1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
   4 * All Rights Reserved.
   5 */
   6#include "xfs.h"
   7#include "xfs_fs.h"
   8#include "xfs_shared.h"
   9#include "xfs_format.h"
  10#include "xfs_log_format.h"
  11#include "xfs_trans_resv.h"
  12#include "xfs_bit.h"
  13#include "xfs_mount.h"
  14#include "xfs_inode.h"
  15#include "xfs_btree.h"
  16#include "xfs_ialloc.h"
  17#include "xfs_ialloc_btree.h"
  18#include "xfs_alloc.h"
  19#include "xfs_errortag.h"
  20#include "xfs_error.h"
  21#include "xfs_bmap.h"
  22#include "xfs_trans.h"
  23#include "xfs_buf_item.h"
  24#include "xfs_icreate_item.h"
  25#include "xfs_icache.h"
  26#include "xfs_trace.h"
  27#include "xfs_log.h"
  28#include "xfs_rmap.h"
  29#include "xfs_ag.h"
  30
  31/*
  32 * Lookup a record by ino in the btree given by cur.
  33 */
  34int					/* error */
  35xfs_inobt_lookup(
  36	struct xfs_btree_cur	*cur,	/* btree cursor */
  37	xfs_agino_t		ino,	/* starting inode of chunk */
  38	xfs_lookup_t		dir,	/* <=, >=, == */
  39	int			*stat)	/* success/failure */
  40{
  41	cur->bc_rec.i.ir_startino = ino;
  42	cur->bc_rec.i.ir_holemask = 0;
  43	cur->bc_rec.i.ir_count = 0;
  44	cur->bc_rec.i.ir_freecount = 0;
  45	cur->bc_rec.i.ir_free = 0;
  46	return xfs_btree_lookup(cur, dir, stat);
  47}
  48
  49/*
  50 * Update the record referred to by cur to the value given.
  51 * This either works (return 0) or gets an EFSCORRUPTED error.
  52 */
  53STATIC int				/* error */
  54xfs_inobt_update(
  55	struct xfs_btree_cur	*cur,	/* btree cursor */
  56	xfs_inobt_rec_incore_t	*irec)	/* btree record */
  57{
  58	union xfs_btree_rec	rec;
  59
  60	rec.inobt.ir_startino = cpu_to_be32(irec->ir_startino);
  61	if (xfs_has_sparseinodes(cur->bc_mp)) {
  62		rec.inobt.ir_u.sp.ir_holemask = cpu_to_be16(irec->ir_holemask);
  63		rec.inobt.ir_u.sp.ir_count = irec->ir_count;
  64		rec.inobt.ir_u.sp.ir_freecount = irec->ir_freecount;
  65	} else {
  66		/* ir_holemask/ir_count not supported on-disk */
  67		rec.inobt.ir_u.f.ir_freecount = cpu_to_be32(irec->ir_freecount);
  68	}
  69	rec.inobt.ir_free = cpu_to_be64(irec->ir_free);
  70	return xfs_btree_update(cur, &rec);
  71}
  72
  73/* Convert on-disk btree record to incore inobt record. */
  74void
  75xfs_inobt_btrec_to_irec(
  76	struct xfs_mount		*mp,
  77	const union xfs_btree_rec	*rec,
  78	struct xfs_inobt_rec_incore	*irec)
  79{
  80	irec->ir_startino = be32_to_cpu(rec->inobt.ir_startino);
  81	if (xfs_has_sparseinodes(mp)) {
  82		irec->ir_holemask = be16_to_cpu(rec->inobt.ir_u.sp.ir_holemask);
  83		irec->ir_count = rec->inobt.ir_u.sp.ir_count;
  84		irec->ir_freecount = rec->inobt.ir_u.sp.ir_freecount;
  85	} else {
  86		/*
  87		 * ir_holemask/ir_count not supported on-disk. Fill in hardcoded
  88		 * values for full inode chunks.
  89		 */
  90		irec->ir_holemask = XFS_INOBT_HOLEMASK_FULL;
  91		irec->ir_count = XFS_INODES_PER_CHUNK;
  92		irec->ir_freecount =
  93				be32_to_cpu(rec->inobt.ir_u.f.ir_freecount);
  94	}
  95	irec->ir_free = be64_to_cpu(rec->inobt.ir_free);
  96}
  97
  98/* Compute the freecount of an incore inode record. */
  99uint8_t
 100xfs_inobt_rec_freecount(
 101	const struct xfs_inobt_rec_incore	*irec)
 102{
 103	uint64_t				realfree = irec->ir_free;
 104
 105	if (xfs_inobt_issparse(irec->ir_holemask))
 106		realfree &= xfs_inobt_irec_to_allocmask(irec);
 107	return hweight64(realfree);
 108}
 109
 110/* Simple checks for inode records. */
 111xfs_failaddr_t
 112xfs_inobt_check_irec(
 113	struct xfs_perag			*pag,
 114	const struct xfs_inobt_rec_incore	*irec)
 115{
 116	/* Record has to be properly aligned within the AG. */
 117	if (!xfs_verify_agino(pag, irec->ir_startino))
 118		return __this_address;
 119	if (!xfs_verify_agino(pag,
 120				irec->ir_startino + XFS_INODES_PER_CHUNK - 1))
 121		return __this_address;
 122	if (irec->ir_count < XFS_INODES_PER_HOLEMASK_BIT ||
 123	    irec->ir_count > XFS_INODES_PER_CHUNK)
 124		return __this_address;
 125	if (irec->ir_freecount > XFS_INODES_PER_CHUNK)
 126		return __this_address;
 127
 128	if (xfs_inobt_rec_freecount(irec) != irec->ir_freecount)
 129		return __this_address;
 130
 131	return NULL;
 132}
 133
 134static inline int
 135xfs_inobt_complain_bad_rec(
 136	struct xfs_btree_cur		*cur,
 137	xfs_failaddr_t			fa,
 138	const struct xfs_inobt_rec_incore *irec)
 139{
 140	struct xfs_mount		*mp = cur->bc_mp;
 141
 142	xfs_warn(mp,
 143		"%s Inode BTree record corruption in AG %d detected at %pS!",
 144		cur->bc_btnum == XFS_BTNUM_INO ? "Used" : "Free",
 145		cur->bc_ag.pag->pag_agno, fa);
 146	xfs_warn(mp,
 147"start inode 0x%x, count 0x%x, free 0x%x freemask 0x%llx, holemask 0x%x",
 148		irec->ir_startino, irec->ir_count, irec->ir_freecount,
 149		irec->ir_free, irec->ir_holemask);
 150	return -EFSCORRUPTED;
 151}
 152
 153/*
 154 * Get the data from the pointed-to record.
 155 */
 156int
 157xfs_inobt_get_rec(
 158	struct xfs_btree_cur		*cur,
 159	struct xfs_inobt_rec_incore	*irec,
 160	int				*stat)
 161{
 162	struct xfs_mount		*mp = cur->bc_mp;
 163	union xfs_btree_rec		*rec;
 164	xfs_failaddr_t			fa;
 165	int				error;
 166
 167	error = xfs_btree_get_rec(cur, &rec, stat);
 168	if (error || *stat == 0)
 169		return error;
 170
 171	xfs_inobt_btrec_to_irec(mp, rec, irec);
 172	fa = xfs_inobt_check_irec(cur->bc_ag.pag, irec);
 173	if (fa)
 174		return xfs_inobt_complain_bad_rec(cur, fa, irec);
 175
 176	return 0;
 177}
 178
 179/*
 180 * Insert a single inobt record. Cursor must already point to desired location.
 181 */
 182int
 183xfs_inobt_insert_rec(
 184	struct xfs_btree_cur	*cur,
 185	uint16_t		holemask,
 186	uint8_t			count,
 187	int32_t			freecount,
 188	xfs_inofree_t		free,
 189	int			*stat)
 190{
 191	cur->bc_rec.i.ir_holemask = holemask;
 192	cur->bc_rec.i.ir_count = count;
 193	cur->bc_rec.i.ir_freecount = freecount;
 194	cur->bc_rec.i.ir_free = free;
 195	return xfs_btree_insert(cur, stat);
 196}
 197
 198/*
 199 * Insert records describing a newly allocated inode chunk into the inobt.
 200 */
 201STATIC int
 202xfs_inobt_insert(
 203	struct xfs_perag	*pag,
 204	struct xfs_trans	*tp,
 205	struct xfs_buf		*agbp,
 206	xfs_agino_t		newino,
 207	xfs_agino_t		newlen,
 208	xfs_btnum_t		btnum)
 209{
 210	struct xfs_btree_cur	*cur;
 211	xfs_agino_t		thisino;
 212	int			i;
 213	int			error;
 214
 215	cur = xfs_inobt_init_cursor(pag, tp, agbp, btnum);
 216
 217	for (thisino = newino;
 218	     thisino < newino + newlen;
 219	     thisino += XFS_INODES_PER_CHUNK) {
 220		error = xfs_inobt_lookup(cur, thisino, XFS_LOOKUP_EQ, &i);
 221		if (error) {
 222			xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 223			return error;
 224		}
 225		ASSERT(i == 0);
 226
 227		error = xfs_inobt_insert_rec(cur, XFS_INOBT_HOLEMASK_FULL,
 228					     XFS_INODES_PER_CHUNK,
 229					     XFS_INODES_PER_CHUNK,
 230					     XFS_INOBT_ALL_FREE, &i);
 231		if (error) {
 232			xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 233			return error;
 234		}
 235		ASSERT(i == 1);
 236	}
 237
 238	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 239
 240	return 0;
 241}
 242
 243/*
 244 * Verify that the number of free inodes in the AGI is correct.
 245 */
 246#ifdef DEBUG
 247static int
 248xfs_check_agi_freecount(
 249	struct xfs_btree_cur	*cur)
 250{
 251	if (cur->bc_nlevels == 1) {
 252		xfs_inobt_rec_incore_t rec;
 253		int		freecount = 0;
 254		int		error;
 255		int		i;
 256
 257		error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
 258		if (error)
 259			return error;
 260
 261		do {
 262			error = xfs_inobt_get_rec(cur, &rec, &i);
 263			if (error)
 264				return error;
 265
 266			if (i) {
 267				freecount += rec.ir_freecount;
 268				error = xfs_btree_increment(cur, 0, &i);
 269				if (error)
 270					return error;
 271			}
 272		} while (i == 1);
 273
 274		if (!xfs_is_shutdown(cur->bc_mp))
 275			ASSERT(freecount == cur->bc_ag.pag->pagi_freecount);
 276	}
 277	return 0;
 278}
 279#else
 280#define xfs_check_agi_freecount(cur)	0
 281#endif
 282
 283/*
 284 * Initialise a new set of inodes. When called without a transaction context
 285 * (e.g. from recovery) we initiate a delayed write of the inode buffers rather
 286 * than logging them (which in a transaction context puts them into the AIL
 287 * for writeback rather than the xfsbufd queue).
 288 */
 289int
 290xfs_ialloc_inode_init(
 291	struct xfs_mount	*mp,
 292	struct xfs_trans	*tp,
 293	struct list_head	*buffer_list,
 294	int			icount,
 295	xfs_agnumber_t		agno,
 296	xfs_agblock_t		agbno,
 297	xfs_agblock_t		length,
 298	unsigned int		gen)
 299{
 300	struct xfs_buf		*fbuf;
 301	struct xfs_dinode	*free;
 302	int			nbufs;
 303	int			version;
 304	int			i, j;
 305	xfs_daddr_t		d;
 306	xfs_ino_t		ino = 0;
 307	int			error;
 308
 309	/*
 310	 * Loop over the new block(s), filling in the inodes.  For small block
 311	 * sizes, manipulate the inodes in buffers  which are multiples of the
 312	 * blocks size.
 313	 */
 314	nbufs = length / M_IGEO(mp)->blocks_per_cluster;
 315
 316	/*
 317	 * Figure out what version number to use in the inodes we create.  If
 318	 * the superblock version has caught up to the one that supports the new
 319	 * inode format, then use the new inode version.  Otherwise use the old
 320	 * version so that old kernels will continue to be able to use the file
 321	 * system.
 322	 *
 323	 * For v3 inodes, we also need to write the inode number into the inode,
 324	 * so calculate the first inode number of the chunk here as
 325	 * XFS_AGB_TO_AGINO() only works within a filesystem block, not
 326	 * across multiple filesystem blocks (such as a cluster) and so cannot
 327	 * be used in the cluster buffer loop below.
 328	 *
 329	 * Further, because we are writing the inode directly into the buffer
 330	 * and calculating a CRC on the entire inode, we have ot log the entire
 331	 * inode so that the entire range the CRC covers is present in the log.
 332	 * That means for v3 inode we log the entire buffer rather than just the
 333	 * inode cores.
 334	 */
 335	if (xfs_has_v3inodes(mp)) {
 336		version = 3;
 337		ino = XFS_AGINO_TO_INO(mp, agno, XFS_AGB_TO_AGINO(mp, agbno));
 338
 339		/*
 340		 * log the initialisation that is about to take place as an
 341		 * logical operation. This means the transaction does not
 342		 * need to log the physical changes to the inode buffers as log
 343		 * recovery will know what initialisation is actually needed.
 344		 * Hence we only need to log the buffers as "ordered" buffers so
 345		 * they track in the AIL as if they were physically logged.
 346		 */
 347		if (tp)
 348			xfs_icreate_log(tp, agno, agbno, icount,
 349					mp->m_sb.sb_inodesize, length, gen);
 350	} else
 351		version = 2;
 352
 353	for (j = 0; j < nbufs; j++) {
 354		/*
 355		 * Get the block.
