUOTA;
1125                 else
1126                         RETURN(-EINVAL);
1127
1128                 switch (valid) {
1129                 case QC_MDTIDX:
1130                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1131                                            sizeof(*qctl), qctl, NULL);
1132                         break;
1133                 case QC_OSTIDX:
1134                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1135                                            sizeof(*qctl), qctl, NULL);
1136                         break;
1137                 case QC_UUID:
1138                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1139                                            sizeof(*qctl), qctl, NULL);
1140                         if (rc == -EAGAIN)
1141                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1142                                                    sbi->ll_dt_exp,
1143                                                    sizeof(*qctl), qctl, NULL);
1144                         break;
1145                 default:
1146                         rc = -EINVAL;
1147                         break;
1148                 }
1149
1150                 if (rc)
1151                         RETURN(rc);
1152
1153                 qctl->qc_cmd = cmd;
1154         } else {
1155                 struct obd_quotactl *oqctl;
1156
1157                 OBD_ALLOC_PTR(oqctl);
1158                 if (oqctl == NULL)
1159                         RETURN(-ENOMEM);
1160
1161                 QCTL_COPY(oqctl, qctl);
1162                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1163                 if (rc) {
1164                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
1165                                 oqctl->qc_cmd = Q_QUOTAOFF;
1166                                 obd_quotactl(sbi->ll_md_exp, oqctl);
1167                         }
1168                         OBD_FREE_PTR(oqctl);
1169                         RETURN(rc);
1170                 }
1171                 /* If QIF_SPACE is not set, client should collect the
1172                  * space usage from OSSs by itself */
1173                 if (cmd == Q_GETQUOTA &&
1174                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1175                     !oqctl->qc_dqblk.dqb_curspace) {
1176                         struct obd_quotactl *oqctl_tmp;
1177
1178                         OBD_ALLOC_PTR(oqctl_tmp);
1179                         if (oqctl_tmp == NULL)
1180                                 GOTO(out, rc = -ENOMEM);
1181
1182                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1183                         oqctl_tmp->qc_id = oqctl->qc_id;
1184                         oqctl_tmp->qc_type = oqctl->qc_type;
1185
1186                         /* collect space usage from OSTs */
1187                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1188                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1189                         if (!rc || rc == -EREMOTEIO) {
1190                                 oqctl->qc_dqblk.dqb_curspace =
1191                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1192                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1193                         }
1194
1195                         /* collect space & inode usage from MDTs */
1196                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1197                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1198                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1199                         if (!rc || rc == -EREMOTEIO) {
1200                                 oqctl->qc_dqblk.dqb_curspace +=
1201                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1202                                 oqctl->qc_dqblk.dqb_curinodes =
1203                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1204                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1205                         } else {
1206                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1207                         }
1208
1209                         OBD_FREE_PTR(oqctl_tmp);
1210                 }
1211 out:
1212                 QCTL_COPY(qctl, oqctl);
1213                 OBD_FREE_PTR(oqctl);
1214         }
1215
1216         RETURN(rc);
1217 }
1218
1219 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1220 {
1221         struct inode *inode = file->f_dentry->d_inode;
1222         struct ll_sb_info *sbi = ll_i2sbi(inode);
1223         struct obd_ioctl_data *data;
1224         int rc = 0;
1225         ENTRY;
1226
1227         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1228                inode->i_ino, inode->i_generation, inode, cmd);
1229
1230         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1231         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1232                 return -ENOTTY;
1233
1234         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1235         switch(cmd) {
1236         case FSFILT_IOC_GETFLAGS:
1237         case FSFILT_IOC_SETFLAGS:
1238                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1239         case FSFILT_IOC_GETVERSION_OLD:
1240         case FSFILT_IOC_GETVERSION:
1241                 RETURN(put_user(inode->i_generation, (int *)arg));
1242         /* We need to special case any other ioctls we want to handle,
1243          * to send them to the MDS/OST as appropriate and to properly
1244          * network encode the arg field.
