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 <asm/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                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 188                 /* Checked by mdc_readpage() */
 189                 LASSERT(body != NULL);
 190
 191                 if (body->valid & OBD_MD_FLSIZE)
 192                         cl_isize_write(inode, body->size);
 193
 194                 nrdpgs = (request->rq_bulk->bd_nob_transferred+PAGE_CACHE_SIZE-1)
 195                          >> PAGE_CACHE_SHIFT;
 196                 SetPageUptodate(page0);
 197         }
 198         unlock_page(page0);
 199         ptlrpc_req_finished(request);
 200
 201         CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages);
 202
 203         ll_pagevec_init(&lru_pvec, 0);
 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                         page_cache_release(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_KERNEL);
 226                 if (ret == 0) {
 227                         unlock_page(page);
 228                         if (ll_pagevec_add(&lru_pvec, page) == 0)
 229                                 ll_pagevec_lru_add_file(&lru_pvec);
 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         ll_pagevec_lru_add_file(&lru_pvec);
 237
 238         if (page_pool != &page0)
 239                 OBD_FREE(page_pool, sizeof(struct page *) * max_pages);
 240         return rc;
 241 }
 242
 243 static void ll_check_page(struct inode *dir, struct page *page)
 244 {
 245         /* XXX: check page format later */
 246         SetPageChecked(page);
 247 }
 248
 249 void ll_release_page(struct page *page, int remove)
 250 {
 251         kunmap(page);
 252         if (remove) {
 253                 lock_page(page);
 254                 if (likely(page->mapping != NULL))
 255                         truncate_complete_page(page->mapping, page);
 256                 unlock_page(page);
 257         }
 258         page_cache_release(page);
 259 }
 260
 261 /*
 262  * Find, kmap and return page that contains given hash.
 263  */
 264 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
 265                                        __u64 *start, __u64 *end)
 266 {
 267         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 268         struct address_space *mapping = dir->i_mapping;
 269         /*
 270          * Complement of hash is used as an index so that
 271          * radix_tree_gang_lookup() can be used to find a page with starting
 272          * hash _smaller_ than one we are looking for.
 273          */
 274         unsigned long offset = hash_x_index(*hash, hash64);
 275         struct page *page;
 276         int found;
 277
 278         spin_lock_irq(&mapping->tree_lock);
 279         found = radix_tree_gang_lookup(&mapping->page_tree,
 280                                        (void **)&page, offset, 1);
 281         if (found > 0) {
 282                 struct lu_dirpage *dp;
 283
 284                 page_cache_get(page);
 285                 spin_unlock_irq(&mapping->tree_lock);
 286                 /*
 287                  * In contrast to find_lock_page() we are sure that directory
 288                  * page cannot be truncated (while DLM lock is held) and,
 289                  * hence, can avoid restart.
 290                  *
 291                  * In fact, page cannot be locked here at all, because
 292                  * ll_dir_filler() does synchronous io.
 293                  */
 294                 wait_on_page_locked(page);
 295                 if (PageUptodate(page)) {
 296                         dp = kmap(page);
 297                         if (BITS_PER_LONG == 32 && hash64) {
 298                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 299                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 300                                 *hash  = *hash >> 32;
 301                         } else {
 302                                 *start = le64_to_cpu(dp->ldp_hash_start);
 303                                 *end   = le64_to_cpu(dp->ldp_hash_end);
 304                         }
 305                         LASSERTF(*start <= *hash, "start = %#llx,end = %#llx,hash = %#llx\n",
 306                                  *start, *end, *hash);
 307                         CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash %llu\n",
 308                                offset, *start, *end, *hash);
 309                         if (*hash > *end) {
 310                                 ll_release_page(page, 0);
 311                                 page = NULL;
 312                         } else if (*end != *start && *hash == *end) {
 313                                 /*
 314                                  * upon hash collision, remove this page,
 315                                  * otherwise put page reference, and
 316                                  * ll_get_dir_page() will issue RPC to fetch
 317                                  * the page we want.
 318                                  */
 319                                 ll_release_page(page,
 320                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
 321                                 page = NULL;
 322                         }
 323                 } else {
 324                         page_cache_release(page);
 325                         page = ERR_PTR(-EIO);
 326                 }
 327
 328         } else {
 329                 spin_unlock_irq(&mapping->tree_lock);
 330                 page = NULL;
 331         }
 332         return page;
 333 }
 334
 335 struct page *ll_get_dir_page(struct inode *dir, __u64 hash,
 336                              struct ll_dir_chain *chain)
 337 {
 338         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 339         struct address_space *mapping = dir->i_mapping;
 340         struct lustre_handle lockh;
 341         struct lu_dirpage *dp;
 342         struct page *page;
 343         ldlm_mode_t mode;
 344         int rc;
 345         __u64 start = 0;
 346         __u64 end = 0;
 347         __u64 lhash = hash;
 348         struct ll_inode_info *lli = ll_i2info(dir);
 349         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 350
 351         mode = LCK_PR;
 352         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 353                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
 354         if (!rc) {
 355                 struct ldlm_enqueue_info einfo = {
 356                         .ei_type = LDLM_IBITS,
 357                         .ei_mode = mode,
 358                         .ei_cb_bl = ll_md_blocking_ast,
 359                         .ei_cb_cp = ldlm_completion_ast,
 360                 };
 361                 struct lookup_intent it = { .it_op = IT_READDIR };
 362                 struct ptlrpc_request *request;
 363                 struct md_op_data *op_data;
 364
 365                 op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0,
 366                 LUSTRE_OPC_ANY, NULL);
 367                 if (IS_ERR(op_data))
 368                         return (void *)op_data;
 369
 370                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
 371                                 op_data, &lockh, NULL, 0, NULL, 0);
 372
 373                 ll_finish_md_op_data(op_data);
 374
 375                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 376                 if (request)
 377                         ptlrpc_req_finished(request);
 378                 if (rc < 0) {
 379                         CERROR("lock enqueue: "DFID" at %llu: rc %d\n",
 380                                 PFID(ll_inode2fid(dir)), hash, rc);
 381                         return ERR_PTR(rc);
 382                 }
 383
 384                 CDEBUG(D_INODE, "setting lr_lvb_inode to inode %p (%lu/%u)\n",
 385                        dir, dir->i_ino, dir->i_generation);
 386                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp,
 387                                  &it.d.lustre.it_lock_handle, dir, NULL);
 388         } else {
 389                 /* for cross-ref object, l_ast_data of the lock may not be set,
 390                  * we reset it here */
 391                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
 392                                  dir, NULL);
 393         }
 394         ldlm_lock_dump_handle(D_OTHER, &lockh);
 395
 396         mutex_lock(&lli->lli_readdir_mutex);
 397         page = ll_dir_page_locate(dir, &lhash, &start, &end);
 398         if (IS_ERR(page)) {
 399                 CERROR("dir page locate: "DFID" at %llu: rc %ld\n",
 400                        PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
 401                 goto out_unlock;
 402         } else if (page != NULL) {
 403                 /*
 404                  * XXX nikita: not entirely correct handling of a corner case:
 405                  * suppose hash chain of entries with hash value HASH crosses
 406                  * border between pages P0 and P1. First both P0 and P1 are
 407                  * cached, seekdir() is called for some entry from the P0 part
 408                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 409                  * happens and finds P1, as it starts with matching hash
 410                  * value. Remaining entries from P0 part of the chain are
 411                  * skipped. (Is that really a bug?)
