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_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the lu_dirpage
 138  * for this integrated page will be adjusted. See 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_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                 if (body->valid & OBD_MD_FLSIZE)
 193                         cl_isize_write(inode, body->size);
 194
 195                 nrdpgs = (request->rq_bulk->bd_nob_transferred+PAGE_SIZE-1)
 196                          >> PAGE_SHIFT;
 197                 SetPageUptodate(page0);
 198         }
 199         unlock_page(page0);
 200         ptlrpc_req_finished(request);
 201
 202         CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages);
 203
 204         for (i = 1; i < npages; i++) {
 205                 unsigned long offset;
 206                 int ret;
 207
 208                 page = page_pool[i];
 209
 210                 if (rc < 0 || i >= nrdpgs) {
 211                         put_page(page);
 212                         continue;
 213                 }
 214
 215                 SetPageUptodate(page);
 216
 217                 dp = kmap(page);
 218                 hash = le64_to_cpu(dp->ldp_hash_start);
 219                 kunmap(page);
 220
 221                 offset = hash_x_index(hash, hash64);
 222
 223                 prefetchw(&page->flags);
 224                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
 225                                             GFP_NOFS);
 226                 if (ret == 0) {
 227                         unlock_page(page);
 228                 } else {
 229                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed: %d\n",
 230                                offset, ret);
 231                 }
 232                 put_page(page);
 233         }
 234
 235         if (page_pool != &page0)
 236                 kfree(page_pool);
 237         return rc;
 238 }
 239
 240 void ll_release_page(struct page *page, int remove)
 241 {
 242         kunmap(page);
 243         if (remove) {
 244                 lock_page(page);
 245                 if (likely(page->mapping))
 246                         truncate_complete_page(page->mapping, page);
 247                 unlock_page(page);
 248         }
 249         put_page(page);
 250 }
 251
 252 /*
 253  * Find, kmap and return page that contains given hash.
 254  */
 255 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
 256                                        __u64 *start, __u64 *end)
 257 {
 258         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 259         struct address_space *mapping = dir->i_mapping;
 260         /*
 261          * Complement of hash is used as an index so that
 262          * radix_tree_gang_lookup() can be used to find a page with starting
 263          * hash _smaller_ than one we are looking for.
 264          */
 265         unsigned long offset = hash_x_index(*hash, hash64);
 266         struct page *page;
 267         int found;
 268
 269         spin_lock_irq(&mapping->tree_lock);
 270         found = radix_tree_gang_lookup(&mapping->page_tree,
 271                                        (void **)&page, offset, 1);
 272         if (found > 0 && !radix_tree_exceptional_entry(page)) {
 273                 struct lu_dirpage *dp;
 274
 275                 get_page(page);
 276                 spin_unlock_irq(&mapping->tree_lock);
 277                 /*
 278                  * In contrast to find_lock_page() we are sure that directory
 279                  * page cannot be truncated (while DLM lock is held) and,
 280                  * hence, can avoid restart.
 281                  *
 282                  * In fact, page cannot be locked here at all, because
 283                  * ll_dir_filler() does synchronous io.
 284                  */
 285                 wait_on_page_locked(page);
 286                 if (PageUptodate(page)) {
 287                         dp = kmap(page);
 288                         if (BITS_PER_LONG == 32 && hash64) {
 289                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 290                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 291                                 *hash  = *hash >> 32;
 292                         } else {
 293                                 *start = le64_to_cpu(dp->ldp_hash_start);
 294                                 *end   = le64_to_cpu(dp->ldp_hash_end);
 295                         }
 296                         LASSERTF(*start <= *hash, "start = %#llx,end = %#llx,hash = %#llx\n",
 297                                  *start, *end, *hash);
 298                         CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash %llu\n",
 299                                offset, *start, *end, *hash);
 300                         if (*hash > *end) {
 301                                 ll_release_page(page, 0);
 302                                 page = NULL;
 303                         } else if (*end != *start && *hash == *end) {
 304                                 /*
 305                                  * upon hash collision, remove this page,
 306                                  * otherwise put page reference, and
 307                                  * ll_get_dir_page() will issue RPC to fetch
 308                                  * the page we want.
 309                                  */
 310                                 ll_release_page(page,
 311                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
 312                                 page = NULL;
 313                         }
 314                 } else {
 315                         put_page(page);
 316                         page = ERR_PTR(-EIO);
 317                 }
 318
 319         } else {
 320                 spin_unlock_irq(&mapping->tree_lock);
 321                 page = NULL;
 322         }
 323         return page;
 324 }
 325
 326 struct page *ll_get_dir_page(struct inode *dir, __u64 hash,
 327                              struct ll_dir_chain *chain)
 328 {
 329         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 330         struct address_space *mapping = dir->i_mapping;
 331         struct lustre_handle lockh;
 332         struct lu_dirpage *dp;
 333         struct page *page;
 334         enum ldlm_mode mode;
 335         int rc;
 336         __u64 start = 0;
 337         __u64 end = 0;
 338         __u64 lhash = hash;
 339         struct ll_inode_info *lli = ll_i2info(dir);
 340         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 341
 342         mode = LCK_PR;
 343         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 344                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
 345         if (!rc) {
 346                 struct ldlm_enqueue_info einfo = {
 347                         .ei_type = LDLM_IBITS,
 348                         .ei_mode = mode,
 349                         .ei_cb_bl = ll_md_blocking_ast,
 350                         .ei_cb_cp = ldlm_completion_ast,
 351                 };
 352                 struct lookup_intent it = { .it_op = IT_READDIR };
 353                 struct ptlrpc_request *request;
 354                 struct md_op_data *op_data;
 355
 356                 op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0,
 357                                              LUSTRE_OPC_ANY, NULL);
 358                 if (IS_ERR(op_data))
 359                         return (void *)op_data;
 360
 361                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
 362                                 op_data, &lockh, NULL, 0, NULL, 0);
 363
 364                 ll_finish_md_op_data(op_data);
 365
 366                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 367                 if (request)
 368                         ptlrpc_req_finished(request);
 369                 if (rc < 0) {
 370                         CERROR("lock enqueue: " DFID " at %llu: rc %d\n",
 371                                PFID(ll_inode2fid(dir)), hash, rc);
 372                         return ERR_PTR(rc);
 373                 }
 374
 375                 CDEBUG(D_INODE, "setting lr_lvb_inode to inode %p (%lu/%u)\n",
 376                        dir, dir->i_ino, dir->i_generation);
 377                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp,
 378                                  &it.d.lustre.it_lock_handle, dir, NULL);
 379         } else {
 380                 /* for cross-ref object, l_ast_data of the lock may not be set,
 381                  * we reset it here
 382                  */
 383                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
 384                                  dir, NULL);
 385         }
 386         ldlm_lock_dump_handle(D_OTHER, &lockh);
 387
 388         mutex_lock(&lli->lli_readdir_mutex);
 389         page = ll_dir_page_locate(dir, &lhash, &start, &end);
 390         if (IS_ERR(page)) {
 391                 CERROR("dir page locate: "DFID" at %llu: rc %ld\n",
 392                        PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
 393                 goto out_unlock;
 394         } else if (page) {
 395                 /*
 396                  * XXX nikita: not entirely correct handling of a corner case:
 397                  * suppose hash chain of entries with hash value HASH crosses
 398                  * border between pages P0 and P1. First both P0 and P1 are
 399                  * cached, seekdir() is called for some entry from the P0 part
 400                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 401                  * happens and finds P1, as it starts with matching hash
 402                  * value. Remaining entries from P0 part of the chain are
 403                  * skipped. (Is that really a bug?)
