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="DFID"(%p) hash %llu\n",
 162                PFID(ll_inode2fid(inode)), inode, hash);
 163
 164         LASSERT(max_pages > 0 && max_pages <= MD_MAX_BRW_PAGES);
 165
 166         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 167                                      LUSTRE_OPC_ANY, NULL);
 168         if (IS_ERR(op_data))
 169                 return PTR_ERR(op_data);
 170
 171         page_pool = kcalloc(max_pages, sizeof(page), GFP_NOFS);
 172         if (page_pool) {
 173                 page_pool[0] = page0;
 174         } else {
 175                 page_pool = &page0;
 176                 max_pages = 1;
 177         }
 178         for (npages = 1; npages < max_pages; npages++) {
 179                 page = page_cache_alloc_cold(inode->i_mapping);
 180                 if (!page)
 181                         break;
 182                 page_pool[npages] = page;
 183         }
 184
 185         op_data->op_npages = npages;
 186         op_data->op_offset = hash;
 187         rc = md_readpage(exp, op_data, page_pool, &request);
 188         ll_finish_md_op_data(op_data);
 189         if (rc < 0) {
 190                 /* page0 is special, which was added into page cache early */
 191                 delete_from_page_cache(page0);
 192         } else if (rc == 0) {
 193                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 194                 /* Checked by mdc_readpage() */
 195                 if (body->valid & OBD_MD_FLSIZE)
 196                         i_size_write(inode, body->size);
 197
 198                 nrdpgs = (request->rq_bulk->bd_nob_transferred+PAGE_SIZE-1)
 199                          >> PAGE_SHIFT;
 200                 SetPageUptodate(page0);
 201         }
 202         unlock_page(page0);
 203         ptlrpc_req_finished(request);
 204
 205         CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages);
 206
 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                         put_page(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_NOFS);
 229                 if (ret == 0) {
 230                         unlock_page(page);
 231                 } else {
 232                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed: %d\n",
 233                                offset, ret);
 234                 }
 235                 put_page(page);
 236         }
 237
 238         if (page_pool != &page0)
 239                 kfree(page_pool);
 240         return rc;
 241 }
 242
 243 void ll_release_page(struct page *page, int remove)
 244 {
 245         kunmap(page);
 246         if (remove) {
 247                 lock_page(page);
 248                 if (likely(page->mapping))
 249                         truncate_complete_page(page->mapping, page);
 250                 unlock_page(page);
 251         }
 252         put_page(page);
 253 }
 254
 255 /*
 256  * Find, kmap and return page that contains given hash.
 257  */
 258 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
 259                                        __u64 *start, __u64 *end)
 260 {
 261         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 262         struct address_space *mapping = dir->i_mapping;
 263         /*
 264          * Complement of hash is used as an index so that
 265          * radix_tree_gang_lookup() can be used to find a page with starting
 266          * hash _smaller_ than one we are looking for.
 267          */
 268         unsigned long offset = hash_x_index(*hash, hash64);
 269         struct page *page;
 270         int found;
 271
 272         spin_lock_irq(&mapping->tree_lock);
 273         found = radix_tree_gang_lookup(&mapping->page_tree,
 274                                        (void **)&page, offset, 1);
 275         if (found > 0 && !radix_tree_exceptional_entry(page)) {
 276                 struct lu_dirpage *dp;
 277
 278                 get_page(page);
 279                 spin_unlock_irq(&mapping->tree_lock);
 280                 /*
 281                  * In contrast to find_lock_page() we are sure that directory
 282                  * page cannot be truncated (while DLM lock is held) and,
 283                  * hence, can avoid restart.
 284                  *
 285                  * In fact, page cannot be locked here at all, because
 286                  * ll_dir_filler() does synchronous io.
 287                  */
 288                 wait_on_page_locked(page);
 289                 if (PageUptodate(page)) {
 290                         dp = kmap(page);
 291                         if (BITS_PER_LONG == 32 && hash64) {
 292                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 293                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 294                                 *hash  = *hash >> 32;
 295                         } else {
 296                                 *start = le64_to_cpu(dp->ldp_hash_start);
 297                                 *end   = le64_to_cpu(dp->ldp_hash_end);
 298                         }
 299                         LASSERTF(*start <= *hash, "start = %#llx,end = %#llx,hash = %#llx\n",
 300                                  *start, *end, *hash);
 301                         CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash %llu\n",
 302                                offset, *start, *end, *hash);
 303                         if (*hash > *end) {
 304                                 ll_release_page(page, 0);
 305                                 page = NULL;
 306                         } else if (*end != *start && *hash == *end) {
 307                                 /*
 308                                  * upon hash collision, remove this page,
 309                                  * otherwise put page reference, and
 310                                  * ll_get_dir_page() will issue RPC to fetch
 311                                  * the page we want.
 312                                  */
 313                                 ll_release_page(page,
 314                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
 315                                 page = NULL;
 316                         }
 317                 } else {
 318                         put_page(page);
 319                         page = ERR_PTR(-EIO);
 320                 }
 321
 322         } else {
 323                 spin_unlock_irq(&mapping->tree_lock);
 324                 page = NULL;
 325         }
 326         return page;
 327 }
 328
 329 struct page *ll_get_dir_page(struct inode *dir, __u64 hash,
 330                              struct ll_dir_chain *chain)
 331 {
 332         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 333         struct address_space *mapping = dir->i_mapping;
 334         struct lustre_handle lockh;
 335         struct lu_dirpage *dp;
 336         struct page *page;
 337         enum ldlm_mode mode;
 338         int rc;
 339         __u64 start = 0;
 340         __u64 end = 0;
 341         __u64 lhash = hash;
 342         struct ll_inode_info *lli = ll_i2info(dir);
 343         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 344
 345         mode = LCK_PR;
 346         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 347                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
 348         if (!rc) {
 349                 struct ldlm_enqueue_info einfo = {
 350                         .ei_type = LDLM_IBITS,
 351                         .ei_mode = mode,
 352                         .ei_cb_bl = ll_md_blocking_ast,
 353                         .ei_cb_cp = ldlm_completion_ast,
 354                 };
 355                 struct lookup_intent it = { .it_op = IT_READDIR };
 356                 struct ptlrpc_request *request;
 357                 struct md_op_data *op_data;
 358
 359                 op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0,
 360                                              LUSTRE_OPC_ANY, NULL);
 361                 if (IS_ERR(op_data))
 362                         return (void *)op_data;
 363
 364                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
 365                                 op_data, &lockh, NULL, 0, NULL, 0);
 366
 367                 ll_finish_md_op_data(op_data);
 368
 369                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 370                 if (request)
 371                         ptlrpc_req_finished(request);
 372                 if (rc < 0) {
 373                         CERROR("lock enqueue: " DFID " at %llu: rc %d\n",
 374                                PFID(ll_inode2fid(dir)), hash, rc);
 375                         return ERR_PTR(rc);
 376                 }
 377
 378                 CDEBUG(D_INODE, "setting lr_lvb_inode to inode "DFID"(%p)\n",
 379                        PFID(ll_inode2fid(dir)), dir);
 380                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp,
 381                                  &it.d.lustre.it_lock_handle, dir, NULL);
 382         } else {
 383                 /* for cross-ref object, l_ast_data of the lock may not be set,
 384                  * we reset it here
 385                  */
 386                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
 387                                  dir, NULL);
 388         }
 389         ldlm_lock_dump_handle(D_OTHER, &lockh);
 390
 391         mutex_lock(&lli->lli_readdir_mutex);
 392         page = ll_dir_page_locate(dir, &lhash, &start, &end);
 393         if (IS_ERR(page)) {
 394                 CERROR("dir page locate: "DFID" at %llu: rc %ld\n",
 395                        PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
 396                 goto out_unlock;
 397         } else if (page) {
 398                 /*
 399                  * XXX nikita: not entirely correct handling of a corner case:
 400                  * suppose hash chain of entries with hash value HASH crosses
 401                  * border between pages P0 and P1. First both P0 and P1 are
 402                  * cached, seekdir() is called for some entry from the P0 part
 403                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 404                  * happens and finds P1, as it starts with matching hash
 405                  * value. Remaining entries from P0 part of the chain are
 406                  * skipped. (Is that really a bug?)