 412                  *
 413                  * Possible solutions: 0. don't cache P1 is such case, handle
 414                  * it as an "overflow" page. 1. invalidate all pages at
 415                  * once. 2. use HASH|1 as an index for P1.
 416                  */
 417                 goto hash_collision;
 418         }
 419
 420         page = read_cache_page(mapping, hash_x_index(hash, hash64),
 421                                ll_dir_filler, &lhash);
 422         if (IS_ERR(page)) {
 423                 CERROR("read cache page: "DFID" at %llu: rc %ld\n",
 424                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 425                 goto out_unlock;
 426         }
 427
 428         wait_on_page_locked(page);
 429         (void)kmap(page);
 430         if (!PageUptodate(page)) {
 431                 CERROR("page not updated: "DFID" at %llu: rc %d\n",
 432                        PFID(ll_inode2fid(dir)), hash, -5);
 433                 goto fail;
 434         }
 435         if (!PageChecked(page))
 436                 ll_check_page(dir, page);
 437         if (PageError(page)) {
 438                 CERROR("page error: "DFID" at %llu: rc %d\n",
 439                        PFID(ll_inode2fid(dir)), hash, -5);
 440                 goto fail;
 441         }
 442 hash_collision:
 443         dp = page_address(page);
 444         if (BITS_PER_LONG == 32 && hash64) {
 445                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 446                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 447                 lhash = hash >> 32;
 448         } else {
 449                 start = le64_to_cpu(dp->ldp_hash_start);
 450                 end   = le64_to_cpu(dp->ldp_hash_end);
 451                 lhash = hash;
 452         }
 453         if (end == start) {
 454                 LASSERT(start == lhash);
 455                 CWARN("Page-wide hash collision: %llu\n", end);
 456                 if (BITS_PER_LONG == 32 && hash64)
 457                         CWARN("Real page-wide hash collision at [%llu %llu] with hash %llu\n",
 458                               le64_to_cpu(dp->ldp_hash_start),
 459                               le64_to_cpu(dp->ldp_hash_end), hash);
 460                 /*
 461                  * Fetch whole overflow chain...
 462                  *
 463                  * XXX not yet.
 464                  */
 465                 goto fail;
 466         }
 467 out_unlock:
 468         mutex_unlock(&lli->lli_readdir_mutex);
 469         ldlm_lock_decref(&lockh, mode);
 470         return page;
 471
 472 fail:
 473         ll_release_page(page, 1);
 474         page = ERR_PTR(-EIO);
 475         goto out_unlock;
 476 }
 477
 478 int ll_dir_read(struct inode *inode, struct dir_context *ctx)
 479 {
 480         struct ll_inode_info *info       = ll_i2info(inode);
 481         struct ll_sb_info    *sbi       = ll_i2sbi(inode);
 482         __u64              pos          = ctx->pos;
 483         int                api32      = ll_need_32bit_api(sbi);
 484         int                hash64     = sbi->ll_flags & LL_SBI_64BIT_HASH;
 485         struct page       *page;
 486         struct ll_dir_chain   chain;
 487         int                done = 0;
 488         int                rc = 0;
 489
 490         ll_dir_chain_init(&chain);
 491
 492         page = ll_get_dir_page(inode, pos, &chain);
 493
 494         while (rc == 0 && !done) {
 495                 struct lu_dirpage *dp;
 496                 struct lu_dirent  *ent;
 497
 498                 if (!IS_ERR(page)) {
 499                         /*
 500                          * If page is empty (end of directory is reached),
 501                          * use this value.
 502                          */
 503                         __u64 hash = MDS_DIR_END_OFF;
 504                         __u64 next;
 505
 506                         dp = page_address(page);
 507                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 508                              ent = lu_dirent_next(ent)) {
 509                                 __u16     type;
 510                                 int         namelen;
 511                                 struct lu_fid  fid;
 512                                 __u64     lhash;
 513                                 __u64     ino;
 514
 515                                 /*
 516                                  * XXX: implement correct swabbing here.
 517                                  */
 518
 519                                 hash = le64_to_cpu(ent->lde_hash);
 520                                 if (hash < pos)
 521                                         /*
 522                                          * Skip until we find target hash
 523                                          * value.
 524                                          */
 525                                         continue;
 526
 527                                 namelen = le16_to_cpu(ent->lde_namelen);
 528                                 if (namelen == 0)
 529                                         /*
 530                                          * Skip dummy record.
 531                                          */
 532                                         continue;
 533
 534                                 if (api32 && hash64)
 535                                         lhash = hash >> 32;
 536                                 else
 537                                         lhash = hash;
 538                                 fid_le_to_cpu(&fid, &ent->lde_fid);
 539                                 ino = cl_fid_build_ino(&fid, api32);
 540                                 type = ll_dirent_type_get(ent);
 541                                 ctx->pos = lhash;
 542                                 /* For 'll_nfs_get_name_filldir()', it will try
 543                                  * to access the 'ent' through its 'lde_name',
 544                                  * so the parameter 'name' for 'ctx->actor()'
 545                                  * must be part of the 'ent'.
 546                                  */
 547                                 done = !dir_emit(ctx, ent->lde_name,
 548                                                  namelen, ino, type);
 549                         }
 550                         next = le64_to_cpu(dp->ldp_hash_end);
 551                         if (!done) {
 552                                 pos = next;
 553                                 if (pos == MDS_DIR_END_OFF) {
 554                                         /*
 555                                          * End of directory reached.
 556                                          */
 557                                         done = 1;
 558                                         ll_release_page(page, 0);
 559                                 } else if (1 /* chain is exhausted*/) {
 560                                         /*
 561                                          * Normal case: continue to the next
 562                                          * page.
 563                                          */
 564                                         ll_release_page(page,
 565                                             le32_to_cpu(dp->ldp_flags) &
 566                                                         LDF_COLLIDE);
 567                                         next = pos;
 568                                         page = ll_get_dir_page(inode, pos,
 569                                                                &chain);
 570                                 } else {
 571                                         /*
 572                                          * go into overflow page.
 573                                          */
 574                                         LASSERT(le32_to_cpu(dp->ldp_flags) &
 575                                                 LDF_COLLIDE);
 576                                         ll_release_page(page, 1);
 577                                 }
 578                         } else {
 579                                 pos = hash;
 580                                 ll_release_page(page, 0);
 581                         }
 582                 } else {
 583                         rc = PTR_ERR(page);
 584                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 585                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 586                 }
 587         }
 588
 589         ctx->pos = pos;
 590         ll_dir_chain_fini(&chain);
 591         return rc;
 592 }
 593
 594 static int ll_readdir(struct file *filp, struct dir_context *ctx)
 595 {
 596         struct inode            *inode  = file_inode(filp);
 597         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 598         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 599         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 600         int                     api32   = ll_need_32bit_api(sbi);
 601         int                     rc;
 602
 603         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu 32bit_api %d\n",
 604                inode->i_ino, inode->i_generation,
 605                inode, (unsigned long)lfd->lfd_pos, i_size_read(inode), api32);
 606
 607         if (lfd->lfd_pos == MDS_DIR_END_OFF) {
 608                 /*
 609                  * end-of-file.
