/linux-2.6.33/kernel/linux-2.6.33/fs/xfs/xfs_attr.c

   1/*
   2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3 * All Rights Reserved.
   4 *
   5 * This program is free software; you can redistribute it and/or
   6 * modify it under the terms of the GNU General Public License as
   7 * published by the Free Software Foundation.
   8 *
   9 * This program is distributed in the hope that it would be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 * GNU General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License
  15 * along with this program; if not, write the Free Software Foundation,
  16 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17 */
  18
  19#include "xfs.h"
  20#include "xfs_fs.h"
  21#include "xfs_types.h"
  22#include "xfs_bit.h"
  23#include "xfs_log.h"
  24#include "xfs_inum.h"
  25#include "xfs_trans.h"
  26#include "xfs_sb.h"
  27#include "xfs_ag.h"
  28#include "xfs_dir2.h"
  29#include "xfs_dmapi.h"
  30#include "xfs_mount.h"
  31#include "xfs_da_btree.h"
  32#include "xfs_bmap_btree.h"
  33#include "xfs_alloc_btree.h"
  34#include "xfs_ialloc_btree.h"
  35#include "xfs_dir2_sf.h"
  36#include "xfs_attr_sf.h"
  37#include "xfs_dinode.h"
  38#include "xfs_inode.h"
  39#include "xfs_alloc.h"
  40#include "xfs_btree.h"
  41#include "xfs_inode_item.h"
  42#include "xfs_bmap.h"
  43#include "xfs_attr.h"
  44#include "xfs_attr_leaf.h"
  45#include "xfs_error.h"
  46#include "xfs_quota.h"
  47#include "xfs_trans_space.h"
  48#include "xfs_rw.h"
  49#include "xfs_vnodeops.h"
  50#include "xfs_trace.h"
  51
  52/*
  53 * xfs_attr.c
  54 *
  55 * Provide the external interfaces to manage attribute lists.
  56 */
  57
  58/*========================================================================
  59 * Function prototypes for the kernel.
  60 *========================================================================*/
  61
  62/*
  63 * Internal routines when attribute list fits inside the inode.
  64 */
  65STATIC int xfs_attr_shortform_addname(xfs_da_args_t *args);
  66
  67/*
  68 * Internal routines when attribute list is one block.
  69 */
  70STATIC int xfs_attr_leaf_get(xfs_da_args_t *args);
  71STATIC int xfs_attr_leaf_addname(xfs_da_args_t *args);
  72STATIC int xfs_attr_leaf_removename(xfs_da_args_t *args);
  73STATIC int xfs_attr_leaf_list(xfs_attr_list_context_t *context);
  74
  75/*
  76 * Internal routines when attribute list is more than one block.
  77 */
  78STATIC int xfs_attr_node_get(xfs_da_args_t *args);
  79STATIC int xfs_attr_node_addname(xfs_da_args_t *args);
  80STATIC int xfs_attr_node_removename(xfs_da_args_t *args);
  81STATIC int xfs_attr_node_list(xfs_attr_list_context_t *context);
  82STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
  83STATIC int xfs_attr_refillstate(xfs_da_state_t *state);
  84
  85/*
  86 * Routines to manipulate out-of-line attribute values.
  87 */
  88STATIC int xfs_attr_rmtval_set(xfs_da_args_t *args);
  89STATIC int xfs_attr_rmtval_remove(xfs_da_args_t *args);
  90
  91#define ATTR_RMTVALUE_MAPSIZE	1	/* # of map entries at once */
  92
  93STATIC int
  94xfs_attr_name_to_xname(
  95	struct xfs_name	*xname,
  96	const char	*aname)
  97{
  98	if (!aname)
  99		return EINVAL;
 100	xname->name = aname;
 101	xname->len = strlen(aname);
 102	if (xname->len >= MAXNAMELEN)
 103		return EFAULT;		/* match IRIX behaviour */
 104
 105	return 0;
 106}
 107
 108STATIC int
 109xfs_inode_hasattr(
 110	struct xfs_inode	*ip)
 111{
 112	if (!XFS_IFORK_Q(ip) ||
 113	    (ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 114	     ip->i_d.di_anextents == 0))
 115		return 0;
 116	return 1;
 117}
 118
 119/*========================================================================
 120 * Overall external interface routines.
 121 *========================================================================*/
 122
 123STATIC int
 124xfs_attr_get_int(
 125	struct xfs_inode	*ip,
 126	struct xfs_name		*name,
 127	char			*value,
 128	int			*valuelenp,
 129	int			flags)
 130{
 131	xfs_da_args_t   args;
 132	int             error;
 133
 134	if (!xfs_inode_hasattr(ip))
 135		return ENOATTR;
 136
 137	/*
 138	 * Fill in the arg structure for this request.
 139	 */
 140	memset((char *)&args, 0, sizeof(args));
 141	args.name = name->name;
 142	args.namelen = name->len;
 143	args.value = value;
 144	args.valuelen = *valuelenp;
 145	args.flags = flags;
 146	args.hashval = xfs_da_hashname(args.name, args.namelen);
 147	args.dp = ip;
 148	args.whichfork = XFS_ATTR_FORK;
 149
 150	/*
 151	 * Decide on what work routines to call based on the inode size.
 152	 */
 153	if (ip->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
 154		error = xfs_attr_shortform_getvalue(&args);
 155	} else if (xfs_bmap_one_block(ip, XFS_ATTR_FORK)) {
 156		error = xfs_attr_leaf_get(&args);
 157	} else {
 158		error = xfs_attr_node_get(&args);
 159	}
 160
 161	/*
 162	 * Return the number of bytes in the value to the caller.
 163	 */
 164	*valuelenp = args.valuelen;
 165
 166	if (error == EEXIST)
 167		error = 0;
 168	return(error);
 169}
 170
 171int
 172xfs_attr_get(
 173	xfs_inode_t	*ip,
 174	const char	*name,
 175	char		*value,
 176	int		*valuelenp,
 177	int		flags)
 178{
 179	int		error;
 180	struct xfs_name	xname;
 181
 182	XFS_STATS_INC(xs_attr_get);
 183
 184	if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 185		return(EIO);
 186
 187	error = xfs_attr_name_to_xname(&xname, name);
 188	if (error)
 189		return error;
 190
 191	xfs_ilock(ip, XFS_ILOCK_SHARED);
 192	error = xfs_attr_get_int(ip, &xname, value, valuelenp, flags);
 193	xfs_iunlock(ip, XFS_ILOCK_SHARED);
 194	return(error);
 195}
 196
 197/*
 198 * Calculate how many blocks we need for the new attribute,
 199 */
 200int
 201xfs_attr_calc_size(
 202	struct xfs_inode 	*ip,
 203	int			namelen,
 204	int			valuelen,
 205	int			*local)
 206{
 207	struct xfs_mount 	*mp = ip->i_mount;
 208	int			size;
 209	int			nblks;
 210
 211	/*
 212	 * Determine space new attribute will use, and if it would be
 213	 * "local" or "remote" (note: local != inline).
 214	 */
 215	size = xfs_attr_leaf_newentsize(namelen, valuelen,
 216					mp->m_sb.sb_blocksize, local);
 217
 218	nblks = XFS_DAENTER_SPACE_RES(mp, XFS_ATTR_FORK);
 219	if (*local) {
 220		if (size > (mp->m_sb.sb_blocksize >> 1)) {
 221			/* Double split possible */
 222			nblks *= 2;
 223		}
 224	} else {
 225		/*
 226		 * Out of line attribute, cannot double split, but
 227		 * make room for the attribute value itself.
 228		 */
 229		uint	dblocks = XFS_B_TO_FSB(mp, valuelen);
 230		nblks += dblocks;
 231		nblks += XFS_NEXTENTADD_SPACE_RES(mp, dblocks, XFS_ATTR_FORK);
 232	}
 233
 234	return nblks;
 235}
 236
 237STATIC int
 238xfs_attr_set_int(xfs_inode_t *dp, struct xfs_name *name,
 239		char *value, int valuelen, int flags)
 240{
 241	xfs_da_args_t	args;
 242	xfs_fsblock_t	firstblock;
 243	xfs_bmap_free_t flist;
 244	int		error, err2, committed;
 245	xfs_mount_t	*mp = dp->i_mount;
 246	int             rsvd = (flags & ATTR_ROOT) != 0;
 247	int		local;
 248
 249	/*
 250	 * Attach the dquots to the inode.
 251	 */
 252	error = xfs_qm_dqattach(dp, 0);
 253	if (error)
 254		return error;
 255
 256	/*
 257	 * If the inode doesn't have an attribute fork, add one.
 258	 * (inode must not be locked when we call this routine)
 259	 */
 260	if (XFS_IFORK_Q(dp) == 0) {
 261		int sf_size = sizeof(xfs_attr_sf_hdr_t) +
 262			      XFS_ATTR_SF_ENTSIZE_BYNAME(name->len, valuelen);
 263
 264		if ((error = xfs_bmap_add_attrfork(dp, sf_size, rsvd)))
 265			return(error);
 266	}
 267
 268	/*
 269	 * Fill in the arg structure for this request.
 270	 */
 271	memset((char *)&args, 0, sizeof(args));
 272	args.name = name->name;
 273	args.namelen = name->len;
 274	args.value = value;
 275	args.valuelen = valuelen;
 276	args.flags = flags;
 277	args.hashval = xfs_da_hashname(args.name, args.namelen);
 278	args.dp = dp;
 279	args.firstblock = &firstblock;
 280	args.flist = &flist;
 281	args.whichfork = XFS_ATTR_FORK;
 282	args.op_flags = XFS_DA_OP_ADDNAME | XFS_DA_OP_OKNOENT;
 283
 284	/* Size is now blocks for attribute data */
 285	args.total = xfs_attr_calc_size(dp, name->len, valuelen, &local);
 286
 287	/*
 288	 * Start our first transaction of the day.
