lustre/llite/dir.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/dir.c
  12  *  linux/fs/ext2/dir.c
  13  *
  14  *  Copyright (C) 1991, 1992  Linus Torvalds
  15  *
  16  *  ext2 directory handling functions
  17  *
  18  *  Big-endian to little-endian byte-swapping/bitmaps by
  19  *        David S. Miller (davem@caip.rutgers.edu), 1995
  20  *
  21  *  All code that works with directory layout had been switched to pagecache
  22  *  and moved here. AV
  23  *
  24  *  Adapted for Lustre Light
  25  *  Copyright (C) 2002-2003, Cluster File Systems, Inc.
  26  *
  27  */
  28
  29 #include <linux/fs.h>
  30 #include <linux/ext2_fs.h>
  31 #include <linux/pagemap.h>
  32 #include <linux/mm.h>
  33 #include <linux/version.h>
  34 #include <linux/smp_lock.h>
  35 #include <asm/uaccess.h>
  36 #include <linux/file.h>
  37 #include <linux/kmod.h>
  38 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0))
  39 # include <linux/locks.h>   // for wait_on_buffer
  40 #else
  41 # include <linux/buffer_head.h>   // for wait_on_buffer
  42 #endif
  43
  44 #define DEBUG_SUBSYSTEM S_LLITE
  45
  46 #include <linux/obd_support.h>
  47 #include <linux/obd_class.h>
  48 #include <linux/lustre_lib.h>
  49 #include <linux/lustre_idl.h>
  50 #include <linux/lustre_mds.h>
  51 #include <linux/lustre_lite.h>
  52 #include <linux/lustre_dlm.h>
  53 #include <linux/lustre_sec.h>
  54 #include <linux/lustre_audit.h>
  55 #include <linux/lustre_acl.h>
  56 #include "llite_internal.h"
  57
  58 typedef struct ext2_dir_entry_2 ext2_dirent;
  59
  60 #define PageChecked(page)        test_bit(PG_checked, &(page)->flags)
  61 #define SetPageChecked(page)     set_bit(PG_checked, &(page)->flags)
  62
  63 /* returns the page unlocked, but with a reference */
  64 static int ll_dir_readpage(struct file *file, struct page *page)
  65 {
  66         struct inode *inode = page->mapping->host;
  67         struct ptlrpc_request *request;
  68         struct mds_body *body;
  69         struct lustre_id id;
  70         __u64 offset;
  71         int rc = 0;
  72         ENTRY;
  73
  74         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p)\n", inode->i_ino,
  75                inode->i_generation, inode);
  76
  77         mdc_pack_id(&id, inode->i_ino, inode->i_generation,
  78                     S_IFDIR, id_group(&ll_i2info(inode)->lli_id),
  79                     id_fid(&ll_i2info(inode)->lli_id));
  80
  81         offset = page->index << PAGE_SHIFT;
  82         rc = md_readpage(ll_i2sbi(inode)->ll_md_exp, &id, offset,
  83                          page, &request);
  84         if (!rc) {
  85                 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
  86                 LASSERT (body != NULL);          /* checked by md_readpage() */
  87                 LASSERT_REPSWABBED (request, 0); /* swabbed by md_readpage() */
  88
  89                 //inode->i_size = body->size;
  90                 SetPageUptodate(page);
  91         }
  92         ptlrpc_req_finished(request);
  93
  94         unlock_page(page);
  95         RETURN(rc);
  96 }
  97
  98 struct address_space_operations ll_dir_aops = {
  99         .readpage  = ll_dir_readpage,
 100 };
 101
 102 /*
 103  * ext2 uses block-sized chunks. Arguably, sector-sized ones would be
 104  * more robust, but we have what we have
 105  */
 106 static inline unsigned ext2_chunk_size(struct inode *inode)
 107 {
 108         return inode->i_sb->s_blocksize;
 109 }
 110
 111 static inline void ext2_put_page(struct page *page)
 112 {
 113         kunmap(page);
 114         page_cache_release(page);
 115 }
 116
 117 static inline unsigned long dir_pages(struct inode *inode)
 118 {
 119         return (inode->i_size+PAGE_CACHE_SIZE-1)>>PAGE_CACHE_SHIFT;
 120 }
 121
 122 static void ext2_check_page(struct page *page)
 123 {
 124         struct inode *dir = page->mapping->host;
 125         unsigned chunk_size = ext2_chunk_size(dir);
 126         char *kaddr = page_address(page);
 127         //      u32 max_inumber = le32_to_cpu(sb->u.ext2_sb.s_es->s_inodes_count);
 128         unsigned offs, rec_len;
 129         unsigned limit = PAGE_CACHE_SIZE;
 130         ext2_dirent *p;
 131         char *error;
 132
 133         if ((dir->i_size >> PAGE_CACHE_SHIFT) == page->index) {
 134                 limit = dir->i_size & ~PAGE_CACHE_MASK;
 135                 if (limit & (chunk_size - 1)) {
 136                         CERROR("limit %d dir size %lld index %ld\n",
 137                                limit, dir->i_size, page->index);
