xref: /openbmc/linux/fs/jfs/jfs_imap.c (revision 4800cd83)
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  */
76 static int diAllocAG(struct inomap *, int, bool, struct inode *);
77 static int diAllocAny(struct inomap *, int, bool, struct inode *);
78 static int diAllocBit(struct inomap *, struct iag *, int);
79 static int diAllocExt(struct inomap *, int, struct inode *);
80 static int diAllocIno(struct inomap *, int, struct inode *);
81 static int diFindFree(u32, int);
82 static int diNewExt(struct inomap *, struct iag *, int);
83 static int diNewIAG(struct inomap *, int *, int, struct metapage **);
84 static void duplicateIXtree(struct super_block *, s64, int, s64 *);
85 
86 static int diIAGRead(struct inomap * imap, int, struct metapage **);
87 static int copy_from_dinode(struct dinode *, struct inode *);
88 static 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  */
107 int 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  */
191 int 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  */
219 int 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  */
304 int 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)->agno = BLKTOAG(agstart, sbi);
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  */
429 struct 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 
517 void 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  */
562 void 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  */
599 int 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  */
863 int 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 = JFS_IP(ip)->agno;
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 preceeding
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 preceeding
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  */
1315 static inline void
1316 diInitInode(struct inode *ip, int iagno, int ino, int extno, struct iag * iagp)
1317 {
1318 	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
1319 	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
1320 
1321 	ip->i_ino = (iagno << L2INOSPERIAG) + ino;
1322 	jfs_ip->ixpxd = iagp->inoext[extno];
1323 	jfs_ip->agno = BLKTOAG(le64_to_cpu(iagp->agstart), sbi);
1324 	jfs_ip->active_ag = -1;
1325 }
1326 
1327 
1328 /*
1329  * NAME:	diAlloc(pip,dir,ip)
1330  *
1331  * FUNCTION:	allocate a disk inode from the inode working map
1332  *		for a fileset or aggregate.
1333  *
1334  * PARAMETERS:
1335  *	pip	- pointer to incore inode for the parent inode.
1336  *	dir	- 'true' if the new disk inode is for a directory.
1337  *	ip	- pointer to a new inode
1338  *
1339  * RETURN VALUES:
1340  *	0	- success.
1341  *	-ENOSPC	- insufficient disk resources.
1342  *	-EIO	- i/o error.
1343  */
1344 int diAlloc(struct inode *pip, bool dir, struct inode *ip)
1345 {
1346 	int rc, ino, iagno, addext, extno, bitno, sword;
1347 	int nwords, rem, i, agno;
1348 	u32 mask, inosmap, extsmap;
1349 	struct inode *ipimap;
1350 	struct metapage *mp;
1351 	ino_t inum;
1352 	struct iag *iagp;
1353 	struct inomap *imap;
1354 
1355 	/* get the pointers to the inode map inode and the
1356 	 * corresponding imap control structure.
1357 	 */
1358 	ipimap = JFS_SBI(pip->i_sb)->ipimap;
1359 	imap = JFS_IP(ipimap)->i_imap;
1360 	JFS_IP(ip)->ipimap = ipimap;
1361 	JFS_IP(ip)->fileset = FILESYSTEM_I;
1362 
1363 	/* for a directory, the allocation policy is to start
1364 	 * at the ag level using the preferred ag.
1365 	 */
1366 	if (dir) {
1367 		agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap);
1368 		AG_LOCK(imap, agno);
1369 		goto tryag;
1370 	}
1371 
1372 	/* for files, the policy starts off by trying to allocate from
1373 	 * the same iag containing the parent disk inode:
1374 	 * try to allocate the new disk inode close to the parent disk
1375 	 * inode, using parent disk inode number + 1 as the allocation
1376 	 * hint.  (we use a left-to-right policy to attempt to avoid
1377 	 * moving backward on the disk.)  compute the hint within the
1378 	 * file system and the iag.
1379 	 */
1380 
1381 	/* get the ag number of this iag */
1382 	agno = JFS_IP(pip)->agno;
1383 
1384 	if (atomic_read(&JFS_SBI(pip->i_sb)->bmap->db_active[agno])) {
1385 		/*
1386 		 * There is an open file actively growing.  We want to
1387 		 * allocate new inodes from a different ag to avoid
1388 		 * fragmentation problems.