 610                  */
 611                 rc = 0;
 612                 goto out;
 613         }
 614
 615         ctx->pos = lfd->lfd_pos;
 616         rc = ll_dir_read(inode, ctx);
 617         lfd->lfd_pos = ctx->pos;
 618         if (ctx->pos == MDS_DIR_END_OFF) {
 619                 if (api32)
 620                         ctx->pos = LL_DIR_END_OFF_32BIT;
 621                 else
 622                         ctx->pos = LL_DIR_END_OFF;
 623         } else {
 624                 if (api32 && hash64)
 625                         ctx->pos >>= 32;
 626         }
 627         filp->f_version = inode->i_version;
 628
 629 out:
 630         if (!rc)
 631                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 632
 633         return rc;
 634 }
 635
 636 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 637 {
 638         struct mgs_send_param *msp;
 639         int rc = 0;
 640
 641         msp = kzalloc(sizeof(*msp), GFP_NOFS);
 642         if (!msp)
 643                 return -ENOMEM;
 644
 645         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 646         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 647                                 sizeof(struct mgs_send_param), msp, NULL);
 648         if (rc)
 649                 CERROR("Failed to set parameter: %d\n", rc);
 650         OBD_FREE_PTR(msp);
 651
 652         return rc;
 653 }
 654
 655 int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
 656                         char *filename)
 657 {
 658         struct ptlrpc_request *request = NULL;
 659         struct md_op_data *op_data;
 660         struct ll_sb_info *sbi = ll_i2sbi(dir);
 661         int mode;
 662         int err;
 663
 664         mode = (0755 & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
 665         op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
 666                                      strlen(filename), mode, LUSTRE_OPC_MKDIR,
 667                                      lump);
 668         if (IS_ERR(op_data)) {
 669                 err = PTR_ERR(op_data);
 670                 goto err_exit;
 671         }
 672
 673         op_data->op_cli_flags |= CLI_SET_MEA;
 674         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 675                         from_kuid(&init_user_ns, current_fsuid()),
 676                         from_kgid(&init_user_ns, current_fsgid()),
 677                         cfs_curproc_cap_pack(), 0, &request);
 678         ll_finish_md_op_data(op_data);
 679         if (err)
 680                 goto err_exit;
 681 err_exit:
 682         ptlrpc_req_finished(request);
 683         return err;
 684 }
 685
 686 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 687                      int set_default)
 688 {
 689         struct ll_sb_info *sbi = ll_i2sbi(inode);
 690         struct md_op_data *op_data;
 691         struct ptlrpc_request *req = NULL;
 692         int rc = 0;
 693         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 694         struct obd_device *mgc = lsi->lsi_mgc;
 695         int lum_size;
 696
 697         if (lump != NULL) {
 698                 /*
 699                  * This is coming from userspace, so should be in
 700                  * local endian.  But the MDS would like it in little
 701                  * endian, so we swab it before we send it.
 702                  */
 703                 switch (lump->lmm_magic) {
 704                 case LOV_USER_MAGIC_V1: {
 705                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 706                                 lustre_swab_lov_user_md_v1(lump);
 707                         lum_size = sizeof(struct lov_user_md_v1);
 708                         break;
 709                 }
 710                 case LOV_USER_MAGIC_V3: {
 711                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 712                                 lustre_swab_lov_user_md_v3(
 713                                         (struct lov_user_md_v3 *)lump);
 714                         lum_size = sizeof(struct lov_user_md_v3);
 715                         break;
 716                 }
 717                 default: {
 718                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
 719                                lump->lmm_magic, LOV_USER_MAGIC_V1,
 720                                LOV_USER_MAGIC_V3);
 721                         return -EINVAL;
 722                 }
 723                 }
 724         } else {
 725                 lum_size = sizeof(struct lov_user_md_v1);
 726         }
 727
 728         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 729                                      LUSTRE_OPC_ANY, NULL);
 730         if (IS_ERR(op_data))
 731                 return PTR_ERR(op_data);
 732
 733         if (lump != NULL && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC))
 734                 op_data->op_cli_flags |= CLI_SET_MEA;
 735
 736         /* swabbing is done in lov_setstripe() on server side */
 737         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 738                         NULL, 0, &req, NULL);
 739         ll_finish_md_op_data(op_data);
 740         ptlrpc_req_finished(req);
 741         if (rc) {
 742                 if (rc != -EPERM && rc != -EACCES)
 743                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 744         }
 745
 746         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 747          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 748          need to make the distinction between the 2 versions */
 749         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 750                 char *param = NULL;
 751                 char *buf;
 752
 753                 param = kzalloc(MGS_PARAM_MAXLEN, GFP_NOFS);
 754                 if (!param) {
 755                         rc = -ENOMEM;
 756                         goto end;
 757                 }
 758
 759                 buf = param;
 760                 /* Get fsname and assume devname to be -MDT0000. */
 761                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 762                 strcat(buf, "-MDT0000.lov");
 763                 buf += strlen(buf);
 764
 765                 /* Set root stripesize */
 766                 sprintf(buf, ".stripesize=%u",
 767                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 768                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 769                 if (rc)
 770                         goto end;
 771
 772                 /* Set root stripecount */
 773                 sprintf(buf, ".stripecount=%hd",
 774                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 775                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 776                 if (rc)
 777                         goto end;
 778
 779                 /* Set root stripeoffset */
 780                 sprintf(buf, ".stripeoffset=%hd",
 781                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 782                         (typeof(lump->lmm_stripe_offset))(-1));
 783                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 784
 785 end:
 786                 if (param != NULL)
 787                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 788         }
 789         return rc;
 790 }
 791
 792 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 793                      int *lmm_size, struct ptlrpc_request **request)
 794 {
 795         struct ll_sb_info *sbi = ll_i2sbi(inode);
 796         struct mdt_body   *body;
 797         struct lov_mds_md *lmm = NULL;
 798         struct ptlrpc_request *req = NULL;
 799         int rc, lmmsize;
 800         struct md_op_data *op_data;
 801
 802         rc = ll_get_default_mdsize(sbi, &lmmsize);
 803         if (rc)
 804                 return rc;
 805
 806         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 807                                      0, lmmsize, LUSTRE_OPC_ANY,
 808                                      NULL);
 809         if (IS_ERR(op_data))
 810                 return PTR_ERR(op_data);
 811
 812         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 813         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 814         ll_finish_md_op_data(op_data);
 815         if (rc < 0) {
 816                 CDEBUG(D_INFO, "md_getattr failed on inode %lu/%u: rc %d\n",
 817                        inode->i_ino,
 818                        inode->i_generation, rc);
 819                 goto out;
 820         }
 821
 822         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 823         LASSERT(body != NULL);
 824
 825         lmmsize = body->eadatasize;
 826
 827         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 828             lmmsize == 0) {
 829                 rc = -ENODATA;
 830                 goto out;
 831         }
 832
 833         lmm = req_capsule_server_sized_get(&req->rq_pill,
 834                                            &RMF_MDT_MD, lmmsize);
 835         LASSERT(lmm != NULL);
 836
 837         /*
 838          * This is coming from the MDS, so is probably in
 839          * little endian.  We convert it to host endian before
 840          * passing it to userspace.
