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

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/dir.c
  37  *
  38  * Directory code for lustre client.
  39  */
  40
  41 #include <linux/fs.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/mm.h>
  44 #include <asm/uaccess.h>
  45 #include <linux/buffer_head.h>   // for wait_on_buffer
  46 #include <linux/pagevec.h>
  47 #include <linux/prefetch.h>
  48
  49 #define DEBUG_SUBSYSTEM S_LLITE
  50
  51 #include <obd_support.h>
  52 #include <obd_class.h>
  53 #include <lustre_lib.h>
  54 #include <lustre/lustre_idl.h>
  55 #include <lustre_lite.h>
  56 #include <lustre_dlm.h>
  57 #include <lustre_fid.h>
  58 #include "llite_internal.h"
  59
  60 /*
  61  * (new) readdir implementation overview.
  62  *
  63  * Original lustre readdir implementation cached exact copy of raw directory
  64  * pages on the client. These pages were indexed in client page cache by
  65  * logical offset in the directory file. This design, while very simple and
  66  * intuitive had some inherent problems:
  67  *
  68  *     . it implies that byte offset to the directory entry serves as a
  69  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  70  *     ext3/htree directory entries may move due to splits, and more
  71  *     importantly,
  72  *
  73  *     . it is incompatible with the design of split directories for cmd3,
  74  *     that assumes that names are distributed across nodes based on their
  75  *     hash, and so readdir should be done in hash order.
  76  *
  77  * New readdir implementation does readdir in hash order, and uses hash of a
  78  * file name as a telldir/seekdir cookie. This led to number of complications:
  79  *
  80  *     . hash is not unique, so it cannot be used to index cached directory
  81  *     pages on the client (note, that it requires a whole pageful of hash
  82  *     collided entries to cause two pages to have identical hashes);
  83  *
  84  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  85  *     entry cookie. ext3/htree has the same problem and lustre implementation
  86  *     mimics their solution: seekdir(hash) positions directory at the first
  87  *     entry with the given hash.
  88  *
  89  * Client side.
  90  *
  91  * 0. caching
  92  *
  93  * Client caches directory pages using hash of the first entry as an index. As
  94  * noted above hash is not unique, so this solution doesn't work as is:
  95  * special processing is needed for "page hash chains" (i.e., sequences of
  96  * pages filled with entries all having the same hash value).
  97  *
  98  * First, such chains have to be detected. To this end, server returns to the
  99  * client the hash of the first entry on the page next to one returned. When
 100  * client detects that this hash is the same as hash of the first entry on the
 101  * returned page, page hash collision has to be handled. Pages in the
 102  * hash chain, except first one, are termed "overflow pages".
 103  *
 104  * Solution to index uniqueness problem is to not cache overflow
 105  * pages. Instead, when page hash collision is detected, all overflow pages
 106  * from emerging chain are immediately requested from the server and placed in
 107  * a special data structure (struct ll_dir_chain). This data structure is used
 108  * by ll_readdir() to process entries from overflow pages. When readdir
 109  * invocation finishes, overflow pages are discarded. If page hash collision
 110  * chain weren't completely processed, next call to readdir will again detect
 111  * page hash collision, again read overflow pages in, process next portion of
 112  * entries and again discard the pages. This is not as wasteful as it looks,
 113  * because, given reasonable hash, page hash collisions are extremely rare.
 114  *
 115  * 1. directory positioning
 116  *
 117  * When seekdir(hash) is called, original
 118  *
 119  *
 120  *
 121  *
 122  *
 123  *
 124  *
 125  *
 126  * Server.
 127  *
 128  * identification of and access to overflow pages
 129  *
 130  * page format
 131  *
 132  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 133  * a header lu_dirpage which describes the start/end hash, and whether this
 134  * page is empty (contains no dir entry) or hash collide with next page.
 135  * After client receives reply, several pages will be integrated into dir page
 136  * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
 137  * lu_dirpage for this integrated page will be adjusted. See
 138  * lmv_adjust_dirpages().
 139  *
 140  */
 141
 142 /* returns the page unlocked, but with a reference */
 143 static int ll_dir_filler(void *_hash, struct page *page0)
 144 {
 145         struct inode *inode = page0->mapping->host;
 146         int hash64 = ll_i2sbi(inode)->ll_flags & LL_SBI_64BIT_HASH;
 147         struct obd_export *exp = ll_i2sbi(inode)->ll_md_exp;
 148         struct ptlrpc_request *request;
 149         struct mdt_body *body;
 150         struct md_op_data *op_data;
 151         __u64 hash = *((__u64 *)_hash);
 152         struct page **page_pool;
 153         struct page *page;
 154         struct lu_dirpage *dp;
 155         int max_pages = ll_i2sbi(inode)->ll_md_brw_size >> PAGE_CACHE_SHIFT;
 156         int nrdpgs = 0; /* number of pages read actually */
 157         int npages;
 158         int i;
 159         int rc;
 160
 161         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) hash "LPU64"\n",
 162                inode->i_ino, inode->i_generation, inode, hash);
 163
 164         LASSERT(max_pages > 0 && max_pages <= MD_MAX_BRW_PAGES);
 165
 166         OBD_ALLOC(page_pool, sizeof(page) * max_pages);
 167         if (page_pool != NULL) {
 168                 page_pool[0] = page0;
 169         } else {
 170                 page_pool = &page0;
 171                 max_pages = 1;
 172         }
 173         for (npages = 1; npages < max_pages; npages++) {
 174                 page = page_cache_alloc_cold(inode->i_mapping);
 175                 if (!page)
 176                         break;
 177                 page_pool[npages] = page;
 178         }
 179
 180         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 181                                      LUSTRE_OPC_ANY, NULL);
 182         op_data->op_npages = npages;
 183         op_data->op_offset = hash;
 184         rc = md_readpage(exp, op_data, page_pool, &request);
 185         ll_finish_md_op_data(op_data);
 186         if (rc == 0) {
 187                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 188                 /* Checked by mdc_readpage() */
 189                 LASSERT(body != NULL);
 190
 191                 if (body->valid & OBD_MD_FLSIZE)
 192                         cl_isize_write(inode, body->size);
 193
 194                 nrdpgs = (request->rq_bulk->bd_nob_transferred+PAGE_CACHE_SIZE-1)
 195                          >> PAGE_CACHE_SHIFT;
 196                 SetPageUptodate(page0);
 197         }
 198         unlock_page(page0);
 199         ptlrpc_req_finished(request);
 200
 201         CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages);
 202
 203         ll_pagevec_init(&lru_pvec, 0);
 204         for (i = 1; i < npages; i++) {
 205                 unsigned long offset;
 206                 int ret;
 207
 208                 page = page_pool[i];
 209
 210                 if (rc < 0 || i >= nrdpgs) {
 211                         page_cache_release(page);
 212                         continue;
 213                 }
 214
 215                 SetPageUptodate(page);
 216
 217                 dp = kmap(page);
 218                 hash = le64_to_cpu(dp->ldp_hash_start);
 219                 kunmap(page);
 220
 221                 offset = hash_x_index(hash, hash64);
 222
 223                 prefetchw(&page->flags);
 224                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
 225                                             GFP_KERNEL);
 226                 if (ret == 0) {
 227                         unlock_page(page);
 228                         if (ll_pagevec_add(&lru_pvec, page) == 0)
 229                                 ll_pagevec_lru_add_file(&lru_pvec);
 230                 } else {
 231                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
 232                                " %d\n", offset, ret);
 233                 }
 234                 page_cache_release(page);
 235         }
 236         ll_pagevec_lru_add_file(&lru_pvec);
 237
 238         if (page_pool != &page0)
 239                 OBD_FREE(page_pool, sizeof(struct page *) * max_pages);
 240         return rc;
 241 }
 242
 243 static void ll_check_page(struct inode *dir, struct page *page)
 244 {
 245         /* XXX: check page format later */
 246         SetPageChecked(page);
 247 }
 248
 249 void ll_release_page(struct page *page, int remove)
 250 {
 251         kunmap(page);
 252         if (remove) {
 253                 lock_page(page);
 254                 if (likely(page->mapping != NULL))
 255                         truncate_complete_page(page->mapping, page);
 256                 unlock_page(page);
 257         }
 258         page_cache_release(page);
 259 }
 260
 261 /*
 262  * Find, kmap and return page that contains given hash.
