Linux Audio

Check our new training course

Loading...
Note: File does not exist in v3.1.
   1/*
   2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3 * All Rights Reserved.
   4 *
   5 * This program is free software; you can redistribute it and/or
   6 * modify it under the terms of the GNU General Public License as
   7 * published by the Free Software Foundation.
   8 *
   9 * This program is distributed in the hope that it would be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12 * GNU General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License
  15 * along with this program; if not, write the Free Software Foundation,
  16 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17 */
  18#include "xfs.h"
  19#include "xfs_fs.h"
  20#include "xfs_shared.h"
  21#include "xfs_format.h"
  22#include "xfs_log_format.h"
  23#include "xfs_trans_resv.h"
  24#include "xfs_bit.h"
  25#include "xfs_mount.h"
  26#include "xfs_defer.h"
  27#include "xfs_da_format.h"
  28#include "xfs_da_btree.h"
  29#include "xfs_attr_sf.h"
  30#include "xfs_inode.h"
  31#include "xfs_alloc.h"
  32#include "xfs_trans.h"
  33#include "xfs_inode_item.h"
  34#include "xfs_bmap.h"
  35#include "xfs_bmap_util.h"
  36#include "xfs_bmap_btree.h"
  37#include "xfs_attr.h"
  38#include "xfs_attr_leaf.h"
  39#include "xfs_attr_remote.h"
  40#include "xfs_error.h"
  41#include "xfs_quota.h"
  42#include "xfs_trans_space.h"
  43#include "xfs_trace.h"
  44
  45/*
  46 * xfs_attr.c
  47 *
  48 * Provide the external interfaces to manage attribute lists.
  49 */
  50
  51/*========================================================================
  52 * Function prototypes for the kernel.
  53 *========================================================================*/
  54
  55/*
  56 * Internal routines when attribute list fits inside the inode.
  57 */
  58STATIC int xfs_attr_shortform_addname(xfs_da_args_t *args);
  59
  60/*
  61 * Internal routines when attribute list is one block.
  62 */
  63STATIC int xfs_attr_leaf_get(xfs_da_args_t *args);
  64STATIC int xfs_attr_leaf_addname(xfs_da_args_t *args);
  65STATIC int xfs_attr_leaf_removename(xfs_da_args_t *args);
  66
  67/*
  68 * Internal routines when attribute list is more than one block.
  69 */
  70STATIC int xfs_attr_node_get(xfs_da_args_t *args);
  71STATIC int xfs_attr_node_addname(xfs_da_args_t *args);
  72STATIC int xfs_attr_node_removename(xfs_da_args_t *args);
  73STATIC int xfs_attr_fillstate(xfs_da_state_t *state);
  74STATIC int xfs_attr_refillstate(xfs_da_state_t *state);
  75
  76
  77STATIC int
  78xfs_attr_args_init(
  79	struct xfs_da_args	*args,
  80	struct xfs_inode	*dp,
  81	const unsigned char	*name,
  82	int			flags)
  83{
  84
  85	if (!name)
  86		return -EINVAL;
  87
  88	memset(args, 0, sizeof(*args));
  89	args->geo = dp->i_mount->m_attr_geo;
  90	args->whichfork = XFS_ATTR_FORK;
  91	args->dp = dp;
  92	args->flags = flags;
  93	args->name = name;
  94	args->namelen = strlen((const char *)name);
  95	if (args->namelen >= MAXNAMELEN)
  96		return -EFAULT;		/* match IRIX behaviour */
  97
  98	args->hashval = xfs_da_hashname(args->name, args->namelen);
  99	return 0;
 100}
 101
 102int
 103xfs_inode_hasattr(
 104	struct xfs_inode	*ip)
 105{
 106	if (!XFS_IFORK_Q(ip) ||
 107	    (ip->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 108	     ip->i_d.di_anextents == 0))
 109		return 0;
 110	return 1;
 111}
 112
 113/*========================================================================
 114 * Overall external interface routines.
 115 *========================================================================*/
 116
 117/* Retrieve an extended attribute and its value.  Must have ilock. */
 118int
 119xfs_attr_get_ilocked(
 120	struct xfs_inode	*ip,
 121	struct xfs_da_args	*args)
 122{
 123	ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED | XFS_ILOCK_EXCL));
 124
 125	if (!xfs_inode_hasattr(ip))
 126		return -ENOATTR;
 127	else if (ip->i_d.di_aformat == XFS_DINODE_FMT_LOCAL)
 128		return xfs_attr_shortform_getvalue(args);
 129	else if (xfs_bmap_one_block(ip, XFS_ATTR_FORK))
 130		return xfs_attr_leaf_get(args);
 131	else
 132		return xfs_attr_node_get(args);
 133}
 134
 135/* Retrieve an extended attribute by name, and its value. */
 136int
 137xfs_attr_get(
 138	struct xfs_inode	*ip,
 139	const unsigned char	*name,
 140	unsigned char		*value,
 141	int			*valuelenp,
 142	int			flags)
 143{
 144	struct xfs_da_args	args;
 145	uint			lock_mode;
 146	int			error;
 147
 148	XFS_STATS_INC(ip->i_mount, xs_attr_get);
 149
 150	if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 151		return -EIO;
 152
 153	error = xfs_attr_args_init(&args, ip, name, flags);
 154	if (error)
 155		return error;
 156
 157	args.value = value;
 158	args.valuelen = *valuelenp;
 159	/* Entirely possible to look up a name which doesn't exist */
 160	args.op_flags = XFS_DA_OP_OKNOENT;
 161
 162	lock_mode = xfs_ilock_attr_map_shared(ip);
 163	error = xfs_attr_get_ilocked(ip, &args);
 164	xfs_iunlock(ip, lock_mode);
 165
 166	*valuelenp = args.valuelen;
 167	return error == -EEXIST ? 0 : error;
 168}
 169
 170/*
 171 * Calculate how many blocks we need for the new attribute,
 172 */
 173STATIC int
 174xfs_attr_calc_size(
 175	struct xfs_da_args	*args,
 176	int			*local)
 177{
 178	struct xfs_mount	*mp = args->dp->i_mount;
 179	int			size;
 180	int			nblks;
 181
 182	/*
 183	 * Determine space new attribute will use, and if it would be
 184	 * "local" or "remote" (note: local != inline).
