fs/xfs/xfs_attr.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18
  19 #include <linux/capability.h>
  20
  21 #include "xfs.h"
  22 #include "xfs_fs.h"
  23 #include "xfs_types.h"
  24 #include "xfs_bit.h"
  25 #include "xfs_log.h"
  26 #include "xfs_inum.h"
  27 #include "xfs_trans.h"
  28 #include "xfs_sb.h"
  29 #include "xfs_ag.h"
  30 #include "xfs_dir2.h"
  31 #include "xfs_dmapi.h"
  32 #include "xfs_mount.h"
  33 #include "xfs_da_btree.h"
  34 #include "xfs_bmap_btree.h"
  35 #include "xfs_alloc_btree.h"
  36 #include "xfs_ialloc_btree.h"
  37 #include "xfs_dir2_sf.h"
  38 #include "xfs_attr_sf.h"
  39 #include "xfs_dinode.h"
  40 #include "xfs_inode.h"
  41 #include "xfs_alloc.h"
  42 #include "xfs_btree.h"
  43 #include "xfs_inode_item.h"
  44 #include "xfs_bmap.h"
  45 #include "xfs_attr.h"
  46 #include "xfs_attr_leaf.h"
  47 #include "xfs_error.h"
  48 #include "xfs_quota.h"
  49 #include "xfs_trans_space.h"
  50 #include "xfs_acl.h"
  51 #include "xfs_rw.h"
  52
  53 /*
  54  * xfs_attr.c
  55  *
  56  * Provide the external interfaces to manage attribute lists.
  57  */
  58
  59 #define ATTR_SYSCOUNT   2
  60 STATIC struct attrnames posix_acl_access;
  61 STATIC struct attrnames posix_acl_default;
  62 STATIC struct attrnames *attr_system_names[ATTR_SYSCOUNT];
  63
  64 /*========================================================================
  65  * Function prototypes for the kernel.
  66  *========================================================================*/
  67
  68 /*
  69  * Internal routines when attribute list fits inside the inode.
  70  */
  71 STATIC int xfs_attr_shortform_addname(xfs_da_args_t *args);
  72
  73 /*
  74  * Internal routines when attribute list is one block.
  75  */
  76 STATIC int xfs_attr_leaf_get(xfs_da_args_t *args);
  77 STATIC int xfs_attr_leaf_addname(xfs_da_args_t *args);
  78 STATIC int xfs_attr_leaf_removename(xfs_da_args_t *args);
  79 STATIC int xfs_attr_leaf_list(xfs_attr_list_context_t *context);
  80
  81 /*
  82  * Internal routines when attribute list is more than one block.
  83  */
  84 STATIC int xfs_attr_node_get(xfs_da_args_t *args);
  85 STATIC int xfs_attr_node_addname(xfs_da_args_t *args);
  86 STATIC int xfs_attr_node_removename(xfs_da_args_t *args);
  87 STATIC int xfs_attr_node_list(xfs_attr_list_context_t *context);
  88 STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
  89 STATIC int xfs_attr_refillstate(xfs_da_state_t *state);
  90
  91 /*
  92  * Routines to manipulate out-of-line attribute values.
  93  */
  94 STATIC int xfs_attr_rmtval_get(xfs_da_args_t *args);
  95 STATIC int xfs_attr_rmtval_set(xfs_da_args_t *args);
  96 STATIC int xfs_attr_rmtval_remove(xfs_da_args_t *args);
  97
  98 #define ATTR_RMTVALUE_MAPSIZE   1       /* # of map entries at once */
  99
 100 #if defined(XFS_ATTR_TRACE)
 101 ktrace_t *xfs_attr_trace_buf;
 102 #endif
 103
 104
 105 /*========================================================================
 106  * Overall external interface routines.
 107  *========================================================================*/
 108
 109 int
 110 xfs_attr_fetch(xfs_inode_t *ip, const char *name, int namelen,
 111                char *value, int *valuelenp, int flags, struct cred *cred)
 112 {
 113         xfs_da_args_t   args;
 114         int             error;
 115
 116         if ((XFS_IFORK_Q(ip) == 0) ||
 117             (ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 118              ip->i_d.di_anextents == 0))
 119                 return(ENOATTR);
 120
 121         /*
 122          * Fill in the arg structure for this request.
 123          */
 124         memset((char *)&args, 0, sizeof(args));
 125         args.name = name;
 126         args.namelen = namelen;
 127         args.value = value;
 128         args.valuelen = *valuelenp;
 129         args.flags = flags;
 130         args.hashval = xfs_da_hashname(args.name, args.namelen);
 131         args.dp = ip;
 132         args.whichfork = XFS_ATTR_FORK;
 133
 134         /*
 135          * Decide on what work routines to call based on the inode size.
 136          */
 137         if (XFS_IFORK_Q(ip) == 0 ||
 138             (ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 139              ip->i_d.di_anextents == 0)) {
 140                 error = XFS_ERROR(ENOATTR);
 141         } else if (ip->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
 142                 error = xfs_attr_shortform_getvalue(&args);
 143         } else if (xfs_bmap_one_block(ip, XFS_ATTR_FORK)) {
 144                 error = xfs_attr_leaf_get(&args);
 145         } else {
 146                 error = xfs_attr_node_get(&args);
 147         }
 148
 149         /*
 150          * Return the number of bytes in the value to the caller.
 151          */
 152         *valuelenp = args.valuelen;
 153
 154         if (error == EEXIST)
 155                 error = 0;
 156         return(error);
 157 }
 158
 159 int
 160 xfs_attr_get(bhv_desc_t *bdp, const char *name, char *value, int *valuelenp,
 161              int flags, struct cred *cred)
 162 {
 163         xfs_inode_t     *ip = XFS_BHVTOI(bdp);
 164         int             error, namelen;
 165
 166         XFS_STATS_INC(xs_attr_get);
 167
 168         if (!name)
 169                 return(EINVAL);
 170         namelen = strlen(name);
 171         if (namelen >= MAXNAMELEN)
 172                 return(EFAULT);         /* match IRIX behaviour */
 173
 174         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 175                 return(EIO);
 176
 177         xfs_ilock(ip, XFS_ILOCK_SHARED);
 178         error = xfs_attr_fetch(ip, name, namelen, value, valuelenp, flags, cred);
 179         xfs_iunlock(ip, XFS_ILOCK_SHARED);
 180         return(error);
 181 }
 182
 183 STATIC int
 184 xfs_attr_set_int(xfs_inode_t *dp, const char *name, int namelen,
 185                  char *value, int valuelen, int flags)
 186 {
 187         xfs_da_args_t   args;
 188         xfs_fsblock_t   firstblock;
 189         xfs_bmap_free_t flist;
 190         int             error, err2, committed;
 191         int             local, size;
 192         uint            nblks;
 193         xfs_mount_t     *mp = dp->i_mount;
 194         int             rsvd = (flags & ATTR_ROOT) != 0;
 195
 196         /*
 197          * Attach the dquots to the inode.
 198          */
 199         if ((error = XFS_QM_DQATTACH(mp, dp, 0)))
 200                 return (error);
 201
 202         /*
 203          * Determine space new attribute will use, and if it would be
 204          * "local" or "remote" (note: local != inline).
 205          */
 206         size = xfs_attr_leaf_newentsize(namelen, valuelen,
 207                                         mp->m_sb.sb_blocksize, &local);
 208
 209         /*
 210          * If the inode doesn't have an attribute fork, add one.
 211          * (inode must not be locked when we call this routine)
 212          */
 213         if (XFS_IFORK_Q(dp) == 0) {
 214                 if ((error = xfs_bmap_add_attrfork(dp, size, rsvd)))
 215                         return(error);
 216         }
 217
 218         /*
 219          * Fill in the arg structure for this request.
 220          */
 221         memset((char *)&args, 0, sizeof(args));
 222         args.name = name;
 223         args.namelen = namelen;
 224         args.value = value;
 225         args.valuelen = valuelen;
 226         args.flags = flags;
 227         args.hashval = xfs_da_hashname(args.name, args.namelen);
 228         args.dp = dp;
 229         args.firstblock = &firstblock;
 230         args.flist = &flist;
 231         args.whichfork = XFS_ATTR_FORK;
 232         args.addname = 1;
 233         args.oknoent = 1;
 234
 235         nblks = XFS_DAENTER_SPACE_RES(mp, XFS_ATTR_FORK);
 236         if (local) {
 237                 if (size > (mp->m_sb.sb_blocksize >> 1)) {
 238                         /* Double split possible */
 239                         nblks <<= 1;
 240                 }
 241         } else {
 242                 uint    dblocks = XFS_B_TO_FSB(mp, valuelen);
 243                 /* Out of line attribute, cannot double split, but make
 244                  * room for the attribute value itself.
 245                  */
 246                 nblks += dblocks;
 247                 nblks += XFS_NEXTENTADD_SPACE_RES(mp, dblocks, XFS_ATTR_FORK);
 248         }
 249
 250         /* Size is now blocks for attribute data */
 251         args.total = nblks;
 252
 253         /*
 254          * Start our first transaction of the day.
 255          *
 256          * All future transactions during this code must be "chained" off
 257          * this one via the trans_dup() call.  All transactions will contain
 258          * the inode, and the inode will always be marked with trans_ihold().
 259          * Since the inode will be locked in all transactions, we must log
 260          * the inode in every transaction to let it float upward through
 261          * the log.
 262          */
 263         args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_SET);
 264
 265         /*
 266          * Root fork attributes can use reserved data blocks for this
 267          * operation if necessary
 268          */
 269
 270         if (rsvd)
 271                 args.trans->t_flags |= XFS_TRANS_RESERVE;
 272
 273         if ((error = xfs_trans_reserve(args.trans, (uint) nblks,
 274                                       XFS_ATTRSET_LOG_RES(mp, nblks),
 275                                       0, XFS_TRANS_PERM_LOG_RES,
 276                                       XFS_ATTRSET_LOG_COUNT))) {
 277                 xfs_trans_cancel(args.trans, 0);
 278                 return(error);
 279         }
 280         xfs_ilock(dp, XFS_ILOCK_EXCL);
 281
 282         error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, args.trans, dp, nblks, 0,
 283                          rsvd ? XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
 284                                 XFS_QMOPT_RES_REGBLKS);
 285         if (error) {
 286                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
 287                 xfs_trans_cancel(args.trans, XFS_TRANS_RELEASE_LOG_RES);
 288                 return (error);
 289         }
 290
 291         xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
 292         xfs_trans_ihold(args.trans, dp);
 293
 294         /*
 295          * If the attribute list is non-existent or a shortform list,
 296          * upgrade it to a single-leaf-block attribute list.
 297          */
 298         if ((dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
 299             ((dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) &&
 300              (dp->i_d.di_anextents == 0))) {
 301
 302                 /*
 303                  * Build initial attribute list (if required).
 304                  */
 305                 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS)
 306                         xfs_attr_shortform_create(&args);
 307
 308                 /*
 309                  * Try to add the attr to the attribute list in
 310                  * the inode.
 311                  */
 312                 error = xfs_attr_shortform_addname(&args);
 313                 if (error != ENOSPC) {
 314                         /*
 315                          * Commit the shortform mods, and we're done.
 316                          * NOTE: this is also the error path (EEXIST, etc).
 317                          */
 318                         ASSERT(args.trans != NULL);
 319
 320                         /*
 321                          * If this is a synchronous mount, make sure that
 322                          * the transaction goes to disk before returning
 323                          * to the user.