 263  */
 264 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
 265                                        __u64 *start, __u64 *end)
 266 {
 267         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 268         struct address_space *mapping = dir->i_mapping;
 269         /*
 270          * Complement of hash is used as an index so that
 271          * radix_tree_gang_lookup() can be used to find a page with starting
 272          * hash _smaller_ than one we are looking for.
 273          */
 274         unsigned long offset = hash_x_index(*hash, hash64);
 275         struct page *page;
 276         int found;
 277
 278         TREE_READ_LOCK_IRQ(mapping);
 279         found = radix_tree_gang_lookup(&mapping->page_tree,
 280                                        (void **)&page, offset, 1);
 281         if (found > 0) {
 282                 struct lu_dirpage *dp;
 283
 284                 page_cache_get(page);
 285                 TREE_READ_UNLOCK_IRQ(mapping);
 286                 /*
 287                  * In contrast to find_lock_page() we are sure that directory
 288                  * page cannot be truncated (while DLM lock is held) and,
 289                  * hence, can avoid restart.
 290                  *
 291                  * In fact, page cannot be locked here at all, because
 292                  * ll_dir_filler() does synchronous io.
 293                  */
 294                 wait_on_page_locked(page);
 295                 if (PageUptodate(page)) {
 296                         dp = kmap(page);
 297                         if (BITS_PER_LONG == 32 && hash64) {
 298                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 299                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 300                                 *hash  = *hash >> 32;
 301                         } else {
 302                                 *start = le64_to_cpu(dp->ldp_hash_start);
 303                                 *end   = le64_to_cpu(dp->ldp_hash_end);
 304                         }
 305                         LASSERTF(*start <= *hash, "start = "LPX64",end = "
 306                                  LPX64",hash = "LPX64"\n", *start, *end, *hash);
 307                         CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash "LPU64"\n",
 308                                offset, *start, *end, *hash);
 309                         if (*hash > *end) {
 310                                 ll_release_page(page, 0);
 311                                 page = NULL;
 312                         } else if (*end != *start && *hash == *end) {
 313                                 /*
 314                                  * upon hash collision, remove this page,
 315                                  * otherwise put page reference, and
 316                                  * ll_get_dir_page() will issue RPC to fetch
 317                                  * the page we want.
 318                                  */
 319                                 ll_release_page(page,
 320                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
 321                                 page = NULL;
 322                         }
 323                 } else {
 324                         page_cache_release(page);
 325                         page = ERR_PTR(-EIO);
 326                 }
 327
 328         } else {
 329                 TREE_READ_UNLOCK_IRQ(mapping);
 330                 page = NULL;
 331         }
 332         return page;
 333 }
 334
 335 struct page *ll_get_dir_page(struct inode *dir, __u64 hash,
 336                              struct ll_dir_chain *chain)
 337 {
 338         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 339         struct address_space *mapping = dir->i_mapping;
 340         struct lustre_handle lockh;
 341         struct lu_dirpage *dp;
 342         struct page *page;
 343         ldlm_mode_t mode;
 344         int rc;
 345         __u64 start = 0;
 346         __u64 end = 0;
 347         __u64 lhash = hash;
 348         struct ll_inode_info *lli = ll_i2info(dir);
 349         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 350
 351         mode = LCK_PR;
 352         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 353                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
 354         if (!rc) {
 355                 struct ldlm_enqueue_info einfo = {
 356                         .ei_type = LDLM_IBITS,
 357                         .ei_mode = mode,
 358                         .ei_cb_bl = ll_md_blocking_ast,
 359                         .ei_cb_cp = ldlm_completion_ast,
 360                 };
 361                 struct lookup_intent it = { .it_op = IT_READDIR };
 362                 struct ptlrpc_request *request;
 363                 struct md_op_data *op_data;
 364
 365                 op_data = ll_prep_md_op_data(NULL, dir, NULL, NULL, 0, 0,
 366                 LUSTRE_OPC_ANY, NULL);
 367                 if (IS_ERR(op_data))
 368                         return (void *)op_data;
 369
 370                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
 371                                 op_data, &lockh, NULL, 0, NULL, 0);
 372
 373                 ll_finish_md_op_data(op_data);
 374
 375                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 376                 if (request)
 377                         ptlrpc_req_finished(request);
 378                 if (rc < 0) {
 379                         CERROR("lock enqueue: "DFID" at "LPU64": rc %d\n",
 380                                 PFID(ll_inode2fid(dir)), hash, rc);
 381                         return ERR_PTR(rc);
 382                 }
 383
 384                 CDEBUG(D_INODE, "setting lr_lvb_inode to inode %p (%lu/%u)\n",
 385                        dir, dir->i_ino, dir->i_generation);
 386                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp,
 387                                  &it.d.lustre.it_lock_handle, dir, NULL);
 388         } else {
 389                 /* for cross-ref object, l_ast_data of the lock may not be set,
 390                  * we reset it here */
 391                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
 392                                  dir, NULL);
 393         }
 394         ldlm_lock_dump_handle(D_OTHER, &lockh);
 395
 396         mutex_lock(&lli->lli_readdir_mutex);
 397         page = ll_dir_page_locate(dir, &lhash, &start, &end);
 398         if (IS_ERR(page)) {
 399                 CERROR("dir page locate: "DFID" at "LPU64": rc %ld\n",
 400                        PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
 401                 GOTO(out_unlock, page);
 402         } else if (page != NULL) {
 403                 /*
 404                  * XXX nikita: not entirely correct handling of a corner case:
 405                  * suppose hash chain of entries with hash value HASH crosses
 406                  * border between pages P0 and P1. First both P0 and P1 are
 407                  * cached, seekdir() is called for some entry from the P0 part
 408                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 409                  * happens and finds P1, as it starts with matching hash
 410                  * value. Remaining entries from P0 part of the chain are
 411                  * skipped. (Is that really a bug?)
 412                  *
 413                  * Possible solutions: 0. don't cache P1 is such case, handle
 414                  * it as an "overflow" page. 1. invalidate all pages at
 415                  * once. 2. use HASH|1 as an index for P1.
 416                  */
 417                 GOTO(hash_collision, page);
 418         }
 419
 420         page = read_cache_page(mapping, hash_x_index(hash, hash64),
 421                                ll_dir_filler, &lhash);
 422         if (IS_ERR(page)) {
 423                 CERROR("read cache page: "DFID" at "LPU64": rc %ld\n",
 424                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 425                 GOTO(out_unlock, page);
 426         }
 427
 428         wait_on_page_locked(page);
 429         (void)kmap(page);
 430         if (!PageUptodate(page)) {
 431                 CERROR("page not updated: "DFID" at "LPU64": rc %d\n",
 432                        PFID(ll_inode2fid(dir)), hash, -5);
 433                 goto fail;
 434         }
 435         if (!PageChecked(page))
 436                 ll_check_page(dir, page);
 437         if (PageError(page)) {
 438                 CERROR("page error: "DFID" at "LPU64": rc %d\n",
 439                        PFID(ll_inode2fid(dir)), hash, -5);
 440                 goto fail;
 441         }
 442 hash_collision:
 443         dp = page_address(page);
 444         if (BITS_PER_LONG == 32 && hash64) {
 445                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 446                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 447                 lhash = hash >> 32;
 448         } else {
 449                 start = le64_to_cpu(dp->ldp_hash_start);
 450                 end   = le64_to_cpu(dp->ldp_hash_end);
 451                 lhash = hash;
 452         }
 453         if (end == start) {
 454                 LASSERT(start == lhash);
 455                 CWARN("Page-wide hash collision: "LPU64"\n", end);
 456                 if (BITS_PER_LONG == 32 && hash64)
 457                         CWARN("Real page-wide hash collision at ["LPU64" "LPU64
 458                               "] with hash "LPU64"\n",
 459                               le64_to_cpu(dp->ldp_hash_start),
 460                               le64_to_cpu(dp->ldp_hash_end), hash);
 461                 /*
 462                  * Fetch whole overflow chain...
 463                  *
 464                  * XXX not yet.
