git.oblomov.eu Git - linux-2.6/blob - 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 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_dir.h"
  28 #include "xfs_dir2.h"
  29 #include "xfs_dmapi.h"
  30 #include "xfs_mount.h"
  31 #include "xfs_da_btree.h"
  32 #include "xfs_bmap_btree.h"
  33 #include "xfs_alloc_btree.h"
  34 #include "xfs_ialloc_btree.h"
  35 #include "xfs_dir_sf.h"
  36 #include "xfs_dir2_sf.h"
  37 #include "xfs_attr_sf.h"
  38 #include "xfs_dinode.h"
  39 #include "xfs_inode.h"
  40 #include "xfs_alloc.h"
  41 #include "xfs_btree.h"
  42 #include "xfs_inode_item.h"
  43 #include "xfs_bmap.h"
  44 #include "xfs_attr.h"
  45 #include "xfs_attr_leaf.h"
  46 #include "xfs_error.h"
  47 #include "xfs_quota.h"
  48 #include "xfs_trans_space.h"
  49 #include "xfs_acl.h"
  50 #include "xfs_rw.h"
  51
  52 /*
  53  * xfs_attr.c
  54  *
  55  * Provide the external interfaces to manage attribute lists.
  56  */
  57
  58 #define ATTR_SYSCOUNT   2
  59 STATIC struct attrnames posix_acl_access;
  60 STATIC struct attrnames posix_acl_default;
  61 STATIC struct attrnames *attr_system_names[ATTR_SYSCOUNT];
  62
  63 /*========================================================================
  64  * Function prototypes for the kernel.
  65  *========================================================================*/
  66
  67 /*
  68  * Internal routines when attribute list fits inside the inode.
  69  */
  70 STATIC int xfs_attr_shortform_addname(xfs_da_args_t *args);
  71
  72 /*
  73  * Internal routines when attribute list is one block.
  74  */
  75 STATIC int xfs_attr_leaf_get(xfs_da_args_t *args);
  76 STATIC int xfs_attr_leaf_addname(xfs_da_args_t *args);
  77 STATIC int xfs_attr_leaf_removename(xfs_da_args_t *args);
  78 STATIC int xfs_attr_leaf_list(xfs_attr_list_context_t *context);
  79
  80 /*
  81  * Internal routines when attribute list is more than one block.
  82  */
  83 STATIC int xfs_attr_node_get(xfs_da_args_t *args);
  84 STATIC int xfs_attr_node_addname(xfs_da_args_t *args);
  85 STATIC int xfs_attr_node_removename(xfs_da_args_t *args);
  86 STATIC int xfs_attr_node_list(xfs_attr_list_context_t *context);
  87 STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
  88 STATIC int xfs_attr_refillstate(xfs_da_state_t *state);
  89
  90 /*
  91  * Routines to manipulate out-of-line attribute values.
  92  */
  93 STATIC int xfs_attr_rmtval_get(xfs_da_args_t *args);
  94 STATIC int xfs_attr_rmtval_set(xfs_da_args_t *args);
  95 STATIC int xfs_attr_rmtval_remove(xfs_da_args_t *args);
  96
  97 #define ATTR_RMTVALUE_MAPSIZE   1       /* # of map entries at once */
  98
  99 #if defined(XFS_ATTR_TRACE)
 100 ktrace_t *xfs_attr_trace_buf;
 101 #endif
 102
 103
 104 /*========================================================================
 105  * Overall external interface routines.
 106  *========================================================================*/
 107
 108 int
 109 xfs_attr_fetch(xfs_inode_t *ip, const char *name, int namelen,
 110                char *value, int *valuelenp, int flags, struct cred *cred)
 111 {
 112         xfs_da_args_t   args;
 113         int             error;
 114
 115         if ((XFS_IFORK_Q(ip) == 0) ||
 116             (ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 117              ip->i_d.di_anextents == 0))
 118                 return(ENOATTR);
 119
 120         if (!(flags & (ATTR_KERNACCESS|ATTR_SECURE))) {
 121                 if ((error = xfs_iaccess(ip, S_IRUSR, cred)))
 122                         return(XFS_ERROR(error));
 123         }
 124
 125         /*
 126          * Fill in the arg structure for this request.
 127          */
 128         memset((char *)&args, 0, sizeof(args));
 129         args.name = name;
 130         args.namelen = namelen;
 131         args.value = value;
 132         args.valuelen = *valuelenp;
 133         args.flags = flags;
 134         args.hashval = xfs_da_hashname(args.name, args.namelen);
 135         args.dp = ip;
 136         args.whichfork = XFS_ATTR_FORK;
 137
 138         /*
 139          * Decide on what work routines to call based on the inode size.
 140          */
 141         if (XFS_IFORK_Q(ip) == 0 ||
 142             (ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 143              ip->i_d.di_anextents == 0)) {
 144                 error = XFS_ERROR(ENOATTR);
 145         } else if (ip->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
 146                 error = xfs_attr_shortform_getvalue(&args);
 147         } else if (xfs_bmap_one_block(ip, XFS_ATTR_FORK)) {
 148                 error = xfs_attr_leaf_get(&args);
 149         } else {
 150                 error = xfs_attr_node_get(&args);
 151         }
 152
 153         /*
 154          * Return the number of bytes in the value to the caller.
 155          */
 156         *valuelenp = args.valuelen;
 157
 158         if (error == EEXIST)
 159                 error = 0;
 160         return(error);
 161 }
 162
 163 int
 164 xfs_attr_get(bhv_desc_t *bdp, const char *name, char *value, int *valuelenp,
 165              int flags, struct cred *cred)
 166 {
 167         xfs_inode_t     *ip = XFS_BHVTOI(bdp);
 168         int             error, namelen;
 169
 170         XFS_STATS_INC(xs_attr_get);
 171
 172         if (!name)
 173                 return(EINVAL);
 174         namelen = strlen(name);
 175         if (namelen >= MAXNAMELEN)
 176                 return(EFAULT);         /* match IRIX behaviour */
 177
 178         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 179                 return(EIO);
 180
 181         xfs_ilock(ip, XFS_ILOCK_SHARED);
 182         error = xfs_attr_fetch(ip, name, namelen, value, valuelenp, flags, cred);
 183         xfs_iunlock(ip, XFS_ILOCK_SHARED);
 184         return(error);
 185 }
 186
 187 STATIC int
 188 xfs_attr_set_int(xfs_inode_t *dp, const char *name, int namelen,
 189                  char *value, int valuelen, int flags)
 190 {
 191         xfs_da_args_t   args;
 192         xfs_fsblock_t   firstblock;
 193         xfs_bmap_free_t flist;
 194         int             error, err2, committed;
 195         int             local, size;
 196         uint            nblks;
 197         xfs_mount_t     *mp = dp->i_mount;
 198         int             rsvd = (flags & ATTR_ROOT) != 0;
 199
 200         /*
 201          * Attach the dquots to the inode.
 202          */
 203         if ((error = XFS_QM_DQATTACH(mp, dp, 0)))
 204                 return (error);
 205
 206         /*
 207          * Determine space new attribute will use, and if it would be
 208          * "local" or "remote" (note: local != inline).
 209          */
 210         size = xfs_attr_leaf_newentsize(namelen, valuelen,
 211                                         mp->m_sb.sb_blocksize, &local);
 212
 213         /*
 214          * If the inode doesn't have an attribute fork, add one.
 215          * (inode must not be locked when we call this routine)
 216          */
 217         if (XFS_IFORK_Q(dp) == 0) {
 218                 if ((error = xfs_bmap_add_attrfork(dp, size, rsvd)))
 219                         return(error);
 220         }
 221
 222         /*
 223          * Fill in the arg structure for this request.
 224          */
 225         memset((char *)&args, 0, sizeof(args));
 226         args.name = name;
 227         args.namelen = namelen;
 228         args.value = value;
 229         args.valuelen = valuelen;
 230         args.flags = flags;
 231         args.hashval = xfs_da_hashname(args.name, args.namelen);
 232         args.dp = dp;
 233         args.firstblock = &firstblock;
 234         args.flist = &flist;
 235         args.whichfork = XFS_ATTR_FORK;
 236         args.addname = 1;
 237         args.oknoent = 1;
 238
 239         nblks = XFS_DAENTER_SPACE_RES(mp, XFS_ATTR_FORK);
 240         if (local) {
 241                 if (size > (mp->m_sb.sb_blocksize >> 1)) {
 242                         /* Double split possible */
 243                         nblks <<= 1;
 244                 }
 245         } else {
 246                 uint    dblocks = XFS_B_TO_FSB(mp, valuelen);
 247                 /* Out of line attribute, cannot double split, but make
 248                  * room for the attribute value itself.
 249                  */
 250                 nblks += dblocks;
 251                 nblks += XFS_NEXTENTADD_SPACE_RES(mp, dblocks, XFS_ATTR_FORK);
 252         }
 253
 254         /* Size is now blocks for attribute data */
 255         args.total = nblks;
 256
 257         /*
 258          * Start our first transaction of the day.
 259          *
 260          * All future transactions during this code must be "chained" off
 261          * this one via the trans_dup() call.  All transactions will contain
 262          * the inode, and the inode will always be marked with trans_ihold().
 263          * Since the inode will be locked in all transactions, we must log
 264          * the inode in every transaction to let it float upward through
 265          * the log.
 266          */
 267         args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_SET);
 268
 269         /*
 270          * Root fork attributes can use reserved data blocks for this
 271          * operation if necessary
 272          */
 273
 274         if (rsvd)
 275                 args.trans->t_flags |= XFS_TRANS_RESERVE;
 276
 277         if ((error = xfs_trans_reserve(args.trans, (uint) nblks,
 278                                       XFS_ATTRSET_LOG_RES(mp, nblks),
 279                                       0, XFS_TRANS_PERM_LOG_RES,
 280                                       XFS_ATTRSET_LOG_COUNT))) {
 281                 xfs_trans_cancel(args.trans, 0);
 282                 return(error);
 283         }
 284         xfs_ilock(dp, XFS_ILOCK_EXCL);
 285
 286         error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, args.trans, dp, nblks, 0,
 287                          rsvd ? XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
 288                                 XFS_QMOPT_RES_REGBLKS);
 289         if (error) {
 290                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
 291                 xfs_trans_cancel(args.trans, XFS_TRANS_RELEASE_LOG_RES);
 292                 return (error);
 293         }
 294
 295         xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
 296         xfs_trans_ihold(args.trans, dp);
 297
 298         /*
 299          * If the attribute list is non-existant or a shortform list,
 300          * upgrade it to a single-leaf-block attribute list.
 301          */
 302         if ((dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
 303             ((dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) &&
 304              (dp->i_d.di_anextents == 0))) {
 305
 306                 /*
 307                  * Build initial attribute list (if required).
 308                  */
 309                 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS)
 310                         xfs_attr_shortform_create(&args);
 311
 312                 /*
 313                  * Try to add the attr to the attribute list in
 314                  * the inode.
 315                  */
 316                 error = xfs_attr_shortform_addname(&args);
 317                 if (error != ENOSPC) {
 318                         /*
 319                          * Commit the shortform mods, and we're done.
 320                          * NOTE: this is also the error path (EEXIST, etc).
 321                          */
 322                         ASSERT(args.trans != NULL);
 323
 324                         /*
 325                          * If this is a synchronous mount, make sure that
 326                          * the transaction goes to disk before returning
 327                          * to the user.
 328                          */
 329                         if (mp->m_flags & XFS_MOUNT_WSYNC) {
 330                                 xfs_trans_set_sync(args.trans);
 331                         }
 332                         err2 = xfs_trans_commit(args.trans,
 333                                                  XFS_TRANS_RELEASE_LOG_RES,
 334                                                  NULL);
 335                         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 336
 337                         /*
 338                          * Hit the inode change time.
