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

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/dir.c
  37  *
  38  * Directory code for lustre client.
  39  */
  40
  41 #include <linux/fs.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/mm.h>
  44 #include <linux/version.h>
  45 #include <asm/uaccess.h>
  46 #include <linux/buffer_head.h>   // for wait_on_buffer
  47 #include <linux/pagevec.h>
  48 #include <linux/prefetch.h>
  49
  50 #define DEBUG_SUBSYSTEM S_LLITE
  51
  52 #include <lustre/lustre_idl.h>
  53 #include <obd_support.h>
  54 #include <obd_class.h>
  55 #include <lustre_lib.h>
  56 #include <lustre/lustre_idl.h>
  57 #include <lustre_lite.h>
  58 #include <lustre_dlm.h>
  59 #include <lustre_fid.h>
  60 #include "llite_internal.h"
  61
  62 /*
  63  * (new) readdir implementation overview.
  64  *
  65  * Original lustre readdir implementation cached exact copy of raw directory
  66  * pages on the client. These pages were indexed in client page cache by
  67  * logical offset in the directory file. This design, while very simple and
  68  * intuitive had some inherent problems:
  69  *
  70  *     . it implies that byte offset to the directory entry serves as a
  71  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  72  *     ext3/htree directory entries may move due to splits, and more
  73  *     importantly,
  74  *
  75  *     . it is incompatible with the design of split directories for cmd3,
  76  *     that assumes that names are distributed across nodes based on their
  77  *     hash, and so readdir should be done in hash order.
  78  *
  79  * New readdir implementation does readdir in hash order, and uses hash of a
  80  * file name as a telldir/seekdir cookie. This led to number of complications:
  81  *
  82  *     . hash is not unique, so it cannot be used to index cached directory
  83  *     pages on the client (note, that it requires a whole pageful of hash
  84  *     collided entries to cause two pages to have identical hashes);
  85  *
  86  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  87  *     entry cookie. ext3/htree has the same problem and lustre implementation
  88  *     mimics their solution: seekdir(hash) positions directory at the first
  89  *     entry with the given hash.
  90  *
  91  * Client side.
  92  *
  93  * 0. caching
  94  *
  95  * Client caches directory pages using hash of the first entry as an index. As
  96  * noted above hash is not unique, so this solution doesn't work as is:
  97  * special processing is needed for "page hash chains" (i.e., sequences of
  98  * pages filled with entries all having the same hash value).
  99  *
 100  * First, such chains have to be detected. To this end, server returns to the
 101  * client the hash of the first entry on the page next to one returned. When
 102  * client detects that this hash is the same as hash of the first entry on the
 103  * returned page, page hash collision has to be handled. Pages in the
 104  * hash chain, except first one, are termed "overflow pages".
 105  *
 106  * Solution to index uniqueness problem is to not cache overflow
 107  * pages. Instead, when page hash collision is detected, all overflow pages
 108  * from emerging chain are immediately requested from the server and placed in
 109  * a special data structure (struct ll_dir_chain). This data structure is used
 110  * by ll_readdir() to process entries from overflow pages. When readdir
 111  * invocation finishes, overflow pages are discarded. If page hash collision
 112  * chain weren't completely processed, next call to readdir will again detect
 113  * page hash collision, again read overflow pages in, process next portion of
 114  * entries and again discard the pages. This is not as wasteful as it looks,
 115  * because, given reasonable hash, page hash collisions are extremely rare.
 116  *
 117  * 1. directory positioning
 118  *
 119  * When seekdir(hash) is called, original
 120  *
 121  *
 122  *
 123  *
 124  *
 125  *
 126  *
 127  *
 128  * Server.
 129  *
 130  * identification of and access to overflow pages
 131  *
 132  * page format
 133  *
 134  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 135  * a header lu_dirpage which describes the start/end hash, and whether this
 136  * page is empty (contains no dir entry) or hash collide with next page.
 137  * After client receives reply, several pages will be integrated into dir page
 138  * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
 139  * lu_dirpage for this integrated page will be adjusted. See
 140  * lmv_adjust_dirpages().
 141  *
 142  */
 143
 144 /* returns the page unlocked, but with a reference */
 145 static int ll_dir_filler(void *_hash, struct page *page0)
 146 {
 147         struct inode *inode = page0->mapping->host;
 148         int hash64 = ll_i2sbi(inode)->ll_flags & LL_SBI_64BIT_HASH;
 149         struct obd_export *exp = ll_i2sbi(inode)->ll_md_exp;
 150         struct ptlrpc_request *request;
 151         struct mdt_body *body;
 152         struct md_op_data *op_data;
 153         __u64 hash = *((__u64 *)_hash);
 154         struct page **page_pool;
 155         struct page *page;
 156         struct lu_dirpage *dp;
 157         int max_pages = ll_i2sbi(inode)->ll_md_brw_size >> PAGE_CACHE_SHIFT;
 158         int nrdpgs = 0; /* number of pages read actually */
 159         int npages;
 160         int i;
 161         int rc;
 162         ENTRY;
 163
 164         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) hash "LPU64"\n",
 165                inode->i_ino, inode->i_generation, inode, hash);
 166
 167         LASSERT(max_pages > 0 && max_pages <= MD_MAX_BRW_PAGES);
 168
 169         OBD_ALLOC(page_pool, sizeof(page) * max_pages);
 170         if (page_pool != NULL) {
 171                 page_pool[0] = page0;
 172         } else {
 173                 page_pool = &page0;
 174                 max_pages = 1;
 175         }
 176         for (npages = 1; npages < max_pages; npages++) {
 177                 page = page_cache_alloc_cold(inode->i_mapping);
 178                 if (!page)
 179                         break;
 180                 page_pool[npages] = page;
 181         }
 182
 183         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 184                                      LUSTRE_OPC_ANY, NULL);
 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                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 191                 /* Checked by mdc_readpage() */
 192                 LASSERT(body != NULL);
 193
 194                 if (body->valid & OBD_MD_FLSIZE)
 195                         cl_isize_write(inode, body->size);
 196
 197                 nrdpgs = (request->rq_bulk->bd_nob_transferred+PAGE_CACHE_SIZE-1)
 198                          >> PAGE_CACHE_SHIFT;
 199                 SetPageUptodate(page0);
 200         }
 201         unlock_page(page0);
 202         ptlrpc_req_finished(request);
 203
 204         CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages);
 205
 206         ll_pagevec_init(&lru_pvec, 0);
 207         for (i = 1; i < npages; i++) {
 208                 unsigned long offset;
 209                 int ret;
 210
 211                 page = page_pool[i];
 212
 213                 if (rc < 0 || i >= nrdpgs) {
 214                         page_cache_release(page);
 215                         continue;
 216                 }
 217
 218                 SetPageUptodate(page);
 219
 220                 dp = kmap(page);
 221                 hash = le64_to_cpu(dp->ldp_hash_start);
 222                 kunmap(page);
 223
 224                 offset = hash_x_index(hash, hash64);
 225
 226                 prefetchw(&page->flags);
 227                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
 228                                             GFP_KERNEL);
 229                 if (ret == 0) {
 230                         unlock_page(page);
 231                         if (ll_pagevec_add(&lru_pvec, page) == 0)
 232                                 ll_pagevec_lru_add_file(&lru_pvec);
 233                 } else {
 234                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
 235                                " %d\n", offset, ret);
 236                 }
 237                 page_cache_release(page);
 238         }
 239         ll_pagevec_lru_add_file(&lru_pvec);
 240
 241         if (page_pool != &page0)
 242                 OBD_FREE(page_pool, sizeof(struct page *) * max_pages);
 243         EXIT;
 244         return rc;
 245 }
 246
 247 static void ll_check_page(struct inode *dir, struct page *page)
 248 {
 249         /* XXX: check page format later */
 250         SetPageChecked(page);
 251 }
 252
 253 void ll_release_page(struct page *page, int remove)
 254 {
 255         kunmap(page);
 256         if (remove) {
 257                 lock_page(page);
 258                 if (likely(page->mapping != NULL))
 259                         truncate_complete_page(page->mapping, page);
 260                 unlock_page(page);
 261         }
 262         page_cache_release(page);
 263 }
 264
 265 /*
 266  * Find, kmap and return page that contains given hash.