 465                  */
 466                 goto fail;
 467         }
 468 out_unlock:
 469         mutex_unlock(&lli->lli_readdir_mutex);
 470         ldlm_lock_decref(&lockh, mode);
 471         return page;
 472
 473 fail:
 474         ll_release_page(page, 1);
 475         page = ERR_PTR(-EIO);
 476         goto out_unlock;
 477 }
 478
 479 int ll_dir_read(struct inode *inode, struct dir_context *ctx)
 480 {
 481         struct ll_inode_info *info       = ll_i2info(inode);
 482         struct ll_sb_info    *sbi       = ll_i2sbi(inode);
 483         __u64              pos          = ctx->pos;
 484         int                api32      = ll_need_32bit_api(sbi);
 485         int                hash64     = sbi->ll_flags & LL_SBI_64BIT_HASH;
 486         struct page       *page;
 487         struct ll_dir_chain   chain;
 488         int                done = 0;
 489         int                rc = 0;
 490
 491         ll_dir_chain_init(&chain);
 492
 493         page = ll_get_dir_page(inode, pos, &chain);
 494
 495         while (rc == 0 && !done) {
 496                 struct lu_dirpage *dp;
 497                 struct lu_dirent  *ent;
 498
 499                 if (!IS_ERR(page)) {
 500                         /*
 501                          * If page is empty (end of directory is reached),
 502                          * use this value.
 503                          */
 504                         __u64 hash = MDS_DIR_END_OFF;
 505                         __u64 next;
 506
 507                         dp = page_address(page);
 508                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 509                              ent = lu_dirent_next(ent)) {
 510                                 __u16     type;
 511                                 int         namelen;
 512                                 struct lu_fid  fid;
 513                                 __u64     lhash;
 514                                 __u64     ino;
 515
 516                                 /*
 517                                  * XXX: implement correct swabbing here.
 518                                  */
 519
 520                                 hash = le64_to_cpu(ent->lde_hash);
 521                                 if (hash < pos)
 522                                         /*
 523                                          * Skip until we find target hash
 524                                          * value.
 525                                          */
 526                                         continue;
 527
 528                                 namelen = le16_to_cpu(ent->lde_namelen);
 529                                 if (namelen == 0)
 530                                         /*
 531                                          * Skip dummy record.
 532                                          */
 533                                         continue;
 534
 535                                 if (api32 && hash64)
 536                                         lhash = hash >> 32;
 537                                 else
 538                                         lhash = hash;
 539                                 fid_le_to_cpu(&fid, &ent->lde_fid);
 540                                 ino = cl_fid_build_ino(&fid, api32);
 541                                 type = ll_dirent_type_get(ent);
 542                                 ctx->pos = lhash;
 543                                 /* For 'll_nfs_get_name_filldir()', it will try
 544                                  * to access the 'ent' through its 'lde_name',
 545                                  * so the parameter 'name' for 'ctx->actor()'
 546                                  * must be part of the 'ent'.
 547                                  */
 548                                 done = !dir_emit(ctx, ent->lde_name,
 549                                                  namelen, ino, type);
 550                         }
 551                         next = le64_to_cpu(dp->ldp_hash_end);
 552                         if (!done) {
 553                                 pos = next;
 554                                 if (pos == MDS_DIR_END_OFF) {
 555                                         /*
 556                                          * End of directory reached.
 557                                          */
 558                                         done = 1;
 559                                         ll_release_page(page, 0);
 560                                 } else if (1 /* chain is exhausted*/) {
 561                                         /*
 562                                          * Normal case: continue to the next
 563                                          * page.
 564                                          */
 565                                         ll_release_page(page,
 566                                             le32_to_cpu(dp->ldp_flags) &
 567                                                         LDF_COLLIDE);
 568                                         next = pos;
 569                                         page = ll_get_dir_page(inode, pos,
 570                                                                &chain);
 571                                 } else {
 572                                         /*
 573                                          * go into overflow page.
 574                                          */
 575                                         LASSERT(le32_to_cpu(dp->ldp_flags) &
 576                                                 LDF_COLLIDE);
 577                                         ll_release_page(page, 1);
 578                                 }
 579                         } else {
 580                                 pos = hash;
 581                                 ll_release_page(page, 0);
 582                         }
 583                 } else {
 584                         rc = PTR_ERR(page);
 585                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 586                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 587                 }
 588         }
 589
 590         ctx->pos = pos;
 591         ll_dir_chain_fini(&chain);
 592         return rc;
 593 }
 594
 595 static int ll_readdir(struct file *filp, struct dir_context *ctx)
 596 {
 597         struct inode            *inode  = filp->f_dentry->d_inode;
 598         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 599         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 600         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 601         int                     api32   = ll_need_32bit_api(sbi);
 602         int                     rc;
 603
 604         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu "
 605                " 32bit_api %d\n", inode->i_ino, inode->i_generation,
 606                inode, (unsigned long)lfd->lfd_pos, i_size_read(inode), api32);
 607
 608         if (lfd->lfd_pos == MDS_DIR_END_OFF)
 609                 /*
 610                  * end-of-file.
 611                  */
 612                 GOTO(out, rc = 0);
 613
 614         ctx->pos = lfd->lfd_pos;
 615         rc = ll_dir_read(inode, ctx);
 616         lfd->lfd_pos = ctx->pos;
 617         if (ctx->pos == MDS_DIR_END_OFF) {
 618                 if (api32)
 619                         ctx->pos = LL_DIR_END_OFF_32BIT;
 620                 else
 621                         ctx->pos = LL_DIR_END_OFF;
 622         } else {
 623                 if (api32 && hash64)
 624                         ctx->pos >>= 32;
 625         }
 626         filp->f_version = inode->i_version;
 627
 628 out:
 629         if (!rc)
 630                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 631
 632         return rc;
 633 }
 634
 635 int ll_send_mgc_param(struct obd_export *mgc, char *string)
 636 {
 637         struct mgs_send_param *msp;
 638         int rc = 0;
 639
 640         OBD_ALLOC_PTR(msp);
 641         if (!msp)
 642                 return -ENOMEM;
 643
 644         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 645         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 646                                 sizeof(struct mgs_send_param), msp, NULL);
 647         if (rc)
 648                 CERROR("Failed to set parameter: %d\n", rc);
 649         OBD_FREE_PTR(msp);
 650
 651         return rc;
 652 }
 653
 654 int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
 655                         char *filename)
 656 {
 657         struct ptlrpc_request *request = NULL;
 658         struct md_op_data *op_data;
 659         struct ll_sb_info *sbi = ll_i2sbi(dir);
 660         int mode;
 661         int err;
 662
 663         mode = (0755 & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
 664         op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
 665                                      strlen(filename), mode, LUSTRE_OPC_MKDIR,
 666                                      lump);
 667         if (IS_ERR(op_data))
 668                 GOTO(err_exit, err = PTR_ERR(op_data));
 669
 670         op_data->op_cli_flags |= CLI_SET_MEA;
 671         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 672                         from_kuid(&init_user_ns, current_fsuid()),
 673                         from_kgid(&init_user_ns, current_fsgid()),
 674                         cfs_curproc_cap_pack(), 0, &request);
 675         ll_finish_md_op_data(op_data);
 676         if (err)
 677                 GOTO(err_exit, err);
 678 err_exit:
 679         ptlrpc_req_finished(request);
 680         return err;
 681 }
 682
 683 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 684                      int set_default)
 685 {
 686         struct ll_sb_info *sbi = ll_i2sbi(inode);
 687         struct md_op_data *op_data;
 688         struct ptlrpc_request *req = NULL;
 689         int rc = 0;
 690         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 691         struct obd_device *mgc = lsi->lsi_mgc;
 692         int lum_size;
 693
 694         if (lump != NULL) {
 695                 /*
 696                  * This is coming from userspace, so should be in
 697                  * local endian.  But the MDS would like it in little
 698                  * endian, so we swab it before we send it.
