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 "llite_internal.h"
  59
  60 /*
  61  * (new) readdir implementation overview.
  62  *
  63  * Original lustre readdir implementation cached exact copy of raw directory
  64  * pages on the client. These pages were indexed in client page cache by
  65  * logical offset in the directory file. This design, while very simple and
  66  * intuitive had some inherent problems:
  67  *
  68  *     . it implies that byte offset to the directory entry serves as a
  69  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  70  *     ext3/htree directory entries may move due to splits, and more
  71  *     importantly,
  72  *
  73  *     . it is incompatible with the design of split directories for cmd3,
  74  *     that assumes that names are distributed across nodes based on their
  75  *     hash, and so readdir should be done in hash order.
  76  *
  77  * New readdir implementation does readdir in hash order, and uses hash of a
  78  * file name as a telldir/seekdir cookie. This led to number of complications:
  79  *
  80  *     . hash is not unique, so it cannot be used to index cached directory
  81  *     pages on the client (note, that it requires a whole pageful of hash
  82  *     collided entries to cause two pages to have identical hashes);
  83  *
  84  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  85  *     entry cookie. ext3/htree has the same problem and lustre implementation
  86  *     mimics their solution: seekdir(hash) positions directory at the first
  87  *     entry with the given hash.
  88  *
  89  * Client side.
  90  *
  91  * 0. caching
  92  *
  93  * Client caches directory pages using hash of the first entry as an index. As
  94  * noted above hash is not unique, so this solution doesn't work as is:
  95  * special processing is needed for "page hash chains" (i.e., sequences of
  96  * pages filled with entries all having the same hash value).
  97  *
  98  * First, such chains have to be detected. To this end, server returns to the
  99  * client the hash of the first entry on the page next to one returned. When
 100  * client detects that this hash is the same as hash of the first entry on the
 101  * returned page, page hash collision has to be handled. Pages in the
 102  * hash chain, except first one, are termed "overflow pages".
 103  *
 104  * Solution to index uniqueness problem is to not cache overflow
 105  * pages. Instead, when page hash collision is detected, all overflow pages
 106  * from emerging chain are immediately requested from the server and placed in
 107  * a special data structure (struct ll_dir_chain). This data structure is used
 108  * by ll_readdir() to process entries from overflow pages. When readdir
 109  * invocation finishes, overflow pages are discarded. If page hash collision
 110  * chain weren't completely processed, next call to readdir will again detect
 111  * page hash collision, again read overflow pages in, process next portion of
 112  * entries and again discard the pages. This is not as wasteful as it looks,
 113  * because, given reasonable hash, page hash collisions are extremely rare.
 114  *
 115  * 1. directory positioning
 116  *
 117  * When seekdir(hash) is called, original
 118  *
 119  *
 120  *
 121  *
 122  *
 123  *
 124  *
 125  *
 126  * Server.
 127  *
 128  * identification of and access to overflow pages
 129  *
 130  * page format
 131  *
 132  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 133  * a header lu_dirpage which describes the start/end hash, and whether this
 134  * page is empty (contains no dir entry) or hash collide with next page.
 135  * After client receives reply, several pages will be integrated into dir page
 136  * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
 137  * lu_dirpage for this integrated page will be adjusted. See
 138  * lmv_adjust_dirpages().
 139  *
 140  */
 141
 142 /* returns the page unlocked, but with a reference */
 143 static int ll_dir_filler(void *_hash, struct page *page0)
 144 {
 145         struct inode *inode = page0->mapping->host;
 146         int hash64 = ll_i2sbi(inode)->ll_flags & LL_SBI_64BIT_HASH;
 147         struct obd_export *exp = ll_i2sbi(inode)->ll_md_exp;
 148         struct ptlrpc_request *request;
 149         struct mdt_body *body;
 150         struct md_op_data *op_data;
 151         __u64 hash = *((__u64 *)_hash);
 152         struct page **page_pool;
 153         struct page *page;
 154         struct lu_dirpage *dp;
 155         int max_pages = ll_i2sbi(inode)->ll_md_brw_size >> PAGE_CACHE_SHIFT;
 156         int nrdpgs = 0; /* number of pages read actually */
 157         int npages;
 158         int i;
 159         int rc;
 160
 161         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) hash %llu\n",
 162                inode->i_ino, inode->i_generation, inode, hash);
 163
 164         LASSERT(max_pages > 0 && max_pages <= MD_MAX_BRW_PAGES);
 165
 166         page_pool = kcalloc(max_pages, sizeof(page), GFP_NOFS);
 167         if (page_pool) {
 168                 page_pool[0] = page0;
 169         } else {
 170                 page_pool = &page0;
 171                 max_pages = 1;
 172         }
 173         for (npages = 1; npages < max_pages; npages++) {
 174                 page = page_cache_alloc_cold(inode->i_mapping);
 175                 if (!page)
 176                         break;
 177                 page_pool[npages] = page;
 178         }
 179
 180         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 181                                      LUSTRE_OPC_ANY, NULL);
 182         op_data->op_npages = npages;
 183         op_data->op_offset = hash;
 184         rc = md_readpage(exp, op_data, page_pool, &request);
 185         ll_finish_md_op_data(op_data);
 186         if (rc < 0) {
 187                 /* page0 is special, which was added into page cache early */
 188                 delete_from_page_cache(page0);
 189         } else if (rc == 0) {
 190                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 191                 /* Checked by mdc_readpage() */
 192                 LASSERT(body != NULL);
 193
 194                 if (body->valid & OBD_MD_FLSIZE)
 195                         cl_isize_write(inode, body->size);
 196
 197                 nrdpgs = (request->rq_bulk->bd_nob_transferred+PAGE_CACHE_SIZE-1)
 198                          >> PAGE_CACHE_SHIFT;
 199                 SetPageUptodate(page0);
 200         }
 201         unlock_page(page0);
 202         ptlrpc_req_finished(request);
 203
 204         CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages);
 205
 206         for (i = 1; i < npages; i++) {
 207                 unsigned long offset;
 208                 int ret;
 209
 210                 page = page_pool[i];
 211
 212                 if (rc < 0 || i >= nrdpgs) {
 213                         page_cache_release(page);
 214                         continue;
 215                 }
 216
 217                 SetPageUptodate(page);
 218
 219                 dp = kmap(page);
 220                 hash = le64_to_cpu(dp->ldp_hash_start);
 221                 kunmap(page);
 222
 223                 offset = hash_x_index(hash, hash64);
 224
 225                 prefetchw(&page->flags);
 226                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
 227                                             GFP_NOFS);
 228                 if (ret == 0) {
 229                         unlock_page(page);
 230                 } else {
 231                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed: %d\n",
 232                                offset, ret);
 233                 }
 234                 page_cache_release(page);
 235         }
 236
 237         if (page_pool != &page0)
 238                 kfree(page_pool);
 239         return rc;
 240 }
 241
 242 void ll_release_page(struct page *page, int remove)
 243 {
 244         kunmap(page);
 245         if (remove) {
 246                 lock_page(page);
 247                 if (likely(page->mapping != NULL))
 248                         truncate_complete_page(page->mapping, page);
 249                 unlock_page(page);
 250         }
 251         page_cache_release(page);
 252 }
 253
 254 /*
 255  * Find, kmap and return page that contains given hash.