1245         case FSFILT_IOC_SETVERSION_OLD:
1246         case FSFILT_IOC_SETVERSION:
1247         */
1248         case LL_IOC_GET_MDTIDX: {
1249                 int mdtidx;
1250
1251                 mdtidx = ll_get_mdt_idx(inode);
1252                 if (mdtidx < 0)
1253                         RETURN(mdtidx);
1254
1255                 if (put_user((int)mdtidx, (int*)arg))
1256                         RETURN(-EFAULT);
1257
1258                 return 0;
1259         }
1260         case IOC_MDC_LOOKUP: {
1261                 struct ptlrpc_request *request = NULL;
1262                 int namelen, len = 0;
1263                 char *buf = NULL;
1264                 char *filename;
1265                 struct md_op_data *op_data;
1266
1267                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1268                 if (rc)
1269                         RETURN(rc);
1270                 data = (void *)buf;
1271
1272                 filename = data->ioc_inlbuf1;
1273                 namelen = strlen(filename);
1274
1275                 if (namelen < 1) {
1276                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1277                         GOTO(out_free, rc = -EINVAL);
1278                 }
1279
1280                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1281                                              0, LUSTRE_OPC_ANY, NULL);
1282                 if (IS_ERR(op_data))
1283                         GOTO(out_free, rc = PTR_ERR(op_data));
1284
1285                 op_data->op_valid = OBD_MD_FLID;
1286                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1287                 ll_finish_md_op_data(op_data);
1288                 if (rc < 0) {
1289                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1290                         GOTO(out_free, rc);
1291                 }
1292                 ptlrpc_req_finished(request);
1293                 EXIT;
1294 out_free:
1295                 obd_ioctl_freedata(buf, len);
1296                 return rc;
1297         }
1298         case LL_IOC_LMV_SETSTRIPE: {
1299                 struct lmv_user_md  *lum;
1300                 char            *buf = NULL;
1301                 char            *filename;
1302                 int              namelen = 0;
1303                 int              lumlen = 0;
1304                 int              len;
1305                 int              rc;
1306
1307                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1308                 if (rc)
1309                         RETURN(rc);
1310
1311                 data = (void *)buf;
1312                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1313                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1314                         GOTO(lmv_out_free, rc = -EINVAL);
1315
1316                 filename = data->ioc_inlbuf1;
1317                 namelen = data->ioc_inllen1;
1318
1319                 if (namelen < 1) {
1320                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1321                         GOTO(lmv_out_free, rc = -EINVAL);
1322                 }
1323                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1324                 lumlen = data->ioc_inllen2;
1325
1326                 if (lum->lum_magic != LMV_USER_MAGIC ||
1327                     lumlen != sizeof(*lum)) {
1328                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1329                                filename, lum->lum_magic, lumlen, -EFAULT);
1330                         GOTO(lmv_out_free, rc = -EINVAL);
1331                 }
1332
1333                 /**
1334                  * ll_dir_setdirstripe will be used to set dir stripe
1335                  *  mdc_create--->mdt_reint_create (with dirstripe)
1336                  */
1337                 rc = ll_dir_setdirstripe(inode, lum, filename);
1338 lmv_out_free:
1339                 obd_ioctl_freedata(buf, len);
1340                 RETURN(rc);
1341
1342         }
1343         case LL_IOC_LOV_SETSTRIPE: {
1344                 struct lov_user_md_v3 lumv3;
1345                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1346                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1347                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1348
1349                 int set_default = 0;
1350
1351                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1352                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1353                         sizeof(lumv3p->lmm_objects[0]));
1354                 /* first try with v1 which is smaller than v3 */
1355                 if (cfs_copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1356                         RETURN(-EFAULT);
1357
1358                 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1359                         if (cfs_copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1360                                 RETURN(-EFAULT);
1361                 }
1362