 185	 */
 186	size = xfs_attr_leaf_newentsize(args, local);
 187	nblks = XFS_DAENTER_SPACE_RES(mp, XFS_ATTR_FORK);
 188	if (*local) {
 189		if (size > (args->geo->blksize / 2)) {
 190			/* Double split possible */
 191			nblks *= 2;
 192		}
 193	} else {
 194		/*
 195		 * Out of line attribute, cannot double split, but
 196		 * make room for the attribute value itself.
 197		 */
 198		uint	dblocks = xfs_attr3_rmt_blocks(mp, args->valuelen);
 199		nblks += dblocks;
 200		nblks += XFS_NEXTENTADD_SPACE_RES(mp, dblocks, XFS_ATTR_FORK);
 201	}
 202
 203	return nblks;
 204}
 205
 206int
 207xfs_attr_set(
 208	struct xfs_inode	*dp,
 209	const unsigned char	*name,
 210	unsigned char		*value,
 211	int			valuelen,
 212	int			flags)
 213{
 214	struct xfs_mount	*mp = dp->i_mount;
 215	struct xfs_buf		*leaf_bp = NULL;
 216	struct xfs_da_args	args;
 217	struct xfs_defer_ops	dfops;
 218	struct xfs_trans_res	tres;
 219	xfs_fsblock_t		firstblock;
 220	int			rsvd = (flags & ATTR_ROOT) != 0;
 221	int			error, err2, local;
 222
 223	XFS_STATS_INC(mp, xs_attr_set);
 224
 225	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
 226		return -EIO;
 227
 228	error = xfs_attr_args_init(&args, dp, name, flags);
 229	if (error)
 230		return error;
 231
 232	args.value = value;
 233	args.valuelen = valuelen;
 234	args.firstblock = &firstblock;
 235	args.dfops = &dfops;
 236	args.op_flags = XFS_DA_OP_ADDNAME | XFS_DA_OP_OKNOENT;
 237	args.total = xfs_attr_calc_size(&args, &local);
 238
 239	error = xfs_qm_dqattach(dp, 0);
 240	if (error)
 241		return error;
 242
 243	/*
 244	 * If the inode doesn't have an attribute fork, add one.
 245	 * (inode must not be locked when we call this routine)
 246	 */
 247	if (XFS_IFORK_Q(dp) == 0) {
 248		int sf_size = sizeof(xfs_attr_sf_hdr_t) +
 249			XFS_ATTR_SF_ENTSIZE_BYNAME(args.namelen, valuelen);
 250
 251		error = xfs_bmap_add_attrfork(dp, sf_size, rsvd);
 252		if (error)
 253			return error;
 254	}
 255
 256	tres.tr_logres = M_RES(mp)->tr_attrsetm.tr_logres +
 257			 M_RES(mp)->tr_attrsetrt.tr_logres * args.total;
 258	tres.tr_logcount = XFS_ATTRSET_LOG_COUNT;
 259	tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
 260
 261	/*
 262	 * Root fork attributes can use reserved data blocks for this
 263	 * operation if necessary
 264	 */
 265	error = xfs_trans_alloc(mp, &tres, args.total, 0,
 266			rsvd ? XFS_TRANS_RESERVE : 0, &args.trans);
 267	if (error)
 268		return error;
 269
 270	xfs_ilock(dp, XFS_ILOCK_EXCL);
 271	error = xfs_trans_reserve_quota_nblks(args.trans, dp, args.total, 0,
 272				rsvd ? XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
 273				       XFS_QMOPT_RES_REGBLKS);
 274	if (error) {
 275		xfs_iunlock(dp, XFS_ILOCK_EXCL);
 276		xfs_trans_cancel(args.trans);
 277		return error;
 278	}
 279
 280	xfs_trans_ijoin(args.trans, dp, 0);
 281
 282	/*
 283	 * If the attribute list is non-existent or a shortform list,
 284	 * upgrade it to a single-leaf-block attribute list.
 285	 */
 286	if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL ||
 287	    (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS &&
 288	     dp->i_d.di_anextents == 0)) {
 289
 290		/*
 291		 * Build initial attribute list (if required).
 292		 */
 293		if (dp->i_d.di_aformat == XFS_DINODE_FMT_EXTENTS)
 294			xfs_attr_shortform_create(&args);
 295
 296		/*
 297		 * Try to add the attr to the attribute list in
 298		 * the inode.
 299		 */
 300		error = xfs_attr_shortform_addname(&args);
 301		if (error != -ENOSPC) {
 302			/*
 303			 * Commit the shortform mods, and we're done.
 304			 * NOTE: this is also the error path (EEXIST, etc).
 305			 */
 306			ASSERT(args.trans != NULL);
 307
 308			/*
 309			 * If this is a synchronous mount, make sure that
 310			 * the transaction goes to disk before returning
 311			 * to the user.
 312			 */
 313			if (mp->m_flags & XFS_MOUNT_WSYNC)
 314				xfs_trans_set_sync(args.trans);
 315
 316			if (!error && (flags & ATTR_KERNOTIME) == 0) {
 317				xfs_trans_ichgtime(args.trans, dp,
 318							XFS_ICHGTIME_CHG);
 319			}
 320			err2 = xfs_trans_commit(args.trans);
 321			xfs_iunlock(dp, XFS_ILOCK_EXCL);
 322
 323			return error ? error : err2;
 324		}
 325
 326		/*
 327		 * It won't fit in the shortform, transform to a leaf block.
 328		 * GROT: another possible req'mt for a double-split btree op.
