Linux Audio

Check our new training course

Loading...
v5.4
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 *   Copyright (C) International Business Machines Corp., 2000-2004
 
 
 
 
 
 
 
 
 
 
 
 
 
 
   4 */
   5
   6/*
   7 *	jfs_imap.c: inode allocation map manager
   8 *
   9 * Serialization:
  10 *   Each AG has a simple lock which is used to control the serialization of
  11 *	the AG level lists.  This lock should be taken first whenever an AG
  12 *	level list will be modified or accessed.
  13 *
  14 *   Each IAG is locked by obtaining the buffer for the IAG page.
  15 *
  16 *   There is also a inode lock for the inode map inode.  A read lock needs to
  17 *	be taken whenever an IAG is read from the map or the global level
  18 *	information is read.  A write lock needs to be taken whenever the global
  19 *	level information is modified or an atomic operation needs to be used.
  20 *
  21 *	If more than one IAG is read at one time, the read lock may not
  22 *	be given up until all of the IAG's are read.  Otherwise, a deadlock
  23 *	may occur when trying to obtain the read lock while another thread
  24 *	holding the read lock is waiting on the IAG already being held.
  25 *
  26 *   The control page of the inode map is read into memory by diMount().
  27 *	Thereafter it should only be modified in memory and then it will be
  28 *	written out when the filesystem is unmounted by diUnmount().
  29 */
  30
  31#include <linux/fs.h>
  32#include <linux/buffer_head.h>
  33#include <linux/pagemap.h>
  34#include <linux/quotaops.h>
  35#include <linux/slab.h>
  36
  37#include "jfs_incore.h"
  38#include "jfs_inode.h"
  39#include "jfs_filsys.h"
  40#include "jfs_dinode.h"
  41#include "jfs_dmap.h"
  42#include "jfs_imap.h"
  43#include "jfs_metapage.h"
  44#include "jfs_superblock.h"
  45#include "jfs_debug.h"
  46
  47/*
  48 * imap locks
  49 */
  50/* iag free list lock */
  51#define IAGFREE_LOCK_INIT(imap)		mutex_init(&imap->im_freelock)
  52#define IAGFREE_LOCK(imap)		mutex_lock(&imap->im_freelock)
  53#define IAGFREE_UNLOCK(imap)		mutex_unlock(&imap->im_freelock)
  54
  55/* per ag iag list locks */
  56#define AG_LOCK_INIT(imap,index)	mutex_init(&(imap->im_aglock[index]))
  57#define AG_LOCK(imap,agno)		mutex_lock(&imap->im_aglock[agno])
  58#define AG_UNLOCK(imap,agno)		mutex_unlock(&imap->im_aglock[agno])
  59
  60/*
  61 * forward references
  62 */
  63static int diAllocAG(struct inomap *, int, bool, struct inode *);
  64static int diAllocAny(struct inomap *, int, bool, struct inode *);
  65static int diAllocBit(struct inomap *, struct iag *, int);
  66static int diAllocExt(struct inomap *, int, struct inode *);
  67static int diAllocIno(struct inomap *, int, struct inode *);
  68static int diFindFree(u32, int);
  69static int diNewExt(struct inomap *, struct iag *, int);
  70static int diNewIAG(struct inomap *, int *, int, struct metapage **);
  71static void duplicateIXtree(struct super_block *, s64, int, s64 *);
  72
  73static int diIAGRead(struct inomap * imap, int, struct metapage **);
  74static int copy_from_dinode(struct dinode *, struct inode *);
  75static void copy_to_dinode(struct dinode *, struct inode *);
  76
  77/*
  78 * NAME:	diMount()
  79 *
  80 * FUNCTION:	initialize the incore inode map control structures for
  81 *		a fileset or aggregate init time.
  82 *
  83 *		the inode map's control structure (dinomap) is
  84 *		brought in from disk and placed in virtual memory.
  85 *
  86 * PARAMETERS:
  87 *	ipimap	- pointer to inode map inode for the aggregate or fileset.
  88 *
  89 * RETURN VALUES:
  90 *	0	- success
  91 *	-ENOMEM	- insufficient free virtual memory.
  92 *	-EIO	- i/o error.
  93 */
  94int diMount(struct inode *ipimap)
  95{
  96	struct inomap *imap;
  97	struct metapage *mp;
  98	int index;
  99	struct dinomap_disk *dinom_le;
 100
 101	/*
 102	 * allocate/initialize the in-memory inode map control structure
 103	 */
 104	/* allocate the in-memory inode map control structure. */
 105	imap = kmalloc(sizeof(struct inomap), GFP_KERNEL);
 106	if (imap == NULL) {
 107		jfs_err("diMount: kmalloc returned NULL!");
 108		return -ENOMEM;
 109	}
 110
 111	/* read the on-disk inode map control structure. */
 112
 113	mp = read_metapage(ipimap,
 114			   IMAPBLKNO << JFS_SBI(ipimap->i_sb)->l2nbperpage,
 115			   PSIZE, 0);
 116	if (mp == NULL) {
 117		kfree(imap);
 118		return -EIO;
 119	}
 120
 121	/* copy the on-disk version to the in-memory version. */
 122	dinom_le = (struct dinomap_disk *) mp->data;
 123	imap->im_freeiag = le32_to_cpu(dinom_le->in_freeiag);
 124	imap->im_nextiag = le32_to_cpu(dinom_le->in_nextiag);
 125	atomic_set(&imap->im_numinos, le32_to_cpu(dinom_le->in_numinos));
 126	atomic_set(&imap->im_numfree, le32_to_cpu(dinom_le->in_numfree));
 127	imap->im_nbperiext = le32_to_cpu(dinom_le->in_nbperiext);
 128	imap->im_l2nbperiext = le32_to_cpu(dinom_le->in_l2nbperiext);
 129	for (index = 0; index < MAXAG; index++) {
 130		imap->im_agctl[index].inofree =
 131		    le32_to_cpu(dinom_le->in_agctl[index].inofree);
 132		imap->im_agctl[index].extfree =
 133		    le32_to_cpu(dinom_le->in_agctl[index].extfree);
 134		imap->im_agctl[index].numinos =
 135		    le32_to_cpu(dinom_le->in_agctl[index].numinos);
 136		imap->im_agctl[index].numfree =
 137		    le32_to_cpu(dinom_le->in_agctl[index].numfree);
 138	}
 139
 140	/* release the buffer. */
 141	release_metapage(mp);
 142
 143	/*
 144	 * allocate/initialize inode allocation map locks
 145	 */
 146	/* allocate and init iag free list lock */
 147	IAGFREE_LOCK_INIT(imap);
 148
 149	/* allocate and init ag list locks */
 150	for (index = 0; index < MAXAG; index++) {
 151		AG_LOCK_INIT(imap, index);
 152	}
 153
 154	/* bind the inode map inode and inode map control structure
 155	 * to each other.
 156	 */
 157	imap->im_ipimap = ipimap;
 158	JFS_IP(ipimap)->i_imap = imap;
 159
 160	return (0);
 161}
 162
 163
 164/*
 165 * NAME:	diUnmount()
 166 *
 167 * FUNCTION:	write to disk the incore inode map control structures for
 168 *		a fileset or aggregate at unmount time.
 169 *
 170 * PARAMETERS:
 171 *	ipimap	- pointer to inode map inode for the aggregate or fileset.
 172 *
 173 * RETURN VALUES:
 174 *	0	- success
 175 *	-ENOMEM	- insufficient free virtual memory.
 176 *	-EIO	- i/o error.
 177 */
 178int diUnmount(struct inode *ipimap, int mounterror)
 179{
 180	struct inomap *imap = JFS_IP(ipimap)->i_imap;
 181
 182	/*
 183	 * update the on-disk inode map control structure
 184	 */
 185
 186	if (!(mounterror || isReadOnly(ipimap)))
 187		diSync(ipimap);
 188
 189	/*
 190	 * Invalidate the page cache buffers
 191	 */
 192	truncate_inode_pages(ipimap->i_mapping, 0);
 193
 194	/*
 195	 * free in-memory control structure
 196	 */
 197	kfree(imap);
 198
 199	return (0);
 200}
 201
 202
 203/*
 204 *	diSync()
 205 */
 206int diSync(struct inode *ipimap)
 207{
 208	struct dinomap_disk *dinom_le;
 209	struct inomap *imp = JFS_IP(ipimap)->i_imap;
 210	struct metapage *mp;
 211	int index;
 212
 213	/*
 214	 * write imap global conrol page
 215	 */
 216	/* read the on-disk inode map control structure */
 217	mp = get_metapage(ipimap,
 218			  IMAPBLKNO << JFS_SBI(ipimap->i_sb)->l2nbperpage,
 219			  PSIZE, 0);
 220	if (mp == NULL) {
 221		jfs_err("diSync: get_metapage failed!");
 222		return -EIO;
 223	}
 224
 225	/* copy the in-memory version to the on-disk version */
 226	dinom_le = (struct dinomap_disk *) mp->data;
 227	dinom_le->in_freeiag = cpu_to_le32(imp->im_freeiag);
 228	dinom_le->in_nextiag = cpu_to_le32(imp->im_nextiag);
 229	dinom_le->in_numinos = cpu_to_le32(atomic_read(&imp->im_numinos));
 230	dinom_le->in_numfree = cpu_to_le32(atomic_read(&imp->im_numfree));
 231	dinom_le->in_nbperiext = cpu_to_le32(imp->im_nbperiext);
 232	dinom_le->in_l2nbperiext = cpu_to_le32(imp->im_l2nbperiext);
 233	for (index = 0; index < MAXAG; index++) {
 234		dinom_le->in_agctl[index].inofree =
 235		    cpu_to_le32(imp->im_agctl[index].inofree);
 236		dinom_le->in_agctl[index].extfree =
 237		    cpu_to_le32(imp->im_agctl[index].extfree);
 238		dinom_le->in_agctl[index].numinos =
 239		    cpu_to_le32(imp->im_agctl[index].numinos);
 240		dinom_le->in_agctl[index].numfree =
 241		    cpu_to_le32(imp->im_agctl[index].numfree);
 242	}
 243
 244	/* write out the control structure */
 245	write_metapage(mp);
 246
 247	/*
 248	 * write out dirty pages of imap
 249	 */
 250	filemap_write_and_wait(ipimap->i_mapping);
 251
 252	diWriteSpecial(ipimap, 0);
 253
 254	return (0);
 255}
 256
 257
 258/*
 259 * NAME:	diRead()
 260 *
 261 * FUNCTION:	initialize an incore inode from disk.
 262 *
 263 *		on entry, the specifed incore inode should itself
 264 *		specify the disk inode number corresponding to the
 265 *		incore inode (i.e. i_number should be initialized).
 266 *
 267 *		this routine handles incore inode initialization for
 268 *		both "special" and "regular" inodes.  special inodes
 269 *		are those required early in the mount process and
 270 *		require special handling since much of the file system
 271 *		is not yet initialized.  these "special" inodes are
 272 *		identified by a NULL inode map inode pointer and are
 273 *		actually initialized by a call to diReadSpecial().
 274 *
 275 *		for regular inodes, the iag describing the disk inode
 276 *		is read from disk to determine the inode extent address
 277 *		for the disk inode.  with the inode extent address in
 278 *		hand, the page of the extent that contains the disk
 279 *		inode is read and the disk inode is copied to the
 280 *		incore inode.
 281 *
 282 * PARAMETERS:
 283 *	ip	-  pointer to incore inode to be initialized from disk.
 284 *
 285 * RETURN VALUES:
 286 *	0	- success
 287 *	-EIO	- i/o error.
 288 *	-ENOMEM	- insufficient memory
 289 *
 290 */
 291int diRead(struct inode *ip)
 292{
 293	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
 294	int iagno, ino, extno, rc;
 295	struct inode *ipimap;
 296	struct dinode *dp;
 297	struct iag *iagp;
 298	struct metapage *mp;
 299	s64 blkno, agstart;
 300	struct inomap *imap;
 301	int block_offset;
 302	int inodes_left;
 303	unsigned long pageno;
 304	int rel_inode;
 305
 306	jfs_info("diRead: ino = %ld", ip->i_ino);
 307
 308	ipimap = sbi->ipimap;
 309	JFS_IP(ip)->ipimap = ipimap;
 310
 311	/* determine the iag number for this inode (number) */
 312	iagno = INOTOIAG(ip->i_ino);
 313
 314	/* read the iag */
 315	imap = JFS_IP(ipimap)->i_imap;
 316	IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
 317	rc = diIAGRead(imap, iagno, &mp);
 318	IREAD_UNLOCK(ipimap);
 319	if (rc) {
 320		jfs_err("diRead: diIAGRead returned %d", rc);
 321		return (rc);
 322	}
 323
 324	iagp = (struct iag *) mp->data;
 325
 326	/* determine inode extent that holds the disk inode */
 327	ino = ip->i_ino & (INOSPERIAG - 1);
 328	extno = ino >> L2INOSPEREXT;
 329
 330	if ((lengthPXD(&iagp->inoext[extno]) != imap->im_nbperiext) ||
 331	    (addressPXD(&iagp->inoext[extno]) == 0)) {
 332		release_metapage(mp);
 333		return -ESTALE;
 334	}
 335
 336	/* get disk block number of the page within the inode extent
 337	 * that holds the disk inode.
 338	 */
 339	blkno = INOPBLK(&iagp->inoext[extno], ino, sbi->l2nbperpage);
 340
 341	/* get the ag for the iag */
 342	agstart = le64_to_cpu(iagp->agstart);
 343
 344	release_metapage(mp);
 345
 346	rel_inode = (ino & (INOSPERPAGE - 1));
 347	pageno = blkno >> sbi->l2nbperpage;
 348
 349	if ((block_offset = ((u32) blkno & (sbi->nbperpage - 1)))) {
 350		/*
 351		 * OS/2 didn't always align inode extents on page boundaries
 352		 */
 353		inodes_left =
 354		     (sbi->nbperpage - block_offset) << sbi->l2niperblk;
 355
 356		if (rel_inode < inodes_left)
 357			rel_inode += block_offset << sbi->l2niperblk;
 358		else {
 359			pageno += 1;
 360			rel_inode -= inodes_left;
 361		}
 362	}
 363
 364	/* read the page of disk inode */
 365	mp = read_metapage(ipimap, pageno << sbi->l2nbperpage, PSIZE, 1);
 366	if (!mp) {
 367		jfs_err("diRead: read_metapage failed");
 368		return -EIO;
 369	}
 370
 371	/* locate the disk inode requested */
 372	dp = (struct dinode *) mp->data;
 373	dp += rel_inode;
 374
 375	if (ip->i_ino != le32_to_cpu(dp->di_number)) {
 376		jfs_error(ip->i_sb, "i_ino != di_number\n");
 377		rc = -EIO;
 378	} else if (le32_to_cpu(dp->di_nlink) == 0)
 379		rc = -ESTALE;
 380	else
 381		/* copy the disk inode to the in-memory inode */
 382		rc = copy_from_dinode(dp, ip);
 383
 384	release_metapage(mp);
 385
 386	/* set the ag for the inode */
 387	JFS_IP(ip)->agstart = agstart;
 388	JFS_IP(ip)->active_ag = -1;
 389
 390	return (rc);
 391}
 392
 393
 394/*
 395 * NAME:	diReadSpecial()
 396 *
 397 * FUNCTION:	initialize a 'special' inode from disk.
 398 *
 399 *		this routines handles aggregate level inodes.  The
 400 *		inode cache cannot differentiate between the
 401 *		aggregate inodes and the filesystem inodes, so we
 402 *		handle these here.  We don't actually use the aggregate
 403 *		inode map, since these inodes are at a fixed location
 404 *		and in some cases the aggregate inode map isn't initialized
 405 *		yet.
 406 *
 407 * PARAMETERS:
 408 *	sb - filesystem superblock
 409 *	inum - aggregate inode number
 410 *	secondary - 1 if secondary aggregate inode table
 411 *
 412 * RETURN VALUES:
 413 *	new inode	- success
 414 *	NULL		- i/o error.
 415 */
 416struct inode *diReadSpecial(struct super_block *sb, ino_t inum, int secondary)
 417{
 418	struct jfs_sb_info *sbi = JFS_SBI(sb);
 419	uint address;
 420	struct dinode *dp;
 421	struct inode *ip;
 422	struct metapage *mp;
 423
 424	ip = new_inode(sb);
 425	if (ip == NULL) {
 426		jfs_err("diReadSpecial: new_inode returned NULL!");
 427		return ip;
 428	}
 429
 430	if (secondary) {
 431		address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage;
 432		JFS_IP(ip)->ipimap = sbi->ipaimap2;
 433	} else {
 434		address = AITBL_OFF >> L2PSIZE;
 435		JFS_IP(ip)->ipimap = sbi->ipaimap;
 436	}
 437
 438	ASSERT(inum < INOSPEREXT);
 439
 440	ip->i_ino = inum;
 441
 442	address += inum >> 3;	/* 8 inodes per 4K page */
 443
 444	/* read the page of fixed disk inode (AIT) in raw mode */
 445	mp = read_metapage(ip, address << sbi->l2nbperpage, PSIZE, 1);
 446	if (mp == NULL) {
 447		set_nlink(ip, 1);	/* Don't want iput() deleting it */
 448		iput(ip);
 449		return (NULL);
 450	}
 451
 452	/* get the pointer to the disk inode of interest */
 453	dp = (struct dinode *) (mp->data);
 454	dp += inum % 8;		/* 8 inodes per 4K page */
 455
 456	/* copy on-disk inode to in-memory inode */
 457	if ((copy_from_dinode(dp, ip)) != 0) {
 458		/* handle bad return by returning NULL for ip */
 459		set_nlink(ip, 1);	/* Don't want iput() deleting it */
 460		iput(ip);
 461		/* release the page */
 462		release_metapage(mp);
 463		return (NULL);
 464
 465	}
 466
 467	ip->i_mapping->a_ops = &jfs_metapage_aops;
 468	mapping_set_gfp_mask(ip->i_mapping, GFP_NOFS);
 469
 470	/* Allocations to metadata inodes should not affect quotas */
 471	ip->i_flags |= S_NOQUOTA;
 472
 473	if ((inum == FILESYSTEM_I) && (JFS_IP(ip)->ipimap == sbi->ipaimap)) {
 474		sbi->gengen = le32_to_cpu(dp->di_gengen);
 475		sbi->inostamp = le32_to_cpu(dp->di_inostamp);
 476	}
 477
 478	/* release the page */
 479	release_metapage(mp);
 480
 481	inode_fake_hash(ip);
 
 
 
 
 
 
 482
 483	return (ip);
 484}
 485
 486/*
 487 * NAME:	diWriteSpecial()
 488 *
 489 * FUNCTION:	Write the special inode to disk
 490 *
 491 * PARAMETERS:
 492 *	ip - special inode
 493 *	secondary - 1 if secondary aggregate inode table
 494 *
 495 * RETURN VALUES: none
 496 */
 497
 498void diWriteSpecial(struct inode *ip, int secondary)
 499{
 500	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
 501	uint address;
 502	struct dinode *dp;
 503	ino_t inum = ip->i_ino;
 504	struct metapage *mp;
 505
 506	if (secondary)
 507		address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage;
 508	else
 509		address = AITBL_OFF >> L2PSIZE;
 510
 511	ASSERT(inum < INOSPEREXT);
 512
 513	address += inum >> 3;	/* 8 inodes per 4K page */
 514
 515	/* read the page of fixed disk inode (AIT) in raw mode */
 516	mp = read_metapage(ip, address << sbi->l2nbperpage, PSIZE, 1);
 517	if (mp == NULL) {
 518		jfs_err("diWriteSpecial: failed to read aggregate inode extent!");
 
 519		return;
 520	}
 521
 522	/* get the pointer to the disk inode of interest */
 523	dp = (struct dinode *) (mp->data);
 524	dp += inum % 8;		/* 8 inodes per 4K page */
 525
 526	/* copy on-disk inode to in-memory inode */
 527	copy_to_dinode(dp, ip);
 528	memcpy(&dp->di_xtroot, &JFS_IP(ip)->i_xtroot, 288);
 529
 530	if (inum == FILESYSTEM_I)
 531		dp->di_gengen = cpu_to_le32(sbi->gengen);
 532
 533	/* write the page */
 534	write_metapage(mp);
 535}
 536
 537/*
 538 * NAME:	diFreeSpecial()
 539 *
 540 * FUNCTION:	Free allocated space for special inode
 541 */
 542void diFreeSpecial(struct inode *ip)
 543{
 544	if (ip == NULL) {
 545		jfs_err("diFreeSpecial called with NULL ip!");
 546		return;
 547	}
 548	filemap_write_and_wait(ip->i_mapping);
 549	truncate_inode_pages(ip->i_mapping, 0);
 550	iput(ip);
 551}
 552
 553
 554
 555/*
 556 * NAME:	diWrite()
 557 *
 558 * FUNCTION:	write the on-disk inode portion of the in-memory inode
 559 *		to its corresponding on-disk inode.
 560 *
 561 *		on entry, the specifed incore inode should itself
 562 *		specify the disk inode number corresponding to the
 563 *		incore inode (i.e. i_number should be initialized).
 564 *
 565 *		the inode contains the inode extent address for the disk
 566 *		inode.  with the inode extent address in hand, the
 567 *		page of the extent that contains the disk inode is
 568 *		read and the disk inode portion of the incore inode
 569 *		is copied to the disk inode.
 570 *
 571 * PARAMETERS:
 572 *	tid -  transacation id
 573 *	ip  -  pointer to incore inode to be written to the inode extent.
 574 *
 575 * RETURN VALUES:
 576 *	0	- success
 577 *	-EIO	- i/o error.
 578 */
 579int diWrite(tid_t tid, struct inode *ip)
 580{
 581	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
 582	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
 583	int rc = 0;
 584	s32 ino;
 585	struct dinode *dp;
 586	s64 blkno;
 587	int block_offset;
 588	int inodes_left;
 589	struct metapage *mp;
 590	unsigned long pageno;
 591	int rel_inode;
 592	int dioffset;
 593	struct inode *ipimap;
 594	uint type;
 595	lid_t lid;
 596	struct tlock *ditlck, *tlck;
 597	struct linelock *dilinelock, *ilinelock;
 598	struct lv *lv;
 599	int n;
 600
 601	ipimap = jfs_ip->ipimap;
 602
 603	ino = ip->i_ino & (INOSPERIAG - 1);
 604
 605	if (!addressPXD(&(jfs_ip->ixpxd)) ||
 606	    (lengthPXD(&(jfs_ip->ixpxd)) !=
 607	     JFS_IP(ipimap)->i_imap->im_nbperiext)) {
 608		jfs_error(ip->i_sb, "ixpxd invalid\n");
 609		return -EIO;
 610	}
 611
 612	/*
 613	 * read the page of disk inode containing the specified inode:
 614	 */
 615	/* compute the block address of the page */
 616	blkno = INOPBLK(&(jfs_ip->ixpxd), ino, sbi->l2nbperpage);
 617
 618	rel_inode = (ino & (INOSPERPAGE - 1));
 619	pageno = blkno >> sbi->l2nbperpage;
 620
 621	if ((block_offset = ((u32) blkno & (sbi->nbperpage - 1)))) {
 622		/*
 623		 * OS/2 didn't always align inode extents on page boundaries
 624		 */
 625		inodes_left =
 626		    (sbi->nbperpage - block_offset) << sbi->l2niperblk;
 627
 628		if (rel_inode < inodes_left)
 629			rel_inode += block_offset << sbi->l2niperblk;
 630		else {
 631			pageno += 1;
 632			rel_inode -= inodes_left;
 633		}
 634	}
 635	/* read the page of disk inode */
 636      retry:
 637	mp = read_metapage(ipimap, pageno << sbi->l2nbperpage, PSIZE, 1);
 638	if (!mp)
 639		return -EIO;
 640
 641	/* get the pointer to the disk inode */
 642	dp = (struct dinode *) mp->data;
 643	dp += rel_inode;
 644
 645	dioffset = (ino & (INOSPERPAGE - 1)) << L2DISIZE;
 646
 647	/*
 648	 * acquire transaction lock on the on-disk inode;
 649	 * N.B. tlock is acquired on ipimap not ip;
 650	 */
 651	if ((ditlck =
 652	     txLock(tid, ipimap, mp, tlckINODE | tlckENTRY)) == NULL)
 653		goto retry;
 654	dilinelock = (struct linelock *) & ditlck->lock;
 655
 656	/*
 657	 * copy btree root from in-memory inode to on-disk inode
 658	 *
 659	 * (tlock is taken from inline B+-tree root in in-memory
 660	 * inode when the B+-tree root is updated, which is pointed
 661	 * by jfs_ip->blid as well as being on tx tlock list)
 662	 *
 663	 * further processing of btree root is based on the copy
 664	 * in in-memory inode, where txLog() will log from, and,
 665	 * for xtree root, txUpdateMap() will update map and reset
 666	 * XAD_NEW bit;
 667	 */
 668
 669	if (S_ISDIR(ip->i_mode) && (lid = jfs_ip->xtlid)) {
 670		/*
 671		 * This is the special xtree inside the directory for storing
 672		 * the directory table
 673		 */
 674		xtpage_t *p, *xp;
 675		xad_t *xad;
 676
 677		jfs_ip->xtlid = 0;
 678		tlck = lid_to_tlock(lid);
 679		assert(tlck->type & tlckXTREE);
 680		tlck->type |= tlckBTROOT;
 681		tlck->mp = mp;
 682		ilinelock = (struct linelock *) & tlck->lock;
 683
 684		/*
 685		 * copy xtree root from inode to dinode:
 686		 */
 687		p = &jfs_ip->i_xtroot;
 688		xp = (xtpage_t *) &dp->di_dirtable;
 689		lv = ilinelock->lv;
 690		for (n = 0; n < ilinelock->index; n++, lv++) {
 691			memcpy(&xp->xad[lv->offset], &p->xad[lv->offset],
 692			       lv->length << L2XTSLOTSIZE);
 693		}
 694
 695		/* reset on-disk (metadata page) xtree XAD_NEW bit */
 696		xad = &xp->xad[XTENTRYSTART];
 697		for (n = XTENTRYSTART;
 698		     n < le16_to_cpu(xp->header.nextindex); n++, xad++)
 699			if (xad->flag & (XAD_NEW | XAD_EXTENDED))
 700				xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
 701	}
 702
 703	if ((lid = jfs_ip->blid) == 0)
 704		goto inlineData;
 705	jfs_ip->blid = 0;
 706
 707	tlck = lid_to_tlock(lid);
 708	type = tlck->type;
 709	tlck->type |= tlckBTROOT;
 710	tlck->mp = mp;
 711	ilinelock = (struct linelock *) & tlck->lock;
 712
 713	/*
 714	 *	regular file: 16 byte (XAD slot) granularity
 715	 */
 716	if (type & tlckXTREE) {
 717		xtpage_t *p, *xp;
 718		xad_t *xad;
 719
 720		/*
 721		 * copy xtree root from inode to dinode:
 722		 */
 723		p = &jfs_ip->i_xtroot;
 724		xp = &dp->di_xtroot;
 725		lv = ilinelock->lv;
 726		for (n = 0; n < ilinelock->index; n++, lv++) {
 727			memcpy(&xp->xad[lv->offset], &p->xad[lv->offset],
 728			       lv->length << L2XTSLOTSIZE);
 729		}
 730
 731		/* reset on-disk (metadata page) xtree XAD_NEW bit */
 732		xad = &xp->xad[XTENTRYSTART];
 733		for (n = XTENTRYSTART;
 734		     n < le16_to_cpu(xp->header.nextindex); n++, xad++)
 735			if (xad->flag & (XAD_NEW | XAD_EXTENDED))
 736				xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
 737	}
 738	/*
 739	 *	directory: 32 byte (directory entry slot) granularity
 740	 */
 741	else if (type & tlckDTREE) {
 742		dtpage_t *p, *xp;
 743
 744		/*
 745		 * copy dtree root from inode to dinode:
 746		 */
 747		p = (dtpage_t *) &jfs_ip->i_dtroot;
 748		xp = (dtpage_t *) & dp->di_dtroot;
 749		lv = ilinelock->lv;
 750		for (n = 0; n < ilinelock->index; n++, lv++) {
 751			memcpy(&xp->slot[lv->offset], &p->slot[lv->offset],
 752			       lv->length << L2DTSLOTSIZE);
 753		}
 754	} else {
 755		jfs_err("diWrite: UFO tlock");
 756	}
 757
 758      inlineData:
 759	/*
 760	 * copy inline symlink from in-memory inode to on-disk inode
 761	 */
 762	if (S_ISLNK(ip->i_mode) && ip->i_size < IDATASIZE) {
 763		lv = & dilinelock->lv[dilinelock->index];
 764		lv->offset = (dioffset + 2 * 128) >> L2INODESLOTSIZE;
 765		lv->length = 2;
 766		memcpy(&dp->di_fastsymlink, jfs_ip->i_inline, IDATASIZE);
 767		dilinelock->index++;
 768	}
 769	/*
 770	 * copy inline data from in-memory inode to on-disk inode:
 771	 * 128 byte slot granularity
 772	 */
 773	if (test_cflag(COMMIT_Inlineea, ip)) {
 774		lv = & dilinelock->lv[dilinelock->index];
 775		lv->offset = (dioffset + 3 * 128) >> L2INODESLOTSIZE;
 776		lv->length = 1;
 777		memcpy(&dp->di_inlineea, jfs_ip->i_inline_ea, INODESLOTSIZE);
 778		dilinelock->index++;
 779
 780		clear_cflag(COMMIT_Inlineea, ip);
 781	}
 782
 783	/*
 784	 *	lock/copy inode base: 128 byte slot granularity
 785	 */
 786	lv = & dilinelock->lv[dilinelock->index];
 787	lv->offset = dioffset >> L2INODESLOTSIZE;
 788	copy_to_dinode(dp, ip);
 789	if (test_and_clear_cflag(COMMIT_Dirtable, ip)) {
 790		lv->length = 2;
 791		memcpy(&dp->di_dirtable, &jfs_ip->i_dirtable, 96);
 792	} else
 793		lv->length = 1;
 794	dilinelock->index++;
 795
 796	/* release the buffer holding the updated on-disk inode.
 797	 * the buffer will be later written by commit processing.
 798	 */
 799	write_metapage(mp);
 800
 801	return (rc);
 802}
 803
 804
 805/*
 806 * NAME:	diFree(ip)
 807 *
 808 * FUNCTION:	free a specified inode from the inode working map
 809 *		for a fileset or aggregate.
 810 *
 811 *		if the inode to be freed represents the first (only)
 812 *		free inode within the iag, the iag will be placed on
 813 *		the ag free inode list.
 814 *
 815 *		freeing the inode will cause the inode extent to be
 816 *		freed if the inode is the only allocated inode within
 817 *		the extent.  in this case all the disk resource backing
 818 *		up the inode extent will be freed. in addition, the iag
 819 *		will be placed on the ag extent free list if the extent
 820 *		is the first free extent in the iag.  if freeing the
 821 *		extent also means that no free inodes will exist for
 822 *		the iag, the iag will also be removed from the ag free
 823 *		inode list.
 824 *
 825 *		the iag describing the inode will be freed if the extent
 826 *		is to be freed and it is the only backed extent within
 827 *		the iag.  in this case, the iag will be removed from the
 828 *		ag free extent list and ag free inode list and placed on
 829 *		the inode map's free iag list.
 830 *
 831 *		a careful update approach is used to provide consistency
 832 *		in the face of updates to multiple buffers.  under this
 833 *		approach, all required buffers are obtained before making
 834 *		any updates and are held until all updates are complete.
 835 *
 836 * PARAMETERS:
 837 *	ip	- inode to be freed.
 838 *
 839 * RETURN VALUES:
 840 *	0	- success
 841 *	-EIO	- i/o error.
 842 */
 843int diFree(struct inode *ip)
 844{
 845	int rc;
 846	ino_t inum = ip->i_ino;
 847	struct iag *iagp, *aiagp, *biagp, *ciagp, *diagp;
 848	struct metapage *mp, *amp, *bmp, *cmp, *dmp;
 849	int iagno, ino, extno, bitno, sword, agno;
 850	int back, fwd;
 851	u32 bitmap, mask;
 852	struct inode *ipimap = JFS_SBI(ip->i_sb)->ipimap;
 853	struct inomap *imap = JFS_IP(ipimap)->i_imap;
 854	pxd_t freepxd;
 855	tid_t tid;
 856	struct inode *iplist[3];
 857	struct tlock *tlck;
 858	struct pxd_lock *pxdlock;
 859
 860	/*
 861	 * This is just to suppress compiler warnings.  The same logic that
 862	 * references these variables is used to initialize them.
 863	 */
 864	aiagp = biagp = ciagp = diagp = NULL;
 865
 866	/* get the iag number containing the inode.
 867	 */
 868	iagno = INOTOIAG(inum);
 869
 870	/* make sure that the iag is contained within
 871	 * the map.
 872	 */
 873	if (iagno >= imap->im_nextiag) {
 874		print_hex_dump(KERN_ERR, "imap: ", DUMP_PREFIX_ADDRESS, 16, 4,
 875			       imap, 32, 0);
 876		jfs_error(ip->i_sb, "inum = %d, iagno = %d, nextiag = %d\n",
 