 267  */
 268 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
 269                                        __u64 *start, __u64 *end)
 270 {
 271         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 272         struct address_space *mapping = dir->i_mapping;
 273         /*
 274          * Complement of hash is used as an index so that
 275          * radix_tree_gang_lookup() can be used to find a page with starting
 276          * hash _smaller_ than one we are looking for.
 277          */
 278         unsigned long offset = hash_x_index(*hash, hash64);
 279         struct page *page;
 280         int found;
 281
 282         TREE_READ_LOCK_IRQ(mapping);
 283         found = radix_tree_gang_lookup(&mapping->page_tree,
 284                                        (void **)&page, offset, 1);
 285         if (found > 0) {
 286                 struct lu_dirpage *dp;
 287
 288                 page_cache_get(page);
 289                 TREE_READ_UNLOCK_IRQ(mapping);
 290                 /*
 291                  * In contrast to find_lock_page() we are sure that directory
 292                  * page cannot be truncated (while DLM lock is held) and,
 293                  * hence, can avoid restart.
 294                  *
 295                  * In fact, page cannot be locked here at all, because
 296                  * ll_dir_filler() does synchronous io.
 297                  */
 298                 wait_on_page_locked(page);
 299                 if (PageUptodate(page)) {
 300                         dp = kmap(page);
 301                         if (BITS_PER_LONG == 32 && hash64) {
 302                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 303                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 304                                 *hash  = *hash >> 32;
 305                         } else {
 306                                 *start = le64_to_cpu(dp->ldp_hash_start);
 307                                 *end   = le64_to_cpu(dp->ldp_hash_end);
 308                         }
 309                         LASSERTF(*start <= *hash, "start = "LPX64",end = "
 310                                  LPX64",hash = "LPX64"\n", *start, *end, *hash);
 311                         CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash "LPU64"\n",
 312                                offset, *start, *end, *hash);
 313                         if (*hash > *end) {
 314                                 ll_release_page(page, 0);
 315                                 page = NULL;
 316                         } else if (*end != *start && *hash == *end) {
 317                                 /*
 318                                  * upon hash collision, remove this page,
 319                                  * otherwise put page reference, and
 320                                  * ll_get_dir_page() will issue RPC to fetch
 321                                  * the page we want.
 322                                  */
 323                                 ll_release_page(page,
 324                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
 325                                 page = NULL;
 326                         }
 327                 } else {
 328                         page_cache_release(page);
 329                         page = ERR_PTR(-EIO);
 330                 }
 331
 332         } else {
 333                 TREE_READ_UNLOCK_IRQ(mapping);
 334                 page = NULL;
 335         }
 336         return page;
 337 }
 338
 339 struct page *ll_get_dir_page(struct inode *dir, __u64 hash,
 340                              struct ll_dir_chain *chain)
 341 {
 342         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 343         struct address_space *mapping = dir->i_mapping;
 344         struct lustre_handle lockh;
 345         struct lu_dirpage *dp;
 346         struct page *page;
 347         ldlm_mode_t mode;
 348         int rc;
 349         __u64 start = 0;
 350         __u64 end = 0;
 351         __u64 lhash = hash;
 352         struct ll_inode_info *lli = ll_i2info(dir);
 353         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 354
 355         mode = LCK_PR;
 356         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 357                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
 358         if (!rc) {
 359                 struct ldlm_enqueue_info einfo = {.ei_type = LDLM_IBITS,
 360                                                   .ei_mode = mode,
 361                                                   .ei_cb_bl =
 362                                                   ll_md_blocking_ast,
 363                                                   .ei_cb_cp =
 364                                                   ldlm_completion_ast,
 365                                                   .ei_cb_gl = NULL,
 366                                                   .ei_cb_wg = NULL,
 367                                                   .ei_cbdata = NULL};
 368                 struct lookup_intent it = { .it_op = IT_READDIR };
 369                 struct ptlrpc_request *request;
 370                 struct md_op_data *op_data;
 371
 372                 op_data = ll_prep_md_op_data(NULL, dir, NULL, NULL, 0, 0,
 373                 LUSTRE_OPC_ANY, NULL);
 374                 if (IS_ERR(op_data))
 375                         return (void *)op_data;
 376
 377                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
 378                                 op_data, &lockh, NULL, 0, NULL, 0);
 379
 380                 ll_finish_md_op_data(op_data);
 381
 382                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 383                 if (request)
 384                         ptlrpc_req_finished(request);
 385                 if (rc < 0) {
 386                         CERROR("lock enqueue: "DFID" at "LPU64": rc %d\n",
 387                                 PFID(ll_inode2fid(dir)), hash, rc);
 388                         return ERR_PTR(rc);
 389                 }
 390
 391                 CDEBUG(D_INODE, "setting lr_lvb_inode to inode %p (%lu/%u)\n",
 392                        dir, dir->i_ino, dir->i_generation);
 393                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp,
 394                                  &it.d.lustre.it_lock_handle, dir, NULL);
 395         } else {
 396                 /* for cross-ref object, l_ast_data of the lock may not be set,
 397                  * we reset it here */
 398                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
 399                                  dir, NULL);
 400         }
 401         ldlm_lock_dump_handle(D_OTHER, &lockh);
 402
 403         mutex_lock(&lli->lli_readdir_mutex);
 404         page = ll_dir_page_locate(dir, &lhash, &start, &end);
 405         if (IS_ERR(page)) {
 406                 CERROR("dir page locate: "DFID" at "LPU64": rc %ld\n",
 407                        PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
 408                 GOTO(out_unlock, page);
 409         } else if (page != NULL) {
 410                 /*
 411                  * XXX nikita: not entirely correct handling of a corner case:
 412                  * suppose hash chain of entries with hash value HASH crosses
 413                  * border between pages P0 and P1. First both P0 and P1 are
 414                  * cached, seekdir() is called for some entry from the P0 part
 415                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 416                  * happens and finds P1, as it starts with matching hash
 417                  * value. Remaining entries from P0 part of the chain are
 418                  * skipped. (Is that really a bug?)
 419                  *
 420                  * Possible solutions: 0. don't cache P1 is such case, handle
 421                  * it as an "overflow" page. 1. invalidate all pages at
 422                  * once. 2. use HASH|1 as an index for P1.
 423                  */
 424                 GOTO(hash_collision, page);
 425         }
 426
 427         page = read_cache_page(mapping, hash_x_index(hash, hash64),
 428                                ll_dir_filler, &lhash);
 429         if (IS_ERR(page)) {
 430                 CERROR("read cache page: "DFID" at "LPU64": rc %ld\n",
 431                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 432                 GOTO(out_unlock, page);
 433         }
 434
 435         wait_on_page_locked(page);
 436         (void)kmap(page);
 437         if (!PageUptodate(page)) {
 438                 CERROR("page not updated: "DFID" at "LPU64": rc %d\n",
 439                        PFID(ll_inode2fid(dir)), hash, -5);
 440                 goto fail;
 441         }
 442         if (!PageChecked(page))
 443                 ll_check_page(dir, page);
 444         if (PageError(page)) {
 445                 CERROR("page error: "DFID" at "LPU64": rc %d\n",
 446                        PFID(ll_inode2fid(dir)), hash, -5);
 447                 goto fail;
 448         }
 449 hash_collision:
 450         dp = page_address(page);
 451         if (BITS_PER_LONG == 32 && hash64) {
 452                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 453                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 454                 lhash = hash >> 32;
 455         } else {
 456                 start = le64_to_cpu(dp->ldp_hash_start);
 457                 end   = le64_to_cpu(dp->ldp_hash_end);
 458                 lhash = hash;
 459         }
 460         if (end == start) {
 461                 LASSERT(start == lhash);
 462                 CWARN("Page-wide hash collision: "LPU64"\n", end);
 463                 if (BITS_PER_LONG == 32 && hash64)
 464                         CWARN("Real page-wide hash collision at ["LPU64" "LPU64
 465                               "] with hash "LPU64"\n",
 466                               le64_to_cpu(dp->ldp_hash_start),
 467                               le64_to_cpu(dp->ldp_hash_end), hash);
 468                 /*
 469                  * Fetch whole overflow chain...