 339                          */
 340                         if (!error && (flags & ATTR_KERNOTIME) == 0) {
 341                                 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
 342                         }
 343                         return(error == 0 ? err2 : error);
 344                 }
 345
 346                 /*
 347                  * It won't fit in the shortform, transform to a leaf block.
 348                  * GROT: another possible req'mt for a double-split btree op.
 349                  */
 350                 XFS_BMAP_INIT(args.flist, args.firstblock);
 351                 error = xfs_attr_shortform_to_leaf(&args);
 352                 if (!error) {
 353                         error = xfs_bmap_finish(&args.trans, args.flist,
 354                                                 *args.firstblock, &committed);
 355                 }
 356                 if (error) {
 357                         ASSERT(committed);
 358                         args.trans = NULL;
 359                         xfs_bmap_cancel(&flist);
 360                         goto out;
 361                 }
 362
 363                 /*
 364                  * bmap_finish() may have committed the last trans and started
 365                  * a new one.  We need the inode to be in all transactions.
 366                  */
 367                 if (committed) {
 368                         xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
 369                         xfs_trans_ihold(args.trans, dp);
 370                 }
 371
 372                 /*
 373                  * Commit the leaf transformation.  We'll need another (linked)
 374                  * transaction to add the new attribute to the leaf.
 375                  */
 376                 if ((error = xfs_attr_rolltrans(&args.trans, dp)))
 377                         goto out;
 378
 379         }
 380
 381         if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
 382                 error = xfs_attr_leaf_addname(&args);
 383         } else {
 384                 error = xfs_attr_node_addname(&args);
 385         }
 386         if (error) {
 387                 goto out;
 388         }
 389
 390         /*
 391          * If this is a synchronous mount, make sure that the
 392          * transaction goes to disk before returning to the user.
 393          */
 394         if (mp->m_flags & XFS_MOUNT_WSYNC) {
 395                 xfs_trans_set_sync(args.trans);
 396         }
 397
 398         /*
 399          * Commit the last in the sequence of transactions.
 400          */
 401         xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
 402         error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES,
 403                                  NULL);
 404         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 405
 406         /*
 407          * Hit the inode change time.
 408          */
 409         if (!error && (flags & ATTR_KERNOTIME) == 0) {
 410                 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
 411         }
 412
 413         return(error);
 414
 415 out:
 416         if (args.trans)
 417                 xfs_trans_cancel(args.trans,
 418                         XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
 419         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 420         return(error);
 421 }
 422
 423 int
 424 xfs_attr_set(bhv_desc_t *bdp, const char *name, char *value, int valuelen, int flags,
 425              struct cred *cred)
 426 {
 427         xfs_inode_t     *dp;
 428         int             namelen, error;
 429
 430         namelen = strlen(name);
 431         if (namelen >= MAXNAMELEN)
 432                 return EFAULT;          /* match IRIX behaviour */
 433
 434         XFS_STATS_INC(xs_attr_set);
 435
 436         dp = XFS_BHVTOI(bdp);
 437         if (XFS_FORCED_SHUTDOWN(dp->i_mount))
 438                 return (EIO);
 439
 440         xfs_ilock(dp, XFS_ILOCK_SHARED);
 441         if (!(flags & ATTR_SECURE) &&
 442              (error = xfs_iaccess(dp, S_IWUSR, cred))) {
 443                 xfs_iunlock(dp, XFS_ILOCK_SHARED);
 444                 return(XFS_ERROR(error));
 445         }
 446         xfs_iunlock(dp, XFS_ILOCK_SHARED);
 447
 448         return xfs_attr_set_int(dp, name, namelen, value, valuelen, flags);
 449 }
 450
 451 /*
 452  * Generic handler routine to remove a name from an attribute list.
 453  * Transitions attribute list from Btree to shortform as necessary.
 454  */
 455 STATIC int
 456 xfs_attr_remove_int(xfs_inode_t *dp, const char *name, int namelen, int flags)
 457 {
 458         xfs_da_args_t   args;
 459         xfs_fsblock_t   firstblock;
 460         xfs_bmap_free_t flist;
 461         int             error;
 462         xfs_mount_t     *mp = dp->i_mount;
 463
 464         /*
 465          * Fill in the arg structure for this request.
 466          */
 467         memset((char *)&args, 0, sizeof(args));
 468         args.name = name;
 469         args.namelen = namelen;
 470         args.flags = flags;
 471         args.hashval = xfs_da_hashname(args.name, args.namelen);
 472         args.dp = dp;
 473         args.firstblock = &firstblock;
 474         args.flist = &flist;
 475         args.total = 0;
 476         args.whichfork = XFS_ATTR_FORK;
 477
 478         /*
 479          * Attach the dquots to the inode.
 480          */
 481         if ((error = XFS_QM_DQATTACH(mp, dp, 0)))
 482                 return (error);
 483
 484         /*
 485          * Start our first transaction of the day.
 486          *
 487          * All future transactions during this code must be "chained" off
 488          * this one via the trans_dup() call.  All transactions will contain
 489          * the inode, and the inode will always be marked with trans_ihold().
 490          * Since the inode will be locked in all transactions, we must log
 491          * the inode in every transaction to let it float upward through
 492          * the log.
 493          */
 494         args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_RM);
 495
 496         /*
 497          * Root fork attributes can use reserved data blocks for this
 498          * operation if necessary
 499          */
 500
 501         if (flags & ATTR_ROOT)
 502                 args.trans->t_flags |= XFS_TRANS_RESERVE;
 503
 504         if ((error = xfs_trans_reserve(args.trans,
 505                                       XFS_ATTRRM_SPACE_RES(mp),
 506                                       XFS_ATTRRM_LOG_RES(mp),
 507                                       0, XFS_TRANS_PERM_LOG_RES,
 508                                       XFS_ATTRRM_LOG_COUNT))) {
 509                 xfs_trans_cancel(args.trans, 0);
 510                 return(error);
 511         }
 512
 513         xfs_ilock(dp, XFS_ILOCK_EXCL);
 514         /*
 515          * No need to make quota reservations here. We expect to release some
 516          * blocks not allocate in the common case.
 517          */
 518         xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
 519         xfs_trans_ihold(args.trans, dp);
 520
 521         /*
 522          * Decide on what work routines to call based on the inode size.
 523          */
 524         if (XFS_IFORK_Q(dp) == 0 ||
 525             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 526              dp->i_d.di_anextents == 0)) {
 527                 error = XFS_ERROR(ENOATTR);
 528                 goto out;
 529         }
 530         if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
 531                 ASSERT(dp->i_afp->if_flags & XFS_IFINLINE);
 532                 error = xfs_attr_shortform_remove(&args);
 533                 if (error) {
 534                         goto out;
 535                 }
 536         } else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
 537                 error = xfs_attr_leaf_removename(&args);
 538         } else {
 539                 error = xfs_attr_node_removename(&args);
 540         }
 541         if (error) {
 542                 goto out;
 543         }
 544
 545         /*
 546          * If this is a synchronous mount, make sure that the
 547          * transaction goes to disk before returning to the user.
 548          */
 549         if (mp->m_flags & XFS_MOUNT_WSYNC) {
 550                 xfs_trans_set_sync(args.trans);
 551         }
 552
 553         /*
 554          * Commit the last in the sequence of transactions.
 555          */
 556         xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
 557         error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES,
 558                                  NULL);
 559         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 560
 561         /*
 562          * Hit the inode change time.
 563          */
 564         if (!error && (flags & ATTR_KERNOTIME) == 0) {
 565                 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
 566         }
 567
 568         return(error);
 569
 570 out:
 571         if (args.trans)
 572                 xfs_trans_cancel(args.trans,
 573                         XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
 574         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 575         return(error);
 576 }
 577
 578 int
 579 xfs_attr_remove(bhv_desc_t *bdp, const char *name, int flags, struct cred *cred)
 580 {
 581         xfs_inode_t         *dp;
 582         int                 namelen, error;
 583
 584         namelen = strlen(name);
 585         if (namelen >= MAXNAMELEN)
 586                 return EFAULT;          /* match IRIX behaviour */
 587
 588         XFS_STATS_INC(xs_attr_remove);
 589
 590         dp = XFS_BHVTOI(bdp);
 591         if (XFS_FORCED_SHUTDOWN(dp->i_mount))
 592                 return (EIO);
 593
 594         xfs_ilock(dp, XFS_ILOCK_SHARED);
 595         if (!(flags & ATTR_SECURE) &&
 596              (error = xfs_iaccess(dp, S_IWUSR, cred))) {
 597                 xfs_iunlock(dp, XFS_ILOCK_SHARED);
 598                 return(XFS_ERROR(error));
 599         } else if (XFS_IFORK_Q(dp) == 0 ||
 600                    (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 601                     dp->i_d.di_anextents == 0)) {
 602                 xfs_iunlock(dp, XFS_ILOCK_SHARED);
 603                 return(XFS_ERROR(ENOATTR));
 604         }
 605         xfs_iunlock(dp, XFS_ILOCK_SHARED);
 606
 607         return xfs_attr_remove_int(dp, name, namelen, flags);
 608 }
 609
 610 /*
 611  * Generate a list of extended attribute names and optionally
 612  * also value lengths.  Positive return value follows the XFS
 613  * convention of being an error, zero or negative return code
 614  * is the length of the buffer returned (negated), indicating
 615  * success.
 616  */
 617 int
 618 xfs_attr_list(bhv_desc_t *bdp, char *buffer, int bufsize, int flags,
 619                       attrlist_cursor_kern_t *cursor, struct cred *cred)
 620 {
 621         xfs_attr_list_context_t context;
 622         xfs_inode_t *dp;
 623         int error;
 624
 625         XFS_STATS_INC(xs_attr_list);
 626
 627         /*
 628          * Validate the cursor.
 629          */
 630         if (cursor->pad1 || cursor->pad2)
 631                 return(XFS_ERROR(EINVAL));
 632         if ((cursor->initted == 0) &&
 633             (cursor->hashval || cursor->blkno || cursor->offset))
 634                 return(XFS_ERROR(EINVAL));
 635
 636         /*
 637          * Check for a properly aligned buffer.
 638          */
 639         if (((long)buffer) & (sizeof(int)-1))
 640                 return(XFS_ERROR(EFAULT));
 641         if (flags & ATTR_KERNOVAL)
 642                 bufsize = 0;
 643
 644         /*
 645          * Initialize the output buffer.
 646          */
 647         context.dp = dp = XFS_BHVTOI(bdp);
 648         context.cursor = cursor;
 649         context.count = 0;
 650         context.dupcnt = 0;
 651         context.resynch = 1;
 652         context.flags = flags;
 653         if (!(flags & ATTR_KERNAMELS)) {
 654                 context.bufsize = (bufsize & ~(sizeof(int)-1));  /* align */
 655                 context.firstu = context.bufsize;
 656                 context.alist = (attrlist_t *)buffer;
 657                 context.alist->al_count = 0;
 658                 context.alist->al_more = 0;
 659                 context.alist->al_offset[0] = context.bufsize;
 660         }
 661         else {
 662                 context.bufsize = bufsize;
 663                 context.firstu = context.bufsize;
 664                 context.alist = (attrlist_t *)buffer;
 665         }
 666
 667         if (XFS_FORCED_SHUTDOWN(dp->i_mount))
 668                 return (EIO);
 669
 670         xfs_ilock(dp, XFS_ILOCK_SHARED);
 671         if (!(flags & ATTR_SECURE) &&
 672              (error = xfs_iaccess(dp, S_IRUSR, cred))) {
 673                 xfs_iunlock(dp, XFS_ILOCK_SHARED);
 674                 return(XFS_ERROR(error));
 675         }
 676
 677         /*
 678          * Decide on what work routines to call based on the inode size.