 841          */
 842         /* We don't swab objects for directories */
 843         switch (le32_to_cpu(lmm->lmm_magic)) {
 844         case LOV_MAGIC_V1:
 845                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 846                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 847                 break;
 848         case LOV_MAGIC_V3:
 849                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 850                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 851                 break;
 852         default:
 853                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 854                 rc = -EPROTO;
 855         }
 856 out:
 857         *lmmp = lmm;
 858         *lmm_size = lmmsize;
 859         *request = req;
 860         return rc;
 861 }
 862
 863 /*
 864  *  Get MDT index for the inode.
 865  */
 866 int ll_get_mdt_idx(struct inode *inode)
 867 {
 868         struct ll_sb_info *sbi = ll_i2sbi(inode);
 869         struct md_op_data *op_data;
 870         int rc, mdtidx;
 871
 872         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 873                                      0, LUSTRE_OPC_ANY, NULL);
 874         if (IS_ERR(op_data))
 875                 return PTR_ERR(op_data);
 876
 877         op_data->op_flags |= MF_GET_MDT_IDX;
 878         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 879         mdtidx = op_data->op_mds;
 880         ll_finish_md_op_data(op_data);
 881         if (rc < 0) {
 882                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 883                 return rc;
 884         }
 885         return mdtidx;
 886 }
 887
 888 /**
 889  * Generic handler to do any pre-copy work.
 890  *
 891  * It send a first hsm_progress (with extent length == 0) to coordinator as a
 892  * first information for it that real work has started.
 893  *
 894  * Moreover, for a ARCHIVE request, it will sample the file data version and
 895  * store it in \a copy.
 896  *
 897  * \return 0 on success.
 898  */
 899 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 900 {
 901         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 902         struct hsm_progress_kernel       hpk;
 903         int                              rc;
 904
 905         /* Forge a hsm_progress based on data from copy. */
 906         hpk.hpk_fid = copy->hc_hai.hai_fid;
 907         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 908         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 909         hpk.hpk_extent.length = 0;
 910         hpk.hpk_flags = 0;
 911         hpk.hpk_errval = 0;
 912         hpk.hpk_data_version = 0;
 913
 914
 915         /* For archive request, we need to read the current file version. */
 916         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 917                 struct inode    *inode;
 918                 __u64            data_version = 0;
 919
 920                 /* Get inode for this fid */
 921                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 922                 if (IS_ERR(inode)) {
 923                         hpk.hpk_flags |= HP_FLAG_RETRY;
 924                         /* hpk_errval is >= 0 */
 925                         hpk.hpk_errval = -PTR_ERR(inode);
 926                         rc = PTR_ERR(inode);
 927                         goto progress;
 928                 }
 929
 930                 /* Read current file data version */
 931                 rc = ll_data_version(inode, &data_version, 1);
 932                 iput(inode);
 933                 if (rc != 0) {
 934                         CDEBUG(D_HSM, "Could not read file data version of "
 935                                       DFID" (rc = %d). Archive request (%#llx) could not be done.\n",
 936                                       PFID(&copy->hc_hai.hai_fid), rc,
 937                                       copy->hc_hai.hai_cookie);
 938                         hpk.hpk_flags |= HP_FLAG_RETRY;
 939                         /* hpk_errval must be >= 0 */
 940                         hpk.hpk_errval = -rc;
 941                         goto progress;
 942                 }
 943
 944                 /* Store it the hsm_copy for later copytool use.
 945                  * Always modified even if no lsm. */
 946                 copy->hc_data_version = data_version;
 947         }
 948
 949 progress:
 950         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 951                            &hpk, NULL);
 952
 953         return rc;
 954 }
 955
 956 /**
 957  * Generic handler to do any post-copy work.
 958  *
 959  * It will send the last hsm_progress update to coordinator to inform it
 960  * that copy is finished and whether it was successful or not.
 961  *
 962  * Moreover,
 963  * - for ARCHIVE request, it will sample the file data version and compare it
 964  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 965  *   will be considered as failed.
 966  * - for RESTORE request, it will sample the file data version and send it to
 967  *   coordinator which is useful if the file was imported as 'released'.
 968  *
 969  * \return 0 on success.
 970  */
 971 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 972 {
 973         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 974         struct hsm_progress_kernel       hpk;
 975         int                              rc;
 976
 977         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 978         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 979          * initialized if copy_end was called with copy == NULL.
 980          */
 981
 982         /* Forge a hsm_progress based on data from copy. */
 983         hpk.hpk_fid = copy->hc_hai.hai_fid;
 984         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 985         hpk.hpk_extent = copy->hc_hai.hai_extent;
 986         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 987         hpk.hpk_errval = copy->hc_errval;
 988         hpk.hpk_data_version = 0;
 989
 990         /* For archive request, we need to check the file data was not changed.
 991          *
 992          * For restore request, we need to send the file data version, this is
 993          * useful when the file was created using hsm_import.
 994          */
 995         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 996              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 997             (copy->hc_errval == 0)) {
 998                 struct inode    *inode;
 999                 __u64            data_version = 0;
1000
1001                 /* Get lsm for this fid */
1002                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
1003                 if (IS_ERR(inode)) {
1004                         hpk.hpk_flags |= HP_FLAG_RETRY;
1005                         /* hpk_errval must be >= 0 */
1006                         hpk.hpk_errval = -PTR_ERR(inode);
1007                         rc = PTR_ERR(inode);
1008                         goto progress;
1009                 }
1010
1011                 rc = ll_data_version(inode, &data_version,
1012                                      copy->hc_hai.hai_action == HSMA_ARCHIVE);
1013                 iput(inode);
1014                 if (rc) {
1015                         CDEBUG(D_HSM, "Could not read file data version. Request could not be confirmed.\n");
1016                         if (hpk.hpk_errval == 0)
1017                                 hpk.hpk_errval = -rc;
1018                         goto progress;
1019                 }
1020
1021                 /* Store it the hsm_copy for later copytool use.
1022                  * Always modified even if no lsm. */
1023                 hpk.hpk_data_version = data_version;
1024
1025                 /* File could have been stripped during archiving, so we need
1026                  * to check anyway. */
1027                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1028                     (copy->hc_data_version != data_version)) {
1029                         CDEBUG(D_HSM, "File data version mismatched. File content was changed during archiving. "
1030                                DFID", start:%#llx current:%#llx\n",
1031                                PFID(&copy->hc_hai.hai_fid),
1032                                copy->hc_data_version, data_version);
1033                         /* File was changed, send error to cdt. Do not ask for
1034                          * retry because if a file is modified frequently,
1035                          * the cdt will loop on retried archive requests.
