lustre/llite/dir.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * Copyright (c) 2011 Whamcloud, Inc.
  34  */
  35 /*
  36  * This file is part of Lustre, http://www.lustre.org/
  37  * Lustre is a trademark of Sun Microsystems, Inc.
  38  *
  39  * lustre/llite/dir.c
  40  *
  41  * Directory code for lustre client.
  42  */
  43
  44 #include <linux/fs.h>
  45 #include <linux/pagemap.h>
  46 #include <linux/mm.h>
  47 #include <linux/version.h>
  48 #include <linux/smp_lock.h>
  49 #include <asm/uaccess.h>
  50 #include <linux/buffer_head.h>   // for wait_on_buffer
  51 #include <linux/pagevec.h>
  52
  53 #define DEBUG_SUBSYSTEM S_LLITE
  54
  55 #include <obd_support.h>
  56 #include <obd_class.h>
  57 #include <lustre_lib.h>
  58 #include <lustre/lustre_idl.h>
  59 #include <lustre_lite.h>
  60 #include <lustre_dlm.h>
  61 #include <lustre_fid.h>
  62 #include "llite_internal.h"
  63
  64 #ifndef HAVE_PAGE_CHECKED
  65 #ifdef HAVE_PG_FS_MISC
  66 #define PageChecked(page)        test_bit(PG_fs_misc, &(page)->flags)
  67 #define SetPageChecked(page)     set_bit(PG_fs_misc, &(page)->flags)
  68 #else
  69 #error PageChecked or PageFsMisc not defined in kernel
  70 #endif
  71 #endif
  72
  73 /*
  74  * (new) readdir implementation overview.
  75  *
  76  * Original lustre readdir implementation cached exact copy of raw directory
  77  * pages on the client. These pages were indexed in client page cache by
  78  * logical offset in the directory file. This design, while very simple and
  79  * intuitive had some inherent problems:
  80  *
  81  *     . it implies that byte offset to the directory entry serves as a
  82  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  83  *     ext3/htree directory entries may move due to splits, and more
  84  *     importantly,
  85  *
  86  *     . it is incompatible with the design of split directories for cmd3,
  87  *     that assumes that names are distributed across nodes based on their
  88  *     hash, and so readdir should be done in hash order.
  89  *
  90  * New readdir implementation does readdir in hash order, and uses hash of a
  91  * file name as a telldir/seekdir cookie. This led to number of complications:
  92  *
  93  *     . hash is not unique, so it cannot be used to index cached directory
  94  *     pages on the client (note, that it requires a whole pageful of hash
  95  *     collided entries to cause two pages to have identical hashes);
  96  *
  97  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  98  *     entry cookie. ext3/htree has the same problem and lustre implementation
  99  *     mimics their solution: seekdir(hash) positions directory at the first
 100  *     entry with the given hash.
 101  *
 102  * Client side.
 103  *
 104  * 0. caching
 105  *
 106  * Client caches directory pages using hash of the first entry as an index. As
 107  * noted above hash is not unique, so this solution doesn't work as is:
 108  * special processing is needed for "page hash chains" (i.e., sequences of
 109  * pages filled with entries all having the same hash value).
 110  *
 111  * First, such chains have to be detected. To this end, server returns to the
 112  * client the hash of the first entry on the page next to one returned. When
 113  * client detects that this hash is the same as hash of the first entry on the
 114  * returned page, page hash collision has to be handled. Pages in the
 115  * hash chain, except first one, are termed "overflow pages".
 116  *
 117  * Solution to index uniqueness problem is to not cache overflow
 118  * pages. Instead, when page hash collision is detected, all overflow pages
 119  * from emerging chain are immediately requested from the server and placed in
 120  * a special data structure (struct ll_dir_chain). This data structure is used
 121  * by ll_readdir() to process entries from overflow pages. When readdir
 122  * invocation finishes, overflow pages are discarded. If page hash collision
 123  * chain weren't completely processed, next call to readdir will again detect
 124  * page hash collision, again read overflow pages in, process next portion of
 125  * entries and again discard the pages. This is not as wasteful as it looks,
 126  * because, given reasonable hash, page hash collisions are extremely rare.
 127  *
 128  * 1. directory positioning
 129  *
 130  * When seekdir(hash) is called, original
 131  *
 132  *
 133  *
 134  *
 135  *
 136  *
 137  *
 138  *
 139  * Server.
 140  *
 141  * identification of and access to overflow pages
 142  *
 143  * page format
 144  *
 145  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 146  * a header lu_dirpage which describes the start/end hash, and whether this
 147  * page is empty (contains no dir entry) or hash collide with next page.
 148  * After client receives reply, several pages will be integrated into dir page
 149  * in CFS_PAGE_SIZE (if CFS_PAGE_SIZE greater than LU_PAGE_SIZE), and the
 150  * lu_dirpage for this integrated page will be adjusted.
 151  *
 152  */
 153
 154 /* returns the page unlocked, but with a reference */
 155 static int ll_dir_readpage(struct file *file, struct page *page0)
 156 {
 157         struct inode *inode = page0->mapping->host;
 158         int hash64 = ll_i2sbi(inode)->ll_flags & LL_SBI_64BIT_HASH;
 159         struct obd_export *exp = ll_i2sbi(inode)->ll_md_exp;
 160         struct ptlrpc_request *request;
 161         struct mdt_body *body;
 162         struct obd_capa *oc;
 163         __u64 hash;
 164         struct page **page_pool;
 165         struct page *page;
 166 #ifndef HAVE_ADD_TO_PAGE_CACHE_LRU
 167         struct pagevec lru_pvec;
 168 #endif
 169         struct lu_dirpage *dp;
 170         int max_pages = ll_i2sbi(inode)->ll_md_brw_size >> CFS_PAGE_SHIFT;
 171         int nrdpgs = 0; /* number of pages read actually */
 172         int npages;
 173         int i;
 174         int rc;
 175         ENTRY;
 176
 177         if (file) {
 178                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
 179
 180                 hash = fd->fd_dir.lfd_next;
 181         } else {
 182                 struct ll_inode_info *lli = ll_i2info(inode);
 183
 184                 cfs_spin_lock(&lli->lli_sa_lock);
 185                 if (lli->lli_sai)
 186                         LASSERT(lli->lli_sai->sai_pid == cfs_curproc_pid());
 187                 else
 188                         LASSERT(lli->lli_opendir_pid == cfs_curproc_pid());
 189                 hash = lli->lli_sa_pos;
 190                 cfs_spin_unlock(&lli->lli_sa_lock);
 191         }
 192         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) hash "LPU64"\n",
 193                inode->i_ino, inode->i_generation, inode, hash);
 194
 195         LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
 196
 197         OBD_ALLOC(page_pool, sizeof(page) * max_pages);
