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, 2013, 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 PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
 138  * lu_dirpage for this integrated page will be adjusted. See
 139  * lmv_adjust_dirpages().
 140  *
 141  */
 142
 143 /* returns the page unlocked, but with a reference */
 144 static int ll_dir_filler(void *_hash, struct page *page0)
 145 {
 146         struct inode *inode = page0->mapping->host;
 147         int hash64 = ll_i2sbi(inode)->ll_flags & LL_SBI_64BIT_HASH;
 148         struct obd_export *exp = ll_i2sbi(inode)->ll_md_exp;
 149         struct ptlrpc_request *request;
 150         struct mdt_body *body;
 151         struct md_op_data *op_data;
 152         __u64 hash = *((__u64 *)_hash);
 153         struct page **page_pool;
 154         struct page *page;
 155         struct lu_dirpage *dp;
 156         int max_pages = ll_i2sbi(inode)->ll_md_brw_size >> PAGE_CACHE_SHIFT;
 157         int nrdpgs = 0; /* number of pages read actually */
 158         int npages;
 159         int i;
 160         int rc;
 161         ENTRY;
 162
 163         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) hash "LPU64"\n",
 164                inode->i_ino, inode->i_generation, inode, hash);
 165
 166         LASSERT(max_pages > 0 && max_pages <= MD_MAX_BRW_PAGES);
 167
 168         OBD_ALLOC(page_pool, sizeof(page) * max_pages);
 169         if (page_pool != NULL) {
 170                 page_pool[0] = page0;
 171         } else {
 172                 page_pool = &page0;
 173                 max_pages = 1;
 174         }
 175         for (npages = 1; npages < max_pages; npages++) {
 176                 page = page_cache_alloc_cold(inode->i_mapping);
 177                 if (!page)
 178                         break;
 179                 page_pool[npages] = page;
 180         }
 181
 182         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 183                                      LUSTRE_OPC_ANY, NULL);
 184         op_data->op_npages = npages;
 185         op_data->op_offset = hash;
 186         rc = md_readpage(exp, op_data, page_pool, &request);
 187         ll_finish_md_op_data(op_data);
 188         if (rc == 0) {
 189                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 190                 /* Checked by mdc_readpage() */
 191                 LASSERT(body != NULL);
 192
 193                 if (body->valid & OBD_MD_FLSIZE)
 194                         cl_isize_write(inode, body->size);
 195
 196                 nrdpgs = (request->rq_bulk->bd_nob_transferred +
 197                           PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 198                 SetPageUptodate(page0);
 199         }
 200         unlock_page(page0);
 201         ptlrpc_req_finished(request);
 202
 203         CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages);
 204
 205         for (i = 1; i < npages; i++) {
 206                 unsigned long offset;
 207                 int ret;
 208
 209                 page = page_pool[i];
 210
 211                 if (rc < 0 || i >= nrdpgs) {
 212                         page_cache_release(page);
 213                         continue;
 214                 }
 215
 216                 SetPageUptodate(page);
 217
 218                 dp = kmap(page);
 219                 hash = le64_to_cpu(dp->ldp_hash_start);
 220                 kunmap(page);
 221
 222                 offset = hash_x_index(hash, hash64);
 223
 224                 prefetchw(&page->flags);
 225                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
 226                                             GFP_KERNEL);
 227                 if (ret == 0)
 228                         unlock_page(page);
 229                 else
 230                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
 231                                " %d\n", offset, ret);
 232                 page_cache_release(page);
 233         }
 234
 235         if (page_pool != &page0)
 236                 OBD_FREE(page_pool, sizeof(struct page *) * max_pages);
 237         EXIT;
 238         return rc;
 239 }
 240
 241 static void ll_check_page(struct inode *dir, struct page *page)
 242 {
 243         /* XXX: check page format later */
 244         SetPageChecked(page);
 245 }
 246
 247 void ll_release_page(struct page *page, int remove)
 248 {
 249         kunmap(page);
 250         if (remove) {
 251                 lock_page(page);
 252                 if (likely(page->mapping != NULL))
 253                         truncate_complete_page(page->mapping, page);
 254                 unlock_page(page);
 255         }
 256         page_cache_release(page);
 257 }
 258
 259 /*
 260  * Find, kmap and return page that contains given hash.
 261  */
 262 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
 263                                        __u64 *start, __u64 *end)
 264 {
 265         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 266         struct address_space *mapping = dir->i_mapping;
 267         /*
 268          * Complement of hash is used as an index so that
 269          * radix_tree_gang_lookup() can be used to find a page with starting
 270          * hash _smaller_ than one we are looking for.
 271          */
 272         unsigned long offset = hash_x_index(*hash, hash64);
 273         struct page *page;
 274         int found;
 275
 276         TREE_READ_LOCK_IRQ(mapping);
 277         found = radix_tree_gang_lookup(&mapping->page_tree,
 278                                        (void **)&page, offset, 1);
 279         if (found > 0) {
 280                 struct lu_dirpage *dp;
 281
 282                 page_cache_get(page);
 283                 TREE_READ_UNLOCK_IRQ(mapping);
 284                 /*
 285                  * In contrast to find_lock_page() we are sure that directory
 286                  * page cannot be truncated (while DLM lock is held) and,
 287                  * hence, can avoid restart.
 288                  *
 289                  * In fact, page cannot be locked here at all, because
 290                  * ll_dir_filler() does synchronous io.
 291                  */
 292                 wait_on_page_locked(page);
 293                 if (PageUptodate(page)) {
 294                         dp = kmap(page);
 295                         if (BITS_PER_LONG == 32 && hash64) {
 296                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 297                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 298                                 *hash  = *hash >> 32;
 299                         } else {
 300                                 *start = le64_to_cpu(dp->ldp_hash_start);
 301                                 *end   = le64_to_cpu(dp->ldp_hash_end);
 302                         }
 303                         LASSERTF(*start <= *hash, "start = "LPX64",end = "
 304                                  LPX64",hash = "LPX64"\n", *start, *end, *hash);
 305                         CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash "LPU64"\n",
 306                                offset, *start, *end, *hash);
 307                         if (*hash > *end) {
 308                                 ll_release_page(page, 0);
 309                                 page = NULL;
 310                         } else if (*end != *start && *hash == *end) {
 311                                 /*
 312                                  * upon hash collision, remove this page,
 313                                  * otherwise put page reference, and
 314                                  * ll_get_dir_page() will issue RPC to fetch
 315                                  * the page we want.