 324                          */
 325                         if (mp->m_flags & XFS_MOUNT_WSYNC) {
 326                                 xfs_trans_set_sync(args.trans);
 327                         }
 328                         err2 = xfs_trans_commit(args.trans,
 329                                                  XFS_TRANS_RELEASE_LOG_RES,
 330                                                  NULL);
 331                         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 332
 333                         /*
 334                          * Hit the inode change time.
 335                          */
 336                         if (!error && (flags & ATTR_KERNOTIME) == 0) {
 337                                 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
 338                         }
 339                         return(error == 0 ? err2 : error);
 340                 }
 341
 342                 /*
 343                  * It won't fit in the shortform, transform to a leaf block.
 344                  * GROT: another possible req'mt for a double-split btree op.
 345                  */
 346                 XFS_BMAP_INIT(args.flist, args.firstblock);
 347                 error = xfs_attr_shortform_to_leaf(&args);
 348                 if (!error) {
 349                         error = xfs_bmap_finish(&args.trans, args.flist,
 350                                                 *args.firstblock, &committed);
 351                 }
 352                 if (error) {
 353                         ASSERT(committed);
 354                         args.trans = NULL;
 355                         xfs_bmap_cancel(&flist);
 356                         goto out;
 357                 }
 358
 359                 /*
 360                  * bmap_finish() may have committed the last trans and started
 361                  * a new one.  We need the inode to be in all transactions.
 362                  */
 363                 if (committed) {
 364                         xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
 365                         xfs_trans_ihold(args.trans, dp);
 366                 }
 367
 368                 /*
 369                  * Commit the leaf transformation.  We'll need another (linked)
 370                  * transaction to add the new attribute to the leaf.
 371                  */
 372                 if ((error = xfs_attr_rolltrans(&args.trans, dp)))
 373                         goto out;
 374
 375         }
 376
 377         if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
 378                 error = xfs_attr_leaf_addname(&args);
 379         } else {
 380                 error = xfs_attr_node_addname(&args);
 381         }
 382         if (error) {
 383                 goto out;
 384         }
 385
 386         /*
 387          * If this is a synchronous mount, make sure that the
 388          * transaction goes to disk before returning to the user.
 389          */
 390         if (mp->m_flags & XFS_MOUNT_WSYNC) {
 391                 xfs_trans_set_sync(args.trans);
 392         }
 393
 394         /*
 395          * Commit the last in the sequence of transactions.
 396          */
 397         xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
 398         error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES,
 399                                  NULL);
 400         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 401
 402         /*
 403          * Hit the inode change time.
 404          */
 405         if (!error && (flags & ATTR_KERNOTIME) == 0) {
 406                 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
 407         }
 408
 409         return(error);
 410
 411 out:
 412         if (args.trans)
 413                 xfs_trans_cancel(args.trans,
 414                         XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
 415         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 416         return(error);
 417 }
 418
 419 int
 420 xfs_attr_set(bhv_desc_t *bdp, const char *name, char *value, int valuelen, int flags,
 421              struct cred *cred)
 422 {
 423         xfs_inode_t     *dp;
 424         int             namelen;
 425
 426         namelen = strlen(name);
 427         if (namelen >= MAXNAMELEN)
 428                 return EFAULT;          /* match IRIX behaviour */
 429
 430         XFS_STATS_INC(xs_attr_set);
 431
 432         dp = XFS_BHVTOI(bdp);
 433         if (XFS_FORCED_SHUTDOWN(dp->i_mount))
 434                 return (EIO);
 435
 436         return xfs_attr_set_int(dp, name, namelen, value, valuelen, flags);
 437 }
 438
 439 /*
 440  * Generic handler routine to remove a name from an attribute list.
 441  * Transitions attribute list from Btree to shortform as necessary.
 442  */
 443 STATIC int
 444 xfs_attr_remove_int(xfs_inode_t *dp, const char *name, int namelen, int flags)
 445 {
 446         xfs_da_args_t   args;
 447         xfs_fsblock_t   firstblock;
 448         xfs_bmap_free_t flist;
 449         int             error;
 450         xfs_mount_t     *mp = dp->i_mount;
 451
 452         /*
 453          * Fill in the arg structure for this request.
 454          */
 455         memset((char *)&args, 0, sizeof(args));
 456         args.name = name;
 457         args.namelen = namelen;
 458         args.flags = flags;
 459         args.hashval = xfs_da_hashname(args.name, args.namelen);
 460         args.dp = dp;
 461         args.firstblock = &firstblock;
 462         args.flist = &flist;
 463         args.total = 0;
 464         args.whichfork = XFS_ATTR_FORK;
 465
 466         /*
 467          * Attach the dquots to the inode.
 468          */
 469         if ((error = XFS_QM_DQATTACH(mp, dp, 0)))
 470                 return (error);
 471
 472         /*
 473          * Start our first transaction of the day.
 474          *
 475          * All future transactions during this code must be "chained" off
 476          * this one via the trans_dup() call.  All transactions will contain
 477          * the inode, and the inode will always be marked with trans_ihold().
 478          * Since the inode will be locked in all transactions, we must log
 479          * the inode in every transaction to let it float upward through
 480          * the log.
 481          */
 482         args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_RM);
 483
 484         /*
 485          * Root fork attributes can use reserved data blocks for this
 486          * operation if necessary
 487          */
 488
 489         if (flags & ATTR_ROOT)
 490                 args.trans->t_flags |= XFS_TRANS_RESERVE;
 491
 492         if ((error = xfs_trans_reserve(args.trans,
 493                                       XFS_ATTRRM_SPACE_RES(mp),
 494                                       XFS_ATTRRM_LOG_RES(mp),
 495                                       0, XFS_TRANS_PERM_LOG_RES,
 496                                       XFS_ATTRRM_LOG_COUNT))) {
 497                 xfs_trans_cancel(args.trans, 0);
 498                 return(error);
 499         }
 500
 501         xfs_ilock(dp, XFS_ILOCK_EXCL);
 502         /*
 503          * No need to make quota reservations here. We expect to release some
 504          * blocks not allocate in the common case.
 505          */
 506         xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
 507         xfs_trans_ihold(args.trans, dp);
 508
 509         /*
 510          * Decide on what work routines to call based on the inode size.
 511          */
 512         if (XFS_IFORK_Q(dp) == 0 ||
 513             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 514              dp->i_d.di_anextents == 0)) {
 515                 error = XFS_ERROR(ENOATTR);
 516                 goto out;
 517         }
 518         if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
 519                 ASSERT(dp->i_afp->if_flags & XFS_IFINLINE);
 520                 error = xfs_attr_shortform_remove(&args);
 521                 if (error) {
 522                         goto out;
 523                 }
 524         } else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
 525                 error = xfs_attr_leaf_removename(&args);
 526         } else {
 527                 error = xfs_attr_node_removename(&args);
 528         }
 529         if (error) {
 530                 goto out;
 531         }
 532
 533         /*
 534          * If this is a synchronous mount, make sure that the
 535          * transaction goes to disk before returning to the user.
 536          */
 537         if (mp->m_flags & XFS_MOUNT_WSYNC) {
 538                 xfs_trans_set_sync(args.trans);
 539         }
 540
 541         /*
 542          * Commit the last in the sequence of transactions.
 543          */
 544         xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
 545         error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES,
 546                                  NULL);
 547         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 548
 549         /*
 550          * Hit the inode change time.
 551          */
 552         if (!error && (flags & ATTR_KERNOTIME) == 0) {
 553                 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
 554         }
 555
 556         return(error);
 557
 558 out:
 559         if (args.trans)
 560                 xfs_trans_cancel(args.trans,
 561                         XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
 562         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 563         return(error);
 564 }
 565
 566 int
 567 xfs_attr_remove(bhv_desc_t *bdp, const char *name, int flags, struct cred *cred)
 568 {
 569         xfs_inode_t         *dp;
 570         int                 namelen;
 571
 572         namelen = strlen(name);
 573         if (namelen >= MAXNAMELEN)
 574                 return EFAULT;          /* match IRIX behaviour */
 575
 576         XFS_STATS_INC(xs_attr_remove);
 577
 578         dp = XFS_BHVTOI(bdp);
 579         if (XFS_FORCED_SHUTDOWN(dp->i_mount))
 580                 return (EIO);
 581
 582         xfs_ilock(dp, XFS_ILOCK_SHARED);
 583         if (XFS_IFORK_Q(dp) == 0 ||
 584                    (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 585                     dp->i_d.di_anextents == 0)) {
 586                 xfs_iunlock(dp, XFS_ILOCK_SHARED);
 587                 return(XFS_ERROR(ENOATTR));
 588         }
 589         xfs_iunlock(dp, XFS_ILOCK_SHARED);
 590
 591         return xfs_attr_remove_int(dp, name, namelen, flags);
 592 }
 593
 594 /*
 595  * Generate a list of extended attribute names and optionally
 596  * also value lengths.  Positive return value follows the XFS
 597  * convention of being an error, zero or negative return code
 598  * is the length of the buffer returned (negated), indicating
 599  * success.
 600  */
 601 int
 602 xfs_attr_list(bhv_desc_t *bdp, char *buffer, int bufsize, int flags,
 603                       attrlist_cursor_kern_t *cursor, struct cred *cred)
 604 {
 605         xfs_attr_list_context_t context;
 606         xfs_inode_t *dp;
 607         int error;
 608
 609         XFS_STATS_INC(xs_attr_list);
 610
 611         /*
 612          * Validate the cursor.
 613          */
 614         if (cursor->pad1 || cursor->pad2)
 615                 return(XFS_ERROR(EINVAL));
 616         if ((cursor->initted == 0) &&
 617             (cursor->hashval || cursor->blkno || cursor->offset))
 618                 return(XFS_ERROR(EINVAL));
 619
 620         /*
 621          * Check for a properly aligned buffer.
 622          */
 623         if (((long)buffer) & (sizeof(int)-1))
 624                 return(XFS_ERROR(EFAULT));
 625         if (flags & ATTR_KERNOVAL)
 626                 bufsize = 0;
 627
 628         /*
 629          * Initialize the output buffer.
 630          */
 631         context.dp = dp = XFS_BHVTOI(bdp);
 632         context.cursor = cursor;
 633         context.count = 0;
 634         context.dupcnt = 0;
 635         context.resynch = 1;
 636         context.flags = flags;
 637         if (!(flags & ATTR_KERNAMELS)) {
 638                 context.bufsize = (bufsize & ~(sizeof(int)-1));  /* align */
 639                 context.firstu = context.bufsize;
 640                 context.alist = (attrlist_t *)buffer;
 641                 context.alist->al_count = 0;
 642                 context.alist->al_more = 0;
 643                 context.alist->al_offset[0] = context.bufsize;
 644         }
 645         else {
 646                 context.bufsize = bufsize;
 647                 context.firstu = context.bufsize;
 648                 context.alist = (attrlist_t *)buffer;
 649         }
 650
 651         if (XFS_FORCED_SHUTDOWN(dp->i_mount))
 652                 return (EIO);
 653
 654         xfs_ilock(dp, XFS_ILOCK_SHARED);
 655         /*
 656          * Decide on what work routines to call based on the inode size.
