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, Whamcloud, Inc.
  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 <linux/smp_lock.h>
  46 #include <asm/uaccess.h>
  47 #include <linux/buffer_head.h>   // for wait_on_buffer
  48 #include <linux/pagevec.h>
  49
  50 #define DEBUG_SUBSYSTEM S_LLITE
  51
  52 #include <obd_support.h>
  53 #include <obd_class.h>
  54 #include <lustre_lib.h>
  55 #include <lustre/lustre_idl.h>
  56 #include <lustre_lite.h>
  57 #include <lustre_dlm.h>
  58 #include <lustre_fid.h>
  59 #include "llite_internal.h"
  60
  61 #ifndef HAVE_PAGE_CHECKED
  62 #ifdef HAVE_PG_FS_MISC
  63 #define PageChecked(page)        test_bit(PG_fs_misc, &(page)->flags)
  64 #define SetPageChecked(page)     set_bit(PG_fs_misc, &(page)->flags)
  65 #else
  66 #error PageChecked or PageFsMisc not defined in kernel
  67 #endif
  68 #endif
  69
  70 /*
  71  * (new) readdir implementation overview.
  72  *
  73  * Original lustre readdir implementation cached exact copy of raw directory
  74  * pages on the client. These pages were indexed in client page cache by
  75  * logical offset in the directory file. This design, while very simple and
  76  * intuitive had some inherent problems:
  77  *
  78  *     . it implies that byte offset to the directory entry serves as a
  79  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  80  *     ext3/htree directory entries may move due to splits, and more
  81  *     importantly,
  82  *
  83  *     . it is incompatible with the design of split directories for cmd3,
  84  *     that assumes that names are distributed across nodes based on their
  85  *     hash, and so readdir should be done in hash order.
  86  *
  87  * New readdir implementation does readdir in hash order, and uses hash of a
  88  * file name as a telldir/seekdir cookie. This led to number of complications:
  89  *
  90  *     . hash is not unique, so it cannot be used to index cached directory
  91  *     pages on the client (note, that it requires a whole pageful of hash
  92  *     collided entries to cause two pages to have identical hashes);
  93  *
  94  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  95  *     entry cookie. ext3/htree has the same problem and lustre implementation
  96  *     mimics their solution: seekdir(hash) positions directory at the first
  97  *     entry with the given hash.
  98  *
  99  * Client side.
 100  *
 101  * 0. caching
 102  *
 103  * Client caches directory pages using hash of the first entry as an index. As
 104  * noted above hash is not unique, so this solution doesn't work as is:
 105  * special processing is needed for "page hash chains" (i.e., sequences of
 106  * pages filled with entries all having the same hash value).
 107  *
 108  * First, such chains have to be detected. To this end, server returns to the
 109  * client the hash of the first entry on the page next to one returned. When
 110  * client detects that this hash is the same as hash of the first entry on the
 111  * returned page, page hash collision has to be handled. Pages in the
 112  * hash chain, except first one, are termed "overflow pages".
 113  *
 114  * Solution to index uniqueness problem is to not cache overflow
 115  * pages. Instead, when page hash collision is detected, all overflow pages
 116  * from emerging chain are immediately requested from the server and placed in
 117  * a special data structure (struct ll_dir_chain). This data structure is used
 118  * by ll_readdir() to process entries from overflow pages. When readdir
 119  * invocation finishes, overflow pages are discarded. If page hash collision
 120  * chain weren't completely processed, next call to readdir will again detect
 121  * page hash collision, again read overflow pages in, process next portion of
 122  * entries and again discard the pages. This is not as wasteful as it looks,
 123  * because, given reasonable hash, page hash collisions are extremely rare.
 124  *
 125  * 1. directory positioning
 126  *
 127  * When seekdir(hash) is called, original
 128  *
 129  *
 130  *
 131  *
 132  *
 133  *
 134  *
 135  *
 136  * Server.
 137  *
 138  * identification of and access to overflow pages
 139  *
 140  * page format
 141  *
 142  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 143  * a header lu_dirpage which describes the start/end hash, and whether this
 144  * page is empty (contains no dir entry) or hash collide with next page.
 145  * After client receives reply, several pages will be integrated into dir page
 146  * in CFS_PAGE_SIZE (if CFS_PAGE_SIZE greater than LU_PAGE_SIZE), and the
 147  * lu_dirpage for this integrated page will be adjusted.
 148  *
 149  */
 150
 151 /* returns the page unlocked, but with a reference */
 152 static int ll_dir_readpage(struct file *file, struct page *page0)
 153 {
 154         struct inode *inode = page0->mapping->host;
 155         int hash64 = ll_i2sbi(inode)->ll_flags & LL_SBI_64BIT_HASH;
 156         struct obd_export *exp = ll_i2sbi(inode)->ll_md_exp;
 157         struct ptlrpc_request *request;
 158         struct mdt_body *body;
 159         struct md_op_data *op_data;
 160         __u64 hash;
 161         struct page **page_pool;
 162         struct page *page;
 163 #ifndef HAVE_ADD_TO_PAGE_CACHE_LRU
 164         struct pagevec lru_pvec;
 165 #endif
 166         struct lu_dirpage *dp;
 167         int max_pages = ll_i2sbi(inode)->ll_md_brw_size >> CFS_PAGE_SHIFT;
 168         int nrdpgs = 0; /* number of pages read actually */
 169         int npages;
 170         int i;
 171         int rc;
 172         ENTRY;
 173
 174         if (file) {
 175                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
 176
 177                 hash = fd->fd_dir.lfd_next;
 178         } else {
 179                 hash = ll_i2info(inode)->lli_sa_pos;
 180         }
 181         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) hash "LPU64"\n",
 182                inode->i_ino, inode->i_generation, inode, hash);
 183
 184         LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
 185
 186         OBD_ALLOC(page_pool, sizeof(page) * max_pages);
 187         if (page_pool != NULL) {
 188                 page_pool[0] = page0;
 189         } else {
 190                 page_pool = &page0;
 191                 max_pages = 1;
 192         }
 193         for (npages = 1; npages < max_pages; npages++) {
 194                 page = page_cache_alloc_cold(inode->i_mapping);
 195                 if (!page)
 196                         break;
 197                 page_pool[npages] = page;
 198         }
 199
 200         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 201                                      LUSTRE_OPC_ANY, NULL);
 202         op_data->op_npages = npages;
 203         op_data->op_offset = hash;
 204         rc = md_readpage(exp, op_data, page_pool, &request);
 205         ll_finish_md_op_data(op_data);
 206         if (rc == 0) {
 207                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 208                 /* Checked by mdc_readpage() */
 209                 LASSERT(body != NULL);
 210
 211                 if (body->valid & OBD_MD_FLSIZE)
 212                         cl_isize_write(inode, body->size);
 213
 214                 nrdpgs = (request->rq_bulk->bd_nob_transferred+CFS_PAGE_SIZE-1)
 215                          >> CFS_PAGE_SHIFT;
 216                 SetPageUptodate(page0);
 217         }
 218         unlock_page(page0);
 219         ptlrpc_req_finished(request);
 220
 221         CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages);
 222
 223         ll_pagevec_init(&lru_pvec, 0);
 224         for (i = 1; i < npages; i++) {
 225                 unsigned long offset;
 226                 int ret;
 227
 228                 page = page_pool[i];
 229
 230                 if (rc < 0 || i >= nrdpgs) {
 231                         page_cache_release(page);
 232                         continue;
 233                 }
 234
 235                 SetPageUptodate(page);
 236
 237                 dp = cfs_kmap(page);
 238                 hash = le64_to_cpu(dp->ldp_hash_start);
 239                 cfs_kunmap(page);
 240
 241                 offset = hash_x_index(hash, hash64);
 242
 243                 prefetchw(&page->flags);
 244                 ret = ll_add_to_page_cache_lru(page, inode->i_mapping, offset,
 245                                                GFP_KERNEL);
 246                 if (ret == 0) {
 247                         unlock_page(page);