 877			  (uint) inum, iagno, imap->im_nextiag);
 878		return -EIO;
 879	}
 880
 881	/* get the allocation group for this ino.
 882	 */
 883	agno = BLKTOAG(JFS_IP(ip)->agstart, JFS_SBI(ip->i_sb));
 884
 885	/* Lock the AG specific inode map information
 886	 */
 887	AG_LOCK(imap, agno);
 888
 889	/* Obtain read lock in imap inode.  Don't release it until we have
 890	 * read all of the IAG's that we are going to.
 891	 */
 892	IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
 893
 894	/* read the iag.
 895	 */
 896	if ((rc = diIAGRead(imap, iagno, &mp))) {
 897		IREAD_UNLOCK(ipimap);
 898		AG_UNLOCK(imap, agno);
 899		return (rc);
 900	}
 901	iagp = (struct iag *) mp->data;
 902
 903	/* get the inode number and extent number of the inode within
 904	 * the iag and the inode number within the extent.
 905	 */
 906	ino = inum & (INOSPERIAG - 1);
 907	extno = ino >> L2INOSPEREXT;
 908	bitno = ino & (INOSPEREXT - 1);
 909	mask = HIGHORDER >> bitno;
 910
 911	if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
 912		jfs_error(ip->i_sb, "wmap shows inode already free\n");
 
 913	}
 914
 915	if (!addressPXD(&iagp->inoext[extno])) {
 916		release_metapage(mp);
 917		IREAD_UNLOCK(ipimap);
 918		AG_UNLOCK(imap, agno);
 919		jfs_error(ip->i_sb, "invalid inoext\n");
 920		return -EIO;
 921	}
 922
 923	/* compute the bitmap for the extent reflecting the freed inode.
 924	 */
 925	bitmap = le32_to_cpu(iagp->wmap[extno]) & ~mask;
 926
 927	if (imap->im_agctl[agno].numfree > imap->im_agctl[agno].numinos) {
 928		release_metapage(mp);
 929		IREAD_UNLOCK(ipimap);
 930		AG_UNLOCK(imap, agno);
 931		jfs_error(ip->i_sb, "numfree > numinos\n");
 932		return -EIO;
 933	}
 934	/*
 935	 *	inode extent still has some inodes or below low water mark:
 936	 *	keep the inode extent;
 937	 */
 938	if (bitmap ||
 939	    imap->im_agctl[agno].numfree < 96 ||
 940	    (imap->im_agctl[agno].numfree < 288 &&
 941	     (((imap->im_agctl[agno].numfree * 100) /
 942	       imap->im_agctl[agno].numinos) <= 25))) {
 943		/* if the iag currently has no free inodes (i.e.,
 944		 * the inode being freed is the first free inode of iag),
 945		 * insert the iag at head of the inode free list for the ag.
 946		 */
 947		if (iagp->nfreeinos == 0) {
 948			/* check if there are any iags on the ag inode
 949			 * free list.  if so, read the first one so that
 950			 * we can link the current iag onto the list at
 951			 * the head.
 952			 */
 953			if ((fwd = imap->im_agctl[agno].inofree) >= 0) {
 954				/* read the iag that currently is the head
 955				 * of the list.
 956				 */
 957				if ((rc = diIAGRead(imap, fwd, &amp))) {
 958					IREAD_UNLOCK(ipimap);
 959					AG_UNLOCK(imap, agno);
 960					release_metapage(mp);
 961					return (rc);
 962				}
 963				aiagp = (struct iag *) amp->data;
 964
 965				/* make current head point back to the iag.
 966				 */
 967				aiagp->inofreeback = cpu_to_le32(iagno);
 968
 969				write_metapage(amp);
 970			}
 971
 972			/* iag points forward to current head and iag
 973			 * becomes the new head of the list.
 974			 */
 975			iagp->inofreefwd =
 976			    cpu_to_le32(imap->im_agctl[agno].inofree);
 977			iagp->inofreeback = cpu_to_le32(-1);
 978			imap->im_agctl[agno].inofree = iagno;
 979		}
 980		IREAD_UNLOCK(ipimap);
 981
 982		/* update the free inode summary map for the extent if
 983		 * freeing the inode means the extent will now have free
 984		 * inodes (i.e., the inode being freed is the first free
 985		 * inode of extent),
 986		 */
 987		if (iagp->wmap[extno] == cpu_to_le32(ONES)) {
 988			sword = extno >> L2EXTSPERSUM;
 989			bitno = extno & (EXTSPERSUM - 1);
 990			iagp->inosmap[sword] &=
 991			    cpu_to_le32(~(HIGHORDER >> bitno));
 992		}
 993
 994		/* update the bitmap.
 995		 */
 996		iagp->wmap[extno] = cpu_to_le32(bitmap);
 997
 998		/* update the free inode counts at the iag, ag and
 999		 * map level.
1000		 */
1001		le32_add_cpu(&iagp->nfreeinos, 1);
1002		imap->im_agctl[agno].numfree += 1;
1003		atomic_inc(&imap->im_numfree);
1004
1005		/* release the AG inode map lock
1006		 */
1007		AG_UNLOCK(imap, agno);
1008
1009		/* write the iag */
1010		write_metapage(mp);
1011
1012		return (0);
1013	}
1014
1015
1016	/*
1017	 *	inode extent has become free and above low water mark:
1018	 *	free the inode extent;
1019	 */
1020
1021	/*
1022	 *	prepare to update iag list(s) (careful update step 1)
1023	 */
1024	amp = bmp = cmp = dmp = NULL;
1025	fwd = back = -1;
1026
1027	/* check if the iag currently has no free extents.  if so,
1028	 * it will be placed on the head of the ag extent free list.
1029	 */
1030	if (iagp->nfreeexts == 0) {
1031		/* check if the ag extent free list has any iags.
1032		 * if so, read the iag at the head of the list now.
1033		 * this (head) iag will be updated later to reflect
1034		 * the addition of the current iag at the head of
1035		 * the list.
1036		 */
1037		if ((fwd = imap->im_agctl[agno].extfree) >= 0) {
1038			if ((rc = diIAGRead(imap, fwd, &amp)))
1039				goto error_out;
1040			aiagp = (struct iag *) amp->data;
1041		}
1042	} else {
1043		/* iag has free extents. check if the addition of a free
1044		 * extent will cause all extents to be free within this
1045		 * iag.  if so, the iag will be removed from the ag extent
1046		 * free list and placed on the inode map's free iag list.
1047		 */
1048		if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) {
1049			/* in preparation for removing the iag from the
1050			 * ag extent free list, read the iags preceding
1051			 * and following the iag on the ag extent free
1052			 * list.
1053			 */
1054			if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) {
1055				if ((rc = diIAGRead(imap, fwd, &amp)))
1056					goto error_out;
1057				aiagp = (struct iag *) amp->data;
1058			}
1059
1060			if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) {
1061				if ((rc = diIAGRead(imap, back, &bmp)))
1062					goto error_out;
1063				biagp = (struct iag *) bmp->data;
1064			}
1065		}
1066	}
1067
1068	/* remove the iag from the ag inode free list if freeing
1069	 * this extent cause the iag to have no free inodes.
1070	 */
1071	if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) {
1072		int inofreeback = le32_to_cpu(iagp->inofreeback);
1073		int inofreefwd = le32_to_cpu(iagp->inofreefwd);
1074
1075		/* in preparation for removing the iag from the
1076		 * ag inode free list, read the iags preceding
1077		 * and following the iag on the ag inode free
1078		 * list.  before reading these iags, we must make
1079		 * sure that we already don't have them in hand
1080		 * from up above, since re-reading an iag (buffer)
1081		 * we are currently holding would cause a deadlock.
1082		 */
1083		if (inofreefwd >= 0) {
1084
1085			if (inofreefwd == fwd)
1086				ciagp = (struct iag *) amp->data;
1087			else if (inofreefwd == back)
1088				ciagp = (struct iag *) bmp->data;
1089			else {
1090				if ((rc =
1091				     diIAGRead(imap, inofreefwd, &cmp)))
1092					goto error_out;
1093				ciagp = (struct iag *) cmp->data;
1094			}
1095			assert(ciagp != NULL);
1096		}
1097
1098		if (inofreeback >= 0) {
1099			if (inofreeback == fwd)
1100				diagp = (struct iag *) amp->data;
1101			else if (inofreeback == back)
1102				diagp = (struct iag *) bmp->data;
1103			else {
1104				if ((rc =
1105				     diIAGRead(imap, inofreeback, &dmp)))
1106					goto error_out;
1107				diagp = (struct iag *) dmp->data;
1108			}
1109			assert(diagp != NULL);
1110		}
1111	}
1112
1113	IREAD_UNLOCK(ipimap);
1114
1115	/*
1116	 * invalidate any page of the inode extent freed from buffer cache;
1117	 */
1118	freepxd = iagp->inoext[extno];
1119	invalidate_pxd_metapages(ip, freepxd);
1120
1121	/*
1122	 *	update iag list(s) (careful update step 2)
1123	 */
1124	/* add the iag to the ag extent free list if this is the
1125	 * first free extent for the iag.
1126	 */
1127	if (iagp->nfreeexts == 0) {
1128		if (fwd >= 0)
1129			aiagp->extfreeback = cpu_to_le32(iagno);
1130
1131		iagp->extfreefwd =
1132		    cpu_to_le32(imap->im_agctl[agno].extfree);
1133		iagp->extfreeback = cpu_to_le32(-1);
1134		imap->im_agctl[agno].extfree = iagno;
1135	} else {
1136		/* remove the iag from the ag extent list if all extents
1137		 * are now free and place it on the inode map iag free list.
1138		 */
1139		if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) {
1140			if (fwd >= 0)
1141				aiagp->extfreeback = iagp->extfreeback;
1142
1143			if (back >= 0)
1144				biagp->extfreefwd = iagp->extfreefwd;
1145			else
1146				imap->im_agctl[agno].extfree =
1147				    le32_to_cpu(iagp->extfreefwd);
1148
1149			iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1);
1150
1151			IAGFREE_LOCK(imap);
1152			iagp->iagfree = cpu_to_le32(imap->im_freeiag);
1153			imap->im_freeiag = iagno;
1154			IAGFREE_UNLOCK(imap);
1155		}
1156	}
1157
1158	/* remove the iag from the ag inode free list if freeing
1159	 * this extent causes the iag to have no free inodes.
1160	 */
1161	if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) {
1162		if ((int) le32_to_cpu(iagp->inofreefwd) >= 0)
1163			ciagp->inofreeback = iagp->inofreeback;
1164
1165		if ((int) le32_to_cpu(iagp->inofreeback) >= 0)
1166			diagp->inofreefwd = iagp->inofreefwd;
1167		else
1168			imap->im_agctl[agno].inofree =
1169			    le32_to_cpu(iagp->inofreefwd);
1170
1171		iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1);
1172	}
1173
1174	/* update the inode extent address and working map
1175	 * to reflect the free extent.
1176	 * the permanent map should have been updated already
1177	 * for the inode being freed.
1178	 */
1179	if (iagp->pmap[extno] != 0) {
1180		jfs_error(ip->i_sb, "the pmap does not show inode free\n");
1181	}
1182	iagp->wmap[extno] = 0;
1183	PXDlength(&iagp->inoext[extno], 0);
1184	PXDaddress(&iagp->inoext[extno], 0);
1185
1186	/* update the free extent and free inode summary maps
1187	 * to reflect the freed extent.
1188	 * the inode summary map is marked to indicate no inodes
1189	 * available for the freed extent.
1190	 */
1191	sword = extno >> L2EXTSPERSUM;
1192	bitno = extno & (EXTSPERSUM - 1);
1193	mask = HIGHORDER >> bitno;
1194	iagp->inosmap[sword] |= cpu_to_le32(mask);
1195	iagp->extsmap[sword] &= cpu_to_le32(~mask);
1196
1197	/* update the number of free inodes and number of free extents
1198	 * for the iag.
1199	 */
1200	le32_add_cpu(&iagp->nfreeinos, -(INOSPEREXT - 1));
1201	le32_add_cpu(&iagp->nfreeexts, 1);
1202
1203	/* update the number of free inodes and backed inodes
1204	 * at the ag and inode map level.
1205	 */
1206	imap->im_agctl[agno].numfree -= (INOSPEREXT - 1);
1207	imap->im_agctl[agno].numinos -= INOSPEREXT;
1208	atomic_sub(INOSPEREXT - 1, &imap->im_numfree);
1209	atomic_sub(INOSPEREXT, &imap->im_numinos);
1210
1211	if (amp)
1212		write_metapage(amp);
1213	if (bmp)
1214		write_metapage(bmp);
1215	if (cmp)
1216		write_metapage(cmp);
1217	if (dmp)
1218		write_metapage(dmp);
1219
1220	/*
1221	 * start transaction to update block allocation map
1222	 * for the inode extent freed;
1223	 *
1224	 * N.B. AG_LOCK is released and iag will be released below, and
1225	 * other thread may allocate inode from/reusing the ixad freed
1226	 * BUT with new/different backing inode extent from the extent
1227	 * to be freed by the transaction;
1228	 */
1229	tid = txBegin(ipimap->i_sb, COMMIT_FORCE);
1230	mutex_lock(&JFS_IP(ipimap)->commit_mutex);
1231
1232	/* acquire tlock of the iag page of the freed ixad
1233	 * to force the page NOHOMEOK (even though no data is
1234	 * logged from the iag page) until NOREDOPAGE|FREEXTENT log
1235	 * for the free of the extent is committed;
1236	 * write FREEXTENT|NOREDOPAGE log record
1237	 * N.B. linelock is overlaid as freed extent descriptor;
1238	 */
1239	tlck = txLock(tid, ipimap, mp, tlckINODE | tlckFREE);
1240	pxdlock = (struct pxd_lock *) & tlck->lock;
1241	pxdlock->flag = mlckFREEPXD;
1242	pxdlock->pxd = freepxd;
1243	pxdlock->index = 1;
1244
1245	write_metapage(mp);
1246
1247	iplist[0] = ipimap;
1248
1249	/*
1250	 * logredo needs the IAG number and IAG extent index in order
1251	 * to ensure that the IMap is consistent.  The least disruptive
1252	 * way to pass these values through  to the transaction manager
1253	 * is in the iplist array.
1254	 *
1255	 * It's not pretty, but it works.
1256	 */
1257	iplist[1] = (struct inode *) (size_t)iagno;
1258	iplist[2] = (struct inode *) (size_t)extno;
1259
1260	rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE);
1261
1262	txEnd(tid);
1263	mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
1264
1265	/* unlock the AG inode map information */
1266	AG_UNLOCK(imap, agno);
1267
1268	return (0);
1269
1270      error_out:
1271	IREAD_UNLOCK(ipimap);
1272
1273	if (amp)
1274		release_metapage(amp);
1275	if (bmp)
1276		release_metapage(bmp);
1277	if (cmp)
1278		release_metapage(cmp);
1279	if (dmp)
1280		release_metapage(dmp);
1281
1282	AG_UNLOCK(imap, agno);
1283
1284	release_metapage(mp);
1285
1286	return (rc);
1287}
1288
1289/*
1290 * There are several places in the diAlloc* routines where we initialize
1291 * the inode.
1292 */
1293static inline void
1294diInitInode(struct inode *ip, int iagno, int ino, int extno, struct iag * iagp)
1295{
1296	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
1297
1298	ip->i_ino = (iagno << L2INOSPERIAG) + ino;
1299	jfs_ip->ixpxd = iagp->inoext[extno];
1300	jfs_ip->agstart = le64_to_cpu(iagp->agstart);
1301	jfs_ip->active_ag = -1;
1302}
1303
1304
1305/*
1306 * NAME:	diAlloc(pip,dir,ip)
1307 *
1308 * FUNCTION:	allocate a disk inode from the inode working map
1309 *		for a fileset or aggregate.
1310 *
1311 * PARAMETERS:
1312 *	pip	- pointer to incore inode for the parent inode.
1313 *	dir	- 'true' if the new disk inode is for a directory.
1314 *	ip	- pointer to a new inode
1315 *
1316 * RETURN VALUES:
1317 *	0	- success.
1318 *	-ENOSPC	- insufficient disk resources.
1319 *	-EIO	- i/o error.
1320 */
1321int diAlloc(struct inode *pip, bool dir, struct inode *ip)
1322{
1323	int rc, ino, iagno, addext, extno, bitno, sword;
1324	int nwords, rem, i, agno;
1325	u32 mask, inosmap, extsmap;
1326	struct inode *ipimap;
1327	struct metapage *mp;
1328	ino_t inum;
1329	struct iag *iagp;
1330	struct inomap *imap;
1331
1332	/* get the pointers to the inode map inode and the
1333	 * corresponding imap control structure.
1334	 */
1335	ipimap = JFS_SBI(pip->i_sb)->ipimap;
1336	imap = JFS_IP(ipimap)->i_imap;
1337	JFS_IP(ip)->ipimap = ipimap;
1338	JFS_IP(ip)->fileset = FILESYSTEM_I;
1339
1340	/* for a directory, the allocation policy is to start
1341	 * at the ag level using the preferred ag.
1342	 */
1343	if (dir) {
1344		agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap);
1345		AG_LOCK(imap, agno);
1346		goto tryag;
1347	}
1348
1349	/* for files, the policy starts off by trying to allocate from
1350	 * the same iag containing the parent disk inode:
1351	 * try to allocate the new disk inode close to the parent disk
1352	 * inode, using parent disk inode number + 1 as the allocation
1353	 * hint.  (we use a left-to-right policy to attempt to avoid
1354	 * moving backward on the disk.)  compute the hint within the
1355	 * file system and the iag.
1356	 */
1357
1358	/* get the ag number of this iag */
1359	agno = BLKTOAG(JFS_IP(pip)->agstart, JFS_SBI(pip->i_sb));
1360
1361	if (atomic_read(&JFS_SBI(pip->i_sb)->bmap->db_active[agno])) {
1362		/*
1363		 * There is an open file actively growing.  We want to
1364		 * allocate new inodes from a different ag to avoid
1365		 * fragmentation problems.
1366		 */
1367		agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap);
1368		AG_LOCK(imap, agno);
1369		goto tryag;
1370	}
1371
1372	inum = pip->i_ino + 1;
1373	ino = inum & (INOSPERIAG - 1);
1374
1375	/* back off the hint if it is outside of the iag */
1376	if (ino == 0)
1377		inum = pip->i_ino;
1378
1379	/* lock the AG inode map information */
1380	AG_LOCK(imap, agno);
1381
1382	/* Get read lock on imap inode */
1383	IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
1384
1385	/* get the iag number and read the iag */
1386	iagno = INOTOIAG(inum);
1387	if ((rc = diIAGRead(imap, iagno, &mp))) {
1388		IREAD_UNLOCK(ipimap);
1389		AG_UNLOCK(imap, agno);
1390		return (rc);
1391	}
1392	iagp = (struct iag *) mp->data;
1393
1394	/* determine if new inode extent is allowed to be added to the iag.
1395	 * new inode extent can be added to the iag if the ag
1396	 * has less than 32 free disk inodes and the iag has free extents.
1397	 */
1398	addext = (imap->im_agctl[agno].numfree < 32 && iagp->nfreeexts);
1399
1400	/*
1401	 *	try to allocate from the IAG
1402	 */
1403	/* check if the inode may be allocated from the iag
1404	 * (i.e. the inode has free inodes or new extent can be added).
1405	 */
1406	if (iagp->nfreeinos || addext) {
1407		/* determine the extent number of the hint.
1408		 */
1409		extno = ino >> L2INOSPEREXT;
1410
1411		/* check if the extent containing the hint has backed
1412		 * inodes.  if so, try to allocate within this extent.
1413		 */
1414		if (addressPXD(&iagp->inoext[extno])) {
1415			bitno = ino & (INOSPEREXT - 1);
1416			if ((bitno =
1417			     diFindFree(le32_to_cpu(iagp->wmap[extno]),
1418					bitno))
1419			    < INOSPEREXT) {
1420				ino = (extno << L2INOSPEREXT) + bitno;
1421
1422				/* a free inode (bit) was found within this
1423				 * extent, so allocate it.
1424				 */
1425				rc = diAllocBit(imap, iagp, ino);
1426				IREAD_UNLOCK(ipimap);
1427				if (rc) {
1428					assert(rc == -EIO);
1429				} else {
1430					/* set the results of the allocation
1431					 * and write the iag.
1432					 */
1433					diInitInode(ip, iagno, ino, extno,
1434						    iagp);
1435					mark_metapage_dirty(mp);
1436				}
1437				release_metapage(mp);
1438
1439				/* free the AG lock and return.
1440				 */
1441				AG_UNLOCK(imap, agno);
1442				return (rc);
1443			}
1444
1445			if (!addext)
1446				extno =
1447				    (extno ==
1448				     EXTSPERIAG - 1) ? 0 : extno + 1;
1449		}
1450
1451		/*
1452		 * no free inodes within the extent containing the hint.
1453		 *
1454		 * try to allocate from the backed extents following
1455		 * hint or, if appropriate (i.e. addext is true), allocate
1456		 * an extent of free inodes at or following the extent
1457		 * containing the hint.
1458		 *
1459		 * the free inode and free extent summary maps are used
1460		 * here, so determine the starting summary map position
1461		 * and the number of words we'll have to examine.  again,
1462		 * the approach is to allocate following the hint, so we
1463		 * might have to initially ignore prior bits of the summary
1464		 * map that represent extents prior to the extent containing
1465		 * the hint and later revisit these bits.
1466		 */
1467		bitno = extno & (EXTSPERSUM - 1);
1468		nwords = (bitno == 0) ? SMAPSZ : SMAPSZ + 1;
1469		sword = extno >> L2EXTSPERSUM;
1470
1471		/* mask any prior bits for the starting words of the
1472		 * summary map.
1473		 */
1474		mask = (bitno == 0) ? 0 : (ONES << (EXTSPERSUM - bitno));
1475		inosmap = le32_to_cpu(iagp->inosmap[sword]) | mask;
1476		extsmap = le32_to_cpu(iagp->extsmap[sword]) | mask;
1477
1478		/* scan the free inode and free extent summary maps for
1479		 * free resources.
1480		 */
1481		for (i = 0; i < nwords; i++) {
1482			/* check if this word of the free inode summary
1483			 * map describes an extent with free inodes.
1484			 */
1485			if (~inosmap) {
1486				/* an extent with free inodes has been
1487				 * found. determine the extent number
1488				 * and the inode number within the extent.
1489				 */
1490				rem = diFindFree(inosmap, 0);
1491				extno = (sword << L2EXTSPERSUM) + rem;
1492				rem = diFindFree(le32_to_cpu(iagp->wmap[extno]),
1493						 0);
1494				if (rem >= INOSPEREXT) {
1495					IREAD_UNLOCK(ipimap);
1496					release_metapage(mp);
1497					AG_UNLOCK(imap, agno);
1498					jfs_error(ip->i_sb,
1499						  "can't find free bit in wmap\n");
 
