fs/xfs/xfs_attr.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18
  19 #include <linux/capability.h>
  20
  21 #include "xfs.h"
  22 #include "xfs_fs.h"
  23 #include "xfs_types.h"
  24 #include "xfs_bit.h"
  25 #include "xfs_log.h"
  26 #include "xfs_inum.h"
  27 #include "xfs_trans.h"
  28 #include "xfs_sb.h"
  29 #include "xfs_ag.h"
  30 #include "xfs_dir2.h"
  31 #include "xfs_dmapi.h"
  32 #include "xfs_mount.h"
  33 #include "xfs_da_btree.h"
  34 #include "xfs_bmap_btree.h"
  35 #include "xfs_alloc_btree.h"
  36 #include "xfs_ialloc_btree.h"
  37 #include "xfs_dir2_sf.h"
  38 #include "xfs_attr_sf.h"
  39 #include "xfs_dinode.h"
  40 #include "xfs_inode.h"
  41 #include "xfs_alloc.h"
  42 #include "xfs_btree.h"
  43 #include "xfs_inode_item.h"
  44 #include "xfs_bmap.h"
  45 #include "xfs_attr.h"
  46 #include "xfs_attr_leaf.h"
  47 #include "xfs_error.h"
  48 #include "xfs_quota.h"
  49 #include "xfs_trans_space.h"
  50 #include "xfs_acl.h"
  51 #include "xfs_rw.h"
  52
  53 /*
  54  * xfs_attr.c
  55  *
  56  * Provide the external interfaces to manage attribute lists.
  57  */
  58
  59 #define ATTR_SYSCOUNT   2
  60 static struct attrnames posix_acl_access;
  61 static struct attrnames posix_acl_default;
  62 static struct attrnames *attr_system_names[ATTR_SYSCOUNT];
  63
  64 /*========================================================================
  65  * Function prototypes for the kernel.
  66  *========================================================================*/
  67
  68 /*
  69  * Internal routines when attribute list fits inside the inode.
  70  */
  71 STATIC int xfs_attr_shortform_addname(xfs_da_args_t *args);
  72
  73 /*
  74  * Internal routines when attribute list is one block.
  75  */
  76 STATIC int xfs_attr_leaf_get(xfs_da_args_t *args);
  77 STATIC int xfs_attr_leaf_addname(xfs_da_args_t *args);
  78 STATIC int xfs_attr_leaf_removename(xfs_da_args_t *args);
  79 STATIC int xfs_attr_leaf_list(xfs_attr_list_context_t *context);
  80
  81 /*
  82  * Internal routines when attribute list is more than one block.
  83  */
  84 STATIC int xfs_attr_node_get(xfs_da_args_t *args);
  85 STATIC int xfs_attr_node_addname(xfs_da_args_t *args);
  86 STATIC int xfs_attr_node_removename(xfs_da_args_t *args);
  87 STATIC int xfs_attr_node_list(xfs_attr_list_context_t *context);
  88 STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
  89 STATIC int xfs_attr_refillstate(xfs_da_state_t *state);
  90
  91 /*
  92  * Routines to manipulate out-of-line attribute values.
  93  */
  94 STATIC int xfs_attr_rmtval_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         /*
 121          * Fill in the arg structure for this request.
 122          */
 123         memset((char *)&args, 0, sizeof(args));
 124         args.name = name;
 125         args.namelen = namelen;
 126         args.value = value;
 127         args.valuelen = *valuelenp;
 128         args.flags = flags;
 129         args.hashval = xfs_da_hashname(args.name, args.namelen);
 130         args.dp = ip;
 131         args.whichfork = XFS_ATTR_FORK;
 132
 133         /*
 134          * Decide on what work routines to call based on the inode size.
 135          */
 136         if (XFS_IFORK_Q(ip) == 0 ||
 137             (ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 138              ip->i_d.di_anextents == 0)) {
 139                 error = XFS_ERROR(ENOATTR);
 140         } else if (ip->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
 141                 error = xfs_attr_shortform_getvalue(&args);
 142         } else if (xfs_bmap_one_block(ip, XFS_ATTR_FORK)) {
 143                 error = xfs_attr_leaf_get(&args);
 144         } else {
 145                 error = xfs_attr_node_get(&args);
 146         }
 147
 148         /*
 149          * Return the number of bytes in the value to the caller.
 150          */
 151         *valuelenp = args.valuelen;
 152
 153         if (error == EEXIST)
 154                 error = 0;
 155         return(error);
 156 }
 157
 158 int
 159 xfs_attr_get(
 160         xfs_inode_t     *ip,
 161         const char      *name,
 162         char            *value,
 163         int             *valuelenp,
 164         int             flags,
 165         cred_t          *cred)
 166 {
 167         int             error, namelen;
 168
 169         XFS_STATS_INC(xs_attr_get);
 170
 171         if (!name)
 172                 return(EINVAL);
 173         namelen = strlen(name);
 174         if (namelen >= MAXNAMELEN)
 175                 return(EFAULT);         /* match IRIX behaviour */
 176
 177         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 178                 return(EIO);
 179
 180         xfs_ilock(ip, XFS_ILOCK_SHARED);
 181         error = xfs_attr_fetch(ip, name, namelen, value, valuelenp, flags, cred);
 182         xfs_iunlock(ip, XFS_ILOCK_SHARED);
 183         return(error);
 184 }
 185
 186 int
 187 xfs_attr_set_int(xfs_inode_t *dp, const char *name, int namelen,
 188                  char *value, int valuelen, int flags)
 189 {
 190         xfs_da_args_t   args;
 191         xfs_fsblock_t   firstblock;
 192         xfs_bmap_free_t flist;
 193         int             error, err2, committed;
 194         int             local, size;
 195         uint            nblks;
 196         xfs_mount_t     *mp = dp->i_mount;
 197         int             rsvd = (flags & ATTR_ROOT) != 0;
 198
 199         /*
 200          * Attach the dquots to the inode.
 201          */
 202         if ((error = XFS_QM_DQATTACH(mp, dp, 0)))
 203                 return (error);
 204
 205         /*
 206          * If the inode doesn't have an attribute fork, add one.
 207          * (inode must not be locked when we call this routine)
 208          */
 209         if (XFS_IFORK_Q(dp) == 0) {
 210                 int sf_size = sizeof(xfs_attr_sf_hdr_t) +
 211                               XFS_ATTR_SF_ENTSIZE_BYNAME(namelen, valuelen);
 212
 213                 if ((error = xfs_bmap_add_attrfork(dp, sf_size, rsvd)))
 214                         return(error);
 215         }
 216
 217         /*
 218          * Fill in the arg structure for this request.
 219          */
 220         memset((char *)&args, 0, sizeof(args));
 221         args.name = name;
 222         args.namelen = namelen;
 223         args.value = value;
 224         args.valuelen = valuelen;
 225         args.flags = flags;
 226         args.hashval = xfs_da_hashname(args.name, args.namelen);
 227         args.dp = dp;
 228         args.firstblock = &firstblock;
 229         args.flist = &flist;
 230         args.whichfork = XFS_ATTR_FORK;
 231         args.addname = 1;
 232         args.oknoent = 1;
 233
 234         /*
 235          * Determine space new attribute will use, and if it would be
 236          * "local" or "remote" (note: local != inline).
 237          */
 238         size = xfs_attr_leaf_newentsize(namelen, valuelen,
 239                                         mp->m_sb.sb_blocksize, &local);
 240
 241         nblks = XFS_DAENTER_SPACE_RES(mp, XFS_ATTR_FORK);
 242         if (local) {
 243                 if (size > (mp->m_sb.sb_blocksize >> 1)) {
 244                         /* Double split possible */
 245                         nblks <<= 1;
 246                 }
 247         } else {
 248                 uint    dblocks = XFS_B_TO_FSB(mp, valuelen);
 249                 /* Out of line attribute, cannot double split, but make
 250                  * room for the attribute value itself.
 251                  */
 252                 nblks += dblocks;
 253                 nblks += XFS_NEXTENTADD_SPACE_RES(mp, dblocks, XFS_ATTR_FORK);
 254         }
 255
 256         /* Size is now blocks for attribute data */
 257         args.total = nblks;
 258
 259         /*
 260          * Start our first transaction of the day.
 261          *
 262          * All future transactions during this code must be "chained" off
 263          * this one via the trans_dup() call.  All transactions will contain
 264          * the inode, and the inode will always be marked with trans_ihold().
 265          * Since the inode will be locked in all transactions, we must log
 266          * the inode in every transaction to let it float upward through
 267          * the log.
 268          */
 269         args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_SET);
 270
 271         /*
 272          * Root fork attributes can use reserved data blocks for this
 273          * operation if necessary
 274          */
 275
 276         if (rsvd)
 277                 args.trans->t_flags |= XFS_TRANS_RESERVE;
 278
 279         if ((error = xfs_trans_reserve(args.trans, (uint) nblks,
 280                                       XFS_ATTRSET_LOG_RES(mp, nblks),
 281                                       0, XFS_TRANS_PERM_LOG_RES,
 282                                       XFS_ATTRSET_LOG_COUNT))) {
 283                 xfs_trans_cancel(args.trans, 0);
 284                 return(error);
 285         }
 286         xfs_ilock(dp, XFS_ILOCK_EXCL);
 287
 288         error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, args.trans, dp, nblks, 0,
 289                          rsvd ? XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
 290                                 XFS_QMOPT_RES_REGBLKS);
 291         if (error) {
 292                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
 293                 xfs_trans_cancel(args.trans, XFS_TRANS_RELEASE_LOG_RES);
 294                 return (error);
 295         }
 296
 297         xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
 298         xfs_trans_ihold(args.trans, dp);
 299
 300         /*
 301          * If the attribute list is non-existent or a shortform list,
 302          * upgrade it to a single-leaf-block attribute list.
 303          */
 304         if ((dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
 305             ((dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) &&
 306              (dp->i_d.di_anextents == 0))) {
 307
 308                 /*
 309                  * Build initial attribute list (if required).
 310                  */
 311                 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS)
 312                         xfs_attr_shortform_create(&args);
 313
 314                 /*
 315                  * Try to add the attr to the attribute list in
 316                  * the inode.
 317                  */
 318                 error = xfs_attr_shortform_addname(&args);
 319                 if (error != ENOSPC) {
 320                         /*
 321                          * Commit the shortform mods, and we're done.
 322                          * NOTE: this is also the error path (EEXIST, etc).
 323                          */
 324                         ASSERT(args.trans != NULL);
 325
 326                         /*
 327                          * If this is a synchronous mount, make sure that
 328                          * the transaction goes to disk before returning
 329                          * to the user.
 330                          */
 331                         if (mp->m_flags & XFS_MOUNT_WSYNC) {
 332                                 xfs_trans_set_sync(args.trans);
 333                         }
 334                         err2 = xfs_trans_commit(args.trans,
 335                                                  XFS_TRANS_RELEASE_LOG_RES);
 336                         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 337
 338                         /*
 339                          * Hit the inode change time.
 340                          */
 341                         if (!error && (flags & ATTR_KERNOTIME) == 0) {
 342                                 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
 343                         }
 344                         return(error == 0 ? err2 : error);
 345                 }
 346
 347                 /*
 348                  * It won't fit in the shortform, transform to a leaf block.
 349                  * GROT: another possible req'mt for a double-split btree op.
 350                  */
 351                 XFS_BMAP_INIT(args.flist, args.firstblock);
 352                 error = xfs_attr_shortform_to_leaf(&args);
 353                 if (!error) {
 354                         error = xfs_bmap_finish(&args.trans, args.flist,
 355                                                 &committed);
 356                 }
 357                 if (error) {
 358                         ASSERT(committed);
 359                         args.trans = NULL;
 360                         xfs_bmap_cancel(&flist);
 361                         goto out;
 362                 }
 363
 364                 /*
 365                  * bmap_finish() may have committed the last trans and started
 366                  * a new one.  We need the inode to be in all transactions.