 329		 */
 330		xfs_defer_init(args.dfops, args.firstblock);
 331		error = xfs_attr_shortform_to_leaf(&args, &leaf_bp);
 332		if (error)
 333			goto out_defer_cancel;
 334		/*
 335		 * Prevent the leaf buffer from being unlocked so that a
 336		 * concurrent AIL push cannot grab the half-baked leaf
 337		 * buffer and run into problems with the write verifier.
 338		 */
 339		xfs_trans_bhold(args.trans, leaf_bp);
 340		xfs_defer_bjoin(args.dfops, leaf_bp);
 341		xfs_defer_ijoin(args.dfops, dp);
 342		error = xfs_defer_finish(&args.trans, args.dfops);
 343		if (error)
 344			goto out_defer_cancel;
 345
 346		/*
 347		 * Commit the leaf transformation.  We'll need another (linked)
 348		 * transaction to add the new attribute to the leaf, which
 349		 * means that we have to hold & join the leaf buffer here too.
 350		 */
 351		error = xfs_trans_roll_inode(&args.trans, dp);
 352		if (error)
 353			goto out;
 354		xfs_trans_bjoin(args.trans, leaf_bp);
 355		leaf_bp = NULL;
 356	}
 357
 358	if (xfs_bmap_one_block(dp, XFS_ATTR_FORK))
 359		error = xfs_attr_leaf_addname(&args);
 360	else
 361		error = xfs_attr_node_addname(&args);
 362	if (error)
 363		goto out;
 364
 365	/*
 366	 * If this is a synchronous mount, make sure that the
 367	 * transaction goes to disk before returning to the user.
 368	 */
 369	if (mp->m_flags & XFS_MOUNT_WSYNC)
 370		xfs_trans_set_sync(args.trans);
 371
 372	if ((flags & ATTR_KERNOTIME) == 0)
 373		xfs_trans_ichgtime(args.trans, dp, XFS_ICHGTIME_CHG);
 374
 375	/*
 376	 * Commit the last in the sequence of transactions.
 377	 */
 378	xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
 379	error = xfs_trans_commit(args.trans);
 380	xfs_iunlock(dp, XFS_ILOCK_EXCL);
 381
 382	return error;
 383
 384out_defer_cancel:
 385	xfs_defer_cancel(&dfops);
 386out:
 387	if (leaf_bp)
 388		xfs_trans_brelse(args.trans, leaf_bp);
 389	if (args.trans)
 390		xfs_trans_cancel(args.trans);
 391	xfs_iunlock(dp, XFS_ILOCK_EXCL);
 392	return error;
 393}
 394
 395/*
 396 * Generic handler routine to remove a name from an attribute list.
 397 * Transitions attribute list from Btree to shortform as necessary.
 398 */
 399int
 400xfs_attr_remove(
 401	struct xfs_inode	*dp,
 402	const unsigned char	*name,
 403	int			flags)
 404{
 405	struct xfs_mount	*mp = dp->i_mount;
 406	struct xfs_da_args	args;
 407	struct xfs_defer_ops	dfops;
 408	xfs_fsblock_t		firstblock;
 409	int			error;
 410
 411	XFS_STATS_INC(mp, xs_attr_remove);
 412
 413	if (XFS_FORCED_SHUTDOWN(dp->i_mount))
 414		return -EIO;
 415
 416	error = xfs_attr_args_init(&args, dp, name, flags);
 417	if (error)
 418		return error;
 419
 420	args.firstblock = &firstblock;
 421	args.dfops = &dfops;
 422
 423	/*
 424	 * we have no control over the attribute names that userspace passes us
 425	 * to remove, so we have to allow the name lookup prior to attribute
 426	 * removal to fail.
 427	 */
 428	args.op_flags = XFS_DA_OP_OKNOENT;
 429
 430	error = xfs_qm_dqattach(dp, 0);
 431	if (error)
 432		return error;
 433
 434	/*
 435	 * Root fork attributes can use reserved data blocks for this
 436	 * operation if necessary
 437	 */
 438	error = xfs_trans_alloc(mp, &M_RES(mp)->tr_attrrm,
 439			XFS_ATTRRM_SPACE_RES(mp), 0,
 440			(flags & ATTR_ROOT) ? XFS_TRANS_RESERVE : 0,
 441			&args.trans);
 442	if (error)
 443		return error;
 444
 445	xfs_ilock(dp, XFS_ILOCK_EXCL);
 446	/*
 447	 * No need to make quota reservations here. We expect to release some
 448	 * blocks not allocate in the common case.
 449	 */
 450	xfs_trans_ijoin(args.trans, dp, 0);
 451
 452	if (!xfs_inode_hasattr(dp)) {
 453		error = -ENOATTR;
 454	} else if (dp->i_d.di_aformat == XFS_DINODE_FMT_LOCAL) {
 455		ASSERT(dp->i_afp->if_flags & XFS_IFINLINE);
 456		error = xfs_attr_shortform_remove(&args);
 457	} else if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
 458		error = xfs_attr_leaf_removename(&args);
 459	} else {
 460		error = xfs_attr_node_removename(&args);
 461	}
 462
 463	if (error)
 464		goto out;
 465
 466	/*
 467	 * If this is a synchronous mount, make sure that the
 468	 * transaction goes to disk before returning to the user.
 469	 */
 470	if (mp->m_flags & XFS_MOUNT_WSYNC)
 471		xfs_trans_set_sync(args.trans);
 472
 473	if ((flags & ATTR_KERNOTIME) == 0)
 474		xfs_trans_ichgtime(args.trans, dp, XFS_ICHGTIME_CHG);
 475
 476	/*
 477	 * Commit the last in the sequence of transactions.
 478	 */
 479	xfs_trans_log_inode(args.trans, dp, XFS_ILOG_CORE);
 480	error = xfs_trans_commit(args.trans);
 481	xfs_iunlock(dp, XFS_ILOCK_EXCL);
 482
 483	return error;
 484
 485out:
 486	if (args.trans)
 487		xfs_trans_cancel(args.trans);
 488	xfs_iunlock(dp, XFS_ILOCK_EXCL);
 489	return error;
 490}
 491
 492/*========================================================================
 493 * External routines when attribute list is inside the inode
 494 *========================================================================*/
 495
 496/*
 497 * Add a name to the shortform attribute list structure
 498 * This is the external routine.
