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         ll_pagevec_init(&lru_pvec, 0);
 207         for (i = 1; i < npages; i++) {
 208                 unsigned long offset;
 209                 int ret;
 210
 211                 page = page_pool[i];
 212
 213                 if (rc < 0 || i >= nrdpgs) {
 214                         page_cache_release(page);
 215                         continue;
 216                 }
 217
 218                 SetPageUptodate(page);
 219
 220                 dp = kmap(page);
 221                 hash = le64_to_cpu(dp->ldp_hash_start);
 222                 kunmap(page);
 223
 224                 offset = hash_x_index(hash, hash64);
 225
 226                 prefetchw(&page->flags);
 227                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
 228                                             GFP_KERNEL);
 229                 if (ret == 0) {
 230                         unlock_page(page);
 231                         if (ll_pagevec_add(&lru_pvec, page) == 0)
 232                                 ll_pagevec_lru_add_file(&lru_pvec);
 233                 } else {
 234                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed: %d\n",
 235                                offset, ret);
 236                 }
 237                 page_cache_release(page);
 238         }
 239         ll_pagevec_lru_add_file(&lru_pvec);
 240
 241         if (page_pool != &page0)
 242                 kfree(page_pool);
 243         return rc;
 244 }
 245
 246 static void ll_check_page(struct inode *dir, struct page *page)
 247 {
 248         /* XXX: check page format later */
 249         SetPageChecked(page);
 250 }
 251
 252 void ll_release_page(struct page *page, int remove)
 253 {
 254         kunmap(page);
 255         if (remove) {
 256                 lock_page(page);
 257                 if (likely(page->mapping != NULL))
 258                         truncate_complete_page(page->mapping, page);
 259                 unlock_page(page);
 260         }
 261         page_cache_release(page);
 262 }
 263
 264 /*
 265  * Find, kmap and return page that contains given hash.
 266  */
 267 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
 268                                        __u64 *start, __u64 *end)
 269 {
 270         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 271         struct address_space *mapping = dir->i_mapping;
 272         /*
 273          * Complement of hash is used as an index so that
 274          * radix_tree_gang_lookup() can be used to find a page with starting
 275          * hash _smaller_ than one we are looking for.
 276          */
 277         unsigned long offset = hash_x_index(*hash, hash64);
 278         struct page *page;
 279         int found;
 280
 281         spin_lock_irq(&mapping->tree_lock);
 282         found = radix_tree_gang_lookup(&mapping->page_tree,
 283                                        (void **)&page, offset, 1);
 284         if (found > 0 && !radix_tree_exceptional_entry(page)) {
 285                 struct lu_dirpage *dp;
 286
 287                 page_cache_get(page);
 288                 spin_unlock_irq(&mapping->tree_lock);
 289                 /*
 290                  * In contrast to find_lock_page() we are sure that directory
 291                  * page cannot be truncated (while DLM lock is held) and,
 292                  * hence, can avoid restart.
 293                  *
 294                  * In fact, page cannot be locked here at all, because
 295                  * ll_dir_filler() does synchronous io.
 296                  */
 297                 wait_on_page_locked(page);
 298                 if (PageUptodate(page)) {
 299                         dp = kmap(page);
 300                         if (BITS_PER_LONG == 32 && hash64) {
 301                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 302                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 303                                 *hash  = *hash >> 32;
 304                         } else {
 305                                 *start = le64_to_cpu(dp->ldp_hash_start);
 306                                 *end   = le64_to_cpu(dp->ldp_hash_end);
 307                         }
 308                         LASSERTF(*start <= *hash, "start = %#llx,end = %#llx,hash = %#llx\n",
 309                                  *start, *end, *hash);
 310                         CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash %llu\n",
 311                                offset, *start, *end, *hash);
 312                         if (*hash > *end) {
 313                                 ll_release_page(page, 0);
 314                                 page = NULL;
 315                         } else if (*end != *start && *hash == *end) {
 316                                 /*
 317                                  * upon hash collision, remove this page,
 318                                  * otherwise put page reference, and
 319                                  * ll_get_dir_page() will issue RPC to fetch
 320                                  * the page we want.