1363                 if (inode->i_sb->s_root == file->f_dentry)
1364                         set_default = 1;
1365
1366                 /* in v1 and v3 cases lumv1 points to data */
1367                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1368
1369                 RETURN(rc);
1370         }
1371         case LL_IOC_LMV_GETSTRIPE: {
1372                 struct lmv_user_md *lump = (struct lmv_user_md *)arg;
1373                 struct lmv_user_md lum;
1374                 struct lmv_user_md *tmp;
1375                 int lum_size;
1376                 int rc = 0;
1377                 int mdtindex;
1378
1379                 if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
1380                         RETURN(-EFAULT);
1381
1382                 if (lum.lum_magic != LMV_MAGIC_V1)
1383                         RETURN(-EINVAL);
1384
1385                 lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
1386                 OBD_ALLOC(tmp, lum_size);
1387                 if (tmp == NULL)
1388                         GOTO(free_lmv, rc = -ENOMEM);
1389
1390                 memcpy(tmp, &lum, sizeof(lum));
1391                 tmp->lum_type = LMV_STRIPE_TYPE;
1392                 tmp->lum_stripe_count = 1;
1393                 mdtindex = ll_get_mdt_idx(inode);
1394                 if (mdtindex < 0)
1395                         GOTO(free_lmv, rc = -ENOMEM);
1396
1397                 tmp->lum_stripe_offset = mdtindex;
1398                 tmp->lum_objects[0].lum_mds = mdtindex;
1399                 memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
1400                        sizeof(struct lu_fid));
1401                 if (copy_to_user((void *)arg, tmp, lum_size))
1402                         GOTO(free_lmv, rc = -EFAULT);
1403 free_lmv:
1404                 if (tmp)
1405                         OBD_FREE(tmp, lum_size);
1406                 RETURN(rc);
1407         }
1408         case LL_IOC_REMOVE_ENTRY: {
1409                 char            *filename = NULL;
1410                 int              namelen = 0;
1411                 int              rc;
1412
1413                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1414                  * unsupported server, which might crash the server(LU-2730),
1415                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1416                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1417                  * server will support REINT_RMENTRY XXX*/
1418                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1419                         return -ENOTSUPP;
1420
1421                 filename = getname((const char *)arg);
1422                 if (IS_ERR(filename))
1423                         RETURN(PTR_ERR(filename));
1424
1425                 namelen = strlen(filename);
1426                 if (namelen < 1)
1427                         GOTO(out_rmdir, rc = -EINVAL);
1428
1429                 rc = ll_rmdir_entry(inode, filename, namelen);
1430 out_rmdir:
1431                 if (filename)
1432                         putname(filename);
1433                 RETURN(rc);
1434         }
1435         case LL_IOC_LOV_SWAP_LAYOUTS:
1436                 RETURN(-EPERM);
1437         case LL_IOC_OBD_STATFS:
1438                 RETURN(ll_obd_statfs(inode, (void *)arg));
1439         case LL_IOC_LOV_GETSTRIPE:
1440         case LL_IOC_MDC_GETINFO:
1441         case IOC_MDC_GETFILEINFO:
1442         case IOC_MDC_GETFILESTRIPE: {
1443                 struct ptlrpc_request *request = NULL;
1444                 struct lov_user_md *lump;
1445                 struct lov_mds_md *lmm = NULL;
1446                 struct mdt_body *body;
1447                 char *filename = NULL;
1448                 int lmmsize;
1449
1450                 if (cmd == IOC_MDC_GETFILEINFO ||
1451                     cmd == IOC_MDC_GETFILESTRIPE) {
1452                         filename = getname((const char *)arg);
1453                         if (IS_ERR(filename))
1454                                 RETURN(PTR_ERR(filename));
1455
1456                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1457                                                       &lmmsize, &request);
1458                 } else {
1459                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1460                 }
1461
1462                 if (request) {
1463                         body = req_capsule_server_get(&request->rq_pill,
1464                                                       &RMF_MDT_BODY);
1465                         LASSERT(body != NULL);
1466                 } else {
1467                         GOTO(out_req, rc);
1468                 }
1469
1470                 if (rc < 0) {
1471                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1472                                                cmd == LL_IOC_MDC_GETINFO))
1473                                 GOTO(skip_lmm, rc = 0);
1474                         else
1475                                 GOTO(out_req, rc);
1476                 }