 657          */
 658         xfs_attr_trace_l_c("syscall start", &context);
 659         if (XFS_IFORK_Q(dp) == 0 ||
 660             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 661              dp->i_d.di_anextents == 0)) {
 662                 error = 0;
 663         } else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
 664                 error = xfs_attr_shortform_list(&context);
 665         } else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
 666                 error = xfs_attr_leaf_list(&context);
 667         } else {
 668                 error = xfs_attr_node_list(&context);
 669         }
 670         xfs_iunlock(dp, XFS_ILOCK_SHARED);
 671         xfs_attr_trace_l_c("syscall end", &context);
 672
 673         if (!(context.flags & (ATTR_KERNOVAL|ATTR_KERNAMELS))) {
 674                 ASSERT(error >= 0);
 675         }
 676         else {  /* must return negated buffer size or the error */
 677                 if (context.count < 0)
 678                         error = XFS_ERROR(ERANGE);
 679                 else
 680                         error = -context.count;
 681         }
 682
 683         return(error);
 684 }
 685
 686 int                                                             /* error */
 687 xfs_attr_inactive(xfs_inode_t *dp)
 688 {
 689         xfs_trans_t *trans;
 690         xfs_mount_t *mp;
 691         int error;
 692
 693         mp = dp->i_mount;
 694         ASSERT(! XFS_NOT_DQATTACHED(mp, dp));
 695
 696         xfs_ilock(dp, XFS_ILOCK_SHARED);
 697         if ((XFS_IFORK_Q(dp) == 0) ||
 698             (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
 699             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 700              dp->i_d.di_anextents == 0)) {
 701                 xfs_iunlock(dp, XFS_ILOCK_SHARED);
 702                 return(0);
 703         }
 704         xfs_iunlock(dp, XFS_ILOCK_SHARED);
 705
 706         /*
 707          * Start our first transaction of the day.
 708          *
 709          * All future transactions during this code must be "chained" off
 710          * this one via the trans_dup() call.  All transactions will contain
 711          * the inode, and the inode will always be marked with trans_ihold().
 712          * Since the inode will be locked in all transactions, we must log
 713          * the inode in every transaction to let it float upward through
 714          * the log.
 715          */
 716         trans = xfs_trans_alloc(mp, XFS_TRANS_ATTRINVAL);
 717         if ((error = xfs_trans_reserve(trans, 0, XFS_ATTRINVAL_LOG_RES(mp), 0,
 718                                       XFS_TRANS_PERM_LOG_RES,
 719                                       XFS_ATTRINVAL_LOG_COUNT))) {
 720                 xfs_trans_cancel(trans, 0);
 721                 return(error);
 722         }
 723         xfs_ilock(dp, XFS_ILOCK_EXCL);
 724
 725         /*
 726          * No need to make quota reservations here. We expect to release some
 727          * blocks, not allocate, in the common case.
 728          */
 729         xfs_trans_ijoin(trans, dp, XFS_ILOCK_EXCL);
 730         xfs_trans_ihold(trans, dp);
 731
 732         /*
 733          * Decide on what work routines to call based on the inode size.
 734          */
 735         if ((XFS_IFORK_Q(dp) == 0) ||
 736             (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
 737             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 738              dp->i_d.di_anextents == 0)) {
 739                 error = 0;
 740                 goto out;
 741         }
 742         error = xfs_attr_root_inactive(&trans, dp);
 743         if (error)
 744                 goto out;
 745         /*
 746          * signal synchronous inactive transactions unless this
 747          * is a synchronous mount filesystem in which case we
 748          * know that we're here because we've been called out of
 749          * xfs_inactive which means that the last reference is gone
 750          * and the unlink transaction has already hit the disk so
 751          * async inactive transactions are safe.
 752          */
 753         if ((error = xfs_itruncate_finish(&trans, dp, 0LL, XFS_ATTR_FORK,
 754                                 (!(mp->m_flags & XFS_MOUNT_WSYNC)
 755                                  ? 1 : 0))))
 756                 goto out;
 757
 758         /*
 759          * Commit the last in the sequence of transactions.
 760          */
 761         xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
 762         error = xfs_trans_commit(trans, XFS_TRANS_RELEASE_LOG_RES,
 763                                  NULL);
 764         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 765
 766         return(error);
 767
 768 out:
 769         xfs_trans_cancel(trans, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
 770         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 771         return(error);
 772 }
 773
 774
 775
 776 /*========================================================================
 777  * External routines when attribute list is inside the inode
 778  *========================================================================*/
 779
 780 /*
 781  * Add a name to the shortform attribute list structure
 782  * This is the external routine.
 783  */
 784 STATIC int
 785 xfs_attr_shortform_addname(xfs_da_args_t *args)
 786 {
 787         int newsize, forkoff, retval;
 788
 789         retval = xfs_attr_shortform_lookup(args);
 790         if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
 791                 return(retval);
 792         } else if (retval == EEXIST) {
 793                 if (args->flags & ATTR_CREATE)
 794                         return(retval);
 795                 retval = xfs_attr_shortform_remove(args);
 796                 ASSERT(retval == 0);
 797         }
 798
 799         if (args->namelen >= XFS_ATTR_SF_ENTSIZE_MAX ||
 800             args->valuelen >= XFS_ATTR_SF_ENTSIZE_MAX)
 801                 return(XFS_ERROR(ENOSPC));
 802
 803         newsize = XFS_ATTR_SF_TOTSIZE(args->dp);
 804         newsize += XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
 805
 806         forkoff = xfs_attr_shortform_bytesfit(args->dp, newsize);
 807         if (!forkoff)
 808                 return(XFS_ERROR(ENOSPC));
 809
 810         xfs_attr_shortform_add(args, forkoff);
 811         return(0);
 812 }
 813
 814
 815 /*========================================================================
 816  * External routines when attribute list is one block
 817  *========================================================================*/
 818
 819 /*
 820  * Add a name to the leaf attribute list structure
 821  *
 822  * This leaf block cannot have a "remote" value, we only call this routine
 823  * if bmap_one_block() says there is only one block (ie: no remote blks).
 824  */
 825 int
 826 xfs_attr_leaf_addname(xfs_da_args_t *args)
 827 {
 828         xfs_inode_t *dp;
 829         xfs_dabuf_t *bp;
 830         int retval, error, committed, forkoff;
 831
 832         /*
 833          * Read the (only) block in the attribute list in.
 834          */
 835         dp = args->dp;
 836         args->blkno = 0;
 837         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
 838                                              XFS_ATTR_FORK);
 839         if (error)
 840                 return(error);
 841         ASSERT(bp != NULL);
 842
 843         /*
 844          * Look up the given attribute in the leaf block.  Figure out if
 845          * the given flags produce an error or call for an atomic rename.
 846          */
 847         retval = xfs_attr_leaf_lookup_int(bp, args);
 848         if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
 849                 xfs_da_brelse(args->trans, bp);
 850                 return(retval);
 851         } else if (retval == EEXIST) {
 852                 if (args->flags & ATTR_CREATE) {        /* pure create op */
 853                         xfs_da_brelse(args->trans, bp);
 854                         return(retval);
 855                 }
 856                 args->rename = 1;                       /* an atomic rename */
 857                 args->blkno2 = args->blkno;             /* set 2nd entry info*/
 858                 args->index2 = args->index;
 859                 args->rmtblkno2 = args->rmtblkno;
 860                 args->rmtblkcnt2 = args->rmtblkcnt;
 861         }
 862
 863         /*
 864          * Add the attribute to the leaf block, transitioning to a Btree
 865          * if required.
 866          */
 867         retval = xfs_attr_leaf_add(bp, args);
 868         xfs_da_buf_done(bp);
 869         if (retval == ENOSPC) {
 870                 /*
 871                  * Promote the attribute list to the Btree format, then
 872                  * Commit that transaction so that the node_addname() call
 873                  * can manage its own transactions.
 874                  */
 875                 XFS_BMAP_INIT(args->flist, args->firstblock);
 876                 error = xfs_attr_leaf_to_node(args);
 877                 if (!error) {
 878                         error = xfs_bmap_finish(&args->trans, args->flist,
 879                                                 *args->firstblock, &committed);
 880                 }
 881                 if (error) {
 882                         ASSERT(committed);
 883                         args->trans = NULL;
 884                         xfs_bmap_cancel(args->flist);
 885                         return(error);
 886                 }
 887
 888                 /*
 889                  * bmap_finish() may have committed the last trans and started
 890                  * a new one.  We need the inode to be in all transactions.
 891                  */
 892                 if (committed) {
 893                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
 894                         xfs_trans_ihold(args->trans, dp);
 895                 }
 896
 897                 /*
 898                  * Commit the current trans (including the inode) and start
 899                  * a new one.
 900                  */
 901                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
 902                         return (error);
 903
 904                 /*
 905                  * Fob the whole rest of the problem off on the Btree code.
 906                  */
 907                 error = xfs_attr_node_addname(args);
 908                 return(error);
 909         }
 910
 911         /*
 912          * Commit the transaction that added the attr name so that
 913          * later routines can manage their own transactions.
 914          */
 915         if ((error = xfs_attr_rolltrans(&args->trans, dp)))
 916                 return (error);
 917
 918         /*
 919          * If there was an out-of-line value, allocate the blocks we
 920          * identified for its storage and copy the value.  This is done
 921          * after we create the attribute so that we don't overflow the
 922          * maximum size of a transaction and/or hit a deadlock.
 923          */
 924         if (args->rmtblkno > 0) {
 925                 error = xfs_attr_rmtval_set(args);
 926                 if (error)
 927                         return(error);
 928         }
 929
 930         /*
 931          * If this is an atomic rename operation, we must "flip" the
 932          * incomplete flags on the "new" and "old" attribute/value pairs
 933          * so that one disappears and one appears atomically.  Then we
 934          * must remove the "old" attribute/value pair.
 935          */
 936         if (args->rename) {
 937                 /*
 938                  * In a separate transaction, set the incomplete flag on the
 939                  * "old" attr and clear the incomplete flag on the "new" attr.
 940                  */
 941                 error = xfs_attr_leaf_flipflags(args);
 942                 if (error)
 943                         return(error);
 944
 945                 /*
 946                  * Dismantle the "old" attribute/value pair by removing
 947                  * a "remote" value (if it exists).
 948                  */
 949                 args->index = args->index2;
 950                 args->blkno = args->blkno2;
 951                 args->rmtblkno = args->rmtblkno2;
 952                 args->rmtblkcnt = args->rmtblkcnt2;
 953                 if (args->rmtblkno) {
 954                         error = xfs_attr_rmtval_remove(args);
 955                         if (error)
 956                                 return(error);
 957                 }
 958
 959                 /*
 960                  * Read in the block containing the "old" attr, then
 961                  * remove the "old" attr from that block (neat, huh!)
 962                  */
 963                 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1,
 964                                                      &bp, XFS_ATTR_FORK);
 965                 if (error)
 966                         return(error);
 967                 ASSERT(bp != NULL);
 968                 (void)xfs_attr_leaf_remove(bp, args);
 969
 970                 /*
 971                  * If the result is small enough, shrink it all into the inode.
 972                  */
 973                 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
 974                         XFS_BMAP_INIT(args->flist, args->firstblock);
 975                         error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
 976                         /* bp is gone due to xfs_da_shrink_inode */
 977                         if (!error) {
 978                                 error = xfs_bmap_finish(&args->trans,
 979                                                         args->flist,
 980                                                         *args->firstblock,
 981                                                         &committed);
 982                         }
 983                         if (error) {
 984                                 ASSERT(committed);
 985                                 args->trans = NULL;
 986                                 xfs_bmap_cancel(args->flist);
 987                                 return(error);
 988                         }
 989
 990                         /*
 991                          * bmap_finish() may have committed the last trans
 992                          * and started a new one.  We need the inode to be
 993                          * in all transactions.
 994                          */
 995                         if (committed) {
 996                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
 997                                 xfs_trans_ihold(args->trans, dp);
 998                         }
 999                 } else
1000                         xfs_da_buf_done(bp);
1001
1002                 /*
1003                  * Commit the remove and start the next trans in series.
1004                  */
1005                 error = xfs_attr_rolltrans(&args->trans, dp);
1006
1007         } else if (args->rmtblkno > 0) {
1008                 /*
1009                  * Added a "remote" value, just clear the incomplete flag.