 321                                  */
 322                                 ll_release_page(page,
 323                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
 324                                 page = NULL;
 325                         }
 326                 } else {
 327                         page_cache_release(page);
 328                         page = ERR_PTR(-EIO);
 329                 }
 330
 331         } else {
 332                 spin_unlock_irq(&mapping->tree_lock);
 333                 page = NULL;
 334         }
 335         return page;
 336 }
 337
 338 struct page *ll_get_dir_page(struct inode *dir, __u64 hash,
 339                              struct ll_dir_chain *chain)
 340 {
 341         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 342         struct address_space *mapping = dir->i_mapping;
 343         struct lustre_handle lockh;
 344         struct lu_dirpage *dp;
 345         struct page *page;
 346         ldlm_mode_t mode;
 347         int rc;
 348         __u64 start = 0;
 349         __u64 end = 0;
 350         __u64 lhash = hash;
 351         struct ll_inode_info *lli = ll_i2info(dir);
 352         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 353
 354         mode = LCK_PR;
 355         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 356                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
 357         if (!rc) {
 358                 struct ldlm_enqueue_info einfo = {
 359                         .ei_type = LDLM_IBITS,
 360                         .ei_mode = mode,
 361                         .ei_cb_bl = ll_md_blocking_ast,
 362                         .ei_cb_cp = ldlm_completion_ast,
 363                 };
 364                 struct lookup_intent it = { .it_op = IT_READDIR };
 365                 struct ptlrpc_request *request;
 366                 struct md_op_data *op_data;
 367
 368                 op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0,
 369                 LUSTRE_OPC_ANY, NULL);
 370                 if (IS_ERR(op_data))
 371                         return (void *)op_data;
 372
 373                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
 374                                 op_data, &lockh, NULL, 0, NULL, 0);
 375
 376                 ll_finish_md_op_data(op_data);
 377
 378                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 379                 if (request)
 380                         ptlrpc_req_finished(request);
 381                 if (rc < 0) {
 382                         CERROR("lock enqueue: "DFID" at %llu: rc %d\n",
 383                                 PFID(ll_inode2fid(dir)), hash, rc);
 384                         return ERR_PTR(rc);
 385                 }
 386
 387                 CDEBUG(D_INODE, "setting lr_lvb_inode to inode %p (%lu/%u)\n",
 388                        dir, dir->i_ino, dir->i_generation);
 389                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp,
 390                                  &it.d.lustre.it_lock_handle, dir, NULL);
 391         } else {
 392                 /* for cross-ref object, l_ast_data of the lock may not be set,
 393                  * we reset it here */
 394                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
 395                                  dir, NULL);
 396         }
 397         ldlm_lock_dump_handle(D_OTHER, &lockh);
 398
 399         mutex_lock(&lli->lli_readdir_mutex);
 400         page = ll_dir_page_locate(dir, &lhash, &start, &end);
 401         if (IS_ERR(page)) {
 402                 CERROR("dir page locate: "DFID" at %llu: rc %ld\n",
 403                        PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
 404                 goto out_unlock;
 405         } else if (page != NULL) {
 406                 /*
 407                  * XXX nikita: not entirely correct handling of a corner case:
 408                  * suppose hash chain of entries with hash value HASH crosses
 409                  * border between pages P0 and P1. First both P0 and P1 are
 410                  * cached, seekdir() is called for some entry from the P0 part
 411                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 412                  * happens and finds P1, as it starts with matching hash
 413                  * value. Remaining entries from P0 part of the chain are
 414                  * skipped. (Is that really a bug?)
 415                  *
 416                  * Possible solutions: 0. don't cache P1 is such case, handle
 417                  * it as an "overflow" page. 1. invalidate all pages at
 418                  * once. 2. use HASH|1 as an index for P1.
 419                  */
 420                 goto hash_collision;
 421         }
 422
 423         page = read_cache_page(mapping, hash_x_index(hash, hash64),
 424                                ll_dir_filler, &lhash);
 425         if (IS_ERR(page)) {
 426                 CERROR("read cache page: "DFID" at %llu: rc %ld\n",
 427                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 428                 goto out_unlock;
 429         }
 430
 431         wait_on_page_locked(page);
 432         (void)kmap(page);
 433         if (!PageUptodate(page)) {
 434                 CERROR("page not updated: "DFID" at %llu: rc %d\n",
 435                        PFID(ll_inode2fid(dir)), hash, -5);
 436                 goto fail;
 437         }
 438         if (!PageChecked(page))
 439                 ll_check_page(dir, page);
 440         if (PageError(page)) {
 441                 CERROR("page error: "DFID" at %llu: rc %d\n",
 442                        PFID(ll_inode2fid(dir)), hash, -5);
 443                 goto fail;
 444         }
 445 hash_collision:
 446         dp = page_address(page);
 447         if (BITS_PER_LONG == 32 && hash64) {
 448                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 449                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 450                 lhash = hash >> 32;
 451         } else {
 452                 start = le64_to_cpu(dp->ldp_hash_start);
 453                 end   = le64_to_cpu(dp->ldp_hash_end);
 454                 lhash = hash;
 455         }
 456         if (end == start) {
 457                 LASSERT(start == lhash);
 458                 CWARN("Page-wide hash collision: %llu\n", end);
 459                 if (BITS_PER_LONG == 32 && hash64)
 460                         CWARN("Real page-wide hash collision at [%llu %llu] with hash %llu\n",
 461                               le64_to_cpu(dp->ldp_hash_start),
 462                               le64_to_cpu(dp->ldp_hash_end), hash);
 463                 /*
 464                  * Fetch whole overflow chain...
 465                  *
 466                  * XXX not yet.
 467                  */
 468                 goto fail;
 469         }
 470 out_unlock:
 471         mutex_unlock(&lli->lli_readdir_mutex);
 472         ldlm_lock_decref(&lockh, mode);
 473         return page;
 474
 475 fail:
 476         ll_release_page(page, 1);
 477         page = ERR_PTR(-EIO);
 478         goto out_unlock;
 479 }
 480
 481 int ll_dir_read(struct inode *inode, struct dir_context *ctx)
 482 {
 483         struct ll_inode_info *info       = ll_i2info(inode);
 484         struct ll_sb_info    *sbi       = ll_i2sbi(inode);
 485         __u64              pos          = ctx->pos;
 486         int                api32      = ll_need_32bit_api(sbi);
 487         int                hash64     = sbi->ll_flags & LL_SBI_64BIT_HASH;
 488         struct page       *page;
 489         struct ll_dir_chain   chain;
 490         int                done = 0;
 491         int                rc = 0;
 492
 493         ll_dir_chain_init(&chain);
 494
 495         page = ll_get_dir_page(inode, pos, &chain);
 496
 497         while (rc == 0 && !done) {
 498                 struct lu_dirpage *dp;
 499                 struct lu_dirent  *ent;
 500
 501                 if (!IS_ERR(page)) {
 502                         /*
 503                          * If page is empty (end of directory is reached),
 504                          * use this value.
