fs/xfs/xfs_attr.c

   1 /*
   2  * Copyright (c) 2000-2003 Silicon Graphics, Inc.  All Rights Reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms of version 2 of the GNU General Public License as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it would be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  11  *
  12  * Further, this software is distributed without any warranty that it is
  13  * free of the rightful claim of any third person regarding infringement
  14  * or the like.  Any license provided herein, whether implied or
  15  * otherwise, applies only to this software file.  Patent licenses, if
  16  * any, provided herein do not apply to combinations of this program with
  17  * other software, or any other product whatsoever.
  18  *
  19  * You should have received a copy of the GNU General Public License along
  20  * with this program; if not, write the Free Software Foundation, Inc., 59
  21  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
  22  *
  23  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
  24  * Mountain View, CA  94043, or:
  25  *
  26  * http://www.sgi.com
  27  *
  28  * For further information regarding this notice, see:
  29  *
  30  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
  31  */
  32
  33 #include "xfs.h"
  34
  35 #include "xfs_macros.h"
  36 #include "xfs_types.h"
  37 #include "xfs_inum.h"
  38 #include "xfs_log.h"
  39 #include "xfs_trans.h"
  40 #include "xfs_sb.h"
  41 #include "xfs_ag.h"
  42 #include "xfs_dir.h"
  43 #include "xfs_dir2.h"
  44 #include "xfs_dmapi.h"
  45 #include "xfs_mount.h"
  46 #include "xfs_alloc_btree.h"
  47 #include "xfs_bmap_btree.h"
  48 #include "xfs_ialloc_btree.h"
  49 #include "xfs_alloc.h"
  50 #include "xfs_btree.h"
  51 #include "xfs_attr_sf.h"
  52 #include "xfs_dir_sf.h"
  53 #include "xfs_dir2_sf.h"
  54 #include "xfs_dinode.h"
  55 #include "xfs_inode_item.h"
  56 #include "xfs_inode.h"
  57 #include "xfs_bmap.h"
  58 #include "xfs_da_btree.h"
  59 #include "xfs_attr.h"
  60 #include "xfs_attr_leaf.h"
  61 #include "xfs_error.h"
  62 #include "xfs_bit.h"
  63 #include "xfs_quota.h"
  64 #include "xfs_rw.h"
  65 #include "xfs_trans_space.h"
  66 #include "xfs_acl.h"
  67
  68 /*
  69  * xfs_attr.c
  70  *
  71  * Provide the external interfaces to manage attribute lists.
  72  */
  73
  74 /*========================================================================
  75  * Function prototypes for the kernel.
  76  *========================================================================*/
  77
  78 /*
  79  * Internal routines when attribute list fits inside the inode.
  80  */
  81 STATIC int xfs_attr_shortform_addname(xfs_da_args_t *args);
  82
  83 /*
  84  * Internal routines when attribute list is one block.
  85  */
  86 STATIC int xfs_attr_leaf_addname(xfs_da_args_t *args);
  87 STATIC int xfs_attr_leaf_removename(xfs_da_args_t *args);
  88 STATIC int xfs_attr_leaf_list(xfs_attr_list_context_t *context);
  89
  90 /*
  91  * Internal routines when attribute list is more than one block.
  92  */
  93 STATIC int xfs_attr_node_addname(xfs_da_args_t *args);
  94 STATIC int xfs_attr_node_removename(xfs_da_args_t *args);
  95 STATIC int xfs_attr_node_list(xfs_attr_list_context_t *context);
  96 STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
  97 STATIC int xfs_attr_refillstate(xfs_da_state_t *state);
  98
  99 /*
 100  * Routines to manipulate out-of-line attribute values.
 101  */
 102 STATIC int xfs_attr_rmtval_get(xfs_da_args_t *args);
 103 STATIC int xfs_attr_rmtval_set(xfs_da_args_t *args);
 104 STATIC int xfs_attr_rmtval_remove(xfs_da_args_t *args);
 105
 106 #define ATTR_RMTVALUE_MAPSIZE   1       /* # of map entries at once */
 107 #define ATTR_RMTVALUE_TRANSBLKS 8       /* max # of blks in a transaction */
 108
 109 #if defined(XFS_ATTR_TRACE)
 110 ktrace_t *xfs_attr_trace_buf;
 111 #endif
 112
 113
 114 /*========================================================================
 115  * Overall external interface routines.
 116  *========================================================================*/
 117
 118 /*ARGSUSED*/
 119 STATIC int
 120 xfs_attr_get_int(xfs_inode_t *ip, char *name, char *value, int *valuelenp,
 121              int flags, int lock, struct cred *cred)
 122 {
 123         xfs_da_args_t   args;
 124         int             error;
 125         int             namelen;
 126
 127         ASSERT(MAXNAMELEN-1 <= 0xff);   /* length is stored in uint8 */
 128         namelen = strlen(name);
 129         if (namelen >= MAXNAMELEN)
 130                 return(EFAULT);         /* match IRIX behaviour */
 131         XFS_STATS_INC(xs_attr_get);
 132
 133         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 134                 return(EIO);
 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                 return(ENOATTR);
 140
 141         if (lock) {
 142                 xfs_ilock(ip, XFS_ILOCK_SHARED);
 143                 /*
 144                  * Do we answer them, or ignore them?
 145                  */
 146                 if ((error = xfs_iaccess(ip, S_IRUSR, cred))) {
 147                         xfs_iunlock(ip, XFS_ILOCK_SHARED);
 148                         return(XFS_ERROR(error));
 149                 }
 150         }
 151
 152         /*
 153          * Fill in the arg structure for this request.
 154          */
 155         memset((char *)&args, 0, sizeof(args));
 156         args.name = name;
 157         args.namelen = namelen;
 158         args.value = value;
 159         args.valuelen = *valuelenp;
 160         args.flags = flags;
 161         args.hashval = xfs_da_hashname(args.name, args.namelen);
 162         args.dp = ip;
 163         args.whichfork = XFS_ATTR_FORK;
 164         args.trans = NULL;
 165
 166         /*
 167          * Decide on what work routines to call based on the inode size.
 168          */
 169         if (XFS_IFORK_Q(ip) == 0 ||
 170             (ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 171              ip->i_d.di_anextents == 0)) {
 172                 error = XFS_ERROR(ENOATTR);
 173         } else if (ip->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
 174                 error = xfs_attr_shortform_getvalue(&args);
 175         } else if (xfs_bmap_one_block(ip, XFS_ATTR_FORK)) {
 176                 error = xfs_attr_leaf_get(&args);
 177         } else {
 178                 error = xfs_attr_node_get(&args);
 179         }
 180
 181         if (lock)
 182                 xfs_iunlock(ip, XFS_ILOCK_SHARED);
 183
 184         /*
 185          * Return the number of bytes in the value to the caller.
 186          */
 187         *valuelenp = args.valuelen;
 188
 189         if (error == EEXIST)
 190                 error = 0;
 191         return(error);
 192 }
 193
 194 int
 195 xfs_attr_fetch(xfs_inode_t *ip, char *name, char *value, int valuelen)
 196 {
 197         return xfs_attr_get_int(ip, name, value, &valuelen, ATTR_ROOT, 0, NULL);
 198 }
 199
 200 int
 201 xfs_attr_get(bhv_desc_t *bdp, char *name, char *value, int *valuelenp,
 202              int flags, struct cred *cred)
 203 {
 204         xfs_inode_t     *ip = XFS_BHVTOI(bdp);
 205
 206         if (!name)
 207                 return(EINVAL);
 208         return xfs_attr_get_int(ip, name, value, valuelenp, flags, 1, cred);
 209 }
 210
 211 /*ARGSUSED*/
 212 int                                                             /* error */
 213 xfs_attr_set(bhv_desc_t *bdp, char *name, char *value, int valuelen, int flags,
 214                      struct cred *cred)
 215 {
 216         xfs_da_args_t   args;
 217         xfs_inode_t     *dp;
 218         xfs_fsblock_t   firstblock;
 219         xfs_bmap_free_t flist;
 220         int             error, err2, committed;
 221         int             local, size;
 222         uint            nblks;
 223         xfs_mount_t     *mp;
 224         int             rsvd = (flags & ATTR_ROOT) != 0;
 225         int             namelen;
 226
 227         ASSERT(MAXNAMELEN-1 <= 0xff); /* length is stored in uint8 */
 228         namelen = strlen(name);
 229         if (namelen >= MAXNAMELEN)
 230                 return EFAULT; /* match irix behaviour */
 231
 232         XFS_STATS_INC(xs_attr_set);
 233         /*
 234          * Do we answer them, or ignore them?
 235          */
 236         dp = XFS_BHVTOI(bdp);
 237         mp = dp->i_mount;
 238         if (XFS_FORCED_SHUTDOWN(mp))
 239                 return (EIO);
 240
 241         xfs_ilock(dp, XFS_ILOCK_SHARED);
 242         if ((error = xfs_iaccess(dp, S_IWUSR, cred))) {
 243                 xfs_iunlock(dp, XFS_ILOCK_SHARED);
 244                 return(XFS_ERROR(error));
 245         }
 246         xfs_iunlock(dp, XFS_ILOCK_SHARED);
 247
 248         /*
 249          * Attach the dquots to the inode.
 250          */
 251         if ((error = XFS_QM_DQATTACH(mp, dp, 0)))
 252                 return (error);
 253
 254         /*
 255          * If the inode doesn't have an attribute fork, add one.
 256          * (inode must not be locked when we call this routine)
 257          */
 258         if (XFS_IFORK_Q(dp) == 0) {
 259                 error = xfs_bmap_add_attrfork(dp, rsvd);
 260                 if (error)
 261                         return(error);
 262         }
 263
 264         /*
 265          * Fill in the arg structure for this request.
 266          */
 267         memset((char *)&args, 0, sizeof(args));
 268         args.name = name;
 269         args.namelen = namelen;
 270         args.value = value;
 271         args.valuelen = valuelen;
 272         args.flags = flags;
 273         args.hashval = xfs_da_hashname(args.name, args.namelen);
 274         args.dp = dp;
 275         args.firstblock = &firstblock;
 276         args.flist = &flist;
 277         args.whichfork = XFS_ATTR_FORK;
 278         args.oknoent = 1;
 279
 280         /* Determine space new attribute will use, and if it will be inline
 281          * or out of line.
 282          */
 283         size = xfs_attr_leaf_newentsize(&args, mp->m_sb.sb_blocksize, &local);
 284
 285         nblks = XFS_DAENTER_SPACE_RES(mp, XFS_ATTR_FORK);
 286         if (local) {
 287                 if (size > (mp->m_sb.sb_blocksize >> 1)) {
 288                         /* Double split possible */
 289                         nblks <<= 1;
 290                 }
 291         } else {
 292                 uint    dblocks = XFS_B_TO_FSB(mp, valuelen);
 293                 /* Out of line attribute, cannot double split, but make
 294                  * room for the attribute value itself.
 295                  */
 296                 nblks += dblocks;
 297                 nblks += XFS_NEXTENTADD_SPACE_RES(mp, dblocks, XFS_ATTR_FORK);
 298         }
 299
 300         /* Size is now blocks for attribute data */
 301         args.total = nblks;
 302
 303         /*
 304          * Start our first transaction of the day.
 305          *
 306          * All future transactions during this code must be "chained" off
 307          * this one via the trans_dup() call.  All transactions will contain
 308          * the inode, and the inode will always be marked with trans_ihold().
 309          * Since the inode will be locked in all transactions, we must log
 310          * the inode in every transaction to let it float upward through
 311          * the log.