 316                                  */
 317                                 ll_release_page(page,
 318                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
 319                                 page = NULL;
 320                         }
 321                 } else {
 322                         page_cache_release(page);
 323                         page = ERR_PTR(-EIO);
 324                 }
 325
 326         } else {
 327                 TREE_READ_UNLOCK_IRQ(mapping);
 328                 page = NULL;
 329         }
 330         return page;
 331 }
 332
 333 struct page *ll_get_dir_page(struct inode *dir, __u64 hash,
 334                              struct ll_dir_chain *chain)
 335 {
 336         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 337         struct address_space *mapping = dir->i_mapping;
 338         struct lustre_handle lockh;
 339         struct lu_dirpage *dp;
 340         struct page *page;
 341         ldlm_mode_t mode;
 342         int rc;
 343         __u64 start = 0;
 344         __u64 end = 0;
 345         __u64 lhash = hash;
 346         struct ll_inode_info *lli = ll_i2info(dir);
 347         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 348
 349         mode = LCK_PR;
 350         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 351                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
 352         if (!rc) {
 353                 struct ldlm_enqueue_info einfo = {
 354                         .ei_type = LDLM_IBITS,
 355                         .ei_mode = mode,
 356                         .ei_cb_bl = ll_md_blocking_ast,
 357                         .ei_cb_cp = ldlm_completion_ast,
 358                 };
 359                 struct lookup_intent it = { .it_op = IT_READDIR };
 360                 struct ptlrpc_request *request;
 361                 struct md_op_data *op_data;
 362
 363                 op_data = ll_prep_md_op_data(NULL, dir, NULL, NULL, 0, 0,
 364                 LUSTRE_OPC_ANY, NULL);
 365                 if (IS_ERR(op_data))
 366                         return (void *)op_data;
 367
 368                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
 369                                 op_data, &lockh, NULL, 0, NULL, 0);
 370
 371                 ll_finish_md_op_data(op_data);
 372
 373                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 374                 if (request)
 375                         ptlrpc_req_finished(request);
 376                 if (rc < 0) {
 377                         CERROR("lock enqueue: "DFID" at "LPU64": rc %d\n",
 378                                 PFID(ll_inode2fid(dir)), hash, rc);
 379                         return ERR_PTR(rc);
 380                 }
 381
 382                 CDEBUG(D_INODE, "setting lr_lvb_inode to inode %p (%lu/%u)\n",
 383                        dir, dir->i_ino, dir->i_generation);
 384                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp,
 385                                  &it.d.lustre.it_lock_handle, dir, NULL);
 386         } else {
 387                 /* for cross-ref object, l_ast_data of the lock may not be set,
 388                  * we reset it here */
 389                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
 390                                  dir, NULL);
 391         }
 392         ldlm_lock_dump_handle(D_OTHER, &lockh);
 393
 394         mutex_lock(&lli->lli_readdir_mutex);
 395         page = ll_dir_page_locate(dir, &lhash, &start, &end);
 396         if (IS_ERR(page)) {
 397                 CERROR("dir page locate: "DFID" at "LPU64": rc %ld\n",
 398                        PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
 399                 GOTO(out_unlock, page);
 400         } else if (page != NULL) {
 401                 /*
 402                  * XXX nikita: not entirely correct handling of a corner case:
 403                  * suppose hash chain of entries with hash value HASH crosses
 404                  * border between pages P0 and P1. First both P0 and P1 are
 405                  * cached, seekdir() is called for some entry from the P0 part
 406                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 407                  * happens and finds P1, as it starts with matching hash
 408                  * value. Remaining entries from P0 part of the chain are
 409                  * skipped. (Is that really a bug?)
 410                  *
 411                  * Possible solutions: 0. don't cache P1 is such case, handle
 412                  * it as an "overflow" page. 1. invalidate all pages at
 413                  * once. 2. use HASH|1 as an index for P1.
 414                  */
 415                 GOTO(hash_collision, page);
 416         }
 417
 418         page = read_cache_page(mapping, hash_x_index(hash, hash64),
 419                                ll_dir_filler, &lhash);
 420         if (IS_ERR(page)) {
 421                 CERROR("read cache page: "DFID" at "LPU64": rc %ld\n",
 422                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 423                 GOTO(out_unlock, page);
 424         }
 425
 426         wait_on_page_locked(page);
 427         (void)kmap(page);
 428         if (!PageUptodate(page)) {
 429                 CERROR("page not updated: "DFID" at "LPU64": rc %d\n",
 430                        PFID(ll_inode2fid(dir)), hash, -5);
 431                 goto fail;
 432         }
 433         if (!PageChecked(page))
 434                 ll_check_page(dir, page);
 435         if (PageError(page)) {
 436                 CERROR("page error: "DFID" at "LPU64": rc %d\n",
 437                        PFID(ll_inode2fid(dir)), hash, -5);
 438                 goto fail;
 439         }
 440 hash_collision:
 441         dp = page_address(page);
 442         if (BITS_PER_LONG == 32 && hash64) {
 443                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 444                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 445                 lhash = hash >> 32;
 446         } else {
 447                 start = le64_to_cpu(dp->ldp_hash_start);
 448                 end   = le64_to_cpu(dp->ldp_hash_end);
 449                 lhash = hash;
 450         }
 451         if (end == start) {
 452                 LASSERT(start == lhash);
 453                 CWARN("Page-wide hash collision: "LPU64"\n", end);
 454                 if (BITS_PER_LONG == 32 && hash64)
 455                         CWARN("Real page-wide hash collision at ["LPU64" "LPU64
 456                               "] with hash "LPU64"\n",
 457                               le64_to_cpu(dp->ldp_hash_start),
 458                               le64_to_cpu(dp->ldp_hash_end), hash);
 459                 /*
 460                  * Fetch whole overflow chain...
 461                  *
 462                  * XXX not yet.
 463                  */
 464                 goto fail;
 465         }
 466 out_unlock:
 467         mutex_unlock(&lli->lli_readdir_mutex);
 468         ldlm_lock_decref(&lockh, mode);
 469         return page;
 470
 471 fail:
 472         ll_release_page(page, 1);
 473         page = ERR_PTR(-EIO);
 474         goto out_unlock;
 475 }
 476
 477 int ll_dir_read(struct inode *inode, __u64 *_pos, void *cookie,
 478                 filldir_t filldir)
 479 {
 480         struct ll_inode_info *info       = ll_i2info(inode);
 481         struct ll_sb_info    *sbi        = ll_i2sbi(inode);
 482         __u64                 pos        = *_pos;
 483         int                   api32      = ll_need_32bit_api(sbi);
 484         int                   hash64     = sbi->ll_flags & LL_SBI_64BIT_HASH;
 485         struct page          *page;
 486         struct ll_dir_chain   chain;
 487         int                   done = 0;
 488         int                   rc = 0;
 489         ENTRY;
 490
 491         ll_dir_chain_init(&chain);
 492
 493         page = ll_get_dir_page(inode, pos, &chain);
 494
 495         while (rc == 0 && !done) {
 496                 struct lu_dirpage *dp;
 497                 struct lu_dirent  *ent;
 498
 499                 if (!IS_ERR(page)) {
 500                         /*
 501                          * If page is empty (end of directory is reached),
 502                          * use this value.
 503                          */
 504                         __u64 hash = MDS_DIR_END_OFF;
 505                         __u64 next;
 506
 507                         dp = page_address(page);
 508                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 509                              ent = lu_dirent_next(ent)) {
 510                                 __u16          type;
 511                                 int            namelen;
 512                                 struct lu_fid  fid;
 513                                 __u64          lhash;
 514                                 __u64          ino;
 515
 516                                 /*
 517                                  * XXX: implement correct swabbing here.
 518                                  */
 519
 520                                 hash = le64_to_cpu(ent->lde_hash);
 521                                 if (hash < pos)
 522                                         /*
 523                                          * Skip until we find target hash
 524                                          * value.
 525                                          */
 526                                         continue;
 527
 528                                 namelen = le16_to_cpu(ent->lde_namelen);
 529                                 if (namelen == 0)
 530                                         /*
 531                                          * Skip dummy record.
 532                                          */
 533                                         continue;
 534
 535                                 if (api32 && hash64)
 536                                         lhash = hash >> 32;
 537                                 else
 538                                         lhash = hash;
 539                                 fid_le_to_cpu(&fid, &ent->lde_fid);
 540                                 ino = cl_fid_build_ino(&fid, api32);
 541                                 type = ll_dirent_type_get(ent);
 542                                 /* For 'll_nfs_get_name_filldir()', it will try
 543                                  * to access the 'ent' through its 'lde_name',
 544                                  * so the parameter 'name' for 'filldir()' must
 545                                  * be part of the 'ent'. */
 546                                 done = filldir(cookie, ent->lde_name, namelen,
 547                                                lhash, ino, type);
 548                         }
 549                         next = le64_to_cpu(dp->ldp_hash_end);
 550                         if (!done) {
 551                                 pos = next;
 552                                 if (pos == MDS_DIR_END_OFF) {
 553                                         /*
 554                                          * End of directory reached.
 555                                          */
 556                                         done = 1;
 557                                         ll_release_page(page, 0);
 558                                 } else if (1 /* chain is exhausted*/) {
 559                                         /*
 560                                          * Normal case: continue to the next
 561                                          * page.
