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="DFID"(%p) hash "LPU64"\n",
 164                PFID(ll_inode2fid(inode)), 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         spin_lock_irq(&mapping->tree_lock);
 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                 spin_unlock_irq(&mapping->tree_lock);
 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                 spin_unlock_irq(&mapping->tree_lock);
 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, dir, 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 "DFID"(%p)\n",
 383                        PFID(ll_inode2fid(dir)), dir);
 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="DFID"(%p) pos %lu/%llu "
 613                " 32bit_api %d\n", PFID(ll_inode2fid(inode)),
 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                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 837                 GOTO(out, rc);
 838         }
 839
 840         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 841         LASSERT(body != NULL);
 842
 843         lmmsize = body->eadatasize;
 844
 845         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 846             lmmsize == 0) {
 847                 GOTO(out, rc = -ENODATA);
 848         }
 849
 850         lmm = req_capsule_server_sized_get(&req->rq_pill,
 851                                            &RMF_MDT_MD, lmmsize);
 852         LASSERT(lmm != NULL);
 853
 854         /*
 855          * This is coming from the MDS, so is probably in
 856          * little endian.  We convert it to host endian before
 857          * passing it to userspace.
 858          */
 859         /* We don't swab objects for directories */
 860         switch (le32_to_cpu(lmm->lmm_magic)) {
 861         case LOV_MAGIC_V1:
 862                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 863                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 864                 break;
 865         case LOV_MAGIC_V3:
 866                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 867                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 868                 break;
 869         default:
 870                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 871                 rc = -EPROTO;
 872         }
 873 out:
 874         *lmmp = lmm;
 875         *lmm_size = lmmsize;
 876         *request = req;
 877         return rc;
 878 }
 879
 880 /*
 881  *  Get MDT index for the inode.
 882  */
 883 int ll_get_mdt_idx(struct inode *inode)
 884 {
 885         struct ll_sb_info *sbi = ll_i2sbi(inode);
 886         struct md_op_data *op_data;
 887         int rc, mdtidx;
 888         ENTRY;
 889
 890         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 891                                      0, LUSTRE_OPC_ANY, NULL);
 892         if (IS_ERR(op_data))
 893                 RETURN(PTR_ERR(op_data));
 894
 895         op_data->op_flags |= MF_GET_MDT_IDX;
 896         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 897         mdtidx = op_data->op_mds;
 898         ll_finish_md_op_data(op_data);
 899         if (rc < 0) {
 900                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 901                 RETURN(rc);
 902         }
 903         return mdtidx;
 904 }
 905
 906 /**
 907  * Generic handler to do any pre-copy work.
 908  *
 909  * It send a first hsm_progress (with extent length == 0) to coordinator as a
 910  * first information for it that real work has started.
 911  *
 912  * Moreover, for a ARCHIVE request, it will sample the file data version and
 913  * store it in \a copy.
 914  *
 915  * \return 0 on success.
 916  */
 917 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 918 {
 919         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 920         struct hsm_progress_kernel       hpk;
 921         int                              rc;
 922         ENTRY;
 923
 924         /* Forge a hsm_progress based on data from copy. */
 925         hpk.hpk_fid = copy->hc_hai.hai_fid;
 926         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 927         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 928         hpk.hpk_extent.length = 0;
 929         hpk.hpk_flags = 0;
 930         hpk.hpk_errval = 0;
 931         hpk.hpk_data_version = 0;
 932
 933
 934         /* For archive request, we need to read the current file version. */
 935         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 936                 struct inode    *inode;
 937                 __u64            data_version = 0;
 938
 939                 /* Get inode for this fid */
 940                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 941                 if (IS_ERR(inode)) {
 942                         hpk.hpk_flags |= HP_FLAG_RETRY;
 943                         /* hpk_errval is >= 0 */
 944                         hpk.hpk_errval = -PTR_ERR(inode);
 945                         GOTO(progress, rc = PTR_ERR(inode));
 946                 }
 947
 948                 /* Read current file data version */
 949                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 950                 iput(inode);
 951                 if (rc != 0) {
 952                         CDEBUG(D_HSM, "Could not read file data version of "
 953                                       DFID" (rc = %d). Archive request ("
 954                                       LPX64") could not be done.\n",
 955                                       PFID(&copy->hc_hai.hai_fid), rc,
 956                                       copy->hc_hai.hai_cookie);
 957                         hpk.hpk_flags |= HP_FLAG_RETRY;
 958                         /* hpk_errval must be >= 0 */
 959                         hpk.hpk_errval = -rc;
 960                         GOTO(progress, rc);
 961                 }
 962
 963                 /* Store it the hsm_copy for later copytool use.
 964                  * Always modified even if no lsm. */
 965                 copy->hc_data_version = data_version;
 966         }
 967
 968 progress:
 969         /* On error, the request should be considered as completed */
 970         if (hpk.hpk_errval > 0)
 971                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 972         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 973                            &hpk, NULL);
 974
 975         RETURN(rc);
 976 }
 977
 978 /**
 979  * Generic handler to do any post-copy work.