 407                  *
 408                  * Possible solutions: 0. don't cache P1 is such case, handle
 409                  * it as an "overflow" page. 1. invalidate all pages at
 410                  * once. 2. use HASH|1 as an index for P1.
 411                  */
 412                 goto hash_collision;
 413         }
 414
 415         page = read_cache_page(mapping, hash_x_index(hash, hash64),
 416                                ll_dir_filler, &lhash);
 417         if (IS_ERR(page)) {
 418                 CERROR("read cache page: "DFID" at %llu: rc %ld\n",
 419                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 420                 goto out_unlock;
 421         }
 422
 423         wait_on_page_locked(page);
 424         (void)kmap(page);
 425         if (!PageUptodate(page)) {
 426                 CERROR("page not updated: "DFID" at %llu: rc %d\n",
 427                        PFID(ll_inode2fid(dir)), hash, -5);
 428                 goto fail;
 429         }
 430         if (!PageChecked(page))
 431                 /* XXX: check page format later */
 432                 SetPageChecked(page);
 433         if (PageError(page)) {
 434                 CERROR("page error: "DFID" at %llu: rc %d\n",
 435                        PFID(ll_inode2fid(dir)), hash, -5);
 436                 goto fail;
 437         }
 438 hash_collision:
 439         dp = page_address(page);
 440         if (BITS_PER_LONG == 32 && hash64) {
 441                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 442                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 443                 lhash = hash >> 32;
 444         } else {
 445                 start = le64_to_cpu(dp->ldp_hash_start);
 446                 end   = le64_to_cpu(dp->ldp_hash_end);
 447                 lhash = hash;
 448         }
 449         if (end == start) {
 450                 LASSERT(start == lhash);
 451                 CWARN("Page-wide hash collision: %llu\n", end);
 452                 if (BITS_PER_LONG == 32 && hash64)
 453                         CWARN("Real page-wide hash collision at [%llu %llu] with hash %llu\n",
 454                               le64_to_cpu(dp->ldp_hash_start),
 455                               le64_to_cpu(dp->ldp_hash_end), hash);
 456                 /*
 457                  * Fetch whole overflow chain...
 458                  *
 459                  * XXX not yet.
 460                  */
 461                 goto fail;
 462         }
 463 out_unlock:
 464         mutex_unlock(&lli->lli_readdir_mutex);
 465         ldlm_lock_decref(&lockh, mode);
 466         return page;
 467
 468 fail:
 469         ll_release_page(page, 1);
 470         page = ERR_PTR(-EIO);
 471         goto out_unlock;
 472 }
 473
 474 /**
 475  * return IF_* type for given lu_dirent entry.
 476  * IF_* flag shld be converted to particular OS file type in
 477  * platform llite module.
 478  */
 479 static __u16 ll_dirent_type_get(struct lu_dirent *ent)
 480 {
 481         __u16 type = 0;
 482         struct luda_type *lt;
 483         int len = 0;
 484
 485         if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
 486                 const unsigned int align = sizeof(struct luda_type) - 1;
 487
 488                 len = le16_to_cpu(ent->lde_namelen);
 489                 len = (len + align) & ~align;
 490                 lt = (void *)ent->lde_name + len;
 491                 type = IFTODT(le16_to_cpu(lt->lt_type));
 492         }
 493         return type;
 494 }
 495
 496 int ll_dir_read(struct inode *inode, struct dir_context *ctx)
 497 {
 498         struct ll_inode_info *info       = ll_i2info(inode);
 499         struct ll_sb_info    *sbi       = ll_i2sbi(inode);
 500         __u64              pos          = ctx->pos;
 501         int                api32      = ll_need_32bit_api(sbi);
 502         int                hash64     = sbi->ll_flags & LL_SBI_64BIT_HASH;
 503         struct page       *page;
 504         struct ll_dir_chain   chain;
 505         int                done = 0;
 506         int                rc = 0;
 507
 508         ll_dir_chain_init(&chain);
 509
 510         page = ll_get_dir_page(inode, pos, &chain);
 511
 512         while (rc == 0 && !done) {
 513                 struct lu_dirpage *dp;
 514                 struct lu_dirent  *ent;
 515
 516                 if (!IS_ERR(page)) {
 517                         /*
 518                          * If page is empty (end of directory is reached),
 519                          * use this value.
 520                          */
 521                         __u64 hash = MDS_DIR_END_OFF;
 522                         __u64 next;
 523
 524                         dp = page_address(page);
 525                         for (ent = lu_dirent_start(dp); ent && !done;
 526                              ent = lu_dirent_next(ent)) {
 527                                 __u16     type;
 528                                 int         namelen;
 529                                 struct lu_fid  fid;
 530                                 __u64     lhash;
 531                                 __u64     ino;
 532
 533                                 /*
 534                                  * XXX: implement correct swabbing here.
 535                                  */
 536
 537                                 hash = le64_to_cpu(ent->lde_hash);
 538                                 if (hash < pos)
 539                                         /*
 540                                          * Skip until we find target hash
 541                                          * value.
 542                                          */
 543                                         continue;
 544
 545                                 namelen = le16_to_cpu(ent->lde_namelen);
 546                                 if (namelen == 0)
 547                                         /*
 548                                          * Skip dummy record.
 549                                          */
 550                                         continue;
 551
 552                                 if (api32 && hash64)
 553                                         lhash = hash >> 32;
 554                                 else
 555                                         lhash = hash;
 556                                 fid_le_to_cpu(&fid, &ent->lde_fid);
 557                                 ino = cl_fid_build_ino(&fid, api32);
 558                                 type = ll_dirent_type_get(ent);
 559                                 ctx->pos = lhash;
 560                                 /* For 'll_nfs_get_name_filldir()', it will try
 561                                  * to access the 'ent' through its 'lde_name',
 562                                  * so the parameter 'name' for 'ctx->actor()'
 563                                  * must be part of the 'ent'.
 564                                  */
 565                                 done = !dir_emit(ctx, ent->lde_name,
 566                                                  namelen, ino, type);
 567                         }
 568                         next = le64_to_cpu(dp->ldp_hash_end);
 569                         if (!done) {
 570                                 pos = next;
 571                                 if (pos == MDS_DIR_END_OFF) {
 572                                         /*
 573                                          * End of directory reached.
 574                                          */
 575                                         done = 1;
 576                                         ll_release_page(page, 0);
 577                                 } else if (1 /* chain is exhausted*/) {
 578                                         /*
 579                                          * Normal case: continue to the next
 580                                          * page.