 679          */
 680         xfs_attr_trace_l_c("syscall start", &context);
 681         if (XFS_IFORK_Q(dp) == 0 ||
 682             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 683              dp->i_d.di_anextents == 0)) {
 684                 error = 0;
 685         } else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
 686                 error = xfs_attr_shortform_list(&context);
 687         } else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
 688                 error = xfs_attr_leaf_list(&context);
 689         } else {
 690                 error = xfs_attr_node_list(&context);
 691         }
 692         xfs_iunlock(dp, XFS_ILOCK_SHARED);
 693         xfs_attr_trace_l_c("syscall end", &context);
 694
 695         if (!(context.flags & (ATTR_KERNOVAL|ATTR_KERNAMELS))) {
 696                 ASSERT(error >= 0);
 697         }
 698         else {  /* must return negated buffer size or the error */
 699                 if (context.count < 0)
 700                         error = XFS_ERROR(ERANGE);
 701                 else
 702                         error = -context.count;
 703         }
 704
 705         return(error);
 706 }
 707
 708 int                                                             /* error */
 709 xfs_attr_inactive(xfs_inode_t *dp)
 710 {
 711         xfs_trans_t *trans;
 712         xfs_mount_t *mp;
 713         int error;
 714
 715         mp = dp->i_mount;
 716         ASSERT(! XFS_NOT_DQATTACHED(mp, dp));
 717
 718         xfs_ilock(dp, XFS_ILOCK_SHARED);
 719         if ((XFS_IFORK_Q(dp) == 0) ||
 720             (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
 721             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 722              dp->i_d.di_anextents == 0)) {
 723                 xfs_iunlock(dp, XFS_ILOCK_SHARED);
 724                 return(0);
 725         }
 726         xfs_iunlock(dp, XFS_ILOCK_SHARED);
 727
 728         /*
 729          * Start our first transaction of the day.
 730          *
 731          * All future transactions during this code must be "chained" off
 732          * this one via the trans_dup() call.  All transactions will contain
 733          * the inode, and the inode will always be marked with trans_ihold().
 734          * Since the inode will be locked in all transactions, we must log
 735          * the inode in every transaction to let it float upward through
 736          * the log.
 737          */
 738         trans = xfs_trans_alloc(mp, XFS_TRANS_ATTRINVAL);
 739         if ((error = xfs_trans_reserve(trans, 0, XFS_ATTRINVAL_LOG_RES(mp), 0,
 740                                       XFS_TRANS_PERM_LOG_RES,
 741                                       XFS_ATTRINVAL_LOG_COUNT))) {
 742                 xfs_trans_cancel(trans, 0);
 743                 return(error);
 744         }
 745         xfs_ilock(dp, XFS_ILOCK_EXCL);
 746
 747         /*
 748          * No need to make quota reservations here. We expect to release some
 749          * blocks, not allocate, in the common case.
 750          */
 751         xfs_trans_ijoin(trans, dp, XFS_ILOCK_EXCL);
 752         xfs_trans_ihold(trans, dp);
 753
 754         /*
 755          * Decide on what work routines to call based on the inode size.
 756          */
 757         if ((XFS_IFORK_Q(dp) == 0) ||
 758             (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
 759             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 760              dp->i_d.di_anextents == 0)) {
 761                 error = 0;
 762                 goto out;
 763         }
 764         error = xfs_attr_root_inactive(&trans, dp);
 765         if (error)
 766                 goto out;
 767         /*
 768          * signal synchronous inactive transactions unless this
 769          * is a synchronous mount filesystem in which case we
 770          * know that we're here because we've been called out of
 771          * xfs_inactive which means that the last reference is gone
 772          * and the unlink transaction has already hit the disk so
 773          * async inactive transactions are safe.
 774          */
 775         if ((error = xfs_itruncate_finish(&trans, dp, 0LL, XFS_ATTR_FORK,
 776                                 (!(mp->m_flags & XFS_MOUNT_WSYNC)
 777                                  ? 1 : 0))))
 778                 goto out;
 779
 780         /*
 781          * Commit the last in the sequence of transactions.
 782          */
 783         xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
 784         error = xfs_trans_commit(trans, XFS_TRANS_RELEASE_LOG_RES,
 785                                  NULL);
 786         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 787
 788         return(error);
 789
 790 out:
 791         xfs_trans_cancel(trans, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
 792         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 793         return(error);
 794 }
 795
 796
 797
 798 /*========================================================================
 799  * External routines when attribute list is inside the inode
 800  *========================================================================*/
 801
 802 /*
 803  * Add a name to the shortform attribute list structure
 804  * This is the external routine.
 805  */
 806 STATIC int
 807 xfs_attr_shortform_addname(xfs_da_args_t *args)
 808 {
 809         int newsize, forkoff, retval;
 810
 811         retval = xfs_attr_shortform_lookup(args);
 812         if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
 813                 return(retval);
 814         } else if (retval == EEXIST) {
 815                 if (args->flags & ATTR_CREATE)
 816                         return(retval);
 817                 retval = xfs_attr_shortform_remove(args);
 818                 ASSERT(retval == 0);
 819         }
 820
 821         if (args->namelen >= XFS_ATTR_SF_ENTSIZE_MAX ||
 822             args->valuelen >= XFS_ATTR_SF_ENTSIZE_MAX)
 823                 return(XFS_ERROR(ENOSPC));
 824
 825         newsize = XFS_ATTR_SF_TOTSIZE(args->dp);
 826         newsize += XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
 827
 828         forkoff = xfs_attr_shortform_bytesfit(args->dp, newsize);
 829         if (!forkoff)
 830                 return(XFS_ERROR(ENOSPC));
 831
 832         xfs_attr_shortform_add(args, forkoff);
 833         return(0);
 834 }
 835
 836
 837 /*========================================================================
 838  * External routines when attribute list is one block
 839  *========================================================================*/
 840
 841 /*
 842  * Add a name to the leaf attribute list structure
 843  *
 844  * This leaf block cannot have a "remote" value, we only call this routine
 845  * if bmap_one_block() says there is only one block (ie: no remote blks).
 846  */
 847 int
 848 xfs_attr_leaf_addname(xfs_da_args_t *args)
 849 {
 850         xfs_inode_t *dp;
 851         xfs_dabuf_t *bp;
 852         int retval, error, committed, forkoff;
 853
 854         /*
 855          * Read the (only) block in the attribute list in.
 856          */
 857         dp = args->dp;
 858         args->blkno = 0;
 859         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
 860                                              XFS_ATTR_FORK);
 861         if (error)
 862                 return(error);
 863         ASSERT(bp != NULL);
 864
 865         /*
 866          * Look up the given attribute in the leaf block.  Figure out if
 867          * the given flags produce an error or call for an atomic rename.
 868          */
 869         retval = xfs_attr_leaf_lookup_int(bp, args);
 870         if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
 871                 xfs_da_brelse(args->trans, bp);
 872                 return(retval);
 873         } else if (retval == EEXIST) {
 874                 if (args->flags & ATTR_CREATE) {        /* pure create op */
 875                         xfs_da_brelse(args->trans, bp);
 876                         return(retval);
 877                 }
 878                 args->rename = 1;                       /* an atomic rename */
 879                 args->blkno2 = args->blkno;             /* set 2nd entry info*/
 880                 args->index2 = args->index;
 881                 args->rmtblkno2 = args->rmtblkno;
 882                 args->rmtblkcnt2 = args->rmtblkcnt;
 883         }
 884
 885         /*
 886          * Add the attribute to the leaf block, transitioning to a Btree
 887          * if required.
 888          */
 889         retval = xfs_attr_leaf_add(bp, args);
 890         xfs_da_buf_done(bp);
 891         if (retval == ENOSPC) {
 892                 /*
 893                  * Promote the attribute list to the Btree format, then
 894                  * Commit that transaction so that the node_addname() call
 895                  * can manage its own transactions.
 896                  */
 897                 XFS_BMAP_INIT(args->flist, args->firstblock);
 898                 error = xfs_attr_leaf_to_node(args);
 899                 if (!error) {
 900                         error = xfs_bmap_finish(&args->trans, args->flist,
 901                                                 *args->firstblock, &committed);
 902                 }
 903                 if (error) {
 904                         ASSERT(committed);
 905                         args->trans = NULL;
 906                         xfs_bmap_cancel(args->flist);
 907                         return(error);
 908                 }
 909
 910                 /*
 911                  * bmap_finish() may have committed the last trans and started
 912                  * a new one.  We need the inode to be in all transactions.
 913                  */
 914                 if (committed) {
 915                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
 916                         xfs_trans_ihold(args->trans, dp);
 917                 }
 918
 919                 /*
 920                  * Commit the current trans (including the inode) and start
 921                  * a new one.
 922                  */
 923                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
 924                         return (error);
 925
 926                 /*
 927                  * Fob the whole rest of the problem off on the Btree code.
 928                  */
 929                 error = xfs_attr_node_addname(args);
 930                 return(error);
 931         }
 932
 933         /*
 934          * Commit the transaction that added the attr name so that
 935          * later routines can manage their own transactions.
 936          */
 937         if ((error = xfs_attr_rolltrans(&args->trans, dp)))
 938                 return (error);
 939
 940         /*
 941          * If there was an out-of-line value, allocate the blocks we
 942          * identified for its storage and copy the value.  This is done
 943          * after we create the attribute so that we don't overflow the
 944          * maximum size of a transaction and/or hit a deadlock.
 945          */
 946         if (args->rmtblkno > 0) {
 947                 error = xfs_attr_rmtval_set(args);
 948                 if (error)
 949                         return(error);
 950         }
 951
 952         /*
 953          * If this is an atomic rename operation, we must "flip" the
 954          * incomplete flags on the "new" and "old" attribute/value pairs
 955          * so that one disappears and one appears atomically.  Then we
 956          * must remove the "old" attribute/value pair.
 957          */
 958         if (args->rename) {
 959                 /*
 960                  * In a separate transaction, set the incomplete flag on the
 961                  * "old" attr and clear the incomplete flag on the "new" attr.
 962                  */
 963                 error = xfs_attr_leaf_flipflags(args);
 964                 if (error)
 965                         return(error);
 966
 967                 /*
 968                  * Dismantle the "old" attribute/value pair by removing
 969                  * a "remote" value (if it exists).
 970                  */
 971                 args->index = args->index2;
 972                 args->blkno = args->blkno2;
 973                 args->rmtblkno = args->rmtblkno2;
 974                 args->rmtblkcnt = args->rmtblkcnt2;
 975                 if (args->rmtblkno) {
 976                         error = xfs_attr_rmtval_remove(args);
 977                         if (error)
 978                                 return(error);
 979                 }
 980
 981                 /*
 982                  * Read in the block containing the "old" attr, then
 983                  * remove the "old" attr from that block (neat, huh!)
 984                  */
 985                 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1,
 986                                                      &bp, XFS_ATTR_FORK);
 987                 if (error)
 988                         return(error);
 989                 ASSERT(bp != NULL);
 990                 (void)xfs_attr_leaf_remove(bp, args);
 991
 992                 /*
 993                  * If the result is small enough, shrink it all into the inode.
 994                  */
 995                 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
 996                         XFS_BMAP_INIT(args->flist, args->firstblock);
 997                         error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
 998                         /* bp is gone due to xfs_da_shrink_inode */
 999                         if (!error) {
1000                                 error = xfs_bmap_finish(&args->trans,
1001                                                         args->flist,
1002                                                         *args->firstblock,
1003                                                         &committed);
1004                         }
1005                         if (error) {
1006                                 ASSERT(committed);
1007                                 args->trans = NULL;
1008                                 xfs_bmap_cancel(args->flist);
1009                                 return(error);
1010                         }
1011
1012                         /*
1013                          * bmap_finish() may have committed the last trans
1014                          * and started a new one.  We need the inode to be
1015                          * in all transactions.
1016                          */
1017                         if (committed) {
1018                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1019                                 xfs_trans_ihold(args->trans, dp);
1020                         }
1021                 } else
1022                         xfs_da_buf_done(bp);
1023
1024                 /*
1025                  * Commit the remove and start the next trans in series.