 367                  */
 368                 if (committed) {
 369                         xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
 370                         xfs_trans_ihold(args.trans, dp);
 371                 }
 372
 373                 /*
 374                  * Commit the leaf transformation.  We'll need another (linked)
 375                  * transaction to add the new attribute to the leaf.
 376                  */
 377                 if ((error = xfs_attr_rolltrans(&args.trans, dp)))
 378                         goto out;
 379
 380         }
 381
 382         if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
 383                 error = xfs_attr_leaf_addname(&args);
 384         } else {
 385                 error = xfs_attr_node_addname(&args);
 386         }
 387         if (error) {
 388                 goto out;
 389         }
 390
 391         /*
 392          * If this is a synchronous mount, make sure that the
 393          * transaction goes to disk before returning to the user.
 394          */
 395         if (mp->m_flags & XFS_MOUNT_WSYNC) {
 396                 xfs_trans_set_sync(args.trans);
 397         }
 398
 399         /*
 400          * Commit the last in the sequence of transactions.
 401          */
 402         xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
 403         error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES);
 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(
 425         xfs_inode_t     *dp,
 426         const char      *name,
 427         char            *value,
 428         int             valuelen,
 429         int             flags)
 430 {
 431         int             namelen;
 432
 433         namelen = strlen(name);
 434         if (namelen >= MAXNAMELEN)
 435                 return EFAULT;          /* match IRIX behaviour */
 436
 437         XFS_STATS_INC(xs_attr_set);
 438
 439         if (XFS_FORCED_SHUTDOWN(dp->i_mount))
 440                 return (EIO);
 441
 442         return xfs_attr_set_int(dp, name, namelen, value, valuelen, flags);
 443 }
 444
 445 /*
 446  * Generic handler routine to remove a name from an attribute list.
 447  * Transitions attribute list from Btree to shortform as necessary.
 448  */
 449 int
 450 xfs_attr_remove_int(xfs_inode_t *dp, const char *name, int namelen, int flags)
 451 {
 452         xfs_da_args_t   args;
 453         xfs_fsblock_t   firstblock;
 454         xfs_bmap_free_t flist;
 455         int             error;
 456         xfs_mount_t     *mp = dp->i_mount;
 457
 458         /*
 459          * Fill in the arg structure for this request.
 460          */
 461         memset((char *)&args, 0, sizeof(args));
 462         args.name = name;
 463         args.namelen = namelen;
 464         args.flags = flags;
 465         args.hashval = xfs_da_hashname(args.name, args.namelen);
 466         args.dp = dp;
 467         args.firstblock = &firstblock;
 468         args.flist = &flist;
 469         args.total = 0;
 470         args.whichfork = XFS_ATTR_FORK;
 471
 472         /*
 473          * Attach the dquots to the inode.
 474          */
 475         if ((error = XFS_QM_DQATTACH(mp, dp, 0)))
 476                 return (error);
 477
 478         /*
 479          * Start our first transaction of the day.
 480          *
 481          * All future transactions during this code must be "chained" off
 482          * this one via the trans_dup() call.  All transactions will contain
 483          * the inode, and the inode will always be marked with trans_ihold().
 484          * Since the inode will be locked in all transactions, we must log
 485          * the inode in every transaction to let it float upward through
 486          * the log.
 487          */
 488         args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_RM);
 489
 490         /*
 491          * Root fork attributes can use reserved data blocks for this
 492          * operation if necessary
 493          */
 494
 495         if (flags & ATTR_ROOT)
 496                 args.trans->t_flags |= XFS_TRANS_RESERVE;
 497
 498         if ((error = xfs_trans_reserve(args.trans,
 499                                       XFS_ATTRRM_SPACE_RES(mp),
 500                                       XFS_ATTRRM_LOG_RES(mp),
 501                                       0, XFS_TRANS_PERM_LOG_RES,
 502                                       XFS_ATTRRM_LOG_COUNT))) {
 503                 xfs_trans_cancel(args.trans, 0);
 504                 return(error);
 505         }
 506
 507         xfs_ilock(dp, XFS_ILOCK_EXCL);
 508         /*
 509          * No need to make quota reservations here. We expect to release some
 510          * blocks not allocate in the common case.
 511          */
 512         xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
 513         xfs_trans_ihold(args.trans, dp);
 514
 515         /*
 516          * Decide on what work routines to call based on the inode size.
 517          */
 518         if (XFS_IFORK_Q(dp) == 0 ||
 519             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 520              dp->i_d.di_anextents == 0)) {
 521                 error = XFS_ERROR(ENOATTR);
 522                 goto out;
 523         }
 524         if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
 525                 ASSERT(dp->i_afp->if_flags & XFS_IFINLINE);
 526                 error = xfs_attr_shortform_remove(&args);
 527                 if (error) {
 528                         goto out;
 529                 }
 530         } else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
 531                 error = xfs_attr_leaf_removename(&args);
 532         } else {
 533                 error = xfs_attr_node_removename(&args);
 534         }
 535         if (error) {
 536                 goto out;
 537         }
 538
 539         /*
 540          * If this is a synchronous mount, make sure that the
 541          * transaction goes to disk before returning to the user.
 542          */
 543         if (mp->m_flags & XFS_MOUNT_WSYNC) {
 544                 xfs_trans_set_sync(args.trans);
 545         }
 546
 547         /*
 548          * Commit the last in the sequence of transactions.
 549          */
 550         xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
 551         error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES);
 552         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 553
 554         /*
 555          * Hit the inode change time.
 556          */
 557         if (!error && (flags & ATTR_KERNOTIME) == 0) {
 558                 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
 559         }
 560
 561         return(error);
 562
 563 out:
 564         if (args.trans)
 565                 xfs_trans_cancel(args.trans,
 566                         XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
 567         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 568         return(error);
 569 }
 570
 571 int
 572 xfs_attr_remove(
 573         xfs_inode_t     *dp,
 574         const char      *name,
 575         int             flags)
 576 {
 577         int             namelen;
 578
 579         namelen = strlen(name);
 580         if (namelen >= MAXNAMELEN)
 581                 return EFAULT;          /* match IRIX behaviour */
 582
 583         XFS_STATS_INC(xs_attr_remove);
 584
 585         if (XFS_FORCED_SHUTDOWN(dp->i_mount))
 586                 return (EIO);
 587
 588         xfs_ilock(dp, XFS_ILOCK_SHARED);
 589         if (XFS_IFORK_Q(dp) == 0 ||
 590                    (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 591                     dp->i_d.di_anextents == 0)) {
 592                 xfs_iunlock(dp, XFS_ILOCK_SHARED);
 593                 return(XFS_ERROR(ENOATTR));
 594         }
 595         xfs_iunlock(dp, XFS_ILOCK_SHARED);
 596
 597         return xfs_attr_remove_int(dp, name, namelen, flags);
 598 }
 599
 600 int                                                             /* error */
 601 xfs_attr_list_int(xfs_attr_list_context_t *context)
 602 {
 603         int error;
 604         xfs_inode_t *dp = context->dp;
 605
 606         /*
 607          * Decide on what work routines to call based on the inode size.
 608          */
 609         if (XFS_IFORK_Q(dp) == 0 ||
 610             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 611              dp->i_d.di_anextents == 0)) {
 612                 error = 0;
 613         } else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
 614                 error = xfs_attr_shortform_list(context);
 615         } else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
 616                 error = xfs_attr_leaf_list(context);
 617         } else {
 618                 error = xfs_attr_node_list(context);
 619         }
 620         return error;
 621 }
 622
 623 #define ATTR_ENTBASESIZE                /* minimum bytes used by an attr */ \
 624         (((struct attrlist_ent *) 0)->a_name - (char *) 0)
 625 #define ATTR_ENTSIZE(namelen)           /* actual bytes used by an attr */ \
 626         ((ATTR_ENTBASESIZE + (namelen) + 1 + sizeof(u_int32_t)-1) \
 627          & ~(sizeof(u_int32_t)-1))
 628
 629 /*
 630  * Format an attribute and copy it out to the user's buffer.
 631  * Take care to check values and protect against them changing later,
 632  * we may be reading them directly out of a user buffer.
 633  */
 634 /*ARGSUSED*/
 635 STATIC int
 636 xfs_attr_put_listent(xfs_attr_list_context_t *context, attrnames_t *namesp,
 637                      char *name, int namelen,
 638                      int valuelen, char *value)
 639 {
 640         attrlist_ent_t *aep;
 641         int arraytop;
 642
 643         ASSERT(!(context->flags & ATTR_KERNOVAL));
 644         ASSERT(context->count >= 0);
 645         ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
 646         ASSERT(context->firstu >= sizeof(*context->alist));
 647         ASSERT(context->firstu <= context->bufsize);
 648
 649         arraytop = sizeof(*context->alist) +
 650                         context->count * sizeof(context->alist->al_offset[0]);
 651         context->firstu -= ATTR_ENTSIZE(namelen);
 652         if (context->firstu < arraytop) {
 653                 xfs_attr_trace_l_c("buffer full", context);
 654                 context->alist->al_more = 1;
 655                 context->seen_enough = 1;
 656                 return 1;
 657         }
 658
 659         aep = (attrlist_ent_t *)&(((char *)context->alist)[ context->firstu ]);
 660         aep->a_valuelen = valuelen;
 661         memcpy(aep->a_name, name, namelen);
 662         aep->a_name[ namelen ] = 0;
 663         context->alist->al_offset[ context->count++ ] = context->firstu;
 664         context->alist->al_count = context->count;
 665         xfs_attr_trace_l_c("add", context);
 666         return 0;
 667 }
 668
 669 STATIC int
 670 xfs_attr_kern_list(xfs_attr_list_context_t *context, attrnames_t *namesp,
 671                      char *name, int namelen,
 672                      int valuelen, char *value)
 673 {
 674         char *offset;
 675         int arraytop;
 676
 677         ASSERT(context->count >= 0);
 678
 679         arraytop = context->count + namesp->attr_namelen + namelen + 1;
 680         if (arraytop > context->firstu) {
 681                 context->count = -1;    /* insufficient space */
 682                 return 1;
 683         }
 684         offset = (char *)context->alist + context->count;
 685         strncpy(offset, namesp->attr_name, namesp->attr_namelen);
 686         offset += namesp->attr_namelen;
 687         strncpy(offset, name, namelen);                 /* real name */
 688         offset += namelen;
 689         *offset = '\0';
 690         context->count += namesp->attr_namelen + namelen + 1;
 691         return 0;
 692 }
 693
 694 /*ARGSUSED*/
 695 STATIC int
 696 xfs_attr_kern_list_sizes(xfs_attr_list_context_t *context, attrnames_t *namesp,
 697                      char *name, int namelen,
 698                      int valuelen, char *value)
 699 {
 700         context->count += namesp->attr_namelen + namelen + 1;
 701         return 0;
 702 }
 703
 704 /*
 705  * Generate a list of extended attribute names and optionally
 706  * also value lengths.  Positive return value follows the XFS
 707  * convention of being an error, zero or negative return code
 708  * is the length of the buffer returned (negated), indicating
 709  * success.
 710  */
 711 int
 712 xfs_attr_list(
 713         xfs_inode_t     *dp,
 714         char            *buffer,
 715         int             bufsize,
 716         int             flags,
 717         attrlist_cursor_kern_t *cursor)
 718 {
 719         xfs_attr_list_context_t context;
 720         int error;
 721
 722         XFS_STATS_INC(xs_attr_list);
 723
 724         /*
 725          * Validate the cursor.
 726          */
 727         if (cursor->pad1 || cursor->pad2)
 728                 return(XFS_ERROR(EINVAL));
 729         if ((cursor->initted == 0) &&
 730             (cursor->hashval || cursor->blkno || cursor->offset))
 731                 return XFS_ERROR(EINVAL);
 732
 733         /*
 734          * Check for a properly aligned buffer.