 980  *
 981  * It will send the last hsm_progress update to coordinator to inform it
 982  * that copy is finished and whether it was successful or not.
 983  *
 984  * Moreover,
 985  * - for ARCHIVE request, it will sample the file data version and compare it
 986  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 987  *   will be considered as failed.
 988  * - for RESTORE request, it will sample the file data version and send it to
 989  *   coordinator which is useful if the file was imported as 'released'.
 990  *
 991  * \return 0 on success.
 992  */
 993 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 994 {
 995         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 996         struct hsm_progress_kernel       hpk;
 997         int                              rc;
 998         ENTRY;
 999
1000         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
1001         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
1002          * initialized if copy_end was called with copy == NULL.
1003          */
1004
1005         /* Forge a hsm_progress based on data from copy. */
1006         hpk.hpk_fid = copy->hc_hai.hai_fid;
1007         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
1008         hpk.hpk_extent = copy->hc_hai.hai_extent;
1009         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
1010         hpk.hpk_errval = copy->hc_errval;
1011         hpk.hpk_data_version = 0;
1012
1013         /* For archive request, we need to check the file data was not changed.
1014          *
1015          * For restore request, we need to send the file data version, this is
1016          * useful when the file was created using hsm_import.
1017          */
1018         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
1019              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
1020             (copy->hc_errval == 0)) {
1021                 struct inode    *inode;
1022                 __u64            data_version = 0;
1023
1024                 /* Get lsm for this fid */
1025                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
1026                 if (IS_ERR(inode)) {
1027                         hpk.hpk_flags |= HP_FLAG_RETRY;
1028                         /* hpk_errval must be >= 0 */
1029                         hpk.hpk_errval = -PTR_ERR(inode);
1030                         GOTO(progress, rc = PTR_ERR(inode));
1031                 }
1032
1033                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
1034                 iput(inode);
1035                 if (rc) {
1036                         CDEBUG(D_HSM, "Could not read file data version. "
1037                                       "Request could not be confirmed.\n");
1038                         if (hpk.hpk_errval == 0)
1039                                 hpk.hpk_errval = -rc;
1040                         GOTO(progress, rc);
1041                 }
1042
1043                 /* Store it the hsm_copy for later copytool use.
1044                  * Always modified even if no lsm. */
1045                 hpk.hpk_data_version = data_version;
1046
1047                 /* File could have been stripped during archiving, so we need
1048                  * to check anyway. */
1049                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1050                     (copy->hc_data_version != data_version)) {
1051                         CDEBUG(D_HSM, "File data version mismatched. "
1052                               "File content was changed during archiving. "
1053                                DFID", start:"LPX64" current:"LPX64"\n",
1054                                PFID(&copy->hc_hai.hai_fid),
1055                                copy->hc_data_version, data_version);
1056                         /* File was changed, send error to cdt. Do not ask for
1057                          * retry because if a file is modified frequently,
1058                          * the cdt will loop on retried archive requests.
1059                          * The policy engine will ask for a new archive later
1060                          * when the file will not be modified for some tunable
1061                          * time */
1062                         /* we do not notify caller */
1063                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
1064                         /* hpk_errval must be >= 0 */
1065                         hpk.hpk_errval = EBUSY;
1066                 }
1067
1068         }
1069
1070 progress:
1071         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1072                            &hpk, NULL);
1073
1074         RETURN(rc);
1075 }
1076
1077
1078 static int copy_and_ioctl(int cmd, struct obd_export *exp,
1079                           const void __user *data, size_t size)
1080 {
1081         void *copy;
1082         int rc;
1083
1084         OBD_ALLOC(copy, size);
1085         if (copy == NULL)
1086                 return -ENOMEM;
1087
1088         if (copy_from_user(copy, data, size)) {
1089                 rc = -EFAULT;
1090                 goto out;
1091         }
1092
1093         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
1094 out:
1095         OBD_FREE(copy, size);
1096
1097         return rc;
1098 }
1099
1100 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1101 {
1102         int cmd = qctl->qc_cmd;
1103         int type = qctl->qc_type;
1104         int id = qctl->qc_id;
1105         int valid = qctl->qc_valid;
1106         int rc = 0;
1107         ENTRY;
1108
1109         switch (cmd) {
1110         case LUSTRE_Q_INVALIDATE:
1111         case LUSTRE_Q_FINVALIDATE:
1112         case Q_QUOTAON:
1113         case Q_QUOTAOFF:
1114         case Q_SETQUOTA:
1115         case Q_SETINFO:
1116                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1117                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1118                         RETURN(-EPERM);
1119                 break;
1120         case Q_GETQUOTA:
1121                 if (((type == USRQUOTA && current_euid() != id) ||
1122                      (type == GRPQUOTA && !in_egroup_p(id))) &&
1123                     (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1124                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
1125                         RETURN(-EPERM);
1126                 break;
1127         case Q_GETINFO:
1128                 break;
1129         default:
1130                 CERROR("unsupported quotactl op: %#x\n", cmd);
1131                 RETURN(-ENOTTY);
1132         }