 499 */
 500STATIC int
 501xfs_attr_shortform_addname(xfs_da_args_t *args)
 502{
 503	int newsize, forkoff, retval;
 504
 505	trace_xfs_attr_sf_addname(args);
 506
 507	retval = xfs_attr_shortform_lookup(args);
 508	if ((args->flags & ATTR_REPLACE) && (retval == -ENOATTR)) {
 509		return retval;
 510	} else if (retval == -EEXIST) {
 511		if (args->flags & ATTR_CREATE)
 512			return retval;
 513		retval = xfs_attr_shortform_remove(args);
 514		if (retval)
 515			return retval;
 516		/*
 517		 * Since we have removed the old attr, clear ATTR_REPLACE so
 518		 * that the leaf format add routine won't trip over the attr
 519		 * not being around.
 520		 */
 521		args->flags &= ~ATTR_REPLACE;
 522	}
 523
 524	if (args->namelen >= XFS_ATTR_SF_ENTSIZE_MAX ||
 525	    args->valuelen >= XFS_ATTR_SF_ENTSIZE_MAX)
 526		return -ENOSPC;
 527
 528	newsize = XFS_ATTR_SF_TOTSIZE(args->dp);
 529	newsize += XFS_ATTR_SF_ENTSIZE_BYNAME(args->namelen, args->valuelen);
 530
 531	forkoff = xfs_attr_shortform_bytesfit(args->dp, newsize);
 532	if (!forkoff)
 533		return -ENOSPC;
 534
 535	xfs_attr_shortform_add(args, forkoff);
 536	return 0;
 537}
 538
 539
 540/*========================================================================
 541 * External routines when attribute list is one block
 542 *========================================================================*/
 543
 544/*
 545 * Add a name to the leaf attribute list structure
 546 *
 547 * This leaf block cannot have a "remote" value, we only call this routine
 548 * if bmap_one_block() says there is only one block (ie: no remote blks).
 549 */
 550STATIC int
 551xfs_attr_leaf_addname(xfs_da_args_t *args)
 552{
 553	xfs_inode_t *dp;
 554	struct xfs_buf *bp;
 555	int retval, error, forkoff;
 556
 557	trace_xfs_attr_leaf_addname(args);
 558
 559	/*
 560	 * Read the (only) block in the attribute list in.
 561	 */
 562	dp = args->dp;
 563	args->blkno = 0;
 564	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
 565	if (error)
 566		return error;
 567
 568	/*
 569	 * Look up the given attribute in the leaf block.  Figure out if
 570	 * the given flags produce an error or call for an atomic rename.
 571	 */
 572	retval = xfs_attr3_leaf_lookup_int(bp, args);
 573	if ((args->flags & ATTR_REPLACE) && (retval == -ENOATTR)) {
 574		xfs_trans_brelse(args->trans, bp);
 575		return retval;
 576	} else if (retval == -EEXIST) {
 577		if (args->flags & ATTR_CREATE) {	/* pure create op */
 578			xfs_trans_brelse(args->trans, bp);
 579			return retval;
 580		}
 581
 582		trace_xfs_attr_leaf_replace(args);
 583
 584		/* save the attribute state for later removal*/
 585		args->op_flags |= XFS_DA_OP_RENAME;	/* an atomic rename */
 586		args->blkno2 = args->blkno;		/* set 2nd entry info*/
 587		args->index2 = args->index;
 588		args->rmtblkno2 = args->rmtblkno;
 589		args->rmtblkcnt2 = args->rmtblkcnt;
 590		args->rmtvaluelen2 = args->rmtvaluelen;
 591
 592		/*
 593		 * clear the remote attr state now that it is saved so that the
 594		 * values reflect the state of the attribute we are about to
 595		 * add, not the attribute we just found and will remove later.
 596		 */
 597		args->rmtblkno = 0;
 598		args->rmtblkcnt = 0;
 599		args->rmtvaluelen = 0;
 600	}
 601
 602	/*
 603	 * Add the attribute to the leaf block, transitioning to a Btree
 604	 * if required.
 605	 */
 606	retval = xfs_attr3_leaf_add(bp, args);
 607	if (retval == -ENOSPC) {
 608		/*
 609		 * Promote the attribute list to the Btree format, then
 610		 * Commit that transaction so that the node_addname() call
 611		 * can manage its own transactions.
 612		 */
 613		xfs_defer_init(args->dfops, args->firstblock);
 614		error = xfs_attr3_leaf_to_node(args);
 615		if (error)
 616			goto out_defer_cancel;
 617		xfs_defer_ijoin(args->dfops, dp);
 618		error = xfs_defer_finish(&args->trans, args->dfops);
 619		if (error)
 620			goto out_defer_cancel;
 621
 622		/*
 623		 * Commit the current trans (including the inode) and start
 624		 * a new one.
 625		 */
 626		error = xfs_trans_roll_inode(&args->trans, dp);
 627		if (error)
 628			return error;
 629
 630		/*
 631		 * Fob the whole rest of the problem off on the Btree code.
 632		 */
 633		error = xfs_attr_node_addname(args);
 634		return error;
 635	}
 636
 637	/*
 638	 * Commit the transaction that added the attr name so that
 639	 * later routines can manage their own transactions.
 640	 */
 641	error = xfs_trans_roll_inode(&args->trans, dp);
 642	if (error)
 643		return error;
 644
 645	/*
 646	 * If there was an out-of-line value, allocate the blocks we
 647	 * identified for its storage and copy the value.  This is done
 648	 * after we create the attribute so that we don't overflow the
 649	 * maximum size of a transaction and/or hit a deadlock.
 650	 */
 651	if (args->rmtblkno > 0) {
 652		error = xfs_attr_rmtval_set(args);
 653		if (error)
 654			return error;
 655	}
 656
 657	/*
 658	 * If this is an atomic rename operation, we must "flip" the
 659	 * incomplete flags on the "new" and "old" attribute/value pairs
 660	 * so that one disappears and one appears atomically.  Then we
 661	 * must remove the "old" attribute/value pair.