 198         if (page_pool != NULL) {
 199                 page_pool[0] = page0;
 200         } else {
 201                 page_pool = &page0;
 202                 max_pages = 1;
 203         }
 204         for (npages = 1; npages < max_pages; npages++) {
 205                 page = page_cache_alloc_cold(inode->i_mapping);
 206                 if (!page)
 207                         break;
 208                 page_pool[npages] = page;
 209         }
 210
 211         oc = ll_mdscapa_get(inode);
 212         rc = md_readpage(exp, ll_inode2fid(inode), oc, hash, page_pool, npages,
 213                          &request);
 214         capa_put(oc);
 215         if (rc == 0) {
 216                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 217                 /* Checked by mdc_readpage() */
 218                 LASSERT(body != NULL);
 219
 220                 if (body->valid & OBD_MD_FLSIZE)
 221                         cl_isize_write(inode, body->size);
 222
 223                 nrdpgs = (request->rq_bulk->bd_nob_transferred+CFS_PAGE_SIZE-1)
 224                          >> CFS_PAGE_SHIFT;
 225                 SetPageUptodate(page0);
 226         }
 227         unlock_page(page0);
 228         ptlrpc_req_finished(request);
 229
 230         CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages);
 231
 232         ll_pagevec_init(&lru_pvec, 0);
 233         for (i = 1; i < npages; i++) {
 234                 unsigned long offset;
 235                 int ret;
 236
 237                 page = page_pool[i];
 238
 239                 if (rc < 0 || i >= nrdpgs) {
 240                         page_cache_release(page);
 241                         continue;
 242                 }
 243
 244                 SetPageUptodate(page);
 245
 246                 dp = cfs_kmap(page);
 247                 hash = le64_to_cpu(dp->ldp_hash_start);
 248                 cfs_kunmap(page);
 249
 250                 offset = hash_x_index(hash, hash64);
 251
 252                 prefetchw(&page->flags);
 253                 ret = ll_add_to_page_cache_lru(page, inode->i_mapping, offset,
 254                                                GFP_KERNEL);
 255                 if (ret == 0) {
 256                         unlock_page(page);
 257                         page_cache_get(page);
 258                         if (ll_pagevec_add(&lru_pvec, page) == 0)
 259                                 ll_pagevec_lru_add_file(&lru_pvec);
 260                 } else {
 261                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
 262                                " %d\n", offset, ret);
 263                 }
 264                 page_cache_release(page);
 265         }
 266         ll_pagevec_lru_add_file(&lru_pvec);
 267
 268         if (page_pool != &page0)
 269                 OBD_FREE(page_pool, sizeof(struct page *) * max_pages);
 270         EXIT;
 271         return rc;
 272 }
 273
 274 #ifndef MS_HAS_NEW_AOPS
 275 struct address_space_operations ll_dir_aops = {
 276         .readpage  = ll_dir_readpage,
 277 };
 278 #else
 279 struct address_space_operations_ext ll_dir_aops = {
 280         .orig_aops.readpage  = ll_dir_readpage,
 281 };
 282 #endif
 283
 284 static void ll_check_page(struct inode *dir, struct page *page)
 285 {
 286         /* XXX: check page format later */
 287         SetPageChecked(page);
 288 }
 289
 290 static void ll_release_page(struct page *page, __u64 hash,
 291                             __u64 start, __u64 end)
 292 {
 293         kunmap(page);
 294         lock_page(page);
 295         if (likely(page->mapping != NULL)) {
 296                 ll_truncate_complete_page(page);
 297                 unlock_page(page);
 298         } else {
 299                 unlock_page(page);
 300                 CWARN("NULL mapping page %p, truncated by others: "
 301                       "hash("LPX64") | start("LPX64") | end("LPX64")\n",
 302                       page, hash, start, end);
 303         }
 304         page_cache_release(page);
 305 }
 306
 307 /*
 308  * Find, kmap and return page that contains given hash.
 309  */
 310 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
 311                                        __u64 *start, __u64 *end)
 312 {
 313         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 314         struct address_space *mapping = dir->i_mapping;
 315         /*
 316          * Complement of hash is used as an index so that
 317          * radix_tree_gang_lookup() can be used to find a page with starting
 318          * hash _smaller_ than one we are looking for.
 319          */
 320         unsigned long offset = hash_x_index(*hash, hash64);
 321         struct page *page;
 322         int found;
 323
 324         TREE_READ_LOCK_IRQ(mapping);
 325         found = radix_tree_gang_lookup(&mapping->page_tree,
 326                                        (void **)&page, offset, 1);
 327         if (found > 0) {
 328                 struct lu_dirpage *dp;
 329
 330                 page_cache_get(page);
 331                 TREE_READ_UNLOCK_IRQ(mapping);
 332                 /*
 333                  * In contrast to find_lock_page() we are sure that directory
 334                  * page cannot be truncated (while DLM lock is held) and,
 335                  * hence, can avoid restart.
 336                  *
 337                  * In fact, page cannot be locked here at all, because
 338                  * ll_dir_readpage() does synchronous io.
 339                  */
 340                 wait_on_page(page);
 341                 if (PageUptodate(page)) {
 342                         dp = cfs_kmap(page);
 343                         if (BITS_PER_LONG == 32 && hash64) {
 344                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 345                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 346                                 *hash  = *hash >> 32;
 347                         } else {
 348                                 *start = le64_to_cpu(dp->ldp_hash_start);
 349                                 *end   = le64_to_cpu(dp->ldp_hash_end);
 350                         }
 351                         LASSERTF(*start <= *hash, "start = "LPX64",end = "
 352                                  LPX64",hash = "LPX64"\n", *start, *end, *hash);
 353                         CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash "LPU64"\n",
 354                                offset, *start, *end, *hash);
 355                         if (*hash > *end || (*end != *start && *hash == *end)) {
 356                                 /*
 357                                  * upon hash collision, remove this page,
 358                                  * otherwise put page reference, and
 359                                  * ll_get_dir_page() will issue RPC to fetch
 360                                  * the page we want.
