drivers/staging/lustre/lustre/llite/dir.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2015, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/dir.c
  37  *
  38  * Directory code for lustre client.
  39  */
  40
  41 #include <linux/fs.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/mm.h>
  44 #include <linux/uaccess.h>
  45 #include <linux/buffer_head.h>   /* for wait_on_buffer */
  46 #include <linux/pagevec.h>
  47 #include <linux/prefetch.h>
  48
  49 #define DEBUG_SUBSYSTEM S_LLITE
  50
  51 #include "../include/obd_support.h"
  52 #include "../include/obd_class.h"
  53 #include "../include/lustre_lib.h"
  54 #include "../include/lustre/lustre_idl.h"
  55 #include "../include/lustre_lite.h"
  56 #include "../include/lustre_dlm.h"
  57 #include "../include/lustre_fid.h"
  58 #include "../include/lustre_kernelcomm.h"
  59 #include "llite_internal.h"
  60
  61 /*
  62  * (new) readdir implementation overview.
  63  *
  64  * Original lustre readdir implementation cached exact copy of raw directory
  65  * pages on the client. These pages were indexed in client page cache by
  66  * logical offset in the directory file. This design, while very simple and
  67  * intuitive had some inherent problems:
  68  *
  69  *     . it implies that byte offset to the directory entry serves as a
  70  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  71  *     ext3/htree directory entries may move due to splits, and more
  72  *     importantly,
  73  *
  74  *     . it is incompatible with the design of split directories for cmd3,
  75  *     that assumes that names are distributed across nodes based on their
  76  *     hash, and so readdir should be done in hash order.
  77  *
  78  * New readdir implementation does readdir in hash order, and uses hash of a
  79  * file name as a telldir/seekdir cookie. This led to number of complications:
  80  *
  81  *     . hash is not unique, so it cannot be used to index cached directory
  82  *     pages on the client (note, that it requires a whole pageful of hash
  83  *     collided entries to cause two pages to have identical hashes);
  84  *
  85  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  86  *     entry cookie. ext3/htree has the same problem and lustre implementation
  87  *     mimics their solution: seekdir(hash) positions directory at the first
  88  *     entry with the given hash.
  89  *
  90  * Client side.
  91  *
  92  * 0. caching
  93  *
  94  * Client caches directory pages using hash of the first entry as an index. As
  95  * noted above hash is not unique, so this solution doesn't work as is:
  96  * special processing is needed for "page hash chains" (i.e., sequences of
  97  * pages filled with entries all having the same hash value).
  98  *
  99  * First, such chains have to be detected. To this end, server returns to the
 100  * client the hash of the first entry on the page next to one returned. When
 101  * client detects that this hash is the same as hash of the first entry on the
 102  * returned page, page hash collision has to be handled. Pages in the
 103  * hash chain, except first one, are termed "overflow pages".
 104  *
 105  * Solution to index uniqueness problem is to not cache overflow
 106  * pages. Instead, when page hash collision is detected, all overflow pages
 107  * from emerging chain are immediately requested from the server and placed in
 108  * a special data structure (struct ll_dir_chain). This data structure is used
 109  * by ll_readdir() to process entries from overflow pages. When readdir
 110  * invocation finishes, overflow pages are discarded. If page hash collision
 111  * chain weren't completely processed, next call to readdir will again detect
 112  * page hash collision, again read overflow pages in, process next portion of
 113  * entries and again discard the pages. This is not as wasteful as it looks,
 114  * because, given reasonable hash, page hash collisions are extremely rare.
 115  *
 116  * 1. directory positioning
 117  *
 118  * When seekdir(hash) is called, original
 119  *
 120  *
 121  *
 122  *
 123  *
 124  *
 125  *
 126  *
 127  * Server.
 128  *
 129  * identification of and access to overflow pages
 130  *
 131  * page format
 132  *
 133  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 134  * a header lu_dirpage which describes the start/end hash, and whether this
 135  * page is empty (contains no dir entry) or hash collide with next page.
 136  * After client receives reply, several pages will be integrated into dir page
 137  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the lu_dirpage
 138  * for this integrated page will be adjusted. See lmv_adjust_dirpages().
 139  *
 140  */
 141
 142 /* returns the page unlocked, but with a reference */
 143 static int ll_dir_filler(void *_hash, struct page *page0)
 144 {
 145         struct inode *inode = page0->mapping->host;
 146         int hash64 = ll_i2sbi(inode)->ll_flags & LL_SBI_64BIT_HASH;
 147         struct obd_export *exp = ll_i2sbi(inode)->ll_md_exp;
 148         struct ptlrpc_request *request;
 149         struct mdt_body *body;
 150         struct md_op_data *op_data;
 151         __u64 hash = *((__u64 *)_hash);
 152         struct page **page_pool;
 153         struct page *page;
 154         struct lu_dirpage *dp;
 155         int max_pages = ll_i2sbi(inode)->ll_md_brw_size >> PAGE_SHIFT;
 156         int nrdpgs = 0; /* number of pages read actually */
 157         int npages;
 158         int i;
 159         int rc;
 160
 161         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) hash %llu\n",
 162                inode->i_ino, inode->i_generation, inode, hash);
 163
 164         LASSERT(max_pages > 0 && max_pages <= MD_MAX_BRW_PAGES);
 165
 166         page_pool = kcalloc(max_pages, sizeof(page), GFP_NOFS);
 167         if (page_pool) {
 168                 page_pool[0] = page0;
 169         } else {
 170                 page_pool = &page0;
 171                 max_pages = 1;
 172         }
 173         for (npages = 1; npages < max_pages; npages++) {
 174                 page = page_cache_alloc_cold(inode->i_mapping);
 175                 if (!page)
 176                         break;
 177                 page_pool[npages] = page;
 178         }
 179
 180         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 181                                      LUSTRE_OPC_ANY, NULL);
 182         op_data->op_npages = npages;
 183         op_data->op_offset = hash;
 184         rc = md_readpage(exp, op_data, page_pool, &request);
 185         ll_finish_md_op_data(op_data);
 186         if (rc < 0) {
 187                 /* page0 is special, which was added into page cache early */
 188                 delete_from_page_cache(page0);
 189         } else if (rc == 0) {
 190                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 191                 /* Checked by mdc_readpage() */
 192                 if (body->valid & OBD_MD_FLSIZE)
 193                         i_size_write(inode, body->size);
 194
 195                 nrdpgs = (request->rq_bulk->bd_nob_transferred+PAGE_SIZE-1)
 196                          >> PAGE_SHIFT;
 197                 SetPageUptodate(page0);
 198         }
 199         unlock_page(page0);
 200         ptlrpc_req_finished(request);
 201
 202         CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages);
 203
 204         for (i = 1; i < npages; i++) {
 205                 unsigned long offset;
 206                 int ret;
 207
 208                 page = page_pool[i];
 209
 210                 if (rc < 0 || i >= nrdpgs) {
 211                         put_page(page);
 212                         continue;
 213                 }
 214
 215                 SetPageUptodate(page);
 216
 217                 dp = kmap(page);
 218                 hash = le64_to_cpu(dp->ldp_hash_start);
 219                 kunmap(page);
 220
 221                 offset = hash_x_index(hash, hash64);
 222
 223                 prefetchw(&page->flags);
 224                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
 225                                             GFP_NOFS);
 226                 if (ret == 0) {
 227                         unlock_page(page);
 228                 } else {
 229                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed: %d\n",
 230                                offset, ret);
 231                 }
 232                 put_page(page);
 233         }
 234
 235         if (page_pool != &page0)
 236                 kfree(page_pool);
 237         return rc;
 238 }
 239
 240 void ll_release_page(struct page *page, int remove)
 241 {
 242         kunmap(page);
 243         if (remove) {
 244                 lock_page(page);
 245                 if (likely(page->mapping))
 246                         truncate_complete_page(page->mapping, page);
 247                 unlock_page(page);
 248         }
 249         put_page(page);
 250 }
 251
 252 /*
 253  * Find, kmap and return page that contains given hash.