 562                                          */
 563                                         ll_release_page(page,
 564                                             le32_to_cpu(dp->ldp_flags) &
 565                                                         LDF_COLLIDE);
 566                                         next = pos;
 567                                         page = ll_get_dir_page(inode, pos,
 568                                                                &chain);
 569                                 } else {
 570                                         /*
 571                                          * go into overflow page.
 572                                          */
 573                                         LASSERT(le32_to_cpu(dp->ldp_flags) &
 574                                                 LDF_COLLIDE);
 575                                         ll_release_page(page, 1);
 576                                 }
 577                         } else {
 578                                 pos = hash;
 579                                 ll_release_page(page, 0);
 580                         }
 581                 } else {
 582                         rc = PTR_ERR(page);
 583                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 584                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 585                 }
 586         }
 587
 588         *_pos = pos;
 589         ll_dir_chain_fini(&chain);
 590         RETURN(rc);
 591 }
 592
 593 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 594 {
 595         struct inode            *inode  = filp->f_dentry->d_inode;
 596         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 597         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 598         __u64                   pos;
 599         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 600         int                     api32   = ll_need_32bit_api(sbi);
 601         int                     rc;
 602 #ifdef HAVE_TOUCH_ATIME_1ARG
 603         struct path             path;
 604 #endif
 605         ENTRY;
 606
 607         if (lfd != NULL)
 608                 pos = lfd->lfd_pos;
 609         else
 610                 pos = 0;
 611
 612         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu "
 613                " 32bit_api %d\n", inode->i_ino, inode->i_generation,
 614                inode, (unsigned long)pos, i_size_read(inode), api32);
 615
 616         if (pos == MDS_DIR_END_OFF)
 617                 /*
 618                  * end-of-file.
 619                  */
 620                 GOTO(out, rc = 0);
 621
 622         rc = ll_dir_read(inode, &pos, cookie, filldir);
 623         if (lfd != NULL)
 624                 lfd->lfd_pos = pos;
 625         if (pos == MDS_DIR_END_OFF) {
 626                 if (api32)
 627                         filp->f_pos = LL_DIR_END_OFF_32BIT;
 628                 else
 629                         filp->f_pos = LL_DIR_END_OFF;
 630         } else {
 631                 if (api32 && hash64)
 632                         filp->f_pos = pos >> 32;
 633                 else
 634                         filp->f_pos = pos;
 635         }
 636         filp->f_version = inode->i_version;
 637 #ifdef HAVE_TOUCH_ATIME_1ARG
 638 #ifdef HAVE_F_PATH_MNT
 639         path.mnt = filp->f_path.mnt;
 640 #else
 641         path.mnt = filp->f_vfsmnt;
 642 #endif
 643         path.dentry = filp->f_dentry;
 644         touch_atime(&path);
 645 #else
 646         touch_atime(filp->f_vfsmnt, filp->f_dentry);
 647 #endif
 648
 649 out:
 650         if (!rc)
 651                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 652
 653         RETURN(rc);
 654 }
 655
 656 int ll_send_mgc_param(struct obd_export *mgc, char *string)
 657 {
 658         struct mgs_send_param *msp;
 659         int rc = 0;
 660
 661         OBD_ALLOC_PTR(msp);
 662         if (!msp)
 663                 return -ENOMEM;
 664
 665         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 666         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 667                                 sizeof(struct mgs_send_param), msp, NULL);
 668         if (rc)
 669                 CERROR("Failed to set parameter: %d\n", rc);
 670         OBD_FREE_PTR(msp);
 671
 672         return rc;
 673 }
 674
 675 int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
 676                         char *filename)
 677 {
 678         struct ptlrpc_request *request = NULL;
 679         struct md_op_data *op_data;
 680         struct ll_sb_info *sbi = ll_i2sbi(dir);
 681         int mode;
 682         int err;
 683
 684         ENTRY;
 685
 686         mode = (0755 & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
 687         op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
 688                                      strlen(filename), mode, LUSTRE_OPC_MKDIR,
 689                                      lump);
 690         if (IS_ERR(op_data))
 691                 GOTO(err_exit, err = PTR_ERR(op_data));
 692
 693         op_data->op_cli_flags |= CLI_SET_MEA;
 694         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 695                         current_fsuid(), current_fsgid(),
 696                         cfs_curproc_cap_pack(), 0, &request);
 697         ll_finish_md_op_data(op_data);
 698         if (err)
 699                 GOTO(err_exit, err);
 700 err_exit:
 701         ptlrpc_req_finished(request);
 702         return err;
 703 }
 704
 705 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 706                      int set_default)
 707 {
 708         struct ll_sb_info *sbi = ll_i2sbi(inode);
 709         struct md_op_data *op_data;
 710         struct ptlrpc_request *req = NULL;
 711         int rc = 0;
 712         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 713         struct obd_device *mgc = lsi->lsi_mgc;
 714         int lum_size;
 715         ENTRY;
 716
 717         if (lump != NULL) {
 718                 /*
 719                  * This is coming from userspace, so should be in
 720                  * local endian.  But the MDS would like it in little
 721                  * endian, so we swab it before we send it.
 722                  */
 723                 switch (lump->lmm_magic) {
 724                 case LOV_USER_MAGIC_V1: {
 725                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 726                                 lustre_swab_lov_user_md_v1(lump);
 727                         lum_size = sizeof(struct lov_user_md_v1);
 728                         break;
 729                 }
 730                 case LOV_USER_MAGIC_V3: {
 731                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 732                                 lustre_swab_lov_user_md_v3(
 733                                         (struct lov_user_md_v3 *)lump);
 734                         lum_size = sizeof(struct lov_user_md_v3);
 735                         break;
 736                 }
 737                 default: {
 738                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 739                                         " %#08x != %#08x nor %#08x\n",
 740                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 741                                         LOV_USER_MAGIC_V3);
 742                         RETURN(-EINVAL);
 743                 }
 744                 }
 745         } else {
 746                 lum_size = sizeof(struct lov_user_md_v1);
 747         }
 748
 749         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 750                                      LUSTRE_OPC_ANY, NULL);
 751         if (IS_ERR(op_data))
 752                 RETURN(PTR_ERR(op_data));
 753
 754         if (lump != NULL && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC))
 755                 op_data->op_cli_flags |= CLI_SET_MEA;
 756
 757         /* swabbing is done in lov_setstripe() on server side */
 758         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 759                         NULL, 0, &req, NULL);
 760         ll_finish_md_op_data(op_data);
 761         ptlrpc_req_finished(req);
 762         if (rc) {
 763                 if (rc != -EPERM && rc != -EACCES)
 764                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 765         }
 766
 767         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 768          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 769          need the make the distiction between the 2 versions */
 770         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 771                 char *param = NULL;
 772                 char *buf;
 773
 774                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 775                 if (param == NULL)
 776                         GOTO(end, rc = -ENOMEM);
 777
 778                 buf = param;
 779                 /* Get fsname and assume devname to be -MDT0000. */
 780                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 781                 strcat(buf, "-MDT0000.lov");
 782                 buf += strlen(buf);
 783
 784                 /* Set root stripesize */
 785                 sprintf(buf, ".stripesize=%u",
 786                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 787                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 788                 if (rc)
 789                         GOTO(end, rc);
 790
 791                 /* Set root stripecount */
 792                 sprintf(buf, ".stripecount=%hd",
 793                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 794                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 795                 if (rc)
 796                         GOTO(end, rc);
 797
 798                 /* Set root stripeoffset */
 799                 sprintf(buf, ".stripeoffset=%hd",
 800                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 801                         (typeof(lump->lmm_stripe_offset))(-1));
 802                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 803
 804 end:
 805                 if (param != NULL)
 806                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 807         }
 808         RETURN(rc);
 809 }
 810
 811 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 812                      int *lmm_size, struct ptlrpc_request **request)
 813 {
 814         struct ll_sb_info *sbi = ll_i2sbi(inode);
 815         struct mdt_body   *body;
 816         struct lov_mds_md *lmm = NULL;
 817         struct ptlrpc_request *req = NULL;
 818         int rc, lmmsize;
 819         struct md_op_data *op_data;
 820
 821         rc = ll_get_max_mdsize(sbi, &lmmsize);
 822         if (rc)
 823                 RETURN(rc);
 824
 825         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 826                                      0, lmmsize, LUSTRE_OPC_ANY,
 827                                      NULL);
 828         if (IS_ERR(op_data))
 829                 RETURN(PTR_ERR(op_data));
 830
 831         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 832         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 833         ll_finish_md_op_data(op_data);
 834         if (rc < 0) {
 835                 CDEBUG(D_INFO, "md_getattr failed on inode "
 836                        "%lu/%u: rc %d\n", inode->i_ino,
 837                        inode->i_generation, rc);
 838                 GOTO(out, rc);
 839         }
 840
 841         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 842         LASSERT(body != NULL);
 843
 844         lmmsize = body->eadatasize;
 845
 846         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 847             lmmsize == 0) {
 848                 GOTO(out, rc = -ENODATA);
 849         }
 850
 851         lmm = req_capsule_server_sized_get(&req->rq_pill,
 852                                            &RMF_MDT_MD, lmmsize);
 853         LASSERT(lmm != NULL);
 854
 855         /*
 856          * This is coming from the MDS, so is probably in
 857          * little endian.  We convert it to host endian before
 858          * passing it to userspace.
