lustre/llite/dir.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2014, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/llite/dir.c
  37  *
  38  * Directory code for lustre client.
  39  */
  40
  41 #include <linux/fs.h>
  42 #include <linux/pagemap.h>
  43 #include <linux/mm.h>
  44 #include <linux/version.h>
  45 #include <linux/user_namespace.h>
  46 #ifdef HAVE_UIDGID_HEADER
  47 # include <linux/uidgid.h>
  48 #endif
  49 #include <asm/uaccess.h>
  50 #include <linux/buffer_head.h>   // for wait_on_buffer
  51 #include <linux/pagevec.h>
  52
  53 #define DEBUG_SUBSYSTEM S_LLITE
  54
  55 #include <lustre/lustre_idl.h>
  56 #include <obd_support.h>
  57 #include <obd_class.h>
  58 #include <lustre_ioctl.h>
  59 #include <lustre_lib.h>
  60 #include <lustre_dlm.h>
  61 #include <lustre_fid.h>
  62 #include <lustre_kernelcomm.h>
  63 #include "llite_internal.h"
  64
  65 /*
  66  * (new) readdir implementation overview.
  67  *
  68  * Original lustre readdir implementation cached exact copy of raw directory
  69  * pages on the client. These pages were indexed in client page cache by
  70  * logical offset in the directory file. This design, while very simple and
  71  * intuitive had some inherent problems:
  72  *
  73  *     . it implies that byte offset to the directory entry serves as a
  74  *     telldir(3)/seekdir(3) cookie, but that offset is not stable: in
  75  *     ext3/htree directory entries may move due to splits, and more
  76  *     importantly,
  77  *
  78  *     . it is incompatible with the design of split directories for cmd3,
  79  *     that assumes that names are distributed across nodes based on their
  80  *     hash, and so readdir should be done in hash order.
  81  *
  82  * New readdir implementation does readdir in hash order, and uses hash of a
  83  * file name as a telldir/seekdir cookie. This led to number of complications:
  84  *
  85  *     . hash is not unique, so it cannot be used to index cached directory
  86  *     pages on the client (note, that it requires a whole pageful of hash
  87  *     collided entries to cause two pages to have identical hashes);
  88  *
  89  *     . hash is not unique, so it cannot, strictly speaking, be used as an
  90  *     entry cookie. ext3/htree has the same problem and lustre implementation
  91  *     mimics their solution: seekdir(hash) positions directory at the first
  92  *     entry with the given hash.
  93  *
  94  * Client side.
  95  *
  96  * 0. caching
  97  *
  98  * Client caches directory pages using hash of the first entry as an index. As
  99  * noted above hash is not unique, so this solution doesn't work as is:
 100  * special processing is needed for "page hash chains" (i.e., sequences of
 101  * pages filled with entries all having the same hash value).
 102  *
 103  * First, such chains have to be detected. To this end, server returns to the
 104  * client the hash of the first entry on the page next to one returned. When
 105  * client detects that this hash is the same as hash of the first entry on the
 106  * returned page, page hash collision has to be handled. Pages in the
 107  * hash chain, except first one, are termed "overflow pages".
 108  *
 109  * Solution to index uniqueness problem is to not cache overflow
 110  * pages. Instead, when page hash collision is detected, all overflow pages
 111  * from emerging chain are immediately requested from the server and placed in
 112  * a special data structure (struct ll_dir_chain). This data structure is used
 113  * by ll_readdir() to process entries from overflow pages. When readdir
 114  * invocation finishes, overflow pages are discarded. If page hash collision
 115  * chain weren't completely processed, next call to readdir will again detect
 116  * page hash collision, again read overflow pages in, process next portion of
 117  * entries and again discard the pages. This is not as wasteful as it looks,
 118  * because, given reasonable hash, page hash collisions are extremely rare.
 119  *
 120  * 1. directory positioning
 121  *
 122  * When seekdir(hash) is called, original
 123  *
 124  *
 125  *
 126  *
 127  *
 128  *
 129  *
 130  *
 131  * Server.
 132  *
 133  * identification of and access to overflow pages
 134  *
 135  * page format
 136  *
 137  * Page in MDS_READPAGE RPC is packed in LU_PAGE_SIZE, and each page contains
 138  * a header lu_dirpage which describes the start/end hash, and whether this
 139  * page is empty (contains no dir entry) or hash collide with next page.
 140  * After client receives reply, several pages will be integrated into dir page
 141  * in PAGE_CACHE_SIZE (if PAGE_CACHE_SIZE greater than LU_PAGE_SIZE), and the
 142  * lu_dirpage for this integrated page will be adjusted. See
 143  * mdc_adjust_dirpages().
 144  *
 145  */
 146 struct page *ll_get_dir_page(struct inode *dir, struct md_op_data *op_data,
 147                              __u64 offset, struct ll_dir_chain *chain)
 148 {
 149         struct md_callback      cb_op;
 150         struct page             *page;
 151         int                     rc;
 152
 153         cb_op.md_blocking_ast = ll_md_blocking_ast;
 154         rc = md_read_page(ll_i2mdexp(dir), op_data, &cb_op, offset, &page);
 155         if (rc != 0)
 156                 return ERR_PTR(rc);
 157
 158         return page;
 159 }
 160
 161 void ll_release_page(struct inode *inode, struct page *page,
 162                      bool remove)
 163 {
 164         kunmap(page);
 165
 166         /* Always remove the page for striped dir, because the page is
 167          * built from temporarily in LMV layer */
 168         if (inode != NULL && S_ISDIR(inode->i_mode) &&
 169             ll_i2info(inode)->lli_lsm_md != NULL) {
 170                 __free_page(page);
 171                 return;
 172         }
 173
 174         if (remove) {
 175                 lock_page(page);
 176                 if (likely(page->mapping != NULL))
 177                         truncate_complete_page(page->mapping, page);
 178                 unlock_page(page);
 179         }
 180         page_cache_release(page);
 181 }
 182
 183 /**
 184  * return IF_* type for given lu_dirent entry.
 185  * IF_* flag shld be converted to particular OS file type in
 186  * platform llite module.
 187  */
 188 static u16 ll_dirent_type_get(struct lu_dirent *ent)
 189 {
 190         u16 type = 0;
 191         struct luda_type *lt;
 192         int len = 0;
 193
 194         if (le32_to_cpu(ent->lde_attrs) & LUDA_TYPE) {
 195                 const unsigned align = sizeof(struct luda_type) - 1;
 196
 197                 len = le16_to_cpu(ent->lde_namelen);
 198                 len = (len + align) & ~align;
 199                 lt = (void *)ent->lde_name + len;
 200                 type = IFTODT(le16_to_cpu(lt->lt_type));
 201         }
 202
 203         return type;
 204 }
 205
 206 #ifdef HAVE_DIR_CONTEXT
 207 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 208                 struct dir_context *ctx)
 209 {
 210 #else
 211 int ll_dir_read(struct inode *inode, __u64 *ppos, struct md_op_data *op_data,
 212                 void *cookie, filldir_t filldir)
 213 {
 214 #endif
 215         struct ll_sb_info    *sbi        = ll_i2sbi(inode);
 216         __u64                 pos        = *ppos;
 217         bool                  is_api32 = ll_need_32bit_api(sbi);
 218         bool                  is_hash64 = sbi->ll_flags & LL_SBI_64BIT_HASH;
 219         struct page          *page;
 220         struct ll_dir_chain   chain;
 221         bool                  done = false;
 222         int                   rc = 0;
 223         ENTRY;
 224
 225         ll_dir_chain_init(&chain);
 226
 227         page = ll_get_dir_page(inode, op_data, pos, &chain);
 228
 229         while (rc == 0 && !done) {
 230                 struct lu_dirpage *dp;
 231                 struct lu_dirent  *ent;
 232                 __u64 hash;
 233                 __u64 next;
 234
 235                 if (IS_ERR(page)) {
 236                         rc = PTR_ERR(page);
 237                         break;
 238                 }
 239
 240                 hash = MDS_DIR_END_OFF;
 241                 dp = page_address(page);
 242                 for (ent = lu_dirent_start(dp); ent != NULL && !done;
 243                      ent = lu_dirent_next(ent)) {
 244                         __u16          type;
 245                         int            namelen;
 246                         struct lu_fid  fid;
 247                         __u64          lhash;
 248                         __u64          ino;
 249
 250                         hash = le64_to_cpu(ent->lde_hash);
 251                         if (hash < pos)
 252                                 /*
 253                                  * Skip until we find target hash
 254                                  * value.
 255                                  */
 256                                 continue;
 257
 258                         namelen = le16_to_cpu(ent->lde_namelen);
 259                         if (namelen == 0)
 260                                 /*
 261                                  * Skip dummy record.