 138                         goto Ebadsize;
 139                 }
 140                 for (offs = limit; offs<PAGE_CACHE_SIZE; offs += chunk_size) {
 141                         ext2_dirent *p = (ext2_dirent*)(kaddr + offs);
 142                         p->rec_len = cpu_to_le16(chunk_size);
 143                         p->name_len = 0;
 144                         p->inode = 0;
 145                 }
 146                 if (!limit)
 147                         goto out;
 148         }
 149         for (offs = 0; offs <= limit - EXT2_DIR_REC_LEN(1); offs += rec_len) {
 150                 p = (ext2_dirent *)(kaddr + offs);
 151                 rec_len = le16_to_cpu(p->rec_len);
 152
 153                 if (rec_len < EXT2_DIR_REC_LEN(1))
 154                         goto Eshort;
 155                 if (rec_len & 3)
 156                         goto Ealign;
 157                 if (rec_len < EXT2_DIR_REC_LEN(p->name_len))
 158                         goto Enamelen;
 159                 if (((offs + rec_len - 1) ^ offs) & ~(chunk_size-1))
 160                         goto Espan;
 161                 //              if (le32_to_cpu(p->inode) > max_inumber)
 162                 //goto Einumber;
 163         }
 164         if (offs != limit)
 165                 goto Eend;
 166 out:
 167         SetPageChecked(page);
 168         return;
 169
 170         /* Too bad, we had an error */
 171
 172 Ebadsize:
 173         CERROR("ext2_check_page"
 174                 "size of directory #%lu is not a multiple of chunk size\n",
 175                 dir->i_ino
 176         );
 177         goto fail;
 178 Eshort:
 179         error = "rec_len is smaller than minimal";
 180         goto bad_entry;
 181 Ealign:
 182         error = "unaligned directory entry";
 183         goto bad_entry;
 184 Enamelen:
 185         error = "rec_len is too small for name_len";
 186         goto bad_entry;
 187 Espan:
 188         error = "directory entry across blocks";
 189         goto bad_entry;
 190         //Einumber:
 191         // error = "inode out of bounds";
 192 bad_entry:
 193         CERROR("ext2_check_page: bad entry in directory #%lu: %s - "
 194                 "offset=%lu+%u, inode=%lu, rec_len=%d, name_len=%d",
 195                 dir->i_ino, error, (page->index<<PAGE_CACHE_SHIFT), offs,
 196                 (unsigned long) le32_to_cpu(p->inode),
 197                 rec_len, p->name_len);
 198         goto fail;
 199 Eend:
 200         p = (ext2_dirent *)(kaddr + offs);
 201         CERROR("ext2_check_page"
 202                 "entry in directory #%lu spans the page boundary"
 203                 "offset=%lu, inode=%lu",
 204                 dir->i_ino, (page->index<<PAGE_CACHE_SHIFT)+offs,
 205                 (unsigned long) le32_to_cpu(p->inode));
 206 fail:
 207         SetPageChecked(page);
 208         SetPageError(page);
 209 }
 210
 211 static struct page *ll_get_dir_page(struct inode *dir, unsigned long n)
 212 {
 213         struct ll_inode_info *li = ll_i2info(dir);
 214         struct ldlm_res_id res_id =
 215                 { .name = { id_fid(&li->lli_id), id_group(&li->lli_id)} };
 216         struct lustre_handle lockh;
 217         struct obd_device *obddev = class_exp2obd(ll_i2sbi(dir)->ll_md_exp);
 218         struct address_space *mapping = dir->i_mapping;
 219         struct page *page;
 220         ldlm_policy_data_t policy = { .l_inodebits = { MDS_INODELOCK_UPDATE } };
 221         int rc;
 222
 223         obddev = md_get_real_obd(ll_i2sbi(dir)->ll_md_exp, &li->lli_id);
 224         rc = ldlm_lock_match(obddev->obd_namespace, LDLM_FL_BLOCK_GRANTED,
 225                              &res_id, LDLM_IBITS, &policy, LCK_PR, &lockh);
 226         if (!rc) {
 227                 struct lookup_intent it = { .it_op = IT_READDIR };
 228                 struct ptlrpc_request *request;
 229                 struct mdc_op_data *op_data;
 230
 231                 OBD_ALLOC(op_data, sizeof(*op_data));
 232                 if (op_data == NULL)
 233                         return ERR_PTR(-ENOMEM);
 234
 235                 ll_inode2mdc_data(op_data, dir, (OBD_MD_FLID | OBD_MD_MEA));
 236
 237                 rc = ll_intent_alloc(&it);
 238                 if (rc)
 239                         return ERR_PTR(rc);
 240
 241                 rc = md_enqueue(ll_i2sbi(dir)->ll_md_exp, LDLM_IBITS, &it,
 242                                 LCK_PR, op_data, &lockh, NULL, 0,
