drivers/staging/lustre/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, 2015, 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/uaccess.h>
  45 #include <linux/buffer_head.h>   /* for wait_on_buffer */
  46 #include <linux/pagevec.h>
  47 #include <linux/prefetch.h>
  48
  49 #define DEBUG_SUBSYSTEM S_LLITE
  50
  51 #include "../include/obd_support.h"
  52 #include "../include/obd_class.h"
  53 #include "../include/lustre_lib.h"
  54 #include "../include/lustre/lustre_idl.h"
  55 #include "../include/lustre_lite.h"
  56 #include "../include/lustre_dlm.h"
  57 #include "../include/lustre_fid.h"
  58 #include "../include/lustre_kernelcomm.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
 162         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) hash %llu\n",
 163                inode->i_ino, inode->i_generation, inode, hash);
 164
 165         LASSERT(max_pages > 0 && max_pages <= MD_MAX_BRW_PAGES);
 166
 167         page_pool = kcalloc(max_pages, sizeof(page), GFP_NOFS);
 168         if (page_pool) {
 169                 page_pool[0] = page0;
 170         } else {
 171                 page_pool = &page0;
 172                 max_pages = 1;
 173         }
 174         for (npages = 1; npages < max_pages; npages++) {
 175                 page = page_cache_alloc_cold(inode->i_mapping);
 176                 if (!page)
 177                         break;
 178                 page_pool[npages] = page;
 179         }
 180
 181         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 182                                      LUSTRE_OPC_ANY, NULL);
 183         op_data->op_npages = npages;
 184         op_data->op_offset = hash;
 185         rc = md_readpage(exp, op_data, page_pool, &request);
 186         ll_finish_md_op_data(op_data);
 187         if (rc < 0) {
 188                 /* page0 is special, which was added into page cache early */
 189                 delete_from_page_cache(page0);
 190         } else if (rc == 0) {
 191                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 192                 /* Checked by mdc_readpage() */
 193                 if (body->valid & OBD_MD_FLSIZE)
 194                         cl_isize_write(inode, body->size);
 195
 196                 nrdpgs = (request->rq_bulk->bd_nob_transferred+PAGE_CACHE_SIZE-1)
 197                          >> 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_NOFS);
 227                 if (ret == 0) {
 228                         unlock_page(page);
 229                 } else {
 230                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed: %d\n",
 231                                offset, ret);
 232                 }
 233                 page_cache_release(page);
 234         }
 235
 236         if (page_pool != &page0)
 237                 kfree(page_pool);
 238         return rc;
 239 }
 240
 241 void ll_release_page(struct page *page, int remove)
 242 {
 243         kunmap(page);
 244         if (remove) {
 245                 lock_page(page);
 246                 if (likely(page->mapping))
 247                         truncate_complete_page(page->mapping, page);
 248                 unlock_page(page);
 249         }
 250         page_cache_release(page);
 251 }
 252
 253 /*
 254  * Find, kmap and return page that contains given hash.
 255  */
 256 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
 257                                        __u64 *start, __u64 *end)
 258 {
 259         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 260         struct address_space *mapping = dir->i_mapping;
 261         /*
 262          * Complement of hash is used as an index so that
 263          * radix_tree_gang_lookup() can be used to find a page with starting
 264          * hash _smaller_ than one we are looking for.
 265          */
 266         unsigned long offset = hash_x_index(*hash, hash64);
 267         struct page *page;
 268         int found;
 269
 270         spin_lock_irq(&mapping->tree_lock);
 271         found = radix_tree_gang_lookup(&mapping->page_tree,
 272                                        (void **)&page, offset, 1);
 273         if (found > 0 && !radix_tree_exceptional_entry(page)) {
 274                 struct lu_dirpage *dp;
 275
 276                 page_cache_get(page);
 277                 spin_unlock_irq(&mapping->tree_lock);
 278                 /*
 279                  * In contrast to find_lock_page() we are sure that directory
 280                  * page cannot be truncated (while DLM lock is held) and,
 281                  * hence, can avoid restart.
 282                  *
 283                  * In fact, page cannot be locked here at all, because
 284                  * ll_dir_filler() does synchronous io.
 285                  */
 286                 wait_on_page_locked(page);
 287                 if (PageUptodate(page)) {
 288                         dp = kmap(page);
 289                         if (BITS_PER_LONG == 32 && hash64) {
 290                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 291                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 292                                 *hash  = *hash >> 32;
 293                         } else {
 294                                 *start = le64_to_cpu(dp->ldp_hash_start);
 295                                 *end   = le64_to_cpu(dp->ldp_hash_end);
 296                         }
 297                         LASSERTF(*start <= *hash, "start = %#llx,end = %#llx,hash = %#llx\n",
 298                                  *start, *end, *hash);
 299                         CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash %llu\n",
 300                                offset, *start, *end, *hash);
 301                         if (*hash > *end) {
 302                                 ll_release_page(page, 0);
 303                                 page = NULL;
 304                         } else if (*end != *start && *hash == *end) {
 305                                 /*
 306                                  * upon hash collision, remove this page,
 307                                  * otherwise put page reference, and
 308                                  * ll_get_dir_page() will issue RPC to fetch
 309                                  * the page we want.
 310                                  */
 311                                 ll_release_page(page,
 312                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
 313                                 page = NULL;
 314                         }
 315                 } else {
 316                         page_cache_release(page);
 317                         page = ERR_PTR(-EIO);
 318                 }
 319
 320         } else {
 321                 spin_unlock_irq(&mapping->tree_lock);
 322                 page = NULL;
 323         }
 324         return page;
 325 }
 326
 327 struct page *ll_get_dir_page(struct inode *dir, __u64 hash,
 328                              struct ll_dir_chain *chain)
 329 {
 330         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 331         struct address_space *mapping = dir->i_mapping;
 332         struct lustre_handle lockh;
 333         struct lu_dirpage *dp;
 334         struct page *page;
 335         enum ldlm_mode mode;
 336         int rc;
 337         __u64 start = 0;
 338         __u64 end = 0;
 339         __u64 lhash = hash;
 340         struct ll_inode_info *lli = ll_i2info(dir);
 341         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 342
 343         mode = LCK_PR;
 344         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 345                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
 346         if (!rc) {
 347                 struct ldlm_enqueue_info einfo = {
 348                         .ei_type = LDLM_IBITS,
 349                         .ei_mode = mode,
 350                         .ei_cb_bl = ll_md_blocking_ast,
 351                         .ei_cb_cp = ldlm_completion_ast,
 352                 };
 353                 struct lookup_intent it = { .it_op = IT_READDIR };
 354                 struct ptlrpc_request *request;
 355                 struct md_op_data *op_data;
 356
 357                 op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0,
 358                                              LUSTRE_OPC_ANY, NULL);
 359                 if (IS_ERR(op_data))
 360                         return (void *)op_data;
 361
 362                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
 363                                 op_data, &lockh, NULL, 0, NULL, 0);
 364
 365                 ll_finish_md_op_data(op_data);
 366
 367                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 368                 if (request)
 369                         ptlrpc_req_finished(request);
 370                 if (rc < 0) {
 371                         CERROR("lock enqueue: " DFID " at %llu: rc %d\n",
 372                                PFID(ll_inode2fid(dir)), hash, rc);
 373                         return ERR_PTR(rc);
 374                 }
 375
 376                 CDEBUG(D_INODE, "setting lr_lvb_inode to inode %p (%lu/%u)\n",
 377                        dir, dir->i_ino, dir->i_generation);
 378                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp,
 379                                  &it.d.lustre.it_lock_handle, dir, NULL);
 380         } else {
 381                 /* for cross-ref object, l_ast_data of the lock may not be set,
 382                  * we reset it here
 383                  */
 384                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
 385                                  dir, NULL);
 386         }
 387         ldlm_lock_dump_handle(D_OTHER, &lockh);
 388
 389         mutex_lock(&lli->lli_readdir_mutex);
 390         page = ll_dir_page_locate(dir, &lhash, &start, &end);
 391         if (IS_ERR(page)) {
 392                 CERROR("dir page locate: "DFID" at %llu: rc %ld\n",
 393                        PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
 394                 goto out_unlock;
 395         } else if (page) {
 396                 /*
 397                  * XXX nikita: not entirely correct handling of a corner case:
 398                  * suppose hash chain of entries with hash value HASH crosses
 399                  * border between pages P0 and P1. First both P0 and P1 are
 400                  * cached, seekdir() is called for some entry from the P0 part
 401                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 402                  * happens and finds P1, as it starts with matching hash
 403                  * value. Remaining entries from P0 part of the chain are
 404                  * skipped. (Is that really a bug?)