1133
1134         if (valid != QC_GENERAL) {
1135                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1136                         RETURN(-EOPNOTSUPP);
1137
1138                 if (cmd == Q_GETINFO)
1139                         qctl->qc_cmd = Q_GETOINFO;
1140                 else if (cmd == Q_GETQUOTA)
1141                         qctl->qc_cmd = Q_GETOQUOTA;
1142                 else
1143                         RETURN(-EINVAL);
1144
1145                 switch (valid) {
1146                 case QC_MDTIDX:
1147                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1148                                            sizeof(*qctl), qctl, NULL);
1149                         break;
1150                 case QC_OSTIDX:
1151                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1152                                            sizeof(*qctl), qctl, NULL);
1153                         break;
1154                 case QC_UUID:
1155                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1156                                            sizeof(*qctl), qctl, NULL);
1157                         if (rc == -EAGAIN)
1158                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1159                                                    sbi->ll_dt_exp,
1160                                                    sizeof(*qctl), qctl, NULL);
1161                         break;
1162                 default:
1163                         rc = -EINVAL;
1164                         break;
1165                 }
1166
1167                 if (rc)
1168                         RETURN(rc);
1169
1170                 qctl->qc_cmd = cmd;
1171         } else {
1172                 struct obd_quotactl *oqctl;
1173
1174                 OBD_ALLOC_PTR(oqctl);
1175                 if (oqctl == NULL)
1176                         RETURN(-ENOMEM);
1177
1178                 QCTL_COPY(oqctl, qctl);
1179                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1180                 if (rc) {
1181                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
1182                                 oqctl->qc_cmd = Q_QUOTAOFF;
1183                                 obd_quotactl(sbi->ll_md_exp, oqctl);
1184                         }
1185                         OBD_FREE_PTR(oqctl);
1186                         RETURN(rc);
1187                 }
1188                 /* If QIF_SPACE is not set, client should collect the
1189                  * space usage from OSSs by itself */
1190                 if (cmd == Q_GETQUOTA &&
1191                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1192                     !oqctl->qc_dqblk.dqb_curspace) {
1193                         struct obd_quotactl *oqctl_tmp;
1194
1195                         OBD_ALLOC_PTR(oqctl_tmp);
1196                         if (oqctl_tmp == NULL)
1197                                 GOTO(out, rc = -ENOMEM);
1198
1199                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1200                         oqctl_tmp->qc_id = oqctl->qc_id;
1201                         oqctl_tmp->qc_type = oqctl->qc_type;
1202
1203                         /* collect space usage from OSTs */
1204                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1205                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1206                         if (!rc || rc == -EREMOTEIO) {
1207                                 oqctl->qc_dqblk.dqb_curspace =
1208                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1209                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1210                         }
1211
1212                         /* collect space & inode usage from MDTs */
1213                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1214                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1215                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1216                         if (!rc || rc == -EREMOTEIO) {
1217                                 oqctl->qc_dqblk.dqb_curspace +=
1218                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1219                                 oqctl->qc_dqblk.dqb_curinodes =
1220                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1221                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1222                         } else {
1223                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1224                         }
1225
1226                         OBD_FREE_PTR(oqctl_tmp);
1227                 }
1228 out:
1229                 QCTL_COPY(qctl, oqctl);
1230                 OBD_FREE_PTR(oqctl);
1231         }
1232
1233         RETURN(rc);
1234 }
1235
1236 static char *
1237 ll_getname(const char __user *filename)
1238 {
1239         int ret = 0, len;
1240         char *tmp = __getname();
1241
1242         if (!tmp)
1243                 return ERR_PTR(-ENOMEM);
1244
1245         len = strncpy_from_user(tmp, filename, PATH_MAX);
1246         if (len == 0)
1247                 ret = -ENOENT;
1248         else if (len > PATH_MAX)
1249                 ret = -ENAMETOOLONG;
1250
1251         if (ret) {
1252                 __putname(tmp);
1253                 tmp =  ERR_PTR(ret);
1254         }
1255         return tmp;
1256 }
1257
1258 #define ll_putname(filename) __putname(filename)
1259
1260 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1261 {
1262         struct inode *inode = file->f_dentry->d_inode;
1263         struct ll_sb_info *sbi = ll_i2sbi(inode);
1264         struct obd_ioctl_data *data;
1265         int rc = 0;
1266         ENTRY;
1267
1268         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1269                PFID(ll_inode2fid(inode)), inode, cmd);
1270
1271         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1272         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1273                 return -ENOTTY;
1274
1275         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1276         switch(cmd) {
1277         case FSFILT_IOC_GETFLAGS:
1278         case FSFILT_IOC_SETFLAGS:
1279                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1280         case FSFILT_IOC_GETVERSION_OLD:
1281         case FSFILT_IOC_GETVERSION:
1282                 RETURN(put_user(inode->i_generation, (int *)arg));
1283         /* We need to special case any other ioctls we want to handle,
1284          * to send them to the MDS/OST as appropriate and to properly
1285          * network encode the arg field.