1389 		 */
1390 		agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap);
1391 		AG_LOCK(imap, agno);
1392 		goto tryag;
1393 	}
1394 
1395 	inum = pip->i_ino + 1;
1396 	ino = inum & (INOSPERIAG - 1);
1397 
1398 	/* back off the hint if it is outside of the iag */
1399 	if (ino == 0)
1400 		inum = pip->i_ino;
1401 
1402 	/* lock the AG inode map information */
1403 	AG_LOCK(imap, agno);
1404 
1405 	/* Get read lock on imap inode */
1406 	IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
1407 
1408 	/* get the iag number and read the iag */
1409 	iagno = INOTOIAG(inum);
1410 	if ((rc = diIAGRead(imap, iagno, &mp))) {
1411 		IREAD_UNLOCK(ipimap);
1412 		AG_UNLOCK(imap, agno);
1413 		return (rc);
1414 	}
1415 	iagp = (struct iag *) mp->data;
1416 
1417 	/* determine if new inode extent is allowed to be added to the iag.
1418 	 * new inode extent can be added to the iag if the ag
1419 	 * has less than 32 free disk inodes and the iag has free extents.
1420 	 */
1421 	addext = (imap->im_agctl[agno].numfree < 32 && iagp->nfreeexts);
1422 
1423 	/*
1424 	 *	try to allocate from the IAG
1425 	 */
1426 	/* check if the inode may be allocated from the iag
1427 	 * (i.e. the inode has free inodes or new extent can be added).
1428 	 */
1429 	if (iagp->nfreeinos || addext) {
1430 		/* determine the extent number of the hint.
1431 		 */
1432 		extno = ino >> L2INOSPEREXT;
1433 
1434 		/* check if the extent containing the hint has backed
1435 		 * inodes.  if so, try to allocate within this extent.
1436 		 */
1437 		if (addressPXD(&iagp->inoext[extno])) {
1438 			bitno = ino & (INOSPEREXT - 1);
1439 			if ((bitno =
1440 			     diFindFree(le32_to_cpu(iagp->wmap[extno]),
1441 					bitno))
1442 			    < INOSPEREXT) {
1443 				ino = (extno << L2INOSPEREXT) + bitno;
1444 
1445 				/* a free inode (bit) was found within this
1446 				 * extent, so allocate it.
1447 				 */
1448 				rc = diAllocBit(imap, iagp, ino);
1449 				IREAD_UNLOCK(ipimap);
1450 				if (rc) {
1451 					assert(rc == -EIO);
1452 				} else {
1453 					/* set the results of the allocation
1454 					 * and write the iag.
1455 					 */
1456 					diInitInode(ip, iagno, ino, extno,
1457 						    iagp);
1458 					mark_metapage_dirty(mp);
1459 				}
1460 				release_metapage(mp);
1461 
1462 				/* free the AG lock and return.
1463 				 */
1464 				AG_UNLOCK(imap, agno);
1465 				return (rc);
1466 			}
1467 
1468 			if (!addext)
1469 				extno =
1470 				    (extno ==
1471 				     EXTSPERIAG - 1) ? 0 : extno + 1;
1472 		}
1473 
1474 		/*
1475 		 * no free inodes within the extent containing the hint.
1476 		 *
1477 		 * try to allocate from the backed extents following
1478 		 * hint or, if appropriate (i.e. addext is true), allocate
1479 		 * an extent of free inodes at or following the extent
1480 		 * containing the hint.
1481 		 *
1482 		 * the free inode and free extent summary maps are used
1483 		 * here, so determine the starting summary map position
1484 		 * and the number of words we'll have to examine.  again,
1485 		 * the approach is to allocate following the hint, so we
1486 		 * might have to initially ignore prior bits of the summary
1487 		 * map that represent extents prior to the extent containing
1488 		 * the hint and later revisit these bits.