 735          */
 736         if (((long)buffer) & (sizeof(int)-1))
 737                 return XFS_ERROR(EFAULT);
 738         if (flags & ATTR_KERNOVAL)
 739                 bufsize = 0;
 740
 741         /*
 742          * Initialize the output buffer.
 743          */
 744         context.dp = dp;
 745         context.cursor = cursor;
 746         context.count = 0;
 747         context.dupcnt = 0;
 748         context.resynch = 1;
 749         context.flags = flags;
 750         context.seen_enough = 0;
 751         context.alist = (attrlist_t *)buffer;
 752         context.put_value = 0;
 753
 754         if (flags & ATTR_KERNAMELS) {
 755                 context.bufsize = bufsize;
 756                 context.firstu = context.bufsize;
 757                 if (flags & ATTR_KERNOVAL)
 758                         context.put_listent = xfs_attr_kern_list_sizes;
 759                 else
 760                         context.put_listent = xfs_attr_kern_list;
 761         } else {
 762                 context.bufsize = (bufsize & ~(sizeof(int)-1));  /* align */
 763                 context.firstu = context.bufsize;
 764                 context.alist->al_count = 0;
 765                 context.alist->al_more = 0;
 766                 context.alist->al_offset[0] = context.bufsize;
 767                 context.put_listent = xfs_attr_put_listent;
 768         }
 769
 770         if (XFS_FORCED_SHUTDOWN(dp->i_mount))
 771                 return EIO;
 772
 773         xfs_ilock(dp, XFS_ILOCK_SHARED);
 774         xfs_attr_trace_l_c("syscall start", &context);
 775
 776         error = xfs_attr_list_int(&context);
 777
 778         xfs_iunlock(dp, XFS_ILOCK_SHARED);
 779         xfs_attr_trace_l_c("syscall end", &context);
 780
 781         if (context.flags & (ATTR_KERNOVAL|ATTR_KERNAMELS)) {
 782                 /* must return negated buffer size or the error */
 783                 if (context.count < 0)
 784                         error = XFS_ERROR(ERANGE);
 785                 else
 786                         error = -context.count;
 787         } else
 788                 ASSERT(error >= 0);
 789
 790         return error;
 791 }
 792
 793 int                                                             /* error */
 794 xfs_attr_inactive(xfs_inode_t *dp)
 795 {
 796         xfs_trans_t *trans;
 797         xfs_mount_t *mp;
 798         int error;
 799
 800         mp = dp->i_mount;
 801         ASSERT(! XFS_NOT_DQATTACHED(mp, dp));
 802
 803         xfs_ilock(dp, XFS_ILOCK_SHARED);
 804         if ((XFS_IFORK_Q(dp) == 0) ||
 805             (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
 806             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 807              dp->i_d.di_anextents == 0)) {
 808                 xfs_iunlock(dp, XFS_ILOCK_SHARED);
 809                 return(0);
 810         }
 811         xfs_iunlock(dp, XFS_ILOCK_SHARED);
 812
 813         /*
 814          * Start our first transaction of the day.
 815          *
 816          * All future transactions during this code must be "chained" off
 817          * this one via the trans_dup() call.  All transactions will contain
 818          * the inode, and the inode will always be marked with trans_ihold().
 819          * Since the inode will be locked in all transactions, we must log
 820          * the inode in every transaction to let it float upward through
 821          * the log.
 822          */
 823         trans = xfs_trans_alloc(mp, XFS_TRANS_ATTRINVAL);
 824         if ((error = xfs_trans_reserve(trans, 0, XFS_ATTRINVAL_LOG_RES(mp), 0,
 825                                       XFS_TRANS_PERM_LOG_RES,
 826                                       XFS_ATTRINVAL_LOG_COUNT))) {
 827                 xfs_trans_cancel(trans, 0);
 828                 return(error);
 829         }
 830         xfs_ilock(dp, XFS_ILOCK_EXCL);
 831
 832         /*
 833          * No need to make quota reservations here. We expect to release some
 834          * blocks, not allocate, in the common case.
 835          */
 836         xfs_trans_ijoin(trans, dp, XFS_ILOCK_EXCL);
 837         xfs_trans_ihold(trans, dp);
 838
 839         /*
 840          * Decide on what work routines to call based on the inode size.
 841          */
 842         if ((XFS_IFORK_Q(dp) == 0) ||
 843             (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
 844             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 845              dp->i_d.di_anextents == 0)) {
 846                 error = 0;
 847                 goto out;
 848         }
 849         error = xfs_attr_root_inactive(&trans, dp);
 850         if (error)
 851                 goto out;
 852         /*
 853          * signal synchronous inactive transactions unless this
 854          * is a synchronous mount filesystem in which case we
 855          * know that we're here because we've been called out of
 856          * xfs_inactive which means that the last reference is gone
 857          * and the unlink transaction has already hit the disk so
 858          * async inactive transactions are safe.
 859          */
 860         if ((error = xfs_itruncate_finish(&trans, dp, 0LL, XFS_ATTR_FORK,
 861                                 (!(mp->m_flags & XFS_MOUNT_WSYNC)
 862                                  ? 1 : 0))))
 863                 goto out;
 864
 865         /*
 866          * Commit the last in the sequence of transactions.
 867          */
 868         xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
 869         error = xfs_trans_commit(trans, XFS_TRANS_RELEASE_LOG_RES);
 870         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 871
 872         return(error);
 873
 874 out:
 875         xfs_trans_cancel(trans, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
 876         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 877         return(error);
 878 }
 879
 880
 881
 882 /*========================================================================
 883  * External routines when attribute list is inside the inode
 884  *========================================================================*/
 885
 886 /*
 887  * Add a name to the shortform attribute list structure
 888  * This is the external routine.
 889  */
 890 STATIC int
 891 xfs_attr_shortform_addname(xfs_da_args_t *args)
 892 {
 893         int newsize, forkoff, retval;
 894
 895         retval = xfs_attr_shortform_lookup(args);
 896         if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
 897                 return(retval);
 898         } else if (retval == EEXIST) {
 899                 if (args->flags & ATTR_CREATE)
 900                         return(retval);
 901                 retval = xfs_attr_shortform_remove(args);
 902                 ASSERT(retval == 0);
 903         }
 904
 905         if (args->namelen >= XFS_ATTR_SF_ENTSIZE_MAX ||
 906             args->valuelen >= XFS_ATTR_SF_ENTSIZE_MAX)
 907                 return(XFS_ERROR(ENOSPC));
 908
 909         newsize = XFS_ATTR_SF_TOTSIZE(args->dp);
 910         newsize += XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
 911
 912         forkoff = xfs_attr_shortform_bytesfit(args->dp, newsize);
 913         if (!forkoff)
 914                 return(XFS_ERROR(ENOSPC));
 915
 916         xfs_attr_shortform_add(args, forkoff);
 917         return(0);
 918 }
 919
 920
 921 /*========================================================================
 922  * External routines when attribute list is one block
 923  *========================================================================*/
 924
 925 /*
 926  * Add a name to the leaf attribute list structure
 927  *
 928  * This leaf block cannot have a "remote" value, we only call this routine
 929  * if bmap_one_block() says there is only one block (ie: no remote blks).
 930  */
 931 int
 932 xfs_attr_leaf_addname(xfs_da_args_t *args)
 933 {
 934         xfs_inode_t *dp;
 935         xfs_dabuf_t *bp;
 936         int retval, error, committed, forkoff;
 937
 938         /*
 939          * Read the (only) block in the attribute list in.
 940          */
 941         dp = args->dp;
 942         args->blkno = 0;
 943         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
 944                                              XFS_ATTR_FORK);
 945         if (error)
 946                 return(error);
 947         ASSERT(bp != NULL);
 948
 949         /*
 950          * Look up the given attribute in the leaf block.  Figure out if
 951          * the given flags produce an error or call for an atomic rename.
 952          */
 953         retval = xfs_attr_leaf_lookup_int(bp, args);
 954         if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
 955                 xfs_da_brelse(args->trans, bp);
 956                 return(retval);
 957         } else if (retval == EEXIST) {
 958                 if (args->flags & ATTR_CREATE) {        /* pure create op */
 959                         xfs_da_brelse(args->trans, bp);
 960                         return(retval);
 961                 }
 962                 args->rename = 1;                       /* an atomic rename */
 963                 args->blkno2 = args->blkno;             /* set 2nd entry info*/
 964                 args->index2 = args->index;
 965                 args->rmtblkno2 = args->rmtblkno;
 966                 args->rmtblkcnt2 = args->rmtblkcnt;
 967         }
 968
 969         /*
 970          * Add the attribute to the leaf block, transitioning to a Btree
 971          * if required.
 972          */
 973         retval = xfs_attr_leaf_add(bp, args);
 974         xfs_da_buf_done(bp);
 975         if (retval == ENOSPC) {
 976                 /*
 977                  * Promote the attribute list to the Btree format, then
 978                  * Commit that transaction so that the node_addname() call
 979                  * can manage its own transactions.
 980                  */
 981                 XFS_BMAP_INIT(args->flist, args->firstblock);
 982                 error = xfs_attr_leaf_to_node(args);
 983                 if (!error) {
 984                         error = xfs_bmap_finish(&args->trans, args->flist,
 985                                                 &committed);
 986                 }
 987                 if (error) {
 988                         ASSERT(committed);
 989                         args->trans = NULL;
 990                         xfs_bmap_cancel(args->flist);
 991                         return(error);
 992                 }
 993
 994                 /*
 995                  * bmap_finish() may have committed the last trans and started
 996                  * a new one.  We need the inode to be in all transactions.
 997                  */
 998                 if (committed) {
 999                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1000                         xfs_trans_ihold(args->trans, dp);
1001                 }
1002
1003                 /*
1004                  * Commit the current trans (including the inode) and start
1005                  * a new one.
1006                  */
1007                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1008                         return (error);
1009
1010                 /*
1011                  * Fob the whole rest of the problem off on the Btree code.
1012                  */
1013                 error = xfs_attr_node_addname(args);
1014                 return(error);
1015         }
1016
1017         /*
1018          * Commit the transaction that added the attr name so that
1019          * later routines can manage their own transactions.
1020          */
1021         if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1022                 return (error);
1023
1024         /*
1025          * If there was an out-of-line value, allocate the blocks we
1026          * identified for its storage and copy the value.  This is done
1027          * after we create the attribute so that we don't overflow the
1028          * maximum size of a transaction and/or hit a deadlock.
1029          */
1030         if (args->rmtblkno > 0) {
1031                 error = xfs_attr_rmtval_set(args);
1032                 if (error)
1033                         return(error);
1034         }
1035
1036         /*
1037          * If this is an atomic rename operation, we must "flip" the
1038          * incomplete flags on the "new" and "old" attribute/value pairs
1039          * so that one disappears and one appears atomically.  Then we
1040          * must remove the "old" attribute/value pair.
1041          */
1042         if (args->rename) {
1043                 /*
1044                  * In a separate transaction, set the incomplete flag on the
1045                  * "old" attr and clear the incomplete flag on the "new" attr.
1046                  */
1047                 error = xfs_attr_leaf_flipflags(args);
1048                 if (error)
1049                         return(error);
1050
1051                 /*
1052                  * Dismantle the "old" attribute/value pair by removing
1053                  * a "remote" value (if it exists).
1054                  */
1055                 args->index = args->index2;
1056                 args->blkno = args->blkno2;
1057                 args->rmtblkno = args->rmtblkno2;
1058                 args->rmtblkcnt = args->rmtblkcnt2;
1059                 if (args->rmtblkno) {
1060                         error = xfs_attr_rmtval_remove(args);
1061                         if (error)
1062                                 return(error);
1063                 }
1064
1065                 /*
1066                  * Read in the block containing the "old" attr, then
1067                  * remove the "old" attr from that block (neat, huh!)