 405                  *
 406                  * Possible solutions: 0. don't cache P1 is such case, handle
 407                  * it as an "overflow" page. 1. invalidate all pages at
 408                  * once. 2. use HASH|1 as an index for P1.
 409                  */
 410                 goto hash_collision;
 411         }
 412
 413         page = read_cache_page(mapping, hash_x_index(hash, hash64),
 414                                ll_dir_filler, &lhash);
 415         if (IS_ERR(page)) {
 416                 CERROR("read cache page: "DFID" at %llu: rc %ld\n",
 417                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 418                 goto out_unlock;
 419         }
 420
 421         wait_on_page_locked(page);
 422         (void)kmap(page);
 423         if (!PageUptodate(page)) {
 424                 CERROR("page not updated: "DFID" at %llu: rc %d\n",
 425                        PFID(ll_inode2fid(dir)), hash, -5);
 426                 goto fail;
 427         }
 428         if (!PageChecked(page))
 429                 /* XXX: check page format later */
 430                 SetPageChecked(page);
 431         if (PageError(page)) {
 432                 CERROR("page error: "DFID" at %llu: rc %d\n",
 433                        PFID(ll_inode2fid(dir)), hash, -5);
 434                 goto fail;
 435         }
 436 hash_collision:
 437         dp = page_address(page);
 438         if (BITS_PER_LONG == 32 && hash64) {
 439                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 440                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 441                 lhash = hash >> 32;
 442         } else {
 443                 start = le64_to_cpu(dp->ldp_hash_start);
 444                 end   = le64_to_cpu(dp->ldp_hash_end);
 445                 lhash = hash;
 446         }
 447         if (end == start) {
 448                 LASSERT(start == lhash);
 449                 CWARN("Page-wide hash collision: %llu\n", end);
 450                 if (BITS_PER_LONG == 32 && hash64)
 451                         CWARN("Real page-wide hash collision at [%llu %llu] with hash %llu\n",
 452                               le64_to_cpu(dp->ldp_hash_start),
 453                               le64_to_cpu(dp->ldp_hash_end), hash);
 454                 /*
 455                  * Fetch whole overflow chain...
 456                  *
 457                  * XXX not yet.
 458                  */
 459                 goto fail;
 460         }
 461 out_unlock:
 462         mutex_unlock(&lli->lli_readdir_mutex);
 463         ldlm_lock_decref(&lockh, mode);
 464         return page;
 465
 466 fail:
 467         ll_release_page(page, 1);
 468         page = ERR_PTR(-EIO);
 469         goto out_unlock;
 470 }
 471
 472 int ll_dir_read(struct inode *inode, struct dir_context *ctx)
 473 {
 474         struct ll_inode_info *info       = ll_i2info(inode);
 475         struct ll_sb_info    *sbi       = ll_i2sbi(inode);
 476         __u64              pos          = ctx->pos;
 477         int                api32      = ll_need_32bit_api(sbi);
 478         int                hash64     = sbi->ll_flags & LL_SBI_64BIT_HASH;
 479         struct page       *page;
 480         struct ll_dir_chain   chain;
 481         int                done = 0;
 482         int                rc = 0;
 483
 484         ll_dir_chain_init(&chain);
 485
 486         page = ll_get_dir_page(inode, pos, &chain);
 487
 488         while (rc == 0 && !done) {
 489                 struct lu_dirpage *dp;
 490                 struct lu_dirent  *ent;
 491
 492                 if (!IS_ERR(page)) {
 493                         /*
 494                          * If page is empty (end of directory is reached),
 495                          * use this value.
 496                          */
 497                         __u64 hash = MDS_DIR_END_OFF;
 498                         __u64 next;
 499
 500                         dp = page_address(page);
 501                         for (ent = lu_dirent_start(dp); ent && !done;
 502                              ent = lu_dirent_next(ent)) {
 503                                 __u16     type;
 504                                 int         namelen;
 505                                 struct lu_fid  fid;
 506                                 __u64     lhash;
 507                                 __u64     ino;
 508
 509                                 /*
 510                                  * XXX: implement correct swabbing here.
 511                                  */
 512
 513                                 hash = le64_to_cpu(ent->lde_hash);
 514                                 if (hash < pos)
 515                                         /*
 516                                          * Skip until we find target hash
 517                                          * value.
 518                                          */
 519                                         continue;
 520
 521                                 namelen = le16_to_cpu(ent->lde_namelen);
 522                                 if (namelen == 0)
 523                                         /*
 524                                          * Skip dummy record.
 525                                          */
 526                                         continue;
 527
 528                                 if (api32 && hash64)
 529                                         lhash = hash >> 32;
 530                                 else
 531                                         lhash = hash;
 532                                 fid_le_to_cpu(&fid, &ent->lde_fid);
 533                                 ino = cl_fid_build_ino(&fid, api32);
 534                                 type = ll_dirent_type_get(ent);
 535                                 ctx->pos = lhash;
 536                                 /* For 'll_nfs_get_name_filldir()', it will try
 537                                  * to access the 'ent' through its 'lde_name',
 538                                  * so the parameter 'name' for 'ctx->actor()'
 539                                  * must be part of the 'ent'.
 540                                  */
 541                                 done = !dir_emit(ctx, ent->lde_name,
 542                                                  namelen, ino, type);
 543                         }
 544                         next = le64_to_cpu(dp->ldp_hash_end);
 545                         if (!done) {
 546                                 pos = next;
 547                                 if (pos == MDS_DIR_END_OFF) {
 548                                         /*
 549                                          * End of directory reached.
 550                                          */
 551                                         done = 1;
 552                                         ll_release_page(page, 0);
 553                                 } else if (1 /* chain is exhausted*/) {
 554                                         /*
 555                                          * Normal case: continue to the next
 556                                          * page.