1489 		 */
1490 		bitno = extno & (EXTSPERSUM - 1);
1491 		nwords = (bitno == 0) ? SMAPSZ : SMAPSZ + 1;
1492 		sword = extno >> L2EXTSPERSUM;
1493 
1494 		/* mask any prior bits for the starting words of the
1495 		 * summary map.
1496 		 */
1497 		mask = ONES << (EXTSPERSUM - bitno);
1498 		inosmap = le32_to_cpu(iagp->inosmap[sword]) | mask;
1499 		extsmap = le32_to_cpu(iagp->extsmap[sword]) | mask;
1500 
1501 		/* scan the free inode and free extent summary maps for
1502 		 * free resources.
1503 		 */
1504 		for (i = 0; i < nwords; i++) {
1505 			/* check if this word of the free inode summary
1506 			 * map describes an extent with free inodes.
1507 			 */
1508 			if (~inosmap) {
1509 				/* an extent with free inodes has been
1510 				 * found. determine the extent number
1511 				 * and the inode number within the extent.
1512 				 */
1513 				rem = diFindFree(inosmap, 0);
1514 				extno = (sword << L2EXTSPERSUM) + rem;
1515 				rem = diFindFree(le32_to_cpu(iagp->wmap[extno]),
1516 						 0);
1517 				if (rem >= INOSPEREXT) {
1518 					IREAD_UNLOCK(ipimap);
1519 					release_metapage(mp);
1520 					AG_UNLOCK(imap, agno);
1521 					jfs_error(ip->i_sb,
1522 						  "diAlloc: can't find free bit "
1523 						  "in wmap");
1524 					return -EIO;
1525 				}
1526 
1527 				/* determine the inode number within the
1528 				 * iag and allocate the inode from the
1529 				 * map.
1530 				 */
1531 				ino = (extno << L2INOSPEREXT) + rem;
1532 				rc = diAllocBit(imap, iagp, ino);
1533 				IREAD_UNLOCK(ipimap);
1534 				if (rc)
1535 					assert(rc == -EIO);
1536 				else {
1537 					/* set the results of the allocation
1538 					 * and write the iag.
1539 					 */
1540 					diInitInode(ip, iagno, ino, extno,
1541 						    iagp);
1542 					mark_metapage_dirty(mp);
1543 				}
1544 				release_metapage(mp);
1545 
1546 				/* free the AG lock and return.
1547 				 */
1548 				AG_UNLOCK(imap, agno);
1549 				return (rc);
1550 
1551 			}
1552 
1553 			/* check if we may allocate an extent of free
1554 			 * inodes and whether this word of the free
1555 			 * extents summary map describes a free extent.
1556 			 */
1557 			if (addext && ~extsmap) {
1558 				/* a free extent has been found.  determine
1559 				 * the extent number.
1560 				 */
1561 				rem = diFindFree(extsmap, 0);
1562 				extno = (sword << L2EXTSPERSUM) + rem;
1563 
1564 				/* allocate an extent of free inodes.
1565 				 */
1566 				if ((rc = diNewExt(imap, iagp, extno))) {
1567 					/* if there is no disk space for a
1568 					 * new extent, try to allocate the
1569 					 * disk inode from somewhere else.
1570 					 */
1571 					if (rc == -ENOSPC)
1572 						break;
1573 
1574 					assert(rc == -EIO);
1575 				} else {
1576 					/* set the results of the allocation
1577 					 * and write the iag.
1578 					 */
1579 					diInitInode(ip, iagno,
1580 						    extno << L2INOSPEREXT,
1581 						    extno, iagp);
1582 					mark_metapage_dirty(mp);
1583 				}
1584 				release_metapage(mp);
1585 				/* free the imap inode & the AG lock & return.
1586 				 */
1587 				IREAD_UNLOCK(ipimap);
1588 				AG_UNLOCK(imap, agno);
1589 				return (rc);
1590 			}
1591 
1592 			/* move on to the next set of summary map words.