 262                                  */
 263                                 continue;
 264
 265                         if (is_api32 && is_hash64)
 266                                 lhash = hash >> 32;
 267                         else
 268                                 lhash = hash;
 269                         fid_le_to_cpu(&fid, &ent->lde_fid);
 270                         ino = cl_fid_build_ino(&fid, is_api32);
 271                         type = ll_dirent_type_get(ent);
 272                         /* For 'll_nfs_get_name_filldir()', it will try
 273                          * to access the 'ent' through its 'lde_name',
 274                          * so the parameter 'name' for 'filldir()' must
 275                          * be part of the 'ent'. */
 276 #ifdef HAVE_DIR_CONTEXT
 277                         ctx->pos = lhash;
 278                         done = !dir_emit(ctx, ent->lde_name, namelen, ino,
 279                                          type);
 280 #else
 281                         done = filldir(cookie, ent->lde_name, namelen, lhash,
 282                                        ino, type);
 283 #endif
 284                 }
 285
 286                 if (done) {
 287                         pos = hash;
 288                         ll_release_page(inode, page, false);
 289                         break;
 290                 }
 291
 292                 next = le64_to_cpu(dp->ldp_hash_end);
 293                 pos = next;
 294                 if (pos == MDS_DIR_END_OFF) {
 295                         /*
 296                          * End of directory reached.
 297                          */
 298                         done = 1;
 299                         ll_release_page(inode, page, false);
 300                 } else {
 301                         /*
 302                          * Normal case: continue to the next
 303                          * page.
 304                          */
 305                         ll_release_page(inode, page,
 306                                         le32_to_cpu(dp->ldp_flags) &
 307                                         LDF_COLLIDE);
 308                         next = pos;
 309                         page = ll_get_dir_page(inode, op_data, pos,
 310                                                &chain);
 311                 }
 312         }
 313 #ifdef HAVE_DIR_CONTEXT
 314         ctx->pos = pos;
 315 #else
 316         *ppos = pos;
 317 #endif
 318         ll_dir_chain_fini(&chain);
 319         RETURN(rc);
 320 }
 321
 322 #ifdef HAVE_DIR_CONTEXT
 323 static int ll_iterate(struct file *filp, struct dir_context *ctx)
 324 #else
 325 static int ll_readdir(struct file *filp, void *cookie, filldir_t filldir)
 326 #endif
 327 {
 328         struct inode            *inode  = filp->f_path.dentry->d_inode;
 329         struct ll_file_data     *lfd    = LUSTRE_FPRIVATE(filp);
 330         struct ll_sb_info       *sbi    = ll_i2sbi(inode);
 331         int                     hash64  = sbi->ll_flags & LL_SBI_64BIT_HASH;
 332         int                     api32   = ll_need_32bit_api(sbi);
 333         struct md_op_data       *op_data;
 334         __u64                   pos;
 335         int                     rc;
 336         ENTRY;
 337
 338         if (lfd != NULL)
 339                 pos = lfd->lfd_pos;
 340         else
 341                 pos = 0;
 342
 343         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) pos/size"
 344                "%lu/%llu 32bit_api %d\n", PFID(ll_inode2fid(inode)),
 345                inode, (unsigned long)pos, i_size_read(inode), api32);
 346
 347         if (pos == MDS_DIR_END_OFF)
 348                 /*
 349                  * end-of-file.
 350                  */
 351                 GOTO(out, rc = 0);
 352
 353         op_data = ll_prep_md_op_data(NULL, inode, inode, NULL, 0, 0,
 354                                      LUSTRE_OPC_ANY, inode);
 355         if (IS_ERR(op_data))
 356                 GOTO(out, rc = PTR_ERR(op_data));
 357
 358         if (unlikely(op_data->op_mea1 != NULL)) {
 359                 /* This is only needed for striped dir to fill ..,
 360                  * see lmv_read_entry */
 361                 if (filp->f_path.dentry->d_parent != NULL &&
 362                     filp->f_path.dentry->d_parent->d_inode != NULL) {
 363                         __u64 ibits = MDS_INODELOCK_UPDATE;
 364                         struct inode *parent =
 365                                 filp->f_path.dentry->d_parent->d_inode;
 366
 367                         if (ll_have_md_lock(parent, &ibits, LCK_MINMODE))
 368                                 op_data->op_fid3 = *ll_inode2fid(parent);
 369                 }
 370
 371                 /* If it can not find in cache, do lookup .. on the master
 372                  * object */
 373                 if (fid_is_zero(&op_data->op_fid3)) {
 374                         rc = ll_dir_get_parent_fid(inode, &op_data->op_fid3);
 375                         if (rc != 0) {
 376                                 ll_finish_md_op_data(op_data);
 377                                 RETURN(rc);
 378                         }
 379                 }
 380         }
 381         op_data->op_max_pages = sbi->ll_md_brw_pages;
 382 #ifdef HAVE_DIR_CONTEXT
 383         ctx->pos = pos;
 384         rc = ll_dir_read(inode, &pos, op_data, ctx);
 385         pos = ctx->pos;
 386 #else
 387         rc = ll_dir_read(inode, &pos, op_data, cookie, filldir);
 388 #endif
 389         if (lfd != NULL)
 390                 lfd->lfd_pos = pos;
 391
 392         if (pos == MDS_DIR_END_OFF) {
 393                 if (api32)
 394                         pos = LL_DIR_END_OFF_32BIT;
 395                 else
 396                         pos = LL_DIR_END_OFF;
 397         } else {
 398                 if (api32 && hash64)
 399                         pos = pos >> 32;
 400         }
 401 #ifdef HAVE_DIR_CONTEXT
 402         ctx->pos = pos;
 403 #else
 404         filp->f_pos = pos;
 405 #endif
 406         ll_finish_md_op_data(op_data);
 407         filp->f_version = inode->i_version;
 408
 409 out:
 410         if (!rc)
 411                 ll_stats_ops_tally(sbi, LPROC_LL_READDIR, 1);
 412
 413         RETURN(rc);
 414 }
 415
 416 static int ll_send_mgc_param(struct obd_export *mgc, char *string)
 417 {
 418         struct mgs_send_param *msp;
 419         int rc = 0;
 420
 421         OBD_ALLOC_PTR(msp);
 422         if (!msp)
 423                 return -ENOMEM;
 424
 425         strlcpy(msp->mgs_param, string, sizeof(msp->mgs_param));
 426         rc = obd_set_info_async(NULL, mgc, sizeof(KEY_SET_INFO), KEY_SET_INFO,
 427                                 sizeof(struct mgs_send_param), msp, NULL);
 428         if (rc)
 429                 CERROR("Failed to set parameter: %d\n", rc);
 430         OBD_FREE_PTR(msp);
 431
 432         return rc;
 433 }
 434
 435 /**
 436  * Create striped directory with specified stripe(@lump)
 437  *
 438  * param[in]parent      the parent of the directory.
 439  * param[in]lump        the specified stripes.
 440  * param[in]dirname     the name of the directory.
 441  * param[in]mode        the specified mode of the directory.
 442  *
 443  * retval               =0 if striped directory is being created successfully.
 444  *                      <0 if the creation is failed.
 445  */
 446 static int ll_dir_setdirstripe(struct inode *parent, struct lmv_user_md *lump,
 447                                const char *dirname, umode_t mode)
 448 {
 449         struct ptlrpc_request *request = NULL;
 450         struct md_op_data *op_data;
 451         struct ll_sb_info *sbi = ll_i2sbi(parent);
 452         int err;
 453         ENTRY;
 454
 455         if (unlikely(lump->lum_magic != LMV_USER_MAGIC))
 456                 RETURN(-EINVAL);
 457
 458         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p) name %s"
 459                "stripe_offset %d, stripe_count: %u\n",
 460                PFID(ll_inode2fid(parent)), parent, dirname,
 461                (int)lump->lum_stripe_offset, lump->lum_stripe_count);
 462
 463         if (lump->lum_stripe_count > 1 &&
 464             !(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_DIR_STRIPE))
 465                 RETURN(-EINVAL);
 466
 467         if (lump->lum_magic != cpu_to_le32(LMV_USER_MAGIC))
 468                 lustre_swab_lmv_user_md(lump);
 469
 470         if (!IS_POSIXACL(parent) || !exp_connect_umask(ll_i2mdexp(parent)))
 471                 mode &= ~current_umask();
 472         mode = (mode & (S_IRWXUGO | S_ISVTX)) | S_IFDIR;
 473         op_data = ll_prep_md_op_data(NULL, parent, NULL, dirname,
 474                                      strlen(dirname), mode, LUSTRE_OPC_MKDIR,
 475                                      lump);
 476         if (IS_ERR(op_data))
 477                 GOTO(err_exit, err = PTR_ERR(op_data));
 478
 479         op_data->op_cli_flags |= CLI_SET_MEA;
 480         err = md_create(sbi->ll_md_exp, op_data, lump, sizeof(*lump), mode,
 481                         from_kuid(&init_user_ns, current_fsuid()),
 482                         from_kgid(&init_user_ns, current_fsgid()),
 483                         cfs_curproc_cap_pack(), 0, &request);
 484         ll_finish_md_op_data(op_data);
 485         if (err)
 486                 GOTO(err_exit, err);
 487 err_exit:
 488         ptlrpc_req_finished(request);
 489         return err;
 490 }
 491
 492 int ll_dir_setstripe(struct inode *inode, struct lov_user_md *lump,
 493                      int set_default)
 494 {
 495         struct ll_sb_info *sbi = ll_i2sbi(inode);
 496         struct md_op_data *op_data;
 497         struct ptlrpc_request *req = NULL;
 498         int rc = 0;
 499         struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
 500         struct obd_device *mgc = lsi->lsi_mgc;
 501         int lum_size;
 502         ENTRY;
 503
 504         if (lump != NULL) {
 505                 /*
 506                  * This is coming from userspace, so should be in
 507                  * local endian.  But the MDS would like it in little
 508                  * endian, so we swab it before we send it.
