Linux Audio

Check our new training course

Loading...
Note: File does not exist in v6.13.7.
   1/*
   2 * Copyright (c) 2000-2002,2005 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_inum.h"
  26#include "xfs_sb.h"
  27#include "xfs_ag.h"
  28#include "xfs_mount.h"
  29#include "xfs_inode.h"
  30#include "xfs_btree.h"
  31#include "xfs_ialloc.h"
  32#include "xfs_ialloc_btree.h"
  33#include "xfs_alloc.h"
  34#include "xfs_rtalloc.h"
  35#include "xfs_error.h"
  36#include "xfs_bmap.h"
  37#include "xfs_cksum.h"
  38#include "xfs_trans.h"
  39#include "xfs_buf_item.h"
  40#include "xfs_icreate_item.h"
  41#include "xfs_icache.h"
  42#include "xfs_dinode.h"
  43#include "xfs_trace.h"
  44
  45
  46/*
  47 * Allocation group level functions.
  48 */
  49static inline int
  50xfs_ialloc_cluster_alignment(
  51	xfs_alloc_arg_t	*args)
  52{
  53	if (xfs_sb_version_hasalign(&args->mp->m_sb) &&
  54	    args->mp->m_sb.sb_inoalignmt >=
  55	     XFS_B_TO_FSBT(args->mp, args->mp->m_inode_cluster_size))
  56		return args->mp->m_sb.sb_inoalignmt;
  57	return 1;
  58}
  59
  60/*
  61 * Lookup a record by ino in the btree given by cur.
  62 */
  63int					/* error */
  64xfs_inobt_lookup(
  65	struct xfs_btree_cur	*cur,	/* btree cursor */
  66	xfs_agino_t		ino,	/* starting inode of chunk */
  67	xfs_lookup_t		dir,	/* <=, >=, == */
  68	int			*stat)	/* success/failure */
  69{
  70	cur->bc_rec.i.ir_startino = ino;
  71	cur->bc_rec.i.ir_freecount = 0;
  72	cur->bc_rec.i.ir_free = 0;
  73	return xfs_btree_lookup(cur, dir, stat);
  74}
  75
  76/*
  77 * Update the record referred to by cur to the value given.
  78 * This either works (return 0) or gets an EFSCORRUPTED error.
  79 */
  80STATIC int				/* error */
  81xfs_inobt_update(
  82	struct xfs_btree_cur	*cur,	/* btree cursor */
  83	xfs_inobt_rec_incore_t	*irec)	/* btree record */
  84{
  85	union xfs_btree_rec	rec;
  86
  87	rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
  88	rec.inobt.ir_freecount = cpu_to_be32(irec->ir_freecount);
  89	rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
  90	return xfs_btree_update(cur, &rec);
  91}
  92
  93/*
  94 * Get the data from the pointed-to record.
  95 */
  96int					/* error */
  97xfs_inobt_get_rec(
  98	struct xfs_btree_cur	*cur,	/* btree cursor */
  99	xfs_inobt_rec_incore_t	*irec,	/* btree record */
 100	int			*stat)	/* output: success/failure */
 101{
 102	union xfs_btree_rec	*rec;
 103	int			error;
 104
 105	error = xfs_btree_get_rec(cur, &rec, stat);
 106	if (!error && *stat == 1) {
 107		irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
 108		irec->ir_freecount = be32_to_cpu(rec->inobt.ir_freecount);
 109		irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
 110	}
 111	return error;
 112}
 113
 114/*
 115 * Verify that the number of free inodes in the AGI is correct.
 116 */
 117#ifdef DEBUG
 118STATIC int
 119xfs_check_agi_freecount(
 120	struct xfs_btree_cur	*cur,
 121	struct xfs_agi		*agi)
 122{
 123	if (cur->bc_nlevels == 1) {
 124		xfs_inobt_rec_incore_t rec;
 125		int		freecount = 0;
 126		int		error;
 127		int		i;
 128
 129		error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
 130		if (error)
 131			return error;
 132
 133		do {
 134			error = xfs_inobt_get_rec(cur, &rec, &i);
 135			if (error)
 136				return error;
 137
 138			if (i) {
 139				freecount += rec.ir_freecount;
 140				error = xfs_btree_increment(cur, 0, &i);
 141				if (error)
 142					return error;
 143			}
 144		} while (i == 1);
 145
 146		if (!XFS_FORCED_SHUTDOWN(cur->bc_mp))
 147			ASSERT(freecount == be32_to_cpu(agi->agi_freecount));
 148	}
 149	return 0;
 150}
 151#else
 152#define xfs_check_agi_freecount(cur, agi)	0
 153#endif
 154
 155/*
 156 * Initialise a new set of inodes. When called without a transaction context
 157 * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
 158 * than logging them (which in a transaction context puts them into the AIL
 159 * for writeback rather than the xfsbufd queue).
 160 */
 161int
 162xfs_ialloc_inode_init(
 163	struct xfs_mount	*mp,
 164	struct xfs_trans	*tp,
 165	struct list_head	*buffer_list,
 166	xfs_agnumber_t		agno,
 167	xfs_agblock_t		agbno,
 168	xfs_agblock_t		length,
 169	unsigned int		gen)
 170{
 171	struct xfs_buf		*fbuf;
 172	struct xfs_dinode	*free;
 173	int			nbufs, blks_per_cluster, inodes_per_cluster;
 174	int			version;
 175	int			i, j;
 176	xfs_daddr_t		d;
 177	xfs_ino_t		ino = 0;
 178
 179	/*
 180	 * Loop over the new block(s), filling in the inodes.  For small block
 181	 * sizes, manipulate the inodes in buffers  which are multiples of the
 182	 * blocks size.
 183	 */
 184	blks_per_cluster = xfs_icluster_size_fsb(mp);
 185	inodes_per_cluster = blks_per_cluster << mp->m_sb.sb_inopblog;
 186	nbufs = length / blks_per_cluster;
 187
 188	/*
 189	 * Figure out what version number to use in the inodes we create.  If
 190	 * the superblock version has caught up to the one that supports the new
 191	 * inode format, then use the new inode version.  Otherwise use the old
 192	 * version so that old kernels will continue to be able to use the file
 193	 * system.
 194	 *
 195	 * For v3 inodes, we also need to write the inode number into the inode,
 196	 * so calculate the first inode number of the chunk here as
 197	 * XFS_OFFBNO_TO_AGINO() only works within a filesystem block, not
 198	 * across multiple filesystem blocks (such as a cluster) and so cannot
 199	 * be used in the cluster buffer loop below.
 200	 *
 201	 * Further, because we are writing the inode directly into the buffer
 202	 * and calculating a CRC on the entire inode, we have ot log the entire
 203	 * inode so that the entire range the CRC covers is present in the log.
 204	 * That means for v3 inode we log the entire buffer rather than just the
 205	 * inode cores.
 206	 */
 207	if (xfs_sb_version_hascrc(&mp->m_sb)) {
 208		version = 3;
 209		ino = XFS_AGINO_TO_INO(mp, agno,
 210				       XFS_OFFBNO_TO_AGINO(mp, agbno, 0));
 211
 212		/*
 213		 * log the initialisation that is about to take place as an
 214		 * logical operation. This means the transaction does not
 215		 * need to log the physical changes to the inode buffers as log
 216		 * recovery will know what initialisation is actually needed.
 217		 * Hence we only need to log the buffers as "ordered" buffers so
 218		 * they track in the AIL as if they were physically logged.