 289	 *
 290	 * All future transactions during this code must be "chained" off
 291	 * this one via the trans_dup() call.  All transactions will contain
 292	 * the inode, and the inode will always be marked with trans_ihold().
 293	 * Since the inode will be locked in all transactions, we must log
 294	 * the inode in every transaction to let it float upward through
 295	 * the log.
 296	 */
 297	args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_SET);
 298
 299	/*
 300	 * Root fork attributes can use reserved data blocks for this
 301	 * operation if necessary
 302	 */
 303
 304	if (rsvd)
 305		args.trans->t_flags |= XFS_TRANS_RESERVE;
 306
 307	if ((error = xfs_trans_reserve(args.trans, args.total,
 308			XFS_ATTRSET_LOG_RES(mp, args.total), 0,
 309			XFS_TRANS_PERM_LOG_RES, XFS_ATTRSET_LOG_COUNT))) {
 310		xfs_trans_cancel(args.trans, 0);
 311		return(error);
 312	}
 313	xfs_ilock(dp, XFS_ILOCK_EXCL);
 314
 315	error = xfs_trans_reserve_quota_nblks(args.trans, dp, args.total, 0,
 316				rsvd ? XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
 317				       XFS_QMOPT_RES_REGBLKS);
 318	if (error) {
 319		xfs_iunlock(dp, XFS_ILOCK_EXCL);
 320		xfs_trans_cancel(args.trans, XFS_TRANS_RELEASE_LOG_RES);
 321		return (error);
 322	}
 323
 324	xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
 325	xfs_trans_ihold(args.trans, dp);
 326
 327	/*
 328	 * If the attribute list is non-existent or a shortform list,
 329	 * upgrade it to a single-leaf-block attribute list.
 330	 */
 331	if ((dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) ||
 332	    ((dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS) &&
 333	     (dp->i_d.di_anextents == 0))) {
 334
 335		/*
 336		 * Build initial attribute list (if required).
 337		 */
 338		if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS)
 339			xfs_attr_shortform_create(&args);
 340
 341		/*
 342		 * Try to add the attr to the attribute list in
 343		 * the inode.
 344		 */
 345		error = xfs_attr_shortform_addname(&args);
 346		if (error != ENOSPC) {
 347			/*
 348			 * Commit the shortform mods, and we're done.
 349			 * NOTE: this is also the error path (EEXIST, etc).
 350			 */
 351			ASSERT(args.trans != NULL);
 352
 353			/*
 354			 * If this is a synchronous mount, make sure that
 355			 * the transaction goes to disk before returning
 356			 * to the user.
 357			 */
 358			if (mp->m_flags & XFS_MOUNT_WSYNC) {
 359				xfs_trans_set_sync(args.trans);
 360			}
 361			err2 = xfs_trans_commit(args.trans,
 362						 XFS_TRANS_RELEASE_LOG_RES);
 363			xfs_iunlock(dp, XFS_ILOCK_EXCL);
 364
 365			/*
 366			 * Hit the inode change time.
 367			 */
 368			if (!error && (flags & ATTR_KERNOTIME) == 0) {
 369				xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
 370			}
 371			return(error == 0 ? err2 : error);
 372		}
 373
 374		/*
 375		 * It won't fit in the shortform, transform to a leaf block.
 376		 * GROT: another possible req'mt for a double-split btree op.
 377		 */
 378		xfs_bmap_init(args.flist, args.firstblock);
 379		error = xfs_attr_shortform_to_leaf(&args);
 380		if (!error) {
 381			error = xfs_bmap_finish(&args.trans, args.flist,
 382						&committed);
 383		}
 384		if (error) {
 385			ASSERT(committed);
 386			args.trans = NULL;
 387			xfs_bmap_cancel(&flist);
 388			goto out;
 389		}
 390
 391		/*
 392		 * bmap_finish() may have committed the last trans and started
 393		 * a new one.  We need the inode to be in all transactions.
 394		 */
 395		if (committed) {
 396			xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
 397			xfs_trans_ihold(args.trans, dp);
 398		}
 399
 400		/*
 401		 * Commit the leaf transformation.  We'll need another (linked)
 402		 * transaction to add the new attribute to the leaf.
 403		 */
 404
 405		error = xfs_trans_roll(&args.trans, dp);
 406		if (error)
 407			goto out;
 408
 409	}
 410
 411	if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
 412		error = xfs_attr_leaf_addname(&args);
 413	} else {
 414		error = xfs_attr_node_addname(&args);
 415	}
 416	if (error) {
 417		goto out;
 418	}
 419
 420	/*
 421	 * If this is a synchronous mount, make sure that the
 422	 * transaction goes to disk before returning to the user.
 423	 */
 424	if (mp->m_flags & XFS_MOUNT_WSYNC) {
 425		xfs_trans_set_sync(args.trans);
 426	}
 427
 428	/*
 429	 * Commit the last in the sequence of transactions.
 430	 */
 431	xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
 432	error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES);
 433	xfs_iunlock(dp, XFS_ILOCK_EXCL);
 434
 435	/*
 436	 * Hit the inode change time.
 437	 */
 438	if (!error && (flags & ATTR_KERNOTIME) == 0) {
 439		xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
 440	}
 441
 442	return(error);
 443
 444out:
 445	if (args.trans)
 446		xfs_trans_cancel(args.trans,
 447			XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
 448	xfs_iunlock(dp, XFS_ILOCK_EXCL);
 449	return(error);
 450}
 451
 452int
 453xfs_attr_set(
 454	xfs_inode_t	*dp,
 455	const char	*name,
 456	char		*value,
 457	int		valuelen,
 458	int		flags)
 459{
 460	int             error;
 461	struct xfs_name	xname;
 462
 463	XFS_STATS_INC(xs_attr_set);
 464
 465	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
 466		return (EIO);
 467
 468	error = xfs_attr_name_to_xname(&xname, name);
 469	if (error)
 470		return error;
 471
 472	return xfs_attr_set_int(dp, &xname, value, valuelen, flags);
 473}
 474
 475/*
 476 * Generic handler routine to remove a name from an attribute list.
 477 * Transitions attribute list from Btree to shortform as necessary.
 478 */
 479STATIC int
 480xfs_attr_remove_int(xfs_inode_t *dp, struct xfs_name *name, int flags)
 481{
 482	xfs_da_args_t	args;
 483	xfs_fsblock_t	firstblock;
 484	xfs_bmap_free_t	flist;
 485	int		error;
 486	xfs_mount_t	*mp = dp->i_mount;
 487
 488	/*
 489	 * Fill in the arg structure for this request.
 490	 */
 491	memset((char *)&args, 0, sizeof(args));
 492	args.name = name->name;
 493	args.namelen = name->len;
 494	args.flags = flags;
 495	args.hashval = xfs_da_hashname(args.name, args.namelen);
 496	args.dp = dp;
 497	args.firstblock = &firstblock;
 498	args.flist = &flist;
 499	args.total = 0;
 500	args.whichfork = XFS_ATTR_FORK;
 501
 502	/*
 503	 * Attach the dquots to the inode.
 504	 */
 505	error = xfs_qm_dqattach(dp, 0);
 506	if (error)
 507		return error;
 508
 509	/*
 510	 * Start our first transaction of the day.
 511	 *
 512	 * All future transactions during this code must be "chained" off
 513	 * this one via the trans_dup() call.  All transactions will contain
 514	 * the inode, and the inode will always be marked with trans_ihold().
 515	 * Since the inode will be locked in all transactions, we must log
 516	 * the inode in every transaction to let it float upward through
 517	 * the log.
 518	 */
 519	args.trans = xfs_trans_alloc(mp, XFS_TRANS_ATTR_RM);
 520
 521	/*
 522	 * Root fork attributes can use reserved data blocks for this
 523	 * operation if necessary
 524	 */
 525
 526	if (flags & ATTR_ROOT)
 527		args.trans->t_flags |= XFS_TRANS_RESERVE;
 528
 529	if ((error = xfs_trans_reserve(args.trans,
 530				      XFS_ATTRRM_SPACE_RES(mp),
 531				      XFS_ATTRRM_LOG_RES(mp),
 532				      0, XFS_TRANS_PERM_LOG_RES,
 533				      XFS_ATTRRM_LOG_COUNT))) {
 534		xfs_trans_cancel(args.trans, 0);
 535		return(error);
 536	}
 537
 538	xfs_ilock(dp, XFS_ILOCK_EXCL);
 539	/*
 540	 * No need to make quota reservations here. We expect to release some
 541	 * blocks not allocate in the common case.
 542	 */
 543	xfs_trans_ijoin(args.trans, dp, XFS_ILOCK_EXCL);
 544	xfs_trans_ihold(args.trans, dp);
 545
 546	/*
 547	 * Decide on what work routines to call based on the inode size.
 548	 */
 549	if (!xfs_inode_hasattr(dp)) {
 550		error = XFS_ERROR(ENOATTR);
 551		goto out;
 552	}
 553	if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
 554		ASSERT(dp->i_afp->if_flags & XFS_IFINLINE);
 555		error = xfs_attr_shortform_remove(&args);
 556		if (error) {
 557			goto out;
 558		}
 559	} else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
 560		error = xfs_attr_leaf_removename(&args);
 561	} else {
 562		error = xfs_attr_node_removename(&args);
 563	}
 564	if (error) {
 565		goto out;
 566	}
 567
 568	/*
 569	 * If this is a synchronous mount, make sure that the
 570	 * transaction goes to disk before returning to the user.
 571	 */
 572	if (mp->m_flags & XFS_MOUNT_WSYNC) {
 573		xfs_trans_set_sync(args.trans);
 574	}
 575
 576	/*
 577	 * Commit the last in the sequence of transactions.
 578	 */
 579	xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
 580	error = xfs_trans_commit(args.trans, XFS_TRANS_RELEASE_LOG_RES);
 581	xfs_iunlock(dp, XFS_ILOCK_EXCL);
 582
 583	/*
 584	 * Hit the inode change time.