1010                  */
1011                 error = xfs_attr_leaf_clearflag(args);
1012         }
1013         return(error);
1014 }
1015
1016 /*
1017  * Remove a name from the leaf attribute list structure
1018  *
1019  * This leaf block cannot have a "remote" value, we only call this routine
1020  * if bmap_one_block() says there is only one block (ie: no remote blks).
1021  */
1022 STATIC int
1023 xfs_attr_leaf_removename(xfs_da_args_t *args)
1024 {
1025         xfs_inode_t *dp;
1026         xfs_dabuf_t *bp;
1027         int error, committed, forkoff;
1028
1029         /*
1030          * Remove the attribute.
1031          */
1032         dp = args->dp;
1033         args->blkno = 0;
1034         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1035                                              XFS_ATTR_FORK);
1036         if (error) {
1037                 return(error);
1038         }
1039
1040         ASSERT(bp != NULL);
1041         error = xfs_attr_leaf_lookup_int(bp, args);
1042         if (error == ENOATTR) {
1043                 xfs_da_brelse(args->trans, bp);
1044                 return(error);
1045         }
1046
1047         (void)xfs_attr_leaf_remove(bp, args);
1048
1049         /*
1050          * If the result is small enough, shrink it all into the inode.
1051          */
1052         if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1053                 XFS_BMAP_INIT(args->flist, args->firstblock);
1054                 error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1055                 /* bp is gone due to xfs_da_shrink_inode */
1056                 if (!error) {
1057                         error = xfs_bmap_finish(&args->trans, args->flist,
1058                                                 *args->firstblock, &committed);
1059                 }
1060                 if (error) {
1061                         ASSERT(committed);
1062                         args->trans = NULL;
1063                         xfs_bmap_cancel(args->flist);
1064                         return(error);
1065                 }
1066
1067                 /*
1068                  * bmap_finish() may have committed the last trans and started
1069                  * a new one.  We need the inode to be in all transactions.
1070                  */
1071                 if (committed) {
1072                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1073                         xfs_trans_ihold(args->trans, dp);
1074                 }
1075         } else
1076                 xfs_da_buf_done(bp);
1077         return(0);
1078 }
1079
1080 /*
1081  * Look up a name in a leaf attribute list structure.
1082  *
1083  * This leaf block cannot have a "remote" value, we only call this routine
1084  * if bmap_one_block() says there is only one block (ie: no remote blks).
1085  */
1086 STATIC int
1087 xfs_attr_leaf_get(xfs_da_args_t *args)
1088 {
1089         xfs_dabuf_t *bp;
1090         int error;
1091
1092         args->blkno = 0;
1093         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1094                                              XFS_ATTR_FORK);
1095         if (error)
1096                 return(error);
1097         ASSERT(bp != NULL);
1098
1099         error = xfs_attr_leaf_lookup_int(bp, args);
1100         if (error != EEXIST)  {
1101                 xfs_da_brelse(args->trans, bp);
1102                 return(error);
1103         }
1104         error = xfs_attr_leaf_getvalue(bp, args);
1105         xfs_da_brelse(args->trans, bp);
1106         if (!error && (args->rmtblkno > 0) && !(args->flags & ATTR_KERNOVAL)) {
1107                 error = xfs_attr_rmtval_get(args);
1108         }
1109         return(error);
1110 }
1111
1112 /*
1113  * Copy out attribute entries for attr_list(), for leaf attribute lists.
1114  */
1115 STATIC int
1116 xfs_attr_leaf_list(xfs_attr_list_context_t *context)
1117 {
1118         xfs_attr_leafblock_t *leaf;
1119         int error;
1120         xfs_dabuf_t *bp;
1121
1122         context->cursor->blkno = 0;
1123         error = xfs_da_read_buf(NULL, context->dp, 0, -1, &bp, XFS_ATTR_FORK);
1124         if (error)
1125                 return(error);
1126         ASSERT(bp != NULL);
1127         leaf = bp->data;
1128         if (unlikely(be16_to_cpu(leaf->hdr.info.magic) != XFS_ATTR_LEAF_MAGIC)) {
1129                 XFS_CORRUPTION_ERROR("xfs_attr_leaf_list", XFS_ERRLEVEL_LOW,
1130                                      context->dp->i_mount, leaf);
1131                 xfs_da_brelse(NULL, bp);
1132                 return(XFS_ERROR(EFSCORRUPTED));
1133         }
1134
1135         (void)xfs_attr_leaf_list_int(bp, context);
1136         xfs_da_brelse(NULL, bp);
1137         return(0);
1138 }
1139
1140
1141 /*========================================================================
1142  * External routines when attribute list size > XFS_LBSIZE(mp).
1143  *========================================================================*/
1144
1145 /*
1146  * Add a name to a Btree-format attribute list.
1147  *
1148  * This will involve walking down the Btree, and may involve splitting
1149  * leaf nodes and even splitting intermediate nodes up to and including
1150  * the root node (a special case of an intermediate node).
1151  *
1152  * "Remote" attribute values confuse the issue and atomic rename operations
1153  * add a whole extra layer of confusion on top of that.
1154  */
1155 STATIC int
1156 xfs_attr_node_addname(xfs_da_args_t *args)
1157 {
1158         xfs_da_state_t *state;
1159         xfs_da_state_blk_t *blk;
1160         xfs_inode_t *dp;
1161         xfs_mount_t *mp;
1162         int committed, retval, error;
1163
1164         /*
1165          * Fill in bucket of arguments/results/context to carry around.
1166          */
1167         dp = args->dp;
1168         mp = dp->i_mount;
1169 restart:
1170         state = xfs_da_state_alloc();
1171         state->args = args;
1172         state->mp = mp;
1173         state->blocksize = state->mp->m_sb.sb_blocksize;
1174         state->node_ents = state->mp->m_attr_node_ents;
1175
1176         /*
1177          * Search to see if name already exists, and get back a pointer
1178          * to where it should go.
1179          */
1180         error = xfs_da_node_lookup_int(state, &retval);
1181         if (error)
1182                 goto out;
1183         blk = &state->path.blk[ state->path.active-1 ];
1184         ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1185         if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
1186                 goto out;
1187         } else if (retval == EEXIST) {
1188                 if (args->flags & ATTR_CREATE)
1189                         goto out;
1190                 args->rename = 1;                       /* atomic rename op */
1191                 args->blkno2 = args->blkno;             /* set 2nd entry info*/
1192                 args->index2 = args->index;
1193                 args->rmtblkno2 = args->rmtblkno;
1194                 args->rmtblkcnt2 = args->rmtblkcnt;
1195                 args->rmtblkno = 0;
1196                 args->rmtblkcnt = 0;
1197         }
1198
1199         retval = xfs_attr_leaf_add(blk->bp, state->args);
1200         if (retval == ENOSPC) {
1201                 if (state->path.active == 1) {
1202                         /*
1203                          * Its really a single leaf node, but it had
1204                          * out-of-line values so it looked like it *might*
1205                          * have been a b-tree.
1206                          */
1207                         xfs_da_state_free(state);
1208                         XFS_BMAP_INIT(args->flist, args->firstblock);
1209                         error = xfs_attr_leaf_to_node(args);
1210                         if (!error) {
1211                                 error = xfs_bmap_finish(&args->trans,
1212                                                         args->flist,
1213                                                         *args->firstblock,
1214                                                         &committed);
1215                         }
1216                         if (error) {
1217                                 ASSERT(committed);
1218                                 args->trans = NULL;
1219                                 xfs_bmap_cancel(args->flist);
1220                                 goto out;
1221                         }
1222
1223                         /*
1224                          * bmap_finish() may have committed the last trans
1225                          * and started a new one.  We need the inode to be
1226                          * in all transactions.
1227                          */
1228                         if (committed) {
1229                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1230                                 xfs_trans_ihold(args->trans, dp);
1231                         }
1232
1233                         /*
1234                          * Commit the node conversion and start the next
1235                          * trans in the chain.
1236                          */
1237                         if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1238                                 goto out;
1239
1240                         goto restart;
1241                 }
1242
1243                 /*
1244                  * Split as many Btree elements as required.
1245                  * This code tracks the new and old attr's location
1246                  * in the index/blkno/rmtblkno/rmtblkcnt fields and
1247                  * in the index2/blkno2/rmtblkno2/rmtblkcnt2 fields.
1248                  */
1249                 XFS_BMAP_INIT(args->flist, args->firstblock);
1250                 error = xfs_da_split(state);
1251                 if (!error) {
1252                         error = xfs_bmap_finish(&args->trans, args->flist,
1253                                                 *args->firstblock, &committed);
1254                 }
1255                 if (error) {
1256                         ASSERT(committed);
1257                         args->trans = NULL;
1258                         xfs_bmap_cancel(args->flist);
1259                         goto out;
1260                 }
1261
1262                 /*
1263                  * bmap_finish() may have committed the last trans and started
1264                  * a new one.  We need the inode to be in all transactions.
1265                  */
1266                 if (committed) {
1267                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1268                         xfs_trans_ihold(args->trans, dp);
1269                 }
1270         } else {
1271                 /*
1272                  * Addition succeeded, update Btree hashvals.
1273                  */
1274                 xfs_da_fixhashpath(state, &state->path);
1275         }
1276
1277         /*
1278          * Kill the state structure, we're done with it and need to
1279          * allow the buffers to come back later.
1280          */
1281         xfs_da_state_free(state);
1282         state = NULL;
1283
1284         /*
1285          * Commit the leaf addition or btree split and start the next
1286          * trans in the chain.
1287          */
1288         if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1289                 goto out;
1290
1291         /*
1292          * If there was an out-of-line value, allocate the blocks we
1293          * identified for its storage and copy the value.  This is done
1294          * after we create the attribute so that we don't overflow the
1295          * maximum size of a transaction and/or hit a deadlock.
1296          */
1297         if (args->rmtblkno > 0) {
1298                 error = xfs_attr_rmtval_set(args);
1299                 if (error)
1300                         return(error);
1301         }
1302
1303         /*
1304          * If this is an atomic rename operation, we must "flip" the
1305          * incomplete flags on the "new" and "old" attribute/value pairs
1306          * so that one disappears and one appears atomically.  Then we
1307          * must remove the "old" attribute/value pair.
1308          */
1309         if (args->rename) {
1310                 /*
1311                  * In a separate transaction, set the incomplete flag on the
1312                  * "old" attr and clear the incomplete flag on the "new" attr.
1313                  */
1314                 error = xfs_attr_leaf_flipflags(args);
1315                 if (error)
1316                         goto out;
1317
1318                 /*
1319                  * Dismantle the "old" attribute/value pair by removing
1320                  * a "remote" value (if it exists).
1321                  */
1322                 args->index = args->index2;
1323                 args->blkno = args->blkno2;
1324                 args->rmtblkno = args->rmtblkno2;
1325                 args->rmtblkcnt = args->rmtblkcnt2;
1326                 if (args->rmtblkno) {
1327                         error = xfs_attr_rmtval_remove(args);
1328                         if (error)
1329                                 return(error);
1330                 }
1331
1332                 /*
1333                  * Re-find the "old" attribute entry after any split ops.
1334                  * The INCOMPLETE flag means that we will find the "old"
1335                  * attr, not the "new" one.
1336                  */
1337                 args->flags |= XFS_ATTR_INCOMPLETE;
1338                 state = xfs_da_state_alloc();
1339                 state->args = args;
1340                 state->mp = mp;
1341                 state->blocksize = state->mp->m_sb.sb_blocksize;
1342                 state->node_ents = state->mp->m_attr_node_ents;
1343                 state->inleaf = 0;
1344                 error = xfs_da_node_lookup_int(state, &retval);
1345                 if (error)
1346                         goto out;
1347
1348                 /*
1349                  * Remove the name and update the hashvals in the tree.