 505                          */
 506                         __u64 hash = MDS_DIR_END_OFF;
 507                         __u64 next;
 508
 509                         dp = page_address(page);
 510                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 511                              ent = lu_dirent_next(ent)) {
 512                                 __u16     type;
 513                                 int         namelen;
 514                                 struct lu_fid  fid;
 515                                 __u64     lhash;
 516                                 __u64     ino;
 517
 518                                 /*
 519                                  * XXX: implement correct swabbing here.
 520                                  */
 521
 522                                 hash = le64_to_cpu(ent->lde_hash);
 523                                 if (hash < pos)
 524                                         /*
 525                                          * Skip until we find target hash
 526                                          * value.
 527                                          */
 528                                         continue;
 529
 530                                 namelen = le16_to_cpu(ent->lde_namelen);
 531                                 if (namelen == 0)
 532                                         /*
 533                                          * Skip dummy record.
 534                                          */
 535                                         continue;
 536
 537                                 if (api32 && hash64)
 538                                         lhash = hash >> 32;
 539                                 else
 540                                         lhash = hash;
 541                                 fid_le_to_cpu(&fid, &ent->lde_fid);
 542                                 ino = cl_fid_build_ino(&fid, api32);
 543                                 type = ll_dirent_type_get(ent);
 544                                 ctx->pos = lhash;
 545                                 /* For 'll_nfs_get_name_filldir()', it will try
 546                                  * to access the 'ent' through its 'lde_name',
 547                                  * so the parameter 'name' for 'ctx->actor()'
 548                                  * must be part of the 'ent'.
 549                                  */
 550                                 done = !dir_emit(ctx, ent->lde_name,
 551                                                  namelen, ino, type);
 552                         }
 553                         next = le64_to_cpu(dp->ldp_hash_end);
 554                         if (!done) {
 555                                 pos = next;
 556                                 if (pos == MDS_DIR_END_OFF) {
 557                                         /*
 558                                          * End of directory reached.
 559                                          */
 560                                         done = 1;
 561                                         ll_release_page(page, 0);
 562                                 } else if (1 /* chain is exhausted*/) {
 563                                         /*
 564                                          * Normal case: continue to the next
 565                                          * page.
 566                                          */
 567                                         ll_release_page(page,
 568                                             le32_to_cpu(dp->ldp_flags) &
 569                                                         LDF_COLLIDE);
 570                                         next = pos;
 571                                         page = ll_get_dir_page(inode, pos,
 572                                                                &chain);
 573                                 } else {
 574                                         /*
 575                                          * go into overflow page.
 576                                          */
 577                                         LASSERT(le32_to_cpu(dp->ldp_flags) &
 578                                                 LDF_COLLIDE);
 579                                         ll_release_page(page, 1);
 580                                 }
 581                         } else {
 582                                 pos = hash;
 583                                 ll_release_page(page, 0);
 584                         }
 585                 } else {
 586                         rc = PTR_ERR(page);
 587                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 588                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 589                 }
 590         }
 591
 592         ctx->pos = pos;
 593         ll_dir_chain_fini(&chain);
 594         return rc;
 595 }
 596
 597 static int ll_readdir(struct file *filp, struct dir_context *ctx)
 598 {
 599         struct inode            *inode  = file_inode(filp);
 600         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 601         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 602         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 603         int                     api32   = ll_need_32bit_api(sbi);
 604         int                     rc;
 605
 606         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu 32bit_api %d\n",
 607                inode->i_ino, inode->i_generation,
 608                inode, (unsigned long)lfd->lfd_pos, i_size_read(inode), api32);
 609
 610         if (lfd->lfd_pos == MDS_DIR_END_OFF) {
 611                 /*
 612                  * end-of-file.
