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

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/dir.c
  37  *
  38  * Directory code for lustre client.
  39  */
  40
  41 #include <linux/fs.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/mm.h>
  44 #include <linux/uaccess.h>
  45 #include <linux/buffer_head.h>   /* for wait_on_buffer */
  46 #include <linux/pagevec.h>
  47 #include <linux/prefetch.h>
  48
  49 #define DEBUG_SUBSYSTEM S_LLITE
  50
  51 #include "../include/obd_support.h"
  52 #include "../include/obd_class.h"
  53 #include "../include/lustre_lib.h"
  54 #include "../include/lustre/lustre_idl.h"
  55 #include "../include/lustre_lite.h"
  56 #include "../include/lustre_dlm.h"
  57 #include "../include/lustre_fid.h"
  58 #include "llite_internal.h"
  59
  60 /*
  61  * (new) readdir implementation overview.
  62  *
  63  * Original lustre readdir implementation cached exact copy of raw directory
  64  * pages on the client. These pages were indexed in client page cache by
  65  * logical offset in the directory file. This design, while very simple and
  66  * intuitive had some inherent problems:
  67  *
  68  *     . it implies that byte offset to the directory entry serves as a
  69  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  70  *     ext3/htree directory entries may move due to splits, and more
  71  *     importantly,
  72  *
  73  *     . it is incompatible with the design of split directories for cmd3,
  74  *     that assumes that names are distributed across nodes based on their
  75  *     hash, and so readdir should be done in hash order.
  76  *
  77  * New readdir implementation does readdir in hash order, and uses hash of a
  78  * file name as a telldir/seekdir cookie. This led to number of complications:
  79  *
  80  *     . hash is not unique, so it cannot be used to index cached directory
  81  *     pages on the client (note, that it requires a whole pageful of hash
  82  *     collided entries to cause two pages to have identical hashes);
  83  *
  84  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  85  *     entry cookie. ext3/htree has the same problem and lustre implementation
  86  *     mimics their solution: seekdir(hash) positions directory at the first
  87  *     entry with the given hash.
  88  *
  89  * Client side.
  90  *
  91  * 0. caching
  92  *
  93  * Client caches directory pages using hash of the first entry as an index. As
  94  * noted above hash is not unique, so this solution doesn't work as is:
  95  * special processing is needed for "page hash chains" (i.e., sequences of
  96  * pages filled with entries all having the same hash value).
  97  *
  98  * First, such chains have to be detected. To this end, server returns to the
  99  * client the hash of the first entry on the page next to one returned. When
 100  * client detects that this hash is the same as hash of the first entry on the
 101  * returned page, page hash collision has to be handled. Pages in the
 102  * hash chain, except first one, are termed "overflow pages".
 103  *
 104  * Solution to index uniqueness problem is to not cache overflow
 105  * pages. Instead, when page hash collision is detected, all overflow pages
 106  * from emerging chain are immediately requested from the server and placed in
 107  * a special data structure (struct ll_dir_chain). This data structure is used
 108  * by ll_readdir() to process entries from overflow pages. When readdir
 109  * invocation finishes, overflow pages are discarded. If page hash collision
 110  * chain weren't completely processed, next call to readdir will again detect
 111  * page hash collision, again read overflow pages in, process next portion of
 112  * entries and again discard the pages. This is not as wasteful as it looks,
 113  * because, given reasonable hash, page hash collisions are extremely rare.
 114  *
 115  * 1. directory positioning
 116  *
 117  * When seekdir(hash) is called, original
 118  *
 119  *
 120  *
 121  *
 122  *
 123  *
 124  *
 125  *
 126  * Server.
 127  *
 128  * identification of and access to overflow pages
 129  *
 130  * page format
 131  *
 132  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 133  * a header lu_dirpage which describes the start/end hash, and whether this
 134  * page is empty (contains no dir entry) or hash collide with next page.
 135  * After client receives reply, several pages will be integrated into dir page
 136  * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
 137  * lu_dirpage for this integrated page will be adjusted. See
 138  * lmv_adjust_dirpages().
 139  *
 140  */
 141
 142 /* returns the page unlocked, but with a reference */
 143 static int ll_dir_filler(void *_hash, struct page *page0)
 144 {
 145         struct inode *inode = page0->mapping->host;
 146         int hash64 = ll_i2sbi(inode)->ll_flags & LL_SBI_64BIT_HASH;
 147         struct obd_export *exp = ll_i2sbi(inode)->ll_md_exp;
 148         struct ptlrpc_request *request;
 149         struct mdt_body *body;
 150         struct md_op_data *op_data;
 151         __u64 hash = *((__u64 *)_hash);
 152         struct page **page_pool;
 153         struct page *page;
 154         struct lu_dirpage *dp;
 155         int max_pages = ll_i2sbi(inode)->ll_md_brw_size >> PAGE_CACHE_SHIFT;
 156         int nrdpgs = 0; /* number of pages read actually */
 157         int npages;
 158         int i;
 159         int rc;
 160
 161         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) hash %llu\n",
 162                inode->i_ino, inode->i_generation, inode, hash);
 163
 164         LASSERT(max_pages > 0 && max_pages <= MD_MAX_BRW_PAGES);
 165
 166         page_pool = kcalloc(max_pages, sizeof(page), GFP_NOFS);
 167         if (page_pool) {
 168                 page_pool[0] = page0;
 169         } else {
 170                 page_pool = &page0;
 171                 max_pages = 1;
 172         }
 173         for (npages = 1; npages < max_pages; npages++) {
 174                 page = page_cache_alloc_cold(inode->i_mapping);
 175                 if (!page)
 176                         break;
 177                 page_pool[npages] = page;
 178         }
 179
 180         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 181                                      LUSTRE_OPC_ANY, NULL);
 182         op_data->op_npages = npages;
 183         op_data->op_offset = hash;
 184         rc = md_readpage(exp, op_data, page_pool, &request);
 185         ll_finish_md_op_data(op_data);
 186         if (rc < 0) {
 187                 /* page0 is special, which was added into page cache early */
 188                 delete_from_page_cache(page0);
 189         } else if (rc == 0) {
 190                 body = req_capsule_server_get(&request->rq_pill, &RMF_MDT_BODY);
 191                 /* Checked by mdc_readpage() */
 192                 LASSERT(body != NULL);
 193
 194                 if (body->valid & OBD_MD_FLSIZE)
 195                         cl_isize_write(inode, body->size);
 196
 197                 nrdpgs = (request->rq_bulk->bd_nob_transferred+PAGE_CACHE_SIZE-1)
 198                          >> PAGE_CACHE_SHIFT;
 199                 SetPageUptodate(page0);
 200         }
 201         unlock_page(page0);
 202         ptlrpc_req_finished(request);
 203
 204         CDEBUG(D_VFSTRACE, "read %d/%d pages\n", nrdpgs, npages);
 205
 206         ll_pagevec_init(&lru_pvec, 0);
 207         for (i = 1; i < npages; i++) {
 208                 unsigned long offset;
 209                 int ret;
 210
 211                 page = page_pool[i];
 212
 213                 if (rc < 0 || i >= nrdpgs) {
 214                         page_cache_release(page);
 215                         continue;
 216                 }
 217
 218                 SetPageUptodate(page);
 219
 220                 dp = kmap(page);
 221                 hash = le64_to_cpu(dp->ldp_hash_start);
 222                 kunmap(page);
 223
 224                 offset = hash_x_index(hash, hash64);
 225
 226                 prefetchw(&page->flags);
 227                 ret = add_to_page_cache_lru(page, inode->i_mapping, offset,
 228                                             GFP_KERNEL);
 229                 if (ret == 0) {
 230                         unlock_page(page);
 231                         if (ll_pagevec_add(&lru_pvec, page) == 0)
 232                                 ll_pagevec_lru_add_file(&lru_pvec);
 233                 } else {
 234                         CDEBUG(D_VFSTRACE, "page %lu add to page cache failed: %d\n",
 235                                offset, ret);
 236                 }
 237                 page_cache_release(page);
 238         }
 239         ll_pagevec_lru_add_file(&lru_pvec);
 240
 241         if (page_pool != &page0)
 242                 OBD_FREE(page_pool, sizeof(struct page *) * max_pages);
 243         return rc;
 244 }
 245
 246 static void ll_check_page(struct inode *dir, struct page *page)
 247 {
 248         /* XXX: check page format later */
 249         SetPageChecked(page);
 250 }
 251
 252 void ll_release_page(struct page *page, int remove)
 253 {
 254         kunmap(page);
 255         if (remove) {
 256                 lock_page(page);
 257                 if (likely(page->mapping != NULL))
 258                         truncate_complete_page(page->mapping, page);
 259                 unlock_page(page);
 260         }
 261         page_cache_release(page);
 262 }
 263
 264 /*
 265  * Find, kmap and return page that contains given hash.