 248                         page_cache_get(page);
 249                         if (ll_pagevec_add(&lru_pvec, page) == 0)
 250                                 ll_pagevec_lru_add_file(&lru_pvec);
 251                 } else {
 252                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
 253                                " %d\n", offset, ret);
 254                 }
 255                 page_cache_release(page);
 256         }
 257         ll_pagevec_lru_add_file(&lru_pvec);
 258
 259         if (page_pool != &page0)
 260                 OBD_FREE(page_pool, sizeof(struct page *) * max_pages);
 261         EXIT;
 262         return rc;
 263 }
 264
 265 #ifndef MS_HAS_NEW_AOPS
 266 struct address_space_operations ll_dir_aops = {
 267         .readpage  = ll_dir_readpage,
 268 };
 269 #else
 270 struct address_space_operations_ext ll_dir_aops = {
 271         .orig_aops.readpage  = ll_dir_readpage,
 272 };
 273 #endif
 274
 275 static void ll_check_page(struct inode *dir, struct page *page)
 276 {
 277         /* XXX: check page format later */
 278         SetPageChecked(page);
 279 }
 280
 281 void ll_release_page(struct page *page, int remove)
 282 {
 283         kunmap(page);
 284         if (remove) {
 285                 lock_page(page);
 286                 if (likely(page->mapping != NULL))
 287                         truncate_complete_page(page->mapping, page);
 288                 unlock_page(page);
 289         }
 290         page_cache_release(page);
 291 }
 292
 293 /*
 294  * Find, kmap and return page that contains given hash.
 295  */
 296 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
 297                                        __u64 *start, __u64 *end)
 298 {
 299         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 300         struct address_space *mapping = dir->i_mapping;
 301         /*
 302          * Complement of hash is used as an index so that
 303          * radix_tree_gang_lookup() can be used to find a page with starting
 304          * hash _smaller_ than one we are looking for.
 305          */
 306         unsigned long offset = hash_x_index(*hash, hash64);
 307         struct page *page;
 308         int found;
 309
 310         TREE_READ_LOCK_IRQ(mapping);
 311         found = radix_tree_gang_lookup(&mapping->page_tree,
 312                                        (void **)&page, offset, 1);
 313         if (found > 0) {
 314                 struct lu_dirpage *dp;
 315
 316                 page_cache_get(page);
 317                 TREE_READ_UNLOCK_IRQ(mapping);
 318                 /*
 319                  * In contrast to find_lock_page() we are sure that directory
 320                  * page cannot be truncated (while DLM lock is held) and,
 321                  * hence, can avoid restart.
 322                  *
 323                  * In fact, page cannot be locked here at all, because
 324                  * ll_dir_readpage() does synchronous io.
 325                  */
 326                 wait_on_page(page);
 327                 if (PageUptodate(page)) {
 328                         dp = cfs_kmap(page);
 329                         if (BITS_PER_LONG == 32 && hash64) {
 330                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 331                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 332                                 *hash  = *hash >> 32;
 333                         } else {
 334                                 *start = le64_to_cpu(dp->ldp_hash_start);
 335                                 *end   = le64_to_cpu(dp->ldp_hash_end);
 336                         }
 337                         LASSERTF(*start <= *hash, "start = "LPX64",end = "
 338                                  LPX64",hash = "LPX64"\n", *start, *end, *hash);
 339                         CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash "LPU64"\n",
 340                                offset, *start, *end, *hash);
 341                         if (*hash > *end) {
 342                                 ll_release_page(page, 0);
 343                                 page = NULL;
 344                         } else if (*end != *start && *hash == *end) {
 345                                 /*
 346                                  * upon hash collision, remove this page,
 347                                  * otherwise put page reference, and
 348                                  * ll_get_dir_page() will issue RPC to fetch
 349                                  * the page we want.
 350                                  */
 351                                 ll_release_page(page,
 352                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
 353                                 page = NULL;
 354                         }
 355                 } else {
 356                         page_cache_release(page);
 357                         page = ERR_PTR(-EIO);
 358                 }
 359
 360         } else {
 361                 TREE_READ_UNLOCK_IRQ(mapping);
 362                 page = NULL;
 363         }
 364         return page;
 365 }
 366
 367 struct page *ll_get_dir_page(struct file *filp, struct inode *dir, __u64 hash,
 368                              struct ll_dir_chain *chain)
 369 {
 370         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 371         struct address_space *mapping = dir->i_mapping;
 372         struct lustre_handle lockh;
 373         struct lu_dirpage *dp;
 374         struct page *page;
 375         ldlm_mode_t mode;
 376         int rc;
 377         __u64 start = 0;
 378         __u64 end = 0;
 379         __u64 lhash = hash;
 380         struct ll_inode_info *lli = ll_i2info(dir);
 381         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 382
 383         mode = LCK_PR;
 384         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 385                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
 386         if (!rc) {
 387                 struct ldlm_enqueue_info einfo = { LDLM_IBITS, mode,
 388                        ll_md_blocking_ast, ldlm_completion_ast,
 389                        NULL, NULL, dir };
 390                 struct lookup_intent it = { .it_op = IT_READDIR };
 391                 struct ptlrpc_request *request;
 392                 struct md_op_data *op_data;
 393
 394                 op_data = ll_prep_md_op_data(NULL, dir, NULL, NULL, 0, 0,
 395                                              LUSTRE_OPC_ANY, NULL);
 396                 if (IS_ERR(op_data))
 397                         return (void *)op_data;
 398
 399                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
 400                                 op_data, &lockh, NULL, 0, NULL, 0);
 401
 402                 ll_finish_md_op_data(op_data);
 403
 404                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 405                 if (request)
 406                         ptlrpc_req_finished(request);
 407                 if (rc < 0) {
 408                         CERROR("lock enqueue: "DFID" at "LPU64": rc %d\n",
 409                                PFID(ll_inode2fid(dir)), hash, rc);
 410                         return ERR_PTR(rc);
 411                 }
 412         } else {
 413                 /* for cross-ref object, l_ast_data of the lock may not be set,
 414                  * we reset it here */
 415                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
 416                                  dir, NULL);
 417         }
 418         ldlm_lock_dump_handle(D_OTHER, &lockh);
 419
 420         cfs_mutex_lock(&lli->lli_readdir_mutex);
 421         page = ll_dir_page_locate(dir, &lhash, &start, &end);
 422         if (IS_ERR(page)) {
 423                 CERROR("dir page locate: "DFID" at "LPU64": rc %ld\n",
 424                        PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
 425                 GOTO(out_unlock, page);
 426         } else if (page != NULL) {
 427                 /*
 428                  * XXX nikita: not entirely correct handling of a corner case:
 429                  * suppose hash chain of entries with hash value HASH crosses
 430                  * border between pages P0 and P1. First both P0 and P1 are
 431                  * cached, seekdir() is called for some entry from the P0 part
 432                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 433                  * happens and finds P1, as it starts with matching hash
 434                  * value. Remaining entries from P0 part of the chain are
 435                  * skipped. (Is that really a bug?)