 312          */
 313         args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_SET);
 314
 315         /*
 316          * Root fork attributes can use reserved data blocks for this
 317          * operation if necessary
 318          */
 319
 320         if (rsvd)
 321                 args.trans->t_flags |= XFS_TRANS_RESERVE;
 322
 323         if ((error = xfs_trans_reserve(args.trans, (uint) nblks,
 324                                       XFS_ATTRSET_LOG_RES(mp, nblks),
 325                                       0, XFS_TRANS_PERM_LOG_RES,
 326                                       XFS_ATTRSET_LOG_COUNT))) {
 327                 xfs_trans_cancel(args.trans, 0);
 328                 return(error);
 329         }
 330         xfs_ilock(dp, XFS_ILOCK_EXCL);
 331
 332         error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, args.trans, dp, nblks, 0,
 333                          rsvd ? XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
 334                                 XFS_QMOPT_RES_REGBLKS);
 335         if (error) {
 336                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
 337                 xfs_trans_cancel(args.trans, XFS_TRANS_RELEASE_LOG_RES);
 338                 return (error);
 339         }
 340
 341         xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
 342         xfs_trans_ihold(args.trans, dp);
 343
 344         /*
 345          * If the attribute list is non-existant or a shortform list,
 346          * upgrade it to a single-leaf-block attribute list.
 347          */
 348         if ((dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
 349             ((dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) &&
 350              (dp->i_d.di_anextents == 0))) {
 351
 352                 /*
 353                  * Build initial attribute list (if required).
 354                  */
 355                 if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS)
 356                         (void)xfs_attr_shortform_create(&args);
 357
 358                 /*
 359                  * Try to add the attr to the attribute list in
 360                  * the inode.
 361                  */
 362                 error = xfs_attr_shortform_addname(&args);
 363                 if (error != ENOSPC) {
 364                         /*
 365                          * Commit the shortform mods, and we're done.
 366                          * NOTE: this is also the error path (EEXIST, etc).
 367                          */
 368                         ASSERT(args.trans != NULL);
 369
 370                         /*
 371                          * If this is a synchronous mount, make sure that
 372                          * the transaction goes to disk before returning
 373                          * to the user.
 374                          */
 375                         if (mp->m_flags & XFS_MOUNT_WSYNC) {
 376                                 xfs_trans_set_sync(args.trans);
 377                         }
 378                         err2 = xfs_trans_commit(args.trans,
 379                                                  XFS_TRANS_RELEASE_LOG_RES,
 380                                                  NULL);
 381                         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 382
 383                         /*
 384                          * Hit the inode change time.
 385                          */
 386                         if (!error && (flags & ATTR_KERNOTIME) == 0) {
 387                                 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
 388                         }
 389                         return(error == 0 ? err2 : error);
 390                 }
 391
 392                 /*
 393                  * It won't fit in the shortform, transform to a leaf block.
 394                  * GROT: another possible req'mt for a double-split btree op.
 395                  */
 396                 XFS_BMAP_INIT(args.flist, args.firstblock);
 397                 error = xfs_attr_shortform_to_leaf(&args);
 398                 if (!error) {
 399                         error = xfs_bmap_finish(&args.trans, args.flist,
 400                                                 *args.firstblock, &committed);
 401                 }
 402                 if (error) {
 403                         ASSERT(committed);
 404                         args.trans = NULL;
 405                         xfs_bmap_cancel(&flist);
 406                         goto out;
 407                 }
 408
 409                 /*
 410                  * bmap_finish() may have committed the last trans and started
 411                  * a new one.  We need the inode to be in all transactions.
 412                  */
 413                 if (committed) {
 414                         xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
 415                         xfs_trans_ihold(args.trans, dp);
 416                 }
 417
 418                 /*
 419                  * Commit the leaf transformation.  We'll need another (linked)
 420                  * transaction to add the new attribute to the leaf.
 421                  */
 422                 if ((error = xfs_attr_rolltrans(&args.trans, dp)))
 423                         goto out;
 424
 425         }
 426
 427         if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
 428                 error = xfs_attr_leaf_addname(&args);
 429         } else {
 430                 error = xfs_attr_node_addname(&args);
 431         }
 432         if (error) {
 433                 goto out;
 434         }
 435
 436         /*
 437          * If this is a synchronous mount, make sure that the
 438          * transaction goes to disk before returning to the user.
 439          */
 440         if (mp->m_flags & XFS_MOUNT_WSYNC) {
 441                 xfs_trans_set_sync(args.trans);
 442         }
 443
 444         /*
 445          * Commit the last in the sequence of transactions.
 446          */
 447         xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
 448         error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES,
 449                                  NULL);
 450         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 451
 452         /*
 453          * Hit the inode change time.
 454          */
 455         if (!error && (flags & ATTR_KERNOTIME) == 0) {
 456                 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
 457         }
 458
 459         return(error);
 460
 461 out:
 462         if (args.trans)
 463                 xfs_trans_cancel(args.trans,
 464                         XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
 465         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 466         return(error);
 467 }
 468
 469 /*
 470  * Generic handler routine to remove a name from an attribute list.
 471  * Transitions attribute list from Btree to shortform as necessary.
 472  */
 473 /*ARGSUSED*/
 474 int                                                             /* error */
 475 xfs_attr_remove(bhv_desc_t *bdp, char *name, int flags, struct cred *cred)
 476 {
 477         xfs_da_args_t       args;
 478         xfs_inode_t         *dp;
 479         xfs_fsblock_t       firstblock;
 480         xfs_bmap_free_t     flist;
 481         int                 error;
 482         xfs_mount_t         *mp;
 483         int                 namelen;
 484
 485         ASSERT(MAXNAMELEN-1<=0xff); /* length is stored in uint8 */
 486         namelen = strlen(name);
 487         if (namelen>=MAXNAMELEN)
 488                 return EFAULT; /* match irix behaviour */
 489
 490         XFS_STATS_INC(xs_attr_remove);
 491
 492         /*
 493          * Do we answer them, or ignore them?
 494          */
 495         dp = XFS_BHVTOI(bdp);
 496         mp = dp->i_mount;
 497         if (XFS_FORCED_SHUTDOWN(mp))
 498                 return (EIO);
 499
 500         xfs_ilock(dp, XFS_ILOCK_SHARED);
 501         if ((error = xfs_iaccess(dp, S_IWUSR, cred))) {
 502                 xfs_iunlock(dp, XFS_ILOCK_SHARED);
 503                 return(XFS_ERROR(error));
 504         } else if (XFS_IFORK_Q(dp) == 0 ||
 505                    (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 506                     dp->i_d.di_anextents == 0)) {
 507                 xfs_iunlock(dp, XFS_ILOCK_SHARED);
 508                 return(XFS_ERROR(ENOATTR));
 509         }
 510         xfs_iunlock(dp, XFS_ILOCK_SHARED);
 511
 512         /*
 513          * Fill in the arg structure for this request.
 514          */
 515         memset((char *)&args, 0, sizeof(args));
 516         args.name = name;
 517         args.namelen = namelen;
 518         args.flags = flags;
 519         args.hashval = xfs_da_hashname(args.name, args.namelen);
 520         args.dp = dp;
 521         args.firstblock = &firstblock;
 522         args.flist = &flist;
 523         args.total = 0;
 524         args.whichfork = XFS_ATTR_FORK;
 525
 526         /*
 527          * Attach the dquots to the inode.
 528          */
 529         if ((error = XFS_QM_DQATTACH(mp, dp, 0)))
 530                 return (error);
 531
 532         /*
 533          * Start our first transaction of the day.
 534          *
 535          * All future transactions during this code must be "chained" off
 536          * this one via the trans_dup() call.  All transactions will contain
 537          * the inode, and the inode will always be marked with trans_ihold().
 538          * Since the inode will be locked in all transactions, we must log
 539          * the inode in every transaction to let it float upward through
 540          * the log.
 541          */
 542         args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_RM);
 543
 544         /*
 545          * Root fork attributes can use reserved data blocks for this
 546          * operation if necessary
 547          */
 548
 549         if (flags & ATTR_ROOT)
 550                 args.trans->t_flags |= XFS_TRANS_RESERVE;
 551
 552         if ((error = xfs_trans_reserve(args.trans,
 553                                       XFS_ATTRRM_SPACE_RES(mp),
 554                                       XFS_ATTRRM_LOG_RES(mp),
 555                                       0, XFS_TRANS_PERM_LOG_RES,
 556                                       XFS_ATTRRM_LOG_COUNT))) {
 557                 xfs_trans_cancel(args.trans, 0);
 558                 return(error);
 559
 560         }
 561
 562         xfs_ilock(dp, XFS_ILOCK_EXCL);
 563         /*
 564          * No need to make quota reservations here. We expect to release some
 565          * blocks not allocate in the common case.
 566          */
 567         xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
 568         xfs_trans_ihold(args.trans, dp);
 569
 570         /*
 571          * Decide on what work routines to call based on the inode size.
 572          */
 573         if (XFS_IFORK_Q(dp) == 0 ||
 574             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 575              dp->i_d.di_anextents == 0)) {
 576                 error = XFS_ERROR(ENOATTR);
 577                 goto out;
 578         }
 579         if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
 580                 ASSERT(dp->i_afp->if_flags & XFS_IFINLINE);
 581                 error = xfs_attr_shortform_remove(&args);
 582                 if (error) {
 583                         goto out;
 584                 }
 585         } else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
 586                 error = xfs_attr_leaf_removename(&args);
 587         } else {
 588                 error = xfs_attr_node_removename(&args);
 589         }
 590         if (error) {
 591                 goto out;
 592         }
 593
 594         /*
 595          * If this is a synchronous mount, make sure that the
 596          * transaction goes to disk before returning to the user.
 597          */
 598         if (mp->m_flags & XFS_MOUNT_WSYNC) {
 599                 xfs_trans_set_sync(args.trans);
 600         }
 601
 602         /*
 603          * Commit the last in the sequence of transactions.
 604          */
 605         xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
 606         error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES,
 607                                  NULL);
 608         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 609
 610         /*
 611          * Hit the inode change time.
 612          */
 613         if (!error && (flags & ATTR_KERNOTIME) == 0) {
 614                 xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
 615         }
 616
 617         return(error);
 618
 619 out:
 620         if (args.trans)
 621                 xfs_trans_cancel(args.trans,
 622                         XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
 623         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 624         return(error);
 625 }
 626
 627 /*
 628  * Generate a list of extended attribute names and optionally
 629  * also value lengths.  Positive return value follows the XFS
 630  * convention of being an error, zero or negative return code
 631  * is the length of the buffer returned (negated), indicating
 632  * success.
 633  */
 634 int
 635 xfs_attr_list(bhv_desc_t *bdp, char *buffer, int bufsize, int flags,
 636                       attrlist_cursor_kern_t *cursor, struct cred *cred)
 637 {
 638         xfs_attr_list_context_t context;
 639         xfs_inode_t *dp;
 640         int error;
 641
 642         XFS_STATS_INC(xs_attr_list);
 643
 644         /*
 645          * Validate the cursor.
 646          */
 647         if (cursor->pad1 || cursor->pad2)
 648                 return(XFS_ERROR(EINVAL));
 649         if ((cursor->initted == 0) &&
 650             (cursor->hashval || cursor->blkno || cursor->offset))
 651                 return(XFS_ERROR(EINVAL));
 652
 653         /*
 654          * Check for a properly aligned buffer.
 655          */
 656         if (((long)buffer) & (sizeof(int)-1))
 657                 return(XFS_ERROR(EFAULT));
 658         if (flags & ATTR_KERNOVAL)
 659                 bufsize = 0;
 660
 661         /*
 662          * Initialize the output buffer.
 663          */
 664         context.dp = dp = XFS_BHVTOI(bdp);
 665         context.cursor = cursor;
 666         context.count = 0;
 667         context.dupcnt = 0;
 668         context.resynch = 1;
 669         context.flags = flags;
 670         if (!(flags & ATTR_KERNAMELS)) {
 671                 context.bufsize = (bufsize & ~(sizeof(int)-1));  /* align */
 672                 context.firstu = context.bufsize;
 673                 context.alist = (attrlist_t *)buffer;
 674                 context.alist->al_count = 0;
 675                 context.alist->al_more = 0;
 676                 context.alist->al_offset[0] = context.bufsize;
 677         }
 678         else {
 679                 context.bufsize = bufsize;
 680                 context.firstu = context.bufsize;
 681                 context.alist = (attrlist_t *)buffer;
 682         }
 683
 684         if (XFS_FORCED_SHUTDOWN(dp->i_mount))
 685                 return (EIO);
 686         /*
 687          * Do they have permission?