 266  */
 267 static struct page *ll_dir_page_locate(struct inode *dir, __u64 *hash,
 268                                        __u64 *start, __u64 *end)
 269 {
 270         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 271         struct address_space *mapping = dir->i_mapping;
 272         /*
 273          * Complement of hash is used as an index so that
 274          * radix_tree_gang_lookup() can be used to find a page with starting
 275          * hash _smaller_ than one we are looking for.
 276          */
 277         unsigned long offset = hash_x_index(*hash, hash64);
 278         struct page *page;
 279         int found;
 280
 281         spin_lock_irq(&mapping->tree_lock);
 282         found = radix_tree_gang_lookup(&mapping->page_tree,
 283                                        (void **)&page, offset, 1);
 284         if (found > 0 && !radix_tree_exceptional_entry(page)) {
 285                 struct lu_dirpage *dp;
 286
 287                 page_cache_get(page);
 288                 spin_unlock_irq(&mapping->tree_lock);
 289                 /*
 290                  * In contrast to find_lock_page() we are sure that directory
 291                  * page cannot be truncated (while DLM lock is held) and,
 292                  * hence, can avoid restart.
 293                  *
 294                  * In fact, page cannot be locked here at all, because
 295                  * ll_dir_filler() does synchronous io.
 296                  */
 297                 wait_on_page_locked(page);
 298                 if (PageUptodate(page)) {
 299                         dp = kmap(page);
 300                         if (BITS_PER_LONG == 32 && hash64) {
 301                                 *start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 302                                 *end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 303                                 *hash  = *hash >> 32;
 304                         } else {
 305                                 *start = le64_to_cpu(dp->ldp_hash_start);
 306                                 *end   = le64_to_cpu(dp->ldp_hash_end);
 307                         }
 308                         LASSERTF(*start <= *hash, "start = %#llx,end = %#llx,hash = %#llx\n",
 309                                  *start, *end, *hash);
 310                         CDEBUG(D_VFSTRACE, "page %lu [%llu %llu], hash %llu\n",
 311                                offset, *start, *end, *hash);
 312                         if (*hash > *end) {
 313                                 ll_release_page(page, 0);
 314                                 page = NULL;
 315                         } else if (*end != *start && *hash == *end) {
 316                                 /*
 317                                  * upon hash collision, remove this page,
 318                                  * otherwise put page reference, and
 319                                  * ll_get_dir_page() will issue RPC to fetch
 320                                  * the page we want.
 321                                  */
 322                                 ll_release_page(page,
 323                                     le32_to_cpu(dp->ldp_flags) & LDF_COLLIDE);
 324                                 page = NULL;
 325                         }
 326                 } else {
 327                         page_cache_release(page);
 328                         page = ERR_PTR(-EIO);
 329                 }
 330
 331         } else {
 332                 spin_unlock_irq(&mapping->tree_lock);
 333                 page = NULL;
 334         }
 335         return page;
 336 }
 337
 338 struct page *ll_get_dir_page(struct inode *dir, __u64 hash,
 339                              struct ll_dir_chain *chain)
 340 {
 341         ldlm_policy_data_t policy = {.l_inodebits = {MDS_INODELOCK_UPDATE} };
 342         struct address_space *mapping = dir->i_mapping;
 343         struct lustre_handle lockh;
 344         struct lu_dirpage *dp;
 345         struct page *page;
 346         ldlm_mode_t mode;
 347         int rc;
 348         __u64 start = 0;
 349         __u64 end = 0;
 350         __u64 lhash = hash;
 351         struct ll_inode_info *lli = ll_i2info(dir);
 352         int hash64 = ll_i2sbi(dir)->ll_flags & LL_SBI_64BIT_HASH;
 353
 354         mode = LCK_PR;
 355         rc = md_lock_match(ll_i2sbi(dir)->ll_md_exp, LDLM_FL_BLOCK_GRANTED,
 356                            ll_inode2fid(dir), LDLM_IBITS, &policy, mode, &lockh);
 357         if (!rc) {
 358                 struct ldlm_enqueue_info einfo = {
 359                         .ei_type = LDLM_IBITS,
 360                         .ei_mode = mode,
 361                         .ei_cb_bl = ll_md_blocking_ast,
 362                         .ei_cb_cp = ldlm_completion_ast,
 363                 };
 364                 struct lookup_intent it = { .it_op = IT_READDIR };
 365                 struct ptlrpc_request *request;
 366                 struct md_op_data *op_data;
 367
 368                 op_data = ll_prep_md_op_data(NULL, dir, dir, NULL, 0, 0,
 369                 LUSTRE_OPC_ANY, NULL);
 370                 if (IS_ERR(op_data))
 371                         return (void *)op_data;
 372
 373                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, &einfo, &it,
 374                                 op_data, &lockh, NULL, 0, NULL, 0);
 375
 376                 ll_finish_md_op_data(op_data);
 377
 378                 request = (struct ptlrpc_request *)it.d.lustre.it_data;
 379                 if (request)
 380                         ptlrpc_req_finished(request);
 381                 if (rc < 0) {
 382                         CERROR("lock enqueue: "DFID" at %llu: rc %d\n",
 383                                 PFID(ll_inode2fid(dir)), hash, rc);
 384                         return ERR_PTR(rc);
 385                 }
 386
 387                 CDEBUG(D_INODE, "setting lr_lvb_inode to inode %p (%lu/%u)\n",
 388                        dir, dir->i_ino, dir->i_generation);
 389                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp,
 390                                  &it.d.lustre.it_lock_handle, dir, NULL);
 391         } else {
 392                 /* for cross-ref object, l_ast_data of the lock may not be set,
 393                  * we reset it here */
 394                 md_set_lock_data(ll_i2sbi(dir)->ll_md_exp, &lockh.cookie,
 395                                  dir, NULL);
 396         }
 397         ldlm_lock_dump_handle(D_OTHER, &lockh);
 398
 399         mutex_lock(&lli->lli_readdir_mutex);
 400         page = ll_dir_page_locate(dir, &lhash, &start, &end);
 401         if (IS_ERR(page)) {
 402                 CERROR("dir page locate: "DFID" at %llu: rc %ld\n",
 403                        PFID(ll_inode2fid(dir)), lhash, PTR_ERR(page));
 404                 goto out_unlock;
 405         } else if (page != NULL) {
 406                 /*
 407                  * XXX nikita: not entirely correct handling of a corner case:
 408                  * suppose hash chain of entries with hash value HASH crosses
 409                  * border between pages P0 and P1. First both P0 and P1 are
 410                  * cached, seekdir() is called for some entry from the P0 part
 411                  * of the chain. Later P0 goes out of cache. telldir(HASH)
 412                  * happens and finds P1, as it starts with matching hash
 413                  * value. Remaining entries from P0 part of the chain are
 414                  * skipped. (Is that really a bug?)
 415                  *
 416                  * Possible solutions: 0. don't cache P1 is such case, handle
 417                  * it as an "overflow" page. 1. invalidate all pages at
 418                  * once. 2. use HASH|1 as an index for P1.
 419                  */
 420                 goto hash_collision;
 421         }
 422
 423         page = read_cache_page(mapping, hash_x_index(hash, hash64),
 424                                ll_dir_filler, &lhash);
 425         if (IS_ERR(page)) {
 426                 CERROR("read cache page: "DFID" at %llu: rc %ld\n",
 427                        PFID(ll_inode2fid(dir)), hash, PTR_ERR(page));
 428                 goto out_unlock;
 429         }
 430
 431         wait_on_page_locked(page);
 432         (void)kmap(page);
 433         if (!PageUptodate(page)) {
 434                 CERROR("page not updated: "DFID" at %llu: rc %d\n",
 435                        PFID(ll_inode2fid(dir)), hash, -5);
 436                 goto fail;
 437         }
 438         if (!PageChecked(page))
 439                 ll_check_page(dir, page);
 440         if (PageError(page)) {
 441                 CERROR("page error: "DFID" at %llu: rc %d\n",
 442                        PFID(ll_inode2fid(dir)), hash, -5);
 443                 goto fail;
 444         }
 445 hash_collision:
 446         dp = page_address(page);
 447         if (BITS_PER_LONG == 32 && hash64) {
 448                 start = le64_to_cpu(dp->ldp_hash_start) >> 32;
 449                 end   = le64_to_cpu(dp->ldp_hash_end) >> 32;
 450                 lhash = hash >> 32;
 451         } else {
 452                 start = le64_to_cpu(dp->ldp_hash_start);
 453                 end   = le64_to_cpu(dp->ldp_hash_end);
 454                 lhash = hash;
 455         }
 456         if (end == start) {
 457                 LASSERT(start == lhash);
 458                 CWARN("Page-wide hash collision: %llu\n", end);
 459                 if (BITS_PER_LONG == 32 && hash64)
 460                         CWARN("Real page-wide hash collision at [%llu %llu] with hash %llu\n",
 461                               le64_to_cpu(dp->ldp_hash_start),
 462                               le64_to_cpu(dp->ldp_hash_end), hash);
 463                 /*
 464                  * Fetch whole overflow chain...