 356		 */
 357		d = XFS_AGB_TO_DADDR(mp, agno, agbno +
 358				(j * M_IGEO(mp)->blocks_per_cluster));
 359		error = xfs_trans_get_buf(tp, mp->m_ddev_targp, d,
 360				mp->m_bsize * M_IGEO(mp)->blocks_per_cluster,
 361				XBF_UNMAPPED, &fbuf);
 362		if (error)
 363			return error;
 364
 365		/* Initialize the inode buffers and log them appropriately. */
 366		fbuf->b_ops = &xfs_inode_buf_ops;
 367		xfs_buf_zero(fbuf, 0, BBTOB(fbuf->b_length));
 368		for (i = 0; i < M_IGEO(mp)->inodes_per_cluster; i++) {
 369			int	ioffset = i << mp->m_sb.sb_inodelog;
 370
 371			free = xfs_make_iptr(mp, fbuf, i);
 372			free->di_magic = cpu_to_be16(XFS_DINODE_MAGIC);
 373			free->di_version = version;
 374			free->di_gen = cpu_to_be32(gen);
 375			free->di_next_unlinked = cpu_to_be32(NULLAGINO);
 376
 377			if (version == 3) {
 378				free->di_ino = cpu_to_be64(ino);
 379				ino++;
 380				uuid_copy(&free->di_uuid,
 381					  &mp->m_sb.sb_meta_uuid);
 382				xfs_dinode_calc_crc(mp, free);
 383			} else if (tp) {
 384				/* just log the inode core */
 385				xfs_trans_log_buf(tp, fbuf, ioffset,
 386					  ioffset + XFS_DINODE_SIZE(mp) - 1);
 387			}
 388		}
 389
 390		if (tp) {
 391			/*
 392			 * Mark the buffer as an inode allocation buffer so it
 393			 * sticks in AIL at the point of this allocation
 394			 * transaction. This ensures the they are on disk before
 395			 * the tail of the log can be moved past this
 396			 * transaction (i.e. by preventing relogging from moving
 397			 * it forward in the log).
 398			 */
 399			xfs_trans_inode_alloc_buf(tp, fbuf);
 400			if (version == 3) {
 401				/*
 402				 * Mark the buffer as ordered so that they are
 403				 * not physically logged in the transaction but
 404				 * still tracked in the AIL as part of the
 405				 * transaction and pin the log appropriately.
 406				 */
 407				xfs_trans_ordered_buf(tp, fbuf);
 408			}
 409		} else {
 410			fbuf->b_flags |= XBF_DONE;
 411			xfs_buf_delwri_queue(fbuf, buffer_list);
 412			xfs_buf_relse(fbuf);
 413		}
 414	}
 415	return 0;
 416}
 417
 418/*
 419 * Align startino and allocmask for a recently allocated sparse chunk such that
 420 * they are fit for insertion (or merge) into the on-disk inode btrees.
 421 *
 422 * Background:
 423 *
 424 * When enabled, sparse inode support increases the inode alignment from cluster
 425 * size to inode chunk size. This means that the minimum range between two
 426 * non-adjacent inode records in the inobt is large enough for a full inode
 427 * record. This allows for cluster sized, cluster aligned block allocation
 428 * without need to worry about whether the resulting inode record overlaps with
 429 * another record in the tree. Without this basic rule, we would have to deal
 430 * with the consequences of overlap by potentially undoing recent allocations in
 431 * the inode allocation codepath.
 432 *
 433 * Because of this alignment rule (which is enforced on mount), there are two
 434 * inobt possibilities for newly allocated sparse chunks. One is that the
 435 * aligned inode record for the chunk covers a range of inodes not already
 436 * covered in the inobt (i.e., it is safe to insert a new sparse record). The
 437 * other is that a record already exists at the aligned startino that considers
 438 * the newly allocated range as sparse. In the latter case, record content is
 439 * merged in hope that sparse inode chunks fill to full chunks over time.
 440 */
 441STATIC void
 442xfs_align_sparse_ino(
 443	struct xfs_mount		*mp,
 444	xfs_agino_t			*startino,
 445	uint16_t			*allocmask)
 446{
 447	xfs_agblock_t			agbno;
 448	xfs_agblock_t			mod;
 449	int				offset;
 450
 451	agbno = XFS_AGINO_TO_AGBNO(mp, *startino);
 452	mod = agbno % mp->m_sb.sb_inoalignmt;
 453	if (!mod)
 454		return;
 455
 456	/* calculate the inode offset and align startino */
 457	offset = XFS_AGB_TO_AGINO(mp, mod);
 458	*startino -= offset;
 459
 460	/*
 461	 * Since startino has been aligned down, left shift allocmask such that
 462	 * it continues to represent the same physical inodes relative to the
 463	 * new startino.
 464	 */
 465	*allocmask <<= offset / XFS_INODES_PER_HOLEMASK_BIT;
 466}
 467
 468/*
 469 * Determine whether the source inode record can merge into the target. Both
 470 * records must be sparse, the inode ranges must match and there must be no
 471 * allocation overlap between the records.
 472 */
 473STATIC bool
 474__xfs_inobt_can_merge(
 475	struct xfs_inobt_rec_incore	*trec,	/* tgt record */
 476	struct xfs_inobt_rec_incore	*srec)	/* src record */
 477{
 478	uint64_t			talloc;
 479	uint64_t			salloc;
 480
 481	/* records must cover the same inode range */
 482	if (trec->ir_startino != srec->ir_startino)
 483		return false;
 484
 485	/* both records must be sparse */
 486	if (!xfs_inobt_issparse(trec->ir_holemask) ||
 487	    !xfs_inobt_issparse(srec->ir_holemask))
 488		return false;
 489
 490	/* both records must track some inodes */
 491	if (!trec->ir_count || !srec->ir_count)
 492		return false;
 493
 494	/* can't exceed capacity of a full record */
 495	if (trec->ir_count + srec->ir_count > XFS_INODES_PER_CHUNK)
 496		return false;
 497
 498	/* verify there is no allocation overlap */
 499	talloc = xfs_inobt_irec_to_allocmask(trec);
 500	salloc = xfs_inobt_irec_to_allocmask(srec);
 501	if (talloc & salloc)
 502		return false;
 503
 504	return true;
 505}
 506
 507/*
 508 * Merge the source inode record into the target. The caller must call
 509 * __xfs_inobt_can_merge() to ensure the merge is valid.
 510 */
 511STATIC void
 512__xfs_inobt_rec_merge(
 513	struct xfs_inobt_rec_incore	*trec,	/* target */
 514	struct xfs_inobt_rec_incore	*srec)	/* src */
 515{
 516	ASSERT(trec->ir_startino == srec->ir_startino);
 517
 518	/* combine the counts */
 519	trec->ir_count += srec->ir_count;
 520	trec->ir_freecount += srec->ir_freecount;
 521
 522	/*
 523	 * Merge the holemask and free mask. For both fields, 0 bits refer to
 524	 * allocated inodes. We combine the allocated ranges with bitwise AND.
 525	 */
 526	trec->ir_holemask &= srec->ir_holemask;
 527	trec->ir_free &= srec->ir_free;
 528}
 529
 530/*
 531 * Insert a new sparse inode chunk into the associated inode btree. The inode
 532 * record for the sparse chunk is pre-aligned to a startino that should match
 533 * any pre-existing sparse inode record in the tree. This allows sparse chunks
 534 * to fill over time.
 535 *
 536 * This function supports two modes of handling preexisting records depending on
 537 * the merge flag. If merge is true, the provided record is merged with the
 538 * existing record and updated in place. The merged record is returned in nrec.
 539 * If merge is false, an existing record is replaced with the provided record.
 540 * If no preexisting record exists, the provided record is always inserted.
 541 *
 542 * It is considered corruption if a merge is requested and not possible. Given
 543 * the sparse inode alignment constraints, this should never happen.
 544 */
 545STATIC int
 546xfs_inobt_insert_sprec(
 547	struct xfs_perag		*pag,
 548	struct xfs_trans		*tp,
 549	struct xfs_buf			*agbp,
 550	int				btnum,
 551	struct xfs_inobt_rec_incore	*nrec,	/* in/out: new/merged rec. */
 552	bool				merge)	/* merge or replace */
 553{
 554	struct xfs_mount		*mp = pag->pag_mount;
 555	struct xfs_btree_cur		*cur;
 556	int				error;
 557	int				i;
 558	struct xfs_inobt_rec_incore	rec;
 559
 560	cur = xfs_inobt_init_cursor(pag, tp, agbp, btnum);
 561
 562	/* the new record is pre-aligned so we know where to look */
 563	error = xfs_inobt_lookup(cur, nrec->ir_startino, XFS_LOOKUP_EQ, &i);
 564	if (error)
 565		goto error;
 566	/* if nothing there, insert a new record and return */
 567	if (i == 0) {
 568		error = xfs_inobt_insert_rec(cur, nrec->ir_holemask,
 569					     nrec->ir_count, nrec->ir_freecount,
 570					     nrec->ir_free, &i);
 571		if (error)
 572			goto error;
 573		if (XFS_IS_CORRUPT(mp, i != 1)) {
 574			error = -EFSCORRUPTED;
 575			goto error;
 576		}
 577
 578		goto out;
 579	}
 580
 581	/*
 582	 * A record exists at this startino. Merge or replace the record
 583	 * depending on what we've been asked to do.
 584	 */
 585	if (merge) {
 586		error = xfs_inobt_get_rec(cur, &rec, &i);
 587		if (error)
 588			goto error;
 589		if (XFS_IS_CORRUPT(mp, i != 1)) {
 590			error = -EFSCORRUPTED;
 591			goto error;
 592		}
 593		if (XFS_IS_CORRUPT(mp, rec.ir_startino != nrec->ir_startino)) {
 594			error = -EFSCORRUPTED;
 595			goto error;
 596		}
 597
 598		/*
 599		 * This should never fail. If we have coexisting records that
 600		 * cannot merge, something is seriously wrong.
 601		 */
 602		if (XFS_IS_CORRUPT(mp, !__xfs_inobt_can_merge(nrec, &rec))) {
 603			error = -EFSCORRUPTED;
 604			goto error;
 605		}
 606
 607		trace_xfs_irec_merge_pre(mp, pag->pag_agno, rec.ir_startino,
 608					 rec.ir_holemask, nrec->ir_startino,
 609					 nrec->ir_holemask);
 610
 611		/* merge to nrec to output the updated record */
 612		__xfs_inobt_rec_merge(nrec, &rec);
 613
 614		trace_xfs_irec_merge_post(mp, pag->pag_agno, nrec->ir_startino,
 615					  nrec->ir_holemask);
 616
 617		error = xfs_inobt_rec_check_count(mp, nrec);
 618		if (error)
 619			goto error;
 620	}
 621
 622	error = xfs_inobt_update(cur, nrec);
 623	if (error)
 624		goto error;
 625
 626out:
 627	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 628	return 0;
 629error:
 630	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 631	return error;
 632}
 633
 634/*
 635 * Allocate new inodes in the allocation group specified by agbp.  Returns 0 if
 636 * inodes were allocated in this AG; -EAGAIN if there was no space in this AG so
 637 * the caller knows it can try another AG, a hard -ENOSPC when over the maximum
 638 * inode count threshold, or the usual negative error code for other errors.
 639 */
 640STATIC int
 641xfs_ialloc_ag_alloc(
 642	struct xfs_perag	*pag,
 643	struct xfs_trans	*tp,
 644	struct xfs_buf		*agbp)
 645{
 646	struct xfs_agi		*agi;
 647	struct xfs_alloc_arg	args;
 648	int			error;
 649	xfs_agino_t		newino;		/* new first inode's number */
 650	xfs_agino_t		newlen;		/* new number of inodes */
 651	int			isaligned = 0;	/* inode allocation at stripe */
 652						/* unit boundary */
 653	/* init. to full chunk */
 654	struct xfs_inobt_rec_incore rec;
 655	struct xfs_ino_geometry	*igeo = M_IGEO(tp->t_mountp);
 656	uint16_t		allocmask = (uint16_t) -1;
 657	int			do_sparse = 0;
 658
 659	memset(&args, 0, sizeof(args));
 660	args.tp = tp;
 661	args.mp = tp->t_mountp;
 662	args.fsbno = NULLFSBLOCK;
 663	args.oinfo = XFS_RMAP_OINFO_INODES;
 664	args.pag = pag;
 665
 666#ifdef DEBUG
 667	/* randomly do sparse inode allocations */
 668	if (xfs_has_sparseinodes(tp->t_mountp) &&
 669	    igeo->ialloc_min_blks < igeo->ialloc_blks)
 670		do_sparse = get_random_u32_below(2);
 671#endif
 672
 673	/*
 674	 * Locking will ensure that we don't have two callers in here
 675	 * at one time.
 676	 */
 677	newlen = igeo->ialloc_inos;
 678	if (igeo->maxicount &&
 679	    percpu_counter_read_positive(&args.mp->m_icount) + newlen >
 680							igeo->maxicount)
 681		return -ENOSPC;
 682	args.minlen = args.maxlen = igeo->ialloc_blks;
 683	/*
 684	 * First try to allocate inodes contiguous with the last-allocated
 685	 * chunk of inodes.  If the filesystem is striped, this will fill
 686	 * an entire stripe unit with inodes.
 687	 */
 688	agi = agbp->b_addr;
 689	newino = be32_to_cpu(agi->agi_newino);
 690	args.agbno = XFS_AGINO_TO_AGBNO(args.mp, newino) +
 691		     igeo->ialloc_blks;
 692	if (do_sparse)
 693		goto sparse_alloc;
 694	if (likely(newino != NULLAGINO &&
 695		  (args.agbno < be32_to_cpu(agi->agi_length)))) {
 696		args.prod = 1;
 697
 698		/*
 699		 * We need to take into account alignment here to ensure that
 700		 * we don't modify the free list if we fail to have an exact
 701		 * block. If we don't have an exact match, and every oher
 702		 * attempt allocation attempt fails, we'll end up cancelling
 703		 * a dirty transaction and shutting down.
 704		 *
 705		 * For an exact allocation, alignment must be 1,
 706		 * however we need to take cluster alignment into account when
 707		 * fixing up the freelist. Use the minalignslop field to
 708		 * indicate that extra blocks might be required for alignment,
 709		 * but not to use them in the actual exact allocation.
 710		 */
 711		args.alignment = 1;
 712		args.minalignslop = igeo->cluster_align - 1;
 713
 714		/* Allow space for the inode btree to split. */
 715		args.minleft = igeo->inobt_maxlevels;
 716		error = xfs_alloc_vextent_exact_bno(&args,
 717				XFS_AGB_TO_FSB(args.mp, pag->pag_agno,
 718						args.agbno));
 719		if (error)
 720			return error;
 721
 722		/*
 723		 * This request might have dirtied the transaction if the AG can
 724		 * satisfy the request, but the exact block was not available.
 725		 * If the allocation did fail, subsequent requests will relax
 726		 * the exact agbno requirement and increase the alignment
 727		 * instead. It is critical that the total size of the request
 728		 * (len + alignment + slop) does not increase from this point
 729		 * on, so reset minalignslop to ensure it is not included in
 730		 * subsequent requests.
 731		 */
 732		args.minalignslop = 0;
 733	}
 734
 735	if (unlikely(args.fsbno == NULLFSBLOCK)) {
 736		/*
 737		 * Set the alignment for the allocation.
 738		 * If stripe alignment is turned on then align at stripe unit
 739		 * boundary.