 361                                  */
 362                                 if (dp->ldp_flags & cpu_to_le32(LDF_COLLIDE)) {
 363                                         ll_release_page(page, *hash, *start, *end);
 364                                 } else {
 365                                         cfs_kunmap(page);
 366                                         page_cache_release(page);
 367                                 }
 368                                 page = NULL;
 369                         }
 370                 } else {
 371                         page_cache_release(page);
 372                         page = ERR_PTR(-EIO);
 373                 }
 374
 375         } else {
 376                 TREE_READ_UNLOCK_IRQ(mapping);
 377                 page = NULL;
 378         }
 379         return page;
 380 }
 381
 382 struct page *ll_get_dir_page(struct file *filp, struct inode *dir, __u64 hash,
 383                              int exact, struct ll_dir_chain *chain)
 384 {
 385         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 386         struct address_space *mapping = dir->i_mapping;
 387         struct lustre_handle lockh;
 388         struct lu_dirpage *dp;
 389         struct page *page;
 390         ldlm_mode_t mode;
 391         int rc;
 392         __u64 start = 0;
 393         __u64 end = 0;
 394         __u64 lhash = hash;
 395         struct ll_inode_info *lli = ll_i2info(dir);
 396         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 397
 398         mode = LCK_PR;
 399         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 400                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
 401         if (!rc) {
 402                 struct ldlm_enqueue_info einfo = { LDLM_IBITS, mode,
 403                        ll_md_blocking_ast, ldlm_completion_ast,
 404                        NULL, NULL, dir };
 405                 struct lookup_intent it = { .it_op = IT_READDIR };
 406                 struct ptlrpc_request *request;
 407                 struct md_op_data *op_data;
 408
 409                 op_data = ll_prep_md_op_data(NULL, dir, NULL, NULL, 0, 0,
 410                                              LUSTRE_OPC_ANY, NULL);
 411                 if (IS_ERR(op_data))
 412                         return (void *)op_data;
 413
 414                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
 415                                 op_data, &lockh, NULL, 0, NULL, 0);
 416
 417                 ll_finish_md_op_data(op_data);
 418
 419                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 420                 if (request)
 421                         ptlrpc_req_finished(request);
 422                 if (rc < 0) {
 423                         CERROR("lock enqueue: "DFID" at "LPU64": rc %d\n",
 424                                PFID(ll_inode2fid(dir)), hash, rc);
 425                         return ERR_PTR(rc);
 426                 }
 427         } else {
 428                 /* for cross-ref object, l_ast_data of the lock may not be set,
 429                  * we reset it here */
 430                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
 431                                  dir, NULL);
 432         }
 433         ldlm_lock_dump_handle(D_OTHER, &lockh);
 434
 435         cfs_down(&lli->lli_readdir_sem);
 436         page = ll_dir_page_locate(dir, &lhash, &start, &end);
 437         if (IS_ERR(page)) {
 438                 CERROR("dir page locate: "DFID" at "LPU64": rc %ld\n",
 439                        PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
 440                 GOTO(out_unlock, page);
 441         }
 442
 443         if (page != NULL) {
 444                 /*
 445                  * XXX nikita: not entirely correct handling of a corner case:
 446                  * suppose hash chain of entries with hash value HASH crosses
 447                  * border between pages P0 and P1. First both P0 and P1 are
 448                  * cached, seekdir() is called for some entry from the P0 part
 449                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 450                  * happens and finds P1, as it starts with matching hash
 451                  * value. Remaining entries from P0 part of the chain are
 452                  * skipped. (Is that really a bug?)
 453                  *
 454                  * Possible solutions: 0. don't cache P1 is such case, handle
 455                  * it as an "overflow" page. 1. invalidate all pages at
 456                  * once. 2. use HASH|1 as an index for P1.
 457                  */
 458                 if (exact && lhash != start) {
 459                         /*
 460                          * readdir asked for a page starting _exactly_ from
 461                          * given hash, but cache contains stale page, with
 462                          * entries with smaller hash values. Stale page should
 463                          * be invalidated, and new one fetched.
 464                          */
 465                         CDEBUG(D_OTHER, "Stale readpage page %p: "
 466                                "start = "LPX64",end = "LPX64"hash ="LPX64"\n",
 467                                page, start, end, lhash);
 468                         ll_release_page(page, lhash, start, end);
 469                 } else {
 470                         GOTO(hash_collision, page);
 471                 }
 472         }
 473
 474         page = read_cache_page(mapping, hash_x_index(hash, hash64),
 475                                (filler_t*)mapping->a_ops->readpage, filp);
 476         if (IS_ERR(page)) {
 477                 CERROR("read cache page: "DFID" at "LPU64": rc %ld\n",
 478                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 479                 GOTO(out_unlock, page);
 480         }
 481
 482         wait_on_page(page);
 483         (void)kmap(page);
 484         if (!PageUptodate(page)) {
 485                 CERROR("page not updated: "DFID" at "LPU64": rc %d\n",
 486                        PFID(ll_inode2fid(dir)), hash, -5);
 487                 goto fail;
 488         }
 489         if (!PageChecked(page))
 490                 ll_check_page(dir, page);
 491         if (PageError(page)) {
 492                 CERROR("page error: "DFID" at "LPU64": rc %d\n",
 493                        PFID(ll_inode2fid(dir)), hash, -5);
 494                 goto fail;
 495         }
 496 hash_collision:
 497         dp = page_address(page);
 498         if (BITS_PER_LONG == 32 && hash64) {
 499                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 500                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 501                 lhash = hash >> 32;
 502         } else {
 503                 start = le64_to_cpu(dp->ldp_hash_start);
 504                 end   = le64_to_cpu(dp->ldp_hash_end);
 505                 lhash = hash;
 506         }
 507         if (end == start) {
 508                 LASSERT(start == lhash);
 509                 CWARN("Page-wide hash collision: "LPU64"\n", end);
 510                 if (BITS_PER_LONG == 32 && hash64)
 511                         CWARN("Real page-wide hash collision at ["LPU64" "LPU64
 512                               "] with hash "LPU64"\n",
 513                               le64_to_cpu(dp->ldp_hash_start),
 514                               le64_to_cpu(dp->ldp_hash_end), hash);
 515                 /*
 516                  * Fetch whole overflow chain...
 517                  *
 518                  * XXX not yet.
 519                  */
 520                 goto fail;
 521         }
 522 out_unlock:
 523         cfs_up(&lli->lli_readdir_sem);
 524         ldlm_lock_decref(&lockh, mode);
 525         return page;
 526
 527 fail:
 528         ll_put_page(page);
 529         page = ERR_PTR(-EIO);
 530         goto out_unlock;
 531 }
 532
 533 int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 534 {
 535         struct inode         *inode      = filp->f_dentry->d_inode;
 536         struct ll_inode_info *info       = ll_i2info(inode);
 537         struct ll_sb_info    *sbi        = ll_i2sbi(inode);
 538         struct ll_file_data  *fd         = LUSTRE_FPRIVATE(filp);
 539         __u64                 pos        = fd->fd_dir.lfd_pos;
 540         int                   api32      = ll_need_32bit_api(sbi);
 541         int                   hash64     = sbi->ll_flags & LL_SBI_64BIT_HASH;
 542         struct page          *page;
 543         struct ll_dir_chain   chain;
 544         int                   done;
 545         int                   shift;
 546         int                   rc;
 547         ENTRY;
 548
 549         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu 32bit_api %d\n",
 550                inode->i_ino, inode->i_generation, inode,
 551                (unsigned long)pos, i_size_read(inode), api32);
 552
 553         if (pos == MDS_DIR_END_OFF)
 554                 /*
 555                  * end-of-file.
 556                  */
 557                 RETURN(0);
 558
 559         rc    = 0;
 560         done  = 0;
 561         shift = 0;
 562         ll_dir_chain_init(&chain);
 563
 564         fd->fd_dir.lfd_next = pos;
 565         page = ll_get_dir_page(filp, inode, pos, 0, &chain);
 566
 567         while (rc == 0 && !done) {
 568                 struct lu_dirpage *dp;
 569                 struct lu_dirent  *ent;
 570
 571                 if (!IS_ERR(page)) {
 572                         /*
 573                          * If page is empty (end of directory is reached),
 574                          * use this value.