1026                  */
1027                 error = xfs_attr_rolltrans(&args->trans, dp);
1028
1029         } else if (args->rmtblkno > 0) {
1030                 /*
1031                  * Added a "remote" value, just clear the incomplete flag.
1032                  */
1033                 error = xfs_attr_leaf_clearflag(args);
1034         }
1035         return(error);
1036 }
1037
1038 /*
1039  * Remove a name from the leaf attribute list structure
1040  *
1041  * This leaf block cannot have a "remote" value, we only call this routine
1042  * if bmap_one_block() says there is only one block (ie: no remote blks).
1043  */
1044 STATIC int
1045 xfs_attr_leaf_removename(xfs_da_args_t *args)
1046 {
1047         xfs_inode_t *dp;
1048         xfs_dabuf_t *bp;
1049         int error, committed, forkoff;
1050
1051         /*
1052          * Remove the attribute.
1053          */
1054         dp = args->dp;
1055         args->blkno = 0;
1056         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1057                                              XFS_ATTR_FORK);
1058         if (error) {
1059                 return(error);
1060         }
1061
1062         ASSERT(bp != NULL);
1063         error = xfs_attr_leaf_lookup_int(bp, args);
1064         if (error == ENOATTR) {
1065                 xfs_da_brelse(args->trans, bp);
1066                 return(error);
1067         }
1068
1069         (void)xfs_attr_leaf_remove(bp, args);
1070
1071         /*
1072          * If the result is small enough, shrink it all into the inode.
1073          */
1074         if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1075                 XFS_BMAP_INIT(args->flist, args->firstblock);
1076                 error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1077                 /* bp is gone due to xfs_da_shrink_inode */
1078                 if (!error) {
1079                         error = xfs_bmap_finish(&args->trans, args->flist,
1080                                                 *args->firstblock, &committed);
1081                 }
1082                 if (error) {
1083                         ASSERT(committed);
1084                         args->trans = NULL;
1085                         xfs_bmap_cancel(args->flist);
1086                         return(error);
1087                 }
1088
1089                 /*
1090                  * bmap_finish() may have committed the last trans and started
1091                  * a new one.  We need the inode to be in all transactions.
1092                  */
1093                 if (committed) {
1094                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1095                         xfs_trans_ihold(args->trans, dp);
1096                 }
1097         } else
1098                 xfs_da_buf_done(bp);
1099         return(0);
1100 }
1101
1102 /*
1103  * Look up a name in a leaf attribute list structure.
1104  *
1105  * This leaf block cannot have a "remote" value, we only call this routine
1106  * if bmap_one_block() says there is only one block (ie: no remote blks).
1107  */
1108 STATIC int
1109 xfs_attr_leaf_get(xfs_da_args_t *args)
1110 {
1111         xfs_dabuf_t *bp;
1112         int error;
1113
1114         args->blkno = 0;
1115         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1116                                              XFS_ATTR_FORK);
1117         if (error)
1118                 return(error);
1119         ASSERT(bp != NULL);
1120
1121         error = xfs_attr_leaf_lookup_int(bp, args);
1122         if (error != EEXIST)  {
1123                 xfs_da_brelse(args->trans, bp);
1124                 return(error);
1125         }
1126         error = xfs_attr_leaf_getvalue(bp, args);
1127         xfs_da_brelse(args->trans, bp);
1128         if (!error && (args->rmtblkno > 0) && !(args->flags & ATTR_KERNOVAL)) {
1129                 error = xfs_attr_rmtval_get(args);
1130         }
1131         return(error);
1132 }
1133
1134 /*
1135  * Copy out attribute entries for attr_list(), for leaf attribute lists.
1136  */
1137 STATIC int
1138 xfs_attr_leaf_list(xfs_attr_list_context_t *context)
1139 {
1140         xfs_attr_leafblock_t *leaf;
1141         int error;
1142         xfs_dabuf_t *bp;
1143
1144         context->cursor->blkno = 0;
1145         error = xfs_da_read_buf(NULL, context->dp, 0, -1, &bp, XFS_ATTR_FORK);
1146         if (error)
1147                 return(error);
1148         ASSERT(bp != NULL);
1149         leaf = bp->data;
1150         if (unlikely(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT)
1151                                                 != XFS_ATTR_LEAF_MAGIC)) {
1152                 XFS_CORRUPTION_ERROR("xfs_attr_leaf_list", XFS_ERRLEVEL_LOW,
1153                                      context->dp->i_mount, leaf);
1154                 xfs_da_brelse(NULL, bp);
1155                 return(XFS_ERROR(EFSCORRUPTED));
1156         }
1157
1158         (void)xfs_attr_leaf_list_int(bp, context);
1159         xfs_da_brelse(NULL, bp);
1160         return(0);
1161 }
1162
1163
1164 /*========================================================================
1165  * External routines when attribute list size > XFS_LBSIZE(mp).
1166  *========================================================================*/
1167
1168 /*
1169  * Add a name to a Btree-format attribute list.
1170  *
1171  * This will involve walking down the Btree, and may involve splitting
1172  * leaf nodes and even splitting intermediate nodes up to and including
1173  * the root node (a special case of an intermediate node).
1174  *
1175  * "Remote" attribute values confuse the issue and atomic rename operations
1176  * add a whole extra layer of confusion on top of that.
1177  */
1178 STATIC int
1179 xfs_attr_node_addname(xfs_da_args_t *args)
1180 {
1181         xfs_da_state_t *state;
1182         xfs_da_state_blk_t *blk;
1183         xfs_inode_t *dp;
1184         xfs_mount_t *mp;
1185         int committed, retval, error;
1186
1187         /*
1188          * Fill in bucket of arguments/results/context to carry around.
1189          */
1190         dp = args->dp;
1191         mp = dp->i_mount;
1192 restart:
1193         state = xfs_da_state_alloc();
1194         state->args = args;
1195         state->mp = mp;
1196         state->blocksize = state->mp->m_sb.sb_blocksize;
1197         state->node_ents = state->mp->m_attr_node_ents;
1198
1199         /*
1200          * Search to see if name already exists, and get back a pointer
1201          * to where it should go.
1202          */
1203         error = xfs_da_node_lookup_int(state, &retval);
1204         if (error)
1205                 goto out;
1206         blk = &state->path.blk[ state->path.active-1 ];
1207         ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1208         if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
1209                 goto out;
1210         } else if (retval == EEXIST) {
1211                 if (args->flags & ATTR_CREATE)
1212                         goto out;
1213                 args->rename = 1;                       /* atomic rename op */
1214                 args->blkno2 = args->blkno;             /* set 2nd entry info*/
1215                 args->index2 = args->index;
1216                 args->rmtblkno2 = args->rmtblkno;
1217                 args->rmtblkcnt2 = args->rmtblkcnt;
1218                 args->rmtblkno = 0;
1219                 args->rmtblkcnt = 0;
1220         }
1221
1222         retval = xfs_attr_leaf_add(blk->bp, state->args);
1223         if (retval == ENOSPC) {
1224                 if (state->path.active == 1) {
1225                         /*
1226                          * Its really a single leaf node, but it had
1227                          * out-of-line values so it looked like it *might*
1228                          * have been a b-tree.
1229                          */
1230                         xfs_da_state_free(state);
1231                         XFS_BMAP_INIT(args->flist, args->firstblock);
1232                         error = xfs_attr_leaf_to_node(args);
1233                         if (!error) {
1234                                 error = xfs_bmap_finish(&args->trans,
1235                                                         args->flist,
1236                                                         *args->firstblock,
1237                                                         &committed);
1238                         }
1239                         if (error) {
1240                                 ASSERT(committed);
1241                                 args->trans = NULL;
1242                                 xfs_bmap_cancel(args->flist);
1243                                 goto out;
1244                         }
1245
1246                         /*
1247                          * bmap_finish() may have committed the last trans
1248                          * and started a new one.  We need the inode to be
1249                          * in all transactions.
1250                          */
1251                         if (committed) {
1252                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1253                                 xfs_trans_ihold(args->trans, dp);
1254                         }
1255
1256                         /*
1257                          * Commit the node conversion and start the next
1258                          * trans in the chain.
1259                          */
1260                         if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1261                                 goto out;
1262
1263                         goto restart;
1264                 }
1265
1266                 /*
1267                  * Split as many Btree elements as required.
1268                  * This code tracks the new and old attr's location
1269                  * in the index/blkno/rmtblkno/rmtblkcnt fields and
1270                  * in the index2/blkno2/rmtblkno2/rmtblkcnt2 fields.
1271                  */
1272                 XFS_BMAP_INIT(args->flist, args->firstblock);
1273                 error = xfs_da_split(state);
1274                 if (!error) {
1275                         error = xfs_bmap_finish(&args->trans, args->flist,
1276                                                 *args->firstblock, &committed);
1277                 }
1278                 if (error) {
1279                         ASSERT(committed);
1280                         args->trans = NULL;
1281                         xfs_bmap_cancel(args->flist);
1282                         goto out;
1283                 }
1284
1285                 /*
1286                  * bmap_finish() may have committed the last trans and started
1287                  * a new one.  We need the inode to be in all transactions.
1288                  */
1289                 if (committed) {
1290                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1291                         xfs_trans_ihold(args->trans, dp);
1292                 }
1293         } else {
1294                 /*
1295                  * Addition succeeded, update Btree hashvals.
1296                  */
1297                 xfs_da_fixhashpath(state, &state->path);
1298         }
1299
1300         /*
1301          * Kill the state structure, we're done with it and need to
1302          * allow the buffers to come back later.
1303          */
1304         xfs_da_state_free(state);
1305         state = NULL;
1306
1307         /*
1308          * Commit the leaf addition or btree split and start the next
1309          * trans in the chain.
1310          */
1311         if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1312                 goto out;
1313
1314         /*
1315          * If there was an out-of-line value, allocate the blocks we
1316          * identified for its storage and copy the value.  This is done
1317          * after we create the attribute so that we don't overflow the
1318          * maximum size of a transaction and/or hit a deadlock.
1319          */
1320         if (args->rmtblkno > 0) {
1321                 error = xfs_attr_rmtval_set(args);
1322                 if (error)
1323                         return(error);
1324         }
1325
1326         /*
1327          * If this is an atomic rename operation, we must "flip" the
1328          * incomplete flags on the "new" and "old" attribute/value pairs
1329          * so that one disappears and one appears atomically.  Then we
1330          * must remove the "old" attribute/value pair.
1331          */
1332         if (args->rename) {
1333                 /*
1334                  * In a separate transaction, set the incomplete flag on the
1335                  * "old" attr and clear the incomplete flag on the "new" attr.
1336                  */
1337                 error = xfs_attr_leaf_flipflags(args);
1338                 if (error)
1339                         goto out;
1340
1341                 /*
1342                  * Dismantle the "old" attribute/value pair by removing
1343                  * a "remote" value (if it exists).
1344                  */
1345                 args->index = args->index2;
1346                 args->blkno = args->blkno2;
1347                 args->rmtblkno = args->rmtblkno2;
1348                 args->rmtblkcnt = args->rmtblkcnt2;
1349                 if (args->rmtblkno) {
1350                         error = xfs_attr_rmtval_remove(args);
1351                         if (error)
1352                                 return(error);
1353                 }
1354
1355                 /*
1356                  * Re-find the "old" attribute entry after any split ops.
1357                  * The INCOMPLETE flag means that we will find the "old"
1358                  * attr, not the "new" one.
1359                  */
1360                 args->flags |= XFS_ATTR_INCOMPLETE;
1361                 state = xfs_da_state_alloc();
1362                 state->args = args;
1363                 state->mp = mp;
1364                 state->blocksize = state->mp->m_sb.sb_blocksize;
1365                 state->node_ents = state->mp->m_attr_node_ents;
1366                 state->inleaf = 0;
1367                 error = xfs_da_node_lookup_int(state, &retval);
1368                 if (error)
1369                         goto out;
1370
1371                 /*
1372                  * Remove the name and update the hashvals in the tree.