 509                  */
 510                 switch (lump->lmm_magic) {
 511                 case LOV_USER_MAGIC_V1: {
 512                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V1))
 513                                 lustre_swab_lov_user_md_v1(lump);
 514                         lum_size = sizeof(struct lov_user_md_v1);
 515                         break;
 516                 }
 517                 case LOV_USER_MAGIC_V3: {
 518                         if (lump->lmm_magic != cpu_to_le32(LOV_USER_MAGIC_V3))
 519                                 lustre_swab_lov_user_md_v3(
 520                                         (struct lov_user_md_v3 *)lump);
 521                         lum_size = sizeof(struct lov_user_md_v3);
 522                         break;
 523                 }
 524                 case LMV_USER_MAGIC: {
 525                         if (lump->lmm_magic != cpu_to_le32(LMV_USER_MAGIC))
 526                                 lustre_swab_lmv_user_md(
 527                                         (struct lmv_user_md *)lump);
 528                         lum_size = sizeof(struct lmv_user_md);
 529                         break;
 530                 }
 531                 default: {
 532                         CDEBUG(D_IOCTL, "bad userland LOV MAGIC:"
 533                                         " %#08x != %#08x nor %#08x\n",
 534                                         lump->lmm_magic, LOV_USER_MAGIC_V1,
 535                                         LOV_USER_MAGIC_V3);
 536                         RETURN(-EINVAL);
 537                 }
 538                 }
 539         } else {
 540                 lum_size = sizeof(struct lov_user_md_v1);
 541         }
 542
 543         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL, 0, 0,
 544                                      LUSTRE_OPC_ANY, NULL);
 545         if (IS_ERR(op_data))
 546                 RETURN(PTR_ERR(op_data));
 547
 548         /* swabbing is done in lov_setstripe() on server side */
 549         rc = md_setattr(sbi->ll_md_exp, op_data, lump, lum_size, &req);
 550         ll_finish_md_op_data(op_data);
 551         ptlrpc_req_finished(req);
 552         if (rc) {
 553                 if (rc != -EPERM && rc != -EACCES)
 554                         CERROR("mdc_setattr fails: rc = %d\n", rc);
 555         }
 556
 557         /* In the following we use the fact that LOV_USER_MAGIC_V1 and
 558          LOV_USER_MAGIC_V3 have the same initial fields so we do not
 559          need the make the distiction between the 2 versions */
 560         if (set_default && mgc->u.cli.cl_mgc_mgsexp) {
 561                 char *param = NULL;
 562                 char *buf;
 563
 564                 OBD_ALLOC(param, MGS_PARAM_MAXLEN);
 565                 if (param == NULL)
 566                         GOTO(end, rc = -ENOMEM);
 567
 568                 buf = param;
 569                 /* Get fsname and assume devname to be -MDT0000. */
 570                 ll_get_fsname(inode->i_sb, buf, MTI_NAME_MAXLEN);
 571                 strcat(buf, "-MDT0000.lov");
 572                 buf += strlen(buf);
 573
 574                 /* Set root stripesize */
 575                 sprintf(buf, ".stripesize=%u",
 576                         lump ? le32_to_cpu(lump->lmm_stripe_size) : 0);
 577                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 578                 if (rc)
 579                         GOTO(end, rc);
 580
 581                 /* Set root stripecount */
 582                 sprintf(buf, ".stripecount=%hd",
 583                         lump ? le16_to_cpu(lump->lmm_stripe_count) : 0);
 584                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 585                 if (rc)
 586                         GOTO(end, rc);
 587
 588                 /* Set root stripeoffset */
 589                 sprintf(buf, ".stripeoffset=%hd",
 590                         lump ? le16_to_cpu(lump->lmm_stripe_offset) :
 591                         (typeof(lump->lmm_stripe_offset))(-1));
 592                 rc = ll_send_mgc_param(mgc->u.cli.cl_mgc_mgsexp, param);
 593
 594 end:
 595                 if (param != NULL)
 596                         OBD_FREE(param, MGS_PARAM_MAXLEN);
 597         }
 598         RETURN(rc);
 599 }
 600
 601 /**
 602  * This function will be used to get default LOV/LMV/Default LMV
 603  * @valid will be used to indicate which stripe it will retrieve
 604  *      OBD_MD_MEA              LMV stripe EA
 605  *      OBD_MD_DEFAULT_MEA      Default LMV stripe EA
 606  *      otherwise               Default LOV EA.
 607  * Each time, it can only retrieve 1 stripe EA
 608  **/
 609 int ll_dir_getstripe(struct inode *inode, void **plmm, int *plmm_size,
 610                      struct ptlrpc_request **request, u64 valid)
 611 {
 612         struct ll_sb_info *sbi = ll_i2sbi(inode);
 613         struct mdt_body   *body;
 614         struct lov_mds_md *lmm = NULL;
 615         struct ptlrpc_request *req = NULL;
 616         int rc, lmm_size;
 617         struct md_op_data *op_data;
 618         ENTRY;
 619
 620         rc = ll_get_default_mdsize(sbi, &lmm_size);
 621         if (rc)
 622                 RETURN(rc);
 623
 624         op_data = ll_prep_md_op_data(NULL, inode, NULL, NULL,
 625                                      0, lmm_size, LUSTRE_OPC_ANY,
 626                                      NULL);
 627         if (IS_ERR(op_data))
 628                 RETURN(PTR_ERR(op_data));
 629
 630         op_data->op_valid = valid | OBD_MD_FLEASIZE | OBD_MD_FLDIREA;
 631         rc = md_getattr(sbi->ll_md_exp, op_data, &req);
 632         ll_finish_md_op_data(op_data);
 633         if (rc < 0) {
 634                 CDEBUG(D_INFO, "md_getattr failed on inode "
 635                        DFID": rc %d\n", PFID(ll_inode2fid(inode)), rc);
 636                 GOTO(out, rc);
 637         }
 638
 639         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
 640         LASSERT(body != NULL);
 641
 642         lmm_size = body->mbo_eadatasize;
 643
 644         if (!(body->mbo_valid & (OBD_MD_FLEASIZE | OBD_MD_FLDIREA)) ||
 645             lmm_size == 0) {
 646                 GOTO(out, rc = -ENODATA);
 647         }
 648
 649         lmm = req_capsule_server_sized_get(&req->rq_pill,
 650                                            &RMF_MDT_MD, lmm_size);
 651         LASSERT(lmm != NULL);
 652
 653         /*
 654          * This is coming from the MDS, so is probably in
 655          * little endian.  We convert it to host endian before
 656          * passing it to userspace.
 657          */
 658         /* We don't swab objects for directories */
 659         switch (le32_to_cpu(lmm->lmm_magic)) {
 660         case LOV_MAGIC_V1:
 661                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 662                         lustre_swab_lov_user_md_v1((struct lov_user_md_v1 *)lmm);
 663                 break;
 664         case LOV_MAGIC_V3:
 665                 if (LOV_MAGIC != cpu_to_le32(LOV_MAGIC))
 666                         lustre_swab_lov_user_md_v3((struct lov_user_md_v3 *)lmm);
 667                 break;
 668         case LMV_MAGIC_V1:
 669                 if (LMV_MAGIC != cpu_to_le32(LMV_MAGIC))
 670                         lustre_swab_lmv_mds_md((union lmv_mds_md *)lmm);
 671                 break;
 672         case LMV_USER_MAGIC:
 673                 if (LMV_USER_MAGIC != cpu_to_le32(LMV_USER_MAGIC))
 674                         lustre_swab_lmv_user_md((struct lmv_user_md *)lmm);
 675                 break;
 676         default:
 677                 CERROR("unknown magic: %lX\n", (unsigned long)lmm->lmm_magic);
 678                 rc = -EPROTO;
 679         }
 680 out:
 681         *plmm = lmm;
 682         *plmm_size = lmm_size;
 683         *request = req;
 684         return rc;
 685 }
 686
 687 int ll_get_mdt_idx_by_fid(struct ll_sb_info *sbi, const struct lu_fid *fid)
 688 {
 689         struct md_op_data       *op_data;
 690         int                     rc;
 691         int                     mdt_index;
 692         ENTRY;
 693
 694         OBD_ALLOC_PTR(op_data);
 695         if (op_data == NULL)
 696                 RETURN(-ENOMEM);
 697
 698         op_data->op_flags |= MF_GET_MDT_IDX;
 699         op_data->op_fid1 = *fid;
 700         rc = md_getattr(sbi->ll_md_exp, op_data, NULL);
 701         mdt_index = op_data->op_mds;
 702         OBD_FREE_PTR(op_data);
 703         if (rc < 0)
 704                 RETURN(rc);
 705
 706         RETURN(mdt_index);
 707 }
 708
 709 /*
 710  *  Get MDT index for the inode.