 404                  *
 405                  * Possible solutions: 0. don't cache P1 is such case, handle
 406                  * it as an "overflow" page. 1. invalidate all pages at
 407                  * once. 2. use HASH|1 as an index for P1.
 408                  */
 409                 goto hash_collision;
 410         }
 411
 412         page = read_cache_page(mapping, hash_x_index(hash, hash64),
 413                                ll_dir_filler, &lhash);
 414         if (IS_ERR(page)) {
 415                 CERROR("read cache page: "DFID" at %llu: rc %ld\n",
 416                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 417                 goto out_unlock;
 418         }
 419
 420         wait_on_page_locked(page);
 421         (void)kmap(page);
 422         if (!PageUptodate(page)) {
 423                 CERROR("page not updated: "DFID" at %llu: rc %d\n",
 424                        PFID(ll_inode2fid(dir)), hash, -5);
 425                 goto fail;
 426         }
 427         if (!PageChecked(page))
 428                 /* XXX: check page format later */
 429                 SetPageChecked(page);
 430         if (PageError(page)) {
 431                 CERROR("page error: "DFID" at %llu: rc %d\n",
 432                        PFID(ll_inode2fid(dir)), hash, -5);
 433                 goto fail;
 434         }
 435 hash_collision:
 436         dp = page_address(page);
 437         if (BITS_PER_LONG == 32 && hash64) {
 438                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 439                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 440                 lhash = hash >> 32;
 441         } else {
 442                 start = le64_to_cpu(dp->ldp_hash_start);
 443                 end   = le64_to_cpu(dp->ldp_hash_end);
 444                 lhash = hash;
 445         }
 446         if (end == start) {
 447                 LASSERT(start == lhash);
 448                 CWARN("Page-wide hash collision: %llu\n", end);
 449                 if (BITS_PER_LONG == 32 && hash64)
 450                         CWARN("Real page-wide hash collision at [%llu %llu] with hash %llu\n",
 451                               le64_to_cpu(dp->ldp_hash_start),
 452                               le64_to_cpu(dp->ldp_hash_end), hash);
 453                 /*
 454                  * Fetch whole overflow chain...
 455                  *
 456                  * XXX not yet.
 457                  */
 458                 goto fail;
 459         }
 460 out_unlock:
 461         mutex_unlock(&lli->lli_readdir_mutex);
 462         ldlm_lock_decref(&lockh, mode);
 463         return page;
 464
 465 fail:
 466         ll_release_page(page, 1);
 467         page = ERR_PTR(-EIO);
 468         goto out_unlock;
 469 }
 470
 471 int ll_dir_read(struct inode *inode, struct dir_context *ctx)
 472 {
 473         struct ll_inode_info *info       = ll_i2info(inode);
 474         struct ll_sb_info    *sbi       = ll_i2sbi(inode);
 475         __u64              pos          = ctx->pos;
 476         int                api32      = ll_need_32bit_api(sbi);
 477         int                hash64     = sbi->ll_flags & LL_SBI_64BIT_HASH;
 478         struct page       *page;
 479         struct ll_dir_chain   chain;
 480         int                done = 0;
 481         int                rc = 0;
 482
 483         ll_dir_chain_init(&chain);
 484
 485         page = ll_get_dir_page(inode, pos, &chain);
 486
 487         while (rc == 0 && !done) {
 488                 struct lu_dirpage *dp;
 489                 struct lu_dirent  *ent;
 490
 491                 if (!IS_ERR(page)) {
 492                         /*
 493                          * If page is empty (end of directory is reached),
 494                          * use this value.
 495                          */
 496                         __u64 hash = MDS_DIR_END_OFF;
 497                         __u64 next;
 498
 499                         dp = page_address(page);
 500                         for (ent = lu_dirent_start(dp); ent && !done;
 501                              ent = lu_dirent_next(ent)) {
 502                                 __u16     type;
 503                                 int         namelen;
 504                                 struct lu_fid  fid;
 505                                 __u64     lhash;
 506                                 __u64     ino;
 507
 508                                 /*
 509                                  * XXX: implement correct swabbing here.
 510                                  */
 511
 512                                 hash = le64_to_cpu(ent->lde_hash);
 513                                 if (hash < pos)
 514                                         /*
 515                                          * Skip until we find target hash
 516                                          * value.
 517                                          */
 518                                         continue;
 519
 520                                 namelen = le16_to_cpu(ent->lde_namelen);
 521                                 if (namelen == 0)
 522                                         /*
 523                                          * Skip dummy record.
 524                                          */
 525                                         continue;
 526
 527                                 if (api32 && hash64)
 528                                         lhash = hash >> 32;
 529                                 else
 530                                         lhash = hash;
 531                                 fid_le_to_cpu(&fid, &ent->lde_fid);
 532                                 ino = cl_fid_build_ino(&fid, api32);
 533                                 type = ll_dirent_type_get(ent);
 534                                 ctx->pos = lhash;
 535                                 /* For 'll_nfs_get_name_filldir()', it will try
 536                                  * to access the 'ent' through its 'lde_name',
 537                                  * so the parameter 'name' for 'ctx->actor()'
 538                                  * must be part of the 'ent'.
 539                                  */
 540                                 done = !dir_emit(ctx, ent->lde_name,
 541                                                  namelen, ino, type);
 542                         }
 543                         next = le64_to_cpu(dp->ldp_hash_end);
 544                         if (!done) {
 545                                 pos = next;
 546                                 if (pos == MDS_DIR_END_OFF) {
 547                                         /*
 548                                          * End of directory reached.
 549                                          */
 550                                         done = 1;
 551                                         ll_release_page(page, 0);
 552                                 } else if (1 /* chain is exhausted*/) {
 553                                         /*
 554                                          * Normal case: continue to the next
 555                                          * page.
 556                                          */
 557                                         ll_release_page(page,
 558                                             le32_to_cpu(dp->ldp_flags) &
 559                                                         LDF_COLLIDE);
 560                                         next = pos;
 561                                         page = ll_get_dir_page(inode, pos,
 562                                                                &chain);
 563                                 } else {
 564                                         /*
 565                                          * go into overflow page.