 688          */
 689         xfs_ilock(dp, XFS_ILOCK_SHARED);
 690         if ((error = xfs_iaccess(dp, S_IRUSR, cred))) {
 691                 xfs_iunlock(dp, XFS_ILOCK_SHARED);
 692                 return(XFS_ERROR(error));
 693         }
 694
 695         /*
 696          * Decide on what work routines to call based on the inode size.
 697          */
 698         xfs_attr_trace_l_c("syscall start", &context);
 699         if (XFS_IFORK_Q(dp) == 0 ||
 700             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 701              dp->i_d.di_anextents == 0)) {
 702                 error = 0;
 703         } else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
 704                 error = xfs_attr_shortform_list(&context);
 705         } else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
 706                 error = xfs_attr_leaf_list(&context);
 707         } else {
 708                 error = xfs_attr_node_list(&context);
 709         }
 710         xfs_iunlock(dp, XFS_ILOCK_SHARED);
 711         xfs_attr_trace_l_c("syscall end", &context);
 712
 713         if (!(context.flags & (ATTR_KERNOVAL|ATTR_KERNAMELS))) {
 714                 ASSERT(error >= 0);
 715         }
 716         else {  /* must return negated buffer size or the error */
 717                 if (context.count < 0)
 718                         error = XFS_ERROR(ERANGE);
 719                 else
 720                         error = -context.count;
 721         }
 722
 723         return(error);
 724 }
 725
 726 int                                                             /* error */
 727 xfs_attr_inactive(xfs_inode_t *dp)
 728 {
 729         xfs_trans_t *trans;
 730         xfs_mount_t *mp;
 731         int error;
 732
 733         mp = dp->i_mount;
 734         ASSERT(! XFS_NOT_DQATTACHED(mp, dp));
 735
 736         /* XXXsup - why on earth are we taking ILOCK_EXCL here??? */
 737         xfs_ilock(dp, XFS_ILOCK_EXCL);
 738         if ((XFS_IFORK_Q(dp) == 0) ||
 739             (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
 740             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 741              dp->i_d.di_anextents == 0)) {
 742                 xfs_iunlock(dp, XFS_ILOCK_EXCL);
 743                 return(0);
 744         }
 745         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 746
 747         /*
 748          * Start our first transaction of the day.
 749          *
 750          * All future transactions during this code must be "chained" off
 751          * this one via the trans_dup() call.  All transactions will contain
 752          * the inode, and the inode will always be marked with trans_ihold().
 753          * Since the inode will be locked in all transactions, we must log
 754          * the inode in every transaction to let it float upward through
 755          * the log.
 756          */
 757         trans = xfs_trans_alloc(mp, XFS_TRANS_ATTRINVAL);
 758         if ((error = xfs_trans_reserve(trans, 0, XFS_ATTRINVAL_LOG_RES(mp), 0,
 759                                       XFS_TRANS_PERM_LOG_RES,
 760                                       XFS_ATTRINVAL_LOG_COUNT))) {
 761                 xfs_trans_cancel(trans, 0);
 762                 return(error);
 763         }
 764         xfs_ilock(dp, XFS_ILOCK_EXCL);
 765
 766         /*
 767          * No need to make quota reservations here. We expect to release some
 768          * blocks, not allocate, in the common case.
 769          */
 770         xfs_trans_ijoin(trans, dp, XFS_ILOCK_EXCL);
 771         xfs_trans_ihold(trans, dp);
 772
 773         /*
 774          * Decide on what work routines to call based on the inode size.
 775          */
 776         if ((XFS_IFORK_Q(dp) == 0) ||
 777             (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
 778             (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 779              dp->i_d.di_anextents == 0)) {
 780                 error = 0;
 781                 goto out;
 782         }
 783         error = xfs_attr_root_inactive(&trans, dp);
 784         if (error)
 785                 goto out;
 786         /*
 787          * signal synchronous inactive transactions unless this
 788          * is a synchronous mount filesystem in which case we
 789          * know that we're here because we've been called out of
 790          * xfs_inactive which means that the last reference is gone
 791          * and the unlink transaction has already hit the disk so
 792          * async inactive transactions are safe.
 793          */
 794         if ((error = xfs_itruncate_finish(&trans, dp, 0LL, XFS_ATTR_FORK,
 795                                 (!(mp->m_flags & XFS_MOUNT_WSYNC)
 796                                  ? 1 : 0))))
 797                 goto out;
 798
 799         /*
 800          * Commit the last in the sequence of transactions.
 801          */
 802         xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
 803         error = xfs_trans_commit(trans, XFS_TRANS_RELEASE_LOG_RES,
 804                                  NULL);
 805         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 806
 807         return(error);
 808
 809 out:
 810         xfs_trans_cancel(trans, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
 811         xfs_iunlock(dp, XFS_ILOCK_EXCL);
 812         return(error);
 813 }
 814
 815
 816
 817 /*========================================================================
 818  * External routines when attribute list is inside the inode
 819  *========================================================================*/
 820
 821 /*
 822  * Add a name to the shortform attribute list structure
 823  * This is the external routine.
 824  */
 825 STATIC int
 826 xfs_attr_shortform_addname(xfs_da_args_t *args)
 827 {
 828         int newsize, retval;
 829
 830         retval = xfs_attr_shortform_lookup(args);
 831         if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
 832                 return(retval);
 833         } else if (retval == EEXIST) {
 834                 if (args->flags & ATTR_CREATE)
 835                         return(retval);
 836                 retval = xfs_attr_shortform_remove(args);
 837                 ASSERT(retval == 0);
 838         }
 839
 840         newsize = XFS_ATTR_SF_TOTSIZE(args->dp);
 841         newsize += XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
 842         if ((newsize <= XFS_IFORK_ASIZE(args->dp)) &&
 843             (args->namelen < XFS_ATTR_SF_ENTSIZE_MAX) &&
 844             (args->valuelen < XFS_ATTR_SF_ENTSIZE_MAX)) {
 845                 retval = xfs_attr_shortform_add(args);
 846                 ASSERT(retval == 0);
 847         } else {
 848                 return(XFS_ERROR(ENOSPC));
 849         }
 850         return(0);
 851 }
 852
 853
 854 /*========================================================================
 855  * External routines when attribute list is one block
 856  *========================================================================*/
 857
 858 /*
 859  * Add a name to the leaf attribute list structure
 860  *
 861  * This leaf block cannot have a "remote" value, we only call this routine
 862  * if bmap_one_block() says there is only one block (ie: no remote blks).
 863  */
 864 int
 865 xfs_attr_leaf_addname(xfs_da_args_t *args)
 866 {
 867         xfs_inode_t *dp;
 868         xfs_dabuf_t *bp;
 869         int retval, error, committed;
 870
 871         /*
 872          * Read the (only) block in the attribute list in.
 873          */
 874         dp = args->dp;
 875         args->blkno = 0;
 876         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
 877                                              XFS_ATTR_FORK);
 878         if (error)
 879                 return(error);
 880         ASSERT(bp != NULL);
 881
 882         /*
 883          * Look up the given attribute in the leaf block.  Figure out if
 884          * the given flags produce an error or call for an atomic rename.
 885          */
 886         retval = xfs_attr_leaf_lookup_int(bp, args);
 887         if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
 888                 xfs_da_brelse(args->trans, bp);
 889                 return(retval);
 890         } else if (retval == EEXIST) {
 891                 if (args->flags & ATTR_CREATE) {        /* pure create op */
 892                         xfs_da_brelse(args->trans, bp);
 893                         return(retval);
 894                 }
 895                 args->rename = 1;                       /* an atomic rename */
 896                 args->blkno2 = args->blkno;             /* set 2nd entry info*/
 897                 args->index2 = args->index;
 898                 args->rmtblkno2 = args->rmtblkno;
 899                 args->rmtblkcnt2 = args->rmtblkcnt;
 900         }
 901
 902         /*
 903          * Add the attribute to the leaf block, transitioning to a Btree
 904          * if required.
 905          */
 906         retval = xfs_attr_leaf_add(bp, args);
 907         xfs_da_buf_done(bp);
 908         if (retval == ENOSPC) {
 909                 /*
 910                  * Promote the attribute list to the Btree format, then
 911                  * Commit that transaction so that the node_addname() call
 912                  * can manage its own transactions.
 913                  */
 914                 XFS_BMAP_INIT(args->flist, args->firstblock);
 915                 error = xfs_attr_leaf_to_node(args);
 916                 if (!error) {
 917                         error = xfs_bmap_finish(&args->trans, args->flist,
 918                                                 *args->firstblock, &committed);
 919                 }
 920                 if (error) {
 921                         ASSERT(committed);
 922                         args->trans = NULL;
 923                         xfs_bmap_cancel(args->flist);
 924                         return(error);
 925                 }
 926
 927                 /*
 928                  * bmap_finish() may have committed the last trans and started
 929                  * a new one.  We need the inode to be in all transactions.
 930                  */
 931                 if (committed) {
 932                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
 933                         xfs_trans_ihold(args->trans, dp);
 934                 }
 935
 936                 /*
 937                  * Commit the current trans (including the inode) and start
 938                  * a new one.
 939                  */
 940                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
 941                         return (error);
 942
 943                 /*
 944                  * Fob the whole rest of the problem off on the Btree code.
 945                  */
 946                 error = xfs_attr_node_addname(args);
 947                 return(error);
 948         }
 949
 950         /*
 951          * Commit the transaction that added the attr name so that
 952          * later routines can manage their own transactions.
 953          */
 954         if ((error = xfs_attr_rolltrans(&args->trans, dp)))
 955                 return (error);
 956
 957         /*
 958          * If there was an out-of-line value, allocate the blocks we
 959          * identified for its storage and copy the value.  This is done
 960          * after we create the attribute so that we don't overflow the
 961          * maximum size of a transaction and/or hit a deadlock.
 962          */
 963         if (args->rmtblkno > 0) {
 964                 error = xfs_attr_rmtval_set(args);
 965                 if (error)
 966                         return(error);
 967         }
 968
 969         /*
 970          * If this is an atomic rename operation, we must "flip" the
 971          * incomplete flags on the "new" and "old" attribute/value pairs
 972          * so that one disappears and one appears atomically.  Then we
 973          * must remove the "old" attribute/value pair.
 974          */
 975         if (args->rename) {
 976                 /*
 977                  * In a separate transaction, set the incomplete flag on the
 978                  * "old" attr and clear the incomplete flag on the "new" attr.
 979                  */
 980                 error = xfs_attr_leaf_flipflags(args);
 981                 if (error)
 982                         return(error);
 983
 984                 /*
 985                  * Dismantle the "old" attribute/value pair by removing
 986                  * a "remote" value (if it exists).
 987                  */
 988                 args->index = args->index2;
 989                 args->blkno = args->blkno2;
 990                 args->rmtblkno = args->rmtblkno2;
 991                 args->rmtblkcnt = args->rmtblkcnt2;
 992                 if (args->rmtblkno) {
 993                         error = xfs_attr_rmtval_remove(args);
 994                         if (error)
 995                                 return(error);
 996                 }
 997
 998                 /*
 999                  * Read in the block containing the "old" attr, then
1000                  * remove the "old" attr from that block (neat, huh!)
1001                  */
1002                 error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1,
1003                                                      &bp, XFS_ATTR_FORK);
1004                 if (error)
1005                         return(error);
1006                 ASSERT(bp != NULL);
1007                 (void)xfs_attr_leaf_remove(bp, args);
1008
1009                 /*
1010                  * If the result is small enough, shrink it all into the inode.
1011                  */
1012                 if (xfs_attr_shortform_allfit(bp, dp)) {
1013                         XFS_BMAP_INIT(args->flist, args->firstblock);
1014                         error = xfs_attr_leaf_to_shortform(bp, args);
1015                         /* bp is gone due to xfs_da_shrink_inode */
1016                         if (!error) {
1017                                 error = xfs_bmap_finish(&args->trans,
1018                                                         args->flist,
1019                                                         *args->firstblock,
1020                                                         &committed);
1021                         }
1022                         if (error) {
1023                                 ASSERT(committed);
1024                                 args->trans = NULL;
1025                                 xfs_bmap_cancel(args->flist);
1026                                 return(error);
1027                         }
1028
1029                         /*
1030                          * bmap_finish() may have committed the last trans
1031                          * and started a new one.  We need the inode to be
1032                          * in all transactions.
