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