 436                  *
 437                  * Possible solutions: 0. don't cache P1 is such case, handle
 438                  * it as an "overflow" page. 1. invalidate all pages at
 439                  * once. 2. use HASH|1 as an index for P1.
 440                  */
 441                 GOTO(hash_collision, page);
 442         }
 443
 444         page = read_cache_page(mapping, hash_x_index(hash, hash64),
 445                                (filler_t*)mapping->a_ops->readpage, filp);
 446         if (IS_ERR(page)) {
 447                 CERROR("read cache page: "DFID" at "LPU64": rc %ld\n",
 448                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 449                 GOTO(out_unlock, page);
 450         }
 451
 452         wait_on_page(page);
 453         (void)kmap(page);
 454         if (!PageUptodate(page)) {
 455                 CERROR("page not updated: "DFID" at "LPU64": rc %d\n",
 456                        PFID(ll_inode2fid(dir)), hash, -5);
 457                 goto fail;
 458         }
 459         if (!PageChecked(page))
 460                 ll_check_page(dir, page);
 461         if (PageError(page)) {
 462                 CERROR("page error: "DFID" at "LPU64": rc %d\n",
 463                        PFID(ll_inode2fid(dir)), hash, -5);
 464                 goto fail;
 465         }
 466 hash_collision:
 467         dp = page_address(page);
 468         if (BITS_PER_LONG == 32 && hash64) {
 469                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 470                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 471                 lhash = hash >> 32;
 472         } else {
 473                 start = le64_to_cpu(dp->ldp_hash_start);
 474                 end   = le64_to_cpu(dp->ldp_hash_end);
 475                 lhash = hash;
 476         }
 477         if (end == start) {
 478                 LASSERT(start == lhash);
 479                 CWARN("Page-wide hash collision: "LPU64"\n", end);
 480                 if (BITS_PER_LONG == 32 && hash64)
 481                         CWARN("Real page-wide hash collision at ["LPU64" "LPU64
 482                               "] with hash "LPU64"\n",
 483                               le64_to_cpu(dp->ldp_hash_start),
 484                               le64_to_cpu(dp->ldp_hash_end), hash);
 485                 /*
 486                  * Fetch whole overflow chain...
 487                  *
 488                  * XXX not yet.
 489                  */
 490                 goto fail;
 491         }
 492 out_unlock:
 493         cfs_mutex_unlock(&lli->lli_readdir_mutex);
 494         ldlm_lock_decref(&lockh, mode);
 495         return page;
 496
 497 fail:
 498         ll_release_page(page, 1);
 499         page = ERR_PTR(-EIO);
 500         goto out_unlock;
 501 }
 502
 503 int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 504 {
 505         struct inode         *inode      = filp->f_dentry->d_inode;
 506         struct ll_inode_info *info       = ll_i2info(inode);
 507         struct ll_sb_info    *sbi        = ll_i2sbi(inode);
 508         struct ll_file_data  *fd         = LUSTRE_FPRIVATE(filp);
 509         __u64                 pos        = fd->fd_dir.lfd_pos;
 510         int                   api32      = ll_need_32bit_api(sbi);
 511         int                   hash64     = sbi->ll_flags & LL_SBI_64BIT_HASH;
 512         struct page          *page;
 513         struct ll_dir_chain   chain;
 514         int                   done;
 515         int                   rc;
 516         ENTRY;
 517
 518         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu 32bit_api %d\n",
 519                inode->i_ino, inode->i_generation, inode,
 520                (unsigned long)pos, i_size_read(inode), api32);
 521
 522         if (pos == MDS_DIR_END_OFF)
 523                 /*
 524                  * end-of-file.
 525                  */
 526                 GOTO(out, rc = 0);
 527
 528         rc    = 0;
 529         done  = 0;
 530         ll_dir_chain_init(&chain);
 531
 532         fd->fd_dir.lfd_next = pos;
 533         page = ll_get_dir_page(filp, inode, pos, &chain);
 534
 535         while (rc == 0 && !done) {
 536                 struct lu_dirpage *dp;
 537                 struct lu_dirent  *ent;
 538
 539                 if (!IS_ERR(page)) {
 540                         /*
 541                          * If page is empty (end of directory is reached),
 542                          * use this value.
 543                          */
 544                         __u64 hash = MDS_DIR_END_OFF;
 545                         __u64 next;
 546
 547                         dp = page_address(page);
 548                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 549                              ent = lu_dirent_next(ent)) {
 550                                 __u16          type;
 551                                 int            namelen;
 552                                 struct lu_fid  fid;
 553                                 __u64          lhash;
 554                                 __u64          ino;
 555
 556                                 /*
 557                                  * XXX: implement correct swabbing here.
 558                                  */
 559
 560                                 hash = le64_to_cpu(ent->lde_hash);
 561                                 if (hash < pos)
 562                                         /*
 563                                          * Skip until we find target hash
 564                                          * value.
 565                                          */
 566                                         continue;
 567
 568                                 namelen = le16_to_cpu(ent->lde_namelen);
 569                                 if (namelen == 0)
 570                                         /*
 571                                          * Skip dummy record.
 572                                          */
 573                                         continue;
 574
 575                                 if (api32 && hash64)
 576                                         lhash = hash >> 32;
 577                                 else
 578                                         lhash = hash;
 579                                 fid_le_to_cpu(&fid, &ent->lde_fid);
 580                                 ino = cl_fid_build_ino(&fid, api32);
 581                                 type = ll_dirent_type_get(ent);
 582                                 /* For 'll_nfs_get_name_filldir()', it will try
 583                                  * to access the 'ent' through its 'lde_name',
 584                                  * so the parameter 'name' for 'filldir()' must
 585                                  * be part of the 'ent'. */
 586                                 done = filldir(cookie, ent->lde_name, namelen,
 587                                                lhash, ino, type);
 588                         }
 589                         next = le64_to_cpu(dp->ldp_hash_end);
 590                         if (!done) {
 591                                 pos = next;
 592                                 if (pos == MDS_DIR_END_OFF) {
 593                                         /*
 594                                          * End of directory reached.
 595                                          */
 596                                         done = 1;
 597                                         ll_release_page(page, 0);
 598                                 } else if (1 /* chain is exhausted*/) {
 599                                         /*
 600                                          * Normal case: continue to the next
 601                                          * page.