 566                                          */
 567                                         LASSERT(le32_to_cpu(dp->ldp_flags) &
 568                                                 LDF_COLLIDE);
 569                                         ll_release_page(page, 1);
 570                                 }
 571                         } else {
 572                                 pos = hash;
 573                                 ll_release_page(page, 0);
 574                         }
 575                 } else {
 576                         rc = PTR_ERR(page);
 577                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 578                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 579                 }
 580         }
 581
 582         ctx->pos = pos;
 583         ll_dir_chain_fini(&chain);
 584         return rc;
 585 }
 586
 587 static int ll_readdir(struct file *filp, struct dir_context *ctx)
 588 {
 589         struct inode            *inode  = file_inode(filp);
 590         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 591         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 592         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 593         int                     api32   = ll_need_32bit_api(sbi);
 594         int                     rc;
 595
 596         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu 32bit_api %d\n",
 597                inode->i_ino, inode->i_generation,
 598                inode, (unsigned long)lfd->lfd_pos, i_size_read(inode), api32);
 599
 600         if (lfd->lfd_pos == MDS_DIR_END_OFF) {
 601                 /*
 602                  * end-of-file.
 603                  */
 604                 rc = 0;
 605                 goto out;
 606         }
 607
 608         ctx->pos = lfd->lfd_pos;
 609         rc = ll_dir_read(inode, ctx);
 610         lfd->lfd_pos = ctx->pos;
 611         if (ctx->pos == MDS_DIR_END_OFF) {
 612                 if (api32)
 613                         ctx->pos = LL_DIR_END_OFF_32BIT;
 614                 else
 615                         ctx->pos = LL_DIR_END_OFF;
 616         } else {
 617                 if (api32 && hash64)
 618                         ctx->pos >>= 32;
 619         }
 620         filp->f_version = inode->i_version;
 621
 622 out:
 623         if (!rc)
 624                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 625
 626         return rc;
 627 }
 628
 629 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 630 {
 631         struct mgs_send_param *msp;
 632         int rc = 0;
 633
 634         msp = kzalloc(sizeof(*msp), GFP_NOFS);
 635         if (!msp)
 636                 return -ENOMEM;
 637
 638         strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
 639         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 640                                 sizeof(struct mgs_send_param), msp, NULL);
 641         if (rc)
 642                 CERROR("Failed to set parameter: %d\n", rc);
 643         kfree(msp);
 644
 645         return rc;
 646 }
 647
 648 static int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
 649                                char *filename)
 650 {
 651         struct ptlrpc_request *request = NULL;
 652         struct md_op_data *op_data;
 653         struct ll_sb_info *sbi = ll_i2sbi(dir);
 654         int mode;
 655         int err;
 656
 657         mode = (~current_umask() & 0755) | S_IFDIR;
 658         op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
 659                                      strlen(filename), mode, LUSTRE_OPC_MKDIR,
 660                                      lump);
 661         if (IS_ERR(op_data)) {
 662                 err = PTR_ERR(op_data);
 663                 goto err_exit;
 664         }
 665
 666         op_data->op_cli_flags |= CLI_SET_MEA;
 667         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 668                         from_kuid(&init_user_ns, current_fsuid()),
 669                         from_kgid(&init_user_ns, current_fsgid()),
 670                         cfs_curproc_cap_pack(), 0, &request);
 671         ll_finish_md_op_data(op_data);
 672         if (err)
 673                 goto err_exit;
 674 err_exit:
 675         ptlrpc_req_finished(request);
 676         return err;
 677 }
 678
 679 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 680                      int set_default)
 681 {
 682         struct ll_sb_info *sbi = ll_i2sbi(inode);
 683         struct md_op_data *op_data;
 684         struct ptlrpc_request *req = NULL;
 685         int rc = 0;
 686         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 687         struct obd_device *mgc = lsi->lsi_mgc;
 688         int lum_size;
 689
 690         if (lump) {
 691                 /*
 692                  * This is coming from userspace, so should be in
 693                  * local endian.  But the MDS would like it in little
 694                  * endian, so we swab it before we send it.
 695                  */
 696                 switch (lump->lmm_magic) {
 697                 case LOV_USER_MAGIC_V1: {
 698                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 699                                 lustre_swab_lov_user_md_v1(lump);
 700                         lum_size = sizeof(struct lov_user_md_v1);
 701                         break;
 702                 }
 703                 case LOV_USER_MAGIC_V3: {
 704                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 705                                 lustre_swab_lov_user_md_v3(
 706                                         (struct lov_user_md_v3 *)lump);
 707                         lum_size = sizeof(struct lov_user_md_v3);
 708                         break;
 709                 }
 710                 default: {
 711                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
 712                                lump->lmm_magic, LOV_USER_MAGIC_V1,
 713                                LOV_USER_MAGIC_V3);
 714                         return -EINVAL;
 715                 }
 716                 }
 717         } else {
 718                 lum_size = sizeof(struct lov_user_md_v1);
 719         }
 720
 721         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 722                                      LUSTRE_OPC_ANY, NULL);
 723         if (IS_ERR(op_data))
 724                 return PTR_ERR(op_data);
 725
 726         if (lump && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC))
 727                 op_data->op_cli_flags |= CLI_SET_MEA;
 728
 729         /* swabbing is done in lov_setstripe() on server side */
 730         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 731                         NULL, 0, &req, NULL);
 732         ll_finish_md_op_data(op_data);
 733         ptlrpc_req_finished(req);
 734         if (rc) {
 735                 if (rc != -EPERM && rc != -EACCES)
 736                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 737         }
 738
 739         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 740          * LOV_USER_MAGIC_V3 have the same initial fields so we do not
 741          * need to make the distinction between the 2 versions
 742          */
 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
 815         lmmsize = body->eadatasize;
 816
 817         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 818             lmmsize == 0) {
 819                 rc = -ENODATA;
 820                 goto out;
 821         }
 822
 823         lmm = req_capsule_server_sized_get(&req->rq_pill,
 824                                            &RMF_MDT_MD, lmmsize);
 825
 826         /*
 827          * This is coming from the MDS, so is probably in
 828          * little endian.  We convert it to host endian before
 829          * passing it to userspace.
 830          */
 831         /* We don't swab objects for directories */
 832         switch (le32_to_cpu(lmm->lmm_magic)) {
 833         case LOV_MAGIC_V1:
 834                 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC)
 835                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 836                 break;
 837         case LOV_MAGIC_V3:
 838                 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC)
 839                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 840                 break;
 841         default:
 842                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 843                 rc = -EPROTO;
 844         }
 845 out:
 846         *lmmp = lmm;
 847         *lmm_size = lmmsize;
 848         *request = req;
 849         return rc;
 850 }
 851
 852 /*
 853  *  Get MDT index for the inode.