 585	 */
 586	if (!error && (flags & ATTR_KERNOTIME) == 0) {
 587		xfs_ichgtime(dp, XFS_ICHGTIME_CHG);
 588	}
 589
 590	return(error);
 591
 592out:
 593	if (args.trans)
 594		xfs_trans_cancel(args.trans,
 595			XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
 596	xfs_iunlock(dp, XFS_ILOCK_EXCL);
 597	return(error);
 598}
 599
 600int
 601xfs_attr_remove(
 602	xfs_inode_t	*dp,
 603	const char	*name,
 604	int		flags)
 605{
 606	int		error;
 607	struct xfs_name	xname;
 608
 609	XFS_STATS_INC(xs_attr_remove);
 610
 611	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
 612		return (EIO);
 613
 614	error = xfs_attr_name_to_xname(&xname, name);
 615	if (error)
 616		return error;
 617
 618	xfs_ilock(dp, XFS_ILOCK_SHARED);
 619	if (!xfs_inode_hasattr(dp)) {
 620		xfs_iunlock(dp, XFS_ILOCK_SHARED);
 621		return XFS_ERROR(ENOATTR);
 622	}
 623	xfs_iunlock(dp, XFS_ILOCK_SHARED);
 624
 625	return xfs_attr_remove_int(dp, &xname, flags);
 626}
 627
 628int
 629xfs_attr_list_int(xfs_attr_list_context_t *context)
 630{
 631	int error;
 632	xfs_inode_t *dp = context->dp;
 633
 634	XFS_STATS_INC(xs_attr_list);
 635
 636	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
 637		return EIO;
 638
 639	xfs_ilock(dp, XFS_ILOCK_SHARED);
 640
 641	/*
 642	 * Decide on what work routines to call based on the inode size.
 643	 */
 644	if (!xfs_inode_hasattr(dp)) {
 645		error = 0;
 646	} else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
 647		error = xfs_attr_shortform_list(context);
 648	} else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
 649		error = xfs_attr_leaf_list(context);
 650	} else {
 651		error = xfs_attr_node_list(context);
 652	}
 653
 654	xfs_iunlock(dp, XFS_ILOCK_SHARED);
 655
 656	return error;
 657}
 658
 659#define	ATTR_ENTBASESIZE		/* minimum bytes used by an attr */ \
 660	(((struct attrlist_ent *) 0)->a_name - (char *) 0)
 661#define	ATTR_ENTSIZE(namelen)		/* actual bytes used by an attr */ \
 662	((ATTR_ENTBASESIZE + (namelen) + 1 + sizeof(u_int32_t)-1) \
 663	 & ~(sizeof(u_int32_t)-1))
 664
 665/*
 666 * Format an attribute and copy it out to the user's buffer.
 667 * Take care to check values and protect against them changing later,
 668 * we may be reading them directly out of a user buffer.
 669 */
 670/*ARGSUSED*/
 671STATIC int
 672xfs_attr_put_listent(xfs_attr_list_context_t *context, int flags,
 673		     char *name, int namelen,
 674		     int valuelen, char *value)
 675{
 676	struct attrlist *alist = (struct attrlist *)context->alist;
 677	attrlist_ent_t *aep;
 678	int arraytop;
 679
 680	ASSERT(!(context->flags & ATTR_KERNOVAL));
 681	ASSERT(context->count >= 0);
 682	ASSERT(context->count < (ATTR_MAX_VALUELEN/8));
 683	ASSERT(context->firstu >= sizeof(*alist));
 684	ASSERT(context->firstu <= context->bufsize);
 685
 686	/*
 687	 * Only list entries in the right namespace.
 688	 */
 689	if (((context->flags & ATTR_SECURE) == 0) !=
 690	    ((flags & XFS_ATTR_SECURE) == 0))
 691		return 0;
 692	if (((context->flags & ATTR_ROOT) == 0) !=
 693	    ((flags & XFS_ATTR_ROOT) == 0))
 694		return 0;
 695
 696	arraytop = sizeof(*alist) +
 697			context->count * sizeof(alist->al_offset[0]);
 698	context->firstu -= ATTR_ENTSIZE(namelen);
 699	if (context->firstu < arraytop) {
 700		trace_xfs_attr_list_full(context);
 701		alist->al_more = 1;
 702		context->seen_enough = 1;
 703		return 1;
 704	}
 705
 706	aep = (attrlist_ent_t *)&context->alist[context->firstu];
 707	aep->a_valuelen = valuelen;
 708	memcpy(aep->a_name, name, namelen);
 709	aep->a_name[namelen] = 0;
 710	alist->al_offset[context->count++] = context->firstu;
 711	alist->al_count = context->count;
 712	trace_xfs_attr_list_add(context);
 713	return 0;
 714}
 715
 716/*
 717 * Generate a list of extended attribute names and optionally
 718 * also value lengths.  Positive return value follows the XFS
 719 * convention of being an error, zero or negative return code
 720 * is the length of the buffer returned (negated), indicating
 721 * success.
 722 */
 723int
 724xfs_attr_list(
 725	xfs_inode_t	*dp,
 726	char		*buffer,
 727	int		bufsize,
 728	int		flags,
 729	attrlist_cursor_kern_t *cursor)
 730{
 731	xfs_attr_list_context_t context;
 732	struct attrlist *alist;
 733	int error;
 734
 735	/*
 736	 * Validate the cursor.
 737	 */
 738	if (cursor->pad1 || cursor->pad2)
 739		return(XFS_ERROR(EINVAL));
 740	if ((cursor->initted == 0) &&
 741	    (cursor->hashval || cursor->blkno || cursor->offset))
 742		return XFS_ERROR(EINVAL);
 743
 744	/*
 745	 * Check for a properly aligned buffer.
 746	 */
 747	if (((long)buffer) & (sizeof(int)-1))
 748		return XFS_ERROR(EFAULT);
 749	if (flags & ATTR_KERNOVAL)
 750		bufsize = 0;
 751
 752	/*
 753	 * Initialize the output buffer.
 754	 */
 755	memset(&context, 0, sizeof(context));
 756	context.dp = dp;
 757	context.cursor = cursor;
 758	context.resynch = 1;
 759	context.flags = flags;
 760	context.alist = buffer;
 761	context.bufsize = (bufsize & ~(sizeof(int)-1));  /* align */
 762	context.firstu = context.bufsize;
 763	context.put_listent = xfs_attr_put_listent;
 764
 765	alist = (struct attrlist *)context.alist;
 766	alist->al_count = 0;
 767	alist->al_more = 0;
 768	alist->al_offset[0] = context.bufsize;
 769
 770	error = xfs_attr_list_int(&context);
 771	ASSERT(error >= 0);
 772	return error;
 773}
 774
 775int								/* error */
 776xfs_attr_inactive(xfs_inode_t *dp)
 777{
 778	xfs_trans_t *trans;
 779	xfs_mount_t *mp;
 780	int error;
 781
 782	mp = dp->i_mount;
 783	ASSERT(! XFS_NOT_DQATTACHED(mp, dp));
 784
 785	xfs_ilock(dp, XFS_ILOCK_SHARED);
 786	if (!xfs_inode_hasattr(dp) ||
 787	    dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
 788		xfs_iunlock(dp, XFS_ILOCK_SHARED);
 789		return 0;
 790	}
 791	xfs_iunlock(dp, XFS_ILOCK_SHARED);
 792
 793	/*
 794	 * Start our first transaction of the day.
 795	 *
 796	 * All future transactions during this code must be "chained" off
 797	 * this one via the trans_dup() call.  All transactions will contain
 798	 * the inode, and the inode will always be marked with trans_ihold().
 799	 * Since the inode will be locked in all transactions, we must log
 800	 * the inode in every transaction to let it float upward through
 801	 * the log.
 802	 */
 803	trans = xfs_trans_alloc(mp, XFS_TRANS_ATTRINVAL);
 804	if ((error = xfs_trans_reserve(trans, 0, XFS_ATTRINVAL_LOG_RES(mp), 0,
 805				      XFS_TRANS_PERM_LOG_RES,
 806				      XFS_ATTRINVAL_LOG_COUNT))) {
 807		xfs_trans_cancel(trans, 0);
 808		return(error);
 809	}
 810	xfs_ilock(dp, XFS_ILOCK_EXCL);
 811
 812	/*
 813	 * No need to make quota reservations here. We expect to release some
 814	 * blocks, not allocate, in the common case.
 815	 */
 816	xfs_trans_ijoin(trans, dp, XFS_ILOCK_EXCL);
 817	xfs_trans_ihold(trans, dp);
 818
 819	/*
 820	 * Decide on what work routines to call based on the inode size.
 821	 */
 822	if (!xfs_inode_hasattr(dp) ||
 823	    dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
 824		error = 0;
 825		goto out;
 826	}
 827	error = xfs_attr_root_inactive(&trans, dp);
 828	if (error)
 829		goto out;
 830	/*
 831	 * signal synchronous inactive transactions unless this
 832	 * is a synchronous mount filesystem in which case we
 833	 * know that we're here because we've been called out of
 834	 * xfs_inactive which means that the last reference is gone
 835	 * and the unlink transaction has already hit the disk so
 836	 * async inactive transactions are safe.
 837	 */
 838	if ((error = xfs_itruncate_finish(&trans, dp, 0LL, XFS_ATTR_FORK,
 839				(!(mp->m_flags & XFS_MOUNT_WSYNC)
 840				 ? 1 : 0))))
 841		goto out;
 842
 843	/*
 844	 * Commit the last in the sequence of transactions.