 699                  */
 700                 switch (lump->lmm_magic) {
 701                 case LOV_USER_MAGIC_V1: {
 702                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 703                                 lustre_swab_lov_user_md_v1(lump);
 704                         lum_size = sizeof(struct lov_user_md_v1);
 705                         break;
 706                 }
 707                 case LOV_USER_MAGIC_V3: {
 708                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 709                                 lustre_swab_lov_user_md_v3(
 710                                         (struct lov_user_md_v3 *)lump);
 711                         lum_size = sizeof(struct lov_user_md_v3);
 712                         break;
 713                 }
 714                 default: {
 715                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 716                                         " %#08x != %#08x nor %#08x\n",
 717                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 718                                         LOV_USER_MAGIC_V3);
 719                         return -EINVAL;
 720                 }
 721                 }
 722         } else {
 723                 lum_size = sizeof(struct lov_user_md_v1);
 724         }
 725
 726         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 727                                      LUSTRE_OPC_ANY, NULL);
 728         if (IS_ERR(op_data))
 729                 return PTR_ERR(op_data);
 730
 731         if (lump != NULL && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC))
 732                 op_data->op_cli_flags |= CLI_SET_MEA;
 733
 734         /* swabbing is done in lov_setstripe() on server side */
 735         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 736                         NULL, 0, &req, NULL);
 737         ll_finish_md_op_data(op_data);
 738         ptlrpc_req_finished(req);
 739         if (rc) {
 740                 if (rc != -EPERM && rc != -EACCES)
 741                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 742         }
 743
 744         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 745          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 746          need to make the distinction between the 2 versions */
 747         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 748                 char *param = NULL;
 749                 char *buf;
 750
 751                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 752                 if (param == NULL)
 753                         GOTO(end, rc = -ENOMEM);
 754
 755                 buf = param;
 756                 /* Get fsname and assume devname to be -MDT0000. */
 757                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 758                 strcat(buf, "-MDT0000.lov");
 759                 buf += strlen(buf);
 760
 761                 /* Set root stripesize */
 762                 sprintf(buf, ".stripesize=%u",
 763                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 764                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 765                 if (rc)
 766                         GOTO(end, rc);
 767
 768                 /* Set root stripecount */
 769                 sprintf(buf, ".stripecount=%hd",
 770                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 771                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 772                 if (rc)
 773                         GOTO(end, rc);
 774
 775                 /* Set root stripeoffset */
 776                 sprintf(buf, ".stripeoffset=%hd",
 777                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 778                         (typeof(lump->lmm_stripe_offset))(-1));
 779                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 780
 781 end:
 782                 if (param != NULL)
 783                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 784         }
 785         return rc;
 786 }
 787
 788 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 789                      int *lmm_size, struct ptlrpc_request **request)
 790 {
 791         struct ll_sb_info *sbi = ll_i2sbi(inode);
 792         struct mdt_body   *body;
 793         struct lov_mds_md *lmm = NULL;
 794         struct ptlrpc_request *req = NULL;
 795         int rc, lmmsize;
 796         struct md_op_data *op_data;
 797
 798         rc = ll_get_max_mdsize(sbi, &lmmsize);
 799         if (rc)
 800                 return rc;
 801
 802         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 803                                      0, lmmsize, LUSTRE_OPC_ANY,
 804                                      NULL);
 805         if (IS_ERR(op_data))
 806                 return PTR_ERR(op_data);
 807
 808         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 809         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 810         ll_finish_md_op_data(op_data);
 811         if (rc < 0) {
 812                 CDEBUG(D_INFO, "md_getattr failed on inode "
 813                        "%lu/%u: rc %d\n", inode->i_ino,
 814                        inode->i_generation, rc);
 815                 GOTO(out, rc);
 816         }
 817
 818         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 819         LASSERT(body != NULL);
 820
 821         lmmsize = body->eadatasize;
 822
 823         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 824             lmmsize == 0) {
 825                 GOTO(out, rc = -ENODATA);
 826         }
 827
 828         lmm = req_capsule_server_sized_get(&req->rq_pill,
 829                                            &RMF_MDT_MD, lmmsize);
 830         LASSERT(lmm != NULL);
 831
 832         /*
 833          * This is coming from the MDS, so is probably in
 834          * little endian.  We convert it to host endian before
 835          * passing it to userspace.
 836          */
 837         /* We don't swab objects for directories */
 838         switch (le32_to_cpu(lmm->lmm_magic)) {
 839         case LOV_MAGIC_V1:
 840                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 841                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 842                 break;
 843         case LOV_MAGIC_V3:
 844                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 845                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 846                 break;
 847         default:
 848                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 849                 rc = -EPROTO;
 850         }
 851 out:
 852         *lmmp = lmm;
 853         *lmm_size = lmmsize;
 854         *request = req;
 855         return rc;
 856 }
 857
 858 /*
 859  *  Get MDT index for the inode.
 860  */
 861 int ll_get_mdt_idx(struct inode *inode)
 862 {
 863         struct ll_sb_info *sbi = ll_i2sbi(inode);
 864         struct md_op_data *op_data;
 865         int rc, mdtidx;
 866
 867         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 868                                      0, LUSTRE_OPC_ANY, NULL);
 869         if (IS_ERR(op_data))
 870                 return PTR_ERR(op_data);
 871
 872         op_data->op_flags |= MF_GET_MDT_IDX;
 873         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 874         mdtidx = op_data->op_mds;
 875         ll_finish_md_op_data(op_data);
 876         if (rc < 0) {
 877                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 878                 return rc;
 879         }
 880         return mdtidx;
 881 }
 882
 883 /**
 884  * Generic handler to do any pre-copy work.
 885  *
 886  * It send a first hsm_progress (with extent length == 0) to coordinator as a
 887  * first information for it that real work has started.
 888  *
 889  * Moreover, for a ARCHIVE request, it will sample the file data version and
 890  * store it in \a copy.
 891  *
 892  * \return 0 on success.
 893  */
 894 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 895 {
 896         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 897         struct hsm_progress_kernel       hpk;
 898         int                              rc;
 899
 900         /* Forge a hsm_progress based on data from copy. */
 901         hpk.hpk_fid = copy->hc_hai.hai_fid;
 902         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 903         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 904         hpk.hpk_extent.length = 0;
 905         hpk.hpk_flags = 0;
 906         hpk.hpk_errval = 0;
 907         hpk.hpk_data_version = 0;
 908
 909
 910         /* For archive request, we need to read the current file version. */
 911         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 912                 struct inode    *inode;
 913                 __u64            data_version = 0;
 914
 915                 /* Get inode for this fid */
 916                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 917                 if (IS_ERR(inode)) {
 918                         hpk.hpk_flags |= HP_FLAG_RETRY;
 919                         /* hpk_errval is >= 0 */
 920                         hpk.hpk_errval = -PTR_ERR(inode);
 921                         GOTO(progress, rc = PTR_ERR(inode));
 922                 }
 923
 924                 /* Read current file data version */
 925                 rc = ll_data_version(inode, &data_version, 1);
 926                 iput(inode);
 927                 if (rc != 0) {
 928                         CDEBUG(D_HSM, "Could not read file data version of "
 929                                       DFID" (rc = %d). Archive request ("
 930                                       LPX64") could not be done.\n",
 931                                       PFID(&copy->hc_hai.hai_fid), rc,
 932                                       copy->hc_hai.hai_cookie);
 933                         hpk.hpk_flags |= HP_FLAG_RETRY;
 934                         /* hpk_errval must be >= 0 */
 935                         hpk.hpk_errval = -rc;
 936                         GOTO(progress, rc);
 937                 }
 938
 939                 /* Store it the hsm_copy for later copytool use.
 940                  * Always modified even if no lsm. */
 941                 copy->hc_data_version = data_version;
 942         }
 943
 944 progress:
 945         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 946                            &hpk, NULL);
 947
 948         return rc;
 949 }
 950
 951 /**
 952  * Generic handler to do any post-copy work.
 953  *
 954  * It will send the last hsm_progress update to coordinator to inform it
 955  * that copy is finished and whether it was successful or not.
 956  *
 957  * Moreover,
 958  * - for ARCHIVE request, it will sample the file data version and compare it
 959  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 960  *   will be considered as failed.
 961  * - for RESTORE request, it will sample the file data version and send it to
 962  *   coordinator which is useful if the file was imported as 'released'.
 963  *
 964  * \return 0 on success.
 965  */
 966 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 967 {
 968         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 969         struct hsm_progress_kernel       hpk;
 970         int                              rc;
 971
 972         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 973         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 974          * initialized if copy_end was called with copy == NULL.