 854  */
 855 int ll_get_mdt_idx(struct inode *inode)
 856 {
 857         struct ll_sb_info *sbi = ll_i2sbi(inode);
 858         struct md_op_data *op_data;
 859         int rc, mdtidx;
 860
 861         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 862                                      0, LUSTRE_OPC_ANY, NULL);
 863         if (IS_ERR(op_data))
 864                 return PTR_ERR(op_data);
 865
 866         op_data->op_flags |= MF_GET_MDT_IDX;
 867         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 868         mdtidx = op_data->op_mds;
 869         ll_finish_md_op_data(op_data);
 870         if (rc < 0) {
 871                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 872                 return rc;
 873         }
 874         return mdtidx;
 875 }
 876
 877 /**
 878  * Generic handler to do any pre-copy work.
 879  *
 880  * It sends a first hsm_progress (with extent length == 0) to coordinator as a
 881  * first information for it that real work has started.
 882  *
 883  * Moreover, for a ARCHIVE request, it will sample the file data version and
 884  * store it in \a copy.
 885  *
 886  * \return 0 on success.
 887  */
 888 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 889 {
 890         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 891         struct hsm_progress_kernel       hpk;
 892         int                              rc;
 893
 894         /* Forge a hsm_progress based on data from copy. */
 895         hpk.hpk_fid = copy->hc_hai.hai_fid;
 896         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 897         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 898         hpk.hpk_extent.length = 0;
 899         hpk.hpk_flags = 0;
 900         hpk.hpk_errval = 0;
 901         hpk.hpk_data_version = 0;
 902
 903         /* For archive request, we need to read the current file version. */
 904         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 905                 struct inode    *inode;
 906                 __u64            data_version = 0;
 907
 908                 /* Get inode for this fid */
 909                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 910                 if (IS_ERR(inode)) {
 911                         hpk.hpk_flags |= HP_FLAG_RETRY;
 912                         /* hpk_errval is >= 0 */
 913                         hpk.hpk_errval = -PTR_ERR(inode);
 914                         rc = PTR_ERR(inode);
 915                         goto progress;
 916                 }
 917
 918                 /* Read current file data version */
 919                 rc = ll_data_version(inode, &data_version, 1);
 920                 iput(inode);
 921                 if (rc != 0) {
 922                         CDEBUG(D_HSM, "Could not read file data version of "
 923                                       DFID" (rc = %d). Archive request (%#llx) could not be done.\n",
 924                                       PFID(&copy->hc_hai.hai_fid), rc,
 925                                       copy->hc_hai.hai_cookie);
 926                         hpk.hpk_flags |= HP_FLAG_RETRY;
 927                         /* hpk_errval must be >= 0 */
 928                         hpk.hpk_errval = -rc;
 929                         goto progress;
 930                 }
 931
 932                 /* Store in the hsm_copy for later copytool use.
 933                  * Always modified even if no lsm.
 934                  */
 935                 copy->hc_data_version = data_version;
 936         }
 937
 938 progress:
 939         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 940                            &hpk, NULL);
 941
 942         return rc;
 943 }
 944
 945 /**
 946  * Generic handler to do any post-copy work.
 947  *
 948  * It will send the last hsm_progress update to coordinator to inform it
 949  * that copy is finished and whether it was successful or not.
 950  *
 951  * Moreover,
 952  * - for ARCHIVE request, it will sample the file data version and compare it
 953  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 954  *   will be considered as failed.
 955  * - for RESTORE request, it will sample the file data version and send it to
 956  *   coordinator which is useful if the file was imported as 'released'.
 957  *
 958  * \return 0 on success.
 959  */
 960 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 961 {
 962         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 963         struct hsm_progress_kernel       hpk;
 964         int                              rc;
 965
 966         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 967         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 968          * initialized if copy_end was called with copy == NULL.
 969          */
 970
 971         /* Forge a hsm_progress based on data from copy. */
 972         hpk.hpk_fid = copy->hc_hai.hai_fid;
 973         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 974         hpk.hpk_extent = copy->hc_hai.hai_extent;
 975         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 976         hpk.hpk_errval = copy->hc_errval;
 977         hpk.hpk_data_version = 0;
 978
 979         /* For archive request, we need to check the file data was not changed.
 980          *
 981          * For restore request, we need to send the file data version, this is
 982          * useful when the file was created using hsm_import.
 983          */
 984         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 985              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 986             (copy->hc_errval == 0)) {
 987                 struct inode    *inode;
 988                 __u64            data_version = 0;
 989
 990                 /* Get lsm for this fid */
 991                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 992                 if (IS_ERR(inode)) {
 993                         hpk.hpk_flags |= HP_FLAG_RETRY;
 994                         /* hpk_errval must be >= 0 */
 995                         hpk.hpk_errval = -PTR_ERR(inode);
 996                         rc = PTR_ERR(inode);
 997                         goto progress;
 998                 }
 999
1000                 rc = ll_data_version(inode, &data_version,
1001                                      copy->hc_hai.hai_action == HSMA_ARCHIVE);
1002                 iput(inode);
1003                 if (rc) {
1004                         CDEBUG(D_HSM, "Could not read file data version. Request could not be confirmed.\n");
1005                         if (hpk.hpk_errval == 0)
1006                                 hpk.hpk_errval = -rc;
1007                         goto progress;
1008                 }
1009
1010                 /* Store in the hsm_copy for later copytool use.
1011                  * Always modified even if no lsm.
1012                  */
1013                 hpk.hpk_data_version = data_version;
1014
1015                 /* File could have been stripped during archiving, so we need
1016                  * to check anyway.
1017                  */
1018                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1019                     (copy->hc_data_version != data_version)) {
1020                         CDEBUG(D_HSM, "File data version mismatched. File content was changed during archiving. "
1021                                DFID", start:%#llx current:%#llx\n",
1022                                PFID(&copy->hc_hai.hai_fid),
1023                                copy->hc_data_version, data_version);
1024                         /* File was changed, send error to cdt. Do not ask for
1025                          * retry because if a file is modified frequently,
1026                          * the cdt will loop on retried archive requests.