 711  */
 712 int ll_get_mdt_idx(struct inode *inode)
 713 {
 714         return ll_get_mdt_idx_by_fid(ll_i2sbi(inode), ll_inode2fid(inode));
 715 }
 716
 717 /**
 718  * Generic handler to do any pre-copy work.
 719  *
 720  * It sends a first hsm_progress (with extent length == 0) to coordinator as a
 721  * first information for it that real work has started.
 722  *
 723  * Moreover, for a ARCHIVE request, it will sample the file data version and
 724  * store it in \a copy.
 725  *
 726  * \return 0 on success.
 727  */
 728 static int ll_ioc_copy_start(struct super_block *sb, struct hsm_copy *copy)
 729 {
 730         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 731         struct hsm_progress_kernel       hpk;
 732         int                              rc = 0;
 733         int                              rc2;
 734         ENTRY;
 735
 736         /* Forge a hsm_progress based on data from copy. */
 737         hpk.hpk_fid = copy->hc_hai.hai_fid;
 738         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 739         hpk.hpk_extent.offset = copy->hc_hai.hai_extent.offset;
 740         hpk.hpk_extent.length = 0;
 741         hpk.hpk_flags = 0;
 742         hpk.hpk_errval = 0;
 743         hpk.hpk_data_version = 0;
 744
 745
 746         /* For archive request, we need to read the current file version. */
 747         if (copy->hc_hai.hai_action == HSMA_ARCHIVE) {
 748                 struct inode    *inode;
 749                 __u64            data_version = 0;
 750
 751                 /* Get inode for this fid */
 752                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 753                 if (IS_ERR(inode)) {
 754                         hpk.hpk_flags |= HP_FLAG_RETRY;
 755                         /* hpk_errval is >= 0 */
 756                         hpk.hpk_errval = -PTR_ERR(inode);
 757                         GOTO(progress, rc = PTR_ERR(inode));
 758                 }
 759
 760                 /* Read current file data version */
 761                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 762                 iput(inode);
 763                 if (rc != 0) {
 764                         CDEBUG(D_HSM, "Could not read file data version of "
 765                                       DFID" (rc = %d). Archive request ("
 766                                       LPX64") could not be done.\n",
 767                                       PFID(&copy->hc_hai.hai_fid), rc,
 768                                       copy->hc_hai.hai_cookie);
 769                         hpk.hpk_flags |= HP_FLAG_RETRY;
 770                         /* hpk_errval must be >= 0 */
 771                         hpk.hpk_errval = -rc;
 772                         GOTO(progress, rc);
 773                 }
 774
 775                 /* Store in the hsm_copy for later copytool use.
 776                  * Always modified even if no lsm. */
 777                 copy->hc_data_version = data_version;
 778         }
 779
 780 progress:
 781         /* On error, the request should be considered as completed */
 782         if (hpk.hpk_errval > 0)
 783                 hpk.hpk_flags |= HP_FLAG_COMPLETED;
 784
 785         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 786                             &hpk, NULL);
 787
 788         /* Return first error */
 789         RETURN(rc != 0 ? rc : rc2);
 790 }
 791
 792 /**
 793  * Generic handler to do any post-copy work.
 794  *
 795  * It will send the last hsm_progress update to coordinator to inform it
 796  * that copy is finished and whether it was successful or not.
 797  *
 798  * Moreover,
 799  * - for ARCHIVE request, it will sample the file data version and compare it
 800  *   with the version saved in ll_ioc_copy_start(). If they do not match, copy
 801  *   will be considered as failed.
 802  * - for RESTORE request, it will sample the file data version and send it to
 803  *   coordinator which is useful if the file was imported as 'released'.
 804  *
 805  * \return 0 on success.
 806  */
 807 static int ll_ioc_copy_end(struct super_block *sb, struct hsm_copy *copy)
 808 {
 809         struct ll_sb_info               *sbi = ll_s2sbi(sb);
 810         struct hsm_progress_kernel       hpk;
 811         int                              rc = 0;
 812         int                              rc2;
 813         ENTRY;
 814
 815         /* If you modify the logic here, also check llapi_hsm_copy_end(). */
 816         /* Take care: copy->hc_hai.hai_action, len, gid and data are not
 817          * initialized if copy_end was called with copy == NULL.
 818          */
 819
 820         /* Forge a hsm_progress based on data from copy. */
 821         hpk.hpk_fid = copy->hc_hai.hai_fid;
 822         hpk.hpk_cookie = copy->hc_hai.hai_cookie;
 823         hpk.hpk_extent = copy->hc_hai.hai_extent;
 824         hpk.hpk_flags = copy->hc_flags | HP_FLAG_COMPLETED;
 825         hpk.hpk_errval = copy->hc_errval;
 826         hpk.hpk_data_version = 0;
 827
 828         /* For archive request, we need to check the file data was not changed.
 829          *
 830          * For restore request, we need to send the file data version, this is
 831          * useful when the file was created using hsm_import.
 832          */
 833         if (((copy->hc_hai.hai_action == HSMA_ARCHIVE) ||
 834              (copy->hc_hai.hai_action == HSMA_RESTORE)) &&
 835             (copy->hc_errval == 0)) {
 836                 struct inode    *inode;
 837                 __u64            data_version = 0;
 838
 839                 /* Get lsm for this fid */
 840                 inode = search_inode_for_lustre(sb, &copy->hc_hai.hai_fid);
 841                 if (IS_ERR(inode)) {
 842                         hpk.hpk_flags |= HP_FLAG_RETRY;
 843                         /* hpk_errval must be >= 0 */
 844                         hpk.hpk_errval = -PTR_ERR(inode);
 845                         GOTO(progress, rc = PTR_ERR(inode));
 846                 }
 847
 848                 rc = ll_data_version(inode, &data_version, LL_DV_RD_FLUSH);
 849                 iput(inode);
 850                 if (rc) {
 851                         CDEBUG(D_HSM, "Could not read file data version. "
 852                                       "Request could not be confirmed.\n");
 853                         if (hpk.hpk_errval == 0)
 854                                 hpk.hpk_errval = -rc;
 855                         GOTO(progress, rc);
 856                 }
 857
 858                 /* Store in the hsm_copy for later copytool use.
 859                  * Always modified even if no lsm. */
 860                 hpk.hpk_data_version = data_version;
 861
 862                 /* File could have been stripped during archiving, so we need
 863                  * to check anyway. */
 864                 if ((copy->hc_hai.hai_action == HSMA_ARCHIVE) &&
 865                     (copy->hc_data_version != data_version)) {
 866                         CDEBUG(D_HSM, "File data version mismatched. "
 867                               "File content was changed during archiving. "
 868                                DFID", start:"LPX64" current:"LPX64"\n",
 869                                PFID(&copy->hc_hai.hai_fid),
 870                                copy->hc_data_version, data_version);
 871                         /* File was changed, send error to cdt. Do not ask for
 872                          * retry because if a file is modified frequently,
 873                          * the cdt will loop on retried archive requests.