 575                          */
 576                         __u64 hash = MDS_DIR_END_OFF;
 577                         __u64 next;
 578
 579                         dp = page_address(page);
 580                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 581                              ent = lu_dirent_next(ent)) {
 582                                 __u16          type;
 583                                 int            namelen;
 584                                 struct lu_fid  fid;
 585                                 __u64          lhash;
 586                                 __u64          ino;
 587
 588                                 /*
 589                                  * XXX: implement correct swabbing here.
 590                                  */
 591
 592                                 hash = le64_to_cpu(ent->lde_hash);
 593                                 if (hash < pos)
 594                                         /*
 595                                          * Skip until we find target hash
 596                                          * value.
 597                                          */
 598                                         continue;
 599
 600                                 namelen = le16_to_cpu(ent->lde_namelen);
 601                                 if (namelen == 0)
 602                                         /*
 603                                          * Skip dummy record.
 604                                          */
 605                                         continue;
 606
 607                                 if (api32 && hash64)
 608                                         lhash = hash >> 32;
 609                                 else
 610                                         lhash = hash;
 611                                 fid_le_to_cpu(&fid, &ent->lde_fid);
 612                                 ino = cl_fid_build_ino(&fid, api32);
 613                                 type = ll_dirent_type_get(ent);
 614                                 /* For 'll_nfs_get_name_filldir()', it will try
 615                                  * to access the 'ent' through its 'lde_name',
 616                                  * so the parameter 'name' for 'filldir()' must
 617                                  * be part of the 'ent'. */
 618                                 done = filldir(cookie, ent->lde_name, namelen,
 619                                                lhash, ino, type);
 620                         }
 621                         next = le64_to_cpu(dp->ldp_hash_end);
 622                         ll_put_page(page);
 623                         if (!done) {
 624                                 pos = next;
 625                                 if (pos == MDS_DIR_END_OFF) {
 626                                         /*
 627                                          * End of directory reached.
 628                                          */
 629                                         done = 1;
 630                                 } else if (1 /* chain is exhausted*/) {
 631                                         /*
 632                                          * Normal case: continue to the next
 633                                          * page.
 634                                          */
 635                                         fd->fd_dir.lfd_next = pos;
 636                                         page = ll_get_dir_page(filp, inode, pos,
 637                                                                1, &chain);
 638                                 } else {
 639                                         /*
 640                                          * go into overflow page.
 641                                          */
 642                                 }
 643                         } else {
 644                                 pos = hash;
 645                         }
 646                 } else {
 647                         rc = PTR_ERR(page);
 648                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 649                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 650                 }
 651         }
 652
 653         fd->fd_dir.lfd_pos = pos;
 654         if (pos == MDS_DIR_END_OFF) {
 655                 if (api32)
 656                         filp->f_pos = LL_DIR_END_OFF_32BIT;
 657                 else
 658                         filp->f_pos = LL_DIR_END_OFF;
 659         } else {
 660                 if (api32 && hash64)
 661                         filp->f_pos = pos >> 32;
 662                 else
 663                         filp->f_pos = pos;
 664         }
 665         filp->f_version = inode->i_version;
 666         touch_atime(filp->f_vfsmnt, filp->f_dentry);
 667
 668         ll_dir_chain_fini(&chain);
 669
 670         RETURN(rc);
 671 }
 672
 673 int ll_send_mgc_param(struct obd_export *mgc, char *string)
 674 {
 675         struct mgs_send_param *msp;
 676         int rc = 0;
 677
 678         OBD_ALLOC_PTR(msp);
 679         if (!msp)
 680                 return -ENOMEM;
 681
 682         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 683         rc = obd_set_info_async(mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 684                                 sizeof(struct mgs_send_param), msp, NULL);
 685         if (rc)
 686                 CERROR("Failed to set parameter: %d\n", rc);
 687         OBD_FREE_PTR(msp);
 688
 689         return rc;
 690 }
 691
 692 char *ll_get_fsname(struct inode *inode)
 693 {
 694         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 695         char *ptr, *fsname;
 696         int len;
 697
 698         OBD_ALLOC(fsname, MGS_PARAM_MAXLEN);
 699         len = strlen(lsi->lsi_lmd->lmd_profile);
 700         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
 701         if (ptr && (strcmp(ptr, "-client") == 0))
 702                 len -= 7;
 703         strncpy(fsname, lsi->lsi_lmd->lmd_profile, len);
 704         fsname[len] = '\0';
 705
 706         return fsname;
 707 }
 708
 709 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 710                      int set_default)
 711 {
 712         struct ll_sb_info *sbi = ll_i2sbi(inode);
 713         struct md_op_data *op_data;
 714         struct ptlrpc_request *req = NULL;
 715         int rc = 0;
 716         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 717         struct obd_device *mgc = lsi->lsi_mgc;
 718         char *fsname = NULL, *param = NULL;
 719         int lum_size;
 720
 721         if (lump != NULL) {
 722                 /*
 723                  * This is coming from userspace, so should be in
 724                  * local endian.  But the MDS would like it in little
 725                  * endian, so we swab it before we send it.