 613                  */
 614                 rc = 0;
 615                 goto out;
 616         }
 617
 618         ctx->pos = lfd->lfd_pos;
 619         rc = ll_dir_read(inode, ctx);
 620         lfd->lfd_pos = ctx->pos;
 621         if (ctx->pos == MDS_DIR_END_OFF) {
 622                 if (api32)
 623                         ctx->pos = LL_DIR_END_OFF_32BIT;
 624                 else
 625                         ctx->pos = LL_DIR_END_OFF;
 626         } else {
 627                 if (api32 && hash64)
 628                         ctx->pos >>= 32;
 629         }
 630         filp->f_version = inode->i_version;
 631
 632 out:
 633         if (!rc)
 634                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 635
 636         return rc;
 637 }
 638
 639 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 640 {
 641         struct mgs_send_param *msp;
 642         int rc = 0;
 643
 644         msp = kzalloc(sizeof(*msp), GFP_NOFS);
 645         if (!msp)
 646                 return -ENOMEM;
 647
 648         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 649         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 650                                 sizeof(struct mgs_send_param), msp, NULL);
 651         if (rc)
 652                 CERROR("Failed to set parameter: %d\n", rc);
 653         kfree(msp);
 654
 655         return rc;
 656 }
 657
 658 static int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
 659                                char *filename)
 660 {
 661         struct ptlrpc_request *request = NULL;
 662         struct md_op_data *op_data;
 663         struct ll_sb_info *sbi = ll_i2sbi(dir);
 664         int mode;
 665         int err;
 666
 667         mode = (0755 & ~current_umask()) | S_IFDIR;
 668         op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
 669                                      strlen(filename), mode, LUSTRE_OPC_MKDIR,
 670                                      lump);
 671         if (IS_ERR(op_data)) {
 672                 err = PTR_ERR(op_data);
 673                 goto err_exit;
 674         }
 675
 676         op_data->op_cli_flags |= CLI_SET_MEA;
 677         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 678                         from_kuid(&init_user_ns, current_fsuid()),
 679                         from_kgid(&init_user_ns, current_fsgid()),
 680                         cfs_curproc_cap_pack(), 0, &request);
 681         ll_finish_md_op_data(op_data);
 682         if (err)
 683                 goto err_exit;
 684 err_exit:
 685         ptlrpc_req_finished(request);
 686         return err;
 687 }
 688
 689 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 690                      int set_default)
 691 {
 692         struct ll_sb_info *sbi = ll_i2sbi(inode);
 693         struct md_op_data *op_data;
 694         struct ptlrpc_request *req = NULL;
 695         int rc = 0;
 696         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 697         struct obd_device *mgc = lsi->lsi_mgc;
 698         int lum_size;
 699
 700         if (lump != NULL) {
 701                 /*
 702                  * This is coming from userspace, so should be in
 703                  * local endian.  But the MDS would like it in little
 704                  * endian, so we swab it before we send it.
 705                  */
 706                 switch (lump->lmm_magic) {
 707                 case LOV_USER_MAGIC_V1: {
 708                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 709                                 lustre_swab_lov_user_md_v1(lump);
 710                         lum_size = sizeof(struct lov_user_md_v1);
 711                         break;
 712                 }
 713                 case LOV_USER_MAGIC_V3: {
 714                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 715                                 lustre_swab_lov_user_md_v3(
 716                                         (struct lov_user_md_v3 *)lump);
 717                         lum_size = sizeof(struct lov_user_md_v3);
 718                         break;
 719                 }
 720                 default: {
 721                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
 722                                lump->lmm_magic, LOV_USER_MAGIC_V1,
 723                                LOV_USER_MAGIC_V3);
 724                         return -EINVAL;
 725                 }
 726                 }
 727         } else {
 728                 lum_size = sizeof(struct lov_user_md_v1);
 729         }
 730
 731         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 732                                      LUSTRE_OPC_ANY, NULL);
 733         if (IS_ERR(op_data))
 734                 return PTR_ERR(op_data);
 735
 736         if (lump != NULL && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC))
 737                 op_data->op_cli_flags |= CLI_SET_MEA;
 738
 739         /* swabbing is done in lov_setstripe() on server side */
 740         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 741                         NULL, 0, &req, NULL);
 742         ll_finish_md_op_data(op_data);
 743         ptlrpc_req_finished(req);
 744         if (rc) {
 745                 if (rc != -EPERM && rc != -EACCES)
 746                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 747         }
 748
 749         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 750          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 751          need to make the distinction between the 2 versions */
 752         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 753                 char *param = NULL;
 754                 char *buf;
 755
 756                 param = kzalloc(MGS_PARAM_MAXLEN, GFP_NOFS);
 757                 if (!param)
 758                         return -ENOMEM;
 759
 760                 buf = param;
 761                 /* Get fsname and assume devname to be -MDT0000. */
 762                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 763                 strcat(buf, "-MDT0000.lov");
 764                 buf += strlen(buf);
 765
 766                 /* Set root stripesize */
 767                 sprintf(buf, ".stripesize=%u",
 768                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 769                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 770                 if (rc)
 771                         goto end;
 772
 773                 /* Set root stripecount */
 774                 sprintf(buf, ".stripecount=%hd",
 775                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 776                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 777                 if (rc)
 778                         goto end;
 779
 780                 /* Set root stripeoffset */
 781                 sprintf(buf, ".stripeoffset=%hd",
 782                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 783                         (typeof(lump->lmm_stripe_offset))(-1));
 784                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 785
 786 end:
 787                 kfree(param);
 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         kfree(copy);
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                         kfree(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                         kfree(oqctl_tmp);
1207                 }
1208 out:
1209                 QCTL_COPY(qctl, oqctl);
1210                 kfree(oqctl);
1211         }
1212
1213         return rc;
1214 }
1215
1216 /* This function tries to get a single name component,
1217  * to send to the server. No actual path traversal involved,
1218  * so we limit to NAME_MAX */
1219 static char *ll_getname(const char __user *filename)
1220 {
1221         int ret = 0, len;
1222         char *tmp;
1223
1224         tmp = kzalloc(NAME_MAX + 1, GFP_KERNEL);
1225         if (!tmp)
1226                 return ERR_PTR(-ENOMEM);
1227
1228         len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1229         if (len < 0)
1230                 ret = len;
1231         else if (len == 0)
1232                 ret = -ENOENT;
1233         else if (len > NAME_MAX && tmp[NAME_MAX] != 0)
1234                 ret = -ENAMETOOLONG;
1235
1236         if (ret) {
1237                 kfree(tmp);
1238                 tmp =  ERR_PTR(ret);
1239         }
1240         return tmp;
1241 }
1242
1243 #define ll_putname(filename) kfree(filename)
1244
1245 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1246 {
1247         struct inode *inode = file_inode(file);
1248         struct ll_sb_info *sbi = ll_i2sbi(inode);
1249         struct obd_ioctl_data *data;
1250         int rc = 0;
1251
1252         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1253                inode->i_ino, inode->i_generation, inode, cmd);
1254
1255         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1256         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1257                 return -ENOTTY;
1258
1259         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1260         switch (cmd) {
1261         case FSFILT_IOC_GETFLAGS:
1262         case FSFILT_IOC_SETFLAGS:
1263                 return ll_iocontrol(inode, file, cmd, arg);
1264         case FSFILT_IOC_GETVERSION_OLD:
1265         case FSFILT_IOC_GETVERSION:
1266                 return put_user(inode->i_generation, (int *)arg);
1267         /* We need to special case any other ioctls we want to handle,
1268          * to send them to the MDS/OST as appropriate and to properly
1269          * network encode the arg field.