 740		 * If the cluster size is smaller than a filesystem block
 741		 * then we're doing I/O for inodes in filesystem block size
 742		 * pieces, so don't need alignment anyway.
 743		 */
 744		isaligned = 0;
 745		if (igeo->ialloc_align) {
 746			ASSERT(!xfs_has_noalign(args.mp));
 747			args.alignment = args.mp->m_dalign;
 748			isaligned = 1;
 749		} else
 750			args.alignment = igeo->cluster_align;
 751		/*
 752		 * Allocate a fixed-size extent of inodes.
 753		 */
 754		args.prod = 1;
 755		/*
 756		 * Allow space for the inode btree to split.
 757		 */
 758		args.minleft = igeo->inobt_maxlevels;
 759		error = xfs_alloc_vextent_near_bno(&args,
 760				XFS_AGB_TO_FSB(args.mp, pag->pag_agno,
 761						be32_to_cpu(agi->agi_root)));
 762		if (error)
 763			return error;
 764	}
 765
 766	/*
 767	 * If stripe alignment is turned on, then try again with cluster
 768	 * alignment.
 769	 */
 770	if (isaligned && args.fsbno == NULLFSBLOCK) {
 771		args.alignment = igeo->cluster_align;
 772		error = xfs_alloc_vextent_near_bno(&args,
 773				XFS_AGB_TO_FSB(args.mp, pag->pag_agno,
 774						be32_to_cpu(agi->agi_root)));
 775		if (error)
 776			return error;
 777	}
 778
 779	/*
 780	 * Finally, try a sparse allocation if the filesystem supports it and
 781	 * the sparse allocation length is smaller than a full chunk.
 782	 */
 783	if (xfs_has_sparseinodes(args.mp) &&
 784	    igeo->ialloc_min_blks < igeo->ialloc_blks &&
 785	    args.fsbno == NULLFSBLOCK) {
 786sparse_alloc:
 787		args.alignment = args.mp->m_sb.sb_spino_align;
 788		args.prod = 1;
 789
 790		args.minlen = igeo->ialloc_min_blks;
 791		args.maxlen = args.minlen;
 792
 793		/*
 794		 * The inode record will be aligned to full chunk size. We must
 795		 * prevent sparse allocation from AG boundaries that result in
 796		 * invalid inode records, such as records that start at agbno 0
 797		 * or extend beyond the AG.
 798		 *
 799		 * Set min agbno to the first aligned, non-zero agbno and max to
 800		 * the last aligned agbno that is at least one full chunk from
 801		 * the end of the AG.
 802		 */
 803		args.min_agbno = args.mp->m_sb.sb_inoalignmt;
 804		args.max_agbno = round_down(args.mp->m_sb.sb_agblocks,
 805					    args.mp->m_sb.sb_inoalignmt) -
 806				 igeo->ialloc_blks;
 807
 808		error = xfs_alloc_vextent_near_bno(&args,
 809				XFS_AGB_TO_FSB(args.mp, pag->pag_agno,
 810						be32_to_cpu(agi->agi_root)));
 811		if (error)
 812			return error;
 813
 814		newlen = XFS_AGB_TO_AGINO(args.mp, args.len);
 815		ASSERT(newlen <= XFS_INODES_PER_CHUNK);
 816		allocmask = (1 << (newlen / XFS_INODES_PER_HOLEMASK_BIT)) - 1;
 817	}
 818
 819	if (args.fsbno == NULLFSBLOCK)
 820		return -EAGAIN;
 821
 822	ASSERT(args.len == args.minlen);
 823
 824	/*
 825	 * Stamp and write the inode buffers.
 826	 *
 827	 * Seed the new inode cluster with a random generation number. This
 828	 * prevents short-term reuse of generation numbers if a chunk is
 829	 * freed and then immediately reallocated. We use random numbers
 830	 * rather than a linear progression to prevent the next generation
 831	 * number from being easily guessable.
 832	 */
 833	error = xfs_ialloc_inode_init(args.mp, tp, NULL, newlen, pag->pag_agno,
 834			args.agbno, args.len, get_random_u32());
 835
 836	if (error)
 837		return error;
 838	/*
 839	 * Convert the results.
 840	 */
 841	newino = XFS_AGB_TO_AGINO(args.mp, args.agbno);
 842
 843	if (xfs_inobt_issparse(~allocmask)) {
 844		/*
 845		 * We've allocated a sparse chunk. Align the startino and mask.
 846		 */
 847		xfs_align_sparse_ino(args.mp, &newino, &allocmask);
 848
 849		rec.ir_startino = newino;
 850		rec.ir_holemask = ~allocmask;
 851		rec.ir_count = newlen;
 852		rec.ir_freecount = newlen;
 853		rec.ir_free = XFS_INOBT_ALL_FREE;
 854
 855		/*
 856		 * Insert the sparse record into the inobt and allow for a merge
 857		 * if necessary. If a merge does occur, rec is updated to the
 858		 * merged record.
 859		 */
 860		error = xfs_inobt_insert_sprec(pag, tp, agbp,
 861				XFS_BTNUM_INO, &rec, true);
 862		if (error == -EFSCORRUPTED) {
 863			xfs_alert(args.mp,
 864	"invalid sparse inode record: ino 0x%llx holemask 0x%x count %u",
 865				  XFS_AGINO_TO_INO(args.mp, pag->pag_agno,
 866						   rec.ir_startino),
 867				  rec.ir_holemask, rec.ir_count);
 868			xfs_force_shutdown(args.mp, SHUTDOWN_CORRUPT_INCORE);
 869		}
 870		if (error)
 871			return error;
 872
 873		/*
 874		 * We can't merge the part we've just allocated as for the inobt
 875		 * due to finobt semantics. The original record may or may not
 876		 * exist independent of whether physical inodes exist in this
 877		 * sparse chunk.
 878		 *
 879		 * We must update the finobt record based on the inobt record.
 880		 * rec contains the fully merged and up to date inobt record
 881		 * from the previous call. Set merge false to replace any
 882		 * existing record with this one.
 883		 */
 884		if (xfs_has_finobt(args.mp)) {
 885			error = xfs_inobt_insert_sprec(pag, tp, agbp,
 886				       XFS_BTNUM_FINO, &rec, false);
 887			if (error)
 888				return error;
 889		}
 890	} else {
 891		/* full chunk - insert new records to both btrees */
 892		error = xfs_inobt_insert(pag, tp, agbp, newino, newlen,
 893					 XFS_BTNUM_INO);
 894		if (error)
 895			return error;
 896
 897		if (xfs_has_finobt(args.mp)) {
 898			error = xfs_inobt_insert(pag, tp, agbp, newino,
 899						 newlen, XFS_BTNUM_FINO);
 900			if (error)
 901				return error;
 902		}
 903	}
 904
 905	/*
 906	 * Update AGI counts and newino.
 907	 */
 908	be32_add_cpu(&agi->agi_count, newlen);
 909	be32_add_cpu(&agi->agi_freecount, newlen);
 910	pag->pagi_freecount += newlen;
 911	pag->pagi_count += newlen;
 912	agi->agi_newino = cpu_to_be32(newino);
 913
 914	/*
 915	 * Log allocation group header fields
 916	 */
 917	xfs_ialloc_log_agi(tp, agbp,
 918		XFS_AGI_COUNT | XFS_AGI_FREECOUNT | XFS_AGI_NEWINO);
 919	/*
 920	 * Modify/log superblock values for inode count and inode free count.
 921	 */
 922	xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, (long)newlen);
 923	xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, (long)newlen);
 924	return 0;
 925}
 926
 927/*
 928 * Try to retrieve the next record to the left/right from the current one.
 929 */
 930STATIC int
 931xfs_ialloc_next_rec(
 932	struct xfs_btree_cur	*cur,
 933	xfs_inobt_rec_incore_t	*rec,
 934	int			*done,
 935	int			left)
 936{
 937	int                     error;
 938	int			i;
 939
 940	if (left)
 941		error = xfs_btree_decrement(cur, 0, &i);
 942	else
 943		error = xfs_btree_increment(cur, 0, &i);
 944
 945	if (error)
 946		return error;
 947	*done = !i;
 948	if (i) {
 949		error = xfs_inobt_get_rec(cur, rec, &i);
 950		if (error)
 951			return error;
 952		if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
 953			return -EFSCORRUPTED;
 954	}
 955
 956	return 0;
 957}
 958
 959STATIC int
 960xfs_ialloc_get_rec(
 961	struct xfs_btree_cur	*cur,
 962	xfs_agino_t		agino,
 963	xfs_inobt_rec_incore_t	*rec,
 964	int			*done)
 965{
 966	int                     error;
 967	int			i;
 968
 969	error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_EQ, &i);
 970	if (error)
 971		return error;
 972	*done = !i;
 973	if (i) {
 974		error = xfs_inobt_get_rec(cur, rec, &i);
 975		if (error)
 976			return error;
 977		if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
 978			return -EFSCORRUPTED;
 979	}
 980
 981	return 0;
 982}
 983
 984/*
 985 * Return the offset of the first free inode in the record. If the inode chunk
 986 * is sparsely allocated, we convert the record holemask to inode granularity
 987 * and mask off the unallocated regions from the inode free mask.
 988 */
 989STATIC int
 990xfs_inobt_first_free_inode(
 991	struct xfs_inobt_rec_incore	*rec)
 992{
 993	xfs_inofree_t			realfree;
 994
 995	/* if there are no holes, return the first available offset */
 996	if (!xfs_inobt_issparse(rec->ir_holemask))
 997		return xfs_lowbit64(rec->ir_free);
 998
 999	realfree = xfs_inobt_irec_to_allocmask(rec);
1000	realfree &= rec->ir_free;
1001
1002	return xfs_lowbit64(realfree);
1003}
1004
1005/*
1006 * Allocate an inode using the inobt-only algorithm.
1007 */
1008STATIC int
1009xfs_dialloc_ag_inobt(
1010	struct xfs_perag	*pag,
1011	struct xfs_trans	*tp,
1012	struct xfs_buf		*agbp,
1013	xfs_ino_t		parent,
1014	xfs_ino_t		*inop)
1015{
1016	struct xfs_mount	*mp = tp->t_mountp;
1017	struct xfs_agi		*agi = agbp->b_addr;
1018	xfs_agnumber_t		pagno = XFS_INO_TO_AGNO(mp, parent);
1019	xfs_agino_t		pagino = XFS_INO_TO_AGINO(mp, parent);
1020	struct xfs_btree_cur	*cur, *tcur;
1021	struct xfs_inobt_rec_incore rec, trec;
1022	xfs_ino_t		ino;
1023	int			error;
1024	int			offset;
1025	int			i, j;
1026	int			searchdistance = 10;
1027
1028	ASSERT(xfs_perag_initialised_agi(pag));
1029	ASSERT(xfs_perag_allows_inodes(pag));
1030	ASSERT(pag->pagi_freecount > 0);
1031
1032 restart_pagno:
1033	cur = xfs_inobt_init_cursor(pag, tp, agbp, XFS_BTNUM_INO);
1034	/*
1035	 * If pagino is 0 (this is the root inode allocation) use newino.
1036	 * This must work because we've just allocated some.
1037	 */
1038	if (!pagino)
1039		pagino = be32_to_cpu(agi->agi_newino);
1040
1041	error = xfs_check_agi_freecount(cur);
1042	if (error)
1043		goto error0;
1044
1045	/*
1046	 * If in the same AG as the parent, try to get near the parent.
1047	 */
1048	if (pagno == pag->pag_agno) {
1049		int		doneleft;	/* done, to the left */
1050		int		doneright;	/* done, to the right */
1051
1052		error = xfs_inobt_lookup(cur, pagino, XFS_LOOKUP_LE, &i);
1053		if (error)
1054			goto error0;
1055		if (XFS_IS_CORRUPT(mp, i != 1)) {
1056			error = -EFSCORRUPTED;
1057			goto error0;
1058		}
1059
1060		error = xfs_inobt_get_rec(cur, &rec, &j);
1061		if (error)
1062			goto error0;
1063		if (XFS_IS_CORRUPT(mp, j != 1)) {
1064			error = -EFSCORRUPTED;
1065			goto error0;
1066		}
1067
1068		if (rec.ir_freecount > 0) {
1069			/*
1070			 * Found a free inode in the same chunk
1071			 * as the parent, done.
1072			 */
1073			goto alloc_inode;
1074		}
1075
1076
1077		/*
1078		 * In the same AG as parent, but parent's chunk is full.
1079		 */
1080
1081		/* duplicate the cursor, search left & right simultaneously */
1082		error = xfs_btree_dup_cursor(cur, &tcur);
1083		if (error)
1084			goto error0;
1085
1086		/*
1087		 * Skip to last blocks looked up if same parent inode.
1088		 */
1089		if (pagino != NULLAGINO &&
1090		    pag->pagl_pagino == pagino &&
1091		    pag->pagl_leftrec != NULLAGINO &&
1092		    pag->pagl_rightrec != NULLAGINO) {
1093			error = xfs_ialloc_get_rec(tcur, pag->pagl_leftrec,
1094						   &trec, &doneleft);
1095			if (error)
1096				goto error1;
1097
1098			error = xfs_ialloc_get_rec(cur, pag->pagl_rightrec,
1099						   &rec, &doneright);
1100			if (error)
1101				goto error1;
1102		} else {
1103			/* search left with tcur, back up 1 record */
1104			error = xfs_ialloc_next_rec(tcur, &trec, &doneleft, 1);
1105			if (error)
1106				goto error1;
1107
1108			/* search right with cur, go forward 1 record. */
1109			error = xfs_ialloc_next_rec(cur, &rec, &doneright, 0);
1110			if (error)
1111				goto error1;
1112		}
1113
1114		/*
1115		 * Loop until we find an inode chunk with a free inode.