1500					return -EIO;
1501				}
1502
1503				/* determine the inode number within the
1504				 * iag and allocate the inode from the
1505				 * map.
1506				 */
1507				ino = (extno << L2INOSPEREXT) + rem;
1508				rc = diAllocBit(imap, iagp, ino);
1509				IREAD_UNLOCK(ipimap);
1510				if (rc)
1511					assert(rc == -EIO);
1512				else {
1513					/* set the results of the allocation
1514					 * and write the iag.
1515					 */
1516					diInitInode(ip, iagno, ino, extno,
1517						    iagp);
1518					mark_metapage_dirty(mp);
1519				}
1520				release_metapage(mp);
1521
1522				/* free the AG lock and return.
1523				 */
1524				AG_UNLOCK(imap, agno);
1525				return (rc);
1526
1527			}
1528
1529			/* check if we may allocate an extent of free
1530			 * inodes and whether this word of the free
1531			 * extents summary map describes a free extent.
1532			 */
1533			if (addext && ~extsmap) {
1534				/* a free extent has been found.  determine
1535				 * the extent number.
1536				 */
1537				rem = diFindFree(extsmap, 0);
1538				extno = (sword << L2EXTSPERSUM) + rem;
1539
1540				/* allocate an extent of free inodes.
1541				 */
1542				if ((rc = diNewExt(imap, iagp, extno))) {
1543					/* if there is no disk space for a
1544					 * new extent, try to allocate the
1545					 * disk inode from somewhere else.
1546					 */
1547					if (rc == -ENOSPC)
1548						break;
1549
1550					assert(rc == -EIO);
1551				} else {
1552					/* set the results of the allocation
1553					 * and write the iag.
1554					 */
1555					diInitInode(ip, iagno,
1556						    extno << L2INOSPEREXT,
1557						    extno, iagp);
1558					mark_metapage_dirty(mp);
1559				}
1560				release_metapage(mp);
1561				/* free the imap inode & the AG lock & return.
1562				 */
1563				IREAD_UNLOCK(ipimap);
1564				AG_UNLOCK(imap, agno);
1565				return (rc);
1566			}
1567
1568			/* move on to the next set of summary map words.
1569			 */
1570			sword = (sword == SMAPSZ - 1) ? 0 : sword + 1;
1571			inosmap = le32_to_cpu(iagp->inosmap[sword]);
1572			extsmap = le32_to_cpu(iagp->extsmap[sword]);
1573		}
1574	}
1575	/* unlock imap inode */
1576	IREAD_UNLOCK(ipimap);
1577
1578	/* nothing doing in this iag, so release it. */
1579	release_metapage(mp);
1580
1581      tryag:
1582	/*
1583	 * try to allocate anywhere within the same AG as the parent inode.
1584	 */
1585	rc = diAllocAG(imap, agno, dir, ip);
1586
1587	AG_UNLOCK(imap, agno);
1588
1589	if (rc != -ENOSPC)
1590		return (rc);
1591
1592	/*
1593	 * try to allocate in any AG.
1594	 */
1595	return (diAllocAny(imap, agno, dir, ip));
1596}
1597
1598
1599/*
1600 * NAME:	diAllocAG(imap,agno,dir,ip)
1601 *
1602 * FUNCTION:	allocate a disk inode from the allocation group.
1603 *
1604 *		this routine first determines if a new extent of free
1605 *		inodes should be added for the allocation group, with
1606 *		the current request satisfied from this extent. if this
1607 *		is the case, an attempt will be made to do just that.  if
1608 *		this attempt fails or it has been determined that a new
1609 *		extent should not be added, an attempt is made to satisfy
1610 *		the request by allocating an existing (backed) free inode
1611 *		from the allocation group.
1612 *
1613 * PRE CONDITION: Already have the AG lock for this AG.
1614 *
1615 * PARAMETERS:
1616 *	imap	- pointer to inode map control structure.
1617 *	agno	- allocation group to allocate from.
1618 *	dir	- 'true' if the new disk inode is for a directory.
1619 *	ip	- pointer to the new inode to be filled in on successful return
1620 *		  with the disk inode number allocated, its extent address
1621 *		  and the start of the ag.
1622 *
1623 * RETURN VALUES:
1624 *	0	- success.
1625 *	-ENOSPC	- insufficient disk resources.
1626 *	-EIO	- i/o error.
1627 */
1628static int
1629diAllocAG(struct inomap * imap, int agno, bool dir, struct inode *ip)
1630{
1631	int rc, addext, numfree, numinos;
1632
1633	/* get the number of free and the number of backed disk
1634	 * inodes currently within the ag.
1635	 */
1636	numfree = imap->im_agctl[agno].numfree;
1637	numinos = imap->im_agctl[agno].numinos;
1638
1639	if (numfree > numinos) {
1640		jfs_error(ip->i_sb, "numfree > numinos\n");
1641		return -EIO;
1642	}
1643
1644	/* determine if we should allocate a new extent of free inodes
1645	 * within the ag: for directory inodes, add a new extent
1646	 * if there are a small number of free inodes or number of free
1647	 * inodes is a small percentage of the number of backed inodes.
1648	 */
1649	if (dir)
1650		addext = (numfree < 64 ||
1651			  (numfree < 256
1652			   && ((numfree * 100) / numinos) <= 20));
1653	else
1654		addext = (numfree == 0);
1655
1656	/*
1657	 * try to allocate a new extent of free inodes.
1658	 */
1659	if (addext) {
1660		/* if free space is not available for this new extent, try
1661		 * below to allocate a free and existing (already backed)
1662		 * inode from the ag.
1663		 */
1664		if ((rc = diAllocExt(imap, agno, ip)) != -ENOSPC)
1665			return (rc);
1666	}
1667
1668	/*
1669	 * try to allocate an existing free inode from the ag.
1670	 */
1671	return (diAllocIno(imap, agno, ip));
1672}
1673
1674
1675/*
1676 * NAME:	diAllocAny(imap,agno,dir,iap)
1677 *
1678 * FUNCTION:	allocate a disk inode from any other allocation group.
1679 *
1680 *		this routine is called when an allocation attempt within
1681 *		the primary allocation group has failed. if attempts to
1682 *		allocate an inode from any allocation group other than the
1683 *		specified primary group.
1684 *
1685 * PARAMETERS:
1686 *	imap	- pointer to inode map control structure.
1687 *	agno	- primary allocation group (to avoid).
1688 *	dir	- 'true' if the new disk inode is for a directory.
1689 *	ip	- pointer to a new inode to be filled in on successful return
1690 *		  with the disk inode number allocated, its extent address
1691 *		  and the start of the ag.
1692 *
1693 * RETURN VALUES:
1694 *	0	- success.
1695 *	-ENOSPC	- insufficient disk resources.
1696 *	-EIO	- i/o error.
1697 */
1698static int
1699diAllocAny(struct inomap * imap, int agno, bool dir, struct inode *ip)
1700{
1701	int ag, rc;
1702	int maxag = JFS_SBI(imap->im_ipimap->i_sb)->bmap->db_maxag;
1703
1704
1705	/* try to allocate from the ags following agno up to
1706	 * the maximum ag number.
1707	 */
1708	for (ag = agno + 1; ag <= maxag; ag++) {
1709		AG_LOCK(imap, ag);
1710
1711		rc = diAllocAG(imap, ag, dir, ip);
1712
1713		AG_UNLOCK(imap, ag);
1714
1715		if (rc != -ENOSPC)
1716			return (rc);
1717	}
1718
1719	/* try to allocate from the ags in front of agno.
1720	 */
1721	for (ag = 0; ag < agno; ag++) {
1722		AG_LOCK(imap, ag);
1723
1724		rc = diAllocAG(imap, ag, dir, ip);
1725
1726		AG_UNLOCK(imap, ag);
1727
1728		if (rc != -ENOSPC)
1729			return (rc);
1730	}
1731
1732	/* no free disk inodes.
1733	 */
1734	return -ENOSPC;
1735}
1736
1737
1738/*
1739 * NAME:	diAllocIno(imap,agno,ip)
1740 *
1741 * FUNCTION:	allocate a disk inode from the allocation group's free
1742 *		inode list, returning an error if this free list is
1743 *		empty (i.e. no iags on the list).
1744 *
1745 *		allocation occurs from the first iag on the list using
1746 *		the iag's free inode summary map to find the leftmost
1747 *		free inode in the iag.
1748 *
1749 * PRE CONDITION: Already have AG lock for this AG.
1750 *
1751 * PARAMETERS:
1752 *	imap	- pointer to inode map control structure.
1753 *	agno	- allocation group.
1754 *	ip	- pointer to new inode to be filled in on successful return
1755 *		  with the disk inode number allocated, its extent address
1756 *		  and the start of the ag.
1757 *
1758 * RETURN VALUES:
1759 *	0	- success.
1760 *	-ENOSPC	- insufficient disk resources.
1761 *	-EIO	- i/o error.
1762 */
1763static int diAllocIno(struct inomap * imap, int agno, struct inode *ip)
1764{
1765	int iagno, ino, rc, rem, extno, sword;
1766	struct metapage *mp;
1767	struct iag *iagp;
1768
1769	/* check if there are iags on the ag's free inode list.
1770	 */
1771	if ((iagno = imap->im_agctl[agno].inofree) < 0)
1772		return -ENOSPC;
1773
1774	/* obtain read lock on imap inode */
1775	IREAD_LOCK(imap->im_ipimap, RDWRLOCK_IMAP);
1776
1777	/* read the iag at the head of the list.
1778	 */
1779	if ((rc = diIAGRead(imap, iagno, &mp))) {
1780		IREAD_UNLOCK(imap->im_ipimap);
1781		return (rc);
1782	}
1783	iagp = (struct iag *) mp->data;
1784
1785	/* better be free inodes in this iag if it is on the
1786	 * list.
1787	 */
1788	if (!iagp->nfreeinos) {
1789		IREAD_UNLOCK(imap->im_ipimap);
1790		release_metapage(mp);
1791		jfs_error(ip->i_sb, "nfreeinos = 0, but iag on freelist\n");
 
1792		return -EIO;
1793	}
1794
1795	/* scan the free inode summary map to find an extent
1796	 * with free inodes.
1797	 */
1798	for (sword = 0;; sword++) {
1799		if (sword >= SMAPSZ) {
1800			IREAD_UNLOCK(imap->im_ipimap);
1801			release_metapage(mp);
1802			jfs_error(ip->i_sb,
1803				  "free inode not found in summary map\n");
1804			return -EIO;
1805		}
1806
1807		if (~iagp->inosmap[sword])
1808			break;
1809	}
1810
1811	/* found a extent with free inodes. determine
1812	 * the extent number.
1813	 */
1814	rem = diFindFree(le32_to_cpu(iagp->inosmap[sword]), 0);
1815	if (rem >= EXTSPERSUM) {
1816		IREAD_UNLOCK(imap->im_ipimap);
1817		release_metapage(mp);
1818		jfs_error(ip->i_sb, "no free extent found\n");
1819		return -EIO;
1820	}
1821	extno = (sword << L2EXTSPERSUM) + rem;
1822
1823	/* find the first free inode in the extent.
1824	 */
1825	rem = diFindFree(le32_to_cpu(iagp->wmap[extno]), 0);
1826	if (rem >= INOSPEREXT) {
1827		IREAD_UNLOCK(imap->im_ipimap);
1828		release_metapage(mp);
1829		jfs_error(ip->i_sb, "free inode not found\n");
1830		return -EIO;
1831	}
1832
1833	/* compute the inode number within the iag.
1834	 */
1835	ino = (extno << L2INOSPEREXT) + rem;
1836
1837	/* allocate the inode.
1838	 */
1839	rc = diAllocBit(imap, iagp, ino);
1840	IREAD_UNLOCK(imap->im_ipimap);
1841	if (rc) {
1842		release_metapage(mp);
1843		return (rc);
1844	}
1845
1846	/* set the results of the allocation and write the iag.
1847	 */
1848	diInitInode(ip, iagno, ino, extno, iagp);
1849	write_metapage(mp);
1850
1851	return (0);
1852}
1853
1854
1855/*
1856 * NAME:	diAllocExt(imap,agno,ip)
1857 *
1858 * FUNCTION:	add a new extent of free inodes to an iag, allocating
1859 *		an inode from this extent to satisfy the current allocation
1860 *		request.
1861 *
1862 *		this routine first tries to find an existing iag with free
1863 *		extents through the ag free extent list.  if list is not
1864 *		empty, the head of the list will be selected as the home
1865 *		of the new extent of free inodes.  otherwise (the list is
1866 *		empty), a new iag will be allocated for the ag to contain
1867 *		the extent.
1868 *
1869 *		once an iag has been selected, the free extent summary map
1870 *		is used to locate a free extent within the iag and diNewExt()
1871 *		is called to initialize the extent, with initialization
1872 *		including the allocation of the first inode of the extent
1873 *		for the purpose of satisfying this request.
1874 *
1875 * PARAMETERS:
1876 *	imap	- pointer to inode map control structure.
1877 *	agno	- allocation group number.
1878 *	ip	- pointer to new inode to be filled in on successful return
1879 *		  with the disk inode number allocated, its extent address
1880 *		  and the start of the ag.
1881 *
1882 * RETURN VALUES:
1883 *	0	- success.
1884 *	-ENOSPC	- insufficient disk resources.
1885 *	-EIO	- i/o error.
1886 */
1887static int diAllocExt(struct inomap * imap, int agno, struct inode *ip)
1888{
1889	int rem, iagno, sword, extno, rc;
1890	struct metapage *mp;
1891	struct iag *iagp;
1892
1893	/* check if the ag has any iags with free extents.  if not,
1894	 * allocate a new iag for the ag.
1895	 */
1896	if ((iagno = imap->im_agctl[agno].extfree) < 0) {
1897		/* If successful, diNewIAG will obtain the read lock on the
1898		 * imap inode.
1899		 */
1900		if ((rc = diNewIAG(imap, &iagno, agno, &mp))) {
1901			return (rc);
1902		}
1903		iagp = (struct iag *) mp->data;
1904
1905		/* set the ag number if this a brand new iag
1906		 */
1907		iagp->agstart =
1908		    cpu_to_le64(AGTOBLK(agno, imap->im_ipimap));
1909	} else {
1910		/* read the iag.
1911		 */
1912		IREAD_LOCK(imap->im_ipimap, RDWRLOCK_IMAP);
1913		if ((rc = diIAGRead(imap, iagno, &mp))) {
1914			IREAD_UNLOCK(imap->im_ipimap);
1915			jfs_error(ip->i_sb, "error reading iag\n");
1916			return rc;
1917		}
1918		iagp = (struct iag *) mp->data;
1919	}
1920
1921	/* using the free extent summary map, find a free extent.
1922	 */
1923	for (sword = 0;; sword++) {
1924		if (sword >= SMAPSZ) {
1925			release_metapage(mp);
1926			IREAD_UNLOCK(imap->im_ipimap);
1927			jfs_error(ip->i_sb, "free ext summary map not found\n");
 
1928			return -EIO;
1929		}
1930		if (~iagp->extsmap[sword])
1931			break;
1932	}
1933
1934	/* determine the extent number of the free extent.
1935	 */
1936	rem = diFindFree(le32_to_cpu(iagp->extsmap[sword]), 0);
1937	if (rem >= EXTSPERSUM) {
1938		release_metapage(mp);
1939		IREAD_UNLOCK(imap->im_ipimap);
1940		jfs_error(ip->i_sb, "free extent not found\n");
1941		return -EIO;
1942	}
1943	extno = (sword << L2EXTSPERSUM) + rem;
1944
1945	/* initialize the new extent.
1946	 */
1947	rc = diNewExt(imap, iagp, extno);
1948	IREAD_UNLOCK(imap->im_ipimap);
1949	if (rc) {
1950		/* something bad happened.  if a new iag was allocated,
1951		 * place it back on the inode map's iag free list, and
1952		 * clear the ag number information.
1953		 */
1954		if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
1955			IAGFREE_LOCK(imap);
1956			iagp->iagfree = cpu_to_le32(imap->im_freeiag);
1957			imap->im_freeiag = iagno;
1958			IAGFREE_UNLOCK(imap);
1959		}
1960		write_metapage(mp);
1961		return (rc);
1962	}
1963
1964	/* set the results of the allocation and write the iag.
1965	 */
1966	diInitInode(ip, iagno, extno << L2INOSPEREXT, extno, iagp);
1967
1968	write_metapage(mp);
1969
1970	return (0);
1971}
1972
1973
1974/*
1975 * NAME:	diAllocBit(imap,iagp,ino)
1976 *
1977 * FUNCTION:	allocate a backed inode from an iag.
1978 *
1979 *		this routine performs the mechanics of allocating a
1980 *		specified inode from a backed extent.
1981 *
1982 *		if the inode to be allocated represents the last free
1983 *		inode within the iag, the iag will be removed from the
1984 *		ag free inode list.
1985 *
1986 *		a careful update approach is used to provide consistency
1987 *		in the face of updates to multiple buffers.  under this
1988 *		approach, all required buffers are obtained before making
1989 *		any updates and are held all are updates are complete.
1990 *
1991 * PRE CONDITION: Already have buffer lock on iagp.  Already have AG lock on
1992 *	this AG.  Must have read lock on imap inode.
1993 *
1994 * PARAMETERS:
1995 *	imap	- pointer to inode map control structure.
1996 *	iagp	- pointer to iag.
1997 *	ino	- inode number to be allocated within the iag.
1998 *
1999 * RETURN VALUES:
2000 *	0	- success.
2001 *	-ENOSPC	- insufficient disk resources.
2002 *	-EIO	- i/o error.
2003 */
2004static int diAllocBit(struct inomap * imap, struct iag * iagp, int ino)
2005{
2006	int extno, bitno, agno, sword, rc;
2007	struct metapage *amp = NULL, *bmp = NULL;
2008	struct iag *aiagp = NULL, *biagp = NULL;
2009	u32 mask;
2010
2011	/* check if this is the last free inode within the iag.
2012	 * if so, it will have to be removed from the ag free
2013	 * inode list, so get the iags preceding and following
2014	 * it on the list.
2015	 */
2016	if (iagp->nfreeinos == cpu_to_le32(1)) {
2017		if ((int) le32_to_cpu(iagp->inofreefwd) >= 0) {
2018			if ((rc =
2019			     diIAGRead(imap, le32_to_cpu(iagp->inofreefwd),
2020				       &amp)))
2021				return (rc);
2022			aiagp = (struct iag *) amp->data;
2023		}
2024
2025		if ((int) le32_to_cpu(iagp->inofreeback) >= 0) {
2026			if ((rc =
2027			     diIAGRead(imap,
2028				       le32_to_cpu(iagp->inofreeback),
2029				       &bmp))) {
2030				if (amp)
2031					release_metapage(amp);
2032				return (rc);
2033			}
2034			biagp = (struct iag *) bmp->data;
2035		}
2036	}
2037
2038	/* get the ag number, extent number, inode number within
2039	 * the extent.
2040	 */
2041	agno = BLKTOAG(le64_to_cpu(iagp->agstart), JFS_SBI(imap->im_ipimap->i_sb));
2042	extno = ino >> L2INOSPEREXT;
2043	bitno = ino & (INOSPEREXT - 1);
2044
2045	/* compute the mask for setting the map.
2046	 */
2047	mask = HIGHORDER >> bitno;
2048
2049	/* the inode should be free and backed.
2050	 */
2051	if (((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) ||
2052	    ((le32_to_cpu(iagp->wmap[extno]) & mask) != 0) ||
2053	    (addressPXD(&iagp->inoext[extno]) == 0)) {
2054		if (amp)
2055			release_metapage(amp);
2056		if (bmp)
2057			release_metapage(bmp);
2058
2059		jfs_error(imap->im_ipimap->i_sb, "iag inconsistent\n");
 
2060		return -EIO;
2061	}
2062
2063	/* mark the inode as allocated in the working map.
2064	 */
2065	iagp->wmap[extno] |= cpu_to_le32(mask);
2066
2067	/* check if all inodes within the extent are now
2068	 * allocated.  if so, update the free inode summary
2069	 * map to reflect this.
2070	 */
2071	if (iagp->wmap[extno] == cpu_to_le32(ONES)) {
2072		sword = extno >> L2EXTSPERSUM;
2073		bitno = extno & (EXTSPERSUM - 1);
2074		iagp->inosmap[sword] |= cpu_to_le32(HIGHORDER >> bitno);
2075	}
2076
2077	/* if this was the last free inode in the iag, remove the
2078	 * iag from the ag free inode list.
2079	 */
2080	if (iagp->nfreeinos == cpu_to_le32(1)) {
2081		if (amp) {
2082			aiagp->inofreeback = iagp->inofreeback;
2083			write_metapage(amp);
2084		}
2085
2086		if (bmp) {
2087			biagp->inofreefwd = iagp->inofreefwd;
2088			write_metapage(bmp);
2089		} else {
2090			imap->im_agctl[agno].inofree =
2091			    le32_to_cpu(iagp->inofreefwd);
2092		}
2093		iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1);
2094	}
2095
2096	/* update the free inode count at the iag, ag, inode
2097	 * map levels.
2098	 */
2099	le32_add_cpu(&iagp->nfreeinos, -1);
2100	imap->im_agctl[agno].numfree -= 1;
2101	atomic_dec(&imap->im_numfree);
2102
2103	return (0);
2104}
2105
2106
2107/*
2108 * NAME:	diNewExt(imap,iagp,extno)
2109 *
2110 * FUNCTION:	initialize a new extent of inodes for an iag, allocating
2111 *		the first inode of the extent for use for the current
2112 *		allocation request.
2113 *
2114 *		disk resources are allocated for the new extent of inodes
2115 *		and the inodes themselves are initialized to reflect their
2116 *		existence within the extent (i.e. their inode numbers and
2117 *		inode extent addresses are set) and their initial state
2118 *		(mode and link count are set to zero).
2119 *
2120 *		if the iag is new, it is not yet on an ag extent free list
2121 *		but will now be placed on this list.
2122 *
2123 *		if the allocation of the new extent causes the iag to
2124 *		have no free extent, the iag will be removed from the
2125 *		ag extent free list.
2126 *
2127 *		if the iag has no free backed inodes, it will be placed
2128 *		on the ag free inode list, since the addition of the new
2129 *		extent will now cause it to have free inodes.
2130 *
2131 *		a careful update approach is used to provide consistency
2132 *		(i.e. list consistency) in the face of updates to multiple
2133 *		buffers.  under this approach, all required buffers are
2134 *		obtained before making any updates and are held until all
2135 *		updates are complete.
2136 *
2137 * PRE CONDITION: Already have buffer lock on iagp.  Already have AG lock on
2138 *	this AG.  Must have read lock on imap inode.
2139 *
2140 * PARAMETERS:
2141 *	imap	- pointer to inode map control structure.
2142 *	iagp	- pointer to iag.
2143 *	extno	- extent number.
2144 *
2145 * RETURN VALUES:
2146 *	0	- success.
2147 *	-ENOSPC	- insufficient disk resources.
2148 *	-EIO	- i/o error.
2149 */
2150static int diNewExt(struct inomap * imap, struct iag * iagp, int extno)
2151{
2152	int agno, iagno, fwd, back, freei = 0, sword, rc;
2153	struct iag *aiagp = NULL, *biagp = NULL, *ciagp = NULL;
2154	struct metapage *amp, *bmp, *cmp, *dmp;
2155	struct inode *ipimap;
2156	s64 blkno, hint;
2157	int i, j;
2158	u32 mask;
2159	ino_t ino;
2160	struct dinode *dp;
2161	struct jfs_sb_info *sbi;
2162
2163	/* better have free extents.
2164	 */
2165	if (!iagp->nfreeexts) {
2166		jfs_error(imap->im_ipimap->i_sb, "no free extents\n");
2167		return -EIO;
2168	}
2169
2170	/* get the inode map inode.
2171	 */
2172	ipimap = imap->im_ipimap;
2173	sbi = JFS_SBI(ipimap->i_sb);
2174
2175	amp = bmp = cmp = NULL;
2176
2177	/* get the ag and iag numbers for this iag.
2178	 */
2179	agno = BLKTOAG(le64_to_cpu(iagp->agstart), sbi);
2180	iagno = le32_to_cpu(iagp->iagnum);
2181
2182	/* check if this is the last free extent within the
2183	 * iag.  if so, the iag must be removed from the ag
2184	 * free extent list, so get the iags preceding and
2185	 * following the iag on this list.
2186	 */
2187	if (iagp->nfreeexts == cpu_to_le32(1)) {
2188		if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) {
2189			if ((rc = diIAGRead(imap, fwd, &amp)))
2190				return (rc);
2191			aiagp = (struct iag *) amp->data;
2192		}
2193
2194		if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) {
2195			if ((rc = diIAGRead(imap, back, &bmp)))
2196				goto error_out;
2197			biagp = (struct iag *) bmp->data;
2198		}
2199	} else {
2200		/* the iag has free extents.  if all extents are free
2201		 * (as is the case for a newly allocated iag), the iag
2202		 * must be added to the ag free extent list, so get
2203		 * the iag at the head of the list in preparation for
2204		 * adding this iag to this list.
2205		 */
2206		fwd = back = -1;
2207		if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2208			if ((fwd = imap->im_agctl[agno].extfree) >= 0) {
2209				if ((rc = diIAGRead(imap, fwd, &amp)))
2210					goto error_out;
2211				aiagp = (struct iag *) amp->data;
2212			}
2213		}
2214	}
2215
2216	/* check if the iag has no free inodes.  if so, the iag
2217	 * will have to be added to the ag free inode list, so get
2218	 * the iag at the head of the list in preparation for
2219	 * adding this iag to this list.  in doing this, we must
2220	 * check if we already have the iag at the head of
2221	 * the list in hand.
2222	 */
2223	if (iagp->nfreeinos == 0) {
2224		freei = imap->im_agctl[agno].inofree;
2225
2226		if (freei >= 0) {
2227			if (freei == fwd) {
2228				ciagp = aiagp;
2229			} else if (freei == back) {
2230				ciagp = biagp;
2231			} else {
2232				if ((rc = diIAGRead(imap, freei, &cmp)))
2233					goto error_out;
2234				ciagp = (struct iag *) cmp->data;
2235			}
2236			if (ciagp == NULL) {
2237				jfs_error(imap->im_ipimap->i_sb,
2238					  "ciagp == NULL\n");
2239				rc = -EIO;
2240				goto error_out;
2241			}
2242		}
2243	}
2244
2245	/* allocate disk space for the inode extent.
2246	 */
2247	if ((extno == 0) || (addressPXD(&iagp->inoext[extno - 1]) == 0))
2248		hint = ((s64) agno << sbi->bmap->db_agl2size) - 1;
2249	else
2250		hint = addressPXD(&iagp->inoext[extno - 1]) +
2251		    lengthPXD(&iagp->inoext[extno - 1]) - 1;
2252
2253	if ((rc = dbAlloc(ipimap, hint, (s64) imap->im_nbperiext, &blkno)))
2254		goto error_out;
2255
2256	/* compute the inode number of the first inode within the
2257	 * extent.
2258	 */
2259	ino = (iagno << L2INOSPERIAG) + (extno << L2INOSPEREXT);
2260
2261	/* initialize the inodes within the newly allocated extent a
2262	 * page at a time.
2263	 */
2264	for (i = 0; i < imap->im_nbperiext; i += sbi->nbperpage) {
2265		/* get a buffer for this page of disk inodes.
2266		 */
2267		dmp = get_metapage(ipimap, blkno + i, PSIZE, 1);
2268		if (dmp == NULL) {
2269			rc = -EIO;
2270			goto error_out;
2271		}
2272		dp = (struct dinode *) dmp->data;
2273
2274		/* initialize the inode number, mode, link count and
2275		 * inode extent address.
2276		 */
2277		for (j = 0; j < INOSPERPAGE; j++, dp++, ino++) {
2278			dp->di_inostamp = cpu_to_le32(sbi->inostamp);
2279			dp->di_number = cpu_to_le32(ino);
2280			dp->di_fileset = cpu_to_le32(FILESYSTEM_I);
2281			dp->di_mode = 0;
2282			dp->di_nlink = 0;
2283			PXDaddress(&(dp->di_ixpxd), blkno);
2284			PXDlength(&(dp->di_ixpxd), imap->im_nbperiext);
2285		}
2286		write_metapage(dmp);
2287	}
2288
2289	/* if this is the last free extent within the iag, remove the
2290	 * iag from the ag free extent list.
2291	 */
2292	if (iagp->nfreeexts == cpu_to_le32(1)) {
2293		if (fwd >= 0)
2294			aiagp->extfreeback = iagp->extfreeback;
2295
2296		if (back >= 0)
2297			biagp->extfreefwd = iagp->extfreefwd;
2298		else
2299			imap->im_agctl[agno].extfree =
2300			    le32_to_cpu(iagp->extfreefwd);
2301
2302		iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1);
2303	} else {
2304		/* if the iag has all free extents (newly allocated iag),
2305		 * add the iag to the ag free extent list.
2306		 */
2307		if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2308			if (fwd >= 0)
2309				aiagp->extfreeback = cpu_to_le32(iagno);
2310
2311			iagp->extfreefwd = cpu_to_le32(fwd);
2312			iagp->extfreeback = cpu_to_le32(-1);
2313			imap->im_agctl[agno].extfree = iagno;
2314		}
2315	}
2316
2317	/* if the iag has no free inodes, add the iag to the
2318	 * ag free inode list.
2319	 */
2320	if (iagp->nfreeinos == 0) {
2321		if (freei >= 0)
2322			ciagp->inofreeback = cpu_to_le32(iagno);
2323
2324		iagp->inofreefwd =
2325		    cpu_to_le32(imap->im_agctl[agno].inofree);
2326		iagp->inofreeback = cpu_to_le32(-1);
2327		imap->im_agctl[agno].inofree = iagno;
2328	}
2329
2330	/* initialize the extent descriptor of the extent. */
2331	PXDlength(&iagp->inoext[extno], imap->im_nbperiext);
2332	PXDaddress(&iagp->inoext[extno], blkno);
2333
2334	/* initialize the working and persistent map of the extent.
2335	 * the working map will be initialized such that
2336	 * it indicates the first inode of the extent is allocated.
2337	 */
2338	iagp->wmap[extno] = cpu_to_le32(HIGHORDER);
2339	iagp->pmap[extno] = 0;
2340
2341	/* update the free inode and free extent summary maps
2342	 * for the extent to indicate the extent has free inodes
2343	 * and no longer represents a free extent.
2344	 */
2345	sword = extno >> L2EXTSPERSUM;
2346	mask = HIGHORDER >> (extno & (EXTSPERSUM - 1));
2347	iagp->extsmap[sword] |= cpu_to_le32(mask);
2348	iagp->inosmap[sword] &= cpu_to_le32(~mask);
2349
2350	/* update the free inode and free extent counts for the
2351	 * iag.
2352	 */
2353	le32_add_cpu(&iagp->nfreeinos, (INOSPEREXT - 1));
2354	le32_add_cpu(&iagp->nfreeexts, -1);
2355
2356	/* update the free and backed inode counts for the ag.
2357	 */
2358	imap->im_agctl[agno].numfree += (INOSPEREXT - 1);
2359	imap->im_agctl[agno].numinos += INOSPEREXT;
2360
2361	/* update the free and backed inode counts for the inode map.
2362	 */
2363	atomic_add(INOSPEREXT - 1, &imap->im_numfree);
2364	atomic_add(INOSPEREXT, &imap->im_numinos);
2365
2366	/* write the iags.
2367	 */
2368	if (amp)
2369		write_metapage(amp);
2370	if (bmp)
2371		write_metapage(bmp);
2372	if (cmp)
2373		write_metapage(cmp);
2374
2375	return (0);
2376
2377      error_out:
2378
2379	/* release the iags.
2380	 */
2381	if (amp)
2382		release_metapage(amp);
2383	if (bmp)
2384		release_metapage(bmp);
2385	if (cmp)
2386		release_metapage(cmp);
2387
2388	return (rc);
2389}
2390
2391
2392/*
2393 * NAME:	diNewIAG(imap,iagnop,agno)
2394 *
2395 * FUNCTION:	allocate a new iag for an allocation group.
2396 *
2397 *		first tries to allocate the iag from the inode map
2398 *		iagfree list:
2399 *		if the list has free iags, the head of the list is removed
2400 *		and returned to satisfy the request.
2401 *		if the inode map's iag free list is empty, the inode map
2402 *		is extended to hold a new iag. this new iag is initialized
2403 *		and returned to satisfy the request.
2404 *
2405 * PARAMETERS:
2406 *	imap	- pointer to inode map control structure.
2407 *	iagnop	- pointer to an iag number set with the number of the
2408 *		  newly allocated iag upon successful return.
2409 *	agno	- allocation group number.
2410 *	bpp	- Buffer pointer to be filled in with new IAG's buffer
2411 *
2412 * RETURN VALUES:
2413 *	0	- success.
2414 *	-ENOSPC	- insufficient disk resources.
2415 *	-EIO	- i/o error.
2416 *
2417 * serialization:
2418 *	AG lock held on entry/exit;
2419 *	write lock on the map is held inside;
2420 *	read lock on the map is held on successful completion;
2421 *
2422 * note: new iag transaction:
2423 * . synchronously write iag;
2424 * . write log of xtree and inode of imap;
2425 * . commit;
2426 * . synchronous write of xtree (right to left, bottom to top);
2427 * . at start of logredo(): init in-memory imap with one additional iag page;
2428 * . at end of logredo(): re-read imap inode to determine
2429 *   new imap size;
2430 */
2431static int
2432diNewIAG(struct inomap * imap, int *iagnop, int agno, struct metapage ** mpp)
2433{
2434	int rc;
2435	int iagno, i, xlen;
2436	struct inode *ipimap;
2437	struct super_block *sb;
2438	struct jfs_sb_info *sbi;
2439	struct metapage *mp;
2440	struct iag *iagp;
2441	s64 xaddr = 0;
2442	s64 blkno;
2443	tid_t tid;
2444	struct inode *iplist[1];
2445
2446	/* pick up pointers to the inode map and mount inodes */
2447	ipimap = imap->im_ipimap;
2448	sb = ipimap->i_sb;
2449	sbi = JFS_SBI(sb);
2450
2451	/* acquire the free iag lock */
2452	IAGFREE_LOCK(imap);
2453
2454	/* if there are any iags on the inode map free iag list,
2455	 * allocate the iag from the head of the list.
2456	 */
2457	if (imap->im_freeiag >= 0) {
2458		/* pick up the iag number at the head of the list */
2459		iagno = imap->im_freeiag;
2460
2461		/* determine the logical block number of the iag */
2462		blkno = IAGTOLBLK(iagno, sbi->l2nbperpage);
2463	} else {
2464		/* no free iags. the inode map will have to be extented
2465		 * to include a new iag.
2466		 */
2467
2468		/* acquire inode map lock */
2469		IWRITE_LOCK(ipimap, RDWRLOCK_IMAP);
2470
2471		if (ipimap->i_size >> L2PSIZE != imap->im_nextiag + 1) {
2472			IWRITE_UNLOCK(ipimap);
2473			IAGFREE_UNLOCK(imap);
2474			jfs_error(imap->im_ipimap->i_sb,
2475				  "ipimap->i_size is wrong\n");
2476			return -EIO;
2477		}
2478
2479
2480		/* get the next available iag number */
2481		iagno = imap->im_nextiag;
2482
2483		/* make sure that we have not exceeded the maximum inode
2484		 * number limit.
2485		 */
2486		if (iagno > (MAXIAGS - 1)) {
2487			/* release the inode map lock */
2488			IWRITE_UNLOCK(ipimap);
2489
2490			rc = -ENOSPC;
2491			goto out;
2492		}
2493
2494		/*
2495		 * synchronously append new iag page.
2496		 */
2497		/* determine the logical address of iag page to append */
2498		blkno = IAGTOLBLK(iagno, sbi->l2nbperpage);
2499
2500		/* Allocate extent for new iag page */
2501		xlen = sbi->nbperpage;
2502		if ((rc = dbAlloc(ipimap, 0, (s64) xlen, &xaddr))) {
2503			/* release the inode map lock */
2504			IWRITE_UNLOCK(ipimap);
2505
2506			goto out;
2507		}
2508
2509		/*
2510		 * start transaction of update of the inode map
2511		 * addressing structure pointing to the new iag page;
2512		 */
2513		tid = txBegin(sb, COMMIT_FORCE);
2514		mutex_lock(&JFS_IP(ipimap)->commit_mutex);
2515
2516		/* update the inode map addressing structure to point to it */
2517		if ((rc =
2518		     xtInsert(tid, ipimap, 0, blkno, xlen, &xaddr, 0))) {
2519			txEnd(tid);
2520			mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
2521			/* Free the blocks allocated for the iag since it was
2522			 * not successfully added to the inode map
2523			 */
2524			dbFree(ipimap, xaddr, (s64) xlen);
2525
2526			/* release the inode map lock */
2527			IWRITE_UNLOCK(ipimap);
2528
2529			goto out;
2530		}
2531
2532		/* update the inode map's inode to reflect the extension */
2533		ipimap->i_size += PSIZE;
2534		inode_add_bytes(ipimap, PSIZE);
2535
2536		/* assign a buffer for the page */
2537		mp = get_metapage(ipimap, blkno, PSIZE, 0);
2538		if (!mp) {
2539			/*
2540			 * This is very unlikely since we just created the
2541			 * extent, but let's try to handle it correctly
2542			 */
2543			xtTruncate(tid, ipimap, ipimap->i_size - PSIZE,
2544				   COMMIT_PWMAP);
2545
2546			txAbort(tid, 0);
2547			txEnd(tid);
2548			mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
2549
2550			/* release the inode map lock */
2551			IWRITE_UNLOCK(ipimap);
2552
2553			rc = -EIO;
2554			goto out;
2555		}
2556		iagp = (struct iag *) mp->data;
2557
2558		/* init the iag */
2559		memset(iagp, 0, sizeof(struct iag));
2560		iagp->iagnum = cpu_to_le32(iagno);
2561		iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1);
2562		iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1);
2563		iagp->iagfree = cpu_to_le32(-1);
2564		iagp->nfreeinos = 0;
2565		iagp->nfreeexts = cpu_to_le32(EXTSPERIAG);
2566
2567		/* initialize the free inode summary map (free extent
2568		 * summary map initialization handled by bzero).
2569		 */
2570		for (i = 0; i < SMAPSZ; i++)
2571			iagp->inosmap[i] = cpu_to_le32(ONES);
2572
2573		/*
2574		 * Write and sync the metapage
2575		 */
2576		flush_metapage(mp);
2577
2578		/*
2579		 * txCommit(COMMIT_FORCE) will synchronously write address
2580		 * index pages and inode after commit in careful update order
2581		 * of address index pages (right to left, bottom up);
2582		 */
2583		iplist[0] = ipimap;
2584		rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE);
2585
2586		txEnd(tid);
2587		mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
2588
2589		duplicateIXtree(sb, blkno, xlen, &xaddr);
2590
2591		/* update the next available iag number */
2592		imap->im_nextiag += 1;
2593
2594		/* Add the iag to the iag free list so we don't lose the iag
2595		 * if a failure happens now.
2596		 */
2597		imap->im_freeiag = iagno;
2598
2599		/* Until we have logredo working, we want the imap inode &
2600		 * control page to be up to date.
2601		 */
2602		diSync(ipimap);
2603
2604		/* release the inode map lock */
2605		IWRITE_UNLOCK(ipimap);
2606	}
2607
2608	/* obtain read lock on map */
2609	IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
2610
2611	/* read the iag */
2612	if ((rc = diIAGRead(imap, iagno, &mp))) {
2613		IREAD_UNLOCK(ipimap);
2614		rc = -EIO;
2615		goto out;
2616	}
2617	iagp = (struct iag *) mp->data;
2618
2619	/* remove the iag from the iag free list */
2620	imap->im_freeiag = le32_to_cpu(iagp->iagfree);
2621	iagp->iagfree = cpu_to_le32(-1);
2622
2623	/* set the return iag number and buffer pointer */
2624	*iagnop = iagno;
2625	*mpp = mp;
2626
2627      out:
2628	/* release the iag free lock */
2629	IAGFREE_UNLOCK(imap);
2630
2631	return (rc);
2632}
2633
2634/*
2635 * NAME:	diIAGRead()
2636 *
2637 * FUNCTION:	get the buffer for the specified iag within a fileset
2638 *		or aggregate inode map.
2639 *
2640 * PARAMETERS:
2641 *	imap	- pointer to inode map control structure.
2642 *	iagno	- iag number.
2643 *	bpp	- point to buffer pointer to be filled in on successful
2644 *		  exit.
2645 *
2646 * SERIALIZATION:
2647 *	must have read lock on imap inode
2648 *	(When called by diExtendFS, the filesystem is quiesced, therefore
2649 *	 the read lock is unnecessary.)
2650 *
2651 * RETURN VALUES:
2652 *	0	- success.
2653 *	-EIO	- i/o error.
2654 */
2655static int diIAGRead(struct inomap * imap, int iagno, struct metapage ** mpp)
2656{
2657	struct inode *ipimap = imap->im_ipimap;
2658	s64 blkno;
2659
2660	/* compute the logical block number of the iag. */
2661	blkno = IAGTOLBLK(iagno, JFS_SBI(ipimap->i_sb)->l2nbperpage);
2662
2663	/* read the iag. */
2664	*mpp = read_metapage(ipimap, blkno, PSIZE, 0);
2665	if (*mpp == NULL) {
2666		return -EIO;
2667	}
2668
2669	return (0);
2670}
2671
2672/*
2673 * NAME:	diFindFree()
2674 *
2675 * FUNCTION:	find the first free bit in a word starting at
2676 *		the specified bit position.
2677 *
2678 * PARAMETERS:
2679 *	word	- word to be examined.
2680 *	start	- starting bit position.
2681 *
2682 * RETURN VALUES:
2683 *	bit position of first free bit in the word or 32 if
2684 *	no free bits were found.
2685 */
2686static int diFindFree(u32 word, int start)
2687{
2688	int bitno;
2689	assert(start < 32);
2690	/* scan the word for the first free bit. */
2691	for (word <<= start, bitno = start; bitno < 32;
2692	     bitno++, word <<= 1) {
2693		if ((word & HIGHORDER) == 0)
2694			break;
2695	}
2696	return (bitno);
2697}
2698
2699/*
2700 * NAME:	diUpdatePMap()
2701 *
2702 * FUNCTION: Update the persistent map in an IAG for the allocation or
2703 *	freeing of the specified inode.
2704 *
2705 * PRE CONDITIONS: Working map has already been updated for allocate.
2706 *
2707 * PARAMETERS:
2708 *	ipimap	- Incore inode map inode
2709 *	inum	- Number of inode to mark in permanent map
2710 *	is_free	- If 'true' indicates inode should be marked freed, otherwise
2711 *		  indicates inode should be marked allocated.
2712 *
2713 * RETURN VALUES:
2714 *		0 for success
2715 */
2716int
2717diUpdatePMap(struct inode *ipimap,
2718	     unsigned long inum, bool is_free, struct tblock * tblk)
2719{
2720	int rc;
2721	struct iag *iagp;
2722	struct metapage *mp;
2723	int iagno, ino, extno, bitno;
2724	struct inomap *imap;
2725	u32 mask;
2726	struct jfs_log *log;
2727	int lsn, difft, diffp;
2728	unsigned long flags;
2729
2730	imap = JFS_IP(ipimap)->i_imap;
2731	/* get the iag number containing the inode */
2732	iagno = INOTOIAG(inum);
2733	/* make sure that the iag is contained within the map */
2734	if (iagno >= imap->im_nextiag) {
2735		jfs_error(ipimap->i_sb, "the iag is outside the map\n");
 