1286         case FSFILT_IOC_SETVERSION_OLD:
1287         case FSFILT_IOC_SETVERSION:
1288         */
1289         case LL_IOC_GET_MDTIDX: {
1290                 int mdtidx;
1291
1292                 mdtidx = ll_get_mdt_idx(inode);
1293                 if (mdtidx < 0)
1294                         RETURN(mdtidx);
1295
1296                 if (put_user((int)mdtidx, (int*)arg))
1297                         RETURN(-EFAULT);
1298
1299                 return 0;
1300         }
1301         case IOC_MDC_LOOKUP: {
1302                 struct ptlrpc_request *request = NULL;
1303                 int namelen, len = 0;
1304                 char *buf = NULL;
1305                 char *filename;
1306                 struct md_op_data *op_data;
1307
1308                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1309                 if (rc)
1310                         RETURN(rc);
1311                 data = (void *)buf;
1312
1313                 filename = data->ioc_inlbuf1;
1314                 namelen = strlen(filename);
1315
1316                 if (namelen < 1) {
1317                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1318                         GOTO(out_free, rc = -EINVAL);
1319                 }
1320
1321                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1322                                              0, LUSTRE_OPC_ANY, NULL);
1323                 if (IS_ERR(op_data))
1324                         GOTO(out_free, rc = PTR_ERR(op_data));
1325
1326                 op_data->op_valid = OBD_MD_FLID;
1327                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1328                 ll_finish_md_op_data(op_data);
1329                 if (rc < 0) {
1330                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1331                         GOTO(out_free, rc);
1332                 }
1333                 ptlrpc_req_finished(request);
1334                 EXIT;
1335 out_free:
1336                 obd_ioctl_freedata(buf, len);
1337                 return rc;
1338         }
1339         case LL_IOC_LMV_SETSTRIPE: {
1340                 struct lmv_user_md  *lum;
1341                 char            *buf = NULL;
1342                 char            *filename;
1343                 int              namelen = 0;
1344                 int              lumlen = 0;
1345                 int              len;
1346                 int              rc;
1347
1348                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1349                 if (rc)
1350                         RETURN(rc);
1351
1352                 data = (void *)buf;
1353                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1354                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1355                         GOTO(lmv_out_free, rc = -EINVAL);
1356
1357                 filename = data->ioc_inlbuf1;
1358                 namelen = data->ioc_inllen1;
1359
1360                 if (namelen < 1) {
1361                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1362                         GOTO(lmv_out_free, rc = -EINVAL);
1363                 }
1364                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1365                 lumlen = data->ioc_inllen2;
1366
1367                 if (lum->lum_magic != LMV_USER_MAGIC ||
1368                     lumlen != sizeof(*lum)) {
1369                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1370                                filename, lum->lum_magic, lumlen, -EFAULT);
1371                         GOTO(lmv_out_free, rc = -EINVAL);
1372                 }
1373
1374                 /**
1375                  * ll_dir_setdirstripe will be used to set dir stripe
1376                  *  mdc_create--->mdt_reint_create (with dirstripe)
1377                  */
1378                 rc = ll_dir_setdirstripe(inode, lum, filename);
1379 lmv_out_free:
1380                 obd_ioctl_freedata(buf, len);
1381                 RETURN(rc);
1382
1383         }
1384         case LL_IOC_LOV_SETSTRIPE: {
1385                 struct lov_user_md_v3 lumv3;
1386                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1387                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1388                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1389
1390                 int set_default = 0;
1391
1392                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1393                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1394                         sizeof(lumv3p->lmm_objects[0]));
1395                 /* first try with v1 which is smaller than v3 */
1396                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1397                         RETURN(-EFAULT);
1398
1399                 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1400                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1401                                 RETURN(-EFAULT);
1402                 }
1403
1404                 if (inode->i_sb->s_root == file->f_dentry)
1405                         set_default = 1;
1406
1407                 /* in v1 and v3 cases lumv1 points to data */
1408                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1409
1410                 RETURN(rc);
1411         }
1412         case LL_IOC_LMV_GETSTRIPE: {
1413                 struct lmv_user_md *lump = (struct lmv_user_md *)arg;
1414                 struct lmv_user_md lum;
1415                 struct lmv_user_md *tmp;
1416                 int lum_size;
1417                 int rc = 0;
1418                 int mdtindex;
1419
1420                 if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
1421                         RETURN(-EFAULT);
1422
1423                 if (lum.lum_magic != LMV_MAGIC_V1)
1424                         RETURN(-EINVAL);
1425
1426                 lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
1427                 OBD_ALLOC(tmp, lum_size);
1428                 if (tmp == NULL)
1429                         GOTO(free_lmv, rc = -ENOMEM);
1430
1431                 memcpy(tmp, &lum, sizeof(lum));
1432                 tmp->lum_type = LMV_STRIPE_TYPE;
1433                 tmp->lum_stripe_count = 1;
1434                 mdtindex = ll_get_mdt_idx(inode);