1116		 */
1117		while (--searchdistance > 0 && (!doneleft || !doneright)) {
1118			int	useleft;  /* using left inode chunk this time */
1119
1120			/* figure out the closer block if both are valid. */
1121			if (!doneleft && !doneright) {
1122				useleft = pagino -
1123				 (trec.ir_startino + XFS_INODES_PER_CHUNK - 1) <
1124				  rec.ir_startino - pagino;
1125			} else {
1126				useleft = !doneleft;
1127			}
1128
1129			/* free inodes to the left? */
1130			if (useleft && trec.ir_freecount) {
1131				xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1132				cur = tcur;
1133
1134				pag->pagl_leftrec = trec.ir_startino;
1135				pag->pagl_rightrec = rec.ir_startino;
1136				pag->pagl_pagino = pagino;
1137				rec = trec;
1138				goto alloc_inode;
1139			}
1140
1141			/* free inodes to the right? */
1142			if (!useleft && rec.ir_freecount) {
1143				xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1144
1145				pag->pagl_leftrec = trec.ir_startino;
1146				pag->pagl_rightrec = rec.ir_startino;
1147				pag->pagl_pagino = pagino;
1148				goto alloc_inode;
1149			}
1150
1151			/* get next record to check */
1152			if (useleft) {
1153				error = xfs_ialloc_next_rec(tcur, &trec,
1154								 &doneleft, 1);
1155			} else {
1156				error = xfs_ialloc_next_rec(cur, &rec,
1157								 &doneright, 0);
1158			}
1159			if (error)
1160				goto error1;
1161		}
1162
1163		if (searchdistance <= 0) {
1164			/*
1165			 * Not in range - save last search
1166			 * location and allocate a new inode
1167			 */
1168			xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1169			pag->pagl_leftrec = trec.ir_startino;
1170			pag->pagl_rightrec = rec.ir_startino;
1171			pag->pagl_pagino = pagino;
1172
1173		} else {
1174			/*
1175			 * We've reached the end of the btree. because
1176			 * we are only searching a small chunk of the
1177			 * btree each search, there is obviously free
1178			 * inodes closer to the parent inode than we
1179			 * are now. restart the search again.
1180			 */
1181			pag->pagl_pagino = NULLAGINO;
1182			pag->pagl_leftrec = NULLAGINO;
1183			pag->pagl_rightrec = NULLAGINO;
1184			xfs_btree_del_cursor(tcur, XFS_BTREE_NOERROR);
1185			xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1186			goto restart_pagno;
1187		}
1188	}
1189
1190	/*
1191	 * In a different AG from the parent.
1192	 * See if the most recently allocated block has any free.
1193	 */
1194	if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1195		error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1196					 XFS_LOOKUP_EQ, &i);
1197		if (error)
1198			goto error0;
1199
1200		if (i == 1) {
1201			error = xfs_inobt_get_rec(cur, &rec, &j);
1202			if (error)
1203				goto error0;
1204
1205			if (j == 1 && rec.ir_freecount > 0) {
1206				/*
1207				 * The last chunk allocated in the group
1208				 * still has a free inode.
1209				 */
1210				goto alloc_inode;
1211			}
1212		}
1213	}
1214
1215	/*
1216	 * None left in the last group, search the whole AG
1217	 */
1218	error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1219	if (error)
1220		goto error0;
1221	if (XFS_IS_CORRUPT(mp, i != 1)) {
1222		error = -EFSCORRUPTED;
1223		goto error0;
1224	}
1225
1226	for (;;) {
1227		error = xfs_inobt_get_rec(cur, &rec, &i);
1228		if (error)
1229			goto error0;
1230		if (XFS_IS_CORRUPT(mp, i != 1)) {
1231			error = -EFSCORRUPTED;
1232			goto error0;
1233		}
1234		if (rec.ir_freecount > 0)
1235			break;
1236		error = xfs_btree_increment(cur, 0, &i);
1237		if (error)
1238			goto error0;
1239		if (XFS_IS_CORRUPT(mp, i != 1)) {
1240			error = -EFSCORRUPTED;
1241			goto error0;
1242		}
1243	}
1244
1245alloc_inode:
1246	offset = xfs_inobt_first_free_inode(&rec);
1247	ASSERT(offset >= 0);
1248	ASSERT(offset < XFS_INODES_PER_CHUNK);
1249	ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1250				   XFS_INODES_PER_CHUNK) == 0);
1251	ino = XFS_AGINO_TO_INO(mp, pag->pag_agno, rec.ir_startino + offset);
1252	rec.ir_free &= ~XFS_INOBT_MASK(offset);
1253	rec.ir_freecount--;
1254	error = xfs_inobt_update(cur, &rec);
1255	if (error)
1256		goto error0;
1257	be32_add_cpu(&agi->agi_freecount, -1);
1258	xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1259	pag->pagi_freecount--;
1260
1261	error = xfs_check_agi_freecount(cur);
1262	if (error)
1263		goto error0;
1264
1265	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1266	xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1267	*inop = ino;
1268	return 0;
1269error1:
1270	xfs_btree_del_cursor(tcur, XFS_BTREE_ERROR);
1271error0:
1272	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1273	return error;
1274}
1275
1276/*
1277 * Use the free inode btree to allocate an inode based on distance from the
1278 * parent. Note that the provided cursor may be deleted and replaced.
1279 */
1280STATIC int
1281xfs_dialloc_ag_finobt_near(
1282	xfs_agino_t			pagino,
1283	struct xfs_btree_cur		**ocur,
1284	struct xfs_inobt_rec_incore	*rec)
1285{
1286	struct xfs_btree_cur		*lcur = *ocur;	/* left search cursor */
1287	struct xfs_btree_cur		*rcur;	/* right search cursor */
1288	struct xfs_inobt_rec_incore	rrec;
1289	int				error;
1290	int				i, j;
1291
1292	error = xfs_inobt_lookup(lcur, pagino, XFS_LOOKUP_LE, &i);
1293	if (error)
1294		return error;
1295
1296	if (i == 1) {
1297		error = xfs_inobt_get_rec(lcur, rec, &i);
1298		if (error)
1299			return error;
1300		if (XFS_IS_CORRUPT(lcur->bc_mp, i != 1))
1301			return -EFSCORRUPTED;
1302
1303		/*
1304		 * See if we've landed in the parent inode record. The finobt
1305		 * only tracks chunks with at least one free inode, so record
1306		 * existence is enough.
1307		 */
1308		if (pagino >= rec->ir_startino &&
1309		    pagino < (rec->ir_startino + XFS_INODES_PER_CHUNK))
1310			return 0;
1311	}
1312
1313	error = xfs_btree_dup_cursor(lcur, &rcur);
1314	if (error)
1315		return error;
1316
1317	error = xfs_inobt_lookup(rcur, pagino, XFS_LOOKUP_GE, &j);
1318	if (error)
1319		goto error_rcur;
1320	if (j == 1) {
1321		error = xfs_inobt_get_rec(rcur, &rrec, &j);
1322		if (error)
1323			goto error_rcur;
1324		if (XFS_IS_CORRUPT(lcur->bc_mp, j != 1)) {
1325			error = -EFSCORRUPTED;
1326			goto error_rcur;
1327		}
1328	}
1329
1330	if (XFS_IS_CORRUPT(lcur->bc_mp, i != 1 && j != 1)) {
1331		error = -EFSCORRUPTED;
1332		goto error_rcur;
1333	}
1334	if (i == 1 && j == 1) {
1335		/*
1336		 * Both the left and right records are valid. Choose the closer
1337		 * inode chunk to the target.
1338		 */
1339		if ((pagino - rec->ir_startino + XFS_INODES_PER_CHUNK - 1) >
1340		    (rrec.ir_startino - pagino)) {
1341			*rec = rrec;
1342			xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1343			*ocur = rcur;
1344		} else {
1345			xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1346		}
1347	} else if (j == 1) {
1348		/* only the right record is valid */
1349		*rec = rrec;
1350		xfs_btree_del_cursor(lcur, XFS_BTREE_NOERROR);
1351		*ocur = rcur;
1352	} else if (i == 1) {
1353		/* only the left record is valid */
1354		xfs_btree_del_cursor(rcur, XFS_BTREE_NOERROR);
1355	}
1356
1357	return 0;
1358
1359error_rcur:
1360	xfs_btree_del_cursor(rcur, XFS_BTREE_ERROR);
1361	return error;
1362}
1363
1364/*
1365 * Use the free inode btree to find a free inode based on a newino hint. If
1366 * the hint is NULL, find the first free inode in the AG.
1367 */
1368STATIC int
1369xfs_dialloc_ag_finobt_newino(
1370	struct xfs_agi			*agi,
1371	struct xfs_btree_cur		*cur,
1372	struct xfs_inobt_rec_incore	*rec)
1373{
1374	int error;
1375	int i;
1376
1377	if (agi->agi_newino != cpu_to_be32(NULLAGINO)) {
1378		error = xfs_inobt_lookup(cur, be32_to_cpu(agi->agi_newino),
1379					 XFS_LOOKUP_EQ, &i);
1380		if (error)
1381			return error;
1382		if (i == 1) {
1383			error = xfs_inobt_get_rec(cur, rec, &i);
1384			if (error)
1385				return error;
1386			if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1387				return -EFSCORRUPTED;
1388			return 0;
1389		}
1390	}
1391
1392	/*
1393	 * Find the first inode available in the AG.
1394	 */
1395	error = xfs_inobt_lookup(cur, 0, XFS_LOOKUP_GE, &i);
1396	if (error)
1397		return error;
1398	if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1399		return -EFSCORRUPTED;
1400
1401	error = xfs_inobt_get_rec(cur, rec, &i);
1402	if (error)
1403		return error;
1404	if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1405		return -EFSCORRUPTED;
1406
1407	return 0;
1408}
1409
1410/*
1411 * Update the inobt based on a modification made to the finobt. Also ensure that
1412 * the records from both trees are equivalent post-modification.
1413 */
1414STATIC int
1415xfs_dialloc_ag_update_inobt(
1416	struct xfs_btree_cur		*cur,	/* inobt cursor */
1417	struct xfs_inobt_rec_incore	*frec,	/* finobt record */
1418	int				offset) /* inode offset */
1419{
1420	struct xfs_inobt_rec_incore	rec;
1421	int				error;
1422	int				i;
1423
1424	error = xfs_inobt_lookup(cur, frec->ir_startino, XFS_LOOKUP_EQ, &i);
1425	if (error)
1426		return error;
1427	if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1428		return -EFSCORRUPTED;
1429
1430	error = xfs_inobt_get_rec(cur, &rec, &i);
1431	if (error)
1432		return error;
1433	if (XFS_IS_CORRUPT(cur->bc_mp, i != 1))
1434		return -EFSCORRUPTED;
1435	ASSERT((XFS_AGINO_TO_OFFSET(cur->bc_mp, rec.ir_startino) %
1436				   XFS_INODES_PER_CHUNK) == 0);
1437
1438	rec.ir_free &= ~XFS_INOBT_MASK(offset);
1439	rec.ir_freecount--;
1440
1441	if (XFS_IS_CORRUPT(cur->bc_mp,
1442			   rec.ir_free != frec->ir_free ||
1443			   rec.ir_freecount != frec->ir_freecount))
1444		return -EFSCORRUPTED;
1445
1446	return xfs_inobt_update(cur, &rec);
1447}
1448
1449/*
1450 * Allocate an inode using the free inode btree, if available. Otherwise, fall
1451 * back to the inobt search algorithm.
1452 *
1453 * The caller selected an AG for us, and made sure that free inodes are
1454 * available.
1455 */
1456static int
1457xfs_dialloc_ag(
1458	struct xfs_perag	*pag,
1459	struct xfs_trans	*tp,
1460	struct xfs_buf		*agbp,
1461	xfs_ino_t		parent,
1462	xfs_ino_t		*inop)
1463{
1464	struct xfs_mount		*mp = tp->t_mountp;
1465	struct xfs_agi			*agi = agbp->b_addr;
1466	xfs_agnumber_t			pagno = XFS_INO_TO_AGNO(mp, parent);
1467	xfs_agino_t			pagino = XFS_INO_TO_AGINO(mp, parent);
1468	struct xfs_btree_cur		*cur;	/* finobt cursor */
1469	struct xfs_btree_cur		*icur;	/* inobt cursor */
1470	struct xfs_inobt_rec_incore	rec;
1471	xfs_ino_t			ino;
1472	int				error;
1473	int				offset;
1474	int				i;
1475
1476	if (!xfs_has_finobt(mp))
1477		return xfs_dialloc_ag_inobt(pag, tp, agbp, parent, inop);
1478
1479	/*
1480	 * If pagino is 0 (this is the root inode allocation) use newino.
1481	 * This must work because we've just allocated some.
1482	 */
1483	if (!pagino)
1484		pagino = be32_to_cpu(agi->agi_newino);
1485
1486	cur = xfs_inobt_init_cursor(pag, tp, agbp, XFS_BTNUM_FINO);
1487
1488	error = xfs_check_agi_freecount(cur);
1489	if (error)
1490		goto error_cur;
1491
1492	/*
1493	 * The search algorithm depends on whether we're in the same AG as the
1494	 * parent. If so, find the closest available inode to the parent. If
1495	 * not, consider the agi hint or find the first free inode in the AG.
1496	 */
1497	if (pag->pag_agno == pagno)
1498		error = xfs_dialloc_ag_finobt_near(pagino, &cur, &rec);
1499	else
1500		error = xfs_dialloc_ag_finobt_newino(agi, cur, &rec);
1501	if (error)
1502		goto error_cur;
1503
1504	offset = xfs_inobt_first_free_inode(&rec);
1505	ASSERT(offset >= 0);
1506	ASSERT(offset < XFS_INODES_PER_CHUNK);
1507	ASSERT((XFS_AGINO_TO_OFFSET(mp, rec.ir_startino) %
1508				   XFS_INODES_PER_CHUNK) == 0);
1509	ino = XFS_AGINO_TO_INO(mp, pag->pag_agno, rec.ir_startino + offset);
1510
1511	/*
1512	 * Modify or remove the finobt record.
1513	 */
1514	rec.ir_free &= ~XFS_INOBT_MASK(offset);
1515	rec.ir_freecount--;
1516	if (rec.ir_freecount)
1517		error = xfs_inobt_update(cur, &rec);
1518	else
1519		error = xfs_btree_delete(cur, &i);
1520	if (error)
1521		goto error_cur;
1522
1523	/*
1524	 * The finobt has now been updated appropriately. We haven't updated the
1525	 * agi and superblock yet, so we can create an inobt cursor and validate
1526	 * the original freecount. If all is well, make the equivalent update to
1527	 * the inobt using the finobt record and offset information.
1528	 */
1529	icur = xfs_inobt_init_cursor(pag, tp, agbp, XFS_BTNUM_INO);
1530
1531	error = xfs_check_agi_freecount(icur);
1532	if (error)
1533		goto error_icur;
1534
1535	error = xfs_dialloc_ag_update_inobt(icur, &rec, offset);
1536	if (error)
1537		goto error_icur;
1538
1539	/*
1540	 * Both trees have now been updated. We must update the perag and
1541	 * superblock before we can check the freecount for each btree.