 602                                          */
 603                                         ll_release_page(page,
 604                                             le32_to_cpu(dp->ldp_flags) &
 605                                                         LDF_COLLIDE);
 606                                         fd->fd_dir.lfd_next = pos;
 607                                         page = ll_get_dir_page(filp, inode, pos,
 608                                                                &chain);
 609                                 } else {
 610                                         /*
 611                                          * go into overflow page.
 612                                          */
 613                                         LASSERT(le32_to_cpu(dp->ldp_flags) &
 614                                                 LDF_COLLIDE);
 615                                         ll_release_page(page, 1);
 616                                 }
 617                         } else {
 618                                 pos = hash;
 619                                 ll_release_page(page, 0);
 620                         }
 621                 } else {
 622                         rc = PTR_ERR(page);
 623                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 624                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 625                 }
 626         }
 627
 628         fd->fd_dir.lfd_pos = pos;
 629         if (pos == MDS_DIR_END_OFF) {
 630                 if (api32)
 631                         filp->f_pos = LL_DIR_END_OFF_32BIT;
 632                 else
 633                         filp->f_pos = LL_DIR_END_OFF;
 634         } else {
 635                 if (api32 && hash64)
 636                         filp->f_pos = pos >> 32;
 637                 else
 638                         filp->f_pos = pos;
 639         }
 640         filp->f_version = inode->i_version;
 641         touch_atime(filp->f_vfsmnt, filp->f_dentry);
 642
 643         ll_dir_chain_fini(&chain);
 644
 645 out:
 646         if (!rc)
 647                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 648
 649         RETURN(rc);
 650 }
 651
 652 int ll_send_mgc_param(struct obd_export *mgc, char *string)
 653 {
 654         struct mgs_send_param *msp;
 655         int rc = 0;
 656
 657         OBD_ALLOC_PTR(msp);
 658         if (!msp)
 659                 return -ENOMEM;
 660
 661         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 662         rc = obd_set_info_async(mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 663                                 sizeof(struct mgs_send_param), msp, NULL);
 664         if (rc)
 665                 CERROR("Failed to set parameter: %d\n", rc);
 666         OBD_FREE_PTR(msp);
 667
 668         return rc;
 669 }
 670
 671 char *ll_get_fsname(struct inode *inode)
 672 {
 673         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 674         char *ptr, *fsname;
 675         int len;
 676
 677         OBD_ALLOC(fsname, MGS_PARAM_MAXLEN);
 678         len = strlen(lsi->lsi_lmd->lmd_profile);
 679         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
 680         if (ptr && (strcmp(ptr, "-client") == 0))
 681                 len -= 7;
 682         strncpy(fsname, lsi->lsi_lmd->lmd_profile, len);
 683         fsname[len] = '\0';
 684
 685         return fsname;
 686 }
 687
 688 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 689                      int set_default)
 690 {
 691         struct ll_sb_info *sbi = ll_i2sbi(inode);
 692         struct md_op_data *op_data;
 693         struct ptlrpc_request *req = NULL;
 694         int rc = 0;
 695         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 696         struct obd_device *mgc = lsi->lsi_mgc;
 697         char *fsname = NULL, *param = NULL;
 698         int lum_size;
 699
 700         if (lump != NULL) {
 701                 /*
 702                  * This is coming from userspace, so should be in
 703                  * local endian.  But the MDS would like it in little
 704                  * endian, so we swab it before we send it.
 705                  */
 706                 switch (lump->lmm_magic) {
 707                 case LOV_USER_MAGIC_V1: {
 708                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 709                                 lustre_swab_lov_user_md_v1(lump);
 710                         lum_size = sizeof(struct lov_user_md_v1);
 711                         break;
 712                         }
 713                 case LOV_USER_MAGIC_V3: {
 714                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 715                                 lustre_swab_lov_user_md_v3(
 716                                         (struct lov_user_md_v3 *)lump);
 717                         lum_size = sizeof(struct lov_user_md_v3);
 718                         break;
 719                         }
 720                 default: {
 721                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 722                                         " %#08x != %#08x nor %#08x\n",
 723                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 724                                         LOV_USER_MAGIC_V3);
 725                         RETURN(-EINVAL);
 726                         }
 727                }
 728         } else {
 729                 lum_size = sizeof(struct lov_user_md_v1);
 730         }
 731
 732         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 733                                      LUSTRE_OPC_ANY, NULL);
 734         if (IS_ERR(op_data))
 735                 RETURN(PTR_ERR(op_data));
 736
 737         /* swabbing is done in lov_setstripe() on server side */
 738         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 739                         NULL, 0, &req, NULL);
 740         ll_finish_md_op_data(op_data);
 741         ptlrpc_req_finished(req);
 742         if (rc) {
 743                 if (rc != -EPERM && rc != -EACCES)
 744                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 745         }
 746
 747         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 748          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 749          need the make the distiction between the 2 versions */
 750         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 751                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 752
 753                 /* Get fsname and assume devname to be -MDT0000. */
 754                 fsname = ll_get_fsname(inode);
 755                 /* Set root stripesize */
 756                 sprintf(param, "%s-MDT0000.lov.stripesize=%u", fsname,
 757                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 758                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 759                 if (rc)
 760                         goto end;
 761
 762                 /* Set root stripecount */
 763                 sprintf(param, "%s-MDT0000.lov.stripecount=%hd", fsname,
 764                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 765                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 766                 if (rc)
 767                         goto end;
 768
 769                 /* Set root stripeoffset */
 770                 sprintf(param, "%s-MDT0000.lov.stripeoffset=%hd", fsname,
 771                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 772                         (typeof(lump->lmm_stripe_offset))(-1));
 773                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 774                 if (rc)
 775                         goto end;
 776 end:
 777                 if (fsname)
 778                         OBD_FREE(fsname, MGS_PARAM_MAXLEN);
 779                 if (param)
 780                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 781         }
 782         return rc;
 783 }
 784
 785 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 786                      int *lmm_size, struct ptlrpc_request **request)
 787 {
 788         struct ll_sb_info *sbi = ll_i2sbi(inode);
 789         struct mdt_body   *body;
 790         struct lov_mds_md *lmm = NULL;
 791         struct ptlrpc_request *req = NULL;
 792         int rc, lmmsize;
 793         struct md_op_data *op_data;
 794
 795         rc = ll_get_max_mdsize(sbi, &lmmsize);
 796         if (rc)
 797                 RETURN(rc);
 798
 799         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 800                                      0, lmmsize, LUSTRE_OPC_ANY,
 801                                      NULL);
 802         if (IS_ERR(op_data))
 803                 RETURN(PTR_ERR(op_data));
 804
 805         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 806         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 807         ll_finish_md_op_data(op_data);
 808         if (rc < 0) {
 809                 CDEBUG(D_INFO, "md_getattr failed on inode "
 810                        "%lu/%u: rc %d\n", inode->i_ino,
 811                        inode->i_generation, rc);
 812                 GOTO(out, rc);
 813         }
 814
 815         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 816         LASSERT(body != NULL);
 817
 818         lmmsize = body->eadatasize;
 819
 820         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 821             lmmsize == 0) {
 822                 GOTO(out, rc = -ENODATA);
 823         }
 824
 825         lmm = req_capsule_server_sized_get(&req->rq_pill,
 826                                            &RMF_MDT_MD, lmmsize);
 827         LASSERT(lmm != NULL);
 828
 829         /*
 830          * This is coming from the MDS, so is probably in
 831          * little endian.  We convert it to host endian before
 832          * passing it to userspace.