1350                  */
1351                 blk = &state->path.blk[ state->path.active-1 ];
1352                 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1353                 error = xfs_attr_leaf_remove(blk->bp, args);
1354                 xfs_da_fixhashpath(state, &state->path);
1355
1356                 /*
1357                  * Check to see if the tree needs to be collapsed.
1358                  */
1359                 if (retval && (state->path.active > 1)) {
1360                         XFS_BMAP_INIT(args->flist, args->firstblock);
1361                         error = xfs_da_join(state);
1362                         if (!error) {
1363                                 error = xfs_bmap_finish(&args->trans,
1364                                                         args->flist,
1365                                                         *args->firstblock,
1366                                                         &committed);
1367                         }
1368                         if (error) {
1369                                 ASSERT(committed);
1370                                 args->trans = NULL;
1371                                 xfs_bmap_cancel(args->flist);
1372                                 goto out;
1373                         }
1374
1375                         /*
1376                          * bmap_finish() may have committed the last trans
1377                          * and started a new one.  We need the inode to be
1378                          * in all transactions.
1379                          */
1380                         if (committed) {
1381                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1382                                 xfs_trans_ihold(args->trans, dp);
1383                         }
1384                 }
1385
1386                 /*
1387                  * Commit and start the next trans in the chain.
1388                  */
1389                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1390                         goto out;
1391
1392         } else if (args->rmtblkno > 0) {
1393                 /*
1394                  * Added a "remote" value, just clear the incomplete flag.
1395                  */
1396                 error = xfs_attr_leaf_clearflag(args);
1397                 if (error)
1398                         goto out;
1399         }
1400         retval = error = 0;
1401
1402 out:
1403         if (state)
1404                 xfs_da_state_free(state);
1405         if (error)
1406                 return(error);
1407         return(retval);
1408 }
1409
1410 /*
1411  * Remove a name from a B-tree attribute list.
1412  *
1413  * This will involve walking down the Btree, and may involve joining
1414  * leaf nodes and even joining intermediate nodes up to and including
1415  * the root node (a special case of an intermediate node).
1416  */
1417 STATIC int
1418 xfs_attr_node_removename(xfs_da_args_t *args)
1419 {
1420         xfs_da_state_t *state;
1421         xfs_da_state_blk_t *blk;
1422         xfs_inode_t *dp;
1423         xfs_dabuf_t *bp;
1424         int retval, error, committed, forkoff;
1425
1426         /*
1427          * Tie a string around our finger to remind us where we are.
1428          */
1429         dp = args->dp;
1430         state = xfs_da_state_alloc();
1431         state->args = args;
1432         state->mp = dp->i_mount;
1433         state->blocksize = state->mp->m_sb.sb_blocksize;
1434         state->node_ents = state->mp->m_attr_node_ents;
1435
1436         /*
1437          * Search to see if name exists, and get back a pointer to it.
1438          */
1439         error = xfs_da_node_lookup_int(state, &retval);
1440         if (error || (retval != EEXIST)) {
1441                 if (error == 0)
1442                         error = retval;
1443                 goto out;
1444         }
1445
1446         /*
1447          * If there is an out-of-line value, de-allocate the blocks.
1448          * This is done before we remove the attribute so that we don't
1449          * overflow the maximum size of a transaction and/or hit a deadlock.
1450          */
1451         blk = &state->path.blk[ state->path.active-1 ];
1452         ASSERT(blk->bp != NULL);
1453         ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1454         if (args->rmtblkno > 0) {
1455                 /*
1456                  * Fill in disk block numbers in the state structure
1457                  * so that we can get the buffers back after we commit
1458                  * several transactions in the following calls.
1459                  */
1460                 error = xfs_attr_fillstate(state);
1461                 if (error)
1462                         goto out;
1463
1464                 /*
1465                  * Mark the attribute as INCOMPLETE, then bunmapi() the
1466                  * remote value.
1467                  */
1468                 error = xfs_attr_leaf_setflag(args);
1469                 if (error)
1470                         goto out;
1471                 error = xfs_attr_rmtval_remove(args);
1472                 if (error)
1473                         goto out;
1474
1475                 /*
1476                  * Refill the state structure with buffers, the prior calls
1477                  * released our buffers.
1478                  */
1479                 error = xfs_attr_refillstate(state);
1480                 if (error)
1481                         goto out;
1482         }
1483
1484         /*
1485          * Remove the name and update the hashvals in the tree.
1486          */
1487         blk = &state->path.blk[ state->path.active-1 ];
1488         ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1489         retval = xfs_attr_leaf_remove(blk->bp, args);
1490         xfs_da_fixhashpath(state, &state->path);
1491
1492         /*
1493          * Check to see if the tree needs to be collapsed.
1494          */
1495         if (retval && (state->path.active > 1)) {
1496                 XFS_BMAP_INIT(args->flist, args->firstblock);
1497                 error = xfs_da_join(state);
1498                 if (!error) {
1499                         error = xfs_bmap_finish(&args->trans, args->flist,
1500                                                 *args->firstblock, &committed);
1501                 }
1502                 if (error) {
1503                         ASSERT(committed);
1504                         args->trans = NULL;
1505                         xfs_bmap_cancel(args->flist);
1506                         goto out;
1507                 }
1508
1509                 /*
1510                  * bmap_finish() may have committed the last trans and started
1511                  * a new one.  We need the inode to be in all transactions.
1512                  */
1513                 if (committed) {
1514                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1515                         xfs_trans_ihold(args->trans, dp);
1516                 }
1517
1518                 /*
1519                  * Commit the Btree join operation and start a new trans.
1520                  */
1521                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1522                         goto out;
1523         }
1524
1525         /*
1526          * If the result is small enough, push it all into the inode.
1527          */
1528         if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
1529                 /*
1530                  * Have to get rid of the copy of this dabuf in the state.
1531                  */
1532                 ASSERT(state->path.active == 1);
1533                 ASSERT(state->path.blk[0].bp);
1534                 xfs_da_buf_done(state->path.blk[0].bp);
1535                 state->path.blk[0].bp = NULL;
1536
1537                 error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp,
1538                                                      XFS_ATTR_FORK);
1539                 if (error)
1540                         goto out;
1541                 ASSERT(be16_to_cpu(((xfs_attr_leafblock_t *)
1542                                       bp->data)->hdr.info.magic)
1543                                                        == XFS_ATTR_LEAF_MAGIC);
1544
1545                 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1546                         XFS_BMAP_INIT(args->flist, args->firstblock);
1547                         error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1548                         /* bp is gone due to xfs_da_shrink_inode */
1549                         if (!error) {
1550                                 error = xfs_bmap_finish(&args->trans,
1551                                                         args->flist,
1552                                                         *args->firstblock,
1553                                                         &committed);
1554                         }
1555                         if (error) {
1556                                 ASSERT(committed);
1557                                 args->trans = NULL;
1558                                 xfs_bmap_cancel(args->flist);
1559                                 goto out;
1560                         }
1561
1562                         /*
1563                          * bmap_finish() may have committed the last trans
1564                          * and started a new one.  We need the inode to be
1565                          * in all transactions.
1566                          */
1567                         if (committed) {
1568                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1569                                 xfs_trans_ihold(args->trans, dp);
1570                         }
1571                 } else
1572                         xfs_da_brelse(args->trans, bp);
1573         }
1574         error = 0;
1575
1576 out:
1577         xfs_da_state_free(state);
1578         return(error);
1579 }
1580
1581 /*
1582  * Fill in the disk block numbers in the state structure for the buffers
1583  * that are attached to the state structure.
1584  * This is done so that we can quickly reattach ourselves to those buffers
1585  * after some set of transaction commits have released these buffers.
1586  */
1587 STATIC int
1588 xfs_attr_fillstate(xfs_da_state_t *state)
1589 {
1590         xfs_da_state_path_t *path;
1591         xfs_da_state_blk_t *blk;
1592         int level;
1593
1594         /*
1595          * Roll down the "path" in the state structure, storing the on-disk
1596          * block number for those buffers in the "path".
1597          */
1598         path = &state->path;
1599         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1600         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1601                 if (blk->bp) {
1602                         blk->disk_blkno = xfs_da_blkno(blk->bp);
1603                         xfs_da_buf_done(blk->bp);
1604                         blk->bp = NULL;
1605                 } else {
1606                         blk->disk_blkno = 0;
1607                 }
1608         }
1609
1610         /*
1611          * Roll down the "altpath" in the state structure, storing the on-disk
1612          * block number for those buffers in the "altpath".
1613          */
1614         path = &state->altpath;
1615         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1616         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1617                 if (blk->bp) {
1618                         blk->disk_blkno = xfs_da_blkno(blk->bp);
1619                         xfs_da_buf_done(blk->bp);
1620                         blk->bp = NULL;
1621                 } else {
1622                         blk->disk_blkno = 0;
1623                 }
1624         }
1625
1626         return(0);
1627 }
1628
1629 /*
1630  * Reattach the buffers to the state structure based on the disk block
1631  * numbers stored in the state structure.
1632  * This is done after some set of transaction commits have released those
1633  * buffers from our grip.
1634  */
1635 STATIC int
1636 xfs_attr_refillstate(xfs_da_state_t *state)
1637 {
1638         xfs_da_state_path_t *path;
1639         xfs_da_state_blk_t *blk;
1640         int level, error;
1641
1642         /*
1643          * Roll down the "path" in the state structure, storing the on-disk
1644          * block number for those buffers in the "path".
1645          */
1646         path = &state->path;
1647         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1648         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1649                 if (blk->disk_blkno) {
1650                         error = xfs_da_read_buf(state->args->trans,
1651                                                 state->args->dp,
1652                                                 blk->blkno, blk->disk_blkno,
1653                                                 &blk->bp, XFS_ATTR_FORK);
1654                         if (error)
1655                                 return(error);
1656                 } else {
1657                         blk->bp = NULL;
1658                 }
1659         }
1660
1661         /*
1662          * Roll down the "altpath" in the state structure, storing the on-disk
1663          * block number for those buffers in the "altpath".
1664          */
1665         path = &state->altpath;
1666         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1667         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1668                 if (blk->disk_blkno) {
1669                         error = xfs_da_read_buf(state->args->trans,
1670                                                 state->args->dp,
1671                                                 blk->blkno, blk->disk_blkno,
1672                                                 &blk->bp, XFS_ATTR_FORK);
1673                         if (error)
1674                                 return(error);
1675                 } else {
1676                         blk->bp = NULL;
1677                 }
1678         }
1679
1680         return(0);
1681 }
1682
1683 /*
1684  * Look up a filename in a node attribute list.
1685  *
1686  * This routine gets called for any attribute fork that has more than one
1687  * block, ie: both true Btree attr lists and for single-leaf-blocks with
1688  * "remote" values taking up more blocks.
1689  */
1690 STATIC int
1691 xfs_attr_node_get(xfs_da_args_t *args)
1692 {
1693         xfs_da_state_t *state;
1694         xfs_da_state_blk_t *blk;
1695         int error, retval;
1696         int i;
1697
1698         state = xfs_da_state_alloc();
1699         state->args = args;
1700         state->mp = args->dp->i_mount;
1701         state->blocksize = state->mp->m_sb.sb_blocksize;
1702         state->node_ents = state->mp->m_attr_node_ents;
1703
1704         /*
1705          * Search to see if name exists, and get back a pointer to it.