 254  */
 255 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
 256                                        __u64 *start, __u64 *end)
 257 {
 258         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 259         struct address_space *mapping = dir->i_mapping;
 260         /*
 261          * Complement of hash is used as an index so that
 262          * radix_tree_gang_lookup() can be used to find a page with starting
 263          * hash _smaller_ than one we are looking for.
 264          */
 265         unsigned long offset = hash_x_index(*hash, hash64);
 266         struct page *page;
 267         int found;
 268
 269         spin_lock_irq(&mapping->tree_lock);
 270         found = radix_tree_gang_lookup(&mapping->page_tree,
 271                                        (void **)&page, offset, 1);
 272         if (found > 0 && !radix_tree_exceptional_entry(page)) {
 273                 struct lu_dirpage *dp;
 274
 275                 get_page(page);
 276                 spin_unlock_irq(&mapping->tree_lock);
 277                 /*
 278                  * In contrast to find_lock_page() we are sure that directory
 279                  * page cannot be truncated (while DLM lock is held) and,
 280                  * hence, can avoid restart.
 281                  *
 282                  * In fact, page cannot be locked here at all, because
 283                  * ll_dir_filler() does synchronous io.
 284                  */
 285                 wait_on_page_locked(page);
 286                 if (PageUptodate(page)) {
 287                         dp = kmap(page);
 288                         if (BITS_PER_LONG == 32 && hash64) {
 289                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 290                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 291                                 *hash  = *hash >> 32;
 292                         } else {
 293                                 *start = le64_to_cpu(dp->ldp_hash_start);
 294                                 *end   = le64_to_cpu(dp->ldp_hash_end);
 295                         }
 296                         LASSERTF(*start <= *hash, "start = %#llx,end = %#llx,hash = %#llx\n",
 297                                  *start, *end, *hash);
 298                         CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash %llu\n",
 299                                offset, *start, *end, *hash);
 300                         if (*hash > *end) {
 301                                 ll_release_page(page, 0);
 302                                 page = NULL;
 303                         } else if (*end != *start && *hash == *end) {
 304                                 /*
 305                                  * upon hash collision, remove this page,
 306                                  * otherwise put page reference, and
 307                                  * ll_get_dir_page() will issue RPC to fetch
 308                                  * the page we want.
 309                                  */
 310                                 ll_release_page(page,
 311                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
 312                                 page = NULL;
 313                         }
 314                 } else {
 315                         put_page(page);
 316                         page = ERR_PTR(-EIO);
 317                 }
 318
 319         } else {
 320                 spin_unlock_irq(&mapping->tree_lock);
 321                 page = NULL;
 322         }
 323         return page;
 324 }
 325
 326 struct page *ll_get_dir_page(struct inode *dir, __u64 hash,
 327                              struct ll_dir_chain *chain)
 328 {
 329         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 330         struct address_space *mapping = dir->i_mapping;
 331         struct lustre_handle lockh;
 332         struct lu_dirpage *dp;
 333         struct page *page;
 334         enum ldlm_mode mode;
 335         int rc;
 336         __u64 start = 0;
 337         __u64 end = 0;
 338         __u64 lhash = hash;
 339         struct ll_inode_info *lli = ll_i2info(dir);
 340         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 341
 342         mode = LCK_PR;
 343         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 344                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
 345         if (!rc) {
 346                 struct ldlm_enqueue_info einfo = {
 347                         .ei_type = LDLM_IBITS,
 348                         .ei_mode = mode,
 349                         .ei_cb_bl = ll_md_blocking_ast,
 350                         .ei_cb_cp = ldlm_completion_ast,
 351                 };
 352                 struct lookup_intent it = { .it_op = IT_READDIR };
 353                 struct ptlrpc_request *request;
 354                 struct md_op_data *op_data;
 355
 356                 op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0,
 357                                              LUSTRE_OPC_ANY, NULL);
 358                 if (IS_ERR(op_data))
 359                         return (void *)op_data;
 360
 361                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
 362                                 op_data, &lockh, NULL, 0, NULL, 0);
 363
 364                 ll_finish_md_op_data(op_data);
 365
 366                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 367                 if (request)
 368                         ptlrpc_req_finished(request);
 369                 if (rc < 0) {
 370                         CERROR("lock enqueue: " DFID " at %llu: rc %d\n",
 371                                PFID(ll_inode2fid(dir)), hash, rc);
 372                         return ERR_PTR(rc);
 373                 }
 374
 375                 CDEBUG(D_INODE, "setting lr_lvb_inode to inode %p (%lu/%u)\n",
 376                        dir, dir->i_ino, dir->i_generation);
 377                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp,
 378                                  &it.d.lustre.it_lock_handle, dir, NULL);
 379         } else {
 380                 /* for cross-ref object, l_ast_data of the lock may not be set,
 381                  * we reset it here
 382                  */
 383                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
 384                                  dir, NULL);
 385         }
 386         ldlm_lock_dump_handle(D_OTHER, &lockh);
 387
 388         mutex_lock(&lli->lli_readdir_mutex);
 389         page = ll_dir_page_locate(dir, &lhash, &start, &end);
 390         if (IS_ERR(page)) {
 391                 CERROR("dir page locate: "DFID" at %llu: rc %ld\n",
 392                        PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
 393                 goto out_unlock;
 394         } else if (page) {
 395                 /*
 396                  * XXX nikita: not entirely correct handling of a corner case:
 397                  * suppose hash chain of entries with hash value HASH crosses
 398                  * border between pages P0 and P1. First both P0 and P1 are
 399                  * cached, seekdir() is called for some entry from the P0 part
 400                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 401                  * happens and finds P1, as it starts with matching hash
 402                  * value. Remaining entries from P0 part of the chain are
 403                  * skipped. (Is that really a bug?)
 404                  *
 405                  * Possible solutions: 0. don't cache P1 is such case, handle
 406                  * it as an "overflow" page. 1. invalidate all pages at
 407                  * once. 2. use HASH|1 as an index for P1.
 408                  */
 409                 goto hash_collision;
 410         }
 411
 412         page = read_cache_page(mapping, hash_x_index(hash, hash64),
 413                                ll_dir_filler, &lhash);
 414         if (IS_ERR(page)) {
 415                 CERROR("read cache page: "DFID" at %llu: rc %ld\n",
 416                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 417                 goto out_unlock;
 418         }
 419
 420         wait_on_page_locked(page);
 421         (void)kmap(page);
 422         if (!PageUptodate(page)) {
 423                 CERROR("page not updated: "DFID" at %llu: rc %d\n",
 424                        PFID(ll_inode2fid(dir)), hash, -5);
 425                 goto fail;
 426         }
 427         if (!PageChecked(page))
 428                 /* XXX: check page format later */
 429                 SetPageChecked(page);
 430         if (PageError(page)) {
 431                 CERROR("page error: "DFID" at %llu: rc %d\n",
 432                        PFID(ll_inode2fid(dir)), hash, -5);
 433                 goto fail;
 434         }
 435 hash_collision:
 436         dp = page_address(page);
 437         if (BITS_PER_LONG == 32 && hash64) {
 438                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 439                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 440                 lhash = hash >> 32;
 441         } else {
 442                 start = le64_to_cpu(dp->ldp_hash_start);
 443                 end   = le64_to_cpu(dp->ldp_hash_end);
 444                 lhash = hash;
 445         }
 446         if (end == start) {
 447                 LASSERT(start == lhash);
 448                 CWARN("Page-wide hash collision: %llu\n", end);
 449                 if (BITS_PER_LONG == 32 && hash64)
 450                         CWARN("Real page-wide hash collision at [%llu %llu] with hash %llu\n",
 451                               le64_to_cpu(dp->ldp_hash_start),
 452                               le64_to_cpu(dp->ldp_hash_end), hash);
 453                 /*
 454                  * Fetch whole overflow chain...
 455                  *
 456                  * XXX not yet.