 581                                          */
 582                                         ll_release_page(page,
 583                                             le32_to_cpu(dp->ldp_flags) &
 584                                                         LDF_COLLIDE);
 585                                         next = pos;
 586                                         page = ll_get_dir_page(inode, pos,
 587                                                                &chain);
 588                                 } else {
 589                                         /*
 590                                          * go into overflow page.
 591                                          */
 592                                         LASSERT(le32_to_cpu(dp->ldp_flags) &
 593                                                 LDF_COLLIDE);
 594                                         ll_release_page(page, 1);
 595                                 }
 596                         } else {
 597                                 pos = hash;
 598                                 ll_release_page(page, 0);
 599                         }
 600                 } else {
 601                         rc = PTR_ERR(page);
 602                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 603                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 604                 }
 605         }
 606
 607         ctx->pos = pos;
 608         ll_dir_chain_fini(&chain);
 609         return rc;
 610 }
 611
 612 static int ll_readdir(struct file *filp, struct dir_context *ctx)
 613 {
 614         struct inode            *inode  = file_inode(filp);
 615         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 616         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 617         __u64 pos = lfd ? lfd->lfd_pos : 0;
 618         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 619         int                     api32   = ll_need_32bit_api(sbi);
 620         int                     rc;
 621
 622         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos %lu/%llu 32bit_api %d\n",
 623                PFID(ll_inode2fid(inode)), inode, (unsigned long)pos,
 624                i_size_read(inode), api32);
 625
 626         if (pos == MDS_DIR_END_OFF) {
 627                 /*
 628                  * end-of-file.
 629                  */
 630                 rc = 0;
 631                 goto out;
 632         }
 633
 634         ctx->pos = pos;
 635         rc = ll_dir_read(inode, ctx);
 636         if (lfd)
 637                 lfd->lfd_pos = ctx->pos;
 638         if (ctx->pos == MDS_DIR_END_OFF) {
 639                 if (api32)
 640                         ctx->pos = LL_DIR_END_OFF_32BIT;
 641                 else
 642                         ctx->pos = LL_DIR_END_OFF;
 643         } else {
 644                 if (api32 && hash64)
 645                         ctx->pos >>= 32;
 646         }
 647         filp->f_version = inode->i_version;
 648
 649 out:
 650         if (!rc)
 651                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 652
 653         return rc;
 654 }
 655
 656 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 657 {
 658         struct mgs_send_param *msp;
 659         int rc = 0;
 660
 661         msp = kzalloc(sizeof(*msp), GFP_NOFS);
 662         if (!msp)
 663                 return -ENOMEM;
 664
 665         strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
 666         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 667                                 sizeof(struct mgs_send_param), msp, NULL);
 668         if (rc)
 669                 CERROR("Failed to set parameter: %d\n", rc);
 670         kfree(msp);
 671
 672         return rc;
 673 }
 674
 675 static int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
 676                                char *filename)
 677 {
 678         struct ptlrpc_request *request = NULL;
 679         struct md_op_data *op_data;
 680         struct ll_sb_info *sbi = ll_i2sbi(dir);
 681         int mode;
 682         int err;
 683
 684         mode = (~current_umask() & 0755) | S_IFDIR;
 685         op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
 686                                      strlen(filename), mode, LUSTRE_OPC_MKDIR,
 687                                      lump);
 688         if (IS_ERR(op_data)) {
 689                 err = PTR_ERR(op_data);
 690                 goto err_exit;
 691         }
 692
 693         op_data->op_cli_flags |= CLI_SET_MEA;
 694         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 695                         from_kuid(&init_user_ns, current_fsuid()),
 696                         from_kgid(&init_user_ns, current_fsgid()),
 697                         cfs_curproc_cap_pack(), 0, &request);
 698         ll_finish_md_op_data(op_data);
 699         if (err)
 700                 goto err_exit;
 701 err_exit:
 702         ptlrpc_req_finished(request);
 703         return err;
 704 }
 705
 706 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 707                      int set_default)
 708 {
 709         struct ll_sb_info *sbi = ll_i2sbi(inode);
 710         struct md_op_data *op_data;
 711         struct ptlrpc_request *req = NULL;
 712         int rc = 0;
 713         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 714         struct obd_device *mgc = lsi->lsi_mgc;
 715         int lum_size;
 716
 717         if (lump) {
 718                 /*
 719                  * This is coming from userspace, so should be in
 720                  * local endian.  But the MDS would like it in little
 721                  * endian, so we swab it before we send it.
 722                  */
 723                 switch (lump->lmm_magic) {
 724                 case LOV_USER_MAGIC_V1: {
 725                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 726                                 lustre_swab_lov_user_md_v1(lump);
 727                         lum_size = sizeof(struct lov_user_md_v1);
 728                         break;
 729                 }
 730                 case LOV_USER_MAGIC_V3: {
 731                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 732                                 lustre_swab_lov_user_md_v3(
 733                                         (struct lov_user_md_v3 *)lump);
 734                         lum_size = sizeof(struct lov_user_md_v3);
 735                         break;
 736                 }
 737                 default: {
 738                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC: %#08x != %#08x nor %#08x\n",
 739                                lump->lmm_magic, LOV_USER_MAGIC_V1,
 740                                LOV_USER_MAGIC_V3);
 741                         return -EINVAL;
 742                 }
 743                 }
 744         } else {
 745                 lum_size = sizeof(struct lov_user_md_v1);
 746         }
 747
 748         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 749                                      LUSTRE_OPC_ANY, NULL);
 750         if (IS_ERR(op_data))
 751                 return PTR_ERR(op_data);
 752
 753         if (lump && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC))
 754                 op_data->op_cli_flags |= CLI_SET_MEA;
 755
 756         /* swabbing is done in lov_setstripe() on server side */
 757         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 758                         NULL, 0, &req, NULL);
 759         ll_finish_md_op_data(op_data);
 760         ptlrpc_req_finished(req);
 761         if (rc) {
 762                 if (rc != -EPERM && rc != -EACCES)
 763                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 764         }
 765
 766         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 767          * LOV_USER_MAGIC_V3 have the same initial fields so we do not
 768          * need to make the distinction between the 2 versions
 769          */
 770         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 771                 char *param = NULL;
 772                 char *buf;
 773
 774                 param = kzalloc(MGS_PARAM_MAXLEN, GFP_NOFS);
 775                 if (!param)
 776                         return -ENOMEM;
 777
 778                 buf = param;
 779                 /* Get fsname and assume devname to be -MDT0000. */
 780                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 781                 strcat(buf, "-MDT0000.lov");
 782                 buf += strlen(buf);
 783
 784                 /* Set root stripesize */
 785                 sprintf(buf, ".stripesize=%u",
 786                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 787                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 788                 if (rc)
 789                         goto end;
 790
 791                 /* Set root stripecount */
 792                 sprintf(buf, ".stripecount=%hd",
 793                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 794                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 795                 if (rc)
 796                         goto end;
 797
 798                 /* Set root stripeoffset */
 799                 sprintf(buf, ".stripeoffset=%hd",
 800                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 801                         (typeof(lump->lmm_stripe_offset))(-1));
 802                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 803
 804 end:
 805                 kfree(param);
 806         }
 807         return rc;
 808 }
 809
 810 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 811                      int *lmm_size, struct ptlrpc_request **request)
 812 {
 813         struct ll_sb_info *sbi = ll_i2sbi(inode);
 814         struct mdt_body   *body;
 815         struct lov_mds_md *lmm = NULL;
 816         struct ptlrpc_request *req = NULL;
 817         int rc, lmmsize;
 818         struct md_op_data *op_data;
 819
 820         rc = ll_get_default_mdsize(sbi, &lmmsize);
 821         if (rc)
 822                 return rc;
 823
 824         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 825                                      0, lmmsize, LUSTRE_OPC_ANY,
 826                                      NULL);
 827         if (IS_ERR(op_data))
 828                 return PTR_ERR(op_data);
 829
 830         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 831         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 832         ll_finish_md_op_data(op_data);
 833         if (rc < 0) {
 834                 CDEBUG(D_INFO, "md_getattr failed on inode "DFID": rc %d\n",
 835                        PFID(ll_inode2fid(inode)), rc);
 836                 goto out;
 837         }
 838
 839         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 840
 841         lmmsize = body->eadatasize;
 842
 843         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 844             lmmsize == 0) {
 845                 rc = -ENODATA;
 846                 goto out;
 847         }
 848
 849         lmm = req_capsule_server_sized_get(&req->rq_pill,
 850                                            &RMF_MDT_MD, lmmsize);
 851
 852         /*
 853          * This is coming from the MDS, so is probably in
 854          * little endian.  We convert it to host endian before
 855          * passing it to userspace.