 256  */
 257 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
 258                                        __u64 *start, __u64 *end)
 259 {
 260         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 261         struct address_space *mapping = dir->i_mapping;
 262         /*
 263          * Complement of hash is used as an index so that
 264          * radix_tree_gang_lookup() can be used to find a page with starting
 265          * hash _smaller_ than one we are looking for.
 266          */
 267         unsigned long offset = hash_x_index(*hash, hash64);
 268         struct page *page;
 269         int found;
 270
 271         spin_lock_irq(&mapping->tree_lock);
 272         found = radix_tree_gang_lookup(&mapping->page_tree,
 273                                        (void **)&page, offset, 1);
 274         if (found > 0 && !radix_tree_exceptional_entry(page)) {
 275                 struct lu_dirpage *dp;
 276
 277                 page_cache_get(page);
 278                 spin_unlock_irq(&mapping->tree_lock);
 279                 /*
 280                  * In contrast to find_lock_page() we are sure that directory
 281                  * page cannot be truncated (while DLM lock is held) and,
 282                  * hence, can avoid restart.
 283                  *
 284                  * In fact, page cannot be locked here at all, because
 285                  * ll_dir_filler() does synchronous io.
 286                  */
 287                 wait_on_page_locked(page);
 288                 if (PageUptodate(page)) {
 289                         dp = kmap(page);
 290                         if (BITS_PER_LONG == 32 && hash64) {
 291                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 292                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 293                                 *hash  = *hash >> 32;
 294                         } else {
 295                                 *start = le64_to_cpu(dp->ldp_hash_start);
 296                                 *end   = le64_to_cpu(dp->ldp_hash_end);
 297                         }
 298                         LASSERTF(*start <= *hash, "start = %#llx,end = %#llx,hash = %#llx\n",
 299                                  *start, *end, *hash);
 300                         CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash %llu\n",
 301                                offset, *start, *end, *hash);
 302                         if (*hash > *end) {
 303                                 ll_release_page(page, 0);
 304                                 page = NULL;
 305                         } else if (*end != *start && *hash == *end) {
 306                                 /*
 307                                  * upon hash collision, remove this page,
 308                                  * otherwise put page reference, and
 309                                  * ll_get_dir_page() will issue RPC to fetch
 310                                  * the page we want.
 311                                  */
 312                                 ll_release_page(page,
 313                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
 314                                 page = NULL;
 315                         }
 316                 } else {
 317                         page_cache_release(page);
 318                         page = ERR_PTR(-EIO);
 319                 }
 320
 321         } else {
 322                 spin_unlock_irq(&mapping->tree_lock);
 323                 page = NULL;
 324         }
 325         return page;
 326 }
 327
 328 struct page *ll_get_dir_page(struct inode *dir, __u64 hash,
 329                              struct ll_dir_chain *chain)
 330 {
 331         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 332         struct address_space *mapping = dir->i_mapping;
 333         struct lustre_handle lockh;
 334         struct lu_dirpage *dp;
 335         struct page *page;
 336         ldlm_mode_t mode;
 337         int rc;
 338         __u64 start = 0;
 339         __u64 end = 0;
 340         __u64 lhash = hash;
 341         struct ll_inode_info *lli = ll_i2info(dir);
 342         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 343
 344         mode = LCK_PR;
 345         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 346                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
 347         if (!rc) {
 348                 struct ldlm_enqueue_info einfo = {
 349                         .ei_type = LDLM_IBITS,
 350                         .ei_mode = mode,
 351                         .ei_cb_bl = ll_md_blocking_ast,
 352                         .ei_cb_cp = ldlm_completion_ast,
 353                 };
 354                 struct lookup_intent it = { .it_op = IT_READDIR };
 355                 struct ptlrpc_request *request;
 356                 struct md_op_data *op_data;
 357
 358                 op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0,
 359                 LUSTRE_OPC_ANY, NULL);
 360                 if (IS_ERR(op_data))
 361                         return (void *)op_data;
 362
 363                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
 364                                 op_data, &lockh, NULL, 0, NULL, 0);
 365
 366                 ll_finish_md_op_data(op_data);
 367
 368                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 369                 if (request)
 370                         ptlrpc_req_finished(request);
 371                 if (rc < 0) {
 372                         CERROR("lock enqueue: "DFID" at %llu: rc %d\n",
 373                                 PFID(ll_inode2fid(dir)), hash, rc);
 374                         return ERR_PTR(rc);
 375                 }
 376
 377                 CDEBUG(D_INODE, "setting lr_lvb_inode to inode %p (%lu/%u)\n",
 378                        dir, dir->i_ino, dir->i_generation);
 379                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp,
 380                                  &it.d.lustre.it_lock_handle, dir, NULL);
 381         } else {
 382                 /* for cross-ref object, l_ast_data of the lock may not be set,
 383                  * we reset it here */
 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 != NULL) {
 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 != NULL && !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 != NULL) {
 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 != NULL && 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         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 744                 char *param = NULL;
 745                 char *buf;
 746
 747                 param = kzalloc(MGS_PARAM_MAXLEN, GFP_NOFS);
 748                 if (!param)
 749                         return -ENOMEM;
 750
 751                 buf = param;
 752                 /* Get fsname and assume devname to be -MDT0000. */
 753                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 754                 strcat(buf, "-MDT0000.lov");