1036                          * The policy engine will ask for a new archive later
1037                          * when the file will not be modified for some tunable
1038                          * time */
1039                         /* we do not notify caller */
1040                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
1041                         /* hpk_errval must be >= 0 */
1042                         hpk.hpk_errval = EBUSY;
1043                 }
1044
1045         }
1046
1047 progress:
1048         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1049                            &hpk, NULL);
1050
1051         return rc;
1052 }
1053
1054
1055 static int copy_and_ioctl(int cmd, struct obd_export *exp,
1056                           const void __user *data, size_t size)
1057 {
1058         void *copy;
1059         int rc;
1060
1061         copy = kzalloc(size, GFP_NOFS);
1062         if (!copy)
1063                 return -ENOMEM;
1064
1065         if (copy_from_user(copy, data, size)) {
1066                 rc = -EFAULT;
1067                 goto out;
1068         }
1069
1070         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
1071 out:
1072         OBD_FREE(copy, size);
1073
1074         return rc;
1075 }
1076
1077 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1078 {
1079         int cmd = qctl->qc_cmd;
1080         int type = qctl->qc_type;
1081         int id = qctl->qc_id;
1082         int valid = qctl->qc_valid;
1083         int rc = 0;
1084
1085         switch (cmd) {
1086         case LUSTRE_Q_INVALIDATE:
1087         case LUSTRE_Q_FINVALIDATE:
1088         case Q_QUOTAON:
1089         case Q_QUOTAOFF:
1090         case Q_SETQUOTA:
1091         case Q_SETINFO:
1092                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1093                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1094                         return -EPERM;
1095                 break;
1096         case Q_GETQUOTA:
1097                 if (((type == USRQUOTA &&
1098                       !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
1099                      (type == GRPQUOTA &&
1100                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
1101                     (!capable(CFS_CAP_SYS_ADMIN) ||
1102                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
1103                         return -EPERM;
1104                 break;
1105         case Q_GETINFO:
1106                 break;
1107         default:
1108                 CERROR("unsupported quotactl op: %#x\n", cmd);
1109                 return -ENOTTY;
1110         }
1111
1112         if (valid != QC_GENERAL) {
1113                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1114                         return -EOPNOTSUPP;
1115
1116                 if (cmd == Q_GETINFO)
1117                         qctl->qc_cmd = Q_GETOINFO;
1118                 else if (cmd == Q_GETQUOTA)
1119                         qctl->qc_cmd = Q_GETOQUOTA;
1120                 else
1121                         return -EINVAL;
1122
1123                 switch (valid) {
1124                 case QC_MDTIDX:
1125                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1126                                            sizeof(*qctl), qctl, NULL);
1127                         break;
1128                 case QC_OSTIDX:
1129                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1130                                            sizeof(*qctl), qctl, NULL);
1131                         break;
1132                 case QC_UUID:
1133                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1134                                            sizeof(*qctl), qctl, NULL);
1135                         if (rc == -EAGAIN)
1136                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1137                                                    sbi->ll_dt_exp,
1138                                                    sizeof(*qctl), qctl, NULL);
1139                         break;
1140                 default:
1141                         rc = -EINVAL;
1142                         break;
1143                 }
1144
1145                 if (rc)
1146                         return rc;
1147
1148                 qctl->qc_cmd = cmd;
1149         } else {
1150                 struct obd_quotactl *oqctl;
1151
1152                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1153                 if (!oqctl)
1154                         return -ENOMEM;
1155
1156                 QCTL_COPY(oqctl, qctl);
1157                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1158                 if (rc) {
1159                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
1160                                 oqctl->qc_cmd = Q_QUOTAOFF;
1161                                 obd_quotactl(sbi->ll_md_exp, oqctl);
1162                         }
1163                         OBD_FREE_PTR(oqctl);
1164                         return rc;
1165                 }
1166                 /* If QIF_SPACE is not set, client should collect the
1167                  * space usage from OSSs by itself */
1168                 if (cmd == Q_GETQUOTA &&
1169                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1170                     !oqctl->qc_dqblk.dqb_curspace) {
1171                         struct obd_quotactl *oqctl_tmp;
1172
1173                         oqctl_tmp = kzalloc(sizeof(*oqctl_tmp), GFP_NOFS);
1174                         if (!oqctl_tmp) {
1175                                 rc = -ENOMEM;
1176                                 goto out;
1177                         }
1178
1179                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1180                         oqctl_tmp->qc_id = oqctl->qc_id;
1181                         oqctl_tmp->qc_type = oqctl->qc_type;
1182
1183                         /* collect space usage from OSTs */
1184                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1185                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1186                         if (!rc || rc == -EREMOTEIO) {
1187                                 oqctl->qc_dqblk.dqb_curspace =
1188                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1189                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1190                         }
1191
1192                         /* collect space & inode usage from MDTs */
1193                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1194                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1195                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1196                         if (!rc || rc == -EREMOTEIO) {
1197                                 oqctl->qc_dqblk.dqb_curspace +=
1198                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1199                                 oqctl->qc_dqblk.dqb_curinodes =
1200                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1201                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1202                         } else {
1203                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1204                         }
1205
1206                         OBD_FREE_PTR(oqctl_tmp);
1207                 }
1208 out:
1209                 QCTL_COPY(qctl, oqctl);
1210                 OBD_FREE_PTR(oqctl);
1211         }
1212
1213         return rc;
1214 }
1215
1216 static char *
1217 ll_getname(const char __user *filename)
1218 {
1219         int ret = 0, len;
1220         char *tmp = __getname();
1221
1222         if (!tmp)
1223                 return ERR_PTR(-ENOMEM);
1224
1225         len = strncpy_from_user(tmp, filename, PATH_MAX);
1226         if (len == 0)
1227                 ret = -ENOENT;
1228         else if (len > PATH_MAX)
1229                 ret = -ENAMETOOLONG;
1230
1231         if (ret) {
1232                 __putname(tmp);
1233                 tmp =  ERR_PTR(ret);
1234         }
1235         return tmp;
1236 }
1237
1238 #define ll_putname(filename) __putname(filename)
1239
1240 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1241 {
1242         struct inode *inode = file_inode(file);
1243         struct ll_sb_info *sbi = ll_i2sbi(inode);
1244         struct obd_ioctl_data *data;
1245         int rc = 0;
1246
1247         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1248                inode->i_ino, inode->i_generation, inode, cmd);
1249
1250         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1251         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1252                 return -ENOTTY;
1253
1254         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1255         switch (cmd) {
1256         case FSFILT_IOC_GETFLAGS:
1257         case FSFILT_IOC_SETFLAGS:
1258                 return ll_iocontrol(inode, file, cmd, arg);
1259         case FSFILT_IOC_GETVERSION_OLD:
1260         case FSFILT_IOC_GETVERSION:
1261                 return put_user(inode->i_generation, (int *)arg);
1262         /* We need to special case any other ioctls we want to handle,
1263          * to send them to the MDS/OST as appropriate and to properly
1264          * network encode the arg field.