 975          */
 976
 977         /* Forge a hsm_progress based on data from copy. */
 978         hpk.hpk_fid = copy->hc_hai.hai_fid;
 979         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 980         hpk.hpk_extent = copy->hc_hai.hai_extent;
 981         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 982         hpk.hpk_errval = copy->hc_errval;
 983         hpk.hpk_data_version = 0;
 984
 985         /* For archive request, we need to check the file data was not changed.
 986          *
 987          * For restore request, we need to send the file data version, this is
 988          * useful when the file was created using hsm_import.
 989          */
 990         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 991              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 992             (copy->hc_errval == 0)) {
 993                 struct inode    *inode;
 994                 __u64            data_version = 0;
 995
 996                 /* Get lsm for this fid */
 997                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 998                 if (IS_ERR(inode)) {
 999                         hpk.hpk_flags |= HP_FLAG_RETRY;
1000                         /* hpk_errval must be >= 0 */
1001                         hpk.hpk_errval = -PTR_ERR(inode);
1002                         GOTO(progress, rc = PTR_ERR(inode));
1003                 }
1004
1005                 rc = ll_data_version(inode, &data_version,
1006                                      copy->hc_hai.hai_action == HSMA_ARCHIVE);
1007                 iput(inode);
1008                 if (rc) {
1009                         CDEBUG(D_HSM, "Could not read file data version. "
1010                                       "Request could not be confirmed.\n");
1011                         if (hpk.hpk_errval == 0)
1012                                 hpk.hpk_errval = -rc;
1013                         GOTO(progress, rc);
1014                 }
1015
1016                 /* Store it the hsm_copy for later copytool use.
1017                  * Always modified even if no lsm. */
1018                 hpk.hpk_data_version = data_version;
1019
1020                 /* File could have been stripped during archiving, so we need
1021                  * to check anyway. */
1022                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1023                     (copy->hc_data_version != data_version)) {
1024                         CDEBUG(D_HSM, "File data version mismatched. "
1025                               "File content was changed during archiving. "
1026                                DFID", start:"LPX64" current:"LPX64"\n",
1027                                PFID(&copy->hc_hai.hai_fid),
1028                                copy->hc_data_version, data_version);
1029                         /* File was changed, send error to cdt. Do not ask for
1030                          * retry because if a file is modified frequently,
1031                          * the cdt will loop on retried archive requests.
1032                          * The policy engine will ask for a new archive later
1033                          * when the file will not be modified for some tunable
1034                          * time */
1035                         /* we do not notify caller */
1036                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
1037                         /* hpk_errval must be >= 0 */
1038                         hpk.hpk_errval = EBUSY;
1039                 }
1040
1041         }
1042
1043 progress:
1044         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1045                            &hpk, NULL);
1046
1047         return rc;
1048 }
1049
1050
1051 static int copy_and_ioctl(int cmd, struct obd_export *exp,
1052                           const void __user *data, size_t size)
1053 {
1054         void *copy;
1055         int rc;
1056
1057         OBD_ALLOC(copy, size);
1058         if (copy == NULL)
1059                 return -ENOMEM;
1060
1061         if (copy_from_user(copy, data, size)) {
1062                 rc = -EFAULT;
1063                 goto out;
1064         }
1065
1066         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
1067 out:
1068         OBD_FREE(copy, size);
1069
1070         return rc;
1071 }
1072
1073 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1074 {
1075         int cmd = qctl->qc_cmd;
1076         int type = qctl->qc_type;
1077         int id = qctl->qc_id;
1078         int valid = qctl->qc_valid;
1079         int rc = 0;
1080
1081         switch (cmd) {
1082         case LUSTRE_Q_INVALIDATE:
1083         case LUSTRE_Q_FINVALIDATE:
1084         case Q_QUOTAON:
1085         case Q_QUOTAOFF:
1086         case Q_SETQUOTA:
1087         case Q_SETINFO:
1088                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1089                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1090                         return -EPERM;
1091                 break;
1092         case Q_GETQUOTA:
1093                 if (((type == USRQUOTA &&
1094                       !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
1095                      (type == GRPQUOTA &&
1096                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
1097                     (!capable(CFS_CAP_SYS_ADMIN) ||
1098                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
1099                         return -EPERM;
1100                 break;
1101         case Q_GETINFO:
1102                 break;
1103         default:
1104                 CERROR("unsupported quotactl op: %#x\n", cmd);
1105                 return -ENOTTY;
1106         }
1107
1108         if (valid != QC_GENERAL) {
1109                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1110                         return -EOPNOTSUPP;
1111
1112                 if (cmd == Q_GETINFO)
1113                         qctl->qc_cmd = Q_GETOINFO;
1114                 else if (cmd == Q_GETQUOTA)
1115                         qctl->qc_cmd = Q_GETOQUOTA;
1116                 else
1117                         return -EINVAL;
1118
1119                 switch (valid) {
1120                 case QC_MDTIDX:
1121                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1122                                            sizeof(*qctl), qctl, NULL);
1123                         break;
1124                 case QC_OSTIDX:
1125                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1126                                            sizeof(*qctl), qctl, NULL);
1127                         break;
1128                 case QC_UUID:
1129                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1130                                            sizeof(*qctl), qctl, NULL);
1131                         if (rc == -EAGAIN)
1132                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1133                                                    sbi->ll_dt_exp,
1134                                                    sizeof(*qctl), qctl, NULL);
1135                         break;
1136                 default:
1137                         rc = -EINVAL;
1138                         break;
1139                 }
1140
1141                 if (rc)
1142                         return rc;
1143
1144                 qctl->qc_cmd = cmd;
1145         } else {
1146                 struct obd_quotactl *oqctl;
1147
1148                 OBD_ALLOC_PTR(oqctl);
1149                 if (oqctl == NULL)
1150                         return -ENOMEM;
1151
1152                 QCTL_COPY(oqctl, qctl);
1153                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1154                 if (rc) {
1155                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
1156                                 oqctl->qc_cmd = Q_QUOTAOFF;
1157                                 obd_quotactl(sbi->ll_md_exp, oqctl);
1158                         }
1159                         OBD_FREE_PTR(oqctl);
1160                         return rc;
1161                 }
1162                 /* If QIF_SPACE is not set, client should collect the
1163                  * space usage from OSSs by itself */
1164                 if (cmd == Q_GETQUOTA &&
1165                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1166                     !oqctl->qc_dqblk.dqb_curspace) {
1167                         struct obd_quotactl *oqctl_tmp;
1168
1169                         OBD_ALLOC_PTR(oqctl_tmp);
1170                         if (oqctl_tmp == NULL)
1171                                 GOTO(out, rc = -ENOMEM);
1172
1173                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1174                         oqctl_tmp->qc_id = oqctl->qc_id;
1175                         oqctl_tmp->qc_type = oqctl->qc_type;
1176
1177                         /* collect space usage from OSTs */
1178                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1179                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1180                         if (!rc || rc == -EREMOTEIO) {
1181                                 oqctl->qc_dqblk.dqb_curspace =
1182                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1183                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1184                         }
1185
1186                         /* collect space & inode usage from MDTs */
1187                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1188                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1189                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1190                         if (!rc || rc == -EREMOTEIO) {
1191                                 oqctl->qc_dqblk.dqb_curspace +=
1192                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1193                                 oqctl->qc_dqblk.dqb_curinodes =
1194                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1195                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1196                         } else {
1197                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1198                         }
1199
1200                         OBD_FREE_PTR(oqctl_tmp);
1201                 }
1202 out:
1203                 QCTL_COPY(qctl, oqctl);
1204                 OBD_FREE_PTR(oqctl);
1205         }
1206
1207         return rc;
1208 }
1209
1210 static char *
1211 ll_getname(const char __user *filename)
1212 {
1213         int ret = 0, len;
1214         char *tmp = __getname();
1215
1216         if (!tmp)
1217                 return ERR_PTR(-ENOMEM);
1218
1219         len = strncpy_from_user(tmp, filename, PATH_MAX);
1220         if (len == 0)
1221                 ret = -ENOENT;
1222         else if (len > PATH_MAX)
1223                 ret = -ENAMETOOLONG;
1224
1225         if (ret) {
1226                 __putname(tmp);
1227                 tmp =  ERR_PTR(ret);
1228         }
1229         return tmp;
1230 }
1231
1232 #define ll_putname(filename) __putname(filename)
1233
1234 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1235 {
1236         struct inode *inode = file->f_dentry->d_inode;
1237         struct ll_sb_info *sbi = ll_i2sbi(inode);
1238         struct obd_ioctl_data *data;
1239         int rc = 0;
1240
1241         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1242                inode->i_ino, inode->i_generation, inode, cmd);
1243
1244         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1245         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1246                 return -ENOTTY;
1247
1248         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1249         switch(cmd) {
1250         case FSFILT_IOC_GETFLAGS:
1251         case FSFILT_IOC_SETFLAGS:
1252                 return ll_iocontrol(inode, file, cmd, arg);
1253         case FSFILT_IOC_GETVERSION_OLD:
1254         case FSFILT_IOC_GETVERSION:
1255                 return put_user(inode->i_generation, (int *)arg);
1256         /* We need to special case any other ioctls we want to handle,
1257          * to send them to the MDS/OST as appropriate and to properly
1258          * network encode the arg field.