 243                                 ldlm_completion_ast, ll_mdc_blocking_ast, dir);
 244                 OBD_FREE(op_data, sizeof(*op_data));
 245
 246                 request = (struct ptlrpc_request *)LUSTRE_IT(&it)->it_data;
 247                 ll_intent_free(&it);
 248
 249                 if (request)
 250                         ptlrpc_req_finished(request);
 251                 if (rc < 0) {
 252                         CERROR("lock enqueue: rc: %d\n", rc);
 253                         return ERR_PTR(rc);
 254                 }
 255         } else {
 256                 if (ptlrpcs_check_cred(obddev->u.cli.cl_import)) {
 257                         /* return immediately if no credential held */
 258                         ldlm_lock_decref(&lockh, LCK_PR);
 259                         return ERR_PTR(-EACCES);
 260                 }
 261         }
 262         ldlm_lock_dump_handle(D_OTHER, &lockh);
 263
 264         page = read_cache_page(mapping, n,
 265                                (filler_t*)mapping->a_ops->readpage, NULL);
 266         if (!IS_ERR(page)) {
 267                 wait_on_page(page);
 268                 (void)kmap(page);
 269                 if (!PageUptodate(page))
 270                         goto fail;
 271                 if (!PageChecked(page))
 272                         ext2_check_page(page);
 273                 if (PageError(page))
 274                         goto fail;
 275         }
 276
 277 out_unlock:
 278         ldlm_lock_decref(&lockh, LCK_PR);
 279         return page;
 280
 281 fail:
 282         ext2_put_page(page);
 283         page = ERR_PTR(-EIO);
 284         goto out_unlock;
 285 }
 286
 287 /*
 288  * p is at least 6 bytes before the end of page
 289  */
 290 static inline ext2_dirent *ext2_next_entry(ext2_dirent *p)
 291 {
 292         return (ext2_dirent *)((char*)p + le16_to_cpu(p->rec_len));
 293 }
 294
 295 static inline unsigned
 296 ext2_validate_entry(char *base, unsigned offset, unsigned mask)
 297 {
 298         ext2_dirent *de = (ext2_dirent*)(base + offset);
 299         ext2_dirent *p = (ext2_dirent*)(base + (offset&mask));
 300         while ((char*)p < (char*)de)
 301                 p = ext2_next_entry(p);
 302         return (char *)p - base;
 303 }
 304
 305 static unsigned char ext2_filetype_table[EXT2_FT_MAX] = {
 306         [EXT2_FT_UNKNOWN]       DT_UNKNOWN,
 307         [EXT2_FT_REG_FILE]      DT_REG,
 308         [EXT2_FT_DIR]           DT_DIR,
 309         [EXT2_FT_CHRDEV]        DT_CHR,
 310         [EXT2_FT_BLKDEV]        DT_BLK,
 311         [EXT2_FT_FIFO]          DT_FIFO,
 312         [EXT2_FT_SOCK]          DT_SOCK,
 313         [EXT2_FT_SYMLINK]       DT_LNK,
 314 };
 315
 316
 317 int ll_readdir(struct file * filp, void * dirent, filldir_t filldir)
 318 {
 319         struct inode *inode = filp->f_dentry->d_inode;
 320         loff_t pos = filp->f_pos;
 321         // XXX struct super_block *sb = inode->i_sb;
 322         unsigned offset = pos & ~PAGE_CACHE_MASK;
 323         unsigned long n = pos >> PAGE_CACHE_SHIFT;
 324         unsigned long npages = dir_pages(inode);
 325         unsigned chunk_mask = ~(ext2_chunk_size(inode)-1);
 326         unsigned char *types = ext2_filetype_table;
 327         int need_revalidate = (filp->f_version != inode->i_version);
 328         int rc = 0;
 329         ENTRY;
 330
 331         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p) pos %llu/%llu\n",
 332                inode->i_ino, inode->i_generation, inode, pos, inode->i_size);
 333
 334         if (pos > inode->i_size - EXT2_DIR_REC_LEN(1))
 335                 RETURN(0);
 336
 337         for ( ; n < npages; n++, offset = 0) {
 338                 char *kaddr, *limit;
 339                 ext2_dirent *de;
 340                 struct page *page;
 341
 342                 CDEBUG(D_EXT2,"read %lu of dir %lu/%u page %lu/%lu size %llu\n",
 343                        PAGE_CACHE_SIZE, inode->i_ino, inode->i_generation,
 344                        n, npages, inode->i_size);
 345                 page = ll_get_dir_page(inode, n);
 346
 347                 /* size might have been updated by md_readpage() */
 348                 npages = dir_pages(inode);
 349
 350                 if (IS_ERR(page)) {
 351                         rc = PTR_ERR(page);
 352                         CERROR("error reading dir %lu/%u page %lu: rc %d\n",
 353                                inode->i_ino, inode->i_generation, n, rc);
 354                         continue;
 355                 }
 356
 357                 kaddr = page_address(page);