 874                          * The policy engine will ask for a new archive later
 875                          * when the file will not be modified for some tunable
 876                          * time */
 877                         /* we do not notify caller */
 878                         hpk.hpk_flags &= ~HP_FLAG_RETRY;
 879                         /* hpk_errval must be >= 0 */
 880                         hpk.hpk_errval = EBUSY;
 881                 }
 882
 883         }
 884
 885 progress:
 886         rc2 = obd_iocontrol(LL_IOC_HSM_PROGRESS, sbi->ll_md_exp, sizeof(hpk),
 887                             &hpk, NULL);
 888
 889         /* Return first error */
 890         RETURN(rc != 0 ? rc : rc2);
 891 }
 892
 893
 894 static int copy_and_ioctl(int cmd, struct obd_export *exp,
 895                           const void __user *data, size_t size)
 896 {
 897         void *copy;
 898         int rc;
 899
 900         OBD_ALLOC(copy, size);
 901         if (copy == NULL)
 902                 return -ENOMEM;
 903
 904         if (copy_from_user(copy, data, size)) {
 905                 rc = -EFAULT;
 906                 goto out;
 907         }
 908
 909         rc = obd_iocontrol(cmd, exp, size, copy, NULL);
 910 out:
 911         OBD_FREE(copy, size);
 912
 913         return rc;
 914 }
 915
 916 static int quotactl_ioctl(struct ll_sb_info *sbi, struct if_quotactl *qctl)
 917 {
 918         int cmd = qctl->qc_cmd;
 919         int type = qctl->qc_type;
 920         int id = qctl->qc_id;
 921         int valid = qctl->qc_valid;
 922         int rc = 0;
 923         ENTRY;
 924
 925         switch (cmd) {
 926         case Q_SETQUOTA:
 927         case Q_SETINFO:
 928                 if (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 929                     sbi->ll_flags & LL_SBI_RMT_CLIENT)
 930                         RETURN(-EPERM);
 931                 break;
 932         case Q_GETQUOTA:
 933                 if (((type == USRQUOTA &&
 934                       !uid_eq(current_euid(), make_kuid(&init_user_ns, id))) ||
 935                      (type == GRPQUOTA &&
 936                       !in_egroup_p(make_kgid(&init_user_ns, id)))) &&
 937                     (!cfs_capable(CFS_CAP_SYS_ADMIN) ||
 938                      sbi->ll_flags & LL_SBI_RMT_CLIENT))
 939                         RETURN(-EPERM);
 940                 break;
 941         case Q_GETINFO:
 942                 break;
 943         default:
 944                 CERROR("unsupported quotactl op: %#x\n", cmd);
 945                 RETURN(-ENOTTY);
 946         }
 947
 948         if (valid != QC_GENERAL) {
 949                 if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 950                         RETURN(-EOPNOTSUPP);
 951
 952                 if (cmd == Q_GETINFO)
 953                         qctl->qc_cmd = Q_GETOINFO;
 954                 else if (cmd == Q_GETQUOTA)
 955                         qctl->qc_cmd = Q_GETOQUOTA;
 956                 else
 957                         RETURN(-EINVAL);
 958
 959                 switch (valid) {
 960                 case QC_MDTIDX:
 961                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 962                                            sizeof(*qctl), qctl, NULL);
 963                         break;
 964                 case QC_OSTIDX:
 965                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_dt_exp,
 966                                            sizeof(*qctl), qctl, NULL);
 967                         break;
 968                 case QC_UUID:
 969                         rc = obd_iocontrol(OBD_IOC_QUOTACTL, sbi->ll_md_exp,
 970                                            sizeof(*qctl), qctl, NULL);
 971                         if (rc == -EAGAIN)
 972                                 rc = obd_iocontrol(OBD_IOC_QUOTACTL,
 973                                                    sbi->ll_dt_exp,
 974                                                    sizeof(*qctl), qctl, NULL);
 975                         break;
 976                 default:
 977                         rc = -EINVAL;
 978                         break;
 979                 }
 980
 981                 if (rc)
 982                         RETURN(rc);
 983
 984                 qctl->qc_cmd = cmd;
 985         } else {
 986                 struct obd_quotactl *oqctl;
 987
 988                 OBD_ALLOC_PTR(oqctl);
 989                 if (oqctl == NULL)
 990                         RETURN(-ENOMEM);
 991
 992                 QCTL_COPY(oqctl, qctl);
 993                 rc = obd_quotactl(sbi->ll_md_exp, oqctl);
 994                 if (rc) {
 995                         OBD_FREE_PTR(oqctl);
 996                         RETURN(rc);
 997                 }
 998                 /* If QIF_SPACE is not set, client should collect the
 999                  * space usage from OSSs by itself */
1000                 if (cmd == Q_GETQUOTA &&
1001                     !(oqctl->qc_dqblk.dqb_valid & QIF_SPACE) &&
1002                     !oqctl->qc_dqblk.dqb_curspace) {
1003                         struct obd_quotactl *oqctl_tmp;
1004
1005                         OBD_ALLOC_PTR(oqctl_tmp);
1006                         if (oqctl_tmp == NULL)
1007                                 GOTO(out, rc = -ENOMEM);
1008
1009                         oqctl_tmp->qc_cmd = Q_GETOQUOTA;
1010                         oqctl_tmp->qc_id = oqctl->qc_id;
1011                         oqctl_tmp->qc_type = oqctl->qc_type;
1012
1013                         /* collect space usage from OSTs */
1014                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1015                         rc = obd_quotactl(sbi->ll_dt_exp, oqctl_tmp);
1016                         if (!rc || rc == -EREMOTEIO) {
1017                                 oqctl->qc_dqblk.dqb_curspace =
1018                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1019                                 oqctl->qc_dqblk.dqb_valid |= QIF_SPACE;
1020                         }
1021
1022                         /* collect space & inode usage from MDTs */
1023                         oqctl_tmp->qc_dqblk.dqb_curspace = 0;
1024                         oqctl_tmp->qc_dqblk.dqb_curinodes = 0;
1025                         rc = obd_quotactl(sbi->ll_md_exp, oqctl_tmp);
1026                         if (!rc || rc == -EREMOTEIO) {
1027                                 oqctl->qc_dqblk.dqb_curspace +=
1028                                         oqctl_tmp->qc_dqblk.dqb_curspace;
1029                                 oqctl->qc_dqblk.dqb_curinodes =
1030                                         oqctl_tmp->qc_dqblk.dqb_curinodes;
1031                                 oqctl->qc_dqblk.dqb_valid |= QIF_INODES;
1032                         } else {
1033                                 oqctl->qc_dqblk.dqb_valid &= ~QIF_SPACE;
1034                         }
1035
1036                         OBD_FREE_PTR(oqctl_tmp);
1037                 }
1038 out:
1039                 QCTL_COPY(qctl, oqctl);
1040                 OBD_FREE_PTR(oqctl);
1041         }
1042
1043         RETURN(rc);
1044 }
1045
1046 static char *
1047 ll_getname(const char __user *filename)
1048 {
1049         int ret = 0, len;
1050         char *tmp = __getname();
1051
1052         if (!tmp)
1053                 return ERR_PTR(-ENOMEM);
1054
1055         len = strncpy_from_user(tmp, filename, PATH_MAX);
1056         if (len == 0)
1057                 ret = -ENOENT;
1058         else if (len > PATH_MAX)
1059                 ret = -ENAMETOOLONG;
1060
1061         if (ret) {
1062                 __putname(tmp);
1063                 tmp =  ERR_PTR(ret);
1064         }
1065         return tmp;
1066 }
1067
1068 #define ll_putname(filename) __putname(filename)
1069
1070 static long ll_dir_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1071 {
1072         struct inode *inode = file->f_path.dentry->d_inode;
1073         struct ll_sb_info *sbi = ll_i2sbi(inode);
1074         struct obd_ioctl_data *data;
1075         int rc = 0;
1076         ENTRY;
1077
1078         CDEBUG(D_VFSTRACE, "VFS Op:inode="DFID"(%p), cmd=%#x\n",
1079                PFID(ll_inode2fid(inode)), inode, cmd);
1080
1081         /* asm-ppc{,64} declares TCGETS, et. al. as type 't' not 'T' */
1082         if (_IOC_TYPE(cmd) == 'T' || _IOC_TYPE(cmd) == 't') /* tty ioctls */
1083                 return -ENOTTY;
1084
1085         ll_stats_ops_tally(ll_i2sbi(inode), LPROC_LL_IOCTL, 1);
1086         switch(cmd) {
1087         case FSFILT_IOC_GETFLAGS:
1088         case FSFILT_IOC_SETFLAGS:
1089                 RETURN(ll_iocontrol(inode, file, cmd, arg));
1090         case FSFILT_IOC_GETVERSION_OLD:
1091         case FSFILT_IOC_GETVERSION:
1092                 RETURN(put_user(inode->i_generation, (int __user *)arg));
1093         /* We need to special case any other ioctls we want to handle,
1094          * to send them to the MDS/OST as appropriate and to properly
1095          * network encode the arg field.