1259         case FSFILT_IOC_SETVERSION_OLD:
1260         case FSFILT_IOC_SETVERSION:
1261         */
1262         case LL_IOC_GET_MDTIDX: {
1263                 int mdtidx;
1264
1265                 mdtidx = ll_get_mdt_idx(inode);
1266                 if (mdtidx < 0)
1267                         return mdtidx;
1268
1269                 if (put_user((int)mdtidx, (int*)arg))
1270                         return -EFAULT;
1271
1272                 return 0;
1273         }
1274         case IOC_MDC_LOOKUP: {
1275                 struct ptlrpc_request *request = NULL;
1276                 int namelen, len = 0;
1277                 char *buf = NULL;
1278                 char *filename;
1279                 struct md_op_data *op_data;
1280
1281                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1282                 if (rc)
1283                         return rc;
1284                 data = (void *)buf;
1285
1286                 filename = data->ioc_inlbuf1;
1287                 namelen = strlen(filename);
1288
1289                 if (namelen < 1) {
1290                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1291                         GOTO(out_free, rc = -EINVAL);
1292                 }
1293
1294                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1295                                              0, LUSTRE_OPC_ANY, NULL);
1296                 if (IS_ERR(op_data))
1297                         GOTO(out_free, rc = PTR_ERR(op_data));
1298
1299                 op_data->op_valid = OBD_MD_FLID;
1300                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1301                 ll_finish_md_op_data(op_data);
1302                 if (rc < 0) {
1303                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1304                         GOTO(out_free, rc);
1305                 }
1306                 ptlrpc_req_finished(request);
1307 out_free:
1308                 obd_ioctl_freedata(buf, len);
1309                 return rc;
1310         }
1311         case LL_IOC_LMV_SETSTRIPE: {
1312                 struct lmv_user_md  *lum;
1313                 char            *buf = NULL;
1314                 char            *filename;
1315                 int              namelen = 0;
1316                 int              lumlen = 0;
1317                 int              len;
1318                 int              rc;
1319
1320                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1321                 if (rc)
1322                         return rc;
1323
1324                 data = (void *)buf;
1325                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1326                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1327                         GOTO(lmv_out_free, rc = -EINVAL);
1328
1329                 filename = data->ioc_inlbuf1;
1330                 namelen = data->ioc_inllen1;
1331
1332                 if (namelen < 1) {
1333                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1334                         GOTO(lmv_out_free, rc = -EINVAL);
1335                 }
1336                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1337                 lumlen = data->ioc_inllen2;
1338
1339                 if (lum->lum_magic != LMV_USER_MAGIC ||
1340                     lumlen != sizeof(*lum)) {
1341                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1342                                filename, lum->lum_magic, lumlen, -EFAULT);
1343                         GOTO(lmv_out_free, rc = -EINVAL);
1344                 }
1345
1346                 /**
1347                  * ll_dir_setdirstripe will be used to set dir stripe
1348                  *  mdc_create--->mdt_reint_create (with dirstripe)
1349                  */
1350                 rc = ll_dir_setdirstripe(inode, lum, filename);
1351 lmv_out_free:
1352                 obd_ioctl_freedata(buf, len);
1353                 return rc;
1354
1355         }
1356         case LL_IOC_LOV_SETSTRIPE: {
1357                 struct lov_user_md_v3 lumv3;
1358                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1359                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1360                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1361
1362                 int set_default = 0;
1363
1364                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1365                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1366                         sizeof(lumv3p->lmm_objects[0]));
1367                 /* first try with v1 which is smaller than v3 */
1368                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1369                         return -EFAULT;
1370
1371                 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1372                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1373                                 return -EFAULT;
1374                 }
1375
1376                 if (inode->i_sb->s_root == file->f_dentry)
1377                         set_default = 1;
1378
1379                 /* in v1 and v3 cases lumv1 points to data */
1380                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1381
1382                 return rc;
1383         }
1384         case LL_IOC_LMV_GETSTRIPE: {
1385                 struct lmv_user_md *lump = (struct lmv_user_md *)arg;
1386                 struct lmv_user_md lum;
1387                 struct lmv_user_md *tmp;
1388                 int lum_size;
1389                 int rc = 0;
1390                 int mdtindex;
1391
1392                 if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
1393                         return -EFAULT;
1394
1395                 if (lum.lum_magic != LMV_MAGIC_V1)
1396                         return -EINVAL;
1397
1398                 lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
1399                 OBD_ALLOC(tmp, lum_size);
1400                 if (tmp == NULL)
1401                         GOTO(free_lmv, rc = -ENOMEM);
1402
1403                 memcpy(tmp, &lum, sizeof(lum));
1404                 tmp->lum_type = LMV_STRIPE_TYPE;
1405                 tmp->lum_stripe_count = 1;
1406                 mdtindex = ll_get_mdt_idx(inode);
1407                 if (mdtindex < 0)
1408                         GOTO(free_lmv, rc = -ENOMEM);
1409
1410                 tmp->lum_stripe_offset = mdtindex;
1411                 tmp->lum_objects[0].lum_mds = mdtindex;
1412                 memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
1413                        sizeof(struct lu_fid));
1414                 if (copy_to_user((void *)arg, tmp, lum_size))
1415                         GOTO(free_lmv, rc = -EFAULT);
1416 free_lmv:
1417                 if (tmp)
1418                         OBD_FREE(tmp, lum_size);
1419                 return rc;
1420         }
1421         case LL_IOC_REMOVE_ENTRY: {
1422                 char            *filename = NULL;
1423                 int              namelen = 0;
1424                 int              rc;
1425
1426                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1427                  * unsupported server, which might crash the server(LU-2730),
1428                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1429                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1430                  * server will support REINT_RMENTRY XXX*/
1431                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1432                         return -ENOTSUPP;
1433
1434                 filename = ll_getname((const char *)arg);
1435                 if (IS_ERR(filename))
1436                         return PTR_ERR(filename);
1437
1438                 namelen = strlen(filename);
1439                 if (namelen < 1)
1440                         GOTO(out_rmdir, rc = -EINVAL);
1441
1442                 rc = ll_rmdir_entry(inode, filename, namelen);
1443 out_rmdir:
1444                 if (filename)
1445                         ll_putname(filename);
1446                 return rc;
1447         }
1448         case LL_IOC_LOV_SWAP_LAYOUTS:
1449                 return -EPERM;
1450         case LL_IOC_OBD_STATFS:
1451                 return ll_obd_statfs(inode, (void *)arg);
1452         case LL_IOC_LOV_GETSTRIPE:
1453         case LL_IOC_MDC_GETINFO:
1454         case IOC_MDC_GETFILEINFO:
1455         case IOC_MDC_GETFILESTRIPE: {
1456                 struct ptlrpc_request *request = NULL;
1457                 struct lov_user_md *lump;
1458                 struct lov_mds_md *lmm = NULL;
1459                 struct mdt_body *body;
1460                 char *filename = NULL;
1461                 int lmmsize;
1462
1463                 if (cmd == IOC_MDC_GETFILEINFO ||
1464                     cmd == IOC_MDC_GETFILESTRIPE) {
1465                         filename = ll_getname((const char *)arg);
1466                         if (IS_ERR(filename))
1467                                 return PTR_ERR(filename);
1468
1469                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1470                                                       &lmmsize, &request);
1471                 } else {
1472                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1473                 }
1474
1475                 if (request) {
1476                         body = req_capsule_server_get(&request->rq_pill,
1477                                                       &RMF_MDT_BODY);
1478                         LASSERT(body != NULL);
1479                 } else {
1480                         GOTO(out_req, rc);
1481                 }
1482
1483                 if (rc < 0) {
1484                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1485                                                cmd == LL_IOC_MDC_GETINFO))
1486                                 GOTO(skip_lmm, rc = 0);
1487                         else
1488                                 GOTO(out_req, rc);
1489                 }
1490
1491                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1492                     cmd == LL_IOC_LOV_GETSTRIPE) {
1493                         lump = (struct lov_user_md *)arg;
1494                 } else {
1495                         struct lov_user_mds_data *lmdp;