 755                 buf += strlen(buf);
 756
 757                 /* Set root stripesize */
 758                 sprintf(buf, ".stripesize=%u",
 759                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 760                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 761                 if (rc)
 762                         goto end;
 763
 764                 /* Set root stripecount */
 765                 sprintf(buf, ".stripecount=%hd",
 766                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 767                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 768                 if (rc)
 769                         goto end;
 770
 771                 /* Set root stripeoffset */
 772                 sprintf(buf, ".stripeoffset=%hd",
 773                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 774                         (typeof(lump->lmm_stripe_offset))(-1));
 775                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 776
 777 end:
 778                 kfree(param);
 779         }
 780         return rc;
 781 }
 782
 783 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 784                      int *lmm_size, struct ptlrpc_request **request)
 785 {
 786         struct ll_sb_info *sbi = ll_i2sbi(inode);
 787         struct mdt_body   *body;
 788         struct lov_mds_md *lmm = NULL;
 789         struct ptlrpc_request *req = NULL;
 790         int rc, lmmsize;
 791         struct md_op_data *op_data;
 792
 793         rc = ll_get_default_mdsize(sbi, &lmmsize);
 794         if (rc)
 795                 return rc;
 796
 797         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 798                                      0, lmmsize, LUSTRE_OPC_ANY,
 799                                      NULL);
 800         if (IS_ERR(op_data))
 801                 return PTR_ERR(op_data);
 802
 803         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 804         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 805         ll_finish_md_op_data(op_data);
 806         if (rc < 0) {
 807                 CDEBUG(D_INFO, "md_getattr failed on inode %lu/%u: rc %d\n",
 808                        inode->i_ino,
 809                        inode->i_generation, rc);
 810                 goto out;
 811         }
 812
 813         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 814         LASSERT(body != NULL);
 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         LASSERT(lmm != NULL);
 827
 828         /*
 829          * This is coming from the MDS, so is probably in
 830          * little endian.  We convert it to host endian before
 831          * passing it to userspace.
 832          */
 833         /* We don't swab objects for directories */
 834         switch (le32_to_cpu(lmm->lmm_magic)) {
 835         case LOV_MAGIC_V1:
 836                 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC)
 837                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 838                 break;
 839         case LOV_MAGIC_V3:
 840                 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC)
 841                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 842                 break;
 843         default:
 844                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 845                 rc = -EPROTO;
 846         }
 847 out:
 848         *lmmp = lmm;
 849         *lmm_size = lmmsize;
 850         *request = req;
 851         return rc;
 852 }
 853
 854 /*
 855  *  Get MDT index for the inode.
 856  */
 857 int ll_get_mdt_idx(struct inode *inode)
 858 {
 859         struct ll_sb_info *sbi = ll_i2sbi(inode);
 860         struct md_op_data *op_data;
 861         int rc, mdtidx;
 862
 863         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 864                                      0, LUSTRE_OPC_ANY, NULL);
 865         if (IS_ERR(op_data))
 866                 return PTR_ERR(op_data);
 867
 868         op_data->op_flags |= MF_GET_MDT_IDX;
 869         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 870         mdtidx = op_data->op_mds;
 871         ll_finish_md_op_data(op_data);
 872         if (rc < 0) {
 873                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 874                 return rc;
 875         }
 876         return mdtidx;
 877 }
 878
 879 /**
 880  * Generic handler to do any pre-copy work.
 881  *
 882  * It send a first hsm_progress (with extent length == 0) to coordinator as a
 883  * first information for it that real work has started.
 884  *
 885  * Moreover, for a ARCHIVE request, it will sample the file data version and
 886  * store it in \a copy.
 887  *
 888  * \return 0 on success.
 889  */
 890 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 891 {
 892         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 893         struct hsm_progress_kernel       hpk;
 894         int                              rc;
 895
 896         /* Forge a hsm_progress based on data from copy. */
 897         hpk.hpk_fid = copy->hc_hai.hai_fid;
 898         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 899         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 900         hpk.hpk_extent.length = 0;
 901         hpk.hpk_flags = 0;
 902         hpk.hpk_errval = 0;
 903         hpk.hpk_data_version = 0;
 904
 905         /* For archive request, we need to read the current file version. */
 906         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 907                 struct inode    *inode;
 908                 __u64            data_version = 0;
 909
 910                 /* Get inode for this fid */
 911                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 912                 if (IS_ERR(inode)) {
 913                         hpk.hpk_flags |= HP_FLAG_RETRY;
 914                         /* hpk_errval is >= 0 */
 915                         hpk.hpk_errval = -PTR_ERR(inode);
 916                         rc = PTR_ERR(inode);
 917                         goto progress;
 918                 }
 919
 920                 /* Read current file data version */
 921                 rc = ll_data_version(inode, &data_version, 1);
 922                 iput(inode);
 923                 if (rc != 0) {
 924                         CDEBUG(D_HSM, "Could not read file data version of "
 925                                       DFID" (rc = %d). Archive request (%#llx) could not be done.\n",
 926                                       PFID(&copy->hc_hai.hai_fid), rc,
 927                                       copy->hc_hai.hai_cookie);
 928                         hpk.hpk_flags |= HP_FLAG_RETRY;
 929                         /* hpk_errval must be >= 0 */
 930                         hpk.hpk_errval = -rc;
 931                         goto progress;
 932                 }
 933
 934                 /* Store it the hsm_copy for later copytool use.