 845	 */
 846	xfs_trans_log_inode(trans, dp, XFS_ILOG_CORE);
 847	error = xfs_trans_commit(trans, XFS_TRANS_RELEASE_LOG_RES);
 848	xfs_iunlock(dp, XFS_ILOCK_EXCL);
 849
 850	return(error);
 851
 852out:
 853	xfs_trans_cancel(trans, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
 854	xfs_iunlock(dp, XFS_ILOCK_EXCL);
 855	return(error);
 856}
 857
 858
 859
 860/*========================================================================
 861 * External routines when attribute list is inside the inode
 862 *========================================================================*/
 863
 864/*
 865 * Add a name to the shortform attribute list structure
 866 * This is the external routine.
 867 */
 868STATIC int
 869xfs_attr_shortform_addname(xfs_da_args_t *args)
 870{
 871	int newsize, forkoff, retval;
 872
 873	retval = xfs_attr_shortform_lookup(args);
 874	if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
 875		return(retval);
 876	} else if (retval == EEXIST) {
 877		if (args->flags & ATTR_CREATE)
 878			return(retval);
 879		retval = xfs_attr_shortform_remove(args);
 880		ASSERT(retval == 0);
 881	}
 882
 883	if (args->namelen >= XFS_ATTR_SF_ENTSIZE_MAX ||
 884	    args->valuelen >= XFS_ATTR_SF_ENTSIZE_MAX)
 885		return(XFS_ERROR(ENOSPC));
 886
 887	newsize = XFS_ATTR_SF_TOTSIZE(args->dp);
 888	newsize += XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
 889
 890	forkoff = xfs_attr_shortform_bytesfit(args->dp, newsize);
 891	if (!forkoff)
 892		return(XFS_ERROR(ENOSPC));
 893
 894	xfs_attr_shortform_add(args, forkoff);
 895	return(0);
 896}
 897
 898
 899/*========================================================================
 900 * External routines when attribute list is one block
 901 *========================================================================*/
 902
 903/*
 904 * Add a name to the leaf attribute list structure
 905 *
 906 * This leaf block cannot have a "remote" value, we only call this routine
 907 * if bmap_one_block() says there is only one block (ie: no remote blks).
 908 */
 909STATIC int
 910xfs_attr_leaf_addname(xfs_da_args_t *args)
 911{
 912	xfs_inode_t *dp;
 913	xfs_dabuf_t *bp;
 914	int retval, error, committed, forkoff;
 915
 916	/*
 917	 * Read the (only) block in the attribute list in.
 918	 */
 919	dp = args->dp;
 920	args->blkno = 0;
 921	error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
 922					     XFS_ATTR_FORK);
 923	if (error)
 924		return(error);
 925	ASSERT(bp != NULL);
 926
 927	/*
 928	 * Look up the given attribute in the leaf block.  Figure out if
 929	 * the given flags produce an error or call for an atomic rename.
 930	 */
 931	retval = xfs_attr_leaf_lookup_int(bp, args);
 932	if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
 933		xfs_da_brelse(args->trans, bp);
 934		return(retval);
 935	} else if (retval == EEXIST) {
 936		if (args->flags & ATTR_CREATE) {	/* pure create op */
 937			xfs_da_brelse(args->trans, bp);
 938			return(retval);
 939		}
 940		args->op_flags |= XFS_DA_OP_RENAME;	/* an atomic rename */
 941		args->blkno2 = args->blkno;		/* set 2nd entry info*/
 942		args->index2 = args->index;
 943		args->rmtblkno2 = args->rmtblkno;
 944		args->rmtblkcnt2 = args->rmtblkcnt;
 945	}
 946
 947	/*
 948	 * Add the attribute to the leaf block, transitioning to a Btree
 949	 * if required.
 950	 */
 951	retval = xfs_attr_leaf_add(bp, args);
 952	xfs_da_buf_done(bp);
 953	if (retval == ENOSPC) {
 954		/*
 955		 * Promote the attribute list to the Btree format, then
 956		 * Commit that transaction so that the node_addname() call
 957		 * can manage its own transactions.
 958		 */
 959		xfs_bmap_init(args->flist, args->firstblock);
 960		error = xfs_attr_leaf_to_node(args);
 961		if (!error) {
 962			error = xfs_bmap_finish(&args->trans, args->flist,
 963						&committed);
 964		}
 965		if (error) {
 966			ASSERT(committed);
 967			args->trans = NULL;
 968			xfs_bmap_cancel(args->flist);
 969			return(error);
 970		}
 971
 972		/*
 973		 * bmap_finish() may have committed the last trans and started
 974		 * a new one.  We need the inode to be in all transactions.
 975		 */
 976		if (committed) {
 977			xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
 978			xfs_trans_ihold(args->trans, dp);
 979		}
 980
 981		/*
 982		 * Commit the current trans (including the inode) and start
 983		 * a new one.
 984		 */
 985		error = xfs_trans_roll(&args->trans, dp);
 986		if (error)
 987			return (error);
 988
 989		/*
 990		 * Fob the whole rest of the problem off on the Btree code.
 991		 */
 992		error = xfs_attr_node_addname(args);
 993		return(error);
 994	}
 995
 996	/*
 997	 * Commit the transaction that added the attr name so that
 998	 * later routines can manage their own transactions.
 999	 */
1000	error = xfs_trans_roll(&args->trans, dp);
1001	if (error)
1002		return (error);
1003
1004	/*
1005	 * If there was an out-of-line value, allocate the blocks we
1006	 * identified for its storage and copy the value.  This is done
1007	 * after we create the attribute so that we don't overflow the
1008	 * maximum size of a transaction and/or hit a deadlock.
1009	 */
1010	if (args->rmtblkno > 0) {
1011		error = xfs_attr_rmtval_set(args);
1012		if (error)
1013			return(error);
1014	}
1015
1016	/*
1017	 * If this is an atomic rename operation, we must "flip" the
1018	 * incomplete flags on the "new" and "old" attribute/value pairs
1019	 * so that one disappears and one appears atomically.  Then we
1020	 * must remove the "old" attribute/value pair.
1021	 */
1022	if (args->op_flags & XFS_DA_OP_RENAME) {
1023		/*
1024		 * In a separate transaction, set the incomplete flag on the
1025		 * "old" attr and clear the incomplete flag on the "new" attr.
1026		 */
1027		error = xfs_attr_leaf_flipflags(args);
1028		if (error)
1029			return(error);
1030
1031		/*
1032		 * Dismantle the "old" attribute/value pair by removing
1033		 * a "remote" value (if it exists).
1034		 */
1035		args->index = args->index2;
1036		args->blkno = args->blkno2;
1037		args->rmtblkno = args->rmtblkno2;
1038		args->rmtblkcnt = args->rmtblkcnt2;
1039		if (args->rmtblkno) {
1040			error = xfs_attr_rmtval_remove(args);
1041			if (error)
1042				return(error);
1043		}
1044
1045		/*
1046		 * Read in the block containing the "old" attr, then
1047		 * remove the "old" attr from that block (neat, huh!)
1048		 */
1049		error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1,
1050						     &bp, XFS_ATTR_FORK);
1051		if (error)
1052			return(error);
1053		ASSERT(bp != NULL);
1054		(void)xfs_attr_leaf_remove(bp, args);
1055
1056		/*
1057		 * If the result is small enough, shrink it all into the inode.
1058		 */
1059		if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1060			xfs_bmap_init(args->flist, args->firstblock);
1061			error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1062			/* bp is gone due to xfs_da_shrink_inode */
1063			if (!error) {
1064				error = xfs_bmap_finish(&args->trans,
1065							args->flist,
1066							&committed);
1067			}
1068			if (error) {
1069				ASSERT(committed);
1070				args->trans = NULL;
1071				xfs_bmap_cancel(args->flist);
1072				return(error);
1073			}
1074
1075			/*
1076			 * bmap_finish() may have committed the last trans
1077			 * and started a new one.  We need the inode to be
1078			 * in all transactions.
1079			 */
1080			if (committed) {
1081				xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1082				xfs_trans_ihold(args->trans, dp);
1083			}
1084		} else
1085			xfs_da_buf_done(bp);
1086
1087		/*
1088		 * Commit the remove and start the next trans in series.
1089		 */
1090		error = xfs_trans_roll(&args->trans, dp);
1091
1092	} else if (args->rmtblkno > 0) {
1093		/*
1094		 * Added a "remote" value, just clear the incomplete flag.
1095		 */
1096		error = xfs_attr_leaf_clearflag(args);
1097	}
1098	return(error);
1099}
1100
1101/*
1102 * Remove a name from the leaf attribute list structure
1103 *
1104 * This leaf block cannot have a "remote" value, we only call this routine
1105 * if bmap_one_block() says there is only one block (ie: no remote blks).
1106 */
1107STATIC int
1108xfs_attr_leaf_removename(xfs_da_args_t *args)
1109{
1110	xfs_inode_t *dp;
1111	xfs_dabuf_t *bp;
1112	int error, committed, forkoff;
1113
1114	/*
1115	 * Remove the attribute.
1116	 */
1117	dp = args->dp;
1118	args->blkno = 0;
1119	error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1120					     XFS_ATTR_FORK);
1121	if (error) {
1122		return(error);
1123	}
1124
1125	ASSERT(bp != NULL);
1126	error = xfs_attr_leaf_lookup_int(bp, args);
1127	if (error == ENOATTR) {
1128		xfs_da_brelse(args->trans, bp);
1129		return(error);
1130	}
1131
1132	(void)xfs_attr_leaf_remove(bp, args);
1133
1134	/*
1135	 * If the result is small enough, shrink it all into the inode.
1136	 */
1137	if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1138		xfs_bmap_init(args->flist, args->firstblock);
1139		error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1140		/* bp is gone due to xfs_da_shrink_inode */
1141		if (!error) {
1142			error = xfs_bmap_finish(&args->trans, args->flist,
1143						&committed);
1144		}
1145		if (error) {
1146			ASSERT(committed);
1147			args->trans = NULL;
1148			xfs_bmap_cancel(args->flist);
1149			return(error);
1150		}
1151
1152		/*
1153		 * bmap_finish() may have committed the last trans and started
1154		 * a new one.  We need the inode to be in all transactions.