 457                  */
 458                 goto fail;
 459         }
 460 out_unlock:
 461         mutex_unlock(&lli->lli_readdir_mutex);
 462         ldlm_lock_decref(&lockh, mode);
 463         return page;
 464
 465 fail:
 466         ll_release_page(page, 1);
 467         page = ERR_PTR(-EIO);
 468         goto out_unlock;
 469 }
 470
 471 /**
 472  * return IF_* type for given lu_dirent entry.
 473  * IF_* flag shld be converted to particular OS file type in
 474  * platform llite module.
 475  */
 476 static __u16 ll_dirent_type_get(struct lu_dirent *ent)
 477 {
 478         __u16 type = 0;
 479         struct luda_type *lt;
 480         int len = 0;
 481
 482         if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
 483                 const unsigned int align = sizeof(struct luda_type) - 1;
 484
 485                 len = le16_to_cpu(ent->lde_namelen);
 486                 len = (len + align) & ~align;
 487                 lt = (void *)ent->lde_name + len;
 488                 type = IFTODT(le16_to_cpu(lt->lt_type));
 489         }
 490         return type;
 491 }
 492
 493 int ll_dir_read(struct inode *inode, struct dir_context *ctx)
 494 {
 495         struct ll_inode_info *info       = ll_i2info(inode);
 496         struct ll_sb_info    *sbi       = ll_i2sbi(inode);
 497         __u64              pos          = ctx->pos;
 498         int                api32      = ll_need_32bit_api(sbi);
 499         int                hash64     = sbi->ll_flags & LL_SBI_64BIT_HASH;
 500         struct page       *page;
 501         struct ll_dir_chain   chain;
 502         int                done = 0;
 503         int                rc = 0;
 504
 505         ll_dir_chain_init(&chain);
 506
 507         page = ll_get_dir_page(inode, pos, &chain);
 508
 509         while (rc == 0 && !done) {
 510                 struct lu_dirpage *dp;
 511                 struct lu_dirent  *ent;
 512
 513                 if (!IS_ERR(page)) {
 514                         /*
 515                          * If page is empty (end of directory is reached),
 516                          * use this value.
 517                          */
 518                         __u64 hash = MDS_DIR_END_OFF;
 519                         __u64 next;
 520
 521                         dp = page_address(page);
 522                         for (ent = lu_dirent_start(dp); ent && !done;
 523                              ent = lu_dirent_next(ent)) {
 524                                 __u16     type;
 525                                 int         namelen;
 526                                 struct lu_fid  fid;
 527                                 __u64     lhash;
 528                                 __u64     ino;
 529
 530                                 /*
 531                                  * XXX: implement correct swabbing here.
 532                                  */
 533
 534                                 hash = le64_to_cpu(ent->lde_hash);
 535                                 if (hash < pos)
 536                                         /*
 537                                          * Skip until we find target hash
 538                                          * value.
 539                                          */
 540                                         continue;
 541
 542                                 namelen = le16_to_cpu(ent->lde_namelen);
 543                                 if (namelen == 0)
 544                                         /*
 545                                          * Skip dummy record.
 546                                          */
 547                                         continue;
 548
 549                                 if (api32 && hash64)
 550                                         lhash = hash >> 32;
 551                                 else
 552                                         lhash = hash;
 553                                 fid_le_to_cpu(&fid, &ent->lde_fid);
 554                                 ino = cl_fid_build_ino(&fid, api32);
 555                                 type = ll_dirent_type_get(ent);
 556                                 ctx->pos = lhash;
 557                                 /* For 'll_nfs_get_name_filldir()', it will try
 558                                  * to access the 'ent' through its 'lde_name',
 559                                  * so the parameter 'name' for 'ctx->actor()'
 560                                  * must be part of the 'ent'.
 561                                  */
 562                                 done = !dir_emit(ctx, ent->lde_name,
 563                                                  namelen, ino, type);
 564                         }
 565                         next = le64_to_cpu(dp->ldp_hash_end);
 566                         if (!done) {
 567                                 pos = next;
 568                                 if (pos == MDS_DIR_END_OFF) {
 569                                         /*
 570                                          * End of directory reached.
 571                                          */
 572                                         done = 1;
 573                                         ll_release_page(page, 0);
 574                                 } else if (1 /* chain is exhausted*/) {
 575                                         /*
 576                                          * Normal case: continue to the next
 577                                          * page.
 578                                          */
 579                                         ll_release_page(page,
 580                                             le32_to_cpu(dp->ldp_flags) &
 581                                                         LDF_COLLIDE);
 582                                         next = pos;
 583                                         page = ll_get_dir_page(inode, pos,
 584                                                                &chain);
 585                                 } else {
 586                                         /*
 587                                          * go into overflow page.
 588                                          */
 589                                         LASSERT(le32_to_cpu(dp->ldp_flags) &
 590                                                 LDF_COLLIDE);
 591                                         ll_release_page(page, 1);
 592                                 }
 593                         } else {
 594                                 pos = hash;
 595                                 ll_release_page(page, 0);
 596                         }
 597                 } else {
 598                         rc = PTR_ERR(page);
 599                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 600                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 601                 }
 602         }
 603
 604         ctx->pos = pos;
 605         ll_dir_chain_fini(&chain);
 606         return rc;
 607 }
 608
 609 static int ll_readdir(struct file *filp, struct dir_context *ctx)
 610 {
 611         struct inode            *inode  = file_inode(filp);
 612         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 613         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 614         __u64 pos = lfd ? lfd->lfd_pos : 0;
 615         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 616         int                     api32   = ll_need_32bit_api(sbi);
 617         int                     rc;
 618
 619         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu 32bit_api %d\n",
 620                inode->i_ino, inode->i_generation,
 621                inode, (unsigned long)pos, i_size_read(inode), api32);
 622
 623         if (pos == MDS_DIR_END_OFF) {
 624                 /*
 625                  * end-of-file.
 626                  */
 627                 rc = 0;
 628                 goto out;
 629         }
 630
 631         ctx->pos = pos;
 632         rc = ll_dir_read(inode, ctx);
 633         if (lfd)
 634                 lfd->lfd_pos = ctx->pos;
 635         if (ctx->pos == MDS_DIR_END_OFF) {
 636                 if (api32)
 637                         ctx->pos = LL_DIR_END_OFF_32BIT;
 638                 else
 639                         ctx->pos = LL_DIR_END_OFF;
 640         } else {
 641                 if (api32 && hash64)
 642                         ctx->pos >>= 32;
 643         }
 644         filp->f_version = inode->i_version;
 645
 646 out:
 647         if (!rc)
 648                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 649
 650         return rc;
 651 }
 652
 653 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 654 {
 655         struct mgs_send_param *msp;
 656         int rc = 0;
 657
 658         msp = kzalloc(sizeof(*msp), GFP_NOFS);
 659         if (!msp)
 660                 return -ENOMEM;
 661
 662         strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
 663         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 664                                 sizeof(struct mgs_send_param), msp, NULL);
 665         if (rc)
 666                 CERROR("Failed to set parameter: %d\n", rc);
 667         kfree(msp);
 668
 669         return rc;
 670 }
 671
 672 static int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
 673                                char *filename)
 674 {
 675         struct ptlrpc_request *request = NULL;
 676         struct md_op_data *op_data;
 677         struct ll_sb_info *sbi = ll_i2sbi(dir);
 678         int mode;
 679         int err;
 680
 681         mode = (~current_umask() & 0755) | S_IFDIR;
 682         op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
 683                                      strlen(filename), mode, LUSTRE_OPC_MKDIR,
 684                                      lump);
 685         if (IS_ERR(op_data)) {
 686                 err = PTR_ERR(op_data);
 687                 goto err_exit;
 688         }
 689
 690         op_data->op_cli_flags |= CLI_SET_MEA;
 691         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 692                         from_kuid(&init_user_ns, current_fsuid()),
 693                         from_kgid(&init_user_ns, current_fsgid()),
 694                         cfs_curproc_cap_pack(), 0, &request);
 695         ll_finish_md_op_data(op_data);
 696         if (err)
 697                 goto err_exit;
 698 err_exit:
 699         ptlrpc_req_finished(request);
 700         return err;
 701 }
 702
 703 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 704                      int set_default)
 705 {
 706         struct ll_sb_info *sbi = ll_i2sbi(inode);
 707         struct md_op_data *op_data;
 708         struct ptlrpc_request *req = NULL;
 709         int rc = 0;
 710         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 711         struct obd_device *mgc = lsi->lsi_mgc;
 712         int lum_size;
 713
 714         if (lump) {
 715                 /*
 716                  * This is coming from userspace, so should be in
 717                  * local endian.  But the MDS would like it in little
 718                  * endian, so we swab it before we send it.