 856          */
 857         /* We don't swab objects for directories */
 858         switch (le32_to_cpu(lmm->lmm_magic)) {
 859         case LOV_MAGIC_V1:
 860                 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC)
 861                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 862                 break;
 863         case LOV_MAGIC_V3:
 864                 if (cpu_to_le32(LOV_MAGIC) != LOV_MAGIC)
 865                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 866                 break;
 867         default:
 868                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 869                 rc = -EPROTO;
 870         }
 871 out:
 872         *lmmp = lmm;
 873         *lmm_size = lmmsize;
 874         *request = req;
 875         return rc;
 876 }
 877
 878 /*
 879  *  Get MDT index for the inode.
 880  */
 881 int ll_get_mdt_idx(struct inode *inode)
 882 {
 883         struct ll_sb_info *sbi = ll_i2sbi(inode);
 884         struct md_op_data *op_data;
 885         int rc, mdtidx;
 886
 887         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 888                                      0, LUSTRE_OPC_ANY, NULL);
 889         if (IS_ERR(op_data))
 890                 return PTR_ERR(op_data);
 891
 892         op_data->op_flags |= MF_GET_MDT_IDX;
 893         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 894         mdtidx = op_data->op_mds;
 895         ll_finish_md_op_data(op_data);
 896         if (rc < 0) {
 897                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 898                 return rc;
 899         }
 900         return mdtidx;
 901 }
 902
 903 /**
 904  * Generic handler to do any pre-copy work.
 905  *
 906  * It sends a first hsm_progress (with extent length == 0) to coordinator as a
 907  * first information for it that real work has started.
 908  *
 909  * Moreover, for a ARCHIVE request, it will sample the file data version and
 910  * store it in \a copy.
 911  *
 912  * \return 0 on success.
 913  */
 914 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 915 {
 916         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 917         struct hsm_progress_kernel       hpk;
 918         int                              rc;
 919
 920         /* Forge a hsm_progress based on data from copy. */
 921         hpk.hpk_fid = copy->hc_hai.hai_fid;
 922         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 923         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 924         hpk.hpk_extent.length = 0;
 925         hpk.hpk_flags = 0;
 926         hpk.hpk_errval = 0;
 927         hpk.hpk_data_version = 0;
 928
 929         /* For archive request, we need to read the current file version. */
 930         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 931                 struct inode    *inode;
 932                 __u64            data_version = 0;
 933
 934                 /* Get inode for this fid */
 935                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 936                 if (IS_ERR(inode)) {
 937                         hpk.hpk_flags |= HP_FLAG_RETRY;
 938                         /* hpk_errval is >= 0 */
 939                         hpk.hpk_errval = -PTR_ERR(inode);
 940                         rc = PTR_ERR(inode);
 941                         goto progress;
 942                 }
 943
 944                 /* Read current file data version */
 945                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 946                 iput(inode);
 947                 if (rc != 0) {
 948                         CDEBUG(D_HSM, "Could not read file data version of "
 949                                       DFID" (rc = %d). Archive request (%#llx) could not be done.\n",
 950                                       PFID(&copy->hc_hai.hai_fid), rc,
 951                                       copy->hc_hai.hai_cookie);
 952                         hpk.hpk_flags |= HP_FLAG_RETRY;
 953                         /* hpk_errval must be >= 0 */
 954                         hpk.hpk_errval = -rc;
 955                         goto progress;
 956                 }
 957
 958                 /* Store in the hsm_copy for later copytool use.
 959                  * Always modified even if no lsm.
 960                  */
 961                 copy->hc_data_version = data_version;
 962         }
 963
 964 progress:
 965         /* On error, the request should be considered as completed */
 966         if (hpk.hpk_errval > 0)
 967                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 968         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 969                            &hpk, NULL);
 970
 971         return rc;
 972 }
 973
 974 /**
 975  * Generic handler to do any post-copy work.
 976  *
 977  * It will send the last hsm_progress update to coordinator to inform it
 978  * that copy is finished and whether it was successful or not.
 979  *
 980  * Moreover,
 981  * - for ARCHIVE request, it will sample the file data version and compare it
 982  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 983  *   will be considered as failed.
 984  * - for RESTORE request, it will sample the file data version and send it to
 985  *   coordinator which is useful if the file was imported as 'released'.
 986  *
 987  * \return 0 on success.
 988  */
 989 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 990 {
 991         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 992         struct hsm_progress_kernel       hpk;
 993         int                              rc;
 994
 995         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 996         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 997          * initialized if copy_end was called with copy == NULL.
 998          */
 999
1000         /* Forge a hsm_progress based on data from copy. */
1001         hpk.hpk_fid = copy->hc_hai.hai_fid;
1002         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
1003         hpk.hpk_extent = copy->hc_hai.hai_extent;
1004         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
1005         hpk.hpk_errval = copy->hc_errval;
1006         hpk.hpk_data_version = 0;
1007
1008         /* For archive request, we need to check the file data was not changed.
1009          *
1010          * For restore request, we need to send the file data version, this is
1011          * useful when the file was created using hsm_import.
1012          */
1013         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
1014              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
1015             (copy->hc_errval == 0)) {
1016                 struct inode    *inode;
1017                 __u64            data_version = 0;
1018
1019                 /* Get lsm for this fid */
1020                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
1021                 if (IS_ERR(inode)) {
1022                         hpk.hpk_flags |= HP_FLAG_RETRY;
1023                         /* hpk_errval must be >= 0 */
1024                         hpk.hpk_errval = -PTR_ERR(inode);
1025                         rc = PTR_ERR(inode);
1026                         goto progress;
1027                 }
1028
1029                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
1030                 iput(inode);
1031                 if (rc) {
1032                         CDEBUG(D_HSM, "Could not read file data version. Request could not be confirmed.\n");
1033                         if (hpk.hpk_errval == 0)
1034                                 hpk.hpk_errval = -rc;
1035                         goto progress;
1036                 }
1037
1038                 /* Store in the hsm_copy for later copytool use.