 465                  *
 466                  * XXX not yet.
 467                  */
 468                 goto fail;
 469         }
 470 out_unlock:
 471         mutex_unlock(&lli->lli_readdir_mutex);
 472         ldlm_lock_decref(&lockh, mode);
 473         return page;
 474
 475 fail:
 476         ll_release_page(page, 1);
 477         page = ERR_PTR(-EIO);
 478         goto out_unlock;
 479 }
 480
 481 int ll_dir_read(struct inode *inode, struct dir_context *ctx)
 482 {
 483         struct ll_inode_info *info       = ll_i2info(inode);
 484         struct ll_sb_info    *sbi       = ll_i2sbi(inode);
 485         __u64              pos          = ctx->pos;
 486         int                api32      = ll_need_32bit_api(sbi);
 487         int                hash64     = sbi->ll_flags & LL_SBI_64BIT_HASH;
 488         struct page       *page;
 489         struct ll_dir_chain   chain;
 490         int                done = 0;
 491         int                rc = 0;
 492
 493         ll_dir_chain_init(&chain);
 494
 495         page = ll_get_dir_page(inode, pos, &chain);
 496
 497         while (rc == 0 && !done) {
 498                 struct lu_dirpage *dp;
 499                 struct lu_dirent  *ent;
 500
 501                 if (!IS_ERR(page)) {
 502                         /*
 503                          * If page is empty (end of directory is reached),
 504                          * use this value.
 505                          */
 506                         __u64 hash = MDS_DIR_END_OFF;
 507                         __u64 next;
 508
 509                         dp = page_address(page);
 510                         for (ent = lu_dirent_start(dp); ent != NULL && !done;
 511                              ent = lu_dirent_next(ent)) {
 512                                 __u16     type;
 513                                 int         namelen;
 514                                 struct lu_fid  fid;
 515                                 __u64     lhash;
 516                                 __u64     ino;
 517
 518                                 /*
 519                                  * XXX: implement correct swabbing here.
 520                                  */
 521
 522                                 hash = le64_to_cpu(ent->lde_hash);
 523                                 if (hash < pos)
 524                                         /*
 525                                          * Skip until we find target hash
 526                                          * value.
 527                                          */
 528                                         continue;
 529
 530                                 namelen = le16_to_cpu(ent->lde_namelen);
 531                                 if (namelen == 0)
 532                                         /*
 533                                          * Skip dummy record.
 534                                          */
 535                                         continue;
 536
 537                                 if (api32 && hash64)
 538                                         lhash = hash >> 32;
 539                                 else
 540                                         lhash = hash;
 541                                 fid_le_to_cpu(&fid, &ent->lde_fid);
 542                                 ino = cl_fid_build_ino(&fid, api32);
 543                                 type = ll_dirent_type_get(ent);
 544                                 ctx->pos = lhash;
 545                                 /* For 'll_nfs_get_name_filldir()', it will try
 546                                  * to access the 'ent' through its 'lde_name',
 547                                  * so the parameter 'name' for 'ctx->actor()'
 548                                  * must be part of the 'ent'.
 549                                  */
 550                                 done = !dir_emit(ctx, ent->lde_name,
 551                                                  namelen, ino, type);
 552                         }
 553                         next = le64_to_cpu(dp->ldp_hash_end);
 554                         if (!done) {
 555                                 pos = next;
 556                                 if (pos == MDS_DIR_END_OFF) {
 557                                         /*
 558                                          * End of directory reached.
 559                                          */
 560                                         done = 1;
 561                                         ll_release_page(page, 0);
 562                                 } else if (1 /* chain is exhausted*/) {
 563                                         /*
 564                                          * Normal case: continue to the next
 565                                          * page.
 566                                          */
 567                                         ll_release_page(page,
 568                                             le32_to_cpu(dp->ldp_flags) &
 569                                                         LDF_COLLIDE);
 570                                         next = pos;
 571                                         page = ll_get_dir_page(inode, pos,
 572                                                                &chain);
 573                                 } else {
 574                                         /*
 575                                          * go into overflow page.
 576                                          */
 577                                         LASSERT(le32_to_cpu(dp->ldp_flags) &
 578                                                 LDF_COLLIDE);
 579                                         ll_release_page(page, 1);
 580                                 }
 581                         } else {
 582                                 pos = hash;
 583                                 ll_release_page(page, 0);
 584                         }
 585                 } else {
 586                         rc = PTR_ERR(page);
 587                         CERROR("error reading dir "DFID" at %lu: rc %d\n",
 588                                PFID(&info->lli_fid), (unsigned long)pos, rc);
 589                 }
 590         }
 591
 592         ctx->pos = pos;
 593         ll_dir_chain_fini(&chain);
 594         return rc;
 595 }
 596
 597 static int ll_readdir(struct file *filp, struct dir_context *ctx)
 598 {
 599         struct inode            *inode  = file_inode(filp);
 600         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 601         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 602         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 603         int                     api32   = ll_need_32bit_api(sbi);
 604         int                     rc;
 605
 606         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %lu/%llu 32bit_api %d\n",
 607                inode->i_ino, inode->i_generation,
 608                inode, (unsigned long)lfd->lfd_pos, i_size_read(inode), api32);
 609
 610         if (lfd->lfd_pos == MDS_DIR_END_OFF) {
 611                 /*
 612                  * end-of-file.
 613                  */
 614                 rc = 0;
 615                 goto out;
 616         }
 617
 618         ctx->pos = lfd->lfd_pos;
 619         rc = ll_dir_read(inode, ctx);
 620         lfd->lfd_pos = ctx->pos;
 621         if (ctx->pos == MDS_DIR_END_OFF) {
 622                 if (api32)
 623                         ctx->pos = LL_DIR_END_OFF_32BIT;
 624                 else
 625                         ctx->pos = LL_DIR_END_OFF;
 626         } else {
 627                 if (api32 && hash64)
 628                         ctx->pos >>= 32;
 629         }
 630         filp->f_version = inode->i_version;
 631
 632 out:
 633         if (!rc)
 634                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 635
 636         return rc;
 637 }
 638
 639 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 640 {
 641         struct mgs_send_param *msp;
 642         int rc = 0;
 643
 644         msp = kzalloc(sizeof(*msp), GFP_NOFS);
 645         if (!msp)
 646                 return -ENOMEM;
 647
 648         strncpy(msp->mgs_param, string, MGS_PARAM_MAXLEN);
 649         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 650                                 sizeof(struct mgs_send_param), msp, NULL);
 651         if (rc)
 652                 CERROR("Failed to set parameter: %d\n", rc);
 653         OBD_FREE_PTR(msp);
 654
 655         return rc;
 656 }
 657
 658 static int ll_dir_setdirstripe(struct inode *dir, struct lmv_user_md *lump,
 659                                char *filename)
 660 {
 661         struct ptlrpc_request *request = NULL;
 662         struct md_op_data *op_data;
 663         struct ll_sb_info *sbi = ll_i2sbi(dir);
 664         int mode;
 665         int err;
 666
 667         mode = (0755 & ~current_umask()) | S_IFDIR;
 668         op_data = ll_prep_md_op_data(NULL, dir, NULL, filename,
 669                                      strlen(filename), mode, LUSTRE_OPC_MKDIR,
 670                                      lump);
 671         if (IS_ERR(op_data)) {
 672                 err = PTR_ERR(op_data);
 673                 goto err_exit;
 674         }
 675
 676         op_data->op_cli_flags |= CLI_SET_MEA;
 677         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 678                         from_kuid(&init_user_ns, current_fsuid()),
 679                         from_kgid(&init_user_ns, current_fsgid()),
 680                         cfs_curproc_cap_pack(), 0, &request);
 681         ll_finish_md_op_data(op_data);
 682         if (err)
 683                 goto err_exit;
 684 err_exit:
 685         ptlrpc_req_finished(request);
 686         return err;
 687 }
 688
 689 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 690                      int set_default)
 691 {
 692         struct ll_sb_info *sbi = ll_i2sbi(inode);
 693         struct md_op_data *op_data;
 694         struct ptlrpc_request *req = NULL;
 695         int rc = 0;
 696         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 697         struct obd_device *mgc = lsi->lsi_mgc;
 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: %#08x != %#08x nor %#08x\n",
 722                                lump->lmm_magic, LOV_USER_MAGIC_V1,
 723                                LOV_USER_MAGIC_V3);
 724                         return -EINVAL;
 725                 }
 726                 }