 726                  */
 727                 switch (lump->lmm_magic) {
 728                 case LOV_USER_MAGIC_V1: {
 729                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 730                                 lustre_swab_lov_user_md_v1(lump);
 731                         lum_size = sizeof(struct lov_user_md_v1);
 732                         break;
 733                         }
 734                 case LOV_USER_MAGIC_V3: {
 735                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 736                                 lustre_swab_lov_user_md_v3(
 737                                         (struct lov_user_md_v3 *)lump);
 738                         lum_size = sizeof(struct lov_user_md_v3);
 739                         break;
 740                         }
 741                 default: {
 742                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 743                                         " %#08x != %#08x nor %#08x\n",
 744                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 745                                         LOV_USER_MAGIC_V3);
 746                         RETURN(-EINVAL);
 747                         }
 748                }
 749         } else {
 750                 lum_size = sizeof(struct lov_user_md_v1);
 751         }
 752
 753         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 754                                      LUSTRE_OPC_ANY, NULL);
 755         if (IS_ERR(op_data))
 756                 RETURN(PTR_ERR(op_data));
 757
 758         /* swabbing is done in lov_setstripe() on server side */
 759         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 760                         NULL, 0, &req, NULL);
 761         ll_finish_md_op_data(op_data);
 762         ptlrpc_req_finished(req);
 763         if (rc) {
 764                 if (rc != -EPERM && rc != -EACCES)
 765                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 766         }
 767
 768         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 769          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 770          need the make the distiction between the 2 versions */
 771         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 772                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 773
 774                 /* Get fsname and assume devname to be -MDT0000. */
 775                 fsname = ll_get_fsname(inode);
 776                 /* Set root stripesize */
 777                 sprintf(param, "%s-MDT0000.lov.stripesize=%u", fsname,
 778                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 779                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 780                 if (rc)
 781                         goto end;
 782
 783                 /* Set root stripecount */
 784                 sprintf(param, "%s-MDT0000.lov.stripecount=%hd", fsname,
 785                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 786                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 787                 if (rc)
 788                         goto end;
 789
 790                 /* Set root stripeoffset */
 791                 sprintf(param, "%s-MDT0000.lov.stripeoffset=%hd", fsname,
 792                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 793                         (typeof(lump->lmm_stripe_offset))(-1));
 794                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 795                 if (rc)
 796                         goto end;
 797 end:
 798                 if (fsname)
 799                         OBD_FREE(fsname, MGS_PARAM_MAXLEN);
 800                 if (param)
 801                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 802         }
 803         return rc;
 804 }
 805
 806 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 807                      int *lmm_size, struct ptlrpc_request **request)
 808 {
 809         struct ll_sb_info *sbi = ll_i2sbi(inode);
 810         struct mdt_body   *body;
 811         struct lov_mds_md *lmm = NULL;
 812         struct ptlrpc_request *req = NULL;
 813         int rc, lmmsize;
 814         struct md_op_data *op_data;
 815
 816         rc = ll_get_max_mdsize(sbi, &lmmsize);
 817         if (rc)
 818                 RETURN(rc);
 819
 820         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 821                                      0, lmmsize, LUSTRE_OPC_ANY,
 822                                      NULL);
 823         if (op_data == NULL)
 824                 RETURN(-ENOMEM);
 825
 826         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 827         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 828         ll_finish_md_op_data(op_data);
 829         if (rc < 0) {
 830                 CDEBUG(D_INFO, "md_getattr failed on inode "
 831                        "%lu/%u: rc %d\n", inode->i_ino,
 832                        inode->i_generation, rc);
 833                 GOTO(out, rc);
 834         }
 835
 836         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 837         LASSERT(body != NULL);
 838
 839         lmmsize = body->eadatasize;
 840
 841         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 842             lmmsize == 0) {
 843                 GOTO(out, rc = -ENODATA);
 844         }
 845
 846         lmm = req_capsule_server_sized_get(&req->rq_pill,
 847                                            &RMF_MDT_MD, lmmsize);
 848         LASSERT(lmm != NULL);
 849
 850         /*
 851          * This is coming from the MDS, so is probably in
 852          * little endian.  We convert it to host endian before
 853          * passing it to userspace.
 854          */
 855         /* We don't swab objects for directories */
 856         switch (le32_to_cpu(lmm->lmm_magic)) {
 857         case LOV_MAGIC_V1:
 858                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 859                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 860                 break;
 861         case LOV_MAGIC_V3:
 862                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 863                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 864                 break;
 865         default:
 866                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 867                 rc = -EPROTO;
 868         }
 869 out:
 870         *lmmp = lmm;
 871         *lmm_size = lmmsize;
 872         *request = req;
 873         return rc;
 874 }
 875
 876 /*
 877  *  Get MDT index for the inode.
 878  */
 879 int ll_get_mdt_idx(struct inode *inode)
 880 {
 881         struct ll_sb_info *sbi = ll_i2sbi(inode);
 882         struct md_op_data *op_data;
 883         int rc, mdtidx;
 884         ENTRY;
 885
 886         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 887                                      0, LUSTRE_OPC_ANY, NULL);
 888         if (op_data == NULL)
 889                 RETURN(-ENOMEM);
 890
 891         op_data->op_valid |= OBD_MD_MDTIDX;
 892         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 893         mdtidx = op_data->op_mds;
 894         ll_finish_md_op_data(op_data);
 895         if (rc < 0) {
 896                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 897                 RETURN(rc);
 898         }
 899         return mdtidx;
 900 }
 901
 902 static int copy_and_ioctl(int cmd, struct obd_export *exp, void *data, int len)
 903 {
 904         void *ptr;
 905         int rc;
 906
 907         OBD_ALLOC(ptr, len);
 908         if (ptr == NULL)
 909                 return -ENOMEM;
 910         if (cfs_copy_from_user(ptr, data, len)) {
 911                 OBD_FREE(ptr, len);
 912                 return -EFAULT;
 913         }
 914         rc = obd_iocontrol(cmd, exp, len, data, NULL);
 915         OBD_FREE(ptr, len);
 916         return rc;
 917 }
 918
 919 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
 920 {
 921         int cmd = qctl->qc_cmd;
 922         int type = qctl->qc_type;
 923         int id = qctl->qc_id;
 924         int valid = qctl->qc_valid;
 925         int rc = 0;
 926         ENTRY;
 927
 928         switch (cmd) {
 929         case LUSTRE_Q_INVALIDATE:
 930         case LUSTRE_Q_FINVALIDATE:
 931         case Q_QUOTAON:
 932         case Q_QUOTAOFF:
 933         case Q_SETQUOTA:
 934         case Q_SETINFO:
 935                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 936                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
 937                         RETURN(-EPERM);
 938                 break;
 939         case Q_GETQUOTA:
 940                 if (((type == USRQUOTA && cfs_curproc_euid() != id) ||
 941                      (type == GRPQUOTA && !in_egroup_p(id))) &&
 942                     (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 943                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
 944                         RETURN(-EPERM);
 945                 break;
 946         case Q_GETINFO:
 947                 break;
 948         default:
 949                 CERROR("unsupported quotactl op: %#x\n", cmd);
 950                 RETURN(-ENOTTY);
 951         }
 952
 953         if (valid != QC_GENERAL) {
 954                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 955                         RETURN(-EOPNOTSUPP);
 956
 957                 if (cmd == Q_GETINFO)
 958                         qctl->qc_cmd = Q_GETOINFO;
 959                 else if (cmd == Q_GETQUOTA)
 960                         qctl->qc_cmd = Q_GETOQUOTA;
 961                 else
 962                         RETURN(-EINVAL);
 963
 964                 switch (valid) {
 965                 case QC_MDTIDX:
 966                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 967                                            sizeof(*qctl), qctl, NULL);
 968                         break;
 969                 case QC_OSTIDX:
 970                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
 971                                            sizeof(*qctl), qctl, NULL);
 972                         break;
 973                 case QC_UUID:
 974                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 975                                            sizeof(*qctl), qctl, NULL);
 976                         if (rc == -EAGAIN)
 977                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
 978                                                    sbi->ll_dt_exp,
 979                                                    sizeof(*qctl), qctl, NULL);
 980                         break;
 981                 default:
 982                         rc = -EINVAL;
 983                         break;
 984                 }
 985
 986                 if (rc)
 987                         RETURN(rc);
 988
 989                 qctl->qc_cmd = cmd;
 990         } else {
 991                 struct obd_quotactl *oqctl;
 992
 993                 OBD_ALLOC_PTR(oqctl);
 994                 if (oqctl == NULL)
 995                         RETURN(-ENOMEM);
 996
 997                 QCTL_COPY(oqctl, qctl);
 998                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
 999                 if (rc) {
1000                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
1001                                 oqctl->qc_cmd = Q_QUOTAOFF;
1002                                 obd_quotactl(sbi->ll_md_exp, oqctl);
1003                         }
1004                         OBD_FREE_PTR(oqctl);
1005                         RETURN(rc);
1006                 }
1007
1008                 QCTL_COPY(qctl, oqctl);
1009                 OBD_FREE_PTR(oqctl);
1010         }
1011
1012         RETURN(rc);
1013 }
1014
1015 static int ll_dir_ioctl(struct inode *inode, struct file *file,
1016                         unsigned int cmd, unsigned long arg)
1017 {
1018         struct ll_sb_info *sbi = ll_i2sbi(inode);
1019         struct obd_ioctl_data *data;
1020         int rc = 0;
1021         ENTRY;
1022
1023         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1024                inode->i_ino, inode->i_generation, inode, cmd);
1025
1026         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1027         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1028                 return -ENOTTY;
1029
1030         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1031         switch(cmd) {
1032         case FSFILT_IOC_GETFLAGS:
1033         case FSFILT_IOC_SETFLAGS:
1034                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1035         case FSFILT_IOC_GETVERSION_OLD:
1036         case FSFILT_IOC_GETVERSION:
1037                 RETURN(put_user(inode->i_generation, (int *)arg));
1038         /* We need to special case any other ioctls we want to handle,
1039          * to send them to the MDS/OST as appropriate and to properly
1040          * network encode the arg field.