1039                  * Always modified even if no lsm.
1040                  */
1041                 hpk.hpk_data_version = data_version;
1042
1043                 /* File could have been stripped during archiving, so we need
1044                  * to check anyway.
1045                  */
1046                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1047                     (copy->hc_data_version != data_version)) {
1048                         CDEBUG(D_HSM, "File data version mismatched. File content was changed during archiving. "
1049                                DFID", start:%#llx current:%#llx\n",
1050                                PFID(&copy->hc_hai.hai_fid),
1051                                copy->hc_data_version, data_version);
1052                         /* File was changed, send error to cdt. Do not ask for
1053                          * retry because if a file is modified frequently,
1054                          * the cdt will loop on retried archive requests.
1055                          * The policy engine will ask for a new archive later
1056                          * when the file will not be modified for some tunable
1057                          * time
1058                          */
1059                         /* we do not notify caller */
1060                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
1061                         /* hpk_errval must be >= 0 */
1062                         hpk.hpk_errval = EBUSY;
1063                 }
1064         }
1065
1066 progress:
1067         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1068                            &hpk, NULL);
1069
1070         return rc;
1071 }
1072
1073 static int copy_and_ioctl(int cmd, struct obd_export *exp,
1074                           const void __user *data, size_t size)
1075 {
1076         void *copy;
1077         int rc;
1078
1079         copy = kzalloc(size, GFP_NOFS);
1080         if (!copy)
1081                 return -ENOMEM;
1082
1083         if (copy_from_user(copy, data, size)) {
1084                 rc = -EFAULT;
1085                 goto out;
1086         }
1087
1088         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
1089 out:
1090         kfree(copy);
1091
1092         return rc;
1093 }
1094
1095 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1096 {
1097         int cmd = qctl->qc_cmd;
1098         int type = qctl->qc_type;
1099         int id = qctl->qc_id;
1100         int valid = qctl->qc_valid;
1101         int rc = 0;
1102
1103         switch (cmd) {
1104         case LUSTRE_Q_INVALIDATE:
1105         case LUSTRE_Q_FINVALIDATE:
1106         case Q_QUOTAON:
1107         case Q_QUOTAOFF:
1108         case Q_SETQUOTA:
1109         case Q_SETINFO:
1110                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1111                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1112                         return -EPERM;
1113                 break;
1114         case Q_GETQUOTA:
1115                 if (((type == USRQUOTA &&
1116                       !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
1117                      (type == GRPQUOTA &&
1118                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
1119                     (!capable(CFS_CAP_SYS_ADMIN) ||
1120                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
1121                         return -EPERM;
1122                 break;
1123         case Q_GETINFO:
1124                 break;
1125         default:
1126                 CERROR("unsupported quotactl op: %#x\n", cmd);
1127                 return -ENOTTY;
1128         }
1129
1130         if (valid != QC_GENERAL) {
1131                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1132                         return -EOPNOTSUPP;
1133
1134                 if (cmd == Q_GETINFO)
1135                         qctl->qc_cmd = Q_GETOINFO;
1136                 else if (cmd == Q_GETQUOTA)
1137                         qctl->qc_cmd = Q_GETOQUOTA;
1138                 else
1139                         return -EINVAL;
1140
1141                 switch (valid) {
1142                 case QC_MDTIDX:
1143                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1144                                            sizeof(*qctl), qctl, NULL);
1145                         break;
1146                 case QC_OSTIDX:
1147                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1148                                            sizeof(*qctl), qctl, NULL);
1149                         break;
1150                 case QC_UUID:
1151                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1152                                            sizeof(*qctl), qctl, NULL);
1153                         if (rc == -EAGAIN)
1154                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1155                                                    sbi->ll_dt_exp,
1156                                                    sizeof(*qctl), qctl, NULL);
1157                         break;
1158                 default:
1159                         rc = -EINVAL;
1160                         break;
1161                 }
1162
1163                 if (rc)
1164                         return rc;
1165
1166                 qctl->qc_cmd = cmd;
1167         } else {
1168                 struct obd_quotactl *oqctl;
1169
1170                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1171                 if (!oqctl)
1172                         return -ENOMEM;
1173
1174                 QCTL_COPY(oqctl, qctl);
1175                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1176                 if (rc) {
1177                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
1178                                 oqctl->qc_cmd = Q_QUOTAOFF;
1179                                 obd_quotactl(sbi->ll_md_exp, oqctl);
1180                         }
1181                         kfree(oqctl);
1182                         return rc;
1183                 }
1184                 /* If QIF_SPACE is not set, client should collect the
1185                  * space usage from OSSs by itself
1186                  */
1187                 if (cmd == Q_GETQUOTA &&
1188                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1189                     !oqctl->qc_dqblk.dqb_curspace) {
1190                         struct obd_quotactl *oqctl_tmp;
1191
1192                         oqctl_tmp = kzalloc(sizeof(*oqctl_tmp), GFP_NOFS);
1193                         if (!oqctl_tmp) {
1194                                 rc = -ENOMEM;
1195                                 goto out;
1196                         }
1197
1198                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1199                         oqctl_tmp->qc_id = oqctl->qc_id;
1200                         oqctl_tmp->qc_type = oqctl->qc_type;
1201
1202                         /* collect space usage from OSTs */
1203                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1204                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1205                         if (!rc || rc == -EREMOTEIO) {
1206                                 oqctl->qc_dqblk.dqb_curspace =
1207                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1208                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1209                         }
1210
1211                         /* collect space & inode usage from MDTs */
1212                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1213                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1214                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1215                         if (!rc || rc == -EREMOTEIO) {
1216                                 oqctl->qc_dqblk.dqb_curspace +=
1217                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1218                                 oqctl->qc_dqblk.dqb_curinodes =
1219                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1220                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1221                         } else {
1222                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1223                         }
1224
1225                         kfree(oqctl_tmp);
1226                 }
1227 out:
1228                 QCTL_COPY(qctl, oqctl);
1229                 kfree(oqctl);
1230         }
1231
1232         return rc;
1233 }
1234
1235 /* This function tries to get a single name component,
1236  * to send to the server. No actual path traversal involved,
1237  * so we limit to NAME_MAX
1238  */
1239 static char *ll_getname(const char __user *filename)
1240 {
1241         int ret = 0, len;
1242         char *tmp;
1243
1244         tmp = kzalloc(NAME_MAX + 1, GFP_KERNEL);
1245         if (!tmp)
1246                 return ERR_PTR(-ENOMEM);
1247
1248         len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1249         if (len < 0)
1250                 ret = len;
1251         else if (len == 0)
1252                 ret = -ENOENT;
1253         else if (len > NAME_MAX && tmp[NAME_MAX] != 0)
1254                 ret = -ENAMETOOLONG;
1255
1256         if (ret) {
1257                 kfree(tmp);
1258                 tmp =  ERR_PTR(ret);
1259         }
1260         return tmp;
1261 }
1262
1263 #define ll_putname(filename) kfree(filename)
1264
1265 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1266 {
1267         struct inode *inode = file_inode(file);
1268         struct ll_sb_info *sbi = ll_i2sbi(inode);
1269         struct obd_ioctl_data *data;
1270         int rc = 0;
1271
1272         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1273                PFID(ll_inode2fid(inode)), inode, cmd);
1274
1275         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1276         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1277                 return -ENOTTY;
1278
1279         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1280         switch (cmd) {
1281         case FSFILT_IOC_GETFLAGS:
1282         case FSFILT_IOC_SETFLAGS:
1283                 return ll_iocontrol(inode, file, cmd, arg);
1284         case FSFILT_IOC_GETVERSION_OLD:
1285         case FSFILT_IOC_GETVERSION:
1286                 return put_user(inode->i_generation, (int __user *)arg);
1287         /* We need to special case any other ioctls we want to handle,
1288          * to send them to the MDS/OST as appropriate and to properly
1289          * network encode the arg field.