 358                 if (need_revalidate) {
 359                         offset = ext2_validate_entry(kaddr, offset, chunk_mask);
 360                         need_revalidate = 0;
 361                 }
 362                 de = (ext2_dirent *)(kaddr+offset);
 363                 limit = kaddr + PAGE_CACHE_SIZE - EXT2_DIR_REC_LEN(1);
 364                 for ( ;(char*)de <= limit; de = ext2_next_entry(de)) {
 365                         if (de->inode) {
 366                                 int over;
 367
 368                                 rc = 0; /* no error if we return something */
 369
 370                                 offset = (char *)de - kaddr;
 371                                 over = filldir(dirent, de->name, de->name_len,
 372                                                (n<<PAGE_CACHE_SHIFT) | offset,
 373                                                le32_to_cpu(de->inode),
 374                                                types[de->file_type &
 375                                                      (EXT2_FT_MAX - 1)]);
 376
 377                                 if (over) {
 378                                         ext2_put_page(page);
 379                                         GOTO(done, rc);
 380                                 }
 381                         }
 382                 }
 383                 ext2_put_page(page);
 384         }
 385
 386         EXIT;
 387 done:
 388         filp->f_pos = (n << PAGE_CACHE_SHIFT) | offset;
 389         filp->f_version = inode->i_version;
 390         update_atime(inode);
 391         return rc;
 392 }
 393
 394 static int ll_mkdir_stripe(struct inode *inode, unsigned long arg)
 395 {
 396         struct ptlrpc_request *request = NULL;
 397         struct ll_sb_info *sbi = ll_i2sbi(inode);
 398         struct ll_user_mkdir_stripe lums;
 399         struct mdc_op_data *op_data;
 400         u16 nstripes;
 401         mode_t mode;
 402         char *name;
 403         int err = 0;
 404         ENTRY;
 405
 406         if (copy_from_user(&lums, (void *)arg, sizeof(lums)))
 407                 RETURN(-EFAULT);
 408
 409         if (lums.lums_namelen <= 0)
 410                 RETURN(-EINVAL);
 411         OBD_ALLOC(name, lums.lums_namelen);
 412         if (!name)
 413                 RETURN(-ENOMEM);
 414
 415         if (copy_from_user(name, lums.lums_name, lums.lums_namelen))
 416                 GOTO(out, err = -EFAULT);
 417
 418         CDEBUG(D_VFSTRACE, "ioctl Op:name=%s,dir=%lu/%u(%p)\n",
 419                name, inode->i_ino, inode->i_generation, inode);
 420         nstripes = lums.lums_nstripes;
 421
 422         mode = lums.lums_mode;
 423         mode = (mode & (S_IRWXUGO|S_ISVTX) & ~current->fs->umask) | S_IFDIR;
 424
 425         OBD_ALLOC(op_data, sizeof(*op_data));
 426         if (op_data == NULL)
 427                 GOTO(out, err = -ENOMEM);
 428         ll_prepare_mdc_data(op_data, inode, NULL, name, lums.lums_namelen, 0);
 429         err = md_create(sbi->ll_md_exp, op_data, &nstripes, sizeof(nstripes),
 430                         mode, current->fsuid, current->fsgid, 0, &request);
 431         OBD_FREE(op_data, sizeof(*op_data));
 432         ptlrpc_req_finished(request);
 433         EXIT;
 434 out:
 435         OBD_FREE(name, lums.lums_namelen);
 436         return err;
 437 }
 438
 439 /*
 440  * we don't call getxattr_internal/setxattr_internal because we
 441  * need more precisely control.
 442  */
 443 static int ll_ioctl_getfacl(struct inode *inode,
 444                             struct file *file,
 445                             struct ll_acl_ioctl_data *ioc)
 446 {
 447         struct ptlrpc_request *req = NULL;
 448         struct mds_body *body;
 449         char *cmd, *res;
 450         struct lustre_id id;
 451         int rc;
 452         ENTRY;
 453
 454         if (!ioc->cmd || !ioc->cmd_len ||
 455             !ioc->res || !ioc->res_len) {
 456                 CERROR("error: cmd %p, len %lu, res %p, len %lu\n",
 457                        ioc->cmd, ioc->cmd_len, ioc->res, ioc->res_len);
 458                 RETURN(-EINVAL);
 459         }
 460
 461         OBD_ALLOC(cmd, ioc->cmd_len);
 462         if (!cmd)
 463                 RETURN(-ENOMEM);
 464         if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
 465                 GOTO(out, rc = -EFAULT);
 466
 467         /* we didn't call ll_getxattr_internal() because we'd like to
 468          * copy from reply buffer to user space directly.