1496                         lmdp = (struct lov_user_mds_data *)arg;
1497                         lump = &lmdp->lmd_lmm;
1498                 }
1499                 if (copy_to_user(lump, lmm, lmmsize)) {
1500                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1501                                 GOTO(out_req, rc = -EFAULT);
1502                         rc = -EOVERFLOW;
1503                 }
1504         skip_lmm:
1505                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1506                         struct lov_user_mds_data *lmdp;
1507                         lstat_t st = { 0 };
1508
1509                         st.st_dev     = inode->i_sb->s_dev;
1510                         st.st_mode    = body->mode;
1511                         st.st_nlink   = body->nlink;
1512                         st.st_uid     = body->uid;
1513                         st.st_gid     = body->gid;
1514                         st.st_rdev    = body->rdev;
1515                         st.st_size    = body->size;
1516                         st.st_blksize = PAGE_CACHE_SIZE;
1517                         st.st_blocks  = body->blocks;
1518                         st.st_atime   = body->atime;
1519                         st.st_mtime   = body->mtime;
1520                         st.st_ctime   = body->ctime;
1521                         st.st_ino     = inode->i_ino;
1522
1523                         lmdp = (struct lov_user_mds_data *)arg;
1524                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1525                                 GOTO(out_req, rc = -EFAULT);
1526                 }
1527
1528         out_req:
1529                 ptlrpc_req_finished(request);
1530                 if (filename)
1531                         ll_putname(filename);
1532                 return rc;
1533         }
1534         case IOC_LOV_GETINFO: {
1535                 struct lov_user_mds_data *lumd;
1536                 struct lov_stripe_md *lsm;
1537                 struct lov_user_md *lum;
1538                 struct lov_mds_md *lmm;
1539                 int lmmsize;
1540                 lstat_t st;
1541
1542                 lumd = (struct lov_user_mds_data *)arg;
1543                 lum = &lumd->lmd_lmm;
1544
1545                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1546                 if (rc)
1547                         return rc;
1548
1549                 OBD_ALLOC_LARGE(lmm, lmmsize);
1550                 if (lmm == NULL)
1551                         return -ENOMEM;
1552                 if (copy_from_user(lmm, lum, lmmsize))
1553                         GOTO(free_lmm, rc = -EFAULT);
1554
1555                 switch (lmm->lmm_magic) {
1556                 case LOV_USER_MAGIC_V1:
1557                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1558                                 break;
1559                         /* swab objects first so that stripes num will be sane */
1560                         lustre_swab_lov_user_md_objects(
1561                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1562                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1563                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1564                         break;
1565                 case LOV_USER_MAGIC_V3:
1566                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1567                                 break;
1568                         /* swab objects first so that stripes num will be sane */
1569                         lustre_swab_lov_user_md_objects(
1570                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1571                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1572                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1573                         break;
1574                 default:
1575                         GOTO(free_lmm, rc = -EINVAL);
1576                 }
1577
1578                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1579                 if (rc < 0)
1580                         GOTO(free_lmm, rc = -ENOMEM);
1581
1582                 /* Perform glimpse_size operation. */
1583                 memset(&st, 0, sizeof(st));
1584
1585                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1586                 if (rc)
1587                         GOTO(free_lsm, rc);
1588
1589                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1590                         GOTO(free_lsm, rc = -EFAULT);
1591
1592         free_lsm:
1593                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1594         free_lmm:
1595                 OBD_FREE_LARGE(lmm, lmmsize);
1596                 return rc;
1597         }
1598         case OBD_IOC_LLOG_CATINFO: {
1599                 return -EOPNOTSUPP;
1600         }
1601         case OBD_IOC_QUOTACHECK: {
1602                 struct obd_quotactl *oqctl;
1603                 int error = 0;
1604
1605                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1606                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1607                         return -EPERM;
1608
1609                 OBD_ALLOC_PTR(oqctl);
1610                 if (!oqctl)
1611                         return -ENOMEM;
1612                 oqctl->qc_type = arg;
1613                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1614                 if (rc < 0) {
1615                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1616                         error = rc;
1617                 }
1618
1619                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1620                 if (rc < 0)
1621                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1622
1623                 OBD_FREE_PTR(oqctl);
1624                 return error ?: rc;
1625         }
1626         case OBD_IOC_POLL_QUOTACHECK: {
1627                 struct if_quotacheck *check;
1628
1629                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1630                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1631                         return -EPERM;
1632
1633                 OBD_ALLOC_PTR(check);
1634                 if (!check)
1635                         return -ENOMEM;
1636
1637                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1638                                    NULL);
1639                 if (rc) {
1640                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1641                         if (copy_to_user((void *)arg, check,
1642                                              sizeof(*check)))
1643                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1644                         GOTO(out_poll, rc);
1645                 }
1646
1647                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1648                                    NULL);
1649                 if (rc) {
1650                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1651                         if (copy_to_user((void *)arg, check,
1652                                              sizeof(*check)))
1653                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1654                         GOTO(out_poll, rc);
1655                 }
1656         out_poll:
1657                 OBD_FREE_PTR(check);
1658                 return rc;
1659         }
1660 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
1661         case LL_IOC_QUOTACTL_18: {
1662                 /* copy the old 1.x quota struct for internal use, then copy
1663                  * back into old format struct.  For 1.8 compatibility. */
1664                 struct if_quotactl_18 *qctl_18;
1665                 struct if_quotactl *qctl_20;
1666
1667                 OBD_ALLOC_PTR(qctl_18);
1668                 if (!qctl_18)
1669                         return -ENOMEM;
1670
1671                 OBD_ALLOC_PTR(qctl_20);
1672                 if (!qctl_20)
1673                         GOTO(out_quotactl_18, rc = -ENOMEM);
1674
1675                 if (copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1676                         GOTO(out_quotactl_20, rc = -ENOMEM);
1677
1678                 QCTL_COPY(qctl_20, qctl_18);
1679                 qctl_20->qc_idx = 0;
1680
1681                 /* XXX: dqb_valid was borrowed as a flag to mark that
1682                  *      only mds quota is wanted */
1683                 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1684                     qctl_18->qc_dqblk.dqb_valid) {
1685                         qctl_20->qc_valid = QC_MDTIDX;
1686                         qctl_20->qc_dqblk.dqb_valid = 0;
1687                 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1688                         qctl_20->qc_valid = QC_UUID;
1689                         qctl_20->obd_uuid = qctl_18->obd_uuid;
1690                 } else {
1691                         qctl_20->qc_valid = QC_GENERAL;
1692                 }
1693
1694                 rc = quotactl_ioctl(sbi, qctl_20);
1695
1696                 if (rc == 0) {
1697                         QCTL_COPY(qctl_18, qctl_20);
1698                         qctl_18->obd_uuid = qctl_20->obd_uuid;
1699
1700                         if (copy_to_user((void *)arg, qctl_18,
1701                                              sizeof(*qctl_18)))
1702                                 rc = -EFAULT;
1703                 }
1704
1705         out_quotactl_20:
1706                 OBD_FREE_PTR(qctl_20);
1707         out_quotactl_18:
1708                 OBD_FREE_PTR(qctl_18);
1709                 return rc;
1710         }
1711 #else
1712 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1713 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
1714         case LL_IOC_QUOTACTL: {
1715                 struct if_quotactl *qctl;
1716
1717                 OBD_ALLOC_PTR(qctl);
1718                 if (!qctl)
1719                         return -ENOMEM;
1720
1721                 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1722                         GOTO(out_quotactl, rc = -EFAULT);
1723
1724                 rc = quotactl_ioctl(sbi, qctl);
1725
1726                 if (rc == 0 && copy_to_user((void *)arg,qctl,sizeof(*qctl)))
1727                         rc = -EFAULT;