1435                 if (mdtindex < 0)
1436                         GOTO(free_lmv, rc = -ENOMEM);
1437
1438                 tmp->lum_stripe_offset = mdtindex;
1439                 tmp->lum_objects[0].lum_mds = mdtindex;
1440                 memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
1441                        sizeof(struct lu_fid));
1442                 if (copy_to_user((void *)arg, tmp, lum_size))
1443                         GOTO(free_lmv, rc = -EFAULT);
1444 free_lmv:
1445                 if (tmp)
1446                         OBD_FREE(tmp, lum_size);
1447                 RETURN(rc);
1448         }
1449         case LL_IOC_REMOVE_ENTRY: {
1450                 char            *filename = NULL;
1451                 int              namelen = 0;
1452                 int              rc;
1453
1454                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1455                  * unsupported server, which might crash the server(LU-2730),
1456                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1457                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1458                  * server will support REINT_RMENTRY XXX*/
1459                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1460                         RETURN(-ENOTSUPP);
1461
1462                 filename = ll_getname((const char *)arg);
1463                 if (IS_ERR(filename))
1464                         RETURN(PTR_ERR(filename));
1465
1466                 namelen = strlen(filename);
1467                 if (namelen < 1)
1468                         GOTO(out_rmdir, rc = -EINVAL);
1469
1470                 rc = ll_rmdir_entry(inode, filename, namelen);
1471 out_rmdir:
1472                 if (filename)
1473                         ll_putname(filename);
1474                 RETURN(rc);
1475         }
1476         case LL_IOC_LOV_SWAP_LAYOUTS:
1477                 RETURN(-EPERM);
1478         case LL_IOC_OBD_STATFS:
1479                 RETURN(ll_obd_statfs(inode, (void *)arg));
1480         case LL_IOC_LOV_GETSTRIPE:
1481         case LL_IOC_MDC_GETINFO:
1482         case IOC_MDC_GETFILEINFO:
1483         case IOC_MDC_GETFILESTRIPE: {
1484                 struct ptlrpc_request *request = NULL;
1485                 struct lov_user_md *lump;
1486                 struct lov_mds_md *lmm = NULL;
1487                 struct mdt_body *body;
1488                 char *filename = NULL;
1489                 int lmmsize;
1490
1491                 if (cmd == IOC_MDC_GETFILEINFO ||
1492                     cmd == IOC_MDC_GETFILESTRIPE) {
1493                         filename = ll_getname((const char *)arg);
1494                         if (IS_ERR(filename))
1495                                 RETURN(PTR_ERR(filename));
1496
1497                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1498                                                       &lmmsize, &request);
1499                 } else {
1500                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1501                 }
1502
1503                 if (request) {
1504                         body = req_capsule_server_get(&request->rq_pill,
1505                                                       &RMF_MDT_BODY);
1506                         LASSERT(body != NULL);
1507                 } else {
1508                         GOTO(out_req, rc);
1509                 }
1510
1511                 if (rc < 0) {
1512                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1513                                                cmd == LL_IOC_MDC_GETINFO))
1514                                 GOTO(skip_lmm, rc = 0);
1515                         else
1516                                 GOTO(out_req, rc);
1517                 }
1518
1519                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1520                     cmd == LL_IOC_LOV_GETSTRIPE) {
1521                         lump = (struct lov_user_md *)arg;
1522                 } else {
1523                         struct lov_user_mds_data *lmdp;
1524                         lmdp = (struct lov_user_mds_data *)arg;
1525                         lump = &lmdp->lmd_lmm;
1526                 }
1527                 if (copy_to_user(lump, lmm, lmmsize)) {
1528                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1529                                 GOTO(out_req, rc = -EFAULT);
1530                         rc = -EOVERFLOW;
1531                 }
1532         skip_lmm:
1533                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1534                         struct lov_user_mds_data *lmdp;
1535                         lstat_t st = { 0 };
1536
1537                         st.st_dev     = inode->i_sb->s_dev;
1538                         st.st_mode    = body->mode;
1539                         st.st_nlink   = body->nlink;
1540                         st.st_uid     = body->uid;
1541                         st.st_gid     = body->gid;
1542                         st.st_rdev    = body->rdev;
1543                         st.st_size    = body->size;
1544                         st.st_blksize = PAGE_CACHE_SIZE;
1545                         st.st_blocks  = body->blocks;
1546                         st.st_atime   = body->atime;
1547                         st.st_mtime   = body->mtime;
1548                         st.st_ctime   = body->ctime;
1549                         st.st_ino     = inode->i_ino;
1550
1551                         lmdp = (struct lov_user_mds_data *)arg;
1552                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1553                                 GOTO(out_req, rc = -EFAULT);
1554                 }
1555
1556                 EXIT;
1557         out_req:
1558                 ptlrpc_req_finished(request);
1559                 if (filename)
1560                         ll_putname(filename);
1561                 return rc;
1562         }
1563         case IOC_LOV_GETINFO: {
1564                 struct lov_user_mds_data *lumd;
1565                 struct lov_stripe_md *lsm;
1566                 struct lov_user_md *lum;
1567                 struct lov_mds_md *lmm;
1568                 int lmmsize;
1569                 lstat_t st;
1570
1571                 lumd = (struct lov_user_mds_data *)arg;