 935                  * Always modified even if no lsm. */
 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 it the hsm_copy for later copytool use.
1012                  * Always modified even if no lsm. */
1013                 hpk.hpk_data_version = data_version;
1014
1015                 /* File could have been stripped during archiving, so we need
1016                  * to check anyway. */
1017                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1018                     (copy->hc_data_version != data_version)) {
1019                         CDEBUG(D_HSM, "File data version mismatched. File content was changed during archiving. "
1020                                DFID", start:%#llx current:%#llx\n",
1021                                PFID(&copy->hc_hai.hai_fid),
1022                                copy->hc_data_version, data_version);
1023                         /* File was changed, send error to cdt. Do not ask for
1024                          * retry because if a file is modified frequently,
1025                          * the cdt will loop on retried archive requests.
1026                          * The policy engine will ask for a new archive later
1027                          * when the file will not be modified for some tunable
1028                          * time */
1029                         /* we do not notify caller */
1030                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
1031                         /* hpk_errval must be >= 0 */
1032                         hpk.hpk_errval = EBUSY;
1033                 }
1034
1035         }
1036
1037 progress:
1038         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1039                            &hpk, NULL);
1040
1041         return rc;
1042 }
1043
1044 static int copy_and_ioctl(int cmd, struct obd_export *exp,
1045                           const void __user *data, size_t size)
1046 {
1047         void *copy;
1048         int rc;
1049
1050         copy = kzalloc(size, GFP_NOFS);
1051         if (!copy)
1052                 return -ENOMEM;
1053
1054         if (copy_from_user(copy, data, size)) {
1055                 rc = -EFAULT;
1056                 goto out;
1057         }
1058
1059         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
1060 out:
1061         kfree(copy);
1062
1063         return rc;
1064 }
1065
1066 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1067 {
1068         int cmd = qctl->qc_cmd;
1069         int type = qctl->qc_type;
1070         int id = qctl->qc_id;
1071         int valid = qctl->qc_valid;
1072         int rc = 0;
1073
1074         switch (cmd) {
1075         case LUSTRE_Q_INVALIDATE:
1076         case LUSTRE_Q_FINVALIDATE:
1077         case Q_QUOTAON:
1078         case Q_QUOTAOFF:
1079         case Q_SETQUOTA:
1080         case Q_SETINFO:
1081                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1082                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1083                         return -EPERM;
1084                 break;
1085         case Q_GETQUOTA:
1086                 if (((type == USRQUOTA &&
1087                       !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
1088                      (type == GRPQUOTA &&
1089                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
1090                     (!capable(CFS_CAP_SYS_ADMIN) ||
1091                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
1092                         return -EPERM;
1093                 break;
1094         case Q_GETINFO:
1095                 break;
1096         default:
1097                 CERROR("unsupported quotactl op: %#x\n", cmd);
1098                 return -ENOTTY;
1099         }
1100
1101         if (valid != QC_GENERAL) {
1102                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1103                         return -EOPNOTSUPP;
1104
1105                 if (cmd == Q_GETINFO)
1106                         qctl->qc_cmd = Q_GETOINFO;
1107                 else if (cmd == Q_GETQUOTA)
1108                         qctl->qc_cmd = Q_GETOQUOTA;
1109                 else
1110                         return -EINVAL;
1111
1112                 switch (valid) {
1113                 case QC_MDTIDX:
1114                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1115                                            sizeof(*qctl), qctl, NULL);
1116                         break;
1117                 case QC_OSTIDX:
1118                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1119                                            sizeof(*qctl), qctl, NULL);
1120                         break;
1121                 case QC_UUID:
1122                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1123                                            sizeof(*qctl), qctl, NULL);
1124                         if (rc == -EAGAIN)
1125                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1126                                                    sbi->ll_dt_exp,
1127                                                    sizeof(*qctl), qctl, NULL);
1128                         break;
1129                 default:
1130                         rc = -EINVAL;
1131                         break;
1132                 }
1133
1134                 if (rc)
1135                         return rc;
1136
1137                 qctl->qc_cmd = cmd;
1138         } else {
1139                 struct obd_quotactl *oqctl;
1140
1141                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1142                 if (!oqctl)
1143                         return -ENOMEM;
1144
1145                 QCTL_COPY(oqctl, qctl);
1146                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1147                 if (rc) {
1148                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
1149                                 oqctl->qc_cmd = Q_QUOTAOFF;
1150                                 obd_quotactl(sbi->ll_md_exp, oqctl);
1151                         }
1152                         kfree(oqctl);
1153                         return rc;
1154                 }
1155                 /* If QIF_SPACE is not set, client should collect the
1156                  * space usage from OSSs by itself */
1157                 if (cmd == Q_GETQUOTA &&
1158                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1159                     !oqctl->qc_dqblk.dqb_curspace) {
1160                         struct obd_quotactl *oqctl_tmp;
1161
1162                         oqctl_tmp = kzalloc(sizeof(*oqctl_tmp), GFP_NOFS);
1163                         if (!oqctl_tmp) {
1164                                 rc = -ENOMEM;
1165                                 goto out;
1166                         }
1167
1168                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1169                         oqctl_tmp->qc_id = oqctl->qc_id;
1170                         oqctl_tmp->qc_type = oqctl->qc_type;
1171
1172                         /* collect space usage from OSTs */
1173                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1174                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1175                         if (!rc || rc == -EREMOTEIO) {
1176                                 oqctl->qc_dqblk.dqb_curspace =
1177                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1178                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1179                         }
1180
1181                         /* collect space & inode usage from MDTs */