1155		 */
1156		if (committed) {
1157			xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1158			xfs_trans_ihold(args->trans, dp);
1159		}
1160	} else
1161		xfs_da_buf_done(bp);
1162	return(0);
1163}
1164
1165/*
1166 * Look up a name in a leaf attribute list structure.
1167 *
1168 * This leaf block cannot have a "remote" value, we only call this routine
1169 * if bmap_one_block() says there is only one block (ie: no remote blks).
1170 */
1171STATIC int
1172xfs_attr_leaf_get(xfs_da_args_t *args)
1173{
1174	xfs_dabuf_t *bp;
1175	int error;
1176
1177	args->blkno = 0;
1178	error = xfs_da_read_buf(args->trans, args->dp, args->blkno, -1, &bp,
1179					     XFS_ATTR_FORK);
1180	if (error)
1181		return(error);
1182	ASSERT(bp != NULL);
1183
1184	error = xfs_attr_leaf_lookup_int(bp, args);
1185	if (error != EEXIST)  {
1186		xfs_da_brelse(args->trans, bp);
1187		return(error);
1188	}
1189	error = xfs_attr_leaf_getvalue(bp, args);
1190	xfs_da_brelse(args->trans, bp);
1191	if (!error && (args->rmtblkno > 0) && !(args->flags & ATTR_KERNOVAL)) {
1192		error = xfs_attr_rmtval_get(args);
1193	}
1194	return(error);
1195}
1196
1197/*
1198 * Copy out attribute entries for attr_list(), for leaf attribute lists.
1199 */
1200STATIC int
1201xfs_attr_leaf_list(xfs_attr_list_context_t *context)
1202{
1203	xfs_attr_leafblock_t *leaf;
1204	int error;
1205	xfs_dabuf_t *bp;
1206
1207	context->cursor->blkno = 0;
1208	error = xfs_da_read_buf(NULL, context->dp, 0, -1, &bp, XFS_ATTR_FORK);
1209	if (error)
1210		return XFS_ERROR(error);
1211	ASSERT(bp != NULL);
1212	leaf = bp->data;
1213	if (unlikely(be16_to_cpu(leaf->hdr.info.magic) != XFS_ATTR_LEAF_MAGIC)) {
1214		XFS_CORRUPTION_ERROR("xfs_attr_leaf_list", XFS_ERRLEVEL_LOW,
1215				     context->dp->i_mount, leaf);
1216		xfs_da_brelse(NULL, bp);
1217		return XFS_ERROR(EFSCORRUPTED);
1218	}
1219
1220	error = xfs_attr_leaf_list_int(bp, context);
1221	xfs_da_brelse(NULL, bp);
1222	return XFS_ERROR(error);
1223}
1224
1225
1226/*========================================================================
1227 * External routines when attribute list size > XFS_LBSIZE(mp).
1228 *========================================================================*/
1229
1230/*
1231 * Add a name to a Btree-format attribute list.
1232 *
1233 * This will involve walking down the Btree, and may involve splitting
1234 * leaf nodes and even splitting intermediate nodes up to and including
1235 * the root node (a special case of an intermediate node).
1236 *
1237 * "Remote" attribute values confuse the issue and atomic rename operations
1238 * add a whole extra layer of confusion on top of that.
1239 */
1240STATIC int
1241xfs_attr_node_addname(xfs_da_args_t *args)
1242{
1243	xfs_da_state_t *state;
1244	xfs_da_state_blk_t *blk;
1245	xfs_inode_t *dp;
1246	xfs_mount_t *mp;
1247	int committed, retval, error;
1248
1249	/*
1250	 * Fill in bucket of arguments/results/context to carry around.
1251	 */
1252	dp = args->dp;
1253	mp = dp->i_mount;
1254restart:
1255	state = xfs_da_state_alloc();
1256	state->args = args;
1257	state->mp = mp;
1258	state->blocksize = state->mp->m_sb.sb_blocksize;
1259	state->node_ents = state->mp->m_attr_node_ents;
1260
1261	/*
1262	 * Search to see if name already exists, and get back a pointer
1263	 * to where it should go.
1264	 */
1265	error = xfs_da_node_lookup_int(state, &retval);
1266	if (error)
1267		goto out;
1268	blk = &state->path.blk[ state->path.active-1 ];
1269	ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1270	if ((args->flags & ATTR_REPLACE) && (retval == ENOATTR)) {
1271		goto out;
1272	} else if (retval == EEXIST) {
1273		if (args->flags & ATTR_CREATE)
1274			goto out;
1275		args->op_flags |= XFS_DA_OP_RENAME;	/* atomic rename op */
1276		args->blkno2 = args->blkno;		/* set 2nd entry info*/
1277		args->index2 = args->index;
1278		args->rmtblkno2 = args->rmtblkno;
1279		args->rmtblkcnt2 = args->rmtblkcnt;
1280		args->rmtblkno = 0;
1281		args->rmtblkcnt = 0;
1282	}
1283
1284	retval = xfs_attr_leaf_add(blk->bp, state->args);
1285	if (retval == ENOSPC) {
1286		if (state->path.active == 1) {
1287			/*
1288			 * Its really a single leaf node, but it had
1289			 * out-of-line values so it looked like it *might*
1290			 * have been a b-tree.
1291			 */
1292			xfs_da_state_free(state);
1293			xfs_bmap_init(args->flist, args->firstblock);
1294			error = xfs_attr_leaf_to_node(args);
1295			if (!error) {
1296				error = xfs_bmap_finish(&args->trans,
1297							args->flist,
1298							&committed);
1299			}
1300			if (error) {
1301				ASSERT(committed);
1302				args->trans = NULL;
1303				xfs_bmap_cancel(args->flist);
1304				goto out;
1305			}
1306
1307			/*
1308			 * bmap_finish() may have committed the last trans
1309			 * and started a new one.  We need the inode to be
1310			 * in all transactions.
1311			 */
1312			if (committed) {
1313				xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1314				xfs_trans_ihold(args->trans, dp);
1315			}
1316
1317			/*
1318			 * Commit the node conversion and start the next
1319			 * trans in the chain.
1320			 */
1321			error = xfs_trans_roll(&args->trans, dp);
1322			if (error)
1323				goto out;
1324
1325			goto restart;
1326		}
1327
1328		/*
1329		 * Split as many Btree elements as required.
1330		 * This code tracks the new and old attr's location
1331		 * in the index/blkno/rmtblkno/rmtblkcnt fields and
1332		 * in the index2/blkno2/rmtblkno2/rmtblkcnt2 fields.
1333		 */
1334		xfs_bmap_init(args->flist, args->firstblock);
1335		error = xfs_da_split(state);
1336		if (!error) {
1337			error = xfs_bmap_finish(&args->trans, args->flist,
1338						&committed);
1339		}
1340		if (error) {
1341			ASSERT(committed);
1342			args->trans = NULL;
1343			xfs_bmap_cancel(args->flist);
1344			goto out;
1345		}
1346
1347		/*
1348		 * bmap_finish() may have committed the last trans and started
1349		 * a new one.  We need the inode to be in all transactions.
1350		 */
1351		if (committed) {
1352			xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1353			xfs_trans_ihold(args->trans, dp);
1354		}
1355	} else {
1356		/*
1357		 * Addition succeeded, update Btree hashvals.
1358		 */
1359		xfs_da_fixhashpath(state, &state->path);
1360	}
1361
1362	/*
1363	 * Kill the state structure, we're done with it and need to
1364	 * allow the buffers to come back later.
1365	 */
1366	xfs_da_state_free(state);
1367	state = NULL;
1368
1369	/*
1370	 * Commit the leaf addition or btree split and start the next
1371	 * trans in the chain.
1372	 */
1373	error = xfs_trans_roll(&args->trans, dp);
1374	if (error)
1375		goto out;
1376
1377	/*
1378	 * If there was an out-of-line value, allocate the blocks we
1379	 * identified for its storage and copy the value.  This is done
1380	 * after we create the attribute so that we don't overflow the
1381	 * maximum size of a transaction and/or hit a deadlock.
1382	 */
1383	if (args->rmtblkno > 0) {
1384		error = xfs_attr_rmtval_set(args);
1385		if (error)
1386			return(error);
1387	}
1388
1389	/*
1390	 * If this is an atomic rename operation, we must "flip" the
1391	 * incomplete flags on the "new" and "old" attribute/value pairs
1392	 * so that one disappears and one appears atomically.  Then we
1393	 * must remove the "old" attribute/value pair.
1394	 */
1395	if (args->op_flags & XFS_DA_OP_RENAME) {
1396		/*
1397		 * In a separate transaction, set the incomplete flag on the
1398		 * "old" attr and clear the incomplete flag on the "new" attr.
1399		 */
1400		error = xfs_attr_leaf_flipflags(args);
1401		if (error)
1402			goto out;
1403
1404		/*
1405		 * Dismantle the "old" attribute/value pair by removing
1406		 * a "remote" value (if it exists).
1407		 */
1408		args->index = args->index2;
1409		args->blkno = args->blkno2;
1410		args->rmtblkno = args->rmtblkno2;
1411		args->rmtblkcnt = args->rmtblkcnt2;
1412		if (args->rmtblkno) {
1413			error = xfs_attr_rmtval_remove(args);
1414			if (error)
1415				return(error);
1416		}
1417
1418		/*
1419		 * Re-find the "old" attribute entry after any split ops.
1420		 * The INCOMPLETE flag means that we will find the "old"
1421		 * attr, not the "new" one.