1265         case FSFILT_IOC_SETVERSION_OLD:
1266         case FSFILT_IOC_SETVERSION:
1267         */
1268         case LL_IOC_GET_MDTIDX: {
1269                 int mdtidx;
1270
1271                 mdtidx = ll_get_mdt_idx(inode);
1272                 if (mdtidx < 0)
1273                         return mdtidx;
1274
1275                 if (put_user((int)mdtidx, (int *)arg))
1276                         return -EFAULT;
1277
1278                 return 0;
1279         }
1280         case IOC_MDC_LOOKUP: {
1281                 struct ptlrpc_request *request = NULL;
1282                 int namelen, len = 0;
1283                 char *buf = NULL;
1284                 char *filename;
1285                 struct md_op_data *op_data;
1286
1287                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1288                 if (rc)
1289                         return rc;
1290                 data = (void *)buf;
1291
1292                 filename = data->ioc_inlbuf1;
1293                 namelen = strlen(filename);
1294
1295                 if (namelen < 1) {
1296                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1297                         rc = -EINVAL;
1298                         goto out_free;
1299                 }
1300
1301                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1302                                              0, LUSTRE_OPC_ANY, NULL);
1303                 if (IS_ERR(op_data)) {
1304                         rc = PTR_ERR(op_data);
1305                         goto out_free;
1306                 }
1307
1308                 op_data->op_valid = OBD_MD_FLID;
1309                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1310                 ll_finish_md_op_data(op_data);
1311                 if (rc < 0) {
1312                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1313                         goto out_free;
1314                 }
1315                 ptlrpc_req_finished(request);
1316 out_free:
1317                 obd_ioctl_freedata(buf, len);
1318                 return rc;
1319         }
1320         case LL_IOC_LMV_SETSTRIPE: {
1321                 struct lmv_user_md  *lum;
1322                 char            *buf = NULL;
1323                 char            *filename;
1324                 int              namelen = 0;
1325                 int              lumlen = 0;
1326                 int              len;
1327                 int              rc;
1328
1329                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1330                 if (rc)
1331                         return rc;
1332
1333                 data = (void *)buf;
1334                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1335                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0) {
1336                         rc = -EINVAL;
1337                         goto lmv_out_free;
1338                 }
1339
1340                 filename = data->ioc_inlbuf1;
1341                 namelen = data->ioc_inllen1;
1342
1343                 if (namelen < 1) {
1344                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1345                         rc = -EINVAL;
1346                         goto lmv_out_free;
1347                 }
1348                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1349                 lumlen = data->ioc_inllen2;
1350
1351                 if (lum->lum_magic != LMV_USER_MAGIC ||
1352                     lumlen != sizeof(*lum)) {
1353                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1354                                filename, lum->lum_magic, lumlen, -EFAULT);
1355                         rc = -EINVAL;
1356                         goto lmv_out_free;
1357                 }
1358
1359                 /**
1360                  * ll_dir_setdirstripe will be used to set dir stripe
1361                  *  mdc_create--->mdt_reint_create (with dirstripe)
1362                  */
1363                 rc = ll_dir_setdirstripe(inode, lum, filename);
1364 lmv_out_free:
1365                 obd_ioctl_freedata(buf, len);
1366                 return rc;
1367
1368         }
1369         case LL_IOC_LOV_SETSTRIPE: {
1370                 struct lov_user_md_v3 lumv3;
1371                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1372                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1373                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1374
1375                 int set_default = 0;
1376
1377                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1378                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1379                         sizeof(lumv3p->lmm_objects[0]));
1380                 /* first try with v1 which is smaller than v3 */
1381                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1382                         return -EFAULT;
1383
1384                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1385                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1386                                 return -EFAULT;
1387                 }
1388
1389                 if (is_root_inode(inode))
1390                         set_default = 1;
1391
1392                 /* in v1 and v3 cases lumv1 points to data */
1393                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1394
1395                 return rc;
1396         }
1397         case LL_IOC_LMV_GETSTRIPE: {
1398                 struct lmv_user_md *lump = (struct lmv_user_md *)arg;
1399                 struct lmv_user_md lum;
1400                 struct lmv_user_md *tmp;
1401                 int lum_size;
1402                 int rc = 0;
1403                 int mdtindex;
1404
1405                 if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
1406                         return -EFAULT;
1407
1408                 if (lum.lum_magic != LMV_MAGIC_V1)
1409                         return -EINVAL;
1410
1411                 lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
1412                 tmp = kzalloc(lum_size, GFP_NOFS);
1413                 if (!tmp) {
1414                         rc = -ENOMEM;
1415                         goto free_lmv;
1416                 }
1417
1418                 *tmp = lum;
1419                 tmp->lum_type = LMV_STRIPE_TYPE;
1420                 tmp->lum_stripe_count = 1;
1421                 mdtindex = ll_get_mdt_idx(inode);
1422                 if (mdtindex < 0) {
1423                         rc = -ENOMEM;
1424                         goto free_lmv;
1425                 }
1426
1427                 tmp->lum_stripe_offset = mdtindex;
1428                 tmp->lum_objects[0].lum_mds = mdtindex;
1429                 memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
1430                        sizeof(struct lu_fid));
1431                 if (copy_to_user((void *)arg, tmp, lum_size)) {
1432                         rc = -EFAULT;
1433                         goto free_lmv;
1434                 }
1435 free_lmv:
1436                 if (tmp)
1437                         OBD_FREE(tmp, lum_size);
1438                 return rc;
1439         }
1440         case LL_IOC_REMOVE_ENTRY: {
1441                 char            *filename = NULL;
1442                 int              namelen = 0;
1443                 int              rc;
1444
1445                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1446                  * unsupported server, which might crash the server(LU-2730),
1447                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1448                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1449                  * server will support REINT_RMENTRY XXX*/
1450                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1451                         return -ENOTSUPP;
1452
1453                 filename = ll_getname((const char *)arg);
1454                 if (IS_ERR(filename))
1455                         return PTR_ERR(filename);
1456
1457                 namelen = strlen(filename);
1458                 if (namelen < 1) {
1459                         rc = -EINVAL;
1460                         goto out_rmdir;
1461                 }
1462
1463                 rc = ll_rmdir_entry(inode, filename, namelen);
1464 out_rmdir:
1465                 if (filename)
1466                         ll_putname(filename);
1467                 return rc;
1468         }
1469         case LL_IOC_LOV_SWAP_LAYOUTS:
1470                 return -EPERM;
1471         case LL_IOC_OBD_STATFS:
1472                 return ll_obd_statfs(inode, (void *)arg);
1473         case LL_IOC_LOV_GETSTRIPE:
1474         case LL_IOC_MDC_GETINFO:
1475         case IOC_MDC_GETFILEINFO:
1476         case IOC_MDC_GETFILESTRIPE: {
1477                 struct ptlrpc_request *request = NULL;
1478                 struct lov_user_md *lump;
1479                 struct lov_mds_md *lmm = NULL;
1480                 struct mdt_body *body;
1481                 char *filename = NULL;
1482                 int lmmsize;
1483
1484                 if (cmd == IOC_MDC_GETFILEINFO ||
1485                     cmd == IOC_MDC_GETFILESTRIPE) {
1486                         filename = ll_getname((const char *)arg);
1487                         if (IS_ERR(filename))
1488                                 return PTR_ERR(filename);
1489
1490                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1491                                                       &lmmsize, &request);
1492                 } else {
1493                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1494                 }
1495
1496                 if (request) {
1497                         body = req_capsule_server_get(&request->rq_pill,
1498                                                       &RMF_MDT_BODY);