 859          */
 860         /* We don't swab objects for directories */
 861         switch (le32_to_cpu(lmm->lmm_magic)) {
 862         case LOV_MAGIC_V1:
 863                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 864                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 865                 break;
 866         case LOV_MAGIC_V3:
 867                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 868                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 869                 break;
 870         default:
 871                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 872                 rc = -EPROTO;
 873         }
 874 out:
 875         *lmmp = lmm;
 876         *lmm_size = lmmsize;
 877         *request = req;
 878         return rc;
 879 }
 880
 881 /*
 882  *  Get MDT index for the inode.
 883  */
 884 int ll_get_mdt_idx(struct inode *inode)
 885 {
 886         struct ll_sb_info *sbi = ll_i2sbi(inode);
 887         struct md_op_data *op_data;
 888         int rc, mdtidx;
 889         ENTRY;
 890
 891         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 892                                      0, LUSTRE_OPC_ANY, NULL);
 893         if (IS_ERR(op_data))
 894                 RETURN(PTR_ERR(op_data));
 895
 896         op_data->op_flags |= MF_GET_MDT_IDX;
 897         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 898         mdtidx = op_data->op_mds;
 899         ll_finish_md_op_data(op_data);
 900         if (rc < 0) {
 901                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 902                 RETURN(rc);
 903         }
 904         return mdtidx;
 905 }
 906
 907 /**
 908  * Generic handler to do any pre-copy work.
 909  *
 910  * It send a first hsm_progress (with extent length == 0) to coordinator as a
 911  * first information for it that real work has started.
 912  *
 913  * Moreover, for a ARCHIVE request, it will sample the file data version and
 914  * store it in \a copy.
 915  *
 916  * \return 0 on success.
 917  */
 918 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 919 {
 920         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 921         struct hsm_progress_kernel       hpk;
 922         int                              rc;
 923         ENTRY;
 924
 925         /* Forge a hsm_progress based on data from copy. */
 926         hpk.hpk_fid = copy->hc_hai.hai_fid;
 927         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 928         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 929         hpk.hpk_extent.length = 0;
 930         hpk.hpk_flags = 0;
 931         hpk.hpk_errval = 0;
 932         hpk.hpk_data_version = 0;
 933
 934
 935         /* For archive request, we need to read the current file version. */
 936         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 937                 struct inode    *inode;
 938                 __u64            data_version = 0;
 939
 940                 /* Get inode for this fid */
 941                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 942                 if (IS_ERR(inode)) {
 943                         hpk.hpk_flags |= HP_FLAG_RETRY;
 944                         /* hpk_errval is >= 0 */
 945                         hpk.hpk_errval = -PTR_ERR(inode);
 946                         GOTO(progress, rc = PTR_ERR(inode));
 947                 }
 948
 949                 /* Read current file data version */
 950                 rc = ll_data_version(inode, &data_version, 1);
 951                 iput(inode);
 952                 if (rc != 0) {
 953                         CDEBUG(D_HSM, "Could not read file data version of "
 954                                       DFID" (rc = %d). Archive request ("
 955                                       LPX64") could not be done.\n",
 956                                       PFID(&copy->hc_hai.hai_fid), rc,
 957                                       copy->hc_hai.hai_cookie);
 958                         hpk.hpk_flags |= HP_FLAG_RETRY;
 959                         /* hpk_errval must be >= 0 */
 960                         hpk.hpk_errval = -rc;
 961                         GOTO(progress, rc);
 962                 }
 963
 964                 /* Store it the hsm_copy for later copytool use.
 965                  * Always modified even if no lsm. */
 966                 copy->hc_data_version = data_version;
 967         }
 968
 969 progress:
 970         /* On error, the request should be considered as completed */
 971         if (hpk.hpk_errval > 0)
 972                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 973         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 974                            &hpk, NULL);
 975
 976         RETURN(rc);
 977 }
 978
 979 /**
 980  * Generic handler to do any post-copy work.
 981  *
 982  * It will send the last hsm_progress update to coordinator to inform it
 983  * that copy is finished and whether it was successful or not.
 984  *
 985  * Moreover,
 986  * - for ARCHIVE request, it will sample the file data version and compare it
 987  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 988  *   will be considered as failed.
 989  * - for RESTORE request, it will sample the file data version and send it to
 990  *   coordinator which is useful if the file was imported as 'released'.
 991  *
 992  * \return 0 on success.
 993  */
 994 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 995 {
 996         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 997         struct hsm_progress_kernel       hpk;
 998         int                              rc;
 999         ENTRY;
1000
1001         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
1002         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
1003          * initialized if copy_end was called with copy == NULL.
1004          */
1005
1006         /* Forge a hsm_progress based on data from copy. */
1007         hpk.hpk_fid = copy->hc_hai.hai_fid;
1008         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
1009         hpk.hpk_extent = copy->hc_hai.hai_extent;
1010         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
1011         hpk.hpk_errval = copy->hc_errval;
1012         hpk.hpk_data_version = 0;
1013
1014         /* For archive request, we need to check the file data was not changed.
1015          *
1016          * For restore request, we need to send the file data version, this is
1017          * useful when the file was created using hsm_import.
1018          */
1019         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
1020              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
1021             (copy->hc_errval == 0)) {
1022                 struct inode    *inode;
1023                 __u64            data_version = 0;
1024
1025                 /* Get lsm for this fid */
1026                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
1027                 if (IS_ERR(inode)) {
1028                         hpk.hpk_flags |= HP_FLAG_RETRY;
1029                         /* hpk_errval must be >= 0 */
1030                         hpk.hpk_errval = -PTR_ERR(inode);
1031                         GOTO(progress, rc = PTR_ERR(inode));
1032                 }
1033
1034                 rc = ll_data_version(inode, &data_version,
1035                                      copy->hc_hai.hai_action == HSMA_ARCHIVE);
1036                 iput(inode);
1037                 if (rc) {
1038                         CDEBUG(D_HSM, "Could not read file data version. "
1039                                       "Request could not be confirmed.\n");
1040                         if (hpk.hpk_errval == 0)
1041                                 hpk.hpk_errval = -rc;
1042                         GOTO(progress, rc);
1043                 }
1044
1045                 /* Store it the hsm_copy for later copytool use.
1046                  * Always modified even if no lsm. */
1047                 hpk.hpk_data_version = data_version;
1048
1049                 /* File could have been stripped during archiving, so we need
1050                  * to check anyway. */
1051                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1052                     (copy->hc_data_version != data_version)) {
1053                         CDEBUG(D_HSM, "File data version mismatched. "
1054                               "File content was changed during archiving. "
1055                                DFID", start:"LPX64" current:"LPX64"\n",
1056                                PFID(&copy->hc_hai.hai_fid),
1057                                copy->hc_data_version, data_version);
1058                         /* File was changed, send error to cdt. Do not ask for
1059                          * retry because if a file is modified frequently,
1060                          * the cdt will loop on retried archive requests.