1182                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1183                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1184                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1185                         if (!rc || rc == -EREMOTEIO) {
1186                                 oqctl->qc_dqblk.dqb_curspace +=
1187                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1188                                 oqctl->qc_dqblk.dqb_curinodes =
1189                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1190                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1191                         } else {
1192                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1193                         }
1194
1195                         kfree(oqctl_tmp);
1196                 }
1197 out:
1198                 QCTL_COPY(qctl, oqctl);
1199                 kfree(oqctl);
1200         }
1201
1202         return rc;
1203 }
1204
1205 /* This function tries to get a single name component,
1206  * to send to the server. No actual path traversal involved,
1207  * so we limit to NAME_MAX */
1208 static char *ll_getname(const char __user *filename)
1209 {
1210         int ret = 0, len;
1211         char *tmp;
1212
1213         tmp = kzalloc(NAME_MAX + 1, GFP_KERNEL);
1214         if (!tmp)
1215                 return ERR_PTR(-ENOMEM);
1216
1217         len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1218         if (len < 0)
1219                 ret = len;
1220         else if (len == 0)
1221                 ret = -ENOENT;
1222         else if (len > NAME_MAX && tmp[NAME_MAX] != 0)
1223                 ret = -ENAMETOOLONG;
1224
1225         if (ret) {
1226                 kfree(tmp);
1227                 tmp =  ERR_PTR(ret);
1228         }
1229         return tmp;
1230 }
1231
1232 #define ll_putname(filename) kfree(filename)
1233
1234 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1235 {
1236         struct inode *inode = file_inode(file);
1237         struct ll_sb_info *sbi = ll_i2sbi(inode);
1238         struct obd_ioctl_data *data;
1239         int rc = 0;
1240
1241         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1242                inode->i_ino, inode->i_generation, inode, cmd);
1243
1244         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1245         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1246                 return -ENOTTY;
1247
1248         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1249         switch (cmd) {
1250         case FSFILT_IOC_GETFLAGS:
1251         case FSFILT_IOC_SETFLAGS:
1252                 return ll_iocontrol(inode, file, cmd, arg);
1253         case FSFILT_IOC_GETVERSION_OLD:
1254         case FSFILT_IOC_GETVERSION:
1255                 return put_user(inode->i_generation, (int *)arg);
1256         /* We need to special case any other ioctls we want to handle,
1257          * to send them to the MDS/OST as appropriate and to properly
1258          * network encode the arg field.
1259         case FSFILT_IOC_SETVERSION_OLD:
1260         case FSFILT_IOC_SETVERSION:
1261         */
1262         case LL_IOC_GET_MDTIDX: {
1263                 int mdtidx;
1264
1265                 mdtidx = ll_get_mdt_idx(inode);
1266                 if (mdtidx < 0)
1267                         return mdtidx;
1268
1269                 if (put_user((int)mdtidx, (int *)arg))
1270                         return -EFAULT;
1271
1272                 return 0;
1273         }
1274         case IOC_MDC_LOOKUP: {
1275                 struct ptlrpc_request *request = NULL;
1276                 int namelen, len = 0;
1277                 char *buf = NULL;
1278                 char *filename;
1279                 struct md_op_data *op_data;
1280
1281                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1282                 if (rc)
1283                         return rc;
1284                 data = (void *)buf;
1285
1286                 filename = data->ioc_inlbuf1;
1287                 namelen = strlen(filename);
1288
1289                 if (namelen < 1) {
1290                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1291                         rc = -EINVAL;
1292                         goto out_free;
1293                 }
1294
1295                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1296                                              0, LUSTRE_OPC_ANY, NULL);
1297                 if (IS_ERR(op_data)) {
1298                         rc = PTR_ERR(op_data);
1299                         goto out_free;
1300                 }
1301
1302                 op_data->op_valid = OBD_MD_FLID;
1303                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1304                 ll_finish_md_op_data(op_data);
1305                 if (rc < 0) {
1306                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1307                         goto out_free;
1308                 }
1309                 ptlrpc_req_finished(request);
1310 out_free:
1311                 obd_ioctl_freedata(buf, len);
1312                 return rc;
1313         }
1314         case LL_IOC_LMV_SETSTRIPE: {
1315                 struct lmv_user_md  *lum;
1316                 char            *buf = NULL;
1317                 char            *filename;
1318                 int              namelen = 0;
1319                 int              lumlen = 0;
1320                 int              len;
1321                 int              rc;
1322
1323                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1324                 if (rc)
1325                         return rc;
1326
1327                 data = (void *)buf;
1328                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1329                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0) {
1330                         rc = -EINVAL;
1331                         goto lmv_out_free;
1332                 }
1333
1334                 filename = data->ioc_inlbuf1;
1335                 namelen = data->ioc_inllen1;
1336
1337                 if (namelen < 1) {
1338                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1339                         rc = -EINVAL;
1340                         goto lmv_out_free;
1341                 }
1342                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1343                 lumlen = data->ioc_inllen2;
1344
1345                 if (lum->lum_magic != LMV_USER_MAGIC ||
1346                     lumlen != sizeof(*lum)) {
1347                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1348                                filename, lum->lum_magic, lumlen, -EFAULT);
1349                         rc = -EINVAL;
1350                         goto lmv_out_free;
1351                 }
1352
1353                 /**
1354                  * ll_dir_setdirstripe will be used to set dir stripe
1355                  *  mdc_create--->mdt_reint_create (with dirstripe)
1356                  */
1357                 rc = ll_dir_setdirstripe(inode, lum, filename);
1358 lmv_out_free:
1359                 obd_ioctl_freedata(buf, len);
1360                 return rc;
1361
1362         }
1363         case LL_IOC_LOV_SETSTRIPE: {
1364                 struct lov_user_md_v3 lumv3;
1365                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1366                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1367                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1368
1369                 int set_default = 0;
1370
1371                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1372                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1373                         sizeof(lumv3p->lmm_objects[0]));