 470                  *
 471                  * XXX not yet.
 472                  */
 473                 goto fail;
 474         }
 475 out_unlock:
 476         mutex_unlock(&lli->lli_readdir_mutex);
 477         ldlm_lock_decref(&lockh, mode);
 478         return page;
 479
 480 fail:
 481         ll_release_page(page, 1);
 482         page = ERR_PTR(-EIO);
 483         goto out_unlock;
 484 }
 485
 486 int ll_dir_read(struct inode *inode, __u64 *_pos, void *cookie,
 487                 filldir_t filldir)
 488 {
 489         struct ll_inode_info *info       = ll_i2info(inode);
 490         struct ll_sb_info    *sbi       = ll_i2sbi(inode);
 491         __u64            pos    = *_pos;
 492         int                api32      = ll_need_32bit_api(sbi);
 493         int                hash64     = sbi->ll_flags & LL_SBI_64BIT_HASH;
 494         struct page       *page;
 495         struct ll_dir_chain   chain;
 496         int                done = 0;
 497         int                rc = 0;
 498         ENTRY;
 499
 500         ll_dir_chain_init(&chain);
 501
 502         page = ll_get_dir_page(inode, pos, &chain);
 503
 504         while (rc == 0 && !done) {
 505                 struct lu_dirpage *dp;
 506                 struct lu_dirent  *ent;
 507
 508                 if (!IS_ERR(page)) {
 509                         /*
 510                          * If page is empty (end of directory is reached),
 511                          * use this value.
 512                          */
 513                         __u64 hash = MDS_DIR_END_OFF;
 514                         __u64 next;
 515
 516                         dp = page_address(page);
 517                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 518                              ent = lu_dirent_next(ent)) {
 519                                 __u16     type;
 520                                 int         namelen;
 521                                 struct lu_fid  fid;
 522                                 __u64     lhash;
 523                                 __u64     ino;
 524
 525                                 /*
 526                                  * XXX: implement correct swabbing here.
 527                                  */
 528
 529                                 hash = le64_to_cpu(ent->lde_hash);
 530                                 if (hash < pos)
 531                                         /*
 532                                          * Skip until we find target hash
 533                                          * value.
 534                                          */
 535                                         continue;
 536
 537                                 namelen = le16_to_cpu(ent->lde_namelen);
 538                                 if (namelen == 0)
 539                                         /*
 540                                          * Skip dummy record.
 541                                          */
 542                                         continue;
 543
 544                                 if (api32 && hash64)
 545                                         lhash = hash >> 32;
 546                                 else
 547                                         lhash = hash;
 548                                 fid_le_to_cpu(&fid, &ent->lde_fid);
 549                                 ino = cl_fid_build_ino(&fid, api32);
 550                                 type = ll_dirent_type_get(ent);
 551                                 /* For 'll_nfs_get_name_filldir()', it will try
 552                                  * to access the 'ent' through its 'lde_name',
 553                                  * so the parameter 'name' for 'filldir()' must
 554                                  * be part of the 'ent'. */
 555                                 done = filldir(cookie, ent->lde_name, namelen,
 556                                                lhash, ino, type);
 557                         }
 558                         next = le64_to_cpu(dp->ldp_hash_end);
 559                         if (!done) {
 560                                 pos = next;
 561                                 if (pos == MDS_DIR_END_OFF) {
 562                                         /*
 563                                          * End of directory reached.
 564                                          */
 565                                         done = 1;
 566                                         ll_release_page(page, 0);
 567                                 } else if (1 /* chain is exhausted*/) {
 568                                         /*
 569                                          * Normal case: continue to the next
 570                                          * page.
 571                                          */
 572                                         ll_release_page(page,
 573                                             le32_to_cpu(dp->ldp_flags) &
 574                                                         LDF_COLLIDE);
 575                                         next = pos;
 576                                         page = ll_get_dir_page(inode, pos,
 577                                                                &chain);
 578                                 } else {
 579                                         /*
 580                                          * go into overflow page.
 581                                          */
 582                                         LASSERT(le32_to_cpu(dp->ldp_flags) &
 583                                                 LDF_COLLIDE);
 584                                         ll_release_page(page, 1);
 585                                 }
 586                         } else {
 587                                 pos = hash;
 588                                 ll_release_page(page, 0);
 589                         }
 590                 } else {
 591                         rc = PTR_ERR(page);
 592                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 593                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 594                 }
 595         }
 596
 597         *_pos = pos;
 598         ll_dir_chain_fini(&chain);
 599         RETURN(rc);
 600 }
 601
 602 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 603 {
 604         struct inode            *inode  = filp->f_dentry->d_inode;
 605         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 606         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 607         __u64                   pos     = lfd->lfd_pos;
 608         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 609         int                     api32   = ll_need_32bit_api(sbi);
 610         int                     rc;
 611         struct path             path;
 612         ENTRY;
 613
 614         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu "
 615                " 32bit_api %d\n", inode->i_ino, inode->i_generation,
 616                inode, (unsigned long)pos, i_size_read(inode), api32);
 617
 618         if (pos == MDS_DIR_END_OFF)
 619                 /*
 620                  * end-of-file.
 621                  */
 622                 GOTO(out, rc = 0);
 623
 624         rc = ll_dir_read(inode, &pos, cookie, filldir);
 625         lfd->lfd_pos = pos;
 626         if (pos == MDS_DIR_END_OFF) {
 627                 if (api32)
 628                         filp->f_pos = LL_DIR_END_OFF_32BIT;
 629                 else
 630                         filp->f_pos = LL_DIR_END_OFF;
 631         } else {
 632                 if (api32 && hash64)
 633                         filp->f_pos = pos >> 32;
 634                 else
 635                         filp->f_pos = pos;
 636         }
 637         filp->f_version = inode->i_version;
 638         path.mnt = filp->f_path.mnt;
 639         path.dentry = filp->f_dentry;
 640         touch_atime(&path);
 641
 642 out:
 643         if (!rc)
 644                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 645
 646         RETURN(rc);
 647 }
 648
 649 int ll_send_mgc_param(struct obd_export *mgc, char *string)
 650 {
 651         struct mgs_send_param *msp;
 652         int rc = 0;
 653
 654         OBD_ALLOC_PTR(msp);
 655         if (!msp)
 656                 return -ENOMEM;
 657
 658         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 659         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 660                                 sizeof(struct mgs_send_param), msp, NULL);
 661         if (rc)
 662                 CERROR("Failed to set parameter: %d\n", rc);
 663         OBD_FREE_PTR(msp);
 664
 665         return rc;
 666 }
 667
 668 int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
 669                         char *filename)
 670 {
 671         struct ptlrpc_request *request = NULL;
 672         struct md_op_data *op_data;
 673         struct ll_sb_info *sbi = ll_i2sbi(dir);
 674         int mode;
 675         int err;
 676
 677         ENTRY;
 678
 679         mode = (0755 & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
 680         op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
 681                                      strlen(filename), mode, LUSTRE_OPC_MKDIR,
 682                                      lump);
 683         if (IS_ERR(op_data))
 684                 GOTO(err_exit, err = PTR_ERR(op_data));
 685
 686         op_data->op_cli_flags |= CLI_SET_MEA;
 687         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 688                         from_kuid(&init_user_ns, current_fsuid()),
 689                         from_kgid(&init_user_ns, current_fsgid()),
 690                         cfs_curproc_cap_pack(), 0, &request);
 691         ll_finish_md_op_data(op_data);
 692         if (err)
 693                 GOTO(err_exit, err);
 694 err_exit:
 695         ptlrpc_req_finished(request);
 696         return err;
 697 }
 698
 699 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 700                      int set_default)
 701 {
 702         struct ll_sb_info *sbi = ll_i2sbi(inode);
 703         struct md_op_data *op_data;
 704         struct ptlrpc_request *req = NULL;
 705         int rc = 0;
 706         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 707         struct obd_device *mgc = lsi->lsi_mgc;
 708         int lum_size;
 709         ENTRY;
 710
 711         if (lump != NULL) {
 712                 /*
 713                  * This is coming from userspace, so should be in
 714                  * local endian.  But the MDS would like it in little
 715                  * endian, so we swab it before we send it.