1572                 lum = &lumd->lmd_lmm;
1573
1574                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1575                 if (rc)
1576                         RETURN(rc);
1577
1578                 OBD_ALLOC_LARGE(lmm, lmmsize);
1579                 if (lmm == NULL)
1580                         RETURN(-ENOMEM);
1581
1582                 if (copy_from_user(lmm, lum, lmmsize))
1583                         GOTO(free_lmm, rc = -EFAULT);
1584
1585                 switch (lmm->lmm_magic) {
1586                 case LOV_USER_MAGIC_V1:
1587                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1588                                 break;
1589                         /* swab objects first so that stripes num will be sane */
1590                         lustre_swab_lov_user_md_objects(
1591                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1592                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1593                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1594                         break;
1595                 case LOV_USER_MAGIC_V3:
1596                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1597                                 break;
1598                         /* swab objects first so that stripes num will be sane */
1599                         lustre_swab_lov_user_md_objects(
1600                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1601                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1602                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1603                         break;
1604                 default:
1605                         GOTO(free_lmm, rc = -EINVAL);
1606                 }
1607
1608                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1609                 if (rc < 0)
1610                         GOTO(free_lmm, rc = -ENOMEM);
1611
1612                 /* Perform glimpse_size operation. */
1613                 memset(&st, 0, sizeof(st));
1614
1615                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1616                 if (rc)
1617                         GOTO(free_lsm, rc);
1618
1619                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1620                         GOTO(free_lsm, rc = -EFAULT);
1621
1622                 EXIT;
1623         free_lsm:
1624                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1625         free_lmm:
1626                 OBD_FREE_LARGE(lmm, lmmsize);
1627                 return rc;
1628         }
1629         case OBD_IOC_LLOG_CATINFO: {
1630                 RETURN(-EOPNOTSUPP);
1631         }
1632         case OBD_IOC_QUOTACHECK: {
1633                 struct obd_quotactl *oqctl;
1634                 int error = 0;
1635
1636                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1637                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1638                         RETURN(-EPERM);
1639
1640                 OBD_ALLOC_PTR(oqctl);
1641                 if (!oqctl)
1642                         RETURN(-ENOMEM);
1643                 oqctl->qc_type = arg;
1644                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1645                 if (rc < 0) {
1646                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1647                         error = rc;
1648                 }
1649
1650                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1651                 if (rc < 0)
1652                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1653
1654                 OBD_FREE_PTR(oqctl);
1655                 return error ?: rc;
1656         }
1657         case OBD_IOC_POLL_QUOTACHECK: {
1658                 struct if_quotacheck *check;
1659
1660                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1661                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1662                         RETURN(-EPERM);
1663
1664                 OBD_ALLOC_PTR(check);
1665                 if (!check)
1666                         RETURN(-ENOMEM);
1667
1668                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1669                                    NULL);
1670                 if (rc) {
1671                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1672                         if (copy_to_user((void *)arg, check,
1673                                              sizeof(*check)))
1674                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1675                         GOTO(out_poll, rc);
1676                 }
1677
1678                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1679                                    NULL);
1680                 if (rc) {
1681                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1682                         if (copy_to_user((void *)arg, check,
1683                                              sizeof(*check)))
1684                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1685                         GOTO(out_poll, rc);
1686                 }
1687         out_poll:
1688                 OBD_FREE_PTR(check);
1689                 RETURN(rc);
1690         }
1691 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
1692         case LL_IOC_QUOTACTL_18: {
1693                 /* copy the old 1.x quota struct for internal use, then copy
1694                  * back into old format struct.  For 1.8 compatibility. */
1695                 struct if_quotactl_18 *qctl_18;
1696                 struct if_quotactl *qctl_20;
1697
1698                 OBD_ALLOC_PTR(qctl_18);
1699                 if (!qctl_18)
1700                         RETURN(-ENOMEM);
1701
1702                 OBD_ALLOC_PTR(qctl_20);
1703                 if (!qctl_20)
1704                         GOTO(out_quotactl_18, rc = -ENOMEM);
1705
1706                 if (copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1707                         GOTO(out_quotactl_20, rc = -ENOMEM);
1708
1709                 QCTL_COPY(qctl_20, qctl_18);
1710                 qctl_20->qc_idx = 0;
1711
1712                 /* XXX: dqb_valid was borrowed as a flag to mark that
1713                  *      only mds quota is wanted */