1033                          */
1034                         if (committed) {
1035                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1036                                 xfs_trans_ihold(args->trans, dp);
1037                         }
1038                 } else
1039                         xfs_da_buf_done(bp);
1040
1041                 /*
1042                  * Commit the remove and start the next trans in series.
1043                  */
1044                 error = xfs_attr_rolltrans(&args->trans, dp);
1045
1046         } else if (args->rmtblkno > 0) {
1047                 /*
1048                  * Added a "remote" value, just clear the incomplete flag.
1049                  */
1050                 error = xfs_attr_leaf_clearflag(args);
1051         }
1052         return(error);
1053 }
1054
1055 /*
1056  * Remove a name from the leaf attribute list structure
1057  *
1058  * This leaf block cannot have a "remote" value, we only call this routine
1059  * if bmap_one_block() says there is only one block (ie: no remote blks).
1060  */
1061 STATIC int
1062 xfs_attr_leaf_removename(xfs_da_args_t *args)
1063 {
1064         xfs_inode_t *dp;
1065         xfs_dabuf_t *bp;
1066         int committed;
1067         int error;
1068
1069         /*
1070          * Remove the attribute.
1071          */
1072         dp = args->dp;
1073         args->blkno = 0;
1074         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1075                                              XFS_ATTR_FORK);
1076         if (error) {
1077                 return(error);
1078         }
1079
1080         ASSERT(bp != NULL);
1081         error = xfs_attr_leaf_lookup_int(bp, args);
1082         if (error == ENOATTR) {
1083                 xfs_da_brelse(args->trans, bp);
1084                 return(error);
1085         }
1086
1087         (void)xfs_attr_leaf_remove(bp, args);
1088
1089         /*
1090          * If the result is small enough, shrink it all into the inode.
1091          */
1092         if (xfs_attr_shortform_allfit(bp, dp)) {
1093                 XFS_BMAP_INIT(args->flist, args->firstblock);
1094                 error = xfs_attr_leaf_to_shortform(bp, args);
1095                 /* bp is gone due to xfs_da_shrink_inode */
1096                 if (!error) {
1097                         error = xfs_bmap_finish(&args->trans, args->flist,
1098                                                 *args->firstblock, &committed);
1099                 }
1100                 if (error) {
1101                         ASSERT(committed);
1102                         args->trans = NULL;
1103                         xfs_bmap_cancel(args->flist);
1104                         return(error);
1105                 }
1106
1107                 /*
1108                  * bmap_finish() may have committed the last trans and started
1109                  * a new one.  We need the inode to be in all transactions.
1110                  */
1111                 if (committed) {
1112                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1113                         xfs_trans_ihold(args->trans, dp);
1114                 }
1115         } else
1116                 xfs_da_buf_done(bp);
1117         return(0);
1118 }
1119
1120 /*
1121  * Look up a name in a leaf attribute list structure.
1122  *
1123  * This leaf block cannot have a "remote" value, we only call this routine
1124  * if bmap_one_block() says there is only one block (ie: no remote blks).
1125  */
1126 int
1127 xfs_attr_leaf_get(xfs_da_args_t *args)
1128 {
1129         xfs_dabuf_t *bp;
1130         int error;
1131
1132         args->blkno = 0;
1133         error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1134                                              XFS_ATTR_FORK);
1135         if (error)
1136                 return(error);
1137         ASSERT(bp != NULL);
1138
1139         error = xfs_attr_leaf_lookup_int(bp, args);
1140         if (error != EEXIST)  {
1141                 xfs_da_brelse(args->trans, bp);
1142                 return(error);
1143         }
1144         error = xfs_attr_leaf_getvalue(bp, args);
1145         xfs_da_brelse(args->trans, bp);
1146         if (!error && (args->rmtblkno > 0) && !(args->flags & ATTR_KERNOVAL)) {
1147                 error = xfs_attr_rmtval_get(args);
1148         }
1149         return(error);
1150 }
1151
1152 /*
1153  * Copy out attribute entries for attr_list(), for leaf attribute lists.
1154  */
1155 STATIC int
1156 xfs_attr_leaf_list(xfs_attr_list_context_t *context)
1157 {
1158         xfs_attr_leafblock_t *leaf;
1159         int error;
1160         xfs_dabuf_t *bp;
1161
1162         context->cursor->blkno = 0;
1163         error = xfs_da_read_buf(NULL, context->dp, 0, -1, &bp, XFS_ATTR_FORK);
1164         if (error)
1165                 return(error);
1166         ASSERT(bp != NULL);
1167         leaf = bp->data;
1168         if (unlikely(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT)
1169                                                 != XFS_ATTR_LEAF_MAGIC)) {
1170                 XFS_CORRUPTION_ERROR("xfs_attr_leaf_list", XFS_ERRLEVEL_LOW,
1171                                      context->dp->i_mount, leaf);
1172                 xfs_da_brelse(NULL, bp);
1173                 return(XFS_ERROR(EFSCORRUPTED));
1174         }
1175
1176         (void)xfs_attr_leaf_list_int(bp, context);
1177         xfs_da_brelse(NULL, bp);
1178         return(0);
1179 }
1180
1181
1182 /*========================================================================
1183  * External routines when attribute list size > XFS_LBSIZE(mp).
1184  *========================================================================*/
1185
1186 /*
1187  * Add a name to a Btree-format attribute list.
1188  *
1189  * This will involve walking down the Btree, and may involve splitting
1190  * leaf nodes and even splitting intermediate nodes up to and including
1191  * the root node (a special case of an intermediate node).
1192  *
1193  * "Remote" attribute values confuse the issue and atomic rename operations
1194  * add a whole extra layer of confusion on top of that.
1195  */
1196 STATIC int
1197 xfs_attr_node_addname(xfs_da_args_t *args)
1198 {
1199         xfs_da_state_t *state;
1200         xfs_da_state_blk_t *blk;
1201         xfs_inode_t *dp;
1202         xfs_mount_t *mp;
1203         int committed, retval, error;
1204
1205         /*
1206          * Fill in bucket of arguments/results/context to carry around.
1207          */
1208         dp = args->dp;
1209         mp = dp->i_mount;
1210 restart:
1211         state = xfs_da_state_alloc();
1212         state->args = args;
1213         state->mp = mp;
1214         state->blocksize = state->mp->m_sb.sb_blocksize;
1215         state->node_ents = state->mp->m_attr_node_ents;
1216
1217         /*
1218          * Search to see if name already exists, and get back a pointer
1219          * to where it should go.
1220          */
1221         error = xfs_da_node_lookup_int(state, &retval);
1222         if (error)
1223                 goto out;
1224         blk = &state->path.blk[ state->path.active-1 ];
1225         ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1226         if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
1227                 goto out;
1228         } else if (retval == EEXIST) {
1229                 if (args->flags & ATTR_CREATE)
1230                         goto out;
1231                 args->rename = 1;                       /* atomic rename op */
1232                 args->blkno2 = args->blkno;             /* set 2nd entry info*/
1233                 args->index2 = args->index;
1234                 args->rmtblkno2 = args->rmtblkno;
1235                 args->rmtblkcnt2 = args->rmtblkcnt;
1236                 args->rmtblkno = 0;
1237                 args->rmtblkcnt = 0;
1238         }
1239
1240         retval = xfs_attr_leaf_add(blk->bp, state->args);
1241         if (retval == ENOSPC) {
1242                 if (state->path.active == 1) {
1243                         /*
1244                          * Its really a single leaf node, but it had
1245                          * out-of-line values so it looked like it *might*
1246                          * have been a b-tree.
1247                          */
1248                         xfs_da_state_free(state);
1249                         XFS_BMAP_INIT(args->flist, args->firstblock);
1250                         error = xfs_attr_leaf_to_node(args);
1251                         if (!error) {
1252                                 error = xfs_bmap_finish(&args->trans,
1253                                                         args->flist,
1254                                                         *args->firstblock,
1255                                                         &committed);
1256                         }
1257                         if (error) {
1258                                 ASSERT(committed);
1259                                 args->trans = NULL;
1260                                 xfs_bmap_cancel(args->flist);
1261                                 goto out;
1262                         }
1263
1264                         /*
1265                          * bmap_finish() may have committed the last trans
1266                          * and started a new one.  We need the inode to be
1267                          * in all transactions.
1268                          */
1269                         if (committed) {
1270                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1271                                 xfs_trans_ihold(args->trans, dp);
1272                         }
1273
1274                         /*
1275                          * Commit the node conversion and start the next
1276                          * trans in the chain.
1277                          */
1278                         if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1279                                 goto out;
1280
1281                         goto restart;
1282                 }
1283
1284                 /*
1285                  * Split as many Btree elements as required.
1286                  * This code tracks the new and old attr's location
1287                  * in the index/blkno/rmtblkno/rmtblkcnt fields and
1288                  * in the index2/blkno2/rmtblkno2/rmtblkcnt2 fields.
1289                  */
1290                 XFS_BMAP_INIT(args->flist, args->firstblock);
1291                 error = xfs_da_split(state);
1292                 if (!error) {
1293                         error = xfs_bmap_finish(&args->trans, args->flist,
1294                                                 *args->firstblock, &committed);
1295                 }
1296                 if (error) {
1297                         ASSERT(committed);
1298                         args->trans = NULL;
1299                         xfs_bmap_cancel(args->flist);
1300                         goto out;
1301                 }
1302
1303                 /*
1304                  * bmap_finish() may have committed the last trans and started
1305                  * a new one.  We need the inode to be in all transactions.
1306                  */
1307                 if (committed) {
1308                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1309                         xfs_trans_ihold(args->trans, dp);
1310                 }
1311         } else {
1312                 /*
1313                  * Addition succeeded, update Btree hashvals.
1314                  */
1315                 xfs_da_fixhashpath(state, &state->path);
1316         }
1317
1318         /*
1319          * Kill the state structure, we're done with it and need to
1320          * allow the buffers to come back later.
1321          */
1322         xfs_da_state_free(state);
1323         state = NULL;
1324
1325         /*
1326          * Commit the leaf addition or btree split and start the next
1327          * trans in the chain.
1328          */
1329         if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1330                 goto out;
1331
1332         /*
1333          * If there was an out-of-line value, allocate the blocks we
1334          * identified for its storage and copy the value.  This is done
1335          * after we create the attribute so that we don't overflow the
1336          * maximum size of a transaction and/or hit a deadlock.
1337          */
1338         if (args->rmtblkno > 0) {
1339                 error = xfs_attr_rmtval_set(args);
1340                 if (error)
1341                         return(error);
1342         }
1343
1344         /*
1345          * If this is an atomic rename operation, we must "flip" the
1346          * incomplete flags on the "new" and "old" attribute/value pairs
1347          * so that one disappears and one appears atomically.  Then we
1348          * must remove the "old" attribute/value pair.
1349          */
1350         if (args->rename) {
1351                 /*
1352                  * In a separate transaction, set the incomplete flag on the
1353                  * "old" attr and clear the incomplete flag on the "new" attr.
1354                  */
1355                 error = xfs_attr_leaf_flipflags(args);
1356                 if (error)
1357                         goto out;
1358
1359                 /*
1360                  * Dismantle the "old" attribute/value pair by removing
1361                  * a "remote" value (if it exists).
1362                  */
1363                 args->index = args->index2;
1364                 args->blkno = args->blkno2;
1365                 args->rmtblkno = args->rmtblkno2;
1366                 args->rmtblkcnt = args->rmtblkcnt2;
1367                 if (args->rmtblkno) {
1368                         error = xfs_attr_rmtval_remove(args);
1369                         if (error)
1370                                 return(error);
1371                 }
1372
1373                 /*
1374                  * Re-find the "old" attribute entry after any split ops.
1375                  * The INCOMPLETE flag means that we will find the "old"
1376                  * attr, not the "new" one.