 716                  */
 717                 switch (lump->lmm_magic) {
 718                 case LOV_USER_MAGIC_V1: {
 719                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 720                                 lustre_swab_lov_user_md_v1(lump);
 721                         lum_size = sizeof(struct lov_user_md_v1);
 722                         break;
 723                 }
 724                 case LOV_USER_MAGIC_V3: {
 725                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 726                                 lustre_swab_lov_user_md_v3(
 727                                         (struct lov_user_md_v3 *)lump);
 728                         lum_size = sizeof(struct lov_user_md_v3);
 729                         break;
 730                 }
 731                 default: {
 732                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 733                                         " %#08x != %#08x nor %#08x\n",
 734                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 735                                         LOV_USER_MAGIC_V3);
 736                         RETURN(-EINVAL);
 737                 }
 738                 }
 739         } else {
 740                 lum_size = sizeof(struct lov_user_md_v1);
 741         }
 742
 743         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 744                                      LUSTRE_OPC_ANY, NULL);
 745         if (IS_ERR(op_data))
 746                 RETURN(PTR_ERR(op_data));
 747
 748         if (lump != NULL && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC))
 749                 op_data->op_cli_flags |= CLI_SET_MEA;
 750
 751         /* swabbing is done in lov_setstripe() on server side */
 752         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 753                         NULL, 0, &req, NULL);
 754         ll_finish_md_op_data(op_data);
 755         ptlrpc_req_finished(req);
 756         if (rc) {
 757                 if (rc != -EPERM && rc != -EACCES)
 758                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 759         }
 760
 761         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 762          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 763          need the make the distiction between the 2 versions */
 764         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 765                 char *param = NULL;
 766                 char *buf;
 767
 768                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 769                 if (param == NULL)
 770                         GOTO(end, rc = -ENOMEM);
 771
 772                 buf = param;
 773                 /* Get fsname and assume devname to be -MDT0000. */
 774                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 775                 strcat(buf, "-MDT0000.lov");
 776                 buf += strlen(buf);
 777
 778                 /* Set root stripesize */
 779                 sprintf(buf, ".stripesize=%u",
 780                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 781                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 782                 if (rc)
 783                         GOTO(end, rc);
 784
 785                 /* Set root stripecount */
 786                 sprintf(buf, ".stripecount=%hd",
 787                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 788                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 789                 if (rc)
 790                         GOTO(end, rc);
 791
 792                 /* Set root stripeoffset */
 793                 sprintf(buf, ".stripeoffset=%hd",
 794                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 795                         (typeof(lump->lmm_stripe_offset))(-1));
 796                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 797
 798 end:
 799                 if (param != NULL)
 800                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 801         }
 802         RETURN(rc);
 803 }
 804
 805 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 806                      int *lmm_size, struct ptlrpc_request **request)
 807 {
 808         struct ll_sb_info *sbi = ll_i2sbi(inode);
 809         struct mdt_body   *body;
 810         struct lov_mds_md *lmm = NULL;
 811         struct ptlrpc_request *req = NULL;
 812         int rc, lmmsize;
 813         struct md_op_data *op_data;
 814
 815         rc = ll_get_max_mdsize(sbi, &lmmsize);
 816         if (rc)
 817                 RETURN(rc);
 818
 819         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 820                                      0, lmmsize, LUSTRE_OPC_ANY,
 821                                      NULL);
 822         if (IS_ERR(op_data))
 823                 RETURN(PTR_ERR(op_data));
 824
 825         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 826         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 827         ll_finish_md_op_data(op_data);
 828         if (rc < 0) {
 829                 CDEBUG(D_INFO, "md_getattr failed on inode "
 830                        "%lu/%u: rc %d\n", inode->i_ino,
 831                        inode->i_generation, rc);
 832                 GOTO(out, rc);
 833         }
 834
 835         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 836         LASSERT(body != NULL);
 837
 838         lmmsize = body->eadatasize;
 839
 840         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 841             lmmsize == 0) {
 842                 GOTO(out, rc = -ENODATA);
 843         }
 844
 845         lmm = req_capsule_server_sized_get(&req->rq_pill,
 846                                            &RMF_MDT_MD, lmmsize);
 847         LASSERT(lmm != NULL);
 848
 849         /*
 850          * This is coming from the MDS, so is probably in
 851          * little endian.  We convert it to host endian before
 852          * passing it to userspace.
 853          */
 854         /* We don't swab objects for directories */
 855         switch (le32_to_cpu(lmm->lmm_magic)) {
 856         case LOV_MAGIC_V1:
 857                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 858                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 859                 break;
 860         case LOV_MAGIC_V3:
 861                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 862                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 863                 break;
 864         default:
 865                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 866                 rc = -EPROTO;
 867         }
 868 out:
 869         *lmmp = lmm;
 870         *lmm_size = lmmsize;
 871         *request = req;
 872         return rc;
 873 }
 874
 875 /*
 876  *  Get MDT index for the inode.
 877  */
 878 int ll_get_mdt_idx(struct inode *inode)
 879 {
 880         struct ll_sb_info *sbi = ll_i2sbi(inode);
 881         struct md_op_data *op_data;
 882         int rc, mdtidx;
 883         ENTRY;
 884
 885         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 886                                      0, LUSTRE_OPC_ANY, NULL);
 887         if (IS_ERR(op_data))
 888                 RETURN(PTR_ERR(op_data));
 889
 890         op_data->op_flags |= MF_GET_MDT_IDX;
 891         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 892         mdtidx = op_data->op_mds;
 893         ll_finish_md_op_data(op_data);
 894         if (rc < 0) {
 895                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 896                 RETURN(rc);
 897         }
 898         return mdtidx;
 899 }
 900
 901 /**
 902  * Generic handler to do any pre-copy work.
 903  *
 904  * It send a first hsm_progress (with extent length == 0) to coordinator as a
 905  * first information for it that real work has started.
 906  *
 907  * Moreover, for a ARCHIVE request, it will sample the file data version and
 908  * store it in \a copy.
 909  *
 910  * \return 0 on success.
 911  */
 912 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 913 {
 914         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 915         struct hsm_progress_kernel       hpk;
 916         int                              rc;
 917         ENTRY;
 918
 919         /* Forge a hsm_progress based on data from copy. */
 920         hpk.hpk_fid = copy->hc_hai.hai_fid;
 921         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 922         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 923         hpk.hpk_extent.length = 0;
 924         hpk.hpk_flags = 0;
 925         hpk.hpk_errval = 0;
 926         hpk.hpk_data_version = 0;
 927
 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                         GOTO(progress, rc = PTR_ERR(inode));
 941                 }
 942
 943                 /* Read current file data version */
 944                 rc = ll_data_version(inode, &data_version, 1);
 945                 iput(inode);
 946                 if (rc != 0) {
 947                         CDEBUG(D_HSM, "Could not read file data version of "
 948                                       DFID" (rc = %d). Archive request ("
 949                                       LPX64") 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, rc);
 956                 }
 957
 958                 /* Store it the hsm_copy for later copytool use.
 959                  * Always modified even if no lsm. */
 960                 copy->hc_data_version = data_version;
 961         }
 962
 963 progress:
 964         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 965                            &hpk, NULL);
 966
 967         RETURN(rc);
 968 }
 969
 970 /**
 971  * Generic handler to do any post-copy work.
 972  *
 973  * It will send the last hsm_progress update to coordinator to inform it
 974  * that copy is finished and whether it was successful or not.
 975  *
 976  * Moreover,
 977  * - for ARCHIVE request, it will sample the file data version and compare it
 978  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 979  *   will be considered as failed.