1270         case FSFILT_IOC_SETVERSION_OLD:
1271         case FSFILT_IOC_SETVERSION:
1272         */
1273         case LL_IOC_GET_MDTIDX: {
1274                 int mdtidx;
1275
1276                 mdtidx = ll_get_mdt_idx(inode);
1277                 if (mdtidx < 0)
1278                         return mdtidx;
1279
1280                 if (put_user((int)mdtidx, (int *)arg))
1281                         return -EFAULT;
1282
1283                 return 0;
1284         }
1285         case IOC_MDC_LOOKUP: {
1286                 struct ptlrpc_request *request = NULL;
1287                 int namelen, len = 0;
1288                 char *buf = NULL;
1289                 char *filename;
1290                 struct md_op_data *op_data;
1291
1292                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1293                 if (rc)
1294                         return rc;
1295                 data = (void *)buf;
1296
1297                 filename = data->ioc_inlbuf1;
1298                 namelen = strlen(filename);
1299
1300                 if (namelen < 1) {
1301                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1302                         rc = -EINVAL;
1303                         goto out_free;
1304                 }
1305
1306                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1307                                              0, LUSTRE_OPC_ANY, NULL);
1308                 if (IS_ERR(op_data)) {
1309                         rc = PTR_ERR(op_data);
1310                         goto out_free;
1311                 }
1312
1313                 op_data->op_valid = OBD_MD_FLID;
1314                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1315                 ll_finish_md_op_data(op_data);
1316                 if (rc < 0) {
1317                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1318                         goto out_free;
1319                 }
1320                 ptlrpc_req_finished(request);
1321 out_free:
1322                 obd_ioctl_freedata(buf, len);
1323                 return rc;
1324         }
1325         case LL_IOC_LMV_SETSTRIPE: {
1326                 struct lmv_user_md  *lum;
1327                 char            *buf = NULL;
1328                 char            *filename;
1329                 int              namelen = 0;
1330                 int              lumlen = 0;
1331                 int              len;
1332                 int              rc;
1333
1334                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1335                 if (rc)
1336                         return rc;
1337
1338                 data = (void *)buf;
1339                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1340                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0) {
1341                         rc = -EINVAL;
1342                         goto lmv_out_free;
1343                 }
1344
1345                 filename = data->ioc_inlbuf1;
1346                 namelen = data->ioc_inllen1;
1347
1348                 if (namelen < 1) {
1349                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1350                         rc = -EINVAL;
1351                         goto lmv_out_free;
1352                 }
1353                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1354                 lumlen = data->ioc_inllen2;
1355
1356                 if (lum->lum_magic != LMV_USER_MAGIC ||
1357                     lumlen != sizeof(*lum)) {
1358                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1359                                filename, lum->lum_magic, lumlen, -EFAULT);
1360                         rc = -EINVAL;
1361                         goto lmv_out_free;
1362                 }
1363
1364                 /**
1365                  * ll_dir_setdirstripe will be used to set dir stripe
1366                  *  mdc_create--->mdt_reint_create (with dirstripe)
1367                  */
1368                 rc = ll_dir_setdirstripe(inode, lum, filename);
1369 lmv_out_free:
1370                 obd_ioctl_freedata(buf, len);
1371                 return rc;
1372
1373         }
1374         case LL_IOC_LOV_SETSTRIPE: {
1375                 struct lov_user_md_v3 lumv3;
1376                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1377                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1378                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1379
1380                 int set_default = 0;
1381
1382                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1383                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1384                         sizeof(lumv3p->lmm_objects[0]));
1385                 /* first try with v1 which is smaller than v3 */
1386                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1387                         return -EFAULT;
1388
1389                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1390                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1391                                 return -EFAULT;
1392                 }
1393
1394                 if (is_root_inode(inode))
1395                         set_default = 1;
1396
1397                 /* in v1 and v3 cases lumv1 points to data */
1398                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1399
1400                 return rc;
1401         }
1402         case LL_IOC_LMV_GETSTRIPE: {
1403                 struct lmv_user_md *lump = (struct lmv_user_md *)arg;
1404                 struct lmv_user_md lum;
1405                 struct lmv_user_md *tmp;
1406                 int lum_size;
1407                 int rc = 0;
1408                 int mdtindex;
1409
1410                 if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
1411                         return -EFAULT;
1412
1413                 if (lum.lum_magic != LMV_MAGIC_V1)
1414                         return -EINVAL;
1415
1416                 lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
1417                 tmp = kzalloc(lum_size, GFP_NOFS);
1418                 if (!tmp) {
1419                         rc = -ENOMEM;
1420                         goto free_lmv;
1421                 }
1422
1423                 *tmp = lum;
1424                 tmp->lum_type = LMV_STRIPE_TYPE;
1425                 tmp->lum_stripe_count = 1;
1426                 mdtindex = ll_get_mdt_idx(inode);
1427                 if (mdtindex < 0) {
1428                         rc = -ENOMEM;
1429                         goto free_lmv;
1430                 }
1431
1432                 tmp->lum_stripe_offset = mdtindex;
1433                 tmp->lum_objects[0].lum_mds = mdtindex;
1434                 memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
1435                        sizeof(struct lu_fid));