1593 			 */
1594 			sword = (sword == SMAPSZ - 1) ? 0 : sword + 1;
1595 			inosmap = le32_to_cpu(iagp->inosmap[sword]);
1596 			extsmap = le32_to_cpu(iagp->extsmap[sword]);
1597 		}
1598 	}
1599 	/* unlock imap inode */
1600 	IREAD_UNLOCK(ipimap);
1601 
1602 	/* nothing doing in this iag, so release it. */
1603 	release_metapage(mp);
1604 
1605       tryag:
1606 	/*
1607 	 * try to allocate anywhere within the same AG as the parent inode.
1608 	 */
1609 	rc = diAllocAG(imap, agno, dir, ip);
1610 
1611 	AG_UNLOCK(imap, agno);
1612 
1613 	if (rc != -ENOSPC)
1614 		return (rc);
1615 
1616 	/*
1617 	 * try to allocate in any AG.
1618 	 */
1619 	return (diAllocAny(imap, agno, dir, ip));
1620 }
1621 
1622 
1623 /*
1624  * NAME:	diAllocAG(imap,agno,dir,ip)
1625  *
1626  * FUNCTION:	allocate a disk inode from the allocation group.
1627  *
1628  *		this routine first determines if a new extent of free
1629  *		inodes should be added for the allocation group, with
1630  *		the current request satisfied from this extent. if this
1631  *		is the case, an attempt will be made to do just that.  if
1632  *		this attempt fails or it has been determined that a new
1633  *		extent should not be added, an attempt is made to satisfy
1634  *		the request by allocating an existing (backed) free inode
1635  *		from the allocation group.
1636  *
1637  * PRE CONDITION: Already have the AG lock for this AG.
1638  *
1639  * PARAMETERS:
1640  *	imap	- pointer to inode map control structure.
1641  *	agno	- allocation group to allocate from.
1642  *	dir	- 'true' if the new disk inode is for a directory.
1643  *	ip	- pointer to the new inode to be filled in on successful return
1644  *		  with the disk inode number allocated, its extent address
1645  *		  and the start of the ag.
1646  *
1647  * RETURN VALUES:
1648  *	0	- success.
1649  *	-ENOSPC	- insufficient disk resources.
1650  *	-EIO	- i/o error.
1651  */
1652 static int
1653 diAllocAG(struct inomap * imap, int agno, bool dir, struct inode *ip)
1654 {
1655 	int rc, addext, numfree, numinos;
1656 
1657 	/* get the number of free and the number of backed disk
1658 	 * inodes currently within the ag.
1659 	 */
1660 	numfree = imap->im_agctl[agno].numfree;
1661 	numinos = imap->im_agctl[agno].numinos;
1662 
1663 	if (numfree > numinos) {
1664 		jfs_error(ip->i_sb, "diAllocAG: numfree > numinos");
1665 		return -EIO;
1666 	}
1667 
1668 	/* determine if we should allocate a new extent of free inodes
1669 	 * within the ag: for directory inodes, add a new extent
1670 	 * if there are a small number of free inodes or number of free
1671 	 * inodes is a small percentage of the number of backed inodes.
1672 	 */
1673 	if (dir)
1674 		addext = (numfree < 64 ||
1675 			  (numfree < 256
1676 			   && ((numfree * 100) / numinos) <= 20));
1677 	else
1678 		addext = (numfree == 0);
1679 
1680 	/*
1681 	 * try to allocate a new extent of free inodes.
1682 	 */
1683 	if (addext) {
1684 		/* if free space is not avaliable for this new extent, try
1685 		 * below to allocate a free and existing (already backed)
1686 		 * inode from the ag.
1687 		 */
1688 		if ((rc = diAllocExt(imap, agno, ip)) != -ENOSPC)
1689 			return (rc);
1690 	}
1691 
1692 	/*
1693 	 * try to allocate an existing free inode from the ag.
1694 	 */
1695 	return (diAllocIno(imap, agno, ip));
1696 }
1697 
1698 
1699 /*
1700  * NAME:	diAllocAny(imap,agno,dir,iap)
1701  *
1702  * FUNCTION:	allocate a disk inode from any other allocation group.
1703  *
1704  *		this routine is called when an allocation attempt within
1705  *		the primary allocation group has failed. if attempts to
1706  *		allocate an inode from any allocation group other than the
1707  *		specified primary group.