1422		 */
1423		args->flags |= XFS_ATTR_INCOMPLETE;
1424		state = xfs_da_state_alloc();
1425		state->args = args;
1426		state->mp = mp;
1427		state->blocksize = state->mp->m_sb.sb_blocksize;
1428		state->node_ents = state->mp->m_attr_node_ents;
1429		state->inleaf = 0;
1430		error = xfs_da_node_lookup_int(state, &retval);
1431		if (error)
1432			goto out;
1433
1434		/*
1435		 * Remove the name and update the hashvals in the tree.
1436		 */
1437		blk = &state->path.blk[ state->path.active-1 ];
1438		ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1439		error = xfs_attr_leaf_remove(blk->bp, args);
1440		xfs_da_fixhashpath(state, &state->path);
1441
1442		/*
1443		 * Check to see if the tree needs to be collapsed.
1444		 */
1445		if (retval && (state->path.active > 1)) {
1446			xfs_bmap_init(args->flist, args->firstblock);
1447			error = xfs_da_join(state);
1448			if (!error) {
1449				error = xfs_bmap_finish(&args->trans,
1450							args->flist,
1451							&committed);
1452			}
1453			if (error) {
1454				ASSERT(committed);
1455				args->trans = NULL;
1456				xfs_bmap_cancel(args->flist);
1457				goto out;
1458			}
1459
1460			/*
1461			 * bmap_finish() may have committed the last trans
1462			 * and started a new one.  We need the inode to be
1463			 * in all transactions.
1464			 */
1465			if (committed) {
1466				xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1467				xfs_trans_ihold(args->trans, dp);
1468			}
1469		}
1470
1471		/*
1472		 * Commit and start the next trans in the chain.
1473		 */
1474		error = xfs_trans_roll(&args->trans, dp);
1475		if (error)
1476			goto out;
1477
1478	} else if (args->rmtblkno > 0) {
1479		/*
1480		 * Added a "remote" value, just clear the incomplete flag.
1481		 */
1482		error = xfs_attr_leaf_clearflag(args);
1483		if (error)
1484			goto out;
1485	}
1486	retval = error = 0;
1487
1488out:
1489	if (state)
1490		xfs_da_state_free(state);
1491	if (error)
1492		return(error);
1493	return(retval);
1494}
1495
1496/*
1497 * Remove a name from a B-tree attribute list.
1498 *
1499 * This will involve walking down the Btree, and may involve joining
1500 * leaf nodes and even joining intermediate nodes up to and including
1501 * the root node (a special case of an intermediate node).
1502 */
1503STATIC int
1504xfs_attr_node_removename(xfs_da_args_t *args)
1505{
1506	xfs_da_state_t *state;
1507	xfs_da_state_blk_t *blk;
1508	xfs_inode_t *dp;
1509	xfs_dabuf_t *bp;
1510	int retval, error, committed, forkoff;
1511
1512	/*
1513	 * Tie a string around our finger to remind us where we are.
1514	 */
1515	dp = args->dp;
1516	state = xfs_da_state_alloc();
1517	state->args = args;
1518	state->mp = dp->i_mount;
1519	state->blocksize = state->mp->m_sb.sb_blocksize;
1520	state->node_ents = state->mp->m_attr_node_ents;
1521
1522	/*
1523	 * Search to see if name exists, and get back a pointer to it.
1524	 */
1525	error = xfs_da_node_lookup_int(state, &retval);
1526	if (error || (retval != EEXIST)) {
1527		if (error == 0)
1528			error = retval;
1529		goto out;
1530	}
1531
1532	/*
1533	 * If there is an out-of-line value, de-allocate the blocks.
1534	 * This is done before we remove the attribute so that we don't
1535	 * overflow the maximum size of a transaction and/or hit a deadlock.
1536	 */
1537	blk = &state->path.blk[ state->path.active-1 ];
1538	ASSERT(blk->bp != NULL);
1539	ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1540	if (args->rmtblkno > 0) {
1541		/*
1542		 * Fill in disk block numbers in the state structure
1543		 * so that we can get the buffers back after we commit
1544		 * several transactions in the following calls.
1545		 */
1546		error = xfs_attr_fillstate(state);
1547		if (error)
1548			goto out;
1549
1550		/*
1551		 * Mark the attribute as INCOMPLETE, then bunmapi() the
1552		 * remote value.
1553		 */
1554		error = xfs_attr_leaf_setflag(args);
1555		if (error)
1556			goto out;
1557		error = xfs_attr_rmtval_remove(args);
1558		if (error)
1559			goto out;
1560
1561		/*
1562		 * Refill the state structure with buffers, the prior calls
1563		 * released our buffers.
1564		 */
1565		error = xfs_attr_refillstate(state);
1566		if (error)
1567			goto out;
1568	}
1569
1570	/*
1571	 * Remove the name and update the hashvals in the tree.
1572	 */
1573	blk = &state->path.blk[ state->path.active-1 ];
1574	ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1575	retval = xfs_attr_leaf_remove(blk->bp, args);
1576	xfs_da_fixhashpath(state, &state->path);
1577
1578	/*
1579	 * Check to see if the tree needs to be collapsed.
1580	 */
1581	if (retval && (state->path.active > 1)) {
1582		xfs_bmap_init(args->flist, args->firstblock);
1583		error = xfs_da_join(state);
1584		if (!error) {
1585			error = xfs_bmap_finish(&args->trans, args->flist,
1586						&committed);
1587		}
1588		if (error) {
1589			ASSERT(committed);
1590			args->trans = NULL;
1591			xfs_bmap_cancel(args->flist);
1592			goto out;
1593		}
1594
1595		/*
1596		 * bmap_finish() may have committed the last trans and started
1597		 * a new one.  We need the inode to be in all transactions.
1598		 */
1599		if (committed) {
1600			xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1601			xfs_trans_ihold(args->trans, dp);
1602		}
1603
1604		/*
1605		 * Commit the Btree join operation and start a new trans.
1606		 */
1607		error = xfs_trans_roll(&args->trans, dp);
1608		if (error)
1609			goto out;
1610	}
1611
1612	/*
1613	 * If the result is small enough, push it all into the inode.
1614	 */
1615	if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
1616		/*
1617		 * Have to get rid of the copy of this dabuf in the state.
1618		 */
1619		ASSERT(state->path.active == 1);
1620		ASSERT(state->path.blk[0].bp);
1621		xfs_da_buf_done(state->path.blk[0].bp);
1622		state->path.blk[0].bp = NULL;
1623
1624		error = xfs_da_read_buf(args->trans, args->dp, 0, -1, &bp,
1625						     XFS_ATTR_FORK);
1626		if (error)
1627			goto out;
1628		ASSERT(be16_to_cpu(((xfs_attr_leafblock_t *)
1629				      bp->data)->hdr.info.magic)
1630						       == XFS_ATTR_LEAF_MAGIC);
1631
1632		if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1633			xfs_bmap_init(args->flist, args->firstblock);
1634			error = xfs_attr_leaf_to_shortform(bp, args, forkoff);
1635			/* bp is gone due to xfs_da_shrink_inode */
1636			if (!error) {
1637				error = xfs_bmap_finish(&args->trans,
1638							args->flist,
1639							&committed);
1640			}
1641			if (error) {
1642				ASSERT(committed);
1643				args->trans = NULL;
1644				xfs_bmap_cancel(args->flist);
1645				goto out;
1646			}
1647
1648			/*
1649			 * bmap_finish() may have committed the last trans
1650			 * and started a new one.  We need the inode to be
1651			 * in all transactions.
1652			 */
1653			if (committed) {
1654				xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
1655				xfs_trans_ihold(args->trans, dp);
1656			}
1657		} else
1658			xfs_da_brelse(args->trans, bp);
1659	}
1660	error = 0;
1661
1662out:
1663	xfs_da_state_free(state);
1664	return(error);
1665}
1666
1667/*
1668 * Fill in the disk block numbers in the state structure for the buffers
1669 * that are attached to the state structure.
1670 * This is done so that we can quickly reattach ourselves to those buffers
1671 * after some set of transaction commits have released these buffers.
1672 */
1673STATIC int
1674xfs_attr_fillstate(xfs_da_state_t *state)
1675{
1676	xfs_da_state_path_t *path;
1677	xfs_da_state_blk_t *blk;
1678	int level;
1679
1680	/*
1681	 * Roll down the "path" in the state structure, storing the on-disk
1682	 * block number for those buffers in the "path".
1683	 */
1684	path = &state->path;
1685	ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1686	for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1687		if (blk->bp) {
1688			blk->disk_blkno = xfs_da_blkno(blk->bp);
1689			xfs_da_buf_done(blk->bp);
1690			blk->bp = NULL;
1691		} else {
1692			blk->disk_blkno = 0;
1693		}
1694	}
1695
1696	/*
1697	 * Roll down the "altpath" in the state structure, storing the on-disk
1698	 * block number for those buffers in the "altpath".
1699	 */
1700	path = &state->altpath;
1701	ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1702	for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1703		if (blk->bp) {
1704			blk->disk_blkno = xfs_da_blkno(blk->bp);
1705			xfs_da_buf_done(blk->bp);
1706			blk->bp = NULL;
1707		} else {
1708			blk->disk_blkno = 0;
1709		}
1710	}
1711
1712	return(0);
1713}
1714
1715/*
1716 * Reattach the buffers to the state structure based on the disk block
1717 * numbers stored in the state structure.
1718 * This is done after some set of transaction commits have released those
1719 * buffers from our grip.
1720 */
1721STATIC int
1722xfs_attr_refillstate(xfs_da_state_t *state)
1723{
1724	xfs_da_state_path_t *path;
1725	xfs_da_state_blk_t *blk;
1726	int level, error;
1727
1728	/*
1729	 * Roll down the "path" in the state structure, storing the on-disk
1730	 * block number for those buffers in the "path".