1436                 if (copy_to_user((void *)arg, tmp, lum_size)) {
1437                         rc = -EFAULT;
1438                         goto free_lmv;
1439                 }
1440 free_lmv:
1441                 kfree(tmp);
1442                 return rc;
1443         }
1444         case LL_IOC_LOV_SWAP_LAYOUTS:
1445                 return -EPERM;
1446         case LL_IOC_OBD_STATFS:
1447                 return ll_obd_statfs(inode, (void *)arg);
1448         case LL_IOC_LOV_GETSTRIPE:
1449         case LL_IOC_MDC_GETINFO:
1450         case IOC_MDC_GETFILEINFO:
1451         case IOC_MDC_GETFILESTRIPE: {
1452                 struct ptlrpc_request *request = NULL;
1453                 struct lov_user_md *lump;
1454                 struct lov_mds_md *lmm = NULL;
1455                 struct mdt_body *body;
1456                 char *filename = NULL;
1457                 int lmmsize;
1458
1459                 if (cmd == IOC_MDC_GETFILEINFO ||
1460                     cmd == IOC_MDC_GETFILESTRIPE) {
1461                         filename = ll_getname((const char *)arg);
1462                         if (IS_ERR(filename))
1463                                 return PTR_ERR(filename);
1464
1465                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1466                                                       &lmmsize, &request);
1467                 } else {
1468                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1469                 }
1470
1471                 if (request) {
1472                         body = req_capsule_server_get(&request->rq_pill,
1473                                                       &RMF_MDT_BODY);
1474                         LASSERT(body != NULL);
1475                 } else {
1476                         goto out_req;
1477                 }
1478
1479                 if (rc < 0) {
1480                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1481                                                cmd == LL_IOC_MDC_GETINFO)) {
1482                                 rc = 0;
1483                                 goto skip_lmm;
1484                         } else
1485                                 goto out_req;
1486                 }
1487
1488                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1489                     cmd == LL_IOC_LOV_GETSTRIPE) {
1490                         lump = (struct lov_user_md *)arg;
1491                 } else {
1492                         struct lov_user_mds_data *lmdp;
1493
1494                         lmdp = (struct lov_user_mds_data *)arg;
1495                         lump = &lmdp->lmd_lmm;
1496                 }
1497                 if (copy_to_user(lump, lmm, lmmsize)) {
1498                         if (copy_to_user(lump, lmm, sizeof(*lump))) {
1499                                 rc = -EFAULT;
1500                                 goto out_req;
1501                         }
1502                         rc = -EOVERFLOW;
1503                 }
1504 skip_lmm:
1505                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1506                         struct lov_user_mds_data *lmdp;
1507                         lstat_t st = { 0 };
1508
1509                         st.st_dev     = inode->i_sb->s_dev;
1510                         st.st_mode    = body->mode;
1511                         st.st_nlink   = body->nlink;
1512                         st.st_uid     = body->uid;
1513                         st.st_gid     = body->gid;
1514                         st.st_rdev    = body->rdev;
1515                         st.st_size    = body->size;
1516                         st.st_blksize = PAGE_CACHE_SIZE;
1517                         st.st_blocks  = body->blocks;
1518                         st.st_atime   = body->atime;
1519                         st.st_mtime   = body->mtime;
1520                         st.st_ctime   = body->ctime;
1521                         st.st_ino     = inode->i_ino;
1522
1523                         lmdp = (struct lov_user_mds_data *)arg;
1524                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st))) {
1525                                 rc = -EFAULT;
1526                                 goto out_req;
1527                         }
1528                 }
1529
1530 out_req:
1531                 ptlrpc_req_finished(request);
1532                 if (filename)
1533                         ll_putname(filename);
1534                 return rc;
1535         }
1536         case IOC_LOV_GETINFO: {
1537                 struct lov_user_mds_data *lumd;
1538                 struct lov_stripe_md *lsm;
1539                 struct lov_user_md *lum;
1540                 struct lov_mds_md *lmm;
1541                 int lmmsize;
1542                 lstat_t st;
1543
1544                 lumd = (struct lov_user_mds_data *)arg;
1545                 lum = &lumd->lmd_lmm;
1546
1547                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1548                 if (rc)
1549                         return rc;
1550
1551                 lmm = libcfs_kvzalloc(lmmsize, GFP_NOFS);
1552                 if (lmm == NULL)
1553                         return -ENOMEM;
1554                 if (copy_from_user(lmm, lum, lmmsize)) {
1555                         rc = -EFAULT;
1556                         goto free_lmm;
1557                 }
1558
1559                 switch (lmm->lmm_magic) {
1560                 case LOV_USER_MAGIC_V1:
1561                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1562                                 break;
1563                         /* swab objects first so that stripes num will be sane */
1564                         lustre_swab_lov_user_md_objects(
1565                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1566                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1567                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1568                         break;
1569                 case LOV_USER_MAGIC_V3:
1570                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1571                                 break;
1572                         /* swab objects first so that stripes num will be sane */
1573                         lustre_swab_lov_user_md_objects(
1574                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1575                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1576                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1577                         break;
1578                 default:
1579                         rc = -EINVAL;
1580                         goto free_lmm;
1581                 }
1582
1583                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1584                 if (rc < 0) {
1585                         rc = -ENOMEM;
1586                         goto free_lmm;