1374                 /* first try with v1 which is smaller than v3 */
1375                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1376                         return -EFAULT;
1377
1378                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1379                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1380                                 return -EFAULT;
1381                 }
1382
1383                 if (is_root_inode(inode))
1384                         set_default = 1;
1385
1386                 /* in v1 and v3 cases lumv1 points to data */
1387                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1388
1389                 return rc;
1390         }
1391         case LL_IOC_LMV_GETSTRIPE: {
1392                 struct lmv_user_md *lump = (struct lmv_user_md *)arg;
1393                 struct lmv_user_md lum;
1394                 struct lmv_user_md *tmp;
1395                 int lum_size;
1396                 int rc = 0;
1397                 int mdtindex;
1398
1399                 if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
1400                         return -EFAULT;
1401
1402                 if (lum.lum_magic != LMV_MAGIC_V1)
1403                         return -EINVAL;
1404
1405                 lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
1406                 tmp = kzalloc(lum_size, GFP_NOFS);
1407                 if (!tmp) {
1408                         rc = -ENOMEM;
1409                         goto free_lmv;
1410                 }
1411
1412                 *tmp = lum;
1413                 tmp->lum_type = LMV_STRIPE_TYPE;
1414                 tmp->lum_stripe_count = 1;
1415                 mdtindex = ll_get_mdt_idx(inode);
1416                 if (mdtindex < 0) {
1417                         rc = -ENOMEM;
1418                         goto free_lmv;
1419                 }
1420
1421                 tmp->lum_stripe_offset = mdtindex;
1422                 tmp->lum_objects[0].lum_mds = mdtindex;
1423                 memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
1424                        sizeof(struct lu_fid));
1425                 if (copy_to_user((void *)arg, tmp, lum_size)) {
1426                         rc = -EFAULT;
1427                         goto free_lmv;
1428                 }
1429 free_lmv:
1430                 kfree(tmp);
1431                 return rc;
1432         }
1433         case LL_IOC_LOV_SWAP_LAYOUTS:
1434                 return -EPERM;
1435         case LL_IOC_OBD_STATFS:
1436                 return ll_obd_statfs(inode, (void *)arg);
1437         case LL_IOC_LOV_GETSTRIPE:
1438         case LL_IOC_MDC_GETINFO:
1439         case IOC_MDC_GETFILEINFO:
1440         case IOC_MDC_GETFILESTRIPE: {
1441                 struct ptlrpc_request *request = NULL;
1442                 struct lov_user_md *lump;
1443                 struct lov_mds_md *lmm = NULL;
1444                 struct mdt_body *body;
1445                 char *filename = NULL;
1446                 int lmmsize;
1447
1448                 if (cmd == IOC_MDC_GETFILEINFO ||
1449                     cmd == IOC_MDC_GETFILESTRIPE) {
1450                         filename = ll_getname((const char *)arg);
1451                         if (IS_ERR(filename))
1452                                 return PTR_ERR(filename);
1453
1454                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1455                                                       &lmmsize, &request);
1456                 } else {
1457                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1458                 }
1459
1460                 if (request) {
1461                         body = req_capsule_server_get(&request->rq_pill,
1462                                                       &RMF_MDT_BODY);
1463                         LASSERT(body != NULL);
1464                 } else {
1465                         goto out_req;
1466                 }
1467
1468                 if (rc < 0) {
1469                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1470                                                cmd == LL_IOC_MDC_GETINFO)) {
1471                                 rc = 0;
1472                                 goto skip_lmm;
1473                         } else
1474                                 goto out_req;
1475                 }
1476
1477                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1478                     cmd == LL_IOC_LOV_GETSTRIPE) {
1479                         lump = (struct lov_user_md *)arg;
1480                 } else {
1481                         struct lov_user_mds_data *lmdp;
1482
1483                         lmdp = (struct lov_user_mds_data *)arg;
1484                         lump = &lmdp->lmd_lmm;
1485                 }
1486                 if (copy_to_user(lump, lmm, lmmsize)) {
1487                         if (copy_to_user(lump, lmm, sizeof(*lump))) {
1488                                 rc = -EFAULT;
1489                                 goto out_req;
1490                         }
1491                         rc = -EOVERFLOW;
1492                 }
1493 skip_lmm:
1494                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1495                         struct lov_user_mds_data *lmdp;
1496                         lstat_t st = { 0 };
1497
1498                         st.st_dev     = inode->i_sb->s_dev;
1499                         st.st_mode    = body->mode;
1500                         st.st_nlink   = body->nlink;
1501                         st.st_uid     = body->uid;
1502                         st.st_gid     = body->gid;
1503                         st.st_rdev    = body->rdev;
1504                         st.st_size    = body->size;
1505                         st.st_blksize = PAGE_CACHE_SIZE;
1506                         st.st_blocks  = body->blocks;
1507                         st.st_atime   = body->atime;
1508                         st.st_mtime   = body->mtime;
1509                         st.st_ctime   = body->ctime;
1510                         st.st_ino     = inode->i_ino;
1511
1512                         lmdp = (struct lov_user_mds_data *)arg;
1513                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st))) {
1514                                 rc = -EFAULT;
1515                                 goto out_req;
1516                         }
1517                 }
1518
1519 out_req:
1520                 ptlrpc_req_finished(request);
1521                 if (filename)
1522                         ll_putname(filename);
1523                 return rc;
1524         }
1525         case IOC_LOV_GETINFO: {
1526                 struct lov_user_mds_data *lumd;
1527                 struct lov_stripe_md *lsm;
1528                 struct lov_user_md *lum;
1529                 struct lov_mds_md *lmm;
1530                 int lmmsize;
1531                 lstat_t st;
1532
1533                 lumd = (struct lov_user_mds_data *)arg;
1534                 lum = &lumd->lmd_lmm;
1535
1536                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1537                 if (rc)
1538                         return rc;
1539
1540                 lmm = libcfs_kvzalloc(lmmsize, GFP_NOFS);
1541                 if (lmm == NULL)
1542                         return -ENOMEM;
1543                 if (copy_from_user(lmm, lum, lmmsize)) {
1544                         rc = -EFAULT;
1545                         goto free_lmm;
1546                 }
1547
1548                 switch (lmm->lmm_magic) {
