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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  *
  32  * lustre/llite/dir.c
  33  *
  34  * Directory code for lustre client.
  35  */
  36
  37 #include <linux/fs.h>
  38 #include <linux/pagemap.h>
  39 #include <linux/mm.h>
  40 #include <linux/version.h>
  41 #include <linux/security.h>
  42 #include <linux/user_namespace.h>
  43 #include <linux/uidgid.h>
  44 #include <linux/uaccess.h>
  45 #include <linux/buffer_head.h>   // for wait_on_buffer
  46 #include <linux/pagevec.h>
  47
  48 #define DEBUG_SUBSYSTEM S_LLITE
  49
  50 #include <obd_support.h>
  51 #include <obd_class.h>
  52 #include <uapi/linux/lustre/lustre_ioctl.h>
  53 #include <uapi/linux/llcrypt.h>
  54 #include <lustre_lib.h>
  55 #include <lustre_dlm.h>
  56 #include <lustre_fid.h>
  57 #include <lustre_kernelcomm.h>
  58 #include <lustre_swab.h>
  59
  60 #include "llite_internal.h"
  61
  62 /*
  63  * (new) readdir implementation overview.
  64  *
  65  * Original lustre readdir implementation cached exact copy of raw directory
  66  * pages on the client. These pages were indexed in client page cache by
  67  * logical offset in the directory file. This design, while very simple and
  68  * intuitive had some inherent problems:
  69  *
  70  *     . it implies that byte offset to the directory entry serves as a
  71  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  72  *     ext3/htree directory entries may move due to splits, and more
  73  *     importantly,
  74  *
  75  *     . it is incompatible with the design of split directories for cmd3,
  76  *     that assumes that names are distributed across nodes based on their
  77  *     hash, and so readdir should be done in hash order.
  78  *
  79  * New readdir implementation does readdir in hash order, and uses hash of a
  80  * file name as a telldir/seekdir cookie. This led to number of complications:
  81  *
  82  *     . hash is not unique, so it cannot be used to index cached directory
  83  *     pages on the client (note, that it requires a whole pageful of hash
  84  *     collided entries to cause two pages to have identical hashes);
  85  *
  86  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  87  *     entry cookie. ext3/htree has the same problem and lustre implementation
  88  *     mimics their solution: seekdir(hash) positions directory at the first
  89  *     entry with the given hash.
  90  *
  91  * Client side.
  92  *
  93  * 0. caching
  94  *
  95  * Client caches directory pages using hash of the first entry as an index. As
  96  * noted above hash is not unique, so this solution doesn't work as is:
  97  * special processing is needed for "page hash chains" (i.e., sequences of
  98  * pages filled with entries all having the same hash value).
  99  *
 100  * First, such chains have to be detected. To this end, server returns to the
 101  * client the hash of the first entry on the page next to one returned. When
 102  * client detects that this hash is the same as hash of the first entry on the
 103  * returned page, page hash collision has to be handled. Pages in the
 104  * hash chain, except first one, are termed "overflow pages".
 105  *
 106  * Solution to index uniqueness problem is to not cache overflow
 107  * pages. Instead, when page hash collision is detected, all overflow pages
 108  * from emerging chain are immediately requested from the server and placed in
 109  * a special data structure (struct ll_dir_chain). This data structure is used
 110  * by ll_readdir() to process entries from overflow pages. When readdir
 111  * invocation finishes, overflow pages are discarded. If page hash collision
 112  * chain weren't completely processed, next call to readdir will again detect
 113  * page hash collision, again read overflow pages in, process next portion of
 114  * entries and again discard the pages. This is not as wasteful as it looks,
 115  * because, given reasonable hash, page hash collisions are extremely rare.
 116  *
 117  * 1. directory positioning
 118  *
 119  * When seekdir(hash) is called, original
 120  *
 121  *
 122  *
 123  *
 124  *
 125  *
 126  *
 127  *
 128  * Server.
 129  *
 130  * identification of and access to overflow pages
 131  *
 132  * page format
 133  *
 134  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 135  * a header lu_dirpage which describes the start/end hash, and whether this
 136  * page is empty (contains no dir entry) or hash collide with next page.
 137  * After client receives reply, several pages will be integrated into dir page
 138  * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
 139  * lu_dirpage for this integrated page will be adjusted. See
 140  * mdc_adjust_dirpages().
 141  *
 142  */
 143 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
 144                              __u64 offset, struct ll_dir_chain *chain)
 145 {
 146         struct md_callback      cb_op;
 147         struct page             *page;
 148         int                     rc;
 149
 150         cb_op.md_blocking_ast = ll_md_blocking_ast;
 151         rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
 152         if (rc != 0)
 153                 return ERR_PTR(rc);
 154
 155         return page;
 156 }
 157
 158 void ll_release_page(struct inode *inode, struct page *page,
 159                      bool remove)
 160 {
 161         kunmap(page);
 162
 163         /* Always remove the page for striped dir, because the page is
 164          * built from temporarily in LMV layer */
 165         if (inode && ll_dir_striped(inode)) {
 166                 __free_page(page);
 167                 return;
 168         }
 169
 170         if (remove) {
 171                 lock_page(page);
 172                 if (likely(page->mapping != NULL))
 173                         delete_from_page_cache(page);
 174                 unlock_page(page);
 175         }
 176         put_page(page);
 177 }
 178
 179 #ifdef HAVE_DIR_CONTEXT
 180 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 181                 struct dir_context *ctx)
 182 {
 183 #else
 184 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 185                 void *cookie, filldir_t filldir)
 186 {
 187 #endif
 188         struct ll_sb_info    *sbi        = ll_i2sbi(inode);
 189         __u64                 pos        = *ppos;
 190         bool                  is_api32 = ll_need_32bit_api(sbi);
 191         bool                  is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
 192         struct page          *page;
 193         struct ll_dir_chain   chain;
 194         bool                  done = false;
 195         int                   rc = 0;
 196         ENTRY;
 197
 198         ll_dir_chain_init(&chain);
 199
 200         page = ll_get_dir_page(inode, op_data, pos, &chain);
 201
 202         while (rc == 0 && !done) {
 203                 struct lu_dirpage *dp;
 204                 struct lu_dirent  *ent;
 205                 __u64 hash;
 206                 __u64 next;
 207
 208                 if (IS_ERR(page)) {
 209                         rc = PTR_ERR(page);
 210                         break;
 211                 }
 212
 213                 hash = MDS_DIR_END_OFF;
 214                 dp = page_address(page);
 215                 for (ent = lu_dirent_start(dp); ent != NULL && !done;
 216                      ent = lu_dirent_next(ent)) {
 217                         __u16          type;
 218                         int            namelen;
 219                         struct lu_fid  fid;
 220                         __u64          lhash;
 221                         __u64          ino;
 222
 223                         hash = le64_to_cpu(ent->lde_hash);
 224                         if (hash < pos) /* Skip until we find target hash */
 225                                 continue;
 226
 227                         namelen = le16_to_cpu(ent->lde_namelen);
 228                         if (namelen == 0) /* Skip dummy record */
 229                                 continue;
 230
 231                         if (is_api32 && is_hash64)
 232                                 lhash = hash >> 32;
 233                         else
 234                                 lhash = hash;
 235                         fid_le_to_cpu(&fid, &ent->lde_fid);
 236                         ino = cl_fid_build_ino(&fid, is_api32);
 237                         type = IFTODT(lu_dirent_type_get(ent));
 238                         /* For ll_nfs_get_name_filldir(), it will try to access
 239                          * 'ent' through 'lde_name', so the parameter 'name'
 240                          * for 'filldir()' must be part of the 'ent'. */
 241 #ifdef HAVE_DIR_CONTEXT
 242                         ctx->pos = lhash;
 243                         done = !dir_emit(ctx, ent->lde_name, namelen, ino,
 244                                          type);
 245 #else
 246                         done = filldir(cookie, ent->lde_name, namelen, lhash,
 247                                        ino, type);
 248 #endif
 249                 }
 250
 251                 if (done) {
 252                         pos = hash;
 253                         ll_release_page(inode, page, false);
 254                         break;
 255                 }
 256
 257                 next = le64_to_cpu(dp->ldp_hash_end);
 258                 pos = next;
 259                 if (pos == MDS_DIR_END_OFF) {
 260                         /*
 261                          * End of directory reached.
 262                          */
 263                         done = 1;
 264                         ll_release_page(inode, page, false);
 265                 } else {
 266                         /*
 267                          * Normal case: continue to the next
 268                          * page.
 269                          */
 270                         ll_release_page(inode, page,
 271                                         le32_to_cpu(dp->ldp_flags) &
 272                                         LDF_COLLIDE);
 273                         next = pos;
 274                         page = ll_get_dir_page(inode, op_data, pos,
 275                                                &chain);
 276                 }
 277         }
 278 #ifdef HAVE_DIR_CONTEXT
 279         ctx->pos = pos;
 280 #else
 281         *ppos = pos;
 282 #endif
 283         ll_dir_chain_fini(&chain);
 284         RETURN(rc);
 285 }
 286
 287 #ifdef HAVE_DIR_CONTEXT
 288 static int ll_iterate(struct file *filp, struct dir_context *ctx)
 289 #else
 290 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 291 #endif
 292 {
 293         struct inode *inode = file_inode(filp);
 294         struct ll_file_data *lfd = filp->private_data;
 295         struct ll_sb_info *sbi = ll_i2sbi(inode);
 296         int hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
 297         int api32 = ll_need_32bit_api(sbi);
 298         struct md_op_data *op_data;
 299         struct lu_fid pfid = { 0 };
 300         ktime_t kstart = ktime_get();
 301         __u64 pos;
 302         int rc;
 303         ENTRY;
 304
 305         if (lfd != NULL)
 306                 pos = lfd->lfd_pos;
 307         else
 308                 pos = 0;
 309
 310         CDEBUG(D_VFSTRACE,
 311                "VFS Op:inode="DFID"(%p) pos/size%lu/%llu 32bit_api %d\n",
 312                PFID(ll_inode2fid(inode)),
 313                inode, (unsigned long)pos, i_size_read(inode), api32);
 314
 315         if (pos == MDS_DIR_END_OFF)
 316                 /*
 317                  * end-of-file.