1373                  */
1374                 blk = &state->path.blk[ state->path.active-1 ];
1375                 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1376                 error = xfs_attr_leaf_remove(blk->bp, args);
1377                 xfs_da_fixhashpath(state, &state->path);
1378
1379                 /*
1380                  * Check to see if the tree needs to be collapsed.
1381                  */
1382                 if (retval && (state->path.active > 1)) {
1383                         XFS_BMAP_INIT(args->flist, args->firstblock);
1384                         error = xfs_da_join(state);
1385                         if (!error) {
1386                                 error = xfs_bmap_finish(&args->trans,
1387                                                         args->flist,
1388                                                         *args->firstblock,
1389                                                         &committed);
1390                         }
1391                         if (error) {
1392                                 ASSERT(committed);
1393                                 args->trans = NULL;
1394                                 xfs_bmap_cancel(args->flist);
1395                                 goto out;
1396                         }
1397
1398                         /*
1399                          * bmap_finish() may have committed the last trans
1400                          * and started a new one.  We need the inode to be
1401                          * in all transactions.
1402                          */
1403                         if (committed) {
1404                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1405                                 xfs_trans_ihold(args->trans, dp);
1406                         }
1407                 }
1408
1409                 /*
1410                  * Commit and start the next trans in the chain.
1411                  */
1412                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1413                         goto out;
1414
1415         } else if (args->rmtblkno > 0) {
1416                 /*
1417                  * Added a "remote" value, just clear the incomplete flag.
1418                  */
1419                 error = xfs_attr_leaf_clearflag(args);
1420                 if (error)
1421                         goto out;
1422         }
1423         retval = error = 0;
1424
1425 out:
1426         if (state)
1427                 xfs_da_state_free(state);
1428         if (error)
1429                 return(error);
1430         return(retval);
1431 }
1432
1433 /*
1434  * Remove a name from a B-tree attribute list.
1435  *
1436  * This will involve walking down the Btree, and may involve joining
1437  * leaf nodes and even joining intermediate nodes up to and including
1438  * the root node (a special case of an intermediate node).
1439  */
1440 STATIC int
1441 xfs_attr_node_removename(xfs_da_args_t *args)
1442 {
1443         xfs_da_state_t *state;
1444         xfs_da_state_blk_t *blk;
1445         xfs_inode_t *dp;
1446         xfs_dabuf_t *bp;
1447         int retval, error, committed, forkoff;
1448
1449         /*
1450          * Tie a string around our finger to remind us where we are.
1451          */
1452         dp = args->dp;
1453         state = xfs_da_state_alloc();
1454         state->args = args;
1455         state->mp = dp->i_mount;
1456         state->blocksize = state->mp->m_sb.sb_blocksize;
1457         state->node_ents = state->mp->m_attr_node_ents;
1458
1459         /*
1460          * Search to see if name exists, and get back a pointer to it.
1461          */
1462         error = xfs_da_node_lookup_int(state, &retval);
1463         if (error || (retval != EEXIST)) {
1464                 if (error == 0)
1465                         error = retval;
1466                 goto out;
1467         }
1468
1469         /*
1470          * If there is an out-of-line value, de-allocate the blocks.
1471          * This is done before we remove the attribute so that we don't
1472          * overflow the maximum size of a transaction and/or hit a deadlock.
1473          */
1474         blk = &state->path.blk[ state->path.active-1 ];
1475         ASSERT(blk->bp != NULL);
1476         ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1477         if (args->rmtblkno > 0) {
1478                 /*
1479                  * Fill in disk block numbers in the state structure
1480                  * so that we can get the buffers back after we commit
1481                  * several transactions in the following calls.
1482                  */
1483                 error = xfs_attr_fillstate(state);
1484                 if (error)
1485                         goto out;
1486
1487                 /*
1488                  * Mark the attribute as INCOMPLETE, then bunmapi() the
1489                  * remote value.
1490                  */
1491                 error = xfs_attr_leaf_setflag(args);
1492                 if (error)
1493                         goto out;
1494                 error = xfs_attr_rmtval_remove(args);
1495                 if (error)
1496                         goto out;
1497
1498                 /*
1499                  * Refill the state structure with buffers, the prior calls
1500                  * released our buffers.
1501                  */
1502                 error = xfs_attr_refillstate(state);
1503                 if (error)
1504                         goto out;
1505         }
1506
1507         /*
1508          * Remove the name and update the hashvals in the tree.
1509          */
1510         blk = &state->path.blk[ state->path.active-1 ];
1511         ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1512         retval = xfs_attr_leaf_remove(blk->bp, args);
1513         xfs_da_fixhashpath(state, &state->path);
1514
1515         /*
1516          * Check to see if the tree needs to be collapsed.
1517          */
1518         if (retval && (state->path.active > 1)) {
1519                 XFS_BMAP_INIT(args->flist, args->firstblock);
1520                 error = xfs_da_join(state);
1521                 if (!error) {
1522                         error = xfs_bmap_finish(&args->trans, args->flist,
1523                                                 *args->firstblock, &committed);
1524                 }
1525                 if (error) {
1526                         ASSERT(committed);
1527                         args->trans = NULL;
1528                         xfs_bmap_cancel(args->flist);
1529                         goto out;
1530                 }
1531
1532                 /*
1533                  * bmap_finish() may have committed the last trans and started
1534                  * a new one.  We need the inode to be in all transactions.
1535                  */
1536                 if (committed) {
1537                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1538                         xfs_trans_ihold(args->trans, dp);
1539                 }
1540
1541                 /*
1542                  * Commit the Btree join operation and start a new trans.
1543                  */
1544                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1545                         goto out;
1546         }
1547
1548         /*
1549          * If the result is small enough, push it all into the inode.
1550          */
1551         if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
1552                 /*
1553                  * Have to get rid of the copy of this dabuf in the state.
1554                  */
1555                 ASSERT(state->path.active == 1);
1556                 ASSERT(state->path.blk[0].bp);
1557                 xfs_da_buf_done(state->path.blk[0].bp);
1558                 state->path.blk[0].bp = NULL;
1559
1560                 error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp,
1561                                                      XFS_ATTR_FORK);
1562                 if (error)
1563                         goto out;
1564                 ASSERT(INT_GET(((xfs_attr_leafblock_t *)
1565                                       bp->data)->hdr.info.magic, ARCH_CONVERT)
1566                                                        == XFS_ATTR_LEAF_MAGIC);
1567
1568                 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1569                         XFS_BMAP_INIT(args->flist, args->firstblock);
1570                         error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1571                         /* bp is gone due to xfs_da_shrink_inode */
1572                         if (!error) {
1573                                 error = xfs_bmap_finish(&args->trans,
1574                                                         args->flist,
1575                                                         *args->firstblock,
1576                                                         &committed);
1577                         }
1578                         if (error) {
1579                                 ASSERT(committed);
1580                                 args->trans = NULL;
1581                                 xfs_bmap_cancel(args->flist);
1582                                 goto out;
1583                         }
1584
1585                         /*
1586                          * bmap_finish() may have committed the last trans
1587                          * and started a new one.  We need the inode to be
1588                          * in all transactions.
1589                          */
1590                         if (committed) {
1591                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1592                                 xfs_trans_ihold(args->trans, dp);
1593                         }
1594                 } else
1595                         xfs_da_brelse(args->trans, bp);
1596         }
1597         error = 0;
1598
1599 out:
1600         xfs_da_state_free(state);
1601         return(error);
1602 }
1603
1604 /*
1605  * Fill in the disk block numbers in the state structure for the buffers
1606  * that are attached to the state structure.
1607  * This is done so that we can quickly reattach ourselves to those buffers
1608  * after some set of transaction commit's has released these buffers.
1609  */
1610 STATIC int
1611 xfs_attr_fillstate(xfs_da_state_t *state)
1612 {
1613         xfs_da_state_path_t *path;
1614         xfs_da_state_blk_t *blk;
1615         int level;
1616
1617         /*
1618          * Roll down the "path" in the state structure, storing the on-disk
1619          * block number for those buffers in the "path".
1620          */
1621         path = &state->path;
1622         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1623         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1624                 if (blk->bp) {
1625                         blk->disk_blkno = xfs_da_blkno(blk->bp);
1626                         xfs_da_buf_done(blk->bp);
1627                         blk->bp = NULL;
1628                 } else {
1629                         blk->disk_blkno = 0;
1630                 }
1631         }
1632
1633         /*
1634          * Roll down the "altpath" in the state structure, storing the on-disk
1635          * block number for those buffers in the "altpath".
1636          */
1637         path = &state->altpath;
1638         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1639         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1640                 if (blk->bp) {
1641                         blk->disk_blkno = xfs_da_blkno(blk->bp);
1642                         xfs_da_buf_done(blk->bp);
1643                         blk->bp = NULL;
1644                 } else {
1645                         blk->disk_blkno = 0;
1646                 }
1647         }
1648
1649         return(0);
1650 }
1651
1652 /*
1653  * Reattach the buffers to the state structure based on the disk block
1654  * numbers stored in the state structure.
1655  * This is done after some set of transaction commit's has released those
1656  * buffers from our grip.
1657  */
1658 STATIC int
1659 xfs_attr_refillstate(xfs_da_state_t *state)
1660 {
1661         xfs_da_state_path_t *path;
1662         xfs_da_state_blk_t *blk;
1663         int level, error;
1664
1665         /*
1666          * Roll down the "path" in the state structure, storing the on-disk
1667          * block number for those buffers in the "path".
1668          */
1669         path = &state->path;
1670         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1671         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1672                 if (blk->disk_blkno) {
1673                         error = xfs_da_read_buf(state->args->trans,
1674                                                 state->args->dp,
1675                                                 blk->blkno, blk->disk_blkno,
1676                                                 &blk->bp, XFS_ATTR_FORK);
1677                         if (error)
1678                                 return(error);
1679                 } else {
1680                         blk->bp = NULL;
1681                 }
1682         }
1683
1684         /*
1685          * Roll down the "altpath" in the state structure, storing the on-disk
1686          * block number for those buffers in the "altpath".
1687          */
1688         path = &state->altpath;
1689         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1690         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1691                 if (blk->disk_blkno) {
1692                         error = xfs_da_read_buf(state->args->trans,
1693                                                 state->args->dp,
1694                                                 blk->blkno, blk->disk_blkno,
1695                                                 &blk->bp, XFS_ATTR_FORK);
1696                         if (error)
1697                                 return(error);
1698                 } else {
1699                         blk->bp = NULL;
1700                 }
1701         }
1702
1703         return(0);
1704 }
1705
1706 /*
1707  * Look up a filename in a node attribute list.
1708  *
1709  * This routine gets called for any attribute fork that has more than one
1710  * block, ie: both true Btree attr lists and for single-leaf-blocks with
1711  * "remote" values taking up more blocks.
1712  */
1713 STATIC int
1714 xfs_attr_node_get(xfs_da_args_t *args)
1715 {
1716         xfs_da_state_t *state;
1717         xfs_da_state_blk_t *blk;
1718         int error, retval;
1719         int i;
1720
1721         state = xfs_da_state_alloc();
1722         state->args = args;
1723         state->mp = args->dp->i_mount;
1724         state->blocksize = state->mp->m_sb.sb_blocksize;
1725         state->node_ents = state->mp->m_attr_node_ents;
1726
1727         /*
1728          * Search to see if name exists, and get back a pointer to it.