 727         } else {
 728                 lum_size = sizeof(struct lov_user_md_v1);
 729         }
 730
 731         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 732                                      LUSTRE_OPC_ANY, NULL);
 733         if (IS_ERR(op_data))
 734                 return PTR_ERR(op_data);
 735
 736         if (lump != NULL && lump->lmm_magic == cpu_to_le32(LMV_USER_MAGIC))
 737                 op_data->op_cli_flags |= CLI_SET_MEA;
 738
 739         /* swabbing is done in lov_setstripe() on server side */
 740         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size,
 741                         NULL, 0, &req, NULL);
 742         ll_finish_md_op_data(op_data);
 743         ptlrpc_req_finished(req);
 744         if (rc) {
 745                 if (rc != -EPERM && rc != -EACCES)
 746                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 747         }
 748
 749         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 750          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 751          need to make the distinction between the 2 versions */
 752         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 753                 char *param = NULL;
 754                 char *buf;
 755
 756                 param = kzalloc(MGS_PARAM_MAXLEN, GFP_NOFS);
 757                 if (!param) {
 758                         rc = -ENOMEM;
 759                         goto end;
 760                 }
 761
 762                 buf = param;
 763                 /* Get fsname and assume devname to be -MDT0000. */
 764                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 765                 strcat(buf, "-MDT0000.lov");
 766                 buf += strlen(buf);
 767
 768                 /* Set root stripesize */
 769                 sprintf(buf, ".stripesize=%u",
 770                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 771                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 772                 if (rc)
 773                         goto end;
 774
 775                 /* Set root stripecount */
 776                 sprintf(buf, ".stripecount=%hd",
 777                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 778                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 779                 if (rc)
 780                         goto end;
 781
 782                 /* Set root stripeoffset */
 783                 sprintf(buf, ".stripeoffset=%hd",
 784                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 785                         (typeof(lump->lmm_stripe_offset))(-1));
 786                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 787
 788 end:
 789                 if (param != NULL)
 790                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 791         }
 792         return rc;
 793 }
 794
 795 int ll_dir_getstripe(struct inode *inode, struct lov_mds_md **lmmp,
 796                      int *lmm_size, struct ptlrpc_request **request)
 797 {
 798         struct ll_sb_info *sbi = ll_i2sbi(inode);
 799         struct mdt_body   *body;
 800         struct lov_mds_md *lmm = NULL;
 801         struct ptlrpc_request *req = NULL;
 802         int rc, lmmsize;
 803         struct md_op_data *op_data;
 804
 805         rc = ll_get_default_mdsize(sbi, &lmmsize);
 806         if (rc)
 807                 return rc;
 808
 809         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 810                                      0, lmmsize, LUSTRE_OPC_ANY,
 811                                      NULL);
 812         if (IS_ERR(op_data))
 813                 return PTR_ERR(op_data);
 814
 815         op_data->op_valid = OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 816         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 817         ll_finish_md_op_data(op_data);
 818         if (rc < 0) {
 819                 CDEBUG(D_INFO, "md_getattr failed on inode %lu/%u: rc %d\n",
 820                        inode->i_ino,
 821                        inode->i_generation, rc);
 822                 goto out;
 823         }
 824
 825         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 826         LASSERT(body != NULL);
 827
 828         lmmsize = body->eadatasize;
 829
 830         if (!(body->valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 831             lmmsize == 0) {
 832                 rc = -ENODATA;
 833                 goto out;
 834         }
 835
 836         lmm = req_capsule_server_sized_get(&req->rq_pill,
 837                                            &RMF_MDT_MD, lmmsize);
 838         LASSERT(lmm != NULL);
 839
 840         /*
 841          * This is coming from the MDS, so is probably in
 842          * little endian.  We convert it to host endian before
 843          * passing it to userspace.
 844          */
 845         /* We don't swab objects for directories */
 846         switch (le32_to_cpu(lmm->lmm_magic)) {
 847         case LOV_MAGIC_V1:
 848                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 849                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 850                 break;
 851         case LOV_MAGIC_V3:
 852                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 853                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 854                 break;
 855         default:
 856                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 857                 rc = -EPROTO;
 858         }
 859 out:
 860         *lmmp = lmm;
 861         *lmm_size = lmmsize;
 862         *request = req;
 863         return rc;
 864 }
 865
 866 /*
 867  *  Get MDT index for the inode.
 868  */
 869 int ll_get_mdt_idx(struct inode *inode)
 870 {
 871         struct ll_sb_info *sbi = ll_i2sbi(inode);
 872         struct md_op_data *op_data;
 873         int rc, mdtidx;
 874
 875         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0,
 876                                      0, LUSTRE_OPC_ANY, NULL);
 877         if (IS_ERR(op_data))
 878                 return PTR_ERR(op_data);
 879
 880         op_data->op_flags |= MF_GET_MDT_IDX;
 881         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 882         mdtidx = op_data->op_mds;
 883         ll_finish_md_op_data(op_data);
 884         if (rc < 0) {
 885                 CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
 886                 return rc;
 887         }
 888         return mdtidx;
 889 }
 890
 891 /**
 892  * Generic handler to do any pre-copy work.
 893  *
 894  * It send a first hsm_progress (with extent length == 0) to coordinator as a
 895  * first information for it that real work has started.
 896  *
 897  * Moreover, for a ARCHIVE request, it will sample the file data version and
 898  * store it in \a copy.
 899  *
 900  * \return 0 on success.
 901  */
 902 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 903 {
 904         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 905         struct hsm_progress_kernel       hpk;
 906         int                              rc;
 907
 908         /* Forge a hsm_progress based on data from copy. */
 909         hpk.hpk_fid = copy->hc_hai.hai_fid;
 910         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 911         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 912         hpk.hpk_extent.length = 0;
 913         hpk.hpk_flags = 0;
 914         hpk.hpk_errval = 0;
 915         hpk.hpk_data_version = 0;
 916
 917
 918         /* For archive request, we need to read the current file version. */
 919         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 920                 struct inode    *inode;
 921                 __u64            data_version = 0;
 922
 923                 /* Get inode for this fid */
 924                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 925                 if (IS_ERR(inode)) {
 926                         hpk.hpk_flags |= HP_FLAG_RETRY;
 927                         /* hpk_errval is >= 0 */
 928                         hpk.hpk_errval = -PTR_ERR(inode);
 929                         rc = PTR_ERR(inode);
 930                         goto progress;
 931                 }
 932
 933                 /* Read current file data version */
 934                 rc = ll_data_version(inode, &data_version, 1);
 935                 iput(inode);
 936                 if (rc != 0) {
 937                         CDEBUG(D_HSM, "Could not read file data version of "
 938                                       DFID" (rc = %d). Archive request (%#llx) could not be done.\n",
 939                                       PFID(&copy->hc_hai.hai_fid), rc,
 940                                       copy->hc_hai.hai_cookie);
 941                         hpk.hpk_flags |= HP_FLAG_RETRY;
 942                         /* hpk_errval must be >= 0 */
 943                         hpk.hpk_errval = -rc;
 944                         goto progress;
 945                 }
 946
 947                 /* Store it the hsm_copy for later copytool use.
 948                  * Always modified even if no lsm. */
 949                 copy->hc_data_version = data_version;
 950         }
 951
 952 progress:
 953         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 954                            &hpk, NULL);
 955
 956         return rc;
 957 }
 958
 959 /**
 960  * Generic handler to do any post-copy work.
 961  *
 962  * It will send the last hsm_progress update to coordinator to inform it
 963  * that copy is finished and whether it was successful or not.
 964  *
 965  * Moreover,
 966  * - for ARCHIVE request, it will sample the file data version and compare it
 967  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 968  *   will be considered as failed.
 969  * - for RESTORE request, it will sample the file data version and send it to
 970  *   coordinator which is useful if the file was imported as 'released'.
 971  *
 972  * \return 0 on success.
 973  */
 974 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 975 {
 976         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 977         struct hsm_progress_kernel       hpk;
 978         int                              rc;
 979
 980         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 981         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 982          * initialized if copy_end was called with copy == NULL.