 318                  */
 319                 GOTO(out, rc = 0);
 320
 321         if (unlikely(ll_dir_striped(inode))) {
 322                 /*
 323                  * This is only needed for striped dir to fill ..,
 324                  * see lmv_read_page()
 325                  */
 326                 if (file_dentry(filp)->d_parent != NULL &&
 327                     file_dentry(filp)->d_parent->d_inode != NULL) {
 328                         __u64 ibits = MDS_INODELOCK_LOOKUP;
 329                         struct inode *parent =
 330                                 file_dentry(filp)->d_parent->d_inode;
 331
 332                         if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
 333                                 pfid = *ll_inode2fid(parent);
 334                 }
 335
 336                 /* If it can not find in cache, do lookup .. on the master
 337                  * object */
 338                 if (fid_is_zero(&pfid)) {
 339                         rc = ll_dir_get_parent_fid(inode, &pfid);
 340                         if (rc != 0)
 341                                 RETURN(rc);
 342                 }
 343         }
 344
 345         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 346                                      LUSTRE_OPC_ANY, inode);
 347         if (IS_ERR(op_data))
 348                 GOTO(out, rc = PTR_ERR(op_data));
 349
 350         /* foreign dirs are browsed out of Lustre */
 351         if (unlikely(op_data->op_mea1 != NULL &&
 352                      op_data->op_mea1->lsm_md_magic == LMV_MAGIC_FOREIGN)) {
 353                 ll_finish_md_op_data(op_data);
 354                 RETURN(-ENODATA);
 355         }
 356
 357         op_data->op_fid3 = pfid;
 358
 359 #ifdef HAVE_DIR_CONTEXT
 360         ctx->pos = pos;
 361         rc = ll_dir_read(inode, &pos, op_data, ctx);
 362         pos = ctx->pos;
 363 #else
 364         rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
 365 #endif
 366         if (lfd != NULL)
 367                 lfd->lfd_pos = pos;
 368
 369         if (pos == MDS_DIR_END_OFF) {
 370                 if (api32)
 371                         pos = LL_DIR_END_OFF_32BIT;
 372                 else
 373                         pos = LL_DIR_END_OFF;
 374         } else {
 375                 if (api32 && hash64)
 376                         pos = pos >> 32;
 377         }
 378 #ifdef HAVE_DIR_CONTEXT
 379         ctx->pos = pos;
 380 #else
 381         filp->f_pos = pos;
 382 #endif
 383         ll_finish_md_op_data(op_data);
 384
 385 out:
 386         if (!rc)
 387                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR,
 388                                    ktime_us_delta(ktime_get(), kstart));
 389
 390         RETURN(rc);
 391 }
 392
 393 /**
 394  * Create striped directory with specified stripe(@lump)
 395  *
 396  * \param[in] dparent   the parent of the directory.
 397  * \param[in] lump      the specified stripes.
 398  * \param[in] dirname   the name of the directory.
 399  * \param[in] mode      the specified mode of the directory.
 400  *
 401  * \retval              =0 if striped directory is being created successfully.
 402  *                      <0 if the creation is failed.
 403  */
 404 static int ll_dir_setdirstripe(struct dentry *dparent, struct lmv_user_md *lump,
 405                                size_t len, const char *dirname, umode_t mode)
 406 {
 407         struct inode *parent = dparent->d_inode;
 408         struct ptlrpc_request *request = NULL;
 409         struct md_op_data *op_data;
 410         struct ll_sb_info *sbi = ll_i2sbi(parent);
 411         struct inode *inode = NULL;
 412         struct dentry dentry = {
 413                 .d_parent = dparent,
 414                 .d_name = {
 415                         .name = dirname,
 416                         .len = strlen(dirname),
 417                         .hash = ll_full_name_hash(dparent, dirname,
 418                                                   strlen(dirname)),
 419                 },
 420         };
 421         bool encrypt = false;
 422         int err;
 423         ENTRY;
 424
 425         if (unlikely(!lmv_user_magic_supported(lump->lum_magic)))
 426                 RETURN(-EINVAL);
 427
 428         if (lump->lum_magic != LMV_MAGIC_FOREIGN) {
 429                 CDEBUG(D_VFSTRACE,
 430                        "VFS Op:inode="DFID"(%p) name %s stripe_offset %d, stripe_count: %u\n",
 431                        PFID(ll_inode2fid(parent)), parent, dirname,
 432                        (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 433         } else {
 434                 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lump;
 435
 436                 CDEBUG(D_VFSTRACE,
 437                        "VFS Op:inode="DFID"(%p) name %s foreign, length %u, value '%.*s'\n",
 438                        PFID(ll_inode2fid(parent)), parent, dirname,
 439                        lfm->lfm_length, lfm->lfm_length, lfm->lfm_value);
 440         }
 441
 442         if (lump->lum_stripe_count > 1 &&
 443             !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
 444                 RETURN(-EINVAL);
 445
 446         if (IS_DEADDIR(parent) &&
 447             !OBD_FAIL_CHECK(OBD_FAIL_LLITE_NO_CHECK_DEAD))
 448                 RETURN(-ENOENT);
 449
 450         if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_CRUSH)) {
 451                 if ((lump->lum_hash_type & LMV_HASH_TYPE_MASK) ==
 452                      LMV_HASH_TYPE_CRUSH) {
 453                         /* if server doesn't support 'crush' hash type,
 454                          * switch to fnv_1a_64.
 455                          */
 456                         lump->lum_hash_type &= ~LMV_HASH_TYPE_MASK;
 457                         lump->lum_hash_type |= LMV_HASH_TYPE_FNV_1A_64;
 458                 } else if ((lump->lum_hash_type & LMV_HASH_TYPE_MASK) ==
 459                      LMV_HASH_TYPE_UNKNOWN) {
 460                         /* from 2.14 MDT will choose default hash type if client
 461                          * doesn't set a valid one, while old server doesn't
 462                          * handle it.
 463                          */
 464                         lump->lum_hash_type &= ~LMV_HASH_TYPE_MASK;
 465                         lump->lum_hash_type |= LMV_HASH_TYPE_DEFAULT;
 466                 }
 467         }
 468
 469         if (unlikely(!lmv_user_magic_supported(cpu_to_le32(lump->lum_magic))))
 470                 lustre_swab_lmv_user_md(lump);
 471
 472         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 473                 mode &= ~current_umask();
 474         mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
 475         op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
 476                                      strlen(dirname), mode, LUSTRE_OPC_MKDIR,
 477                                      lump);
 478         if (IS_ERR(op_data))
 479                 RETURN(PTR_ERR(op_data));
 480
 481         if (IS_ENCRYPTED(parent) ||
 482             unlikely(llcrypt_dummy_context_enabled(parent))) {
 483                 err = llcrypt_get_encryption_info(parent);
 484                 if (err)
 485                         GOTO(out_op_data, err);
 486                 if (!llcrypt_has_encryption_key(parent))
 487                         GOTO(out_op_data, err = -ENOKEY);
 488                 encrypt = true;
 489         }
 490
 491         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
 492                 /* selinux_dentry_init_security() uses dentry->d_parent and name
 493                  * to determine the security context for the file. So our fake
 494                  * dentry should be real enough for this purpose. */
 495                 err = ll_dentry_init_security(&dentry, mode, &dentry.d_name,
 496                                               &op_data->op_file_secctx_name,
 497                                               &op_data->op_file_secctx,
 498                                               &op_data->op_file_secctx_size);
 499                 if (err < 0)
 500                         GOTO(out_op_data, err);
 501         }
 502
 503         op_data->op_cli_flags |= CLI_SET_MEA;
 504         err = md_create(sbi->ll_md_exp, op_data, lump, len, mode,
 505                         from_kuid(&init_user_ns, current_fsuid()),
 506                         from_kgid(&init_user_ns, current_fsgid()),
 507                         cfs_curproc_cap_pack(), 0, &request);
 508         if (err)
 509                 GOTO(out_request, err);
 510
 511         CFS_FAIL_TIMEOUT(OBD_FAIL_LLITE_SETDIRSTRIPE_PAUSE, cfs_fail_val);
 512
 513         err = ll_prep_inode(&inode, request, parent->i_sb, NULL);
 514         if (err)
 515                 GOTO(out_inode, err);
 516
 517         dentry.d_inode = inode;
 518
 519         if (sbi->ll_flags & LL_SBI_FILE_SECCTX) {
 520                 /* no need to protect selinux_inode_setsecurity() by
 521                  * inode_lock. Taking it would lead to a client deadlock
 522                  * LU-13617
 523                  */
 524                 err = security_inode_notifysecctx(inode,
 525                                                   op_data->op_file_secctx,
 526                                                   op_data->op_file_secctx_size);
 527         } else {
 528                 err = ll_inode_init_security(&dentry, inode, parent);
 529         }
 530         if (err)
 531                 GOTO(out_inode, err);
 532
 533         if (encrypt) {
 534                 err = llcrypt_inherit_context(parent, inode, NULL, false);
 535                 if (err)
 536                         GOTO(out_inode, err);
 537         }
 538
 539 out_inode:
 540         if (inode != NULL)
 541                 iput(inode);
 542 out_request:
 543         ptlrpc_req_finished(request);
 544 out_op_data:
 545         ll_finish_md_op_data(op_data);
 546
 547         return err;
 548 }
 549
 550 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 551                      int set_default)
 552 {
 553         struct ll_sb_info *sbi = ll_i2sbi(inode);
 554         struct md_op_data *op_data;
 555         struct ptlrpc_request *req = NULL;
 556         int rc = 0;
 557         int lum_size;
 558         ENTRY;
 559
 560         if (lump != NULL) {
 561                 switch (lump->lmm_magic) {
 562                 case LOV_USER_MAGIC_V1:
 563                         lum_size = sizeof(struct lov_user_md_v1);
 564                         break;
 565                 case LOV_USER_MAGIC_V3:
 566                         lum_size = sizeof(struct lov_user_md_v3);
 567                         break;
 568                 case LOV_USER_MAGIC_COMP_V1:
 569                         lum_size = ((struct lov_comp_md_v1 *)lump)->lcm_size;
 570                         break;
 571                 case LMV_USER_MAGIC:
 572                         if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
 573                                 lustre_swab_lmv_user_md(
 574                                         (struct lmv_user_md *)lump);
 575                         lum_size = sizeof(struct lmv_user_md);
 576                         break;
 577                 case LOV_USER_MAGIC_SPECIFIC: {
 578                         struct lov_user_md_v3 *v3 =
 579                                 (struct lov_user_md_v3 *)lump;
 580                         if (v3->lmm_stripe_count > LOV_MAX_STRIPE_COUNT)
 581                                 RETURN(-EINVAL);
 582                         lum_size = lov_user_md_size(v3->lmm_stripe_count,
 583                                                     LOV_USER_MAGIC_SPECIFIC);
 584                         break;
 585                 }
 586                 default:
 587                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 588                                         " %#08x != %#08x nor %#08x\n",
 589                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 590                                         LOV_USER_MAGIC_V3);
 591                         RETURN(-EINVAL);
 592                 }
 593
 594                 /*
 595                  * This is coming from userspace, so should be in
 596                  * local endian.  But the MDS would like it in little
 597                  * endian, so we swab it before we send it.