 719                  */
 720                 switch (lump->lmm_magic) {
 721                 case LOV_USER_MAGIC_V1: {
 722                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 723                                 lustre_swab_lov_user_md_v1(lump);
 724                         lum_size = sizeof(struct lov_user_md_v1);
 725                         break;
 726                 }
 727                 case LOV_USER_MAGIC_V3: {
 728                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 729                                 lustre_swab_lov_user_md_v3(
 730                                         (struct lov_user_md_v3 *)lump);
 731                         lum_size = sizeof(struct lov_user_md_v3);
 732                         break;
 733                 }
 734                 default: {
 735                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
 736                                lump->lmm_magic, LOV_USER_MAGIC_V1,
 737                                LOV_USER_MAGIC_V3);
 738                         return -EINVAL;
 739                 }
 740                 }
 741         } else {
 742                 lum_size = sizeof(struct lov_user_md_v1);
 743         }
 744
 745         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 746                                      LUSTRE_OPC_ANY, NULL);
 747         if (IS_ERR(op_data))
 748                 return PTR_ERR(op_data);
 749
 750         if (lump && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC))
 751                 op_data->op_cli_flags |= CLI_SET_MEA;
 752
 753         /* swabbing is done in lov_setstripe() on server side */
 754         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 755                         NULL, 0, &req, NULL);
 756         ll_finish_md_op_data(op_data);
 757         ptlrpc_req_finished(req);
 758         if (rc) {
 759                 if (rc != -EPERM && rc != -EACCES)
 760                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 761         }
 762
 763         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 764          * LOV_USER_MAGIC_V3 have the same initial fields so we do not
 765          * need to make the distinction between the 2 versions
 766          */
 767         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 768                 char *param = NULL;
 769                 char *buf;
 770
 771                 param = kzalloc(MGS_PARAM_MAXLEN, GFP_NOFS);
 772                 if (!param)
 773                         return -ENOMEM;
 774
 775                 buf = param;
 776                 /* Get fsname and assume devname to be -MDT0000. */
 777                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 778                 strcat(buf, "-MDT0000.lov");
 779                 buf += strlen(buf);
 780
 781                 /* Set root stripesize */
 782                 sprintf(buf, ".stripesize=%u",
 783                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 784                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 785                 if (rc)
 786                         goto end;
 787
 788                 /* Set root stripecount */
 789                 sprintf(buf, ".stripecount=%hd",
 790                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 791                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 792                 if (rc)
 793                         goto end;
 794
 795                 /* Set root stripeoffset */
 796                 sprintf(buf, ".stripeoffset=%hd",
 797                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 798                         (typeof(lump->lmm_stripe_offset))(-1));
 799                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 800
 801 end:
 802                 kfree(param);
 803         }
 804         return rc;
 805 }
 806
 807 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 808                      int *lmm_size, struct ptlrpc_request **request)
 809 {
 810         struct ll_sb_info *sbi = ll_i2sbi(inode);
 811         struct mdt_body   *body;
 812         struct lov_mds_md *lmm = NULL;
 813         struct ptlrpc_request *req = NULL;
 814         int rc, lmmsize;
 815         struct md_op_data *op_data;
 816
 817         rc = ll_get_default_mdsize(sbi, &lmmsize);
 818         if (rc)
 819                 return rc;
 820
 821         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 822                                      0, lmmsize, LUSTRE_OPC_ANY,
 823                                      NULL);
 824         if (IS_ERR(op_data))
 825                 return PTR_ERR(op_data);
 826
 827         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 828         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 829         ll_finish_md_op_data(op_data);
 830         if (rc < 0) {
 831                 CDEBUG(D_INFO, "md_getattr failed on inode %lu/%u: rc %d\n",
 832                        inode->i_ino,
 833                        inode->i_generation, rc);
 834                 goto out;
 835         }
 836
 837         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 838
 839         lmmsize = body->eadatasize;
 840
 841         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 842             lmmsize == 0) {
 843                 rc = -ENODATA;
 844                 goto out;
 845         }
 846
 847         lmm = req_capsule_server_sized_get(&req->rq_pill,
 848                                            &RMF_MDT_MD, lmmsize);
 849
 850         /*
 851          * This is coming from the MDS, so is probably in
 852          * little endian.  We convert it to host endian before
 853          * passing it to userspace.
 854          */
 855         /* We don't swab objects for directories */
 856         switch (le32_to_cpu(lmm->lmm_magic)) {
 857         case LOV_MAGIC_V1:
 858                 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC)
 859                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 860                 break;
 861         case LOV_MAGIC_V3:
 862                 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC)
 863                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 864                 break;
 865         default:
 866                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 867                 rc = -EPROTO;
 868         }
 869 out:
 870         *lmmp = lmm;
 871         *lmm_size = lmmsize;
 872         *request = req;
 873         return rc;
 874 }
 875
 876 /*
 877  *  Get MDT index for the inode.
 878  */
 879 int ll_get_mdt_idx(struct inode *inode)
 880 {
 881         struct ll_sb_info *sbi = ll_i2sbi(inode);
 882         struct md_op_data *op_data;
 883         int rc, mdtidx;
 884
 885         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 886                                      0, LUSTRE_OPC_ANY, NULL);
 887         if (IS_ERR(op_data))
 888                 return PTR_ERR(op_data);
 889
 890         op_data->op_flags |= MF_GET_MDT_IDX;
 891         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 892         mdtidx = op_data->op_mds;
 893         ll_finish_md_op_data(op_data);
 894         if (rc < 0) {
 895                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 896                 return rc;
 897         }
 898         return mdtidx;
 899 }
 900
 901 /**
 902  * Generic handler to do any pre-copy work.
 903  *
 904  * It sends a first hsm_progress (with extent length == 0) to coordinator as a
 905  * first information for it that real work has started.
 906  *
 907  * Moreover, for a ARCHIVE request, it will sample the file data version and
 908  * store it in \a copy.
 909  *
 910  * \return 0 on success.
 911  */
 912 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 913 {
 914         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 915         struct hsm_progress_kernel       hpk;
 916         int                              rc;
 917
 918         /* Forge a hsm_progress based on data from copy. */
 919         hpk.hpk_fid = copy->hc_hai.hai_fid;
 920         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 921         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 922         hpk.hpk_extent.length = 0;
 923         hpk.hpk_flags = 0;
 924         hpk.hpk_errval = 0;
 925         hpk.hpk_data_version = 0;
 926
 927         /* For archive request, we need to read the current file version. */
 928         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 929                 struct inode    *inode;
 930                 __u64            data_version = 0;
 931
 932                 /* Get inode for this fid */
 933                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 934                 if (IS_ERR(inode)) {
 935                         hpk.hpk_flags |= HP_FLAG_RETRY;
 936                         /* hpk_errval is >= 0 */
 937                         hpk.hpk_errval = -PTR_ERR(inode);
 938                         rc = PTR_ERR(inode);
 939                         goto progress;
 940                 }
 941
 942                 /* Read current file data version */
 943                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 944                 iput(inode);
 945                 if (rc != 0) {
 946                         CDEBUG(D_HSM, "Could not read file data version of "
 947                                       DFID" (rc = %d). Archive request (%#llx) could not be done.\n",
 948                                       PFID(&copy->hc_hai.hai_fid), rc,
 949                                       copy->hc_hai.hai_cookie);
 950                         hpk.hpk_flags |= HP_FLAG_RETRY;
 951                         /* hpk_errval must be >= 0 */
 952                         hpk.hpk_errval = -rc;
 953                         goto progress;
 954                 }
 955
 956                 /* Store in the hsm_copy for later copytool use.