1542	 */
1543	be32_add_cpu(&agi->agi_freecount, -1);
1544	xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
1545	pag->pagi_freecount--;
1546
1547	xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -1);
1548
1549	error = xfs_check_agi_freecount(icur);
1550	if (error)
1551		goto error_icur;
1552	error = xfs_check_agi_freecount(cur);
1553	if (error)
1554		goto error_icur;
1555
1556	xfs_btree_del_cursor(icur, XFS_BTREE_NOERROR);
1557	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1558	*inop = ino;
1559	return 0;
1560
1561error_icur:
1562	xfs_btree_del_cursor(icur, XFS_BTREE_ERROR);
1563error_cur:
1564	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1565	return error;
1566}
1567
1568static int
1569xfs_dialloc_roll(
1570	struct xfs_trans	**tpp,
1571	struct xfs_buf		*agibp)
1572{
1573	struct xfs_trans	*tp = *tpp;
1574	struct xfs_dquot_acct	*dqinfo;
1575	int			error;
1576
1577	/*
1578	 * Hold to on to the agibp across the commit so no other allocation can
1579	 * come in and take the free inodes we just allocated for our caller.
1580	 */
1581	xfs_trans_bhold(tp, agibp);
1582
1583	/*
1584	 * We want the quota changes to be associated with the next transaction,
1585	 * NOT this one. So, detach the dqinfo from this and attach it to the
1586	 * next transaction.
1587	 */
1588	dqinfo = tp->t_dqinfo;
1589	tp->t_dqinfo = NULL;
1590
1591	error = xfs_trans_roll(&tp);
1592
1593	/* Re-attach the quota info that we detached from prev trx. */
1594	tp->t_dqinfo = dqinfo;
1595
1596	/*
1597	 * Join the buffer even on commit error so that the buffer is released
1598	 * when the caller cancels the transaction and doesn't have to handle
1599	 * this error case specially.
1600	 */
1601	xfs_trans_bjoin(tp, agibp);
1602	*tpp = tp;
1603	return error;
1604}
1605
1606static bool
1607xfs_dialloc_good_ag(
1608	struct xfs_perag	*pag,
1609	struct xfs_trans	*tp,
1610	umode_t			mode,
1611	int			flags,
1612	bool			ok_alloc)
1613{
1614	struct xfs_mount	*mp = tp->t_mountp;
1615	xfs_extlen_t		ineed;
1616	xfs_extlen_t		longest = 0;
1617	int			needspace;
1618	int			error;
1619
1620	if (!pag)
1621		return false;
1622	if (!xfs_perag_allows_inodes(pag))
1623		return false;
1624
1625	if (!xfs_perag_initialised_agi(pag)) {
1626		error = xfs_ialloc_read_agi(pag, tp, NULL);
1627		if (error)
1628			return false;
1629	}
1630
1631	if (pag->pagi_freecount)
1632		return true;
1633	if (!ok_alloc)
1634		return false;
1635
1636	if (!xfs_perag_initialised_agf(pag)) {
1637		error = xfs_alloc_read_agf(pag, tp, flags, NULL);
1638		if (error)
1639			return false;
1640	}
1641
1642	/*
1643	 * Check that there is enough free space for the file plus a chunk of
1644	 * inodes if we need to allocate some. If this is the first pass across
1645	 * the AGs, take into account the potential space needed for alignment
1646	 * of inode chunks when checking the longest contiguous free space in
1647	 * the AG - this prevents us from getting ENOSPC because we have free
1648	 * space larger than ialloc_blks but alignment constraints prevent us
1649	 * from using it.
1650	 *
1651	 * If we can't find an AG with space for full alignment slack to be
1652	 * taken into account, we must be near ENOSPC in all AGs.  Hence we
1653	 * don't include alignment for the second pass and so if we fail
1654	 * allocation due to alignment issues then it is most likely a real
1655	 * ENOSPC condition.
1656	 *
1657	 * XXX(dgc): this calculation is now bogus thanks to the per-ag
1658	 * reservations that xfs_alloc_fix_freelist() now does via
1659	 * xfs_alloc_space_available(). When the AG fills up, pagf_freeblks will
1660	 * be more than large enough for the check below to succeed, but
1661	 * xfs_alloc_space_available() will fail because of the non-zero
1662	 * metadata reservation and hence we won't actually be able to allocate
1663	 * more inodes in this AG. We do soooo much unnecessary work near ENOSPC
1664	 * because of this.
1665	 */
1666	ineed = M_IGEO(mp)->ialloc_min_blks;
1667	if (flags && ineed > 1)
1668		ineed += M_IGEO(mp)->cluster_align;
1669	longest = pag->pagf_longest;
1670	if (!longest)
1671		longest = pag->pagf_flcount > 0;
1672	needspace = S_ISDIR(mode) || S_ISREG(mode) || S_ISLNK(mode);
1673
1674	if (pag->pagf_freeblks < needspace + ineed || longest < ineed)
1675		return false;
1676	return true;
1677}
1678
1679static int
1680xfs_dialloc_try_ag(
1681	struct xfs_perag	*pag,
1682	struct xfs_trans	**tpp,
1683	xfs_ino_t		parent,
1684	xfs_ino_t		*new_ino,
1685	bool			ok_alloc)
1686{
1687	struct xfs_buf		*agbp;
1688	xfs_ino_t		ino;
1689	int			error;
1690
1691	/*
1692	 * Then read in the AGI buffer and recheck with the AGI buffer
1693	 * lock held.
1694	 */
1695	error = xfs_ialloc_read_agi(pag, *tpp, &agbp);
1696	if (error)
1697		return error;
1698
1699	if (!pag->pagi_freecount) {
1700		if (!ok_alloc) {
1701			error = -EAGAIN;
1702			goto out_release;
1703		}
1704
1705		error = xfs_ialloc_ag_alloc(pag, *tpp, agbp);
1706		if (error < 0)
1707			goto out_release;
1708
1709		/*
1710		 * We successfully allocated space for an inode cluster in this
1711		 * AG.  Roll the transaction so that we can allocate one of the
1712		 * new inodes.
1713		 */
1714		ASSERT(pag->pagi_freecount > 0);
1715		error = xfs_dialloc_roll(tpp, agbp);
1716		if (error)
1717			goto out_release;
1718	}
1719
1720	/* Allocate an inode in the found AG */
1721	error = xfs_dialloc_ag(pag, *tpp, agbp, parent, &ino);
1722	if (!error)
1723		*new_ino = ino;
1724	return error;
1725
1726out_release:
1727	xfs_trans_brelse(*tpp, agbp);
1728	return error;
1729}
1730
1731/*
1732 * Allocate an on-disk inode.
1733 *
1734 * Mode is used to tell whether the new inode is a directory and hence where to
1735 * locate it. The on-disk inode that is allocated will be returned in @new_ino
1736 * on success, otherwise an error will be set to indicate the failure (e.g.
1737 * -ENOSPC).
1738 */
1739int
1740xfs_dialloc(
1741	struct xfs_trans	**tpp,
1742	xfs_ino_t		parent,
1743	umode_t			mode,
1744	xfs_ino_t		*new_ino)
1745{
1746	struct xfs_mount	*mp = (*tpp)->t_mountp;
1747	xfs_agnumber_t		agno;
1748	int			error = 0;
1749	xfs_agnumber_t		start_agno;
1750	struct xfs_perag	*pag;
1751	struct xfs_ino_geometry	*igeo = M_IGEO(mp);
1752	bool			ok_alloc = true;
1753	bool			low_space = false;
1754	int			flags;
1755	xfs_ino_t		ino = NULLFSINO;
1756
1757	/*
1758	 * Directories, symlinks, and regular files frequently allocate at least
1759	 * one block, so factor that potential expansion when we examine whether
1760	 * an AG has enough space for file creation.
1761	 */
1762	if (S_ISDIR(mode))
1763		start_agno = (atomic_inc_return(&mp->m_agirotor) - 1) %
1764				mp->m_maxagi;
1765	else {
1766		start_agno = XFS_INO_TO_AGNO(mp, parent);
1767		if (start_agno >= mp->m_maxagi)
1768			start_agno = 0;
1769	}
1770
1771	/*
1772	 * If we have already hit the ceiling of inode blocks then clear
1773	 * ok_alloc so we scan all available agi structures for a free
1774	 * inode.
1775	 *
1776	 * Read rough value of mp->m_icount by percpu_counter_read_positive,
1777	 * which will sacrifice the preciseness but improve the performance.
1778	 */
1779	if (igeo->maxicount &&
1780	    percpu_counter_read_positive(&mp->m_icount) + igeo->ialloc_inos
1781							> igeo->maxicount) {
1782		ok_alloc = false;
1783	}
1784
1785	/*
1786	 * If we are near to ENOSPC, we want to prefer allocation from AGs that
1787	 * have free inodes in them rather than use up free space allocating new
1788	 * inode chunks. Hence we turn off allocation for the first non-blocking
1789	 * pass through the AGs if we are near ENOSPC to consume free inodes
1790	 * that we can immediately allocate, but then we allow allocation on the
1791	 * second pass if we fail to find an AG with free inodes in it.
1792	 */
1793	if (percpu_counter_read_positive(&mp->m_fdblocks) <
1794			mp->m_low_space[XFS_LOWSP_1_PCNT]) {
1795		ok_alloc = false;
1796		low_space = true;
1797	}
1798
1799	/*
1800	 * Loop until we find an allocation group that either has free inodes
1801	 * or in which we can allocate some inodes.  Iterate through the
1802	 * allocation groups upward, wrapping at the end.
1803	 */
1804	flags = XFS_ALLOC_FLAG_TRYLOCK;
1805retry:
1806	for_each_perag_wrap_at(mp, start_agno, mp->m_maxagi, agno, pag) {
1807		if (xfs_dialloc_good_ag(pag, *tpp, mode, flags, ok_alloc)) {
1808			error = xfs_dialloc_try_ag(pag, tpp, parent,
1809					&ino, ok_alloc);
1810			if (error != -EAGAIN)
1811				break;
1812			error = 0;
1813		}
1814
1815		if (xfs_is_shutdown(mp)) {
1816			error = -EFSCORRUPTED;
1817			break;
1818		}
1819	}
1820	if (pag)
1821		xfs_perag_rele(pag);
1822	if (error)
1823		return error;
1824	if (ino == NULLFSINO) {
1825		if (flags) {
1826			flags = 0;
1827			if (low_space)
1828				ok_alloc = true;
1829			goto retry;
1830		}
1831		return -ENOSPC;
1832	}
1833	*new_ino = ino;
1834	return 0;
1835}
1836
1837/*
1838 * Free the blocks of an inode chunk. We must consider that the inode chunk
1839 * might be sparse and only free the regions that are allocated as part of the
1840 * chunk.
1841 */
1842static int
1843xfs_difree_inode_chunk(
1844	struct xfs_trans		*tp,
1845	xfs_agnumber_t			agno,
1846	struct xfs_inobt_rec_incore	*rec)
1847{
1848	struct xfs_mount		*mp = tp->t_mountp;
1849	xfs_agblock_t			sagbno = XFS_AGINO_TO_AGBNO(mp,
1850							rec->ir_startino);
1851	int				startidx, endidx;
1852	int				nextbit;
1853	xfs_agblock_t			agbno;
1854	int				contigblk;
1855	DECLARE_BITMAP(holemask, XFS_INOBT_HOLEMASK_BITS);
1856
1857	if (!xfs_inobt_issparse(rec->ir_holemask)) {
1858		/* not sparse, calculate extent info directly */
1859		return xfs_free_extent_later(tp,
1860				XFS_AGB_TO_FSB(mp, agno, sagbno),
1861				M_IGEO(mp)->ialloc_blks, &XFS_RMAP_OINFO_INODES,
1862				XFS_AG_RESV_NONE, false);
1863	}
1864
1865	/* holemask is only 16-bits (fits in an unsigned long) */
1866	ASSERT(sizeof(rec->ir_holemask) <= sizeof(holemask[0]));
1867	holemask[0] = rec->ir_holemask;
1868
1869	/*
1870	 * Find contiguous ranges of zeroes (i.e., allocated regions) in the
1871	 * holemask and convert the start/end index of each range to an extent.
1872	 * We start with the start and end index both pointing at the first 0 in
1873	 * the mask.
1874	 */
1875	startidx = endidx = find_first_zero_bit(holemask,
1876						XFS_INOBT_HOLEMASK_BITS);
1877	nextbit = startidx + 1;
1878	while (startidx < XFS_INOBT_HOLEMASK_BITS) {
1879		int error;
1880
1881		nextbit = find_next_zero_bit(holemask, XFS_INOBT_HOLEMASK_BITS,
1882					     nextbit);
1883		/*
1884		 * If the next zero bit is contiguous, update the end index of
1885		 * the current range and continue.
1886		 */
1887		if (nextbit != XFS_INOBT_HOLEMASK_BITS &&
1888		    nextbit == endidx + 1) {
1889			endidx = nextbit;
1890			goto next;
1891		}
1892
1893		/*
1894		 * nextbit is not contiguous with the current end index. Convert
1895		 * the current start/end to an extent and add it to the free
1896		 * list.
1897		 */
1898		agbno = sagbno + (startidx * XFS_INODES_PER_HOLEMASK_BIT) /
1899				  mp->m_sb.sb_inopblock;
1900		contigblk = ((endidx - startidx + 1) *
1901			     XFS_INODES_PER_HOLEMASK_BIT) /
1902			    mp->m_sb.sb_inopblock;
1903
1904		ASSERT(agbno % mp->m_sb.sb_spino_align == 0);
1905		ASSERT(contigblk % mp->m_sb.sb_spino_align == 0);
1906		error = xfs_free_extent_later(tp,
1907				XFS_AGB_TO_FSB(mp, agno, agbno), contigblk,
1908				&XFS_RMAP_OINFO_INODES, XFS_AG_RESV_NONE,
1909				false);
1910		if (error)
1911			return error;
1912
1913		/* reset range to current bit and carry on... */
1914		startidx = endidx = nextbit;
1915
1916next:
1917		nextbit++;
1918	}
1919	return 0;
1920}
1921
1922STATIC int
1923xfs_difree_inobt(
1924	struct xfs_perag		*pag,
1925	struct xfs_trans		*tp,
1926	struct xfs_buf			*agbp,
1927	xfs_agino_t			agino,
1928	struct xfs_icluster		*xic,
1929	struct xfs_inobt_rec_incore	*orec)
1930{
1931	struct xfs_mount		*mp = pag->pag_mount;
1932	struct xfs_agi			*agi = agbp->b_addr;
1933	struct xfs_btree_cur		*cur;
1934	struct xfs_inobt_rec_incore	rec;
1935	int				ilen;
1936	int				error;
1937	int				i;
1938	int				off;
1939
1940	ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
1941	ASSERT(XFS_AGINO_TO_AGBNO(mp, agino) < be32_to_cpu(agi->agi_length));
1942
1943	/*
1944	 * Initialize the cursor.