1728
1729         out_quotactl:
1730                 OBD_FREE_PTR(qctl);
1731                 return rc;
1732         }
1733         case OBD_IOC_GETDTNAME:
1734         case OBD_IOC_GETMDNAME:
1735                 return ll_get_obd_name(inode, cmd, arg);
1736         case LL_IOC_FLUSHCTX:
1737                 return ll_flush_ctx(inode);
1738 #ifdef CONFIG_FS_POSIX_ACL
1739         case LL_IOC_RMTACL: {
1740             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1741                 inode == inode->i_sb->s_root->d_inode) {
1742                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1743
1744                 LASSERT(fd != NULL);
1745                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1746                 if (!rc)
1747                         fd->fd_flags |= LL_FILE_RMTACL;
1748                 return rc;
1749             } else
1750                 return 0;
1751         }
1752 #endif
1753         case LL_IOC_GETOBDCOUNT: {
1754                 int count, vallen;
1755                 struct obd_export *exp;
1756
1757                 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1758                         return -EFAULT;
1759
1760                 /* get ost count when count is zero, get mdt count otherwise */
1761                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1762                 vallen = sizeof(count);
1763                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1764                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1765                 if (rc) {
1766                         CERROR("get target count failed: %d\n", rc);
1767                         return rc;
1768                 }
1769
1770                 if (copy_to_user((int *)arg, &count, sizeof(int)))
1771                         return -EFAULT;
1772
1773                 return 0;
1774         }
1775         case LL_IOC_PATH2FID:
1776                 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1777                                      sizeof(struct lu_fid)))
1778                         return -EFAULT;
1779                 return 0;
1780         case LL_IOC_GET_CONNECT_FLAGS: {
1781                 return obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg);
1782         }
1783         case OBD_IOC_CHANGELOG_SEND:
1784         case OBD_IOC_CHANGELOG_CLEAR:
1785                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1786                                     sizeof(struct ioc_changelog));
1787                 return rc;
1788         case OBD_IOC_FID2PATH:
1789                 return ll_fid2path(inode, (void *)arg);
1790         case LL_IOC_HSM_REQUEST: {
1791                 struct hsm_user_request *hur;
1792                 int                      totalsize;
1793
1794                 OBD_ALLOC_PTR(hur);
1795                 if (hur == NULL)
1796                         return -ENOMEM;
1797
1798                 /* We don't know the true size yet; copy the fixed-size part */
1799                 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1800                         OBD_FREE_PTR(hur);
1801                         return -EFAULT;
1802                 }
1803
1804                 /* Compute the whole struct size */
1805                 totalsize = hur_len(hur);
1806                 OBD_FREE_PTR(hur);
1807                 OBD_ALLOC_LARGE(hur, totalsize);
1808                 if (hur == NULL)
1809                         return -ENOMEM;
1810
1811                 /* Copy the whole struct */
1812                 if (copy_from_user(hur, (void *)arg, totalsize)) {
1813                         OBD_FREE_LARGE(hur, totalsize);
1814                         return -EFAULT;
1815                 }
1816
1817                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1818                         const struct lu_fid *fid;
1819                         struct inode *f;
1820                         int i;
1821
1822                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1823                                 fid = &hur->hur_user_item[i].hui_fid;
1824                                 f = search_inode_for_lustre(inode->i_sb, fid);
1825                                 if (IS_ERR(f)) {
1826                                         rc = PTR_ERR(f);
1827                                         break;
1828                                 }
1829
1830                                 rc = ll_hsm_release(f);
1831                                 iput(f);
1832                                 if (rc != 0)
1833                                         break;
1834                         }
1835                 } else {
1836                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1837                                            hur, NULL);
1838                 }
1839
1840                 OBD_FREE_LARGE(hur, totalsize);
1841
1842                 return rc;
1843         }
1844         case LL_IOC_HSM_PROGRESS: {
1845                 struct hsm_progress_kernel      hpk;
1846                 struct hsm_progress             hp;
1847
1848                 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1849                         return -EFAULT;
1850
1851                 hpk.hpk_fid = hp.hp_fid;
1852                 hpk.hpk_cookie = hp.hp_cookie;
1853                 hpk.hpk_extent = hp.hp_extent;
1854                 hpk.hpk_flags = hp.hp_flags;
1855                 hpk.hpk_errval = hp.hp_errval;
1856                 hpk.hpk_data_version = 0;
1857
1858                 /* File may not exist in Lustre; all progress
1859                  * reported to Lustre root */
1860                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1861                                    NULL);
1862                 return rc;
1863         }
1864         case LL_IOC_HSM_CT_START:
1865                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1866                                     sizeof(struct lustre_kernelcomm));
1867                 return rc;
1868
1869         case LL_IOC_HSM_COPY_START: {
1870                 struct hsm_copy *copy;
1871                 int              rc;
1872
1873                 OBD_ALLOC_PTR(copy);
1874                 if (copy == NULL)
1875                         return -ENOMEM;
1876                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1877                         OBD_FREE_PTR(copy);
1878                         return -EFAULT;
1879                 }
1880
1881                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1882                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1883                         rc = -EFAULT;
1884
1885                 OBD_FREE_PTR(copy);
1886                 return rc;
1887         }
1888         case LL_IOC_HSM_COPY_END: {
1889                 struct hsm_copy *copy;
1890                 int              rc;
1891
1892                 OBD_ALLOC_PTR(copy);
1893                 if (copy == NULL)
1894                         return -ENOMEM;
1895                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1896                         OBD_FREE_PTR(copy);
1897                         return -EFAULT;
1898                 }
1899
1900                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1901                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1902                         rc = -EFAULT;
1903
1904                 OBD_FREE_PTR(copy);
1905                 return rc;
1906         }
1907         default:
1908                 return obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL, (void *)arg);
1909         }
1910 }
1911
1912 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1913 {
1914         struct inode *inode = file->f_mapping->host;
1915         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1916         struct ll_sb_info *sbi = ll_i2sbi(inode);
1917         int api32 = ll_need_32bit_api(sbi);
1918         loff_t ret = -EINVAL;
1919
1920         mutex_lock(&inode->i_mutex);
1921         switch (origin) {
1922                 case SEEK_SET:
1923                         break;
1924                 case SEEK_CUR:
1925                         offset += file->f_pos;
1926                         break;
1927                 case SEEK_END:
1928                         if (offset > 0)
1929                                 GOTO(out, ret);
1930                         if (api32)
1931                                 offset += LL_DIR_END_OFF_32BIT;
1932                         else
1933                                 offset += LL_DIR_END_OFF;
1934                         break;
1935                 default:
1936                         GOTO(out, ret);
1937         }
1938
1939         if (offset >= 0 &&
1940             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1941              (!api32 && offset <= LL_DIR_END_OFF))) {
1942                 if (offset != file->f_pos) {
1943                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1944                             (!api32 && offset == LL_DIR_END_OFF))
1945                                 fd->lfd_pos = MDS_DIR_END_OFF;
1946                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1947                                 fd->lfd_pos = offset << 32;
1948                         else
1949                                 fd->lfd_pos = offset;
1950                         file->f_pos = offset;
1951                         file->f_version = 0;
1952                 }
1953                 ret = offset;
1954         }
1955         GOTO(out, ret);
1956
1957 out:
1958         mutex_unlock(&inode->i_mutex);
1959         return ret;
1960 }
1961
1962 int ll_dir_open(struct inode *inode, struct file *file)
1963 {
1964         return ll_file_open(inode, file);
1965 }
1966
1967 int ll_dir_release(struct inode *inode, struct file *file)
1968 {
1969         return ll_file_release(inode, file);
1970 }
1971
1972 struct file_operations ll_dir_operations = {
1973         .llseek   = ll_dir_seek,
1974         .open     = ll_dir_open,
1975         .release  = ll_dir_release,
1976         .read     = generic_read_dir,
1977         .iterate  = ll_readdir,
1978         .unlocked_ioctl   = ll_dir_ioctl,
1979         .fsync    = ll_fsync,
1980 };