 957                  * Always modified even if no lsm.
 958                  */
 959                 copy->hc_data_version = data_version;
 960         }
 961
 962 progress:
 963         /* On error, the request should be considered as completed */
 964         if (hpk.hpk_errval > 0)
 965                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 966         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 967                            &hpk, NULL);
 968
 969         return rc;
 970 }
 971
 972 /**
 973  * Generic handler to do any post-copy work.
 974  *
 975  * It will send the last hsm_progress update to coordinator to inform it
 976  * that copy is finished and whether it was successful or not.
 977  *
 978  * Moreover,
 979  * - for ARCHIVE request, it will sample the file data version and compare it
 980  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 981  *   will be considered as failed.
 982  * - for RESTORE request, it will sample the file data version and send it to
 983  *   coordinator which is useful if the file was imported as 'released'.
 984  *
 985  * \return 0 on success.
 986  */
 987 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 988 {
 989         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 990         struct hsm_progress_kernel       hpk;
 991         int                              rc;
 992
 993         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 994         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 995          * initialized if copy_end was called with copy == NULL.
 996          */
 997
 998         /* Forge a hsm_progress based on data from copy. */
 999         hpk.hpk_fid = copy->hc_hai.hai_fid;
1000         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
1001         hpk.hpk_extent = copy->hc_hai.hai_extent;
1002         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
1003         hpk.hpk_errval = copy->hc_errval;
1004         hpk.hpk_data_version = 0;
1005
1006         /* For archive request, we need to check the file data was not changed.
1007          *
1008          * For restore request, we need to send the file data version, this is
1009          * useful when the file was created using hsm_import.
1010          */
1011         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
1012              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
1013             (copy->hc_errval == 0)) {
1014                 struct inode    *inode;
1015                 __u64            data_version = 0;
1016
1017                 /* Get lsm for this fid */
1018                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
1019                 if (IS_ERR(inode)) {
1020                         hpk.hpk_flags |= HP_FLAG_RETRY;
1021                         /* hpk_errval must be >= 0 */
1022                         hpk.hpk_errval = -PTR_ERR(inode);
1023                         rc = PTR_ERR(inode);
1024                         goto progress;
1025                 }
1026
1027                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
1028                 iput(inode);
1029                 if (rc) {
1030                         CDEBUG(D_HSM, "Could not read file data version. Request could not be confirmed.\n");
1031                         if (hpk.hpk_errval == 0)
1032                                 hpk.hpk_errval = -rc;
1033                         goto progress;
1034                 }
1035
1036                 /* Store in the hsm_copy for later copytool use.
1037                  * Always modified even if no lsm.
1038                  */
1039                 hpk.hpk_data_version = data_version;
1040
1041                 /* File could have been stripped during archiving, so we need
1042                  * to check anyway.
1043                  */
1044                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1045                     (copy->hc_data_version != data_version)) {
1046                         CDEBUG(D_HSM, "File data version mismatched. File content was changed during archiving. "
1047                                DFID", start:%#llx current:%#llx\n",
1048                                PFID(&copy->hc_hai.hai_fid),
1049                                copy->hc_data_version, data_version);
1050                         /* File was changed, send error to cdt. Do not ask for
1051                          * retry because if a file is modified frequently,
1052                          * the cdt will loop on retried archive requests.
1053                          * The policy engine will ask for a new archive later
1054                          * when the file will not be modified for some tunable
1055                          * time
1056                          */
1057                         /* we do not notify caller */
1058                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
1059                         /* hpk_errval must be >= 0 */
1060                         hpk.hpk_errval = EBUSY;
1061                 }
1062         }
1063
1064 progress:
1065         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1066                            &hpk, NULL);
1067
1068         return rc;
1069 }
1070
1071 static int copy_and_ioctl(int cmd, struct obd_export *exp,
1072                           const void __user *data, size_t size)
1073 {
1074         void *copy;
1075         int rc;
1076
1077         copy = kzalloc(size, GFP_NOFS);
1078         if (!copy)
1079                 return -ENOMEM;
1080
1081         if (copy_from_user(copy, data, size)) {
1082                 rc = -EFAULT;
1083                 goto out;
1084         }
1085
1086         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
1087 out:
1088         kfree(copy);
1089
1090         return rc;
1091 }
1092
1093 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1094 {
1095         int cmd = qctl->qc_cmd;
1096         int type = qctl->qc_type;
1097         int id = qctl->qc_id;
1098         int valid = qctl->qc_valid;
1099         int rc = 0;
1100
1101         switch (cmd) {
1102         case LUSTRE_Q_INVALIDATE:
1103         case LUSTRE_Q_FINVALIDATE:
1104         case Q_QUOTAON:
1105         case Q_QUOTAOFF:
1106         case Q_SETQUOTA:
1107         case Q_SETINFO:
1108                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1109                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1110                         return -EPERM;
1111                 break;
1112         case Q_GETQUOTA:
1113                 if (((type == USRQUOTA &&
1114                       !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
1115                      (type == GRPQUOTA &&
1116                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
1117                     (!capable(CFS_CAP_SYS_ADMIN) ||
1118                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
1119                         return -EPERM;
1120                 break;
1121         case Q_GETINFO:
1122                 break;
1123         default:
1124                 CERROR("unsupported quotactl op: %#x\n", cmd);
1125                 return -ENOTTY;
1126         }
1127
1128         if (valid != QC_GENERAL) {
1129                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1130                         return -EOPNOTSUPP;
1131
1132                 if (cmd == Q_GETINFO)
1133                         qctl->qc_cmd = Q_GETOINFO;
1134                 else if (cmd == Q_GETQUOTA)
1135                         qctl->qc_cmd = Q_GETOQUOTA;
1136                 else
1137                         return -EINVAL;
1138
1139                 switch (valid) {
1140                 case QC_MDTIDX:
1141                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1142                                            sizeof(*qctl), qctl, NULL);
1143                         break;
1144                 case QC_OSTIDX:
1145                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1146                                            sizeof(*qctl), qctl, NULL);
1147                         break;
1148                 case QC_UUID:
1149                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1150                                            sizeof(*qctl), qctl, NULL);
1151                         if (rc == -EAGAIN)
1152                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1153                                                    sbi->ll_dt_exp,
1154                                                    sizeof(*qctl), qctl, NULL);
1155                         break;
1156                 default:
1157                         rc = -EINVAL;
1158                         break;
1159                 }
1160
1161                 if (rc)
1162                         return rc;
1163
1164                 qctl->qc_cmd = cmd;
1165         } else {
1166                 struct obd_quotactl *oqctl;
1167
1168                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1169                 if (!oqctl)
1170                         return -ENOMEM;
1171
1172                 QCTL_COPY(oqctl, qctl);
1173                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1174                 if (rc) {
1175                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
1176                                 oqctl->qc_cmd = Q_QUOTAOFF;
1177                                 obd_quotactl(sbi->ll_md_exp, oqctl);
1178                         }
1179                         kfree(oqctl);
1180                         return rc;
1181                 }
1182                 /* If QIF_SPACE is not set, client should collect the
1183                  * space usage from OSSs by itself
1184                  */
1185                 if (cmd == Q_GETQUOTA &&
1186                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1187                     !oqctl->qc_dqblk.dqb_curspace) {
1188                         struct obd_quotactl *oqctl_tmp;
1189
1190                         oqctl_tmp = kzalloc(sizeof(*oqctl_tmp), GFP_NOFS);
1191                         if (!oqctl_tmp) {
1192                                 rc = -ENOMEM;
1193                                 goto out;
1194                         }
1195
1196                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1197                         oqctl_tmp->qc_id = oqctl->qc_id;
1198                         oqctl_tmp->qc_type = oqctl->qc_type;
1199
1200                         /* collect space usage from OSTs */