1945	 */
1946	cur = xfs_inobt_init_cursor(pag, tp, agbp, XFS_BTNUM_INO);
1947
1948	error = xfs_check_agi_freecount(cur);
1949	if (error)
1950		goto error0;
1951
1952	/*
1953	 * Look for the entry describing this inode.
1954	 */
1955	if ((error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i))) {
1956		xfs_warn(mp, "%s: xfs_inobt_lookup() returned error %d.",
1957			__func__, error);
1958		goto error0;
1959	}
1960	if (XFS_IS_CORRUPT(mp, i != 1)) {
1961		error = -EFSCORRUPTED;
1962		goto error0;
1963	}
1964	error = xfs_inobt_get_rec(cur, &rec, &i);
1965	if (error) {
1966		xfs_warn(mp, "%s: xfs_inobt_get_rec() returned error %d.",
1967			__func__, error);
1968		goto error0;
1969	}
1970	if (XFS_IS_CORRUPT(mp, i != 1)) {
1971		error = -EFSCORRUPTED;
1972		goto error0;
1973	}
1974	/*
1975	 * Get the offset in the inode chunk.
1976	 */
1977	off = agino - rec.ir_startino;
1978	ASSERT(off >= 0 && off < XFS_INODES_PER_CHUNK);
1979	ASSERT(!(rec.ir_free & XFS_INOBT_MASK(off)));
1980	/*
1981	 * Mark the inode free & increment the count.
1982	 */
1983	rec.ir_free |= XFS_INOBT_MASK(off);
1984	rec.ir_freecount++;
1985
1986	/*
1987	 * When an inode chunk is free, it becomes eligible for removal. Don't
1988	 * remove the chunk if the block size is large enough for multiple inode
1989	 * chunks (that might not be free).
1990	 */
1991	if (!xfs_has_ikeep(mp) && rec.ir_free == XFS_INOBT_ALL_FREE &&
1992	    mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) {
1993		xic->deleted = true;
1994		xic->first_ino = XFS_AGINO_TO_INO(mp, pag->pag_agno,
1995				rec.ir_startino);
1996		xic->alloc = xfs_inobt_irec_to_allocmask(&rec);
1997
1998		/*
1999		 * Remove the inode cluster from the AGI B+Tree, adjust the
2000		 * AGI and Superblock inode counts, and mark the disk space
2001		 * to be freed when the transaction is committed.
2002		 */
2003		ilen = rec.ir_freecount;
2004		be32_add_cpu(&agi->agi_count, -ilen);
2005		be32_add_cpu(&agi->agi_freecount, -(ilen - 1));
2006		xfs_ialloc_log_agi(tp, agbp, XFS_AGI_COUNT | XFS_AGI_FREECOUNT);
2007		pag->pagi_freecount -= ilen - 1;
2008		pag->pagi_count -= ilen;
2009		xfs_trans_mod_sb(tp, XFS_TRANS_SB_ICOUNT, -ilen);
2010		xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, -(ilen - 1));
2011
2012		if ((error = xfs_btree_delete(cur, &i))) {
2013			xfs_warn(mp, "%s: xfs_btree_delete returned error %d.",
2014				__func__, error);
2015			goto error0;
2016		}
2017
2018		error = xfs_difree_inode_chunk(tp, pag->pag_agno, &rec);
2019		if (error)
2020			goto error0;
2021	} else {
2022		xic->deleted = false;
2023
2024		error = xfs_inobt_update(cur, &rec);
2025		if (error) {
2026			xfs_warn(mp, "%s: xfs_inobt_update returned error %d.",
2027				__func__, error);
2028			goto error0;
2029		}
2030
2031		/*
2032		 * Change the inode free counts and log the ag/sb changes.
2033		 */
2034		be32_add_cpu(&agi->agi_freecount, 1);
2035		xfs_ialloc_log_agi(tp, agbp, XFS_AGI_FREECOUNT);
2036		pag->pagi_freecount++;
2037		xfs_trans_mod_sb(tp, XFS_TRANS_SB_IFREE, 1);
2038	}
2039
2040	error = xfs_check_agi_freecount(cur);
2041	if (error)
2042		goto error0;
2043
2044	*orec = rec;
2045	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2046	return 0;
2047
2048error0:
2049	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2050	return error;
2051}
2052
2053/*
2054 * Free an inode in the free inode btree.
2055 */
2056STATIC int
2057xfs_difree_finobt(
2058	struct xfs_perag		*pag,
2059	struct xfs_trans		*tp,
2060	struct xfs_buf			*agbp,
2061	xfs_agino_t			agino,
2062	struct xfs_inobt_rec_incore	*ibtrec) /* inobt record */
2063{
2064	struct xfs_mount		*mp = pag->pag_mount;
2065	struct xfs_btree_cur		*cur;
2066	struct xfs_inobt_rec_incore	rec;
2067	int				offset = agino - ibtrec->ir_startino;
2068	int				error;
2069	int				i;
2070
2071	cur = xfs_inobt_init_cursor(pag, tp, agbp, XFS_BTNUM_FINO);
2072
2073	error = xfs_inobt_lookup(cur, ibtrec->ir_startino, XFS_LOOKUP_EQ, &i);
2074	if (error)
2075		goto error;
2076	if (i == 0) {
2077		/*
2078		 * If the record does not exist in the finobt, we must have just
2079		 * freed an inode in a previously fully allocated chunk. If not,
2080		 * something is out of sync.
2081		 */
2082		if (XFS_IS_CORRUPT(mp, ibtrec->ir_freecount != 1)) {
2083			error = -EFSCORRUPTED;
2084			goto error;
2085		}
2086
2087		error = xfs_inobt_insert_rec(cur, ibtrec->ir_holemask,
2088					     ibtrec->ir_count,
2089					     ibtrec->ir_freecount,
2090					     ibtrec->ir_free, &i);
2091		if (error)
2092			goto error;
2093		ASSERT(i == 1);
2094
2095		goto out;
2096	}
2097
2098	/*
2099	 * Read and update the existing record. We could just copy the ibtrec
2100	 * across here, but that would defeat the purpose of having redundant
2101	 * metadata. By making the modifications independently, we can catch
2102	 * corruptions that we wouldn't see if we just copied from one record
2103	 * to another.
2104	 */
2105	error = xfs_inobt_get_rec(cur, &rec, &i);
2106	if (error)
2107		goto error;
2108	if (XFS_IS_CORRUPT(mp, i != 1)) {
2109		error = -EFSCORRUPTED;
2110		goto error;
2111	}
2112
2113	rec.ir_free |= XFS_INOBT_MASK(offset);
2114	rec.ir_freecount++;
2115
2116	if (XFS_IS_CORRUPT(mp,
2117			   rec.ir_free != ibtrec->ir_free ||
2118			   rec.ir_freecount != ibtrec->ir_freecount)) {
2119		error = -EFSCORRUPTED;
2120		goto error;
2121	}
2122
2123	/*
2124	 * The content of inobt records should always match between the inobt
2125	 * and finobt. The lifecycle of records in the finobt is different from
2126	 * the inobt in that the finobt only tracks records with at least one
2127	 * free inode. Hence, if all of the inodes are free and we aren't
2128	 * keeping inode chunks permanently on disk, remove the record.
2129	 * Otherwise, update the record with the new information.
2130	 *
2131	 * Note that we currently can't free chunks when the block size is large
2132	 * enough for multiple chunks. Leave the finobt record to remain in sync
2133	 * with the inobt.
2134	 */
2135	if (!xfs_has_ikeep(mp) && rec.ir_free == XFS_INOBT_ALL_FREE &&
2136	    mp->m_sb.sb_inopblock <= XFS_INODES_PER_CHUNK) {
2137		error = xfs_btree_delete(cur, &i);
2138		if (error)
2139			goto error;
2140		ASSERT(i == 1);
2141	} else {
2142		error = xfs_inobt_update(cur, &rec);
2143		if (error)
2144			goto error;
2145	}
2146
2147out:
2148	error = xfs_check_agi_freecount(cur);
2149	if (error)
2150		goto error;
2151
2152	xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
2153	return 0;
2154
2155error:
2156	xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
2157	return error;
2158}
2159
2160/*
2161 * Free disk inode.  Carefully avoids touching the incore inode, all
2162 * manipulations incore are the caller's responsibility.
2163 * The on-disk inode is not changed by this operation, only the
2164 * btree (free inode mask) is changed.
2165 */
2166int
2167xfs_difree(
2168	struct xfs_trans	*tp,
2169	struct xfs_perag	*pag,
2170	xfs_ino_t		inode,
2171	struct xfs_icluster	*xic)
2172{
2173	/* REFERENCED */
2174	xfs_agblock_t		agbno;	/* block number containing inode */
2175	struct xfs_buf		*agbp;	/* buffer for allocation group header */
2176	xfs_agino_t		agino;	/* allocation group inode number */
2177	int			error;	/* error return value */
2178	struct xfs_mount	*mp = tp->t_mountp;
2179	struct xfs_inobt_rec_incore rec;/* btree record */
2180
2181	/*
2182	 * Break up inode number into its components.
2183	 */
2184	if (pag->pag_agno != XFS_INO_TO_AGNO(mp, inode)) {
2185		xfs_warn(mp, "%s: agno != pag->pag_agno (%d != %d).",
2186			__func__, XFS_INO_TO_AGNO(mp, inode), pag->pag_agno);
2187		ASSERT(0);
2188		return -EINVAL;
2189	}
2190	agino = XFS_INO_TO_AGINO(mp, inode);
2191	if (inode != XFS_AGINO_TO_INO(mp, pag->pag_agno, agino))  {
2192		xfs_warn(mp, "%s: inode != XFS_AGINO_TO_INO() (%llu != %llu).",
2193			__func__, (unsigned long long)inode,
2194			(unsigned long long)XFS_AGINO_TO_INO(mp, pag->pag_agno, agino));
2195		ASSERT(0);
2196		return -EINVAL;
2197	}
2198	agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2199	if (agbno >= mp->m_sb.sb_agblocks)  {
2200		xfs_warn(mp, "%s: agbno >= mp->m_sb.sb_agblocks (%d >= %d).",
2201			__func__, agbno, mp->m_sb.sb_agblocks);
2202		ASSERT(0);
2203		return -EINVAL;
2204	}
2205	/*
2206	 * Get the allocation group header.
2207	 */
2208	error = xfs_ialloc_read_agi(pag, tp, &agbp);
2209	if (error) {
2210		xfs_warn(mp, "%s: xfs_ialloc_read_agi() returned error %d.",
2211			__func__, error);
2212		return error;
2213	}
2214
2215	/*
2216	 * Fix up the inode allocation btree.
2217	 */
2218	error = xfs_difree_inobt(pag, tp, agbp, agino, xic, &rec);
2219	if (error)
2220		goto error0;
2221
2222	/*
2223	 * Fix up the free inode btree.
2224	 */
2225	if (xfs_has_finobt(mp)) {
2226		error = xfs_difree_finobt(pag, tp, agbp, agino, &rec);
2227		if (error)
2228			goto error0;
2229	}
2230
2231	return 0;
2232
2233error0:
2234	return error;
2235}
2236
2237STATIC int
2238xfs_imap_lookup(
2239	struct xfs_perag	*pag,
2240	struct xfs_trans	*tp,
2241	xfs_agino_t		agino,
2242	xfs_agblock_t		agbno,
2243	xfs_agblock_t		*chunk_agbno,
2244	xfs_agblock_t		*offset_agbno,
2245	int			flags)
2246{
2247	struct xfs_mount	*mp = pag->pag_mount;
2248	struct xfs_inobt_rec_incore rec;
2249	struct xfs_btree_cur	*cur;
2250	struct xfs_buf		*agbp;
2251	int			error;
2252	int			i;
2253
2254	error = xfs_ialloc_read_agi(pag, tp, &agbp);
2255	if (error) {
2256		xfs_alert(mp,
2257			"%s: xfs_ialloc_read_agi() returned error %d, agno %d",
2258			__func__, error, pag->pag_agno);
2259		return error;
2260	}
2261
2262	/*
2263	 * Lookup the inode record for the given agino. If the record cannot be
2264	 * found, then it's an invalid inode number and we should abort. Once
2265	 * we have a record, we need to ensure it contains the inode number
2266	 * we are looking up.
2267	 */
2268	cur = xfs_inobt_init_cursor(pag, tp, agbp, XFS_BTNUM_INO);
2269	error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &i);
2270	if (!error) {
2271		if (i)
2272			error = xfs_inobt_get_rec(cur, &rec, &i);
2273		if (!error && i == 0)
2274			error = -EINVAL;
2275	}
2276
2277	xfs_trans_brelse(tp, agbp);
2278	xfs_btree_del_cursor(cur, error);
2279	if (error)
2280		return error;
2281
2282	/* check that the returned record contains the required inode */
2283	if (rec.ir_startino > agino ||
2284	    rec.ir_startino + M_IGEO(mp)->ialloc_inos <= agino)
2285		return -EINVAL;
2286
2287	/* for untrusted inodes check it is allocated first */
2288	if ((flags & XFS_IGET_UNTRUSTED) &&
2289	    (rec.ir_free & XFS_INOBT_MASK(agino - rec.ir_startino)))
2290		return -EINVAL;
2291
2292	*chunk_agbno = XFS_AGINO_TO_AGBNO(mp, rec.ir_startino);
2293	*offset_agbno = agbno - *chunk_agbno;
2294	return 0;
2295}
2296
2297/*
2298 * Return the location of the inode in imap, for mapping it into a buffer.
2299 */
2300int
2301xfs_imap(
2302	struct xfs_perag	*pag,
2303	struct xfs_trans	*tp,
2304	xfs_ino_t		ino,	/* inode to locate */
2305	struct xfs_imap		*imap,	/* location map structure */
2306	uint			flags)	/* flags for inode btree lookup */
2307{
2308	struct xfs_mount	*mp = pag->pag_mount;
2309	xfs_agblock_t		agbno;	/* block number of inode in the alloc group */
2310	xfs_agino_t		agino;	/* inode number within alloc group */
2311	xfs_agblock_t		chunk_agbno;	/* first block in inode chunk */
2312	xfs_agblock_t		cluster_agbno;	/* first block in inode cluster */
2313	int			error;	/* error code */
2314	int			offset;	/* index of inode in its buffer */
2315	xfs_agblock_t		offset_agbno;	/* blks from chunk start to inode */
2316
2317	ASSERT(ino != NULLFSINO);
2318
2319	/*
2320	 * Split up the inode number into its parts.