 219		 */
 220		if (tp)
 221			xfs_icreate_log(tp, agno, agbno, mp->m_ialloc_inos,
 222					mp->m_sb.sb_inodesize, length, gen);
 223	} else if (xfs_sb_version_hasnlink(&mp->m_sb))
 224		version = 2;
 225	else
 226		version = 1;
 227
 228	for (j = 0; j < nbufs; j++) {
 229		/*
 230		 * Get the block.
 231		 */
 232		d = XFS_AGB_TO_DADDR(mp, agno, agbno + (j * blks_per_cluster));
 233		fbuf = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
 234					 mp->m_bsize * blks_per_cluster,
 235					 XBF_UNMAPPED);
 236		if (!fbuf)
 237			return ENOMEM;
 238
 239		/* Initialize the inode buffers and log them appropriately. */
 240		fbuf->b_ops = &xfs_inode_buf_ops;
 241		xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length));
 242		for (i = 0; i < inodes_per_cluster; i++) {
 243			int	ioffset = i << mp->m_sb.sb_inodelog;
 244			uint	isize = xfs_dinode_size(version);
 245
 246			free = xfs_make_iptr(mp, fbuf, i);
 247			free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
 248			free->di_version = version;
 249			free->di_gen = cpu_to_be32(gen);
 250			free->di_next_unlinked = cpu_to_be32(NULLAGINO);
 251
 252			if (version == 3) {
 253				free->di_ino = cpu_to_be64(ino);
 254				ino++;
 255				uuid_copy(&free->di_uuid, &mp->m_sb.sb_uuid);
 256				xfs_dinode_calc_crc(mp, free);
 257			} else if (tp) {
 258				/* just log the inode core */
 259				xfs_trans_log_buf(tp, fbuf, ioffset,
 260						  ioffset + isize - 1);
 261			}
 262		}
 263
 264		if (tp) {
 265			/*
 266			 * Mark the buffer as an inode allocation buffer so it
 267			 * sticks in AIL at the point of this allocation
 268			 * transaction. This ensures the they are on disk before
 269			 * the tail of the log can be moved past this
 270			 * transaction (i.e. by preventing relogging from moving
 271			 * it forward in the log).
 272			 */
 273			xfs_trans_inode_alloc_buf(tp, fbuf);
 274			if (version == 3) {
 275				/*
 276				 * Mark the buffer as ordered so that they are
 277				 * not physically logged in the transaction but
 278				 * still tracked in the AIL as part of the
 279				 * transaction and pin the log appropriately.
 280				 */
 281				xfs_trans_ordered_buf(tp, fbuf);
 282				xfs_trans_log_buf(tp, fbuf, 0,
 283						  BBTOB(fbuf->b_length) - 1);
 284			}
 285		} else {
 286			fbuf->b_flags |= XBF_DONE;
 287			xfs_buf_delwri_queue(fbuf, buffer_list);
 288			xfs_buf_relse(fbuf);
 289		}
 290	}
 291	return 0;
 292}
 293
 294/*
 295 * Allocate new inodes in the allocation group specified by agbp.
 296 * Return 0 for success, else error code.
 297 */
 298STATIC int				/* error code or 0 */
 299xfs_ialloc_ag_alloc(
 300	xfs_trans_t	*tp,		/* transaction pointer */
 301	xfs_buf_t	*agbp,		/* alloc group buffer */
 302	int		*alloc)
 303{
 304	xfs_agi_t	*agi;		/* allocation group header */
 305	xfs_alloc_arg_t	args;		/* allocation argument structure */
 306	xfs_btree_cur_t	*cur;		/* inode btree cursor */
 307	xfs_agnumber_t	agno;
 308	int		error;
 309	int		i;
 310	xfs_agino_t	newino;		/* new first inode's number */
 311	xfs_agino_t	newlen;		/* new number of inodes */
 312	xfs_agino_t	thisino;	/* current inode number, for loop */
 313	int		isaligned = 0;	/* inode allocation at stripe unit */
 314					/* boundary */
 315	struct xfs_perag *pag;
 316
 317	memset(&args, 0, sizeof(args));
 318	args.tp = tp;
 319	args.mp = tp->t_mountp;
 320
 321	/*
 322	 * Locking will ensure that we don't have two callers in here
 323	 * at one time.
 324	 */
 325	newlen = args.mp->m_ialloc_inos;
 326	if (args.mp->m_maxicount &&
 327	    args.mp->m_sb.sb_icount + newlen > args.mp->m_maxicount)
 328		return XFS_ERROR(ENOSPC);
 329	args.minlen = args.maxlen = args.mp->m_ialloc_blks;
 330	/*
 331	 * First try to allocate inodes contiguous with the last-allocated
 332	 * chunk of inodes.  If the filesystem is striped, this will fill
 333	 * an entire stripe unit with inodes.
 334	 */
 335	agi = XFS_BUF_TO_AGI(agbp);
 336	newino = be32_to_cpu(agi->agi_newino);
 337	agno = be32_to_cpu(agi->agi_seqno);
 338	args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
 339		     args.mp->m_ialloc_blks;
 340	if (likely(newino != NULLAGINO &&
 341		  (args.agbno < be32_to_cpu(agi->agi_length)))) {
 342		args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
 343		args.type = XFS_ALLOCTYPE_THIS_BNO;
 344		args.prod = 1;
 345
 346		/*
 347		 * We need to take into account alignment here to ensure that
 348		 * we don't modify the free list if we fail to have an exact
 349		 * block. If we don't have an exact match, and every oher
 350		 * attempt allocation attempt fails, we'll end up cancelling
 351		 * a dirty transaction and shutting down.
 352		 *
 353		 * For an exact allocation, alignment must be 1,
 354		 * however we need to take cluster alignment into account when
 355		 * fixing up the freelist. Use the minalignslop field to
 356		 * indicate that extra blocks might be required for alignment,
 357		 * but not to use them in the actual exact allocation.
 358		 */
 359		args.alignment = 1;
 360		args.minalignslop = xfs_ialloc_cluster_alignment(&args) - 1;
 361
 362		/* Allow space for the inode btree to split. */
 363		args.minleft = args.mp->m_in_maxlevels - 1;
 364		if ((error = xfs_alloc_vextent(&args)))
 365			return error;
 366
 367		/*
 368		 * This request might have dirtied the transaction if the AG can
 369		 * satisfy the request, but the exact block was not available.
 370		 * If the allocation did fail, subsequent requests will relax
 371		 * the exact agbno requirement and increase the alignment
 372		 * instead. It is critical that the total size of the request
 373		 * (len + alignment + slop) does not increase from this point
 374		 * on, so reset minalignslop to ensure it is not included in
 375		 * subsequent requests.
 376		 */
 377		args.minalignslop = 0;
 378	} else
 379		args.fsbno = NULLFSBLOCK;
 380
 381	if (unlikely(args.fsbno == NULLFSBLOCK)) {
 382		/*
 383		 * Set the alignment for the allocation.
 384		 * If stripe alignment is turned on then align at stripe unit
 385		 * boundary.
 386		 * If the cluster size is smaller than a filesystem block
 387		 * then we're doing I/O for inodes in filesystem block size
 388		 * pieces, so don't need alignment anyway.
 389		 */
 390		isaligned = 0;
 391		if (args.mp->m_sinoalign) {
 392			ASSERT(!(args.mp->m_flags & XFS_MOUNT_NOALIGN));
 393			args.alignment = args.mp->m_dalign;
 394			isaligned = 1;
 395		} else
 396			args.alignment = xfs_ialloc_cluster_alignment(&args);
 397		/*
 398		 * Need to figure out where to allocate the inode blocks.
 399		 * Ideally they should be spaced out through the a.g.