1061                          * The policy engine will ask for a new archive later
1062                          * when the file will not be modified for some tunable
1063                          * time */
1064                         /* we do not notify caller */
1065                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
1066                         /* hpk_errval must be >= 0 */
1067                         hpk.hpk_errval = EBUSY;
1068                 }
1069
1070         }
1071
1072 progress:
1073         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1074                            &hpk, NULL);
1075
1076         RETURN(rc);
1077 }
1078
1079
1080 static int copy_and_ioctl(int cmd, struct obd_export *exp,
1081                           const void __user *data, size_t size)
1082 {
1083         void *copy;
1084         int rc;
1085
1086         OBD_ALLOC(copy, size);
1087         if (copy == NULL)
1088                 return -ENOMEM;
1089
1090         if (copy_from_user(copy, data, size)) {
1091                 rc = -EFAULT;
1092                 goto out;
1093         }
1094
1095         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
1096 out:
1097         OBD_FREE(copy, size);
1098
1099         return rc;
1100 }
1101
1102 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1103 {
1104         int cmd = qctl->qc_cmd;
1105         int type = qctl->qc_type;
1106         int id = qctl->qc_id;
1107         int valid = qctl->qc_valid;
1108         int rc = 0;
1109         ENTRY;
1110
1111         switch (cmd) {
1112         case LUSTRE_Q_INVALIDATE:
1113         case LUSTRE_Q_FINVALIDATE:
1114         case Q_QUOTAON:
1115         case Q_QUOTAOFF:
1116         case Q_SETQUOTA:
1117         case Q_SETINFO:
1118                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1119                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1120                         RETURN(-EPERM);
1121                 break;
1122         case Q_GETQUOTA:
1123                 if (((type == USRQUOTA && current_euid() != id) ||
1124                      (type == GRPQUOTA && !in_egroup_p(id))) &&
1125                     (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1126                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
1127                         RETURN(-EPERM);
1128                 break;
1129         case Q_GETINFO:
1130                 break;
1131         default:
1132                 CERROR("unsupported quotactl op: %#x\n", cmd);
1133                 RETURN(-ENOTTY);
1134         }
1135
1136         if (valid != QC_GENERAL) {
1137                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1138                         RETURN(-EOPNOTSUPP);
1139
1140                 if (cmd == Q_GETINFO)
1141                         qctl->qc_cmd = Q_GETOINFO;
1142                 else if (cmd == Q_GETQUOTA)
1143                         qctl->qc_cmd = Q_GETOQUOTA;
1144                 else
1145                         RETURN(-EINVAL);
1146
1147                 switch (valid) {
1148                 case QC_MDTIDX:
1149                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1150                                            sizeof(*qctl), qctl, NULL);
1151                         break;
1152                 case QC_OSTIDX:
1153                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1154                                            sizeof(*qctl), qctl, NULL);
1155                         break;
1156                 case QC_UUID:
1157                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1158                                            sizeof(*qctl), qctl, NULL);
1159                         if (rc == -EAGAIN)
1160                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1161                                                    sbi->ll_dt_exp,
1162                                                    sizeof(*qctl), qctl, NULL);
1163                         break;
1164                 default:
1165                         rc = -EINVAL;
1166                         break;
1167                 }
1168
1169                 if (rc)
1170                         RETURN(rc);
1171
1172                 qctl->qc_cmd = cmd;
1173         } else {
1174                 struct obd_quotactl *oqctl;
1175
1176                 OBD_ALLOC_PTR(oqctl);
1177                 if (oqctl == NULL)
1178                         RETURN(-ENOMEM);
1179
1180                 QCTL_COPY(oqctl, qctl);
1181                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1182                 if (rc) {
1183                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
1184                                 oqctl->qc_cmd = Q_QUOTAOFF;
1185                                 obd_quotactl(sbi->ll_md_exp, oqctl);
1186                         }
1187                         OBD_FREE_PTR(oqctl);
1188                         RETURN(rc);
1189                 }
1190                 /* If QIF_SPACE is not set, client should collect the
1191                  * space usage from OSSs by itself */
1192                 if (cmd == Q_GETQUOTA &&
1193                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1194                     !oqctl->qc_dqblk.dqb_curspace) {
1195                         struct obd_quotactl *oqctl_tmp;
1196
1197                         OBD_ALLOC_PTR(oqctl_tmp);
1198                         if (oqctl_tmp == NULL)
1199                                 GOTO(out, rc = -ENOMEM);
1200
1201                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1202                         oqctl_tmp->qc_id = oqctl->qc_id;
1203                         oqctl_tmp->qc_type = oqctl->qc_type;
1204
1205                         /* collect space usage from OSTs */
1206                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1207                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1208                         if (!rc || rc == -EREMOTEIO) {
1209                                 oqctl->qc_dqblk.dqb_curspace =
1210                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1211                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1212                         }
1213
1214                         /* collect space & inode usage from MDTs */
1215                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1216                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1217                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1218                         if (!rc || rc == -EREMOTEIO) {
1219                                 oqctl->qc_dqblk.dqb_curspace +=
1220                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1221                                 oqctl->qc_dqblk.dqb_curinodes =
1222                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1223                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1224                         } else {
1225                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1226                         }
1227
1228                         OBD_FREE_PTR(oqctl_tmp);
1229                 }
1230 out:
1231                 QCTL_COPY(qctl, oqctl);
1232                 OBD_FREE_PTR(oqctl);
1233         }
1234
1235         RETURN(rc);
1236 }
1237
1238 static char *
1239 ll_getname(const char __user *filename)
1240 {
1241         int ret = 0, len;
1242         char *tmp = __getname();
1243
1244         if (!tmp)
1245                 return ERR_PTR(-ENOMEM);
1246
1247         len = strncpy_from_user(tmp, filename, PATH_MAX);
1248         if (len == 0)
1249                 ret = -ENOENT;
1250         else if (len > PATH_MAX)
1251                 ret = -ENAMETOOLONG;
1252
1253         if (ret) {
1254                 __putname(tmp);
1255                 tmp =  ERR_PTR(ret);
1256         }
1257         return tmp;
1258 }
1259
1260 #define ll_putname(filename) __putname(filename)
1261
1262 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1263 {
1264         struct inode *inode = file->f_dentry->d_inode;
1265         struct ll_sb_info *sbi = ll_i2sbi(inode);
1266         struct obd_ioctl_data *data;
1267         int rc = 0;
1268         ENTRY;
1269
1270         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1271                inode->i_ino, inode->i_generation, inode, cmd);
1272
1273         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1274         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1275                 return -ENOTTY;
1276
1277         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1278         switch(cmd) {
1279         case FSFILT_IOC_GETFLAGS:
1280         case FSFILT_IOC_SETFLAGS:
1281                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1282         case FSFILT_IOC_GETVERSION_OLD:
1283         case FSFILT_IOC_GETVERSION:
1284                 RETURN(put_user(inode->i_generation, (int *)arg));
1285         /* We need to special case any other ioctls we want to handle,
1286          * to send them to the MDS/OST as appropriate and to properly
1287          * network encode the arg field.