1477
1478                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1479                     cmd == LL_IOC_LOV_GETSTRIPE) {
1480                         lump = (struct lov_user_md *)arg;
1481                 } else {
1482                         struct lov_user_mds_data *lmdp;
1483                         lmdp = (struct lov_user_mds_data *)arg;
1484                         lump = &lmdp->lmd_lmm;
1485                 }
1486                 if (cfs_copy_to_user(lump, lmm, lmmsize)) {
1487                         if (cfs_copy_to_user(lump, lmm, sizeof(*lump)))
1488                                 GOTO(out_req, rc = -EFAULT);
1489                         rc = -EOVERFLOW;
1490                 }
1491         skip_lmm:
1492                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1493                         struct lov_user_mds_data *lmdp;
1494                         lstat_t st = { 0 };
1495
1496                         st.st_dev     = inode->i_sb->s_dev;
1497                         st.st_mode    = body->mode;
1498                         st.st_nlink   = body->nlink;
1499                         st.st_uid     = body->uid;
1500                         st.st_gid     = body->gid;
1501                         st.st_rdev    = body->rdev;
1502                         st.st_size    = body->size;
1503                         st.st_blksize = CFS_PAGE_SIZE;
1504                         st.st_blocks  = body->blocks;
1505                         st.st_atime   = body->atime;
1506                         st.st_mtime   = body->mtime;
1507                         st.st_ctime   = body->ctime;
1508                         st.st_ino     = inode->i_ino;
1509
1510                         lmdp = (struct lov_user_mds_data *)arg;
1511                         if (cfs_copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1512                                 GOTO(out_req, rc = -EFAULT);
1513                 }
1514
1515                 EXIT;
1516         out_req:
1517                 ptlrpc_req_finished(request);
1518                 if (filename)
1519                         putname(filename);
1520                 return rc;
1521         }
1522         case IOC_LOV_GETINFO: {
1523                 struct lov_user_mds_data *lumd;
1524                 struct lov_stripe_md *lsm;
1525                 struct lov_user_md *lum;
1526                 struct lov_mds_md *lmm;
1527                 int lmmsize;
1528                 lstat_t st;
1529
1530                 lumd = (struct lov_user_mds_data *)arg;
1531                 lum = &lumd->lmd_lmm;
1532
1533                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1534                 if (rc)
1535                         RETURN(rc);
1536
1537                 OBD_ALLOC_LARGE(lmm, lmmsize);
1538                 if (cfs_copy_from_user(lmm, lum, lmmsize))
1539                         GOTO(free_lmm, rc = -EFAULT);
1540
1541                 switch (lmm->lmm_magic) {
1542                 case LOV_USER_MAGIC_V1:
1543                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1544                                 break;
1545                         /* swab objects first so that stripes num will be sane */
1546                         lustre_swab_lov_user_md_objects(
1547                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1548                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1549                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1550                         break;
1551                 case LOV_USER_MAGIC_V3:
1552                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1553                                 break;
1554                         /* swab objects first so that stripes num will be sane */
1555                         lustre_swab_lov_user_md_objects(
1556                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1557                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1558                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1559                         break;
1560                 default:
1561                         GOTO(free_lmm, rc = -EINVAL);
1562                 }
1563
1564                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1565                 if (rc < 0)
1566                         GOTO(free_lmm, rc = -ENOMEM);
1567
1568                 /* Perform glimpse_size operation. */
1569                 memset(&st, 0, sizeof(st));
1570
1571                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1572                 if (rc)
1573                         GOTO(free_lsm, rc);
1574
1575                 if (cfs_copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1576                         GOTO(free_lsm, rc = -EFAULT);
1577
1578                 EXIT;
1579         free_lsm:
1580                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1581         free_lmm:
1582                 OBD_FREE_LARGE(lmm, lmmsize);
1583                 return rc;
1584         }