 598                  */
 599                 if ((__swab32(lump->lmm_magic) & le32_to_cpu(LOV_MAGIC_MASK)) ==
 600                     le32_to_cpu(LOV_MAGIC_MAGIC))
 601                         lustre_swab_lov_user_md(lump, 0);
 602         } else {
 603                 lum_size = sizeof(struct lov_user_md_v1);
 604         }
 605
 606         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 607                                      LUSTRE_OPC_ANY, NULL);
 608         if (IS_ERR(op_data))
 609                 RETURN(PTR_ERR(op_data));
 610
 611         /* swabbing is done in lov_setstripe() on server side */
 612         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
 613         ll_finish_md_op_data(op_data);
 614         ptlrpc_req_finished(req);
 615         if (rc)
 616                 RETURN(rc);
 617
 618         RETURN(rc);
 619 }
 620
 621 static int ll_dir_get_default_layout(struct inode *inode, void **plmm,
 622                                      int *plmm_size,
 623                                      struct ptlrpc_request **request, u64 valid,
 624                                      enum get_default_layout_type type)
 625 {
 626         struct ll_sb_info *sbi = ll_i2sbi(inode);
 627         struct mdt_body   *body;
 628         struct lov_mds_md *lmm = NULL;
 629         struct ptlrpc_request *req = NULL;
 630         int rc, lmm_size;
 631         struct md_op_data *op_data;
 632         struct lu_fid fid;
 633         ENTRY;
 634
 635         rc = ll_get_default_mdsize(sbi, &lmm_size);
 636         if (rc)
 637                 RETURN(rc);
 638
 639         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 640                                      0, lmm_size, LUSTRE_OPC_ANY,
 641                                      NULL);
 642         if (IS_ERR(op_data))
 643                 RETURN(PTR_ERR(op_data));
 644
 645         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 646
 647         if (type == GET_DEFAULT_LAYOUT_ROOT) {
 648                 lu_root_fid(&op_data->op_fid1);
 649                 fid = op_data->op_fid1;
 650         } else {
 651                 fid = *ll_inode2fid(inode);
 652         }
 653
 654         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 655         ll_finish_md_op_data(op_data);
 656         if (rc < 0) {
 657                 CDEBUG(D_INFO, "md_getattr failed on inode "DFID": rc %d\n",
 658                        PFID(&fid), rc);
 659                 GOTO(out, rc);
 660         }
 661
 662         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 663         LASSERT(body != NULL);
 664
 665         lmm_size = body->mbo_eadatasize;
 666
 667         if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 668             lmm_size == 0) {
 669                 GOTO(out, rc = -ENODATA);
 670         }
 671
 672         lmm = req_capsule_server_sized_get(&req->rq_pill,
 673                                            &RMF_MDT_MD, lmm_size);
 674         LASSERT(lmm != NULL);
 675
 676         /*
 677          * This is coming from the MDS, so is probably in
 678          * little endian.  We convert it to host endian before
 679          * passing it to userspace.
 680          */
 681         /* We don't swab objects for directories */
 682         switch (le32_to_cpu(lmm->lmm_magic)) {
 683         case LOV_MAGIC_V1:
 684         case LOV_MAGIC_V3:
 685         case LOV_MAGIC_COMP_V1:
 686         case LOV_USER_MAGIC_SPECIFIC:
 687                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 688                         lustre_swab_lov_user_md((struct lov_user_md *)lmm, 0);
 689                 break;
 690         case LMV_MAGIC_V1:
 691                 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
 692                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 693                 break;
 694         case LMV_USER_MAGIC:
 695                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 696                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 697                 break;
 698         case LMV_MAGIC_FOREIGN: {
 699                 struct lmv_foreign_md *lfm = (struct lmv_foreign_md *)lmm;
 700
 701                 if (LMV_MAGIC_FOREIGN != cpu_to_le32(LMV_MAGIC_FOREIGN)) {
 702                         __swab32s(&lfm->lfm_magic);
 703                         __swab32s(&lfm->lfm_length);
 704                         __swab32s(&lfm->lfm_type);
 705                         __swab32s(&lfm->lfm_flags);
 706                 }
 707                 break;
 708         }
 709         default:
 710                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 711                 rc = -EPROTO;
 712         }
 713 out:
 714         *plmm = lmm;
 715         *plmm_size = lmm_size;
 716         *request = req;
 717         return rc;
 718 }
 719
 720 /**
 721  * This function will be used to get default LOV/LMV/Default LMV
 722  * @valid will be used to indicate which stripe it will retrieve.
 723  * If the directory does not have its own default layout, then the
 724  * function will request the default layout from root FID.
 725  *      OBD_MD_MEA              LMV stripe EA
 726  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 727  *      otherwise               Default LOV EA.
 728  * Each time, it can only retrieve 1 stripe EA
 729  **/
 730 int ll_dir_getstripe_default(struct inode *inode, void **plmm, int *plmm_size,
 731                              struct ptlrpc_request **request,
 732                              struct ptlrpc_request **root_request,
 733                              u64 valid)
 734 {
 735         struct ptlrpc_request *req = NULL;
 736         struct ptlrpc_request *root_req = NULL;
 737         struct lov_mds_md *lmm = NULL;
 738         int lmm_size = 0;
 739         int rc = 0;
 740         ENTRY;
 741
 742         rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
 743                                        &req, valid, 0);
 744         if (rc == -ENODATA && !fid_is_root(ll_inode2fid(inode)) &&
 745             !(valid & (OBD_MD_MEA|OBD_MD_DEFAULT_MEA)) && root_request != NULL)
 746                 rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
 747                                                &root_req, valid,
 748                                                GET_DEFAULT_LAYOUT_ROOT);
 749
 750         *plmm = lmm;
 751         *plmm_size = lmm_size;
 752         *request = req;
 753         if (root_request != NULL)
 754                 *root_request = root_req;
 755
 756         RETURN(rc);
 757 }
 758
 759 /**
 760  * This function will be used to get default LOV/LMV/Default LMV
 761  * @valid will be used to indicate which stripe it will retrieve
 762  *      OBD_MD_MEA              LMV stripe EA
 763  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 764  *      otherwise               Default LOV EA.
 765  * Each time, it can only retrieve 1 stripe EA
 766  **/
 767 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 768                      struct ptlrpc_request **request, u64 valid)
 769 {
 770         struct ptlrpc_request *req = NULL;
 771         struct lov_mds_md *lmm = NULL;
 772         int lmm_size = 0;
 773         int rc = 0;
 774         ENTRY;
 775
 776         rc = ll_dir_get_default_layout(inode, (void **)&lmm, &lmm_size,
 777                                        &req, valid, 0);
 778
 779         *plmm = lmm;
 780         *plmm_size = lmm_size;
 781         *request = req;
 782
 783         RETURN(rc);
 784 }
 785
 786 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 787 {
 788         struct md_op_data       *op_data;
 789         int                     rc;
 790         int                     mdt_index;
 791         ENTRY;
 792
 793         OBD_ALLOC_PTR(op_data);
 794         if (op_data == NULL)
 795                 RETURN(-ENOMEM);
 796
 797         op_data->op_flags |= MF_GET_MDT_IDX;
 798         op_data->op_fid1 = *fid;
 799         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 800         mdt_index = op_data->op_mds;
 801         OBD_FREE_PTR(op_data);
 802         if (rc < 0)
 803                 RETURN(rc);
 804
 805         RETURN(mdt_index);
 806 }
 807
 808 /*
 809  *  Get MDT index for the inode.
 810  */
 811 int ll_get_mdt_idx(struct inode *inode)
 812 {
 813         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 814 }
 815
 816 /**
 817  * Generic handler to do any pre-copy work.
 818  *
 819  * It sends a first hsm_progress (with extent length == 0) to coordinator as a
 820  * first information for it that real work has started.
 821  *
 822  * Moreover, for a ARCHIVE request, it will sample the file data version and
 823  * store it in \a copy.
 824  *
 825  * \return 0 on success.
 826  */
 827 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 828 {
 829         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 830         struct hsm_progress_kernel       hpk;
 831         int                              rc = 0;
 832         int                              rc2;
 833         ENTRY;
 834
 835         /* Forge a hsm_progress based on data from copy. */
 836         hpk.hpk_fid = copy->hc_hai.hai_fid;
 837         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 838         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 839         hpk.hpk_extent.length = 0;
 840         hpk.hpk_flags = 0;
 841         hpk.hpk_errval = 0;
 842         hpk.hpk_data_version = 0;
 843
 844
 845         /* For archive request, we need to read the current file version. */
 846         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 847                 struct inode    *inode;
 848                 __u64            data_version = 0;
 849
 850                 /* Get inode for this fid */
 851                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 852                 if (IS_ERR(inode)) {
 853                         hpk.hpk_flags |= HP_FLAG_RETRY;
 854                         /* hpk_errval is >= 0 */
 855                         hpk.hpk_errval = -PTR_ERR(inode);
 856                         GOTO(progress, rc = PTR_ERR(inode));
 857                 }
 858
 859                 /* Read current file data version */
 860                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 861                 iput(inode);
 862                 if (rc != 0) {
 863                         CDEBUG(D_HSM, "Could not read file data version of "
 864                                       DFID" (rc = %d). Archive request ("
 865                                       "%#llx) could not be done.\n",
 866                                       PFID(&copy->hc_hai.hai_fid), rc,
 867                                       copy->hc_hai.hai_cookie);
 868                         hpk.hpk_flags |= HP_FLAG_RETRY;
 869                         /* hpk_errval must be >= 0 */
 870                         hpk.hpk_errval = -rc;
 871                         GOTO(progress, rc);
 872                 }
 873
 874                 /* Store in the hsm_copy for later copytool use.
 875                  * Always modified even if no lsm. */
 876                 copy->hc_data_version = data_version;
 877         }
 878
 879 progress:
 880         /* On error, the request should be considered as completed */
 881         if (hpk.hpk_errval > 0)
 882                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 883
 884         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 885                             &hpk, NULL);
 886
 887         /* Return first error */
 888         RETURN(rc != 0 ? rc : rc2);
 889 }
 890
 891 /**
 892  * Generic handler to do any post-copy work.
 893  *
 894  * It will send the last hsm_progress update to coordinator to inform it
 895  * that copy is finished and whether it was successful or not.