 662	 */
 663	if (args->op_flags & XFS_DA_OP_RENAME) {
 664		/*
 665		 * In a separate transaction, set the incomplete flag on the
 666		 * "old" attr and clear the incomplete flag on the "new" attr.
 667		 */
 668		error = xfs_attr3_leaf_flipflags(args);
 669		if (error)
 670			return error;
 671
 672		/*
 673		 * Dismantle the "old" attribute/value pair by removing
 674		 * a "remote" value (if it exists).
 675		 */
 676		args->index = args->index2;
 677		args->blkno = args->blkno2;
 678		args->rmtblkno = args->rmtblkno2;
 679		args->rmtblkcnt = args->rmtblkcnt2;
 680		args->rmtvaluelen = args->rmtvaluelen2;
 681		if (args->rmtblkno) {
 682			error = xfs_attr_rmtval_remove(args);
 683			if (error)
 684				return error;
 685		}
 686
 687		/*
 688		 * Read in the block containing the "old" attr, then
 689		 * remove the "old" attr from that block (neat, huh!)
 690		 */
 691		error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno,
 692					   -1, &bp);
 693		if (error)
 694			return error;
 695
 696		xfs_attr3_leaf_remove(bp, args);
 697
 698		/*
 699		 * If the result is small enough, shrink it all into the inode.
 700		 */
 701		if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
 702			xfs_defer_init(args->dfops, args->firstblock);
 703			error = xfs_attr3_leaf_to_shortform(bp, args, forkoff);
 704			/* bp is gone due to xfs_da_shrink_inode */
 705			if (error)
 706				goto out_defer_cancel;
 707			xfs_defer_ijoin(args->dfops, dp);
 708			error = xfs_defer_finish(&args->trans, args->dfops);
 709			if (error)
 710				goto out_defer_cancel;
 711		}
 712
 713		/*
 714		 * Commit the remove and start the next trans in series.
 715		 */
 716		error = xfs_trans_roll_inode(&args->trans, dp);
 717
 718	} else if (args->rmtblkno > 0) {
 719		/*
 720		 * Added a "remote" value, just clear the incomplete flag.
 721		 */
 722		error = xfs_attr3_leaf_clearflag(args);
 723	}
 724	return error;
 725out_defer_cancel:
 726	xfs_defer_cancel(args->dfops);
 727	return error;
 728}
 729
 730/*
 731 * Remove a name from the leaf attribute list structure
 732 *
 733 * This leaf block cannot have a "remote" value, we only call this routine
 734 * if bmap_one_block() says there is only one block (ie: no remote blks).
 735 */
 736STATIC int
 737xfs_attr_leaf_removename(xfs_da_args_t *args)
 738{
 739	xfs_inode_t *dp;
 740	struct xfs_buf *bp;
 741	int error, forkoff;
 742
 743	trace_xfs_attr_leaf_removename(args);
 744
 745	/*
 746	 * Remove the attribute.
 747	 */
 748	dp = args->dp;
 749	args->blkno = 0;
 750	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
 751	if (error)
 752		return error;
 753
 754	error = xfs_attr3_leaf_lookup_int(bp, args);
 755	if (error == -ENOATTR) {
 756		xfs_trans_brelse(args->trans, bp);
 757		return error;
 758	}
 759
 760	xfs_attr3_leaf_remove(bp, args);
 761
 762	/*
 763	 * If the result is small enough, shrink it all into the inode.
 764	 */
 765	if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
 766		xfs_defer_init(args->dfops, args->firstblock);
 767		error = xfs_attr3_leaf_to_shortform(bp, args, forkoff);
 768		/* bp is gone due to xfs_da_shrink_inode */
 769		if (error)
 770			goto out_defer_cancel;
 771		xfs_defer_ijoin(args->dfops, dp);
 772		error = xfs_defer_finish(&args->trans, args->dfops);
 773		if (error)
 774			goto out_defer_cancel;
 775	}
 776	return 0;
 777out_defer_cancel:
 778	xfs_defer_cancel(args->dfops);
 779	return error;
 780}
 781
 782/*
 783 * Look up a name in a leaf attribute list structure.
 784 *
 785 * This leaf block cannot have a "remote" value, we only call this routine
 786 * if bmap_one_block() says there is only one block (ie: no remote blks).
 787 */
 788STATIC int
 789xfs_attr_leaf_get(xfs_da_args_t *args)
 790{
 791	struct xfs_buf *bp;
 792	int error;
 793
 794	trace_xfs_attr_leaf_get(args);
 795
 796	args->blkno = 0;
 797	error = xfs_attr3_leaf_read(args->trans, args->dp, args->blkno, -1, &bp);
 798	if (error)
 799		return error;
 800
 801	error = xfs_attr3_leaf_lookup_int(bp, args);
 802	if (error != -EEXIST)  {
 803		xfs_trans_brelse(args->trans, bp);
 804		return error;
 805	}
 806	error = xfs_attr3_leaf_getvalue(bp, args);
 807	xfs_trans_brelse(args->trans, bp);
 808	if (!error && (args->rmtblkno > 0) && !(args->flags & ATTR_KERNOVAL)) {
 809		error = xfs_attr_rmtval_get(args);
 810	}
 811	return error;
 812}
 813
 814/*========================================================================
 815 * External routines when attribute list size > geo->blksize
 816 *========================================================================*/
 817
 818/*
 819 * Add a name to a Btree-format attribute list.
 820 *
 821 * This will involve walking down the Btree, and may involve splitting
 822 * leaf nodes and even splitting intermediate nodes up to and including
 823 * the root node (a special case of an intermediate node).
 824 *
 825 * "Remote" attribute values confuse the issue and atomic rename operations
 826 * add a whole extra layer of confusion on top of that.