 983          */
 984
 985         /* Forge a hsm_progress based on data from copy. */
 986         hpk.hpk_fid = copy->hc_hai.hai_fid;
 987         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 988         hpk.hpk_extent = copy->hc_hai.hai_extent;
 989         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 990         hpk.hpk_errval = copy->hc_errval;
 991         hpk.hpk_data_version = 0;
 992
 993         /* For archive request, we need to check the file data was not changed.
 994          *
 995          * For restore request, we need to send the file data version, this is
 996          * useful when the file was created using hsm_import.
 997          */
 998         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 999              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
1000             (copy->hc_errval == 0)) {
1001                 struct inode    *inode;
1002                 __u64            data_version = 0;
1003
1004                 /* Get lsm for this fid */
1005                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
1006                 if (IS_ERR(inode)) {
1007                         hpk.hpk_flags |= HP_FLAG_RETRY;
1008                         /* hpk_errval must be >= 0 */
1009                         hpk.hpk_errval = -PTR_ERR(inode);
1010                         rc = PTR_ERR(inode);
1011                         goto progress;
1012                 }
1013
1014                 rc = ll_data_version(inode, &data_version,
1015                                      copy->hc_hai.hai_action == HSMA_ARCHIVE);
1016                 iput(inode);
1017                 if (rc) {
1018                         CDEBUG(D_HSM, "Could not read file data version. Request could not be confirmed.\n");
1019                         if (hpk.hpk_errval == 0)
1020                                 hpk.hpk_errval = -rc;
1021                         goto progress;
1022                 }
1023
1024                 /* Store it the hsm_copy for later copytool use.
1025                  * Always modified even if no lsm. */
1026                 hpk.hpk_data_version = data_version;
1027
1028                 /* File could have been stripped during archiving, so we need
1029                  * to check anyway. */
1030                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
1031                     (copy->hc_data_version != data_version)) {
1032                         CDEBUG(D_HSM, "File data version mismatched. File content was changed during archiving. "
1033                                DFID", start:%#llx current:%#llx\n",
1034                                PFID(&copy->hc_hai.hai_fid),
1035                                copy->hc_data_version, data_version);
1036                         /* File was changed, send error to cdt. Do not ask for
1037                          * retry because if a file is modified frequently,
1038                          * the cdt will loop on retried archive requests.
1039                          * The policy engine will ask for a new archive later
1040                          * when the file will not be modified for some tunable
1041                          * time */
1042                         /* we do not notify caller */
1043                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
1044                         /* hpk_errval must be >= 0 */
1045                         hpk.hpk_errval = EBUSY;
1046                 }
1047
1048         }
1049
1050 progress:
1051         rc = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
1052                            &hpk, NULL);
1053
1054         return rc;
1055 }
1056
1057
1058 static int copy_and_ioctl(int cmd, struct obd_export *exp,
1059                           const void __user *data, size_t size)
1060 {
1061         void *copy;
1062         int rc;
1063
1064         copy = kzalloc(size, GFP_NOFS);
1065         if (!copy)
1066                 return -ENOMEM;
1067
1068         if (copy_from_user(copy, data, size)) {
1069                 rc = -EFAULT;
1070                 goto out;
1071         }
1072
1073         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
1074 out:
1075         OBD_FREE(copy, size);
1076
1077         return rc;
1078 }
1079
1080 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1081 {
1082         int cmd = qctl->qc_cmd;
1083         int type = qctl->qc_type;
1084         int id = qctl->qc_id;
1085         int valid = qctl->qc_valid;
1086         int rc = 0;
1087
1088         switch (cmd) {
1089         case LUSTRE_Q_INVALIDATE:
1090         case LUSTRE_Q_FINVALIDATE:
1091         case Q_QUOTAON:
1092         case Q_QUOTAOFF:
1093         case Q_SETQUOTA:
1094         case Q_SETINFO:
1095                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1096                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1097                         return -EPERM;
1098                 break;
1099         case Q_GETQUOTA:
1100                 if (((type == USRQUOTA &&
1101                       !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
1102                      (type == GRPQUOTA &&
1103                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
1104                     (!capable(CFS_CAP_SYS_ADMIN) ||
1105                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
1106                         return -EPERM;
1107                 break;
1108         case Q_GETINFO:
1109                 break;
1110         default:
1111                 CERROR("unsupported quotactl op: %#x\n", cmd);
1112                 return -ENOTTY;
1113         }
1114
1115         if (valid != QC_GENERAL) {
1116                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
1117                         return -EOPNOTSUPP;
1118
1119                 if (cmd == Q_GETINFO)
1120                         qctl->qc_cmd = Q_GETOINFO;
1121                 else if (cmd == Q_GETQUOTA)
1122                         qctl->qc_cmd = Q_GETOQUOTA;
1123                 else
1124                         return -EINVAL;
1125
1126                 switch (valid) {
1127                 case QC_MDTIDX:
1128                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1129                                            sizeof(*qctl), qctl, NULL);
1130                         break;
1131                 case QC_OSTIDX:
1132                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1133                                            sizeof(*qctl), qctl, NULL);
1134                         break;
1135                 case QC_UUID:
1136                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1137                                            sizeof(*qctl), qctl, NULL);
1138                         if (rc == -EAGAIN)
1139                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1140                                                    sbi->ll_dt_exp,
1141                                                    sizeof(*qctl), qctl, NULL);
1142                         break;
1143                 default:
1144                         rc = -EINVAL;
1145                         break;
1146                 }
1147
1148                 if (rc)
1149                         return rc;
1150
1151                 qctl->qc_cmd = cmd;
1152         } else {
1153                 struct obd_quotactl *oqctl;
1154
1155                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1156                 if (!oqctl)
1157                         return -ENOMEM;
1158
1159                 QCTL_COPY(oqctl, qctl);
1160                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1161                 if (rc) {
1162                         if (rc != -EALREADY && cmd == Q_QUOTAON) {
1163                                 oqctl->qc_cmd = Q_QUOTAOFF;
1164                                 obd_quotactl(sbi->ll_md_exp, oqctl);
1165                         }
1166                         OBD_FREE_PTR(oqctl);
1167                         return rc;
1168                 }
1169                 /* If QIF_SPACE is not set, client should collect the
1170                  * space usage from OSSs by itself */
1171                 if (cmd == Q_GETQUOTA &&
1172                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1173                     !oqctl->qc_dqblk.dqb_curspace) {
1174                         struct obd_quotactl *oqctl_tmp;
1175
1176                         oqctl_tmp = kzalloc(sizeof(*oqctl_tmp), GFP_NOFS);
1177                         if (!oqctl_tmp) {
1178                                 rc = -ENOMEM;
1179                                 goto out;
1180                         }
1181
1182                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1183                         oqctl_tmp->qc_id = oqctl->qc_id;
1184                         oqctl_tmp->qc_type = oqctl->qc_type;
1185
1186                         /* collect space usage from OSTs */
1187                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1188                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1189                         if (!rc || rc == -EREMOTEIO) {
1190                                 oqctl->qc_dqblk.dqb_curspace =
1191                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1192                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1193                         }
1194
1195                         /* collect space & inode usage from MDTs */
1196                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1197                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1198                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1199                         if (!rc || rc == -EREMOTEIO) {
1200                                 oqctl->qc_dqblk.dqb_curspace +=
1201                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1202                                 oqctl->qc_dqblk.dqb_curinodes =
1203                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1204                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1205                         } else {
1206                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1207                         }
1208
1209                         OBD_FREE_PTR(oqctl_tmp);
1210                 }
1211 out:
1212                 QCTL_COPY(qctl, oqctl);
1213                 OBD_FREE_PTR(oqctl);
1214         }
1215
1216         return rc;
1217 }
1218
1219 static char *
1220 ll_getname(const char __user *filename)
1221 {
1222         int ret = 0, len;
1223         char *tmp = __getname();
1224
1225         if (!tmp)
1226                 return ERR_PTR(-ENOMEM);
1227
1228         len = strncpy_from_user(tmp, filename, PATH_MAX);
1229         if (len == 0)
1230                 ret = -ENOENT;
1231         else if (len > PATH_MAX)
1232                 ret = -ENAMETOOLONG;
1233
1234         if (ret) {
1235                 __putname(tmp);
1236                 tmp =  ERR_PTR(ret);
1237         }
1238         return tmp;
1239 }
1240
1241 #define ll_putname(filename) __putname(filename)
1242
1243 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1244 {
1245         struct inode *inode = file_inode(file);
1246         struct ll_sb_info *sbi = ll_i2sbi(inode);
1247         struct obd_ioctl_data *data;
1248         int rc = 0;
1249
1250         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
1251                inode->i_ino, inode->i_generation, inode, cmd);
1252
1253         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1254         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1255                 return -ENOTTY;
1256
1257         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1258         switch (cmd) {
1259         case FSFILT_IOC_GETFLAGS:
1260         case FSFILT_IOC_SETFLAGS:
1261                 return ll_iocontrol(inode, file, cmd, arg);
1262         case FSFILT_IOC_GETVERSION_OLD:
1263         case FSFILT_IOC_GETVERSION:
1264                 return put_user(inode->i_generation, (int *)arg);
1265         /* We need to special case any other ioctls we want to handle,
1266          * to send them to the MDS/OST as appropriate and to properly
1267          * network encode the arg field.