1706          */
1707         error = xfs_da_node_lookup_int(state, &retval);
1708         if (error) {
1709                 retval = error;
1710         } else if (retval == EEXIST) {
1711                 blk = &state->path.blk[ state->path.active-1 ];
1712                 ASSERT(blk->bp != NULL);
1713                 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1714
1715                 /*
1716                  * Get the value, local or "remote"
1717                  */
1718                 retval = xfs_attr_leaf_getvalue(blk->bp, args);
1719                 if (!retval && (args->rmtblkno > 0)
1720                     && !(args->flags & ATTR_KERNOVAL)) {
1721                         retval = xfs_attr_rmtval_get(args);
1722                 }
1723         }
1724
1725         /*
1726          * If not in a transaction, we have to release all the buffers.
1727          */
1728         for (i = 0; i < state->path.active; i++) {
1729                 xfs_da_brelse(args->trans, state->path.blk[i].bp);
1730                 state->path.blk[i].bp = NULL;
1731         }
1732
1733         xfs_da_state_free(state);
1734         return(retval);
1735 }
1736
1737 STATIC int                                                      /* error */
1738 xfs_attr_node_list(xfs_attr_list_context_t *context)
1739 {
1740         attrlist_cursor_kern_t *cursor;
1741         xfs_attr_leafblock_t *leaf;
1742         xfs_da_intnode_t *node;
1743         xfs_da_node_entry_t *btree;
1744         int error, i;
1745         xfs_dabuf_t *bp;
1746
1747         cursor = context->cursor;
1748         cursor->initted = 1;
1749
1750         /*
1751          * Do all sorts of validation on the passed-in cursor structure.
1752          * If anything is amiss, ignore the cursor and look up the hashval
1753          * starting from the btree root.
1754          */
1755         bp = NULL;
1756         if (cursor->blkno > 0) {
1757                 error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1758                                               &bp, XFS_ATTR_FORK);
1759                 if ((error != 0) && (error != EFSCORRUPTED))
1760                         return(error);
1761                 if (bp) {
1762                         node = bp->data;
1763                         switch (be16_to_cpu(node->hdr.info.magic)) {
1764                         case XFS_DA_NODE_MAGIC:
1765                                 xfs_attr_trace_l_cn("wrong blk", context, node);
1766                                 xfs_da_brelse(NULL, bp);
1767                                 bp = NULL;
1768                                 break;
1769                         case XFS_ATTR_LEAF_MAGIC:
1770                                 leaf = bp->data;
1771                                 if (cursor->hashval > be32_to_cpu(leaf->entries[
1772                                     be16_to_cpu(leaf->hdr.count)-1].hashval)) {
1773                                         xfs_attr_trace_l_cl("wrong blk",
1774                                                            context, leaf);
1775                                         xfs_da_brelse(NULL, bp);
1776                                         bp = NULL;
1777                                 } else if (cursor->hashval <=
1778                                              be32_to_cpu(leaf->entries[0].hashval)) {
1779                                         xfs_attr_trace_l_cl("maybe wrong blk",
1780                                                            context, leaf);
1781                                         xfs_da_brelse(NULL, bp);
1782                                         bp = NULL;
1783                                 }
1784                                 break;
1785                         default:
1786                                 xfs_attr_trace_l_c("wrong blk - ??", context);
1787                                 xfs_da_brelse(NULL, bp);
1788                                 bp = NULL;
1789                         }
1790                 }
1791         }
1792
1793         /*
1794          * We did not find what we expected given the cursor's contents,
1795          * so we start from the top and work down based on the hash value.
1796          * Note that start of node block is same as start of leaf block.
1797          */
1798         if (bp == NULL) {
1799                 cursor->blkno = 0;
1800                 for (;;) {
1801                         error = xfs_da_read_buf(NULL, context->dp,
1802                                                       cursor->blkno, -1, &bp,
1803                                                       XFS_ATTR_FORK);
1804                         if (error)
1805                                 return(error);
1806                         if (unlikely(bp == NULL)) {
1807                                 XFS_ERROR_REPORT("xfs_attr_node_list(2)",
1808                                                  XFS_ERRLEVEL_LOW,
1809                                                  context->dp->i_mount);
1810                                 return(XFS_ERROR(EFSCORRUPTED));
1811                         }
1812                         node = bp->data;
1813                         if (be16_to_cpu(node->hdr.info.magic)
1814                                                         == XFS_ATTR_LEAF_MAGIC)
1815                                 break;
1816                         if (unlikely(be16_to_cpu(node->hdr.info.magic)
1817                                                         != XFS_DA_NODE_MAGIC)) {
1818                                 XFS_CORRUPTION_ERROR("xfs_attr_node_list(3)",
1819                                                      XFS_ERRLEVEL_LOW,
1820                                                      context->dp->i_mount,
1821                                                      node);
1822                                 xfs_da_brelse(NULL, bp);
1823                                 return(XFS_ERROR(EFSCORRUPTED));
1824                         }
1825                         btree = node->btree;
1826                         for (i = 0; i < be16_to_cpu(node->hdr.count);
1827                                                                 btree++, i++) {
1828                                 if (cursor->hashval
1829                                                 <= be32_to_cpu(btree->hashval)) {
1830                                         cursor->blkno = be32_to_cpu(btree->before);
1831                                         xfs_attr_trace_l_cb("descending",
1832                                                             context, btree);
1833                                         break;
1834                                 }
1835                         }
1836                         if (i == be16_to_cpu(node->hdr.count)) {
1837                                 xfs_da_brelse(NULL, bp);
1838                                 return(0);
1839                         }
1840                         xfs_da_brelse(NULL, bp);
1841                 }
1842         }
1843         ASSERT(bp != NULL);
1844
1845         /*
1846          * Roll upward through the blocks, processing each leaf block in
1847          * order.  As long as there is space in the result buffer, keep
1848          * adding the information.
1849          */
1850         for (;;) {
1851                 leaf = bp->data;
1852                 if (unlikely(be16_to_cpu(leaf->hdr.info.magic)
1853                                                 != XFS_ATTR_LEAF_MAGIC)) {
1854                         XFS_CORRUPTION_ERROR("xfs_attr_node_list(4)",
1855                                              XFS_ERRLEVEL_LOW,
1856                                              context->dp->i_mount, leaf);
1857                         xfs_da_brelse(NULL, bp);
1858                         return(XFS_ERROR(EFSCORRUPTED));
1859                 }
1860                 error = xfs_attr_leaf_list_int(bp, context);
1861                 if (error || !leaf->hdr.info.forw)
1862                         break;  /* not really an error, buffer full or EOF */
1863                 cursor->blkno = be32_to_cpu(leaf->hdr.info.forw);
1864                 xfs_da_brelse(NULL, bp);
1865                 error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1866                                               &bp, XFS_ATTR_FORK);
1867                 if (error)
1868                         return(error);
1869                 if (unlikely((bp == NULL))) {
1870                         XFS_ERROR_REPORT("xfs_attr_node_list(5)",
1871                                          XFS_ERRLEVEL_LOW,
1872                                          context->dp->i_mount);
1873                         return(XFS_ERROR(EFSCORRUPTED));
1874                 }
1875         }
1876         xfs_da_brelse(NULL, bp);
1877         return(0);
1878 }
1879
1880
1881 /*========================================================================
1882  * External routines for manipulating out-of-line attribute values.
1883  *========================================================================*/
1884
1885 /*
1886  * Read the value associated with an attribute from the out-of-line buffer
1887  * that we stored it in.
1888  */
1889 STATIC int
1890 xfs_attr_rmtval_get(xfs_da_args_t *args)
1891 {
1892         xfs_bmbt_irec_t map[ATTR_RMTVALUE_MAPSIZE];
1893         xfs_mount_t *mp;
1894         xfs_daddr_t dblkno;
1895         xfs_caddr_t dst;
1896         xfs_buf_t *bp;
1897         int nmap, error, tmp, valuelen, blkcnt, i;
1898         xfs_dablk_t lblkno;
1899
1900         ASSERT(!(args->flags & ATTR_KERNOVAL));
1901
1902         mp = args->dp->i_mount;
1903         dst = args->value;
1904         valuelen = args->valuelen;
1905         lblkno = args->rmtblkno;
1906         while (valuelen > 0) {
1907                 nmap = ATTR_RMTVALUE_MAPSIZE;
1908                 error = xfs_bmapi(args->trans, args->dp, (xfs_fileoff_t)lblkno,
1909                                   args->rmtblkcnt,
1910                                   XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
1911                                   NULL, 0, map, &nmap, NULL, NULL);
1912                 if (error)
1913                         return(error);
1914                 ASSERT(nmap >= 1);
1915
1916                 for (i = 0; (i < nmap) && (valuelen > 0); i++) {
1917                         ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
1918                                (map[i].br_startblock != HOLESTARTBLOCK));
1919                         dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
1920                         blkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
1921                         error = xfs_read_buf(mp, mp->m_ddev_targp, dblkno,
1922                                              blkcnt, XFS_BUF_LOCK, &bp);
1923                         if (error)
1924                                 return(error);
1925
1926                         tmp = (valuelen < XFS_BUF_SIZE(bp))
1927                                 ? valuelen : XFS_BUF_SIZE(bp);
1928                         xfs_biomove(bp, 0, tmp, dst, XFS_B_READ);
1929                         xfs_buf_relse(bp);
1930                         dst += tmp;
1931                         valuelen -= tmp;
1932
1933                         lblkno += map[i].br_blockcount;
1934                 }
1935         }
1936         ASSERT(valuelen == 0);
1937         return(0);
1938 }
1939
1940 /*
1941  * Write the value associated with an attribute into the out-of-line buffer
1942  * that we have defined for it.
1943  */
1944 STATIC int
1945 xfs_attr_rmtval_set(xfs_da_args_t *args)
1946 {
1947         xfs_mount_t *mp;
1948         xfs_fileoff_t lfileoff;
1949         xfs_inode_t *dp;
1950         xfs_bmbt_irec_t map;
1951         xfs_daddr_t dblkno;
1952         xfs_caddr_t src;
1953         xfs_buf_t *bp;
1954         xfs_dablk_t lblkno;
1955         int blkcnt, valuelen, nmap, error, tmp, committed;
1956
1957         dp = args->dp;
1958         mp = dp->i_mount;
1959         src = args->value;
1960
1961         /*
1962          * Find a "hole" in the attribute address space large enough for
1963          * us to drop the new attribute's value into.
1964          */
1965         blkcnt = XFS_B_TO_FSB(mp, args->valuelen);
1966         lfileoff = 0;
1967         error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
1968                                                    XFS_ATTR_FORK);
1969         if (error) {
1970                 return(error);
1971         }
1972         args->rmtblkno = lblkno = (xfs_dablk_t)lfileoff;
1973         args->rmtblkcnt = blkcnt;
1974
1975         /*
1976          * Roll through the "value", allocating blocks on disk as required.
1977          */
1978         while (blkcnt > 0) {
1979                 /*
1980                  * Allocate a single extent, up to the size of the value.
1981                  */
1982                 XFS_BMAP_INIT(args->flist, args->firstblock);
1983                 nmap = 1;
1984                 error = xfs_bmapi(args->trans, dp, (xfs_fileoff_t)lblkno,
1985                                   blkcnt,
1986                                   XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA |
1987                                                         XFS_BMAPI_WRITE,
1988                                   args->firstblock, args->total, &map, &nmap,
1989                                   args->flist, NULL);
1990                 if (!error) {
1991                         error = xfs_bmap_finish(&args->trans, args->flist,
1992                                                 *args->firstblock, &committed);
1993                 }
1994                 if (error) {
1995                         ASSERT(committed);
1996                         args->trans = NULL;
1997                         xfs_bmap_cancel(args->flist);
1998                         return(error);
1999                 }
2000
2001                 /*
2002                  * bmap_finish() may have committed the last trans and started
2003                  * a new one.  We need the inode to be in all transactions.