 469          */
 470         ll_inode2id(&id, inode);
 471         rc = md_getattr(ll_i2sbi(inode)->ll_md_exp, &id, OBD_MD_FLXATTR,
 472                         XATTR_NAME_LUSTRE_ACL,
 473                         cmd, ioc->cmd_len, ioc->res_len, NULL, &req);
 474         if (rc < 0) {
 475                 CERROR("rc: %d\n", rc);
 476                 GOTO(out, rc);
 477         }
 478
 479         res = lustre_msg_buf(req->rq_repmsg, 1, ioc->res_len);
 480         LASSERT(res);
 481         if (copy_to_user(ioc->res, res, ioc->res_len))
 482                 rc = -EFAULT;
 483
 484         body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
 485         LASSERT(body);
 486         ioc->status = (__s32) body->flags;
 487
 488         EXIT;
 489 out:
 490         if (req)
 491                 ptlrpc_req_finished(req);
 492         OBD_FREE(cmd, ioc->cmd_len);
 493
 494         return rc;
 495 }
 496
 497 static int ll_ioctl_setfacl(struct inode *inode,
 498                             struct file *file,
 499                             struct ll_acl_ioctl_data *ioc)
 500 {
 501         struct ptlrpc_request *req = NULL;
 502         struct mdc_op_data *op_data;
 503         struct mds_body *body;
 504         struct iattr attr;
 505         char *cmd;
 506         int replen, rc;
 507         ENTRY;
 508
 509         if (!ioc->cmd || !ioc->cmd_len) {
 510                 CERROR("error: cmd %p, len %lu\n", ioc->cmd, ioc->cmd_len);
 511                 RETURN(-EINVAL);
 512         }
 513
 514         OBD_ALLOC(cmd, ioc->cmd_len);
 515         if (!cmd)
 516                 RETURN(-ENOMEM);
 517         if (copy_from_user(cmd, ioc->cmd, ioc->cmd_len))
 518                 GOTO(out, rc = -EFAULT);
 519
 520         memset(&attr, 0x0, sizeof(attr));
 521         attr.ia_valid |= ATTR_EA;
 522         attr.ia_attr_flags = 0;
 523
 524         OBD_ALLOC(op_data, sizeof(*op_data));
 525         if (!op_data)
 526                 GOTO(out, rc = -ENOMEM);
 527
 528         ll_inode2mdc_data(op_data, inode, (OBD_MD_FLID | OBD_MD_MEA));
 529
 530         rc = md_setattr(ll_i2sbi(inode)->ll_md_exp, op_data, &attr,
 531                         (void *)XATTR_NAME_LUSTRE_ACL,
 532                         sizeof(XATTR_NAME_LUSTRE_ACL),
 533                         (void *)cmd, ioc->cmd_len, NULL, 0, &req);
 534         OBD_FREE(op_data, sizeof(*op_data));
 535
 536         if (rc) {
 537                 CERROR("md_setattr fails: rc = %d\n", rc);
 538                 GOTO(out, rc);
 539         }
 540
 541         body = lustre_msg_buf(req->rq_repmsg, 0, sizeof(*body));
 542         LASSERT(body);
 543         ioc->status = (__s32) body->flags;
 544
 545         LASSERT(req->rq_repmsg->bufcount == 2);
 546         replen = req->rq_repmsg->buflens[1];
 547         LASSERT(replen <= LUSTRE_ACL_SIZE_MAX);
 548         if (replen) {
 549                 if (replen > ioc->res_len)
 550                         replen = ioc->res_len;
 551                 if (copy_to_user(ioc->res,
 552                                  lustre_msg_buf(req->rq_repmsg, 1, replen),
 553                                  replen))
 554                         rc = -EFAULT;
 555         }
 556         EXIT;
 557 out:
 558         if (req)
 559                 ptlrpc_req_finished(req);
 560         OBD_FREE(cmd, ioc->cmd_len);
 561
 562         return rc;
 563 }
 564
 565 static int ll_dir_ioctl(struct inode *inode, struct file *file,
 566                         unsigned int cmd, unsigned long arg)
 567 {
 568         struct ll_sb_info *sbi = ll_i2sbi(inode);
 569         struct obd_ioctl_data *data;
 570         ENTRY;
 571
 572         CDEBUG(D_VFSTRACE, "VFS Op:inode=%lu/%u(%p), cmd=%#x\n",
 573                inode->i_ino, inode->i_generation, inode, cmd);
 574
 575         if (_IOC_TYPE(cmd) == 'T') /* tty ioctls */
 576                 return -ENOTTY;
 577
 578         lprocfs_counter_incr(ll_i2sbi(inode)->ll_stats, LPROC_LL_IOCTL);
 579         switch(cmd) {
 580         case EXT3_IOC_GETFLAGS:
 581         case EXT3_IOC_SETFLAGS:
 582                 RETURN(ll_iocontrol(inode, file, cmd, arg));
 583         case IOC_MDC_LOOKUP: {
 584                 struct ptlrpc_request *request = NULL;
 585                 struct lustre_id id;
 586                 char *buf = NULL;
 587                 char *filename;
 588                 int namelen, rc, len = 0;
 589                 __u64 valid;
 590
 591                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 592                 if (rc)
 593                         RETURN(rc);
 594                 data = (void *)buf;
 595
 596                 filename = data->ioc_inlbuf1;