 896  *
 897  * Moreover,
 898  * - for ARCHIVE request, it will sample the file data version and compare it
 899  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 900  *   will be considered as failed.
 901  * - for RESTORE request, it will sample the file data version and send it to
 902  *   coordinator which is useful if the file was imported as 'released'.
 903  *
 904  * \return 0 on success.
 905  */
 906 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 907 {
 908         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 909         struct hsm_progress_kernel       hpk;
 910         int                              rc = 0;
 911         int                              rc2;
 912         ENTRY;
 913
 914         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 915         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 916          * initialized if copy_end was called with copy == NULL.
 917          */
 918
 919         /* Forge a hsm_progress based on data from copy. */
 920         hpk.hpk_fid = copy->hc_hai.hai_fid;
 921         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 922         hpk.hpk_extent = copy->hc_hai.hai_extent;
 923         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 924         hpk.hpk_errval = copy->hc_errval;
 925         hpk.hpk_data_version = 0;
 926
 927         /* For archive request, we need to check the file data was not changed.
 928          *
 929          * For restore request, we need to send the file data version, this is
 930          * useful when the file was created using hsm_import.
 931          */
 932         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 933              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 934             (copy->hc_errval == 0)) {
 935                 struct inode    *inode;
 936                 __u64            data_version = 0;
 937
 938                 /* Get lsm for this fid */
 939                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 940                 if (IS_ERR(inode)) {
 941                         hpk.hpk_flags |= HP_FLAG_RETRY;
 942                         /* hpk_errval must be >= 0 */
 943                         hpk.hpk_errval = -PTR_ERR(inode);
 944                         GOTO(progress, rc = PTR_ERR(inode));
 945                 }
 946
 947                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 948                 iput(inode);
 949                 if (rc) {
 950                         CDEBUG(D_HSM, "Could not read file data version. "
 951                                       "Request could not be confirmed.\n");
 952                         if (hpk.hpk_errval == 0)
 953                                 hpk.hpk_errval = -rc;
 954                         GOTO(progress, rc);
 955                 }
 956
 957                 /* Store in the hsm_copy for later copytool use.
 958                  * Always modified even if no lsm. */
 959                 hpk.hpk_data_version = data_version;
 960
 961                 /* File could have been stripped during archiving, so we need
 962                  * to check anyway. */
 963                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 964                     (copy->hc_data_version != data_version)) {
 965                         CDEBUG(D_HSM, "File data version mismatched. "
 966                               "File content was changed during archiving. "
 967                                DFID", start:%#llx current:%#llx\n",
 968                                PFID(&copy->hc_hai.hai_fid),
 969                                copy->hc_data_version, data_version);
 970                         /* File was changed, send error to cdt. Do not ask for
 971                          * retry because if a file is modified frequently,
 972                          * the cdt will loop on retried archive requests.
 973                          * The policy engine will ask for a new archive later
 974                          * when the file will not be modified for some tunable
 975                          * time */
 976                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 977                         rc = -EBUSY;
 978                         /* hpk_errval must be >= 0 */
 979                         hpk.hpk_errval = -rc;
 980                         GOTO(progress, rc);
 981                 }
 982
 983         }
 984
 985 progress:
 986         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 987                             &hpk, NULL);
 988
 989         /* Return first error */
 990         RETURN(rc != 0 ? rc : rc2);
 991 }
 992
 993
 994 static int copy_and_ct_start(int cmd, struct obd_export *exp,
 995                              const struct lustre_kernelcomm __user *data)
 996 {
 997         struct lustre_kernelcomm *lk;
 998         struct lustre_kernelcomm *tmp;
 999         size_t size = sizeof(*lk);
1000         size_t new_size;
1001         int i;
1002         int rc;
1003
1004         /* copy data from userspace to get numbers of archive_id */
1005         OBD_ALLOC(lk, size);
1006         if (lk == NULL)
1007                 return -ENOMEM;
1008
1009         if (copy_from_user(lk, data, size))
1010                 GOTO(out_lk, rc = -EFAULT);
1011
1012         if (lk->lk_flags & LK_FLG_STOP)
1013                 goto do_ioctl;
1014
1015         if (!(lk->lk_flags & LK_FLG_DATANR)) {
1016                 __u32 archive_mask = lk->lk_data_count;
1017                 int count;
1018
1019                 /* old hsm agent to old MDS */
1020                 if (!exp_connect_archive_id_array(exp))
1021                         goto do_ioctl;
1022
1023                 /* old hsm agent to new MDS */
1024                 lk->lk_flags |= LK_FLG_DATANR;
1025
1026                 if (archive_mask == 0)
1027                         goto do_ioctl;
1028
1029                 count = hweight32(archive_mask);
1030                 new_size = offsetof(struct lustre_kernelcomm, lk_data[count]);
1031                 OBD_ALLOC(tmp, new_size);
1032                 if (tmp == NULL)
1033                         GOTO(out_lk, rc = -ENOMEM);
1034
1035                 memcpy(tmp, lk, size);
1036                 tmp->lk_data_count = count;
1037                 OBD_FREE(lk, size);
1038                 lk = tmp;
1039                 size = new_size;
1040
1041                 count = 0;
1042                 for (i = 0; i < sizeof(archive_mask) * 8; i++) {
1043                         if (BIT(i) & archive_mask) {
1044                                 lk->lk_data[count] = i + 1;
1045                                 count++;
1046                         }
1047                 }
1048                 goto do_ioctl;
1049         }
1050
1051         /* new hsm agent to new mds */
1052         if (lk->lk_data_count > 0) {
1053                 new_size = offsetof(struct lustre_kernelcomm,
1054                                     lk_data[lk->lk_data_count]);
1055                 OBD_ALLOC(tmp, new_size);
1056                 if (tmp == NULL)
1057                         GOTO(out_lk, rc = -ENOMEM);
1058
1059                 OBD_FREE(lk, size);
1060                 lk = tmp;
1061                 size = new_size;
1062
1063                 if (copy_from_user(lk, data, size))
1064                         GOTO(out_lk, rc = -EFAULT);
1065         }
1066
1067         /* new hsm agent to old MDS */
1068         if (!exp_connect_archive_id_array(exp)) {
1069                 __u32 archives = 0;
1070
1071                 if (lk->lk_data_count > LL_HSM_ORIGIN_MAX_ARCHIVE)
1072                         GOTO(out_lk, rc = -EINVAL);
1073
1074                 for (i = 0; i < lk->lk_data_count; i++) {
1075                         if (lk->lk_data[i] > LL_HSM_ORIGIN_MAX_ARCHIVE) {
1076                                 rc = -EINVAL;
1077                                 CERROR("%s: archive id %d requested but only "
1078                                        "[0 - %zu] supported: rc = %d\n",
1079                                        exp->exp_obd->obd_name, lk->lk_data[i],
1080                                        LL_HSM_ORIGIN_MAX_ARCHIVE, rc);
1081                                 GOTO(out_lk, rc);
1082                         }
1083
1084                         if (lk->lk_data[i] == 0) {
1085                                 archives = 0;
1086                                 break;
1087                         }
1088
1089                         archives |= (1 << (lk->lk_data[i] - 1));
1090                 }
1091                 lk->lk_flags &= ~LK_FLG_DATANR;
1092                 lk->lk_data_count = archives;
1093         }
1094 do_ioctl:
1095         rc = obd_iocontrol(cmd, exp, size, lk, NULL);
1096 out_lk:
1097         OBD_FREE(lk, size);
1098         return rc;
1099 }
1100
1101 static int check_owner(int type, int id)
1102 {
1103         switch (type) {
1104         case USRQUOTA:
1105                 if (!uid_eq(current_euid(), make_kuid(&init_user_ns, id)))
1106                         return -EPERM;
1107                 break;
1108         case GRPQUOTA:
1109                 if (!in_egroup_p(make_kgid(&init_user_ns, id)))
1110                         return -EPERM;
1111                 break;
1112         case PRJQUOTA:
1113                 break;
1114         }
1115         return 0;
1116 }
1117
1118 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
1119 {
1120         int cmd = qctl->qc_cmd;
1121         int type = qctl->qc_type;
1122         int id = qctl->qc_id;
1123         int valid = qctl->qc_valid;
1124         int rc = 0;
1125         ENTRY;
1126
1127         switch (cmd) {
1128         case Q_SETQUOTA:
1129         case Q_SETINFO:
1130         case LUSTRE_Q_SETDEFAULT:
1131         case LUSTRE_Q_SETQUOTAPOOL:
1132         case LUSTRE_Q_SETINFOPOOL:
1133                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1134                         RETURN(-EPERM);
1135                 break;
1136         case Q_GETQUOTA:
1137         case LUSTRE_Q_GETDEFAULT:
1138         case LUSTRE_Q_GETQUOTAPOOL:
1139                 if (check_owner(type, id) &&
1140                     (!cfs_capable(CFS_CAP_SYS_ADMIN)))
1141                         RETURN(-EPERM);
1142                 break;
1143         case Q_GETINFO:
1144         case LUSTRE_Q_GETINFOPOOL:
1145                 break;
1146         default:
1147                 CERROR("unsupported quotactl op: %#x\n", cmd);
1148                 RETURN(-ENOTSUPP);
1149         }
1150
1151         if (valid != QC_GENERAL) {
1152                 if (cmd == Q_GETINFO)
1153                         qctl->qc_cmd = Q_GETOINFO;
1154                 else if (cmd == Q_GETQUOTA ||
1155                          cmd == LUSTRE_Q_GETQUOTAPOOL)
1156                         qctl->qc_cmd = Q_GETOQUOTA;
1157                 else
1158                         RETURN(-EINVAL);
1159
1160                 switch (valid) {
1161                 case QC_MDTIDX:
1162                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1163                                            sizeof(*qctl), qctl, NULL);
1164                         break;
1165                 case QC_OSTIDX:
1166                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
1167                                            sizeof(*qctl), qctl, NULL);
1168                         break;
1169                 case QC_UUID:
1170                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
1171                                            sizeof(*qctl), qctl, NULL);
1172                         if (rc == -EAGAIN)
1173                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
1174                                                    sbi->ll_dt_exp,
1175                                                    sizeof(*qctl), qctl, NULL);
1176                         break;
1177                 default:
1178                         rc = -EINVAL;
1179                         break;
1180                 }