1041         case FSFILT_IOC_SETVERSION_OLD:
1042         case FSFILT_IOC_SETVERSION:
1043         */
1044         case LL_IOC_GET_MDTIDX: {
1045                 int mdtidx;
1046
1047                 mdtidx = ll_get_mdt_idx(inode);
1048                 if (mdtidx < 0)
1049                         RETURN(mdtidx);
1050
1051                 if (put_user((int)mdtidx, (int*)arg))
1052                         RETURN(-EFAULT);
1053
1054                 return 0;
1055         }
1056         case IOC_MDC_LOOKUP: {
1057                 struct ptlrpc_request *request = NULL;
1058                 int namelen, len = 0;
1059                 char *buf = NULL;
1060                 char *filename;
1061                 struct md_op_data *op_data;
1062
1063                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1064                 if (rc)
1065                         RETURN(rc);
1066                 data = (void *)buf;
1067
1068                 filename = data->ioc_inlbuf1;
1069                 namelen = strlen(filename);
1070
1071                 if (namelen < 1) {
1072                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1073                         GOTO(out_free, rc = -EINVAL);
1074                 }
1075
1076                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1077                                              0, LUSTRE_OPC_ANY, NULL);
1078                 if (op_data == NULL)
1079                         GOTO(out_free, rc = -ENOMEM);
1080
1081                 op_data->op_valid = OBD_MD_FLID;
1082                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1083                 ll_finish_md_op_data(op_data);
1084                 if (rc < 0) {
1085                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1086                         GOTO(out_free, rc);
1087                 }
1088                 ptlrpc_req_finished(request);
1089                 EXIT;
1090 out_free:
1091                 obd_ioctl_freedata(buf, len);
1092                 return rc;
1093         }
1094         case LL_IOC_LOV_SETSTRIPE: {
1095                 struct lov_user_md_v3 lumv3;
1096                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1097                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1098                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1099
1100                 int set_default = 0;
1101
1102                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1103                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1104                         sizeof(lumv3p->lmm_objects[0]));
1105                 /* first try with v1 which is smaller than v3 */
1106                 if (cfs_copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1107                         RETURN(-EFAULT);
1108
1109                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1110                         if (cfs_copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1111                                 RETURN(-EFAULT);
1112                 }
1113
1114                 if (inode->i_sb->s_root == file->f_dentry)
1115                         set_default = 1;
1116
1117                 /* in v1 and v3 cases lumv1 points to data */
1118                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1119
1120                 RETURN(rc);
1121         }
1122         case LL_IOC_OBD_STATFS:
1123                 RETURN(ll_obd_statfs(inode, (void *)arg));
1124         case LL_IOC_LOV_GETSTRIPE:
1125         case LL_IOC_MDC_GETINFO:
1126         case IOC_MDC_GETFILEINFO:
1127         case IOC_MDC_GETFILESTRIPE: {
1128                 struct ptlrpc_request *request = NULL;
1129                 struct lov_user_md *lump;
1130                 struct lov_mds_md *lmm = NULL;
1131                 struct mdt_body *body;
1132                 char *filename = NULL;
1133                 int lmmsize;
1134
1135                 if (cmd == IOC_MDC_GETFILEINFO ||
1136                     cmd == IOC_MDC_GETFILESTRIPE) {
1137                         filename = getname((const char *)arg);
1138                         if (IS_ERR(filename))
1139                                 RETURN(PTR_ERR(filename));
1140
1141                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1142                                                       &lmmsize, &request);
1143                 } else {
1144                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1145                 }
1146
1147                 if (request) {
1148                         body = req_capsule_server_get(&request->rq_pill,
1149                                                       &RMF_MDT_BODY);
1150                         LASSERT(body != NULL);
1151                 } else {
1152                         GOTO(out_req, rc);
1153                 }
1154
1155                 if (rc < 0) {
1156                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1157                                                cmd == LL_IOC_MDC_GETINFO))
1158                                 GOTO(skip_lmm, rc = 0);
1159                         else
1160                                 GOTO(out_req, rc);
1161                 }
1162
1163                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1164                     cmd == LL_IOC_LOV_GETSTRIPE) {
1165                         lump = (struct lov_user_md *)arg;
1166                 } else {
1167                         struct lov_user_mds_data *lmdp;
1168                         lmdp = (struct lov_user_mds_data *)arg;
1169                         lump = &lmdp->lmd_lmm;
1170                 }
1171                 if (cfs_copy_to_user(lump, lmm, lmmsize)) {
1172                         if (cfs_copy_to_user(lump, lmm, sizeof(*lump)))
1173                                 GOTO(out_req, rc = -EFAULT);
1174                         rc = -EOVERFLOW;
1175                 }
1176         skip_lmm:
1177                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1178                         struct lov_user_mds_data *lmdp;
1179                         lstat_t st = { 0 };
1180
1181                         st.st_dev     = inode->i_sb->s_dev;
1182                         st.st_mode    = body->mode;
1183                         st.st_nlink   = body->nlink;
1184                         st.st_uid     = body->uid;
1185                         st.st_gid     = body->gid;
1186                         st.st_rdev    = body->rdev;
1187                         st.st_size    = body->size;
1188                         st.st_blksize = CFS_PAGE_SIZE;
1189                         st.st_blocks  = body->blocks;
1190                         st.st_atime   = body->atime;
1191                         st.st_mtime   = body->mtime;
1192                         st.st_ctime   = body->ctime;
1193                         st.st_ino     = inode->i_ino;
1194
1195                         lmdp = (struct lov_user_mds_data *)arg;
1196                         if (cfs_copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1197                                 GOTO(out_req, rc = -EFAULT);
1198                 }
1199
1200                 EXIT;
1201         out_req:
1202                 ptlrpc_req_finished(request);
1203                 if (filename)
1204                         putname(filename);
1205                 return rc;
1206         }
1207         case IOC_LOV_GETINFO: {
1208                 struct lov_user_mds_data *lumd;
1209                 struct lov_stripe_md *lsm;
1210                 struct lov_user_md *lum;
1211                 struct lov_mds_md *lmm;
1212                 int lmmsize;
1213                 lstat_t st;
1214
1215                 lumd = (struct lov_user_mds_data *)arg;
1216                 lum = &lumd->lmd_lmm;
1217
1218                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1219                 if (rc)
1220                         RETURN(rc);
1221
1222                 OBD_ALLOC_LARGE(lmm, lmmsize);
1223                 if (cfs_copy_from_user(lmm, lum, lmmsize))
1224                         GOTO(free_lmm, rc = -EFAULT);
1225
1226                 switch (lmm->lmm_magic) {