 833          */
 834         /* We don't swab objects for directories */
 835         switch (le32_to_cpu(lmm->lmm_magic)) {
 836         case LOV_MAGIC_V1:
 837                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 838                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 839                 break;
 840         case LOV_MAGIC_V3:
 841                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 842                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 843                 break;
 844         default:
 845                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 846                 rc = -EPROTO;
 847         }
 848 out:
 849         *lmmp = lmm;
 850         *lmm_size = lmmsize;
 851         *request = req;
 852         return rc;
 853 }
 854
 855 /*
 856  *  Get MDT index for the inode.
 857  */
 858 int ll_get_mdt_idx(struct inode *inode)
 859 {
 860         struct ll_sb_info *sbi = ll_i2sbi(inode);
 861         struct md_op_data *op_data;
 862         int rc, mdtidx;
 863         ENTRY;
 864
 865         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 866                                      0, LUSTRE_OPC_ANY, NULL);
 867         if (IS_ERR(op_data))
 868                 RETURN(PTR_ERR(op_data));
 869
 870         op_data->op_valid |= OBD_MD_MDTIDX;
 871         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 872         mdtidx = op_data->op_mds;
 873         ll_finish_md_op_data(op_data);
 874         if (rc < 0) {
 875                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 876                 RETURN(rc);
 877         }
 878         return mdtidx;
 879 }
 880
 881 static int copy_and_ioctl(int cmd, struct obd_export *exp, void *data, int len)
 882 {
 883         void *ptr;
 884         int rc;
 885
 886         OBD_ALLOC(ptr, len);
 887         if (ptr == NULL)
 888                 return -ENOMEM;
 889         if (cfs_copy_from_user(ptr, data, len)) {
 890                 OBD_FREE(ptr, len);
 891                 return -EFAULT;
 892         }
 893         rc = obd_iocontrol(cmd, exp, len, data, NULL);
 894         OBD_FREE(ptr, len);
 895         return rc;
 896 }
 897
 898 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
 899 {
 900         int cmd = qctl->qc_cmd;
 901         int type = qctl->qc_type;
 902         int id = qctl->qc_id;
 903         int valid = qctl->qc_valid;
 904         int rc = 0;
 905         ENTRY;
 906
 907         switch (cmd) {
 908         case LUSTRE_Q_INVALIDATE:
 909         case LUSTRE_Q_FINVALIDATE:
 910         case Q_QUOTAON:
 911         case Q_QUOTAOFF:
 912         case Q_SETQUOTA:
 913         case Q_SETINFO:
 914                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 915                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
 916                         RETURN(-EPERM);
 917                 break;
 918         case Q_GETQUOTA:
 919                 if (((type == USRQUOTA && cfs_curproc_euid() != id) ||
 920                      (type == GRPQUOTA && !in_egroup_p(id))) &&
 921                     (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 922                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
 923                         RETURN(-EPERM);
 924                 break;
 925         case Q_GETINFO:
 926                 break;
 927         default:
 928                 CERROR("unsupported quotactl op: %#x\n", cmd);
 929                 RETURN(-ENOTTY);
 930         }
 931
 932         if (valid != QC_GENERAL) {
 933                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 934                         RETURN(-EOPNOTSUPP);
 935
 936                 if (cmd == Q_GETINFO)
 937                         qctl->qc_cmd = Q_GETOINFO;
 938                 else if (cmd == Q_GETQUOTA)
 939                         qctl->qc_cmd = Q_GETOQUOTA;
 940                 else
 941                         RETURN(-EINVAL);
 942
 943                 switch (valid) {
 944                 case QC_MDTIDX:
 945                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 946                                            sizeof(*qctl), qctl, NULL);
 947                         break;
 948                 case QC_OSTIDX:
 949                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
 950                                            sizeof(*qctl), qctl, NULL);
 951                         break;
 952                 case QC_UUID:
 953                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 954                                            sizeof(*qctl), qctl, NULL);
 955                         if (rc == -EAGAIN)
 956                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
 957                                                    sbi->ll_dt_exp,
 958                                                    sizeof(*qctl), qctl, NULL);
 959                         break;
 960                 default:
 961                         rc = -EINVAL;
 962                         break;
 963                 }
 964
 965                 if (rc)
 966                         RETURN(rc);
 967
 968                 qctl->qc_cmd = cmd;
 969         } else {
 970                 struct obd_quotactl *oqctl;
 971
 972                 OBD_ALLOC_PTR(oqctl);
 973                 if (oqctl == NULL)
 974                         RETURN(-ENOMEM);
 975
 976                 QCTL_COPY(oqctl, qctl);
 977                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
 978                 if (rc) {
 979                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
 980                                 oqctl->qc_cmd = Q_QUOTAOFF;
 981                                 obd_quotactl(sbi->ll_md_exp, oqctl);
 982                         }
 983                         OBD_FREE_PTR(oqctl);
 984                         RETURN(rc);
 985                 }
 986                 /* If QIF_SPACE is not set, client should collect the
 987                  * space usage from OSSs by itself */
 988                 if (cmd == Q_GETQUOTA &&
 989                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
 990                     !oqctl->qc_dqblk.dqb_curspace) {
 991                         struct obd_quotactl *oqctl_tmp;
 992
 993                         OBD_ALLOC_PTR(oqctl_tmp);
 994                         if (oqctl_tmp == NULL)
 995                                 GOTO(out, rc = -ENOMEM);
 996
 997                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
 998                         oqctl_tmp->qc_id = oqctl->qc_id;
 999                         oqctl_tmp->qc_type = oqctl->qc_type;
1000
1001                         /* collect space usage from OSTs */
1002                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1003                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1004                         if (!rc || rc == -EREMOTEIO) {
1005                                 oqctl->qc_dqblk.dqb_curspace =
1006                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1007                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1008                         }
1009
1010                         /* collect space & inode usage from MDTs */
1011                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1012                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1013                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1014                         if (!rc || rc == -EREMOTEIO) {
1015                                 oqctl->qc_dqblk.dqb_curspace +=
1016                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1017                                 oqctl->qc_dqblk.dqb_curinodes =
1018                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1019                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1020                         } else {
1021                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1022                         }
1023
1024                         OBD_FREE_PTR(oqctl_tmp);
1025                 }
1026 out:
1027                 QCTL_COPY(qctl, oqctl);
1028                 OBD_FREE_PTR(oqctl);
1029         }
1030
1031         RETURN(rc);
1032 }
1033
1034 #ifdef HAVE_UNLOCKED_IOCTL
1035 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1036 #else
1037 static int  ll_dir_ioctl(struct inode *unuse, struct file *file,
1038                         unsigned int cmd, unsigned long arg)
1039 #endif
1040 {
1041         struct inode *inode = file->f_dentry->d_inode;
1042         struct ll_sb_info *sbi = ll_i2sbi(inode);
1043         struct obd_ioctl_data *data;
1044         int rc = 0;
1045         ENTRY;
1046
1047         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1048                inode->i_ino, inode->i_generation, inode, cmd);
1049
1050         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1051         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1052                 return -ENOTTY;
1053
1054         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1055         switch(cmd) {
1056         case FSFILT_IOC_GETFLAGS:
1057         case FSFILT_IOC_SETFLAGS:
1058                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1059         case FSFILT_IOC_GETVERSION_OLD:
1060         case FSFILT_IOC_GETVERSION:
1061                 RETURN(put_user(inode->i_generation, (int *)arg));
1062         /* We need to special case any other ioctls we want to handle,
1063          * to send them to the MDS/OST as appropriate and to properly
1064          * network encode the arg field.