1027                          * The policy engine will ask for a new archive later
1028                          * when the file will not be modified for some tunable
1029                          * time
1030                          */
1031                         /* we do not notify caller */
1032                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
1033                         /* hpk_errval must be >= 0 */
1034                         hpk.hpk_errval = EBUSY;
1035                 }
1036
1037         }
1038
1039 progress:
1040         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1041                            &hpk, NULL);
1042
1043         return rc;
1044 }
1045
1046 static int copy_and_ioctl(int cmd, struct obd_export *exp,
1047                           const void __user *data, size_t size)
1048 {
1049         void *copy;
1050         int rc;
1051
1052         copy = kzalloc(size, GFP_NOFS);
1053         if (!copy)
1054                 return -ENOMEM;
1055
1056         if (copy_from_user(copy, data, size)) {
1057                 rc = -EFAULT;
1058                 goto out;
1059         }
1060
1061         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
1062 out:
1063         kfree(copy);
1064
1065         return rc;
1066 }
1067
1068 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1069 {
1070         int cmd = qctl->qc_cmd;
1071         int type = qctl->qc_type;
1072         int id = qctl->qc_id;
1073         int valid = qctl->qc_valid;
1074         int rc = 0;
1075
1076         switch (cmd) {
1077         case LUSTRE_Q_INVALIDATE:
1078         case LUSTRE_Q_FINVALIDATE:
1079         case Q_QUOTAON:
1080         case Q_QUOTAOFF:
1081         case Q_SETQUOTA:
1082         case Q_SETINFO:
1083                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1084                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1085                         return -EPERM;
1086                 break;
1087         case Q_GETQUOTA:
1088                 if (((type == USRQUOTA &&
1089                       !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
1090                      (type == GRPQUOTA &&
1091                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
1092                     (!capable(CFS_CAP_SYS_ADMIN) ||
1093                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
1094                         return -EPERM;
1095                 break;
1096         case Q_GETINFO:
1097                 break;
1098         default:
1099                 CERROR("unsupported quotactl op: %#x\n", cmd);
1100                 return -ENOTTY;
1101         }
1102
1103         if (valid != QC_GENERAL) {
1104                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1105                         return -EOPNOTSUPP;
1106
1107                 if (cmd == Q_GETINFO)
1108                         qctl->qc_cmd = Q_GETOINFO;
1109                 else if (cmd == Q_GETQUOTA)
1110                         qctl->qc_cmd = Q_GETOQUOTA;
1111                 else
1112                         return -EINVAL;
1113
1114                 switch (valid) {
1115                 case QC_MDTIDX:
1116                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1117                                            sizeof(*qctl), qctl, NULL);
1118                         break;
1119                 case QC_OSTIDX:
1120                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1121                                            sizeof(*qctl), qctl, NULL);
1122                         break;
1123                 case QC_UUID:
1124                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1125                                            sizeof(*qctl), qctl, NULL);
1126                         if (rc == -EAGAIN)
1127                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1128                                                    sbi->ll_dt_exp,
1129                                                    sizeof(*qctl), qctl, NULL);
1130                         break;
1131                 default:
1132                         rc = -EINVAL;
1133                         break;
1134                 }
1135
1136                 if (rc)
1137                         return rc;
1138
1139                 qctl->qc_cmd = cmd;
1140         } else {
1141                 struct obd_quotactl *oqctl;
1142
1143                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1144                 if (!oqctl)
1145                         return -ENOMEM;
1146
1147                 QCTL_COPY(oqctl, qctl);
1148                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1149                 if (rc) {
1150                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
1151                                 oqctl->qc_cmd = Q_QUOTAOFF;
1152                                 obd_quotactl(sbi->ll_md_exp, oqctl);
1153                         }
1154                         kfree(oqctl);
1155                         return rc;
1156                 }
1157                 /* If QIF_SPACE is not set, client should collect the
1158                  * space usage from OSSs by itself
1159                  */
1160                 if (cmd == Q_GETQUOTA &&
1161                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1162                     !oqctl->qc_dqblk.dqb_curspace) {
1163                         struct obd_quotactl *oqctl_tmp;
1164
1165                         oqctl_tmp = kzalloc(sizeof(*oqctl_tmp), GFP_NOFS);
1166                         if (!oqctl_tmp) {
1167                                 rc = -ENOMEM;
1168                                 goto out;
1169                         }
1170
1171                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1172                         oqctl_tmp->qc_id = oqctl->qc_id;
1173                         oqctl_tmp->qc_type = oqctl->qc_type;
1174
1175                         /* collect space usage from OSTs */
1176                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1177                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1178                         if (!rc || rc == -EREMOTEIO) {
1179                                 oqctl->qc_dqblk.dqb_curspace =
1180                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1181                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1182                         }
1183
1184                         /* collect space & inode usage from MDTs */
1185                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1186                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1187                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1188                         if (!rc || rc == -EREMOTEIO) {
1189                                 oqctl->qc_dqblk.dqb_curspace +=
1190                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1191                                 oqctl->qc_dqblk.dqb_curinodes =
1192                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1193                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1194                         } else {
1195                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1196                         }
1197
1198                         kfree(oqctl_tmp);
1199                 }
1200 out:
1201                 QCTL_COPY(qctl, oqctl);
1202                 kfree(oqctl);
1203         }
1204
1205         return rc;
1206 }
1207
1208 /* This function tries to get a single name component,
1209  * to send to the server. No actual path traversal involved,
1210  * so we limit to NAME_MAX
1211  */
1212 static char *ll_getname(const char __user *filename)
1213 {
1214         int ret = 0, len;
1215         char *tmp;
1216
1217         tmp = kzalloc(NAME_MAX + 1, GFP_KERNEL);
1218         if (!tmp)
1219                 return ERR_PTR(-ENOMEM);
1220
1221         len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1222         if (len < 0)
1223                 ret = len;
1224         else if (len == 0)
1225                 ret = -ENOENT;
1226         else if (len > NAME_MAX && tmp[NAME_MAX] != 0)
1227                 ret = -ENAMETOOLONG;
1228
1229         if (ret) {
1230                 kfree(tmp);
1231                 tmp =  ERR_PTR(ret);
1232         }
1233         return tmp;
1234 }
1235
1236 #define ll_putname(filename) kfree(filename)
1237
1238 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1239 {
1240         struct inode *inode = file_inode(file);
1241         struct ll_sb_info *sbi = ll_i2sbi(inode);
1242         struct obd_ioctl_data *data;
1243         int rc = 0;
1244
1245         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1246                inode->i_ino, inode->i_generation, inode, cmd);
1247
1248         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1249         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1250                 return -ENOTTY;
1251
1252         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1253         switch (cmd) {
1254         case FSFILT_IOC_GETFLAGS:
1255         case FSFILT_IOC_SETFLAGS:
1256                 return ll_iocontrol(inode, file, cmd, arg);
1257         case FSFILT_IOC_GETVERSION_OLD:
1258         case FSFILT_IOC_GETVERSION:
1259                 return put_user(inode->i_generation, (int __user *)arg);
1260         /* We need to special case any other ioctls we want to handle,
1261          * to send them to the MDS/OST as appropriate and to properly
1262          * network encode the arg field.