 557                                          */
 558                                         ll_release_page(page,
 559                                             le32_to_cpu(dp->ldp_flags) &
 560                                                         LDF_COLLIDE);
 561                                         next = pos;
 562                                         page = ll_get_dir_page(inode, pos,
 563                                                                &chain);
 564                                 } else {
 565                                         /*
 566                                          * go into overflow page.
 567                                          */
 568                                         LASSERT(le32_to_cpu(dp->ldp_flags) &
 569                                                 LDF_COLLIDE);
 570                                         ll_release_page(page, 1);
 571                                 }
 572                         } else {
 573                                 pos = hash;
 574                                 ll_release_page(page, 0);
 575                         }
 576                 } else {
 577                         rc = PTR_ERR(page);
 578                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 579                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 580                 }
 581         }
 582
 583         ctx->pos = pos;
 584         ll_dir_chain_fini(&chain);
 585         return rc;
 586 }
 587
 588 static int ll_readdir(struct file *filp, struct dir_context *ctx)
 589 {
 590         struct inode            *inode  = file_inode(filp);
 591         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 592         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 593         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 594         int                     api32   = ll_need_32bit_api(sbi);
 595         int                     rc;
 596
 597         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu 32bit_api %d\n",
 598                inode->i_ino, inode->i_generation,
 599                inode, (unsigned long)lfd->lfd_pos, i_size_read(inode), api32);
 600
 601         if (lfd->lfd_pos == MDS_DIR_END_OFF) {
 602                 /*
 603                  * end-of-file.
 604                  */
 605                 rc = 0;
 606                 goto out;
 607         }
 608
 609         ctx->pos = lfd->lfd_pos;
 610         rc = ll_dir_read(inode, ctx);
 611         lfd->lfd_pos = ctx->pos;
 612         if (ctx->pos == MDS_DIR_END_OFF) {
 613                 if (api32)
 614                         ctx->pos = LL_DIR_END_OFF_32BIT;
 615                 else
 616                         ctx->pos = LL_DIR_END_OFF;
 617         } else {
 618                 if (api32 && hash64)
 619                         ctx->pos >>= 32;
 620         }
 621         filp->f_version = inode->i_version;
 622
 623 out:
 624         if (!rc)
 625                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 626
 627         return rc;
 628 }
 629
 630 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 631 {
 632         struct mgs_send_param *msp;
 633         int rc = 0;
 634
 635         msp = kzalloc(sizeof(*msp), GFP_NOFS);
 636         if (!msp)
 637                 return -ENOMEM;
 638
 639         strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
 640         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 641                                 sizeof(struct mgs_send_param), msp, NULL);
 642         if (rc)
 643                 CERROR("Failed to set parameter: %d\n", rc);
 644         kfree(msp);
 645
 646         return rc;
 647 }
 648
 649 static int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
 650                                char *filename)
 651 {
 652         struct ptlrpc_request *request = NULL;
 653         struct md_op_data *op_data;
 654         struct ll_sb_info *sbi = ll_i2sbi(dir);
 655         int mode;
 656         int err;
 657
 658         mode = (~current_umask() & 0755) | S_IFDIR;
 659         op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
 660                                      strlen(filename), mode, LUSTRE_OPC_MKDIR,
 661                                      lump);
 662         if (IS_ERR(op_data)) {
 663                 err = PTR_ERR(op_data);
 664                 goto err_exit;
 665         }
 666
 667         op_data->op_cli_flags |= CLI_SET_MEA;
 668         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 669                         from_kuid(&init_user_ns, current_fsuid()),
 670                         from_kgid(&init_user_ns, current_fsgid()),
 671                         cfs_curproc_cap_pack(), 0, &request);
 672         ll_finish_md_op_data(op_data);
 673         if (err)
 674                 goto err_exit;
 675 err_exit:
 676         ptlrpc_req_finished(request);
 677         return err;
 678 }
 679
 680 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 681                      int set_default)
 682 {
 683         struct ll_sb_info *sbi = ll_i2sbi(inode);
 684         struct md_op_data *op_data;
 685         struct ptlrpc_request *req = NULL;
 686         int rc = 0;
 687         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 688         struct obd_device *mgc = lsi->lsi_mgc;
 689         int lum_size;
 690
 691         if (lump) {
 692                 /*
 693                  * This is coming from userspace, so should be in
 694                  * local endian.  But the MDS would like it in little
 695                  * endian, so we swab it before we send it.
 696                  */
 697                 switch (lump->lmm_magic) {
 698                 case LOV_USER_MAGIC_V1: {
 699                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 700                                 lustre_swab_lov_user_md_v1(lump);
 701                         lum_size = sizeof(struct lov_user_md_v1);
 702                         break;
 703                 }
 704                 case LOV_USER_MAGIC_V3: {
 705                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 706                                 lustre_swab_lov_user_md_v3(
 707                                         (struct lov_user_md_v3 *)lump);
 708                         lum_size = sizeof(struct lov_user_md_v3);
 709                         break;
 710                 }
 711                 default: {
 712                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
 713                                lump->lmm_magic, LOV_USER_MAGIC_V1,
 714                                LOV_USER_MAGIC_V3);
 715                         return -EINVAL;
 716                 }
 717                 }
 718         } else {
 719                 lum_size = sizeof(struct lov_user_md_v1);
 720         }
 721
 722         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 723                                      LUSTRE_OPC_ANY, NULL);
 724         if (IS_ERR(op_data))
 725                 return PTR_ERR(op_data);
 726
 727         if (lump && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC))
 728                 op_data->op_cli_flags |= CLI_SET_MEA;
 729
 730         /* swabbing is done in lov_setstripe() on server side */
 731         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 732                         NULL, 0, &req, NULL);
 733         ll_finish_md_op_data(op_data);
 734         ptlrpc_req_finished(req);
 735         if (rc) {
 736                 if (rc != -EPERM && rc != -EACCES)
 737                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 738         }
 739
 740         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 741          * LOV_USER_MAGIC_V3 have the same initial fields so we do not
 742          * need to make the distinction between the 2 versions
 743          */
 744         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 745                 char *param = NULL;
 746                 char *buf;
 747
 748                 param = kzalloc(MGS_PARAM_MAXLEN, GFP_NOFS);
 749                 if (!param)
 750                         return -ENOMEM;
 751
 752                 buf = param;
 753                 /* Get fsname and assume devname to be -MDT0000. */
 754                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 755                 strcat(buf, "-MDT0000.lov");
 756                 buf += strlen(buf);
 757
 758                 /* Set root stripesize */
 759                 sprintf(buf, ".stripesize=%u",
 760                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 761                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 762                 if (rc)
 763                         goto end;
 764
 765                 /* Set root stripecount */
 766                 sprintf(buf, ".stripecount=%hd",
 767                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 768                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 769                 if (rc)
 770                         goto end;
 771
 772                 /* Set root stripeoffset */
 773                 sprintf(buf, ".stripeoffset=%hd",
 774                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 775                         (typeof(lump->lmm_stripe_offset))(-1));
 776                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 777
 778 end:
 779                 kfree(param);
 780         }
 781         return rc;
 782 }
 783
 784 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 785                      int *lmm_size, struct ptlrpc_request **request)
 786 {
 787         struct ll_sb_info *sbi = ll_i2sbi(inode);
 788         struct mdt_body   *body;
 789         struct lov_mds_md *lmm = NULL;
 790         struct ptlrpc_request *req = NULL;
 791         int rc, lmmsize;
 792         struct md_op_data *op_data;
 793
 794         rc = ll_get_default_mdsize(sbi, &lmmsize);
 795         if (rc)
 796                 return rc;
 797
 798         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 799                                      0, lmmsize, LUSTRE_OPC_ANY,
 800                                      NULL);
 801         if (IS_ERR(op_data))
 802                 return PTR_ERR(op_data);
 803
 804         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 805         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 806         ll_finish_md_op_data(op_data);
 807         if (rc < 0) {
 808                 CDEBUG(D_INFO, "md_getattr failed on inode %lu/%u: rc %d\n",
 809                        inode->i_ino,
 810                        inode->i_generation, rc);
 811                 goto out;
 812         }
 813
 814         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 815
 816         lmmsize = body->eadatasize;
 817
 818         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 819             lmmsize == 0) {
 820                 rc = -ENODATA;
 821                 goto out;
 822         }
 823
 824         lmm = req_capsule_server_sized_get(&req->rq_pill,
 825                                            &RMF_MDT_MD, lmmsize);
 826
 827         /*
 828          * This is coming from the MDS, so is probably in
 829          * little endian.  We convert it to host endian before
 830          * passing it to userspace.