1731	 */
1732	path = &state->path;
1733	ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1734	for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1735		if (blk->disk_blkno) {
1736			error = xfs_da_read_buf(state->args->trans,
1737						state->args->dp,
1738						blk->blkno, blk->disk_blkno,
1739						&blk->bp, XFS_ATTR_FORK);
1740			if (error)
1741				return(error);
1742		} else {
1743			blk->bp = NULL;
1744		}
1745	}
1746
1747	/*
1748	 * Roll down the "altpath" in the state structure, storing the on-disk
1749	 * block number for those buffers in the "altpath".
1750	 */
1751	path = &state->altpath;
1752	ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1753	for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1754		if (blk->disk_blkno) {
1755			error = xfs_da_read_buf(state->args->trans,
1756						state->args->dp,
1757						blk->blkno, blk->disk_blkno,
1758						&blk->bp, XFS_ATTR_FORK);
1759			if (error)
1760				return(error);
1761		} else {
1762			blk->bp = NULL;
1763		}
1764	}
1765
1766	return(0);
1767}
1768
1769/*
1770 * Look up a filename in a node attribute list.
1771 *
1772 * This routine gets called for any attribute fork that has more than one
1773 * block, ie: both true Btree attr lists and for single-leaf-blocks with
1774 * "remote" values taking up more blocks.
1775 */
1776STATIC int
1777xfs_attr_node_get(xfs_da_args_t *args)
1778{
1779	xfs_da_state_t *state;
1780	xfs_da_state_blk_t *blk;
1781	int error, retval;
1782	int i;
1783
1784	state = xfs_da_state_alloc();
1785	state->args = args;
1786	state->mp = args->dp->i_mount;
1787	state->blocksize = state->mp->m_sb.sb_blocksize;
1788	state->node_ents = state->mp->m_attr_node_ents;
1789
1790	/*
1791	 * Search to see if name exists, and get back a pointer to it.
1792	 */
1793	error = xfs_da_node_lookup_int(state, &retval);
1794	if (error) {
1795		retval = error;
1796	} else if (retval == EEXIST) {
1797		blk = &state->path.blk[ state->path.active-1 ];
1798		ASSERT(blk->bp != NULL);
1799		ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1800
1801		/*
1802		 * Get the value, local or "remote"
1803		 */
1804		retval = xfs_attr_leaf_getvalue(blk->bp, args);
1805		if (!retval && (args->rmtblkno > 0)
1806		    && !(args->flags & ATTR_KERNOVAL)) {
1807			retval = xfs_attr_rmtval_get(args);
1808		}
1809	}
1810
1811	/*
1812	 * If not in a transaction, we have to release all the buffers.
1813	 */
1814	for (i = 0; i < state->path.active; i++) {
1815		xfs_da_brelse(args->trans, state->path.blk[i].bp);
1816		state->path.blk[i].bp = NULL;
1817	}
1818
1819	xfs_da_state_free(state);
1820	return(retval);
1821}
1822
1823STATIC int							/* error */
1824xfs_attr_node_list(xfs_attr_list_context_t *context)
1825{
1826	attrlist_cursor_kern_t *cursor;
1827	xfs_attr_leafblock_t *leaf;
1828	xfs_da_intnode_t *node;
1829	xfs_da_node_entry_t *btree;
1830	int error, i;
1831	xfs_dabuf_t *bp;
1832
1833	cursor = context->cursor;
1834	cursor->initted = 1;
1835
1836	/*
1837	 * Do all sorts of validation on the passed-in cursor structure.
1838	 * If anything is amiss, ignore the cursor and look up the hashval
1839	 * starting from the btree root.
1840	 */
1841	bp = NULL;
1842	if (cursor->blkno > 0) {
1843		error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1844					      &bp, XFS_ATTR_FORK);
1845		if ((error != 0) && (error != EFSCORRUPTED))
1846			return(error);
1847		if (bp) {
1848			node = bp->data;
1849			switch (be16_to_cpu(node->hdr.info.magic)) {
1850			case XFS_DA_NODE_MAGIC:
1851				trace_xfs_attr_list_wrong_blk(context);
1852				xfs_da_brelse(NULL, bp);
1853				bp = NULL;
1854				break;
1855			case XFS_ATTR_LEAF_MAGIC:
1856				leaf = bp->data;
1857				if (cursor->hashval > be32_to_cpu(leaf->entries[
1858				    be16_to_cpu(leaf->hdr.count)-1].hashval)) {
1859					trace_xfs_attr_list_wrong_blk(context);
1860					xfs_da_brelse(NULL, bp);
1861					bp = NULL;
1862				} else if (cursor->hashval <=
1863					     be32_to_cpu(leaf->entries[0].hashval)) {
1864					trace_xfs_attr_list_wrong_blk(context);
1865					xfs_da_brelse(NULL, bp);
1866					bp = NULL;
1867				}
1868				break;
1869			default:
1870				trace_xfs_attr_list_wrong_blk(context);
1871				xfs_da_brelse(NULL, bp);
1872				bp = NULL;
1873			}
1874		}
1875	}
1876
1877	/*
1878	 * We did not find what we expected given the cursor's contents,
1879	 * so we start from the top and work down based on the hash value.
1880	 * Note that start of node block is same as start of leaf block.
1881	 */
1882	if (bp == NULL) {
1883		cursor->blkno = 0;
1884		for (;;) {
1885			error = xfs_da_read_buf(NULL, context->dp,
1886						      cursor->blkno, -1, &bp,
1887						      XFS_ATTR_FORK);
1888			if (error)
1889				return(error);
1890			if (unlikely(bp == NULL)) {
1891				XFS_ERROR_REPORT("xfs_attr_node_list(2)",
1892						 XFS_ERRLEVEL_LOW,
1893						 context->dp->i_mount);
1894				return(XFS_ERROR(EFSCORRUPTED));
1895			}
1896			node = bp->data;
1897			if (be16_to_cpu(node->hdr.info.magic)
1898							== XFS_ATTR_LEAF_MAGIC)
1899				break;
1900			if (unlikely(be16_to_cpu(node->hdr.info.magic)
1901							!= XFS_DA_NODE_MAGIC)) {
1902				XFS_CORRUPTION_ERROR("xfs_attr_node_list(3)",
1903						     XFS_ERRLEVEL_LOW,
1904						     context->dp->i_mount,
1905						     node);
1906				xfs_da_brelse(NULL, bp);
1907				return(XFS_ERROR(EFSCORRUPTED));
1908			}
1909			btree = node->btree;
1910			for (i = 0; i < be16_to_cpu(node->hdr.count);
1911								btree++, i++) {
1912				if (cursor->hashval
1913						<= be32_to_cpu(btree->hashval)) {
1914					cursor->blkno = be32_to_cpu(btree->before);
1915					trace_xfs_attr_list_node_descend(context,
1916									 btree);
1917					break;
1918				}
1919			}
1920			if (i == be16_to_cpu(node->hdr.count)) {
1921				xfs_da_brelse(NULL, bp);
1922				return(0);
1923			}
1924			xfs_da_brelse(NULL, bp);
1925		}
1926	}
1927	ASSERT(bp != NULL);
1928
1929	/*
1930	 * Roll upward through the blocks, processing each leaf block in
1931	 * order.  As long as there is space in the result buffer, keep
1932	 * adding the information.
1933	 */
1934	for (;;) {
1935		leaf = bp->data;
1936		if (unlikely(be16_to_cpu(leaf->hdr.info.magic)
1937						!= XFS_ATTR_LEAF_MAGIC)) {
1938			XFS_CORRUPTION_ERROR("xfs_attr_node_list(4)",
1939					     XFS_ERRLEVEL_LOW,
1940					     context->dp->i_mount, leaf);
1941			xfs_da_brelse(NULL, bp);
1942			return(XFS_ERROR(EFSCORRUPTED));
1943		}
1944		error = xfs_attr_leaf_list_int(bp, context);
1945		if (error) {
1946			xfs_da_brelse(NULL, bp);
1947			return error;
1948		}
1949		if (context->seen_enough || leaf->hdr.info.forw == 0)
1950			break;
1951		cursor->blkno = be32_to_cpu(leaf->hdr.info.forw);
1952		xfs_da_brelse(NULL, bp);
1953		error = xfs_da_read_buf(NULL, context->dp, cursor->blkno, -1,
1954					      &bp, XFS_ATTR_FORK);
1955		if (error)
1956			return(error);
1957		if (unlikely((bp == NULL))) {
1958			XFS_ERROR_REPORT("xfs_attr_node_list(5)",
1959					 XFS_ERRLEVEL_LOW,
1960					 context->dp->i_mount);
1961			return(XFS_ERROR(EFSCORRUPTED));
1962		}
1963	}
1964	xfs_da_brelse(NULL, bp);
1965	return(0);
1966}
1967
1968
1969/*========================================================================
1970 * External routines for manipulating out-of-line attribute values.
1971 *========================================================================*/
1972
1973/*
1974 * Read the value associated with an attribute from the out-of-line buffer
1975 * that we stored it in.