 597                 namelen = data->ioc_inllen1;
 598
 599                 if (namelen < 1) {
 600                         CDEBUG(D_INFO, "IOC_MDC_LOOKUP missing filename\n");
 601                         GOTO(out, rc = -EINVAL);
 602                 }
 603
 604                 valid = OBD_MD_FLID;
 605                 ll_inode2id(&id, inode);
 606                 rc = md_getattr_lock(sbi->ll_md_exp, &id,
 607                                      filename, namelen, valid, 0, &request);
 608                 if (rc < 0) {
 609                         CDEBUG(D_INFO, "md_getattr_lock: %d\n", rc);
 610                         GOTO(out, rc);
 611                 }
 612
 613                 ptlrpc_req_finished(request);
 614
 615                 EXIT;
 616         out:
 617                 obd_ioctl_freedata(buf, len);
 618                 return rc;
 619         }
 620         case LL_IOC_KEY_TYPE: {
 621                 char *buf = NULL;
 622                 char *type;
 623                 int typelen, rc, len = 0;
 624
 625                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 626                 if (rc)
 627                         RETURN(rc);
 628                 data = (void *)buf;
 629
 630                 type = data->ioc_inlbuf1;
 631                 typelen = data->ioc_inllen1;
 632
 633                 if (typelen < 1) {
 634                         CDEBUG(D_INFO, "LL_IOC_KEY_TYPE missing filename\n");
 635                         GOTO(out_free, rc = -EINVAL);
 636                 }
 637                 ll_set_sb_gksinfo(inode->i_sb, type);
 638                 EXIT;
 639         out_free:
 640                 obd_ioctl_freedata(buf, len);
 641                 RETURN(rc);
 642         }
 643         case LL_IOC_MDC_MKDIRSTRIPE:
 644                 RETURN(ll_mkdir_stripe(inode, arg));
 645         case LL_IOC_LOV_SETSTRIPE: {
 646                 struct ptlrpc_request *request = NULL;
 647                 struct mdc_op_data *op_data;
 648                 struct iattr attr = { 0 };
 649                 struct lov_user_md lum, *lump = (struct lov_user_md *)arg;
 650                 int rc = 0;
 651
 652                 LASSERT(sizeof(lum) == sizeof(*lump));
 653                 LASSERT(sizeof(lum.lmm_objects[0]) ==
 654                         sizeof(lump->lmm_objects[0]));
 655                 rc = copy_from_user(&lum, lump, sizeof(lum));
 656                 if (rc)
 657                         RETURN(-EFAULT);
 658
 659                 if (lum.lmm_magic != LOV_USER_MAGIC)
 660                         RETURN(-EINVAL);
 661
 662                 OBD_ALLOC(op_data, sizeof(*op_data));
 663                 if (op_data == NULL)
 664                         RETURN(-ENOMEM);
 665
 666                 ll_inode2mdc_data(op_data, inode, (OBD_MD_FLID | OBD_MD_MEA));
 667
 668                 rc = md_setattr(sbi->ll_md_exp, op_data, &attr, &lum,
 669                                 sizeof(lum), NULL, 0, NULL, 0, &request);
 670                 OBD_FREE(op_data, sizeof(*op_data));
 671                 ptlrpc_req_finished(request);
 672
 673                 if (rc) {
 674                         if (rc != -EPERM && rc != -EACCES)
 675                                 CERROR("md_setattr fails: rc = %d\n", rc);
 676                 }
 677                 RETURN(rc);
 678         }
 679         case LL_IOC_LOV_GETSTRIPE: {
 680                 struct ptlrpc_request *request = NULL;
 681                 struct lov_user_md *lump = (struct lov_user_md *)arg;
 682                 struct lov_mds_md *lmm;
 683                 struct lustre_id id;
 684                 struct mds_body *body;
 685                 __u64 valid = 0;
 686                 int rc, lmmsize;
 687
 688                 valid |= OBD_MD_FLDIREA;
 689
 690                 ll_inode2id(&id, inode);
 691                 rc = md_getattr(sbi->ll_md_exp, &id, valid, NULL, NULL, 0,
 692                                 obd_size_diskmd(sbi->ll_dt_exp, NULL),
 693                                 NULL, &request);
 694                 if (rc < 0) {
 695                         CDEBUG(D_INFO, "md_getattr failed: rc = %d\n", rc);
 696                         RETURN(rc);
 697                 }
 698
 699                 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof(*body));
 700                 LASSERT(body != NULL);         /* checked by md_getattr_lock */
 701                 LASSERT_REPSWABBED(request, 0);/* swabbed by md_getattr_lock */
 702
 703                 lmmsize = body->eadatasize;
 704                 if (lmmsize == 0)
 705                         GOTO(out_get, rc = -ENODATA);
 706
 707                 lmm = lustre_msg_buf(request->rq_repmsg, 1, lmmsize);