1499                         LASSERT(body != NULL);
1500                 } else {
1501                         goto out_req;
1502                 }
1503
1504                 if (rc < 0) {
1505                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1506                                                cmd == LL_IOC_MDC_GETINFO)) {
1507                                 rc = 0;
1508                                 goto skip_lmm;
1509                         } else
1510                                 goto out_req;
1511                 }
1512
1513                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1514                     cmd == LL_IOC_LOV_GETSTRIPE) {
1515                         lump = (struct lov_user_md *)arg;
1516                 } else {
1517                         struct lov_user_mds_data *lmdp;
1518                         lmdp = (struct lov_user_mds_data *)arg;
1519                         lump = &lmdp->lmd_lmm;
1520                 }
1521                 if (copy_to_user(lump, lmm, lmmsize)) {
1522                         if (copy_to_user(lump, lmm, sizeof(*lump))) {
1523                                 rc = -EFAULT;
1524                                 goto out_req;
1525                         }
1526                         rc = -EOVERFLOW;
1527                 }
1528 skip_lmm:
1529                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1530                         struct lov_user_mds_data *lmdp;
1531                         lstat_t st = { 0 };
1532
1533                         st.st_dev     = inode->i_sb->s_dev;
1534                         st.st_mode    = body->mode;
1535                         st.st_nlink   = body->nlink;
1536                         st.st_uid     = body->uid;
1537                         st.st_gid     = body->gid;
1538                         st.st_rdev    = body->rdev;
1539                         st.st_size    = body->size;
1540                         st.st_blksize = PAGE_CACHE_SIZE;
1541                         st.st_blocks  = body->blocks;
1542                         st.st_atime   = body->atime;
1543                         st.st_mtime   = body->mtime;
1544                         st.st_ctime   = body->ctime;
1545                         st.st_ino     = inode->i_ino;
1546
1547                         lmdp = (struct lov_user_mds_data *)arg;
1548                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st))) {
1549                                 rc = -EFAULT;
1550                                 goto out_req;
1551                         }
1552                 }
1553
1554 out_req:
1555                 ptlrpc_req_finished(request);
1556                 if (filename)
1557                         ll_putname(filename);
1558                 return rc;
1559         }
1560         case IOC_LOV_GETINFO: {
1561                 struct lov_user_mds_data *lumd;
1562                 struct lov_stripe_md *lsm;
1563                 struct lov_user_md *lum;
1564                 struct lov_mds_md *lmm;
1565                 int lmmsize;
1566                 lstat_t st;
1567
1568                 lumd = (struct lov_user_mds_data *)arg;
1569                 lum = &lumd->lmd_lmm;
1570
1571                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1572                 if (rc)
1573                         return rc;
1574
1575                 OBD_ALLOC_LARGE(lmm, lmmsize);
1576                 if (lmm == NULL)
1577                         return -ENOMEM;
1578                 if (copy_from_user(lmm, lum, lmmsize)) {
1579                         rc = -EFAULT;
1580                         goto free_lmm;
1581                 }
1582
1583                 switch (lmm->lmm_magic) {
1584                 case LOV_USER_MAGIC_V1:
1585                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1586                                 break;
1587                         /* swab objects first so that stripes num will be sane */
1588                         lustre_swab_lov_user_md_objects(
1589                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1590                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1591                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1592                         break;
1593                 case LOV_USER_MAGIC_V3:
1594                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1595                                 break;
1596                         /* swab objects first so that stripes num will be sane */
1597                         lustre_swab_lov_user_md_objects(
1598                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1599                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1600                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1601                         break;
1602                 default:
1603                         rc = -EINVAL;
1604                         goto free_lmm;
1605                 }
1606
1607                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1608                 if (rc < 0) {
1609                         rc = -ENOMEM;
1610                         goto free_lmm;
1611                 }
1612
1613                 /* Perform glimpse_size operation. */
1614                 memset(&st, 0, sizeof(st));
1615
1616                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1617                 if (rc)
1618                         goto free_lsm;
1619
1620                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st))) {
1621                         rc = -EFAULT;
1622                         goto free_lsm;
1623                 }
1624
1625 free_lsm:
1626                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1627 free_lmm:
1628                 OBD_FREE_LARGE(lmm, lmmsize);
1629                 return rc;
1630         }
1631         case OBD_IOC_LLOG_CATINFO: {
1632                 return -EOPNOTSUPP;
1633         }
1634         case OBD_IOC_QUOTACHECK: {
1635                 struct obd_quotactl *oqctl;
1636                 int error = 0;
1637
1638                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1639                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1640                         return -EPERM;
1641
1642                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1643                 if (!oqctl)
1644                         return -ENOMEM;
1645                 oqctl->qc_type = arg;
1646                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1647                 if (rc < 0) {
1648                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1649                         error = rc;
1650                 }
1651
1652                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1653                 if (rc < 0)
1654                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1655
1656                 OBD_FREE_PTR(oqctl);
1657                 return error ?: rc;
1658         }
1659         case OBD_IOC_POLL_QUOTACHECK: {
1660                 struct if_quotacheck *check;
1661
1662                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1663                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1664                         return -EPERM;
1665
1666                 check = kzalloc(sizeof(*check), GFP_NOFS);
1667                 if (!check)
1668                         return -ENOMEM;
1669
1670                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1671                                    NULL);
1672                 if (rc) {
1673                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1674                         if (copy_to_user((void *)arg, check,
1675                                              sizeof(*check)))
1676                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1677                         goto out_poll;
1678                 }
1679
1680                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1681                                    NULL);
1682                 if (rc) {
1683                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1684                         if (copy_to_user((void *)arg, check,
1685                                              sizeof(*check)))
1686                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1687                         goto out_poll;
1688                 }
1689 out_poll:
1690                 OBD_FREE_PTR(check);
1691                 return rc;
1692         }
1693         case LL_IOC_QUOTACTL: {
1694                 struct if_quotactl *qctl;
1695
1696                 qctl = kzalloc(sizeof(*qctl), GFP_NOFS);
1697                 if (!qctl)
1698                         return -ENOMEM;
1699
1700                 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl))) {
1701                         rc = -EFAULT;
1702                         goto out_quotactl;
1703                 }
1704
1705                 rc = quotactl_ioctl(sbi, qctl);
1706
1707                 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1708                         rc = -EFAULT;
1709
1710 out_quotactl:
1711                 OBD_FREE_PTR(qctl);
1712                 return rc;
1713         }
1714         case OBD_IOC_GETDTNAME:
1715         case OBD_IOC_GETMDNAME:
1716                 return ll_get_obd_name(inode, cmd, arg);
1717         case LL_IOC_FLUSHCTX:
1718                 return ll_flush_ctx(inode);
1719 #ifdef CONFIG_FS_POSIX_ACL
1720         case LL_IOC_RMTACL: {
1721             if (sbi->ll_flags & LL_SBI_RMT_CLIENT && is_root_inode(inode)) {
1722                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1723
1724                 LASSERT(fd != NULL);
1725                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1726                 if (!rc)