1181
1182                 if (rc)
1183                         RETURN(rc);
1184
1185                 qctl->qc_cmd = cmd;
1186         } else {
1187                 struct obd_quotactl *oqctl;
1188                 int oqctl_len = sizeof(*oqctl);
1189
1190                 if (LUSTRE_Q_CMD_IS_POOL(cmd))
1191                         oqctl_len += LOV_MAXPOOLNAME + 1;
1192
1193                 OBD_ALLOC(oqctl, oqctl_len);
1194                 if (oqctl == NULL)
1195                         RETURN(-ENOMEM);
1196
1197                 QCTL_COPY(oqctl, qctl);
1198                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
1199                 if (rc) {
1200                         OBD_FREE(oqctl, oqctl_len);
1201                         RETURN(rc);
1202                 }
1203                 /* If QIF_SPACE is not set, client should collect the
1204                  * space usage from OSSs by itself */
1205                 if ((cmd == Q_GETQUOTA || cmd == LUSTRE_Q_GETQUOTAPOOL) &&
1206                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1207                     !oqctl->qc_dqblk.dqb_curspace) {
1208                         struct obd_quotactl *oqctl_tmp;
1209
1210                         OBD_ALLOC_PTR(oqctl_tmp);
1211                         if (oqctl_tmp == NULL)
1212                                 GOTO(out, rc = -ENOMEM);
1213
1214                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1215                         oqctl_tmp->qc_id = oqctl->qc_id;
1216                         oqctl_tmp->qc_type = oqctl->qc_type;
1217
1218                         /* collect space usage from OSTs */
1219                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1220                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1221                         if (!rc || rc == -EREMOTEIO) {
1222                                 oqctl->qc_dqblk.dqb_curspace =
1223                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1224                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1225                         }
1226
1227                         /* collect space & inode usage from MDTs */
1228                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1229                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1230                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1231                         if (!rc || rc == -EREMOTEIO) {
1232                                 oqctl->qc_dqblk.dqb_curspace +=
1233                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1234                                 oqctl->qc_dqblk.dqb_curinodes =
1235                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1236                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1237                         } else {
1238                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1239                         }
1240
1241                         OBD_FREE_PTR(oqctl_tmp);
1242                 }
1243 out:
1244                 QCTL_COPY(qctl, oqctl);
1245                 OBD_FREE(oqctl, oqctl_len);
1246         }
1247
1248         RETURN(rc);
1249 }
1250
1251 int ll_rmfid(struct file *file, void __user *arg)
1252 {
1253         const struct fid_array __user *ufa = arg;
1254         struct fid_array *lfa = NULL;
1255         size_t size;
1256         unsigned nr;
1257         int i, rc, *rcs = NULL;
1258         ENTRY;
1259
1260         if (!cfs_capable(CFS_CAP_DAC_READ_SEARCH) &&
1261             !(ll_i2sbi(file_inode(file))->ll_flags & LL_SBI_USER_FID2PATH))
1262                 RETURN(-EPERM);
1263         /* Only need to get the buflen */
1264         if (get_user(nr, &ufa->fa_nr))
1265                 RETURN(-EFAULT);
1266         /* DoS protection */
1267         if (nr > OBD_MAX_FIDS_IN_ARRAY)
1268                 RETURN(-E2BIG);
1269
1270         size = offsetof(struct fid_array, fa_fids[nr]);
1271         OBD_ALLOC(lfa, size);
1272         if (!lfa)
1273                 RETURN(-ENOMEM);
1274         OBD_ALLOC_PTR_ARRAY(rcs, nr);
1275         if (!rcs)
1276                 GOTO(free_lfa, rc = -ENOMEM);
1277
1278         if (copy_from_user(lfa, arg, size))
1279                 GOTO(free_rcs, rc = -EFAULT);
1280
1281         /* Call mdc_iocontrol */
1282         rc = md_rmfid(ll_i2mdexp(file_inode(file)), lfa, rcs, NULL);
1283         if (!rc) {
1284                 for (i = 0; i < nr; i++)
1285                         if (rcs[i])
1286                                 lfa->fa_fids[i].f_ver = rcs[i];
1287                 if (copy_to_user(arg, lfa, size))
1288                         rc = -EFAULT;
1289         }
1290
1291 free_rcs:
1292         OBD_FREE_PTR_ARRAY(rcs, nr);
1293 free_lfa:
1294         OBD_FREE(lfa, size);
1295
1296         RETURN(rc);
1297 }
1298
1299 /* This function tries to get a single name component,
1300  * to send to the server. No actual path traversal involved,
1301  * so we limit to NAME_MAX */
1302 static char *ll_getname(const char __user *filename)
1303 {
1304         int ret = 0, len;
1305         char *tmp;
1306
1307         OBD_ALLOC(tmp, NAME_MAX + 1);
1308
1309         if (!tmp)
1310                 return ERR_PTR(-ENOMEM);
1311
1312         len = strncpy_from_user(tmp, filename, NAME_MAX + 1);
1313         if (len < 0)
1314                 ret = -ENOENT;
1315         else if (len > NAME_MAX)
1316                 ret = -ENAMETOOLONG;
1317
1318         if (ret) {
1319                 OBD_FREE(tmp, NAME_MAX + 1);
1320                 tmp =  ERR_PTR(ret);
1321         }
1322         return tmp;
1323 }
1324
1325 #define ll_putname(filename) OBD_FREE(filename, NAME_MAX + 1);
1326
1327 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1328 {
1329         struct dentry *dentry = file_dentry(file);
1330         struct inode *inode = file_inode(file);
1331         struct ll_sb_info *sbi = ll_i2sbi(inode);
1332         struct obd_ioctl_data *data;
1333         int rc = 0;
1334         ENTRY;
1335
1336         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1337                PFID(ll_inode2fid(inode)), inode, cmd);
1338
1339         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1340         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1341                 return -ENOTTY;
1342
1343         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1344         switch (cmd) {
1345         case FS_IOC_GETFLAGS:
1346         case FS_IOC_SETFLAGS:
1347                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1348         case FSFILT_IOC_GETVERSION:
1349         case FS_IOC_GETVERSION:
1350                 RETURN(put_user(inode->i_generation, (int __user *)arg));
1351         /* We need to special case any other ioctls we want to handle,
1352          * to send them to the MDS/OST as appropriate and to properly
1353          * network encode the arg field. */
1354         case FS_IOC_SETVERSION:
1355                 RETURN(-ENOTSUPP);
1356
1357         case LL_IOC_GET_MDTIDX: {
1358                 int mdtidx;
1359
1360                 mdtidx = ll_get_mdt_idx(inode);
1361                 if (mdtidx < 0)
1362                         RETURN(mdtidx);
1363
1364                 if (put_user((int)mdtidx, (int __user *)arg))
1365                         RETURN(-EFAULT);
1366
1367                 return 0;
1368         }
1369         case IOC_MDC_LOOKUP: {
1370                                      int namelen, len = 0;
1371                 char *buf = NULL;
1372                 char *filename;
1373
1374                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1375                 if (rc != 0)
1376                         RETURN(rc);
1377                 data = (void *)buf;
1378
1379                 filename = data->ioc_inlbuf1;
1380                 namelen = strlen(filename);
1381                 if (namelen < 1) {
1382                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1383                         GOTO(out_free, rc = -EINVAL);
1384                 }
1385
1386                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL, NULL);
1387                 if (rc < 0) {
1388                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1389                                sbi->ll_fsname, namelen, filename, rc);
1390                         GOTO(out_free, rc);
1391                 }
1392 out_free:
1393                 OBD_FREE_LARGE(buf, len);
1394                 return rc;
1395         }
1396         case LL_IOC_LMV_SETSTRIPE: {
1397                 struct lmv_user_md  *lum;
1398                 char            *buf = NULL;
1399                 char            *filename;
1400                 int              namelen = 0;
1401                 int              lumlen = 0;
1402                 umode_t          mode;
1403                 int              len;
1404                 int              rc;
1405
1406                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1407                 if (rc)
1408                         RETURN(rc);
1409
1410                 data = (void *)buf;
1411                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1412                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1413                         GOTO(lmv_out_free, rc = -EINVAL);
1414
1415                 filename = data->ioc_inlbuf1;
1416                 namelen = data->ioc_inllen1;
1417
1418                 if (namelen < 1) {
1419                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1420                         GOTO(lmv_out_free, rc = -EINVAL);
1421                 }
1422                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1423                 lumlen = data->ioc_inllen2;
1424
1425                 if (!lmv_user_magic_supported(lum->lum_magic)) {
1426                         CERROR("%s: wrong lum magic %x : rc = %d\n", filename,
1427                                lum->lum_magic, -EINVAL);
1428                         GOTO(lmv_out_free, rc = -EINVAL);
1429                 }
1430
1431                 if ((lum->lum_magic == LMV_USER_MAGIC ||
1432                      lum->lum_magic == LMV_USER_MAGIC_SPECIFIC) &&
1433                     lumlen < sizeof(*lum)) {
1434                         CERROR("%s: wrong lum size %d for magic %x : rc = %d\n",
1435                                filename, lumlen, lum->lum_magic, -EINVAL);
1436                         GOTO(lmv_out_free, rc = -EINVAL);
1437                 }
1438
1439                 if (lum->lum_magic == LMV_MAGIC_FOREIGN &&
1440                     lumlen < sizeof(struct lmv_foreign_md)) {
1441                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1442                                filename, lum->lum_magic, lumlen, -EFAULT);
1443                         GOTO(lmv_out_free, rc = -EINVAL);
1444                 }
1445
1446                 mode = data->ioc_type;
1447                 rc = ll_dir_setdirstripe(dentry, lum, lumlen, filename, mode);
1448 lmv_out_free:
1449                 OBD_FREE_LARGE(buf, len);
1450                 RETURN(rc);
1451
1452         }
1453         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1454                 struct lmv_user_md        lum;
1455                 struct lmv_user_md __user *ulump =
1456                                         (struct lmv_user_md __user *)arg;
1457                 int                       rc;
1458
1459                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1460                         RETURN(-EFAULT);
1461
1462                 if (lum.lum_magic != LMV_USER_MAGIC)
1463                         RETURN(-EINVAL);
1464
1465                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1466
1467                 RETURN(rc);
1468         }
1469         case LL_IOC_LOV_SETSTRIPE_NEW:
1470         case LL_IOC_LOV_SETSTRIPE: {
1471                 struct lov_user_md_v3 *lumv3 = NULL;
1472                 struct lov_user_md_v1 lumv1;
1473                 struct lov_user_md_v1 *lumv1_ptr = &lumv1;