1068                  */
1069                 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1,
1070                                                      &bp, XFS_ATTR_FORK);
1071                 if (error)
1072                         return(error);
1073                 ASSERT(bp != NULL);
1074                 (void)xfs_attr_leaf_remove(bp, args);
1075
1076                 /*
1077                  * If the result is small enough, shrink it all into the inode.
1078                  */
1079                 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1080                         XFS_BMAP_INIT(args->flist, args->firstblock);
1081                         error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1082                         /* bp is gone due to xfs_da_shrink_inode */
1083                         if (!error) {
1084                                 error = xfs_bmap_finish(&args->trans,
1085                                                         args->flist,
1086                                                         &committed);
1087                         }
1088                         if (error) {
1089                                 ASSERT(committed);
1090                                 args->trans = NULL;
1091                                 xfs_bmap_cancel(args->flist);
1092                                 return(error);
1093                         }
1094
1095                         /*
1096                          * bmap_finish() may have committed the last trans
1097                          * and started a new one.  We need the inode to be
1098                          * in all transactions.
1099                          */
1100                         if (committed) {
1101                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1102                                 xfs_trans_ihold(args->trans, dp);
1103                         }
1104                 } else
1105                         xfs_da_buf_done(bp);
1106
1107                 /*
1108                  * Commit the remove and start the next trans in series.
1109                  */
1110                 error = xfs_attr_rolltrans(&args->trans, dp);
1111
1112         } else if (args->rmtblkno > 0) {
1113                 /*
1114                  * Added a "remote" value, just clear the incomplete flag.
1115                  */
1116                 error = xfs_attr_leaf_clearflag(args);
1117         }
1118         return(error);
1119 }
1120
1121 /*
1122  * Remove a name from the leaf attribute list structure
1123  *
1124  * This leaf block cannot have a "remote" value, we only call this routine
1125  * if bmap_one_block() says there is only one block (ie: no remote blks).
1126  */
1127 STATIC int
1128 xfs_attr_leaf_removename(xfs_da_args_t *args)
1129 {
1130         xfs_inode_t *dp;
1131         xfs_dabuf_t *bp;
1132         int error, committed, forkoff;
1133
1134         /*
1135          * Remove the attribute.
1136          */
1137         dp = args->dp;
1138         args->blkno = 0;
1139         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1140                                              XFS_ATTR_FORK);
1141         if (error) {
1142                 return(error);
1143         }
1144
1145         ASSERT(bp != NULL);
1146         error = xfs_attr_leaf_lookup_int(bp, args);
1147         if (error == ENOATTR) {
1148                 xfs_da_brelse(args->trans, bp);
1149                 return(error);
1150         }
1151
1152         (void)xfs_attr_leaf_remove(bp, args);
1153
1154         /*
1155          * If the result is small enough, shrink it all into the inode.
1156          */
1157         if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1158                 XFS_BMAP_INIT(args->flist, args->firstblock);
1159                 error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1160                 /* bp is gone due to xfs_da_shrink_inode */
1161                 if (!error) {
1162                         error = xfs_bmap_finish(&args->trans, args->flist,
1163                                                 &committed);
1164                 }
1165                 if (error) {
1166                         ASSERT(committed);
1167                         args->trans = NULL;
1168                         xfs_bmap_cancel(args->flist);
1169                         return(error);
1170                 }
1171
1172                 /*
1173                  * bmap_finish() may have committed the last trans and started
1174                  * a new one.  We need the inode to be in all transactions.
1175                  */
1176                 if (committed) {
1177                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1178                         xfs_trans_ihold(args->trans, dp);
1179                 }
1180         } else
1181                 xfs_da_buf_done(bp);
1182         return(0);
1183 }
1184
1185 /*
1186  * Look up a name in a leaf attribute list structure.
1187  *
1188  * This leaf block cannot have a "remote" value, we only call this routine
1189  * if bmap_one_block() says there is only one block (ie: no remote blks).
1190  */
1191 STATIC int
1192 xfs_attr_leaf_get(xfs_da_args_t *args)
1193 {
1194         xfs_dabuf_t *bp;
1195         int error;
1196
1197         args->blkno = 0;
1198         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1199                                              XFS_ATTR_FORK);
1200         if (error)
1201                 return(error);
1202         ASSERT(bp != NULL);
1203
1204         error = xfs_attr_leaf_lookup_int(bp, args);
1205         if (error != EEXIST)  {
1206                 xfs_da_brelse(args->trans, bp);
1207                 return(error);
1208         }
1209         error = xfs_attr_leaf_getvalue(bp, args);
1210         xfs_da_brelse(args->trans, bp);
1211         if (!error && (args->rmtblkno > 0) && !(args->flags & ATTR_KERNOVAL)) {
1212                 error = xfs_attr_rmtval_get(args);
1213         }
1214         return(error);
1215 }
1216
1217 /*
1218  * Copy out attribute entries for attr_list(), for leaf attribute lists.
1219  */
1220 STATIC int
1221 xfs_attr_leaf_list(xfs_attr_list_context_t *context)
1222 {
1223         xfs_attr_leafblock_t *leaf;
1224         int error;
1225         xfs_dabuf_t *bp;
1226
1227         context->cursor->blkno = 0;
1228         error = xfs_da_read_buf(NULL, context->dp, 0, -1, &bp, XFS_ATTR_FORK);
1229         if (error)
1230                 return XFS_ERROR(error);
1231         ASSERT(bp != NULL);
1232         leaf = bp->data;
1233         if (unlikely(be16_to_cpu(leaf->hdr.info.magic) != XFS_ATTR_LEAF_MAGIC)) {
1234                 XFS_CORRUPTION_ERROR("xfs_attr_leaf_list", XFS_ERRLEVEL_LOW,
1235                                      context->dp->i_mount, leaf);
1236                 xfs_da_brelse(NULL, bp);
1237                 return XFS_ERROR(EFSCORRUPTED);
1238         }
1239
1240         error = xfs_attr_leaf_list_int(bp, context);
1241         xfs_da_brelse(NULL, bp);
1242         return XFS_ERROR(error);
1243 }
1244
1245
1246 /*========================================================================
1247  * External routines when attribute list size > XFS_LBSIZE(mp).
1248  *========================================================================*/
1249
1250 /*
1251  * Add a name to a Btree-format attribute list.
1252  *
1253  * This will involve walking down the Btree, and may involve splitting
1254  * leaf nodes and even splitting intermediate nodes up to and including
1255  * the root node (a special case of an intermediate node).
1256  *
1257  * "Remote" attribute values confuse the issue and atomic rename operations
1258  * add a whole extra layer of confusion on top of that.
1259  */
1260 STATIC int
1261 xfs_attr_node_addname(xfs_da_args_t *args)
1262 {
1263         xfs_da_state_t *state;
1264         xfs_da_state_blk_t *blk;
1265         xfs_inode_t *dp;
1266         xfs_mount_t *mp;
1267         int committed, retval, error;
1268
1269         /*
1270          * Fill in bucket of arguments/results/context to carry around.
1271          */
1272         dp = args->dp;
1273         mp = dp->i_mount;
1274 restart:
1275         state = xfs_da_state_alloc();
1276         state->args = args;
1277         state->mp = mp;
1278         state->blocksize = state->mp->m_sb.sb_blocksize;
1279         state->node_ents = state->mp->m_attr_node_ents;
1280
1281         /*
1282          * Search to see if name already exists, and get back a pointer
1283          * to where it should go.
1284          */
1285         error = xfs_da_node_lookup_int(state, &retval);
1286         if (error)
1287                 goto out;
1288         blk = &state->path.blk[ state->path.active-1 ];
1289         ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1290         if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
1291                 goto out;
1292         } else if (retval == EEXIST) {
1293                 if (args->flags & ATTR_CREATE)
1294                         goto out;
1295                 args->rename = 1;                       /* atomic rename op */
1296                 args->blkno2 = args->blkno;             /* set 2nd entry info*/
1297                 args->index2 = args->index;
1298                 args->rmtblkno2 = args->rmtblkno;
1299                 args->rmtblkcnt2 = args->rmtblkcnt;
1300                 args->rmtblkno = 0;
1301                 args->rmtblkcnt = 0;
1302         }
1303
1304         retval = xfs_attr_leaf_add(blk->bp, state->args);
1305         if (retval == ENOSPC) {
1306                 if (state->path.active == 1) {
1307                         /*
1308                          * Its really a single leaf node, but it had
1309                          * out-of-line values so it looked like it *might*
1310                          * have been a b-tree.
1311                          */
1312                         xfs_da_state_free(state);
1313                         XFS_BMAP_INIT(args->flist, args->firstblock);
1314                         error = xfs_attr_leaf_to_node(args);
1315                         if (!error) {
1316                                 error = xfs_bmap_finish(&args->trans,
1317                                                         args->flist,
1318                                                         &committed);
1319                         }
1320                         if (error) {
1321                                 ASSERT(committed);
1322                                 args->trans = NULL;
1323                                 xfs_bmap_cancel(args->flist);
1324                                 goto out;
1325                         }
1326
1327                         /*
1328                          * bmap_finish() may have committed the last trans
1329                          * and started a new one.  We need the inode to be
1330                          * in all transactions.
1331                          */
1332                         if (committed) {
1333                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1334                                 xfs_trans_ihold(args->trans, dp);
1335                         }
1336
1337                         /*
1338                          * Commit the node conversion and start the next
1339                          * trans in the chain.
1340                          */
1341                         if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1342                                 goto out;
1343
1344                         goto restart;
1345                 }
1346
1347                 /*
1348                  * Split as many Btree elements as required.
1349                  * This code tracks the new and old attr's location
1350                  * in the index/blkno/rmtblkno/rmtblkcnt fields and
1351                  * in the index2/blkno2/rmtblkno2/rmtblkcnt2 fields.
1352                  */
1353                 XFS_BMAP_INIT(args->flist, args->firstblock);
1354                 error = xfs_da_split(state);
1355                 if (!error) {
1356                         error = xfs_bmap_finish(&args->trans, args->flist,
1357                                                 &committed);
1358                 }
1359                 if (error) {
1360                         ASSERT(committed);
1361                         args->trans = NULL;
1362                         xfs_bmap_cancel(args->flist);
1363                         goto out;
1364                 }
1365
1366                 /*
1367                  * bmap_finish() may have committed the last trans and started
1368                  * a new one.  We need the inode to be in all transactions.
1369                  */
1370                 if (committed) {
1371                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1372                         xfs_trans_ihold(args->trans, dp);
1373                 }
1374         } else {
1375                 /*
1376                  * Addition succeeded, update Btree hashvals.
1377                  */
1378                 xfs_da_fixhashpath(state, &state->path);
1379         }
1380
1381         /*
1382          * Kill the state structure, we're done with it and need to
1383          * allow the buffers to come back later.
1384          */
1385         xfs_da_state_free(state);
1386         state = NULL;
1387
1388         /*
1389          * Commit the leaf addition or btree split and start the next
1390          * trans in the chain.
1391          */
1392         if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1393                 goto out;
1394
1395         /*
1396          * If there was an out-of-line value, allocate the blocks we
1397          * identified for its storage and copy the value.  This is done
1398          * after we create the attribute so that we don't overflow the
1399          * maximum size of a transaction and/or hit a deadlock.
1400          */
1401         if (args->rmtblkno > 0) {
1402                 error = xfs_attr_rmtval_set(args);
1403                 if (error)
1404                         return(error);
1405         }
1406
1407         /*
1408          * If this is an atomic rename operation, we must "flip" the
1409          * incomplete flags on the "new" and "old" attribute/value pairs
1410          * so that one disappears and one appears atomically.  Then we
1411          * must remove the "old" attribute/value pair.
1412          */
1413         if (args->rename) {
1414                 /*
1415                  * In a separate transaction, set the incomplete flag on the
1416                  * "old" attr and clear the incomplete flag on the "new" attr.