 400		 * For now, just allocate blocks up front.
 401		 */
 402		args.agbno = be32_to_cpu(agi->agi_root);
 403		args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
 404		/*
 405		 * Allocate a fixed-size extent of inodes.
 406		 */
 407		args.type = XFS_ALLOCTYPE_NEAR_BNO;
 408		args.prod = 1;
 409		/*
 410		 * Allow space for the inode btree to split.
 411		 */
 412		args.minleft = args.mp->m_in_maxlevels - 1;
 413		if ((error = xfs_alloc_vextent(&args)))
 414			return error;
 415	}
 416
 417	/*
 418	 * If stripe alignment is turned on, then try again with cluster
 419	 * alignment.
 420	 */
 421	if (isaligned && args.fsbno == NULLFSBLOCK) {
 422		args.type = XFS_ALLOCTYPE_NEAR_BNO;
 423		args.agbno = be32_to_cpu(agi->agi_root);
 424		args.fsbno = XFS_AGB_TO_FSB(args.mp, agno, args.agbno);
 425		args.alignment = xfs_ialloc_cluster_alignment(&args);
 426		if ((error = xfs_alloc_vextent(&args)))
 427			return error;
 428	}
 429
 430	if (args.fsbno == NULLFSBLOCK) {
 431		*alloc = 0;
 432		return 0;
 433	}
 434	ASSERT(args.len == args.minlen);
 435
 436	/*
 437	 * Stamp and write the inode buffers.
 438	 *
 439	 * Seed the new inode cluster with a random generation number. This
 440	 * prevents short-term reuse of generation numbers if a chunk is
 441	 * freed and then immediately reallocated. We use random numbers
 442	 * rather than a linear progression to prevent the next generation
 443	 * number from being easily guessable.
 444	 */
 445	error = xfs_ialloc_inode_init(args.mp, tp, NULL, agno, args.agbno,
 446			args.len, prandom_u32());
 447
 448	if (error)
 449		return error;
 450	/*
 451	 * Convert the results.
 452	 */
 453	newino = XFS_OFFBNO_TO_AGINO(args.mp, args.agbno, 0);
 454	be32_add_cpu(&agi->agi_count, newlen);
 455	be32_add_cpu(&agi->agi_freecount, newlen);
 456	pag = xfs_perag_get(args.mp, agno);
 457	pag->pagi_freecount += newlen;
 458	xfs_perag_put(pag);
 459	agi->agi_newino = cpu_to_be32(newino);
 460
 461	/*
 462	 * Insert records describing the new inode chunk into the btree.
 463	 */
 464	cur = xfs_inobt_init_cursor(args.mp, tp, agbp, agno);
 465	for (thisino = newino;
 466	     thisino < newino + newlen;
 467	     thisino += XFS_INODES_PER_CHUNK) {
 468		cur->bc_rec.i.ir_startino = thisino;
 469		cur->bc_rec.i.ir_freecount = XFS_INODES_PER_CHUNK;
 470		cur->bc_rec.i.ir_free = XFS_INOBT_ALL_FREE;
 471		error = xfs_btree_lookup(cur, XFS_LOOKUP_EQ, &i);
 472		if (error) {
 473			xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 474			return error;
 475		}
 476		ASSERT(i == 0);
 477		error = xfs_btree_insert(cur, &i);
 478		if (error) {
 479			xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 480			return error;
 481		}
 482		ASSERT(i == 1);
 483	}
 484	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 485	/*
 486	 * Log allocation group header fields
 487	 */
 488	xfs_ialloc_log_agi(tp, agbp,
 489		XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
 490	/*
 491	 * Modify/log superblock values for inode count and inode free count.
 492	 */
 493	xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
 494	xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
 495	*alloc = 1;
 496	return 0;
 497}
 498
 499STATIC xfs_agnumber_t
 500xfs_ialloc_next_ag(
 501	xfs_mount_t	*mp)
 502{
 503	xfs_agnumber_t	agno;
 504
 505	spin_lock(&mp->m_agirotor_lock);
 506	agno = mp->m_agirotor;
 507	if (++mp->m_agirotor >= mp->m_maxagi)
 508		mp->m_agirotor = 0;
 509	spin_unlock(&mp->m_agirotor_lock);
 510
 511	return agno;
 512}
 513
 514/*
 515 * Select an allocation group to look for a free inode in, based on the parent
 516 * inode and the mode.  Return the allocation group buffer.
 517 */
 518STATIC xfs_agnumber_t
 519xfs_ialloc_ag_select(
 520	xfs_trans_t	*tp,		/* transaction pointer */
 521	xfs_ino_t	parent,		/* parent directory inode number */
 522	umode_t		mode,		/* bits set to indicate file type */
 523	int		okalloc)	/* ok to allocate more space */
 524{
 525	xfs_agnumber_t	agcount;	/* number of ag's in the filesystem */
 526	xfs_agnumber_t	agno;		/* current ag number */
 527	int		flags;		/* alloc buffer locking flags */
 528	xfs_extlen_t	ineed;		/* blocks needed for inode allocation */
 529	xfs_extlen_t	longest = 0;	/* longest extent available */
 530	xfs_mount_t	*mp;		/* mount point structure */
 531	int		needspace;	/* file mode implies space allocated */
 532	xfs_perag_t	*pag;		/* per allocation group data */
 533	xfs_agnumber_t	pagno;		/* parent (starting) ag number */
 534	int		error;
 535
 536	/*
 537	 * Files of these types need at least one block if length > 0
 538	 * (and they won't fit in the inode, but that's hard to figure out).
 539	 */
 540	needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
 541	mp = tp->t_mountp;
 542	agcount = mp->m_maxagi;
 543	if (S_ISDIR(mode))
 544		pagno = xfs_ialloc_next_ag(mp);
 545	else {
 546		pagno = XFS_INO_TO_AGNO(mp, parent);
 547		if (pagno >= agcount)
 548			pagno = 0;
 549	}
 550
 551	ASSERT(pagno < agcount);
 552
 553	/*
 554	 * Loop through allocation groups, looking for one with a little
 555	 * free space in it.  Note we don't look for free inodes, exactly.
 556	 * Instead, we include whether there is a need to allocate inodes
 557	 * to mean that blocks must be allocated for them,
 558	 * if none are currently free.
 559	 */
 560	agno = pagno;
 561	flags = XFS_ALLOC_FLAG_TRYLOCK;
 562	for (;;) {
 563		pag = xfs_perag_get(mp, agno);
 564		if (!pag->pagi_inodeok) {
 565			xfs_ialloc_next_ag(mp);
 566			goto nextag;
 567		}
 568
 569		if (!pag->pagi_init) {
 570			error = xfs_ialloc_pagi_init(mp, tp, agno);
 571			if (error)
 572				goto nextag;
 573		}
 574
 575		if (pag->pagi_freecount) {
 576			xfs_perag_put(pag);
 577			return agno;
 578		}
 579
 580		if (!okalloc)
 581			goto nextag;
 582
 583		if (!pag->pagf_init) {
 584			error = xfs_alloc_pagf_init(mp, tp, agno, flags);
 585			if (error)
 586				goto nextag;
 587		}
 588
 589		/*
 590		 * Is there enough free space for the file plus a block of
 591		 * inodes? (if we need to allocate some)?
 592		 */
 593		ineed = mp->m_ialloc_blks;
 594		longest = pag->pagf_longest;
 595		if (!longest)
 596			longest = pag->pagf_flcount > 0;
 597
 598		if (pag->pagf_freeblks >= needspace + ineed &&
 599		    longest >= ineed) {
 600			xfs_perag_put(pag);
 601			return agno;
 602		}
 603nextag:
 604		xfs_perag_put(pag);
 605		/*
 606		 * No point in iterating over the rest, if we're shutting
 607		 * down.