1065         case FSFILT_IOC_SETVERSION_OLD:
1066         case FSFILT_IOC_SETVERSION:
1067         */
1068         case LL_IOC_GET_MDTIDX: {
1069                 int mdtidx;
1070
1071                 mdtidx = ll_get_mdt_idx(inode);
1072                 if (mdtidx < 0)
1073                         RETURN(mdtidx);
1074
1075                 if (put_user((int)mdtidx, (int*)arg))
1076                         RETURN(-EFAULT);
1077
1078                 return 0;
1079         }
1080         case IOC_MDC_LOOKUP: {
1081                 struct ptlrpc_request *request = NULL;
1082                 int namelen, len = 0;
1083                 char *buf = NULL;
1084                 char *filename;
1085                 struct md_op_data *op_data;
1086
1087                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1088                 if (rc)
1089                         RETURN(rc);
1090                 data = (void *)buf;
1091
1092                 filename = data->ioc_inlbuf1;
1093                 namelen = strlen(filename);
1094
1095                 if (namelen < 1) {
1096                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1097                         GOTO(out_free, rc = -EINVAL);
1098                 }
1099
1100                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1101                                              0, LUSTRE_OPC_ANY, NULL);
1102                 if (IS_ERR(op_data))
1103                         GOTO(out_free, rc = PTR_ERR(op_data));
1104
1105                 op_data->op_valid = OBD_MD_FLID;
1106                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1107                 ll_finish_md_op_data(op_data);
1108                 if (rc < 0) {
1109                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1110                         GOTO(out_free, rc);
1111                 }
1112                 ptlrpc_req_finished(request);
1113                 EXIT;
1114 out_free:
1115                 obd_ioctl_freedata(buf, len);
1116                 return rc;
1117         }
1118         case LL_IOC_LOV_SETSTRIPE: {
1119                 struct lov_user_md_v3 lumv3;
1120                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1121                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1122                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1123
1124                 int set_default = 0;
1125
1126                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1127                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1128                         sizeof(lumv3p->lmm_objects[0]));
1129                 /* first try with v1 which is smaller than v3 */
1130                 if (cfs_copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1131                         RETURN(-EFAULT);
1132
1133                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1134                         if (cfs_copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1135                                 RETURN(-EFAULT);
1136                 }
1137
1138                 if (inode->i_sb->s_root == file->f_dentry)
1139                         set_default = 1;
1140
1141                 /* in v1 and v3 cases lumv1 points to data */
1142                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1143
1144                 RETURN(rc);
1145         }
1146         case LL_IOC_OBD_STATFS:
1147                 RETURN(ll_obd_statfs(inode, (void *)arg));
1148         case LL_IOC_LOV_GETSTRIPE:
1149         case LL_IOC_MDC_GETINFO:
1150         case IOC_MDC_GETFILEINFO:
1151         case IOC_MDC_GETFILESTRIPE: {
1152                 struct ptlrpc_request *request = NULL;
1153                 struct lov_user_md *lump;
1154                 struct lov_mds_md *lmm = NULL;
1155                 struct mdt_body *body;
1156                 char *filename = NULL;
1157                 int lmmsize;
1158
1159                 if (cmd == IOC_MDC_GETFILEINFO ||
1160                     cmd == IOC_MDC_GETFILESTRIPE) {
1161                         filename = getname((const char *)arg);
1162                         if (IS_ERR(filename))
1163                                 RETURN(PTR_ERR(filename));
1164
1165                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1166                                                       &lmmsize, &request);
1167                 } else {
1168                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1169                 }
1170
1171                 if (request) {
1172                         body = req_capsule_server_get(&request->rq_pill,
1173                                                       &RMF_MDT_BODY);
1174                         LASSERT(body != NULL);
1175                 } else {
1176                         GOTO(out_req, rc);
1177                 }
1178
1179                 if (rc < 0) {
1180                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1181                                                cmd == LL_IOC_MDC_GETINFO))
1182                                 GOTO(skip_lmm, rc = 0);
1183                         else
1184                                 GOTO(out_req, rc);
1185                 }
1186
1187                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1188                     cmd == LL_IOC_LOV_GETSTRIPE) {
1189                         lump = (struct lov_user_md *)arg;
1190                 } else {
1191                         struct lov_user_mds_data *lmdp;
1192                         lmdp = (struct lov_user_mds_data *)arg;
1193                         lump = &lmdp->lmd_lmm;
1194                 }
1195                 if (cfs_copy_to_user(lump, lmm, lmmsize)) {
1196                         if (cfs_copy_to_user(lump, lmm, sizeof(*lump)))
1197                                 GOTO(out_req, rc = -EFAULT);
1198                         rc = -EOVERFLOW;
1199                 }
1200         skip_lmm:
1201                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1202                         struct lov_user_mds_data *lmdp;
1203                         lstat_t st = { 0 };
1204
1205                         st.st_dev     = inode->i_sb->s_dev;
1206                         st.st_mode    = body->mode;
1207                         st.st_nlink   = body->nlink;
1208                         st.st_uid     = body->uid;
1209                         st.st_gid     = body->gid;
1210                         st.st_rdev    = body->rdev;
1211                         st.st_size    = body->size;
1212                         st.st_blksize = CFS_PAGE_SIZE;
1213                         st.st_blocks  = body->blocks;
1214                         st.st_atime   = body->atime;
1215                         st.st_mtime   = body->mtime;
1216                         st.st_ctime   = body->ctime;
1217                         st.st_ino     = inode->i_ino;
1218
1219                         lmdp = (struct lov_user_mds_data *)arg;
1220                         if (cfs_copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1221                                 GOTO(out_req, rc = -EFAULT);
1222                 }
1223
1224                 EXIT;
1225         out_req:
1226                 ptlrpc_req_finished(request);
1227                 if (filename)
1228                         putname(filename);
1229                 return rc;
1230         }
1231         case IOC_LOV_GETINFO: {
1232                 struct lov_user_mds_data *lumd;
1233                 struct lov_stripe_md *lsm;
1234                 struct lov_user_md *lum;
1235                 struct lov_mds_md *lmm;
1236                 int lmmsize;
1237                 lstat_t st;
1238
1239                 lumd = (struct lov_user_mds_data *)arg;
1240                 lum = &lumd->lmd_lmm;
1241
1242                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1243                 if (rc)
1244                         RETURN(rc);
1245
1246                 OBD_ALLOC_LARGE(lmm, lmmsize);
1247                 if (cfs_copy_from_user(lmm, lum, lmmsize))