1288         case FSFILT_IOC_SETVERSION_OLD:
1289         case FSFILT_IOC_SETVERSION:
1290         */
1291         case LL_IOC_GET_MDTIDX: {
1292                 int mdtidx;
1293
1294                 mdtidx = ll_get_mdt_idx(inode);
1295                 if (mdtidx < 0)
1296                         RETURN(mdtidx);
1297
1298                 if (put_user((int)mdtidx, (int*)arg))
1299                         RETURN(-EFAULT);
1300
1301                 return 0;
1302         }
1303         case IOC_MDC_LOOKUP: {
1304                 struct ptlrpc_request *request = NULL;
1305                 int namelen, len = 0;
1306                 char *buf = NULL;
1307                 char *filename;
1308                 struct md_op_data *op_data;
1309
1310                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1311                 if (rc)
1312                         RETURN(rc);
1313                 data = (void *)buf;
1314
1315                 filename = data->ioc_inlbuf1;
1316                 namelen = strlen(filename);
1317
1318                 if (namelen < 1) {
1319                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1320                         GOTO(out_free, rc = -EINVAL);
1321                 }
1322
1323                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1324                                              0, LUSTRE_OPC_ANY, NULL);
1325                 if (IS_ERR(op_data))
1326                         GOTO(out_free, rc = PTR_ERR(op_data));
1327
1328                 op_data->op_valid = OBD_MD_FLID;
1329                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1330                 ll_finish_md_op_data(op_data);
1331                 if (rc < 0) {
1332                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1333                         GOTO(out_free, rc);
1334                 }
1335                 ptlrpc_req_finished(request);
1336                 EXIT;
1337 out_free:
1338                 obd_ioctl_freedata(buf, len);
1339                 return rc;
1340         }
1341         case LL_IOC_LMV_SETSTRIPE: {
1342                 struct lmv_user_md  *lum;
1343                 char            *buf = NULL;
1344                 char            *filename;
1345                 int              namelen = 0;
1346                 int              lumlen = 0;
1347                 int              len;
1348                 int              rc;
1349
1350                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1351                 if (rc)
1352                         RETURN(rc);
1353
1354                 data = (void *)buf;
1355                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1356                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1357                         GOTO(lmv_out_free, rc = -EINVAL);
1358
1359                 filename = data->ioc_inlbuf1;
1360                 namelen = data->ioc_inllen1;
1361
1362                 if (namelen < 1) {
1363                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1364                         GOTO(lmv_out_free, rc = -EINVAL);
1365                 }
1366                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1367                 lumlen = data->ioc_inllen2;
1368
1369                 if (lum->lum_magic != LMV_USER_MAGIC ||
1370                     lumlen != sizeof(*lum)) {
1371                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1372                                filename, lum->lum_magic, lumlen, -EFAULT);
1373                         GOTO(lmv_out_free, rc = -EINVAL);
1374                 }
1375
1376                 /**
1377                  * ll_dir_setdirstripe will be used to set dir stripe
1378                  *  mdc_create--->mdt_reint_create (with dirstripe)
1379                  */
1380                 rc = ll_dir_setdirstripe(inode, lum, filename);
1381 lmv_out_free:
1382                 obd_ioctl_freedata(buf, len);
1383                 RETURN(rc);
1384
1385         }
1386         case LL_IOC_LOV_SETSTRIPE: {
1387                 struct lov_user_md_v3 lumv3;
1388                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1389                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1390                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1391
1392                 int set_default = 0;
1393
1394                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1395                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1396                         sizeof(lumv3p->lmm_objects[0]));
1397                 /* first try with v1 which is smaller than v3 */
1398                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1399                         RETURN(-EFAULT);
1400
1401                 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1402                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1403                                 RETURN(-EFAULT);
1404                 }
1405
1406                 if (inode->i_sb->s_root == file->f_dentry)
1407                         set_default = 1;
1408
1409                 /* in v1 and v3 cases lumv1 points to data */
1410                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1411
1412                 RETURN(rc);
1413         }
1414         case LL_IOC_LMV_GETSTRIPE: {
1415                 struct lmv_user_md *lump = (struct lmv_user_md *)arg;
1416                 struct lmv_user_md lum;
1417                 struct lmv_user_md *tmp;
1418                 int lum_size;
1419                 int rc = 0;
1420                 int mdtindex;
1421
1422                 if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
1423                         RETURN(-EFAULT);
1424
1425                 if (lum.lum_magic != LMV_MAGIC_V1)
1426                         RETURN(-EINVAL);
1427
1428                 lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
1429                 OBD_ALLOC(tmp, lum_size);
1430                 if (tmp == NULL)
1431                         GOTO(free_lmv, rc = -ENOMEM);
1432
1433                 memcpy(tmp, &lum, sizeof(lum));
1434                 tmp->lum_type = LMV_STRIPE_TYPE;
1435                 tmp->lum_stripe_count = 1;
1436                 mdtindex = ll_get_mdt_idx(inode);
1437                 if (mdtindex < 0)
1438                         GOTO(free_lmv, rc = -ENOMEM);
1439
1440                 tmp->lum_stripe_offset = mdtindex;
1441                 tmp->lum_objects[0].lum_mds = mdtindex;
1442                 memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
1443                        sizeof(struct lu_fid));
1444                 if (copy_to_user((void *)arg, tmp, lum_size))
1445                         GOTO(free_lmv, rc = -EFAULT);
1446 free_lmv:
1447                 if (tmp)
1448                         OBD_FREE(tmp, lum_size);
1449                 RETURN(rc);
1450         }
1451         case LL_IOC_REMOVE_ENTRY: {
1452                 char            *filename = NULL;
1453                 int              namelen = 0;
1454                 int              rc;
1455
1456                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1457                  * unsupported server, which might crash the server(LU-2730),
1458                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1459                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1460                  * server will support REINT_RMENTRY XXX*/
1461                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1462                         RETURN(-ENOTSUPP);
1463
1464                 filename = ll_getname((const char *)arg);
1465                 if (IS_ERR(filename))
1466                         RETURN(PTR_ERR(filename));
1467
1468                 namelen = strlen(filename);
1469                 if (namelen < 1)
1470                         GOTO(out_rmdir, rc = -EINVAL);
1471
1472                 rc = ll_rmdir_entry(inode, filename, namelen);
1473 out_rmdir:
1474                 if (filename)
1475                         ll_putname(filename);
1476                 RETURN(rc);
1477         }
1478         case LL_IOC_LOV_SWAP_LAYOUTS:
1479                 RETURN(-EPERM);
1480         case LL_IOC_OBD_STATFS:
1481                 RETURN(ll_obd_statfs(inode, (void *)arg));
1482         case LL_IOC_LOV_GETSTRIPE:
1483         case LL_IOC_MDC_GETINFO:
1484         case IOC_MDC_GETFILEINFO:
1485         case IOC_MDC_GETFILESTRIPE: {
1486                 struct ptlrpc_request *request = NULL;
1487                 struct lov_user_md *lump;
1488                 struct lov_mds_md *lmm = NULL;
1489                 struct mdt_body *body;
1490                 char *filename = NULL;
1491                 int lmmsize;
1492
1493                 if (cmd == IOC_MDC_GETFILEINFO ||
1494                     cmd == IOC_MDC_GETFILESTRIPE) {
1495                         filename = ll_getname((const char *)arg);
1496                         if (IS_ERR(filename))
1497                                 RETURN(PTR_ERR(filename));
1498
1499                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1500                                                       &lmmsize, &request);
1501                 } else {
1502                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1503                 }
1504
1505                 if (request) {
1506                         body = req_capsule_server_get(&request->rq_pill,
1507                                                       &RMF_MDT_BODY);
1508                         LASSERT(body != NULL);
1509                 } else {
1510                         GOTO(out_req, rc);
1511                 }
1512
1513                 if (rc < 0) {
1514                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1515                                                cmd == LL_IOC_MDC_GETINFO))
1516                                 GOTO(skip_lmm, rc = 0);
1517                         else
1518                                 GOTO(out_req, rc);
1519                 }
1520
1521                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1522                     cmd == LL_IOC_LOV_GETSTRIPE) {
1523                         lump = (struct lov_user_md *)arg;
1524                 } else {
1525                         struct lov_user_mds_data *lmdp;
1526                         lmdp = (struct lov_user_mds_data *)arg;