 608		 */
 609		if (XFS_FORCED_SHUTDOWN(mp))
 610			return NULLAGNUMBER;
 611		agno++;
 612		if (agno >= agcount)
 613			agno = 0;
 614		if (agno == pagno) {
 615			if (flags == 0)
 616				return NULLAGNUMBER;
 617			flags = 0;
 618		}
 619	}
 620}
 621
 622/*
 623 * Try to retrieve the next record to the left/right from the current one.
 624 */
 625STATIC int
 626xfs_ialloc_next_rec(
 627	struct xfs_btree_cur	*cur,
 628	xfs_inobt_rec_incore_t	*rec,
 629	int			*done,
 630	int			left)
 631{
 632	int                     error;
 633	int			i;
 634
 635	if (left)
 636		error = xfs_btree_decrement(cur, 0, &i);
 637	else
 638		error = xfs_btree_increment(cur, 0, &i);
 639
 640	if (error)
 641		return error;
 642	*done = !i;
 643	if (i) {
 644		error = xfs_inobt_get_rec(cur, rec, &i);
 645		if (error)
 646			return error;
 647		XFS_WANT_CORRUPTED_RETURN(i == 1);
 648	}
 649
 650	return 0;
 651}
 652
 653STATIC int
 654xfs_ialloc_get_rec(
 655	struct xfs_btree_cur	*cur,
 656	xfs_agino_t		agino,
 657	xfs_inobt_rec_incore_t	*rec,
 658	int			*done)
 659{
 660	int                     error;
 661	int			i;
 662
 663	error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
 664	if (error)
 665		return error;
 666	*done = !i;
 667	if (i) {
 668		error = xfs_inobt_get_rec(cur, rec, &i);
 669		if (error)
 670			return error;
 671		XFS_WANT_CORRUPTED_RETURN(i == 1);
 672	}
 673
 674	return 0;
 675}
 676
 677/*
 678 * Allocate an inode.
 679 *
 680 * The caller selected an AG for us, and made sure that free inodes are
 681 * available.
 682 */
 683STATIC int
 684xfs_dialloc_ag(
 685	struct xfs_trans	*tp,
 686	struct xfs_buf		*agbp,
 687	xfs_ino_t		parent,
 688	xfs_ino_t		*inop)
 689{
 690	struct xfs_mount	*mp = tp->t_mountp;
 691	struct xfs_agi		*agi = XFS_BUF_TO_AGI(agbp);
 692	xfs_agnumber_t		agno = be32_to_cpu(agi->agi_seqno);
 693	xfs_agnumber_t		pagno = XFS_INO_TO_AGNO(mp, parent);
 694	xfs_agino_t		pagino = XFS_INO_TO_AGINO(mp, parent);
 695	struct xfs_perag	*pag;
 696	struct xfs_btree_cur	*cur, *tcur;
 697	struct xfs_inobt_rec_incore rec, trec;
 698	xfs_ino_t		ino;
 699	int			error;
 700	int			offset;
 701	int			i, j;
 702
 703	pag = xfs_perag_get(mp, agno);
 704
 705	ASSERT(pag->pagi_init);
 706	ASSERT(pag->pagi_inodeok);
 707	ASSERT(pag->pagi_freecount > 0);
 708
 709 restart_pagno:
 710	cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
 711	/*
 712	 * If pagino is 0 (this is the root inode allocation) use newino.
 713	 * This must work because we've just allocated some.
 714	 */
 715	if (!pagino)
 716		pagino = be32_to_cpu(agi->agi_newino);
 717
 718	error = xfs_check_agi_freecount(cur, agi);
 719	if (error)
 720		goto error0;
 721
 722	/*
 723	 * If in the same AG as the parent, try to get near the parent.
 724	 */
 725	if (pagno == agno) {
 726		int		doneleft;	/* done, to the left */
 727		int		doneright;	/* done, to the right */
 728		int		searchdistance = 10;
 729
 730		error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
 731		if (error)
 732			goto error0;
 733		XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
 734
 735		error = xfs_inobt_get_rec(cur, &rec, &j);
 736		if (error)
 737			goto error0;
 738		XFS_WANT_CORRUPTED_GOTO(j == 1, error0);
 739
 740		if (rec.ir_freecount > 0) {
 741			/*
 742			 * Found a free inode in the same chunk
 743			 * as the parent, done.
 744			 */
 745			goto alloc_inode;
 746		}
 747
 748
 749		/*
 750		 * In the same AG as parent, but parent's chunk is full.
 751		 */
 752
 753		/* duplicate the cursor, search left & right simultaneously */
 754		error = xfs_btree_dup_cursor(cur, &tcur);
 755		if (error)
 756			goto error0;
 757
 758		/*
 759		 * Skip to last blocks looked up if same parent inode.
 760		 */
 761		if (pagino != NULLAGINO &&
 762		    pag->pagl_pagino == pagino &&
 763		    pag->pagl_leftrec != NULLAGINO &&
 764		    pag->pagl_rightrec != NULLAGINO) {
 765			error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
 766						   &trec, &doneleft);
 767			if (error)
 768				goto error1;
 769
 770			error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
 771						   &rec, &doneright);
 772			if (error)
 773				goto error1;
 774		} else {
 775			/* search left with tcur, back up 1 record */
 776			error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
 777			if (error)
 778				goto error1;
 779
 780			/* search right with cur, go forward 1 record. */
 781			error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
 782			if (error)
 783				goto error1;
 784		}
 785
 786		/*
 787		 * Loop until we find an inode chunk with a free inode.
 788		 */
 789		while (!doneleft || !doneright) {
 790			int	useleft;  /* using left inode chunk this time */
 791
 792			if (!--searchdistance) {
 793				/*
 794				 * Not in range - save last search
 795				 * location and allocate a new inode
 796				 */
 797				xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
 798				pag->pagl_leftrec = trec.ir_startino;
 799				pag->pagl_rightrec = rec.ir_startino;
 800				pag->pagl_pagino = pagino;
 801				goto newino;
 802			}
 803
 804			/* figure out the closer block if both are valid. */
 805			if (!doneleft && !doneright) {
 806				useleft = pagino -
 807				 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
 808				  rec.ir_startino - pagino;
 809			} else {
 810				useleft = !doneleft;
 811			}
 812
 813			/* free inodes to the left? */
 814			if (useleft && trec.ir_freecount) {
 815				rec = trec;
 816				xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 817				cur = tcur;
 818
 819				pag->pagl_leftrec = trec.ir_startino;
 820				pag->pagl_rightrec = rec.ir_startino;
 821				pag->pagl_pagino = pagino;
 822				goto alloc_inode;
 823			}
 824
 825			/* free inodes to the right? */
 826			if (!useleft && rec.ir_freecount) {
 827				xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
 828
 829				pag->pagl_leftrec = trec.ir_startino;
 830				pag->pagl_rightrec = rec.ir_startino;
 831				pag->pagl_pagino = pagino;
 832				goto alloc_inode;
 833			}
 834
 835			/* get next record to check */
 836			if (useleft) {
 837				error = xfs_ialloc_next_rec(tcur, &trec,
 838								 &doneleft, 1);
 839			} else {
 840				error = xfs_ialloc_next_rec(cur, &rec,
 841								 &doneright, 0);
 842			}
 843			if (error)
 844				goto error1;
 845		}
 846
 847		/*
 848		 * We've reached the end of the btree. because
 849		 * we are only searching a small chunk of the
 850		 * btree each search, there is obviously free
 851		 * inodes closer to the parent inode than we
 852		 * are now. restart the search again.