1227                 case LOV_USER_MAGIC_V1:
1228                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1229                                 break;
1230                         /* swab objects first so that stripes num will be sane */
1231                         lustre_swab_lov_user_md_objects(
1232                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1233                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1234                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1235                         break;
1236                 case LOV_USER_MAGIC_V3:
1237                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1238                                 break;
1239                         /* swab objects first so that stripes num will be sane */
1240                         lustre_swab_lov_user_md_objects(
1241                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1242                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1243                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1244                         break;
1245                 default:
1246                         GOTO(free_lmm, rc = -EINVAL);
1247                 }
1248
1249                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1250                 if (rc < 0)
1251                         GOTO(free_lmm, rc = -ENOMEM);
1252
1253                 /* Perform glimpse_size operation. */
1254                 memset(&st, 0, sizeof(st));
1255
1256                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1257                 if (rc)
1258                         GOTO(free_lsm, rc);
1259
1260                 if (cfs_copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1261                         GOTO(free_lsm, rc = -EFAULT);
1262
1263                 EXIT;
1264         free_lsm:
1265                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1266         free_lmm:
1267                 OBD_FREE_LARGE(lmm, lmmsize);
1268                 return rc;
1269         }
1270         case OBD_IOC_LLOG_CATINFO: {
1271                 struct ptlrpc_request *req = NULL;
1272                 char                  *buf = NULL;
1273                 char                  *str;
1274                 int                    len = 0;
1275
1276                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1277                 if (rc)
1278                         RETURN(rc);
1279                 data = (void *)buf;
1280
1281                 if (!data->ioc_inlbuf1) {
1282                         obd_ioctl_freedata(buf, len);
1283                         RETURN(-EINVAL);
1284                 }
1285
1286                 req = ptlrpc_request_alloc(sbi2mdc(sbi)->cl_import,
1287                                            &RQF_LLOG_CATINFO);
1288                 if (req == NULL)
1289                         GOTO(out_catinfo, rc = -ENOMEM);
1290
1291                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
1292                                      data->ioc_inllen1);
1293                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_CLIENT,
1294                                      data->ioc_inllen2);
1295
1296                 rc = ptlrpc_request_pack(req, LUSTRE_LOG_VERSION, LLOG_CATINFO);
1297                 if (rc) {
1298                         ptlrpc_request_free(req);
1299                         GOTO(out_catinfo, rc);
1300                 }
1301
1302                 str = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
1303                 memcpy(str, data->ioc_inlbuf1, data->ioc_inllen1);
1304                 if (data->ioc_inllen2) {
1305                         str = req_capsule_client_get(&req->rq_pill,
1306                                                      &RMF_STRING);
1307                         memcpy(str, data->ioc_inlbuf2, data->ioc_inllen2);
1308                 }
1309
1310                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_SERVER,
1311                                      data->ioc_plen1);
1312                 ptlrpc_request_set_replen(req);
1313
1314                 rc = ptlrpc_queue_wait(req);
1315                 if (!rc) {
1316                         str = req_capsule_server_get(&req->rq_pill,
1317                                                      &RMF_STRING);
1318                         if (cfs_copy_to_user(data->ioc_pbuf1, str,
1319                                              data->ioc_plen1))
1320                                 rc = -EFAULT;
1321                 }
1322                 ptlrpc_req_finished(req);
1323         out_catinfo:
1324                 obd_ioctl_freedata(buf, len);
1325                 RETURN(rc);
1326         }
1327         case OBD_IOC_QUOTACHECK: {
1328                 struct obd_quotactl *oqctl;
1329                 int error = 0;
1330
1331                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1332                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1333                         RETURN(-EPERM);
1334
1335                 OBD_ALLOC_PTR(oqctl);
1336                 if (!oqctl)
1337                         RETURN(-ENOMEM);
1338                 oqctl->qc_type = arg;
1339                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1340                 if (rc < 0) {
1341                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1342                         error = rc;
1343                 }
1344
1345                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1346                 if (rc < 0)
1347                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1348
1349                 OBD_FREE_PTR(oqctl);
1350                 return error ?: rc;
1351         }
1352         case OBD_IOC_POLL_QUOTACHECK: {
1353                 struct if_quotacheck *check;
1354
1355                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1356                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1357                         RETURN(-EPERM);
1358
1359                 OBD_ALLOC_PTR(check);
1360                 if (!check)
1361                         RETURN(-ENOMEM);
1362
1363                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1364                                    NULL);
1365                 if (rc) {
1366                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1367                         if (cfs_copy_to_user((void *)arg, check,
1368                                              sizeof(*check)))
1369                                 CDEBUG(D_QUOTA, "cfs_copy_to_user failed\n");
1370                         GOTO(out_poll, rc);
1371                 }
1372
1373                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1374                                    NULL);
1375                 if (rc) {
1376                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1377                         if (cfs_copy_to_user((void *)arg, check,
1378                                              sizeof(*check)))
1379                                 CDEBUG(D_QUOTA, "cfs_copy_to_user failed\n");
1380                         GOTO(out_poll, rc);
1381                 }
1382         out_poll:
1383                 OBD_FREE_PTR(check);
1384                 RETURN(rc);
1385         }
1386 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0)
1387         case LL_IOC_QUOTACTL_18: {
1388                 /* copy the old 1.x quota struct for internal use, then copy
1389                  * back into old format struct.  For 1.8 compatibility. */
1390                 struct if_quotactl_18 *qctl_18;
1391                 struct if_quotactl *qctl_20;
1392
1393                 OBD_ALLOC_PTR(qctl_18);
1394                 if (!qctl_18)
1395                         RETURN(-ENOMEM);
1396
1397                 OBD_ALLOC_PTR(qctl_20);
1398                 if (!qctl_20)
1399                         GOTO(out_quotactl_18, rc = -ENOMEM);
1400
1401                 if (cfs_copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1402                         GOTO(out_quotactl_20, rc = -ENOMEM);