1727                         fd->fd_flags |= LL_FILE_RMTACL;
1728                 return rc;
1729             } else
1730                 return 0;
1731         }
1732 #endif
1733         case LL_IOC_GETOBDCOUNT: {
1734                 int count, vallen;
1735                 struct obd_export *exp;
1736
1737                 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1738                         return -EFAULT;
1739
1740                 /* get ost count when count is zero, get mdt count otherwise */
1741                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1742                 vallen = sizeof(count);
1743                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1744                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1745                 if (rc) {
1746                         CERROR("get target count failed: %d\n", rc);
1747                         return rc;
1748                 }
1749
1750                 if (copy_to_user((int *)arg, &count, sizeof(int)))
1751                         return -EFAULT;
1752
1753                 return 0;
1754         }
1755         case LL_IOC_PATH2FID:
1756                 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1757                                      sizeof(struct lu_fid)))
1758                         return -EFAULT;
1759                 return 0;
1760         case LL_IOC_GET_CONNECT_FLAGS: {
1761                 return obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void *)arg);
1762         }
1763         case OBD_IOC_CHANGELOG_SEND:
1764         case OBD_IOC_CHANGELOG_CLEAR:
1765                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1766                                     sizeof(struct ioc_changelog));
1767                 return rc;
1768         case OBD_IOC_FID2PATH:
1769                 return ll_fid2path(inode, (void *)arg);
1770         case LL_IOC_HSM_REQUEST: {
1771                 struct hsm_user_request *hur;
1772                 ssize_t                  totalsize;
1773
1774                 hur = kzalloc(sizeof(*hur), GFP_NOFS);
1775                 if (!hur)
1776                         return -ENOMEM;
1777
1778                 /* We don't know the true size yet; copy the fixed-size part */
1779                 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1780                         OBD_FREE_PTR(hur);
1781                         return -EFAULT;
1782                 }
1783
1784                 /* Compute the whole struct size */
1785                 totalsize = hur_len(hur);
1786                 OBD_FREE_PTR(hur);
1787                 if (totalsize < 0)
1788                         return -E2BIG;
1789
1790                 /* Final size will be more than double totalsize */
1791                 if (totalsize >= MDS_MAXREQSIZE / 3)
1792                         return -E2BIG;
1793
1794                 OBD_ALLOC_LARGE(hur, totalsize);
1795                 if (hur == NULL)
1796                         return -ENOMEM;
1797
1798                 /* Copy the whole struct */
1799                 if (copy_from_user(hur, (void *)arg, totalsize)) {
1800                         OBD_FREE_LARGE(hur, totalsize);
1801                         return -EFAULT;
1802                 }
1803
1804                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1805                         const struct lu_fid *fid;
1806                         struct inode *f;
1807                         int i;
1808
1809                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1810                                 fid = &hur->hur_user_item[i].hui_fid;
1811                                 f = search_inode_for_lustre(inode->i_sb, fid);
1812                                 if (IS_ERR(f)) {
1813                                         rc = PTR_ERR(f);
1814                                         break;
1815                                 }
1816
1817                                 rc = ll_hsm_release(f);
1818                                 iput(f);
1819                                 if (rc != 0)
1820                                         break;
1821                         }
1822                 } else {
1823                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1824                                            hur, NULL);
1825                 }
1826
1827                 OBD_FREE_LARGE(hur, totalsize);
1828
1829                 return rc;
1830         }
1831         case LL_IOC_HSM_PROGRESS: {
1832                 struct hsm_progress_kernel      hpk;
1833                 struct hsm_progress             hp;
1834
1835                 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1836                         return -EFAULT;
1837
1838                 hpk.hpk_fid = hp.hp_fid;
1839                 hpk.hpk_cookie = hp.hp_cookie;
1840                 hpk.hpk_extent = hp.hp_extent;
1841                 hpk.hpk_flags = hp.hp_flags;
1842                 hpk.hpk_errval = hp.hp_errval;
1843                 hpk.hpk_data_version = 0;
1844
1845                 /* File may not exist in Lustre; all progress
1846                  * reported to Lustre root */
1847                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1848                                    NULL);
1849                 return rc;
1850         }
1851         case LL_IOC_HSM_CT_START:
1852                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1853                                     sizeof(struct lustre_kernelcomm));
1854                 return rc;
1855
1856         case LL_IOC_HSM_COPY_START: {
1857                 struct hsm_copy *copy;
1858                 int              rc;
1859
1860                 copy = kzalloc(sizeof(*copy), GFP_NOFS);
1861                 if (!copy)
1862                         return -ENOMEM;
1863                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1864                         OBD_FREE_PTR(copy);
1865                         return -EFAULT;
1866                 }
1867
1868                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1869                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1870                         rc = -EFAULT;
1871
1872                 OBD_FREE_PTR(copy);
1873                 return rc;
1874         }
1875         case LL_IOC_HSM_COPY_END: {
1876                 struct hsm_copy *copy;
1877                 int              rc;
1878
1879                 copy = kzalloc(sizeof(*copy), GFP_NOFS);
1880                 if (!copy)
1881                         return -ENOMEM;
1882                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1883                         OBD_FREE_PTR(copy);
1884                         return -EFAULT;
1885                 }
1886
1887                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1888                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1889                         rc = -EFAULT;
1890
1891                 OBD_FREE_PTR(copy);
1892                 return rc;
1893         }
1894         default:
1895                 return obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL, (void *)arg);
1896         }
1897 }
1898
1899 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1900 {
1901         struct inode *inode = file->f_mapping->host;
1902         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1903         struct ll_sb_info *sbi = ll_i2sbi(inode);
1904         int api32 = ll_need_32bit_api(sbi);
1905         loff_t ret = -EINVAL;
1906
1907         mutex_lock(&inode->i_mutex);
1908         switch (origin) {
1909                 case SEEK_SET:
1910                         break;
1911                 case SEEK_CUR:
1912                         offset += file->f_pos;
1913                         break;
1914                 case SEEK_END:
1915                         if (offset > 0)
1916                                 goto out;
1917                         if (api32)
1918                                 offset += LL_DIR_END_OFF_32BIT;
1919                         else
1920                                 offset += LL_DIR_END_OFF;
1921                         break;
1922                 default:
1923                         goto out;
1924         }
1925
1926         if (offset >= 0 &&
1927             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1928              (!api32 && offset <= LL_DIR_END_OFF))) {
1929                 if (offset != file->f_pos) {
1930                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1931                             (!api32 && offset == LL_DIR_END_OFF))
1932                                 fd->lfd_pos = MDS_DIR_END_OFF;
1933                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1934                                 fd->lfd_pos = offset << 32;
1935                         else
1936                                 fd->lfd_pos = offset;
1937                         file->f_pos = offset;
1938                         file->f_version = 0;
1939                 }
1940                 ret = offset;
1941         }
1942         goto out;
1943
1944 out:
1945         mutex_unlock(&inode->i_mutex);
1946         return ret;
1947 }
1948
1949 static int ll_dir_open(struct inode *inode, struct file *file)
1950 {
1951         return ll_file_open(inode, file);
1952 }
1953
1954 static int ll_dir_release(struct inode *inode, struct file *file)
1955 {
1956         return ll_file_release(inode, file);
1957 }
1958
1959 const struct file_operations ll_dir_operations = {
1960         .llseek   = ll_dir_seek,
1961         .open     = ll_dir_open,
1962         .release  = ll_dir_release,
1963         .read     = generic_read_dir,
1964         .iterate  = ll_readdir,
1965         .unlocked_ioctl   = ll_dir_ioctl,
1966         .fsync    = ll_fsync,
1967 };