 853		 */
 854		pag->pagl_pagino = NULLAGINO;
 855		pag->pagl_leftrec = NULLAGINO;
 856		pag->pagl_rightrec = NULLAGINO;
 857		xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
 858		xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 859		goto restart_pagno;
 860	}
 861
 862	/*
 863	 * In a different AG from the parent.
 864	 * See if the most recently allocated block has any free.
 865	 */
 866newino:
 867	if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
 868		error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
 869					 XFS_LOOKUP_EQ, &i);
 870		if (error)
 871			goto error0;
 872
 873		if (i == 1) {
 874			error = xfs_inobt_get_rec(cur, &rec, &j);
 875			if (error)
 876				goto error0;
 877
 878			if (j == 1 && rec.ir_freecount > 0) {
 879				/*
 880				 * The last chunk allocated in the group
 881				 * still has a free inode.
 882				 */
 883				goto alloc_inode;
 884			}
 885		}
 886	}
 887
 888	/*
 889	 * None left in the last group, search the whole AG
 890	 */
 891	error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
 892	if (error)
 893		goto error0;
 894	XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
 895
 896	for (;;) {
 897		error = xfs_inobt_get_rec(cur, &rec, &i);
 898		if (error)
 899			goto error0;
 900		XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
 901		if (rec.ir_freecount > 0)
 902			break;
 903		error = xfs_btree_increment(cur, 0, &i);
 904		if (error)
 905			goto error0;
 906		XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
 907	}
 908
 909alloc_inode:
 910	offset = xfs_lowbit64(rec.ir_free);
 911	ASSERT(offset >= 0);
 912	ASSERT(offset < XFS_INODES_PER_CHUNK);
 913	ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
 914				   XFS_INODES_PER_CHUNK) == 0);
 915	ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino + offset);
 916	rec.ir_free &= ~XFS_INOBT_MASK(offset);
 917	rec.ir_freecount--;
 918	error = xfs_inobt_update(cur, &rec);
 919	if (error)
 920		goto error0;
 921	be32_add_cpu(&agi->agi_freecount, -1);
 922	xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
 923	pag->pagi_freecount--;
 924
 925	error = xfs_check_agi_freecount(cur, agi);
 926	if (error)
 927		goto error0;
 928
 929	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 930	xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
 931	xfs_perag_put(pag);
 932	*inop = ino;
 933	return 0;
 934error1:
 935	xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
 936error0:
 937	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 938	xfs_perag_put(pag);
 939	return error;
 940}
 941
 942/*
 943 * Allocate an inode on disk.
 944 *
 945 * Mode is used to tell whether the new inode will need space, and whether it
 946 * is a directory.
 947 *
 948 * This function is designed to be called twice if it has to do an allocation
 949 * to make more free inodes.  On the first call, *IO_agbp should be set to NULL.
 950 * If an inode is available without having to performn an allocation, an inode
 951 * number is returned.  In this case, *IO_agbp is set to NULL.  If an allocation
 952 * needs to be done, xfs_dialloc returns the current AGI buffer in *IO_agbp.
 953 * The caller should then commit the current transaction, allocate a
 954 * new transaction, and call xfs_dialloc() again, passing in the previous value
 955 * of *IO_agbp.  IO_agbp should be held across the transactions. Since the AGI
 956 * buffer is locked across the two calls, the second call is guaranteed to have
 957 * a free inode available.
 958 *
 959 * Once we successfully pick an inode its number is returned and the on-disk
 960 * data structures are updated.  The inode itself is not read in, since doing so
 961 * would break ordering constraints with xfs_reclaim.
 962 */
 963int
 964xfs_dialloc(
 965	struct xfs_trans	*tp,
 966	xfs_ino_t		parent,
 967	umode_t			mode,
 968	int			okalloc,
 969	struct xfs_buf		**IO_agbp,
 970	xfs_ino_t		*inop)
 971{
 972	struct xfs_mount	*mp = tp->t_mountp;
 973	struct xfs_buf		*agbp;
 974	xfs_agnumber_t		agno;
 975	int			error;
 976	int			ialloced;
 977	int			noroom = 0;
 978	xfs_agnumber_t		start_agno;
 979	struct xfs_perag	*pag;
 980
 981	if (*IO_agbp) {
 982		/*
 983		 * If the caller passes in a pointer to the AGI buffer,
 984		 * continue where we left off before.  In this case, we
 985		 * know that the allocation group has free inodes.
 986		 */
 987		agbp = *IO_agbp;
 988		goto out_alloc;
 989	}
 990
 991	/*
 992	 * We do not have an agbp, so select an initial allocation
 993	 * group for inode allocation.
 994	 */
 995	start_agno = xfs_ialloc_ag_select(tp, parent, mode, okalloc);
 996	if (start_agno == NULLAGNUMBER) {
 997		*inop = NULLFSINO;
 998		return 0;
 999	}
1000
1001	/*
1002	 * If we have already hit the ceiling of inode blocks then clear
1003	 * okalloc so we scan all available agi structures for a free
1004	 * inode.
1005	 */
1006	if (mp->m_maxicount &&
1007	    mp->m_sb.sb_icount + mp->m_ialloc_inos > mp->m_maxicount) {
1008		noroom = 1;
1009		okalloc = 0;
1010	}
1011
1012	/*
1013	 * Loop until we find an allocation group that either has free inodes
1014	 * or in which we can allocate some inodes.  Iterate through the
1015	 * allocation groups upward, wrapping at the end.
1016	 */
1017	agno = start_agno;
1018	for (;;) {
1019		pag = xfs_perag_get(mp, agno);
1020		if (!pag->pagi_inodeok) {
1021			xfs_ialloc_next_ag(mp);
1022			goto nextag;
1023		}
1024
1025		if (!pag->pagi_init) {
1026			error = xfs_ialloc_pagi_init(mp, tp, agno);
1027			if (error)
1028				goto out_error;
1029		}
1030
1031		/*
1032		 * Do a first racy fast path check if this AG is usable.
1033		 */
1034		if (!pag->pagi_freecount && !okalloc)
1035			goto nextag;
1036
1037		/*
1038		 * Then read in the AGI buffer and recheck with the AGI buffer
1039		 * lock held.
1040		 */
1041		error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1042		if (error)
1043			goto out_error;
1044
1045		if (pag->pagi_freecount) {
1046			xfs_perag_put(pag);
1047			goto out_alloc;
1048		}
1049
1050		if (!okalloc)
1051			goto nextag_relse_buffer;
1052
1053
1054		error = xfs_ialloc_ag_alloc(tp, agbp, &ialloced);
1055		if (error) {
1056			xfs_trans_brelse(tp, agbp);
1057
1058			if (error != ENOSPC)
1059				goto out_error;
1060
1061			xfs_perag_put(pag);
1062			*inop = NULLFSINO;
1063			return 0;
1064		}
1065
1066		if (ialloced) {
1067			/*
1068			 * We successfully allocated some inodes, return
1069			 * the current context to the caller so that it
1070			 * can commit the current transaction and call
1071			 * us again where we left off.