 980  * - for RESTORE request, it will sample the file data version and send it to
 981  *   coordinator which is useful if the file was imported as 'released'.
 982  *
 983  * \return 0 on success.
 984  */
 985 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 986 {
 987         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 988         struct hsm_progress_kernel       hpk;
 989         int                              rc;
 990         ENTRY;
 991
 992         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 993         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 994          * initialized if copy_end was called with copy == NULL.
 995          */
 996
 997         /* Forge a hsm_progress based on data from copy. */
 998         hpk.hpk_fid = copy->hc_hai.hai_fid;
 999         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
1000         hpk.hpk_extent = copy->hc_hai.hai_extent;
1001         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
1002         hpk.hpk_errval = copy->hc_errval;
1003         hpk.hpk_data_version = 0;
1004
1005         /* For archive request, we need to check the file data was not changed.
1006          *
1007          * For restore request, we need to send the file data version, this is
1008          * useful when the file was created using hsm_import.
1009          */
1010         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
1011              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
1012             (copy->hc_errval == 0)) {
1013                 struct inode    *inode;
1014                 __u64            data_version = 0;
1015
1016                 /* Get lsm for this fid */
1017                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
1018                 if (IS_ERR(inode)) {
1019                         hpk.hpk_flags |= HP_FLAG_RETRY;
1020                         /* hpk_errval must be >= 0 */
1021                         hpk.hpk_errval = -PTR_ERR(inode);
1022                         GOTO(progress, rc = PTR_ERR(inode));
1023                 }
1024
1025                 rc = ll_data_version(inode, &data_version,
1026                                      copy->hc_hai.hai_action == HSMA_ARCHIVE);
1027                 iput(inode);
1028                 if (rc) {
1029                         CDEBUG(D_HSM, "Could not read file data version. "
1030                                       "Request could not be confirmed.\n");
1031                         if (hpk.hpk_errval == 0)
1032                                 hpk.hpk_errval = -rc;
1033                         GOTO(progress, rc);
1034                 }
1035
1036                 /* Store it the hsm_copy for later copytool use.
1037                  * Always modified even if no lsm. */
1038                 hpk.hpk_data_version = data_version;
1039
1040                 /* File could have been stripped during archiving, so we need
1041                  * to check anyway. */
1042                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1043                     (copy->hc_data_version != data_version)) {
1044                         CDEBUG(D_HSM, "File data version mismatched. "
1045                               "File content was changed during archiving. "
1046                                DFID", start:"LPX64" current:"LPX64"\n",
1047                                PFID(&copy->hc_hai.hai_fid),
1048                                copy->hc_data_version, data_version);
1049                         /* File was changed, send error to cdt. Do not ask for
1050                          * retry because if a file is modified frequently,
1051                          * the cdt will loop on retried archive requests.
1052                          * The policy engine will ask for a new archive later
1053                          * when the file will not be modified for some tunable
1054                          * time */
1055                         /* we do not notify caller */
1056                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
1057                         /* hpk_errval must be >= 0 */
1058                         hpk.hpk_errval = EBUSY;
1059                 }
1060
1061         }
1062
1063 progress:
1064         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1065                            &hpk, NULL);
1066
1067         RETURN(rc);
1068 }
1069
1070
1071 static int copy_and_ioctl(int cmd, struct obd_export *exp, void *data, int len)
1072 {
1073         void *ptr;
1074         int rc;
1075
1076         OBD_ALLOC(ptr, len);
1077         if (ptr == NULL)
1078                 return -ENOMEM;
1079         if (copy_from_user(ptr, data, len)) {
1080                 OBD_FREE(ptr, len);
1081                 return -EFAULT;
1082         }
1083         rc = obd_iocontrol(cmd, exp, len, data, NULL);
1084         OBD_FREE(ptr, len);
1085         return rc;
1086 }
1087
1088 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1089 {
1090         int cmd = qctl->qc_cmd;
1091         int type = qctl->qc_type;
1092         int id = qctl->qc_id;
1093         int valid = qctl->qc_valid;
1094         int rc = 0;
1095         ENTRY;
1096
1097         switch (cmd) {
1098         case LUSTRE_Q_INVALIDATE:
1099         case LUSTRE_Q_FINVALIDATE:
1100         case Q_QUOTAON:
1101         case Q_QUOTAOFF:
1102         case Q_SETQUOTA:
1103         case Q_SETINFO:
1104                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1105                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1106                         RETURN(-EPERM);
1107                 break;
1108         case Q_GETQUOTA:
1109                 if (((type == USRQUOTA &&
1110                       uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
1111                      (type == GRPQUOTA &&
1112                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
1113                     (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1114                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
1115                         RETURN(-EPERM);
1116                 break;
1117         case Q_GETINFO:
1118                 break;
1119         default:
1120                 CERROR("unsupported quotactl op: %#x\n", cmd);
1121                 RETURN(-ENOTTY);
1122         }
1123
1124         if (valid != QC_GENERAL) {
1125                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1126                         RETURN(-EOPNOTSUPP);
1127
1128                 if (cmd == Q_GETINFO)
1129                         qctl->qc_cmd = Q_GETOINFO;
1130                 else if (cmd == Q_GETQUOTA)
1131                         qctl->qc_cmd = Q_GETOQUOTA;
1132                 else
1133                         RETURN(-EINVAL);
1134
1135                 switch (valid) {
1136                 case QC_MDTIDX:
1137                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1138                                            sizeof(*qctl), qctl, NULL);
1139                         break;
1140                 case QC_OSTIDX:
1141                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1142                                            sizeof(*qctl), qctl, NULL);
1143                         break;
1144                 case QC_UUID:
1145                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1146                                            sizeof(*qctl), qctl, NULL);
1147                         if (rc == -EAGAIN)
1148                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1149                                                    sbi->ll_dt_exp,
1150                                                    sizeof(*qctl), qctl, NULL);
1151                         break;
1152                 default:
1153                         rc = -EINVAL;
1154                         break;
1155                 }
1156
1157                 if (rc)
1158                         RETURN(rc);
1159
1160                 qctl->qc_cmd = cmd;
1161         } else {
1162                 struct obd_quotactl *oqctl;
1163
1164                 OBD_ALLOC_PTR(oqctl);
1165                 if (oqctl == NULL)
1166                         RETURN(-ENOMEM);
1167
1168                 QCTL_COPY(oqctl, qctl);
1169                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1170                 if (rc) {
1171                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
1172                                 oqctl->qc_cmd = Q_QUOTAOFF;
1173                                 obd_quotactl(sbi->ll_md_exp, oqctl);
1174                         }
1175                         OBD_FREE_PTR(oqctl);
1176                         RETURN(rc);
1177                 }
1178                 /* If QIF_SPACE is not set, client should collect the
1179                  * space usage from OSSs by itself */
1180                 if (cmd == Q_GETQUOTA &&
1181                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1182                     !oqctl->qc_dqblk.dqb_curspace) {
1183                         struct obd_quotactl *oqctl_tmp;
1184
1185                         OBD_ALLOC_PTR(oqctl_tmp);
1186                         if (oqctl_tmp == NULL)
1187                                 GOTO(out, rc = -ENOMEM);
1188
1189                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1190                         oqctl_tmp->qc_id = oqctl->qc_id;
1191                         oqctl_tmp->qc_type = oqctl->qc_type;
1192
1193                         /* collect space usage from OSTs */
1194                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1195                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1196                         if (!rc || rc == -EREMOTEIO) {
1197                                 oqctl->qc_dqblk.dqb_curspace =
1198                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1199                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1200                         }
1201
1202                         /* collect space & inode usage from MDTs */
1203                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1204                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1205                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1206                         if (!rc || rc == -EREMOTEIO) {
1207                                 oqctl->qc_dqblk.dqb_curspace +=
1208                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1209                                 oqctl->qc_dqblk.dqb_curinodes =
1210                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1211                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1212                         } else {
1213                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1214                         }
1215
1216                         OBD_FREE_PTR(oqctl_tmp);
1217                 }
1218 out:
1219                 QCTL_COPY(qctl, oqctl);
1220                 OBD_FREE_PTR(oqctl);
1221         }
1222
1223         RETURN(rc);
1224 }
1225
1226 static char *
1227 ll_getname(const char __user *filename)
1228 {
1229         int ret = 0, len;
1230         char *tmp = __getname();
1231
1232         if (!tmp)
1233                 return ERR_PTR(-ENOMEM);
1234
1235         len = strncpy_from_user(tmp, filename, PATH_MAX);
1236         if (len == 0)
1237                 ret = -ENOENT;
1238         else if (len > PATH_MAX)
1239                 ret = -ENAMETOOLONG;
1240
1241         if (ret) {
1242                 __putname(tmp);
1243                 tmp =  ERR_PTR(ret);
1244         }
1245         return tmp;
1246 }
1247
1248 #define ll_putname(filename) __putname(filename)
1249
1250 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1251 {
1252         struct inode *inode = file->f_dentry->d_inode;
1253         struct ll_sb_info *sbi = ll_i2sbi(inode);
1254         struct obd_ioctl_data *data;
1255         int rc = 0;
1256         ENTRY;
1257
1258         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1259                inode->i_ino, inode->i_generation, inode, cmd);
1260
1261         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1262         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1263                 return -ENOTTY;
1264
1265         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1266         switch(cmd) {
1267         case FSFILT_IOC_GETFLAGS:
1268         case FSFILT_IOC_SETFLAGS:
1269                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1270         case FSFILT_IOC_GETVERSION_OLD:
1271         case FSFILT_IOC_GETVERSION:
1272                 RETURN(put_user(inode->i_generation, (int *)arg));
1273         /* We need to special case any other ioctls we want to handle,
1274          * to send them to the MDS/OST as appropriate and to properly
1275          * network encode the arg field.