2004                  */
2005                 if (committed) {
2006                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
2007                         xfs_trans_ihold(args->trans, dp);
2008                 }
2009
2010                 ASSERT(nmap == 1);
2011                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2012                        (map.br_startblock != HOLESTARTBLOCK));
2013                 lblkno += map.br_blockcount;
2014                 blkcnt -= map.br_blockcount;
2015
2016                 /*
2017                  * Start the next trans in the chain.
2018                  */
2019                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
2020                         return (error);
2021         }
2022
2023         /*
2024          * Roll through the "value", copying the attribute value to the
2025          * already-allocated blocks.  Blocks are written synchronously
2026          * so that we can know they are all on disk before we turn off
2027          * the INCOMPLETE flag.
2028          */
2029         lblkno = args->rmtblkno;
2030         valuelen = args->valuelen;
2031         while (valuelen > 0) {
2032                 /*
2033                  * Try to remember where we decided to put the value.
2034                  */
2035                 XFS_BMAP_INIT(args->flist, args->firstblock);
2036                 nmap = 1;
2037                 error = xfs_bmapi(NULL, dp, (xfs_fileoff_t)lblkno,
2038                                   args->rmtblkcnt,
2039                                   XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2040                                   args->firstblock, 0, &map, &nmap,
2041                                   NULL, NULL);
2042                 if (error) {
2043                         return(error);
2044                 }
2045                 ASSERT(nmap == 1);
2046                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2047                        (map.br_startblock != HOLESTARTBLOCK));
2048
2049                 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2050                 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2051
2052                 bp = xfs_buf_get_flags(mp->m_ddev_targp, dblkno,
2053                                                         blkcnt, XFS_BUF_LOCK);
2054                 ASSERT(bp);
2055                 ASSERT(!XFS_BUF_GETERROR(bp));
2056
2057                 tmp = (valuelen < XFS_BUF_SIZE(bp)) ? valuelen :
2058                                                         XFS_BUF_SIZE(bp);
2059                 xfs_biomove(bp, 0, tmp, src, XFS_B_WRITE);
2060                 if (tmp < XFS_BUF_SIZE(bp))
2061                         xfs_biozero(bp, tmp, XFS_BUF_SIZE(bp) - tmp);
2062                 if ((error = xfs_bwrite(mp, bp))) {/* GROT: NOTE: synchronous write */
2063                         return (error);
2064                 }
2065                 src += tmp;
2066                 valuelen -= tmp;
2067
2068                 lblkno += map.br_blockcount;
2069         }
2070         ASSERT(valuelen == 0);
2071         return(0);
2072 }
2073
2074 /*
2075  * Remove the value associated with an attribute by deleting the
2076  * out-of-line buffer that it is stored on.
2077  */
2078 STATIC int
2079 xfs_attr_rmtval_remove(xfs_da_args_t *args)
2080 {
2081         xfs_mount_t *mp;
2082         xfs_bmbt_irec_t map;
2083         xfs_buf_t *bp;
2084         xfs_daddr_t dblkno;
2085         xfs_dablk_t lblkno;
2086         int valuelen, blkcnt, nmap, error, done, committed;
2087
2088         mp = args->dp->i_mount;
2089
2090         /*
2091          * Roll through the "value", invalidating the attribute value's
2092          * blocks.
2093          */
2094         lblkno = args->rmtblkno;
2095         valuelen = args->rmtblkcnt;
2096         while (valuelen > 0) {
2097                 /*
2098                  * Try to remember where we decided to put the value.
2099                  */
2100                 XFS_BMAP_INIT(args->flist, args->firstblock);
2101                 nmap = 1;
2102                 error = xfs_bmapi(NULL, args->dp, (xfs_fileoff_t)lblkno,
2103                                         args->rmtblkcnt,
2104                                         XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2105                                         args->firstblock, 0, &map, &nmap,
2106                                         args->flist, NULL);
2107                 if (error) {
2108                         return(error);
2109                 }
2110                 ASSERT(nmap == 1);
2111                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2112                        (map.br_startblock != HOLESTARTBLOCK));
2113
2114                 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2115                 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2116
2117                 /*
2118                  * If the "remote" value is in the cache, remove it.
2119                  */
2120                 bp = xfs_incore(mp->m_ddev_targp, dblkno, blkcnt,
2121                                 XFS_INCORE_TRYLOCK);
2122                 if (bp) {
2123                         XFS_BUF_STALE(bp);
2124                         XFS_BUF_UNDELAYWRITE(bp);
2125                         xfs_buf_relse(bp);
2126                         bp = NULL;
2127                 }
2128
2129                 valuelen -= map.br_blockcount;
2130
2131                 lblkno += map.br_blockcount;
2132         }
2133
2134         /*
2135          * Keep de-allocating extents until the remote-value region is gone.
2136          */
2137         lblkno = args->rmtblkno;
2138         blkcnt = args->rmtblkcnt;
2139         done = 0;
2140         while (!done) {
2141                 XFS_BMAP_INIT(args->flist, args->firstblock);
2142                 error = xfs_bunmapi(args->trans, args->dp, lblkno, blkcnt,
2143                                     XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2144                                     1, args->firstblock, args->flist,
2145                                     NULL, &done);
2146                 if (!error) {
2147                         error = xfs_bmap_finish(&args->trans, args->flist,
2148                                                 *args->firstblock, &committed);
2149                 }
2150                 if (error) {
2151                         ASSERT(committed);
2152                         args->trans = NULL;
2153                         xfs_bmap_cancel(args->flist);
2154                         return(error);
2155                 }
2156
2157                 /*
2158                  * bmap_finish() may have committed the last trans and started
2159                  * a new one.  We need the inode to be in all transactions.
2160                  */
2161                 if (committed) {
2162                         xfs_trans_ijoin(args->trans, args->dp, XFS_ILOCK_EXCL);
2163                         xfs_trans_ihold(args->trans, args->dp);
2164                 }
2165
2166                 /*
2167                  * Close out trans and start the next one in the chain.
2168                  */
2169                 if ((error = xfs_attr_rolltrans(&args->trans, args->dp)))
2170                         return (error);
2171         }
2172         return(0);
2173 }
2174
2175 #if defined(XFS_ATTR_TRACE)
2176 /*
2177  * Add a trace buffer entry for an attr_list context structure.
2178  */
2179 void
2180 xfs_attr_trace_l_c(char *where, struct xfs_attr_list_context *context)
2181 {
2182         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_C, where,
2183                 (__psunsigned_t)context->dp,
2184                 (__psunsigned_t)context->cursor->hashval,
2185                 (__psunsigned_t)context->cursor->blkno,
2186                 (__psunsigned_t)context->cursor->offset,
2187                 (__psunsigned_t)context->alist,
2188                 (__psunsigned_t)context->bufsize,
2189                 (__psunsigned_t)context->count,
2190                 (__psunsigned_t)context->firstu,
2191                 (__psunsigned_t)
2192                         ((context->count > 0) &&
2193                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2194                                 ? (ATTR_ENTRY(context->alist,
2195                                               context->count-1)->a_valuelen)
2196                                 : 0,
2197                 (__psunsigned_t)context->dupcnt,
2198                 (__psunsigned_t)context->flags,
2199                 (__psunsigned_t)NULL,
2200                 (__psunsigned_t)NULL,
2201                 (__psunsigned_t)NULL);
2202 }
2203
2204 /*
2205  * Add a trace buffer entry for a context structure and a Btree node.
2206  */
2207 void
2208 xfs_attr_trace_l_cn(char *where, struct xfs_attr_list_context *context,
2209                          struct xfs_da_intnode *node)
2210 {
2211         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_CN, where,
2212                 (__psunsigned_t)context->dp,
2213                 (__psunsigned_t)context->cursor->hashval,
2214                 (__psunsigned_t)context->cursor->blkno,
2215                 (__psunsigned_t)context->cursor->offset,
2216                 (__psunsigned_t)context->alist,
2217                 (__psunsigned_t)context->bufsize,
2218                 (__psunsigned_t)context->count,
2219                 (__psunsigned_t)context->firstu,
2220                 (__psunsigned_t)
2221                         ((context->count > 0) &&
2222                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2223                                 ? (ATTR_ENTRY(context->alist,
2224                                               context->count-1)->a_valuelen)
2225                                 : 0,
2226                 (__psunsigned_t)context->dupcnt,
2227                 (__psunsigned_t)context->flags,
2228                 (__psunsigned_t)be16_to_cpu(node->hdr.count),
2229                 (__psunsigned_t)be32_to_cpu(node->btree[0].hashval),
2230                 (__psunsigned_t)be32_to_cpu(node->btree[
2231                                     be16_to_cpu(node->hdr.count)-1].hashval));
2232 }
2233
2234 /*
2235  * Add a trace buffer entry for a context structure and a Btree element.
2236  */
2237 void
2238 xfs_attr_trace_l_cb(char *where, struct xfs_attr_list_context *context,
2239                           struct xfs_da_node_entry *btree)
2240 {
2241         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_CB, where,
2242                 (__psunsigned_t)context->dp,
2243                 (__psunsigned_t)context->cursor->hashval,
2244                 (__psunsigned_t)context->cursor->blkno,
2245                 (__psunsigned_t)context->cursor->offset,
2246                 (__psunsigned_t)context->alist,
2247                 (__psunsigned_t)context->bufsize,
2248                 (__psunsigned_t)context->count,
2249                 (__psunsigned_t)context->firstu,
2250                 (__psunsigned_t)
2251                         ((context->count > 0) &&
2252                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2253                                 ? (ATTR_ENTRY(context->alist,
2254                                               context->count-1)->a_valuelen)
2255                                 : 0,
2256                 (__psunsigned_t)context->dupcnt,
2257                 (__psunsigned_t)context->flags,
2258                 (__psunsigned_t)be32_to_cpu(btree->hashval),
2259                 (__psunsigned_t)be32_to_cpu(btree->before),
2260                 (__psunsigned_t)NULL);
2261 }
2262
2263 /*
2264  * Add a trace buffer entry for a context structure and a leaf block.
2265  */
2266 void
2267 xfs_attr_trace_l_cl(char *where, struct xfs_attr_list_context *context,
2268                               struct xfs_attr_leafblock *leaf)
2269 {
2270         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_CL, where,
2271                 (__psunsigned_t)context->dp,
2272                 (__psunsigned_t)context->cursor->hashval,
2273                 (__psunsigned_t)context->cursor->blkno,
2274                 (__psunsigned_t)context->cursor->offset,
2275                 (__psunsigned_t)context->alist,
2276                 (__psunsigned_t)context->bufsize,
2277                 (__psunsigned_t)context->count,
2278                 (__psunsigned_t)context->firstu,
2279                 (__psunsigned_t)
2280                         ((context->count > 0) &&
2281                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2282                                 ? (ATTR_ENTRY(context->alist,
2283                                               context->count-1)->a_valuelen)
2284                                 : 0,
2285                 (__psunsigned_t)context->dupcnt,
2286                 (__psunsigned_t)context->flags,
2287                 (__psunsigned_t)be16_to_cpu(leaf->hdr.count),
2288                 (__psunsigned_t)be32_to_cpu(leaf->entries[0].hashval),
2289                 (__psunsigned_t)be32_to_cpu(leaf->entries[
2290                                 be16_to_cpu(leaf->hdr.count)-1].hashval));
2291 }
2292
2293 /*
2294  * Add a trace buffer entry for the arguments given to the routine,
2295  * generic form.