1585         case OBD_IOC_LLOG_CATINFO: {
1586                 RETURN(-EOPNOTSUPP);
1587         }
1588         case OBD_IOC_QUOTACHECK: {
1589                 struct obd_quotactl *oqctl;
1590                 int error = 0;
1591
1592                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1593                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1594                         RETURN(-EPERM);
1595
1596                 OBD_ALLOC_PTR(oqctl);
1597                 if (!oqctl)
1598                         RETURN(-ENOMEM);
1599                 oqctl->qc_type = arg;
1600                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1601                 if (rc < 0) {
1602                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1603                         error = rc;
1604                 }
1605
1606                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1607                 if (rc < 0)
1608                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1609
1610                 OBD_FREE_PTR(oqctl);
1611                 return error ?: rc;
1612         }
1613         case OBD_IOC_POLL_QUOTACHECK: {
1614                 struct if_quotacheck *check;
1615
1616                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1617                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1618                         RETURN(-EPERM);
1619
1620                 OBD_ALLOC_PTR(check);
1621                 if (!check)
1622                         RETURN(-ENOMEM);
1623
1624                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1625                                    NULL);
1626                 if (rc) {
1627                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1628                         if (cfs_copy_to_user((void *)arg, check,
1629                                              sizeof(*check)))
1630                                 CDEBUG(D_QUOTA, "cfs_copy_to_user failed\n");
1631                         GOTO(out_poll, rc);
1632                 }
1633
1634                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1635                                    NULL);
1636                 if (rc) {
1637                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1638                         if (cfs_copy_to_user((void *)arg, check,
1639                                              sizeof(*check)))
1640                                 CDEBUG(D_QUOTA, "cfs_copy_to_user failed\n");
1641                         GOTO(out_poll, rc);
1642                 }
1643         out_poll:
1644                 OBD_FREE_PTR(check);
1645                 RETURN(rc);
1646         }
1647 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
1648         case LL_IOC_QUOTACTL_18: {
1649                 /* copy the old 1.x quota struct for internal use, then copy
1650                  * back into old format struct.  For 1.8 compatibility. */
1651                 struct if_quotactl_18 *qctl_18;
1652                 struct if_quotactl *qctl_20;
1653
1654                 OBD_ALLOC_PTR(qctl_18);
1655                 if (!qctl_18)
1656                         RETURN(-ENOMEM);
1657
1658                 OBD_ALLOC_PTR(qctl_20);
1659                 if (!qctl_20)
1660                         GOTO(out_quotactl_18, rc = -ENOMEM);
1661
1662                 if (cfs_copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1663                         GOTO(out_quotactl_20, rc = -ENOMEM);
1664
1665                 QCTL_COPY(qctl_20, qctl_18);
1666                 qctl_20->qc_idx = 0;
1667
1668                 /* XXX: dqb_valid was borrowed as a flag to mark that
1669                  *      only mds quota is wanted */
1670                 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1671                     qctl_18->qc_dqblk.dqb_valid) {
1672                         qctl_20->qc_valid = QC_MDTIDX;
1673                         qctl_20->qc_dqblk.dqb_valid = 0;
1674                 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1675                         qctl_20->qc_valid = QC_UUID;
1676                         qctl_20->obd_uuid = qctl_18->obd_uuid;
1677                 } else {
1678                         qctl_20->qc_valid = QC_GENERAL;
1679                 }
1680
1681                 rc = quotactl_ioctl(sbi, qctl_20);
1682
1683                 if (rc == 0) {
1684                         QCTL_COPY(qctl_18, qctl_20);
1685                         qctl_18->obd_uuid = qctl_20->obd_uuid;
1686
1687                         if (cfs_copy_to_user((void *)arg, qctl_18,
1688                                              sizeof(*qctl_18)))
1689                                 rc = -EFAULT;
1690                 }
1691
1692         out_quotactl_20:
1693                 OBD_FREE_PTR(qctl_20);
1694         out_quotactl_18:
1695                 OBD_FREE_PTR(qctl_18);
1696                 RETURN(rc);
1697         }
1698 #else
1699 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1700 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