1268         case FSFILT_IOC_SETVERSION_OLD:
1269         case FSFILT_IOC_SETVERSION:
1270         */
1271         case LL_IOC_GET_MDTIDX: {
1272                 int mdtidx;
1273
1274                 mdtidx = ll_get_mdt_idx(inode);
1275                 if (mdtidx < 0)
1276                         return mdtidx;
1277
1278                 if (put_user((int)mdtidx, (int *)arg))
1279                         return -EFAULT;
1280
1281                 return 0;
1282         }
1283         case IOC_MDC_LOOKUP: {
1284                 struct ptlrpc_request *request = NULL;
1285                 int namelen, len = 0;
1286                 char *buf = NULL;
1287                 char *filename;
1288                 struct md_op_data *op_data;
1289
1290                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1291                 if (rc)
1292                         return rc;
1293                 data = (void *)buf;
1294
1295                 filename = data->ioc_inlbuf1;
1296                 namelen = strlen(filename);
1297
1298                 if (namelen < 1) {
1299                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1300                         rc = -EINVAL;
1301                         goto out_free;
1302                 }
1303
1304                 op_data = ll_prep_md_op_data(NULL, inode, NULL, filename, namelen,
1305                                              0, LUSTRE_OPC_ANY, NULL);
1306                 if (IS_ERR(op_data)) {
1307                         rc = PTR_ERR(op_data);
1308                         goto out_free;
1309                 }
1310
1311                 op_data->op_valid = OBD_MD_FLID;
1312                 rc = md_getattr_name(sbi->ll_md_exp, op_data, &request);
1313                 ll_finish_md_op_data(op_data);
1314                 if (rc < 0) {
1315                         CDEBUG(D_INFO, "md_getattr_name: %d\n", rc);
1316                         goto out_free;
1317                 }
1318                 ptlrpc_req_finished(request);
1319 out_free:
1320                 obd_ioctl_freedata(buf, len);
1321                 return rc;
1322         }
1323         case LL_IOC_LMV_SETSTRIPE: {
1324                 struct lmv_user_md  *lum;
1325                 char            *buf = NULL;
1326                 char            *filename;
1327                 int              namelen = 0;
1328                 int              lumlen = 0;
1329                 int              len;
1330                 int              rc;
1331
1332                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
1333                 if (rc)
1334                         return rc;
1335
1336                 data = (void *)buf;
1337                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1338                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0) {
1339                         rc = -EINVAL;
1340                         goto lmv_out_free;
1341                 }
1342
1343                 filename = data->ioc_inlbuf1;
1344                 namelen = data->ioc_inllen1;
1345
1346                 if (namelen < 1) {
1347                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1348                         rc = -EINVAL;
1349                         goto lmv_out_free;
1350                 }
1351                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1352                 lumlen = data->ioc_inllen2;
1353
1354                 if (lum->lum_magic != LMV_USER_MAGIC ||
1355                     lumlen != sizeof(*lum)) {
1356                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1357                                filename, lum->lum_magic, lumlen, -EFAULT);
1358                         rc = -EINVAL;
1359                         goto lmv_out_free;
1360                 }
1361
1362                 /**
1363                  * ll_dir_setdirstripe will be used to set dir stripe
1364                  *  mdc_create--->mdt_reint_create (with dirstripe)
1365                  */
1366                 rc = ll_dir_setdirstripe(inode, lum, filename);
1367 lmv_out_free:
1368                 obd_ioctl_freedata(buf, len);
1369                 return rc;
1370
1371         }
1372         case LL_IOC_LOV_SETSTRIPE: {
1373                 struct lov_user_md_v3 lumv3;
1374                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1375                 struct lov_user_md_v1 *lumv1p = (struct lov_user_md_v1 *)arg;
1376                 struct lov_user_md_v3 *lumv3p = (struct lov_user_md_v3 *)arg;
1377
1378                 int set_default = 0;
1379
1380                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1381                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1382                         sizeof(lumv3p->lmm_objects[0]));
1383                 /* first try with v1 which is smaller than v3 */
1384                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1385                         return -EFAULT;
1386
1387                 if (lumv1->lmm_magic == LOV_USER_MAGIC_V3) {
1388                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1389                                 return -EFAULT;
1390                 }
1391
1392                 if (is_root_inode(inode))
1393                         set_default = 1;
1394
1395                 /* in v1 and v3 cases lumv1 points to data */
1396                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1397
1398                 return rc;
1399         }
1400         case LL_IOC_LMV_GETSTRIPE: {
1401                 struct lmv_user_md *lump = (struct lmv_user_md *)arg;
1402                 struct lmv_user_md lum;
1403                 struct lmv_user_md *tmp;
1404                 int lum_size;
1405                 int rc = 0;
1406                 int mdtindex;
1407
1408                 if (copy_from_user(&lum, lump, sizeof(struct lmv_user_md)))
1409                         return -EFAULT;
1410
1411                 if (lum.lum_magic != LMV_MAGIC_V1)
1412                         return -EINVAL;
1413
1414                 lum_size = lmv_user_md_size(1, LMV_MAGIC_V1);
1415                 tmp = kzalloc(lum_size, GFP_NOFS);
1416                 if (!tmp) {
1417                         rc = -ENOMEM;
1418                         goto free_lmv;
1419                 }
1420
1421                 *tmp = lum;
1422                 tmp->lum_type = LMV_STRIPE_TYPE;
1423                 tmp->lum_stripe_count = 1;
1424                 mdtindex = ll_get_mdt_idx(inode);
1425                 if (mdtindex < 0) {
1426                         rc = -ENOMEM;
1427                         goto free_lmv;
1428                 }
1429
1430                 tmp->lum_stripe_offset = mdtindex;
1431                 tmp->lum_objects[0].lum_mds = mdtindex;
1432                 memcpy(&tmp->lum_objects[0].lum_fid, ll_inode2fid(inode),
1433                        sizeof(struct lu_fid));
1434                 if (copy_to_user((void *)arg, tmp, lum_size)) {
1435                         rc = -EFAULT;
1436                         goto free_lmv;
1437                 }
1438 free_lmv:
1439                 if (tmp)
1440                         OBD_FREE(tmp, lum_size);
1441                 return rc;
1442         }
1443         case LL_IOC_REMOVE_ENTRY: {
1444                 char            *filename = NULL;
1445                 int              namelen = 0;
1446                 int              rc;
1447
1448                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1449                  * unsupported server, which might crash the server(LU-2730),
1450                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1451                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1452                  * server will support REINT_RMENTRY XXX*/
1453                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1454                         return -ENOTSUPP;
1455
1456                 filename = ll_getname((const char *)arg);
1457                 if (IS_ERR(filename))
1458                         return PTR_ERR(filename);
1459
1460                 namelen = strlen(filename);
1461                 if (namelen < 1) {
1462                         rc = -EINVAL;
1463                         goto out_rmdir;
1464                 }
1465
1466                 rc = ll_rmdir_entry(inode, filename, namelen);
1467 out_rmdir:
1468                 if (filename)
1469                         ll_putname(filename);
1470                 return rc;
1471         }
1472         case LL_IOC_LOV_SWAP_LAYOUTS:
1473                 return -EPERM;
1474         case LL_IOC_OBD_STATFS:
1475                 return ll_obd_statfs(inode, (void *)arg);
1476         case LL_IOC_LOV_GETSTRIPE:
1477         case LL_IOC_MDC_GETINFO:
1478         case IOC_MDC_GETFILEINFO:
1479         case IOC_MDC_GETFILESTRIPE: {
1480                 struct ptlrpc_request *request = NULL;
1481                 struct lov_user_md *lump;
1482                 struct lov_mds_md *lmm = NULL;
1483                 struct mdt_body *body;
1484                 char *filename = NULL;
1485                 int lmmsize;
1486
1487                 if (cmd == IOC_MDC_GETFILEINFO ||
1488                     cmd == IOC_MDC_GETFILESTRIPE) {
1489                         filename = ll_getname((const char *)arg);
1490                         if (IS_ERR(filename))
1491                                 return PTR_ERR(filename);
1492
1493                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1494                                                       &lmmsize, &request);
1495                 } else {
1496                         rc = ll_dir_getstripe(inode, &lmm, &lmmsize, &request);
1497                 }
1498
1499                 if (request) {
1500                         body = req_capsule_server_get(&request->rq_pill,