1276         case FSFILT_IOC_SETVERSION_OLD:
1277         case FSFILT_IOC_SETVERSION:
1278         */
1279         case LL_IOC_GET_MDTIDX: {
1280                 int mdtidx;
1281
1282                 mdtidx = ll_get_mdt_idx(inode);
1283                 if (mdtidx < 0)
1284                         RETURN(mdtidx);
1285
1286                 if (put_user((int)mdtidx, (int*)arg))
1287                         RETURN(-EFAULT);
1288
1289                 return 0;
1290         }
1291         case IOC_MDC_LOOKUP: {
1292                 struct ptlrpc_request *request = NULL;
1293                 int namelen, len = 0;
1294                 char *buf = NULL;
1295                 char *filename;
1296                 struct md_op_data *op_data;
1297
1298                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1299                 if (rc)
1300                         RETURN(rc);
1301                 data = (void *)buf;
1302
1303                 filename = data->ioc_inlbuf1;
1304                 namelen = strlen(filename);
1305
1306                 if (namelen < 1) {
1307                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1308                         GOTO(out_free, rc = -EINVAL);
1309                 }
1310
1311                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1312                                              0, LUSTRE_OPC_ANY, NULL);
1313                 if (IS_ERR(op_data))
1314                         GOTO(out_free, rc = PTR_ERR(op_data));
1315
1316                 op_data->op_valid = OBD_MD_FLID;
1317                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1318                 ll_finish_md_op_data(op_data);
1319                 if (rc < 0) {
1320                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1321                         GOTO(out_free, rc);
1322                 }
1323                 ptlrpc_req_finished(request);
1324                 EXIT;
1325 out_free:
1326                 obd_ioctl_freedata(buf, len);
1327                 return rc;
1328         }
1329         case LL_IOC_LMV_SETSTRIPE: {
1330                 struct lmv_user_md  *lum;
1331                 char            *buf = NULL;
1332                 char            *filename;
1333                 int              namelen = 0;
1334                 int              lumlen = 0;
1335                 int              len;
1336                 int              rc;
1337
1338                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1339                 if (rc)
1340                         RETURN(rc);
1341
1342                 data = (void *)buf;
1343                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1344                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1345                         GOTO(lmv_out_free, rc = -EINVAL);
1346
1347                 filename = data->ioc_inlbuf1;
1348                 namelen = data->ioc_inllen1;
1349
1350                 if (namelen < 1) {
1351                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1352                         GOTO(lmv_out_free, rc = -EINVAL);
1353                 }
1354                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1355                 lumlen = data->ioc_inllen2;
1356
1357                 if (lum->lum_magic != LMV_USER_MAGIC ||
1358                     lumlen != sizeof(*lum)) {
1359                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1360                                filename, lum->lum_magic, lumlen, -EFAULT);
1361                         GOTO(lmv_out_free, rc = -EINVAL);
1362                 }
1363
1364                 /**
1365                  * ll_dir_setdirstripe will be used to set dir stripe
1366                  *  mdc_create--->mdt_reint_create (with dirstripe)
1367                  */
1368                 rc = ll_dir_setdirstripe(inode, lum, filename);
1369 lmv_out_free:
1370                 obd_ioctl_freedata(buf, len);
1371                 RETURN(rc);
1372
1373         }
1374         case LL_IOC_LOV_SETSTRIPE: {
1375                 struct lov_user_md_v3 lumv3;
1376                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1377                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1378                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1379
1380                 int set_default = 0;
1381
1382                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1383                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1384                         sizeof(lumv3p->lmm_objects[0]));
1385                 /* first try with v1 which is smaller than v3 */
1386                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1387                         RETURN(-EFAULT);
1388
1389                 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1390                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1391                                 RETURN(-EFAULT);
1392                 }
1393
1394                 if (inode->i_sb->s_root == file->f_dentry)
1395                         set_default = 1;
1396
1397                 /* in v1 and v3 cases lumv1 points to data */
1398                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1399
1400                 RETURN(rc);
1401         }
1402         case LL_IOC_LMV_GETSTRIPE: {
1403                 struct lmv_user_md *lump = (struct lmv_user_md *)arg;
1404                 struct lmv_user_md lum;
1405                 struct lmv_user_md *tmp;
1406                 int lum_size;
1407                 int rc = 0;
1408                 int mdtindex;
1409
1410                 if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
1411                         RETURN(-EFAULT);
1412
1413                 if (lum.lum_magic != LMV_MAGIC_V1)
1414                         RETURN(-EINVAL);
1415
1416                 lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
1417                 OBD_ALLOC(tmp, lum_size);
1418                 if (tmp == NULL)
1419                         GOTO(free_lmv, rc = -ENOMEM);
1420
1421                 memcpy(tmp, &lum, sizeof(lum));
1422                 tmp->lum_type = LMV_STRIPE_TYPE;
1423                 tmp->lum_stripe_count = 1;
1424                 mdtindex = ll_get_mdt_idx(inode);
1425                 if (mdtindex < 0)
1426                         GOTO(free_lmv, rc = -ENOMEM);
1427
1428                 tmp->lum_stripe_offset = mdtindex;
1429                 tmp->lum_objects[0].lum_mds = mdtindex;
1430                 memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
1431                        sizeof(struct lu_fid));
1432                 if (copy_to_user((void *)arg, tmp, lum_size))
1433                         GOTO(free_lmv, rc = -EFAULT);
1434 free_lmv:
1435                 if (tmp)
1436                         OBD_FREE(tmp, lum_size);
1437                 RETURN(rc);
1438         }
1439         case LL_IOC_REMOVE_ENTRY: {
1440                 char            *filename = NULL;
1441                 int              namelen = 0;
1442                 int              rc;
1443
1444                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1445                  * unsupported server, which might crash the server(LU-2730),
1446                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1447                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1448                  * server will support REINT_RMENTRY XXX*/
1449                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1450                         return -ENOTSUPP;
1451
1452                 filename = ll_getname((const char *)arg);
1453                 if (IS_ERR(filename))
1454                         RETURN(PTR_ERR(filename));
1455
1456                 namelen = strlen(filename);
1457                 if (namelen < 1)
1458                         GOTO(out_rmdir, rc = -EINVAL);
1459
1460                 rc = ll_rmdir_entry(inode, filename, namelen);
1461 out_rmdir:
1462                 if (filename)
1463                         ll_putname(filename);
1464                 RETURN(rc);
1465         }
1466         case LL_IOC_LOV_SWAP_LAYOUTS:
1467                 RETURN(-EPERM);
1468         case LL_IOC_OBD_STATFS:
1469                 RETURN(ll_obd_statfs(inode, (void *)arg));
1470         case LL_IOC_LOV_GETSTRIPE:
1471         case LL_IOC_MDC_GETINFO:
1472         case IOC_MDC_GETFILEINFO:
1473         case IOC_MDC_GETFILESTRIPE: {
1474                 struct ptlrpc_request *request = NULL;
1475                 struct lov_user_md *lump;
1476                 struct lov_mds_md *lmm = NULL;
1477                 struct mdt_body *body;
1478                 char *filename = NULL;
1479                 int lmmsize;
1480
1481                 if (cmd == IOC_MDC_GETFILEINFO ||
1482                     cmd == IOC_MDC_GETFILESTRIPE) {
1483                         filename = ll_getname((const char *)arg);
1484                         if (IS_ERR(filename))
1485                                 RETURN(PTR_ERR(filename));
1486
1487                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1488                                                       &lmmsize, &request);
1489                 } else {
1490                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1491                 }
1492
1493                 if (request) {