 708                 LASSERT(lmm != NULL);
 709                 LASSERT_REPSWABBED(request, 1);
 710                 rc = copy_to_user(lump, lmm, lmmsize);
 711                 if (rc)
 712                         GOTO(out_get, rc = -EFAULT);
 713
 714                 EXIT;
 715         out_get:
 716                 ptlrpc_req_finished(request);
 717                 RETURN(rc);
 718         }
 719         case IOC_MDC_SHOWFID: {
 720                 struct lustre_id *idp = (struct lustre_id *)arg;
 721                 struct lustre_id id;
 722                 char *filename;
 723                 int rc;
 724
 725                 filename = getname((const char *)arg);
 726                 if (IS_ERR(filename))
 727                         RETURN(PTR_ERR(filename));
 728
 729                 ll_inode2id(&id, inode);
 730
 731                 rc = ll_get_fid(sbi->ll_md_exp, &id, filename, &id);
 732                 if (rc < 0)
 733                         GOTO(out_filename, rc);
 734
 735                 rc = copy_to_user(idp, &id, sizeof(*idp));
 736                 if (rc)
 737                         GOTO(out_filename, rc = -EFAULT);
 738
 739                 EXIT;
 740         out_filename:
 741                 putname(filename);
 742                 return rc;
 743         }
 744         case IOC_MDC_GETSTRIPE: {
 745                 struct lov_user_md *lump = (struct lov_user_md *)arg;
 746                 struct ptlrpc_request *request = NULL;
 747                 struct lov_mds_md *lmm;
 748                 struct mds_body *body;
 749                 struct lustre_id id;
 750                 char *filename;
 751                 int rc, lmmsize;
 752
 753                 filename = getname((const char *)arg);
 754                 if (IS_ERR(filename))
 755                         RETURN(PTR_ERR(filename));
 756
 757                 ll_inode2id(&id, inode);
 758                 rc = md_getattr_lock(sbi->ll_md_exp, &id, filename,
 759                                      strlen(filename) + 1, OBD_MD_FLEASIZE,
 760                                      obd_size_diskmd(sbi->ll_dt_exp, NULL),
 761                                      &request);
 762                 if (rc < 0) {
 763                         CDEBUG(D_INFO, "md_getattr_lock failed on %s: rc %d\n",
 764                                filename, rc);
 765                         GOTO(out_name, rc);
 766                 }
 767
 768                 body = lustre_msg_buf(request->rq_repmsg, 0, sizeof (*body));
 769                 LASSERT(body != NULL);         /* checked by md_getattr_lock */
 770                 LASSERT_REPSWABBED(request, 0);/* swabbed by md_getattr_lock */
 771
 772                 lmmsize = body->eadatasize;
 773
 774                 if (!(body->valid & OBD_MD_FLEASIZE) || lmmsize == 0)
 775                         GOTO(out_req, rc = -ENODATA);
 776
 777                 if (lmmsize > 4096)
 778                         GOTO(out_req, rc = -EFBIG);
 779
 780                 lmm = lustre_msg_buf(request->rq_repmsg, 1, lmmsize);
 781                 LASSERT(lmm != NULL);
 782                 LASSERT_REPSWABBED(request, 1);
 783
 784                 rc = copy_to_user(lump, lmm, lmmsize);
 785                 if (rc)
 786                         GOTO(out_req, rc = -EFAULT);
 787
 788                 EXIT;
 789         out_req:
 790                 ptlrpc_req_finished(request);
 791         out_name:
 792                 putname(filename);
 793                 return rc;
 794         }
 795         case OBD_IOC_PING: {
 796                 struct ptlrpc_request *req = NULL;
 797                 char *buf = NULL;
 798                 int rc, len=0;
 799                 struct client_obd *cli;
 800                 struct obd_device *obd;
 801
 802                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 803                 if (rc)
 804                         RETURN(rc);
 805                 data = (void *)buf;
 806
 807                 obd = class_name2obd(data->ioc_inlbuf1);
 808
 809                 if (!obd )
 810                         GOTO(out_ping, rc = -ENODEV);
 811
 812                 if (!obd->obd_attached) {
 813                         CERROR("Device %d not attached\n", obd->obd_minor);
 814                         GOTO(out_ping, rc = -ENODEV);
 815                 }
 816                 if (!obd->obd_set_up) {
 817                         CERROR("Device %d still not setup\n", obd->obd_minor);
 818                         GOTO(out_ping, rc = -ENODEV);
 819                 }
 820                 cli = &obd->u.cli;
 821                 req = ptlrpc_prep_req(cli->cl_import, LUSTRE_OBD_VERSION,