1201                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1202                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1203                         if (!rc || rc == -EREMOTEIO) {
1204                                 oqctl->qc_dqblk.dqb_curspace =
1205                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1206                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1207                         }
1208
1209                         /* collect space & inode usage from MDTs */
1210                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1211                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1212                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1213                         if (!rc || rc == -EREMOTEIO) {
1214                                 oqctl->qc_dqblk.dqb_curspace +=
1215                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1216                                 oqctl->qc_dqblk.dqb_curinodes =
1217                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1218                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1219                         } else {
1220                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1221                         }
1222
1223                         kfree(oqctl_tmp);
1224                 }
1225 out:
1226                 QCTL_COPY(qctl, oqctl);
1227                 kfree(oqctl);
1228         }
1229
1230         return rc;
1231 }
1232
1233 /* This function tries to get a single name component,
1234  * to send to the server. No actual path traversal involved,
1235  * so we limit to NAME_MAX
1236  */
1237 static char *ll_getname(const char __user *filename)
1238 {
1239         int ret = 0, len;
1240         char *tmp;
1241
1242         tmp = kzalloc(NAME_MAX + 1, GFP_KERNEL);
1243         if (!tmp)
1244                 return ERR_PTR(-ENOMEM);
1245
1246         len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1247         if (len < 0)
1248                 ret = len;
1249         else if (len == 0)
1250                 ret = -ENOENT;
1251         else if (len > NAME_MAX && tmp[NAME_MAX] != 0)
1252                 ret = -ENAMETOOLONG;
1253
1254         if (ret) {
1255                 kfree(tmp);
1256                 tmp =  ERR_PTR(ret);
1257         }
1258         return tmp;
1259 }
1260
1261 #define ll_putname(filename) kfree(filename)
1262
1263 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1264 {
1265         struct inode *inode = file_inode(file);
1266         struct ll_sb_info *sbi = ll_i2sbi(inode);
1267         struct obd_ioctl_data *data;
1268         int rc = 0;
1269
1270         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1271                inode->i_ino, inode->i_generation, inode, cmd);
1272
1273         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1274         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1275                 return -ENOTTY;
1276
1277         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1278         switch (cmd) {
1279         case FSFILT_IOC_GETFLAGS:
1280         case FSFILT_IOC_SETFLAGS:
1281                 return ll_iocontrol(inode, file, cmd, arg);
1282         case FSFILT_IOC_GETVERSION_OLD:
1283         case FSFILT_IOC_GETVERSION:
1284                 return put_user(inode->i_generation, (int __user *)arg);
1285         /* We need to special case any other ioctls we want to handle,
1286          * to send them to the MDS/OST as appropriate and to properly
1287          * network encode the arg field.
1288         case FSFILT_IOC_SETVERSION_OLD:
1289         case FSFILT_IOC_SETVERSION:
1290         */
1291         case LL_IOC_GET_MDTIDX: {
1292                 int mdtidx;
1293
1294                 mdtidx = ll_get_mdt_idx(inode);
1295                 if (mdtidx < 0)
1296                         return mdtidx;
1297
1298                 if (put_user((int)mdtidx, (int __user *)arg))
1299                         return -EFAULT;
1300
1301                 return 0;
1302         }
1303         case IOC_MDC_LOOKUP: {
1304                 struct ptlrpc_request *request = NULL;
1305                 int namelen, len = 0;
1306                 char *buf = NULL;
1307                 char *filename;
1308                 struct md_op_data *op_data;
1309
1310                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1311                 if (rc)
1312                         return rc;
1313                 data = (void *)buf;
1314
1315                 filename = data->ioc_inlbuf1;
1316                 namelen = strlen(filename);
1317
1318                 if (namelen < 1) {
1319                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1320                         rc = -EINVAL;
1321                         goto out_free;
1322                 }
1323
1324                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1325                                              0, LUSTRE_OPC_ANY, NULL);
1326                 if (IS_ERR(op_data)) {
1327                         rc = PTR_ERR(op_data);
1328                         goto out_free;
1329                 }
1330
1331                 op_data->op_valid = OBD_MD_FLID;
1332                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1333                 ll_finish_md_op_data(op_data);
1334                 if (rc < 0) {
1335                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1336                         goto out_free;
1337                 }
1338                 ptlrpc_req_finished(request);
1339 out_free:
1340                 obd_ioctl_freedata(buf, len);
1341                 return rc;
1342         }
1343         case LL_IOC_LMV_SETSTRIPE: {
1344                 struct lmv_user_md  *lum;
1345                 char            *buf = NULL;
1346                 char            *filename;
1347                 int              namelen = 0;
1348                 int              lumlen = 0;
1349                 int              len;
1350                 int              rc;
1351
1352                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1353                 if (rc)
1354                         return rc;
1355
1356                 data = (void *)buf;
1357                 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
1358                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0) {
1359                         rc = -EINVAL;
1360                         goto lmv_out_free;
1361                 }
1362
1363                 filename = data->ioc_inlbuf1;
1364                 namelen = data->ioc_inllen1;
1365
1366                 if (namelen < 1) {
1367                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1368                         rc = -EINVAL;
1369                         goto lmv_out_free;
1370                 }
1371                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1372                 lumlen = data->ioc_inllen2;
1373
1374                 if (lum->lum_magic != LMV_USER_MAGIC ||
1375                     lumlen != sizeof(*lum)) {
1376                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1377                                filename, lum->lum_magic, lumlen, -EFAULT);
1378                         rc = -EINVAL;
1379                         goto lmv_out_free;
1380                 }
1381
1382                 /**
1383                  * ll_dir_setdirstripe will be used to set dir stripe
1384                  *  mdc_create--->mdt_reint_create (with dirstripe)
1385                  */
1386                 rc = ll_dir_setdirstripe(inode, lum, filename);
1387 lmv_out_free:
1388                 obd_ioctl_freedata(buf, len);
1389                 return rc;
1390         }
1391         case LL_IOC_LOV_SETSTRIPE: {
1392                 struct lov_user_md_v3 lumv3;
1393                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1394                 struct lov_user_md_v1 __user *lumv1p = (void __user *)arg;
1395                 struct lov_user_md_v3 __user *lumv3p = (void __user *)arg;
1396
1397                 int set_default = 0;
1398
1399                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1400                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1401                         sizeof(lumv3p->lmm_objects[0]));
1402                 /* first try with v1 which is smaller than v3 */
1403                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1404                         return -EFAULT;
1405
1406                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1407                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1408                                 return -EFAULT;
1409                 }
1410
1411                 if (is_root_inode(inode))
1412                         set_default = 1;
1413
1414                 /* in v1 and v3 cases lumv1 points to data */
1415                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1416
1417                 return rc;
1418         }
1419         case LL_IOC_LMV_GETSTRIPE: {
1420                 struct lmv_user_md __user *lump = (void __user *)arg;
1421                 struct lmv_user_md lum;
1422                 struct lmv_user_md *tmp;
1423                 int lum_size;
1424                 int rc = 0;
1425                 int mdtindex;
1426
1427                 if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
1428                         return -EFAULT;
1429
1430                 if (lum.lum_magic != LMV_MAGIC_V1)
1431                         return -EINVAL;
1432
1433                 lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
1434                 tmp = kzalloc(lum_size, GFP_NOFS);
1435                 if (!tmp) {
1436                         rc = -ENOMEM;
1437                         goto free_lmv;
1438                 }
1439
1440                 *tmp = lum;
1441                 tmp->lum_type = LMV_STRIPE_TYPE;
1442                 tmp->lum_stripe_count = 1;
1443                 mdtindex = ll_get_mdt_idx(inode);
1444                 if (mdtindex < 0) {
1445                         rc = -ENOMEM;
1446                         goto free_lmv;
1447                 }
1448
1449                 tmp->lum_stripe_offset = mdtindex;
1450                 tmp->lum_objects[0].lum_mds = mdtindex;
1451                 memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
1452                        sizeof(struct lu_fid));