1729          */
1730         error = xfs_da_node_lookup_int(state, &retval);
1731         if (error) {
1732                 retval = error;
1733         } else if (retval == EEXIST) {
1734                 blk = &state->path.blk[ state->path.active-1 ];
1735                 ASSERT(blk->bp != NULL);
1736                 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1737
1738                 /*
1739                  * Get the value, local or "remote"
1740                  */
1741                 retval = xfs_attr_leaf_getvalue(blk->bp, args);
1742                 if (!retval && (args->rmtblkno > 0)
1743                     && !(args->flags & ATTR_KERNOVAL)) {
1744                         retval = xfs_attr_rmtval_get(args);
1745                 }
1746         }
1747
1748         /*
1749          * If not in a transaction, we have to release all the buffers.
1750          */
1751         for (i = 0; i < state->path.active; i++) {
1752                 xfs_da_brelse(args->trans, state->path.blk[i].bp);
1753                 state->path.blk[i].bp = NULL;
1754         }
1755
1756         xfs_da_state_free(state);
1757         return(retval);
1758 }
1759
1760 STATIC int                                                      /* error */
1761 xfs_attr_node_list(xfs_attr_list_context_t *context)
1762 {
1763         attrlist_cursor_kern_t *cursor;
1764         xfs_attr_leafblock_t *leaf;
1765         xfs_da_intnode_t *node;
1766         xfs_da_node_entry_t *btree;
1767         int error, i;
1768         xfs_dabuf_t *bp;
1769
1770         cursor = context->cursor;
1771         cursor->initted = 1;
1772
1773         /*
1774          * Do all sorts of validation on the passed-in cursor structure.
1775          * If anything is amiss, ignore the cursor and look up the hashval
1776          * starting from the btree root.
1777          */
1778         bp = NULL;
1779         if (cursor->blkno > 0) {
1780                 error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1781                                               &bp, XFS_ATTR_FORK);
1782                 if ((error != 0) && (error != EFSCORRUPTED))
1783                         return(error);
1784                 if (bp) {
1785                         node = bp->data;
1786                         switch (INT_GET(node->hdr.info.magic, ARCH_CONVERT)) {
1787                         case XFS_DA_NODE_MAGIC:
1788                                 xfs_attr_trace_l_cn("wrong blk", context, node);
1789                                 xfs_da_brelse(NULL, bp);
1790                                 bp = NULL;
1791                                 break;
1792                         case XFS_ATTR_LEAF_MAGIC:
1793                                 leaf = bp->data;
1794                                 if (cursor->hashval >
1795                                     INT_GET(leaf->entries[
1796                                          INT_GET(leaf->hdr.count,
1797                                                 ARCH_CONVERT)-1].hashval,
1798                                                         ARCH_CONVERT)) {
1799                                         xfs_attr_trace_l_cl("wrong blk",
1800                                                            context, leaf);
1801                                         xfs_da_brelse(NULL, bp);
1802                                         bp = NULL;
1803                                 } else if (cursor->hashval <=
1804                                              INT_GET(leaf->entries[0].hashval,
1805                                                         ARCH_CONVERT)) {
1806                                         xfs_attr_trace_l_cl("maybe wrong blk",
1807                                                            context, leaf);
1808                                         xfs_da_brelse(NULL, bp);
1809                                         bp = NULL;
1810                                 }
1811                                 break;
1812                         default:
1813                                 xfs_attr_trace_l_c("wrong blk - ??", context);
1814                                 xfs_da_brelse(NULL, bp);
1815                                 bp = NULL;
1816                         }
1817                 }
1818         }
1819
1820         /*
1821          * We did not find what we expected given the cursor's contents,
1822          * so we start from the top and work down based on the hash value.
1823          * Note that start of node block is same as start of leaf block.
1824          */
1825         if (bp == NULL) {
1826                 cursor->blkno = 0;
1827                 for (;;) {
1828                         error = xfs_da_read_buf(NULL, context->dp,
1829                                                       cursor->blkno, -1, &bp,
1830                                                       XFS_ATTR_FORK);
1831                         if (error)
1832                                 return(error);
1833                         if (unlikely(bp == NULL)) {
1834                                 XFS_ERROR_REPORT("xfs_attr_node_list(2)",
1835                                                  XFS_ERRLEVEL_LOW,
1836                                                  context->dp->i_mount);
1837                                 return(XFS_ERROR(EFSCORRUPTED));
1838                         }
1839                         node = bp->data;
1840                         if (INT_GET(node->hdr.info.magic, ARCH_CONVERT)
1841                                                         == XFS_ATTR_LEAF_MAGIC)
1842                                 break;
1843                         if (unlikely(INT_GET(node->hdr.info.magic, ARCH_CONVERT)
1844                                                         != XFS_DA_NODE_MAGIC)) {
1845                                 XFS_CORRUPTION_ERROR("xfs_attr_node_list(3)",
1846                                                      XFS_ERRLEVEL_LOW,
1847                                                      context->dp->i_mount,
1848                                                      node);
1849                                 xfs_da_brelse(NULL, bp);
1850                                 return(XFS_ERROR(EFSCORRUPTED));
1851                         }
1852                         btree = node->btree;
1853                         for (i = 0;
1854                                 i < INT_GET(node->hdr.count, ARCH_CONVERT);
1855                                                                 btree++, i++) {
1856                                 if (cursor->hashval
1857                                                 <= INT_GET(btree->hashval,
1858                                                             ARCH_CONVERT)) {
1859                                         cursor->blkno = INT_GET(btree->before, ARCH_CONVERT);
1860                                         xfs_attr_trace_l_cb("descending",
1861                                                             context, btree);
1862                                         break;
1863                                 }
1864                         }
1865                         if (i == INT_GET(node->hdr.count, ARCH_CONVERT)) {
1866                                 xfs_da_brelse(NULL, bp);
1867                                 return(0);
1868                         }
1869                         xfs_da_brelse(NULL, bp);
1870                 }
1871         }
1872         ASSERT(bp != NULL);
1873
1874         /*
1875          * Roll upward through the blocks, processing each leaf block in
1876          * order.  As long as there is space in the result buffer, keep
1877          * adding the information.
1878          */
1879         for (;;) {
1880                 leaf = bp->data;
1881                 if (unlikely(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT)
1882                                                 != XFS_ATTR_LEAF_MAGIC)) {
1883                         XFS_CORRUPTION_ERROR("xfs_attr_node_list(4)",
1884                                              XFS_ERRLEVEL_LOW,
1885                                              context->dp->i_mount, leaf);
1886                         xfs_da_brelse(NULL, bp);
1887                         return(XFS_ERROR(EFSCORRUPTED));
1888                 }
1889                 error = xfs_attr_leaf_list_int(bp, context);
1890                 if (error || !leaf->hdr.info.forw)
1891                         break;  /* not really an error, buffer full or EOF */
1892                 cursor->blkno = INT_GET(leaf->hdr.info.forw, ARCH_CONVERT);
1893                 xfs_da_brelse(NULL, bp);
1894                 error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1895                                               &bp, XFS_ATTR_FORK);
1896                 if (error)
1897                         return(error);
1898                 if (unlikely((bp == NULL))) {
1899                         XFS_ERROR_REPORT("xfs_attr_node_list(5)",
1900                                          XFS_ERRLEVEL_LOW,
1901                                          context->dp->i_mount);
1902                         return(XFS_ERROR(EFSCORRUPTED));
1903                 }
1904         }
1905         xfs_da_brelse(NULL, bp);
1906         return(0);
1907 }
1908
1909
1910 /*========================================================================
1911  * External routines for manipulating out-of-line attribute values.
1912  *========================================================================*/
1913
1914 /*
1915  * Read the value associated with an attribute from the out-of-line buffer
1916  * that we stored it in.
1917  */
1918 STATIC int
1919 xfs_attr_rmtval_get(xfs_da_args_t *args)
1920 {
1921         xfs_bmbt_irec_t map[ATTR_RMTVALUE_MAPSIZE];
1922         xfs_mount_t *mp;
1923         xfs_daddr_t dblkno;
1924         xfs_caddr_t dst;
1925         xfs_buf_t *bp;
1926         int nmap, error, tmp, valuelen, blkcnt, i;
1927         xfs_dablk_t lblkno;
1928
1929         ASSERT(!(args->flags & ATTR_KERNOVAL));
1930
1931         mp = args->dp->i_mount;
1932         dst = args->value;
1933         valuelen = args->valuelen;
1934         lblkno = args->rmtblkno;
1935         while (valuelen > 0) {
1936                 nmap = ATTR_RMTVALUE_MAPSIZE;
1937                 error = xfs_bmapi(args->trans, args->dp, (xfs_fileoff_t)lblkno,
1938                                   args->rmtblkcnt,
1939                                   XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
1940                                   NULL, 0, map, &nmap, NULL);
1941                 if (error)
1942                         return(error);
1943                 ASSERT(nmap >= 1);
1944
1945                 for (i = 0; (i < nmap) && (valuelen > 0); i++) {
1946                         ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
1947                                (map[i].br_startblock != HOLESTARTBLOCK));
1948                         dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
1949                         blkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
1950                         error = xfs_read_buf(mp, mp->m_ddev_targp, dblkno,
1951                                              blkcnt, XFS_BUF_LOCK, &bp);
1952                         if (error)
1953                                 return(error);
1954
1955                         tmp = (valuelen < XFS_BUF_SIZE(bp))
1956                                 ? valuelen : XFS_BUF_SIZE(bp);
1957                         xfs_biomove(bp, 0, tmp, dst, XFS_B_READ);
1958                         xfs_buf_relse(bp);
1959                         dst += tmp;
1960                         valuelen -= tmp;
1961
1962                         lblkno += map[i].br_blockcount;
1963                 }
1964         }
1965         ASSERT(valuelen == 0);
1966         return(0);
1967 }
1968
1969 /*
1970  * Write the value associated with an attribute into the out-of-line buffer
1971  * that we have defined for it.
1972  */
1973 STATIC int
1974 xfs_attr_rmtval_set(xfs_da_args_t *args)
1975 {
1976         xfs_mount_t *mp;
1977         xfs_fileoff_t lfileoff;
1978         xfs_inode_t *dp;
1979         xfs_bmbt_irec_t map;
1980         xfs_daddr_t dblkno;
1981         xfs_caddr_t src;
1982         xfs_buf_t *bp;
1983         xfs_dablk_t lblkno;
1984         int blkcnt, valuelen, nmap, error, tmp, committed;
1985
1986         dp = args->dp;
1987         mp = dp->i_mount;
1988         src = args->value;
1989
1990         /*
1991          * Find a "hole" in the attribute address space large enough for
1992          * us to drop the new attribute's value into.
1993          */
1994         blkcnt = XFS_B_TO_FSB(mp, args->valuelen);
1995         lfileoff = 0;
1996         error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
1997                                                    XFS_ATTR_FORK);
1998         if (error) {
1999                 return(error);
2000         }
2001         args->rmtblkno = lblkno = (xfs_dablk_t)lfileoff;
2002         args->rmtblkcnt = blkcnt;
2003
2004         /*
2005          * Roll through the "value", allocating blocks on disk as required.
2006          */
2007         while (blkcnt > 0) {
2008                 /*
2009                  * Allocate a single extent, up to the size of the value.
2010                  */
2011                 XFS_BMAP_INIT(args->flist, args->firstblock);
2012                 nmap = 1;
2013                 error = xfs_bmapi(args->trans, dp, (xfs_fileoff_t)lblkno,
2014                                   blkcnt,
2015                                   XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA |
2016                                                         XFS_BMAPI_WRITE,
2017                                   args->firstblock, args->total, &map, &nmap,
2018                                   args->flist);
2019                 if (!error) {
2020                         error = xfs_bmap_finish(&args->trans, args->flist,
2021                                                 *args->firstblock, &committed);
2022                 }
2023                 if (error) {
2024                         ASSERT(committed);
2025                         args->trans = NULL;
2026                         xfs_bmap_cancel(args->flist);
2027                         return(error);
2028                 }
2029
2030                 /*
2031                  * bmap_finish() may have committed the last trans and started
2032                  * a new one.  We need the inode to be in all transactions.