1248                         GOTO(free_lmm, rc = -EFAULT);
1249
1250                 switch (lmm->lmm_magic) {
1251                 case LOV_USER_MAGIC_V1:
1252                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1253                                 break;
1254                         /* swab objects first so that stripes num will be sane */
1255                         lustre_swab_lov_user_md_objects(
1256                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1257                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1258                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1259                         break;
1260                 case LOV_USER_MAGIC_V3:
1261                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1262                                 break;
1263                         /* swab objects first so that stripes num will be sane */
1264                         lustre_swab_lov_user_md_objects(
1265                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1266                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1267                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1268                         break;
1269                 default:
1270                         GOTO(free_lmm, rc = -EINVAL);
1271                 }
1272
1273                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1274                 if (rc < 0)
1275                         GOTO(free_lmm, rc = -ENOMEM);
1276
1277                 /* Perform glimpse_size operation. */
1278                 memset(&st, 0, sizeof(st));
1279
1280                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1281                 if (rc)
1282                         GOTO(free_lsm, rc);
1283
1284                 if (cfs_copy_to_user(&lumd->lmd_st, &st, sizeof(st)))
1285                         GOTO(free_lsm, rc = -EFAULT);
1286
1287                 EXIT;
1288         free_lsm:
1289                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1290         free_lmm:
1291                 OBD_FREE_LARGE(lmm, lmmsize);
1292                 return rc;
1293         }
1294         case OBD_IOC_LLOG_CATINFO: {
1295                 struct ptlrpc_request *req = NULL;
1296                 char                  *buf = NULL;
1297                 char                  *str;
1298                 int                    len = 0;
1299
1300                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1301                 if (rc)
1302                         RETURN(rc);
1303                 data = (void *)buf;
1304
1305                 if (!data->ioc_inlbuf1) {
1306                         obd_ioctl_freedata(buf, len);
1307                         RETURN(-EINVAL);
1308                 }
1309
1310                 req = ptlrpc_request_alloc(sbi2mdc(sbi)->cl_import,
1311                                            &RQF_LLOG_CATINFO);
1312                 if (req == NULL)
1313                         GOTO(out_catinfo, rc = -ENOMEM);
1314
1315                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
1316                                      data->ioc_inllen1);
1317                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_CLIENT,
1318                                      data->ioc_inllen2);
1319
1320                 rc = ptlrpc_request_pack(req, LUSTRE_LOG_VERSION, LLOG_CATINFO);
1321                 if (rc) {
1322                         ptlrpc_request_free(req);
1323                         GOTO(out_catinfo, rc);
1324                 }
1325
1326                 str = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
1327                 memcpy(str, data->ioc_inlbuf1, data->ioc_inllen1);
1328                 if (data->ioc_inllen2) {
1329                         str = req_capsule_client_get(&req->rq_pill,
1330                                                      &RMF_STRING);
1331                         memcpy(str, data->ioc_inlbuf2, data->ioc_inllen2);
1332                 }
1333
1334                 req_capsule_set_size(&req->rq_pill, &RMF_STRING, RCL_SERVER,
1335                                      data->ioc_plen1);
1336                 ptlrpc_request_set_replen(req);
1337
1338                 rc = ptlrpc_queue_wait(req);
1339                 if (!rc) {
1340                         str = req_capsule_server_get(&req->rq_pill,
1341                                                      &RMF_STRING);
1342                         if (cfs_copy_to_user(data->ioc_pbuf1, str,
1343                                              data->ioc_plen1))
1344                                 rc = -EFAULT;
1345                 }
1346                 ptlrpc_req_finished(req);
1347         out_catinfo:
1348                 obd_ioctl_freedata(buf, len);
1349                 RETURN(rc);
1350         }
1351         case OBD_IOC_QUOTACHECK: {
1352                 struct obd_quotactl *oqctl;
1353                 int error = 0;
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(oqctl);
1360                 if (!oqctl)
1361                         RETURN(-ENOMEM);
1362                 oqctl->qc_type = arg;
1363                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1364                 if (rc < 0) {
1365                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1366                         error = rc;
1367                 }
1368
1369                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1370                 if (rc < 0)
1371                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1372
1373                 OBD_FREE_PTR(oqctl);
1374                 return error ?: rc;
1375         }
1376         case OBD_IOC_POLL_QUOTACHECK: {
1377                 struct if_quotacheck *check;
1378
1379                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
1380                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1381                         RETURN(-EPERM);
1382
1383                 OBD_ALLOC_PTR(check);
1384                 if (!check)
1385                         RETURN(-ENOMEM);
1386
1387                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1388                                    NULL);
1389                 if (rc) {
1390                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1391                         if (cfs_copy_to_user((void *)arg, check,
1392                                              sizeof(*check)))
1393                                 CDEBUG(D_QUOTA, "cfs_copy_to_user failed\n");
1394                         GOTO(out_poll, rc);
1395                 }
1396
1397                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1398                                    NULL);
1399                 if (rc) {
1400                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1401                         if (cfs_copy_to_user((void *)arg, check,
1402                                              sizeof(*check)))
1403                                 CDEBUG(D_QUOTA, "cfs_copy_to_user failed\n");
1404                         GOTO(out_poll, rc);
1405                 }
1406         out_poll:
1407                 OBD_FREE_PTR(check);
1408                 RETURN(rc);
1409         }
1410 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0)
1411         case LL_IOC_QUOTACTL_18: {
1412                 /* copy the old 1.x quota struct for internal use, then copy
1413                  * back into old format struct.  For 1.8 compatibility. */
1414                 struct if_quotactl_18 *qctl_18;
1415                 struct if_quotactl *qctl_20;
1416
1417                 OBD_ALLOC_PTR(qctl_18);
1418                 if (!qctl_18)
1419                         RETURN(-ENOMEM);
1420
1421                 OBD_ALLOC_PTR(qctl_20);
1422                 if (!qctl_20)
1423                         GOTO(out_quotactl_18, rc = -ENOMEM);
1424
1425                 if (cfs_copy_from_user(qctl_18, (void *)arg, sizeof(*qctl_18)))