2736		return -EIO;
2737	}
2738	/* read the iag */
2739	IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
2740	rc = diIAGRead(imap, iagno, &mp);
2741	IREAD_UNLOCK(ipimap);
2742	if (rc)
2743		return (rc);
2744	metapage_wait_for_io(mp);
2745	iagp = (struct iag *) mp->data;
2746	/* get the inode number and extent number of the inode within
2747	 * the iag and the inode number within the extent.
2748	 */
2749	ino = inum & (INOSPERIAG - 1);
2750	extno = ino >> L2INOSPEREXT;
2751	bitno = ino & (INOSPEREXT - 1);
2752	mask = HIGHORDER >> bitno;
2753	/*
2754	 * mark the inode free in persistent map:
2755	 */
2756	if (is_free) {
2757		/* The inode should have been allocated both in working
2758		 * map and in persistent map;
2759		 * the inode will be freed from working map at the release
2760		 * of last reference release;
2761		 */
2762		if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
2763			jfs_error(ipimap->i_sb,
2764				  "inode %ld not marked as allocated in wmap!\n",
2765				  inum);
2766		}
2767		if (!(le32_to_cpu(iagp->pmap[extno]) & mask)) {
2768			jfs_error(ipimap->i_sb,
2769				  "inode %ld not marked as allocated in pmap!\n",
2770				  inum);
2771		}
2772		/* update the bitmap for the extent of the freed inode */
2773		iagp->pmap[extno] &= cpu_to_le32(~mask);
2774	}
2775	/*
2776	 * mark the inode allocated in persistent map:
2777	 */
2778	else {
2779		/* The inode should be already allocated in the working map
2780		 * and should be free in persistent map;
2781		 */
2782		if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
2783			release_metapage(mp);
2784			jfs_error(ipimap->i_sb,
2785				  "the inode is not allocated in the working map\n");
 
2786			return -EIO;
2787		}
2788		if ((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) {
2789			release_metapage(mp);
2790			jfs_error(ipimap->i_sb,
2791				  "the inode is not free in the persistent map\n");
 
2792			return -EIO;
2793		}
2794		/* update the bitmap for the extent of the allocated inode */
2795		iagp->pmap[extno] |= cpu_to_le32(mask);
2796	}
2797	/*
2798	 * update iag lsn
2799	 */
2800	lsn = tblk->lsn;
2801	log = JFS_SBI(tblk->sb)->log;
2802	LOGSYNC_LOCK(log, flags);
2803	if (mp->lsn != 0) {
2804		/* inherit older/smaller lsn */
2805		logdiff(difft, lsn, log);
2806		logdiff(diffp, mp->lsn, log);
2807		if (difft < diffp) {
2808			mp->lsn = lsn;
2809			/* move mp after tblock in logsync list */
2810			list_move(&mp->synclist, &tblk->synclist);
2811		}
2812		/* inherit younger/larger clsn */
2813		assert(mp->clsn);
2814		logdiff(difft, tblk->clsn, log);
2815		logdiff(diffp, mp->clsn, log);
2816		if (difft > diffp)
2817			mp->clsn = tblk->clsn;
2818	} else {
2819		mp->log = log;
2820		mp->lsn = lsn;
2821		/* insert mp after tblock in logsync list */
2822		log->count++;
2823		list_add(&mp->synclist, &tblk->synclist);
2824		mp->clsn = tblk->clsn;
2825	}
2826	LOGSYNC_UNLOCK(log, flags);
2827	write_metapage(mp);
2828	return (0);
2829}
2830
2831/*
2832 *	diExtendFS()
2833 *
2834 * function: update imap for extendfs();
2835 *
2836 * note: AG size has been increased s.t. each k old contiguous AGs are
2837 * coalesced into a new AG;
2838 */
2839int diExtendFS(struct inode *ipimap, struct inode *ipbmap)
2840{
2841	int rc, rcx = 0;
2842	struct inomap *imap = JFS_IP(ipimap)->i_imap;
2843	struct iag *iagp = NULL, *hiagp = NULL;
2844	struct bmap *mp = JFS_SBI(ipbmap->i_sb)->bmap;
2845	struct metapage *bp, *hbp;
2846	int i, n, head;
2847	int numinos, xnuminos = 0, xnumfree = 0;
2848	s64 agstart;
2849
2850	jfs_info("diExtendFS: nextiag:%d numinos:%d numfree:%d",
2851		   imap->im_nextiag, atomic_read(&imap->im_numinos),
2852		   atomic_read(&imap->im_numfree));
2853
2854	/*
2855	 *	reconstruct imap
2856	 *
2857	 * coalesce contiguous k (newAGSize/oldAGSize) AGs;
2858	 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn;
2859	 * note: new AG size = old AG size * (2**x).
2860	 */
2861
2862	/* init per AG control information im_agctl[] */
2863	for (i = 0; i < MAXAG; i++) {
2864		imap->im_agctl[i].inofree = -1;
2865		imap->im_agctl[i].extfree = -1;
2866		imap->im_agctl[i].numinos = 0;	/* number of backed inodes */
2867		imap->im_agctl[i].numfree = 0;	/* number of free backed inodes */
2868	}
2869
2870	/*
2871	 *	process each iag page of the map.
2872	 *
2873	 * rebuild AG Free Inode List, AG Free Inode Extent List;
2874	 */
2875	for (i = 0; i < imap->im_nextiag; i++) {
2876		if ((rc = diIAGRead(imap, i, &bp))) {
2877			rcx = rc;
2878			continue;
2879		}
2880		iagp = (struct iag *) bp->data;
2881		if (le32_to_cpu(iagp->iagnum) != i) {
2882			release_metapage(bp);
2883			jfs_error(ipimap->i_sb, "unexpected value of iagnum\n");
 
2884			return -EIO;
2885		}
2886
2887		/* leave free iag in the free iag list */
2888		if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2889			release_metapage(bp);
2890			continue;
2891		}
2892
2893		agstart = le64_to_cpu(iagp->agstart);
2894		n = agstart >> mp->db_agl2size;
2895		iagp->agstart = cpu_to_le64((s64)n << mp->db_agl2size);
2896
2897		/* compute backed inodes */
2898		numinos = (EXTSPERIAG - le32_to_cpu(iagp->nfreeexts))
2899		    << L2INOSPEREXT;
2900		if (numinos > 0) {
2901			/* merge AG backed inodes */
2902			imap->im_agctl[n].numinos += numinos;
2903			xnuminos += numinos;
2904		}
2905
2906		/* if any backed free inodes, insert at AG free inode list */
2907		if ((int) le32_to_cpu(iagp->nfreeinos) > 0) {
2908			if ((head = imap->im_agctl[n].inofree) == -1) {
2909				iagp->inofreefwd = cpu_to_le32(-1);
2910				iagp->inofreeback = cpu_to_le32(-1);
2911			} else {
2912				if ((rc = diIAGRead(imap, head, &hbp))) {
2913					rcx = rc;
2914					goto nextiag;
2915				}
2916				hiagp = (struct iag *) hbp->data;
2917				hiagp->inofreeback = iagp->iagnum;
2918				iagp->inofreefwd = cpu_to_le32(head);
2919				iagp->inofreeback = cpu_to_le32(-1);
2920				write_metapage(hbp);
2921			}
2922
2923			imap->im_agctl[n].inofree =
2924			    le32_to_cpu(iagp->iagnum);
2925
2926			/* merge AG backed free inodes */
2927			imap->im_agctl[n].numfree +=
2928			    le32_to_cpu(iagp->nfreeinos);
2929			xnumfree += le32_to_cpu(iagp->nfreeinos);
2930		}
2931
2932		/* if any free extents, insert at AG free extent list */
2933		if (le32_to_cpu(iagp->nfreeexts) > 0) {
2934			if ((head = imap->im_agctl[n].extfree) == -1) {
2935				iagp->extfreefwd = cpu_to_le32(-1);
2936				iagp->extfreeback = cpu_to_le32(-1);
2937			} else {
2938				if ((rc = diIAGRead(imap, head, &hbp))) {
2939					rcx = rc;
2940					goto nextiag;
2941				}
2942				hiagp = (struct iag *) hbp->data;
2943				hiagp->extfreeback = iagp->iagnum;
2944				iagp->extfreefwd = cpu_to_le32(head);
2945				iagp->extfreeback = cpu_to_le32(-1);
2946				write_metapage(hbp);
2947			}
2948
2949			imap->im_agctl[n].extfree =
2950			    le32_to_cpu(iagp->iagnum);
2951		}
2952
2953	      nextiag:
2954		write_metapage(bp);
2955	}
2956
2957	if (xnuminos != atomic_read(&imap->im_numinos) ||
2958	    xnumfree != atomic_read(&imap->im_numfree)) {
2959		jfs_error(ipimap->i_sb, "numinos or numfree incorrect\n");
 
2960		return -EIO;
2961	}
2962
2963	return rcx;
2964}
2965
2966
2967/*
2968 *	duplicateIXtree()
2969 *
2970 * serialization: IWRITE_LOCK held on entry/exit
2971 *
2972 * note: shadow page with regular inode (rel.2);
2973 */
2974static void duplicateIXtree(struct super_block *sb, s64 blkno,
2975			    int xlen, s64 *xaddr)
2976{
2977	struct jfs_superblock *j_sb;
2978	struct buffer_head *bh;
2979	struct inode *ip;
2980	tid_t tid;
2981
2982	/* if AIT2 ipmap2 is bad, do not try to update it */
2983	if (JFS_SBI(sb)->mntflag & JFS_BAD_SAIT)	/* s_flag */
2984		return;
2985	ip = diReadSpecial(sb, FILESYSTEM_I, 1);
2986	if (ip == NULL) {
2987		JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT;
2988		if (readSuper(sb, &bh))
2989			return;
2990		j_sb = (struct jfs_superblock *)bh->b_data;
2991		j_sb->s_flag |= cpu_to_le32(JFS_BAD_SAIT);
2992
2993		mark_buffer_dirty(bh);
2994		sync_dirty_buffer(bh);
2995		brelse(bh);
2996		return;
2997	}
2998
2999	/* start transaction */
3000	tid = txBegin(sb, COMMIT_FORCE);
3001	/* update the inode map addressing structure to point to it */
3002	if (xtInsert(tid, ip, 0, blkno, xlen, xaddr, 0)) {
3003		JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT;
3004		txAbort(tid, 1);
3005		goto cleanup;
3006
3007	}
3008	/* update the inode map's inode to reflect the extension */
3009	ip->i_size += PSIZE;
3010	inode_add_bytes(ip, PSIZE);
3011	txCommit(tid, 1, &ip, COMMIT_FORCE);
3012      cleanup:
3013	txEnd(tid);
3014	diFreeSpecial(ip);
3015}
3016
3017/*
3018 * NAME:	copy_from_dinode()
3019 *
3020 * FUNCTION:	Copies inode info from disk inode to in-memory inode
3021 *
3022 * RETURN VALUES:
3023 *	0	- success
3024 *	-ENOMEM	- insufficient memory
3025 */
3026static int copy_from_dinode(struct dinode * dip, struct inode *ip)
3027{
3028	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
3029	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
3030
3031	jfs_ip->fileset = le32_to_cpu(dip->di_fileset);
3032	jfs_ip->mode2 = le32_to_cpu(dip->di_mode);
3033	jfs_set_inode_flags(ip);
3034
3035	ip->i_mode = le32_to_cpu(dip->di_mode) & 0xffff;
3036	if (sbi->umask != -1) {
3037		ip->i_mode = (ip->i_mode & ~0777) | (0777 & ~sbi->umask);
3038		/* For directories, add x permission if r is allowed by umask */
3039		if (S_ISDIR(ip->i_mode)) {
3040			if (ip->i_mode & 0400)
3041				ip->i_mode |= 0100;
3042			if (ip->i_mode & 0040)
3043				ip->i_mode |= 0010;
3044			if (ip->i_mode & 0004)
3045				ip->i_mode |= 0001;
3046		}
3047	}
3048	set_nlink(ip, le32_to_cpu(dip->di_nlink));
3049
3050	jfs_ip->saved_uid = make_kuid(&init_user_ns, le32_to_cpu(dip->di_uid));
3051	if (!uid_valid(sbi->uid))
3052		ip->i_uid = jfs_ip->saved_uid;
3053	else {
3054		ip->i_uid = sbi->uid;
3055	}
3056
3057	jfs_ip->saved_gid = make_kgid(&init_user_ns, le32_to_cpu(dip->di_gid));
3058	if (!gid_valid(sbi->gid))
3059		ip->i_gid = jfs_ip->saved_gid;
3060	else {
3061		ip->i_gid = sbi->gid;
3062	}
3063
3064	ip->i_size = le64_to_cpu(dip->di_size);
3065	ip->i_atime.tv_sec = le32_to_cpu(dip->di_atime.tv_sec);
3066	ip->i_atime.tv_nsec = le32_to_cpu(dip->di_atime.tv_nsec);
3067	ip->i_mtime.tv_sec = le32_to_cpu(dip->di_mtime.tv_sec);
3068	ip->i_mtime.tv_nsec = le32_to_cpu(dip->di_mtime.tv_nsec);
3069	ip->i_ctime.tv_sec = le32_to_cpu(dip->di_ctime.tv_sec);
3070	ip->i_ctime.tv_nsec = le32_to_cpu(dip->di_ctime.tv_nsec);
3071	ip->i_blocks = LBLK2PBLK(ip->i_sb, le64_to_cpu(dip->di_nblocks));
3072	ip->i_generation = le32_to_cpu(dip->di_gen);
3073
3074	jfs_ip->ixpxd = dip->di_ixpxd;	/* in-memory pxd's are little-endian */
3075	jfs_ip->acl = dip->di_acl;	/* as are dxd's */
3076	jfs_ip->ea = dip->di_ea;
3077	jfs_ip->next_index = le32_to_cpu(dip->di_next_index);
3078	jfs_ip->otime = le32_to_cpu(dip->di_otime.tv_sec);
3079	jfs_ip->acltype = le32_to_cpu(dip->di_acltype);
3080
3081	if (S_ISCHR(ip->i_mode) || S_ISBLK(ip->i_mode)) {
3082		jfs_ip->dev = le32_to_cpu(dip->di_rdev);
3083		ip->i_rdev = new_decode_dev(jfs_ip->dev);
3084	}
3085
3086	if (S_ISDIR(ip->i_mode)) {
3087		memcpy(&jfs_ip->i_dirtable, &dip->di_dirtable, 384);
3088	} else if (S_ISREG(ip->i_mode) || S_ISLNK(ip->i_mode)) {
3089		memcpy(&jfs_ip->i_xtroot, &dip->di_xtroot, 288);
3090	} else
3091		memcpy(&jfs_ip->i_inline_ea, &dip->di_inlineea, 128);
3092
3093	/* Zero the in-memory-only stuff */
3094	jfs_ip->cflag = 0;
3095	jfs_ip->btindex = 0;
3096	jfs_ip->btorder = 0;
3097	jfs_ip->bxflag = 0;
3098	jfs_ip->blid = 0;
3099	jfs_ip->atlhead = 0;
3100	jfs_ip->atltail = 0;
3101	jfs_ip->xtlid = 0;
3102	return (0);
3103}
3104
3105/*
3106 * NAME:	copy_to_dinode()
3107 *
3108 * FUNCTION:	Copies inode info from in-memory inode to disk inode
3109 */
3110static void copy_to_dinode(struct dinode * dip, struct inode *ip)
3111{
3112	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
3113	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
3114
3115	dip->di_fileset = cpu_to_le32(jfs_ip->fileset);
3116	dip->di_inostamp = cpu_to_le32(sbi->inostamp);
3117	dip->di_number = cpu_to_le32(ip->i_ino);
3118	dip->di_gen = cpu_to_le32(ip->i_generation);
3119	dip->di_size = cpu_to_le64(ip->i_size);
3120	dip->di_nblocks = cpu_to_le64(PBLK2LBLK(ip->i_sb, ip->i_blocks));
3121	dip->di_nlink = cpu_to_le32(ip->i_nlink);
3122	if (!uid_valid(sbi->uid))
3123		dip->di_uid = cpu_to_le32(i_uid_read(ip));
3124	else
3125		dip->di_uid =cpu_to_le32(from_kuid(&init_user_ns,
3126						   jfs_ip->saved_uid));
3127	if (!gid_valid(sbi->gid))
3128		dip->di_gid = cpu_to_le32(i_gid_read(ip));
3129	else
3130		dip->di_gid = cpu_to_le32(from_kgid(&init_user_ns,
3131						    jfs_ip->saved_gid));
3132	/*
3133	 * mode2 is only needed for storing the higher order bits.
3134	 * Trust i_mode for the lower order ones
3135	 */
3136	if (sbi->umask == -1)
3137		dip->di_mode = cpu_to_le32((jfs_ip->mode2 & 0xffff0000) |
3138					   ip->i_mode);
3139	else /* Leave the original permissions alone */
3140		dip->di_mode = cpu_to_le32(jfs_ip->mode2);
3141
3142	dip->di_atime.tv_sec = cpu_to_le32(ip->i_atime.tv_sec);
3143	dip->di_atime.tv_nsec = cpu_to_le32(ip->i_atime.tv_nsec);
3144	dip->di_ctime.tv_sec = cpu_to_le32(ip->i_ctime.tv_sec);
3145	dip->di_ctime.tv_nsec = cpu_to_le32(ip->i_ctime.tv_nsec);
3146	dip->di_mtime.tv_sec = cpu_to_le32(ip->i_mtime.tv_sec);
3147	dip->di_mtime.tv_nsec = cpu_to_le32(ip->i_mtime.tv_nsec);
3148	dip->di_ixpxd = jfs_ip->ixpxd;	/* in-memory pxd's are little-endian */
3149	dip->di_acl = jfs_ip->acl;	/* as are dxd's */
3150	dip->di_ea = jfs_ip->ea;
3151	dip->di_next_index = cpu_to_le32(jfs_ip->next_index);
3152	dip->di_otime.tv_sec = cpu_to_le32(jfs_ip->otime);
3153	dip->di_otime.tv_nsec = 0;
3154	dip->di_acltype = cpu_to_le32(jfs_ip->acltype);
3155	if (S_ISCHR(ip->i_mode) || S_ISBLK(ip->i_mode))
3156		dip->di_rdev = cpu_to_le32(jfs_ip->dev);
3157}
v3.1
 