1708  *
1709  * PARAMETERS:
1710  *	imap	- pointer to inode map control structure.
1711  *	agno	- primary allocation group (to avoid).
1712  *	dir	- 'true' if the new disk inode is for a directory.
1713  *	ip	- pointer to a new inode to be filled in on successful return
1714  *		  with the disk inode number allocated, its extent address
1715  *		  and the start of the ag.
1716  *
1717  * RETURN VALUES:
1718  *	0	- success.
1719  *	-ENOSPC	- insufficient disk resources.
1720  *	-EIO	- i/o error.
1721  */
1722 static int
1723 diAllocAny(struct inomap * imap, int agno, bool dir, struct inode *ip)
1724 {
1725 	int ag, rc;
1726 	int maxag = JFS_SBI(imap->im_ipimap->i_sb)->bmap->db_maxag;
1727 
1728 
1729 	/* try to allocate from the ags following agno up to
1730 	 * the maximum ag number.
1731 	 */
1732 	for (ag = agno + 1; ag <= maxag; ag++) {
1733 		AG_LOCK(imap, ag);
1734 
1735 		rc = diAllocAG(imap, ag, dir, ip);
1736 
1737 		AG_UNLOCK(imap, ag);
1738 
1739 		if (rc != -ENOSPC)
1740 			return (rc);
1741 	}
1742 
1743 	/* try to allocate from the ags in front of agno.
1744 	 */
1745 	for (ag = 0; ag < agno; ag++) {
1746 		AG_LOCK(imap, ag);
1747 
1748 		rc = diAllocAG(imap, ag, dir, ip);
1749 
1750 		AG_UNLOCK(imap, ag);
1751 
1752 		if (rc != -ENOSPC)
1753 			return (rc);
1754 	}
1755 
1756 	/* no free disk inodes.
1757 	 */
1758 	return -ENOSPC;
1759 }
1760 
1761 
1762 /*
1763  * NAME:	diAllocIno(imap,agno,ip)
1764  *
1765  * FUNCTION:	allocate a disk inode from the allocation group's free
1766  *		inode list, returning an error if this free list is
1767  *		empty (i.e. no iags on the list).
1768  *
1769  *		allocation occurs from the first iag on the list using
1770  *		the iag's free inode summary map to find the leftmost
1771  *		free inode in the iag.
1772  *
1773  * PRE CONDITION: Already have AG lock for this AG.
1774  *
1775  * PARAMETERS:
1776  *	imap	- pointer to inode map control structure.
1777  *	agno	- allocation group.
1778  *	ip	- pointer to new inode to be filled in on successful return
1779  *		  with the disk inode number allocated, its extent address
1780  *		  and the start of the ag.
1781  *
1782  * RETURN VALUES:
1783  *	0	- success.
1784  *	-ENOSPC	- insufficient disk resources.
1785  *	-EIO	- i/o error.
1786  */
1787 static int diAllocIno(struct inomap * imap, int agno, struct inode *ip)
1788 {
1789 	int iagno, ino, rc, rem, extno, sword;
1790 	struct metapage *mp;
1791 	struct iag *iagp;
1792 
1793 	/* check if there are iags on the ag's free inode list.
1794 	 */
1795 	if ((iagno = imap->im_agctl[agno].inofree) < 0)
1796 		return -ENOSPC;
1797 
1798 	/* obtain read lock on imap inode */
1799 	IREAD_LOCK(imap->im_ipimap, RDWRLOCK_IMAP);
1800 
1801 	/* read the iag at the head of the list.
1802 	 */
1803 	if ((rc = diIAGRead(imap, iagno, &mp))) {
1804 		IREAD_UNLOCK(imap->im_ipimap);
1805 		return (rc);
1806 	}
1807 	iagp = (struct iag *) mp->data;
1808 
1809 	/* better be free inodes in this iag if it is on the
1810 	 * list.
1811 	 */
1812 	if (!iagp->nfreeinos) {
1813 		IREAD_UNLOCK(imap->im_ipimap);
1814 		release_metapage(mp);
1815 		jfs_error(ip->i_sb,
1816 			  "diAllocIno: nfreeinos = 0, but iag on freelist");
1817 		return -EIO;
1818 	}
1819 
1820 	/* scan the free inode summary map to find an extent
1821 	 * with free inodes.