1701         case LL_IOC_QUOTACTL: {
1702                 struct if_quotactl *qctl;
1703
1704                 OBD_ALLOC_PTR(qctl);
1705                 if (!qctl)
1706                         RETURN(-ENOMEM);
1707
1708                 if (cfs_copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1709                         GOTO(out_quotactl, rc = -EFAULT);
1710
1711                 rc = quotactl_ioctl(sbi, qctl);
1712
1713                 if (rc == 0 && cfs_copy_to_user((void *)arg,qctl,sizeof(*qctl)))
1714                         rc = -EFAULT;
1715
1716         out_quotactl:
1717                 OBD_FREE_PTR(qctl);
1718                 RETURN(rc);
1719         }
1720         case OBD_IOC_GETDTNAME:
1721         case OBD_IOC_GETMDNAME:
1722                 RETURN(ll_get_obd_name(inode, cmd, arg));
1723         case LL_IOC_FLUSHCTX:
1724                 RETURN(ll_flush_ctx(inode));
1725 #ifdef CONFIG_FS_POSIX_ACL
1726         case LL_IOC_RMTACL: {
1727             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1728                 inode == inode->i_sb->s_root->d_inode) {
1729                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1730
1731                 LASSERT(fd != NULL);
1732                 rc = rct_add(&sbi->ll_rct, cfs_curproc_pid(), arg);
1733                 if (!rc)
1734                         fd->fd_flags |= LL_FILE_RMTACL;
1735                 RETURN(rc);
1736             } else
1737                 RETURN(0);
1738         }
1739 #endif
1740         case LL_IOC_GETOBDCOUNT: {
1741                 int count, vallen;
1742                 struct obd_export *exp;
1743
1744                 if (cfs_copy_from_user(&count, (int *)arg, sizeof(int)))
1745                         RETURN(-EFAULT);
1746
1747                 /* get ost count when count is zero, get mdt count otherwise */
1748                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1749                 vallen = sizeof(count);
1750                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1751                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1752                 if (rc) {
1753                         CERROR("get target count failed: %d\n", rc);
1754                         RETURN(rc);
1755                 }
1756
1757                 if (cfs_copy_to_user((int *)arg, &count, sizeof(int)))
1758                         RETURN(-EFAULT);
1759
1760                 RETURN(0);
1761         }
1762         case LL_IOC_PATH2FID:
1763                 if (cfs_copy_to_user((void *)arg, ll_inode2fid(inode),
1764                                      sizeof(struct lu_fid)))
1765                         RETURN(-EFAULT);
1766                 RETURN(0);
1767         case LL_IOC_GET_CONNECT_FLAGS: {
1768                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1769         }
1770         case OBD_IOC_CHANGELOG_SEND:
1771         case OBD_IOC_CHANGELOG_CLEAR:
1772                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1773                                     sizeof(struct ioc_changelog));
1774                 RETURN(rc);
1775         case OBD_IOC_FID2PATH:
1776                 RETURN(ll_fid2path(inode, (void *)arg));
1777         case LL_IOC_HSM_REQUEST: {
1778                 struct hsm_user_request *hur;
1779                 int                      totalsize;
1780
1781                 OBD_ALLOC_PTR(hur);
1782                 if (hur == NULL)
1783                         RETURN(-ENOMEM);
1784
1785                 /* We don't know the true size yet; copy the fixed-size part */
1786                 if (cfs_copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1787                         OBD_FREE_PTR(hur);
1788                         RETURN(-EFAULT);
1789                 }
1790
1791                 /* Compute the whole struct size */
1792                 totalsize = hur_len(hur);
1793                 OBD_FREE_PTR(hur);
1794                 OBD_ALLOC_LARGE(hur, totalsize);
1795                 if (hur == NULL)
1796                         RETURN(-ENOMEM);
1797
1798                 /* Copy the whole struct */
1799                 if (cfs_copy_from_user(hur, (void *)arg, totalsize)) {
1800                         OBD_FREE_LARGE(hur, totalsize);
1801                         RETURN(-EFAULT);
1802                 }
1803
1804                 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1805                                    hur, NULL);
1806
1807                 OBD_FREE_LARGE(hur, totalsize);
1808
1809                 RETURN(rc);
1810         }
1811         case LL_IOC_HSM_PROGRESS: {
1812                 struct hsm_progress_kernel      hpk;
1813                 struct hsm_progress             hp;
1814
1815                 if (cfs_copy_from_user(&hp, (void *)arg, sizeof(hp)))
1816                         RETURN(-EFAULT);
1817
1818                 hpk.hpk_fid = hp.hp_fid;
1819                 hpk.hpk_cookie = hp.hp_cookie;
1820                 hpk.hpk_extent = hp.hp_extent;