1714                 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1715                     qctl_18->qc_dqblk.dqb_valid) {
1716                         qctl_20->qc_valid = QC_MDTIDX;
1717                         qctl_20->qc_dqblk.dqb_valid = 0;
1718                 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1719                         qctl_20->qc_valid = QC_UUID;
1720                         qctl_20->obd_uuid = qctl_18->obd_uuid;
1721                 } else {
1722                         qctl_20->qc_valid = QC_GENERAL;
1723                 }
1724
1725                 rc = quotactl_ioctl(sbi, qctl_20);
1726
1727                 if (rc == 0) {
1728                         QCTL_COPY(qctl_18, qctl_20);
1729                         qctl_18->obd_uuid = qctl_20->obd_uuid;
1730
1731                         if (copy_to_user((void *)arg, qctl_18,
1732                                              sizeof(*qctl_18)))
1733                                 rc = -EFAULT;
1734                 }
1735
1736         out_quotactl_20:
1737                 OBD_FREE_PTR(qctl_20);
1738         out_quotactl_18:
1739                 OBD_FREE_PTR(qctl_18);
1740                 RETURN(rc);
1741         }
1742 #else
1743 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1744 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
1745         case LL_IOC_QUOTACTL: {
1746                 struct if_quotactl *qctl;
1747
1748                 OBD_ALLOC_PTR(qctl);
1749                 if (!qctl)
1750                         RETURN(-ENOMEM);
1751
1752                 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1753                         GOTO(out_quotactl, rc = -EFAULT);
1754
1755                 rc = quotactl_ioctl(sbi, qctl);
1756
1757                 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1758                         rc = -EFAULT;
1759
1760         out_quotactl:
1761                 OBD_FREE_PTR(qctl);
1762                 RETURN(rc);
1763         }
1764         case OBD_IOC_GETDTNAME:
1765         case OBD_IOC_GETMDNAME:
1766                 RETURN(ll_get_obd_name(inode, cmd, arg));
1767         case LL_IOC_FLUSHCTX:
1768                 RETURN(ll_flush_ctx(inode));
1769 #ifdef CONFIG_FS_POSIX_ACL
1770         case LL_IOC_RMTACL: {
1771             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1772                 inode == inode->i_sb->s_root->d_inode) {
1773                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1774
1775                 LASSERT(fd != NULL);
1776                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1777                 if (!rc)
1778                         fd->fd_flags |= LL_FILE_RMTACL;
1779                 RETURN(rc);
1780             } else
1781                 RETURN(0);
1782         }
1783 #endif
1784         case LL_IOC_GETOBDCOUNT: {
1785                 int count, vallen;
1786                 struct obd_export *exp;
1787
1788                 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1789                         RETURN(-EFAULT);
1790
1791                 /* get ost count when count is zero, get mdt count otherwise */
1792                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1793                 vallen = sizeof(count);
1794                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1795                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1796                 if (rc) {
1797                         CERROR("get target count failed: %d\n", rc);
1798                         RETURN(rc);
1799                 }
1800
1801                 if (copy_to_user((int *)arg, &count, sizeof(int)))
1802                         RETURN(-EFAULT);
1803
1804                 RETURN(0);
1805         }
1806         case LL_IOC_PATH2FID:
1807                 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1808                                      sizeof(struct lu_fid)))
1809                         RETURN(-EFAULT);
1810                 RETURN(0);
1811         case LL_IOC_GET_CONNECT_FLAGS: {
1812                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1813         }
1814         case OBD_IOC_CHANGELOG_SEND:
1815         case OBD_IOC_CHANGELOG_CLEAR:
1816                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1817                                     sizeof(struct ioc_changelog));
1818                 RETURN(rc);
1819         case OBD_IOC_FID2PATH:
1820                 RETURN(ll_fid2path(inode, (void *)arg));
1821         case LL_IOC_HSM_REQUEST: {
1822                 struct hsm_user_request *hur;
1823                 int                      totalsize;
1824
1825                 OBD_ALLOC_PTR(hur);
1826                 if (hur == NULL)
1827                         RETURN(-ENOMEM);
1828
1829                 /* We don't know the true size yet; copy the fixed-size part */
1830                 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1831                         OBD_FREE_PTR(hur);
1832                         RETURN(-EFAULT);
1833                 }
1834
1835                 /* Compute the whole struct size */
1836                 totalsize = hur_len(hur);
1837                 OBD_FREE_PTR(hur);
1838
1839                 /* Make sure the size is reasonable */
1840                 if (totalsize >= MDS_MAXREQSIZE)
1841                         RETURN(-E2BIG);
1842
1843                 OBD_ALLOC_LARGE(hur, totalsize);
1844                 if (hur == NULL)
1845                         RETURN(-ENOMEM);
1846
1847                 /* Copy the whole struct */
1848                 if (copy_from_user(hur, (void *)arg, totalsize)) {
1849                         OBD_FREE_LARGE(hur, totalsize);
1850                         RETURN(-EFAULT);
1851                 }
1852
1853                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1854                         const struct lu_fid *fid;
1855                         struct inode *f;
1856                         int i;
1857
1858                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1859                                 fid = &hur->hur_user_item[i].hui_fid;
1860                                 f = search_inode_for_lustre(inode->i_sb, fid);
1861                                 if (IS_ERR(f)) {