1417                  */
1418                 error = xfs_attr_leaf_flipflags(args);
1419                 if (error)
1420                         goto out;
1421
1422                 /*
1423                  * Dismantle the "old" attribute/value pair by removing
1424                  * a "remote" value (if it exists).
1425                  */
1426                 args->index = args->index2;
1427                 args->blkno = args->blkno2;
1428                 args->rmtblkno = args->rmtblkno2;
1429                 args->rmtblkcnt = args->rmtblkcnt2;
1430                 if (args->rmtblkno) {
1431                         error = xfs_attr_rmtval_remove(args);
1432                         if (error)
1433                                 return(error);
1434                 }
1435
1436                 /*
1437                  * Re-find the "old" attribute entry after any split ops.
1438                  * The INCOMPLETE flag means that we will find the "old"
1439                  * attr, not the "new" one.
1440                  */
1441                 args->flags |= XFS_ATTR_INCOMPLETE;
1442                 state = xfs_da_state_alloc();
1443                 state->args = args;
1444                 state->mp = mp;
1445                 state->blocksize = state->mp->m_sb.sb_blocksize;
1446                 state->node_ents = state->mp->m_attr_node_ents;
1447                 state->inleaf = 0;
1448                 error = xfs_da_node_lookup_int(state, &retval);
1449                 if (error)
1450                         goto out;
1451
1452                 /*
1453                  * Remove the name and update the hashvals in the tree.
1454                  */
1455                 blk = &state->path.blk[ state->path.active-1 ];
1456                 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1457                 error = xfs_attr_leaf_remove(blk->bp, args);
1458                 xfs_da_fixhashpath(state, &state->path);
1459
1460                 /*
1461                  * Check to see if the tree needs to be collapsed.
1462                  */
1463                 if (retval && (state->path.active > 1)) {
1464                         XFS_BMAP_INIT(args->flist, args->firstblock);
1465                         error = xfs_da_join(state);
1466                         if (!error) {
1467                                 error = xfs_bmap_finish(&args->trans,
1468                                                         args->flist,
1469                                                         &committed);
1470                         }
1471                         if (error) {
1472                                 ASSERT(committed);
1473                                 args->trans = NULL;
1474                                 xfs_bmap_cancel(args->flist);
1475                                 goto out;
1476                         }
1477
1478                         /*
1479                          * bmap_finish() may have committed the last trans
1480                          * and started a new one.  We need the inode to be
1481                          * in all transactions.
1482                          */
1483                         if (committed) {
1484                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1485                                 xfs_trans_ihold(args->trans, dp);
1486                         }
1487                 }
1488
1489                 /*
1490                  * Commit and start the next trans in the chain.
1491                  */
1492                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1493                         goto out;
1494
1495         } else if (args->rmtblkno > 0) {
1496                 /*
1497                  * Added a "remote" value, just clear the incomplete flag.
1498                  */
1499                 error = xfs_attr_leaf_clearflag(args);
1500                 if (error)
1501                         goto out;
1502         }
1503         retval = error = 0;
1504
1505 out:
1506         if (state)
1507                 xfs_da_state_free(state);
1508         if (error)
1509                 return(error);
1510         return(retval);
1511 }
1512
1513 /*
1514  * Remove a name from a B-tree attribute list.
1515  *
1516  * This will involve walking down the Btree, and may involve joining
1517  * leaf nodes and even joining intermediate nodes up to and including
1518  * the root node (a special case of an intermediate node).
1519  */
1520 STATIC int
1521 xfs_attr_node_removename(xfs_da_args_t *args)
1522 {
1523         xfs_da_state_t *state;
1524         xfs_da_state_blk_t *blk;
1525         xfs_inode_t *dp;
1526         xfs_dabuf_t *bp;
1527         int retval, error, committed, forkoff;
1528
1529         /*
1530          * Tie a string around our finger to remind us where we are.
1531          */
1532         dp = args->dp;
1533         state = xfs_da_state_alloc();
1534         state->args = args;
1535         state->mp = dp->i_mount;
1536         state->blocksize = state->mp->m_sb.sb_blocksize;
1537         state->node_ents = state->mp->m_attr_node_ents;
1538
1539         /*
1540          * Search to see if name exists, and get back a pointer to it.
1541          */
1542         error = xfs_da_node_lookup_int(state, &retval);
1543         if (error || (retval != EEXIST)) {
1544                 if (error == 0)
1545                         error = retval;
1546                 goto out;
1547         }
1548
1549         /*
1550          * If there is an out-of-line value, de-allocate the blocks.
1551          * This is done before we remove the attribute so that we don't
1552          * overflow the maximum size of a transaction and/or hit a deadlock.
1553          */
1554         blk = &state->path.blk[ state->path.active-1 ];
1555         ASSERT(blk->bp != NULL);
1556         ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1557         if (args->rmtblkno > 0) {
1558                 /*
1559                  * Fill in disk block numbers in the state structure
1560                  * so that we can get the buffers back after we commit
1561                  * several transactions in the following calls.
1562                  */
1563                 error = xfs_attr_fillstate(state);
1564                 if (error)
1565                         goto out;
1566
1567                 /*
1568                  * Mark the attribute as INCOMPLETE, then bunmapi() the
1569                  * remote value.
1570                  */
1571                 error = xfs_attr_leaf_setflag(args);
1572                 if (error)
1573                         goto out;
1574                 error = xfs_attr_rmtval_remove(args);
1575                 if (error)
1576                         goto out;
1577
1578                 /*
1579                  * Refill the state structure with buffers, the prior calls
1580                  * released our buffers.
1581                  */
1582                 error = xfs_attr_refillstate(state);
1583                 if (error)
1584                         goto out;
1585         }
1586
1587         /*
1588          * Remove the name and update the hashvals in the tree.
1589          */
1590         blk = &state->path.blk[ state->path.active-1 ];
1591         ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1592         retval = xfs_attr_leaf_remove(blk->bp, args);
1593         xfs_da_fixhashpath(state, &state->path);
1594
1595         /*
1596          * Check to see if the tree needs to be collapsed.
1597          */
1598         if (retval && (state->path.active > 1)) {
1599                 XFS_BMAP_INIT(args->flist, args->firstblock);
1600                 error = xfs_da_join(state);
1601                 if (!error) {
1602                         error = xfs_bmap_finish(&args->trans, args->flist,
1603                                                 &committed);
1604                 }
1605                 if (error) {
1606                         ASSERT(committed);
1607                         args->trans = NULL;
1608                         xfs_bmap_cancel(args->flist);
1609                         goto out;
1610                 }
1611
1612                 /*
1613                  * bmap_finish() may have committed the last trans and started
1614                  * a new one.  We need the inode to be in all transactions.
1615                  */
1616                 if (committed) {
1617                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1618                         xfs_trans_ihold(args->trans, dp);
1619                 }
1620
1621                 /*
1622                  * Commit the Btree join operation and start a new trans.
1623                  */
1624                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1625                         goto out;
1626         }
1627
1628         /*
1629          * If the result is small enough, push it all into the inode.
1630          */
1631         if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
1632                 /*
1633                  * Have to get rid of the copy of this dabuf in the state.
1634                  */
1635                 ASSERT(state->path.active == 1);
1636                 ASSERT(state->path.blk[0].bp);
1637                 xfs_da_buf_done(state->path.blk[0].bp);
1638                 state->path.blk[0].bp = NULL;
1639
1640                 error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp,
1641                                                      XFS_ATTR_FORK);
1642                 if (error)
1643                         goto out;
1644                 ASSERT(be16_to_cpu(((xfs_attr_leafblock_t *)
1645                                       bp->data)->hdr.info.magic)
1646                                                        == XFS_ATTR_LEAF_MAGIC);
1647
1648                 if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1649                         XFS_BMAP_INIT(args->flist, args->firstblock);
1650                         error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1651                         /* bp is gone due to xfs_da_shrink_inode */
1652                         if (!error) {
1653                                 error = xfs_bmap_finish(&args->trans,
1654                                                         args->flist,
1655                                                         &committed);
1656                         }
1657                         if (error) {
1658                                 ASSERT(committed);
1659                                 args->trans = NULL;
1660                                 xfs_bmap_cancel(args->flist);
1661                                 goto out;
1662                         }
1663
1664                         /*
1665                          * bmap_finish() may have committed the last trans
1666                          * and started a new one.  We need the inode to be
1667                          * in all transactions.
1668                          */
1669                         if (committed) {
1670                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1671                                 xfs_trans_ihold(args->trans, dp);
1672                         }
1673                 } else
1674                         xfs_da_brelse(args->trans, bp);
1675         }
1676         error = 0;
1677
1678 out:
1679         xfs_da_state_free(state);
1680         return(error);
1681 }
1682
1683 /*
1684  * Fill in the disk block numbers in the state structure for the buffers
1685  * that are attached to the state structure.
1686  * This is done so that we can quickly reattach ourselves to those buffers
1687  * after some set of transaction commits have released these buffers.
1688  */
1689 STATIC int
1690 xfs_attr_fillstate(xfs_da_state_t *state)
1691 {
1692         xfs_da_state_path_t *path;
1693         xfs_da_state_blk_t *blk;
1694         int level;
1695
1696         /*
1697          * Roll down the "path" in the state structure, storing the on-disk
1698          * block number for those buffers in the "path".
1699          */
1700         path = &state->path;
1701         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1702         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1703                 if (blk->bp) {
1704                         blk->disk_blkno = xfs_da_blkno(blk->bp);
1705                         xfs_da_buf_done(blk->bp);
1706                         blk->bp = NULL;
1707                 } else {
1708                         blk->disk_blkno = 0;
1709                 }
1710         }
1711
1712         /*
1713          * Roll down the "altpath" in the state structure, storing the on-disk
1714          * block number for those buffers in the "altpath".
1715          */
1716         path = &state->altpath;
1717         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1718         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1719                 if (blk->bp) {
1720                         blk->disk_blkno = xfs_da_blkno(blk->bp);
1721                         xfs_da_buf_done(blk->bp);
1722                         blk->bp = NULL;
1723                 } else {
1724                         blk->disk_blkno = 0;
1725                 }
1726         }
1727
1728         return(0);
1729 }
1730
1731 /*
1732  * Reattach the buffers to the state structure based on the disk block
1733  * numbers stored in the state structure.
1734  * This is done after some set of transaction commits have released those
1735  * buffers from our grip.
1736  */
1737 STATIC int
1738 xfs_attr_refillstate(xfs_da_state_t *state)
1739 {
1740         xfs_da_state_path_t *path;
1741         xfs_da_state_blk_t *blk;
1742         int level, error;
1743
1744         /*
1745          * Roll down the "path" in the state structure, storing the on-disk
1746          * block number for those buffers in the "path".
1747          */
1748         path = &state->path;
1749         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1750         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1751                 if (blk->disk_blkno) {
1752                         error = xfs_da_read_buf(state->args->trans,
1753                                                 state->args->dp,
1754                                                 blk->blkno, blk->disk_blkno,
1755                                                 &blk->bp, XFS_ATTR_FORK);
1756                         if (error)
1757                                 return(error);
1758                 } else {
1759                         blk->bp = NULL;
1760                 }
1761         }
1762
1763         /*
1764          * Roll down the "altpath" in the state structure, storing the on-disk
1765          * block number for those buffers in the "altpath".
1766          */
1767         path = &state->altpath;
1768         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1769         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1770                 if (blk->disk_blkno) {
1771                         error = xfs_da_read_buf(state->args->trans,
1772                                                 state->args->dp,
1773                                                 blk->blkno, blk->disk_blkno,
1774                                                 &blk->bp, XFS_ATTR_FORK);
1775                         if (error)
1776                                 return(error);
1777                 } else {
1778                         blk->bp = NULL;
1779                 }
1780         }
1781
1782         return(0);
1783 }
1784
1785 /*
1786  * Look up a filename in a node attribute list.