 831          */
 832         /* We don't swab objects for directories */
 833         switch (le32_to_cpu(lmm->lmm_magic)) {
 834         case LOV_MAGIC_V1:
 835                 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC)
 836                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 837                 break;
 838         case LOV_MAGIC_V3:
 839                 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC)
 840                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 841                 break;
 842         default:
 843                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 844                 rc = -EPROTO;
 845         }
 846 out:
 847         *lmmp = lmm;
 848         *lmm_size = lmmsize;
 849         *request = req;
 850         return rc;
 851 }
 852
 853 /*
 854  *  Get MDT index for the inode.
 855  */
 856 int ll_get_mdt_idx(struct inode *inode)
 857 {
 858         struct ll_sb_info *sbi = ll_i2sbi(inode);
 859         struct md_op_data *op_data;
 860         int rc, mdtidx;
 861
 862         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 863                                      0, LUSTRE_OPC_ANY, NULL);
 864         if (IS_ERR(op_data))
 865                 return PTR_ERR(op_data);
 866
 867         op_data->op_flags |= MF_GET_MDT_IDX;
 868         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 869         mdtidx = op_data->op_mds;
 870         ll_finish_md_op_data(op_data);
 871         if (rc < 0) {
 872                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 873                 return rc;
 874         }
 875         return mdtidx;
 876 }
 877
 878 /**
 879  * Generic handler to do any pre-copy work.
 880  *
 881  * It sends a first hsm_progress (with extent length == 0) to coordinator as a
 882  * first information for it that real work has started.
 883  *
 884  * Moreover, for a ARCHIVE request, it will sample the file data version and
 885  * store it in \a copy.
 886  *
 887  * \return 0 on success.
 888  */
 889 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 890 {
 891         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 892         struct hsm_progress_kernel       hpk;
 893         int                              rc;
 894
 895         /* Forge a hsm_progress based on data from copy. */
 896         hpk.hpk_fid = copy->hc_hai.hai_fid;
 897         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 898         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 899         hpk.hpk_extent.length = 0;
 900         hpk.hpk_flags = 0;
 901         hpk.hpk_errval = 0;
 902         hpk.hpk_data_version = 0;
 903
 904         /* For archive request, we need to read the current file version. */
 905         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 906                 struct inode    *inode;
 907                 __u64            data_version = 0;
 908
 909                 /* Get inode for this fid */
 910                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 911                 if (IS_ERR(inode)) {
 912                         hpk.hpk_flags |= HP_FLAG_RETRY;
 913                         /* hpk_errval is >= 0 */
 914                         hpk.hpk_errval = -PTR_ERR(inode);
 915                         rc = PTR_ERR(inode);
 916                         goto progress;
 917                 }
 918
 919                 /* Read current file data version */
 920                 rc = ll_data_version(inode, &data_version, 1);
 921                 iput(inode);
 922                 if (rc != 0) {
 923                         CDEBUG(D_HSM, "Could not read file data version of "
 924                                       DFID" (rc = %d). Archive request (%#llx) could not be done.\n",
 925                                       PFID(&copy->hc_hai.hai_fid), rc,
 926                                       copy->hc_hai.hai_cookie);
 927                         hpk.hpk_flags |= HP_FLAG_RETRY;
 928                         /* hpk_errval must be >= 0 */
 929                         hpk.hpk_errval = -rc;
 930                         goto progress;
 931                 }
 932
 933                 /* Store in the hsm_copy for later copytool use.
 934                  * Always modified even if no lsm.
 935                  */
 936                 copy->hc_data_version = data_version;
 937         }
 938
 939 progress:
 940         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 941                            &hpk, NULL);
 942
 943         return rc;
 944 }
 945
 946 /**
 947  * Generic handler to do any post-copy work.
 948  *
 949  * It will send the last hsm_progress update to coordinator to inform it
 950  * that copy is finished and whether it was successful or not.
 951  *
 952  * Moreover,
 953  * - for ARCHIVE request, it will sample the file data version and compare it
 954  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 955  *   will be considered as failed.
 956  * - for RESTORE request, it will sample the file data version and send it to
 957  *   coordinator which is useful if the file was imported as 'released'.
 958  *
 959  * \return 0 on success.
 960  */
 961 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 962 {
 963         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 964         struct hsm_progress_kernel       hpk;
 965         int                              rc;
 966
 967         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 968         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 969          * initialized if copy_end was called with copy == NULL.
 970          */
 971
 972         /* Forge a hsm_progress based on data from copy. */
 973         hpk.hpk_fid = copy->hc_hai.hai_fid;
 974         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 975         hpk.hpk_extent = copy->hc_hai.hai_extent;
 976         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 977         hpk.hpk_errval = copy->hc_errval;
 978         hpk.hpk_data_version = 0;
 979
 980         /* For archive request, we need to check the file data was not changed.
 981          *
 982          * For restore request, we need to send the file data version, this is
 983          * useful when the file was created using hsm_import.
 984          */
 985         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 986              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 987             (copy->hc_errval == 0)) {
 988                 struct inode    *inode;
 989                 __u64            data_version = 0;
 990
 991                 /* Get lsm for this fid */
 992                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 993                 if (IS_ERR(inode)) {
 994                         hpk.hpk_flags |= HP_FLAG_RETRY;
 995                         /* hpk_errval must be >= 0 */
 996                         hpk.hpk_errval = -PTR_ERR(inode);
 997                         rc = PTR_ERR(inode);
 998                         goto progress;
 999                 }
1000
1001                 rc = ll_data_version(inode, &data_version,
1002                                      copy->hc_hai.hai_action == HSMA_ARCHIVE);
1003                 iput(inode);
1004                 if (rc) {
1005                         CDEBUG(D_HSM, "Could not read file data version. Request could not be confirmed.\n");
1006                         if (hpk.hpk_errval == 0)
1007                                 hpk.hpk_errval = -rc;
1008                         goto progress;
1009                 }
1010
1011                 /* Store in the hsm_copy for later copytool use.
1012                  * Always modified even if no lsm.