1096         case FSFILT_IOC_SETVERSION_OLD:
1097         case FSFILT_IOC_SETVERSION:
1098         */
1099         case LL_IOC_GET_MDTIDX: {
1100                 int mdtidx;
1101
1102                 mdtidx = ll_get_mdt_idx(inode);
1103                 if (mdtidx < 0)
1104                         RETURN(mdtidx);
1105
1106                 if (put_user((int)mdtidx, (int __user *)arg))
1107                         RETURN(-EFAULT);
1108
1109                 return 0;
1110         }
1111         case IOC_MDC_LOOKUP: {
1112                 int namelen, len = 0;
1113                 char *buf = NULL;
1114                 char *filename;
1115
1116                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1117                 if (rc != 0)
1118                         RETURN(rc);
1119                 data = (void *)buf;
1120
1121                 filename = data->ioc_inlbuf1;
1122                 namelen = strlen(filename);
1123                 if (namelen < 1) {
1124                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1125                         GOTO(out_free, rc = -EINVAL);
1126                 }
1127
1128                 rc = ll_get_fid_by_name(inode, filename, namelen, NULL);
1129                 if (rc < 0) {
1130                         CERROR("%s: lookup %.*s failed: rc = %d\n",
1131                                ll_get_fsname(inode->i_sb, NULL, 0), namelen,
1132                                filename, rc);
1133                         GOTO(out_free, rc);
1134                 }
1135 out_free:
1136                 obd_ioctl_freedata(buf, len);
1137                 return rc;
1138         }
1139         case LL_IOC_LMV_SETSTRIPE: {
1140                 struct lmv_user_md  *lum;
1141                 char            *buf = NULL;
1142                 char            *filename;
1143                 int              namelen = 0;
1144                 int              lumlen = 0;
1145                 umode_t          mode;
1146                 int              len;
1147                 int              rc;
1148
1149                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1150                 if (rc)
1151                         RETURN(rc);
1152
1153                 data = (void *)buf;
1154                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1155                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1156                         GOTO(lmv_out_free, rc = -EINVAL);
1157
1158                 filename = data->ioc_inlbuf1;
1159                 namelen = data->ioc_inllen1;
1160
1161                 if (namelen < 1) {
1162                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
1163                         GOTO(lmv_out_free, rc = -EINVAL);
1164                 }
1165                 lum = (struct lmv_user_md *)data->ioc_inlbuf2;
1166                 lumlen = data->ioc_inllen2;
1167
1168                 if (lum->lum_magic != LMV_USER_MAGIC ||
1169                     lumlen != sizeof(*lum)) {
1170                         CERROR("%s: wrong lum magic %x or size %d: rc = %d\n",
1171                                filename, lum->lum_magic, lumlen, -EFAULT);
1172                         GOTO(lmv_out_free, rc = -EINVAL);
1173                 }
1174
1175 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 50, 0)
1176                 mode = data->ioc_type != 0 ? data->ioc_type : S_IRWXUGO;
1177 #else
1178                 mode = data->ioc_type;
1179 #endif
1180                 rc = ll_dir_setdirstripe(inode, lum, filename, mode);
1181 lmv_out_free:
1182                 obd_ioctl_freedata(buf, len);
1183                 RETURN(rc);
1184
1185         }
1186         case LL_IOC_LMV_SET_DEFAULT_STRIPE: {
1187                 struct lmv_user_md        lum;
1188                 struct lmv_user_md __user *ulump =
1189                                         (struct lmv_user_md __user *)arg;
1190                 int                       rc;
1191
1192                 if (copy_from_user(&lum, ulump, sizeof(lum)))
1193                         RETURN(-EFAULT);
1194
1195                 if (lum.lum_magic != LMV_USER_MAGIC)
1196                         RETURN(-EINVAL);
1197
1198                 rc = ll_dir_setstripe(inode, (struct lov_user_md *)&lum, 0);
1199
1200                 RETURN(rc);
1201         }
1202         case LL_IOC_LOV_SETSTRIPE: {
1203                 struct lov_user_md_v3 lumv3;
1204                 struct lov_user_md_v1 *lumv1 = (struct lov_user_md_v1 *)&lumv3;
1205                 struct lov_user_md_v1 __user *lumv1p =
1206                         (struct lov_user_md_v1 __user *)arg;
1207                 struct lov_user_md_v3 __user *lumv3p =
1208                         (struct lov_user_md_v3 __user *)arg;
1209
1210                 int set_default = 0;
1211
1212                 LASSERT(sizeof(lumv3) == sizeof(*lumv3p));
1213                 LASSERT(sizeof(lumv3.lmm_objects[0]) ==
1214                         sizeof(lumv3p->lmm_objects[0]));
1215                 /* first try with v1 which is smaller than v3 */
1216                 if (copy_from_user(lumv1, lumv1p, sizeof(*lumv1)))
1217                         RETURN(-EFAULT);
1218
1219                 if ((lumv1->lmm_magic == LOV_USER_MAGIC_V3) ) {
1220                         if (copy_from_user(&lumv3, lumv3p, sizeof(lumv3)))
1221                                 RETURN(-EFAULT);
1222                 }
1223
1224                 if (inode->i_sb->s_root == file->f_path.dentry)
1225                         set_default = 1;
1226
1227                 /* in v1 and v3 cases lumv1 points to data */
1228                 rc = ll_dir_setstripe(inode, lumv1, set_default);
1229
1230                 RETURN(rc);
1231         }
1232         case LL_IOC_LMV_GETSTRIPE: {
1233                 struct lmv_user_md __user *ulmv =
1234                                         (struct lmv_user_md __user *)arg;
1235                 struct lmv_user_md      lum;
1236                 struct ptlrpc_request   *request = NULL;
1237                 union lmv_mds_md        *lmm = NULL;
1238                 int                     lmmsize;
1239                 u64                     valid = 0;
1240                 struct lmv_user_md      *tmp = NULL;
1241                 int                     mdt_index;
1242                 int                     lum_size;
1243                 int                     stripe_count;
1244                 int                     i;
1245                 int                     rc;
1246
1247                 if (copy_from_user(&lum, ulmv, sizeof(*ulmv)))
1248                         RETURN(-EFAULT);
1249
1250                 /* lum_magic will indicate which stripe the ioctl will like
1251                  * to get, LMV_MAGIC_V1 is for normal LMV stripe, LMV_USER_MAGIC
1252                  * is for default LMV stripe */
1253                 if (lum.lum_magic == LMV_MAGIC_V1)
1254                         valid |= OBD_MD_MEA;
1255                 else if (lum.lum_magic == LMV_USER_MAGIC)
1256                         valid |= OBD_MD_DEFAULT_MEA;
1257                 else
1258                         RETURN(-EINVAL);
1259
1260                 rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize, &request,
1261                                       valid);
1262                 if (rc != 0)
1263                         GOTO(finish_req, rc);
1264
1265                 /* Get default LMV EA */
1266                 if (lum.lum_magic == LMV_USER_MAGIC) {
1267                         if (rc != 0)
1268                                 GOTO(finish_req, rc);
1269
1270                         if (lmmsize > sizeof(*ulmv))
1271                                 GOTO(finish_req, rc = -EINVAL);
1272
1273                         if (copy_to_user(ulmv, lmm, lmmsize))
1274                                 GOTO(finish_req, rc = -EFAULT);
1275
1276                         GOTO(finish_req, rc);
1277                 }
1278
1279                 stripe_count = lmv_mds_md_stripe_count_get(lmm);
1280                 lum_size = lmv_user_md_size(stripe_count, LMV_MAGIC_V1);
1281                 OBD_ALLOC(tmp, lum_size);
1282                 if (tmp == NULL)
1283                         GOTO(finish_req, rc = -ENOMEM);
1284
1285                 mdt_index = ll_get_mdt_idx(inode);
1286                 if (mdt_index < 0)
1287                         GOTO(out_tmp, rc = -ENOMEM);
1288
1289                 tmp->lum_magic = LMV_MAGIC_V1;
1290                 tmp->lum_stripe_count = 0;
1291                 tmp->lum_stripe_offset = mdt_index;
1292                 for (i = 0; i < stripe_count; i++) {
1293                         struct lu_fid   fid;
1294
1295                         fid_le_to_cpu(&fid, &lmm->lmv_md_v1.lmv_stripe_fids[i]);
1296                         mdt_index = ll_get_mdt_idx_by_fid(sbi, &fid);
1297                         if (mdt_index < 0)
1298                                 GOTO(out_tmp, rc = mdt_index);
1299
1300                         tmp->lum_objects[i].lum_mds = mdt_index;
1301                         tmp->lum_objects[i].lum_fid = fid;
1302                         tmp->lum_stripe_count++;
1303                 }
1304
1305                 if (copy_to_user(ulmv, tmp, lum_size))
1306                         GOTO(out_tmp, rc = -EFAULT);
1307 out_tmp:
1308                 OBD_FREE(tmp, lum_size);
1309 finish_req:
1310                 ptlrpc_req_finished(request);
1311                 return rc;
1312         }
1313
1314         case LL_IOC_REMOVE_ENTRY: {
1315                 char            *filename = NULL;
1316                 int              namelen = 0;
1317                 int              rc;