1494                         body = req_capsule_server_get(&request->rq_pill,
1495                                                       &RMF_MDT_BODY);
1496                         LASSERT(body != NULL);
1497                 } else {
1498                         GOTO(out_req, rc);
1499                 }
1500
1501                 if (rc < 0) {
1502                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1503                                                cmd == LL_IOC_MDC_GETINFO))
1504                                 GOTO(skip_lmm, rc = 0);
1505                         else
1506                                 GOTO(out_req, rc);
1507                 }
1508
1509                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1510                     cmd == LL_IOC_LOV_GETSTRIPE) {
1511                         lump = (struct lov_user_md *)arg;
1512                 } else {
1513                         struct lov_user_mds_data *lmdp;
1514                         lmdp = (struct lov_user_mds_data *)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                                 GOTO(out_req, rc = -EFAULT);
1520                         rc = -EOVERFLOW;
1521                 }
1522         skip_lmm:
1523                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1524                         struct lov_user_mds_data *lmdp;
1525                         lstat_t st = { 0 };
1526
1527                         st.st_dev     = inode->i_sb->s_dev;
1528                         st.st_mode    = body->mode;
1529                         st.st_nlink   = body->nlink;
1530                         st.st_uid     = body->uid;
1531                         st.st_gid     = body->gid;
1532                         st.st_rdev    = body->rdev;
1533                         st.st_size    = body->size;
1534                         st.st_blksize = PAGE_CACHE_SIZE;
1535                         st.st_blocks  = body->blocks;
1536                         st.st_atime   = body->atime;
1537                         st.st_mtime   = body->mtime;
1538                         st.st_ctime   = body->ctime;
1539                         st.st_ino     = inode->i_ino;
1540
1541                         lmdp = (struct lov_user_mds_data *)arg;
1542                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1543                                 GOTO(out_req, rc = -EFAULT);
1544                 }
1545
1546                 EXIT;
1547         out_req:
1548                 ptlrpc_req_finished(request);
1549                 if (filename)
1550                         ll_putname(filename);
1551                 return rc;
1552         }
1553         case IOC_LOV_GETINFO: {
1554                 struct lov_user_mds_data *lumd;
1555                 struct lov_stripe_md *lsm;
1556                 struct lov_user_md *lum;
1557                 struct lov_mds_md *lmm;
1558                 int lmmsize;
1559                 lstat_t st;
1560
1561                 lumd = (struct lov_user_mds_data *)arg;
1562                 lum = &lumd->lmd_lmm;
1563
1564                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1565                 if (rc)
1566                         RETURN(rc);
1567
1568                 OBD_ALLOC_LARGE(lmm, lmmsize);
1569                 if (copy_from_user(lmm, lum, lmmsize))
1570                         GOTO(free_lmm, rc = -EFAULT);
1571
1572                 switch (lmm->lmm_magic) {
1573                 case LOV_USER_MAGIC_V1:
1574                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1575                                 break;
1576                         /* swab objects first so that stripes num will be sane */
1577                         lustre_swab_lov_user_md_objects(
1578                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1579                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1580                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1581                         break;
1582                 case LOV_USER_MAGIC_V3:
1583                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1584                                 break;
1585                         /* swab objects first so that stripes num will be sane */
1586                         lustre_swab_lov_user_md_objects(
1587                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1588                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1589                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1590                         break;
1591                 default:
1592                         GOTO(free_lmm, rc = -EINVAL);
1593                 }
1594
1595                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1596                 if (rc < 0)
1597                         GOTO(free_lmm, rc = -ENOMEM);
1598
1599                 /* Perform glimpse_size operation. */
1600                 memset(&st, 0, sizeof(st));
1601
1602                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1603                 if (rc)
1604                         GOTO(free_lsm, rc);
1605
1606                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1607                         GOTO(free_lsm, rc = -EFAULT);
1608
1609                 EXIT;
1610         free_lsm:
1611                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1612         free_lmm:
1613                 OBD_FREE_LARGE(lmm, lmmsize);
1614                 return rc;
1615         }
1616         case OBD_IOC_LLOG_CATINFO: {
1617                 RETURN(-EOPNOTSUPP);
1618         }
1619         case OBD_IOC_QUOTACHECK: {
1620                 struct obd_quotactl *oqctl;
1621                 int error = 0;
1622
1623                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1624                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1625                         RETURN(-EPERM);
1626
1627                 OBD_ALLOC_PTR(oqctl);
1628                 if (!oqctl)
1629                         RETURN(-ENOMEM);
1630                 oqctl->qc_type = arg;
1631                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1632                 if (rc < 0) {
1633                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1634                         error = rc;
1635                 }
1636
1637                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1638                 if (rc < 0)
1639                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1640
1641                 OBD_FREE_PTR(oqctl);
1642                 return error ?: rc;
1643         }
1644         case OBD_IOC_POLL_QUOTACHECK: {
1645                 struct if_quotacheck *check;
1646
1647                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1648                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1649                         RETURN(-EPERM);
1650
1651                 OBD_ALLOC_PTR(check);
1652                 if (!check)
1653                         RETURN(-ENOMEM);
1654
1655                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1656                                    NULL);
1657                 if (rc) {
1658                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1659                         if (copy_to_user((void *)arg, check,
1660                                              sizeof(*check)))
1661                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1662                         GOTO(out_poll, rc);
1663                 }
1664
1665                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1666                                    NULL);
1667                 if (rc) {
1668                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1669                         if (copy_to_user((void *)arg, check,
1670                                              sizeof(*check)))
1671                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1672                         GOTO(out_poll, rc);
1673                 }
1674         out_poll:
1675                 OBD_FREE_PTR(check);
1676                 RETURN(rc);
1677         }
1678 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0)
1679         case LL_IOC_QUOTACTL_18: {
1680                 /* copy the old 1.x quota struct for internal use, then copy
1681                  * back into old format struct.  For 1.8 compatibility. */
1682                 struct if_quotactl_18 *qctl_18;
1683                 struct if_quotactl *qctl_20;
1684
1685                 OBD_ALLOC_PTR(qctl_18);
1686                 if (!qctl_18)
1687                         RETURN(-ENOMEM);
1688
1689                 OBD_ALLOC_PTR(qctl_20);
1690                 if (!qctl_20)
1691                         GOTO(out_quotactl_18, rc = -ENOMEM);
1692
1693                 if (copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1694                         GOTO(out_quotactl_20, rc = -ENOMEM);
1695
1696                 QCTL_COPY(qctl_20, qctl_18);
1697                 qctl_20->qc_idx = 0;
1698
1699                 /* XXX: dqb_valid was borrowed as a flag to mark that
1700                  *      only mds quota is wanted */
1701                 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1702                     qctl_18->qc_dqblk.dqb_valid) {
1703                         qctl_20->qc_valid = QC_MDTIDX;