1377                  */
1378                 args->flags |= XFS_ATTR_INCOMPLETE;
1379                 state = xfs_da_state_alloc();
1380                 state->args = args;
1381                 state->mp = mp;
1382                 state->blocksize = state->mp->m_sb.sb_blocksize;
1383                 state->node_ents = state->mp->m_attr_node_ents;
1384                 state->inleaf = 0;
1385                 error = xfs_da_node_lookup_int(state, &retval);
1386                 if (error)
1387                         goto out;
1388
1389                 /*
1390                  * Remove the name and update the hashvals in the tree.
1391                  */
1392                 blk = &state->path.blk[ state->path.active-1 ];
1393                 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1394                 error = xfs_attr_leaf_remove(blk->bp, args);
1395                 xfs_da_fixhashpath(state, &state->path);
1396
1397                 /*
1398                  * Check to see if the tree needs to be collapsed.
1399                  */
1400                 if (retval && (state->path.active > 1)) {
1401                         XFS_BMAP_INIT(args->flist, args->firstblock);
1402                         error = xfs_da_join(state);
1403                         if (!error) {
1404                                 error = xfs_bmap_finish(&args->trans,
1405                                                         args->flist,
1406                                                         *args->firstblock,
1407                                                         &committed);
1408                         }
1409                         if (error) {
1410                                 ASSERT(committed);
1411                                 args->trans = NULL;
1412                                 xfs_bmap_cancel(args->flist);
1413                                 goto out;
1414                         }
1415
1416                         /*
1417                          * bmap_finish() may have committed the last trans
1418                          * and started a new one.  We need the inode to be
1419                          * in all transactions.
1420                          */
1421                         if (committed) {
1422                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1423                                 xfs_trans_ihold(args->trans, dp);
1424                         }
1425                 }
1426
1427                 /*
1428                  * Commit and start the next trans in the chain.
1429                  */
1430                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1431                         goto out;
1432
1433         } else if (args->rmtblkno > 0) {
1434                 /*
1435                  * Added a "remote" value, just clear the incomplete flag.
1436                  */
1437                 error = xfs_attr_leaf_clearflag(args);
1438                 if (error)
1439                         goto out;
1440         }
1441         retval = error = 0;
1442
1443 out:
1444         if (state)
1445                 xfs_da_state_free(state);
1446         if (error)
1447                 return(error);
1448         return(retval);
1449 }
1450
1451 /*
1452  * Remove a name from a B-tree attribute list.
1453  *
1454  * This will involve walking down the Btree, and may involve joining
1455  * leaf nodes and even joining intermediate nodes up to and including
1456  * the root node (a special case of an intermediate node).
1457  */
1458 STATIC int
1459 xfs_attr_node_removename(xfs_da_args_t *args)
1460 {
1461         xfs_da_state_t *state;
1462         xfs_da_state_blk_t *blk;
1463         xfs_inode_t *dp;
1464         xfs_dabuf_t *bp;
1465         int retval, error, committed;
1466
1467         /*
1468          * Tie a string around our finger to remind us where we are.
1469          */
1470         dp = args->dp;
1471         state = xfs_da_state_alloc();
1472         state->args = args;
1473         state->mp = dp->i_mount;
1474         state->blocksize = state->mp->m_sb.sb_blocksize;
1475         state->node_ents = state->mp->m_attr_node_ents;
1476
1477         /*
1478          * Search to see if name exists, and get back a pointer to it.
1479          */
1480         error = xfs_da_node_lookup_int(state, &retval);
1481         if (error || (retval != EEXIST)) {
1482                 if (error == 0)
1483                         error = retval;
1484                 goto out;
1485         }
1486
1487         /*
1488          * If there is an out-of-line value, de-allocate the blocks.
1489          * This is done before we remove the attribute so that we don't
1490          * overflow the maximum size of a transaction and/or hit a deadlock.
1491          */
1492         blk = &state->path.blk[ state->path.active-1 ];
1493         ASSERT(blk->bp != NULL);
1494         ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1495         if (args->rmtblkno > 0) {
1496                 /*
1497                  * Fill in disk block numbers in the state structure
1498                  * so that we can get the buffers back after we commit
1499                  * several transactions in the following calls.
1500                  */
1501                 error = xfs_attr_fillstate(state);
1502                 if (error)
1503                         goto out;
1504
1505                 /*
1506                  * Mark the attribute as INCOMPLETE, then bunmapi() the
1507                  * remote value.
1508                  */
1509                 error = xfs_attr_leaf_setflag(args);
1510                 if (error)
1511                         goto out;
1512                 error = xfs_attr_rmtval_remove(args);
1513                 if (error)
1514                         goto out;
1515
1516                 /*
1517                  * Refill the state structure with buffers, the prior calls
1518                  * released our buffers.
1519                  */
1520                 error = xfs_attr_refillstate(state);
1521                 if (error)
1522                         goto out;
1523         }
1524
1525         /*
1526          * Remove the name and update the hashvals in the tree.
1527          */
1528         blk = &state->path.blk[ state->path.active-1 ];
1529         ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1530         retval = xfs_attr_leaf_remove(blk->bp, args);
1531         xfs_da_fixhashpath(state, &state->path);
1532
1533         /*
1534          * Check to see if the tree needs to be collapsed.
1535          */
1536         if (retval && (state->path.active > 1)) {
1537                 XFS_BMAP_INIT(args->flist, args->firstblock);
1538                 error = xfs_da_join(state);
1539                 if (!error) {
1540                         error = xfs_bmap_finish(&args->trans, args->flist,
1541                                                 *args->firstblock, &committed);
1542                 }
1543                 if (error) {
1544                         ASSERT(committed);
1545                         args->trans = NULL;
1546                         xfs_bmap_cancel(args->flist);
1547                         goto out;
1548                 }
1549
1550                 /*
1551                  * bmap_finish() may have committed the last trans and started
1552                  * a new one.  We need the inode to be in all transactions.
1553                  */
1554                 if (committed) {
1555                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1556                         xfs_trans_ihold(args->trans, dp);
1557                 }
1558
1559                 /*
1560                  * Commit the Btree join operation and start a new trans.
1561                  */
1562                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
1563                         goto out;
1564         }
1565
1566         /*
1567          * If the result is small enough, push it all into the inode.
1568          */
1569         if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
1570                 /*
1571                  * Have to get rid of the copy of this dabuf in the state.
1572                  */
1573                 ASSERT(state->path.active == 1);
1574                 ASSERT(state->path.blk[0].bp);
1575                 xfs_da_buf_done(state->path.blk[0].bp);
1576                 state->path.blk[0].bp = NULL;
1577
1578                 error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp,
1579                                                      XFS_ATTR_FORK);
1580                 if (error)
1581                         goto out;
1582                 ASSERT(INT_GET(((xfs_attr_leafblock_t *)
1583                                       bp->data)->hdr.info.magic, ARCH_CONVERT)
1584                                                        == XFS_ATTR_LEAF_MAGIC);
1585
1586                 if (xfs_attr_shortform_allfit(bp, dp)) {
1587                         XFS_BMAP_INIT(args->flist, args->firstblock);
1588                         error = xfs_attr_leaf_to_shortform(bp, args);
1589                         /* bp is gone due to xfs_da_shrink_inode */
1590                         if (!error) {
1591                                 error = xfs_bmap_finish(&args->trans,
1592                                                         args->flist,
1593                                                         *args->firstblock,
1594                                                         &committed);
1595                         }
1596                         if (error) {
1597                                 ASSERT(committed);
1598                                 args->trans = NULL;
1599                                 xfs_bmap_cancel(args->flist);
1600                                 goto out;
1601                         }
1602
1603                         /*
1604                          * bmap_finish() may have committed the last trans
1605                          * and started a new one.  We need the inode to be
1606                          * in all transactions.
1607                          */
1608                         if (committed) {
1609                                 xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1610                                 xfs_trans_ihold(args->trans, dp);
1611                         }
1612                 } else
1613                         xfs_da_brelse(args->trans, bp);
1614         }
1615         error = 0;
1616
1617 out:
1618         xfs_da_state_free(state);
1619         return(error);
1620 }
1621
1622 /*
1623  * Fill in the disk block numbers in the state structure for the buffers
1624  * that are attached to the state structure.
1625  * This is done so that we can quickly reattach ourselves to those buffers
1626  * after some set of transaction commit's has released these buffers.
1627  */
1628 STATIC int
1629 xfs_attr_fillstate(xfs_da_state_t *state)
1630 {
1631         xfs_da_state_path_t *path;
1632         xfs_da_state_blk_t *blk;
1633         int level;
1634
1635         /*
1636          * Roll down the "path" in the state structure, storing the on-disk
1637          * block number for those buffers in the "path".
1638          */
1639         path = &state->path;
1640         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1641         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1642                 if (blk->bp) {
1643                         blk->disk_blkno = xfs_da_blkno(blk->bp);
1644                         xfs_da_buf_done(blk->bp);
1645                         blk->bp = NULL;
1646                 } else {
1647                         blk->disk_blkno = 0;
1648                 }
1649         }
1650
1651         /*
1652          * Roll down the "altpath" in the state structure, storing the on-disk
1653          * block number for those buffers in the "altpath".
1654          */
1655         path = &state->altpath;
1656         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1657         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1658                 if (blk->bp) {
1659                         blk->disk_blkno = xfs_da_blkno(blk->bp);
1660                         xfs_da_buf_done(blk->bp);
1661                         blk->bp = NULL;
1662                 } else {
1663                         blk->disk_blkno = 0;
1664                 }
1665         }
1666
1667         return(0);
1668 }
1669
1670 /*
1671  * Reattach the buffers to the state structure based on the disk block
1672  * numbers stored in the state structure.
1673  * This is done after some set of transaction commit's has released those
1674  * buffers from our grip.
1675  */
1676 STATIC int
1677 xfs_attr_refillstate(xfs_da_state_t *state)
1678 {
1679         xfs_da_state_path_t *path;
1680         xfs_da_state_blk_t *blk;
1681         int level, error;
1682
1683         /*
1684          * Roll down the "path" in the state structure, storing the on-disk
1685          * block number for those buffers in the "path".
1686          */
1687         path = &state->path;
1688         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1689         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1690                 if (blk->disk_blkno) {
1691                         error = xfs_da_read_buf(state->args->trans,
1692                                                 state->args->dp,
1693                                                 blk->blkno, blk->disk_blkno,
1694                                                 &blk->bp, XFS_ATTR_FORK);
1695                         if (error)
1696                                 return(error);
1697                 } else {
1698                         blk->bp = NULL;
1699                 }
1700         }
1701
1702         /*
1703          * Roll down the "altpath" in the state structure, storing the on-disk
1704          * block number for those buffers in the "altpath".
1705          */
1706         path = &state->altpath;
1707         ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1708         for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1709                 if (blk->disk_blkno) {
1710                         error = xfs_da_read_buf(state->args->trans,
1711                                                 state->args->dp,
1712                                                 blk->blkno, blk->disk_blkno,
1713                                                 &blk->bp, XFS_ATTR_FORK);
1714                         if (error)
1715                                 return(error);
1716                 } else {
1717                         blk->bp = NULL;
1718                 }
1719         }
1720
1721         return(0);
1722 }
1723
1724 /*
1725  * Look up a filename in a node attribute list.
1726  *
1727  * This routine gets called for any attribute fork that has more than one
1728  * block, ie: both true Btree attr lists and for single-leaf-blocks with
1729  * "remote" values taking up more blocks.
1730  */
1731 int
1732 xfs_attr_node_get(xfs_da_args_t *args)
1733 {
1734         xfs_da_state_t *state;
1735         xfs_da_state_blk_t *blk;
1736         int error, retval;
1737         int i;
1738
1739         state = xfs_da_state_alloc();
1740         state->args = args;
1741         state->mp = args->dp->i_mount;
1742         state->blocksize = state->mp->m_sb.sb_blocksize;
1743         state->node_ents = state->mp->m_attr_node_ents;
1744
1745         /*
1746          * Search to see if name exists, and get back a pointer to it.