1403
1404                 QCTL_COPY(qctl_20, qctl_18);
1405                 qctl_20->qc_idx = 0;
1406
1407                 /* XXX: dqb_valid was borrowed as a flag to mark that
1408                  *      only mds quota is wanted */
1409                 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1410                     qctl_18->qc_dqblk.dqb_valid) {
1411                         qctl_20->qc_valid = QC_MDTIDX;
1412                         qctl_20->qc_dqblk.dqb_valid = 0;
1413                 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1414                         qctl_20->qc_valid = QC_UUID;
1415                         qctl_20->obd_uuid = qctl_18->obd_uuid;
1416                 } else {
1417                         qctl_20->qc_valid = QC_GENERAL;
1418                 }
1419
1420                 rc = quotactl_ioctl(sbi, qctl_20);
1421
1422                 if (rc == 0) {
1423                         QCTL_COPY(qctl_18, qctl_20);
1424                         qctl_18->obd_uuid = qctl_20->obd_uuid;
1425
1426                         if (cfs_copy_to_user((void *)arg, qctl_18,
1427                                              sizeof(*qctl_18)))
1428                                 rc = -EFAULT;
1429                 }
1430
1431         out_quotactl_20:
1432                 OBD_FREE_PTR(qctl_20);
1433         out_quotactl_18:
1434                 OBD_FREE_PTR(qctl_18);
1435                 RETURN(rc);
1436         }
1437 #else
1438 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1439 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0) */
1440         case LL_IOC_QUOTACTL: {
1441                 struct if_quotactl *qctl;
1442
1443                 OBD_ALLOC_PTR(qctl);
1444                 if (!qctl)
1445                         RETURN(-ENOMEM);
1446
1447                 if (cfs_copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1448                         GOTO(out_quotactl, rc = -EFAULT);
1449
1450                 rc = quotactl_ioctl(sbi, qctl);
1451
1452                 if (rc == 0 && cfs_copy_to_user((void *)arg,qctl,sizeof(*qctl)))
1453                         rc = -EFAULT;
1454
1455         out_quotactl:
1456                 OBD_FREE_PTR(qctl);
1457                 RETURN(rc);
1458         }
1459         case OBD_IOC_GETNAME: {
1460                 struct obd_device *obd = class_exp2obd(sbi->ll_dt_exp);
1461                 if (!obd)
1462                         RETURN(-EFAULT);
1463                 if (cfs_copy_to_user((void *)arg, obd->obd_name,
1464                                      strlen(obd->obd_name) + 1))
1465                         RETURN (-EFAULT);
1466                 RETURN(0);
1467         }
1468         case LL_IOC_FLUSHCTX:
1469                 RETURN(ll_flush_ctx(inode));
1470 #ifdef CONFIG_FS_POSIX_ACL
1471         case LL_IOC_RMTACL: {
1472             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1473                 inode == inode->i_sb->s_root->d_inode) {
1474                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1475
1476                 LASSERT(fd != NULL);
1477                 rc = rct_add(&sbi->ll_rct, cfs_curproc_pid(), arg);
1478                 if (!rc)
1479                         fd->fd_flags |= LL_FILE_RMTACL;
1480                 RETURN(rc);
1481             } else
1482                 RETURN(0);
1483         }
1484 #endif
1485         case LL_IOC_GETOBDCOUNT: {
1486                 int count, vallen;
1487                 struct obd_export *exp;
1488
1489                 if (cfs_copy_from_user(&count, (int *)arg, sizeof(int)))
1490                         RETURN(-EFAULT);
1491
1492                 /* get ost count when count is zero, get mdt count otherwise */
1493                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1494                 vallen = sizeof(count);
1495                 rc = obd_get_info(exp, sizeof(KEY_TGT_COUNT), KEY_TGT_COUNT,
1496                                   &vallen, &count, NULL);
1497                 if (rc) {
1498                         CERROR("get target count failed: %d\n", rc);
1499                         RETURN(rc);
1500                 }
1501
1502                 if (cfs_copy_to_user((int *)arg, &count, sizeof(int)))
1503                         RETURN(-EFAULT);
1504
1505                 RETURN(0);
1506         }
1507         case LL_IOC_PATH2FID:
1508                 if (cfs_copy_to_user((void *)arg, ll_inode2fid(inode),
1509                                      sizeof(struct lu_fid)))
1510                         RETURN(-EFAULT);
1511                 RETURN(0);
1512         case LL_IOC_GET_CONNECT_FLAGS: {
1513                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1514         }
1515         case OBD_IOC_CHANGELOG_SEND:
1516         case OBD_IOC_CHANGELOG_CLEAR:
1517                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1518                                     sizeof(struct ioc_changelog));
1519                 RETURN(rc);
1520         case OBD_IOC_FID2PATH:
1521                 RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg));
1522         case LL_IOC_HSM_CT_START:
1523                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1524                                     sizeof(struct lustre_kernelcomm));
1525                 RETURN(rc);
1526
1527         default:
1528                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp,0,NULL,(void *)arg));
1529         }
1530 }
1531
1532 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1533 {
1534         struct inode *inode = file->f_mapping->host;
1535         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1536         struct ll_sb_info *sbi = ll_i2sbi(inode);
1537         int api32 = ll_need_32bit_api(sbi);
1538         loff_t ret = -EINVAL;
1539         ENTRY;
1540
1541         cfs_mutex_lock(&inode->i_mutex);
1542         switch (origin) {
1543                 case SEEK_SET:
1544                         break;
1545                 case SEEK_CUR:
1546                         offset += file->f_pos;
1547                         break;
1548                 case SEEK_END:
1549                         if (offset > 0)
1550                                 GOTO(out, ret);
1551                         if (api32)
1552                                 offset += LL_DIR_END_OFF_32BIT;
1553                         else
1554                                 offset += LL_DIR_END_OFF;
1555                         break;
1556                 default:
1557                         GOTO(out, ret);
1558         }
1559
1560         if (offset >= 0 &&
1561             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1562              (!api32 && offset <= LL_DIR_END_OFF))) {
1563                 if (offset != file->f_pos) {
1564                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1565                             (!api32 && offset == LL_DIR_END_OFF))
1566                                 fd->fd_dir.lfd_pos = MDS_DIR_END_OFF;
1567                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1568                                 fd->fd_dir.lfd_pos = offset << 32;
1569                         else
1570                                 fd->fd_dir.lfd_pos = offset;
1571                         file->f_pos = offset;
1572                         file->f_version = 0;
1573                 }
1574                 ret = offset;
1575         }
1576         GOTO(out, ret);
1577
1578 out:
1579         cfs_mutex_unlock(&inode->i_mutex);
1580         return ret;
1581 }
1582
1583 int ll_dir_open(struct inode *inode, struct file *file)
1584 {
1585         ENTRY;
1586         RETURN(ll_file_open(inode, file));
1587 }
1588
1589 int ll_dir_release(struct inode *inode, struct file *file)
1590 {
1591         ENTRY;
1592         RETURN(ll_file_release(inode, file));
1593 }
1594
1595 struct file_operations ll_dir_operations = {
1596         .llseek   = ll_dir_seek,
1597         .open     = ll_dir_open,
1598         .release  = ll_dir_release,
1599         .read     = generic_read_dir,
1600         .readdir  = ll_readdir,
1601         .ioctl    = ll_dir_ioctl,
1602         .fsync    = ll_fsync
1603 };