1787  *
1788  * This routine gets called for any attribute fork that has more than one
1789  * block, ie: both true Btree attr lists and for single-leaf-blocks with
1790  * "remote" values taking up more blocks.
1791  */
1792 STATIC int
1793 xfs_attr_node_get(xfs_da_args_t *args)
1794 {
1795         xfs_da_state_t *state;
1796         xfs_da_state_blk_t *blk;
1797         int error, retval;
1798         int i;
1799
1800         state = xfs_da_state_alloc();
1801         state->args = args;
1802         state->mp = args->dp->i_mount;
1803         state->blocksize = state->mp->m_sb.sb_blocksize;
1804         state->node_ents = state->mp->m_attr_node_ents;
1805
1806         /*
1807          * Search to see if name exists, and get back a pointer to it.
1808          */
1809         error = xfs_da_node_lookup_int(state, &retval);
1810         if (error) {
1811                 retval = error;
1812         } else if (retval == EEXIST) {
1813                 blk = &state->path.blk[ state->path.active-1 ];
1814                 ASSERT(blk->bp != NULL);
1815                 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1816
1817                 /*
1818                  * Get the value, local or "remote"
1819                  */
1820                 retval = xfs_attr_leaf_getvalue(blk->bp, args);
1821                 if (!retval && (args->rmtblkno > 0)
1822                     && !(args->flags & ATTR_KERNOVAL)) {
1823                         retval = xfs_attr_rmtval_get(args);
1824                 }
1825         }
1826
1827         /*
1828          * If not in a transaction, we have to release all the buffers.
1829          */
1830         for (i = 0; i < state->path.active; i++) {
1831                 xfs_da_brelse(args->trans, state->path.blk[i].bp);
1832                 state->path.blk[i].bp = NULL;
1833         }
1834
1835         xfs_da_state_free(state);
1836         return(retval);
1837 }
1838
1839 STATIC int                                                      /* error */
1840 xfs_attr_node_list(xfs_attr_list_context_t *context)
1841 {
1842         attrlist_cursor_kern_t *cursor;
1843         xfs_attr_leafblock_t *leaf;
1844         xfs_da_intnode_t *node;
1845         xfs_da_node_entry_t *btree;
1846         int error, i;
1847         xfs_dabuf_t *bp;
1848
1849         cursor = context->cursor;
1850         cursor->initted = 1;
1851
1852         /*
1853          * Do all sorts of validation on the passed-in cursor structure.
1854          * If anything is amiss, ignore the cursor and look up the hashval
1855          * starting from the btree root.
1856          */
1857         bp = NULL;
1858         if (cursor->blkno > 0) {
1859                 error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1860                                               &bp, XFS_ATTR_FORK);
1861                 if ((error != 0) && (error != EFSCORRUPTED))
1862                         return(error);
1863                 if (bp) {
1864                         node = bp->data;
1865                         switch (be16_to_cpu(node->hdr.info.magic)) {
1866                         case XFS_DA_NODE_MAGIC:
1867                                 xfs_attr_trace_l_cn("wrong blk", context, node);
1868                                 xfs_da_brelse(NULL, bp);
1869                                 bp = NULL;
1870                                 break;
1871                         case XFS_ATTR_LEAF_MAGIC:
1872                                 leaf = bp->data;
1873                                 if (cursor->hashval > be32_to_cpu(leaf->entries[
1874                                     be16_to_cpu(leaf->hdr.count)-1].hashval)) {
1875                                         xfs_attr_trace_l_cl("wrong blk",
1876                                                            context, leaf);
1877                                         xfs_da_brelse(NULL, bp);
1878                                         bp = NULL;
1879                                 } else if (cursor->hashval <=
1880                                              be32_to_cpu(leaf->entries[0].hashval)) {
1881                                         xfs_attr_trace_l_cl("maybe wrong blk",
1882                                                            context, leaf);
1883                                         xfs_da_brelse(NULL, bp);
1884                                         bp = NULL;
1885                                 }
1886                                 break;
1887                         default:
1888                                 xfs_attr_trace_l_c("wrong blk - ??", context);
1889                                 xfs_da_brelse(NULL, bp);
1890                                 bp = NULL;
1891                         }
1892                 }
1893         }
1894
1895         /*
1896          * We did not find what we expected given the cursor's contents,
1897          * so we start from the top and work down based on the hash value.
1898          * Note that start of node block is same as start of leaf block.
1899          */
1900         if (bp == NULL) {
1901                 cursor->blkno = 0;
1902                 for (;;) {
1903                         error = xfs_da_read_buf(NULL, context->dp,
1904                                                       cursor->blkno, -1, &bp,
1905                                                       XFS_ATTR_FORK);
1906                         if (error)
1907                                 return(error);
1908                         if (unlikely(bp == NULL)) {
1909                                 XFS_ERROR_REPORT("xfs_attr_node_list(2)",
1910                                                  XFS_ERRLEVEL_LOW,
1911                                                  context->dp->i_mount);
1912                                 return(XFS_ERROR(EFSCORRUPTED));
1913                         }
1914                         node = bp->data;
1915                         if (be16_to_cpu(node->hdr.info.magic)
1916                                                         == XFS_ATTR_LEAF_MAGIC)
1917                                 break;
1918                         if (unlikely(be16_to_cpu(node->hdr.info.magic)
1919                                                         != XFS_DA_NODE_MAGIC)) {
1920                                 XFS_CORRUPTION_ERROR("xfs_attr_node_list(3)",
1921                                                      XFS_ERRLEVEL_LOW,
1922                                                      context->dp->i_mount,
1923                                                      node);
1924                                 xfs_da_brelse(NULL, bp);
1925                                 return(XFS_ERROR(EFSCORRUPTED));
1926                         }
1927                         btree = node->btree;
1928                         for (i = 0; i < be16_to_cpu(node->hdr.count);
1929                                                                 btree++, i++) {
1930                                 if (cursor->hashval
1931                                                 <= be32_to_cpu(btree->hashval)) {
1932                                         cursor->blkno = be32_to_cpu(btree->before);
1933                                         xfs_attr_trace_l_cb("descending",
1934                                                             context, btree);
1935                                         break;
1936                                 }
1937                         }
1938                         if (i == be16_to_cpu(node->hdr.count)) {
1939                                 xfs_da_brelse(NULL, bp);
1940                                 return(0);
1941                         }
1942                         xfs_da_brelse(NULL, bp);
1943                 }
1944         }
1945         ASSERT(bp != NULL);
1946
1947         /*
1948          * Roll upward through the blocks, processing each leaf block in
1949          * order.  As long as there is space in the result buffer, keep
1950          * adding the information.
1951          */
1952         for (;;) {
1953                 leaf = bp->data;
1954                 if (unlikely(be16_to_cpu(leaf->hdr.info.magic)
1955                                                 != XFS_ATTR_LEAF_MAGIC)) {
1956                         XFS_CORRUPTION_ERROR("xfs_attr_node_list(4)",
1957                                              XFS_ERRLEVEL_LOW,
1958                                              context->dp->i_mount, leaf);
1959                         xfs_da_brelse(NULL, bp);
1960                         return(XFS_ERROR(EFSCORRUPTED));
1961                 }
1962                 error = xfs_attr_leaf_list_int(bp, context);
1963                 if (error) {
1964                         xfs_da_brelse(NULL, bp);
1965                         return error;
1966                 }
1967                 if (context->seen_enough || leaf->hdr.info.forw == 0)
1968                         break;
1969                 cursor->blkno = be32_to_cpu(leaf->hdr.info.forw);
1970                 xfs_da_brelse(NULL, bp);
1971                 error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1972                                               &bp, XFS_ATTR_FORK);
1973                 if (error)
1974                         return(error);
1975                 if (unlikely((bp == NULL))) {
1976                         XFS_ERROR_REPORT("xfs_attr_node_list(5)",
1977                                          XFS_ERRLEVEL_LOW,
1978                                          context->dp->i_mount);
1979                         return(XFS_ERROR(EFSCORRUPTED));
1980                 }
1981         }
1982         xfs_da_brelse(NULL, bp);
1983         return(0);
1984 }
1985
1986
1987 /*========================================================================
1988  * External routines for manipulating out-of-line attribute values.
1989  *========================================================================*/
1990
1991 /*
1992  * Read the value associated with an attribute from the out-of-line buffer
1993  * that we stored it in.
1994  */
1995 int
1996 xfs_attr_rmtval_get(xfs_da_args_t *args)
1997 {
1998         xfs_bmbt_irec_t map[ATTR_RMTVALUE_MAPSIZE];
1999         xfs_mount_t *mp;
2000         xfs_daddr_t dblkno;
2001         xfs_caddr_t dst;
2002         xfs_buf_t *bp;
2003         int nmap, error, tmp, valuelen, blkcnt, i;
2004         xfs_dablk_t lblkno;
2005
2006         ASSERT(!(args->flags & ATTR_KERNOVAL));
2007
2008         mp = args->dp->i_mount;
2009         dst = args->value;
2010         valuelen = args->valuelen;
2011         lblkno = args->rmtblkno;
2012         while (valuelen > 0) {
2013                 nmap = ATTR_RMTVALUE_MAPSIZE;
2014                 error = xfs_bmapi(args->trans, args->dp, (xfs_fileoff_t)lblkno,
2015                                   args->rmtblkcnt,
2016                                   XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2017                                   NULL, 0, map, &nmap, NULL, NULL);
2018                 if (error)
2019                         return(error);
2020                 ASSERT(nmap >= 1);
2021
2022                 for (i = 0; (i < nmap) && (valuelen > 0); i++) {
2023                         ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
2024                                (map[i].br_startblock != HOLESTARTBLOCK));
2025                         dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
2026                         blkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
2027                         error = xfs_read_buf(mp, mp->m_ddev_targp, dblkno,
2028                                              blkcnt, XFS_BUF_LOCK, &bp);
2029                         if (error)
2030                                 return(error);
2031
2032                         tmp = (valuelen < XFS_BUF_SIZE(bp))
2033                                 ? valuelen : XFS_BUF_SIZE(bp);
2034                         xfs_biomove(bp, 0, tmp, dst, XFS_B_READ);
2035                         xfs_buf_relse(bp);
2036                         dst += tmp;
2037                         valuelen -= tmp;
2038
2039                         lblkno += map[i].br_blockcount;
2040                 }
2041         }
2042         ASSERT(valuelen == 0);
2043         return(0);
2044 }
2045
2046 /*
2047  * Write the value associated with an attribute into the out-of-line buffer
2048  * that we have defined for it.
2049  */
2050 STATIC int
2051 xfs_attr_rmtval_set(xfs_da_args_t *args)
2052 {
2053         xfs_mount_t *mp;
2054         xfs_fileoff_t lfileoff;
2055         xfs_inode_t *dp;
2056         xfs_bmbt_irec_t map;
2057         xfs_daddr_t dblkno;
2058         xfs_caddr_t src;
2059         xfs_buf_t *bp;
2060         xfs_dablk_t lblkno;
2061         int blkcnt, valuelen, nmap, error, tmp, committed;
2062
2063         dp = args->dp;
2064         mp = dp->i_mount;
2065         src = args->value;
2066
2067         /*
2068          * Find a "hole" in the attribute address space large enough for
2069          * us to drop the new attribute's value into.
2070          */
2071         blkcnt = XFS_B_TO_FSB(mp, args->valuelen);
2072         lfileoff = 0;
2073         error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
2074                                                    XFS_ATTR_FORK);
2075         if (error) {
2076                 return(error);
2077         }
2078         args->rmtblkno = lblkno = (xfs_dablk_t)lfileoff;
2079         args->rmtblkcnt = blkcnt;
2080
2081         /*
2082          * Roll through the "value", allocating blocks on disk as required.
2083          */
2084         while (blkcnt > 0) {
2085                 /*
2086                  * Allocate a single extent, up to the size of the value.