1587                 }
1588
1589                 /* Perform glimpse_size operation. */
1590                 memset(&st, 0, sizeof(st));
1591
1592                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1593                 if (rc)
1594                         goto free_lsm;
1595
1596                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st))) {
1597                         rc = -EFAULT;
1598                         goto free_lsm;
1599                 }
1600
1601 free_lsm:
1602                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1603 free_lmm:
1604                 kvfree(lmm);
1605                 return rc;
1606         }
1607         case OBD_IOC_LLOG_CATINFO: {
1608                 return -EOPNOTSUPP;
1609         }
1610         case OBD_IOC_QUOTACHECK: {
1611                 struct obd_quotactl *oqctl;
1612                 int error = 0;
1613
1614                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1615                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1616                         return -EPERM;
1617
1618                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1619                 if (!oqctl)
1620                         return -ENOMEM;
1621                 oqctl->qc_type = arg;
1622                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1623                 if (rc < 0) {
1624                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1625                         error = rc;
1626                 }
1627
1628                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1629                 if (rc < 0)
1630                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1631
1632                 kfree(oqctl);
1633                 return error ?: rc;
1634         }
1635         case OBD_IOC_POLL_QUOTACHECK: {
1636                 struct if_quotacheck *check;
1637
1638                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1639                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1640                         return -EPERM;
1641
1642                 check = kzalloc(sizeof(*check), GFP_NOFS);
1643                 if (!check)
1644                         return -ENOMEM;
1645
1646                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1647                                    NULL);
1648                 if (rc) {
1649                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1650                         if (copy_to_user((void *)arg, check,
1651                                              sizeof(*check)))
1652                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1653                         goto out_poll;
1654                 }
1655
1656                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1657                                    NULL);
1658                 if (rc) {
1659                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1660                         if (copy_to_user((void *)arg, check,
1661                                              sizeof(*check)))
1662                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1663                         goto out_poll;
1664                 }
1665 out_poll:
1666                 kfree(check);
1667                 return rc;
1668         }
1669         case LL_IOC_QUOTACTL: {
1670                 struct if_quotactl *qctl;
1671
1672                 qctl = kzalloc(sizeof(*qctl), GFP_NOFS);
1673                 if (!qctl)
1674                         return -ENOMEM;
1675
1676                 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl))) {
1677                         rc = -EFAULT;
1678                         goto out_quotactl;
1679                 }
1680
1681                 rc = quotactl_ioctl(sbi, qctl);
1682
1683                 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1684                         rc = -EFAULT;
1685
1686 out_quotactl:
1687                 kfree(qctl);
1688                 return rc;
1689         }
1690         case OBD_IOC_GETDTNAME:
1691         case OBD_IOC_GETMDNAME:
1692                 return ll_get_obd_name(inode, cmd, arg);
1693         case LL_IOC_FLUSHCTX:
1694                 return ll_flush_ctx(inode);
1695 #ifdef CONFIG_FS_POSIX_ACL
1696         case LL_IOC_RMTACL: {
1697             if (sbi->ll_flags & LL_SBI_RMT_CLIENT && is_root_inode(inode)) {
1698                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1699
1700                 LASSERT(fd != NULL);
1701                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1702                 if (!rc)
1703                         fd->fd_flags |= LL_FILE_RMTACL;
1704                 return rc;
1705             } else
1706                 return 0;
1707         }
1708 #endif
1709         case LL_IOC_GETOBDCOUNT: {
1710                 int count, vallen;
1711                 struct obd_export *exp;
1712
1713                 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1714                         return -EFAULT;
1715
1716                 /* get ost count when count is zero, get mdt count otherwise */
1717                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1718                 vallen = sizeof(count);
1719                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1720                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1721                 if (rc) {
1722                         CERROR("get target count failed: %d\n", rc);
1723                         return rc;
1724                 }
1725
1726                 if (copy_to_user((int *)arg, &count, sizeof(int)))
1727                         return -EFAULT;
1728
1729                 return 0;
1730         }
1731         case LL_IOC_PATH2FID:
1732                 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1733                                      sizeof(struct lu_fid)))
1734                         return -EFAULT;
1735                 return 0;
1736         case LL_IOC_GET_CONNECT_FLAGS: {
1737                 return obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void *)arg);
1738         }
1739         case OBD_IOC_CHANGELOG_SEND:
1740         case OBD_IOC_CHANGELOG_CLEAR:
1741                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1742                                     sizeof(struct ioc_changelog));
1743                 return rc;
1744         case OBD_IOC_FID2PATH:
1745                 return ll_fid2path(inode, (void *)arg);
1746         case LL_IOC_HSM_REQUEST: {
1747                 struct hsm_user_request *hur;
1748                 ssize_t                  totalsize;
1749
1750                 hur = memdup_user((void *)arg, sizeof(*hur));
1751                 if (IS_ERR(hur))
1752                         return PTR_ERR(hur);
1753