1318
1319                 /* Here is a little hack to avoid sending REINT_RMENTRY to
1320                  * unsupported server, which might crash the server(LU-2730),
1321                  * Because both LVB_TYPE and REINT_RMENTRY will be supported
1322                  * on 2.4, we use OBD_CONNECT_LVB_TYPE to detect whether the
1323                  * server will support REINT_RMENTRY XXX*/
1324                 if (!(exp_connect_flags(sbi->ll_md_exp) & OBD_CONNECT_LVB_TYPE))
1325                         RETURN(-ENOTSUPP);
1326
1327                 filename = ll_getname((const char __user *)arg);
1328                 if (IS_ERR(filename))
1329                         RETURN(PTR_ERR(filename));
1330
1331                 namelen = strlen(filename);
1332                 if (namelen < 1)
1333                         GOTO(out_rmdir, rc = -EINVAL);
1334
1335                 rc = ll_rmdir_entry(inode, filename, namelen);
1336 out_rmdir:
1337                 if (filename)
1338                         ll_putname(filename);
1339                 RETURN(rc);
1340         }
1341         case LL_IOC_LOV_SWAP_LAYOUTS:
1342                 RETURN(-EPERM);
1343         case IOC_OBD_STATFS:
1344                 RETURN(ll_obd_statfs(inode, (void __user *)arg));
1345         case LL_IOC_LOV_GETSTRIPE:
1346         case LL_IOC_MDC_GETINFO:
1347         case IOC_MDC_GETFILEINFO:
1348         case IOC_MDC_GETFILESTRIPE: {
1349                 struct ptlrpc_request *request = NULL;
1350                 struct lov_user_md __user *lump;
1351                 struct lov_mds_md *lmm = NULL;
1352                 struct mdt_body *body;
1353                 char *filename = NULL;
1354                 int lmmsize;
1355
1356                 if (cmd == IOC_MDC_GETFILEINFO ||
1357                     cmd == IOC_MDC_GETFILESTRIPE) {
1358                         filename = ll_getname((const char __user *)arg);
1359                         if (IS_ERR(filename))
1360                                 RETURN(PTR_ERR(filename));
1361
1362                         rc = ll_lov_getstripe_ea_info(inode, filename, &lmm,
1363                                                       &lmmsize, &request);
1364                 } else {
1365                         rc = ll_dir_getstripe(inode, (void **)&lmm, &lmmsize,
1366                                               &request, 0);
1367                 }
1368
1369                 if (request) {
1370                         body = req_capsule_server_get(&request->rq_pill,
1371                                                       &RMF_MDT_BODY);
1372                         LASSERT(body != NULL);
1373                 } else {
1374                         GOTO(out_req, rc);
1375                 }
1376
1377                 if (rc < 0) {
1378                         if (rc == -ENODATA && (cmd == IOC_MDC_GETFILEINFO ||
1379                                                cmd == LL_IOC_MDC_GETINFO))
1380                                 GOTO(skip_lmm, rc = 0);
1381                         else
1382                                 GOTO(out_req, rc);
1383                 }
1384
1385                 if (cmd == IOC_MDC_GETFILESTRIPE ||
1386                     cmd == LL_IOC_LOV_GETSTRIPE) {
1387                         lump = (struct lov_user_md __user *)arg;
1388                 } else {
1389                         struct lov_user_mds_data __user *lmdp;
1390                         lmdp = (struct lov_user_mds_data __user *)arg;
1391                         lump = &lmdp->lmd_lmm;
1392                 }
1393                 if (copy_to_user(lump, lmm, lmmsize)) {
1394                         if (copy_to_user(lump, lmm, sizeof(*lump)))
1395                                 GOTO(out_req, rc = -EFAULT);
1396                         rc = -EOVERFLOW;
1397                 }
1398         skip_lmm:
1399                 if (cmd == IOC_MDC_GETFILEINFO || cmd == LL_IOC_MDC_GETINFO) {
1400                         struct lov_user_mds_data __user *lmdp;
1401                         lstat_t st = { 0 };
1402
1403                         st.st_dev       = inode->i_sb->s_dev;
1404                         st.st_mode      = body->mbo_mode;
1405                         st.st_nlink     = body->mbo_nlink;
1406                         st.st_uid       = body->mbo_uid;
1407                         st.st_gid       = body->mbo_gid;
1408                         st.st_rdev      = body->mbo_rdev;
1409                         st.st_size      = body->mbo_size;
1410                         st.st_blksize   = PAGE_CACHE_SIZE;
1411                         st.st_blocks    = body->mbo_blocks;
1412                         st.st_atime     = body->mbo_atime;
1413                         st.st_mtime     = body->mbo_mtime;
1414                         st.st_ctime     = body->mbo_ctime;
1415                         st.st_ino       = inode->i_ino;
1416
1417                         lmdp = (struct lov_user_mds_data __user *)arg;
1418                         if (copy_to_user(&lmdp->lmd_st, &st, sizeof(st)))
1419                                 GOTO(out_req, rc = -EFAULT);
1420                 }
1421
1422                 EXIT;
1423         out_req:
1424                 ptlrpc_req_finished(request);
1425                 if (filename)
1426                         ll_putname(filename);
1427                 return rc;
1428         }
1429         case OBD_IOC_QUOTACTL: {
1430                 struct if_quotactl *qctl;
1431
1432                 OBD_ALLOC_PTR(qctl);
1433                 if (!qctl)
1434                         RETURN(-ENOMEM);
1435
1436                 if (copy_from_user(qctl, (void __user *)arg, sizeof(*qctl)))
1437                         GOTO(out_quotactl, rc = -EFAULT);
1438
1439                 rc = quotactl_ioctl(sbi, qctl);
1440
1441                 if (rc == 0 &&
1442                     copy_to_user((void __user *)arg, qctl, sizeof(*qctl)))
1443                         rc = -EFAULT;
1444
1445         out_quotactl:
1446                 OBD_FREE_PTR(qctl);
1447                 RETURN(rc);
1448         }
1449         case OBD_IOC_GETDTNAME:
1450         case OBD_IOC_GETMDNAME:
1451                 RETURN(ll_get_obd_name(inode, cmd, arg));
1452         case LL_IOC_FLUSHCTX:
1453                 RETURN(ll_flush_ctx(inode));
1454 #ifdef CONFIG_FS_POSIX_ACL
1455         case LL_IOC_RMTACL: {
1456             if (sbi->ll_flags & LL_SBI_RMT_CLIENT &&
1457                 inode == inode->i_sb->s_root->d_inode) {
1458                 struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1459
1460                 LASSERT(fd != NULL);
1461                 rc = rct_add(&sbi->ll_rct, current_pid(), arg);
1462                 if (!rc)
1463                         fd->fd_flags |= LL_FILE_RMTACL;
1464                 RETURN(rc);
1465             } else
1466                 RETURN(0);
1467         }
1468 #endif
1469         case LL_IOC_GETOBDCOUNT: {
1470                 u32 count, vallen;
1471                 struct obd_export *exp;
1472
1473                 if (copy_from_user(&count, (int __user *)arg, sizeof(int)))
1474                         RETURN(-EFAULT);
1475
1476                 /* get ost count when count is zero, get mdt count otherwise */
1477                 exp = count ? sbi->ll_md_exp : sbi->ll_dt_exp;
1478                 vallen = sizeof(count);
1479                 rc = obd_get_info(NULL, exp, sizeof(KEY_TGT_COUNT),
1480                                   KEY_TGT_COUNT, &vallen, &count);
1481                 if (rc) {
1482                         CERROR("get target count failed: %d\n", rc);
1483                         RETURN(rc);
1484                 }
1485
1486                 if (copy_to_user((int __user *)arg, &count, sizeof(int)))
1487                         RETURN(-EFAULT);
1488
1489                 RETURN(0);
1490         }
1491         case LL_IOC_PATH2FID:
1492                 if (copy_to_user((void __user *)arg, ll_inode2fid(inode),
1493                                      sizeof(struct lu_fid)))
1494                         RETURN(-EFAULT);
1495                 RETURN(0);
1496         case LL_IOC_GET_CONNECT_FLAGS: {
1497                 RETURN(obd_iocontrol(cmd, sbi->ll_md_exp, 0, NULL,
1498                                      (void __user *)arg));
1499         }
1500         case OBD_IOC_CHANGELOG_SEND:
1501         case OBD_IOC_CHANGELOG_CLEAR:
1502                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1503                                     sizeof(struct ioc_changelog));
1504                 RETURN(rc);
1505         case OBD_IOC_FID2PATH:
1506                 RETURN(ll_fid2path(inode, (void __user *)arg));
1507         case LL_IOC_GETPARENT:
1508                 RETURN(ll_getparent(file, (void __user *)arg));
1509         case LL_IOC_FID2MDTIDX: {
1510                 struct obd_export *exp = ll_i2mdexp(inode);
1511                 struct lu_fid     fid;
1512                 __u32             index;
1513
1514                 if (copy_from_user(&fid, (const struct lu_fid __user *)arg,
1515                                    sizeof(fid)))
1516                         RETURN(-EFAULT);
1517
1518                 /* Call mdc_iocontrol */
1519                 rc = obd_iocontrol(LL_IOC_FID2MDTIDX, exp, sizeof(fid), &fid,
1520                                    (__u32 __user *)&index);
1521                 if (rc != 0)
1522                         RETURN(rc);
1523
1524                 RETURN(index);
1525         }
1526         case LL_IOC_HSM_REQUEST: {
1527                 struct hsm_user_request *hur;
1528                 ssize_t                  totalsize;
1529