2087                  */
2088                 XFS_BMAP_INIT(args->flist, args->firstblock);
2089                 nmap = 1;
2090                 error = xfs_bmapi(args->trans, dp, (xfs_fileoff_t)lblkno,
2091                                   blkcnt,
2092                                   XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA |
2093                                                         XFS_BMAPI_WRITE,
2094                                   args->firstblock, args->total, &map, &nmap,
2095                                   args->flist, NULL);
2096                 if (!error) {
2097                         error = xfs_bmap_finish(&args->trans, args->flist,
2098                                                 &committed);
2099                 }
2100                 if (error) {
2101                         ASSERT(committed);
2102                         args->trans = NULL;
2103                         xfs_bmap_cancel(args->flist);
2104                         return(error);
2105                 }
2106
2107                 /*
2108                  * bmap_finish() may have committed the last trans and started
2109                  * a new one.  We need the inode to be in all transactions.
2110                  */
2111                 if (committed) {
2112                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
2113                         xfs_trans_ihold(args->trans, dp);
2114                 }
2115
2116                 ASSERT(nmap == 1);
2117                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2118                        (map.br_startblock != HOLESTARTBLOCK));
2119                 lblkno += map.br_blockcount;
2120                 blkcnt -= map.br_blockcount;
2121
2122                 /*
2123                  * Start the next trans in the chain.
2124                  */
2125                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
2126                         return (error);
2127         }
2128
2129         /*
2130          * Roll through the "value", copying the attribute value to the
2131          * already-allocated blocks.  Blocks are written synchronously
2132          * so that we can know they are all on disk before we turn off
2133          * the INCOMPLETE flag.
2134          */
2135         lblkno = args->rmtblkno;
2136         valuelen = args->valuelen;
2137         while (valuelen > 0) {
2138                 /*
2139                  * Try to remember where we decided to put the value.
2140                  */
2141                 XFS_BMAP_INIT(args->flist, args->firstblock);
2142                 nmap = 1;
2143                 error = xfs_bmapi(NULL, dp, (xfs_fileoff_t)lblkno,
2144                                   args->rmtblkcnt,
2145                                   XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2146                                   args->firstblock, 0, &map, &nmap,
2147                                   NULL, NULL);
2148                 if (error) {
2149                         return(error);
2150                 }
2151                 ASSERT(nmap == 1);
2152                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2153                        (map.br_startblock != HOLESTARTBLOCK));
2154
2155                 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2156                 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2157
2158                 bp = xfs_buf_get_flags(mp->m_ddev_targp, dblkno,
2159                                                         blkcnt, XFS_BUF_LOCK);
2160                 ASSERT(bp);
2161                 ASSERT(!XFS_BUF_GETERROR(bp));
2162
2163                 tmp = (valuelen < XFS_BUF_SIZE(bp)) ? valuelen :
2164                                                         XFS_BUF_SIZE(bp);
2165                 xfs_biomove(bp, 0, tmp, src, XFS_B_WRITE);
2166                 if (tmp < XFS_BUF_SIZE(bp))
2167                         xfs_biozero(bp, tmp, XFS_BUF_SIZE(bp) - tmp);
2168                 if ((error = xfs_bwrite(mp, bp))) {/* GROT: NOTE: synchronous write */
2169                         return (error);
2170                 }
2171                 src += tmp;
2172                 valuelen -= tmp;
2173
2174                 lblkno += map.br_blockcount;
2175         }
2176         ASSERT(valuelen == 0);
2177         return(0);
2178 }
2179
2180 /*
2181  * Remove the value associated with an attribute by deleting the
2182  * out-of-line buffer that it is stored on.
2183  */
2184 STATIC int
2185 xfs_attr_rmtval_remove(xfs_da_args_t *args)
2186 {
2187         xfs_mount_t *mp;
2188         xfs_bmbt_irec_t map;
2189         xfs_buf_t *bp;
2190         xfs_daddr_t dblkno;
2191         xfs_dablk_t lblkno;
2192         int valuelen, blkcnt, nmap, error, done, committed;
2193
2194         mp = args->dp->i_mount;
2195
2196         /*
2197          * Roll through the "value", invalidating the attribute value's
2198          * blocks.
2199          */
2200         lblkno = args->rmtblkno;
2201         valuelen = args->rmtblkcnt;
2202         while (valuelen > 0) {
2203                 /*
2204                  * Try to remember where we decided to put the value.
2205                  */
2206                 XFS_BMAP_INIT(args->flist, args->firstblock);
2207                 nmap = 1;
2208                 error = xfs_bmapi(NULL, args->dp, (xfs_fileoff_t)lblkno,
2209                                         args->rmtblkcnt,
2210                                         XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2211                                         args->firstblock, 0, &map, &nmap,
2212                                         args->flist, NULL);
2213                 if (error) {
2214                         return(error);
2215                 }
2216                 ASSERT(nmap == 1);
2217                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2218                        (map.br_startblock != HOLESTARTBLOCK));
2219
2220                 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2221                 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2222
2223                 /*
2224                  * If the "remote" value is in the cache, remove it.
2225                  */
2226                 bp = xfs_incore(mp->m_ddev_targp, dblkno, blkcnt,
2227                                 XFS_INCORE_TRYLOCK);
2228                 if (bp) {
2229                         XFS_BUF_STALE(bp);
2230                         XFS_BUF_UNDELAYWRITE(bp);
2231                         xfs_buf_relse(bp);
2232                         bp = NULL;
2233                 }
2234
2235                 valuelen -= map.br_blockcount;
2236
2237                 lblkno += map.br_blockcount;
2238         }
2239
2240         /*
2241          * Keep de-allocating extents until the remote-value region is gone.
2242          */
2243         lblkno = args->rmtblkno;
2244         blkcnt = args->rmtblkcnt;
2245         done = 0;
2246         while (!done) {
2247                 XFS_BMAP_INIT(args->flist, args->firstblock);
2248                 error = xfs_bunmapi(args->trans, args->dp, lblkno, blkcnt,
2249                                     XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2250                                     1, args->firstblock, args->flist,
2251                                     NULL, &done);
2252                 if (!error) {
2253                         error = xfs_bmap_finish(&args->trans, args->flist,
2254                                                 &committed);
2255                 }
2256                 if (error) {
2257                         ASSERT(committed);
2258                         args->trans = NULL;
2259                         xfs_bmap_cancel(args->flist);
2260                         return(error);
2261                 }
2262
2263                 /*
2264                  * bmap_finish() may have committed the last trans and started
2265                  * a new one.  We need the inode to be in all transactions.
2266                  */
2267                 if (committed) {
2268                         xfs_trans_ijoin(args->trans, args->dp, XFS_ILOCK_EXCL);
2269                         xfs_trans_ihold(args->trans, args->dp);
2270                 }
2271
2272                 /*
2273                  * Close out trans and start the next one in the chain.
2274                  */
2275                 if ((error = xfs_attr_rolltrans(&args->trans, args->dp)))
2276                         return (error);
2277         }
2278         return(0);
2279 }
2280
2281 #if defined(XFS_ATTR_TRACE)
2282 /*
2283  * Add a trace buffer entry for an attr_list context structure.
2284  */
2285 void
2286 xfs_attr_trace_l_c(char *where, struct xfs_attr_list_context *context)
2287 {
2288         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_C, where,
2289                 (__psunsigned_t)context->dp,
2290                 (__psunsigned_t)context->cursor->hashval,
2291                 (__psunsigned_t)context->cursor->blkno,
2292                 (__psunsigned_t)context->cursor->offset,
2293                 (__psunsigned_t)context->alist,
2294                 (__psunsigned_t)context->bufsize,
2295                 (__psunsigned_t)context->count,
2296                 (__psunsigned_t)context->firstu,
2297                 (__psunsigned_t)
2298                         ((context->count > 0) &&
2299                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2300                                 ? (ATTR_ENTRY(context->alist,
2301                                               context->count-1)->a_valuelen)
2302                                 : 0,
2303                 (__psunsigned_t)context->dupcnt,
2304                 (__psunsigned_t)context->flags,
2305                 (__psunsigned_t)NULL,
2306                 (__psunsigned_t)NULL,
2307                 (__psunsigned_t)NULL);
2308 }
2309
2310 /*
2311  * Add a trace buffer entry for a context structure and a Btree node.
2312  */
2313 void
2314 xfs_attr_trace_l_cn(char *where, struct xfs_attr_list_context *context,
2315                          struct xfs_da_intnode *node)
2316 {
2317         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_CN, where,
2318                 (__psunsigned_t)context->dp,
2319                 (__psunsigned_t)context->cursor->hashval,
2320                 (__psunsigned_t)context->cursor->blkno,
2321                 (__psunsigned_t)context->cursor->offset,
2322                 (__psunsigned_t)context->alist,
2323                 (__psunsigned_t)context->bufsize,
2324                 (__psunsigned_t)context->count,
2325                 (__psunsigned_t)context->firstu,
2326                 (__psunsigned_t)
2327                         ((context->count > 0) &&
2328                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2329                                 ? (ATTR_ENTRY(context->alist,
2330                                               context->count-1)->a_valuelen)
2331                                 : 0,
2332                 (__psunsigned_t)context->dupcnt,
2333                 (__psunsigned_t)context->flags,
2334                 (__psunsigned_t)be16_to_cpu(node->hdr.count),
2335                 (__psunsigned_t)be32_to_cpu(node->btree[0].hashval),
2336                 (__psunsigned_t)be32_to_cpu(node->btree[
2337                                     be16_to_cpu(node->hdr.count)-1].hashval));
2338 }
2339
2340 /*
2341  * Add a trace buffer entry for a context structure and a Btree element.
2342  */
2343 void
2344 xfs_attr_trace_l_cb(char *where, struct xfs_attr_list_context *context,
2345                           struct xfs_da_node_entry *btree)
2346 {
2347         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_CB, where,
2348                 (__psunsigned_t)context->dp,
2349                 (__psunsigned_t)context->cursor->hashval,
2350                 (__psunsigned_t)context->cursor->blkno,
2351                 (__psunsigned_t)context->cursor->offset,
2352                 (__psunsigned_t)context->alist,
2353                 (__psunsigned_t)context->bufsize,
2354                 (__psunsigned_t)context->count,
2355                 (__psunsigned_t)context->firstu,
2356                 (__psunsigned_t)
2357                         ((context->count > 0) &&
2358                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2359                                 ? (ATTR_ENTRY(context->alist,
2360                                               context->count-1)->a_valuelen)
2361                                 : 0,
2362                 (__psunsigned_t)context->dupcnt,
2363                 (__psunsigned_t)context->flags,
2364                 (__psunsigned_t)be32_to_cpu(btree->hashval),
2365                 (__psunsigned_t)be32_to_cpu(btree->before),
2366                 (__psunsigned_t)NULL);
2367 }
2368
2369 /*
2370  * Add a trace buffer entry for a context structure and a leaf block.
2371  */
2372 void
2373 xfs_attr_trace_l_cl(char *where, struct xfs_attr_list_context *context,
2374                               struct xfs_attr_leafblock *leaf)
2375 {
2376         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_CL, where,
2377                 (__psunsigned_t)context->dp,
2378                 (__psunsigned_t)context->cursor->hashval,
2379                 (__psunsigned_t)context->cursor->blkno,
2380                 (__psunsigned_t)context->cursor->offset,
2381                 (__psunsigned_t)context->alist,
2382                 (__psunsigned_t)context->bufsize,
2383                 (__psunsigned_t)context->count,
2384                 (__psunsigned_t)context->firstu,
2385                 (__psunsigned_t)
2386                         ((context->count > 0) &&
2387                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2388                                 ? (ATTR_ENTRY(context->alist,
2389                                               context->count-1)->a_valuelen)
2390                                 : 0,
2391                 (__psunsigned_t)context->dupcnt,
2392                 (__psunsigned_t)context->flags,
2393                 (__psunsigned_t)be16_to_cpu(leaf->hdr.count),
2394                 (__psunsigned_t)be32_to_cpu(leaf->entries[0].hashval),
2395                 (__psunsigned_t)be32_to_cpu(leaf->entries[
2396                                 be16_to_cpu(leaf->hdr.count)-1].hashval));
2397 }
2398
2399 /*
2400  * Add a trace buffer entry for the arguments given to the routine,
2401  * generic form.