2321	 */
2322	agino = XFS_INO_TO_AGINO(mp, ino);
2323	agbno = XFS_AGINO_TO_AGBNO(mp, agino);
2324	if (agbno >= mp->m_sb.sb_agblocks ||
2325	    ino != XFS_AGINO_TO_INO(mp, pag->pag_agno, agino)) {
2326		error = -EINVAL;
2327#ifdef DEBUG
2328		/*
2329		 * Don't output diagnostic information for untrusted inodes
2330		 * as they can be invalid without implying corruption.
2331		 */
2332		if (flags & XFS_IGET_UNTRUSTED)
2333			return error;
2334		if (agbno >= mp->m_sb.sb_agblocks) {
2335			xfs_alert(mp,
2336		"%s: agbno (0x%llx) >= mp->m_sb.sb_agblocks (0x%lx)",
2337				__func__, (unsigned long long)agbno,
2338				(unsigned long)mp->m_sb.sb_agblocks);
2339		}
2340		if (ino != XFS_AGINO_TO_INO(mp, pag->pag_agno, agino)) {
2341			xfs_alert(mp,
2342		"%s: ino (0x%llx) != XFS_AGINO_TO_INO() (0x%llx)",
2343				__func__, ino,
2344				XFS_AGINO_TO_INO(mp, pag->pag_agno, agino));
2345		}
2346		xfs_stack_trace();
2347#endif /* DEBUG */
2348		return error;
2349	}
2350
2351	/*
2352	 * For bulkstat and handle lookups, we have an untrusted inode number
2353	 * that we have to verify is valid. We cannot do this just by reading
2354	 * the inode buffer as it may have been unlinked and removed leaving
2355	 * inodes in stale state on disk. Hence we have to do a btree lookup
2356	 * in all cases where an untrusted inode number is passed.
2357	 */
2358	if (flags & XFS_IGET_UNTRUSTED) {
2359		error = xfs_imap_lookup(pag, tp, agino, agbno,
2360					&chunk_agbno, &offset_agbno, flags);
2361		if (error)
2362			return error;
2363		goto out_map;
2364	}
2365
2366	/*
2367	 * If the inode cluster size is the same as the blocksize or
2368	 * smaller we get to the buffer by simple arithmetics.
2369	 */
2370	if (M_IGEO(mp)->blocks_per_cluster == 1) {
2371		offset = XFS_INO_TO_OFFSET(mp, ino);
2372		ASSERT(offset < mp->m_sb.sb_inopblock);
2373
2374		imap->im_blkno = XFS_AGB_TO_DADDR(mp, pag->pag_agno, agbno);
2375		imap->im_len = XFS_FSB_TO_BB(mp, 1);
2376		imap->im_boffset = (unsigned short)(offset <<
2377							mp->m_sb.sb_inodelog);
2378		return 0;
2379	}
2380
2381	/*
2382	 * If the inode chunks are aligned then use simple maths to
2383	 * find the location. Otherwise we have to do a btree
2384	 * lookup to find the location.
2385	 */
2386	if (M_IGEO(mp)->inoalign_mask) {
2387		offset_agbno = agbno & M_IGEO(mp)->inoalign_mask;
2388		chunk_agbno = agbno - offset_agbno;
2389	} else {
2390		error = xfs_imap_lookup(pag, tp, agino, agbno,
2391					&chunk_agbno, &offset_agbno, flags);
2392		if (error)
2393			return error;
2394	}
2395
2396out_map:
2397	ASSERT(agbno >= chunk_agbno);
2398	cluster_agbno = chunk_agbno +
2399		((offset_agbno / M_IGEO(mp)->blocks_per_cluster) *
2400		 M_IGEO(mp)->blocks_per_cluster);
2401	offset = ((agbno - cluster_agbno) * mp->m_sb.sb_inopblock) +
2402		XFS_INO_TO_OFFSET(mp, ino);
2403
2404	imap->im_blkno = XFS_AGB_TO_DADDR(mp, pag->pag_agno, cluster_agbno);
2405	imap->im_len = XFS_FSB_TO_BB(mp, M_IGEO(mp)->blocks_per_cluster);
2406	imap->im_boffset = (unsigned short)(offset << mp->m_sb.sb_inodelog);
2407
2408	/*
2409	 * If the inode number maps to a block outside the bounds
2410	 * of the file system then return NULL rather than calling
2411	 * read_buf and panicing when we get an error from the
2412	 * driver.
2413	 */
2414	if ((imap->im_blkno + imap->im_len) >
2415	    XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks)) {
2416		xfs_alert(mp,
2417	"%s: (im_blkno (0x%llx) + im_len (0x%llx)) > sb_dblocks (0x%llx)",
2418			__func__, (unsigned long long) imap->im_blkno,
2419			(unsigned long long) imap->im_len,
2420			XFS_FSB_TO_BB(mp, mp->m_sb.sb_dblocks));
2421		return -EINVAL;
2422	}
2423	return 0;
2424}
2425
2426/*
2427 * Log specified fields for the ag hdr (inode section). The growth of the agi
2428 * structure over time requires that we interpret the buffer as two logical
2429 * regions delineated by the end of the unlinked list. This is due to the size
2430 * of the hash table and its location in the middle of the agi.
2431 *
2432 * For example, a request to log a field before agi_unlinked and a field after
2433 * agi_unlinked could cause us to log the entire hash table and use an excessive
2434 * amount of log space. To avoid this behavior, log the region up through
2435 * agi_unlinked in one call and the region after agi_unlinked through the end of
2436 * the structure in another.
2437 */
2438void
2439xfs_ialloc_log_agi(
2440	struct xfs_trans	*tp,
2441	struct xfs_buf		*bp,
2442	uint32_t		fields)
2443{
2444	int			first;		/* first byte number */
2445	int			last;		/* last byte number */
2446	static const short	offsets[] = {	/* field starting offsets */
2447					/* keep in sync with bit definitions */
2448		offsetof(xfs_agi_t, agi_magicnum),
2449		offsetof(xfs_agi_t, agi_versionnum),
2450		offsetof(xfs_agi_t, agi_seqno),
2451		offsetof(xfs_agi_t, agi_length),
2452		offsetof(xfs_agi_t, agi_count),
2453		offsetof(xfs_agi_t, agi_root),
2454		offsetof(xfs_agi_t, agi_level),
2455		offsetof(xfs_agi_t, agi_freecount),
2456		offsetof(xfs_agi_t, agi_newino),
2457		offsetof(xfs_agi_t, agi_dirino),
2458		offsetof(xfs_agi_t, agi_unlinked),
2459		offsetof(xfs_agi_t, agi_free_root),
2460		offsetof(xfs_agi_t, agi_free_level),
2461		offsetof(xfs_agi_t, agi_iblocks),
2462		sizeof(xfs_agi_t)
2463	};
2464#ifdef DEBUG
2465	struct xfs_agi		*agi = bp->b_addr;
2466
2467	ASSERT(agi->agi_magicnum == cpu_to_be32(XFS_AGI_MAGIC));
2468#endif
2469
2470	/*
2471	 * Compute byte offsets for the first and last fields in the first
2472	 * region and log the agi buffer. This only logs up through
2473	 * agi_unlinked.
2474	 */
2475	if (fields & XFS_AGI_ALL_BITS_R1) {
2476		xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R1,
2477				  &first, &last);
2478		xfs_trans_log_buf(tp, bp, first, last);
2479	}
2480
2481	/*
2482	 * Mask off the bits in the first region and calculate the first and
2483	 * last field offsets for any bits in the second region.
2484	 */
2485	fields &= ~XFS_AGI_ALL_BITS_R1;
2486	if (fields) {
2487		xfs_btree_offsets(fields, offsets, XFS_AGI_NUM_BITS_R2,
2488				  &first, &last);
2489		xfs_trans_log_buf(tp, bp, first, last);
2490	}
2491}
2492
2493static xfs_failaddr_t
2494xfs_agi_verify(
2495	struct xfs_buf		*bp)
2496{
2497	struct xfs_mount	*mp = bp->b_mount;
2498	struct xfs_agi		*agi = bp->b_addr;
2499	xfs_failaddr_t		fa;
2500	uint32_t		agi_seqno = be32_to_cpu(agi->agi_seqno);
2501	uint32_t		agi_length = be32_to_cpu(agi->agi_length);
2502	int			i;
2503
2504	if (xfs_has_crc(mp)) {
2505		if (!uuid_equal(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid))
2506			return __this_address;
2507		if (!xfs_log_check_lsn(mp, be64_to_cpu(agi->agi_lsn)))
2508			return __this_address;
2509	}
2510
2511	/*
2512	 * Validate the magic number of the agi block.
2513	 */
2514	if (!xfs_verify_magic(bp, agi->agi_magicnum))
2515		return __this_address;
2516	if (!XFS_AGI_GOOD_VERSION(be32_to_cpu(agi->agi_versionnum)))
2517		return __this_address;
2518
2519	fa = xfs_validate_ag_length(bp, agi_seqno, agi_length);
2520	if (fa)
2521		return fa;
2522
2523	if (be32_to_cpu(agi->agi_level) < 1 ||
2524	    be32_to_cpu(agi->agi_level) > M_IGEO(mp)->inobt_maxlevels)
2525		return __this_address;
2526
2527	if (xfs_has_finobt(mp) &&
2528	    (be32_to_cpu(agi->agi_free_level) < 1 ||
2529	     be32_to_cpu(agi->agi_free_level) > M_IGEO(mp)->inobt_maxlevels))
2530		return __this_address;
2531
2532	for (i = 0; i < XFS_AGI_UNLINKED_BUCKETS; i++) {
2533		if (agi->agi_unlinked[i] == cpu_to_be32(NULLAGINO))
2534			continue;
2535		if (!xfs_verify_ino(mp, be32_to_cpu(agi->agi_unlinked[i])))
2536			return __this_address;
2537	}
2538
2539	return NULL;
2540}
2541
2542static void
2543xfs_agi_read_verify(
2544	struct xfs_buf	*bp)
2545{
2546	struct xfs_mount *mp = bp->b_mount;
2547	xfs_failaddr_t	fa;
2548
2549	if (xfs_has_crc(mp) &&
2550	    !xfs_buf_verify_cksum(bp, XFS_AGI_CRC_OFF))
2551		xfs_verifier_error(bp, -EFSBADCRC, __this_address);
2552	else {
2553		fa = xfs_agi_verify(bp);
2554		if (XFS_TEST_ERROR(fa, mp, XFS_ERRTAG_IALLOC_READ_AGI))
2555			xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2556	}
2557}
2558
2559static void
2560xfs_agi_write_verify(
2561	struct xfs_buf	*bp)
2562{
2563	struct xfs_mount	*mp = bp->b_mount;
2564	struct xfs_buf_log_item	*bip = bp->b_log_item;
2565	struct xfs_agi		*agi = bp->b_addr;
2566	xfs_failaddr_t		fa;
2567
2568	fa = xfs_agi_verify(bp);
2569	if (fa) {
2570		xfs_verifier_error(bp, -EFSCORRUPTED, fa);
2571		return;
2572	}
2573
2574	if (!xfs_has_crc(mp))
2575		return;
2576
2577	if (bip)
2578		agi->agi_lsn = cpu_to_be64(bip->bli_item.li_lsn);
2579	xfs_buf_update_cksum(bp, XFS_AGI_CRC_OFF);
2580}
2581
2582const struct xfs_buf_ops xfs_agi_buf_ops = {
2583	.name = "xfs_agi",
2584	.magic = { cpu_to_be32(XFS_AGI_MAGIC), cpu_to_be32(XFS_AGI_MAGIC) },
2585	.verify_read = xfs_agi_read_verify,
2586	.verify_write = xfs_agi_write_verify,
2587	.verify_struct = xfs_agi_verify,
2588};
2589
2590/*
2591 * Read in the allocation group header (inode allocation section)
2592 */
2593int
2594xfs_read_agi(
2595	struct xfs_perag	*pag,
2596	struct xfs_trans	*tp,
2597	struct xfs_buf		**agibpp)
2598{
2599	struct xfs_mount	*mp = pag->pag_mount;
2600	int			error;
2601
2602	trace_xfs_read_agi(pag->pag_mount, pag->pag_agno);
2603
2604	error = xfs_trans_read_buf(mp, tp, mp->m_ddev_targp,
2605			XFS_AG_DADDR(mp, pag->pag_agno, XFS_AGI_DADDR(mp)),
2606			XFS_FSS_TO_BB(mp, 1), 0, agibpp, &xfs_agi_buf_ops);
2607	if (error)
2608		return error;
2609	if (tp)
2610		xfs_trans_buf_set_type(tp, *agibpp, XFS_BLFT_AGI_BUF);
2611
2612	xfs_buf_set_ref(*agibpp, XFS_AGI_REF);
2613	return 0;
2614}
2615
2616/*
2617 * Read in the agi and initialise the per-ag data. If the caller supplies a
2618 * @agibpp, return the locked AGI buffer to them, otherwise release it.
2619 */
2620int
2621xfs_ialloc_read_agi(
2622	struct xfs_perag	*pag,
2623	struct xfs_trans	*tp,
2624	struct xfs_buf		**agibpp)
2625{
2626	struct xfs_buf		*agibp;
2627	struct xfs_agi		*agi;
2628	int			error;
2629
2630	trace_xfs_ialloc_read_agi(pag->pag_mount, pag->pag_agno);
2631
2632	error = xfs_read_agi(pag, tp, &agibp);
2633	if (error)
2634		return error;
2635
2636	agi = agibp->b_addr;
2637	if (!xfs_perag_initialised_agi(pag)) {
2638		pag->pagi_freecount = be32_to_cpu(agi->agi_freecount);
2639		pag->pagi_count = be32_to_cpu(agi->agi_count);
2640		set_bit(XFS_AGSTATE_AGI_INIT, &pag->pag_opstate);
2641	}
2642
2643	/*
2644	 * It's possible for these to be out of sync if
2645	 * we are in the middle of a forced shutdown.