1754                 /* Compute the whole struct size */
1755                 totalsize = hur_len(hur);
1756                 kfree(hur);
1757                 if (totalsize < 0)
1758                         return -E2BIG;
1759
1760                 /* Final size will be more than double totalsize */
1761                 if (totalsize >= MDS_MAXREQSIZE / 3)
1762                         return -E2BIG;
1763
1764                 hur = libcfs_kvzalloc(totalsize, GFP_NOFS);
1765                 if (hur == NULL)
1766                         return -ENOMEM;
1767
1768                 /* Copy the whole struct */
1769                 if (copy_from_user(hur, (void *)arg, totalsize)) {
1770                         kvfree(hur);
1771                         return -EFAULT;
1772                 }
1773
1774                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1775                         const struct lu_fid *fid;
1776                         struct inode *f;
1777                         int i;
1778
1779                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1780                                 fid = &hur->hur_user_item[i].hui_fid;
1781                                 f = search_inode_for_lustre(inode->i_sb, fid);
1782                                 if (IS_ERR(f)) {
1783                                         rc = PTR_ERR(f);
1784                                         break;
1785                                 }
1786
1787                                 rc = ll_hsm_release(f);
1788                                 iput(f);
1789                                 if (rc != 0)
1790                                         break;
1791                         }
1792                 } else {
1793                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1794                                            hur, NULL);
1795                 }
1796
1797                 kvfree(hur);
1798
1799                 return rc;
1800         }
1801         case LL_IOC_HSM_PROGRESS: {
1802                 struct hsm_progress_kernel      hpk;
1803                 struct hsm_progress             hp;
1804
1805                 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1806                         return -EFAULT;
1807
1808                 hpk.hpk_fid = hp.hp_fid;
1809                 hpk.hpk_cookie = hp.hp_cookie;
1810                 hpk.hpk_extent = hp.hp_extent;
1811                 hpk.hpk_flags = hp.hp_flags;
1812                 hpk.hpk_errval = hp.hp_errval;
1813                 hpk.hpk_data_version = 0;
1814
1815                 /* File may not exist in Lustre; all progress
1816                  * reported to Lustre root */
1817                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1818                                    NULL);
1819                 return rc;
1820         }
1821         case LL_IOC_HSM_CT_START:
1822                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1823                                     sizeof(struct lustre_kernelcomm));
1824                 return rc;
1825
1826         case LL_IOC_HSM_COPY_START: {
1827                 struct hsm_copy *copy;
1828                 int              rc;
1829
1830                 copy = memdup_user((char *)arg, sizeof(*copy));
1831                 if (IS_ERR(copy))
1832                         return PTR_ERR(copy);
1833
1834                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1835                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1836                         rc = -EFAULT;
1837
1838                 kfree(copy);
1839                 return rc;
1840         }
1841         case LL_IOC_HSM_COPY_END: {
1842                 struct hsm_copy *copy;
1843                 int              rc;
1844
1845                 copy = memdup_user((char *)arg, sizeof(*copy));
1846                 if (IS_ERR(copy))
1847                         return PTR_ERR(copy);
1848
1849                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1850                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1851                         rc = -EFAULT;
1852
1853                 kfree(copy);
1854                 return rc;
1855         }
1856         default:
1857                 return obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL, (void *)arg);
1858         }
1859 }
1860
1861 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1862 {
1863         struct inode *inode = file->f_mapping->host;
1864         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1865         struct ll_sb_info *sbi = ll_i2sbi(inode);
1866         int api32 = ll_need_32bit_api(sbi);
1867         loff_t ret = -EINVAL;
1868
1869         mutex_lock(&inode->i_mutex);
1870         switch (origin) {
1871         case SEEK_SET:
1872                 break;
1873         case SEEK_CUR:
1874                 offset += file->f_pos;
1875                 break;
1876         case SEEK_END:
1877                 if (offset > 0)
1878                         goto out;
1879                 if (api32)
1880                         offset += LL_DIR_END_OFF_32BIT;
1881                 else
1882                         offset += LL_DIR_END_OFF;
1883                 break;
1884         default:
1885                 goto out;
1886         }
1887
1888         if (offset >= 0 &&
1889             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1890              (!api32 && offset <= LL_DIR_END_OFF))) {
1891                 if (offset != file->f_pos) {
1892                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1893                             (!api32 && offset == LL_DIR_END_OFF))
1894                                 fd->lfd_pos = MDS_DIR_END_OFF;
1895                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1896                                 fd->lfd_pos = offset << 32;
1897                         else
1898                                 fd->lfd_pos = offset;
1899                         file->f_pos = offset;
1900                         file->f_version = 0;
1901                 }
1902                 ret = offset;
1903         }
1904         goto out;
1905
1906 out:
1907         mutex_unlock(&inode->i_mutex);
1908         return ret;
1909 }
1910
1911 static int ll_dir_open(struct inode *inode, struct file *file)
1912 {
1913         return ll_file_open(inode, file);
1914 }
1915
1916 static int ll_dir_release(struct inode *inode, struct file *file)
1917 {
1918         return ll_file_release(inode, file);
1919 }
1920
1921 const struct file_operations ll_dir_operations = {
1922         .llseek   = ll_dir_seek,
1923         .open     = ll_dir_open,
1924         .release  = ll_dir_release,
1925         .read     = generic_read_dir,
1926         .iterate  = ll_readdir,
1927         .unlocked_ioctl   = ll_dir_ioctl,
1928         .fsync    = ll_fsync,
1929 };