1704                         qctl_20->qc_dqblk.dqb_valid = 0;
1705                 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1706                         qctl_20->qc_valid = QC_UUID;
1707                         qctl_20->obd_uuid = qctl_18->obd_uuid;
1708                 } else {
1709                         qctl_20->qc_valid = QC_GENERAL;
1710                 }
1711
1712                 rc = quotactl_ioctl(sbi, qctl_20);
1713
1714                 if (rc == 0) {
1715                         QCTL_COPY(qctl_18, qctl_20);
1716                         qctl_18->obd_uuid = qctl_20->obd_uuid;
1717
1718                         if (copy_to_user((void *)arg, qctl_18,
1719                                              sizeof(*qctl_18)))
1720                                 rc = -EFAULT;
1721                 }
1722
1723         out_quotactl_20:
1724                 OBD_FREE_PTR(qctl_20);
1725         out_quotactl_18:
1726                 OBD_FREE_PTR(qctl_18);
1727                 RETURN(rc);
1728         }
1729 #else
1730 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1731 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 50, 0) */
1732         case LL_IOC_QUOTACTL: {
1733                 struct if_quotactl *qctl;
1734
1735                 OBD_ALLOC_PTR(qctl);
1736                 if (!qctl)
1737                         RETURN(-ENOMEM);
1738
1739                 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1740                         GOTO(out_quotactl, rc = -EFAULT);
1741
1742                 rc = quotactl_ioctl(sbi, qctl);
1743
1744                 if (rc == 0 && copy_to_user((void *)arg,qctl,sizeof(*qctl)))
1745                         rc = -EFAULT;
1746
1747         out_quotactl:
1748                 OBD_FREE_PTR(qctl);
1749                 RETURN(rc);
1750         }
1751         case OBD_IOC_GETDTNAME:
1752         case OBD_IOC_GETMDNAME:
1753                 RETURN(ll_get_obd_name(inode, cmd, arg));
1754         case LL_IOC_FLUSHCTX:
1755                 RETURN(ll_flush_ctx(inode));
1756 #ifdef CONFIG_FS_POSIX_ACL
1757         case LL_IOC_RMTACL: {
1758             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1759                 inode == inode->i_sb->s_root->d_inode) {
1760                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1761
1762                 LASSERT(fd != NULL);
1763                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1764                 if (!rc)
1765                         fd->fd_flags |= LL_FILE_RMTACL;
1766                 RETURN(rc);
1767             } else
1768                 RETURN(0);
1769         }
1770 #endif
1771         case LL_IOC_GETOBDCOUNT: {
1772                 int count, vallen;
1773                 struct obd_export *exp;
1774
1775                 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1776                         RETURN(-EFAULT);
1777
1778                 /* get ost count when count is zero, get mdt count otherwise */
1779                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1780                 vallen = sizeof(count);
1781                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1782                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1783                 if (rc) {
1784                         CERROR("get target count failed: %d\n", rc);
1785                         RETURN(rc);
1786                 }
1787
1788                 if (copy_to_user((int *)arg, &count, sizeof(int)))
1789                         RETURN(-EFAULT);
1790
1791                 RETURN(0);
1792         }
1793         case LL_IOC_PATH2FID:
1794                 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1795                                      sizeof(struct lu_fid)))
1796                         RETURN(-EFAULT);
1797                 RETURN(0);
1798         case LL_IOC_GET_CONNECT_FLAGS: {
1799                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1800         }
1801         case OBD_IOC_CHANGELOG_SEND:
1802         case OBD_IOC_CHANGELOG_CLEAR:
1803                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1804                                     sizeof(struct ioc_changelog));
1805                 RETURN(rc);
1806         case OBD_IOC_FID2PATH:
1807                 RETURN(ll_fid2path(inode, (void *)arg));
1808         case LL_IOC_HSM_REQUEST: {
1809                 struct hsm_user_request *hur;
1810                 int                      totalsize;
1811
1812                 OBD_ALLOC_PTR(hur);
1813                 if (hur == NULL)
1814                         RETURN(-ENOMEM);
1815
1816                 /* We don't know the true size yet; copy the fixed-size part */
1817                 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1818                         OBD_FREE_PTR(hur);
1819                         RETURN(-EFAULT);
1820                 }
1821
1822                 /* Compute the whole struct size */
1823                 totalsize = hur_len(hur);
1824                 OBD_FREE_PTR(hur);
1825                 OBD_ALLOC_LARGE(hur, totalsize);
1826                 if (hur == NULL)
1827                         RETURN(-ENOMEM);
1828
1829                 /* Copy the whole struct */
1830                 if (copy_from_user(hur, (void *)arg, totalsize)) {
1831                         OBD_FREE_LARGE(hur, totalsize);
1832                         RETURN(-EFAULT);
1833                 }
1834
1835                 rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1836                                    hur, NULL);
1837
1838                 OBD_FREE_LARGE(hur, totalsize);
1839
1840                 RETURN(rc);
1841         }
1842         case LL_IOC_HSM_PROGRESS: {
1843                 struct hsm_progress_kernel      hpk;
1844                 struct hsm_progress             hp;
1845
1846                 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1847                         RETURN(-EFAULT);
1848
1849                 hpk.hpk_fid = hp.hp_fid;
1850                 hpk.hpk_cookie = hp.hp_cookie;
1851                 hpk.hpk_extent = hp.hp_extent;
1852                 hpk.hpk_flags = hp.hp_flags;
1853                 hpk.hpk_errval = hp.hp_errval;
1854                 hpk.hpk_data_version = 0;
1855
1856                 /* File may not exist in Lustre; all progress
1857                  * reported to Lustre root */
1858                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1859                                    NULL);
1860                 RETURN(rc);
1861         }
1862         case LL_IOC_HSM_CT_START:
1863                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1864                                     sizeof(struct lustre_kernelcomm));
1865                 RETURN(rc);
1866
1867         case LL_IOC_HSM_COPY_START: {
1868                 struct hsm_copy *copy;
1869                 int              rc;
1870
1871                 OBD_ALLOC_PTR(copy);
1872                 if (copy == NULL)
1873                         RETURN(-ENOMEM);
1874                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1875                         OBD_FREE_PTR(copy);
1876                         RETURN(-EFAULT);
1877                 }
1878
1879                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1880                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1881                         rc = -EFAULT;
1882
1883                 OBD_FREE_PTR(copy);
1884                 RETURN(rc);
1885         }
1886         case LL_IOC_HSM_COPY_END: {
1887                 struct hsm_copy *copy;
1888                 int              rc;
1889
1890                 OBD_ALLOC_PTR(copy);
1891                 if (copy == NULL)
1892                         RETURN(-ENOMEM);
1893                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1894                         OBD_FREE_PTR(copy);
1895                         RETURN(-EFAULT);
1896                 }
1897
1898                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1899                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1900                         rc = -EFAULT;
1901
1902                 OBD_FREE_PTR(copy);
1903                 RETURN(rc);
1904         }
1905         default:
1906                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1907                                      (void *)arg));
1908         }
1909 }
1910
1911 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1912 {
1913         struct inode *inode = file->f_mapping->host;
1914         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1915         struct ll_sb_info *sbi = ll_i2sbi(inode);
1916         int api32 = ll_need_32bit_api(sbi);
1917         loff_t ret = -EINVAL;
1918         ENTRY;
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         ENTRY;
1965         RETURN(ll_file_open(inode, file));
1966 }
1967
1968 int ll_dir_release(struct inode *inode, struct file *file)
1969 {
1970         ENTRY;
1971         RETURN(ll_file_release(inode, file));
1972 }
1973
1974 struct file_operations ll_dir_operations = {
1975         .llseek   = ll_dir_seek,
1976         .open     = ll_dir_open,
1977         .release  = ll_dir_release,
1978         .read     = generic_read_dir,
1979         .readdir  = ll_readdir,
1980         .unlocked_ioctl   = ll_dir_ioctl,
1981         .fsync    = ll_fsync,
1982 };