   1/*
   2 *   Copyright (C) International Business Machines Corp., 2000-2004
   3 *
   4 *   This program is free software;  you can redistribute it and/or modify
   5 *   it under the terms of the GNU General Public License as published by
   6 *   the Free Software Foundation; either version 2 of the License, or
   7 *   (at your option) any later version.
   8 *
   9 *   This program is distributed in the hope that it will be useful,
  10 *   but WITHOUT ANY WARRANTY;  without even the implied warranty of
  11 *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See
  12 *   the GNU General Public License for more details.
  13 *
  14 *   You should have received a copy of the GNU General Public License
  15 *   along with this program;  if not, write to the Free Software
  16 *   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  17 */
  18
  19/*
  20 *	jfs_imap.c: inode allocation map manager
  21 *
  22 * Serialization:
  23 *   Each AG has a simple lock which is used to control the serialization of
  24 *	the AG level lists.  This lock should be taken first whenever an AG
  25 *	level list will be modified or accessed.
  26 *
  27 *   Each IAG is locked by obtaining the buffer for the IAG page.
  28 *
  29 *   There is also a inode lock for the inode map inode.  A read lock needs to
  30 *	be taken whenever an IAG is read from the map or the global level
  31 *	information is read.  A write lock needs to be taken whenever the global
  32 *	level information is modified or an atomic operation needs to be used.
  33 *
  34 *	If more than one IAG is read at one time, the read lock may not
  35 *	be given up until all of the IAG's are read.  Otherwise, a deadlock
  36 *	may occur when trying to obtain the read lock while another thread
  37 *	holding the read lock is waiting on the IAG already being held.
  38 *
  39 *   The control page of the inode map is read into memory by diMount().
  40 *	Thereafter it should only be modified in memory and then it will be
  41 *	written out when the filesystem is unmounted by diUnmount().
  42 */
  43
  44#include <linux/fs.h>
  45#include <linux/buffer_head.h>
  46#include <linux/pagemap.h>
  47#include <linux/quotaops.h>
  48#include <linux/slab.h>
  49
  50#include "jfs_incore.h"
  51#include "jfs_inode.h"
  52#include "jfs_filsys.h"
  53#include "jfs_dinode.h"
  54#include "jfs_dmap.h"
  55#include "jfs_imap.h"
  56#include "jfs_metapage.h"
  57#include "jfs_superblock.h"
  58#include "jfs_debug.h"
  59
  60/*
  61 * imap locks
  62 */
  63/* iag free list lock */
  64#define IAGFREE_LOCK_INIT(imap)		mutex_init(&imap->im_freelock)
  65#define IAGFREE_LOCK(imap)		mutex_lock(&imap->im_freelock)
  66#define IAGFREE_UNLOCK(imap)		mutex_unlock(&imap->im_freelock)
  67
  68/* per ag iag list locks */
  69#define AG_LOCK_INIT(imap,index)	mutex_init(&(imap->im_aglock[index]))
  70#define AG_LOCK(imap,agno)		mutex_lock(&imap->im_aglock[agno])
  71#define AG_UNLOCK(imap,agno)		mutex_unlock(&imap->im_aglock[agno])
  72
  73/*
  74 * forward references
  75 */
  76static int diAllocAG(struct inomap *, int, bool, struct inode *);
  77static int diAllocAny(struct inomap *, int, bool, struct inode *);
  78static int diAllocBit(struct inomap *, struct iag *, int);
  79static int diAllocExt(struct inomap *, int, struct inode *);
  80static int diAllocIno(struct inomap *, int, struct inode *);
  81static int diFindFree(u32, int);
  82static int diNewExt(struct inomap *, struct iag *, int);
  83static int diNewIAG(struct inomap *, int *, int, struct metapage **);
  84static void duplicateIXtree(struct super_block *, s64, int, s64 *);
  85
  86static int diIAGRead(struct inomap * imap, int, struct metapage **);
  87static int copy_from_dinode(struct dinode *, struct inode *);
  88static void copy_to_dinode(struct dinode *, struct inode *);
  89
  90/*
  91 * NAME:	diMount()
  92 *
  93 * FUNCTION:	initialize the incore inode map control structures for
  94 *		a fileset or aggregate init time.
  95 *
  96 *		the inode map's control structure (dinomap) is
  97 *		brought in from disk and placed in virtual memory.
  98 *
  99 * PARAMETERS:
 100 *	ipimap	- pointer to inode map inode for the aggregate or fileset.
 101 *
 102 * RETURN VALUES:
 103 *	0	- success
 104 *	-ENOMEM	- insufficient free virtual memory.
 105 *	-EIO	- i/o error.
 106 */
 107int diMount(struct inode *ipimap)
 108{
 109	struct inomap *imap;
 110	struct metapage *mp;
 111	int index;
 112	struct dinomap_disk *dinom_le;
 113
 114	/*
 115	 * allocate/initialize the in-memory inode map control structure
 116	 */
 117	/* allocate the in-memory inode map control structure. */
 118	imap = kmalloc(sizeof(struct inomap), GFP_KERNEL);
 119	if (imap == NULL) {
 120		jfs_err("diMount: kmalloc returned NULL!");
 121		return -ENOMEM;
 122	}
 123
 124	/* read the on-disk inode map control structure. */
 125
 126	mp = read_metapage(ipimap,
 127			   IMAPBLKNO << JFS_SBI(ipimap->i_sb)->l2nbperpage,
 128			   PSIZE, 0);
 129	if (mp == NULL) {
 130		kfree(imap);
 131		return -EIO;
 132	}
 133
 134	/* copy the on-disk version to the in-memory version. */
 135	dinom_le = (struct dinomap_disk *) mp->data;
 136	imap->im_freeiag = le32_to_cpu(dinom_le->in_freeiag);
 137	imap->im_nextiag = le32_to_cpu(dinom_le->in_nextiag);
 138	atomic_set(&imap->im_numinos, le32_to_cpu(dinom_le->in_numinos));
 139	atomic_set(&imap->im_numfree, le32_to_cpu(dinom_le->in_numfree));
 140	imap->im_nbperiext = le32_to_cpu(dinom_le->in_nbperiext);
 141	imap->im_l2nbperiext = le32_to_cpu(dinom_le->in_l2nbperiext);
 142	for (index = 0; index < MAXAG; index++) {
 143		imap->im_agctl[index].inofree =
 144		    le32_to_cpu(dinom_le->in_agctl[index].inofree);
 145		imap->im_agctl[index].extfree =
 146		    le32_to_cpu(dinom_le->in_agctl[index].extfree);
 147		imap->im_agctl[index].numinos =
 148		    le32_to_cpu(dinom_le->in_agctl[index].numinos);
 149		imap->im_agctl[index].numfree =
 150		    le32_to_cpu(dinom_le->in_agctl[index].numfree);
 151	}
 152
 153	/* release the buffer. */
 154	release_metapage(mp);
 155
 156	/*
 157	 * allocate/initialize inode allocation map locks
 158	 */
 159	/* allocate and init iag free list lock */
 160	IAGFREE_LOCK_INIT(imap);
 161
 162	/* allocate and init ag list locks */
 163	for (index = 0; index < MAXAG; index++) {
 164		AG_LOCK_INIT(imap, index);
 165	}
 166
 167	/* bind the inode map inode and inode map control structure
 168	 * to each other.
 169	 */
 170	imap->im_ipimap = ipimap;
 171	JFS_IP(ipimap)->i_imap = imap;
 172
 173	return (0);
 174}
 175
 176
 177/*
 178 * NAME:	diUnmount()
 179 *
 180 * FUNCTION:	write to disk the incore inode map control structures for
 181 *		a fileset or aggregate at unmount time.
 182 *
 183 * PARAMETERS:
 184 *	ipimap	- pointer to inode map inode for the aggregate or fileset.
 185 *
 186 * RETURN VALUES:
 187 *	0	- success
 188 *	-ENOMEM	- insufficient free virtual memory.
 189 *	-EIO	- i/o error.
 190 */
 191int diUnmount(struct inode *ipimap, int mounterror)
 192{
 193	struct inomap *imap = JFS_IP(ipimap)->i_imap;
 194
 195	/*
 196	 * update the on-disk inode map control structure
 197	 */
 198
 199	if (!(mounterror || isReadOnly(ipimap)))
 200		diSync(ipimap);
 201
 202	/*
 203	 * Invalidate the page cache buffers
 204	 */
 205	truncate_inode_pages(ipimap->i_mapping, 0);
 206
 207	/*
 208	 * free in-memory control structure
 209	 */
 210	kfree(imap);
 211
 212	return (0);
 213}
 214
 215
 216/*
 217 *	diSync()
 218 */
 219int diSync(struct inode *ipimap)
 220{
 221	struct dinomap_disk *dinom_le;
 222	struct inomap *imp = JFS_IP(ipimap)->i_imap;
 223	struct metapage *mp;
 224	int index;
 225
 226	/*
 227	 * write imap global conrol page
 228	 */
 229	/* read the on-disk inode map control structure */
 230	mp = get_metapage(ipimap,
 231			  IMAPBLKNO << JFS_SBI(ipimap->i_sb)->l2nbperpage,
 232			  PSIZE, 0);
 233	if (mp == NULL) {
 234		jfs_err("diSync: get_metapage failed!");
 235		return -EIO;
 236	}
 237
 238	/* copy the in-memory version to the on-disk version */
 239	dinom_le = (struct dinomap_disk *) mp->data;
 240	dinom_le->in_freeiag = cpu_to_le32(imp->im_freeiag);
 241	dinom_le->in_nextiag = cpu_to_le32(imp->im_nextiag);
 242	dinom_le->in_numinos = cpu_to_le32(atomic_read(&imp->im_numinos));
 243	dinom_le->in_numfree = cpu_to_le32(atomic_read(&imp->im_numfree));
 244	dinom_le->in_nbperiext = cpu_to_le32(imp->im_nbperiext);
 245	dinom_le->in_l2nbperiext = cpu_to_le32(imp->im_l2nbperiext);
 246	for (index = 0; index < MAXAG; index++) {
 247		dinom_le->in_agctl[index].inofree =
 248		    cpu_to_le32(imp->im_agctl[index].inofree);
 249		dinom_le->in_agctl[index].extfree =
 250		    cpu_to_le32(imp->im_agctl[index].extfree);
 251		dinom_le->in_agctl[index].numinos =
 252		    cpu_to_le32(imp->im_agctl[index].numinos);
 253		dinom_le->in_agctl[index].numfree =
 254		    cpu_to_le32(imp->im_agctl[index].numfree);
 255	}
 256
 257	/* write out the control structure */
 258	write_metapage(mp);
 259
 260	/*
 261	 * write out dirty pages of imap
 262	 */
 263	filemap_write_and_wait(ipimap->i_mapping);
 264
 265	diWriteSpecial(ipimap, 0);
 266
 267	return (0);
 268}
 269
 270
 271/*
 272 * NAME:	diRead()
 273 *
 274 * FUNCTION:	initialize an incore inode from disk.
 275 *
 276 *		on entry, the specifed incore inode should itself
 277 *		specify the disk inode number corresponding to the
 278 *		incore inode (i.e. i_number should be initialized).
 279 *
 280 *		this routine handles incore inode initialization for
 281 *		both "special" and "regular" inodes.  special inodes
 282 *		are those required early in the mount process and
 283 *		require special handling since much of the file system
 284 *		is not yet initialized.  these "special" inodes are
 285 *		identified by a NULL inode map inode pointer and are
 286 *		actually initialized by a call to diReadSpecial().
 287 *
 288 *		for regular inodes, the iag describing the disk inode
 289 *		is read from disk to determine the inode extent address
 290 *		for the disk inode.  with the inode extent address in
 291 *		hand, the page of the extent that contains the disk
 292 *		inode is read and the disk inode is copied to the
 293 *		incore inode.
 294 *
 295 * PARAMETERS:
 296 *	ip	-  pointer to incore inode to be initialized from disk.
 297 *
 298 * RETURN VALUES:
 299 *	0	- success
 300 *	-EIO	- i/o error.
 301 *	-ENOMEM	- insufficient memory
 302 *
 303 */
 304int diRead(struct inode *ip)
 305{
 306	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
 307	int iagno, ino, extno, rc;
 308	struct inode *ipimap;
 309	struct dinode *dp;
 310	struct iag *iagp;
 311	struct metapage *mp;
 312	s64 blkno, agstart;
 313	struct inomap *imap;
 314	int block_offset;
 315	int inodes_left;
 316	unsigned long pageno;
 317	int rel_inode;
 318
 319	jfs_info("diRead: ino = %ld", ip->i_ino);
 320
 321	ipimap = sbi->ipimap;
 322	JFS_IP(ip)->ipimap = ipimap;
 323
 324	/* determine the iag number for this inode (number) */
 325	iagno = INOTOIAG(ip->i_ino);
 326
 327	/* read the iag */
 328	imap = JFS_IP(ipimap)->i_imap;
 329	IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
 330	rc = diIAGRead(imap, iagno, &mp);
 331	IREAD_UNLOCK(ipimap);
 332	if (rc) {
 333		jfs_err("diRead: diIAGRead returned %d", rc);
 334		return (rc);
 335	}
 336
 337	iagp = (struct iag *) mp->data;
 338
 339	/* determine inode extent that holds the disk inode */
 340	ino = ip->i_ino & (INOSPERIAG - 1);
 341	extno = ino >> L2INOSPEREXT;
 342
 343	if ((lengthPXD(&iagp->inoext[extno]) != imap->im_nbperiext) ||
 344	    (addressPXD(&iagp->inoext[extno]) == 0)) {
 345		release_metapage(mp);
 346		return -ESTALE;
 347	}
 348
 349	/* get disk block number of the page within the inode extent
 350	 * that holds the disk inode.
 351	 */
 352	blkno = INOPBLK(&iagp->inoext[extno], ino, sbi->l2nbperpage);
 353
 354	/* get the ag for the iag */
 355	agstart = le64_to_cpu(iagp->agstart);
 356
 357	release_metapage(mp);
 358
 359	rel_inode = (ino & (INOSPERPAGE - 1));
 360	pageno = blkno >> sbi->l2nbperpage;
 361
 362	if ((block_offset = ((u32) blkno & (sbi->nbperpage - 1)))) {
 363		/*
 364		 * OS/2 didn't always align inode extents on page boundaries
 365		 */
 366		inodes_left =
 367		     (sbi->nbperpage - block_offset) << sbi->l2niperblk;
 368
 369		if (rel_inode < inodes_left)
 370			rel_inode += block_offset << sbi->l2niperblk;
 371		else {
 372			pageno += 1;
 373			rel_inode -= inodes_left;
 374		}
 375	}
 376
 377	/* read the page of disk inode */
 378	mp = read_metapage(ipimap, pageno << sbi->l2nbperpage, PSIZE, 1);
 379	if (!mp) {
 380		jfs_err("diRead: read_metapage failed");
 381		return -EIO;
 382	}
 383
 384	/* locate the disk inode requested */
 385	dp = (struct dinode *) mp->data;
 386	dp += rel_inode;
 387
 388	if (ip->i_ino != le32_to_cpu(dp->di_number)) {
 389		jfs_error(ip->i_sb, "diRead: i_ino != di_number");
 390		rc = -EIO;
 391	} else if (le32_to_cpu(dp->di_nlink) == 0)
 392		rc = -ESTALE;
 393	else
 394		/* copy the disk inode to the in-memory inode */
 395		rc = copy_from_dinode(dp, ip);
 396
 397	release_metapage(mp);
 398
 399	/* set the ag for the inode */
 400	JFS_IP(ip)->agstart = agstart;
 401	JFS_IP(ip)->active_ag = -1;
 402
 403	return (rc);
 404}
 405
 406
 407/*
 408 * NAME:	diReadSpecial()
 409 *
 410 * FUNCTION:	initialize a 'special' inode from disk.
 411 *
 412 *		this routines handles aggregate level inodes.  The
 413 *		inode cache cannot differentiate between the
 414 *		aggregate inodes and the filesystem inodes, so we
 415 *		handle these here.  We don't actually use the aggregate
 416 *		inode map, since these inodes are at a fixed location
 417 *		and in some cases the aggregate inode map isn't initialized
 418 *		yet.
 419 *
 420 * PARAMETERS:
 421 *	sb - filesystem superblock
 422 *	inum - aggregate inode number
 423 *	secondary - 1 if secondary aggregate inode table
 424 *
 425 * RETURN VALUES:
 426 *	new inode	- success
 427 *	NULL		- i/o error.
 428 */
 429struct inode *diReadSpecial(struct super_block *sb, ino_t inum, int secondary)
 430{
 431	struct jfs_sb_info *sbi = JFS_SBI(sb);
 432	uint address;
 433	struct dinode *dp;
 434	struct inode *ip;
 435	struct metapage *mp;
 436
 437	ip = new_inode(sb);
 438	if (ip == NULL) {
 439		jfs_err("diReadSpecial: new_inode returned NULL!");
 440		return ip;
 441	}
 442
 443	if (secondary) {
 444		address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage;
 445		JFS_IP(ip)->ipimap = sbi->ipaimap2;
 446	} else {
 447		address = AITBL_OFF >> L2PSIZE;
 448		JFS_IP(ip)->ipimap = sbi->ipaimap;
 449	}
 450
 451	ASSERT(inum < INOSPEREXT);
 452
 453	ip->i_ino = inum;
 454
 455	address += inum >> 3;	/* 8 inodes per 4K page */
 456
 457	/* read the page of fixed disk inode (AIT) in raw mode */
 458	mp = read_metapage(ip, address << sbi->l2nbperpage, PSIZE, 1);
 459	if (mp == NULL) {
 460		ip->i_nlink = 1;	/* Don't want iput() deleting it */
 461		iput(ip);
 462		return (NULL);
 463	}
 464
 465	/* get the pointer to the disk inode of interest */
 466	dp = (struct dinode *) (mp->data);
 467	dp += inum % 8;		/* 8 inodes per 4K page */
 468
 469	/* copy on-disk inode to in-memory inode */
 470	if ((copy_from_dinode(dp, ip)) != 0) {
 471		/* handle bad return by returning NULL for ip */
 472		ip->i_nlink = 1;	/* Don't want iput() deleting it */
 473		iput(ip);
 474		/* release the page */
 475		release_metapage(mp);
 476		return (NULL);
 477
 478	}
 479
 480	ip->i_mapping->a_ops = &jfs_metapage_aops;
 481	mapping_set_gfp_mask(ip->i_mapping, GFP_NOFS);
 482
 483	/* Allocations to metadata inodes should not affect quotas */
 484	ip->i_flags |= S_NOQUOTA;
 485
 486	if ((inum == FILESYSTEM_I) && (JFS_IP(ip)->ipimap == sbi->ipaimap)) {
 487		sbi->gengen = le32_to_cpu(dp->di_gengen);
 488		sbi->inostamp = le32_to_cpu(dp->di_inostamp);
 489	}
 490
 491	/* release the page */
 492	release_metapage(mp);
 493
 494	/*
 495	 * __mark_inode_dirty expects inodes to be hashed.  Since we don't
 496	 * want special inodes in the fileset inode space, we make them
 497	 * appear hashed, but do not put on any lists.  hlist_del()
 498	 * will work fine and require no locking.
 499	 */
 500	hlist_add_fake(&ip->i_hash);
 501
 502	return (ip);
 503}
 504
 505/*
 506 * NAME:	diWriteSpecial()
 507 *
 508 * FUNCTION:	Write the special inode to disk
 509 *
 510 * PARAMETERS:
 511 *	ip - special inode
 512 *	secondary - 1 if secondary aggregate inode table
 513 *
 514 * RETURN VALUES: none
 515 */
 516
 517void diWriteSpecial(struct inode *ip, int secondary)
 518{
 519	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
 520	uint address;
 521	struct dinode *dp;
 522	ino_t inum = ip->i_ino;
 523	struct metapage *mp;
 524
 525	if (secondary)
 526		address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage;
 527	else
 528		address = AITBL_OFF >> L2PSIZE;
 529
 530	ASSERT(inum < INOSPEREXT);
 531
 532	address += inum >> 3;	/* 8 inodes per 4K page */
 533
 534	/* read the page of fixed disk inode (AIT) in raw mode */
 535	mp = read_metapage(ip, address << sbi->l2nbperpage, PSIZE, 1);
 536	if (mp == NULL) {
 537		jfs_err("diWriteSpecial: failed to read aggregate inode "
 538			"extent!");
 539		return;
 540	}
 541
 542	/* get the pointer to the disk inode of interest */
 543	dp = (struct dinode *) (mp->data);
 544	dp += inum % 8;		/* 8 inodes per 4K page */
 545
 546	/* copy on-disk inode to in-memory inode */
 547	copy_to_dinode(dp, ip);
 548	memcpy(&dp->di_xtroot, &JFS_IP(ip)->i_xtroot, 288);
 549
 550	if (inum == FILESYSTEM_I)
 551		dp->di_gengen = cpu_to_le32(sbi->gengen);
 552
 553	/* write the page */
 554	write_metapage(mp);
 555}
 556
 557/*
 558 * NAME:	diFreeSpecial()
 559 *
 560 * FUNCTION:	Free allocated space for special inode
 561 */
 562void diFreeSpecial(struct inode *ip)
 563{
 564	if (ip == NULL) {
 565		jfs_err("diFreeSpecial called with NULL ip!");
 566		return;
 567	}
 568	filemap_write_and_wait(ip->i_mapping);
 569	truncate_inode_pages(ip->i_mapping, 0);
 570	iput(ip);
 571}
 572
 573
 574
 575/*
 576 * NAME:	diWrite()
 577 *
 578 * FUNCTION:	write the on-disk inode portion of the in-memory inode
 579 *		to its corresponding on-disk inode.
 580 *
 581 *		on entry, the specifed incore inode should itself
 582 *		specify the disk inode number corresponding to the
 583 *		incore inode (i.e. i_number should be initialized).
 584 *
 585 *		the inode contains the inode extent address for the disk
 586 *		inode.  with the inode extent address in hand, the
 587 *		page of the extent that contains the disk inode is
 588 *		read and the disk inode portion of the incore inode
 589 *		is copied to the disk inode.
 590 *
 591 * PARAMETERS:
 592 *	tid -  transacation id
 593 *	ip  -  pointer to incore inode to be written to the inode extent.
 594 *
 595 * RETURN VALUES:
 596 *	0	- success
 597 *	-EIO	- i/o error.
 598 */
 599int diWrite(tid_t tid, struct inode *ip)
 600{
 601	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
 602	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
 603	int rc = 0;
 604	s32 ino;
 605	struct dinode *dp;
 606	s64 blkno;
 607	int block_offset;
 608	int inodes_left;
 609	struct metapage *mp;
 610	unsigned long pageno;
 611	int rel_inode;
 612	int dioffset;
 613	struct inode *ipimap;
 614	uint type;
 615	lid_t lid;
 616	struct tlock *ditlck, *tlck;
 617	struct linelock *dilinelock, *ilinelock;
 618	struct lv *lv;
 619	int n;
 620
 621	ipimap = jfs_ip->ipimap;
 622
 623	ino = ip->i_ino & (INOSPERIAG - 1);
 624
 625	if (!addressPXD(&(jfs_ip->ixpxd)) ||
 626	    (lengthPXD(&(jfs_ip->ixpxd)) !=
 627	     JFS_IP(ipimap)->i_imap->im_nbperiext)) {
 628		jfs_error(ip->i_sb, "diWrite: ixpxd invalid");
 629		return -EIO;
 630	}
 631
 632	/*
 633	 * read the page of disk inode containing the specified inode:
 634	 */
 635	/* compute the block address of the page */
 636	blkno = INOPBLK(&(jfs_ip->ixpxd), ino, sbi->l2nbperpage);
 637
 638	rel_inode = (ino & (INOSPERPAGE - 1));
 639	pageno = blkno >> sbi->l2nbperpage;
 640
 641	if ((block_offset = ((u32) blkno & (sbi->nbperpage - 1)))) {
 642		/*
 643		 * OS/2 didn't always align inode extents on page boundaries
 644		 */
 645		inodes_left =
 646		    (sbi->nbperpage - block_offset) << sbi->l2niperblk;
 647
 648		if (rel_inode < inodes_left)
 649			rel_inode += block_offset << sbi->l2niperblk;
 650		else {
 651			pageno += 1;
 652			rel_inode -= inodes_left;
 653		}
 654	}
 655	/* read the page of disk inode */
 656      retry:
 657	mp = read_metapage(ipimap, pageno << sbi->l2nbperpage, PSIZE, 1);
 658	if (!mp)
 659		return -EIO;
 660
 661	/* get the pointer to the disk inode */
 662	dp = (struct dinode *) mp->data;
 663	dp += rel_inode;
 664
 665	dioffset = (ino & (INOSPERPAGE - 1)) << L2DISIZE;
 666
 667	/*
 668	 * acquire transaction lock on the on-disk inode;
 669	 * N.B. tlock is acquired on ipimap not ip;
 670	 */
 671	if ((ditlck =
 672	     txLock(tid, ipimap, mp, tlckINODE | tlckENTRY)) == NULL)
 673		goto retry;
 674	dilinelock = (struct linelock *) & ditlck->lock;
 675
 676	/*
 677	 * copy btree root from in-memory inode to on-disk inode
 678	 *
 679	 * (tlock is taken from inline B+-tree root in in-memory
 680	 * inode when the B+-tree root is updated, which is pointed
 681	 * by jfs_ip->blid as well as being on tx tlock list)
 682	 *
 683	 * further processing of btree root is based on the copy
 684	 * in in-memory inode, where txLog() will log from, and,
 685	 * for xtree root, txUpdateMap() will update map and reset
 686	 * XAD_NEW bit;
 687	 */
 688
 689	if (S_ISDIR(ip->i_mode) && (lid = jfs_ip->xtlid)) {
 690		/*
 691		 * This is the special xtree inside the directory for storing
 692		 * the directory table
 693		 */
 694		xtpage_t *p, *xp;
 695		xad_t *xad;
 696
 697		jfs_ip->xtlid = 0;
 698		tlck = lid_to_tlock(lid);
 699		assert(tlck->type & tlckXTREE);
 700		tlck->type |= tlckBTROOT;
 701		tlck->mp = mp;
 702		ilinelock = (struct linelock *) & tlck->lock;
 703
 704		/*
 705		 * copy xtree root from inode to dinode:
 706		 */
 707		p = &jfs_ip->i_xtroot;
 708		xp = (xtpage_t *) &dp->di_dirtable;
 709		lv = ilinelock->lv;
 710		for (n = 0; n < ilinelock->index; n++, lv++) {
 711			memcpy(&xp->xad[lv->offset], &p->xad[lv->offset],
 712			       lv->length << L2XTSLOTSIZE);
 713		}
 714
 715		/* reset on-disk (metadata page) xtree XAD_NEW bit */
 716		xad = &xp->xad[XTENTRYSTART];
 717		for (n = XTENTRYSTART;
 718		     n < le16_to_cpu(xp->header.nextindex); n++, xad++)
 719			if (xad->flag & (XAD_NEW | XAD_EXTENDED))
 720				xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
 721	}
 722
 723	if ((lid = jfs_ip->blid) == 0)
 724		goto inlineData;
 725	jfs_ip->blid = 0;
 726
 727	tlck = lid_to_tlock(lid);
 728	type = tlck->type;
 729	tlck->type |= tlckBTROOT;
 730	tlck->mp = mp;
 731	ilinelock = (struct linelock *) & tlck->lock;
 732
 733	/*
 734	 *	regular file: 16 byte (XAD slot) granularity
 735	 */
 736	if (type & tlckXTREE) {
 737		xtpage_t *p, *xp;
 738		xad_t *xad;
 739
 740		/*
 741		 * copy xtree root from inode to dinode:
 742		 */
 743		p = &jfs_ip->i_xtroot;
 744		xp = &dp->di_xtroot;
 745		lv = ilinelock->lv;
 746		for (n = 0; n < ilinelock->index; n++, lv++) {
 747			memcpy(&xp->xad[lv->offset], &p->xad[lv->offset],
 748			       lv->length << L2XTSLOTSIZE);
 749		}
 750
 751		/* reset on-disk (metadata page) xtree XAD_NEW bit */
 752		xad = &xp->xad[XTENTRYSTART];
 753		for (n = XTENTRYSTART;
 754		     n < le16_to_cpu(xp->header.nextindex); n++, xad++)
 755			if (xad->flag & (XAD_NEW | XAD_EXTENDED))
 756				xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
 757	}
 758	/*
 759	 *	directory: 32 byte (directory entry slot) granularity
 760	 */
 761	else if (type & tlckDTREE) {
 762		dtpage_t *p, *xp;
 763
 764		/*
 765		 * copy dtree root from inode to dinode:
 766		 */
 767		p = (dtpage_t *) &jfs_ip->i_dtroot;
 768		xp = (dtpage_t *) & dp->di_dtroot;
 769		lv = ilinelock->lv;
 770		for (n = 0; n < ilinelock->index; n++, lv++) {
 771			memcpy(&xp->slot[lv->offset], &p->slot[lv->offset],
 772			       lv->length << L2DTSLOTSIZE);
 773		}
 774	} else {
 775		jfs_err("diWrite: UFO tlock");
 776	}
 777
 778      inlineData:
 779	/*
 780	 * copy inline symlink from in-memory inode to on-disk inode
 781	 */
 782	if (S_ISLNK(ip->i_mode) && ip->i_size < IDATASIZE) {
 783		lv = & dilinelock->lv[dilinelock->index];
 784		lv->offset = (dioffset + 2 * 128) >> L2INODESLOTSIZE;
 785		lv->length = 2;
 786		memcpy(&dp->di_fastsymlink, jfs_ip->i_inline, IDATASIZE);
 787		dilinelock->index++;
 788	}
 789	/*
 790	 * copy inline data from in-memory inode to on-disk inode:
 791	 * 128 byte slot granularity
 792	 */
 793	if (test_cflag(COMMIT_Inlineea, ip)) {
 794		lv = & dilinelock->lv[dilinelock->index];
 795		lv->offset = (dioffset + 3 * 128) >> L2INODESLOTSIZE;
 796		lv->length = 1;
 797		memcpy(&dp->di_inlineea, jfs_ip->i_inline_ea, INODESLOTSIZE);
 798		dilinelock->index++;
 799
 800		clear_cflag(COMMIT_Inlineea, ip);
 801	}
 802
 803	/*
 804	 *	lock/copy inode base: 128 byte slot granularity
 805	 */
 806	lv = & dilinelock->lv[dilinelock->index];
 807	lv->offset = dioffset >> L2INODESLOTSIZE;
 808	copy_to_dinode(dp, ip);
 809	if (test_and_clear_cflag(COMMIT_Dirtable, ip)) {
 810		lv->length = 2;
 811		memcpy(&dp->di_dirtable, &jfs_ip->i_dirtable, 96);
 812	} else
 813		lv->length = 1;
 814	dilinelock->index++;
 815
 816	/* release the buffer holding the updated on-disk inode.
 817	 * the buffer will be later written by commit processing.
 818	 */
 819	write_metapage(mp);
 820
 821	return (rc);
 822}
 823
 824
 825/*
 826 * NAME:	diFree(ip)
 827 *
 828 * FUNCTION:	free a specified inode from the inode working map
 829 *		for a fileset or aggregate.
 830 *
 831 *		if the inode to be freed represents the first (only)
 832 *		free inode within the iag, the iag will be placed on
 833 *		the ag free inode list.
 834 *
 835 *		freeing the inode will cause the inode extent to be
 836 *		freed if the inode is the only allocated inode within
 837 *		the extent.  in this case all the disk resource backing
 838 *		up the inode extent will be freed. in addition, the iag
 839 *		will be placed on the ag extent free list if the extent
 840 *		is the first free extent in the iag.  if freeing the
 841 *		extent also means that no free inodes will exist for
 842 *		the iag, the iag will also be removed from the ag free
 843 *		inode list.
 844 *
 845 *		the iag describing the inode will be freed if the extent
 846 *		is to be freed and it is the only backed extent within
 847 *		the iag.  in this case, the iag will be removed from the
 848 *		ag free extent list and ag free inode list and placed on
 849 *		the inode map's free iag list.
 850 *
 851 *		a careful update approach is used to provide consistency
 852 *		in the face of updates to multiple buffers.  under this
 853 *		approach, all required buffers are obtained before making
 854 *		any updates and are held until all updates are complete.
 855 *
 856 * PARAMETERS:
 857 *	ip	- inode to be freed.
 858 *
 859 * RETURN VALUES:
 860 *	0	- success
 861 *	-EIO	- i/o error.
 862 */
 863int diFree(struct inode *ip)
 864{
 865	int rc;
 866	ino_t inum = ip->i_ino;
 867	struct iag *iagp, *aiagp, *biagp, *ciagp, *diagp;
 868	struct metapage *mp, *amp, *bmp, *cmp, *dmp;
 869	int iagno, ino, extno, bitno, sword, agno;
 870	int back, fwd;
 871	u32 bitmap, mask;
 872	struct inode *ipimap = JFS_SBI(ip->i_sb)->ipimap;
 873	struct inomap *imap = JFS_IP(ipimap)->i_imap;
 874	pxd_t freepxd;
 875	tid_t tid;
 876	struct inode *iplist[3];
 877	struct tlock *tlck;
 878	struct pxd_lock *pxdlock;
 879
 880	/*
 881	 * This is just to suppress compiler warnings.  The same logic that
 882	 * references these variables is used to initialize them.
 883	 */
 884	aiagp = biagp = ciagp = diagp = NULL;
 885
 886	/* get the iag number containing the inode.
 887	 */
 888	iagno = INOTOIAG(inum);
 889
 890	/* make sure that the iag is contained within
 891	 * the map.
 892	 */
 893	if (iagno >= imap->im_nextiag) {
 894		print_hex_dump(KERN_ERR, "imap: ", DUMP_PREFIX_ADDRESS, 16, 4,
 895			       imap, 32, 0);
 896		jfs_error(ip->i_sb,
 897			  "diFree: inum = %d, iagno = %d, nextiag = %d",
 898			  (uint) inum, iagno, imap->im_nextiag);
 899		return -EIO;
 900	}
 901
 902	/* get the allocation group for this ino.
 903	 */
 904	agno = BLKTOAG(JFS_IP(ip)->agstart, JFS_SBI(ip->i_sb));
 905
 906	/* Lock the AG specific inode map information
 907	 */
 908	AG_LOCK(imap, agno);
 909
 910	/* Obtain read lock in imap inode.  Don't release it until we have
 911	 * read all of the IAG's that we are going to.
 912	 */
 913	IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
 914
 915	/* read the iag.
 916	 */
 917	if ((rc = diIAGRead(imap, iagno, &mp))) {
 918		IREAD_UNLOCK(ipimap);
 919		AG_UNLOCK(imap, agno);
 920		return (rc);
 921	}
 922	iagp = (struct iag *) mp->data;
 923
 924	/* get the inode number and extent number of the inode within
 925	 * the iag and the inode number within the extent.
 926	 */
 927	ino = inum & (INOSPERIAG - 1);
 928	extno = ino >> L2INOSPEREXT;
 929	bitno = ino & (INOSPEREXT - 1);
 930	mask = HIGHORDER >> bitno;
 931
 932	if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
 933		jfs_error(ip->i_sb,
 934			  "diFree: wmap shows inode already free");
 935	}
 936
 937	if (!addressPXD(&iagp->inoext[extno])) {
 938		release_metapage(mp);
 939		IREAD_UNLOCK(ipimap);
 940		AG_UNLOCK(imap, agno);
 941		jfs_error(ip->i_sb, "diFree: invalid inoext");
 942		return -EIO;
 943	}
 944
 945	/* compute the bitmap for the extent reflecting the freed inode.
 946	 */
 947	bitmap = le32_to_cpu(iagp->wmap[extno]) & ~mask;
 948
 949	if (imap->im_agctl[agno].numfree > imap->im_agctl[agno].numinos) {
 950		release_metapage(mp);
 951		IREAD_UNLOCK(ipimap);
 952		AG_UNLOCK(imap, agno);
 953		jfs_error(ip->i_sb, "diFree: numfree > numinos");
 954		return -EIO;
 955	}
 956	/*
 957	 *	inode extent still has some inodes or below low water mark:
 958	 *	keep the inode extent;
 959	 */
 960	if (bitmap ||
 961	    imap->im_agctl[agno].numfree < 96 ||
 962	    (imap->im_agctl[agno].numfree < 288 &&
 963	     (((imap->im_agctl[agno].numfree * 100) /
 964	       imap->im_agctl[agno].numinos) <= 25))) {
 965		/* if the iag currently has no free inodes (i.e.,
 966		 * the inode being freed is the first free inode of iag),
 967		 * insert the iag at head of the inode free list for the ag.
 968		 */
 969		if (iagp->nfreeinos == 0) {
 970			/* check if there are any iags on the ag inode
 971			 * free list.  if so, read the first one so that
 972			 * we can link the current iag onto the list at
 973			 * the head.
 974			 */
 975			if ((fwd = imap->im_agctl[agno].inofree) >= 0) {
 976				/* read the iag that currently is the head
 977				 * of the list.
 978				 */
 979				if ((rc = diIAGRead(imap, fwd, &amp))) {
 980					IREAD_UNLOCK(ipimap);
 981					AG_UNLOCK(imap, agno);
 982					release_metapage(mp);
 983					return (rc);
 984				}
 985				aiagp = (struct iag *) amp->data;
 986
 987				/* make current head point back to the iag.
 988				 */
 989				aiagp->inofreeback = cpu_to_le32(iagno);
 990
 991				write_metapage(amp);
 992			}
 993
 994			/* iag points forward to current head and iag
 995			 * becomes the new head of the list.
 996			 */
 997			iagp->inofreefwd =
 998			    cpu_to_le32(imap->im_agctl[agno].inofree);
 999			iagp->inofreeback = cpu_to_le32(-1);
1000			imap->im_agctl[agno].inofree = iagno;
1001		}
1002		IREAD_UNLOCK(ipimap);
1003
1004		/* update the free inode summary map for the extent if
1005		 * freeing the inode means the extent will now have free
1006		 * inodes (i.e., the inode being freed is the first free
1007		 * inode of extent),
1008		 */
1009		if (iagp->wmap[extno] == cpu_to_le32(ONES)) {
1010			sword = extno >> L2EXTSPERSUM;
1011			bitno = extno & (EXTSPERSUM - 1);
1012			iagp->inosmap[sword] &=
1013			    cpu_to_le32(~(HIGHORDER >> bitno));
1014		}
1015
1016		/* update the bitmap.
1017		 */
1018		iagp->wmap[extno] = cpu_to_le32(bitmap);
1019
1020		/* update the free inode counts at the iag, ag and
1021		 * map level.
1022		 */
1023		le32_add_cpu(&iagp->nfreeinos, 1);
1024		imap->im_agctl[agno].numfree += 1;
1025		atomic_inc(&imap->im_numfree);
1026
1027		/* release the AG inode map lock
1028		 */
1029		AG_UNLOCK(imap, agno);
1030
1031		/* write the iag */
1032		write_metapage(mp);
1033
1034		return (0);
1035	}
1036
1037
1038	/*
1039	 *	inode extent has become free and above low water mark:
1040	 *	free the inode extent;
1041	 */
1042
1043	/*
1044	 *	prepare to update iag list(s) (careful update step 1)
1045	 */
1046	amp = bmp = cmp = dmp = NULL;
1047	fwd = back = -1;
1048