1530                 OBD_ALLOC_PTR(hur);
1531                 if (hur == NULL)
1532                         RETURN(-ENOMEM);
1533
1534                 /* We don't know the true size yet; copy the fixed-size part */
1535                 if (copy_from_user(hur, (void __user *)arg, sizeof(*hur))) {
1536                         OBD_FREE_PTR(hur);
1537                         RETURN(-EFAULT);
1538                 }
1539
1540                 /* Compute the whole struct size */
1541                 totalsize = hur_len(hur);
1542                 OBD_FREE_PTR(hur);
1543                 if (totalsize < 0)
1544                         RETURN(-E2BIG);
1545
1546                 /* Final size will be more than double totalsize */
1547                 if (totalsize >= MDS_MAXREQSIZE / 3)
1548                         RETURN(-E2BIG);
1549
1550                 OBD_ALLOC_LARGE(hur, totalsize);
1551                 if (hur == NULL)
1552                         RETURN(-ENOMEM);
1553
1554                 /* Copy the whole struct */
1555                 if (copy_from_user(hur, (void __user *)arg, totalsize)) {
1556                         OBD_FREE_LARGE(hur, totalsize);
1557                         RETURN(-EFAULT);
1558                 }
1559
1560                 if (hur->hur_request.hr_action == HUA_RELEASE) {
1561                         const struct lu_fid *fid;
1562                         struct inode *f;
1563                         int i;
1564
1565                         for (i = 0; i < hur->hur_request.hr_itemcount; i++) {
1566                                 fid = &hur->hur_user_item[i].hui_fid;
1567                                 f = search_inode_for_lustre(inode->i_sb, fid);
1568                                 if (IS_ERR(f)) {
1569                                         rc = PTR_ERR(f);
1570                                         break;
1571                                 }
1572
1573                                 rc = ll_hsm_release(f);
1574                                 iput(f);
1575                                 if (rc != 0)
1576                                         break;
1577                         }
1578                 } else {
1579                         rc = obd_iocontrol(cmd, ll_i2mdexp(inode), totalsize,
1580                                            hur, NULL);
1581                 }
1582
1583                 OBD_FREE_LARGE(hur, totalsize);
1584
1585                 RETURN(rc);
1586         }
1587         case LL_IOC_HSM_PROGRESS: {
1588                 struct hsm_progress_kernel      hpk;
1589                 struct hsm_progress             hp;
1590
1591                 if (copy_from_user(&hp, (void __user *)arg, sizeof(hp)))
1592                         RETURN(-EFAULT);
1593
1594                 hpk.hpk_fid = hp.hp_fid;
1595                 hpk.hpk_cookie = hp.hp_cookie;
1596                 hpk.hpk_extent = hp.hp_extent;
1597                 hpk.hpk_flags = hp.hp_flags;
1598                 hpk.hpk_errval = hp.hp_errval;
1599                 hpk.hpk_data_version = 0;
1600
1601                 /* File may not exist in Lustre; all progress
1602                  * reported to Lustre root */
1603                 rc = obd_iocontrol(cmd, sbi->ll_md_exp, sizeof(hpk), &hpk,
1604                                    NULL);
1605                 RETURN(rc);
1606         }
1607         case LL_IOC_HSM_CT_START:
1608                 if (!cfs_capable(CFS_CAP_SYS_ADMIN))
1609                         RETURN(-EPERM);
1610
1611                 rc = copy_and_ioctl(cmd, sbi->ll_md_exp, (void __user *)arg,
1612                                     sizeof(struct lustre_kernelcomm));
1613                 RETURN(rc);
1614
1615         case LL_IOC_HSM_COPY_START: {
1616                 struct hsm_copy *copy;
1617                 int              rc;
1618
1619                 OBD_ALLOC_PTR(copy);
1620                 if (copy == NULL)
1621                         RETURN(-ENOMEM);
1622                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1623                         OBD_FREE_PTR(copy);
1624                         RETURN(-EFAULT);
1625                 }
1626
1627                 rc = ll_ioc_copy_start(inode->i_sb, copy);
1628                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1629                         rc = -EFAULT;
1630
1631                 OBD_FREE_PTR(copy);
1632                 RETURN(rc);
1633         }
1634         case LL_IOC_HSM_COPY_END: {
1635                 struct hsm_copy *copy;
1636                 int              rc;
1637
1638                 OBD_ALLOC_PTR(copy);
1639                 if (copy == NULL)
1640                         RETURN(-ENOMEM);
1641                 if (copy_from_user(copy, (char __user *)arg, sizeof(*copy))) {
1642                         OBD_FREE_PTR(copy);
1643                         RETURN(-EFAULT);
1644                 }
1645
1646                 rc = ll_ioc_copy_end(inode->i_sb, copy);
1647                 if (copy_to_user((char __user *)arg, copy, sizeof(*copy)))
1648                         rc = -EFAULT;
1649
1650                 OBD_FREE_PTR(copy);
1651                 RETURN(rc);
1652         }
1653         case LL_IOC_MIGRATE: {
1654                 char            *buf = NULL;
1655                 const char      *filename;
1656                 int             namelen = 0;
1657                 int             len;
1658                 int             rc;
1659                 int             mdtidx;
1660
1661                 rc = obd_ioctl_getdata(&buf, &len, (void __user *)arg);
1662                 if (rc < 0)
1663                         RETURN(rc);
1664
1665                 data = (struct obd_ioctl_data *)buf;
1666                 if (data->ioc_inlbuf1 == NULL || data->ioc_inlbuf2 == NULL ||
1667                     data->ioc_inllen1 == 0 || data->ioc_inllen2 == 0)
1668                         GOTO(migrate_free, rc = -EINVAL);
1669
1670                 filename = data->ioc_inlbuf1;
1671                 namelen = data->ioc_inllen1;
1672                 /* \0 is packed at the end of filename */
1673                 if (namelen < 1 || namelen != strlen(filename) + 1)
1674                         GOTO(migrate_free, rc = -EINVAL);
1675
1676                 if (data->ioc_inllen2 != sizeof(mdtidx))
1677                         GOTO(migrate_free, rc = -EINVAL);
1678                 mdtidx = *(int *)data->ioc_inlbuf2;
1679
1680                 rc = ll_migrate(inode, file, mdtidx, filename, namelen - 1);
1681 migrate_free:
1682                 obd_ioctl_freedata(buf, len);
1683
1684                 RETURN(rc);
1685         }
1686         default:
1687                 RETURN(obd_iocontrol(cmd, sbi->ll_dt_exp, 0, NULL,
1688                                      (void __user *)arg));
1689         }
1690 }
1691
1692 static loff_t ll_dir_seek(struct file *file, loff_t offset, int origin)
1693 {
1694         struct inode *inode = file->f_mapping->host;
1695         struct ll_file_data *fd = LUSTRE_FPRIVATE(file);
1696         struct ll_sb_info *sbi = ll_i2sbi(inode);
1697         int api32 = ll_need_32bit_api(sbi);
1698         loff_t ret = -EINVAL;
1699         ENTRY;
1700
1701         mutex_lock(&inode->i_mutex);
1702         switch (origin) {
1703                 case SEEK_SET:
1704                         break;
1705                 case SEEK_CUR:
1706                         offset += file->f_pos;
1707                         break;
1708                 case SEEK_END:
1709                         if (offset > 0)
1710                                 GOTO(out, ret);
1711                         if (api32)
1712                                 offset += LL_DIR_END_OFF_32BIT;
1713                         else
1714                                 offset += LL_DIR_END_OFF;
1715                         break;
1716                 default:
1717                         GOTO(out, ret);
1718         }
1719
1720         if (offset >= 0 &&
1721             ((api32 && offset <= LL_DIR_END_OFF_32BIT) ||
1722              (!api32 && offset <= LL_DIR_END_OFF))) {
1723                 if (offset != file->f_pos) {
1724                         if ((api32 && offset == LL_DIR_END_OFF_32BIT) ||
1725                             (!api32 && offset == LL_DIR_END_OFF))
1726                                 fd->lfd_pos = MDS_DIR_END_OFF;
1727                         else if (api32 && sbi->ll_flags & LL_SBI_64BIT_HASH)
1728                                 fd->lfd_pos = offset << 32;
1729                         else
1730                                 fd->lfd_pos = offset;
1731                         file->f_pos = offset;
1732                         file->f_version = 0;
1733                 }
1734                 ret = offset;
1735         }
1736         GOTO(out, ret);
1737
1738 out:
1739         mutex_unlock(&inode->i_mutex);
1740         return ret;
1741 }
1742
1743 static int ll_dir_open(struct inode *inode, struct file *file)
1744 {
1745         ENTRY;
1746         RETURN(ll_file_open(inode, file));
1747 }
1748
1749 static int ll_dir_release(struct inode *inode, struct file *file)
1750 {
1751         ENTRY;
1752         RETURN(ll_file_release(inode, file));
1753 }
1754
1755 const struct file_operations ll_dir_operations = {
1756         .llseek         = ll_dir_seek,
1757         .open           = ll_dir_open,
1758         .release        = ll_dir_release,
1759         .read           = generic_read_dir,
1760 #ifdef HAVE_DIR_CONTEXT
1761         .iterate        = ll_iterate,
1762 #else
1763         .readdir        = ll_readdir,
1764 #endif
1765         .unlocked_ioctl = ll_dir_ioctl,
1766         .fsync          = ll_fsync,
1767 };