 827 */
 828STATIC int
 829xfs_attr_node_addname(xfs_da_args_t *args)
 830{
 831	xfs_da_state_t *state;
 832	xfs_da_state_blk_t *blk;
 833	xfs_inode_t *dp;
 834	xfs_mount_t *mp;
 835	int retval, error;
 836
 837	trace_xfs_attr_node_addname(args);
 838
 839	/*
 840	 * Fill in bucket of arguments/results/context to carry around.
 841	 */
 842	dp = args->dp;
 843	mp = dp->i_mount;
 844restart:
 845	state = xfs_da_state_alloc();
 846	state->args = args;
 847	state->mp = mp;
 848
 849	/*
 850	 * Search to see if name already exists, and get back a pointer
 851	 * to where it should go.
 852	 */
 853	error = xfs_da3_node_lookup_int(state, &retval);
 854	if (error)
 855		goto out;
 856	blk = &state->path.blk[ state->path.active-1 ];
 857	ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
 858	if ((args->flags & ATTR_REPLACE) && (retval == -ENOATTR)) {
 859		goto out;
 860	} else if (retval == -EEXIST) {
 861		if (args->flags & ATTR_CREATE)
 862			goto out;
 863
 864		trace_xfs_attr_node_replace(args);
 865
 866		/* save the attribute state for later removal*/
 867		args->op_flags |= XFS_DA_OP_RENAME;	/* atomic rename op */
 868		args->blkno2 = args->blkno;		/* set 2nd entry info*/
 869		args->index2 = args->index;
 870		args->rmtblkno2 = args->rmtblkno;
 871		args->rmtblkcnt2 = args->rmtblkcnt;
 872		args->rmtvaluelen2 = args->rmtvaluelen;
 873
 874		/*
 875		 * clear the remote attr state now that it is saved so that the
 876		 * values reflect the state of the attribute we are about to
 877		 * add, not the attribute we just found and will remove later.
 878		 */
 879		args->rmtblkno = 0;
 880		args->rmtblkcnt = 0;
 881		args->rmtvaluelen = 0;
 882	}
 883
 884	retval = xfs_attr3_leaf_add(blk->bp, state->args);
 885	if (retval == -ENOSPC) {
 886		if (state->path.active == 1) {
 887			/*
 888			 * Its really a single leaf node, but it had
 889			 * out-of-line values so it looked like it *might*
 890			 * have been a b-tree.
 891			 */
 892			xfs_da_state_free(state);
 893			state = NULL;
 894			xfs_defer_init(args->dfops, args->firstblock);
 895			error = xfs_attr3_leaf_to_node(args);
 896			if (error)
 897				goto out_defer_cancel;
 898			xfs_defer_ijoin(args->dfops, dp);
 899			error = xfs_defer_finish(&args->trans, args->dfops);
 900			if (error)
 901				goto out_defer_cancel;
 902
 903			/*
 904			 * Commit the node conversion and start the next
 905			 * trans in the chain.
 906			 */
 907			error = xfs_trans_roll_inode(&args->trans, dp);
 908			if (error)
 909				goto out;
 910
 911			goto restart;
 912		}
 913
 914		/*
 915		 * Split as many Btree elements as required.
 916		 * This code tracks the new and old attr's location
 917		 * in the index/blkno/rmtblkno/rmtblkcnt fields and
 918		 * in the index2/blkno2/rmtblkno2/rmtblkcnt2 fields.
 919		 */
 920		xfs_defer_init(args->dfops, args->firstblock);
 921		error = xfs_da3_split(state);
 922		if (error)
 923			goto out_defer_cancel;
 924		xfs_defer_ijoin(args->dfops, dp);
 925		error = xfs_defer_finish(&args->trans, args->dfops);
 926		if (error)
 927			goto out_defer_cancel;
 928	} else {
 929		/*
 930		 * Addition succeeded, update Btree hashvals.
 931		 */
 932		xfs_da3_fixhashpath(state, &state->path);
 933	}
 934
 935	/*
 936	 * Kill the state structure, we're done with it and need to
 937	 * allow the buffers to come back later.
 938	 */
 939	xfs_da_state_free(state);
 940	state = NULL;
 941
 942	/*
 943	 * Commit the leaf addition or btree split and start the next
 944	 * trans in the chain.
 945	 */
 946	error = xfs_trans_roll_inode(&args->trans, dp);
 947	if (error)
 948		goto out;
 949
 950	/*
 951	 * If there was an out-of-line value, allocate the blocks we
 952	 * identified for its storage and copy the value.  This is done
 953	 * after we create the attribute so that we don't overflow the
 954	 * maximum size of a transaction and/or hit a deadlock.
 955	 */
 956	if (args->rmtblkno > 0) {
 957		error = xfs_attr_rmtval_set(args);
 958		if (error)
 959			return error;
 960	}
 961
 962	/*
 963	 * If this is an atomic rename operation, we must "flip" the
 964	 * incomplete flags on the "new" and "old" attribute/value pairs
 965	 * so that one disappears and one appears atomically.  Then we
 966	 * must remove the "old" attribute/value pair.
 967	 */
 968	if (args->op_flags & XFS_DA_OP_RENAME) {
 969		/*
 970		 * In a separate transaction, set the incomplete flag on the
 971		 * "old" attr and clear the incomplete flag on the "new" attr.
 972		 */
 973		error = xfs_attr3_leaf_flipflags(args);
 974		if (error)
 975			goto out;
 976
 977		/*
 978		 * Dismantle the "old" attribute/value pair by removing
 979		 * a "remote" value (if it exists).
 980		 */
 981		args->index = args->index2;
 982		args->blkno = args->blkno2;
 983		args->rmtblkno = args->rmtblkno2;
 984		args->rmtblkcnt = args->rmtblkcnt2;
 985		args->rmtvaluelen = args->rmtvaluelen2;
 986		if (args->rmtblkno) {
 987			error = xfs_attr_rmtval_remove(args);
 988			if (error)
 989				return error;
 990		}
 991
 992		/*
 993		 * Re-find the "old" attribute entry after any split ops.