1527                         lump = &lmdp->lmd_lmm;
1528                 }
1529                 if (copy_to_user(lump, lmm, lmmsize)) {
1530                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1531                                 GOTO(out_req, rc = -EFAULT);
1532                         rc = -EOVERFLOW;
1533                 }
1534         skip_lmm:
1535                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1536                         struct lov_user_mds_data *lmdp;
1537                         lstat_t st = { 0 };
1538
1539                         st.st_dev     = inode->i_sb->s_dev;
1540                         st.st_mode    = body->mode;
1541                         st.st_nlink   = body->nlink;
1542                         st.st_uid     = body->uid;
1543                         st.st_gid     = body->gid;
1544                         st.st_rdev    = body->rdev;
1545                         st.st_size    = body->size;
1546                         st.st_blksize = PAGE_CACHE_SIZE;
1547                         st.st_blocks  = body->blocks;
1548                         st.st_atime   = body->atime;
1549                         st.st_mtime   = body->mtime;
1550                         st.st_ctime   = body->ctime;
1551                         st.st_ino     = inode->i_ino;
1552
1553                         lmdp = (struct lov_user_mds_data *)arg;
1554                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1555                                 GOTO(out_req, rc = -EFAULT);
1556                 }
1557
1558                 EXIT;
1559         out_req:
1560                 ptlrpc_req_finished(request);
1561                 if (filename)
1562                         ll_putname(filename);
1563                 return rc;
1564         }
1565         case IOC_LOV_GETINFO: {
1566                 struct lov_user_mds_data *lumd;
1567                 struct lov_stripe_md *lsm;
1568                 struct lov_user_md *lum;
1569                 struct lov_mds_md *lmm;
1570                 int lmmsize;
1571                 lstat_t st;
1572
1573                 lumd = (struct lov_user_mds_data *)arg;
1574                 lum = &lumd->lmd_lmm;
1575
1576                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1577                 if (rc)
1578                         RETURN(rc);
1579
1580                 OBD_ALLOC_LARGE(lmm, lmmsize);
1581                 if (lmm == NULL)
1582                         RETURN(-ENOMEM);
1583
1584                 if (copy_from_user(lmm, lum, lmmsize))
1585                         GOTO(free_lmm, rc = -EFAULT);
1586
1587                 switch (lmm->lmm_magic) {
1588                 case LOV_USER_MAGIC_V1:
1589                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1590                                 break;
1591                         /* swab objects first so that stripes num will be sane */
1592                         lustre_swab_lov_user_md_objects(
1593                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1594                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1595                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1596                         break;
1597                 case LOV_USER_MAGIC_V3:
1598                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1599                                 break;
1600                         /* swab objects first so that stripes num will be sane */
1601                         lustre_swab_lov_user_md_objects(
1602                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1603                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1604                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1605                         break;
1606                 default:
1607                         GOTO(free_lmm, rc = -EINVAL);
1608                 }
1609
1610                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1611                 if (rc < 0)
1612                         GOTO(free_lmm, rc = -ENOMEM);
1613
1614                 /* Perform glimpse_size operation. */
1615                 memset(&st, 0, sizeof(st));
1616
1617                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1618                 if (rc)
1619                         GOTO(free_lsm, rc);
1620
1621                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1622                         GOTO(free_lsm, rc = -EFAULT);
1623
1624                 EXIT;
1625         free_lsm:
1626                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1627         free_lmm:
1628                 OBD_FREE_LARGE(lmm, lmmsize);
1629                 return rc;
1630         }
1631         case OBD_IOC_LLOG_CATINFO: {
1632                 RETURN(-EOPNOTSUPP);
1633         }
1634         case OBD_IOC_QUOTACHECK: {
1635                 struct obd_quotactl *oqctl;
1636                 int error = 0;
1637
1638                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1639                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1640                         RETURN(-EPERM);
1641
1642                 OBD_ALLOC_PTR(oqctl);
1643                 if (!oqctl)
1644                         RETURN(-ENOMEM);
1645                 oqctl->qc_type = arg;
1646                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1647                 if (rc < 0) {
1648                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1649                         error = rc;
1650                 }
1651
1652                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1653                 if (rc < 0)
1654                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1655
1656                 OBD_FREE_PTR(oqctl);
1657                 return error ?: rc;
1658         }
1659         case OBD_IOC_POLL_QUOTACHECK: {
1660                 struct if_quotacheck *check;
1661
1662                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1663                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1664                         RETURN(-EPERM);
1665
1666                 OBD_ALLOC_PTR(check);
1667                 if (!check)
1668                         RETURN(-ENOMEM);
1669
1670                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1671                                    NULL);
1672                 if (rc) {
1673                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1674                         if (copy_to_user((void *)arg, check,
1675                                              sizeof(*check)))
1676                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1677                         GOTO(out_poll, rc);
1678                 }
1679
1680                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1681                                    NULL);
1682                 if (rc) {
1683                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1684                         if (copy_to_user((void *)arg, check,
1685                                              sizeof(*check)))
1686                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1687                         GOTO(out_poll, rc);
1688                 }
1689         out_poll:
1690                 OBD_FREE_PTR(check);
1691                 RETURN(rc);
1692         }
1693 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
1694         case LL_IOC_QUOTACTL_18: {
1695                 /* copy the old 1.x quota struct for internal use, then copy
1696                  * back into old format struct.  For 1.8 compatibility. */
1697                 struct if_quotactl_18 *qctl_18;
1698                 struct if_quotactl *qctl_20;
1699
1700                 OBD_ALLOC_PTR(qctl_18);
1701                 if (!qctl_18)
1702                         RETURN(-ENOMEM);
1703
1704                 OBD_ALLOC_PTR(qctl_20);
1705                 if (!qctl_20)
1706                         GOTO(out_quotactl_18, rc = -ENOMEM);
1707
1708                 if (copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1709                         GOTO(out_quotactl_20, rc = -ENOMEM);
1710
1711                 QCTL_COPY(qctl_20, qctl_18);
1712                 qctl_20->qc_idx = 0;
1713
1714                 /* XXX: dqb_valid was borrowed as a flag to mark that
1715                  *      only mds quota is wanted */
1716                 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1717                     qctl_18->qc_dqblk.dqb_valid) {
1718                         qctl_20->qc_valid = QC_MDTIDX;
1719                         qctl_20->qc_dqblk.dqb_valid = 0;
1720                 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1721                         qctl_20->qc_valid = QC_UUID;
1722                         qctl_20->obd_uuid = qctl_18->obd_uuid;
1723                 } else {
1724                         qctl_20->qc_valid = QC_GENERAL;
1725                 }
1726
1727                 rc = quotactl_ioctl(sbi, qctl_20);
1728
1729                 if (rc == 0) {
1730                         QCTL_COPY(qctl_18, qctl_20);
1731                         qctl_18->obd_uuid = qctl_20->obd_uuid;
1732
1733                         if (copy_to_user((void *)arg, qctl_18,
1734                                              sizeof(*qctl_18)))
1735                                 rc = -EFAULT;
1736                 }
1737
1738         out_quotactl_20:
1739                 OBD_FREE_PTR(qctl_20);
1740         out_quotactl_18:
1741                 OBD_FREE_PTR(qctl_18);
1742                 RETURN(rc);
1743         }
1744 #else
1745 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1746 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
1747         case LL_IOC_QUOTACTL: {
1748                 struct if_quotactl *qctl;
1749
1750                 OBD_ALLOC_PTR(qctl);
1751                 if (!qctl)
1752                         RETURN(-ENOMEM);
1753
1754                 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1755                         GOTO(out_quotactl, rc = -EFAULT);
1756
1757                 rc = quotactl_ioctl(sbi, qctl);
1758
1759                 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1760                         rc = -EFAULT;
1761
1762         out_quotactl:
1763                 OBD_FREE_PTR(qctl);
1764                 RETURN(rc);
1765         }