1072			 */
1073			ASSERT(pag->pagi_freecount > 0);
1074			xfs_perag_put(pag);
1075
1076			*IO_agbp = agbp;
1077			*inop = NULLFSINO;
1078			return 0;
1079		}
1080
1081nextag_relse_buffer:
1082		xfs_trans_brelse(tp, agbp);
1083nextag:
1084		xfs_perag_put(pag);
1085		if (++agno == mp->m_sb.sb_agcount)
1086			agno = 0;
1087		if (agno == start_agno) {
1088			*inop = NULLFSINO;
1089			return noroom ? ENOSPC : 0;
1090		}
1091	}
1092
1093out_alloc:
1094	*IO_agbp = NULL;
1095	return xfs_dialloc_ag(tp, agbp, parent, inop);
1096out_error:
1097	xfs_perag_put(pag);
1098	return XFS_ERROR(error);
1099}
1100
1101/*
1102 * Free disk inode.  Carefully avoids touching the incore inode, all
1103 * manipulations incore are the caller's responsibility.
1104 * The on-disk inode is not changed by this operation, only the
1105 * btree (free inode mask) is changed.
1106 */
1107int
1108xfs_difree(
1109	xfs_trans_t	*tp,		/* transaction pointer */
1110	xfs_ino_t	inode,		/* inode to be freed */
1111	xfs_bmap_free_t	*flist,		/* extents to free */
1112	int		*delete,	/* set if inode cluster was deleted */
1113	xfs_ino_t	*first_ino)	/* first inode in deleted cluster */
1114{
1115	/* REFERENCED */
1116	xfs_agblock_t	agbno;	/* block number containing inode */
1117	xfs_buf_t	*agbp;	/* buffer containing allocation group header */
1118	xfs_agino_t	agino;	/* inode number relative to allocation group */
1119	xfs_agnumber_t	agno;	/* allocation group number */
1120	xfs_agi_t	*agi;	/* allocation group header */
1121	xfs_btree_cur_t	*cur;	/* inode btree cursor */
1122	int		error;	/* error return value */
1123	int		i;	/* result code */
1124	int		ilen;	/* inodes in an inode cluster */
1125	xfs_mount_t	*mp;	/* mount structure for filesystem */
1126	int		off;	/* offset of inode in inode chunk */
1127	xfs_inobt_rec_incore_t rec;	/* btree record */
1128	struct xfs_perag *pag;
1129
1130	mp = tp->t_mountp;
1131
1132	/*
1133	 * Break up inode number into its components.
1134	 */
1135	agno = XFS_INO_TO_AGNO(mp, inode);
1136	if (agno >= mp->m_sb.sb_agcount)  {
1137		xfs_warn(mp, "%s: agno >= mp->m_sb.sb_agcount (%d >= %d).",
1138			__func__, agno, mp->m_sb.sb_agcount);
1139		ASSERT(0);
1140		return XFS_ERROR(EINVAL);
1141	}
1142	agino = XFS_INO_TO_AGINO(mp, inode);
1143	if (inode != XFS_AGINO_TO_INO(mp, agno, agino))  {
1144		xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
1145			__func__, (unsigned long long)inode,
1146			(unsigned long long)XFS_AGINO_TO_INO(mp, agno, agino));
1147		ASSERT(0);
1148		return XFS_ERROR(EINVAL);
1149	}
1150	agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1151	if (agbno >= mp->m_sb.sb_agblocks)  {
1152		xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
1153			__func__, agbno, mp->m_sb.sb_agblocks);
1154		ASSERT(0);
1155		return XFS_ERROR(EINVAL);
1156	}
1157	/*
1158	 * Get the allocation group header.
1159	 */
1160	error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1161	if (error) {
1162		xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
1163			__func__, error);
1164		return error;
1165	}
1166	agi = XFS_BUF_TO_AGI(agbp);
1167	ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1168	ASSERT(agbno < be32_to_cpu(agi->agi_length));
1169	/*
1170	 * Initialize the cursor.
1171	 */
1172	cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1173
1174	error = xfs_check_agi_freecount(cur, agi);
1175	if (error)
1176		goto error0;
1177
1178	/*
1179	 * Look for the entry describing this inode.
1180	 */
1181	if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1182		xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1183			__func__, error);
1184		goto error0;
1185	}
1186	XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1187	error = xfs_inobt_get_rec(cur, &rec, &i);
1188	if (error) {
1189		xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1190			__func__, error);
1191		goto error0;
1192	}
1193	XFS_WANT_CORRUPTED_GOTO(i == 1, error0);
1194	/*
1195	 * Get the offset in the inode chunk.
1196	 */
1197	off = agino - rec.ir_startino;
1198	ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1199	ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1200	/*
1201	 * Mark the inode free & increment the count.
1202	 */
1203	rec.ir_free |= XFS_INOBT_MASK(off);
1204	rec.ir_freecount++;
1205
1206	/*
1207	 * When an inode cluster is free, it becomes eligible for removal
1208	 */
1209	if (!(mp->m_flags & XFS_MOUNT_IKEEP) &&
1210	    (rec.ir_freecount == mp->m_ialloc_inos)) {
1211
1212		*delete = 1;
1213		*first_ino = XFS_AGINO_TO_INO(mp, agno, rec.ir_startino);
1214
1215		/*
1216		 * Remove the inode cluster from the AGI B+Tree, adjust the
1217		 * AGI and Superblock inode counts, and mark the disk space
1218		 * to be freed when the transaction is committed.
1219		 */
1220		ilen = mp->m_ialloc_inos;
1221		be32_add_cpu(&agi->agi_count, -ilen);
1222		be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
1223		xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
1224		pag = xfs_perag_get(mp, agno);
1225		pag->pagi_freecount -= ilen - 1;
1226		xfs_perag_put(pag);
1227		xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
1228		xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
1229
1230		if ((error = xfs_btree_delete(cur, &i))) {
1231			xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
1232				__func__, error);
1233			goto error0;
1234		}
1235
1236		xfs_bmap_add_free(XFS_AGB_TO_FSB(mp, agno,
1237				  XFS_AGINO_TO_AGBNO(mp, rec.ir_startino)),
1238				  mp->m_ialloc_blks, flist, mp);
1239	} else {
1240		*delete = 0;
1241
1242		error = xfs_inobt_update(cur, &rec);
1243		if (error) {
1244			xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
1245				__func__, error);
1246			goto error0;
1247		}
1248
1249		/* 
1250		 * Change the inode free counts and log the ag/sb changes.
1251		 */
1252		be32_add_cpu(&agi->agi_freecount, 1);
1253		xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1254		pag = xfs_perag_get(mp, agno);
1255		pag->pagi_freecount++;
1256		xfs_perag_put(pag);
1257		xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
1258	}
1259
1260	error = xfs_check_agi_freecount(cur, agi);
1261	if (error)
1262		goto error0;
1263
1264	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1265	return 0;
1266
1267error0:
1268	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1269	return error;
1270}
1271
1272STATIC int
1273xfs_imap_lookup(
1274	struct xfs_mount	*mp,
1275	struct xfs_trans	*tp,
1276	xfs_agnumber_t		agno,
1277	xfs_agino_t		agino,
1278	xfs_agblock_t		agbno,
1279	xfs_agblock_t		*chunk_agbno,
1280	xfs_agblock_t		*offset_agbno,
1281	int			flags)
1282{
1283	struct xfs_inobt_rec_incore rec;
1284	struct xfs_btree_cur	*cur;
1285	struct xfs_buf		*agbp;
1286	int			error;
1287	int			i;
1288
1289	error = xfs_ialloc_read_agi(mp, tp, agno, &agbp);
1290	if (error) {
1291		xfs_alert(mp,
1292			"%s: xfs_ialloc_read_agi() returned error %d, agno %d",
1293			__func__, error, agno);
1294		return error;
1295	}
1296
1297	/*
1298	 * Lookup the inode record for the given agino. If the record cannot be
1299	 * found, then it's an invalid inode number and we should abort. Once
1300	 * we have a record, we need to ensure it contains the inode number
1301	 * we are looking up.