2033                  */
2034                 if (committed) {
2035                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
2036                         xfs_trans_ihold(args->trans, dp);
2037                 }
2038
2039                 ASSERT(nmap == 1);
2040                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2041                        (map.br_startblock != HOLESTARTBLOCK));
2042                 lblkno += map.br_blockcount;
2043                 blkcnt -= map.br_blockcount;
2044
2045                 /*
2046                  * Start the next trans in the chain.
2047                  */
2048                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
2049                         return (error);
2050         }
2051
2052         /*
2053          * Roll through the "value", copying the attribute value to the
2054          * already-allocated blocks.  Blocks are written synchronously
2055          * so that we can know they are all on disk before we turn off
2056          * the INCOMPLETE flag.
2057          */
2058         lblkno = args->rmtblkno;
2059         valuelen = args->valuelen;
2060         while (valuelen > 0) {
2061                 /*
2062                  * Try to remember where we decided to put the value.
2063                  */
2064                 XFS_BMAP_INIT(args->flist, args->firstblock);
2065                 nmap = 1;
2066                 error = xfs_bmapi(NULL, dp, (xfs_fileoff_t)lblkno,
2067                                   args->rmtblkcnt,
2068                                   XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2069                                   args->firstblock, 0, &map, &nmap, NULL);
2070                 if (error) {
2071                         return(error);
2072                 }
2073                 ASSERT(nmap == 1);
2074                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2075                        (map.br_startblock != HOLESTARTBLOCK));
2076
2077                 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2078                 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2079
2080                 bp = xfs_buf_get_flags(mp->m_ddev_targp, dblkno,
2081                                                         blkcnt, XFS_BUF_LOCK);
2082                 ASSERT(bp);
2083                 ASSERT(!XFS_BUF_GETERROR(bp));
2084
2085                 tmp = (valuelen < XFS_BUF_SIZE(bp)) ? valuelen :
2086                                                         XFS_BUF_SIZE(bp);
2087                 xfs_biomove(bp, 0, tmp, src, XFS_B_WRITE);
2088                 if (tmp < XFS_BUF_SIZE(bp))
2089                         xfs_biozero(bp, tmp, XFS_BUF_SIZE(bp) - tmp);
2090                 if ((error = xfs_bwrite(mp, bp))) {/* GROT: NOTE: synchronous write */
2091                         return (error);
2092                 }
2093                 src += tmp;
2094                 valuelen -= tmp;
2095
2096                 lblkno += map.br_blockcount;
2097         }
2098         ASSERT(valuelen == 0);
2099         return(0);
2100 }
2101
2102 /*
2103  * Remove the value associated with an attribute by deleting the
2104  * out-of-line buffer that it is stored on.
2105  */
2106 STATIC int
2107 xfs_attr_rmtval_remove(xfs_da_args_t *args)
2108 {
2109         xfs_mount_t *mp;
2110         xfs_bmbt_irec_t map;
2111         xfs_buf_t *bp;
2112         xfs_daddr_t dblkno;
2113         xfs_dablk_t lblkno;
2114         int valuelen, blkcnt, nmap, error, done, committed;
2115
2116         mp = args->dp->i_mount;
2117
2118         /*
2119          * Roll through the "value", invalidating the attribute value's
2120          * blocks.
2121          */
2122         lblkno = args->rmtblkno;
2123         valuelen = args->rmtblkcnt;
2124         while (valuelen > 0) {
2125                 /*
2126                  * Try to remember where we decided to put the value.
2127                  */
2128                 XFS_BMAP_INIT(args->flist, args->firstblock);
2129                 nmap = 1;
2130                 error = xfs_bmapi(NULL, args->dp, (xfs_fileoff_t)lblkno,
2131                                         args->rmtblkcnt,
2132                                         XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2133                                         args->firstblock, 0, &map, &nmap,
2134                                         args->flist);
2135                 if (error) {
2136                         return(error);
2137                 }
2138                 ASSERT(nmap == 1);
2139                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2140                        (map.br_startblock != HOLESTARTBLOCK));
2141
2142                 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2143                 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2144
2145                 /*
2146                  * If the "remote" value is in the cache, remove it.
2147                  */
2148                 bp = xfs_incore(mp->m_ddev_targp, dblkno, blkcnt,
2149                                 XFS_INCORE_TRYLOCK);
2150                 if (bp) {
2151                         XFS_BUF_STALE(bp);
2152                         XFS_BUF_UNDELAYWRITE(bp);
2153                         xfs_buf_relse(bp);
2154                         bp = NULL;
2155                 }
2156
2157                 valuelen -= map.br_blockcount;
2158
2159                 lblkno += map.br_blockcount;
2160         }
2161
2162         /*
2163          * Keep de-allocating extents until the remote-value region is gone.
2164          */
2165         lblkno = args->rmtblkno;
2166         blkcnt = args->rmtblkcnt;
2167         done = 0;
2168         while (!done) {
2169                 XFS_BMAP_INIT(args->flist, args->firstblock);
2170                 error = xfs_bunmapi(args->trans, args->dp, lblkno, blkcnt,
2171                                     XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2172                                     1, args->firstblock, args->flist, &done);
2173                 if (!error) {
2174                         error = xfs_bmap_finish(&args->trans, args->flist,
2175                                                 *args->firstblock, &committed);
2176                 }
2177                 if (error) {
2178                         ASSERT(committed);
2179                         args->trans = NULL;
2180                         xfs_bmap_cancel(args->flist);
2181                         return(error);
2182                 }
2183
2184                 /*
2185                  * bmap_finish() may have committed the last trans and started
2186                  * a new one.  We need the inode to be in all transactions.
2187                  */
2188                 if (committed) {
2189                         xfs_trans_ijoin(args->trans, args->dp, XFS_ILOCK_EXCL);
2190                         xfs_trans_ihold(args->trans, args->dp);
2191                 }
2192
2193                 /*
2194                  * Close out trans and start the next one in the chain.
2195                  */
2196                 if ((error = xfs_attr_rolltrans(&args->trans, args->dp)))
2197                         return (error);
2198         }
2199         return(0);
2200 }
2201
2202 #if defined(XFS_ATTR_TRACE)
2203 /*
2204  * Add a trace buffer entry for an attr_list context structure.
2205  */
2206 void
2207 xfs_attr_trace_l_c(char *where, struct xfs_attr_list_context *context)
2208 {
2209         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_C, where,
2210                 (__psunsigned_t)context->dp,
2211                 (__psunsigned_t)context->cursor->hashval,
2212                 (__psunsigned_t)context->cursor->blkno,
2213                 (__psunsigned_t)context->cursor->offset,
2214                 (__psunsigned_t)context->alist,
2215                 (__psunsigned_t)context->bufsize,
2216                 (__psunsigned_t)context->count,
2217                 (__psunsigned_t)context->firstu,
2218                 (__psunsigned_t)
2219                         ((context->count > 0) &&
2220                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2221                                 ? (ATTR_ENTRY(context->alist,
2222                                               context->count-1)->a_valuelen)
2223                                 : 0,
2224                 (__psunsigned_t)context->dupcnt,
2225                 (__psunsigned_t)context->flags,
2226                 (__psunsigned_t)NULL,
2227                 (__psunsigned_t)NULL,
2228                 (__psunsigned_t)NULL);
2229 }
2230
2231 /*
2232  * Add a trace buffer entry for a context structure and a Btree node.
2233  */
2234 void
2235 xfs_attr_trace_l_cn(char *where, struct xfs_attr_list_context *context,
2236                          struct xfs_da_intnode *node)
2237 {
2238         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_CN, where,
2239                 (__psunsigned_t)context->dp,
2240                 (__psunsigned_t)context->cursor->hashval,
2241                 (__psunsigned_t)context->cursor->blkno,
2242                 (__psunsigned_t)context->cursor->offset,
2243                 (__psunsigned_t)context->alist,
2244                 (__psunsigned_t)context->bufsize,
2245                 (__psunsigned_t)context->count,
2246                 (__psunsigned_t)context->firstu,
2247                 (__psunsigned_t)
2248                         ((context->count > 0) &&
2249                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2250                                 ? (ATTR_ENTRY(context->alist,
2251                                               context->count-1)->a_valuelen)
2252                                 : 0,
2253                 (__psunsigned_t)context->dupcnt,
2254                 (__psunsigned_t)context->flags,
2255                 (__psunsigned_t)INT_GET(node->hdr.count, ARCH_CONVERT),
2256                 (__psunsigned_t)INT_GET(node->btree[0].hashval, ARCH_CONVERT),
2257                 (__psunsigned_t)INT_GET(node->btree[INT_GET(node->hdr.count, ARCH_CONVERT)-1].hashval, ARCH_CONVERT));
2258 }
2259
2260 /*
2261  * Add a trace buffer entry for a context structure and a Btree element.
2262  */
2263 void
2264 xfs_attr_trace_l_cb(char *where, struct xfs_attr_list_context *context,
2265                           struct xfs_da_node_entry *btree)
2266 {
2267         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_CB, where,
2268                 (__psunsigned_t)context->dp,
2269                 (__psunsigned_t)context->cursor->hashval,
2270                 (__psunsigned_t)context->cursor->blkno,
2271                 (__psunsigned_t)context->cursor->offset,
2272                 (__psunsigned_t)context->alist,
2273                 (__psunsigned_t)context->bufsize,
2274                 (__psunsigned_t)context->count,
2275                 (__psunsigned_t)context->firstu,
2276                 (__psunsigned_t)
2277                         ((context->count > 0) &&
2278                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2279                                 ? (ATTR_ENTRY(context->alist,
2280                                               context->count-1)->a_valuelen)
2281                                 : 0,
2282                 (__psunsigned_t)context->dupcnt,
2283                 (__psunsigned_t)context->flags,
2284                 (__psunsigned_t)INT_GET(btree->hashval, ARCH_CONVERT),
2285                 (__psunsigned_t)INT_GET(btree->before, ARCH_CONVERT),
2286                 (__psunsigned_t)NULL);
2287 }
2288
2289 /*
2290  * Add a trace buffer entry for a context structure and a leaf block.
2291  */
2292 void
2293 xfs_attr_trace_l_cl(char *where, struct xfs_attr_list_context *context,
2294                               struct xfs_attr_leafblock *leaf)
2295 {
2296         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_CL, where,
2297                 (__psunsigned_t)context->dp,
2298                 (__psunsigned_t)context->cursor->hashval,
2299                 (__psunsigned_t)context->cursor->blkno,
2300                 (__psunsigned_t)context->cursor->offset,
2301                 (__psunsigned_t)context->alist,
2302                 (__psunsigned_t)context->bufsize,
2303                 (__psunsigned_t)context->count,
2304                 (__psunsigned_t)context->firstu,
2305                 (__psunsigned_t)
2306                         ((context->count > 0) &&
2307                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2308                                 ? (ATTR_ENTRY(context->alist,
2309                                               context->count-1)->a_valuelen)
2310                                 : 0,
2311                 (__psunsigned_t)context->dupcnt,
2312                 (__psunsigned_t)context->flags,
2313                 (__psunsigned_t)INT_GET(leaf->hdr.count, ARCH_CONVERT),
2314                 (__psunsigned_t)INT_GET(leaf->entries[0].hashval, ARCH_CONVERT),
2315                 (__psunsigned_t)INT_GET(leaf->entries[INT_GET(leaf->hdr.count, ARCH_CONVERT)-1].hashval, ARCH_CONVERT));
2316 }
2317
2318 /*
2319  * Add a trace buffer entry for the arguments given to the routine,
2320  * generic form.