1766         case OBD_IOC_GETDTNAME:
1767         case OBD_IOC_GETMDNAME:
1768                 RETURN(ll_get_obd_name(inode, cmd, arg));
1769         case LL_IOC_FLUSHCTX:
1770                 RETURN(ll_flush_ctx(inode));
1771 #ifdef CONFIG_FS_POSIX_ACL
1772         case LL_IOC_RMTACL: {
1773             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1774                 inode == inode->i_sb->s_root->d_inode) {
1775                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1776
1777                 LASSERT(fd != NULL);
1778                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1779                 if (!rc)
1780                         fd->fd_flags |= LL_FILE_RMTACL;
1781                 RETURN(rc);
1782             } else
1783                 RETURN(0);
1784         }
1785 #endif
1786         case LL_IOC_GETOBDCOUNT: {
1787                 int count, vallen;
1788                 struct obd_export *exp;
1789
1790                 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1791                         RETURN(-EFAULT);
1792
1793                 /* get ost count when count is zero, get mdt count otherwise */
1794                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1795                 vallen = sizeof(count);
1796                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1797                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1798                 if (rc) {
1799                         CERROR("get target count failed: %d\n", rc);
1800                         RETURN(rc);
1801                 }
1802
1803                 if (copy_to_user((int *)arg, &count, sizeof(int)))
1804                         RETURN(-EFAULT);
1805
1806                 RETURN(0);
1807         }
1808         case LL_IOC_PATH2FID:
1809                 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1810                                      sizeof(struct lu_fid)))
1811                         RETURN(-EFAULT);
1812                 RETURN(0);
1813         case LL_IOC_GET_CONNECT_FLAGS: {
1814                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1815         }
1816         case OBD_IOC_CHANGELOG_SEND:
1817         case OBD_IOC_CHANGELOG_CLEAR:
1818                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1819                                     sizeof(struct ioc_changelog));
1820                 RETURN(rc);
1821         case OBD_IOC_FID2PATH:
1822                 RETURN(ll_fid2path(inode, (void *)arg));
1823         case LL_IOC_HSM_REQUEST: {
1824                 struct hsm_user_request *hur;
1825                 int                      totalsize;
1826
1827                 OBD_ALLOC_PTR(hur);
1828                 if (hur == NULL)
1829                         RETURN(-ENOMEM);
1830
1831                 /* We don't know the true size yet; copy the fixed-size part */
1832                 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1833                         OBD_FREE_PTR(hur);
1834                         RETURN(-EFAULT);
1835                 }
1836
1837                 /* Compute the whole struct size */
1838                 totalsize = hur_len(hur);
1839                 OBD_FREE_PTR(hur);
1840
1841                 /* Make sure the size is reasonable */
1842                 if (totalsize >= MDS_MAXREQSIZE)
1843                         RETURN(-E2BIG);
1844
1845                 OBD_ALLOC_LARGE(hur, totalsize);
1846                 if (hur == NULL)
1847                         RETURN(-ENOMEM);
1848
1849                 /* Copy the whole struct */
1850                 if (copy_from_user(hur, (void *)arg, totalsize)) {
1851                         OBD_FREE_LARGE(hur, totalsize);
1852                         RETURN(-EFAULT);
1853                 }
1854
1855                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1856                         const struct lu_fid *fid;
1857                         struct inode *f;
1858                         int i;
1859
1860                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1861                                 fid = &hur->hur_user_item[i].hui_fid;
1862                                 f = search_inode_for_lustre(inode->i_sb, fid);
1863                                 if (IS_ERR(f)) {
1864                                         rc = PTR_ERR(f);
1865                                         break;
1866                                 }
1867
1868                                 rc = ll_hsm_release(f);
1869                                 iput(f);
1870                                 if (rc != 0)
1871                                         break;
1872                         }
1873                 } else {
1874                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1875                                            hur, NULL);
1876                 }
1877
1878                 OBD_FREE_LARGE(hur, totalsize);
1879
1880                 RETURN(rc);
1881         }
1882         case LL_IOC_HSM_PROGRESS: {
1883                 struct hsm_progress_kernel      hpk;
1884                 struct hsm_progress             hp;
1885
1886                 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1887                         RETURN(-EFAULT);
1888
1889                 hpk.hpk_fid = hp.hp_fid;
1890                 hpk.hpk_cookie = hp.hp_cookie;
1891                 hpk.hpk_extent = hp.hp_extent;
1892                 hpk.hpk_flags = hp.hp_flags;
1893                 hpk.hpk_errval = hp.hp_errval;
1894                 hpk.hpk_data_version = 0;
1895
1896                 /* File may not exist in Lustre; all progress
1897                  * reported to Lustre root */
1898                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1899                                    NULL);
1900                 RETURN(rc);
1901         }
1902         case LL_IOC_HSM_CT_START:
1903                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1904                                     sizeof(struct lustre_kernelcomm));
1905                 RETURN(rc);
1906
1907         case LL_IOC_HSM_COPY_START: {
1908                 struct hsm_copy *copy;
1909                 int              rc;
1910
1911                 OBD_ALLOC_PTR(copy);
1912                 if (copy == NULL)
1913                         RETURN(-ENOMEM);
1914                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1915                         OBD_FREE_PTR(copy);
1916                         RETURN(-EFAULT);
1917                 }
1918
1919                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1920                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1921                         rc = -EFAULT;
1922
1923                 OBD_FREE_PTR(copy);
1924                 RETURN(rc);
1925         }
1926         case LL_IOC_HSM_COPY_END: {
1927                 struct hsm_copy *copy;
1928                 int              rc;
1929
1930                 OBD_ALLOC_PTR(copy);
1931                 if (copy == NULL)
1932                         RETURN(-ENOMEM);
1933                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1934                         OBD_FREE_PTR(copy);
1935                         RETURN(-EFAULT);
1936                 }
1937
1938                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1939                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1940                         rc = -EFAULT;
1941
1942                 OBD_FREE_PTR(copy);
1943                 RETURN(rc);
1944         }
1945         default:
1946                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1947                                      (void *)arg));
1948         }
1949 }
1950
1951 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1952 {
1953         struct inode *inode = file->f_mapping->host;
1954         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1955         struct ll_sb_info *sbi = ll_i2sbi(inode);
1956         int api32 = ll_need_32bit_api(sbi);
1957         loff_t ret = -EINVAL;
1958         ENTRY;
1959
1960         mutex_lock(&inode->i_mutex);
1961         switch (origin) {
1962                 case SEEK_SET:
1963                         break;
1964                 case SEEK_CUR:
1965                         offset += file->f_pos;
1966                         break;
1967                 case SEEK_END:
1968                         if (offset > 0)
1969                                 GOTO(out, ret);
1970                         if (api32)
1971                                 offset += LL_DIR_END_OFF_32BIT;
1972                         else
1973                                 offset += LL_DIR_END_OFF;
1974                         break;
1975                 default:
1976                         GOTO(out, ret);
1977         }
1978
1979         if (offset >= 0 &&
1980             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1981              (!api32 && offset <= LL_DIR_END_OFF))) {
1982                 if (offset != file->f_pos) {
1983                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1984                             (!api32 && offset == LL_DIR_END_OFF))
1985                                 fd->lfd_pos = MDS_DIR_END_OFF;
1986                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1987                                 fd->lfd_pos = offset << 32;
1988                         else
1989                                 fd->lfd_pos = offset;
1990                         file->f_pos = offset;
1991                         file->f_version = 0;
1992                 }
1993                 ret = offset;
1994         }
1995         GOTO(out, ret);
1996
1997 out:
1998         mutex_unlock(&inode->i_mutex);
1999         return ret;
2000 }
2001
2002 int ll_dir_open(struct inode *inode, struct file *file)
2003 {
2004         ENTRY;
2005         RETURN(ll_file_open(inode, file));
2006 }
2007
2008 int ll_dir_release(struct inode *inode, struct file *file)
2009 {
2010         ENTRY;
2011         RETURN(ll_file_release(inode, file));
2012 }
2013
2014 struct file_operations ll_dir_operations = {
2015         .llseek   = ll_dir_seek,
2016         .open     = ll_dir_open,
2017         .release  = ll_dir_release,
2018         .read     = generic_read_dir,
2019         .readdir  = ll_readdir,
2020         .unlocked_ioctl   = ll_dir_ioctl,
2021         .fsync    = ll_fsync,
2022 };