 994		 * The INCOMPLETE flag means that we will find the "old"
 995		 * attr, not the "new" one.
 996		 */
 997		args->flags |= XFS_ATTR_INCOMPLETE;
 998		state = xfs_da_state_alloc();
 999		state->args = args;
1000		state->mp = mp;
1001		state->inleaf = 0;
1002		error = xfs_da3_node_lookup_int(state, &retval);
1003		if (error)
1004			goto out;
1005
1006		/*
1007		 * Remove the name and update the hashvals in the tree.
1008		 */
1009		blk = &state->path.blk[ state->path.active-1 ];
1010		ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1011		error = xfs_attr3_leaf_remove(blk->bp, args);
1012		xfs_da3_fixhashpath(state, &state->path);
1013
1014		/*
1015		 * Check to see if the tree needs to be collapsed.
1016		 */
1017		if (retval && (state->path.active > 1)) {
1018			xfs_defer_init(args->dfops, args->firstblock);
1019			error = xfs_da3_join(state);
1020			if (error)
1021				goto out_defer_cancel;
1022			xfs_defer_ijoin(args->dfops, dp);
1023			error = xfs_defer_finish(&args->trans, args->dfops);
1024			if (error)
1025				goto out_defer_cancel;
1026		}
1027
1028		/*
1029		 * Commit and start the next trans in the chain.
1030		 */
1031		error = xfs_trans_roll_inode(&args->trans, dp);
1032		if (error)
1033			goto out;
1034
1035	} else if (args->rmtblkno > 0) {
1036		/*
1037		 * Added a "remote" value, just clear the incomplete flag.
1038		 */
1039		error = xfs_attr3_leaf_clearflag(args);
1040		if (error)
1041			goto out;
1042	}
1043	retval = error = 0;
1044
1045out:
1046	if (state)
1047		xfs_da_state_free(state);
1048	if (error)
1049		return error;
1050	return retval;
1051out_defer_cancel:
1052	xfs_defer_cancel(args->dfops);
1053	goto out;
1054}
1055
1056/*
1057 * Remove a name from a B-tree attribute list.
1058 *
1059 * This will involve walking down the Btree, and may involve joining
1060 * leaf nodes and even joining intermediate nodes up to and including
1061 * the root node (a special case of an intermediate node).
1062 */
1063STATIC int
1064xfs_attr_node_removename(xfs_da_args_t *args)
1065{
1066	xfs_da_state_t *state;
1067	xfs_da_state_blk_t *blk;
1068	xfs_inode_t *dp;
1069	struct xfs_buf *bp;
1070	int retval, error, forkoff;
1071
1072	trace_xfs_attr_node_removename(args);
1073
1074	/*
1075	 * Tie a string around our finger to remind us where we are.
1076	 */
1077	dp = args->dp;
1078	state = xfs_da_state_alloc();
1079	state->args = args;
1080	state->mp = dp->i_mount;
1081
1082	/*
1083	 * Search to see if name exists, and get back a pointer to it.
1084	 */
1085	error = xfs_da3_node_lookup_int(state, &retval);
1086	if (error || (retval != -EEXIST)) {
1087		if (error == 0)
1088			error = retval;
1089		goto out;
1090	}
1091
1092	/*
1093	 * If there is an out-of-line value, de-allocate the blocks.
1094	 * This is done before we remove the attribute so that we don't
1095	 * overflow the maximum size of a transaction and/or hit a deadlock.
1096	 */
1097	blk = &state->path.blk[ state->path.active-1 ];
1098	ASSERT(blk->bp != NULL);
1099	ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1100	if (args->rmtblkno > 0) {
1101		/*
1102		 * Fill in disk block numbers in the state structure
1103		 * so that we can get the buffers back after we commit
1104		 * several transactions in the following calls.
1105		 */
1106		error = xfs_attr_fillstate(state);
1107		if (error)
1108			goto out;
1109
1110		/*
1111		 * Mark the attribute as INCOMPLETE, then bunmapi() the
1112		 * remote value.
1113		 */
1114		error = xfs_attr3_leaf_setflag(args);
1115		if (error)
1116			goto out;
1117		error = xfs_attr_rmtval_remove(args);
1118		if (error)
1119			goto out;
1120
1121		/*
1122		 * Refill the state structure with buffers, the prior calls
1123		 * released our buffers.
1124		 */
1125		error = xfs_attr_refillstate(state);
1126		if (error)
1127			goto out;
1128	}
1129
1130	/*
1131	 * Remove the name and update the hashvals in the tree.
1132	 */
1133	blk = &state->path.blk[ state->path.active-1 ];
1134	ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1135	retval = xfs_attr3_leaf_remove(blk->bp, args);
1136	xfs_da3_fixhashpath(state, &state->path);
1137
1138	/*
1139	 * Check to see if the tree needs to be collapsed.
1140	 */
1141	if (retval && (state->path.active > 1)) {
1142		xfs_defer_init(args->dfops, args->firstblock);
1143		error = xfs_da3_join(state);
1144		if (error)
1145			goto out_defer_cancel;
1146		xfs_defer_ijoin(args->dfops, dp);
1147		error = xfs_defer_finish(&args->trans, args->dfops);
1148		if (error)
1149			goto out_defer_cancel;
1150		/*
1151		 * Commit the Btree join operation and start a new trans.
1152		 */
1153		error = xfs_trans_roll_inode(&args->trans, dp);
1154		if (error)
1155			goto out;
1156	}
1157
1158	/*
1159	 * If the result is small enough, push it all into the inode.
1160	 */
1161	if (xfs_bmap_one_block(dp, XFS_ATTR_FORK)) {
1162		/*
1163		 * Have to get rid of the copy of this dabuf in the state.