1747          */
1748         error = xfs_da_node_lookup_int(state, &retval);
1749         if (error) {
1750                 retval = error;
1751         } else if (retval == EEXIST) {
1752                 blk = &state->path.blk[ state->path.active-1 ];
1753                 ASSERT(blk->bp != NULL);
1754                 ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1755
1756                 /*
1757                  * Get the value, local or "remote"
1758                  */
1759                 retval = xfs_attr_leaf_getvalue(blk->bp, args);
1760                 if (!retval && (args->rmtblkno > 0)
1761                     && !(args->flags & ATTR_KERNOVAL)) {
1762                         retval = xfs_attr_rmtval_get(args);
1763                 }
1764         }
1765
1766         /*
1767          * If not in a transaction, we have to release all the buffers.
1768          */
1769         for (i = 0; i < state->path.active; i++) {
1770                 xfs_da_brelse(args->trans, state->path.blk[i].bp);
1771                 state->path.blk[i].bp = NULL;
1772         }
1773
1774         xfs_da_state_free(state);
1775         return(retval);
1776 }
1777
1778 STATIC int                                                      /* error */
1779 xfs_attr_node_list(xfs_attr_list_context_t *context)
1780 {
1781         attrlist_cursor_kern_t *cursor;
1782         xfs_attr_leafblock_t *leaf;
1783         xfs_da_intnode_t *node;
1784         xfs_da_node_entry_t *btree;
1785         int error, i;
1786         xfs_dabuf_t *bp;
1787
1788         cursor = context->cursor;
1789         cursor->initted = 1;
1790
1791         /*
1792          * Do all sorts of validation on the passed-in cursor structure.
1793          * If anything is amiss, ignore the cursor and look up the hashval
1794          * starting from the btree root.
1795          */
1796         bp = NULL;
1797         if (cursor->blkno > 0) {
1798                 error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1799                                               &bp, XFS_ATTR_FORK);
1800                 if ((error != 0) && (error != EFSCORRUPTED))
1801                         return(error);
1802                 if (bp) {
1803                         node = bp->data;
1804                         switch (INT_GET(node->hdr.info.magic, ARCH_CONVERT)) {
1805                         case XFS_DA_NODE_MAGIC:
1806                                 xfs_attr_trace_l_cn("wrong blk", context, node);
1807                                 xfs_da_brelse(NULL, bp);
1808                                 bp = NULL;
1809                                 break;
1810                         case XFS_ATTR_LEAF_MAGIC:
1811                                 leaf = bp->data;
1812                                 if (cursor->hashval >
1813                                     INT_GET(leaf->entries[
1814                                          INT_GET(leaf->hdr.count,
1815                                                 ARCH_CONVERT)-1].hashval,
1816                                                         ARCH_CONVERT)) {
1817                                         xfs_attr_trace_l_cl("wrong blk",
1818                                                            context, leaf);
1819                                         xfs_da_brelse(NULL, bp);
1820                                         bp = NULL;
1821                                 } else if (cursor->hashval <=
1822                                              INT_GET(leaf->entries[0].hashval,
1823                                                         ARCH_CONVERT)) {
1824                                         xfs_attr_trace_l_cl("maybe wrong blk",
1825                                                            context, leaf);
1826                                         xfs_da_brelse(NULL, bp);
1827                                         bp = NULL;
1828                                 }
1829                                 break;
1830                         default:
1831                                 xfs_attr_trace_l_c("wrong blk - ??", context);
1832                                 xfs_da_brelse(NULL, bp);
1833                                 bp = NULL;
1834                         }
1835                 }
1836         }
1837
1838         /*
1839          * We did not find what we expected given the cursor's contents,
1840          * so we start from the top and work down based on the hash value.
1841          * Note that start of node block is same as start of leaf block.
1842          */
1843         if (bp == NULL) {
1844                 cursor->blkno = 0;
1845                 for (;;) {
1846                         error = xfs_da_read_buf(NULL, context->dp,
1847                                                       cursor->blkno, -1, &bp,
1848                                                       XFS_ATTR_FORK);
1849                         if (error)
1850                                 return(error);
1851                         if (unlikely(bp == NULL)) {
1852                                 XFS_ERROR_REPORT("xfs_attr_node_list(2)",
1853                                                  XFS_ERRLEVEL_LOW,
1854                                                  context->dp->i_mount);
1855                                 return(XFS_ERROR(EFSCORRUPTED));
1856                         }
1857                         node = bp->data;
1858                         if (INT_GET(node->hdr.info.magic, ARCH_CONVERT)
1859                                                         == XFS_ATTR_LEAF_MAGIC)
1860                                 break;
1861                         if (unlikely(INT_GET(node->hdr.info.magic, ARCH_CONVERT)
1862                                                         != XFS_DA_NODE_MAGIC)) {
1863                                 XFS_CORRUPTION_ERROR("xfs_attr_node_list(3)",
1864                                                      XFS_ERRLEVEL_LOW,
1865                                                      context->dp->i_mount,
1866                                                      node);
1867                                 xfs_da_brelse(NULL, bp);
1868                                 return(XFS_ERROR(EFSCORRUPTED));
1869                         }
1870                         btree = node->btree;
1871                         for (i = 0;
1872                                 i < INT_GET(node->hdr.count, ARCH_CONVERT);
1873                                                                 btree++, i++) {
1874                                 if (cursor->hashval
1875                                                 <= INT_GET(btree->hashval,
1876                                                             ARCH_CONVERT)) {
1877                                         cursor->blkno = INT_GET(btree->before, ARCH_CONVERT);
1878                                         xfs_attr_trace_l_cb("descending",
1879                                                             context, btree);
1880                                         break;
1881                                 }
1882                         }
1883                         if (i == INT_GET(node->hdr.count, ARCH_CONVERT)) {
1884                                 xfs_da_brelse(NULL, bp);
1885                                 return(0);
1886                         }
1887                         xfs_da_brelse(NULL, bp);
1888                 }
1889         }
1890         ASSERT(bp != NULL);
1891
1892         /*
1893          * Roll upward through the blocks, processing each leaf block in
1894          * order.  As long as there is space in the result buffer, keep
1895          * adding the information.
1896          */
1897         for (;;) {
1898                 leaf = bp->data;
1899                 if (unlikely(INT_GET(leaf->hdr.info.magic, ARCH_CONVERT)
1900                                                 != XFS_ATTR_LEAF_MAGIC)) {
1901                         XFS_CORRUPTION_ERROR("xfs_attr_node_list(4)",
1902                                              XFS_ERRLEVEL_LOW,
1903                                              context->dp->i_mount, leaf);
1904                         xfs_da_brelse(NULL, bp);
1905                         return(XFS_ERROR(EFSCORRUPTED));
1906                 }
1907                 error = xfs_attr_leaf_list_int(bp, context);
1908                 if (error || (INT_ISZERO(leaf->hdr.info.forw, ARCH_CONVERT)))
1909                         break;  /* not really an error, buffer full or EOF */
1910                 cursor->blkno = INT_GET(leaf->hdr.info.forw, ARCH_CONVERT);
1911                 xfs_da_brelse(NULL, bp);
1912                 error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1913                                               &bp, XFS_ATTR_FORK);
1914                 if (error)
1915                         return(error);
1916                 if (unlikely((bp == NULL))) {
1917                         XFS_ERROR_REPORT("xfs_attr_node_list(5)",
1918                                          XFS_ERRLEVEL_LOW,
1919                                          context->dp->i_mount);
1920                         return(XFS_ERROR(EFSCORRUPTED));
1921                 }
1922         }
1923         xfs_da_brelse(NULL, bp);
1924         return(0);
1925 }
1926
1927
1928 /*========================================================================
1929  * External routines for manipulating out-of-line attribute values.
1930  *========================================================================*/
1931
1932 /*
1933  * Read the value associated with an attribute from the out-of-line buffer
1934  * that we stored it in.
1935  */
1936 STATIC int
1937 xfs_attr_rmtval_get(xfs_da_args_t *args)
1938 {
1939         xfs_bmbt_irec_t map[ATTR_RMTVALUE_MAPSIZE];
1940         xfs_mount_t *mp;
1941         xfs_daddr_t dblkno;
1942         xfs_caddr_t dst;
1943         xfs_buf_t *bp;
1944         int nmap, error, tmp, valuelen, blkcnt, i;
1945         xfs_dablk_t lblkno;
1946
1947         ASSERT(!(args->flags & ATTR_KERNOVAL));
1948
1949         mp = args->dp->i_mount;
1950         dst = args->value;
1951         valuelen = args->valuelen;
1952         lblkno = args->rmtblkno;
1953         while (valuelen > 0) {
1954                 nmap = ATTR_RMTVALUE_MAPSIZE;
1955                 error = xfs_bmapi(args->trans, args->dp, (xfs_fileoff_t)lblkno,
1956                                   args->rmtblkcnt,
1957                                   XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
1958                                   NULL, 0, map, &nmap, NULL);
1959                 if (error)
1960                         return(error);
1961                 ASSERT(nmap >= 1);
1962
1963                 for (i = 0; (i < nmap) && (valuelen > 0); i++) {
1964                         ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
1965                                (map[i].br_startblock != HOLESTARTBLOCK));
1966                         dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
1967                         blkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
1968                         error = xfs_read_buf(mp, mp->m_ddev_targp, dblkno,
1969                                              blkcnt, XFS_BUF_LOCK, &bp);
1970                         if (error)
1971                                 return(error);
1972
1973                         tmp = (valuelen < XFS_BUF_SIZE(bp))
1974                                 ? valuelen : XFS_BUF_SIZE(bp);
1975                         xfs_biomove(bp, 0, tmp, dst, XFS_B_READ);
1976                         xfs_buf_relse(bp);
1977                         dst += tmp;
1978                         valuelen -= tmp;
1979
1980                         lblkno += map[i].br_blockcount;
1981                 }
1982         }
1983         ASSERT(valuelen == 0);
1984         return(0);
1985 }
1986
1987 /*
1988  * Write the value associated with an attribute into the out-of-line buffer
1989  * that we have defined for it.
1990  */
1991 STATIC int
1992 xfs_attr_rmtval_set(xfs_da_args_t *args)
1993 {
1994         xfs_mount_t *mp;
1995         xfs_fileoff_t lfileoff;
1996         xfs_inode_t *dp;
1997         xfs_bmbt_irec_t map;
1998         xfs_daddr_t dblkno;
1999         xfs_caddr_t src;
2000         xfs_buf_t *bp;
2001         xfs_dablk_t lblkno;
2002         int blkcnt, valuelen, nmap, error, tmp, committed;
2003
2004         dp = args->dp;
2005         mp = dp->i_mount;
2006         src = args->value;
2007
2008         /*
2009          * Find a "hole" in the attribute address space large enough for
2010          * us to drop the new attribute's value into.
2011          */
2012         blkcnt = XFS_B_TO_FSB(mp, args->valuelen);
2013         lfileoff = 0;
2014         error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
2015                                                    XFS_ATTR_FORK);
2016         if (error) {
2017                 return(error);
2018         }
2019         args->rmtblkno = lblkno = (xfs_dablk_t)lfileoff;
2020         args->rmtblkcnt = blkcnt;
2021
2022         /*
2023          * Roll through the "value", allocating blocks on disk as required.
2024          */
2025         while (blkcnt > 0) {
2026                 /*
2027                  * Allocate a single extent, up to the size of the value.
2028                  */
2029                 XFS_BMAP_INIT(args->flist, args->firstblock);
2030                 nmap = 1;
2031                 error = xfs_bmapi(args->trans, dp, (xfs_fileoff_t)lblkno,
2032                                   blkcnt,
2033                                   XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA |
2034                                                         XFS_BMAPI_WRITE,
2035                                   args->firstblock, args->total, &map, &nmap,
2036                                   args->flist);
2037                 if (!error) {
2038                         error = xfs_bmap_finish(&args->trans, args->flist,
2039                                                 *args->firstblock, &committed);
2040                 }
2041                 if (error) {
2042                         ASSERT(committed);
2043                         args->trans = NULL;
2044                         xfs_bmap_cancel(args->flist);
2045                         return(error);
2046                 }
2047
2048                 /*
2049                  * bmap_finish() may have committed the last trans and started
2050                  * a new one.  We need the inode to be in all transactions.