1302	 */
1303	cur = xfs_inobt_init_cursor(mp, tp, agbp, agno);
1304	error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
1305	if (!error) {
1306		if (i)
1307			error = xfs_inobt_get_rec(cur, &rec, &i);
1308		if (!error && i == 0)
1309			error = EINVAL;
1310	}
1311
1312	xfs_trans_brelse(tp, agbp);
1313	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1314	if (error)
1315		return error;
1316
1317	/* check that the returned record contains the required inode */
1318	if (rec.ir_startino > agino ||
1319	    rec.ir_startino + mp->m_ialloc_inos <= agino)
1320		return EINVAL;
1321
1322	/* for untrusted inodes check it is allocated first */
1323	if ((flags & XFS_IGET_UNTRUSTED) &&
1324	    (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
1325		return EINVAL;
1326
1327	*chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
1328	*offset_agbno = agbno - *chunk_agbno;
1329	return 0;
1330}
1331
1332/*
1333 * Return the location of the inode in imap, for mapping it into a buffer.
1334 */
1335int
1336xfs_imap(
1337	xfs_mount_t	 *mp,	/* file system mount structure */
1338	xfs_trans_t	 *tp,	/* transaction pointer */
1339	xfs_ino_t	ino,	/* inode to locate */
1340	struct xfs_imap	*imap,	/* location map structure */
1341	uint		flags)	/* flags for inode btree lookup */
1342{
1343	xfs_agblock_t	agbno;	/* block number of inode in the alloc group */
1344	xfs_agino_t	agino;	/* inode number within alloc group */
1345	xfs_agnumber_t	agno;	/* allocation group number */
1346	int		blks_per_cluster; /* num blocks per inode cluster */
1347	xfs_agblock_t	chunk_agbno;	/* first block in inode chunk */
1348	xfs_agblock_t	cluster_agbno;	/* first block in inode cluster */
1349	int		error;	/* error code */
1350	int		offset;	/* index of inode in its buffer */
1351	xfs_agblock_t	offset_agbno;	/* blks from chunk start to inode */
1352
1353	ASSERT(ino != NULLFSINO);
1354
1355	/*
1356	 * Split up the inode number into its parts.
1357	 */
1358	agno = XFS_INO_TO_AGNO(mp, ino);
1359	agino = XFS_INO_TO_AGINO(mp, ino);
1360	agbno = XFS_AGINO_TO_AGBNO(mp, agino);
1361	if (agno >= mp->m_sb.sb_agcount || agbno >= mp->m_sb.sb_agblocks ||
1362	    ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1363#ifdef DEBUG
1364		/*
1365		 * Don't output diagnostic information for untrusted inodes
1366		 * as they can be invalid without implying corruption.
1367		 */
1368		if (flags & XFS_IGET_UNTRUSTED)
1369			return XFS_ERROR(EINVAL);
1370		if (agno >= mp->m_sb.sb_agcount) {
1371			xfs_alert(mp,
1372				"%s: agno (%d) >= mp->m_sb.sb_agcount (%d)",
1373				__func__, agno, mp->m_sb.sb_agcount);
1374		}
1375		if (agbno >= mp->m_sb.sb_agblocks) {
1376			xfs_alert(mp,
1377		"%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
1378				__func__, (unsigned long long)agbno,
1379				(unsigned long)mp->m_sb.sb_agblocks);
1380		}
1381		if (ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
1382			xfs_alert(mp,
1383		"%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
1384				__func__, ino,
1385				XFS_AGINO_TO_INO(mp, agno, agino));
1386		}
1387		xfs_stack_trace();
1388#endif /* DEBUG */
1389		return XFS_ERROR(EINVAL);
1390	}
1391
1392	blks_per_cluster = xfs_icluster_size_fsb(mp);
1393
1394	/*
1395	 * For bulkstat and handle lookups, we have an untrusted inode number
1396	 * that we have to verify is valid. We cannot do this just by reading
1397	 * the inode buffer as it may have been unlinked and removed leaving
1398	 * inodes in stale state on disk. Hence we have to do a btree lookup
1399	 * in all cases where an untrusted inode number is passed.
1400	 */
1401	if (flags & XFS_IGET_UNTRUSTED) {
1402		error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1403					&chunk_agbno, &offset_agbno, flags);
1404		if (error)
1405			return error;
1406		goto out_map;
1407	}
1408
1409	/*
1410	 * If the inode cluster size is the same as the blocksize or
1411	 * smaller we get to the buffer by simple arithmetics.
1412	 */
1413	if (blks_per_cluster == 1) {
1414		offset = XFS_INO_TO_OFFSET(mp, ino);
1415		ASSERT(offset < mp->m_sb.sb_inopblock);
1416
1417		imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, agbno);
1418		imap->im_len = XFS_FSB_TO_BB(mp, 1);
1419		imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1420		return 0;
1421	}
1422
1423	/*
1424	 * If the inode chunks are aligned then use simple maths to
1425	 * find the location. Otherwise we have to do a btree
1426	 * lookup to find the location.
1427	 */
1428	if (mp->m_inoalign_mask) {
1429		offset_agbno = agbno & mp->m_inoalign_mask;
1430		chunk_agbno = agbno - offset_agbno;
1431	} else {
1432		error = xfs_imap_lookup(mp, tp, agno, agino, agbno,
1433					&chunk_agbno, &offset_agbno, flags);
1434		if (error)
1435			return error;
1436	}
1437
1438out_map:
1439	ASSERT(agbno >= chunk_agbno);
1440	cluster_agbno = chunk_agbno +
1441		((offset_agbno / blks_per_cluster) * blks_per_cluster);
1442	offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
1443		XFS_INO_TO_OFFSET(mp, ino);
1444
1445	imap->im_blkno = XFS_AGB_TO_DADDR(mp, agno, cluster_agbno);
1446	imap->im_len = XFS_FSB_TO_BB(mp, blks_per_cluster);
1447	imap->im_boffset = (ushort)(offset << mp->m_sb.sb_inodelog);
1448
1449	/*
1450	 * If the inode number maps to a block outside the bounds
1451	 * of the file system then return NULL rather than calling
1452	 * read_buf and panicing when we get an error from the
1453	 * driver.
1454	 */
1455	if ((imap->im_blkno + imap->im_len) >
1456	    XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
1457		xfs_alert(mp,
1458	"%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
1459			__func__, (unsigned long long) imap->im_blkno,
1460			(unsigned long long) imap->im_len,
1461			XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
1462		return XFS_ERROR(EINVAL);
1463	}
1464	return 0;
1465}
1466
1467/*
1468 * Compute and fill in value of m_in_maxlevels.