1263         case FSFILT_IOC_SETVERSION_OLD:
1264         case FSFILT_IOC_SETVERSION:
1265         */
1266         case LL_IOC_GET_MDTIDX: {
1267                 int mdtidx;
1268
1269                 mdtidx = ll_get_mdt_idx(inode);
1270                 if (mdtidx < 0)
1271                         return mdtidx;
1272
1273                 if (put_user((int)mdtidx, (int __user *)arg))
1274                         return -EFAULT;
1275
1276                 return 0;
1277         }
1278         case IOC_MDC_LOOKUP: {
1279                 struct ptlrpc_request *request = NULL;
1280                 int namelen, len = 0;
1281                 char *buf = NULL;
1282                 char *filename;
1283                 struct md_op_data *op_data;
1284
1285                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1286                 if (rc)
1287                         return rc;
1288                 data = (void *)buf;
1289
1290                 filename = data->ioc_inlbuf1;
1291                 namelen = strlen(filename);
1292
1293                 if (namelen < 1) {
1294                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1295                         rc = -EINVAL;
1296                         goto out_free;
1297                 }
1298
1299                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1300                                              0, LUSTRE_OPC_ANY, NULL);
1301                 if (IS_ERR(op_data)) {
1302                         rc = PTR_ERR(op_data);
1303                         goto out_free;
1304                 }
1305
1306                 op_data->op_valid = OBD_MD_FLID;
1307                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1308                 ll_finish_md_op_data(op_data);
1309                 if (rc < 0) {
1310                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1311                         goto out_free;
1312                 }
1313                 ptlrpc_req_finished(request);
1314 out_free:
1315                 obd_ioctl_freedata(buf, len);
1316                 return rc;
1317         }
1318         case LL_IOC_LMV_SETSTRIPE: {
1319                 struct lmv_user_md  *lum;
1320                 char            *buf = NULL;
1321                 char            *filename;
1322                 int              namelen = 0;
1323                 int              lumlen = 0;
1324                 int              len;
1325                 int              rc;
1326
1327                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1328                 if (rc)
1329                         return rc;
1330
1331                 data = (void *)buf;
1332                 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
1333                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0) {
1334                         rc = -EINVAL;
1335                         goto lmv_out_free;
1336                 }
1337
1338                 filename = data->ioc_inlbuf1;
1339                 namelen = data->ioc_inllen1;
1340
1341                 if (namelen < 1) {
1342                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1343                         rc = -EINVAL;
1344                         goto lmv_out_free;
1345                 }
1346                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1347                 lumlen = data->ioc_inllen2;
1348
1349                 if (lum->lum_magic != LMV_USER_MAGIC ||
1350                     lumlen != sizeof(*lum)) {
1351                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1352                                filename, lum->lum_magic, lumlen, -EFAULT);
1353                         rc = -EINVAL;
1354                         goto lmv_out_free;
1355                 }
1356
1357                 /**
1358                  * ll_dir_setdirstripe will be used to set dir stripe
1359                  *  mdc_create--->mdt_reint_create (with dirstripe)
1360                  */
1361                 rc = ll_dir_setdirstripe(inode, lum, filename);
1362 lmv_out_free:
1363                 obd_ioctl_freedata(buf, len);
1364                 return rc;
1365
1366         }
1367         case LL_IOC_LOV_SETSTRIPE: {
1368                 struct lov_user_md_v3 lumv3;
1369                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1370                 struct lov_user_md_v1 __user *lumv1p = (void __user *)arg;
1371                 struct lov_user_md_v3 __user *lumv3p = (void __user *)arg;
1372
1373                 int set_default = 0;
1374
1375                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1376                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1377                         sizeof(lumv3p->lmm_objects[0]));
1378                 /* first try with v1 which is smaller than v3 */
1379                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1380                         return -EFAULT;
1381
1382                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1383                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1384                                 return -EFAULT;
1385                 }
1386
1387                 if (is_root_inode(inode))
1388                         set_default = 1;
1389
1390                 /* in v1 and v3 cases lumv1 points to data */
1391                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1392
1393                 return rc;
1394         }
1395         case LL_IOC_LMV_GETSTRIPE: {
1396                 struct lmv_user_md __user *lump = (void __user *)arg;
1397                 struct lmv_user_md lum;
1398                 struct lmv_user_md *tmp;
1399                 int lum_size;
1400                 int rc = 0;
1401                 int mdtindex;
1402
1403                 if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
1404                         return -EFAULT;
1405
1406                 if (lum.lum_magic != LMV_MAGIC_V1)
1407                         return -EINVAL;
1408
1409                 lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
1410                 tmp = kzalloc(lum_size, GFP_NOFS);
1411                 if (!tmp) {
1412                         rc = -ENOMEM;
1413                         goto free_lmv;
1414                 }
1415
1416                 *tmp = lum;
1417                 tmp->lum_type = LMV_STRIPE_TYPE;
1418                 tmp->lum_stripe_count = 1;
1419                 mdtindex = ll_get_mdt_idx(inode);
1420                 if (mdtindex < 0) {
1421                         rc = -ENOMEM;
1422                         goto free_lmv;
1423                 }
1424
1425                 tmp->lum_stripe_offset = mdtindex;
1426                 tmp->lum_objects[0].lum_mds = mdtindex;
1427                 memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
1428                        sizeof(struct lu_fid));
1429                 if (copy_to_user((void __user *)arg, tmp, lum_size)) {
1430                         rc = -EFAULT;
1431                         goto free_lmv;
1432                 }
1433 free_lmv:
1434                 kfree(tmp);
1435                 return rc;
1436         }
1437         case LL_IOC_LOV_SWAP_LAYOUTS:
1438                 return -EPERM;
1439         case LL_IOC_OBD_STATFS:
1440                 return ll_obd_statfs(inode, (void __user *)arg);
1441         case LL_IOC_LOV_GETSTRIPE:
1442         case LL_IOC_MDC_GETINFO:
1443         case IOC_MDC_GETFILEINFO:
1444         case IOC_MDC_GETFILESTRIPE: {
1445                 struct ptlrpc_request *request = NULL;
1446                 struct lov_user_md __user *lump;
1447                 struct lov_mds_md *lmm = NULL;
1448                 struct mdt_body *body;
1449                 char *filename = NULL;
1450                 int lmmsize;
1451
1452                 if (cmd == IOC_MDC_GETFILEINFO ||
1453                     cmd == IOC_MDC_GETFILESTRIPE) {
1454                         filename = ll_getname((const char __user *)arg);
1455                         if (IS_ERR(filename))
1456                                 return PTR_ERR(filename);
1457
1458                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1459                                                       &lmmsize, &request);
1460                 } else {
1461                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1462                 }
1463
1464                 if (request) {
1465                         body = req_capsule_server_get(&request->rq_pill,
1466                                                       &RMF_MDT_BODY);
1467                         LASSERT(body);
1468                 } else {
1469                         goto out_req;
1470                 }
1471
1472                 if (rc < 0) {
1473                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1474                                                cmd == LL_IOC_MDC_GETINFO)) {
1475                                 rc = 0;
1476                                 goto skip_lmm;
1477                         } else
1478                                 goto out_req;
1479                 }
1480
1481                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1482                     cmd == LL_IOC_LOV_GETSTRIPE) {