1013                  */
1014                 hpk.hpk_data_version = data_version;
1015
1016                 /* File could have been stripped during archiving, so we need
1017                  * to check anyway.
1018                  */
1019                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1020                     (copy->hc_data_version != data_version)) {
1021                         CDEBUG(D_HSM, "File data version mismatched. File content was changed during archiving. "
1022                                DFID", start:%#llx current:%#llx\n",
1023                                PFID(&copy->hc_hai.hai_fid),
1024                                copy->hc_data_version, data_version);
1025                         /* File was changed, send error to cdt. Do not ask for
1026                          * retry because if a file is modified frequently,
1027                          * the cdt will loop on retried archive requests.
1028                          * The policy engine will ask for a new archive later
1029                          * when the file will not be modified for some tunable
1030                          * time
1031                          */
1032                         /* we do not notify caller */
1033                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
1034                         /* hpk_errval must be >= 0 */
1035                         hpk.hpk_errval = EBUSY;
1036                 }
1037
1038         }
1039
1040 progress:
1041         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1042                            &hpk, NULL);
1043
1044         return rc;
1045 }
1046
1047 static int copy_and_ioctl(int cmd, struct obd_export *exp,
1048                           const void __user *data, size_t size)
1049 {
1050         void *copy;
1051         int rc;
1052
1053         copy = kzalloc(size, GFP_NOFS);
1054         if (!copy)
1055                 return -ENOMEM;
1056
1057         if (copy_from_user(copy, data, size)) {
1058                 rc = -EFAULT;
1059                 goto out;
1060         }
1061
1062         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
1063 out:
1064         kfree(copy);
1065
1066         return rc;
1067 }
1068
1069 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1070 {
1071         int cmd = qctl->qc_cmd;
1072         int type = qctl->qc_type;
1073         int id = qctl->qc_id;
1074         int valid = qctl->qc_valid;
1075         int rc = 0;
1076
1077         switch (cmd) {
1078         case LUSTRE_Q_INVALIDATE:
1079         case LUSTRE_Q_FINVALIDATE:
1080         case Q_QUOTAON:
1081         case Q_QUOTAOFF:
1082         case Q_SETQUOTA:
1083         case Q_SETINFO:
1084                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1085                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1086                         return -EPERM;
1087                 break;
1088         case Q_GETQUOTA:
1089                 if (((type == USRQUOTA &&
1090                       !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
1091                      (type == GRPQUOTA &&
1092                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
1093                     (!capable(CFS_CAP_SYS_ADMIN) ||
1094                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
1095                         return -EPERM;
1096                 break;
1097         case Q_GETINFO:
1098                 break;
1099         default:
1100                 CERROR("unsupported quotactl op: %#x\n", cmd);
1101                 return -ENOTTY;
1102         }
1103
1104         if (valid != QC_GENERAL) {
1105                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1106                         return -EOPNOTSUPP;
1107
1108                 if (cmd == Q_GETINFO)
1109                         qctl->qc_cmd = Q_GETOINFO;
1110                 else if (cmd == Q_GETQUOTA)
1111                         qctl->qc_cmd = Q_GETOQUOTA;
1112                 else
1113                         return -EINVAL;
1114
1115                 switch (valid) {
1116                 case QC_MDTIDX:
1117                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1118                                            sizeof(*qctl), qctl, NULL);
1119                         break;
1120                 case QC_OSTIDX:
1121                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1122                                            sizeof(*qctl), qctl, NULL);
1123                         break;
1124                 case QC_UUID:
1125                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1126                                            sizeof(*qctl), qctl, NULL);
1127                         if (rc == -EAGAIN)
1128                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1129                                                    sbi->ll_dt_exp,
1130                                                    sizeof(*qctl), qctl, NULL);
1131                         break;
1132                 default:
1133                         rc = -EINVAL;
1134                         break;
1135                 }
1136
1137                 if (rc)
1138                         return rc;
1139
1140                 qctl->qc_cmd = cmd;
1141         } else {
1142                 struct obd_quotactl *oqctl;
1143
1144                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1145                 if (!oqctl)
1146                         return -ENOMEM;
1147
1148                 QCTL_COPY(oqctl, qctl);
1149                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1150                 if (rc) {
1151                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
1152                                 oqctl->qc_cmd = Q_QUOTAOFF;
1153                                 obd_quotactl(sbi->ll_md_exp, oqctl);
1154                         }
1155                         kfree(oqctl);
1156                         return rc;
1157                 }
1158                 /* If QIF_SPACE is not set, client should collect the
1159                  * space usage from OSSs by itself
1160                  */
1161                 if (cmd == Q_GETQUOTA &&
1162                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1163                     !oqctl->qc_dqblk.dqb_curspace) {
1164                         struct obd_quotactl *oqctl_tmp;
1165
1166                         oqctl_tmp = kzalloc(sizeof(*oqctl_tmp), GFP_NOFS);
1167                         if (!oqctl_tmp) {
1168                                 rc = -ENOMEM;
1169                                 goto out;
1170                         }
1171
1172                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1173                         oqctl_tmp->qc_id = oqctl->qc_id;
1174                         oqctl_tmp->qc_type = oqctl->qc_type;
1175
1176                         /* collect space usage from OSTs */
1177                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1178                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1179                         if (!rc || rc == -EREMOTEIO) {
1180                                 oqctl->qc_dqblk.dqb_curspace =
1181                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1182                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1183                         }
1184
1185                         /* collect space & inode usage from MDTs */
1186                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1187                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1188                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1189                         if (!rc || rc == -EREMOTEIO) {
1190                                 oqctl->qc_dqblk.dqb_curspace +=
1191                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1192                                 oqctl->qc_dqblk.dqb_curinodes =
1193                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1194                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1195                         } else {
1196                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1197                         }
1198
1199                         kfree(oqctl_tmp);
1200                 }
1201 out:
1202                 QCTL_COPY(qctl, oqctl);
1203                 kfree(oqctl);
1204         }
1205
1206         return rc;
1207 }
1208
1209 /* This function tries to get a single name component,
1210  * to send to the server. No actual path traversal involved,
1211  * so we limit to NAME_MAX
1212  */
1213 static char *ll_getname(const char __user *filename)
1214 {
1215         int ret = 0, len;
1216         char *tmp;
1217
1218         tmp = kzalloc(NAME_MAX + 1, GFP_KERNEL);
1219         if (!tmp)
1220                 return ERR_PTR(-ENOMEM);
1221
1222         len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1223         if (len < 0)
1224                 ret = len;
1225         else if (len == 0)
1226                 ret = -ENOENT;
1227         else if (len > NAME_MAX && tmp[NAME_MAX] != 0)
1228                 ret = -ENAMETOOLONG;
1229
1230         if (ret) {
1231                 kfree(tmp);
1232                 tmp =  ERR_PTR(ret);
1233         }
1234         return tmp;
1235 }
1236
1237 #define ll_putname(filename) kfree(filename)
1238
1239 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1240 {
1241         struct inode *inode = file_inode(file);
1242         struct ll_sb_info *sbi = ll_i2sbi(inode);
1243         struct obd_ioctl_data *data;
1244         int rc = 0;
1245
1246         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1247                inode->i_ino, inode->i_generation, inode, cmd);
1248
1249         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1250         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1251                 return -ENOTTY;
1252
1253         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1254         switch (cmd) {
1255         case FSFILT_IOC_GETFLAGS:
1256         case FSFILT_IOC_SETFLAGS:
1257                 return ll_iocontrol(inode, file, cmd, arg);
1258         case FSFILT_IOC_GETVERSION_OLD:
1259         case FSFILT_IOC_GETVERSION:
1260                 return put_user(inode->i_generation, (int __user *)arg);
1261         /* We need to special case any other ioctls we want to handle,
1262          * to send them to the MDS/OST as appropriate and to properly
1263          * network encode the arg field.