1290         case FSFILT_IOC_SETVERSION_OLD:
1291         case FSFILT_IOC_SETVERSION:
1292         */
1293         case LL_IOC_GET_MDTIDX: {
1294                 int mdtidx;
1295
1296                 mdtidx = ll_get_mdt_idx(inode);
1297                 if (mdtidx < 0)
1298                         return mdtidx;
1299
1300                 if (put_user((int)mdtidx, (int __user *)arg))
1301                         return -EFAULT;
1302
1303                 return 0;
1304         }
1305         case IOC_MDC_LOOKUP: {
1306                 struct ptlrpc_request *request = NULL;
1307                 int namelen, len = 0;
1308                 char *buf = NULL;
1309                 char *filename;
1310                 struct md_op_data *op_data;
1311
1312                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1313                 if (rc)
1314                         return rc;
1315                 data = (void *)buf;
1316
1317                 filename = data->ioc_inlbuf1;
1318                 namelen = strlen(filename);
1319
1320                 if (namelen < 1) {
1321                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1322                         rc = -EINVAL;
1323                         goto out_free;
1324                 }
1325
1326                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1327                                              0, LUSTRE_OPC_ANY, NULL);
1328                 if (IS_ERR(op_data)) {
1329                         rc = PTR_ERR(op_data);
1330                         goto out_free;
1331                 }
1332
1333                 op_data->op_valid = OBD_MD_FLID;
1334                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1335                 ll_finish_md_op_data(op_data);
1336                 if (rc < 0) {
1337                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1338                         goto out_free;
1339                 }
1340                 ptlrpc_req_finished(request);
1341 out_free:
1342                 obd_ioctl_freedata(buf, len);
1343                 return rc;
1344         }
1345         case LL_IOC_LMV_SETSTRIPE: {
1346                 struct lmv_user_md  *lum;
1347                 char            *buf = NULL;
1348                 char            *filename;
1349                 int              namelen = 0;
1350                 int              lumlen = 0;
1351                 int              len;
1352                 int              rc;
1353
1354                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1355                 if (rc)
1356                         return rc;
1357
1358                 data = (void *)buf;
1359                 if (!data->ioc_inlbuf1 || !data->ioc_inlbuf2 ||
1360                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0) {
1361                         rc = -EINVAL;
1362                         goto lmv_out_free;
1363                 }
1364
1365                 filename = data->ioc_inlbuf1;
1366                 namelen = data->ioc_inllen1;
1367
1368                 if (namelen < 1) {
1369                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1370                         rc = -EINVAL;
1371                         goto lmv_out_free;
1372                 }
1373                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1374                 lumlen = data->ioc_inllen2;
1375
1376                 if (lum->lum_magic != LMV_USER_MAGIC ||
1377                     lumlen != sizeof(*lum)) {
1378                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1379                                filename, lum->lum_magic, lumlen, -EFAULT);
1380                         rc = -EINVAL;
1381                         goto lmv_out_free;
1382                 }
1383
1384                 /**
1385                  * ll_dir_setdirstripe will be used to set dir stripe
1386                  *  mdc_create--->mdt_reint_create (with dirstripe)
1387                  */
1388                 rc = ll_dir_setdirstripe(inode, lum, filename);
1389 lmv_out_free:
1390                 obd_ioctl_freedata(buf, len);
1391                 return rc;
1392         }
1393         case LL_IOC_LOV_SETSTRIPE: {
1394                 struct lov_user_md_v3 lumv3;
1395                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1396                 struct lov_user_md_v1 __user *lumv1p = (void __user *)arg;
1397                 struct lov_user_md_v3 __user *lumv3p = (void __user *)arg;
1398
1399                 int set_default = 0;
1400
1401                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1402                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1403                         sizeof(lumv3p->lmm_objects[0]));
1404                 /* first try with v1 which is smaller than v3 */
1405                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1406                         return -EFAULT;
1407
1408                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1409                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1410                                 return -EFAULT;
1411                 }
1412
1413                 if (is_root_inode(inode))
1414                         set_default = 1;
1415
1416                 /* in v1 and v3 cases lumv1 points to data */
1417                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1418
1419                 return rc;
1420         }
1421         case LL_IOC_LMV_GETSTRIPE: {
1422                 struct lmv_user_md __user *lump = (void __user *)arg;
1423                 struct lmv_user_md lum;
1424                 struct lmv_user_md *tmp;
1425                 int lum_size;
1426                 int rc = 0;
1427                 int mdtindex;
1428
1429                 if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
1430                         return -EFAULT;
1431
1432                 if (lum.lum_magic != LMV_MAGIC_V1)
1433                         return -EINVAL;
1434
1435                 lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
1436                 tmp = kzalloc(lum_size, GFP_NOFS);
1437                 if (!tmp) {
1438                         rc = -ENOMEM;
1439                         goto free_lmv;
1440                 }
1441
1442                 *tmp = lum;
1443                 tmp->lum_type = LMV_STRIPE_TYPE;
1444                 tmp->lum_stripe_count = 1;
1445                 mdtindex = ll_get_mdt_idx(inode);
1446                 if (mdtindex < 0) {
1447                         rc = -ENOMEM;
1448                         goto free_lmv;
1449                 }
1450
1451                 tmp->lum_stripe_offset = mdtindex;
1452                 tmp->lum_objects[0].lum_mds = mdtindex;
1453                 memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
1454                        sizeof(struct lu_fid));
1455                 if (copy_to_user((void __user *)arg, tmp, lum_size)) {
1456                         rc = -EFAULT;
1457                         goto free_lmv;
1458                 }
1459 free_lmv:
1460                 kfree(tmp);
1461                 return rc;
1462         }
1463         case LL_IOC_LOV_SWAP_LAYOUTS:
1464                 return -EPERM;
1465         case LL_IOC_OBD_STATFS:
1466                 return ll_obd_statfs(inode, (void __user *)arg);
1467         case LL_IOC_LOV_GETSTRIPE:
1468         case LL_IOC_MDC_GETINFO:
1469         case IOC_MDC_GETFILEINFO:
1470         case IOC_MDC_GETFILESTRIPE: {
1471                 struct ptlrpc_request *request = NULL;
1472                 struct lov_user_md __user *lump;
1473                 struct lov_mds_md *lmm = NULL;
1474                 struct mdt_body *body;
1475                 char *filename = NULL;
1476                 int lmmsize;
1477
1478                 if (cmd == IOC_MDC_GETFILEINFO ||
1479                     cmd == IOC_MDC_GETFILESTRIPE) {
1480                         filename = ll_getname((const char __user *)arg);
1481                         if (IS_ERR(filename))
1482                                 return PTR_ERR(filename);
1483
1484                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1485                                                       &lmmsize, &request);
1486                 } else {
1487                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1488                 }
1489
1490                 if (request) {
1491                         body = req_capsule_server_get(&request->rq_pill,
1492                                                       &RMF_MDT_BODY);
1493                         LASSERT(body);
1494                 } else {
1495                         goto out_req;
1496                 }
1497
1498                 if (rc < 0) {
1499                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1500                                                cmd == LL_IOC_MDC_GETINFO)) {
1501                                 rc = 0;
1502                                 goto skip_lmm;
1503                         } else {
1504                                 goto out_req;
1505                         }
1506                 }
1507
1508                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1509                     cmd == LL_IOC_LOV_GETSTRIPE) {
1510                         lump = (struct lov_user_md __user *)arg;