1822 	 */
1823 	for (sword = 0;; sword++) {
1824 		if (sword >= SMAPSZ) {
1825 			IREAD_UNLOCK(imap->im_ipimap);
1826 			release_metapage(mp);
1827 			jfs_error(ip->i_sb,
1828 				  "diAllocIno: free inode not found in summary map");
1829 			return -EIO;
1830 		}
1831 
1832 		if (~iagp->inosmap[sword])
1833 			break;
1834 	}
1835 
1836 	/* found a extent with free inodes. determine
1837 	 * the extent number.
1838 	 */
1839 	rem = diFindFree(le32_to_cpu(iagp->inosmap[sword]), 0);
1840 	if (rem >= EXTSPERSUM) {
1841 		IREAD_UNLOCK(imap->im_ipimap);
1842 		release_metapage(mp);
1843 		jfs_error(ip->i_sb, "diAllocIno: no free extent found");
1844 		return -EIO;
1845 	}
1846 	extno = (sword << L2EXTSPERSUM) + rem;
1847 
1848 	/* find the first free inode in the extent.
1849 	 */
1850 	rem = diFindFree(le32_to_cpu(iagp->wmap[extno]), 0);
1851 	if (rem >= INOSPEREXT) {
1852 		IREAD_UNLOCK(imap->im_ipimap);
1853 		release_metapage(mp);
1854 		jfs_error(ip->i_sb, "diAllocIno: free inode not found");
1855 		return -EIO;
1856 	}
1857 
1858 	/* compute the inode number within the iag.
1859 	 */
1860 	ino = (extno << L2INOSPEREXT) + rem;
1861 
1862 	/* allocate the inode.
1863 	 */
1864 	rc = diAllocBit(imap, iagp, ino);
1865 	IREAD_UNLOCK(imap->im_ipimap);
1866 	if (rc) {
1867 		release_metapage(mp);
1868 		return (rc);
1869 	}
1870 
1871 	/* set the results of the allocation and write the iag.
1872 	 */
1873 	diInitInode(ip, iagno, ino, extno, iagp);
1874 	write_metapage(mp);
1875 
1876 	return (0);
1877 }
1878 
1879 
1880 /*
1881  * NAME:	diAllocExt(imap,agno,ip)
1882  *
1883  * FUNCTION:	add a new extent of free inodes to an iag, allocating
1884  *		an inode from this extent to satisfy the current allocation
1885  *		request.
1886  *
1887  *		this routine first tries to find an existing iag with free
1888  *		extents through the ag free extent list.  if list is not
1889  *		empty, the head of the list will be selected as the home
1890  *		of the new extent of free inodes.  otherwise (the list is
1891  *		empty), a new iag will be allocated for the ag to contain
1892  *		the extent.
1893  *
1894  *		once an iag has been selected, the free extent summary map
1895  *		is used to locate a free extent within the iag and diNewExt()
1896  *		is called to initialize the extent, with initialization
1897  *		including the allocation of the first inode of the extent
1898  *		for the purpose of satisfying this request.
1899  *
1900  * PARAMETERS:
1901  *	imap	- pointer to inode map control structure.
1902  *	agno	- allocation group number.
1903  *	ip	- pointer to new inode to be filled in on successful return
1904  *		  with the disk inode number allocated, its extent address
1905  *		  and the start of the ag.
1906  *
1907  * RETURN VALUES:
1908  *	0	- success.
1909  *	-ENOSPC	- insufficient disk resources.
1910  *	-EIO	- i/o error.
1911  */
1912 static int diAllocExt(struct inomap * imap, int agno, struct inode *ip)
1913 {
1914 	int rem, iagno, sword, extno, rc;
1915 	struct metapage *mp;
1916 	struct iag *iagp;
1917 
1918 	/* check if the ag has any iags with free extents.  if not,
1919 	 * allocate a new iag for the ag.