1474                 struct lov_user_md_v1 __user *lumv1p =
1475                         (struct lov_user_md_v1 __user *)arg;
1476                 struct lov_user_md_v3 __user *lumv3p =
1477                         (struct lov_user_md_v3 __user *)arg;
1478                 int lum_size = 0;
1479
1480                 int set_default = 0;
1481
1482                 BUILD_BUG_ON(sizeof(struct lov_user_md_v3) <=
1483                              sizeof(struct lov_comp_md_v1));
1484                 BUILD_BUG_ON(sizeof(*lumv3) != sizeof(*lumv3p));
1485                 /* first try with v1 which is smaller than v3 */
1486                 if (copy_from_user(&lumv1, lumv1p, sizeof(lumv1)))
1487                         RETURN(-EFAULT);
1488
1489                 if (inode->i_sb->s_root == file_dentry(file))
1490                         set_default = 1;
1491
1492                 switch (lumv1.lmm_magic) {
1493                 case LOV_USER_MAGIC_V3:
1494                 case LOV_USER_MAGIC_SPECIFIC:
1495                         lum_size = ll_lov_user_md_size(&lumv1);
1496                         if (lum_size < 0)
1497                                 RETURN(lum_size);
1498                         OBD_ALLOC(lumv3, lum_size);
1499                         if (!lumv3)
1500                                 RETURN(-ENOMEM);
1501                         if (copy_from_user(lumv3, lumv3p, lum_size))
1502                                 GOTO(out, rc = -EFAULT);
1503                         lumv1_ptr = (struct lov_user_md_v1 *)lumv3;
1504                         break;
1505                 case LOV_USER_MAGIC_V1:
1506                         break;
1507                 default:
1508                         GOTO(out, rc = -ENOTSUPP);
1509                 }
1510
1511                 /* in v1 and v3 cases lumv1 points to data */
1512                 rc = ll_dir_setstripe(inode, lumv1_ptr, set_default);
1513 out:
1514                 if (lumv3)
1515                         OBD_FREE(lumv3, lum_size);
1516                 RETURN(rc);
1517         }
1518         case LL_IOC_LMV_GETSTRIPE: {
1519                 struct lmv_user_md __user *ulmv =
1520                                         (struct lmv_user_md __user *)arg;
1521                 struct lmv_user_md      lum;
1522                 struct ptlrpc_request   *request = NULL;
1523                 struct ptlrpc_request   *root_request = NULL;
1524                 union lmv_mds_md        *lmm = NULL;
1525                 int                     lmmsize;
1526                 u64                     valid = 0;
1527                 struct lmv_user_md      *tmp = NULL;
1528                 int                     mdt_index;
1529                 int                     lum_size;
1530                 int                     stripe_count;
1531                 int                     max_stripe_count;
1532                 int                     i;
1533                 int                     rc;
1534
1535                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1536                         RETURN(-EFAULT);
1537
1538                 max_stripe_count = lum.lum_stripe_count;
1539                 /* lum_magic will indicate which stripe the ioctl will like
1540                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1541                  * is for default LMV stripe */
1542                 if (lum.lum_magic == LMV_MAGIC_V1)
1543                         valid |= OBD_MD_MEA;
1544                 else if (lum.lum_magic == LMV_USER_MAGIC)
1545                         valid |= OBD_MD_DEFAULT_MEA;
1546                 else
1547                         RETURN(-EINVAL);
1548
1549                 rc = ll_dir_getstripe_default(inode, (void **)&lmm, &lmmsize,
1550                                               &request, &root_request, valid);
1551                 if (rc != 0)
1552                         GOTO(finish_req, rc);
1553
1554                 /* Get default LMV EA */
1555                 if (lum.lum_magic == LMV_USER_MAGIC) {
1556                         if (lmmsize > sizeof(*ulmv))
1557                                 GOTO(finish_req, rc = -EINVAL);
1558
1559                         if (copy_to_user(ulmv, lmm, lmmsize))
1560                                 GOTO(finish_req, rc = -EFAULT);
1561
1562                         GOTO(finish_req, rc);
1563                 }
1564
1565                 /* if foreign LMV case, fake stripes number */
1566                 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1567                         struct lmv_foreign_md *lfm;
1568
1569                         lfm = (struct lmv_foreign_md *)lmm;
1570                         if (lfm->lfm_length < XATTR_SIZE_MAX -
1571                             offsetof(typeof(*lfm), lfm_value)) {
1572                                 __u32 size = lfm->lfm_length +
1573                                              offsetof(typeof(*lfm), lfm_value);
1574
1575                                 stripe_count = lmv_foreign_to_md_stripes(size);
1576                         } else {
1577                                 CERROR("invalid %d foreign size returned\n",
1578                                             lfm->lfm_length);
1579                                 return -EINVAL;
1580                         }
1581                 } else {
1582                         stripe_count = lmv_mds_md_stripe_count_get(lmm);
1583                 }
1584                 if (max_stripe_count < stripe_count) {
1585                         lum.lum_stripe_count = stripe_count;
1586                         if (copy_to_user(ulmv, &lum, sizeof(lum)))
1587                                 GOTO(finish_req, rc = -EFAULT);
1588                         GOTO(finish_req, rc = -E2BIG);
1589                 }
1590
1591                 /* enough room on user side and foreign case */
1592                 if (lmm->lmv_magic == LMV_MAGIC_FOREIGN) {
1593                         struct lmv_foreign_md *lfm;
1594                         __u32 size;
1595
1596                         lfm = (struct lmv_foreign_md *)lmm;
1597                         size = lfm->lfm_length +
1598                                offsetof(struct lmv_foreign_md, lfm_value);
1599                         if (copy_to_user(ulmv, lfm, size))
1600                                 GOTO(finish_req, rc = -EFAULT);
1601                         GOTO(finish_req, rc);
1602                 }
1603
1604                 lum_size = lmv_user_md_size(stripe_count,
1605                                             LMV_USER_MAGIC_SPECIFIC);
1606                 OBD_ALLOC(tmp, lum_size);
1607                 if (tmp == NULL)
1608                         GOTO(finish_req, rc = -ENOMEM);
1609
1610                 mdt_index = ll_get_mdt_idx(inode);
1611                 if (mdt_index < 0)
1612                         GOTO(out_tmp, rc = -ENOMEM);
1613
1614                 tmp->lum_magic = LMV_MAGIC_V1;
1615                 tmp->lum_stripe_count = 0;
1616                 tmp->lum_stripe_offset = mdt_index;
1617                 tmp->lum_hash_type = lmv_mds_md_hash_type_get(lmm);
1618                 for (i = 0; i < stripe_count; i++) {
1619                         struct lu_fid   fid;
1620
1621                         fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1622                         if (fid_is_sane(&fid)) {
1623                                 mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1624                                 if (mdt_index < 0)
1625                                         GOTO(out_tmp, rc = mdt_index);
1626
1627                                 tmp->lum_objects[i].lum_mds = mdt_index;
1628                                 tmp->lum_objects[i].lum_fid = fid;
1629                         }
1630
1631                         tmp->lum_stripe_count++;
1632                 }
1633
1634                 if (copy_to_user(ulmv, tmp, lum_size))
1635                         GOTO(out_tmp, rc = -EFAULT);
1636 out_tmp:
1637                 OBD_FREE(tmp, lum_size);
1638 finish_req:
1639                 ptlrpc_req_finished(request);
1640                 ptlrpc_req_finished(root_request);
1641                 return rc;
1642         }
1643
1644         case LL_IOC_REMOVE_ENTRY: {
1645                 char            *filename = NULL;
1646                 int              namelen = 0;
1647                 int              rc;
1648
1649                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1650                  * unsupported server, which might crash the server(LU-2730),
1651                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1652                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1653                  * server will support REINT_RMENTRY XXX*/
1654                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1655                         RETURN(-ENOTSUPP);
1656
1657                 filename = ll_getname((const char __user *)arg);
1658                 if (IS_ERR(filename))
1659                         RETURN(PTR_ERR(filename));
1660
1661                 namelen = strlen(filename);
1662                 if (namelen < 1)
1663                         GOTO(out_rmdir, rc = -EINVAL);
1664
1665                 rc = ll_rmdir_entry(inode, filename, namelen);
1666 out_rmdir:
1667                 if (filename)
1668                         ll_putname(filename);
1669                 RETURN(rc);
1670         }
1671         case LL_IOC_RMFID:
1672                 RETURN(ll_rmfid(file, (void __user *)arg));
1673         case LL_IOC_LOV_SWAP_LAYOUTS:
1674                 RETURN(-EPERM);
1675         case IOC_OBD_STATFS:
1676                 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1677         case LL_IOC_LOV_GETSTRIPE:
1678         case LL_IOC_LOV_GETSTRIPE_NEW:
1679         case LL_IOC_MDC_GETINFO:
1680         case LL_IOC_MDC_GETINFO_OLD:
1681         case IOC_MDC_GETFILEINFO:
1682         case IOC_MDC_GETFILEINFO_OLD:
1683         case IOC_MDC_GETFILESTRIPE: {
1684                 struct ptlrpc_request *request = NULL;
1685                 struct ptlrpc_request *root_request = NULL;
1686                 struct lov_user_md __user *lump;
1687                 struct lov_mds_md *lmm = NULL;
1688                 struct mdt_body *body;
1689                 char *filename = NULL;
1690                 lstat_t __user *statp = NULL;
1691                 lstatx_t __user *stxp = NULL;
1692                 __u64 __user *flagsp = NULL;
1693                 __u32 __user *lmmsizep = NULL;
1694                 struct lu_fid __user *fidp = NULL;
1695                 int lmmsize;
1696
1697                 if (cmd == IOC_MDC_GETFILEINFO_OLD ||
1698                     cmd == IOC_MDC_GETFILEINFO ||
1699                     cmd == IOC_MDC_GETFILESTRIPE) {
1700                         filename = ll_getname((const char __user *)arg);
1701                         if (IS_ERR(filename))
1702                                 RETURN(PTR_ERR(filename));
1703
1704                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1705                                                       &lmmsize, &request);
1706                 } else {
1707                         rc = ll_dir_getstripe_default(inode, (void **)&lmm,
1708                                                       &lmmsize, &request,
1709                                                       &root_request, 0);
1710                 }
1711
1712                 if (request) {
1713                         body = req_capsule_server_get(&request->rq_pill,
1714                                                       &RMF_MDT_BODY);
1715                         LASSERT(body != NULL);
1716                 } else {
1717                         GOTO(out_req, rc);
1718                 }
1719
1720                 if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1721                                        cmd == LL_IOC_MDC_GETINFO ||
1722                                        cmd == IOC_MDC_GETFILEINFO_OLD ||