1511                 } else {
1512                         struct lov_user_mds_data __user *lmdp;
1513
1514                         lmdp = (struct lov_user_mds_data __user *)arg;
1515                         lump = &lmdp->lmd_lmm;
1516                 }
1517                 if (copy_to_user(lump, lmm, lmmsize)) {
1518                         if (copy_to_user(lump, lmm, sizeof(*lump))) {
1519                                 rc = -EFAULT;
1520                                 goto out_req;
1521                         }
1522                         rc = -EOVERFLOW;
1523                 }
1524 skip_lmm:
1525                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1526                         struct lov_user_mds_data __user *lmdp;
1527                         lstat_t st = { 0 };
1528
1529                         st.st_dev     = inode->i_sb->s_dev;
1530                         st.st_mode    = body->mode;
1531                         st.st_nlink   = body->nlink;
1532                         st.st_uid     = body->uid;
1533                         st.st_gid     = body->gid;
1534                         st.st_rdev    = body->rdev;
1535                         st.st_size    = body->size;
1536                         st.st_blksize = PAGE_SIZE;
1537                         st.st_blocks  = body->blocks;
1538                         st.st_atime   = body->atime;
1539                         st.st_mtime   = body->mtime;
1540                         st.st_ctime   = body->ctime;
1541                         st.st_ino     = inode->i_ino;
1542
1543                         lmdp = (struct lov_user_mds_data __user *)arg;
1544                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st))) {
1545                                 rc = -EFAULT;
1546                                 goto out_req;
1547                         }
1548                 }
1549
1550 out_req:
1551                 ptlrpc_req_finished(request);
1552                 if (filename)
1553                         ll_putname(filename);
1554                 return rc;
1555         }
1556         case IOC_LOV_GETINFO: {
1557                 struct lov_user_mds_data __user *lumd;
1558                 struct lov_stripe_md *lsm;
1559                 struct lov_user_md __user *lum;
1560                 struct lov_mds_md *lmm;
1561                 int lmmsize;
1562                 lstat_t st;
1563
1564                 lumd = (struct lov_user_mds_data __user *)arg;
1565                 lum = &lumd->lmd_lmm;
1566
1567                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1568                 if (rc)
1569                         return rc;
1570
1571                 lmm = libcfs_kvzalloc(lmmsize, GFP_NOFS);
1572                 if (!lmm)
1573                         return -ENOMEM;
1574                 if (copy_from_user(lmm, lum, lmmsize)) {
1575                         rc = -EFAULT;
1576                         goto free_lmm;
1577                 }
1578
1579                 switch (lmm->lmm_magic) {
1580                 case LOV_USER_MAGIC_V1:
1581                         if (cpu_to_le32(LOV_USER_MAGIC_V1) == LOV_USER_MAGIC_V1)
1582                                 break;
1583                         /* swab objects first so that stripes num will be sane */
1584                         lustre_swab_lov_user_md_objects(
1585                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1586                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1587                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1588                         break;
1589                 case LOV_USER_MAGIC_V3:
1590                         if (cpu_to_le32(LOV_USER_MAGIC_V3) == LOV_USER_MAGIC_V3)
1591                                 break;
1592                         /* swab objects first so that stripes num will be sane */
1593                         lustre_swab_lov_user_md_objects(
1594                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1595                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1596                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1597                         break;
1598                 default:
1599                         rc = -EINVAL;
1600                         goto free_lmm;
1601                 }
1602
1603                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1604                 if (rc < 0) {
1605                         rc = -ENOMEM;
1606                         goto free_lmm;
1607                 }
1608
1609                 /* Perform glimpse_size operation. */
1610                 memset(&st, 0, sizeof(st));
1611
1612                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1613                 if (rc)
1614                         goto free_lsm;
1615
1616                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st))) {
1617                         rc = -EFAULT;
1618                         goto free_lsm;
1619                 }
1620
1621 free_lsm:
1622                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1623 free_lmm:
1624                 kvfree(lmm);
1625                 return rc;
1626         }
1627         case OBD_IOC_LLOG_CATINFO: {
1628                 return -EOPNOTSUPP;
1629         }
1630         case OBD_IOC_QUOTACHECK: {
1631                 struct obd_quotactl *oqctl;
1632                 int error = 0;
1633
1634                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1635                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1636                         return -EPERM;
1637
1638                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1639                 if (!oqctl)
1640                         return -ENOMEM;
1641                 oqctl->qc_type = arg;
1642                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1643                 if (rc < 0) {
1644                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1645                         error = rc;
1646                 }
1647
1648                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1649                 if (rc < 0)
1650                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1651
1652                 kfree(oqctl);
1653                 return error ?: rc;
1654         }
1655         case OBD_IOC_POLL_QUOTACHECK: {
1656                 struct if_quotacheck *check;
1657
1658                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1659                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1660                         return -EPERM;
1661
1662                 check = kzalloc(sizeof(*check), GFP_NOFS);
1663                 if (!check)
1664                         return -ENOMEM;
1665
1666                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1667                                    NULL);
1668                 if (rc) {
1669                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1670                         if (copy_to_user((void __user *)arg, check,
1671                                          sizeof(*check)))
1672                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1673                         goto out_poll;
1674                 }
1675
1676                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1677                                    NULL);
1678                 if (rc) {
1679                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1680                         if (copy_to_user((void __user *)arg, check,
1681                                          sizeof(*check)))
1682                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1683                         goto out_poll;
1684                 }
1685 out_poll:
1686                 kfree(check);
1687                 return rc;
1688         }
1689         case LL_IOC_QUOTACTL: {
1690                 struct if_quotactl *qctl;
1691
1692                 qctl = kzalloc(sizeof(*qctl), GFP_NOFS);
1693                 if (!qctl)
1694                         return -ENOMEM;
1695
1696                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl))) {
1697                         rc = -EFAULT;
1698                         goto out_quotactl;
1699                 }
1700
1701                 rc = quotactl_ioctl(sbi, qctl);
1702
1703                 if (rc == 0 && copy_to_user((void __user *)arg, qctl,
1704                                             sizeof(*qctl)))
1705                         rc = -EFAULT;
1706
1707 out_quotactl:
1708                 kfree(qctl);
1709                 return rc;
1710         }
1711         case OBD_IOC_GETDTNAME:
1712         case OBD_IOC_GETMDNAME:
1713                 return ll_get_obd_name(inode, cmd, arg);
1714         case LL_IOC_FLUSHCTX:
1715                 return ll_flush_ctx(inode);
1716 #ifdef CONFIG_FS_POSIX_ACL
1717         case LL_IOC_RMTACL: {
1718                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT && is_root_inode(inode)) {
1719                         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1720
1721                         rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1722                         if (!rc)
1723                                 fd->fd_flags |= LL_FILE_RMTACL;