1469 */
1470void
1471xfs_ialloc_compute_maxlevels(
1472	xfs_mount_t	*mp)		/* file system mount structure */
1473{
1474	int		level;
1475	uint		maxblocks;
1476	uint		maxleafents;
1477	int		minleafrecs;
1478	int		minnoderecs;
1479
1480	maxleafents = (1LL << XFS_INO_AGINO_BITS(mp)) >>
1481		XFS_INODES_PER_CHUNK_LOG;
1482	minleafrecs = mp->m_alloc_mnr[0];
1483	minnoderecs = mp->m_alloc_mnr[1];
1484	maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
1485	for (level = 1; maxblocks > 1; level++)
1486		maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
1487	mp->m_in_maxlevels = level;
1488}
1489
1490/*
1491 * Log specified fields for the ag hdr (inode section)
1492 */
1493void
1494xfs_ialloc_log_agi(
1495	xfs_trans_t	*tp,		/* transaction pointer */
1496	xfs_buf_t	*bp,		/* allocation group header buffer */
1497	int		fields)		/* bitmask of fields to log */
1498{
1499	int			first;		/* first byte number */
1500	int			last;		/* last byte number */
1501	static const short	offsets[] = {	/* field starting offsets */
1502					/* keep in sync with bit definitions */
1503		offsetof(xfs_agi_t, agi_magicnum),
1504		offsetof(xfs_agi_t, agi_versionnum),
1505		offsetof(xfs_agi_t, agi_seqno),
1506		offsetof(xfs_agi_t, agi_length),
1507		offsetof(xfs_agi_t, agi_count),
1508		offsetof(xfs_agi_t, agi_root),
1509		offsetof(xfs_agi_t, agi_level),
1510		offsetof(xfs_agi_t, agi_freecount),
1511		offsetof(xfs_agi_t, agi_newino),
1512		offsetof(xfs_agi_t, agi_dirino),
1513		offsetof(xfs_agi_t, agi_unlinked),
1514		sizeof(xfs_agi_t)
1515	};
1516#ifdef DEBUG
1517	xfs_agi_t		*agi;	/* allocation group header */
1518
1519	agi = XFS_BUF_TO_AGI(bp);
1520	ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1521#endif
1522	/*
1523	 * Compute byte offsets for the first and last fields.
1524	 */
1525	xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS, &first, &last);
1526	/*
1527	 * Log the allocation group inode header buffer.
1528	 */
1529	xfs_trans_buf_set_type(tp, bp, XFS_BLFT_AGI_BUF);
1530	xfs_trans_log_buf(tp, bp, first, last);
1531}
1532
1533#ifdef DEBUG
1534STATIC void
1535xfs_check_agi_unlinked(
1536	struct xfs_agi		*agi)
1537{
1538	int			i;
1539
1540	for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++)
1541		ASSERT(agi->agi_unlinked[i]);
1542}
1543#else
1544#define xfs_check_agi_unlinked(agi)
1545#endif
1546
1547static bool
1548xfs_agi_verify(
1549	struct xfs_buf	*bp)
1550{
1551	struct xfs_mount *mp = bp->b_target->bt_mount;
1552	struct xfs_agi	*agi = XFS_BUF_TO_AGI(bp);
1553
1554	if (xfs_sb_version_hascrc(&mp->m_sb) &&
1555	    !uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_uuid))
1556			return false;
1557	/*
1558	 * Validate the magic number of the agi block.
1559	 */
1560	if (agi->agi_magicnum != cpu_to_be32(XFS_AGI_MAGIC))
1561		return false;
1562	if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)))
1563		return false;
1564
1565	/*
1566	 * during growfs operations, the perag is not fully initialised,
1567	 * so we can't use it for any useful checking. growfs ensures we can't
1568	 * use it by using uncached buffers that don't have the perag attached
1569	 * so we can detect and avoid this problem.
1570	 */
1571	if (bp->b_pag && be32_to_cpu(agi->agi_seqno) != bp->b_pag->pag_agno)
1572		return false;
1573
1574	xfs_check_agi_unlinked(agi);
1575	return true;
1576}
1577
1578static void
1579xfs_agi_read_verify(
1580	struct xfs_buf	*bp)
1581{
1582	struct xfs_mount *mp = bp->b_target->bt_mount;
1583
1584	if (xfs_sb_version_hascrc(&mp->m_sb) &&
1585	    !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF))
1586		xfs_buf_ioerror(bp, EFSBADCRC);
1587	else if (XFS_TEST_ERROR(!xfs_agi_verify(bp), mp,
1588				XFS_ERRTAG_IALLOC_READ_AGI,
1589				XFS_RANDOM_IALLOC_READ_AGI))
1590		xfs_buf_ioerror(bp, EFSCORRUPTED);
1591
1592	if (bp->b_error)
1593		xfs_verifier_error(bp);
1594}
1595
1596static void
1597xfs_agi_write_verify(
1598	struct xfs_buf	*bp)
1599{
1600	struct xfs_mount *mp = bp->b_target->bt_mount;
1601	struct xfs_buf_log_item	*bip = bp->b_fspriv;
1602
1603	if (!xfs_agi_verify(bp)) {
1604		xfs_buf_ioerror(bp, EFSCORRUPTED);
1605		xfs_verifier_error(bp);
1606		return;
1607	}
1608
1609	if (!xfs_sb_version_hascrc(&mp->m_sb))
1610		return;
1611
1612	if (bip)
1613		XFS_BUF_TO_AGI(bp)->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn);
1614	xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF);
1615}
1616
1617const struct xfs_buf_ops xfs_agi_buf_ops = {
1618	.verify_read = xfs_agi_read_verify,
1619	.verify_write = xfs_agi_write_verify,
1620};
1621
1622/*
1623 * Read in the allocation group header (inode allocation section)
1624 */
1625int
1626xfs_read_agi(
1627	struct xfs_mount	*mp,	/* file system mount structure */
1628	struct xfs_trans	*tp,	/* transaction pointer */
1629	xfs_agnumber_t		agno,	/* allocation group number */
1630	struct xfs_buf		**bpp)	/* allocation group hdr buf */
1631{
1632	int			error;
1633
1634	trace_xfs_read_agi(mp, agno);
1635
1636	ASSERT(agno != NULLAGNUMBER);
1637	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
1638			XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
1639			XFS_FSS_TO_BB(mp, 1), 0, bpp, &xfs_agi_buf_ops);
1640	if (error)
1641		return error;
1642
1643	ASSERT(!xfs_buf_geterror(*bpp));
1644	xfs_buf_set_ref(*bpp, XFS_AGI_REF);
1645	return 0;
1646}
1647
1648int
1649xfs_ialloc_read_agi(
1650	struct xfs_mount	*mp,	/* file system mount structure */
1651	struct xfs_trans	*tp,	/* transaction pointer */
1652	xfs_agnumber_t		agno,	/* allocation group number */
1653	struct xfs_buf		**bpp)	/* allocation group hdr buf */
1654{
1655	struct xfs_agi		*agi;	/* allocation group header */
1656	struct xfs_perag	*pag;	/* per allocation group data */
1657	int			error;
1658
1659	trace_xfs_ialloc_read_agi(mp, agno);
1660
1661	error = xfs_read_agi(mp, tp, agno, bpp);
1662	if (error)
1663		return error;
1664
1665	agi = XFS_BUF_TO_AGI(*bpp);
1666	pag = xfs_perag_get(mp, agno);
1667	if (!pag->pagi_init) {
1668		pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
1669		pag->pagi_count = be32_to_cpu(agi->agi_count);
1670		pag->pagi_init = 1;
1671	}
1672
1673	/*
1674	 * It's possible for these to be out of sync if
1675	 * we are in the middle of a forced shutdown.
1676	 */
1677	ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
1678		XFS_FORCED_SHUTDOWN(mp));
1679	xfs_perag_put(pag);
1680	return 0;
1681}
1682
1683/*
1684 * Read in the agi to initialise the per-ag data in the mount structure
1685 */
1686int
1687xfs_ialloc_pagi_init(
1688	xfs_mount_t	*mp,		/* file system mount structure */
1689	xfs_trans_t	*tp,		/* transaction pointer */
1690	xfs_agnumber_t	agno)		/* allocation group number */
1691{
1692	xfs_buf_t	*bp = NULL;
1693	int		error;
1694
1695	error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
1696	if (error)
1697		return error;
1698	if (bp)
1699		xfs_trans_brelse(tp, bp);
1700	return 0;
1701}