1164		 */
1165		ASSERT(state->path.active == 1);
1166		ASSERT(state->path.blk[0].bp);
1167		state->path.blk[0].bp = NULL;
1168
1169		error = xfs_attr3_leaf_read(args->trans, args->dp, 0, -1, &bp);
1170		if (error)
1171			goto out;
1172
1173		if ((forkoff = xfs_attr_shortform_allfit(bp, dp))) {
1174			xfs_defer_init(args->dfops, args->firstblock);
1175			error = xfs_attr3_leaf_to_shortform(bp, args, forkoff);
1176			/* bp is gone due to xfs_da_shrink_inode */
1177			if (error)
1178				goto out_defer_cancel;
1179			xfs_defer_ijoin(args->dfops, dp);
1180			error = xfs_defer_finish(&args->trans, args->dfops);
1181			if (error)
1182				goto out_defer_cancel;
1183		} else
1184			xfs_trans_brelse(args->trans, bp);
1185	}
1186	error = 0;
1187
1188out:
1189	xfs_da_state_free(state);
1190	return error;
1191out_defer_cancel:
1192	xfs_defer_cancel(args->dfops);
1193	goto out;
1194}
1195
1196/*
1197 * Fill in the disk block numbers in the state structure for the buffers
1198 * that are attached to the state structure.
1199 * This is done so that we can quickly reattach ourselves to those buffers
1200 * after some set of transaction commits have released these buffers.
1201 */
1202STATIC int
1203xfs_attr_fillstate(xfs_da_state_t *state)
1204{
1205	xfs_da_state_path_t *path;
1206	xfs_da_state_blk_t *blk;
1207	int level;
1208
1209	trace_xfs_attr_fillstate(state->args);
1210
1211	/*
1212	 * Roll down the "path" in the state structure, storing the on-disk
1213	 * block number for those buffers in the "path".
1214	 */
1215	path = &state->path;
1216	ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1217	for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1218		if (blk->bp) {
1219			blk->disk_blkno = XFS_BUF_ADDR(blk->bp);
1220			blk->bp = NULL;
1221		} else {
1222			blk->disk_blkno = 0;
1223		}
1224	}
1225
1226	/*
1227	 * Roll down the "altpath" in the state structure, storing the on-disk
1228	 * block number for those buffers in the "altpath".
1229	 */
1230	path = &state->altpath;
1231	ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1232	for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1233		if (blk->bp) {
1234			blk->disk_blkno = XFS_BUF_ADDR(blk->bp);
1235			blk->bp = NULL;
1236		} else {
1237			blk->disk_blkno = 0;
1238		}
1239	}
1240
1241	return 0;
1242}
1243
1244/*
1245 * Reattach the buffers to the state structure based on the disk block
1246 * numbers stored in the state structure.
1247 * This is done after some set of transaction commits have released those
1248 * buffers from our grip.
1249 */
1250STATIC int
1251xfs_attr_refillstate(xfs_da_state_t *state)
1252{
1253	xfs_da_state_path_t *path;
1254	xfs_da_state_blk_t *blk;
1255	int level, error;
1256
1257	trace_xfs_attr_refillstate(state->args);
1258
1259	/*
1260	 * Roll down the "path" in the state structure, storing the on-disk
1261	 * block number for those buffers in the "path".
1262	 */
1263	path = &state->path;
1264	ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1265	for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1266		if (blk->disk_blkno) {
1267			error = xfs_da3_node_read(state->args->trans,
1268						state->args->dp,
1269						blk->blkno, blk->disk_blkno,
1270						&blk->bp, XFS_ATTR_FORK);
1271			if (error)
1272				return error;
1273		} else {
1274			blk->bp = NULL;
1275		}
1276	}
1277
1278	/*
1279	 * Roll down the "altpath" in the state structure, storing the on-disk
1280	 * block number for those buffers in the "altpath".
1281	 */
1282	path = &state->altpath;
1283	ASSERT((path->active >= 0) && (path->active < XFS_DA_NODE_MAXDEPTH));
1284	for (blk = path->blk, level = 0; level < path->active; blk++, level++) {
1285		if (blk->disk_blkno) {
1286			error = xfs_da3_node_read(state->args->trans,
1287						state->args->dp,
1288						blk->blkno, blk->disk_blkno,
1289						&blk->bp, XFS_ATTR_FORK);
1290			if (error)
1291				return error;
1292		} else {
1293			blk->bp = NULL;
1294		}
1295	}
1296
1297	return 0;
1298}
1299
1300/*
1301 * Look up a filename in a node attribute list.
1302 *
1303 * This routine gets called for any attribute fork that has more than one
1304 * block, ie: both true Btree attr lists and for single-leaf-blocks with
1305 * "remote" values taking up more blocks.
1306 */
1307STATIC int
1308xfs_attr_node_get(xfs_da_args_t *args)
1309{
1310	xfs_da_state_t *state;
1311	xfs_da_state_blk_t *blk;
1312	int error, retval;
1313	int i;
1314
1315	trace_xfs_attr_node_get(args);
1316
1317	state = xfs_da_state_alloc();
1318	state->args = args;
1319	state->mp = args->dp->i_mount;
1320
1321	/*
1322	 * Search to see if name exists, and get back a pointer to it.
1323	 */
1324	error = xfs_da3_node_lookup_int(state, &retval);
1325	if (error) {
1326		retval = error;
1327	} else if (retval == -EEXIST) {
1328		blk = &state->path.blk[ state->path.active-1 ];
1329		ASSERT(blk->bp != NULL);
1330		ASSERT(blk->magic == XFS_ATTR_LEAF_MAGIC);
1331
1332		/*
1333		 * Get the value, local or "remote"
1334		 */
1335		retval = xfs_attr3_leaf_getvalue(blk->bp, args);
1336		if (!retval && (args->rmtblkno > 0)
1337		    && !(args->flags & ATTR_KERNOVAL)) {
1338			retval = xfs_attr_rmtval_get(args);
1339		}
1340	}
1341
1342	/*
1343	 * If not in a transaction, we have to release all the buffers.
1344	 */
1345	for (i = 0; i < state->path.active; i++) {
1346		xfs_trans_brelse(args->trans, state->path.blk[i].bp);
1347		state->path.blk[i].bp = NULL;
1348	}
1349
1350	xfs_da_state_free(state);
1351	return retval;
1352}