1426                         GOTO(out_quotactl_20, rc = -ENOMEM);
1427
1428                 QCTL_COPY(qctl_20, qctl_18);
1429                 qctl_20->qc_idx = 0;
1430
1431                 /* XXX: dqb_valid was borrowed as a flag to mark that
1432                  *      only mds quota is wanted */
1433                 if (qctl_18->qc_cmd == Q_GETQUOTA &&
1434                     qctl_18->qc_dqblk.dqb_valid) {
1435                         qctl_20->qc_valid = QC_MDTIDX;
1436                         qctl_20->qc_dqblk.dqb_valid = 0;
1437                 } else if (qctl_18->obd_uuid.uuid[0] != '\0') {
1438                         qctl_20->qc_valid = QC_UUID;
1439                         qctl_20->obd_uuid = qctl_18->obd_uuid;
1440                 } else {
1441                         qctl_20->qc_valid = QC_GENERAL;
1442                 }
1443
1444                 rc = quotactl_ioctl(sbi, qctl_20);
1445
1446                 if (rc == 0) {
1447                         QCTL_COPY(qctl_18, qctl_20);
1448                         qctl_18->obd_uuid = qctl_20->obd_uuid;
1449
1450                         if (cfs_copy_to_user((void *)arg, qctl_18,
1451                                              sizeof(*qctl_18)))
1452                                 rc = -EFAULT;
1453                 }
1454
1455         out_quotactl_20:
1456                 OBD_FREE_PTR(qctl_20);
1457         out_quotactl_18:
1458                 OBD_FREE_PTR(qctl_18);
1459                 RETURN(rc);
1460         }
1461 #else
1462 #warning "remove old LL_IOC_QUOTACTL_18 compatibility code"
1463 #endif /* LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2,7,50,0) */
1464         case LL_IOC_QUOTACTL: {
1465                 struct if_quotactl *qctl;
1466
1467                 OBD_ALLOC_PTR(qctl);
1468                 if (!qctl)
1469                         RETURN(-ENOMEM);
1470
1471                 if (cfs_copy_from_user(qctl, (void *)arg, sizeof(*qctl)))
1472                         GOTO(out_quotactl, rc = -EFAULT);
1473
1474                 rc = quotactl_ioctl(sbi, qctl);
1475
1476                 if (rc == 0 && cfs_copy_to_user((void *)arg,qctl,sizeof(*qctl)))
1477                         rc = -EFAULT;
1478
1479         out_quotactl:
1480                 OBD_FREE_PTR(qctl);
1481                 RETURN(rc);
1482         }
1483         case OBD_IOC_GETDTNAME:
1484         case OBD_IOC_GETMDNAME:
1485                 RETURN(ll_get_obd_name(inode, cmd, arg));
1486         case LL_IOC_FLUSHCTX:
1487                 RETURN(ll_flush_ctx(inode));
1488 #ifdef CONFIG_FS_POSIX_ACL
1489         case LL_IOC_RMTACL: {
1490             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1491                 inode == inode->i_sb->s_root->d_inode) {
1492                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1493
1494                 LASSERT(fd != NULL);
1495                 rc = rct_add(&sbi->ll_rct, cfs_curproc_pid(), arg);
1496                 if (!rc)
1497                         fd->fd_flags |= LL_FILE_RMTACL;
1498                 RETURN(rc);
1499             } else
1500                 RETURN(0);
1501         }
1502 #endif
1503         case LL_IOC_GETOBDCOUNT: {
1504                 int count, vallen;
1505                 struct obd_export *exp;
1506
1507                 if (cfs_copy_from_user(&count, (int *)arg, sizeof(int)))
1508                         RETURN(-EFAULT);
1509
1510                 /* get ost count when count is zero, get mdt count otherwise */
1511                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1512                 vallen = sizeof(count);
1513                 rc = obd_get_info(exp, sizeof(KEY_TGT_COUNT), KEY_TGT_COUNT,
1514                                   &vallen, &count, NULL);
1515                 if (rc) {
1516                         CERROR("get target count failed: %d\n", rc);
1517                         RETURN(rc);
1518                 }
1519
1520                 if (cfs_copy_to_user((int *)arg, &count, sizeof(int)))
1521                         RETURN(-EFAULT);
1522
1523                 RETURN(0);
1524         }
1525         case LL_IOC_PATH2FID:
1526                 if (cfs_copy_to_user((void *)arg, ll_inode2fid(inode),
1527                                      sizeof(struct lu_fid)))
1528                         RETURN(-EFAULT);
1529                 RETURN(0);
1530         case LL_IOC_GET_CONNECT_FLAGS: {
1531                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void*)arg));
1532         }
1533         case OBD_IOC_CHANGELOG_SEND:
1534         case OBD_IOC_CHANGELOG_CLEAR:
1535                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1536                                     sizeof(struct ioc_changelog));
1537                 RETURN(rc);
1538         case OBD_IOC_FID2PATH:
1539                 RETURN(ll_fid2path(ll_i2mdexp(inode), (void *)arg));
1540         case LL_IOC_HSM_CT_START:
1541                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1542                                     sizeof(struct lustre_kernelcomm));
1543                 RETURN(rc);
1544
1545         default:
1546                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp,0,NULL,(void *)arg));
1547         }
1548 }
1549
1550 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1551 {
1552         struct inode *inode = file->f_mapping->host;
1553         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1554         struct ll_sb_info *sbi = ll_i2sbi(inode);
1555         int api32 = ll_need_32bit_api(sbi);
1556         loff_t ret = -EINVAL;
1557         ENTRY;
1558
1559         cfs_mutex_lock(&inode->i_mutex);
1560         switch (origin) {
1561                 case SEEK_SET:
1562                         break;
1563                 case SEEK_CUR:
1564                         offset += file->f_pos;
1565                         break;
1566                 case SEEK_END:
1567                         if (offset > 0)
1568                                 GOTO(out, ret);
1569                         if (api32)
1570                                 offset += LL_DIR_END_OFF_32BIT;
1571                         else
1572                                 offset += LL_DIR_END_OFF;
1573                         break;
1574                 default:
1575                         GOTO(out, ret);
1576         }
1577
1578         if (offset >= 0 &&
1579             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1580              (!api32 && offset <= LL_DIR_END_OFF))) {
1581                 if (offset != file->f_pos) {
1582                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1583                             (!api32 && offset == LL_DIR_END_OFF))
1584                                 fd->fd_dir.lfd_pos = MDS_DIR_END_OFF;
1585                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1586                                 fd->fd_dir.lfd_pos = offset << 32;
1587                         else
1588                                 fd->fd_dir.lfd_pos = offset;
1589                         file->f_pos = offset;
1590                         file->f_version = 0;
1591                 }
1592                 ret = offset;
1593         }
1594         GOTO(out, ret);
1595
1596 out:
1597         cfs_mutex_unlock(&inode->i_mutex);
1598         return ret;
1599 }
1600
1601 int ll_dir_open(struct inode *inode, struct file *file)
1602 {
1603         ENTRY;
1604         RETURN(ll_file_open(inode, file));
1605 }
1606
1607 int ll_dir_release(struct inode *inode, struct file *file)
1608 {
1609         ENTRY;
1610         RETURN(ll_file_release(inode, file));
1611 }
1612
1613 struct file_operations ll_dir_operations = {
1614         .llseek   = ll_dir_seek,
1615         .open     = ll_dir_open,
1616         .release  = ll_dir_release,
1617         .read     = generic_read_dir,
1618         .readdir  = ll_readdir,
1619 #ifdef HAVE_UNLOCKED_IOCTL
1620         .unlocked_ioctl   = ll_dir_ioctl,
1621 #else
1622         .ioctl          = ll_dir_ioctl,
1623 #endif
1624         .fsync    = ll_fsync,
1625 };