1723                                        cmd == LL_IOC_MDC_GETINFO_OLD)) {
1724                         lmmsize = 0;
1725                         rc = 0;
1726                 }
1727
1728                 if (rc < 0)
1729                         GOTO(out_req, rc);
1730
1731                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1732                     cmd == LL_IOC_LOV_GETSTRIPE ||
1733                     cmd == LL_IOC_LOV_GETSTRIPE_NEW) {
1734                         lump = (struct lov_user_md __user *)arg;
1735                 } else if (cmd == IOC_MDC_GETFILEINFO_OLD ||
1736                            cmd == LL_IOC_MDC_GETINFO_OLD){
1737                         struct lov_user_mds_data_v1 __user *lmdp;
1738
1739                         lmdp = (struct lov_user_mds_data_v1 __user *)arg;
1740                         statp = &lmdp->lmd_st;
1741                         lump = &lmdp->lmd_lmm;
1742                 } else {
1743                         struct lov_user_mds_data __user *lmdp;
1744
1745                         lmdp = (struct lov_user_mds_data __user *)arg;
1746                         fidp = &lmdp->lmd_fid;
1747                         stxp = &lmdp->lmd_stx;
1748                         flagsp = &lmdp->lmd_flags;
1749                         lmmsizep = &lmdp->lmd_lmmsize;
1750                         lump = &lmdp->lmd_lmm;
1751                 }
1752
1753                 if (lmmsize == 0) {
1754                         /* If the file has no striping then zero out *lump so
1755                          * that the caller isn't confused by garbage. */
1756                         if (clear_user(lump, sizeof(*lump)))
1757                                 GOTO(out_req, rc = -EFAULT);
1758                 } else if (copy_to_user(lump, lmm, lmmsize)) {
1759                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1760                                 GOTO(out_req, rc = -EFAULT);
1761                         rc = -EOVERFLOW;
1762                 }
1763
1764                 if (cmd == IOC_MDC_GETFILEINFO_OLD ||
1765                     cmd == LL_IOC_MDC_GETINFO_OLD) {
1766                         lstat_t st = { 0 };
1767
1768                         st.st_dev       = inode->i_sb->s_dev;
1769                         st.st_mode      = body->mbo_mode;
1770                         st.st_nlink     = body->mbo_nlink;
1771                         st.st_uid       = body->mbo_uid;
1772                         st.st_gid       = body->mbo_gid;
1773                         st.st_rdev      = body->mbo_rdev;
1774                         st.st_size      = body->mbo_size;
1775                         st.st_blksize   = PAGE_SIZE;
1776                         st.st_blocks    = body->mbo_blocks;
1777                         st.st_atime     = body->mbo_atime;
1778                         st.st_mtime     = body->mbo_mtime;
1779                         st.st_ctime     = body->mbo_ctime;
1780                         st.st_ino       = cl_fid_build_ino(&body->mbo_fid1,
1781                                                 sbi->ll_flags &
1782                                                 LL_SBI_32BIT_API);
1783
1784                         if (copy_to_user(statp, &st, sizeof(st)))
1785                                 GOTO(out_req, rc = -EFAULT);
1786                 } else if (cmd == IOC_MDC_GETFILEINFO ||
1787                            cmd == LL_IOC_MDC_GETINFO) {
1788                         lstatx_t stx = { 0 };
1789                         __u64 valid = body->mbo_valid;
1790
1791                         stx.stx_blksize = PAGE_SIZE;
1792                         stx.stx_nlink = body->mbo_nlink;
1793                         stx.stx_uid = body->mbo_uid;
1794                         stx.stx_gid = body->mbo_gid;
1795                         stx.stx_mode = body->mbo_mode;
1796                         stx.stx_ino = cl_fid_build_ino(&body->mbo_fid1,
1797                                                        sbi->ll_flags &
1798                                                        LL_SBI_32BIT_API);
1799                         stx.stx_size = body->mbo_size;
1800                         stx.stx_blocks = body->mbo_blocks;
1801                         stx.stx_atime.tv_sec = body->mbo_atime;
1802                         stx.stx_ctime.tv_sec = body->mbo_ctime;
1803                         stx.stx_mtime.tv_sec = body->mbo_mtime;
1804                         stx.stx_btime.tv_sec = body->mbo_btime;
1805                         stx.stx_rdev_major = MAJOR(body->mbo_rdev);
1806                         stx.stx_rdev_minor = MINOR(body->mbo_rdev);
1807                         stx.stx_dev_major = MAJOR(inode->i_sb->s_dev);
1808                         stx.stx_dev_minor = MINOR(inode->i_sb->s_dev);
1809                         stx.stx_mask |= STATX_BASIC_STATS | STATX_BTIME;
1810
1811                         /*
1812                          * For a striped directory, the size and blocks returned
1813                          * from MDT is not correct.
1814                          * The size and blocks are aggregated by client across
1815                          * all stripes.
1816                          * Thus for a striped directory, do not return the valid
1817                          * FLSIZE and FLBLOCKS flags to the caller.
1818                          * However, this whould be better decided by the MDS
1819                          * instead of the client.
1820                          */
1821                         if (cmd == LL_IOC_MDC_GETINFO &&
1822                             ll_i2info(inode)->lli_lsm_md != NULL)
1823                                 valid &= ~(OBD_MD_FLSIZE | OBD_MD_FLBLOCKS);
1824
1825                         if (flagsp && copy_to_user(flagsp, &valid,
1826                                                    sizeof(*flagsp)))
1827                                 GOTO(out_req, rc = -EFAULT);
1828
1829                         if (fidp && copy_to_user(fidp, &body->mbo_fid1,
1830                                                  sizeof(*fidp)))
1831                                 GOTO(out_req, rc = -EFAULT);
1832
1833                         if (!(valid & OBD_MD_FLSIZE))
1834                                 stx.stx_mask &= ~STATX_SIZE;
1835                         if (!(valid & OBD_MD_FLBLOCKS))
1836                                 stx.stx_mask &= ~STATX_BLOCKS;
1837
1838                         if (stxp && copy_to_user(stxp, &stx, sizeof(stx)))
1839                                 GOTO(out_req, rc = -EFAULT);
1840
1841                         if (lmmsizep && copy_to_user(lmmsizep, &lmmsize,
1842                                                      sizeof(*lmmsizep)))
1843                                 GOTO(out_req, rc = -EFAULT);
1844                 }
1845
1846                 EXIT;
1847 out_req:
1848                 ptlrpc_req_finished(request);
1849                 ptlrpc_req_finished(root_request);
1850                 if (filename)
1851                         ll_putname(filename);
1852                 return rc;
1853         }
1854         case OBD_IOC_QUOTACTL: {
1855                 struct if_quotactl *qctl;
1856                 int qctl_len = sizeof(*qctl) + LOV_MAXPOOLNAME + 1;
1857
1858                 OBD_ALLOC(qctl, qctl_len);
1859                 if (!qctl)
1860                         RETURN(-ENOMEM);
1861
1862                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1863                         GOTO(out_quotactl, rc = -EFAULT);
1864
1865                 if (LUSTRE_Q_CMD_IS_POOL(qctl->qc_cmd)) {
1866                         char __user *from = (char __user *)arg +
1867                                         offsetof(typeof(*qctl), qc_poolname);
1868                         if (copy_from_user(qctl->qc_poolname, from,
1869                                            LOV_MAXPOOLNAME + 1))
1870                                 GOTO(out_quotactl, rc = -EFAULT);
1871                 }
1872
1873                 rc = quotactl_ioctl(sbi, qctl);
1874                 if (rc == 0 &&
1875                     copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1876                         rc = -EFAULT;
1877
1878 out_quotactl:
1879                 OBD_FREE(qctl, qctl_len);
1880                 RETURN(rc);
1881         }
1882         case OBD_IOC_GETDTNAME:
1883         case OBD_IOC_GETMDNAME:
1884                 RETURN(ll_get_obd_name(inode, cmd, arg));
1885         case LL_IOC_FLUSHCTX:
1886                 RETURN(ll_flush_ctx(inode));
1887         case LL_IOC_GETOBDCOUNT: {
1888                 u32 count, vallen;
1889                 struct obd_export *exp;
1890
1891                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1892                         RETURN(-EFAULT);
1893
1894                 /* get ost count when count is zero, get mdt count otherwise */
1895                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1896                 vallen = sizeof(count);
1897                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1898                                   KEY_TGT_COUNT, &vallen, &count);
1899                 if (rc) {
1900                         CERROR("get target count failed: %d\n", rc);
1901                         RETURN(rc);
1902                 }
1903
1904                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1905                         RETURN(-EFAULT);
1906
1907                 RETURN(0);
1908         }
1909         case LL_IOC_PATH2FID:
1910                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1911                                      sizeof(struct lu_fid)))
1912                         RETURN(-EFAULT);
1913                 RETURN(0);
1914         case LL_IOC_GET_CONNECT_FLAGS: {
1915                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1916                                      (void __user *)arg));
1917         }
1918         case OBD_IOC_FID2PATH:
1919                 RETURN(ll_fid2path(inode, (void __user *)arg));
1920         case LL_IOC_GETPARENT:
1921                 RETURN(ll_getparent(file, (void __user *)arg));
1922         case LL_IOC_FID2MDTIDX: {
1923                 struct obd_export *exp = ll_i2mdexp(inode);
1924                 struct lu_fid     fid;
1925                 __u32             index;
1926
1927                 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1928                                    sizeof(fid)))
1929                         RETURN(-EFAULT);
1930
1931                 /* Call mdc_iocontrol */
1932                 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1933                                    (__u32 __user *)&index);
1934                 if (rc != 0)
1935                         RETURN(rc);
1936
1937                 RETURN(index);
1938         }
1939         case LL_IOC_HSM_REQUEST: {
1940                 struct hsm_user_request *hur;
1941                 ssize_t                  totalsize;
1942
1943                 OBD_ALLOC_PTR(hur);
1944                 if (hur == NULL)
1945                         RETURN(-ENOMEM);
1946
1947                 /* We don't know the true size yet; copy the fixed-size part */
1948                 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1949                         OBD_FREE_PTR(hur);
1950                         RETURN(-EFAULT);
1951                 }
1952
1953                 /* Compute the whole struct size */
1954                 totalsize = hur_len(hur);
1955                 OBD_FREE_PTR(hur);
1956                 if (totalsize < 0)
1957                         RETURN(-E2BIG);
1958
1959                 /* Final size will be more than double totalsize */
1960                 if (totalsize >= MDS_MAXREQSIZE / 3)
1961                         RETURN(-E2BIG);
1962
1963                 OBD_ALLOC_LARGE(hur, totalsize);
1964                 if (hur == NULL)
1965                         RETURN(-ENOMEM);
1966
1967                 /* Copy the whole struct */