1501                                                       &RMF_MDT_BODY);
1502                         LASSERT(body != NULL);
1503                 } else {
1504                         goto out_req;
1505                 }
1506
1507                 if (rc < 0) {
1508                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1509                                                cmd == LL_IOC_MDC_GETINFO)) {
1510                                 rc = 0;
1511                                 goto skip_lmm;
1512                         } else
1513                                 goto out_req;
1514                 }
1515
1516                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1517                     cmd == LL_IOC_LOV_GETSTRIPE) {
1518                         lump = (struct lov_user_md *)arg;
1519                 } else {
1520                         struct lov_user_mds_data *lmdp;
1521
1522                         lmdp = (struct lov_user_mds_data *)arg;
1523                         lump = &lmdp->lmd_lmm;
1524                 }
1525                 if (copy_to_user(lump, lmm, lmmsize)) {
1526                         if (copy_to_user(lump, lmm, sizeof(*lump))) {
1527                                 rc = -EFAULT;
1528                                 goto out_req;
1529                         }
1530                         rc = -EOVERFLOW;
1531                 }
1532 skip_lmm:
1533                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1534                         struct lov_user_mds_data *lmdp;
1535                         lstat_t st = { 0 };
1536
1537                         st.st_dev     = inode->i_sb->s_dev;
1538                         st.st_mode    = body->mode;
1539                         st.st_nlink   = body->nlink;
1540                         st.st_uid     = body->uid;
1541                         st.st_gid     = body->gid;
1542                         st.st_rdev    = body->rdev;
1543                         st.st_size    = body->size;
1544                         st.st_blksize = PAGE_CACHE_SIZE;
1545                         st.st_blocks  = body->blocks;
1546                         st.st_atime   = body->atime;
1547                         st.st_mtime   = body->mtime;
1548                         st.st_ctime   = body->ctime;
1549                         st.st_ino     = inode->i_ino;
1550
1551                         lmdp = (struct lov_user_mds_data *)arg;
1552                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st))) {
1553                                 rc = -EFAULT;
1554                                 goto out_req;
1555                         }
1556                 }
1557
1558 out_req:
1559                 ptlrpc_req_finished(request);
1560                 if (filename)
1561                         ll_putname(filename);
1562                 return rc;
1563         }
1564         case IOC_LOV_GETINFO: {
1565                 struct lov_user_mds_data *lumd;
1566                 struct lov_stripe_md *lsm;
1567                 struct lov_user_md *lum;
1568                 struct lov_mds_md *lmm;
1569                 int lmmsize;
1570                 lstat_t st;
1571
1572                 lumd = (struct lov_user_mds_data *)arg;
1573                 lum = &lumd->lmd_lmm;
1574
1575                 rc = ll_get_max_mdsize(sbi, &lmmsize);
1576                 if (rc)
1577                         return rc;
1578
1579                 OBD_ALLOC_LARGE(lmm, lmmsize);
1580                 if (lmm == NULL)
1581                         return -ENOMEM;
1582                 if (copy_from_user(lmm, lum, lmmsize)) {
1583                         rc = -EFAULT;
1584                         goto free_lmm;
1585                 }
1586
1587                 switch (lmm->lmm_magic) {
1588                 case LOV_USER_MAGIC_V1:
1589                         if (LOV_USER_MAGIC_V1 == cpu_to_le32(LOV_USER_MAGIC_V1))
1590                                 break;
1591                         /* swab objects first so that stripes num will be sane */
1592                         lustre_swab_lov_user_md_objects(
1593                                 ((struct lov_user_md_v1 *)lmm)->lmm_objects,
1594                                 ((struct lov_user_md_v1 *)lmm)->lmm_stripe_count);
1595                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
1596                         break;
1597                 case LOV_USER_MAGIC_V3:
1598                         if (LOV_USER_MAGIC_V3 == cpu_to_le32(LOV_USER_MAGIC_V3))
1599                                 break;
1600                         /* swab objects first so that stripes num will be sane */
1601                         lustre_swab_lov_user_md_objects(
1602                                 ((struct lov_user_md_v3 *)lmm)->lmm_objects,
1603                                 ((struct lov_user_md_v3 *)lmm)->lmm_stripe_count);
1604                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
1605                         break;
1606                 default:
1607                         rc = -EINVAL;
1608                         goto free_lmm;
1609                 }
1610
1611                 rc = obd_unpackmd(sbi->ll_dt_exp, &lsm, lmm, lmmsize);
1612                 if (rc < 0) {
1613                         rc = -ENOMEM;
1614                         goto free_lmm;
1615                 }
1616
1617                 /* Perform glimpse_size operation. */
1618                 memset(&st, 0, sizeof(st));
1619
1620                 rc = ll_glimpse_ioctl(sbi, lsm, &st);
1621                 if (rc)
1622                         goto free_lsm;
1623
1624                 if (copy_to_user(&lumd->lmd_st, &st, sizeof(st))) {
1625                         rc = -EFAULT;
1626                         goto free_lsm;
1627                 }
1628
1629 free_lsm:
1630                 obd_free_memmd(sbi->ll_dt_exp, &lsm);
1631 free_lmm:
1632                 OBD_FREE_LARGE(lmm, lmmsize);
1633                 return rc;
1634         }
1635         case OBD_IOC_LLOG_CATINFO: {
1636                 return -EOPNOTSUPP;
1637         }
1638         case OBD_IOC_QUOTACHECK: {
1639                 struct obd_quotactl *oqctl;
1640                 int error = 0;
1641
1642                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1643                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1644                         return -EPERM;
1645
1646                 oqctl = kzalloc(sizeof(*oqctl), GFP_NOFS);
1647                 if (!oqctl)
1648                         return -ENOMEM;
1649                 oqctl->qc_type = arg;
1650                 rc = obd_quotacheck(sbi->ll_md_exp, oqctl);
1651                 if (rc < 0) {
1652                         CDEBUG(D_INFO, "md_quotacheck failed: rc %d\n", rc);
1653                         error = rc;
1654                 }
1655
1656                 rc = obd_quotacheck(sbi->ll_dt_exp, oqctl);
1657                 if (rc < 0)
1658                         CDEBUG(D_INFO, "obd_quotacheck failed: rc %d\n", rc);
1659
1660                 OBD_FREE_PTR(oqctl);
1661                 return error ?: rc;
1662         }
1663         case OBD_IOC_POLL_QUOTACHECK: {
1664                 struct if_quotacheck *check;
1665
1666                 if (!capable(CFS_CAP_SYS_ADMIN) ||
1667                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
1668                         return -EPERM;
1669
1670                 check = kzalloc(sizeof(*check), GFP_NOFS);
1671                 if (!check)
1672                         return -ENOMEM;
1673
1674                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, 0, (void *)check,
1675                                    NULL);
1676                 if (rc) {
1677                         CDEBUG(D_QUOTA, "mdc ioctl %d failed: %d\n", cmd, rc);
1678                         if (copy_to_user((void *)arg, check,
1679                                              sizeof(*check)))
1680                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1681                         goto out_poll;
1682                 }
1683
1684                 rc = obd_iocontrol(cmd, sbi->ll_dt_exp, 0, (void *)check,
1685                                    NULL);
1686                 if (rc) {
1687                         CDEBUG(D_QUOTA, "osc ioctl %d failed: %d\n", cmd, rc);
1688                         if (copy_to_user((void *)arg, check,
1689                                              sizeof(*check)))
1690                                 CDEBUG(D_QUOTA, "copy_to_user failed\n");
1691                         goto out_poll;
1692                 }
1693 out_poll:
1694                 OBD_FREE_PTR(check);
1695                 return rc;
1696         }
1697         case LL_IOC_QUOTACTL: {
1698                 struct if_quotactl *qctl;
1699
1700                 qctl = kzalloc(sizeof(*qctl), GFP_NOFS);
1701                 if (!qctl)
1702                         return -ENOMEM;
1703
1704                 if (copy_from_user(qctl, (void *)arg, sizeof(*qctl))) {
1705                         rc = -EFAULT;
1706                         goto out_quotactl;
1707                 }
1708
1709                 rc = quotactl_ioctl(sbi, qctl);
1710
1711                 if (rc == 0 && copy_to_user((void *)arg, qctl, sizeof(*qctl)))
1712                         rc = -EFAULT;
1713
1714 out_quotactl:
1715                 OBD_FREE_PTR(qctl);
1716                 return rc;
1717         }
1718         case OBD_IOC_GETDTNAME:
1719         case OBD_IOC_GETMDNAME:
1720                 return ll_get_obd_name(inode, cmd, arg);
1721         case LL_IOC_FLUSHCTX:
1722                 return ll_flush_ctx(inode);
1723 #ifdef CONFIG_FS_POSIX_ACL
1724         case LL_IOC_RMTACL: {
1725             if (sbi->ll_flags & LL_SBI_RMT_CLIENT && is_root_inode(inode)) {
1726                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1727