2296  */
2297 void
2298 xfs_attr_trace_enter(int type, char *where,
2299                          __psunsigned_t a2, __psunsigned_t a3,
2300                          __psunsigned_t a4, __psunsigned_t a5,
2301                          __psunsigned_t a6, __psunsigned_t a7,
2302                          __psunsigned_t a8, __psunsigned_t a9,
2303                          __psunsigned_t a10, __psunsigned_t a11,
2304                          __psunsigned_t a12, __psunsigned_t a13,
2305                          __psunsigned_t a14, __psunsigned_t a15)
2306 {
2307         ASSERT(xfs_attr_trace_buf);
2308         ktrace_enter(xfs_attr_trace_buf, (void *)((__psunsigned_t)type),
2309                                          (void *)where,
2310                                          (void *)a2,  (void *)a3,  (void *)a4,
2311                                          (void *)a5,  (void *)a6,  (void *)a7,
2312                                          (void *)a8,  (void *)a9,  (void *)a10,
2313                                          (void *)a11, (void *)a12, (void *)a13,
2314                                          (void *)a14, (void *)a15);
2315 }
2316 #endif  /* XFS_ATTR_TRACE */
2317
2318
2319 /*========================================================================
2320  * System (pseudo) namespace attribute interface routines.
2321  *========================================================================*/
2322
2323 STATIC int
2324 posix_acl_access_set(
2325         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2326 {
2327         return xfs_acl_vset(vp, data, size, _ACL_TYPE_ACCESS);
2328 }
2329
2330 STATIC int
2331 posix_acl_access_remove(
2332         bhv_vnode_t *vp, char *name, int xflags)
2333 {
2334         return xfs_acl_vremove(vp, _ACL_TYPE_ACCESS);
2335 }
2336
2337 STATIC int
2338 posix_acl_access_get(
2339         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2340 {
2341         return xfs_acl_vget(vp, data, size, _ACL_TYPE_ACCESS);
2342 }
2343
2344 STATIC int
2345 posix_acl_access_exists(
2346         bhv_vnode_t *vp)
2347 {
2348         return xfs_acl_vhasacl_access(vp);
2349 }
2350
2351 STATIC int
2352 posix_acl_default_set(
2353         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2354 {
2355         return xfs_acl_vset(vp, data, size, _ACL_TYPE_DEFAULT);
2356 }
2357
2358 STATIC int
2359 posix_acl_default_get(
2360         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2361 {
2362         return xfs_acl_vget(vp, data, size, _ACL_TYPE_DEFAULT);
2363 }
2364
2365 STATIC int
2366 posix_acl_default_remove(
2367         bhv_vnode_t *vp, char *name, int xflags)
2368 {
2369         return xfs_acl_vremove(vp, _ACL_TYPE_DEFAULT);
2370 }
2371
2372 STATIC int
2373 posix_acl_default_exists(
2374         bhv_vnode_t *vp)
2375 {
2376         return xfs_acl_vhasacl_default(vp);
2377 }
2378
2379 STATIC struct attrnames posix_acl_access = {
2380         .attr_name      = "posix_acl_access",
2381         .attr_namelen   = sizeof("posix_acl_access") - 1,
2382         .attr_get       = posix_acl_access_get,
2383         .attr_set       = posix_acl_access_set,
2384         .attr_remove    = posix_acl_access_remove,
2385         .attr_exists    = posix_acl_access_exists,
2386 };
2387
2388 STATIC struct attrnames posix_acl_default = {
2389         .attr_name      = "posix_acl_default",
2390         .attr_namelen   = sizeof("posix_acl_default") - 1,
2391         .attr_get       = posix_acl_default_get,
2392         .attr_set       = posix_acl_default_set,
2393         .attr_remove    = posix_acl_default_remove,
2394         .attr_exists    = posix_acl_default_exists,
2395 };
2396
2397 STATIC struct attrnames *attr_system_names[] =
2398         { &posix_acl_access, &posix_acl_default };
2399
2400
2401 /*========================================================================
2402  * Namespace-prefix-style attribute name interface routines.
2403  *========================================================================*/
2404
2405 STATIC int
2406 attr_generic_set(
2407         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2408 {
2409         return -bhv_vop_attr_set(vp, name, data, size, xflags, NULL);
2410 }
2411
2412 STATIC int
2413 attr_generic_get(
2414         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2415 {
2416         int     error, asize = size;
2417
2418         error = bhv_vop_attr_get(vp, name, data, &asize, xflags, NULL);
2419         if (!error)
2420                 return asize;
2421         return -error;
2422 }
2423
2424 STATIC int
2425 attr_generic_remove(
2426         bhv_vnode_t *vp, char *name, int xflags)
2427 {
2428         return -bhv_vop_attr_remove(vp, name, xflags, NULL);
2429 }
2430
2431 STATIC int
2432 attr_generic_listadd(
2433         attrnames_t             *prefix,
2434         attrnames_t             *namesp,
2435         void                    *data,
2436         size_t                  size,
2437         ssize_t                 *result)
2438 {
2439         char                    *p = data + *result;
2440
2441         *result += prefix->attr_namelen;
2442         *result += namesp->attr_namelen + 1;
2443         if (!size)
2444                 return 0;
2445         if (*result > size)
2446                 return -ERANGE;
2447         strcpy(p, prefix->attr_name);
2448         p += prefix->attr_namelen;
2449         strcpy(p, namesp->attr_name);
2450         p += namesp->attr_namelen + 1;
2451         return 0;
2452 }
2453
2454 STATIC int
2455 attr_system_list(
2456         bhv_vnode_t             *vp,
2457         void                    *data,
2458         size_t                  size,
2459         ssize_t                 *result)
2460 {
2461         attrnames_t             *namesp;
2462         int                     i, error = 0;
2463
2464         for (i = 0; i < ATTR_SYSCOUNT; i++) {
2465                 namesp = attr_system_names[i];
2466                 if (!namesp->attr_exists || !namesp->attr_exists(vp))
2467                         continue;
2468                 error = attr_generic_listadd(&attr_system, namesp,
2469                                                 data, size, result);
2470                 if (error)
2471                         break;
2472         }
2473         return error;
2474 }
2475
2476 int
2477 attr_generic_list(
2478         bhv_vnode_t *vp, void *data, size_t size, int xflags, ssize_t *result)
2479 {
2480         attrlist_cursor_kern_t  cursor = { 0 };
2481         int                     error;
2482
2483         error = bhv_vop_attr_list(vp, data, size, xflags, &cursor, NULL);
2484         if (error > 0)
2485                 return -error;
2486         *result = -error;
2487         return attr_system_list(vp, data, size, result);
2488 }
2489
2490 attrnames_t *
2491 attr_lookup_namespace(
2492         char                    *name,
2493         struct attrnames        **names,
2494         int                     nnames)
2495 {
2496         int                     i;
2497
2498         for (i = 0; i < nnames; i++)
2499                 if (!strncmp(name, names[i]->attr_name, names[i]->attr_namelen))
2500                         return names[i];
2501         return NULL;
2502 }
2503
2504 /*
2505  * Some checks to prevent people abusing EAs to get over quota:
2506  * - Don't allow modifying user EAs on devices/symlinks;
2507  * - Don't allow modifying user EAs if sticky bit set;
2508  */
2509 STATIC int
2510 attr_user_capable(
2511         bhv_vnode_t     *vp,
2512         cred_t          *cred)
2513 {
2514         struct inode    *inode = vn_to_inode(vp);
2515
2516         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
2517                 return -EPERM;
2518         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode) &&
2519             !capable(CAP_SYS_ADMIN))
2520                 return -EPERM;
2521         if (S_ISDIR(inode->i_mode) && (inode->i_mode & S_ISVTX) &&
2522             (current_fsuid(cred) != inode->i_uid) && !capable(CAP_FOWNER))
2523                 return -EPERM;
2524         return 0;
2525 }
2526
2527 STATIC int
2528 attr_trusted_capable(
2529         bhv_vnode_t     *vp,
2530         cred_t          *cred)
2531 {
2532         struct inode    *inode = vn_to_inode(vp);
2533
2534         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
2535                 return -EPERM;
2536         if (!capable(CAP_SYS_ADMIN))
2537                 return -EPERM;
2538         return 0;
2539 }
2540
2541 STATIC int
2542 attr_secure_capable(
2543         bhv_vnode_t     *vp,
2544         cred_t          *cred)
2545 {
2546         return -ENOSECURITY;
2547 }
2548
2549 STATIC int
2550 attr_system_set(
2551         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2552 {
2553         attrnames_t     *namesp;
2554         int             error;
2555
2556         if (xflags & ATTR_CREATE)
2557                 return -EINVAL;
2558
2559         namesp = attr_lookup_namespace(name, attr_system_names, ATTR_SYSCOUNT);
2560         if (!namesp)
2561                 return -EOPNOTSUPP;
2562         error = namesp->attr_set(vp, name, data, size, xflags);
2563         if (!error)
2564                 error = vn_revalidate(vp);
2565         return error;
2566 }
2567
2568 STATIC int
2569 attr_system_get(
2570         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2571 {
2572         attrnames_t     *namesp;
2573
2574         namesp = attr_lookup_namespace(name, attr_system_names, ATTR_SYSCOUNT);
2575         if (!namesp)
2576                 return -EOPNOTSUPP;
2577         return namesp->attr_get(vp, name, data, size, xflags);
2578 }
2579
2580 STATIC int
2581 attr_system_remove(
2582         bhv_vnode_t *vp, char *name, int xflags)
2583 {
2584         attrnames_t     *namesp;
2585
2586         namesp = attr_lookup_namespace(name, attr_system_names, ATTR_SYSCOUNT);
2587         if (!namesp)
2588                 return -EOPNOTSUPP;
2589         return namesp->attr_remove(vp, name, xflags);
2590 }
2591
2592 struct attrnames attr_system = {
2593         .attr_name      = "system.",
2594         .attr_namelen   = sizeof("system.") - 1,
2595         .attr_flag      = ATTR_SYSTEM,
2596         .attr_get       = attr_system_get,
2597         .attr_set       = attr_system_set,
2598         .attr_remove    = attr_system_remove,
2599         .attr_capable   = (attrcapable_t)fs_noerr,
2600 };
2601
2602 struct attrnames attr_trusted = {
2603         .attr_name      = "trusted.",
2604         .attr_namelen   = sizeof("trusted.") - 1,
2605         .attr_flag      = ATTR_ROOT,
2606         .attr_get       = attr_generic_get,
2607         .attr_set       = attr_generic_set,
2608         .attr_remove    = attr_generic_remove,
2609         .attr_capable   = attr_trusted_capable,
2610 };
2611
2612 struct attrnames attr_secure = {
2613         .attr_name      = "security.",
2614         .attr_namelen   = sizeof("security.") - 1,
2615         .attr_flag      = ATTR_SECURE,
2616         .attr_get       = attr_generic_get,
2617         .attr_set       = attr_generic_set,
2618         .attr_remove    = attr_generic_remove,
2619         .attr_capable   = attr_secure_capable,
2620 };
2621
2622 struct attrnames attr_user = {
2623         .attr_name      = "user.",
2624         .attr_namelen   = sizeof("user.") - 1,
2625         .attr_get       = attr_generic_get,
2626         .attr_set       = attr_generic_set,
2627         .attr_remove    = attr_generic_remove,
2628         .attr_capable   = attr_user_capable,
2629 };
2630
2631 struct attrnames *attr_namespaces[] =
2632         { &attr_system, &attr_trusted, &attr_secure, &attr_user };