1049	/* check if the iag currently has no free extents.  if so,
1050	 * it will be placed on the head of the ag extent free list.
1051	 */
1052	if (iagp->nfreeexts == 0) {
1053		/* check if the ag extent free list has any iags.
1054		 * if so, read the iag at the head of the list now.
1055		 * this (head) iag will be updated later to reflect
1056		 * the addition of the current iag at the head of
1057		 * the list.
1058		 */
1059		if ((fwd = imap->im_agctl[agno].extfree) >= 0) {
1060			if ((rc = diIAGRead(imap, fwd, &amp)))
1061				goto error_out;
1062			aiagp = (struct iag *) amp->data;
1063		}
1064	} else {
1065		/* iag has free extents. check if the addition of a free
1066		 * extent will cause all extents to be free within this
1067		 * iag.  if so, the iag will be removed from the ag extent
1068		 * free list and placed on the inode map's free iag list.
1069		 */
1070		if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) {
1071			/* in preparation for removing the iag from the
1072			 * ag extent free list, read the iags preceding
1073			 * and following the iag on the ag extent free
1074			 * list.
1075			 */
1076			if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) {
1077				if ((rc = diIAGRead(imap, fwd, &amp)))
1078					goto error_out;
1079				aiagp = (struct iag *) amp->data;
1080			}
1081
1082			if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) {
1083				if ((rc = diIAGRead(imap, back, &bmp)))
1084					goto error_out;
1085				biagp = (struct iag *) bmp->data;
1086			}
1087		}
1088	}
1089
1090	/* remove the iag from the ag inode free list if freeing
1091	 * this extent cause the iag to have no free inodes.
1092	 */
1093	if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) {
1094		int inofreeback = le32_to_cpu(iagp->inofreeback);
1095		int inofreefwd = le32_to_cpu(iagp->inofreefwd);
1096
1097		/* in preparation for removing the iag from the
1098		 * ag inode free list, read the iags preceding
1099		 * and following the iag on the ag inode free
1100		 * list.  before reading these iags, we must make
1101		 * sure that we already don't have them in hand
1102		 * from up above, since re-reading an iag (buffer)
1103		 * we are currently holding would cause a deadlock.
1104		 */
1105		if (inofreefwd >= 0) {
1106
1107			if (inofreefwd == fwd)
1108				ciagp = (struct iag *) amp->data;
1109			else if (inofreefwd == back)
1110				ciagp = (struct iag *) bmp->data;
1111			else {
1112				if ((rc =
1113				     diIAGRead(imap, inofreefwd, &cmp)))
1114					goto error_out;
1115				ciagp = (struct iag *) cmp->data;
1116			}
1117			assert(ciagp != NULL);
1118		}
1119
1120		if (inofreeback >= 0) {
1121			if (inofreeback == fwd)
1122				diagp = (struct iag *) amp->data;
1123			else if (inofreeback == back)
1124				diagp = (struct iag *) bmp->data;
1125			else {
1126				if ((rc =
1127				     diIAGRead(imap, inofreeback, &dmp)))
1128					goto error_out;
1129				diagp = (struct iag *) dmp->data;
1130			}
1131			assert(diagp != NULL);
1132		}
1133	}
1134
1135	IREAD_UNLOCK(ipimap);
1136
1137	/*
1138	 * invalidate any page of the inode extent freed from buffer cache;
1139	 */
1140	freepxd = iagp->inoext[extno];
1141	invalidate_pxd_metapages(ip, freepxd);
1142
1143	/*
1144	 *	update iag list(s) (careful update step 2)
1145	 */
1146	/* add the iag to the ag extent free list if this is the
1147	 * first free extent for the iag.
1148	 */
1149	if (iagp->nfreeexts == 0) {
1150		if (fwd >= 0)
1151			aiagp->extfreeback = cpu_to_le32(iagno);
1152
1153		iagp->extfreefwd =
1154		    cpu_to_le32(imap->im_agctl[agno].extfree);
1155		iagp->extfreeback = cpu_to_le32(-1);
1156		imap->im_agctl[agno].extfree = iagno;
1157	} else {
1158		/* remove the iag from the ag extent list if all extents
1159		 * are now free and place it on the inode map iag free list.
1160		 */
1161		if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) {
1162			if (fwd >= 0)
1163				aiagp->extfreeback = iagp->extfreeback;
1164
1165			if (back >= 0)
1166				biagp->extfreefwd = iagp->extfreefwd;
1167			else
1168				imap->im_agctl[agno].extfree =
1169				    le32_to_cpu(iagp->extfreefwd);
1170
1171			iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1);
1172
1173			IAGFREE_LOCK(imap);
1174			iagp->iagfree = cpu_to_le32(imap->im_freeiag);
1175			imap->im_freeiag = iagno;
1176			IAGFREE_UNLOCK(imap);
1177		}
1178	}
1179
1180	/* remove the iag from the ag inode free list if freeing
1181	 * this extent causes the iag to have no free inodes.
1182	 */
1183	if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) {
1184		if ((int) le32_to_cpu(iagp->inofreefwd) >= 0)
1185			ciagp->inofreeback = iagp->inofreeback;
1186
1187		if ((int) le32_to_cpu(iagp->inofreeback) >= 0)
1188			diagp->inofreefwd = iagp->inofreefwd;
1189		else
1190			imap->im_agctl[agno].inofree =
1191			    le32_to_cpu(iagp->inofreefwd);
1192
1193		iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1);
1194	}
1195
1196	/* update the inode extent address and working map
1197	 * to reflect the free extent.
1198	 * the permanent map should have been updated already
1199	 * for the inode being freed.
1200	 */
1201	if (iagp->pmap[extno] != 0) {
1202		jfs_error(ip->i_sb, "diFree: the pmap does not show inode free");
1203	}
1204	iagp->wmap[extno] = 0;
1205	PXDlength(&iagp->inoext[extno], 0);
1206	PXDaddress(&iagp->inoext[extno], 0);
1207
1208	/* update the free extent and free inode summary maps
1209	 * to reflect the freed extent.
1210	 * the inode summary map is marked to indicate no inodes
1211	 * available for the freed extent.
1212	 */
1213	sword = extno >> L2EXTSPERSUM;
1214	bitno = extno & (EXTSPERSUM - 1);
1215	mask = HIGHORDER >> bitno;
1216	iagp->inosmap[sword] |= cpu_to_le32(mask);
1217	iagp->extsmap[sword] &= cpu_to_le32(~mask);
1218
1219	/* update the number of free inodes and number of free extents
1220	 * for the iag.
1221	 */
1222	le32_add_cpu(&iagp->nfreeinos, -(INOSPEREXT - 1));
1223	le32_add_cpu(&iagp->nfreeexts, 1);
1224
1225	/* update the number of free inodes and backed inodes
1226	 * at the ag and inode map level.
1227	 */
1228	imap->im_agctl[agno].numfree -= (INOSPEREXT - 1);
1229	imap->im_agctl[agno].numinos -= INOSPEREXT;
1230	atomic_sub(INOSPEREXT - 1, &imap->im_numfree);
1231	atomic_sub(INOSPEREXT, &imap->im_numinos);
1232
1233	if (amp)
1234		write_metapage(amp);
1235	if (bmp)
1236		write_metapage(bmp);
1237	if (cmp)
1238		write_metapage(cmp);
1239	if (dmp)
1240		write_metapage(dmp);
1241
1242	/*
1243	 * start transaction to update block allocation map
1244	 * for the inode extent freed;
1245	 *
1246	 * N.B. AG_LOCK is released and iag will be released below, and
1247	 * other thread may allocate inode from/reusing the ixad freed
1248	 * BUT with new/different backing inode extent from the extent
1249	 * to be freed by the transaction;
1250	 */
1251	tid = txBegin(ipimap->i_sb, COMMIT_FORCE);
1252	mutex_lock(&JFS_IP(ipimap)->commit_mutex);
1253
1254	/* acquire tlock of the iag page of the freed ixad
1255	 * to force the page NOHOMEOK (even though no data is
1256	 * logged from the iag page) until NOREDOPAGE|FREEXTENT log
1257	 * for the free of the extent is committed;
1258	 * write FREEXTENT|NOREDOPAGE log record
1259	 * N.B. linelock is overlaid as freed extent descriptor;
1260	 */
1261	tlck = txLock(tid, ipimap, mp, tlckINODE | tlckFREE);
1262	pxdlock = (struct pxd_lock *) & tlck->lock;
1263	pxdlock->flag = mlckFREEPXD;
1264	pxdlock->pxd = freepxd;
1265	pxdlock->index = 1;
1266
1267	write_metapage(mp);
1268
1269	iplist[0] = ipimap;
1270
1271	/*
1272	 * logredo needs the IAG number and IAG extent index in order
1273	 * to ensure that the IMap is consistent.  The least disruptive
1274	 * way to pass these values through  to the transaction manager
1275	 * is in the iplist array.
1276	 *
1277	 * It's not pretty, but it works.
1278	 */
1279	iplist[1] = (struct inode *) (size_t)iagno;
1280	iplist[2] = (struct inode *) (size_t)extno;
1281
1282	rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE);
1283
1284	txEnd(tid);
1285	mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
1286
1287	/* unlock the AG inode map information */
1288	AG_UNLOCK(imap, agno);
1289
1290	return (0);
1291
1292      error_out:
1293	IREAD_UNLOCK(ipimap);
1294
1295	if (amp)
1296		release_metapage(amp);
1297	if (bmp)
1298		release_metapage(bmp);
1299	if (cmp)
1300		release_metapage(cmp);
1301	if (dmp)
1302		release_metapage(dmp);
1303
1304	AG_UNLOCK(imap, agno);
1305
1306	release_metapage(mp);
1307
1308	return (rc);
1309}
1310
1311/*
1312 * There are several places in the diAlloc* routines where we initialize
1313 * the inode.
1314 */
1315static inline void
1316diInitInode(struct inode *ip, int iagno, int ino, int extno, struct iag * iagp)
1317{
1318	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
1319
1320	ip->i_ino = (iagno << L2INOSPERIAG) + ino;
1321	jfs_ip->ixpxd = iagp->inoext[extno];
1322	jfs_ip->agstart = le64_to_cpu(iagp->agstart);
1323	jfs_ip->active_ag = -1;
1324}
1325
1326
1327/*
1328 * NAME:	diAlloc(pip,dir,ip)
1329 *
1330 * FUNCTION:	allocate a disk inode from the inode working map
1331 *		for a fileset or aggregate.
1332 *
1333 * PARAMETERS:
1334 *	pip	- pointer to incore inode for the parent inode.
1335 *	dir	- 'true' if the new disk inode is for a directory.
1336 *	ip	- pointer to a new inode
1337 *
1338 * RETURN VALUES:
1339 *	0	- success.
1340 *	-ENOSPC	- insufficient disk resources.
1341 *	-EIO	- i/o error.
1342 */
1343int diAlloc(struct inode *pip, bool dir, struct inode *ip)
1344{
1345	int rc, ino, iagno, addext, extno, bitno, sword;
1346	int nwords, rem, i, agno;
1347	u32 mask, inosmap, extsmap;
1348	struct inode *ipimap;
1349	struct metapage *mp;
1350	ino_t inum;
1351	struct iag *iagp;
1352	struct inomap *imap;
1353
1354	/* get the pointers to the inode map inode and the
1355	 * corresponding imap control structure.
1356	 */
1357	ipimap = JFS_SBI(pip->i_sb)->ipimap;
1358	imap = JFS_IP(ipimap)->i_imap;
1359	JFS_IP(ip)->ipimap = ipimap;
1360	JFS_IP(ip)->fileset = FILESYSTEM_I;
1361
1362	/* for a directory, the allocation policy is to start
1363	 * at the ag level using the preferred ag.
1364	 */
1365	if (dir) {
1366		agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap);
1367		AG_LOCK(imap, agno);
1368		goto tryag;
1369	}
1370
1371	/* for files, the policy starts off by trying to allocate from
1372	 * the same iag containing the parent disk inode:
1373	 * try to allocate the new disk inode close to the parent disk
1374	 * inode, using parent disk inode number + 1 as the allocation
1375	 * hint.  (we use a left-to-right policy to attempt to avoid
1376	 * moving backward on the disk.)  compute the hint within the
1377	 * file system and the iag.
1378	 */
1379
1380	/* get the ag number of this iag */
1381	agno = BLKTOAG(JFS_IP(pip)->agstart, JFS_SBI(pip->i_sb));
1382
1383	if (atomic_read(&JFS_SBI(pip->i_sb)->bmap->db_active[agno])) {
1384		/*
1385		 * There is an open file actively growing.  We want to
1386		 * allocate new inodes from a different ag to avoid
1387		 * fragmentation problems.
1388		 */
1389		agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap);
1390		AG_LOCK(imap, agno);
1391		goto tryag;
1392	}
1393
1394	inum = pip->i_ino + 1;
1395	ino = inum & (INOSPERIAG - 1);
1396
1397	/* back off the hint if it is outside of the iag */
1398	if (ino == 0)
1399		inum = pip->i_ino;
1400
1401	/* lock the AG inode map information */
1402	AG_LOCK(imap, agno);
1403
1404	/* Get read lock on imap inode */
1405	IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
1406
1407	/* get the iag number and read the iag */
1408	iagno = INOTOIAG(inum);
1409	if ((rc = diIAGRead(imap, iagno, &mp))) {
1410		IREAD_UNLOCK(ipimap);
1411		AG_UNLOCK(imap, agno);
1412		return (rc);
1413	}
1414	iagp = (struct iag *) mp->data;
1415
1416	/* determine if new inode extent is allowed to be added to the iag.
1417	 * new inode extent can be added to the iag if the ag
1418	 * has less than 32 free disk inodes and the iag has free extents.
1419	 */
1420	addext = (imap->im_agctl[agno].numfree < 32 && iagp->nfreeexts);
1421
1422	/*
1423	 *	try to allocate from the IAG
1424	 */
1425	/* check if the inode may be allocated from the iag
1426	 * (i.e. the inode has free inodes or new extent can be added).
1427	 */
1428	if (iagp->nfreeinos || addext) {
1429		/* determine the extent number of the hint.
1430		 */
1431		extno = ino >> L2INOSPEREXT;
1432
1433		/* check if the extent containing the hint has backed
1434		 * inodes.  if so, try to allocate within this extent.
1435		 */
1436		if (addressPXD(&iagp->inoext[extno])) {
1437			bitno = ino & (INOSPEREXT - 1);
1438			if ((bitno =
1439			     diFindFree(le32_to_cpu(iagp->wmap[extno]),
1440					bitno))
1441			    < INOSPEREXT) {
1442				ino = (extno << L2INOSPEREXT) + bitno;
1443
1444				/* a free inode (bit) was found within this
1445				 * extent, so allocate it.
1446				 */
1447				rc = diAllocBit(imap, iagp, ino);
1448				IREAD_UNLOCK(ipimap);
1449				if (rc) {
1450					assert(rc == -EIO);
1451				} else {
1452					/* set the results of the allocation
1453					 * and write the iag.
1454					 */
1455					diInitInode(ip, iagno, ino, extno,
1456						    iagp);
1457					mark_metapage_dirty(mp);
1458				}
1459				release_metapage(mp);
1460
1461				/* free the AG lock and return.
1462				 */
1463				AG_UNLOCK(imap, agno);
1464				return (rc);
1465			}
1466
1467			if (!addext)
1468				extno =
1469				    (extno ==
1470				     EXTSPERIAG - 1) ? 0 : extno + 1;
1471		}
1472
1473		/*
1474		 * no free inodes within the extent containing the hint.
1475		 *
1476		 * try to allocate from the backed extents following
1477		 * hint or, if appropriate (i.e. addext is true), allocate
1478		 * an extent of free inodes at or following the extent
1479		 * containing the hint.
1480		 *
1481		 * the free inode and free extent summary maps are used
1482		 * here, so determine the starting summary map position
1483		 * and the number of words we'll have to examine.  again,
1484		 * the approach is to allocate following the hint, so we
1485		 * might have to initially ignore prior bits of the summary
1486		 * map that represent extents prior to the extent containing
1487		 * the hint and later revisit these bits.
1488		 */
1489		bitno = extno & (EXTSPERSUM - 1);
1490		nwords = (bitno == 0) ? SMAPSZ : SMAPSZ + 1;
1491		sword = extno >> L2EXTSPERSUM;
1492
1493		/* mask any prior bits for the starting words of the
1494		 * summary map.
1495		 */
1496		mask = ONES << (EXTSPERSUM - bitno);
1497		inosmap = le32_to_cpu(iagp->inosmap[sword]) | mask;
1498		extsmap = le32_to_cpu(iagp->extsmap[sword]) | mask;
1499
1500		/* scan the free inode and free extent summary maps for
1501		 * free resources.
1502		 */
1503		for (i = 0; i < nwords; i++) {
1504			/* check if this word of the free inode summary
1505			 * map describes an extent with free inodes.
1506			 */
1507			if (~inosmap) {
1508				/* an extent with free inodes has been
1509				 * found. determine the extent number
1510				 * and the inode number within the extent.
1511				 */
1512				rem = diFindFree(inosmap, 0);
1513				extno = (sword << L2EXTSPERSUM) + rem;
1514				rem = diFindFree(le32_to_cpu(iagp->wmap[extno]),
1515						 0);
1516				if (rem >= INOSPEREXT) {
1517					IREAD_UNLOCK(ipimap);
1518					release_metapage(mp);
1519					AG_UNLOCK(imap, agno);
1520					jfs_error(ip->i_sb,
1521						  "diAlloc: can't find free bit "
1522						  "in wmap");
1523					return -EIO;
1524				}
1525
1526				/* determine the inode number within the
1527				 * iag and allocate the inode from the
1528				 * map.
1529				 */
1530				ino = (extno << L2INOSPEREXT) + rem;
1531				rc = diAllocBit(imap, iagp, ino);
1532				IREAD_UNLOCK(ipimap);
1533				if (rc)
1534					assert(rc == -EIO);
1535				else {
1536					/* set the results of the allocation
1537					 * and write the iag.
1538					 */
1539					diInitInode(ip, iagno, ino, extno,
1540						    iagp);
1541					mark_metapage_dirty(mp);
1542				}
1543				release_metapage(mp);
1544
1545				/* free the AG lock and return.
1546				 */
1547				AG_UNLOCK(imap, agno);
1548				return (rc);
1549
1550			}
1551
1552			/* check if we may allocate an extent of free
1553			 * inodes and whether this word of the free
1554			 * extents summary map describes a free extent.
1555			 */
1556			if (addext && ~extsmap) {
1557				/* a free extent has been found.  determine
1558				 * the extent number.
1559				 */
1560				rem = diFindFree(extsmap, 0);
1561				extno = (sword << L2EXTSPERSUM) + rem;
1562
1563				/* allocate an extent of free inodes.
1564				 */
1565				if ((rc = diNewExt(imap, iagp, extno))) {
1566					/* if there is no disk space for a
1567					 * new extent, try to allocate the
1568					 * disk inode from somewhere else.
1569					 */
1570					if (rc == -ENOSPC)
1571						break;
1572
1573					assert(rc == -EIO);
1574				} else {
1575					/* set the results of the allocation
1576					 * and write the iag.
1577					 */
1578					diInitInode(ip, iagno,
1579						    extno << L2INOSPEREXT,
1580						    extno, iagp);
1581					mark_metapage_dirty(mp);
1582				}
1583				release_metapage(mp);
1584				/* free the imap inode & the AG lock & return.
1585				 */
1586				IREAD_UNLOCK(ipimap);
1587				AG_UNLOCK(imap, agno);
1588				return (rc);
1589			}
1590
1591			/* move on to the next set of summary map words.
1592			 */
1593			sword = (sword == SMAPSZ - 1) ? 0 : sword + 1;
1594			inosmap = le32_to_cpu(iagp->inosmap[sword]);
1595			extsmap = le32_to_cpu(iagp->extsmap[sword]);
1596		}
1597	}
1598	/* unlock imap inode */
1599	IREAD_UNLOCK(ipimap);
1600
1601	/* nothing doing in this iag, so release it. */
1602	release_metapage(mp);
1603
1604      tryag:
1605	/*
1606	 * try to allocate anywhere within the same AG as the parent inode.
1607	 */
1608	rc = diAllocAG(imap, agno, dir, ip);
1609
1610	AG_UNLOCK(imap, agno);
1611
1612	if (rc != -ENOSPC)
1613		return (rc);
1614
1615	/*
1616	 * try to allocate in any AG.
1617	 */
1618	return (diAllocAny(imap, agno, dir, ip));
1619}
1620
1621
1622/*
1623 * NAME:	diAllocAG(imap,agno,dir,ip)
1624 *
1625 * FUNCTION:	allocate a disk inode from the allocation group.
1626 *
1627 *		this routine first determines if a new extent of free
1628 *		inodes should be added for the allocation group, with
1629 *		the current request satisfied from this extent. if this
1630 *		is the case, an attempt will be made to do just that.  if
1631 *		this attempt fails or it has been determined that a new
1632 *		extent should not be added, an attempt is made to satisfy
1633 *		the request by allocating an existing (backed) free inode
1634 *		from the allocation group.
1635 *
1636 * PRE CONDITION: Already have the AG lock for this AG.
1637 *
1638 * PARAMETERS:
1639 *	imap	- pointer to inode map control structure.
1640 *	agno	- allocation group to allocate from.
1641 *	dir	- 'true' if the new disk inode is for a directory.
1642 *	ip	- pointer to the new inode to be filled in on successful return
1643 *		  with the disk inode number allocated, its extent address
1644 *		  and the start of the ag.
1645 *
1646 * RETURN VALUES:
1647 *	0	- success.
1648 *	-ENOSPC	- insufficient disk resources.
1649 *	-EIO	- i/o error.
1650 */
1651static int
1652diAllocAG(struct inomap * imap, int agno, bool dir, struct inode *ip)
1653{
1654	int rc, addext, numfree, numinos;
1655
1656	/* get the number of free and the number of backed disk
1657	 * inodes currently within the ag.
1658	 */
1659	numfree = imap->im_agctl[agno].numfree;
1660	numinos = imap->im_agctl[agno].numinos;
1661
1662	if (numfree > numinos) {
1663		jfs_error(ip->i_sb, "diAllocAG: numfree > numinos");
1664		return -EIO;
1665	}
1666
1667	/* determine if we should allocate a new extent of free inodes
1668	 * within the ag: for directory inodes, add a new extent
1669	 * if there are a small number of free inodes or number of free
1670	 * inodes is a small percentage of the number of backed inodes.
1671	 */
1672	if (dir)
1673		addext = (numfree < 64 ||
1674			  (numfree < 256
1675			   && ((numfree * 100) / numinos) <= 20));
1676	else
1677		addext = (numfree == 0);
1678
1679	/*
1680	 * try to allocate a new extent of free inodes.
1681	 */
1682	if (addext) {
1683		/* if free space is not available for this new extent, try
1684		 * below to allocate a free and existing (already backed)
1685		 * inode from the ag.
1686		 */
1687		if ((rc = diAllocExt(imap, agno, ip)) != -ENOSPC)
1688			return (rc);
1689	}
1690
1691	/*
1692	 * try to allocate an existing free inode from the ag.
1693	 */
1694	return (diAllocIno(imap, agno, ip));
1695}
1696
1697
1698/*
1699 * NAME:	diAllocAny(imap,agno,dir,iap)
1700 *
1701 * FUNCTION:	allocate a disk inode from any other allocation group.
1702 *
1703 *		this routine is called when an allocation attempt within
1704 *		the primary allocation group has failed. if attempts to
1705 *		allocate an inode from any allocation group other than the
1706 *		specified primary group.
1707 *
1708 * PARAMETERS:
1709 *	imap	- pointer to inode map control structure.
1710 *	agno	- primary allocation group (to avoid).
1711 *	dir	- 'true' if the new disk inode is for a directory.
1712 *	ip	- pointer to a new inode to be filled in on successful return
1713 *		  with the disk inode number allocated, its extent address
1714 *		  and the start of the ag.
1715 *
1716 * RETURN VALUES:
1717 *	0	- success.
1718 *	-ENOSPC	- insufficient disk resources.
1719 *	-EIO	- i/o error.
1720 */
1721static int
1722diAllocAny(struct inomap * imap, int agno, bool dir, struct inode *ip)
1723{
1724	int ag, rc;
1725	int maxag = JFS_SBI(imap->im_ipimap->i_sb)->bmap->db_maxag;
1726
1727
1728	/* try to allocate from the ags following agno up to
1729	 * the maximum ag number.
1730	 */
1731	for (ag = agno + 1; ag <= maxag; ag++) {
1732		AG_LOCK(imap, ag);
1733
1734		rc = diAllocAG(imap, ag, dir, ip);
1735
1736		AG_UNLOCK(imap, ag);
1737
1738		if (rc != -ENOSPC)
1739			return (rc);
1740	}
1741
1742	/* try to allocate from the ags in front of agno.
1743	 */
1744	for (ag = 0; ag < agno; ag++) {
1745		AG_LOCK(imap, ag);
1746
1747		rc = diAllocAG(imap, ag, dir, ip);
1748
1749		AG_UNLOCK(imap, ag);
1750
1751		if (rc != -ENOSPC)
1752			return (rc);
1753	}
1754
1755	/* no free disk inodes.
1756	 */
1757	return -ENOSPC;
1758}
1759
1760
1761/*
1762 * NAME:	diAllocIno(imap,agno,ip)
1763 *
1764 * FUNCTION:	allocate a disk inode from the allocation group's free
1765 *		inode list, returning an error if this free list is
1766 *		empty (i.e. no iags on the list).
1767 *
1768 *		allocation occurs from the first iag on the list using
1769 *		the iag's free inode summary map to find the leftmost
1770 *		free inode in the iag.
1771 *
1772 * PRE CONDITION: Already have AG lock for this AG.
1773 *
1774 * PARAMETERS:
1775 *	imap	- pointer to inode map control structure.
1776 *	agno	- allocation group.
1777 *	ip	- pointer to new inode to be filled in on successful return
1778 *		  with the disk inode number allocated, its extent address
1779 *		  and the start of the ag.
1780 *
1781 * RETURN VALUES:
1782 *	0	- success.
1783 *	-ENOSPC	- insufficient disk resources.
1784 *	-EIO	- i/o error.
1785 */
1786static int diAllocIno(struct inomap * imap, int agno, struct inode *ip)
1787{
1788	int iagno, ino, rc, rem, extno, sword;
1789	struct metapage *mp;
1790	struct iag *iagp;
1791
1792	/* check if there are iags on the ag's free inode list.
1793	 */
1794	if ((iagno = imap->im_agctl[agno].inofree) < 0)
1795		return -ENOSPC;
1796
1797	/* obtain read lock on imap inode */
1798	IREAD_LOCK(imap->im_ipimap, RDWRLOCK_IMAP);
1799
1800	/* read the iag at the head of the list.
1801	 */
1802	if ((rc = diIAGRead(imap, iagno, &mp))) {
1803		IREAD_UNLOCK(imap->im_ipimap);
1804		return (rc);
1805	}
1806	iagp = (struct iag *) mp->data;
1807
1808	/* better be free inodes in this iag if it is on the
1809	 * list.
1810	 */
1811	if (!iagp->nfreeinos) {
1812		IREAD_UNLOCK(imap->im_ipimap);
1813		release_metapage(mp);
1814		jfs_error(ip->i_sb,
1815			  "diAllocIno: nfreeinos = 0, but iag on freelist");
1816		return -EIO;
1817	}
1818
1819	/* scan the free inode summary map to find an extent
1820	 * with free inodes.
1821	 */
1822	for (sword = 0;; sword++) {
1823		if (sword >= SMAPSZ) {
1824			IREAD_UNLOCK(imap->im_ipimap);
1825			release_metapage(mp);
1826			jfs_error(ip->i_sb,
1827				  "diAllocIno: free inode not found in summary map");
1828			return -EIO;
1829		}
1830
1831		if (~iagp->inosmap[sword])
1832			break;
1833	}
1834
1835	/* found a extent with free inodes. determine
1836	 * the extent number.
1837	 */
1838	rem = diFindFree(le32_to_cpu(iagp->inosmap[sword]), 0);
1839	if (rem >= EXTSPERSUM) {
1840		IREAD_UNLOCK(imap->im_ipimap);
1841		release_metapage(mp);
1842		jfs_error(ip->i_sb, "diAllocIno: no free extent found");
1843		return -EIO;
1844	}
1845	extno = (sword << L2EXTSPERSUM) + rem;
1846
1847	/* find the first free inode in the extent.
1848	 */
1849	rem = diFindFree(le32_to_cpu(iagp->wmap[extno]), 0);
1850	if (rem >= INOSPEREXT) {
1851		IREAD_UNLOCK(imap->im_ipimap);
1852		release_metapage(mp);
1853		jfs_error(ip->i_sb, "diAllocIno: free inode not found");
1854		return -EIO;
1855	}
1856
1857	/* compute the inode number within the iag.
1858	 */
1859	ino = (extno << L2INOSPEREXT) + rem;
1860
1861	/* allocate the inode.
1862	 */
1863	rc = diAllocBit(imap, iagp, ino);
1864	IREAD_UNLOCK(imap->im_ipimap);
1865	if (rc) {
1866		release_metapage(mp);
1867		return (rc);
1868	}
1869
1870	/* set the results of the allocation and write the iag.
1871	 */
1872	diInitInode(ip, iagno, ino, extno, iagp);
1873	write_metapage(mp);
1874
1875	return (0);
1876}
1877
1878
1879/*
1880 * NAME:	diAllocExt(imap,agno,ip)
1881 *
1882 * FUNCTION:	add a new extent of free inodes to an iag, allocating
1883 *		an inode from this extent to satisfy the current allocation
1884 *		request.
1885 *
1886 *		this routine first tries to find an existing iag with free
1887 *		extents through the ag free extent list.  if list is not
1888 *		empty, the head of the list will be selected as the home
1889 *		of the new extent of free inodes.  otherwise (the list is
1890 *		empty), a new iag will be allocated for the ag to contain
1891 *		the extent.
1892 *
1893 *		once an iag has been selected, the free extent summary map
1894 *		is used to locate a free extent within the iag and diNewExt()
1895 *		is called to initialize the extent, with initialization
1896 *		including the allocation of the first inode of the extent
1897 *		for the purpose of satisfying this request.
1898 *
1899 * PARAMETERS:
1900 *	imap	- pointer to inode map control structure.
1901 *	agno	- allocation group number.
1902 *	ip	- pointer to new inode to be filled in on successful return
1903 *		  with the disk inode number allocated, its extent address
1904 *		  and the start of the ag.
1905 *
1906 * RETURN VALUES:
1907 *	0	- success.
1908 *	-ENOSPC	- insufficient disk resources.
1909 *	-EIO	- i/o error.
1910 */
1911static int diAllocExt(struct inomap * imap, int agno, struct inode *ip)
1912{
1913	int rem, iagno, sword, extno, rc;
1914	struct metapage *mp;
1915	struct iag *iagp;
1916
1917	/* check if the ag has any iags with free extents.  if not,
1918	 * allocate a new iag for the ag.
1919	 */
1920	if ((iagno = imap->im_agctl[agno].extfree) < 0) {
1921		/* If successful, diNewIAG will obtain the read lock on the
1922		 * imap inode.
1923		 */
1924		if ((rc = diNewIAG(imap, &iagno, agno, &mp))) {
1925			return (rc);
1926		}
1927		iagp = (struct iag *) mp->data;
1928
1929		/* set the ag number if this a brand new iag
1930		 */
1931		iagp->agstart =
1932		    cpu_to_le64(AGTOBLK(agno, imap->im_ipimap));
1933	} else {
1934		/* read the iag.
1935		 */
1936		IREAD_LOCK(imap->im_ipimap, RDWRLOCK_IMAP);
1937		if ((rc = diIAGRead(imap, iagno, &mp))) {
1938			IREAD_UNLOCK(imap->im_ipimap);
1939			jfs_error(ip->i_sb, "diAllocExt: error reading iag");
1940			return rc;
1941		}
1942		iagp = (struct iag *) mp->data;
1943	}
1944
1945	/* using the free extent summary map, find a free extent.
1946	 */
1947	for (sword = 0;; sword++) {
1948		if (sword >= SMAPSZ) {
1949			release_metapage(mp);
1950			IREAD_UNLOCK(imap->im_ipimap);
1951			jfs_error(ip->i_sb,
1952				  "diAllocExt: free ext summary map not found");
1953			return -EIO;
1954		}
1955		if (~iagp->extsmap[sword])
1956			break;
1957	}
1958
1959	/* determine the extent number of the free extent.
1960	 */
1961	rem = diFindFree(le32_to_cpu(iagp->extsmap[sword]), 0);
1962	if (rem >= EXTSPERSUM) {
1963		release_metapage(mp);
1964		IREAD_UNLOCK(imap->im_ipimap);
1965		jfs_error(ip->i_sb, "diAllocExt: free extent not found");
1966		return -EIO;
1967	}
1968	extno = (sword << L2EXTSPERSUM) + rem;
1969
1970	/* initialize the new extent.
1971	 */
1972	rc = diNewExt(imap, iagp, extno);
1973	IREAD_UNLOCK(imap->im_ipimap);
1974	if (rc) {
1975		/* something bad happened.  if a new iag was allocated,
1976		 * place it back on the inode map's iag free list, and
1977		 * clear the ag number information.
1978		 */
1979		if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
1980			IAGFREE_LOCK(imap);
1981			iagp->iagfree = cpu_to_le32(imap->im_freeiag);
1982			imap->im_freeiag = iagno;
1983			IAGFREE_UNLOCK(imap);
1984		}
1985		write_metapage(mp);
1986		return (rc);
1987	}
1988
1989	/* set the results of the allocation and write the iag.
1990	 */
1991	diInitInode(ip, iagno, extno << L2INOSPEREXT, extno, iagp);
1992
1993	write_metapage(mp);
1994
1995	return (0);
1996}
1997
1998
1999/*
2000 * NAME:	diAllocBit(imap,iagp,ino)
2001 *
2002 * FUNCTION:	allocate a backed inode from an iag.
2003 *
2004 *		this routine performs the mechanics of allocating a
2005 *		specified inode from a backed extent.
2006 *
2007 *		if the inode to be allocated represents the last free
2008 *		inode within the iag, the iag will be removed from the
2009 *		ag free inode list.
2010 *
2011 *		a careful update approach is used to provide consistency
2012 *		in the face of updates to multiple buffers.  under this
2013 *		approach, all required buffers are obtained before making
2014 *		any updates and are held all are updates are complete.
2015 *
2016 * PRE CONDITION: Already have buffer lock on iagp.  Already have AG lock on
2017 *	this AG.  Must have read lock on imap inode.
2018 *
2019 * PARAMETERS:
2020 *	imap	- pointer to inode map control structure.
2021 *	iagp	- pointer to iag.
2022 *	ino	- inode number to be allocated within the iag.
2023 *
2024 * RETURN VALUES:
2025 *	0	- success.
2026 *	-ENOSPC	- insufficient disk resources.
2027 *	-EIO	- i/o error.
2028 */
2029static int diAllocBit(struct inomap * imap, struct iag * iagp, int ino)
2030{
2031	int extno, bitno, agno, sword, rc;
2032	struct metapage *amp = NULL, *bmp = NULL;
2033	struct iag *aiagp = NULL, *biagp = NULL;
2034	u32 mask;
2035
2036	/* check if this is the last free inode within the iag.
2037	 * if so, it will have to be removed from the ag free
2038	 * inode list, so get the iags preceding and following
2039	 * it on the list.
2040	 */
2041	if (iagp->nfreeinos == cpu_to_le32(1)) {
2042		if ((int) le32_to_cpu(iagp->inofreefwd) >= 0) {
2043			if ((rc =
2044			     diIAGRead(imap, le32_to_cpu(iagp->inofreefwd),
2045				       &amp)))
2046				return (rc);
2047			aiagp = (struct iag *) amp->data;
2048		}
2049
2050		if ((int) le32_to_cpu(iagp->inofreeback) >= 0) {
2051			if ((rc =
2052			     diIAGRead(imap,
2053				       le32_to_cpu(iagp->inofreeback),
2054				       &bmp))) {
2055				if (amp)
2056					release_metapage(amp);
2057				return (rc);
2058			}
2059			biagp = (struct iag *) bmp->data;
2060		}
2061	}
2062
2063	/* get the ag number, extent number, inode number within
2064	 * the extent.
2065	 */
2066	agno = BLKTOAG(le64_to_cpu(iagp->agstart), JFS_SBI(imap->im_ipimap->i_sb));
2067	extno = ino >> L2INOSPEREXT;
2068	bitno = ino & (INOSPEREXT - 1);
2069
2070	/* compute the mask for setting the map.
2071	 */
2072	mask = HIGHORDER >> bitno;
2073
2074	/* the inode should be free and backed.
2075	 */
2076	if (((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) ||
2077	    ((le32_to_cpu(iagp->wmap[extno]) & mask) != 0) ||
2078	    (addressPXD(&iagp->inoext[extno]) == 0)) {
2079		if (amp)
2080			release_metapage(amp);
2081		if (bmp)
2082			release_metapage(bmp);
2083
2084		jfs_error(imap->im_ipimap->i_sb,
2085			  "diAllocBit: iag inconsistent");
2086		return -EIO;
2087	}
2088
2089	/* mark the inode as allocated in the working map.
2090	 */
2091	iagp->wmap[extno] |= cpu_to_le32(mask);
2092
2093	/* check if all inodes within the extent are now
2094	 * allocated.  if so, update the free inode summary
2095	 * map to reflect this.
2096	 */
2097	if (iagp->wmap[extno] == cpu_to_le32(ONES)) {
2098		sword = extno >> L2EXTSPERSUM;
2099		bitno = extno & (EXTSPERSUM - 1);