1728                 LASSERT(fd != NULL);
1729                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1730                 if (!rc)
1731                         fd->fd_flags |= LL_FILE_RMTACL;
1732                 return rc;
1733             } else
1734                 return 0;
1735         }
1736 #endif
1737         case LL_IOC_GETOBDCOUNT: {
1738                 int count, vallen;
1739                 struct obd_export *exp;
1740
1741                 if (copy_from_user(&count, (int *)arg, sizeof(int)))
1742                         return -EFAULT;
1743
1744                 /* get ost count when count is zero, get mdt count otherwise */
1745                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1746                 vallen = sizeof(count);
1747                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1748                                   KEY_TGT_COUNT, &vallen, &count, NULL);
1749                 if (rc) {
1750                         CERROR("get target count failed: %d\n", rc);
1751                         return rc;
1752                 }
1753
1754                 if (copy_to_user((int *)arg, &count, sizeof(int)))
1755                         return -EFAULT;
1756
1757                 return 0;
1758         }
1759         case LL_IOC_PATH2FID:
1760                 if (copy_to_user((void *)arg, ll_inode2fid(inode),
1761                                      sizeof(struct lu_fid)))
1762                         return -EFAULT;
1763                 return 0;
1764         case LL_IOC_GET_CONNECT_FLAGS: {
1765                 return obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL, (void *)arg);
1766         }
1767         case OBD_IOC_CHANGELOG_SEND:
1768         case OBD_IOC_CHANGELOG_CLEAR:
1769                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1770                                     sizeof(struct ioc_changelog));
1771                 return rc;
1772         case OBD_IOC_FID2PATH:
1773                 return ll_fid2path(inode, (void *)arg);
1774         case LL_IOC_HSM_REQUEST: {
1775                 struct hsm_user_request *hur;
1776                 ssize_t                  totalsize;
1777
1778                 hur = kzalloc(sizeof(*hur), GFP_NOFS);
1779                 if (!hur)
1780                         return -ENOMEM;
1781
1782                 /* We don't know the true size yet; copy the fixed-size part */
1783                 if (copy_from_user(hur, (void *)arg, sizeof(*hur))) {
1784                         OBD_FREE_PTR(hur);
1785                         return -EFAULT;
1786                 }
1787
1788                 /* Compute the whole struct size */
1789                 totalsize = hur_len(hur);
1790                 OBD_FREE_PTR(hur);
1791                 if (totalsize < 0)
1792                         return -E2BIG;
1793
1794                 /* Final size will be more than double totalsize */
1795                 if (totalsize >= MDS_MAXREQSIZE / 3)
1796                         return -E2BIG;
1797
1798                 OBD_ALLOC_LARGE(hur, totalsize);
1799                 if (hur == NULL)
1800                         return -ENOMEM;
1801
1802                 /* Copy the whole struct */
1803                 if (copy_from_user(hur, (void *)arg, totalsize)) {
1804                         OBD_FREE_LARGE(hur, totalsize);
1805                         return -EFAULT;
1806                 }
1807
1808                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1809                         const struct lu_fid *fid;
1810                         struct inode *f;
1811                         int i;
1812
1813                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1814                                 fid = &hur->hur_user_item[i].hui_fid;
1815                                 f = search_inode_for_lustre(inode->i_sb, fid);
1816                                 if (IS_ERR(f)) {
1817                                         rc = PTR_ERR(f);
1818                                         break;
1819                                 }
1820
1821                                 rc = ll_hsm_release(f);
1822                                 iput(f);
1823                                 if (rc != 0)
1824                                         break;
1825                         }
1826                 } else {
1827                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1828                                            hur, NULL);
1829                 }
1830
1831                 OBD_FREE_LARGE(hur, totalsize);
1832
1833                 return rc;
1834         }
1835         case LL_IOC_HSM_PROGRESS: {
1836                 struct hsm_progress_kernel      hpk;
1837                 struct hsm_progress             hp;
1838
1839                 if (copy_from_user(&hp, (void *)arg, sizeof(hp)))
1840                         return -EFAULT;
1841
1842                 hpk.hpk_fid = hp.hp_fid;
1843                 hpk.hpk_cookie = hp.hp_cookie;
1844                 hpk.hpk_extent = hp.hp_extent;
1845                 hpk.hpk_flags = hp.hp_flags;
1846                 hpk.hpk_errval = hp.hp_errval;
1847                 hpk.hpk_data_version = 0;
1848
1849                 /* File may not exist in Lustre; all progress
1850                  * reported to Lustre root */
1851                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1852                                    NULL);
1853                 return rc;
1854         }
1855         case LL_IOC_HSM_CT_START:
1856                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void *)arg,
1857                                     sizeof(struct lustre_kernelcomm));
1858                 return rc;
1859
1860         case LL_IOC_HSM_COPY_START: {
1861                 struct hsm_copy *copy;
1862                 int              rc;
1863
1864                 copy = kzalloc(sizeof(*copy), GFP_NOFS);
1865                 if (!copy)
1866                         return -ENOMEM;
1867                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1868                         OBD_FREE_PTR(copy);
1869                         return -EFAULT;
1870                 }
1871
1872                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1873                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1874                         rc = -EFAULT;
1875
1876                 OBD_FREE_PTR(copy);
1877                 return rc;
1878         }
1879         case LL_IOC_HSM_COPY_END: {
1880                 struct hsm_copy *copy;
1881                 int              rc;
1882
1883                 copy = kzalloc(sizeof(*copy), GFP_NOFS);
1884                 if (!copy)
1885                         return -ENOMEM;
1886                 if (copy_from_user(copy, (char *)arg, sizeof(*copy))) {
1887                         OBD_FREE_PTR(copy);
1888                         return -EFAULT;
1889                 }
1890
1891                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1892                 if (copy_to_user((char *)arg, copy, sizeof(*copy)))
1893                         rc = -EFAULT;
1894
1895                 OBD_FREE_PTR(copy);
1896                 return rc;
1897         }
1898         default:
1899                 return obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL, (void *)arg);
1900         }
1901 }
1902
1903 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1904 {
1905         struct inode *inode = file->f_mapping->host;
1906         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1907         struct ll_sb_info *sbi = ll_i2sbi(inode);
1908         int api32 = ll_need_32bit_api(sbi);
1909         loff_t ret = -EINVAL;
1910
1911         mutex_lock(&inode->i_mutex);
1912         switch (origin) {
1913         case SEEK_SET:
1914                 break;
1915         case SEEK_CUR:
1916                 offset += file->f_pos;
1917                 break;
1918         case SEEK_END:
1919                 if (offset > 0)
1920                         goto out;
1921                 if (api32)
1922                         offset += LL_DIR_END_OFF_32BIT;
1923                 else
1924                         offset += LL_DIR_END_OFF;
1925                 break;
1926         default:
1927                 goto out;
1928         }
1929
1930         if (offset >= 0 &&
1931             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1932              (!api32 && offset <= LL_DIR_END_OFF))) {
1933                 if (offset != file->f_pos) {
1934                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1935                             (!api32 && offset == LL_DIR_END_OFF))
1936                                 fd->lfd_pos = MDS_DIR_END_OFF;
1937                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1938                                 fd->lfd_pos = offset << 32;
1939                         else
1940                                 fd->lfd_pos = offset;
1941                         file->f_pos = offset;
1942                         file->f_version = 0;
1943                 }
1944                 ret = offset;
1945         }
1946         goto out;
1947
1948 out:
1949         mutex_unlock(&inode->i_mutex);
1950         return ret;
1951 }
1952
1953 static int ll_dir_open(struct inode *inode, struct file *file)
1954 {
1955         return ll_file_open(inode, file);
1956 }
1957
1958 static int ll_dir_release(struct inode *inode, struct file *file)
1959 {
1960         return ll_file_release(inode, file);
1961 }
1962
1963 const struct file_operations ll_dir_operations = {
1964         .llseek   = ll_dir_seek,
1965         .open     = ll_dir_open,
1966         .release  = ll_dir_release,
1967         .read     = generic_read_dir,
1968         .iterate  = ll_readdir,
1969         .unlocked_ioctl   = ll_dir_ioctl,
1970         .fsync    = ll_fsync,
1971 };