1453                 if (copy_to_user((void __user *)arg, tmp, lum_size)) {
1454                         rc = -EFAULT;
1455                         goto free_lmv;
1456                 }
1457 free_lmv:
1458                 kfree(tmp);
1459                 return rc;
1460         }
1461         case LL_IOC_LOV_SWAP_LAYOUTS:
1462                 return -EPERM;
1463         case LL_IOC_OBD_STATFS:
1464                 return ll_obd_statfs(inode, (void __user *)arg);
1465         case LL_IOC_LOV_GETSTRIPE:
1466         case LL_IOC_MDC_GETINFO:
1467         case IOC_MDC_GETFILEINFO:
1468         case IOC_MDC_GETFILESTRIPE: {
1469                 struct ptlrpc_request *request = NULL;
1470                 struct lov_user_md __user *lump;
1471                 struct lov_mds_md *lmm = NULL;
1472                 struct mdt_body *body;
1473                 char *filename = NULL;
1474                 int lmmsize;
1475
1476                 if (cmd == IOC_MDC_GETFILEINFO ||
1477                     cmd == IOC_MDC_GETFILESTRIPE) {
1478                         filename = ll_getname((const char __user *)arg);
1479                         if (IS_ERR(filename))
1480                                 return PTR_ERR(filename);
1481
1482                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1483                                                       &lmmsize, &request);
1484                 } else {
1485                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1486                 }
1487
1488                 if (request) {
1489                         body = req_capsule_server_get(&request->rq_pill,
1490                                                       &RMF_MDT_BODY);
1491                         LASSERT(body);
1492                 } else {
1493                         goto out_req;
1494                 }
1495
1496                 if (rc < 0) {
1497                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1498                                                cmd == LL_IOC_MDC_GETINFO)) {
1499                                 rc = 0;
1500                                 goto skip_lmm;
1501                         } else {
1502                                 goto out_req;
1503                         }
1504                 }
1505
1506                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1507                     cmd == LL_IOC_LOV_GETSTRIPE) {
1508                         lump = (struct lov_user_md __user *)arg;
1509                 } else {
1510                         struct lov_user_mds_data __user *lmdp;
1511
1512                         lmdp = (struct lov_user_mds_data __user *)arg;
1513                         lump = &lmdp->lmd_lmm;
1514                 }
1515                 if (copy_to_user(lump, lmm, lmmsize)) {
1516                         if (copy_to_user(lump, lmm, sizeof(*lump))) {
1517                                 rc = -EFAULT;
1518                                 goto out_req;
1519                         }
1520                         rc = -EOVERFLOW;
1521                 }
1522 skip_lmm:
1523                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1524                         struct lov_user_mds_data __user *lmdp;
1525                         lstat_t st = { 0 };
1526
1527                         st.st_dev     = inode->i_sb->s_dev;
1528                         st.st_mode    = body->mode;
1529                         st.st_nlink   = body->nlink;
1530                         st.st_uid     = body->uid;
1531                         st.st_gid     = body->gid;
1532                         st.st_rdev    = body->rdev;
1533                         st.st_size    = body->size;
1534                         st.st_blksize = PAGE_SIZE;
1535                         st.st_blocks  = body->blocks;
1536                         st.st_atime   = body->atime;
1537                         st.st_mtime   = body->mtime;
1538                         st.st_ctime   = body->ctime;
1539                         st.st_ino     = inode->i_ino;
1540
1541                         lmdp = (struct lov_user_mds_data __user *)arg;
1542                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st))) {
1543                                 rc = -EFAULT;
1544                                 goto out_req;
1545                         }
1546                 }
1547
1548 out_req:
1549                 ptlrpc_req_finished(request);
1550                 if (filename)
1551                         ll_putname(filename);
1552                 return rc;
1553         }
1554         case IOC_LOV_GETINFO: {
1555                 struct lov_user_mds_data __user *lumd;
1556                 struct lov_stripe_md *lsm;
1557                 struct lov_user_md __user *lum;
1558                 struct lov_mds_md *lmm;
1559                 int lmmsize;
1560                 lstat_t st;
1561
1562                 lumd = (struct lov_user_mds_data __user *)arg;
1563                 lum = &lumd->lmd_lmm;
1564
1565                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1566                 if (rc)
1567                         return rc;
1568
1569                 lmm = libcfs_kvzalloc(lmmsize, GFP_NOFS);
1570                 if (!lmm)
1571                         return -ENOMEM;
1572                 if (copy_from_user(lmm, lum, lmmsize)) {
1573                         rc = -EFAULT;
1574                         goto free_lmm;
1575                 }
1576
1577                 switch (lmm->lmm_magic) {
1578                 case LOV_USER_MAGIC_V1:
1579                         if (cpu_to_le32(LOV_USER_MAGIC_V1) == LOV_USER_MAGIC_V1)
1580                                 break;
1581                         /* swab objects first so that stripes num will be sane */
1582                         lustre_swab_lov_user_md_objects(
1583                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1584                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1585                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1586                         break;
1587                 case LOV_USER_MAGIC_V3:
1588                         if (cpu_to_le32(LOV_USER_MAGIC_V3) == LOV_USER_MAGIC_V3)
1589                                 break;
1590                         /* swab objects first so that stripes num will be sane */
1591                         lustre_swab_lov_user_md_objects(
1592                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1593                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1594                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1595                         break;
1596                 default:
1597                         rc = -EINVAL;
1598                         goto free_lmm;
1599                 }
1600
1601                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1602                 if (rc < 0) {
1603                         rc = -ENOMEM;
1604                         goto free_lmm;
1605                 }
1606
1607                 /* Perform glimpse_size operation. */
1608                 memset(&st, 0, sizeof(st));
1609
1610                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1611                 if (rc)
1612                         goto free_lsm;
1613
1614                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st))) {
1615                         rc = -EFAULT;
1616                         goto free_lsm;
1617                 }
1618
1619 free_lsm:
1620                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1621 free_lmm:
1622                 kvfree(lmm);
1623                 return rc;
1624         }
1625         case OBD_IOC_LLOG_CATINFO: {
1626                 return -EOPNOTSUPP;
1627         }
1628         case OBD_IOC_QUOTACHECK: {
1629                 struct obd_quotactl *oqctl;
1630                 int error = 0;
1631
1632                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1633                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1634                         return -EPERM;
1635
1636                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1637                 if (!oqctl)
1638                         return -ENOMEM;
1639                 oqctl->qc_type = arg;
1640                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1641                 if (rc < 0) {
1642                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1643                         error = rc;
1644                 }
1645
1646                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1647                 if (rc < 0)
1648                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1649
1650                 kfree(oqctl);
1651                 return error ?: rc;
1652         }
1653         case OBD_IOC_POLL_QUOTACHECK: {
1654                 struct if_quotacheck *check;
1655
1656                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1657                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1658                         return -EPERM;
1659
1660                 check = kzalloc(sizeof(*check), GFP_NOFS);
1661                 if (!check)
1662                         return -ENOMEM;
1663
1664                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1665                                    NULL);
1666                 if (rc) {
1667                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1668                         if (copy_to_user((void __user *)arg, check,
1669                                          sizeof(*check)))
1670                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1671                         goto out_poll;
1672                 }
1673
1674                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1675                                    NULL);
1676                 if (rc) {
1677                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1678                         if (copy_to_user((void __user *)arg, check,
1679                                          sizeof(*check)))
1680                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1681                         goto out_poll;
1682                 }
1683 out_poll:
1684                 kfree(check);
1685                 return rc;
1686         }
1687         case LL_IOC_QUOTACTL: {
1688                 struct if_quotactl *qctl;
1689
1690                 qctl = kzalloc(sizeof(*qctl), GFP_NOFS);
1691                 if (!qctl)