1968                 if (copy_from_user(hur, (void __user *)arg, totalsize))
1969                         GOTO(out_hur, rc = -EFAULT);
1970
1971                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1972                         const struct lu_fid *fid;
1973                         struct inode *f;
1974                         int i;
1975
1976                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1977                                 fid = &hur->hur_user_item[i].hui_fid;
1978                                 f = search_inode_for_lustre(inode->i_sb, fid);
1979                                 if (IS_ERR(f)) {
1980                                         rc = PTR_ERR(f);
1981                                         break;
1982                                 }
1983
1984                                 rc = ll_hsm_release(f);
1985                                 iput(f);
1986                                 if (rc != 0)
1987                                         break;
1988                         }
1989                 } else {
1990                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1991                                            hur, NULL);
1992                 }
1993
1994 out_hur:
1995                 OBD_FREE_LARGE(hur, totalsize);
1996
1997                 RETURN(rc);
1998         }
1999         case LL_IOC_HSM_PROGRESS: {
2000                 struct hsm_progress_kernel      hpk;
2001                 struct hsm_progress             hp;
2002
2003                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
2004                         RETURN(-EFAULT);
2005
2006                 hpk.hpk_fid = hp.hp_fid;
2007                 hpk.hpk_cookie = hp.hp_cookie;
2008                 hpk.hpk_extent = hp.hp_extent;
2009                 hpk.hpk_flags = hp.hp_flags;
2010                 hpk.hpk_errval = hp.hp_errval;
2011                 hpk.hpk_data_version = 0;
2012
2013                 /* File may not exist in Lustre; all progress
2014                  * reported to Lustre root */
2015                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
2016                                    NULL);
2017                 RETURN(rc);
2018         }
2019         case LL_IOC_HSM_CT_START:
2020                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
2021                         RETURN(-EPERM);
2022
2023                 rc = copy_and_ct_start(cmd, sbi->ll_md_exp,
2024                                        (struct lustre_kernelcomm __user *)arg);
2025                 RETURN(rc);
2026
2027         case LL_IOC_HSM_COPY_START: {
2028                 struct hsm_copy *copy;
2029                 int              rc;
2030
2031                 OBD_ALLOC_PTR(copy);
2032                 if (copy == NULL)
2033                         RETURN(-ENOMEM);
2034                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
2035                         OBD_FREE_PTR(copy);
2036                         RETURN(-EFAULT);
2037                 }
2038
2039                 rc = ll_ioc_copy_start(inode->i_sb, copy);
2040                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
2041                         rc = -EFAULT;
2042
2043                 OBD_FREE_PTR(copy);
2044                 RETURN(rc);
2045         }
2046         case LL_IOC_HSM_COPY_END: {
2047                 struct hsm_copy *copy;
2048                 int              rc;
2049
2050                 OBD_ALLOC_PTR(copy);
2051                 if (copy == NULL)
2052                         RETURN(-ENOMEM);
2053                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
2054                         OBD_FREE_PTR(copy);
2055                         RETURN(-EFAULT);
2056                 }
2057
2058                 rc = ll_ioc_copy_end(inode->i_sb, copy);
2059                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
2060                         rc = -EFAULT;
2061
2062                 OBD_FREE_PTR(copy);
2063                 RETURN(rc);
2064         }
2065         case LL_IOC_MIGRATE: {
2066                 struct lmv_user_md *lum;
2067                 char *buf = NULL;
2068                 int len;
2069                 char *filename;
2070                 int namelen = 0;
2071                 int rc;
2072
2073                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
2074                 if (rc)
2075                         RETURN(rc);
2076
2077                 data = (struct obd_ioctl_data *)buf;
2078                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
2079                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
2080                         GOTO(migrate_free, rc = -EINVAL);
2081
2082                 filename = data->ioc_inlbuf1;
2083                 namelen = data->ioc_inllen1;
2084
2085                 if (namelen < 1 || namelen != strlen(filename) + 1) {
2086                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
2087                         GOTO(migrate_free, rc = -EINVAL);
2088                 }
2089
2090                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
2091                 if (lum->lum_magic != LMV_USER_MAGIC &&
2092                     lum->lum_magic != LMV_USER_MAGIC_SPECIFIC) {
2093                         rc = -EINVAL;
2094                         CERROR("%s: wrong lum magic %x: rc = %d\n",
2095                                filename, lum->lum_magic, rc);
2096                         GOTO(migrate_free, rc);
2097                 }
2098
2099                 rc = ll_migrate(inode, file, lum, filename);
2100 migrate_free:
2101                 OBD_FREE_LARGE(buf, len);
2102
2103                 RETURN(rc);
2104         }
2105         case FS_IOC_FSGETXATTR:
2106                 RETURN(ll_ioctl_fsgetxattr(inode, cmd, arg));
2107         case FS_IOC_FSSETXATTR:
2108                 RETURN(ll_ioctl_fssetxattr(inode, cmd, arg));
2109         case LL_IOC_PCC_DETACH_BY_FID: {
2110                 struct lu_pcc_detach_fid *detach;
2111                 struct lu_fid *fid;
2112                 struct inode *inode2;
2113                 unsigned long ino;
2114
2115                 OBD_ALLOC_PTR(detach);
2116                 if (detach == NULL)
2117                         RETURN(-ENOMEM);
2118
2119                 if (copy_from_user(detach,
2120                                    (const struct lu_pcc_detach_fid __user *)arg,
2121                                    sizeof(*detach)))
2122                         GOTO(out_detach, rc = -EFAULT);
2123
2124                 fid = &detach->pccd_fid;
2125                 ino = cl_fid_build_ino(fid, ll_need_32bit_api(sbi));
2126                 inode2 = ilookup5(inode->i_sb, ino, ll_test_inode_by_fid, fid);
2127                 if (inode2 == NULL)
2128                         /* Target inode is not in inode cache, and PCC file
2129                          * has aleady released, return immdiately.
2130                          */
2131                         GOTO(out_detach, rc = 0);
2132
2133                 if (!S_ISREG(inode2->i_mode))
2134                         GOTO(out_iput, rc = -EINVAL);
2135
2136                 if (!inode_owner_or_capable(inode2))
2137                         GOTO(out_iput, rc = -EPERM);
2138
2139                 rc = pcc_ioctl_detach(inode2, detach->pccd_opt);
2140 out_iput:
2141                 iput(inode2);
2142 out_detach:
2143                 OBD_FREE_PTR(detach);
2144                 RETURN(rc);
2145         }
2146 #ifdef HAVE_LUSTRE_CRYPTO
2147         case LL_IOC_SET_ENCRYPTION_POLICY:
2148                 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2149                         return -EOPNOTSUPP;
2150                 return llcrypt_ioctl_set_policy(file, (const void __user *)arg);
2151         case LL_IOC_GET_ENCRYPTION_POLICY_EX:
2152                 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2153                         return -EOPNOTSUPP;
2154                 return llcrypt_ioctl_get_policy_ex(file, (void __user *)arg);
2155         case LL_IOC_ADD_ENCRYPTION_KEY:
2156                 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2157                         return -EOPNOTSUPP;
2158                 return llcrypt_ioctl_add_key(file, (void __user *)arg);
2159         case LL_IOC_REMOVE_ENCRYPTION_KEY:
2160                 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2161                         return -EOPNOTSUPP;
2162                 return llcrypt_ioctl_remove_key(file, (void __user *)arg);
2163         case LL_IOC_REMOVE_ENCRYPTION_KEY_ALL_USERS:
2164                 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2165                         return -EOPNOTSUPP;
2166                 return llcrypt_ioctl_remove_key_all_users(file,
2167                                                           (void __user *)arg);
2168         case LL_IOC_GET_ENCRYPTION_KEY_STATUS:
2169                 if (!ll_sbi_has_encrypt(ll_i2sbi(inode)))
2170                         return -EOPNOTSUPP;
2171                 return llcrypt_ioctl_get_key_status(file, (void __user *)arg);
2172 #endif
2173         default:
2174                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
2175                                      (void __user *)arg));
2176         }
2177 }
2178
2179 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
2180 {
2181         struct inode *inode = file->f_mapping->host;
2182         struct ll_file_data *fd = file->private_data;
2183         struct ll_sb_info *sbi = ll_i2sbi(inode);
2184         int api32 = ll_need_32bit_api(sbi);
2185         loff_t ret = -EINVAL;
2186         ENTRY;
2187
2188         inode_lock(inode);
2189         switch (origin) {
2190         case SEEK_SET:
2191                 break;
2192         case SEEK_CUR:
2193                 offset += file->f_pos;
2194                 break;
2195         case SEEK_END:
2196                 if (offset > 0)
2197                         GOTO(out, ret);
2198                 if (api32)
2199                         offset += LL_DIR_END_OFF_32BIT;
2200                 else
2201                         offset += LL_DIR_END_OFF;
2202                 break;
2203         default:
2204                 GOTO(out, ret);
2205         }
2206
2207         if (offset >= 0 &&
2208             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
2209              (!api32 && offset <= LL_DIR_END_OFF))) {
2210                 if (offset != file->f_pos) {
2211                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
2212                             (!api32 && offset == LL_DIR_END_OFF))
2213                                 fd->lfd_pos = MDS_DIR_END_OFF;
2214                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
2215                                 fd->lfd_pos = offset << 32;
2216                         else
2217                                 fd->lfd_pos = offset;
2218                         file->f_pos = offset;
2219                         file->f_version = 0;
2220                 }
2221                 ret = offset;
2222         }
2223         GOTO(out, ret);
2224
2225 out:
2226         inode_unlock(inode);
2227         return ret;
2228 }
2229
2230 static int ll_dir_open(struct inode *inode, struct file *file)
2231 {
2232         ENTRY;
2233         RETURN(ll_file_open(inode, file));
2234 }
2235
2236 static int ll_dir_release(struct inode *inode, struct file *file)
2237 {
2238         ENTRY;
2239         RETURN(ll_file_release(inode, file));
2240 }
2241
2242 const struct file_operations ll_dir_operations = {
2243         .llseek         = ll_dir_seek,
2244         .open           = ll_dir_open,
2245         .release        = ll_dir_release,
2246         .read           = generic_read_dir,
2247 #ifdef HAVE_DIR_CONTEXT
2248         .iterate_shared = ll_iterate,
2249 #else
2250         .readdir        = ll_readdir,
2251 #endif
2252         .unlocked_ioctl = ll_dir_ioctl,
2253         .fsync          = ll_fsync,
2254 };