1862                                         rc = PTR_ERR(f);
1863                                         break;
1864                                 }
1865
1866                                 rc = ll_hsm_release(f);
1867                                 iput(f);
1868                                 if (rc != 0)
1869                                         break;
1870                         }
1871                 } else {
1872                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1873                                            hur, NULL);
1874                 }
1875
1876                 OBD_FREE_LARGE(hur, totalsize);
1877
1878                 RETURN(rc);
1879         }
1880         case LL_IOC_HSM_PROGRESS: {
1881                 struct hsm_progress_kernel      hpk;
1882                 struct hsm_progress             hp;
1883
1884                 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1885                         RETURN(-EFAULT);
1886
1887                 hpk.hpk_fid = hp.hp_fid;
1888                 hpk.hpk_cookie = hp.hp_cookie;
1889                 hpk.hpk_extent = hp.hp_extent;
1890                 hpk.hpk_flags = hp.hp_flags;
1891                 hpk.hpk_errval = hp.hp_errval;
1892                 hpk.hpk_data_version = 0;
1893
1894                 /* File may not exist in Lustre; all progress
1895                  * reported to Lustre root */
1896                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1897                                    NULL);
1898                 RETURN(rc);
1899         }
1900         case LL_IOC_HSM_CT_START:
1901                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1902                         RETURN(-EPERM);
1903
1904                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1905                                     sizeof(struct lustre_kernelcomm));
1906                 RETURN(rc);
1907
1908         case LL_IOC_HSM_COPY_START: {
1909                 struct hsm_copy *copy;
1910                 int              rc;
1911
1912                 OBD_ALLOC_PTR(copy);
1913                 if (copy == NULL)
1914                         RETURN(-ENOMEM);
1915                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1916                         OBD_FREE_PTR(copy);
1917                         RETURN(-EFAULT);
1918                 }
1919
1920                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1921                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1922                         rc = -EFAULT;
1923
1924                 OBD_FREE_PTR(copy);
1925                 RETURN(rc);
1926         }
1927         case LL_IOC_HSM_COPY_END: {
1928                 struct hsm_copy *copy;
1929                 int              rc;
1930
1931                 OBD_ALLOC_PTR(copy);
1932                 if (copy == NULL)
1933                         RETURN(-ENOMEM);
1934                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1935                         OBD_FREE_PTR(copy);
1936                         RETURN(-EFAULT);
1937                 }
1938
1939                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1940                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1941                         rc = -EFAULT;
1942
1943                 OBD_FREE_PTR(copy);
1944                 RETURN(rc);
1945         }
1946         default:
1947                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1948                                      (void *)arg));
1949         }
1950 }
1951
1952 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1953 {
1954         struct inode *inode = file->f_mapping->host;
1955         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1956         struct ll_sb_info *sbi = ll_i2sbi(inode);
1957         int api32 = ll_need_32bit_api(sbi);
1958         loff_t ret = -EINVAL;
1959         ENTRY;
1960
1961         mutex_lock(&inode->i_mutex);
1962         switch (origin) {
1963                 case SEEK_SET:
1964                         break;
1965                 case SEEK_CUR:
1966                         offset += file->f_pos;
1967                         break;
1968                 case SEEK_END:
1969                         if (offset > 0)
1970                                 GOTO(out, ret);
1971                         if (api32)
1972                                 offset += LL_DIR_END_OFF_32BIT;
1973                         else
1974                                 offset += LL_DIR_END_OFF;
1975                         break;
1976                 default:
1977                         GOTO(out, ret);
1978         }
1979
1980         if (offset >= 0 &&
1981             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1982              (!api32 && offset <= LL_DIR_END_OFF))) {
1983                 if (offset != file->f_pos) {
1984                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1985                             (!api32 && offset == LL_DIR_END_OFF))
1986                                 fd->lfd_pos = MDS_DIR_END_OFF;
1987                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1988                                 fd->lfd_pos = offset << 32;
1989                         else
1990                                 fd->lfd_pos = offset;
1991                         file->f_pos = offset;
1992                         file->f_version = 0;
1993                 }
1994                 ret = offset;
1995         }
1996         GOTO(out, ret);
1997
1998 out:
1999         mutex_unlock(&inode->i_mutex);
2000         return ret;
2001 }
2002
2003 int ll_dir_open(struct inode *inode, struct file *file)
2004 {
2005         ENTRY;
2006         RETURN(ll_file_open(inode, file));
2007 }
2008
2009 int ll_dir_release(struct inode *inode, struct file *file)
2010 {
2011         ENTRY;
2012         RETURN(ll_file_release(inode, file));
2013 }
2014
2015 struct file_operations ll_dir_operations = {
2016         .llseek   = ll_dir_seek,
2017         .open     = ll_dir_open,
2018         .release  = ll_dir_release,
2019         .read     = generic_read_dir,
2020         .readdir  = ll_readdir,
2021         .unlocked_ioctl   = ll_dir_ioctl,
2022         .fsync    = ll_fsync,
2023 };