1724                         return rc;
1725                 } else {
1726                         return 0;
1727                 }
1728         }
1729 #endif
1730         case LL_IOC_GETOBDCOUNT: {
1731                 int count, vallen;
1732                 struct obd_export *exp;
1733
1734                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1735                         return -EFAULT;
1736
1737                 /* get ost count when count is zero, get mdt count otherwise */
1738                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1739                 vallen = sizeof(count);
1740                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1741                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1742                 if (rc) {
1743                         CERROR("get target count failed: %d\n", rc);
1744                         return rc;
1745                 }
1746
1747                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1748                         return -EFAULT;
1749
1750                 return 0;
1751         }
1752         case LL_IOC_PATH2FID:
1753                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1754                                  sizeof(struct lu_fid)))
1755                         return -EFAULT;
1756                 return 0;
1757         case LL_IOC_GET_CONNECT_FLAGS: {
1758                 return obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1759                                      (void __user *)arg);
1760         }
1761         case OBD_IOC_CHANGELOG_SEND:
1762         case OBD_IOC_CHANGELOG_CLEAR:
1763                 if (!capable(CFS_CAP_SYS_ADMIN))
1764                         return -EPERM;
1765
1766                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1767                                     sizeof(struct ioc_changelog));
1768                 return rc;
1769         case OBD_IOC_FID2PATH:
1770                 return ll_fid2path(inode, (void __user *)arg);
1771         case LL_IOC_HSM_REQUEST: {
1772                 struct hsm_user_request *hur;
1773                 ssize_t                  totalsize;
1774
1775                 hur = memdup_user((void __user *)arg, sizeof(*hur));
1776                 if (IS_ERR(hur))
1777                         return PTR_ERR(hur);
1778
1779                 /* Compute the whole struct size */
1780                 totalsize = hur_len(hur);
1781                 kfree(hur);
1782                 if (totalsize < 0)
1783                         return -E2BIG;
1784
1785                 /* Final size will be more than double totalsize */
1786                 if (totalsize >= MDS_MAXREQSIZE / 3)
1787                         return -E2BIG;
1788
1789                 hur = libcfs_kvzalloc(totalsize, GFP_NOFS);
1790                 if (!hur)
1791                         return -ENOMEM;
1792
1793                 /* Copy the whole struct */
1794                 if (copy_from_user(hur, (void __user *)arg, totalsize)) {
1795                         kvfree(hur);
1796                         return -EFAULT;
1797                 }
1798
1799                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1800                         const struct lu_fid *fid;
1801                         struct inode *f;
1802                         int i;
1803
1804                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1805                                 fid = &hur->hur_user_item[i].hui_fid;
1806                                 f = search_inode_for_lustre(inode->i_sb, fid);
1807                                 if (IS_ERR(f)) {
1808                                         rc = PTR_ERR(f);
1809                                         break;
1810                                 }
1811
1812                                 rc = ll_hsm_release(f);
1813                                 iput(f);
1814                                 if (rc != 0)
1815                                         break;
1816                         }
1817                 } else {
1818                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1819                                            hur, NULL);
1820                 }
1821
1822                 kvfree(hur);
1823
1824                 return rc;
1825         }
1826         case LL_IOC_HSM_PROGRESS: {
1827                 struct hsm_progress_kernel      hpk;
1828                 struct hsm_progress             hp;
1829
1830                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1831                         return -EFAULT;
1832
1833                 hpk.hpk_fid = hp.hp_fid;
1834                 hpk.hpk_cookie = hp.hp_cookie;
1835                 hpk.hpk_extent = hp.hp_extent;
1836                 hpk.hpk_flags = hp.hp_flags;
1837                 hpk.hpk_errval = hp.hp_errval;
1838                 hpk.hpk_data_version = 0;
1839
1840                 /* File may not exist in Lustre; all progress
1841                  * reported to Lustre root
1842                  */
1843                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1844                                    NULL);
1845                 return rc;
1846         }
1847         case LL_IOC_HSM_CT_START:
1848                 if (!capable(CFS_CAP_SYS_ADMIN))
1849                         return -EPERM;
1850
1851                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1852                                     sizeof(struct lustre_kernelcomm));
1853                 return rc;
1854
1855         case LL_IOC_HSM_COPY_START: {
1856                 struct hsm_copy *copy;
1857                 int              rc;
1858
1859                 copy = memdup_user((char __user *)arg, sizeof(*copy));
1860                 if (IS_ERR(copy))
1861                         return PTR_ERR(copy);
1862
1863                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1864                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1865                         rc = -EFAULT;
1866
1867                 kfree(copy);
1868                 return rc;
1869         }
1870         case LL_IOC_HSM_COPY_END: {
1871                 struct hsm_copy *copy;
1872                 int              rc;
1873
1874                 copy = memdup_user((char __user *)arg, sizeof(*copy));
1875                 if (IS_ERR(copy))
1876                         return PTR_ERR(copy);
1877
1878                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1879                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1880                         rc = -EFAULT;
1881
1882                 kfree(copy);
1883                 return rc;
1884         }
1885         default:
1886                 return obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1887                                      (void __user *)arg);
1888         }
1889 }
1890
1891 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1892 {
1893         struct inode *inode = file->f_mapping->host;
1894         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1895         struct ll_sb_info *sbi = ll_i2sbi(inode);
1896         int api32 = ll_need_32bit_api(sbi);
1897         loff_t ret = -EINVAL;
1898
1899         switch (origin) {
1900         case SEEK_SET:
1901                 break;
1902         case SEEK_CUR:
1903                 offset += file->f_pos;
1904                 break;
1905         case SEEK_END:
1906                 if (offset > 0)
1907                         goto out;
1908                 if (api32)
1909                         offset += LL_DIR_END_OFF_32BIT;
1910                 else
1911                         offset += LL_DIR_END_OFF;
1912                 break;
1913         default:
1914                 goto out;
1915         }
1916
1917         if (offset >= 0 &&
1918             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1919              (!api32 && offset <= LL_DIR_END_OFF))) {
1920                 if (offset != file->f_pos) {
1921                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1922                             (!api32 && offset == LL_DIR_END_OFF))
1923                                 fd->lfd_pos = MDS_DIR_END_OFF;
1924                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1925                                 fd->lfd_pos = offset << 32;
1926                         else
1927                                 fd->lfd_pos = offset;
1928                         file->f_pos = offset;
1929                         file->f_version = 0;
1930                 }
1931                 ret = offset;
1932         }
1933         goto out;
1934
1935 out:
1936         return ret;
1937 }
1938
1939 static int ll_dir_open(struct inode *inode, struct file *file)
1940 {
1941         return ll_file_open(inode, file);
1942 }
1943
1944 static int ll_dir_release(struct inode *inode, struct file *file)
1945 {
1946         return ll_file_release(inode, file);
1947 }
1948
1949 const struct file_operations ll_dir_operations = {
1950         .llseek   = ll_dir_seek,
1951         .open     = ll_dir_open,
1952         .release  = ll_dir_release,
1953         .read     = generic_read_dir,
1954         .iterate_shared  = ll_readdir,
1955         .unlocked_ioctl   = ll_dir_ioctl,
1956         .fsync    = ll_fsync,
1957 };