1692                         return -ENOMEM;
1693
1694                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl))) {
1695                         rc = -EFAULT;
1696                         goto out_quotactl;
1697                 }
1698
1699                 rc = quotactl_ioctl(sbi, qctl);
1700
1701                 if (rc == 0 && copy_to_user((void __user *)arg, qctl,
1702                                             sizeof(*qctl)))
1703                         rc = -EFAULT;
1704
1705 out_quotactl:
1706                 kfree(qctl);
1707                 return rc;
1708         }
1709         case OBD_IOC_GETDTNAME:
1710         case OBD_IOC_GETMDNAME:
1711                 return ll_get_obd_name(inode, cmd, arg);
1712         case LL_IOC_FLUSHCTX:
1713                 return ll_flush_ctx(inode);
1714 #ifdef CONFIG_FS_POSIX_ACL
1715         case LL_IOC_RMTACL: {
1716                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT && is_root_inode(inode)) {
1717                         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1718
1719                         rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1720                         if (!rc)
1721                                 fd->fd_flags |= LL_FILE_RMTACL;
1722                         return rc;
1723                 } else {
1724                         return 0;
1725                 }
1726         }
1727 #endif
1728         case LL_IOC_GETOBDCOUNT: {
1729                 int count, vallen;
1730                 struct obd_export *exp;
1731
1732                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1733                         return -EFAULT;
1734
1735                 /* get ost count when count is zero, get mdt count otherwise */
1736                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1737                 vallen = sizeof(count);
1738                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1739                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1740                 if (rc) {
1741                         CERROR("get target count failed: %d\n", rc);
1742                         return rc;
1743                 }
1744
1745                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1746                         return -EFAULT;
1747
1748                 return 0;
1749         }
1750         case LL_IOC_PATH2FID:
1751                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1752                                  sizeof(struct lu_fid)))
1753                         return -EFAULT;
1754                 return 0;
1755         case LL_IOC_GET_CONNECT_FLAGS: {
1756                 return obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1757                                      (void __user *)arg);
1758         }
1759         case OBD_IOC_CHANGELOG_SEND:
1760         case OBD_IOC_CHANGELOG_CLEAR:
1761                 if (!capable(CFS_CAP_SYS_ADMIN))
1762                         return -EPERM;
1763
1764                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1765                                     sizeof(struct ioc_changelog));
1766                 return rc;
1767         case OBD_IOC_FID2PATH:
1768                 return ll_fid2path(inode, (void __user *)arg);
1769         case LL_IOC_HSM_REQUEST: {
1770                 struct hsm_user_request *hur;
1771                 ssize_t                  totalsize;
1772
1773                 hur = memdup_user((void __user *)arg, sizeof(*hur));
1774                 if (IS_ERR(hur))
1775                         return PTR_ERR(hur);
1776
1777                 /* Compute the whole struct size */
1778                 totalsize = hur_len(hur);
1779                 kfree(hur);
1780                 if (totalsize < 0)
1781                         return -E2BIG;
1782
1783                 /* Final size will be more than double totalsize */
1784                 if (totalsize >= MDS_MAXREQSIZE / 3)
1785                         return -E2BIG;
1786
1787                 hur = libcfs_kvzalloc(totalsize, GFP_NOFS);
1788                 if (!hur)
1789                         return -ENOMEM;
1790
1791                 /* Copy the whole struct */
1792                 if (copy_from_user(hur, (void __user *)arg, totalsize)) {
1793                         kvfree(hur);
1794                         return -EFAULT;
1795                 }
1796
1797                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1798                         const struct lu_fid *fid;
1799                         struct inode *f;
1800                         int i;
1801
1802                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1803                                 fid = &hur->hur_user_item[i].hui_fid;
1804                                 f = search_inode_for_lustre(inode->i_sb, fid);
1805                                 if (IS_ERR(f)) {
1806                                         rc = PTR_ERR(f);
1807                                         break;
1808                                 }
1809
1810                                 rc = ll_hsm_release(f);
1811                                 iput(f);
1812                                 if (rc != 0)
1813                                         break;
1814                         }
1815                 } else {
1816                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1817                                            hur, NULL);
1818                 }
1819
1820                 kvfree(hur);
1821
1822                 return rc;
1823         }
1824         case LL_IOC_HSM_PROGRESS: {
1825                 struct hsm_progress_kernel      hpk;
1826                 struct hsm_progress             hp;
1827
1828                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1829                         return -EFAULT;
1830
1831                 hpk.hpk_fid = hp.hp_fid;
1832                 hpk.hpk_cookie = hp.hp_cookie;
1833                 hpk.hpk_extent = hp.hp_extent;
1834                 hpk.hpk_flags = hp.hp_flags;
1835                 hpk.hpk_errval = hp.hp_errval;
1836                 hpk.hpk_data_version = 0;
1837
1838                 /* File may not exist in Lustre; all progress
1839                  * reported to Lustre root
1840                  */
1841                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1842                                    NULL);
1843                 return rc;
1844         }
1845         case LL_IOC_HSM_CT_START:
1846                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1847                                     sizeof(struct lustre_kernelcomm));
1848                 return rc;
1849
1850         case LL_IOC_HSM_COPY_START: {
1851                 struct hsm_copy *copy;
1852                 int              rc;
1853
1854                 copy = memdup_user((char __user *)arg, sizeof(*copy));
1855                 if (IS_ERR(copy))
1856                         return PTR_ERR(copy);
1857
1858                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1859                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1860                         rc = -EFAULT;
1861
1862                 kfree(copy);
1863                 return rc;
1864         }
1865         case LL_IOC_HSM_COPY_END: {
1866                 struct hsm_copy *copy;
1867                 int              rc;
1868
1869                 copy = memdup_user((char __user *)arg, sizeof(*copy));
1870                 if (IS_ERR(copy))
1871                         return PTR_ERR(copy);
1872
1873                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1874                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1875                         rc = -EFAULT;
1876
1877                 kfree(copy);
1878                 return rc;
1879         }
1880         default:
1881                 return obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1882                                      (void __user *)arg);
1883         }
1884 }
1885
1886 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1887 {
1888         struct inode *inode = file->f_mapping->host;
1889         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1890         struct ll_sb_info *sbi = ll_i2sbi(inode);
1891         int api32 = ll_need_32bit_api(sbi);
1892         loff_t ret = -EINVAL;
1893
1894         inode_lock(inode);
1895         switch (origin) {
1896         case SEEK_SET:
1897                 break;
1898         case SEEK_CUR:
1899                 offset += file->f_pos;
1900                 break;
1901         case SEEK_END:
1902                 if (offset > 0)
1903                         goto out;
1904                 if (api32)
1905                         offset += LL_DIR_END_OFF_32BIT;
1906                 else
1907                         offset += LL_DIR_END_OFF;
1908                 break;
1909         default:
1910                 goto out;
1911         }
1912
1913         if (offset >= 0 &&
1914             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1915              (!api32 && offset <= LL_DIR_END_OFF))) {
1916                 if (offset != file->f_pos) {
1917                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1918                             (!api32 && offset == LL_DIR_END_OFF))
1919                                 fd->lfd_pos = MDS_DIR_END_OFF;
1920                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1921                                 fd->lfd_pos = offset << 32;
1922                         else
1923                                 fd->lfd_pos = offset;
1924                         file->f_pos = offset;
1925                         file->f_version = 0;
1926                 }
1927                 ret = offset;
1928         }
1929         goto out;
1930
1931 out:
1932         inode_unlock(inode);
1933         return ret;
1934 }
1935
1936 static int ll_dir_open(struct inode *inode, struct file *file)
1937 {
1938         return ll_file_open(inode, file);
1939 }
1940
1941 static int ll_dir_release(struct inode *inode, struct file *file)
1942 {
1943         return ll_file_release(inode, file);
1944 }
1945
1946 const struct file_operations ll_dir_operations = {
1947         .llseek   = ll_dir_seek,
1948         .open     = ll_dir_open,
1949         .release  = ll_dir_release,
1950         .read     = generic_read_dir,
1951         .iterate  = ll_readdir,
1952         .unlocked_ioctl   = ll_dir_ioctl,
1953         .fsync    = ll_fsync,
1954 };