1920 	 */
1921 	if ((iagno = imap->im_agctl[agno].extfree) < 0) {
1922 		/* If successful, diNewIAG will obtain the read lock on the
1923 		 * imap inode.
1924 		 */
1925 		if ((rc = diNewIAG(imap, &iagno, agno, &mp))) {
1926 			return (rc);
1927 		}
1928 		iagp = (struct iag *) mp->data;
1929 
1930 		/* set the ag number if this a brand new iag
1931 		 */
1932 		iagp->agstart =
1933 		    cpu_to_le64(AGTOBLK(agno, imap->im_ipimap));
1934 	} else {
1935 		/* read the iag.
1936 		 */
1937 		IREAD_LOCK(imap->im_ipimap, RDWRLOCK_IMAP);
1938 		if ((rc = diIAGRead(imap, iagno, &mp))) {
1939 			IREAD_UNLOCK(imap->im_ipimap);
1940 			jfs_error(ip->i_sb, "diAllocExt: error reading iag");
1941 			return rc;
1942 		}
1943 		iagp = (struct iag *) mp->data;
1944 	}
1945 
1946 	/* using the free extent summary map, find a free extent.
1947 	 */
1948 	for (sword = 0;; sword++) {
1949 		if (sword >= SMAPSZ) {
1950 			release_metapage(mp);
1951 			IREAD_UNLOCK(imap->im_ipimap);
1952 			jfs_error(ip->i_sb,
1953 				  "diAllocExt: free ext summary map not found");
1954 			return -EIO;
1955 		}
1956 		if (~iagp->extsmap[sword])
1957 			break;
1958 	}
1959 
1960 	/* determine the extent number of the free extent.
1961 	 */
1962 	rem = diFindFree(le32_to_cpu(iagp->extsmap[sword]), 0);
1963 	if (rem >= EXTSPERSUM) {
1964 		release_metapage(mp);
1965 		IREAD_UNLOCK(imap->im_ipimap);
1966 		jfs_error(ip->i_sb, "diAllocExt: free extent not found");
1967 		return -EIO;
1968 	}
1969 	extno = (sword << L2EXTSPERSUM) + rem;
1970 
1971 	/* initialize the new extent.
1972 	 */
1973 	rc = diNewExt(imap, iagp, extno);
1974 	IREAD_UNLOCK(imap->im_ipimap);
1975 	if (rc) {
1976 		/* something bad happened.  if a new iag was allocated,
1977 		 * place it back on the inode map's iag free list, and
1978 		 * clear the ag number information.
1979 		 */
1980 		if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
1981 			IAGFREE_LOCK(imap);
1982 			iagp->iagfree = cpu_to_le32(imap->im_freeiag);
1983 			imap->im_freeiag = iagno;
1984 			IAGFREE_UNLOCK(imap);
1985 		}
1986 		write_metapage(mp);
1987 		return (rc);
1988 	}
1989 
1990 	/* set the results of the allocation and write the iag.
1991 	 */
1992 	diInitInode(ip, iagno, extno << L2INOSPEREXT, extno, iagp);
1993 
1994 	write_metapage(mp);
1995 
1996 	return (0);
1997 }
1998 
1999 
2000 /*
2001  * NAME:	diAllocBit(imap,iagp,ino)
2002  *
2003  * FUNCTION:	allocate a backed inode from an iag.
2004  *
2005  *		this routine performs the mechanics of allocating a
2006  *		specified inode from a backed extent.
2007  *
2008  *		if the inode to be allocated represents the last free
2009  *		inode within the iag, the iag will be removed from the
2010  *		ag free inode list.
2011  *
2012  *		a careful update approach is used to provide consistency
2013  *		in the face of updates to multiple buffers.  under this
2014  *		approach, all required buffers are obtained before making
2015  *		any updates and are held all are updates are complete.
2016  *
2017  * PRE CONDITION: Already have buffer lock on iagp.  Already have AG lock on
2018  *	this AG.  Must have read lock on imap inode.
2019  *
2020  * PARAMETERS:
2021  *	imap	- pointer to inode map control structure.
2022  *	iagp	- pointer to iag.
2023  *	ino	- inode number to be allocated within the iag.