1976 */
1977int
1978xfs_attr_rmtval_get(xfs_da_args_t *args)
1979{
1980	xfs_bmbt_irec_t map[ATTR_RMTVALUE_MAPSIZE];
1981	xfs_mount_t *mp;
1982	xfs_daddr_t dblkno;
1983	xfs_caddr_t dst;
1984	xfs_buf_t *bp;
1985	int nmap, error, tmp, valuelen, blkcnt, i;
1986	xfs_dablk_t lblkno;
1987
1988	ASSERT(!(args->flags & ATTR_KERNOVAL));
1989
1990	mp = args->dp->i_mount;
1991	dst = args->value;
1992	valuelen = args->valuelen;
1993	lblkno = args->rmtblkno;
1994	while (valuelen > 0) {
1995		nmap = ATTR_RMTVALUE_MAPSIZE;
1996		error = xfs_bmapi(args->trans, args->dp, (xfs_fileoff_t)lblkno,
1997				  args->rmtblkcnt,
1998				  XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
1999				  NULL, 0, map, &nmap, NULL, NULL);
2000		if (error)
2001			return(error);
2002		ASSERT(nmap >= 1);
2003
2004		for (i = 0; (i < nmap) && (valuelen > 0); i++) {
2005			ASSERT((map[i].br_startblock != DELAYSTARTBLOCK) &&
2006			       (map[i].br_startblock != HOLESTARTBLOCK));
2007			dblkno = XFS_FSB_TO_DADDR(mp, map[i].br_startblock);
2008			blkcnt = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
2009			error = xfs_read_buf(mp, mp->m_ddev_targp, dblkno,
2010					     blkcnt,
2011					     XFS_BUF_LOCK | XBF_DONT_BLOCK,
2012					     &bp);
2013			if (error)
2014				return(error);
2015
2016			tmp = (valuelen < XFS_BUF_SIZE(bp))
2017				? valuelen : XFS_BUF_SIZE(bp);
2018			xfs_biomove(bp, 0, tmp, dst, XFS_B_READ);
2019			xfs_buf_relse(bp);
2020			dst += tmp;
2021			valuelen -= tmp;
2022
2023			lblkno += map[i].br_blockcount;
2024		}
2025	}
2026	ASSERT(valuelen == 0);
2027	return(0);
2028}
2029
2030/*
2031 * Write the value associated with an attribute into the out-of-line buffer
2032 * that we have defined for it.
2033 */
2034STATIC int
2035xfs_attr_rmtval_set(xfs_da_args_t *args)
2036{
2037	xfs_mount_t *mp;
2038	xfs_fileoff_t lfileoff;
2039	xfs_inode_t *dp;
2040	xfs_bmbt_irec_t map;
2041	xfs_daddr_t dblkno;
2042	xfs_caddr_t src;
2043	xfs_buf_t *bp;
2044	xfs_dablk_t lblkno;
2045	int blkcnt, valuelen, nmap, error, tmp, committed;
2046
2047	dp = args->dp;
2048	mp = dp->i_mount;
2049	src = args->value;
2050
2051	/*
2052	 * Find a "hole" in the attribute address space large enough for
2053	 * us to drop the new attribute's value into.
2054	 */
2055	blkcnt = XFS_B_TO_FSB(mp, args->valuelen);
2056	lfileoff = 0;
2057	error = xfs_bmap_first_unused(args->trans, args->dp, blkcnt, &lfileoff,
2058						   XFS_ATTR_FORK);
2059	if (error) {
2060		return(error);
2061	}
2062	args->rmtblkno = lblkno = (xfs_dablk_t)lfileoff;
2063	args->rmtblkcnt = blkcnt;
2064
2065	/*
2066	 * Roll through the "value", allocating blocks on disk as required.
2067	 */
2068	while (blkcnt > 0) {
2069		/*
2070		 * Allocate a single extent, up to the size of the value.
2071		 */
2072		xfs_bmap_init(args->flist, args->firstblock);
2073		nmap = 1;
2074		error = xfs_bmapi(args->trans, dp, (xfs_fileoff_t)lblkno,
2075				  blkcnt,
2076				  XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA |
2077							XFS_BMAPI_WRITE,
2078				  args->firstblock, args->total, &map, &nmap,
2079				  args->flist, NULL);
2080		if (!error) {
2081			error = xfs_bmap_finish(&args->trans, args->flist,
2082						&committed);
2083		}
2084		if (error) {
2085			ASSERT(committed);
2086			args->trans = NULL;
2087			xfs_bmap_cancel(args->flist);
2088			return(error);
2089		}
2090
2091		/*
2092		 * bmap_finish() may have committed the last trans and started
2093		 * a new one.  We need the inode to be in all transactions.
2094		 */
2095		if (committed) {
2096			xfs_trans_ijoin(args->trans, dp, XFS_ILOCK_EXCL);
2097			xfs_trans_ihold(args->trans, dp);
2098		}
2099
2100		ASSERT(nmap == 1);
2101		ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2102		       (map.br_startblock != HOLESTARTBLOCK));
2103		lblkno += map.br_blockcount;
2104		blkcnt -= map.br_blockcount;
2105
2106		/*
2107		 * Start the next trans in the chain.
2108		 */
2109		error = xfs_trans_roll(&args->trans, dp);
2110		if (error)
2111			return (error);
2112	}
2113
2114	/*
2115	 * Roll through the "value", copying the attribute value to the
2116	 * already-allocated blocks.  Blocks are written synchronously
2117	 * so that we can know they are all on disk before we turn off
2118	 * the INCOMPLETE flag.
2119	 */
2120	lblkno = args->rmtblkno;
2121	valuelen = args->valuelen;
2122	while (valuelen > 0) {
2123		/*
2124		 * Try to remember where we decided to put the value.
2125		 */
2126		xfs_bmap_init(args->flist, args->firstblock);
2127		nmap = 1;
2128		error = xfs_bmapi(NULL, dp, (xfs_fileoff_t)lblkno,
2129				  args->rmtblkcnt,
2130				  XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2131				  args->firstblock, 0, &map, &nmap,
2132				  NULL, NULL);
2133		if (error) {
2134			return(error);
2135		}
2136		ASSERT(nmap == 1);
2137		ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2138		       (map.br_startblock != HOLESTARTBLOCK));
2139
2140		dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2141		blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2142
2143		bp = xfs_buf_get(mp->m_ddev_targp, dblkno, blkcnt,
2144				 XFS_BUF_LOCK | XBF_DONT_BLOCK);
2145		ASSERT(bp);
2146		ASSERT(!XFS_BUF_GETERROR(bp));
2147
2148		tmp = (valuelen < XFS_BUF_SIZE(bp)) ? valuelen :
2149							XFS_BUF_SIZE(bp);
2150		xfs_biomove(bp, 0, tmp, src, XFS_B_WRITE);
2151		if (tmp < XFS_BUF_SIZE(bp))
2152			xfs_biozero(bp, tmp, XFS_BUF_SIZE(bp) - tmp);
2153		if ((error = xfs_bwrite(mp, bp))) {/* GROT: NOTE: synchronous write */
2154			return (error);
2155		}
2156		src += tmp;
2157		valuelen -= tmp;
2158
2159		lblkno += map.br_blockcount;
2160	}
2161	ASSERT(valuelen == 0);
2162	return(0);
2163}
2164
2165/*
2166 * Remove the value associated with an attribute by deleting the
2167 * out-of-line buffer that it is stored on.
2168 */
2169STATIC int
2170xfs_attr_rmtval_remove(xfs_da_args_t *args)
2171{
2172	xfs_mount_t *mp;
2173	xfs_bmbt_irec_t map;
2174	xfs_buf_t *bp;
2175	xfs_daddr_t dblkno;
2176	xfs_dablk_t lblkno;
2177	int valuelen, blkcnt, nmap, error, done, committed;
2178
2179	mp = args->dp->i_mount;
2180
2181	/*
2182	 * Roll through the "value", invalidating the attribute value's
2183	 * blocks.
2184	 */
2185	lblkno = args->rmtblkno;
2186	valuelen = args->rmtblkcnt;
2187	while (valuelen > 0) {
2188		/*
2189		 * Try to remember where we decided to put the value.
2190		 */
2191		xfs_bmap_init(args->flist, args->firstblock);
2192		nmap = 1;
2193		error = xfs_bmapi(NULL, args->dp, (xfs_fileoff_t)lblkno,
2194					args->rmtblkcnt,
2195					XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2196					args->firstblock, 0, &map, &nmap,
2197					args->flist, NULL);
2198		if (error) {
2199			return(error);
2200		}
2201		ASSERT(nmap == 1);
2202		ASSERT((map.br_startblock != DELAYSTARTBLOCK) &&
2203		       (map.br_startblock != HOLESTARTBLOCK));
2204
2205		dblkno = XFS_FSB_TO_DADDR(mp, map.br_startblock),
2206		blkcnt = XFS_FSB_TO_BB(mp, map.br_blockcount);
2207
2208		/*
2209		 * If the "remote" value is in the cache, remove it.
2210		 */
2211		bp = xfs_incore(mp->m_ddev_targp, dblkno, blkcnt,
2212				XFS_INCORE_TRYLOCK);
2213		if (bp) {
2214			XFS_BUF_STALE(bp);
2215			XFS_BUF_UNDELAYWRITE(bp);
2216			xfs_buf_relse(bp);
2217			bp = NULL;
2218		}
2219
2220		valuelen -= map.br_blockcount;
2221
2222		lblkno += map.br_blockcount;
2223	}
2224
2225	/*
2226	 * Keep de-allocating extents until the remote-value region is gone.
2227	 */
2228	lblkno = args->rmtblkno;
2229	blkcnt = args->rmtblkcnt;
2230	done = 0;
2231	while (!done) {
2232		xfs_bmap_init(args->flist, args->firstblock);
2233		error = xfs_bunmapi(args->trans, args->dp, lblkno, blkcnt,
2234				    XFS_BMAPI_ATTRFORK | XFS_BMAPI_METADATA,
2235				    1, args->firstblock, args->flist,
2236				    NULL, &done);
2237		if (!error) {
2238			error = xfs_bmap_finish(&args->trans, args->flist,
2239						&committed);
2240		}
2241		if (error) {
2242			ASSERT(committed);
2243			args->trans = NULL;
2244			xfs_bmap_cancel(args->flist);
2245			return(error);
2246		}
2247
2248		/*
2249		 * bmap_finish() may have committed the last trans and started
2250		 * a new one.  We need the inode to be in all transactions.
2251		 */
2252		if (committed) {
2253			xfs_trans_ijoin(args->trans, args->dp, XFS_ILOCK_EXCL);
2254			xfs_trans_ihold(args->trans, args->dp);
2255		}
2256
2257		/*
2258		 * Close out trans and start the next one in the chain.
2259		 */
2260		error = xfs_trans_roll(&args->trans, args->dp);
2261		if (error)
2262			return (error);
2263	}
2264	return(0);
2265}