1549                 case LOV_USER_MAGIC_V1:
1550                         if (cpu_to_le32(LOV_USER_MAGIC_V1) == LOV_USER_MAGIC_V1)
1551                                 break;
1552                         /* swab objects first so that stripes num will be sane */
1553                         lustre_swab_lov_user_md_objects(
1554                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1555                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1556                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1557                         break;
1558                 case LOV_USER_MAGIC_V3:
1559                         if (cpu_to_le32(LOV_USER_MAGIC_V3) == LOV_USER_MAGIC_V3)
1560                                 break;
1561                         /* swab objects first so that stripes num will be sane */
1562                         lustre_swab_lov_user_md_objects(
1563                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1564                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1565                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1566                         break;
1567                 default:
1568                         rc = -EINVAL;
1569                         goto free_lmm;
1570                 }
1571
1572                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1573                 if (rc < 0) {
1574                         rc = -ENOMEM;
1575                         goto free_lmm;
1576                 }
1577
1578                 /* Perform glimpse_size operation. */
1579                 memset(&st, 0, sizeof(st));
1580
1581                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1582                 if (rc)
1583                         goto free_lsm;
1584
1585                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st))) {
1586                         rc = -EFAULT;
1587                         goto free_lsm;
1588                 }
1589
1590 free_lsm:
1591                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1592 free_lmm:
1593                 kvfree(lmm);
1594                 return rc;
1595         }
1596         case OBD_IOC_LLOG_CATINFO: {
1597                 return -EOPNOTSUPP;
1598         }
1599         case OBD_IOC_QUOTACHECK: {
1600                 struct obd_quotactl *oqctl;
1601                 int error = 0;
1602
1603                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1604                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1605                         return -EPERM;
1606
1607                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1608                 if (!oqctl)
1609                         return -ENOMEM;
1610                 oqctl->qc_type = arg;
1611                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1612                 if (rc < 0) {
1613                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1614                         error = rc;
1615                 }
1616
1617                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1618                 if (rc < 0)
1619                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1620
1621                 kfree(oqctl);
1622                 return error ?: rc;
1623         }
1624         case OBD_IOC_POLL_QUOTACHECK: {
1625                 struct if_quotacheck *check;
1626
1627                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1628                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1629                         return -EPERM;
1630
1631                 check = kzalloc(sizeof(*check), GFP_NOFS);
1632                 if (!check)
1633                         return -ENOMEM;
1634
1635                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1636                                    NULL);
1637                 if (rc) {
1638                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1639                         if (copy_to_user((void *)arg, check,
1640                                              sizeof(*check)))
1641                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1642                         goto out_poll;
1643                 }
1644
1645                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1646                                    NULL);
1647                 if (rc) {
1648                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1649                         if (copy_to_user((void *)arg, check,
1650                                              sizeof(*check)))
1651                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1652                         goto out_poll;
1653                 }
1654 out_poll:
1655                 kfree(check);
1656                 return rc;
1657         }
1658         case LL_IOC_QUOTACTL: {
1659                 struct if_quotactl *qctl;
1660
1661                 qctl = kzalloc(sizeof(*qctl), GFP_NOFS);
1662                 if (!qctl)
1663                         return -ENOMEM;
1664
1665                 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl))) {
1666                         rc = -EFAULT;
1667                         goto out_quotactl;
1668                 }
1669
1670                 rc = quotactl_ioctl(sbi, qctl);
1671
1672                 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1673                         rc = -EFAULT;
1674
1675 out_quotactl:
1676                 kfree(qctl);
1677                 return rc;
1678         }
1679         case OBD_IOC_GETDTNAME:
1680         case OBD_IOC_GETMDNAME:
1681                 return ll_get_obd_name(inode, cmd, arg);
1682         case LL_IOC_FLUSHCTX:
1683                 return ll_flush_ctx(inode);
1684 #ifdef CONFIG_FS_POSIX_ACL
1685         case LL_IOC_RMTACL: {
1686             if (sbi->ll_flags & LL_SBI_RMT_CLIENT && is_root_inode(inode)) {
1687                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1688
1689                 LASSERT(fd != NULL);
1690                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1691                 if (!rc)
1692                         fd->fd_flags |= LL_FILE_RMTACL;
1693                 return rc;
1694             } else
1695                 return 0;
1696         }
1697 #endif
1698         case LL_IOC_GETOBDCOUNT: {
1699                 int count, vallen;
1700                 struct obd_export *exp;
1701
1702                 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1703                         return -EFAULT;
1704
1705                 /* get ost count when count is zero, get mdt count otherwise */
1706                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1707                 vallen = sizeof(count);
1708                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1709                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1710                 if (rc) {
1711                         CERROR("get target count failed: %d\n", rc);
1712                         return rc;
1713                 }
1714
1715                 if (copy_to_user((int *)arg, &count, sizeof(int)))
1716                         return -EFAULT;
1717
1718                 return 0;
1719         }
1720         case LL_IOC_PATH2FID:
1721                 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1722                                      sizeof(struct lu_fid)))
1723                         return -EFAULT;
1724                 return 0;
1725         case LL_IOC_GET_CONNECT_FLAGS: {
1726                 return obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void *)arg);