2321  */
2322 void
2323 xfs_attr_trace_enter(int type, char *where,
2324                          __psunsigned_t a2, __psunsigned_t a3,
2325                          __psunsigned_t a4, __psunsigned_t a5,
2326                          __psunsigned_t a6, __psunsigned_t a7,
2327                          __psunsigned_t a8, __psunsigned_t a9,
2328                          __psunsigned_t a10, __psunsigned_t a11,
2329                          __psunsigned_t a12, __psunsigned_t a13,
2330                          __psunsigned_t a14, __psunsigned_t a15)
2331 {
2332         ASSERT(xfs_attr_trace_buf);
2333         ktrace_enter(xfs_attr_trace_buf, (void *)((__psunsigned_t)type),
2334                                          (void *)where,
2335                                          (void *)a2,  (void *)a3,  (void *)a4,
2336                                          (void *)a5,  (void *)a6,  (void *)a7,
2337                                          (void *)a8,  (void *)a9,  (void *)a10,
2338                                          (void *)a11, (void *)a12, (void *)a13,
2339                                          (void *)a14, (void *)a15);
2340 }
2341 #endif  /* XFS_ATTR_TRACE */
2342
2343
2344 /*========================================================================
2345  * System (pseudo) namespace attribute interface routines.
2346  *========================================================================*/
2347
2348 STATIC int
2349 posix_acl_access_set(
2350         vnode_t *vp, char *name, void *data, size_t size, int xflags)
2351 {
2352         return xfs_acl_vset(vp, data, size, _ACL_TYPE_ACCESS);
2353 }
2354
2355 STATIC int
2356 posix_acl_access_remove(
2357         struct vnode *vp, char *name, int xflags)
2358 {
2359         return xfs_acl_vremove(vp, _ACL_TYPE_ACCESS);
2360 }
2361
2362 STATIC int
2363 posix_acl_access_get(
2364         vnode_t *vp, char *name, void *data, size_t size, int xflags)
2365 {
2366         return xfs_acl_vget(vp, data, size, _ACL_TYPE_ACCESS);
2367 }
2368
2369 STATIC int
2370 posix_acl_access_exists(
2371         vnode_t *vp)
2372 {
2373         return xfs_acl_vhasacl_access(vp);
2374 }
2375
2376 STATIC int
2377 posix_acl_default_set(
2378         vnode_t *vp, char *name, void *data, size_t size, int xflags)
2379 {
2380         return xfs_acl_vset(vp, data, size, _ACL_TYPE_DEFAULT);
2381 }
2382
2383 STATIC int
2384 posix_acl_default_get(
2385         vnode_t *vp, char *name, void *data, size_t size, int xflags)
2386 {
2387         return xfs_acl_vget(vp, data, size, _ACL_TYPE_DEFAULT);
2388 }
2389
2390 STATIC int
2391 posix_acl_default_remove(
2392         struct vnode *vp, char *name, int xflags)
2393 {
2394         return xfs_acl_vremove(vp, _ACL_TYPE_DEFAULT);
2395 }
2396
2397 STATIC int
2398 posix_acl_default_exists(
2399         vnode_t *vp)
2400 {
2401         return xfs_acl_vhasacl_default(vp);
2402 }
2403
2404 STATIC struct attrnames posix_acl_access = {
2405         .attr_name      = "posix_acl_access",
2406         .attr_namelen   = sizeof("posix_acl_access") - 1,
2407         .attr_get       = posix_acl_access_get,
2408         .attr_set       = posix_acl_access_set,
2409         .attr_remove    = posix_acl_access_remove,
2410         .attr_exists    = posix_acl_access_exists,
2411 };
2412
2413 STATIC struct attrnames posix_acl_default = {
2414         .attr_name      = "posix_acl_default",
2415         .attr_namelen   = sizeof("posix_acl_default") - 1,
2416         .attr_get       = posix_acl_default_get,
2417         .attr_set       = posix_acl_default_set,
2418         .attr_remove    = posix_acl_default_remove,
2419         .attr_exists    = posix_acl_default_exists,
2420 };
2421
2422 STATIC struct attrnames *attr_system_names[] =
2423         { &posix_acl_access, &posix_acl_default };
2424
2425
2426 /*========================================================================
2427  * Namespace-prefix-style attribute name interface routines.
2428  *========================================================================*/
2429
2430 STATIC int
2431 attr_generic_set(
2432         struct vnode *vp, char *name, void *data, size_t size, int xflags)
2433 {
2434         int     error;
2435
2436         VOP_ATTR_SET(vp, name, data, size, xflags, NULL, error);
2437         return -error;
2438 }
2439
2440 STATIC int
2441 attr_generic_get(
2442         struct vnode *vp, char *name, void *data, size_t size, int xflags)
2443 {
2444         int     error, asize = size;
2445
2446         VOP_ATTR_GET(vp, name, data, &asize, xflags, NULL, error);
2447         if (!error)
2448                 return asize;
2449         return -error;
2450 }
2451
2452 STATIC int
2453 attr_generic_remove(
2454         struct vnode *vp, char *name, int xflags)
2455 {
2456         int     error;
2457
2458         VOP_ATTR_REMOVE(vp, name, xflags, NULL, error);
2459         return -error;
2460 }
2461
2462 STATIC int
2463 attr_generic_listadd(
2464         attrnames_t             *prefix,
2465         attrnames_t             *namesp,
2466         void                    *data,
2467         size_t                  size,
2468         ssize_t                 *result)
2469 {
2470         char                    *p = data + *result;
2471
2472         *result += prefix->attr_namelen;
2473         *result += namesp->attr_namelen + 1;
2474         if (!size)
2475                 return 0;
2476         if (*result > size)
2477                 return -ERANGE;
2478         strcpy(p, prefix->attr_name);
2479         p += prefix->attr_namelen;
2480         strcpy(p, namesp->attr_name);
2481         p += namesp->attr_namelen + 1;
2482         return 0;
2483 }
2484
2485 STATIC int
2486 attr_system_list(
2487         struct vnode            *vp,
2488         void                    *data,
2489         size_t                  size,
2490         ssize_t                 *result)
2491 {
2492         attrnames_t             *namesp;
2493         int                     i, error = 0;
2494
2495         for (i = 0; i < ATTR_SYSCOUNT; i++) {
2496                 namesp = attr_system_names[i];
2497                 if (!namesp->attr_exists || !namesp->attr_exists(vp))
2498                         continue;
2499                 error = attr_generic_listadd(&attr_system, namesp,
2500                                                 data, size, result);
2501                 if (error)
2502                         break;
2503         }
2504         return error;
2505 }
2506
2507 int
2508 attr_generic_list(
2509         struct vnode *vp, void *data, size_t size, int xflags, ssize_t *result)
2510 {
2511         attrlist_cursor_kern_t  cursor = { 0 };
2512         int                     error;
2513
2514         VOP_ATTR_LIST(vp, data, size, xflags, &cursor, NULL, error);
2515         if (error > 0)
2516                 return -error;
2517         *result = -error;
2518         return attr_system_list(vp, data, size, result);
2519 }
2520
2521 attrnames_t *
2522 attr_lookup_namespace(
2523         char                    *name,
2524         struct attrnames        **names,
2525         int                     nnames)
2526 {
2527         int                     i;
2528
2529         for (i = 0; i < nnames; i++)
2530                 if (!strncmp(name, names[i]->attr_name, names[i]->attr_namelen))
2531                         return names[i];
2532         return NULL;
2533 }
2534
2535 /*
2536  * Some checks to prevent people abusing EAs to get over quota:
2537  * - Don't allow modifying user EAs on devices/symlinks;
2538  * - Don't allow modifying user EAs if sticky bit set;
2539  */
2540 STATIC int
2541 attr_user_capable(
2542         struct vnode    *vp,
2543         cred_t          *cred)
2544 {
2545         struct inode    *inode = LINVFS_GET_IP(vp);
2546
2547         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
2548                 return -EPERM;
2549         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode) &&
2550             !capable(CAP_SYS_ADMIN))
2551                 return -EPERM;
2552         if (S_ISDIR(inode->i_mode) && (inode->i_mode & S_ISVTX) &&
2553             (current_fsuid(cred) != inode->i_uid) && !capable(CAP_FOWNER))
2554                 return -EPERM;
2555         return 0;
2556 }
2557
2558 STATIC int
2559 attr_trusted_capable(
2560         struct vnode    *vp,
2561         cred_t          *cred)
2562 {
2563         struct inode    *inode = LINVFS_GET_IP(vp);
2564
2565         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
2566                 return -EPERM;
2567         if (!capable(CAP_SYS_ADMIN))
2568                 return -EPERM;
2569         return 0;
2570 }
2571
2572 STATIC int
2573 attr_secure_capable(
2574         struct vnode    *vp,
2575         cred_t          *cred)
2576 {
2577         return -ENOSECURITY;
2578 }
2579
2580 STATIC int
2581 attr_system_set(
2582         struct vnode *vp, char *name, void *data, size_t size, int xflags)
2583 {
2584         attrnames_t     *namesp;
2585         int             error;
2586
2587         if (xflags & ATTR_CREATE)
2588                 return -EINVAL;
2589
2590         namesp = attr_lookup_namespace(name, attr_system_names, ATTR_SYSCOUNT);
2591         if (!namesp)
2592                 return -EOPNOTSUPP;
2593         error = namesp->attr_set(vp, name, data, size, xflags);
2594         if (!error)
2595                 error = vn_revalidate(vp);
2596         return error;
2597 }
2598
2599 STATIC int
2600 attr_system_get(
2601         struct vnode *vp, char *name, void *data, size_t size, int xflags)
2602 {
2603         attrnames_t     *namesp;
2604
2605         namesp = attr_lookup_namespace(name, attr_system_names, ATTR_SYSCOUNT);
2606         if (!namesp)
2607                 return -EOPNOTSUPP;
2608         return namesp->attr_get(vp, name, data, size, xflags);
2609 }
2610
2611 STATIC int
2612 attr_system_remove(
2613         struct vnode *vp, char *name, int xflags)
2614 {
2615         attrnames_t     *namesp;
2616
2617         namesp = attr_lookup_namespace(name, attr_system_names, ATTR_SYSCOUNT);
2618         if (!namesp)
2619                 return -EOPNOTSUPP;
2620         return namesp->attr_remove(vp, name, xflags);
2621 }
2622
2623 struct attrnames attr_system = {
2624         .attr_name      = "system.",
2625         .attr_namelen   = sizeof("system.") - 1,
2626         .attr_flag      = ATTR_SYSTEM,
2627         .attr_get       = attr_system_get,
2628         .attr_set       = attr_system_set,
2629         .attr_remove    = attr_system_remove,
2630         .attr_capable   = (attrcapable_t)fs_noerr,
2631 };
2632
2633 struct attrnames attr_trusted = {
2634         .attr_name      = "trusted.",
2635         .attr_namelen   = sizeof("trusted.") - 1,
2636         .attr_flag      = ATTR_ROOT,
2637         .attr_get       = attr_generic_get,
2638         .attr_set       = attr_generic_set,
2639         .attr_remove    = attr_generic_remove,
2640         .attr_capable   = attr_trusted_capable,
2641 };
2642
2643 struct attrnames attr_secure = {
2644         .attr_name      = "security.",
2645         .attr_namelen   = sizeof("security.") - 1,
2646         .attr_flag      = ATTR_SECURE,
2647         .attr_get       = attr_generic_get,
2648         .attr_set       = attr_generic_set,
2649         .attr_remove    = attr_generic_remove,
2650         .attr_capable   = attr_secure_capable,
2651 };
2652
2653 struct attrnames attr_user = {
2654         .attr_name      = "user.",
2655         .attr_namelen   = sizeof("user.") - 1,
2656         .attr_get       = attr_generic_get,
2657         .attr_set       = attr_generic_set,
2658         .attr_remove    = attr_generic_remove,
2659         .attr_capable   = attr_user_capable,
2660 };
2661
2662 struct attrnames *attr_namespaces[] =
2663         { &attr_system, &attr_trusted, &attr_secure, &attr_user };