1821                 hpk.hpk_flags = hp.hp_flags;
1822                 hpk.hpk_errval = hp.hp_errval;
1823                 hpk.hpk_data_version = 0;
1824
1825                 /* File may not exist in Lustre; all progress
1826                  * reported to Lustre root */
1827                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1828                                    NULL);
1829                 RETURN(rc);
1830         }
1831         case LL_IOC_HSM_CT_START:
1832                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1833                                     sizeof(struct lustre_kernelcomm));
1834                 RETURN(rc);
1835
1836         case LL_IOC_HSM_COPY_START: {
1837                 struct hsm_copy *copy;
1838                 int              rc;
1839
1840                 OBD_ALLOC_PTR(copy);
1841                 if (copy == NULL)
1842                         RETURN(-ENOMEM);
1843                 if (cfs_copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1844                         OBD_FREE_PTR(copy);
1845                         RETURN(-EFAULT);
1846                 }
1847
1848                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1849                 if (cfs_copy_to_user((char *)arg, copy, sizeof(*copy)))
1850                         rc = -EFAULT;
1851
1852                 OBD_FREE_PTR(copy);
1853                 RETURN(rc);
1854         }
1855         case LL_IOC_HSM_COPY_END: {
1856                 struct hsm_copy *copy;
1857                 int              rc;
1858
1859                 OBD_ALLOC_PTR(copy);
1860                 if (copy == NULL)
1861                         RETURN(-ENOMEM);
1862                 if (cfs_copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1863                         OBD_FREE_PTR(copy);
1864                         RETURN(-EFAULT);
1865                 }
1866
1867                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1868                 if (cfs_copy_to_user((char *)arg, copy, sizeof(*copy)))
1869                         rc = -EFAULT;
1870
1871                 OBD_FREE_PTR(copy);
1872                 RETURN(rc);
1873         }
1874         default:
1875                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1876                                      (void *)arg));
1877         }
1878 }
1879
1880 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1881 {
1882         struct inode *inode = file->f_mapping->host;
1883         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1884         struct ll_sb_info *sbi = ll_i2sbi(inode);
1885         int api32 = ll_need_32bit_api(sbi);
1886         loff_t ret = -EINVAL;
1887         ENTRY;
1888
1889         mutex_lock(&inode->i_mutex);
1890         switch (origin) {
1891                 case SEEK_SET:
1892                         break;
1893                 case SEEK_CUR:
1894                         offset += file->f_pos;
1895                         break;
1896                 case SEEK_END:
1897                         if (offset > 0)
1898                                 GOTO(out, ret);
1899                         if (api32)
1900                                 offset += LL_DIR_END_OFF_32BIT;
1901                         else
1902                                 offset += LL_DIR_END_OFF;
1903                         break;
1904                 default:
1905                         GOTO(out, ret);
1906         }
1907
1908         if (offset >= 0 &&
1909             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1910              (!api32 && offset <= LL_DIR_END_OFF))) {
1911                 if (offset != file->f_pos) {
1912                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1913                             (!api32 && offset == LL_DIR_END_OFF))
1914                                 fd->lfd_pos = MDS_DIR_END_OFF;
1915                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1916                                 fd->lfd_pos = offset << 32;
1917                         else
1918                                 fd->lfd_pos = offset;
1919                         file->f_pos = offset;
1920                         file->f_version = 0;
1921                 }
1922                 ret = offset;
1923         }
1924         GOTO(out, ret);
1925
1926 out:
1927         mutex_unlock(&inode->i_mutex);
1928         return ret;
1929 }
1930
1931 int ll_dir_open(struct inode *inode, struct file *file)
1932 {
1933         ENTRY;
1934         RETURN(ll_file_open(inode, file));
1935 }
1936
1937 int ll_dir_release(struct inode *inode, struct file *file)
1938 {
1939         ENTRY;
1940         RETURN(ll_file_release(inode, file));
1941 }
1942
1943 struct file_operations ll_dir_operations = {
1944         .llseek   = ll_dir_seek,
1945         .open     = ll_dir_open,
1946         .release  = ll_dir_release,
1947         .read     = generic_read_dir,
1948         .readdir  = ll_readdir,
1949         .unlocked_ioctl   = ll_dir_ioctl,
1950         .fsync    = ll_fsync,
1951 };