2402  */
2403 void
2404 xfs_attr_trace_enter(int type, char *where,
2405                          __psunsigned_t a2, __psunsigned_t a3,
2406                          __psunsigned_t a4, __psunsigned_t a5,
2407                          __psunsigned_t a6, __psunsigned_t a7,
2408                          __psunsigned_t a8, __psunsigned_t a9,
2409                          __psunsigned_t a10, __psunsigned_t a11,
2410                          __psunsigned_t a12, __psunsigned_t a13,
2411                          __psunsigned_t a14, __psunsigned_t a15)
2412 {
2413         ASSERT(xfs_attr_trace_buf);
2414         ktrace_enter(xfs_attr_trace_buf, (void *)((__psunsigned_t)type),
2415                                          (void *)where,
2416                                          (void *)a2,  (void *)a3,  (void *)a4,
2417                                          (void *)a5,  (void *)a6,  (void *)a7,
2418                                          (void *)a8,  (void *)a9,  (void *)a10,
2419                                          (void *)a11, (void *)a12, (void *)a13,
2420                                          (void *)a14, (void *)a15);
2421 }
2422 #endif  /* XFS_ATTR_TRACE */
2423
2424
2425 /*========================================================================
2426  * System (pseudo) namespace attribute interface routines.
2427  *========================================================================*/
2428
2429 STATIC int
2430 posix_acl_access_set(
2431         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2432 {
2433         return xfs_acl_vset(vp, data, size, _ACL_TYPE_ACCESS);
2434 }
2435
2436 STATIC int
2437 posix_acl_access_remove(
2438         bhv_vnode_t *vp, char *name, int xflags)
2439 {
2440         return xfs_acl_vremove(vp, _ACL_TYPE_ACCESS);
2441 }
2442
2443 STATIC int
2444 posix_acl_access_get(
2445         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2446 {
2447         return xfs_acl_vget(vp, data, size, _ACL_TYPE_ACCESS);
2448 }
2449
2450 STATIC int
2451 posix_acl_access_exists(
2452         bhv_vnode_t *vp)
2453 {
2454         return xfs_acl_vhasacl_access(vp);
2455 }
2456
2457 STATIC int
2458 posix_acl_default_set(
2459         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2460 {
2461         return xfs_acl_vset(vp, data, size, _ACL_TYPE_DEFAULT);
2462 }
2463
2464 STATIC int
2465 posix_acl_default_get(
2466         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2467 {
2468         return xfs_acl_vget(vp, data, size, _ACL_TYPE_DEFAULT);
2469 }
2470
2471 STATIC int
2472 posix_acl_default_remove(
2473         bhv_vnode_t *vp, char *name, int xflags)
2474 {
2475         return xfs_acl_vremove(vp, _ACL_TYPE_DEFAULT);
2476 }
2477
2478 STATIC int
2479 posix_acl_default_exists(
2480         bhv_vnode_t *vp)
2481 {
2482         return xfs_acl_vhasacl_default(vp);
2483 }
2484
2485 static struct attrnames posix_acl_access = {
2486         .attr_name      = "posix_acl_access",
2487         .attr_namelen   = sizeof("posix_acl_access") - 1,
2488         .attr_get       = posix_acl_access_get,
2489         .attr_set       = posix_acl_access_set,
2490         .attr_remove    = posix_acl_access_remove,
2491         .attr_exists    = posix_acl_access_exists,
2492 };
2493
2494 static struct attrnames posix_acl_default = {
2495         .attr_name      = "posix_acl_default",
2496         .attr_namelen   = sizeof("posix_acl_default") - 1,
2497         .attr_get       = posix_acl_default_get,
2498         .attr_set       = posix_acl_default_set,
2499         .attr_remove    = posix_acl_default_remove,
2500         .attr_exists    = posix_acl_default_exists,
2501 };
2502
2503 static struct attrnames *attr_system_names[] =
2504         { &posix_acl_access, &posix_acl_default };
2505
2506
2507 /*========================================================================
2508  * Namespace-prefix-style attribute name interface routines.
2509  *========================================================================*/
2510
2511 STATIC int
2512 attr_generic_set(
2513         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2514 {
2515         return -bhv_vop_attr_set(vp, name, data, size, xflags, NULL);
2516 }
2517
2518 STATIC int
2519 attr_generic_get(
2520         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2521 {
2522         int     error, asize = size;
2523
2524         error = bhv_vop_attr_get(vp, name, data, &asize, xflags, NULL);
2525         if (!error)
2526                 return asize;
2527         return -error;
2528 }
2529
2530 STATIC int
2531 attr_generic_remove(
2532         bhv_vnode_t *vp, char *name, int xflags)
2533 {
2534         return -bhv_vop_attr_remove(vp, name, xflags, NULL);
2535 }
2536
2537 STATIC int
2538 attr_generic_listadd(
2539         attrnames_t             *prefix,
2540         attrnames_t             *namesp,
2541         void                    *data,
2542         size_t                  size,
2543         ssize_t                 *result)
2544 {
2545         char                    *p = data + *result;
2546
2547         *result += prefix->attr_namelen;
2548         *result += namesp->attr_namelen + 1;
2549         if (!size)
2550                 return 0;
2551         if (*result > size)
2552                 return -ERANGE;
2553         strcpy(p, prefix->attr_name);
2554         p += prefix->attr_namelen;
2555         strcpy(p, namesp->attr_name);
2556         p += namesp->attr_namelen + 1;
2557         return 0;
2558 }
2559
2560 STATIC int
2561 attr_system_list(
2562         bhv_vnode_t             *vp,
2563         void                    *data,
2564         size_t                  size,
2565         ssize_t                 *result)
2566 {
2567         attrnames_t             *namesp;
2568         int                     i, error = 0;
2569
2570         for (i = 0; i < ATTR_SYSCOUNT; i++) {
2571                 namesp = attr_system_names[i];
2572                 if (!namesp->attr_exists || !namesp->attr_exists(vp))
2573                         continue;
2574                 error = attr_generic_listadd(&attr_system, namesp,
2575                                                 data, size, result);
2576                 if (error)
2577                         break;
2578         }
2579         return error;
2580 }
2581
2582 int
2583 attr_generic_list(
2584         bhv_vnode_t *vp, void *data, size_t size, int xflags, ssize_t *result)
2585 {
2586         attrlist_cursor_kern_t  cursor = { 0 };
2587         int                     error;
2588
2589         error = bhv_vop_attr_list(vp, data, size, xflags, &cursor, NULL);
2590         if (error > 0)
2591                 return -error;
2592         *result = -error;
2593         return attr_system_list(vp, data, size, result);
2594 }
2595
2596 attrnames_t *
2597 attr_lookup_namespace(
2598         char                    *name,
2599         struct attrnames        **names,
2600         int                     nnames)
2601 {
2602         int                     i;
2603
2604         for (i = 0; i < nnames; i++)
2605                 if (!strncmp(name, names[i]->attr_name, names[i]->attr_namelen))
2606                         return names[i];
2607         return NULL;
2608 }
2609
2610 /*
2611  * Some checks to prevent people abusing EAs to get over quota:
2612  * - Don't allow modifying user EAs on devices/symlinks;
2613  * - Don't allow modifying user EAs if sticky bit set;
2614  */
2615 STATIC int
2616 attr_user_capable(
2617         bhv_vnode_t     *vp,
2618         cred_t          *cred)
2619 {
2620         struct inode    *inode = vn_to_inode(vp);
2621
2622         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
2623                 return -EPERM;
2624         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode) &&
2625             !capable(CAP_SYS_ADMIN))
2626                 return -EPERM;
2627         if (S_ISDIR(inode->i_mode) && (inode->i_mode & S_ISVTX) &&
2628             (current_fsuid(cred) != inode->i_uid) && !capable(CAP_FOWNER))
2629                 return -EPERM;
2630         return 0;
2631 }
2632
2633 STATIC int
2634 attr_trusted_capable(
2635         bhv_vnode_t     *vp,
2636         cred_t          *cred)
2637 {
2638         struct inode    *inode = vn_to_inode(vp);
2639
2640         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
2641                 return -EPERM;
2642         if (!capable(CAP_SYS_ADMIN))
2643                 return -EPERM;
2644         return 0;
2645 }
2646
2647 STATIC int
2648 attr_secure_capable(
2649         bhv_vnode_t     *vp,
2650         cred_t          *cred)
2651 {
2652         return -ENOSECURITY;
2653 }
2654
2655 STATIC int
2656 attr_system_set(
2657         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2658 {
2659         attrnames_t     *namesp;
2660         int             error;
2661
2662         if (xflags & ATTR_CREATE)
2663                 return -EINVAL;
2664
2665         namesp = attr_lookup_namespace(name, attr_system_names, ATTR_SYSCOUNT);
2666         if (!namesp)
2667                 return -EOPNOTSUPP;
2668         error = namesp->attr_set(vp, name, data, size, xflags);
2669         if (!error)
2670                 error = vn_revalidate(vp);
2671         return error;
2672 }
2673
2674 STATIC int
2675 attr_system_get(
2676         bhv_vnode_t *vp, char *name, void *data, size_t size, int xflags)
2677 {
2678         attrnames_t     *namesp;
2679
2680         namesp = attr_lookup_namespace(name, attr_system_names, ATTR_SYSCOUNT);
2681         if (!namesp)
2682                 return -EOPNOTSUPP;
2683         return namesp->attr_get(vp, name, data, size, xflags);
2684 }
2685
2686 STATIC int
2687 attr_system_remove(
2688         bhv_vnode_t *vp, char *name, int xflags)
2689 {
2690         attrnames_t     *namesp;
2691
2692         namesp = attr_lookup_namespace(name, attr_system_names, ATTR_SYSCOUNT);
2693         if (!namesp)
2694                 return -EOPNOTSUPP;
2695         return namesp->attr_remove(vp, name, xflags);
2696 }
2697
2698 struct attrnames attr_system = {
2699         .attr_name      = "system.",
2700         .attr_namelen   = sizeof("system.") - 1,
2701         .attr_flag      = ATTR_SYSTEM,
2702         .attr_get       = attr_system_get,
2703         .attr_set       = attr_system_set,
2704         .attr_remove    = attr_system_remove,
2705         .attr_capable   = (attrcapable_t)fs_noerr,
2706 };
2707
2708 struct attrnames attr_trusted = {
2709         .attr_name      = "trusted.",
2710         .attr_namelen   = sizeof("trusted.") - 1,
2711         .attr_flag      = ATTR_ROOT,
2712         .attr_get       = attr_generic_get,
2713         .attr_set       = attr_generic_set,
2714         .attr_remove    = attr_generic_remove,
2715         .attr_capable   = attr_trusted_capable,
2716 };
2717
2718 struct attrnames attr_secure = {
2719         .attr_name      = "security.",
2720         .attr_namelen   = sizeof("security.") - 1,
2721         .attr_flag      = ATTR_SECURE,
2722         .attr_get       = attr_generic_get,
2723         .attr_set       = attr_generic_set,
2724         .attr_remove    = attr_generic_remove,
2725         .attr_capable   = attr_secure_capable,
2726 };
2727
2728 struct attrnames attr_user = {
2729         .attr_name      = "user.",
2730         .attr_namelen   = sizeof("user.") - 1,
2731         .attr_get       = attr_generic_get,
2732         .attr_set       = attr_generic_set,
2733         .attr_remove    = attr_generic_remove,
2734         .attr_capable   = attr_user_capable,
2735 };
2736
2737 struct attrnames *attr_namespaces[] =
2738         { &attr_system, &attr_trusted, &attr_secure, &attr_user };