1483                         lump = (struct lov_user_md __user *)arg;
1484                 } else {
1485                         struct lov_user_mds_data __user *lmdp;
1486
1487                         lmdp = (struct lov_user_mds_data __user *)arg;
1488                         lump = &lmdp->lmd_lmm;
1489                 }
1490                 if (copy_to_user(lump, lmm, lmmsize)) {
1491                         if (copy_to_user(lump, lmm, sizeof(*lump))) {
1492                                 rc = -EFAULT;
1493                                 goto out_req;
1494                         }
1495                         rc = -EOVERFLOW;
1496                 }
1497 skip_lmm:
1498                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1499                         struct lov_user_mds_data __user *lmdp;
1500                         lstat_t st = { 0 };
1501
1502                         st.st_dev     = inode->i_sb->s_dev;
1503                         st.st_mode    = body->mode;
1504                         st.st_nlink   = body->nlink;
1505                         st.st_uid     = body->uid;
1506                         st.st_gid     = body->gid;
1507                         st.st_rdev    = body->rdev;
1508                         st.st_size    = body->size;
1509                         st.st_blksize = PAGE_SIZE;
1510                         st.st_blocks  = body->blocks;
1511                         st.st_atime   = body->atime;
1512                         st.st_mtime   = body->mtime;
1513                         st.st_ctime   = body->ctime;
1514                         st.st_ino     = inode->i_ino;
1515
1516                         lmdp = (struct lov_user_mds_data __user *)arg;
1517                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st))) {
1518                                 rc = -EFAULT;
1519                                 goto out_req;
1520                         }
1521                 }
1522
1523 out_req:
1524                 ptlrpc_req_finished(request);
1525                 if (filename)
1526                         ll_putname(filename);
1527                 return rc;
1528         }
1529         case IOC_LOV_GETINFO: {
1530                 struct lov_user_mds_data __user *lumd;
1531                 struct lov_stripe_md *lsm;
1532                 struct lov_user_md __user *lum;
1533                 struct lov_mds_md *lmm;
1534                 int lmmsize;
1535                 lstat_t st;
1536
1537                 lumd = (struct lov_user_mds_data __user *)arg;
1538                 lum = &lumd->lmd_lmm;
1539
1540                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1541                 if (rc)
1542                         return rc;
1543
1544                 lmm = libcfs_kvzalloc(lmmsize, GFP_NOFS);
1545                 if (!lmm)
1546                         return -ENOMEM;
1547                 if (copy_from_user(lmm, lum, lmmsize)) {
1548                         rc = -EFAULT;
1549                         goto free_lmm;
1550                 }
1551
1552                 switch (lmm->lmm_magic) {
1553                 case LOV_USER_MAGIC_V1:
1554                         if (cpu_to_le32(LOV_USER_MAGIC_V1) == LOV_USER_MAGIC_V1)
1555                                 break;
1556                         /* swab objects first so that stripes num will be sane */
1557                         lustre_swab_lov_user_md_objects(
1558                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1559                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1560                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1561                         break;
1562                 case LOV_USER_MAGIC_V3:
1563                         if (cpu_to_le32(LOV_USER_MAGIC_V3) == LOV_USER_MAGIC_V3)
1564                                 break;
1565                         /* swab objects first so that stripes num will be sane */
1566                         lustre_swab_lov_user_md_objects(
1567                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1568                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1569                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1570                         break;
1571                 default:
1572                         rc = -EINVAL;
1573                         goto free_lmm;
1574                 }
1575
1576                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1577                 if (rc < 0) {
1578                         rc = -ENOMEM;
1579                         goto free_lmm;
1580                 }
1581
1582                 /* Perform glimpse_size operation. */
1583                 memset(&st, 0, sizeof(st));
1584
1585                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1586                 if (rc)
1587                         goto free_lsm;
1588
1589                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st))) {
1590                         rc = -EFAULT;
1591                         goto free_lsm;
1592                 }
1593
1594 free_lsm:
1595                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1596 free_lmm:
1597                 kvfree(lmm);
1598                 return rc;
1599         }
1600         case OBD_IOC_LLOG_CATINFO: {
1601                 return -EOPNOTSUPP;
1602         }
1603         case OBD_IOC_QUOTACHECK: {
1604                 struct obd_quotactl *oqctl;
1605                 int error = 0;
1606
1607                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1608                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1609                         return -EPERM;
1610
1611                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1612                 if (!oqctl)
1613                         return -ENOMEM;
1614                 oqctl->qc_type = arg;
1615                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1616                 if (rc < 0) {
1617                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1618                         error = rc;
1619                 }
1620
1621                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1622                 if (rc < 0)
1623                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1624
1625                 kfree(oqctl);
1626                 return error ?: rc;
1627         }
1628         case OBD_IOC_POLL_QUOTACHECK: {
1629                 struct if_quotacheck *check;
1630
1631                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1632                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1633                         return -EPERM;
1634
1635                 check = kzalloc(sizeof(*check), GFP_NOFS);
1636                 if (!check)
1637                         return -ENOMEM;
1638
1639                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1640                                    NULL);
1641                 if (rc) {
1642                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1643                         if (copy_to_user((void __user *)arg, check,
1644                                          sizeof(*check)))
1645                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1646                         goto out_poll;
1647                 }
1648
1649                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1650                                    NULL);
1651                 if (rc) {
1652                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1653                         if (copy_to_user((void __user *)arg, check,
1654                                          sizeof(*check)))
1655                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1656                         goto out_poll;
1657                 }
1658 out_poll:
1659                 kfree(check);
1660                 return rc;
1661         }
1662         case LL_IOC_QUOTACTL: {
1663                 struct if_quotactl *qctl;
1664
1665                 qctl = kzalloc(sizeof(*qctl), GFP_NOFS);
1666                 if (!qctl)
1667                         return -ENOMEM;
1668
1669                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl))) {
1670                         rc = -EFAULT;
1671                         goto out_quotactl;
1672                 }
1673
1674                 rc = quotactl_ioctl(sbi, qctl);
1675
1676                 if (rc == 0 && copy_to_user((void __user *)arg, qctl,
1677                                             sizeof(*qctl)))
1678                         rc = -EFAULT;
1679
1680 out_quotactl:
1681                 kfree(qctl);
1682                 return rc;
1683         }
1684         case OBD_IOC_GETDTNAME:
1685         case OBD_IOC_GETMDNAME:
1686                 return ll_get_obd_name(inode, cmd, arg);
1687         case LL_IOC_FLUSHCTX:
1688                 return ll_flush_ctx(inode);
1689 #ifdef CONFIG_FS_POSIX_ACL
1690         case LL_IOC_RMTACL: {
1691             if (sbi->ll_flags & LL_SBI_RMT_CLIENT && is_root_inode(inode)) {
1692                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1693
1694                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1695                 if (!rc)