1264         case FSFILT_IOC_SETVERSION_OLD:
1265         case FSFILT_IOC_SETVERSION:
1266         */
1267         case LL_IOC_GET_MDTIDX: {
1268                 int mdtidx;
1269
1270                 mdtidx = ll_get_mdt_idx(inode);
1271                 if (mdtidx < 0)
1272                         return mdtidx;
1273
1274                 if (put_user((int)mdtidx, (int __user *)arg))
1275                         return -EFAULT;
1276
1277                 return 0;
1278         }
1279         case IOC_MDC_LOOKUP: {
1280                 struct ptlrpc_request *request = NULL;
1281                 int namelen, len = 0;
1282                 char *buf = NULL;
1283                 char *filename;
1284                 struct md_op_data *op_data;
1285
1286                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1287                 if (rc)
1288                         return rc;
1289                 data = (void *)buf;
1290
1291                 filename = data->ioc_inlbuf1;
1292                 namelen = strlen(filename);
1293
1294                 if (namelen < 1) {
1295                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1296                         rc = -EINVAL;
1297                         goto out_free;
1298                 }
1299
1300                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1301                                              0, LUSTRE_OPC_ANY, NULL);
1302                 if (IS_ERR(op_data)) {
1303                         rc = PTR_ERR(op_data);
1304                         goto out_free;
1305                 }
1306
1307                 op_data->op_valid = OBD_MD_FLID;
1308                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1309                 ll_finish_md_op_data(op_data);
1310                 if (rc < 0) {
1311                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1312                         goto out_free;
1313                 }
1314                 ptlrpc_req_finished(request);
1315 out_free:
1316                 obd_ioctl_freedata(buf, len);
1317                 return rc;
1318         }
1319         case LL_IOC_LMV_SETSTRIPE: {
1320                 struct lmv_user_md  *lum;
1321                 char            *buf = NULL;
1322                 char            *filename;
1323                 int              namelen = 0;
1324                 int              lumlen = 0;
1325                 int              len;
1326                 int              rc;
1327
1328                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1329                 if (rc)
1330                         return rc;
1331
1332                 data = (void *)buf;
1333                 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
1334                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0) {
1335                         rc = -EINVAL;
1336                         goto lmv_out_free;
1337                 }
1338
1339                 filename = data->ioc_inlbuf1;
1340                 namelen = data->ioc_inllen1;
1341
1342                 if (namelen < 1) {
1343                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1344                         rc = -EINVAL;
1345                         goto lmv_out_free;
1346                 }
1347                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1348                 lumlen = data->ioc_inllen2;
1349
1350                 if (lum->lum_magic != LMV_USER_MAGIC ||
1351                     lumlen != sizeof(*lum)) {
1352                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1353                                filename, lum->lum_magic, lumlen, -EFAULT);
1354                         rc = -EINVAL;
1355                         goto lmv_out_free;
1356                 }
1357
1358                 /**
1359                  * ll_dir_setdirstripe will be used to set dir stripe
1360                  *  mdc_create--->mdt_reint_create (with dirstripe)
1361                  */
1362                 rc = ll_dir_setdirstripe(inode, lum, filename);
1363 lmv_out_free:
1364                 obd_ioctl_freedata(buf, len);
1365                 return rc;
1366
1367         }
1368         case LL_IOC_LOV_SETSTRIPE: {
1369                 struct lov_user_md_v3 lumv3;
1370                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1371                 struct lov_user_md_v1 __user *lumv1p = (void __user *)arg;
1372                 struct lov_user_md_v3 __user *lumv3p = (void __user *)arg;
1373
1374                 int set_default = 0;
1375
1376                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1377                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1378                         sizeof(lumv3p->lmm_objects[0]));
1379                 /* first try with v1 which is smaller than v3 */
1380                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1381                         return -EFAULT;
1382
1383                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1384                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1385                                 return -EFAULT;
1386                 }
1387
1388                 if (is_root_inode(inode))
1389                         set_default = 1;
1390
1391                 /* in v1 and v3 cases lumv1 points to data */
1392                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1393
1394                 return rc;
1395         }
1396         case LL_IOC_LMV_GETSTRIPE: {
1397                 struct lmv_user_md __user *lump = (void __user *)arg;
1398                 struct lmv_user_md lum;
1399                 struct lmv_user_md *tmp;
1400                 int lum_size;
1401                 int rc = 0;
1402                 int mdtindex;
1403
1404                 if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
1405                         return -EFAULT;
1406
1407                 if (lum.lum_magic != LMV_MAGIC_V1)
1408                         return -EINVAL;
1409
1410                 lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
1411                 tmp = kzalloc(lum_size, GFP_NOFS);
1412                 if (!tmp) {
1413                         rc = -ENOMEM;
1414                         goto free_lmv;
1415                 }
1416
1417                 *tmp = lum;
1418                 tmp->lum_type = LMV_STRIPE_TYPE;
1419                 tmp->lum_stripe_count = 1;
1420                 mdtindex = ll_get_mdt_idx(inode);
1421                 if (mdtindex < 0) {
1422                         rc = -ENOMEM;
1423                         goto free_lmv;
1424                 }
1425
1426                 tmp->lum_stripe_offset = mdtindex;
1427                 tmp->lum_objects[0].lum_mds = mdtindex;
1428                 memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
1429                        sizeof(struct lu_fid));
1430                 if (copy_to_user((void __user *)arg, tmp, lum_size)) {
1431                         rc = -EFAULT;
1432                         goto free_lmv;
1433                 }
1434 free_lmv:
1435                 kfree(tmp);
1436                 return rc;
1437         }
1438         case LL_IOC_LOV_SWAP_LAYOUTS:
1439                 return -EPERM;
1440         case LL_IOC_OBD_STATFS:
1441                 return ll_obd_statfs(inode, (void __user *)arg);
1442         case LL_IOC_LOV_GETSTRIPE:
1443         case LL_IOC_MDC_GETINFO:
1444         case IOC_MDC_GETFILEINFO:
1445         case IOC_MDC_GETFILESTRIPE: {
1446                 struct ptlrpc_request *request = NULL;
1447                 struct lov_user_md __user *lump;
1448                 struct lov_mds_md *lmm = NULL;
1449                 struct mdt_body *body;
1450                 char *filename = NULL;
1451                 int lmmsize;
1452
1453                 if (cmd == IOC_MDC_GETFILEINFO ||
1454                     cmd == IOC_MDC_GETFILESTRIPE) {
1455                         filename = ll_getname((const char __user *)arg);
1456                         if (IS_ERR(filename))
1457                                 return PTR_ERR(filename);
1458
1459                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1460                                                       &lmmsize, &request);
1461                 } else {
1462                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1463                 }
1464
1465                 if (request) {
1466                         body = req_capsule_server_get(&request->rq_pill,
1467                                                       &RMF_MDT_BODY);
1468                         LASSERT(body);
1469                 } else {
1470                         goto out_req;
1471                 }
1472
1473                 if (rc < 0) {
1474                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1475                                                cmd == LL_IOC_MDC_GETINFO)) {
1476                                 rc = 0;
1477                                 goto skip_lmm;
1478                         } else
1479                                 goto out_req;
1480                 }
1481
1482                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1483                     cmd == LL_IOC_LOV_GETSTRIPE) {