1727         }
1728         case OBD_IOC_CHANGELOG_SEND:
1729         case OBD_IOC_CHANGELOG_CLEAR:
1730                 if (!capable(CFS_CAP_SYS_ADMIN))
1731                         return -EPERM;
1732
1733                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1734                                     sizeof(struct ioc_changelog));
1735                 return rc;
1736         case OBD_IOC_FID2PATH:
1737                 return ll_fid2path(inode, (void *)arg);
1738         case LL_IOC_HSM_REQUEST: {
1739                 struct hsm_user_request *hur;
1740                 ssize_t                  totalsize;
1741
1742                 hur = memdup_user((void *)arg, sizeof(*hur));
1743                 if (IS_ERR(hur))
1744                         return PTR_ERR(hur);
1745
1746                 /* Compute the whole struct size */
1747                 totalsize = hur_len(hur);
1748                 kfree(hur);
1749                 if (totalsize < 0)
1750                         return -E2BIG;
1751
1752                 /* Final size will be more than double totalsize */
1753                 if (totalsize >= MDS_MAXREQSIZE / 3)
1754                         return -E2BIG;
1755
1756                 hur = libcfs_kvzalloc(totalsize, GFP_NOFS);
1757                 if (hur == NULL)
1758                         return -ENOMEM;
1759
1760                 /* Copy the whole struct */
1761                 if (copy_from_user(hur, (void *)arg, totalsize)) {
1762                         kvfree(hur);
1763                         return -EFAULT;
1764                 }
1765
1766                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1767                         const struct lu_fid *fid;
1768                         struct inode *f;
1769                         int i;
1770
1771                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1772                                 fid = &hur->hur_user_item[i].hui_fid;
1773                                 f = search_inode_for_lustre(inode->i_sb, fid);
1774                                 if (IS_ERR(f)) {
1775                                         rc = PTR_ERR(f);
1776                                         break;
1777                                 }
1778
1779                                 rc = ll_hsm_release(f);
1780                                 iput(f);
1781                                 if (rc != 0)
1782                                         break;
1783                         }
1784                 } else {
1785                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1786                                            hur, NULL);
1787                 }
1788
1789                 kvfree(hur);
1790
1791                 return rc;
1792         }
1793         case LL_IOC_HSM_PROGRESS: {
1794                 struct hsm_progress_kernel      hpk;
1795                 struct hsm_progress             hp;
1796
1797                 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1798                         return -EFAULT;
1799
1800                 hpk.hpk_fid = hp.hp_fid;
1801                 hpk.hpk_cookie = hp.hp_cookie;
1802                 hpk.hpk_extent = hp.hp_extent;
1803                 hpk.hpk_flags = hp.hp_flags;
1804                 hpk.hpk_errval = hp.hp_errval;
1805                 hpk.hpk_data_version = 0;
1806
1807                 /* File may not exist in Lustre; all progress
1808                  * reported to Lustre root */
1809                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1810                                    NULL);
1811                 return rc;
1812         }
1813         case LL_IOC_HSM_CT_START:
1814                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1815                                     sizeof(struct lustre_kernelcomm));
1816                 return rc;
1817
1818         case LL_IOC_HSM_COPY_START: {
1819                 struct hsm_copy *copy;
1820                 int              rc;
1821
1822                 copy = memdup_user((char *)arg, sizeof(*copy));
1823                 if (IS_ERR(copy))
1824                         return PTR_ERR(copy);
1825
1826                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1827                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1828                         rc = -EFAULT;
1829
1830                 kfree(copy);
1831                 return rc;
1832         }
1833         case LL_IOC_HSM_COPY_END: {
1834                 struct hsm_copy *copy;
1835                 int              rc;
1836
1837                 copy = memdup_user((char *)arg, sizeof(*copy));
1838                 if (IS_ERR(copy))
1839                         return PTR_ERR(copy);
1840
1841                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1842                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1843                         rc = -EFAULT;
1844
1845                 kfree(copy);
1846                 return rc;
1847         }
1848         default:
1849                 return obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL, (void *)arg);
1850         }
1851 }
1852
1853 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1854 {
1855         struct inode *inode = file->f_mapping->host;
1856         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1857         struct ll_sb_info *sbi = ll_i2sbi(inode);
1858         int api32 = ll_need_32bit_api(sbi);
1859         loff_t ret = -EINVAL;
1860
1861         mutex_lock(&inode->i_mutex);
1862         switch (origin) {
1863         case SEEK_SET:
1864                 break;
1865         case SEEK_CUR:
1866                 offset += file->f_pos;
1867                 break;
1868         case SEEK_END:
1869                 if (offset > 0)
1870                         goto out;
1871                 if (api32)
1872                         offset += LL_DIR_END_OFF_32BIT;
1873                 else
1874                         offset += LL_DIR_END_OFF;
1875                 break;
1876         default:
1877                 goto out;
1878         }
1879
1880         if (offset >= 0 &&
1881             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1882              (!api32 && offset <= LL_DIR_END_OFF))) {
1883                 if (offset != file->f_pos) {
1884                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1885                             (!api32 && offset == LL_DIR_END_OFF))
1886                                 fd->lfd_pos = MDS_DIR_END_OFF;
1887                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1888                                 fd->lfd_pos = offset << 32;
1889                         else
1890                                 fd->lfd_pos = offset;
1891                         file->f_pos = offset;
1892                         file->f_version = 0;
1893                 }
1894                 ret = offset;
1895         }
1896         goto out;
1897
1898 out:
1899         mutex_unlock(&inode->i_mutex);
1900         return ret;
1901 }
1902
1903 static int ll_dir_open(struct inode *inode, struct file *file)
1904 {
1905         return ll_file_open(inode, file);
1906 }
1907
1908 static int ll_dir_release(struct inode *inode, struct file *file)
1909 {
1910         return ll_file_release(inode, file);
1911 }
1912
1913 const struct file_operations ll_dir_operations = {
1914         .llseek   = ll_dir_seek,
1915         .open     = ll_dir_open,
1916         .release  = ll_dir_release,
1917         .read     = generic_read_dir,
1918         .iterate  = ll_readdir,
1919         .unlocked_ioctl   = ll_dir_ioctl,
1920         .fsync    = ll_fsync,
1921 };