2100		iagp->inosmap[sword] |= cpu_to_le32(HIGHORDER >> bitno);
2101	}
2102
2103	/* if this was the last free inode in the iag, remove the
2104	 * iag from the ag free inode list.
2105	 */
2106	if (iagp->nfreeinos == cpu_to_le32(1)) {
2107		if (amp) {
2108			aiagp->inofreeback = iagp->inofreeback;
2109			write_metapage(amp);
2110		}
2111
2112		if (bmp) {
2113			biagp->inofreefwd = iagp->inofreefwd;
2114			write_metapage(bmp);
2115		} else {
2116			imap->im_agctl[agno].inofree =
2117			    le32_to_cpu(iagp->inofreefwd);
2118		}
2119		iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1);
2120	}
2121
2122	/* update the free inode count at the iag, ag, inode
2123	 * map levels.
2124	 */
2125	le32_add_cpu(&iagp->nfreeinos, -1);
2126	imap->im_agctl[agno].numfree -= 1;
2127	atomic_dec(&imap->im_numfree);
2128
2129	return (0);
2130}
2131
2132
2133/*
2134 * NAME:	diNewExt(imap,iagp,extno)
2135 *
2136 * FUNCTION:	initialize a new extent of inodes for an iag, allocating
2137 *		the first inode of the extent for use for the current
2138 *		allocation request.
2139 *
2140 *		disk resources are allocated for the new extent of inodes
2141 *		and the inodes themselves are initialized to reflect their
2142 *		existence within the extent (i.e. their inode numbers and
2143 *		inode extent addresses are set) and their initial state
2144 *		(mode and link count are set to zero).
2145 *
2146 *		if the iag is new, it is not yet on an ag extent free list
2147 *		but will now be placed on this list.
2148 *
2149 *		if the allocation of the new extent causes the iag to
2150 *		have no free extent, the iag will be removed from the
2151 *		ag extent free list.
2152 *
2153 *		if the iag has no free backed inodes, it will be placed
2154 *		on the ag free inode list, since the addition of the new
2155 *		extent will now cause it to have free inodes.
2156 *
2157 *		a careful update approach is used to provide consistency
2158 *		(i.e. list consistency) in the face of updates to multiple
2159 *		buffers.  under this approach, all required buffers are
2160 *		obtained before making any updates and are held until all
2161 *		updates are complete.
2162 *
2163 * PRE CONDITION: Already have buffer lock on iagp.  Already have AG lock on
2164 *	this AG.  Must have read lock on imap inode.
2165 *
2166 * PARAMETERS:
2167 *	imap	- pointer to inode map control structure.
2168 *	iagp	- pointer to iag.
2169 *	extno	- extent number.
2170 *
2171 * RETURN VALUES:
2172 *	0	- success.
2173 *	-ENOSPC	- insufficient disk resources.
2174 *	-EIO	- i/o error.
2175 */
2176static int diNewExt(struct inomap * imap, struct iag * iagp, int extno)
2177{
2178	int agno, iagno, fwd, back, freei = 0, sword, rc;
2179	struct iag *aiagp = NULL, *biagp = NULL, *ciagp = NULL;
2180	struct metapage *amp, *bmp, *cmp, *dmp;
2181	struct inode *ipimap;
2182	s64 blkno, hint;
2183	int i, j;
2184	u32 mask;
2185	ino_t ino;
2186	struct dinode *dp;
2187	struct jfs_sb_info *sbi;
2188
2189	/* better have free extents.
2190	 */
2191	if (!iagp->nfreeexts) {
2192		jfs_error(imap->im_ipimap->i_sb, "diNewExt: no free extents");
2193		return -EIO;
2194	}
2195
2196	/* get the inode map inode.
2197	 */
2198	ipimap = imap->im_ipimap;
2199	sbi = JFS_SBI(ipimap->i_sb);
2200
2201	amp = bmp = cmp = NULL;
2202
2203	/* get the ag and iag numbers for this iag.
2204	 */
2205	agno = BLKTOAG(le64_to_cpu(iagp->agstart), sbi);
2206	iagno = le32_to_cpu(iagp->iagnum);
2207
2208	/* check if this is the last free extent within the
2209	 * iag.  if so, the iag must be removed from the ag
2210	 * free extent list, so get the iags preceding and
2211	 * following the iag on this list.
2212	 */
2213	if (iagp->nfreeexts == cpu_to_le32(1)) {
2214		if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) {
2215			if ((rc = diIAGRead(imap, fwd, &amp)))
2216				return (rc);
2217			aiagp = (struct iag *) amp->data;
2218		}
2219
2220		if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) {
2221			if ((rc = diIAGRead(imap, back, &bmp)))
2222				goto error_out;
2223			biagp = (struct iag *) bmp->data;
2224		}
2225	} else {
2226		/* the iag has free extents.  if all extents are free
2227		 * (as is the case for a newly allocated iag), the iag
2228		 * must be added to the ag free extent list, so get
2229		 * the iag at the head of the list in preparation for
2230		 * adding this iag to this list.
2231		 */
2232		fwd = back = -1;
2233		if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2234			if ((fwd = imap->im_agctl[agno].extfree) >= 0) {
2235				if ((rc = diIAGRead(imap, fwd, &amp)))
2236					goto error_out;
2237				aiagp = (struct iag *) amp->data;
2238			}
2239		}
2240	}
2241
2242	/* check if the iag has no free inodes.  if so, the iag
2243	 * will have to be added to the ag free inode list, so get
2244	 * the iag at the head of the list in preparation for
2245	 * adding this iag to this list.  in doing this, we must
2246	 * check if we already have the iag at the head of
2247	 * the list in hand.
2248	 */
2249	if (iagp->nfreeinos == 0) {
2250		freei = imap->im_agctl[agno].inofree;
2251
2252		if (freei >= 0) {
2253			if (freei == fwd) {
2254				ciagp = aiagp;
2255			} else if (freei == back) {
2256				ciagp = biagp;
2257			} else {
2258				if ((rc = diIAGRead(imap, freei, &cmp)))
2259					goto error_out;
2260				ciagp = (struct iag *) cmp->data;
2261			}
2262			if (ciagp == NULL) {
2263				jfs_error(imap->im_ipimap->i_sb,
2264					  "diNewExt: ciagp == NULL");
2265				rc = -EIO;
2266				goto error_out;
2267			}
2268		}
2269	}
2270
2271	/* allocate disk space for the inode extent.
2272	 */
2273	if ((extno == 0) || (addressPXD(&iagp->inoext[extno - 1]) == 0))
2274		hint = ((s64) agno << sbi->bmap->db_agl2size) - 1;
2275	else
2276		hint = addressPXD(&iagp->inoext[extno - 1]) +
2277		    lengthPXD(&iagp->inoext[extno - 1]) - 1;
2278
2279	if ((rc = dbAlloc(ipimap, hint, (s64) imap->im_nbperiext, &blkno)))
2280		goto error_out;
2281
2282	/* compute the inode number of the first inode within the
2283	 * extent.
2284	 */
2285	ino = (iagno << L2INOSPERIAG) + (extno << L2INOSPEREXT);
2286
2287	/* initialize the inodes within the newly allocated extent a
2288	 * page at a time.
2289	 */
2290	for (i = 0; i < imap->im_nbperiext; i += sbi->nbperpage) {
2291		/* get a buffer for this page of disk inodes.
2292		 */
2293		dmp = get_metapage(ipimap, blkno + i, PSIZE, 1);
2294		if (dmp == NULL) {
2295			rc = -EIO;
2296			goto error_out;
2297		}
2298		dp = (struct dinode *) dmp->data;
2299
2300		/* initialize the inode number, mode, link count and
2301		 * inode extent address.
2302		 */
2303		for (j = 0; j < INOSPERPAGE; j++, dp++, ino++) {
2304			dp->di_inostamp = cpu_to_le32(sbi->inostamp);
2305			dp->di_number = cpu_to_le32(ino);
2306			dp->di_fileset = cpu_to_le32(FILESYSTEM_I);
2307			dp->di_mode = 0;
2308			dp->di_nlink = 0;
2309			PXDaddress(&(dp->di_ixpxd), blkno);
2310			PXDlength(&(dp->di_ixpxd), imap->im_nbperiext);
2311		}
2312		write_metapage(dmp);
2313	}
2314
2315	/* if this is the last free extent within the iag, remove the
2316	 * iag from the ag free extent list.
2317	 */
2318	if (iagp->nfreeexts == cpu_to_le32(1)) {
2319		if (fwd >= 0)
2320			aiagp->extfreeback = iagp->extfreeback;
2321
2322		if (back >= 0)
2323			biagp->extfreefwd = iagp->extfreefwd;
2324		else
2325			imap->im_agctl[agno].extfree =
2326			    le32_to_cpu(iagp->extfreefwd);
2327
2328		iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1);
2329	} else {
2330		/* if the iag has all free extents (newly allocated iag),
2331		 * add the iag to the ag free extent list.
2332		 */
2333		if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2334			if (fwd >= 0)
2335				aiagp->extfreeback = cpu_to_le32(iagno);
2336
2337			iagp->extfreefwd = cpu_to_le32(fwd);
2338			iagp->extfreeback = cpu_to_le32(-1);
2339			imap->im_agctl[agno].extfree = iagno;
2340		}
2341	}
2342
2343	/* if the iag has no free inodes, add the iag to the
2344	 * ag free inode list.
2345	 */
2346	if (iagp->nfreeinos == 0) {
2347		if (freei >= 0)
2348			ciagp->inofreeback = cpu_to_le32(iagno);
2349
2350		iagp->inofreefwd =
2351		    cpu_to_le32(imap->im_agctl[agno].inofree);
2352		iagp->inofreeback = cpu_to_le32(-1);
2353		imap->im_agctl[agno].inofree = iagno;
2354	}
2355
2356	/* initialize the extent descriptor of the extent. */
2357	PXDlength(&iagp->inoext[extno], imap->im_nbperiext);
2358	PXDaddress(&iagp->inoext[extno], blkno);
2359
2360	/* initialize the working and persistent map of the extent.
2361	 * the working map will be initialized such that
2362	 * it indicates the first inode of the extent is allocated.
2363	 */
2364	iagp->wmap[extno] = cpu_to_le32(HIGHORDER);
2365	iagp->pmap[extno] = 0;
2366
2367	/* update the free inode and free extent summary maps
2368	 * for the extent to indicate the extent has free inodes
2369	 * and no longer represents a free extent.
2370	 */
2371	sword = extno >> L2EXTSPERSUM;
2372	mask = HIGHORDER >> (extno & (EXTSPERSUM - 1));
2373	iagp->extsmap[sword] |= cpu_to_le32(mask);
2374	iagp->inosmap[sword] &= cpu_to_le32(~mask);
2375
2376	/* update the free inode and free extent counts for the
2377	 * iag.
2378	 */
2379	le32_add_cpu(&iagp->nfreeinos, (INOSPEREXT - 1));
2380	le32_add_cpu(&iagp->nfreeexts, -1);
2381
2382	/* update the free and backed inode counts for the ag.
2383	 */
2384	imap->im_agctl[agno].numfree += (INOSPEREXT - 1);
2385	imap->im_agctl[agno].numinos += INOSPEREXT;
2386
2387	/* update the free and backed inode counts for the inode map.
2388	 */
2389	atomic_add(INOSPEREXT - 1, &imap->im_numfree);
2390	atomic_add(INOSPEREXT, &imap->im_numinos);
2391
2392	/* write the iags.
2393	 */
2394	if (amp)
2395		write_metapage(amp);
2396	if (bmp)
2397		write_metapage(bmp);
2398	if (cmp)
2399		write_metapage(cmp);
2400
2401	return (0);
2402
2403      error_out:
2404
2405	/* release the iags.
2406	 */
2407	if (amp)
2408		release_metapage(amp);
2409	if (bmp)
2410		release_metapage(bmp);
2411	if (cmp)
2412		release_metapage(cmp);
2413
2414	return (rc);
2415}
2416
2417
2418/*
2419 * NAME:	diNewIAG(imap,iagnop,agno)
2420 *
2421 * FUNCTION:	allocate a new iag for an allocation group.
2422 *
2423 *		first tries to allocate the iag from the inode map
2424 *		iagfree list:
2425 *		if the list has free iags, the head of the list is removed
2426 *		and returned to satisfy the request.
2427 *		if the inode map's iag free list is empty, the inode map
2428 *		is extended to hold a new iag. this new iag is initialized
2429 *		and returned to satisfy the request.
2430 *
2431 * PARAMETERS:
2432 *	imap	- pointer to inode map control structure.
2433 *	iagnop	- pointer to an iag number set with the number of the
2434 *		  newly allocated iag upon successful return.
2435 *	agno	- allocation group number.
2436 *	bpp	- Buffer pointer to be filled in with new IAG's buffer
2437 *
2438 * RETURN VALUES:
2439 *	0	- success.
2440 *	-ENOSPC	- insufficient disk resources.
2441 *	-EIO	- i/o error.
2442 *
2443 * serialization:
2444 *	AG lock held on entry/exit;
2445 *	write lock on the map is held inside;
2446 *	read lock on the map is held on successful completion;
2447 *
2448 * note: new iag transaction:
2449 * . synchronously write iag;
2450 * . write log of xtree and inode of imap;
2451 * . commit;
2452 * . synchronous write of xtree (right to left, bottom to top);
2453 * . at start of logredo(): init in-memory imap with one additional iag page;
2454 * . at end of logredo(): re-read imap inode to determine
2455 *   new imap size;
2456 */
2457static int
2458diNewIAG(struct inomap * imap, int *iagnop, int agno, struct metapage ** mpp)
2459{
2460	int rc;
2461	int iagno, i, xlen;
2462	struct inode *ipimap;
2463	struct super_block *sb;
2464	struct jfs_sb_info *sbi;
2465	struct metapage *mp;
2466	struct iag *iagp;
2467	s64 xaddr = 0;
2468	s64 blkno;
2469	tid_t tid;
2470	struct inode *iplist[1];
2471
2472	/* pick up pointers to the inode map and mount inodes */
2473	ipimap = imap->im_ipimap;
2474	sb = ipimap->i_sb;
2475	sbi = JFS_SBI(sb);
2476
2477	/* acquire the free iag lock */
2478	IAGFREE_LOCK(imap);
2479
2480	/* if there are any iags on the inode map free iag list,
2481	 * allocate the iag from the head of the list.
2482	 */
2483	if (imap->im_freeiag >= 0) {
2484		/* pick up the iag number at the head of the list */
2485		iagno = imap->im_freeiag;
2486
2487		/* determine the logical block number of the iag */
2488		blkno = IAGTOLBLK(iagno, sbi->l2nbperpage);
2489	} else {
2490		/* no free iags. the inode map will have to be extented
2491		 * to include a new iag.
2492		 */
2493
2494		/* acquire inode map lock */
2495		IWRITE_LOCK(ipimap, RDWRLOCK_IMAP);
2496
2497		if (ipimap->i_size >> L2PSIZE != imap->im_nextiag + 1) {
2498			IWRITE_UNLOCK(ipimap);
2499			IAGFREE_UNLOCK(imap);
2500			jfs_error(imap->im_ipimap->i_sb,
2501				  "diNewIAG: ipimap->i_size is wrong");
2502			return -EIO;
2503		}
2504
2505
2506		/* get the next available iag number */
2507		iagno = imap->im_nextiag;
2508
2509		/* make sure that we have not exceeded the maximum inode
2510		 * number limit.
2511		 */
2512		if (iagno > (MAXIAGS - 1)) {
2513			/* release the inode map lock */
2514			IWRITE_UNLOCK(ipimap);
2515
2516			rc = -ENOSPC;
2517			goto out;
2518		}
2519
2520		/*
2521		 * synchronously append new iag page.
2522		 */
2523		/* determine the logical address of iag page to append */
2524		blkno = IAGTOLBLK(iagno, sbi->l2nbperpage);
2525
2526		/* Allocate extent for new iag page */
2527		xlen = sbi->nbperpage;
2528		if ((rc = dbAlloc(ipimap, 0, (s64) xlen, &xaddr))) {
2529			/* release the inode map lock */
2530			IWRITE_UNLOCK(ipimap);
2531
2532			goto out;
2533		}
2534
2535		/*
2536		 * start transaction of update of the inode map
2537		 * addressing structure pointing to the new iag page;
2538		 */
2539		tid = txBegin(sb, COMMIT_FORCE);
2540		mutex_lock(&JFS_IP(ipimap)->commit_mutex);
2541
2542		/* update the inode map addressing structure to point to it */
2543		if ((rc =
2544		     xtInsert(tid, ipimap, 0, blkno, xlen, &xaddr, 0))) {
2545			txEnd(tid);
2546			mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
2547			/* Free the blocks allocated for the iag since it was
2548			 * not successfully added to the inode map
2549			 */
2550			dbFree(ipimap, xaddr, (s64) xlen);
2551
2552			/* release the inode map lock */
2553			IWRITE_UNLOCK(ipimap);
2554
2555			goto out;
2556		}
2557
2558		/* update the inode map's inode to reflect the extension */
2559		ipimap->i_size += PSIZE;
2560		inode_add_bytes(ipimap, PSIZE);
2561
2562		/* assign a buffer for the page */
2563		mp = get_metapage(ipimap, blkno, PSIZE, 0);
2564		if (!mp) {
2565			/*
2566			 * This is very unlikely since we just created the
2567			 * extent, but let's try to handle it correctly
2568			 */
2569			xtTruncate(tid, ipimap, ipimap->i_size - PSIZE,
2570				   COMMIT_PWMAP);
2571
2572			txAbort(tid, 0);
2573			txEnd(tid);
2574			mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
2575
2576			/* release the inode map lock */
2577			IWRITE_UNLOCK(ipimap);
2578
2579			rc = -EIO;
2580			goto out;
2581		}
2582		iagp = (struct iag *) mp->data;
2583
2584		/* init the iag */
2585		memset(iagp, 0, sizeof(struct iag));
2586		iagp->iagnum = cpu_to_le32(iagno);
2587		iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1);
2588		iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1);
2589		iagp->iagfree = cpu_to_le32(-1);
2590		iagp->nfreeinos = 0;
2591		iagp->nfreeexts = cpu_to_le32(EXTSPERIAG);
2592
2593		/* initialize the free inode summary map (free extent
2594		 * summary map initialization handled by bzero).
2595		 */
2596		for (i = 0; i < SMAPSZ; i++)
2597			iagp->inosmap[i] = cpu_to_le32(ONES);
2598
2599		/*
2600		 * Write and sync the metapage
2601		 */
2602		flush_metapage(mp);
2603
2604		/*
2605		 * txCommit(COMMIT_FORCE) will synchronously write address
2606		 * index pages and inode after commit in careful update order
2607		 * of address index pages (right to left, bottom up);
2608		 */
2609		iplist[0] = ipimap;
2610		rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE);
2611
2612		txEnd(tid);
2613		mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
2614
2615		duplicateIXtree(sb, blkno, xlen, &xaddr);
2616
2617		/* update the next available iag number */
2618		imap->im_nextiag += 1;
2619
2620		/* Add the iag to the iag free list so we don't lose the iag
2621		 * if a failure happens now.
2622		 */
2623		imap->im_freeiag = iagno;
2624
2625		/* Until we have logredo working, we want the imap inode &
2626		 * control page to be up to date.
2627		 */
2628		diSync(ipimap);
2629
2630		/* release the inode map lock */
2631		IWRITE_UNLOCK(ipimap);
2632	}
2633
2634	/* obtain read lock on map */
2635	IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
2636
2637	/* read the iag */
2638	if ((rc = diIAGRead(imap, iagno, &mp))) {
2639		IREAD_UNLOCK(ipimap);
2640		rc = -EIO;
2641		goto out;
2642	}
2643	iagp = (struct iag *) mp->data;
2644
2645	/* remove the iag from the iag free list */
2646	imap->im_freeiag = le32_to_cpu(iagp->iagfree);
2647	iagp->iagfree = cpu_to_le32(-1);
2648
2649	/* set the return iag number and buffer pointer */
2650	*iagnop = iagno;
2651	*mpp = mp;
2652
2653      out:
2654	/* release the iag free lock */
2655	IAGFREE_UNLOCK(imap);
2656
2657	return (rc);
2658}
2659
2660/*
2661 * NAME:	diIAGRead()
2662 *
2663 * FUNCTION:	get the buffer for the specified iag within a fileset
2664 *		or aggregate inode map.
2665 *
2666 * PARAMETERS:
2667 *	imap	- pointer to inode map control structure.
2668 *	iagno	- iag number.
2669 *	bpp	- point to buffer pointer to be filled in on successful
2670 *		  exit.
2671 *
2672 * SERIALIZATION:
2673 *	must have read lock on imap inode
2674 *	(When called by diExtendFS, the filesystem is quiesced, therefore
2675 *	 the read lock is unnecessary.)
2676 *
2677 * RETURN VALUES:
2678 *	0	- success.
2679 *	-EIO	- i/o error.
2680 */
2681static int diIAGRead(struct inomap * imap, int iagno, struct metapage ** mpp)
2682{
2683	struct inode *ipimap = imap->im_ipimap;
2684	s64 blkno;
2685
2686	/* compute the logical block number of the iag. */
2687	blkno = IAGTOLBLK(iagno, JFS_SBI(ipimap->i_sb)->l2nbperpage);
2688
2689	/* read the iag. */
2690	*mpp = read_metapage(ipimap, blkno, PSIZE, 0);
2691	if (*mpp == NULL) {
2692		return -EIO;
2693	}
2694
2695	return (0);
2696}
2697
2698/*
2699 * NAME:	diFindFree()
2700 *
2701 * FUNCTION:	find the first free bit in a word starting at
2702 *		the specified bit position.
2703 *
2704 * PARAMETERS:
2705 *	word	- word to be examined.
2706 *	start	- starting bit position.
2707 *
2708 * RETURN VALUES:
2709 *	bit position of first free bit in the word or 32 if
2710 *	no free bits were found.
2711 */
2712static int diFindFree(u32 word, int start)
2713{
2714	int bitno;
2715	assert(start < 32);
2716	/* scan the word for the first free bit. */
2717	for (word <<= start, bitno = start; bitno < 32;
2718	     bitno++, word <<= 1) {
2719		if ((word & HIGHORDER) == 0)
2720			break;
2721	}
2722	return (bitno);
2723}
2724
2725/*
2726 * NAME:	diUpdatePMap()
2727 *
2728 * FUNCTION: Update the persistent map in an IAG for the allocation or
2729 *	freeing of the specified inode.
2730 *
2731 * PRE CONDITIONS: Working map has already been updated for allocate.
2732 *
2733 * PARAMETERS:
2734 *	ipimap	- Incore inode map inode
2735 *	inum	- Number of inode to mark in permanent map
2736 *	is_free	- If 'true' indicates inode should be marked freed, otherwise
2737 *		  indicates inode should be marked allocated.
2738 *
2739 * RETURN VALUES:
2740 *		0 for success
2741 */
2742int
2743diUpdatePMap(struct inode *ipimap,
2744	     unsigned long inum, bool is_free, struct tblock * tblk)
2745{
2746	int rc;
2747	struct iag *iagp;
2748	struct metapage *mp;
2749	int iagno, ino, extno, bitno;
2750	struct inomap *imap;
2751	u32 mask;
2752	struct jfs_log *log;
2753	int lsn, difft, diffp;
2754	unsigned long flags;
2755
2756	imap = JFS_IP(ipimap)->i_imap;
2757	/* get the iag number containing the inode */
2758	iagno = INOTOIAG(inum);
2759	/* make sure that the iag is contained within the map */
2760	if (iagno >= imap->im_nextiag) {
2761		jfs_error(ipimap->i_sb,
2762			  "diUpdatePMap: the iag is outside the map");
2763		return -EIO;
2764	}
2765	/* read the iag */
2766	IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
2767	rc = diIAGRead(imap, iagno, &mp);
2768	IREAD_UNLOCK(ipimap);
2769	if (rc)
2770		return (rc);
2771	metapage_wait_for_io(mp);
2772	iagp = (struct iag *) mp->data;
2773	/* get the inode number and extent number of the inode within
2774	 * the iag and the inode number within the extent.
2775	 */
2776	ino = inum & (INOSPERIAG - 1);
2777	extno = ino >> L2INOSPEREXT;
2778	bitno = ino & (INOSPEREXT - 1);
2779	mask = HIGHORDER >> bitno;
2780	/*
2781	 * mark the inode free in persistent map:
2782	 */
2783	if (is_free) {
2784		/* The inode should have been allocated both in working
2785		 * map and in persistent map;
2786		 * the inode will be freed from working map at the release
2787		 * of last reference release;
2788		 */
2789		if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
2790			jfs_error(ipimap->i_sb,
2791				  "diUpdatePMap: inode %ld not marked as "
2792				  "allocated in wmap!", inum);
2793		}
2794		if (!(le32_to_cpu(iagp->pmap[extno]) & mask)) {
2795			jfs_error(ipimap->i_sb,
2796				  "diUpdatePMap: inode %ld not marked as "
2797				  "allocated in pmap!", inum);
2798		}
2799		/* update the bitmap for the extent of the freed inode */
2800		iagp->pmap[extno] &= cpu_to_le32(~mask);
2801	}
2802	/*
2803	 * mark the inode allocated in persistent map:
2804	 */
2805	else {
2806		/* The inode should be already allocated in the working map
2807		 * and should be free in persistent map;
2808		 */
2809		if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
2810			release_metapage(mp);
2811			jfs_error(ipimap->i_sb,
2812				  "diUpdatePMap: the inode is not allocated in "
2813				  "the working map");
2814			return -EIO;
2815		}
2816		if ((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) {
2817			release_metapage(mp);
2818			jfs_error(ipimap->i_sb,
2819				  "diUpdatePMap: the inode is not free in the "
2820				  "persistent map");
2821			return -EIO;
2822		}
2823		/* update the bitmap for the extent of the allocated inode */
2824		iagp->pmap[extno] |= cpu_to_le32(mask);
2825	}
2826	/*
2827	 * update iag lsn
2828	 */
2829	lsn = tblk->lsn;
2830	log = JFS_SBI(tblk->sb)->log;
2831	LOGSYNC_LOCK(log, flags);
2832	if (mp->lsn != 0) {
2833		/* inherit older/smaller lsn */
2834		logdiff(difft, lsn, log);
2835		logdiff(diffp, mp->lsn, log);
2836		if (difft < diffp) {
2837			mp->lsn = lsn;
2838			/* move mp after tblock in logsync list */
2839			list_move(&mp->synclist, &tblk->synclist);
2840		}
2841		/* inherit younger/larger clsn */
2842		assert(mp->clsn);
2843		logdiff(difft, tblk->clsn, log);
2844		logdiff(diffp, mp->clsn, log);
2845		if (difft > diffp)
2846			mp->clsn = tblk->clsn;
2847	} else {
2848		mp->log = log;
2849		mp->lsn = lsn;
2850		/* insert mp after tblock in logsync list */
2851		log->count++;
2852		list_add(&mp->synclist, &tblk->synclist);
2853		mp->clsn = tblk->clsn;
2854	}
2855	LOGSYNC_UNLOCK(log, flags);
2856	write_metapage(mp);
2857	return (0);
2858}
2859
2860/*
2861 *	diExtendFS()
2862 *
2863 * function: update imap for extendfs();
2864 *
2865 * note: AG size has been increased s.t. each k old contiguous AGs are
2866 * coalesced into a new AG;
2867 */
2868int diExtendFS(struct inode *ipimap, struct inode *ipbmap)
2869{
2870	int rc, rcx = 0;
2871	struct inomap *imap = JFS_IP(ipimap)->i_imap;
2872	struct iag *iagp = NULL, *hiagp = NULL;
2873	struct bmap *mp = JFS_SBI(ipbmap->i_sb)->bmap;
2874	struct metapage *bp, *hbp;
2875	int i, n, head;
2876	int numinos, xnuminos = 0, xnumfree = 0;
2877	s64 agstart;
2878
2879	jfs_info("diExtendFS: nextiag:%d numinos:%d numfree:%d",
2880		   imap->im_nextiag, atomic_read(&imap->im_numinos),
2881		   atomic_read(&imap->im_numfree));
2882
2883	/*
2884	 *	reconstruct imap
2885	 *
2886	 * coalesce contiguous k (newAGSize/oldAGSize) AGs;
2887	 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn;
2888	 * note: new AG size = old AG size * (2**x).
2889	 */
2890
2891	/* init per AG control information im_agctl[] */
2892	for (i = 0; i < MAXAG; i++) {
2893		imap->im_agctl[i].inofree = -1;
2894		imap->im_agctl[i].extfree = -1;
2895		imap->im_agctl[i].numinos = 0;	/* number of backed inodes */
2896		imap->im_agctl[i].numfree = 0;	/* number of free backed inodes */
2897	}
2898
2899	/*
2900	 *	process each iag page of the map.
2901	 *
2902	 * rebuild AG Free Inode List, AG Free Inode Extent List;
2903	 */
2904	for (i = 0; i < imap->im_nextiag; i++) {
2905		if ((rc = diIAGRead(imap, i, &bp))) {
2906			rcx = rc;
2907			continue;
2908		}
2909		iagp = (struct iag *) bp->data;
2910		if (le32_to_cpu(iagp->iagnum) != i) {
2911			release_metapage(bp);
2912			jfs_error(ipimap->i_sb,
2913				  "diExtendFs: unexpected value of iagnum");
2914			return -EIO;
2915		}
2916
2917		/* leave free iag in the free iag list */
2918		if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2919			release_metapage(bp);
2920			continue;
2921		}
2922
2923		agstart = le64_to_cpu(iagp->agstart);
2924		n = agstart >> mp->db_agl2size;
2925		iagp->agstart = cpu_to_le64((s64)n << mp->db_agl2size);
2926
2927		/* compute backed inodes */
2928		numinos = (EXTSPERIAG - le32_to_cpu(iagp->nfreeexts))
2929		    << L2INOSPEREXT;
2930		if (numinos > 0) {
2931			/* merge AG backed inodes */
2932			imap->im_agctl[n].numinos += numinos;
2933			xnuminos += numinos;
2934		}
2935
2936		/* if any backed free inodes, insert at AG free inode list */
2937		if ((int) le32_to_cpu(iagp->nfreeinos) > 0) {
2938			if ((head = imap->im_agctl[n].inofree) == -1) {
2939				iagp->inofreefwd = cpu_to_le32(-1);
2940				iagp->inofreeback = cpu_to_le32(-1);
2941			} else {
2942				if ((rc = diIAGRead(imap, head, &hbp))) {
2943					rcx = rc;
2944					goto nextiag;
2945				}
2946				hiagp = (struct iag *) hbp->data;
2947				hiagp->inofreeback = iagp->iagnum;
2948				iagp->inofreefwd = cpu_to_le32(head);
2949				iagp->inofreeback = cpu_to_le32(-1);
2950				write_metapage(hbp);
2951			}
2952
2953			imap->im_agctl[n].inofree =
2954			    le32_to_cpu(iagp->iagnum);
2955
2956			/* merge AG backed free inodes */
2957			imap->im_agctl[n].numfree +=
2958			    le32_to_cpu(iagp->nfreeinos);
2959			xnumfree += le32_to_cpu(iagp->nfreeinos);
2960		}
2961
2962		/* if any free extents, insert at AG free extent list */
2963		if (le32_to_cpu(iagp->nfreeexts) > 0) {
2964			if ((head = imap->im_agctl[n].extfree) == -1) {
2965				iagp->extfreefwd = cpu_to_le32(-1);
2966				iagp->extfreeback = cpu_to_le32(-1);
2967			} else {
2968				if ((rc = diIAGRead(imap, head, &hbp))) {
2969					rcx = rc;
2970					goto nextiag;
2971				}
2972				hiagp = (struct iag *) hbp->data;
2973				hiagp->extfreeback = iagp->iagnum;
2974				iagp->extfreefwd = cpu_to_le32(head);
2975				iagp->extfreeback = cpu_to_le32(-1);
2976				write_metapage(hbp);
2977			}
2978
2979			imap->im_agctl[n].extfree =
2980			    le32_to_cpu(iagp->iagnum);
2981		}
2982
2983	      nextiag:
2984		write_metapage(bp);
2985	}
2986
2987	if (xnuminos != atomic_read(&imap->im_numinos) ||
2988	    xnumfree != atomic_read(&imap->im_numfree)) {
2989		jfs_error(ipimap->i_sb,
2990			  "diExtendFs: numinos or numfree incorrect");
2991		return -EIO;
2992	}
2993
2994	return rcx;
2995}
2996
2997
2998/*
2999 *	duplicateIXtree()
3000 *
3001 * serialization: IWRITE_LOCK held on entry/exit
3002 *
3003 * note: shadow page with regular inode (rel.2);
3004 */
3005static void duplicateIXtree(struct super_block *sb, s64 blkno,
3006			    int xlen, s64 *xaddr)
3007{
3008	struct jfs_superblock *j_sb;
3009	struct buffer_head *bh;
3010	struct inode *ip;
3011	tid_t tid;
3012
3013	/* if AIT2 ipmap2 is bad, do not try to update it */
3014	if (JFS_SBI(sb)->mntflag & JFS_BAD_SAIT)	/* s_flag */
3015		return;
3016	ip = diReadSpecial(sb, FILESYSTEM_I, 1);
3017	if (ip == NULL) {
3018		JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT;
3019		if (readSuper(sb, &bh))
3020			return;
3021		j_sb = (struct jfs_superblock *)bh->b_data;
3022		j_sb->s_flag |= cpu_to_le32(JFS_BAD_SAIT);
3023
3024		mark_buffer_dirty(bh);
3025		sync_dirty_buffer(bh);
3026		brelse(bh);
3027		return;
3028	}
3029
3030	/* start transaction */
3031	tid = txBegin(sb, COMMIT_FORCE);
3032	/* update the inode map addressing structure to point to it */
3033	if (xtInsert(tid, ip, 0, blkno, xlen, xaddr, 0)) {
3034		JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT;
3035		txAbort(tid, 1);
3036		goto cleanup;
3037
3038	}
3039	/* update the inode map's inode to reflect the extension */
3040	ip->i_size += PSIZE;
3041	inode_add_bytes(ip, PSIZE);
3042	txCommit(tid, 1, &ip, COMMIT_FORCE);
3043      cleanup:
3044	txEnd(tid);
3045	diFreeSpecial(ip);
3046}
3047
3048/*
3049 * NAME:	copy_from_dinode()
3050 *
3051 * FUNCTION:	Copies inode info from disk inode to in-memory inode
3052 *
3053 * RETURN VALUES:
3054 *	0	- success
3055 *	-ENOMEM	- insufficient memory
3056 */
3057static int copy_from_dinode(struct dinode * dip, struct inode *ip)
3058{
3059	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
3060	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
3061
3062	jfs_ip->fileset = le32_to_cpu(dip->di_fileset);
3063	jfs_ip->mode2 = le32_to_cpu(dip->di_mode);
3064	jfs_set_inode_flags(ip);
3065
3066	ip->i_mode = le32_to_cpu(dip->di_mode) & 0xffff;
3067	if (sbi->umask != -1) {
3068		ip->i_mode = (ip->i_mode & ~0777) | (0777 & ~sbi->umask);
3069		/* For directories, add x permission if r is allowed by umask */
3070		if (S_ISDIR(ip->i_mode)) {
3071			if (ip->i_mode & 0400)
3072				ip->i_mode |= 0100;
3073			if (ip->i_mode & 0040)
3074				ip->i_mode |= 0010;
3075			if (ip->i_mode & 0004)
3076				ip->i_mode |= 0001;
3077		}
3078	}
3079	ip->i_nlink = le32_to_cpu(dip->di_nlink);
3080
3081	jfs_ip->saved_uid = le32_to_cpu(dip->di_uid);
3082	if (sbi->uid == -1)
3083		ip->i_uid = jfs_ip->saved_uid;
3084	else {
3085		ip->i_uid = sbi->uid;
3086	}
3087
3088	jfs_ip->saved_gid = le32_to_cpu(dip->di_gid);
3089	if (sbi->gid == -1)
3090		ip->i_gid = jfs_ip->saved_gid;
3091	else {
3092		ip->i_gid = sbi->gid;
3093	}
3094
3095	ip->i_size = le64_to_cpu(dip->di_size);
3096	ip->i_atime.tv_sec = le32_to_cpu(dip->di_atime.tv_sec);
3097	ip->i_atime.tv_nsec = le32_to_cpu(dip->di_atime.tv_nsec);
3098	ip->i_mtime.tv_sec = le32_to_cpu(dip->di_mtime.tv_sec);
3099	ip->i_mtime.tv_nsec = le32_to_cpu(dip->di_mtime.tv_nsec);
3100	ip->i_ctime.tv_sec = le32_to_cpu(dip->di_ctime.tv_sec);
3101	ip->i_ctime.tv_nsec = le32_to_cpu(dip->di_ctime.tv_nsec);
3102	ip->i_blocks = LBLK2PBLK(ip->i_sb, le64_to_cpu(dip->di_nblocks));
3103	ip->i_generation = le32_to_cpu(dip->di_gen);
3104
3105	jfs_ip->ixpxd = dip->di_ixpxd;	/* in-memory pxd's are little-endian */
3106	jfs_ip->acl = dip->di_acl;	/* as are dxd's */
3107	jfs_ip->ea = dip->di_ea;
3108	jfs_ip->next_index = le32_to_cpu(dip->di_next_index);
3109	jfs_ip->otime = le32_to_cpu(dip->di_otime.tv_sec);
3110	jfs_ip->acltype = le32_to_cpu(dip->di_acltype);
3111
3112	if (S_ISCHR(ip->i_mode) || S_ISBLK(ip->i_mode)) {
3113		jfs_ip->dev = le32_to_cpu(dip->di_rdev);
3114		ip->i_rdev = new_decode_dev(jfs_ip->dev);
3115	}
3116
3117	if (S_ISDIR(ip->i_mode)) {
3118		memcpy(&jfs_ip->i_dirtable, &dip->di_dirtable, 384);
3119	} else if (S_ISREG(ip->i_mode) || S_ISLNK(ip->i_mode)) {
3120		memcpy(&jfs_ip->i_xtroot, &dip->di_xtroot, 288);
3121	} else
3122		memcpy(&jfs_ip->i_inline_ea, &dip->di_inlineea, 128);
3123
3124	/* Zero the in-memory-only stuff */
3125	jfs_ip->cflag = 0;
3126	jfs_ip->btindex = 0;
3127	jfs_ip->btorder = 0;
3128	jfs_ip->bxflag = 0;
3129	jfs_ip->blid = 0;
3130	jfs_ip->atlhead = 0;
3131	jfs_ip->atltail = 0;
3132	jfs_ip->xtlid = 0;
3133	return (0);
3134}
3135
3136/*
3137 * NAME:	copy_to_dinode()
3138 *
3139 * FUNCTION:	Copies inode info from in-memory inode to disk inode
3140 */
3141static void copy_to_dinode(struct dinode * dip, struct inode *ip)
3142{
3143	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
3144	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
3145
3146	dip->di_fileset = cpu_to_le32(jfs_ip->fileset);
3147	dip->di_inostamp = cpu_to_le32(sbi->inostamp);
3148	dip->di_number = cpu_to_le32(ip->i_ino);
3149	dip->di_gen = cpu_to_le32(ip->i_generation);
3150	dip->di_size = cpu_to_le64(ip->i_size);
3151	dip->di_nblocks = cpu_to_le64(PBLK2LBLK(ip->i_sb, ip->i_blocks));
3152	dip->di_nlink = cpu_to_le32(ip->i_nlink);
3153	if (sbi->uid == -1)
3154		dip->di_uid = cpu_to_le32(ip->i_uid);
3155	else
3156		dip->di_uid = cpu_to_le32(jfs_ip->saved_uid);
3157	if (sbi->gid == -1)
3158		dip->di_gid = cpu_to_le32(ip->i_gid);
 