1484                         lump = (struct lov_user_md __user *)arg;
1485                 } else {
1486                         struct lov_user_mds_data __user *lmdp;
1487
1488                         lmdp = (struct lov_user_mds_data __user *)arg;
1489                         lump = &lmdp->lmd_lmm;
1490                 }
1491                 if (copy_to_user(lump, lmm, lmmsize)) {
1492                         if (copy_to_user(lump, lmm, sizeof(*lump))) {
1493                                 rc = -EFAULT;
1494                                 goto out_req;
1495                         }
1496                         rc = -EOVERFLOW;
1497                 }
1498 skip_lmm:
1499                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1500                         struct lov_user_mds_data __user *lmdp;
1501                         lstat_t st = { 0 };
1502
1503                         st.st_dev     = inode->i_sb->s_dev;
1504                         st.st_mode    = body->mode;
1505                         st.st_nlink   = body->nlink;
1506                         st.st_uid     = body->uid;
1507                         st.st_gid     = body->gid;
1508                         st.st_rdev    = body->rdev;
1509                         st.st_size    = body->size;
1510                         st.st_blksize = PAGE_CACHE_SIZE;
1511                         st.st_blocks  = body->blocks;
1512                         st.st_atime   = body->atime;
1513                         st.st_mtime   = body->mtime;
1514                         st.st_ctime   = body->ctime;
1515                         st.st_ino     = inode->i_ino;
1516
1517                         lmdp = (struct lov_user_mds_data __user *)arg;
1518                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st))) {
1519                                 rc = -EFAULT;
1520                                 goto out_req;
1521                         }
1522                 }
1523
1524 out_req:
1525                 ptlrpc_req_finished(request);
1526                 if (filename)
1527                         ll_putname(filename);
1528                 return rc;
1529         }
1530         case IOC_LOV_GETINFO: {
1531                 struct lov_user_mds_data __user *lumd;
1532                 struct lov_stripe_md *lsm;
1533                 struct lov_user_md __user *lum;
1534                 struct lov_mds_md *lmm;
1535                 int lmmsize;
1536                 lstat_t st;
1537
1538                 lumd = (struct lov_user_mds_data __user *)arg;
1539                 lum = &lumd->lmd_lmm;
1540
1541                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1542                 if (rc)
1543                         return rc;
1544
1545                 lmm = libcfs_kvzalloc(lmmsize, GFP_NOFS);
1546                 if (!lmm)
1547                         return -ENOMEM;
1548                 if (copy_from_user(lmm, lum, lmmsize)) {
1549                         rc = -EFAULT;
1550                         goto free_lmm;
1551                 }
1552
1553                 switch (lmm->lmm_magic) {
1554                 case LOV_USER_MAGIC_V1:
1555                         if (cpu_to_le32(LOV_USER_MAGIC_V1) == LOV_USER_MAGIC_V1)
1556                                 break;
1557                         /* swab objects first so that stripes num will be sane */
1558                         lustre_swab_lov_user_md_objects(
1559                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1560                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1561                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1562                         break;
1563                 case LOV_USER_MAGIC_V3:
1564                         if (cpu_to_le32(LOV_USER_MAGIC_V3) == LOV_USER_MAGIC_V3)
1565                                 break;
1566                         /* swab objects first so that stripes num will be sane */
1567                         lustre_swab_lov_user_md_objects(
1568                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1569                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1570                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1571                         break;
1572                 default:
1573                         rc = -EINVAL;
1574                         goto free_lmm;
1575                 }
1576
1577                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1578                 if (rc < 0) {
1579                         rc = -ENOMEM;
1580                         goto free_lmm;
1581                 }
1582
1583                 /* Perform glimpse_size operation. */
1584                 memset(&st, 0, sizeof(st));
1585
1586                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1587                 if (rc)
1588                         goto free_lsm;
1589
1590                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st))) {
1591                         rc = -EFAULT;
1592                         goto free_lsm;
1593                 }
1594
1595 free_lsm:
1596                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1597 free_lmm:
1598                 kvfree(lmm);
1599                 return rc;
1600         }
1601         case OBD_IOC_LLOG_CATINFO: {
1602                 return -EOPNOTSUPP;
1603         }
1604         case OBD_IOC_QUOTACHECK: {
1605                 struct obd_quotactl *oqctl;
1606                 int error = 0;
1607
1608                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1609                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1610                         return -EPERM;
1611
1612                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1613                 if (!oqctl)
1614                         return -ENOMEM;
1615                 oqctl->qc_type = arg;
1616                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1617                 if (rc < 0) {
1618                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1619                         error = rc;
1620                 }
1621
1622                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1623                 if (rc < 0)
1624                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1625
1626                 kfree(oqctl);
1627                 return error ?: rc;
1628         }
1629         case OBD_IOC_POLL_QUOTACHECK: {
1630                 struct if_quotacheck *check;
1631
1632                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1633                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1634                         return -EPERM;
1635
1636                 check = kzalloc(sizeof(*check), GFP_NOFS);
1637                 if (!check)
1638                         return -ENOMEM;
1639
1640                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1641                                    NULL);
1642                 if (rc) {
1643                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1644                         if (copy_to_user((void __user *)arg, check,
1645                                          sizeof(*check)))
1646                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1647                         goto out_poll;
1648                 }
1649
1650                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1651                                    NULL);
1652                 if (rc) {
1653                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1654                         if (copy_to_user((void __user *)arg, check,
1655                                          sizeof(*check)))
1656                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1657                         goto out_poll;
1658                 }
1659 out_poll:
1660                 kfree(check);
1661                 return rc;
1662         }
1663         case LL_IOC_QUOTACTL: {
1664                 struct if_quotactl *qctl;
1665
1666                 qctl = kzalloc(sizeof(*qctl), GFP_NOFS);
1667                 if (!qctl)
1668                         return -ENOMEM;
1669
1670                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl))) {
1671                         rc = -EFAULT;
1672                         goto out_quotactl;
1673                 }
1674
1675                 rc = quotactl_ioctl(sbi, qctl);
1676
1677                 if (rc == 0 && copy_to_user((void __user *)arg, qctl,
1678                                             sizeof(*qctl)))
1679                         rc = -EFAULT;
1680
1681 out_quotactl:
1682                 kfree(qctl);
1683                 return rc;
1684         }
1685         case OBD_IOC_GETDTNAME:
1686         case OBD_IOC_GETMDNAME:
1687                 return ll_get_obd_name(inode, cmd, arg);
1688         case LL_IOC_FLUSHCTX:
1689                 return ll_flush_ctx(inode);
1690 #ifdef CONFIG_FS_POSIX_ACL
1691         case LL_IOC_RMTACL: {
1692             if (sbi->ll_flags & LL_SBI_RMT_CLIENT && is_root_inode(inode)) {
1693                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1694
1695                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1696                 if (!rc)