 822                                       OBD_PING, 0, NULL, NULL);
 823                 if (!req)
 824                         GOTO(out_ping, rc = -ENOMEM);
 825
 826                 req->rq_replen = lustre_msg_size(0, NULL);
 827                 req->rq_send_state = LUSTRE_IMP_FULL;
 828
 829                 rc = ptlrpc_queue_wait(req);
 830
 831                 ptlrpc_req_finished(req);
 832         out_ping:
 833                 obd_ioctl_freedata(buf, len);
 834                 return rc;
 835         }
 836         case OBD_IOC_LLOG_CATINFO: {
 837                 struct ptlrpc_request *req = NULL;
 838                 char *buf = NULL;
 839                 int rc, len = 0;
 840                 char *bufs[2], *str;
 841                 int lens[2], size;
 842
 843                 rc = obd_ioctl_getdata(&buf, &len, (void *)arg);
 844                 if (rc)
 845                         RETURN(rc);
 846                 data = (void *)buf;
 847
 848                 if (!data->ioc_inlbuf1) {
 849                         obd_ioctl_freedata(buf, len);
 850                         RETURN(-EINVAL);
 851                 }
 852
 853                 lens[0] = data->ioc_inllen1;
 854                 bufs[0] = data->ioc_inlbuf1;
 855                 if (data->ioc_inllen2) {
 856                         lens[1] = data->ioc_inllen2;
 857                         bufs[1] = data->ioc_inlbuf2;
 858                 } else {
 859                         lens[1] = 0;
 860                         bufs[1] = NULL;
 861                 }
 862                 size = data->ioc_plen1;
 863                 req = ptlrpc_prep_req(sbi2md(sbi)->cl_import,
 864                                       LUSTRE_LOG_VERSION, LLOG_CATINFO,
 865                                       2, lens, bufs);
 866                 if (!req)
 867                         GOTO(out_catinfo, rc = -ENOMEM);
 868
 869                 req->rq_replen = lustre_msg_size(1, &size);
 870
 871                 rc = ptlrpc_queue_wait(req);
 872                 str = lustre_msg_string(req->rq_repmsg, 0, data->ioc_plen1);
 873                 if (!rc)
 874                         rc = copy_to_user(data->ioc_pbuf1, str,
 875                                           data->ioc_plen1);
 876                 ptlrpc_req_finished(req);
 877         out_catinfo:
 878                 obd_ioctl_freedata(buf, len);
 879                 RETURN(rc);
 880         }
 881         case LL_IOC_GETFACL: {
 882                 struct ll_acl_ioctl_data ioc, *uioc;
 883                 int rc;
 884
 885                 if (copy_from_user(&ioc, (void *) arg, sizeof(ioc)))
 886                         RETURN(-EFAULT);
 887
 888                 rc = ll_ioctl_getfacl(inode, file, &ioc);
 889                 if (!rc) {
 890                         uioc = (struct ll_acl_ioctl_data *) arg;
 891                         if (copy_to_user(&uioc->status, &ioc.status,
 892                                          sizeof(ioc.status)))
 893                                 rc = -EFAULT;
 894                 }
 895                 RETURN(rc);
 896         }
 897         case LL_IOC_SETFACL: {
 898                 struct ll_acl_ioctl_data ioc, *uioc;
 899                 int rc;
 900
 901                 if (copy_from_user(&ioc, (void *) arg, sizeof(ioc)))
 902                         RETURN(-EFAULT);
 903
 904                 rc = ll_ioctl_setfacl(inode, file, &ioc);
 905                 if (!rc) {
 906                         uioc = (struct ll_acl_ioctl_data *) arg;
 907                         if (copy_to_user(&uioc->status, &ioc.status,
 908                                          sizeof(ioc.status)))
 909                                 rc = -EFAULT;
 910                 }
 911                 RETURN(rc);
 912         }
 913         case LL_IOC_FLUSH_CRED:
 914                 RETURN(ll_flush_cred(inode));
 915         case LL_IOC_AUDIT:
 916                 RETURN(ll_set_audit(inode, SET_AUDIT_OP(arg, AUDIT_DIR)));
 917         default:
 918                 return obd_iocontrol(cmd, sbi->ll_dt_exp, 0,
 919                                      NULL, (void *)arg);
 920         }
 921 }
 922
 923 int ll_dir_open(struct inode *inode, struct file *file)
 924 {
 925         return ll_file_open(inode, file);
 926 }
 927
 928 int ll_dir_release(struct inode *inode, struct file *file)
 929 {
 930         return ll_file_release(inode, file);
 931 }
 932
 933 struct file_operations ll_dir_operations = {
 934         .open     = ll_dir_open,
 935         .release  = ll_dir_release,
 936         .read     = generic_read_dir,
 937         .readdir  = ll_readdir,
 938         .ioctl    = ll_dir_ioctl
 939 };
 940