1696                         fd->fd_flags |= LL_FILE_RMTACL;
1697                 return rc;
1698             } else
1699                 return 0;
1700         }
1701 #endif
1702         case LL_IOC_GETOBDCOUNT: {
1703                 int count, vallen;
1704                 struct obd_export *exp;
1705
1706                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1707                         return -EFAULT;
1708
1709                 /* get ost count when count is zero, get mdt count otherwise */
1710                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1711                 vallen = sizeof(count);
1712                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1713                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1714                 if (rc) {
1715                         CERROR("get target count failed: %d\n", rc);
1716                         return rc;
1717                 }
1718
1719                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1720                         return -EFAULT;
1721
1722                 return 0;
1723         }
1724         case LL_IOC_PATH2FID:
1725                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1726                                  sizeof(struct lu_fid)))
1727                         return -EFAULT;
1728                 return 0;
1729         case LL_IOC_GET_CONNECT_FLAGS: {
1730                 return obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1731                                      (void __user *)arg);
1732         }
1733         case OBD_IOC_CHANGELOG_SEND:
1734         case OBD_IOC_CHANGELOG_CLEAR:
1735                 if (!capable(CFS_CAP_SYS_ADMIN))
1736                         return -EPERM;
1737
1738                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1739                                     sizeof(struct ioc_changelog));
1740                 return rc;
1741         case OBD_IOC_FID2PATH:
1742                 return ll_fid2path(inode, (void __user *)arg);
1743         case LL_IOC_HSM_REQUEST: {
1744                 struct hsm_user_request *hur;
1745                 ssize_t                  totalsize;
1746
1747                 hur = memdup_user((void __user *)arg, sizeof(*hur));
1748                 if (IS_ERR(hur))
1749                         return PTR_ERR(hur);
1750
1751                 /* Compute the whole struct size */
1752                 totalsize = hur_len(hur);
1753                 kfree(hur);
1754                 if (totalsize < 0)
1755                         return -E2BIG;
1756
1757                 /* Final size will be more than double totalsize */
1758                 if (totalsize >= MDS_MAXREQSIZE / 3)
1759                         return -E2BIG;
1760
1761                 hur = libcfs_kvzalloc(totalsize, GFP_NOFS);
1762                 if (!hur)
1763                         return -ENOMEM;
1764
1765                 /* Copy the whole struct */
1766                 if (copy_from_user(hur, (void __user *)arg, totalsize)) {
1767                         kvfree(hur);
1768                         return -EFAULT;
1769                 }
1770
1771                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1772                         const struct lu_fid *fid;
1773                         struct inode *f;
1774                         int i;
1775
1776                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1777                                 fid = &hur->hur_user_item[i].hui_fid;
1778                                 f = search_inode_for_lustre(inode->i_sb, fid);
1779                                 if (IS_ERR(f)) {
1780                                         rc = PTR_ERR(f);
1781                                         break;
1782                                 }
1783
1784                                 rc = ll_hsm_release(f);
1785                                 iput(f);
1786                                 if (rc != 0)
1787                                         break;
1788                         }
1789                 } else {
1790                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1791                                            hur, NULL);
1792                 }
1793
1794                 kvfree(hur);
1795
1796                 return rc;
1797         }
1798         case LL_IOC_HSM_PROGRESS: {
1799                 struct hsm_progress_kernel      hpk;
1800                 struct hsm_progress             hp;
1801
1802                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1803                         return -EFAULT;
1804
1805                 hpk.hpk_fid = hp.hp_fid;
1806                 hpk.hpk_cookie = hp.hp_cookie;
1807                 hpk.hpk_extent = hp.hp_extent;
1808                 hpk.hpk_flags = hp.hp_flags;
1809                 hpk.hpk_errval = hp.hp_errval;
1810                 hpk.hpk_data_version = 0;
1811
1812                 /* File may not exist in Lustre; all progress
1813                  * reported to Lustre root
1814                  */
1815                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1816                                    NULL);
1817                 return rc;
1818         }
1819         case LL_IOC_HSM_CT_START:
1820                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1821                                     sizeof(struct lustre_kernelcomm));
1822                 return rc;
1823
1824         case LL_IOC_HSM_COPY_START: {
1825                 struct hsm_copy *copy;
1826                 int              rc;
1827
1828                 copy = memdup_user((char __user *)arg, sizeof(*copy));
1829                 if (IS_ERR(copy))
1830                         return PTR_ERR(copy);
1831
1832                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1833                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1834                         rc = -EFAULT;
1835
1836                 kfree(copy);
1837                 return rc;
1838         }
1839         case LL_IOC_HSM_COPY_END: {
1840                 struct hsm_copy *copy;
1841                 int              rc;
1842
1843                 copy = memdup_user((char __user *)arg, sizeof(*copy));
1844                 if (IS_ERR(copy))
1845                         return PTR_ERR(copy);
1846
1847                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1848                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1849                         rc = -EFAULT;
1850
1851                 kfree(copy);
1852                 return rc;
1853         }
1854         default:
1855                 return obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1856                                      (void __user *)arg);
1857         }
1858 }
1859
1860 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1861 {
1862         struct inode *inode = file->f_mapping->host;
1863         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1864         struct ll_sb_info *sbi = ll_i2sbi(inode);
1865         int api32 = ll_need_32bit_api(sbi);
1866         loff_t ret = -EINVAL;
1867
1868         inode_lock(inode);
1869         switch (origin) {
1870         case SEEK_SET:
1871                 break;
1872         case SEEK_CUR:
1873                 offset += file->f_pos;
1874                 break;
1875         case SEEK_END:
1876                 if (offset > 0)
1877                         goto out;
1878                 if (api32)
1879                         offset += LL_DIR_END_OFF_32BIT;
1880                 else
1881                         offset += LL_DIR_END_OFF;
1882                 break;
1883         default:
1884                 goto out;
1885         }
1886
1887         if (offset >= 0 &&
1888             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1889              (!api32 && offset <= LL_DIR_END_OFF))) {
1890                 if (offset != file->f_pos) {
1891                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1892                             (!api32 && offset == LL_DIR_END_OFF))
1893                                 fd->lfd_pos = MDS_DIR_END_OFF;
1894                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1895                                 fd->lfd_pos = offset << 32;
1896                         else
1897                                 fd->lfd_pos = offset;
1898                         file->f_pos = offset;
1899                         file->f_version = 0;
1900                 }
1901                 ret = offset;
1902         }
1903         goto out;
1904
1905 out:
1906         inode_unlock(inode);
1907         return ret;
1908 }
1909
1910 static int ll_dir_open(struct inode *inode, struct file *file)
1911 {
1912         return ll_file_open(inode, file);
1913 }
1914
1915 static int ll_dir_release(struct inode *inode, struct file *file)
1916 {
1917         return ll_file_release(inode, file);
1918 }
1919
1920 const struct file_operations ll_dir_operations = {
1921         .llseek   = ll_dir_seek,
1922         .open     = ll_dir_open,
1923         .release  = ll_dir_release,
1924         .read     = generic_read_dir,
1925         .iterate  = ll_readdir,
1926         .unlocked_ioctl   = ll_dir_ioctl,
1927         .fsync    = ll_fsync,
1928 };