2646	 */
2647	ASSERT(pag->pagi_freecount == be32_to_cpu(agi->agi_freecount) ||
2648		xfs_is_shutdown(pag->pag_mount));
2649	if (agibpp)
2650		*agibpp = agibp;
2651	else
2652		xfs_trans_brelse(tp, agibp);
2653	return 0;
2654}
2655
2656/* How many inodes are backed by inode clusters ondisk? */
2657STATIC int
2658xfs_ialloc_count_ondisk(
2659	struct xfs_btree_cur		*cur,
2660	xfs_agino_t			low,
2661	xfs_agino_t			high,
2662	unsigned int			*allocated)
2663{
2664	struct xfs_inobt_rec_incore	irec;
2665	unsigned int			ret = 0;
2666	int				has_record;
2667	int				error;
2668
2669	error = xfs_inobt_lookup(cur, low, XFS_LOOKUP_LE, &has_record);
2670	if (error)
2671		return error;
2672
2673	while (has_record) {
2674		unsigned int		i, hole_idx;
2675
2676		error = xfs_inobt_get_rec(cur, &irec, &has_record);
2677		if (error)
2678			return error;
2679		if (irec.ir_startino > high)
2680			break;
2681
2682		for (i = 0; i < XFS_INODES_PER_CHUNK; i++) {
2683			if (irec.ir_startino + i < low)
2684				continue;
2685			if (irec.ir_startino + i > high)
2686				break;
2687
2688			hole_idx = i / XFS_INODES_PER_HOLEMASK_BIT;
2689			if (!(irec.ir_holemask & (1U << hole_idx)))
2690				ret++;
2691		}
2692
2693		error = xfs_btree_increment(cur, 0, &has_record);
2694		if (error)
2695			return error;
2696	}
2697
2698	*allocated = ret;
2699	return 0;
2700}
2701
2702/* Is there an inode record covering a given extent? */
2703int
2704xfs_ialloc_has_inodes_at_extent(
2705	struct xfs_btree_cur	*cur,
2706	xfs_agblock_t		bno,
2707	xfs_extlen_t		len,
2708	enum xbtree_recpacking	*outcome)
2709{
2710	xfs_agino_t		agino;
2711	xfs_agino_t		last_agino;
2712	unsigned int		allocated;
2713	int			error;
2714
2715	agino = XFS_AGB_TO_AGINO(cur->bc_mp, bno);
2716	last_agino = XFS_AGB_TO_AGINO(cur->bc_mp, bno + len) - 1;
2717
2718	error = xfs_ialloc_count_ondisk(cur, agino, last_agino, &allocated);
2719	if (error)
2720		return error;
2721
2722	if (allocated == 0)
2723		*outcome = XBTREE_RECPACKING_EMPTY;
2724	else if (allocated == last_agino - agino + 1)
2725		*outcome = XBTREE_RECPACKING_FULL;
2726	else
2727		*outcome = XBTREE_RECPACKING_SPARSE;
2728	return 0;
2729}
2730
2731struct xfs_ialloc_count_inodes {
2732	xfs_agino_t			count;
2733	xfs_agino_t			freecount;
2734};
2735
2736/* Record inode counts across all inobt records. */
2737STATIC int
2738xfs_ialloc_count_inodes_rec(
2739	struct xfs_btree_cur		*cur,
2740	const union xfs_btree_rec	*rec,
2741	void				*priv)
2742{
2743	struct xfs_inobt_rec_incore	irec;
2744	struct xfs_ialloc_count_inodes	*ci = priv;
2745	xfs_failaddr_t			fa;
2746
2747	xfs_inobt_btrec_to_irec(cur->bc_mp, rec, &irec);
2748	fa = xfs_inobt_check_irec(cur->bc_ag.pag, &irec);
2749	if (fa)
2750		return xfs_inobt_complain_bad_rec(cur, fa, &irec);
2751
2752	ci->count += irec.ir_count;
2753	ci->freecount += irec.ir_freecount;
2754
2755	return 0;
2756}
2757
2758/* Count allocated and free inodes under an inobt. */
2759int
2760xfs_ialloc_count_inodes(
2761	struct xfs_btree_cur		*cur,
2762	xfs_agino_t			*count,
2763	xfs_agino_t			*freecount)
2764{
2765	struct xfs_ialloc_count_inodes	ci = {0};
2766	int				error;
2767
2768	ASSERT(cur->bc_btnum == XFS_BTNUM_INO);
2769	error = xfs_btree_query_all(cur, xfs_ialloc_count_inodes_rec, &ci);
2770	if (error)
2771		return error;
2772
2773	*count = ci.count;
2774	*freecount = ci.freecount;
2775	return 0;
2776}
2777
2778/*
2779 * Initialize inode-related geometry information.
2780 *
2781 * Compute the inode btree min and max levels and set maxicount.
2782 *
2783 * Set the inode cluster size.  This may still be overridden by the file
2784 * system block size if it is larger than the chosen cluster size.
2785 *
2786 * For v5 filesystems, scale the cluster size with the inode size to keep a
2787 * constant ratio of inode per cluster buffer, but only if mkfs has set the
2788 * inode alignment value appropriately for larger cluster sizes.
2789 *
2790 * Then compute the inode cluster alignment information.
2791 */
2792void
2793xfs_ialloc_setup_geometry(
2794	struct xfs_mount	*mp)
2795{
2796	struct xfs_sb		*sbp = &mp->m_sb;
2797	struct xfs_ino_geometry	*igeo = M_IGEO(mp);
2798	uint64_t		icount;
2799	uint			inodes;
2800
2801	igeo->new_diflags2 = 0;
2802	if (xfs_has_bigtime(mp))
2803		igeo->new_diflags2 |= XFS_DIFLAG2_BIGTIME;
2804	if (xfs_has_large_extent_counts(mp))
2805		igeo->new_diflags2 |= XFS_DIFLAG2_NREXT64;
2806
2807	/* Compute inode btree geometry. */
2808	igeo->agino_log = sbp->sb_inopblog + sbp->sb_agblklog;
2809	igeo->inobt_mxr[0] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 1);
2810	igeo->inobt_mxr[1] = xfs_inobt_maxrecs(mp, sbp->sb_blocksize, 0);
2811	igeo->inobt_mnr[0] = igeo->inobt_mxr[0] / 2;
2812	igeo->inobt_mnr[1] = igeo->inobt_mxr[1] / 2;
2813
2814	igeo->ialloc_inos = max_t(uint16_t, XFS_INODES_PER_CHUNK,
2815			sbp->sb_inopblock);
2816	igeo->ialloc_blks = igeo->ialloc_inos >> sbp->sb_inopblog;
2817
2818	if (sbp->sb_spino_align)
2819		igeo->ialloc_min_blks = sbp->sb_spino_align;
2820	else
2821		igeo->ialloc_min_blks = igeo->ialloc_blks;
2822
2823	/* Compute and fill in value of m_ino_geo.inobt_maxlevels. */
2824	inodes = (1LL << XFS_INO_AGINO_BITS(mp)) >> XFS_INODES_PER_CHUNK_LOG;
2825	igeo->inobt_maxlevels = xfs_btree_compute_maxlevels(igeo->inobt_mnr,
2826			inodes);
2827	ASSERT(igeo->inobt_maxlevels <= xfs_iallocbt_maxlevels_ondisk());
2828
2829	/*
2830	 * Set the maximum inode count for this filesystem, being careful not
2831	 * to use obviously garbage sb_inopblog/sb_inopblock values.  Regular
2832	 * users should never get here due to failing sb verification, but
2833	 * certain users (xfs_db) need to be usable even with corrupt metadata.
2834	 */
2835	if (sbp->sb_imax_pct && igeo->ialloc_blks) {
2836		/*
2837		 * Make sure the maximum inode count is a multiple
2838		 * of the units we allocate inodes in.
2839		 */
2840		icount = sbp->sb_dblocks * sbp->sb_imax_pct;
2841		do_div(icount, 100);
2842		do_div(icount, igeo->ialloc_blks);
2843		igeo->maxicount = XFS_FSB_TO_INO(mp,
2844				icount * igeo->ialloc_blks);
2845	} else {
2846		igeo->maxicount = 0;
2847	}
2848
2849	/*
2850	 * Compute the desired size of an inode cluster buffer size, which
2851	 * starts at 8K and (on v5 filesystems) scales up with larger inode
2852	 * sizes.
2853	 *
2854	 * Preserve the desired inode cluster size because the sparse inodes
2855	 * feature uses that desired size (not the actual size) to compute the
2856	 * sparse inode alignment.  The mount code validates this value, so we
2857	 * cannot change the behavior.
2858	 */
2859	igeo->inode_cluster_size_raw = XFS_INODE_BIG_CLUSTER_SIZE;
2860	if (xfs_has_v3inodes(mp)) {
2861		int	new_size = igeo->inode_cluster_size_raw;
2862
2863		new_size *= mp->m_sb.sb_inodesize / XFS_DINODE_MIN_SIZE;
2864		if (mp->m_sb.sb_inoalignmt >= XFS_B_TO_FSBT(mp, new_size))
2865			igeo->inode_cluster_size_raw = new_size;
2866	}
2867
2868	/* Calculate inode cluster ratios. */
2869	if (igeo->inode_cluster_size_raw > mp->m_sb.sb_blocksize)
2870		igeo->blocks_per_cluster = XFS_B_TO_FSBT(mp,
2871				igeo->inode_cluster_size_raw);
2872	else
2873		igeo->blocks_per_cluster = 1;
2874	igeo->inode_cluster_size = XFS_FSB_TO_B(mp, igeo->blocks_per_cluster);
2875	igeo->inodes_per_cluster = XFS_FSB_TO_INO(mp, igeo->blocks_per_cluster);
2876
2877	/* Calculate inode cluster alignment. */
2878	if (xfs_has_align(mp) &&
2879	    mp->m_sb.sb_inoalignmt >= igeo->blocks_per_cluster)
2880		igeo->cluster_align = mp->m_sb.sb_inoalignmt;
2881	else
2882		igeo->cluster_align = 1;
2883	igeo->inoalign_mask = igeo->cluster_align - 1;
2884	igeo->cluster_align_inodes = XFS_FSB_TO_INO(mp, igeo->cluster_align);
2885
2886	/*
2887	 * If we are using stripe alignment, check whether
2888	 * the stripe unit is a multiple of the inode alignment
2889	 */
2890	if (mp->m_dalign && igeo->inoalign_mask &&
2891	    !(mp->m_dalign & igeo->inoalign_mask))
2892		igeo->ialloc_align = mp->m_dalign;
2893	else
2894		igeo->ialloc_align = 0;
2895}
2896
2897/* Compute the location of the root directory inode that is laid out by mkfs. */
2898xfs_ino_t
2899xfs_ialloc_calc_rootino(
2900	struct xfs_mount	*mp,
2901	int			sunit)
2902{
2903	struct xfs_ino_geometry	*igeo = M_IGEO(mp);
2904	xfs_agblock_t		first_bno;
2905
2906	/*
2907	 * Pre-calculate the geometry of AG 0.  We know what it looks like
2908	 * because libxfs knows how to create allocation groups now.
2909	 *
2910	 * first_bno is the first block in which mkfs could possibly have
2911	 * allocated the root directory inode, once we factor in the metadata
2912	 * that mkfs formats before it.  Namely, the four AG headers...
2913	 */
2914	first_bno = howmany(4 * mp->m_sb.sb_sectsize, mp->m_sb.sb_blocksize);
2915
2916	/* ...the two free space btree roots... */
2917	first_bno += 2;
2918
2919	/* ...the inode btree root... */
2920	first_bno += 1;
2921
2922	/* ...the initial AGFL... */
2923	first_bno += xfs_alloc_min_freelist(mp, NULL);
2924
2925	/* ...the free inode btree root... */
2926	if (xfs_has_finobt(mp))
2927		first_bno++;
2928
2929	/* ...the reverse mapping btree root... */
2930	if (xfs_has_rmapbt(mp))
2931		first_bno++;
2932
2933	/* ...the reference count btree... */
2934	if (xfs_has_reflink(mp))
2935		first_bno++;
2936
2937	/*
2938	 * ...and the log, if it is allocated in the first allocation group.
2939	 *
2940	 * This can happen with filesystems that only have a single
2941	 * allocation group, or very odd geometries created by old mkfs
2942	 * versions on very small filesystems.
2943	 */
2944	if (xfs_ag_contains_log(mp, 0))
2945		 first_bno += mp->m_sb.sb_logblocks;
2946
2947	/*
2948	 * Now round first_bno up to whatever allocation alignment is given
2949	 * by the filesystem or was passed in.
2950	 */
2951	if (xfs_has_dalign(mp) && igeo->ialloc_align > 0)
2952		first_bno = roundup(first_bno, sunit);
2953	else if (xfs_has_align(mp) &&
2954			mp->m_sb.sb_inoalignmt > 1)
2955		first_bno = roundup(first_bno, mp->m_sb.sb_inoalignmt);
2956
2957	return XFS_AGINO_TO_INO(mp, 0, XFS_AGB_TO_AGINO(mp, first_bno));
2958}
2959
2960/*
2961 * Ensure there are not sparse inode clusters that cross the new EOAG.
2962 *
2963 * This is a no-op for non-spinode filesystems since clusters are always fully
2964 * allocated and checking the bnobt suffices.  However, a spinode filesystem
2965 * could have a record where the upper inodes are free blocks.  If those blocks
2966 * were removed from the filesystem, the inode record would extend beyond EOAG,
2967 * which will be flagged as corruption.
2968 */
2969int
2970xfs_ialloc_check_shrink(
2971	struct xfs_perag	*pag,
2972	struct xfs_trans	*tp,
2973	struct xfs_buf		*agibp,
2974	xfs_agblock_t		new_length)
2975{
2976	struct xfs_inobt_rec_incore rec;
2977	struct xfs_btree_cur	*cur;
2978	xfs_agino_t		agino;
2979	int			has;
2980	int			error;
2981
2982	if (!xfs_has_sparseinodes(pag->pag_mount))
2983		return 0;
2984
2985	cur = xfs_inobt_init_cursor(pag, tp, agibp, XFS_BTNUM_INO);
2986
2987	/* Look up the inobt record that would correspond to the new EOFS. */
2988	agino = XFS_AGB_TO_AGINO(pag->pag_mount, new_length);
2989	error = xfs_inobt_lookup(cur, agino, XFS_LOOKUP_LE, &has);
2990	if (error || !has)
2991		goto out;
2992
2993	error = xfs_inobt_get_rec(cur, &rec, &has);
2994	if (error)
2995		goto out;
2996
2997	if (!has) {
2998		error = -EFSCORRUPTED;
2999		goto out;
3000	}
3001
3002	/* If the record covers inodes that would be beyond EOFS, bail out. */
3003	if (rec.ir_startino + XFS_INODES_PER_CHUNK > agino) {
3004		error = -ENOSPC;
3005		goto out;
3006	}
3007out:
3008	xfs_btree_del_cursor(cur, error);
3009	return error;
3010}