2051                  */
2052                 if (committed) {
2053                         xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
2054                         xfs_trans_ihold(args->trans, dp);
2055                 }
2056
2057                 ASSERT(nmap == 1);
2058                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2059                        (map.br_startblock != HOLESTARTBLOCK));
2060                 lblkno += map.br_blockcount;
2061                 blkcnt -= map.br_blockcount;
2062
2063                 /*
2064                  * Start the next trans in the chain.
2065                  */
2066                 if ((error = xfs_attr_rolltrans(&args->trans, dp)))
2067                         return (error);
2068         }
2069
2070         /*
2071          * Roll through the "value", copying the attribute value to the
2072          * already-allocated blocks.  Blocks are written synchronously
2073          * so that we can know they are all on disk before we turn off
2074          * the INCOMPLETE flag.
2075          */
2076         lblkno = args->rmtblkno;
2077         valuelen = args->valuelen;
2078         while (valuelen > 0) {
2079                 /*
2080                  * Try to remember where we decided to put the value.
2081                  */
2082                 XFS_BMAP_INIT(args->flist, args->firstblock);
2083                 nmap = 1;
2084                 error = xfs_bmapi(NULL, dp, (xfs_fileoff_t)lblkno,
2085                                   args->rmtblkcnt,
2086                                   XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2087                                   args->firstblock, 0, &map, &nmap, NULL);
2088                 if (error) {
2089                         return(error);
2090                 }
2091                 ASSERT(nmap == 1);
2092                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2093                        (map.br_startblock != HOLESTARTBLOCK));
2094
2095                 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2096                 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2097
2098                 bp = xfs_buf_get_flags(mp->m_ddev_targp, dblkno,
2099                                                         blkcnt, XFS_BUF_LOCK);
2100                 ASSERT(bp);
2101                 ASSERT(!XFS_BUF_GETERROR(bp));
2102
2103                 tmp = (valuelen < XFS_BUF_SIZE(bp)) ? valuelen :
2104                                                         XFS_BUF_SIZE(bp);
2105                 xfs_biomove(bp, 0, tmp, src, XFS_B_WRITE);
2106                 if (tmp < XFS_BUF_SIZE(bp))
2107                         xfs_biozero(bp, tmp, XFS_BUF_SIZE(bp) - tmp);
2108                 if ((error = xfs_bwrite(mp, bp))) {/* GROT: NOTE: synchronous write */
2109                         return (error);
2110                 }
2111                 src += tmp;
2112                 valuelen -= tmp;
2113
2114                 lblkno += map.br_blockcount;
2115         }
2116         ASSERT(valuelen == 0);
2117         return(0);
2118 }
2119
2120 /*
2121  * Remove the value associated with an attribute by deleting the
2122  * out-of-line buffer that it is stored on.
2123  */
2124 STATIC int
2125 xfs_attr_rmtval_remove(xfs_da_args_t *args)
2126 {
2127         xfs_mount_t *mp;
2128         xfs_bmbt_irec_t map;
2129         xfs_buf_t *bp;
2130         xfs_daddr_t dblkno;
2131         xfs_dablk_t lblkno;
2132         int valuelen, blkcnt, nmap, error, done, committed;
2133
2134         mp = args->dp->i_mount;
2135
2136         /*
2137          * Roll through the "value", invalidating the attribute value's
2138          * blocks.
2139          */
2140         lblkno = args->rmtblkno;
2141         valuelen = args->rmtblkcnt;
2142         while (valuelen > 0) {
2143                 /*
2144                  * Try to remember where we decided to put the value.
2145                  */
2146                 XFS_BMAP_INIT(args->flist, args->firstblock);
2147                 nmap = 1;
2148                 error = xfs_bmapi(NULL, args->dp, (xfs_fileoff_t)lblkno,
2149                                         args->rmtblkcnt,
2150                                         XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2151                                         args->firstblock, 0, &map, &nmap,
2152                                         args->flist);
2153                 if (error) {
2154                         return(error);
2155                 }
2156                 ASSERT(nmap == 1);
2157                 ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2158                        (map.br_startblock != HOLESTARTBLOCK));
2159
2160                 dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2161                 blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2162
2163                 /*
2164                  * If the "remote" value is in the cache, remove it.
2165                  */
2166                 /* bp = incore(mp->m_dev, dblkno, blkcnt, 1); */
2167                 bp = xfs_incore(mp->m_ddev_targp, dblkno, blkcnt, 1);
2168                 if (bp) {
2169                         XFS_BUF_STALE(bp);
2170                         XFS_BUF_UNDELAYWRITE(bp);
2171                         xfs_buf_relse(bp);
2172                         bp = NULL;
2173                 }
2174
2175                 valuelen -= map.br_blockcount;
2176
2177                 lblkno += map.br_blockcount;
2178         }
2179
2180         /*
2181          * Keep de-allocating extents until the remote-value region is gone.
2182          */
2183         lblkno = args->rmtblkno;
2184         blkcnt = args->rmtblkcnt;
2185         done = 0;
2186         while (!done) {
2187                 XFS_BMAP_INIT(args->flist, args->firstblock);
2188                 error = xfs_bunmapi(args->trans, args->dp, lblkno, blkcnt,
2189                                     XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2190                                     1, args->firstblock, args->flist, &done);
2191                 if (!error) {
2192                         error = xfs_bmap_finish(&args->trans, args->flist,
2193                                                 *args->firstblock, &committed);
2194                 }
2195                 if (error) {
2196                         ASSERT(committed);
2197                         args->trans = NULL;
2198                         xfs_bmap_cancel(args->flist);
2199                         return(error);
2200                 }
2201
2202                 /*
2203                  * bmap_finish() may have committed the last trans and started
2204                  * a new one.  We need the inode to be in all transactions.
2205                  */
2206                 if (committed) {
2207                         xfs_trans_ijoin(args->trans, args->dp, XFS_ILOCK_EXCL);
2208                         xfs_trans_ihold(args->trans, args->dp);
2209                 }
2210
2211                 /*
2212                  * Close out trans and start the next one in the chain.
2213                  */
2214                 if ((error = xfs_attr_rolltrans(&args->trans, args->dp)))
2215                         return (error);
2216         }
2217         return(0);
2218 }
2219
2220 #if defined(XFS_ATTR_TRACE)
2221 /*
2222  * Add a trace buffer entry for an attr_list context structure.
2223  */
2224 void
2225 xfs_attr_trace_l_c(char *where, struct xfs_attr_list_context *context)
2226 {
2227         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_C, where,
2228                 (__psunsigned_t)context->dp,
2229                 (__psunsigned_t)context->cursor->hashval,
2230                 (__psunsigned_t)context->cursor->blkno,
2231                 (__psunsigned_t)context->cursor->offset,
2232                 (__psunsigned_t)context->alist,
2233                 (__psunsigned_t)context->bufsize,
2234                 (__psunsigned_t)context->count,
2235                 (__psunsigned_t)context->firstu,
2236                 (__psunsigned_t)
2237                         ((context->count > 0) &&
2238                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2239                                 ? (ATTR_ENTRY(context->alist,
2240                                               context->count-1)->a_valuelen)
2241                                 : 0,
2242                 (__psunsigned_t)context->dupcnt,
2243                 (__psunsigned_t)context->flags,
2244                 (__psunsigned_t)NULL,
2245                 (__psunsigned_t)NULL,
2246                 (__psunsigned_t)NULL);
2247 }
2248
2249 /*
2250  * Add a trace buffer entry for a context structure and a Btree node.
2251  */
2252 void
2253 xfs_attr_trace_l_cn(char *where, struct xfs_attr_list_context *context,
2254                          struct xfs_da_intnode *node)
2255 {
2256         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_CN, where,
2257                 (__psunsigned_t)context->dp,
2258                 (__psunsigned_t)context->cursor->hashval,
2259                 (__psunsigned_t)context->cursor->blkno,
2260                 (__psunsigned_t)context->cursor->offset,
2261                 (__psunsigned_t)context->alist,
2262                 (__psunsigned_t)context->bufsize,
2263                 (__psunsigned_t)context->count,
2264                 (__psunsigned_t)context->firstu,
2265                 (__psunsigned_t)
2266                         ((context->count > 0) &&
2267                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2268                                 ? (ATTR_ENTRY(context->alist,
2269                                               context->count-1)->a_valuelen)
2270                                 : 0,
2271                 (__psunsigned_t)context->dupcnt,
2272                 (__psunsigned_t)context->flags,
2273                 (__psunsigned_t)INT_GET(node->hdr.count, ARCH_CONVERT),
2274                 (__psunsigned_t)INT_GET(node->btree[0].hashval, ARCH_CONVERT),
2275                 (__psunsigned_t)INT_GET(node->btree[INT_GET(node->hdr.count, ARCH_CONVERT)-1].hashval, ARCH_CONVERT));
2276 }
2277
2278 /*
2279  * Add a trace buffer entry for a context structure and a Btree element.
2280  */
2281 void
2282 xfs_attr_trace_l_cb(char *where, struct xfs_attr_list_context *context,
2283                           struct xfs_da_node_entry *btree)
2284 {
2285         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_CB, where,
2286                 (__psunsigned_t)context->dp,
2287                 (__psunsigned_t)context->cursor->hashval,
2288                 (__psunsigned_t)context->cursor->blkno,
2289                 (__psunsigned_t)context->cursor->offset,
2290                 (__psunsigned_t)context->alist,
2291                 (__psunsigned_t)context->bufsize,
2292                 (__psunsigned_t)context->count,
2293                 (__psunsigned_t)context->firstu,
2294                 (__psunsigned_t)
2295                         ((context->count > 0) &&
2296                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2297                                 ? (ATTR_ENTRY(context->alist,
2298                                               context->count-1)->a_valuelen)
2299                                 : 0,
2300                 (__psunsigned_t)context->dupcnt,
2301                 (__psunsigned_t)context->flags,
2302                 (__psunsigned_t)INT_GET(btree->hashval, ARCH_CONVERT),
2303                 (__psunsigned_t)INT_GET(btree->before, ARCH_CONVERT),
2304                 (__psunsigned_t)NULL);
2305 }
2306
2307 /*
2308  * Add a trace buffer entry for a context structure and a leaf block.
2309  */
2310 void
2311 xfs_attr_trace_l_cl(char *where, struct xfs_attr_list_context *context,
2312                               struct xfs_attr_leafblock *leaf)
2313 {
2314         xfs_attr_trace_enter(XFS_ATTR_KTRACE_L_CL, where,
2315                 (__psunsigned_t)context->dp,
2316                 (__psunsigned_t)context->cursor->hashval,
2317                 (__psunsigned_t)context->cursor->blkno,
2318                 (__psunsigned_t)context->cursor->offset,
2319                 (__psunsigned_t)context->alist,
2320                 (__psunsigned_t)context->bufsize,
2321                 (__psunsigned_t)context->count,
2322                 (__psunsigned_t)context->firstu,
2323                 (__psunsigned_t)
2324                         ((context->count > 0) &&
2325                         !(context->flags & (ATTR_KERNAMELS|ATTR_KERNOVAL)))
2326                                 ? (ATTR_ENTRY(context->alist,
2327                                               context->count-1)->a_valuelen)
2328                                 : 0,
2329                 (__psunsigned_t)context->dupcnt,
2330                 (__psunsigned_t)context->flags,
2331                 (__psunsigned_t)INT_GET(leaf->hdr.count, ARCH_CONVERT),
2332                 (__psunsigned_t)INT_GET(leaf->entries[0].hashval, ARCH_CONVERT),
2333                 (__psunsigned_t)INT_GET(leaf->entries[INT_GET(leaf->hdr.count, ARCH_CONVERT)-1].hashval, ARCH_CONVERT));
2334 }
2335
2336 /*
2337  * Add a trace buffer entry for the arguments given to the routine,
2338  * generic form.