3159	else
3160		dip->di_gid = cpu_to_le32(jfs_ip->saved_gid);
3161	jfs_get_inode_flags(jfs_ip);
3162	/*
3163	 * mode2 is only needed for storing the higher order bits.
3164	 * Trust i_mode for the lower order ones
3165	 */
3166	if (sbi->umask == -1)
3167		dip->di_mode = cpu_to_le32((jfs_ip->mode2 & 0xffff0000) |
3168					   ip->i_mode);
3169	else /* Leave the original permissions alone */
3170		dip->di_mode = cpu_to_le32(jfs_ip->mode2);
3171
3172	dip->di_atime.tv_sec = cpu_to_le32(ip->i_atime.tv_sec);
3173	dip->di_atime.tv_nsec = cpu_to_le32(ip->i_atime.tv_nsec);
3174	dip->di_ctime.tv_sec = cpu_to_le32(ip->i_ctime.tv_sec);
3175	dip->di_ctime.tv_nsec = cpu_to_le32(ip->i_ctime.tv_nsec);
3176	dip->di_mtime.tv_sec = cpu_to_le32(ip->i_mtime.tv_sec);
3177	dip->di_mtime.tv_nsec = cpu_to_le32(ip->i_mtime.tv_nsec);
3178	dip->di_ixpxd = jfs_ip->ixpxd;	/* in-memory pxd's are little-endian */
3179	dip->di_acl = jfs_ip->acl;	/* as are dxd's */
3180	dip->di_ea = jfs_ip->ea;
3181	dip->di_next_index = cpu_to_le32(jfs_ip->next_index);
3182	dip->di_otime.tv_sec = cpu_to_le32(jfs_ip->otime);
3183	dip->di_otime.tv_nsec = 0;
3184	dip->di_acltype = cpu_to_le32(jfs_ip->acltype);
3185	if (S_ISCHR(ip->i_mode) || S_ISBLK(ip->i_mode))
3186		dip->di_rdev = cpu_to_le32(jfs_ip->dev);
3187}