1697                         fd->fd_flags |= LL_FILE_RMTACL;
1698                 return rc;
1699             } else
1700                 return 0;
1701         }
1702 #endif
1703         case LL_IOC_GETOBDCOUNT: {
1704                 int count, vallen;
1705                 struct obd_export *exp;
1706
1707                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1708                         return -EFAULT;
1709
1710                 /* get ost count when count is zero, get mdt count otherwise */
1711                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1712                 vallen = sizeof(count);
1713                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1714                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1715                 if (rc) {
1716                         CERROR("get target count failed: %d\n", rc);
1717                         return rc;
1718                 }
1719
1720                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1721                         return -EFAULT;
1722
1723                 return 0;
1724         }
1725         case LL_IOC_PATH2FID:
1726                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1727                                  sizeof(struct lu_fid)))
1728                         return -EFAULT;
1729                 return 0;
1730         case LL_IOC_GET_CONNECT_FLAGS: {
1731                 return obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1732                                      (void __user *)arg);
1733         }
1734         case OBD_IOC_CHANGELOG_SEND:
1735         case OBD_IOC_CHANGELOG_CLEAR:
1736                 if (!capable(CFS_CAP_SYS_ADMIN))
1737                         return -EPERM;
1738
1739                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1740                                     sizeof(struct ioc_changelog));
1741                 return rc;
1742         case OBD_IOC_FID2PATH:
1743                 return ll_fid2path(inode, (void __user *)arg);
1744         case LL_IOC_HSM_REQUEST: {
1745                 struct hsm_user_request *hur;
1746                 ssize_t                  totalsize;
1747
1748                 hur = memdup_user((void __user *)arg, sizeof(*hur));
1749                 if (IS_ERR(hur))
1750                         return PTR_ERR(hur);
1751
1752                 /* Compute the whole struct size */
1753                 totalsize = hur_len(hur);
1754                 kfree(hur);
1755                 if (totalsize < 0)
1756                         return -E2BIG;
1757
1758                 /* Final size will be more than double totalsize */
1759                 if (totalsize >= MDS_MAXREQSIZE / 3)
1760                         return -E2BIG;
1761
1762                 hur = libcfs_kvzalloc(totalsize, GFP_NOFS);
1763                 if (!hur)
1764                         return -ENOMEM;
1765
1766                 /* Copy the whole struct */
1767                 if (copy_from_user(hur, (void __user *)arg, totalsize)) {
1768                         kvfree(hur);
1769                         return -EFAULT;
1770                 }
1771
1772                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1773                         const struct lu_fid *fid;
1774                         struct inode *f;
1775                         int i;
1776
1777                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1778                                 fid = &hur->hur_user_item[i].hui_fid;
1779                                 f = search_inode_for_lustre(inode->i_sb, fid);
1780                                 if (IS_ERR(f)) {
1781                                         rc = PTR_ERR(f);
1782                                         break;
1783                                 }
1784
1785                                 rc = ll_hsm_release(f);
1786                                 iput(f);
1787                                 if (rc != 0)
1788                                         break;
1789                         }
1790                 } else {
1791                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1792                                            hur, NULL);
1793                 }
1794
1795                 kvfree(hur);
1796
1797                 return rc;
1798         }
1799         case LL_IOC_HSM_PROGRESS: {
1800                 struct hsm_progress_kernel      hpk;
1801                 struct hsm_progress             hp;
1802
1803                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1804                         return -EFAULT;
1805
1806                 hpk.hpk_fid = hp.hp_fid;
1807                 hpk.hpk_cookie = hp.hp_cookie;
1808                 hpk.hpk_extent = hp.hp_extent;
1809                 hpk.hpk_flags = hp.hp_flags;
1810                 hpk.hpk_errval = hp.hp_errval;
1811                 hpk.hpk_data_version = 0;
1812
1813                 /* File may not exist in Lustre; all progress
1814                  * reported to Lustre root
1815                  */
1816                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1817                                    NULL);
1818                 return rc;
1819         }
1820         case LL_IOC_HSM_CT_START:
1821                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1822                                     sizeof(struct lustre_kernelcomm));
1823                 return rc;
1824
1825         case LL_IOC_HSM_COPY_START: {
1826                 struct hsm_copy *copy;
1827                 int              rc;
1828
1829                 copy = memdup_user((char __user *)arg, sizeof(*copy));
1830                 if (IS_ERR(copy))
1831                         return PTR_ERR(copy);
1832
1833                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1834                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1835                         rc = -EFAULT;
1836
1837                 kfree(copy);
1838                 return rc;
1839         }
1840         case LL_IOC_HSM_COPY_END: {
1841                 struct hsm_copy *copy;
1842                 int              rc;
1843
1844                 copy = memdup_user((char __user *)arg, sizeof(*copy));
1845                 if (IS_ERR(copy))
1846                         return PTR_ERR(copy);
1847
1848                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1849                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1850                         rc = -EFAULT;
1851
1852                 kfree(copy);
1853                 return rc;
1854         }
1855         default:
1856                 return obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1857                                      (void __user *)arg);
1858         }
1859 }
1860
1861 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1862 {
1863         struct inode *inode = file->f_mapping->host;
1864         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1865         struct ll_sb_info *sbi = ll_i2sbi(inode);
1866         int api32 = ll_need_32bit_api(sbi);
1867         loff_t ret = -EINVAL;
1868
1869         inode_lock(inode);
1870         switch (origin) {
1871         case SEEK_SET:
1872                 break;
1873         case SEEK_CUR:
1874                 offset += file->f_pos;
1875                 break;
1876         case SEEK_END:
1877                 if (offset > 0)
1878                         goto out;
1879                 if (api32)
1880                         offset += LL_DIR_END_OFF_32BIT;
1881                 else
1882                         offset += LL_DIR_END_OFF;
1883                 break;
1884         default:
1885                 goto out;
1886         }
1887
1888         if (offset >= 0 &&
1889             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1890              (!api32 && offset <= LL_DIR_END_OFF))) {
1891                 if (offset != file->f_pos) {
1892                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1893                             (!api32 && offset == LL_DIR_END_OFF))
1894                                 fd->lfd_pos = MDS_DIR_END_OFF;
1895                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1896                                 fd->lfd_pos = offset << 32;
1897                         else
1898                                 fd->lfd_pos = offset;
1899                         file->f_pos = offset;
1900                         file->f_version = 0;
1901                 }
1902                 ret = offset;
1903         }
1904         goto out;
1905
1906 out:
1907         inode_unlock(inode);
1908         return ret;
1909 }
1910
1911 static int ll_dir_open(struct inode *inode, struct file *file)
1912 {
1913         return ll_file_open(inode, file);
1914 }
1915
1916 static int ll_dir_release(struct inode *inode, struct file *file)
1917 {
1918         return ll_file_release(inode, file);
1919 }
1920
1921 const struct file_operations ll_dir_operations = {
1922         .llseek   = ll_dir_seek,
1923         .open     = ll_dir_open,
1924         .release  = ll_dir_release,
1925         .read     = generic_read_dir,
1926         .iterate  = ll_readdir,
1927         .unlocked_ioctl   = ll_dir_ioctl,
1928         .fsync    = ll_fsync,
1929 };