2339  */
2340 void
2341 xfs_attr_trace_enter(int type, char *where,
2342                          __psunsigned_t a2, __psunsigned_t a3,
2343                          __psunsigned_t a4, __psunsigned_t a5,
2344                          __psunsigned_t a6, __psunsigned_t a7,
2345                          __psunsigned_t a8, __psunsigned_t a9,
2346                          __psunsigned_t a10, __psunsigned_t a11,
2347                          __psunsigned_t a12, __psunsigned_t a13,
2348                          __psunsigned_t a14, __psunsigned_t a15)
2349 {
2350         ASSERT(xfs_attr_trace_buf);
2351         ktrace_enter(xfs_attr_trace_buf, (void *)((__psunsigned_t)type),
2352                                          (void *)where,
2353                                          (void *)a2,  (void *)a3,  (void *)a4,
2354                                          (void *)a5,  (void *)a6,  (void *)a7,
2355                                          (void *)a8,  (void *)a9,  (void *)a10,
2356                                          (void *)a11, (void *)a12, (void *)a13,
2357                                          (void *)a14, (void *)a15);
2358 }
2359 #endif  /* XFS_ATTR_TRACE */
2360
2361
2362 /*========================================================================
2363  * System (pseudo) namespace attribute interface routines.
2364  *========================================================================*/
2365
2366 STATIC int
2367 posix_acl_access_set(
2368         vnode_t *vp, char *name, void *data, size_t size, int xflags)
2369 {
2370         return xfs_acl_vset(vp, data, size, _ACL_TYPE_ACCESS);
2371 }
2372
2373 STATIC int
2374 posix_acl_access_remove(
2375         struct vnode *vp, char *name, int xflags)
2376 {
2377         return xfs_acl_vremove(vp, _ACL_TYPE_ACCESS);
2378 }
2379
2380 STATIC int
2381 posix_acl_access_get(
2382         vnode_t *vp, char *name, void *data, size_t size, int xflags)
2383 {
2384         return xfs_acl_vget(vp, data, size, _ACL_TYPE_ACCESS);
2385 }
2386
2387 STATIC int
2388 posix_acl_access_exists(
2389         vnode_t *vp)
2390 {
2391         return xfs_acl_vhasacl_access(vp);
2392 }
2393
2394 STATIC int
2395 posix_acl_default_set(
2396         vnode_t *vp, char *name, void *data, size_t size, int xflags)
2397 {
2398         return xfs_acl_vset(vp, data, size, _ACL_TYPE_DEFAULT);
2399 }
2400
2401 STATIC int
2402 posix_acl_default_get(
2403         vnode_t *vp, char *name, void *data, size_t size, int xflags)
2404 {
2405         return xfs_acl_vget(vp, data, size, _ACL_TYPE_DEFAULT);
2406 }
2407
2408 STATIC int
2409 posix_acl_default_remove(
2410         struct vnode *vp, char *name, int xflags)
2411 {
2412         return xfs_acl_vremove(vp, _ACL_TYPE_DEFAULT);
2413 }
2414
2415 STATIC int
2416 posix_acl_default_exists(
2417         vnode_t *vp)
2418 {
2419         return xfs_acl_vhasacl_default(vp);
2420 }
2421
2422 struct attrnames posix_acl_access = {
2423         .attr_name      = "posix_acl_access",
2424         .attr_namelen   = sizeof("posix_acl_access") - 1,
2425         .attr_get       = posix_acl_access_get,
2426         .attr_set       = posix_acl_access_set,
2427         .attr_remove    = posix_acl_access_remove,
2428         .attr_exists    = posix_acl_access_exists,
2429 };
2430
2431 struct attrnames posix_acl_default = {
2432         .attr_name      = "posix_acl_default",
2433         .attr_namelen   = sizeof("posix_acl_default") - 1,
2434         .attr_get       = posix_acl_default_get,
2435         .attr_set       = posix_acl_default_set,
2436         .attr_remove    = posix_acl_default_remove,
2437         .attr_exists    = posix_acl_default_exists,
2438 };
2439
2440 struct attrnames *attr_system_names[] =
2441         { &posix_acl_access, &posix_acl_default };
2442
2443
2444 /*========================================================================
2445  * Namespace-prefix-style attribute name interface routines.
2446  *========================================================================*/
2447
2448 STATIC int
2449 attr_generic_set(
2450         struct vnode *vp, char *name, void *data, size_t size, int xflags)
2451 {
2452         int     error;
2453
2454         VOP_ATTR_SET(vp, name, data, size, xflags, NULL, error);
2455         return -error;
2456 }
2457
2458 STATIC int
2459 attr_generic_get(
2460         struct vnode *vp, char *name, void *data, size_t size, int xflags)
2461 {
2462         int     error, asize = size;
2463
2464         VOP_ATTR_GET(vp, name, data, &asize, xflags, NULL, error);
2465         if (!error)
2466                 return asize;
2467         return -error;
2468 }
2469
2470 STATIC int
2471 attr_generic_remove(
2472         struct vnode *vp, char *name, int xflags)
2473 {
2474         int     error;
2475
2476         VOP_ATTR_REMOVE(vp, name, xflags, NULL, error);
2477         return -error;
2478 }
2479
2480 STATIC int
2481 attr_generic_listadd(
2482         attrnames_t             *prefix,
2483         attrnames_t             *namesp,
2484         void                    *data,
2485         size_t                  size,
2486         ssize_t                 *result)
2487 {
2488         char                    *p = data + *result;
2489
2490         *result += prefix->attr_namelen;
2491         *result += namesp->attr_namelen + 1;
2492         if (!size)
2493                 return 0;
2494         if (*result > size)
2495                 return -ERANGE;
2496         strcpy(p, prefix->attr_name);
2497         p += prefix->attr_namelen;
2498         strcpy(p, namesp->attr_name);
2499         p += namesp->attr_namelen + 1;
2500         return 0;
2501 }
2502
2503 STATIC int
2504 attr_system_list(
2505         struct vnode            *vp,
2506         void                    *data,
2507         size_t                  size,
2508         ssize_t                 *result)
2509 {
2510         attrnames_t             *namesp;
2511         int                     i, error = 0;
2512
2513         for (i = 0; i < ATTR_SYSCOUNT; i++) {
2514                 namesp = attr_system_names[i];
2515                 if (!namesp->attr_exists || !namesp->attr_exists(vp))
2516                         continue;
2517                 error = attr_generic_listadd(&attr_system, namesp,
2518                                                 data, size, result);
2519                 if (error)
2520                         break;
2521         }
2522         return error;
2523 }
2524
2525 int
2526 attr_generic_list(
2527         struct vnode *vp, void *data, size_t size, int xflags, ssize_t *result)
2528 {
2529         attrlist_cursor_kern_t  cursor = { 0 };
2530         int                     error;
2531
2532         VOP_ATTR_LIST(vp, data, size, xflags, &cursor, NULL, error);
2533         if (error > 0)
2534                 return -error;
2535         *result = -error;
2536         return attr_system_list(vp, data, size, result);
2537 }
2538
2539 attrnames_t *
2540 attr_lookup_namespace(
2541         char                    *name,
2542         struct attrnames        **names,
2543         int                     nnames)
2544 {
2545         int                     i;
2546
2547         for (i = 0; i < nnames; i++)
2548                 if (!strncmp(name, names[i]->attr_name, names[i]->attr_namelen))
2549                         return names[i];
2550         return NULL;
2551 }
2552
2553 /*
2554  * Some checks to prevent people abusing EAs to get over quota:
2555  * - Don't allow modifying user EAs on devices/symlinks;
2556  * - Don't allow modifying user EAs if sticky bit set;
2557  */
2558 STATIC int
2559 attr_user_capable(
2560         struct vnode    *vp,
2561         cred_t          *cred)
2562 {
2563         struct inode    *inode = LINVFS_GET_IP(vp);
2564
2565         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
2566                 return -EPERM;
2567         if (!S_ISREG(inode->i_mode) && !S_ISDIR(inode->i_mode) &&
2568             !capable(CAP_SYS_ADMIN))
2569                 return -EPERM;
2570         if (S_ISDIR(inode->i_mode) && (inode->i_mode & S_ISVTX) &&
2571             (current_fsuid(cred) != inode->i_uid) && !capable(CAP_FOWNER))
2572                 return -EPERM;
2573         return 0;
2574 }
2575
2576 STATIC int
2577 attr_trusted_capable(
2578         struct vnode    *vp,
2579         cred_t          *cred)
2580 {
2581         struct inode    *inode = LINVFS_GET_IP(vp);
2582
2583         if (IS_IMMUTABLE(inode) || IS_APPEND(inode))
2584                 return -EPERM;
2585         if (!capable(CAP_SYS_ADMIN))
2586                 return -EPERM;
2587         return 0;
2588 }
2589
2590 STATIC int
2591 attr_secure_capable(
2592         struct vnode    *vp,
2593         cred_t          *cred)
2594 {
2595         return -ENOSECURITY;
2596 }
2597
2598 STATIC int
2599 attr_system_set(
2600         struct vnode *vp, char *name, void *data, size_t size, int xflags)
2601 {
2602         attrnames_t     *namesp;
2603         int             error;
2604
2605         if (xflags & ATTR_CREATE)
2606                 return -EINVAL;
2607
2608         namesp = attr_lookup_namespace(name, attr_system_names, ATTR_SYSCOUNT);
2609         if (!namesp)
2610                 return -EOPNOTSUPP;
2611         error = namesp->attr_set(vp, name, data, size, xflags);
2612         if (!error)
2613                 error = vn_revalidate(vp);
2614         return error;
2615 }
2616
2617 STATIC int
2618 attr_system_get(
2619         struct vnode *vp, char *name, void *data, size_t size, int xflags)
2620 {
2621         attrnames_t     *namesp;
2622
2623         namesp = attr_lookup_namespace(name, attr_system_names, ATTR_SYSCOUNT);
2624         if (!namesp)
2625                 return -EOPNOTSUPP;
2626         return namesp->attr_get(vp, name, data, size, xflags);
2627 }
2628
2629 STATIC int
2630 attr_system_remove(
2631         struct vnode *vp, char *name, int xflags)
2632 {
2633         attrnames_t     *namesp;
2634
2635         namesp = attr_lookup_namespace(name, attr_system_names, ATTR_SYSCOUNT);
2636         if (!namesp)
2637                 return -EOPNOTSUPP;
2638         return namesp->attr_remove(vp, name, xflags);
2639 }
2640
2641 struct attrnames attr_system = {
2642         .attr_name      = "system.",
2643         .attr_namelen   = sizeof("system.") - 1,
2644         .attr_flag      = ATTR_SYSTEM,
2645         .attr_get       = attr_system_get,
2646         .attr_set       = attr_system_set,
2647         .attr_remove    = attr_system_remove,
2648         .attr_capable   = (attrcapable_t)fs_noerr,
2649 };
2650
2651 struct attrnames attr_trusted = {
2652         .attr_name      = "trusted.",
2653         .attr_namelen   = sizeof("trusted.") - 1,
2654         .attr_flag      = ATTR_ROOT,
2655         .attr_get       = attr_generic_get,
2656         .attr_set       = attr_generic_set,
2657         .attr_remove    = attr_generic_remove,
2658         .attr_capable   = attr_trusted_capable,
2659 };
2660
2661 struct attrnames attr_secure = {
2662         .attr_name      = "security.",
2663         .attr_namelen   = sizeof("security.") - 1,
2664         .attr_flag      = ATTR_SECURE,
2665         .attr_get       = attr_generic_get,
2666         .attr_set       = attr_generic_set,
2667         .attr_remove    = attr_generic_remove,
2668         .attr_capable   = attr_secure_capable,
2669 };
2670
2671 struct attrnames attr_user = {
2672         .attr_name      = "user.",
2673         .attr_namelen   = sizeof("user.") - 1,
2674         .attr_get       = attr_generic_get,
2675         .attr_set       = attr_generic_set,
2676         .attr_remove    = attr_generic_remove,
2677         .attr_capable   = attr_user_capable,
2678 };
2679
2680 struct attrnames *attr_namespaces[] =
2681         { &attr_system, &attr_trusted, &attr_secure, &attr_user };