2024  *
2025  * RETURN VALUES:
2026  *	0	- success.
2027  *	-ENOSPC	- insufficient disk resources.
2028  *	-EIO	- i/o error.
2029  */
2030 static int diAllocBit(struct inomap * imap, struct iag * iagp, int ino)
2031 {
2032 	int extno, bitno, agno, sword, rc;
2033 	struct metapage *amp = NULL, *bmp = NULL;
2034 	struct iag *aiagp = NULL, *biagp = NULL;
2035 	u32 mask;
2036 
2037 	/* check if this is the last free inode within the iag.
2038 	 * if so, it will have to be removed from the ag free
2039 	 * inode list, so get the iags preceeding and following
2040 	 * it on the list.
2041 	 */
2042 	if (iagp->nfreeinos == cpu_to_le32(1)) {
2043 		if ((int) le32_to_cpu(iagp->inofreefwd) >= 0) {
2044 			if ((rc =
2045 			     diIAGRead(imap, le32_to_cpu(iagp->inofreefwd),
2046 				       &amp)))
2047 				return (rc);
2048 			aiagp = (struct iag *) amp->data;
2049 		}
2050 
2051 		if ((int) le32_to_cpu(iagp->inofreeback) >= 0) {
2052 			if ((rc =
2053 			     diIAGRead(imap,
2054 				       le32_to_cpu(iagp->inofreeback),
2055 				       &bmp))) {
2056 				if (amp)
2057 					release_metapage(amp);
2058 				return (rc);
2059 			}
2060 			biagp = (struct iag *) bmp->data;
2061 		}
2062 	}
2063 
2064 	/* get the ag number, extent number, inode number within
2065 	 * the extent.
2066 	 */
2067 	agno = BLKTOAG(le64_to_cpu(iagp->agstart), JFS_SBI(imap->im_ipimap->i_sb));
2068 	extno = ino >> L2INOSPEREXT;
2069 	bitno = ino & (INOSPEREXT - 1);
2070 
2071 	/* compute the mask for setting the map.
2072 	 */
2073 	mask = HIGHORDER >> bitno;
2074 
2075 	/* the inode should be free and backed.
2076 	 */
2077 	if (((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) ||
2078 	    ((le32_to_cpu(iagp->wmap[extno]) & mask) != 0) ||
2079 	    (addressPXD(&iagp->inoext[extno]) == 0)) {
2080 		if (amp)
2081 			release_metapage(amp);
2082 		if (bmp)
2083 			release_metapage(bmp);
2084 
2085 		jfs_error(imap->im_ipimap->i_sb,
2086 			  "diAllocBit: iag inconsistent");
2087 		return -EIO;
2088 	}
2089 
2090 	/* mark the inode as allocated in the working map.
2091 	 */
2092 	iagp->wmap[extno] |= cpu_to_le32(mask);
2093 
2094 	/* check if all inodes within the extent are now
2095 	 * allocated.  if so, update the free inode summary
2096 	 * map to reflect this.
2097 	 */
2098 	if (iagp->wmap[extno] == cpu_to_le32(ONES)) {
2099 		sword = extno >> L2EXTSPERSUM;
2100 		bitno = extno & (EXTSPERSUM - 1);
2101 		iagp->inosmap[sword] |= cpu_to_le32(HIGHORDER >> bitno);
2102 	}
2103 
2104 	/* if this was the last free inode in the iag, remove the
2105 	 * iag from the ag free inode list.
2106 	 */
2107 	if (iagp->nfreeinos == cpu_to_le32(1)) {
2108 		if (amp) {
2109 			aiagp->inofreeback = iagp->inofreeback;
2110 			write_metapage(amp);
2111 		}
2112 
2113 		if (bmp) {
2114 			biagp->inofreefwd = iagp->inofreefwd;
2115 			write_metapage(bmp);
2116 		} else {
2117 			imap->im_agctl[agno].inofree =
2118 			    le32_to_cpu(iagp->inofreefwd);
2119 		}
2120 		iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1);
2121 	}
2122 
2123 	/* update the free inode count at the iag, ag, inode
2124 	 * map levels.
2125 	 */
2126 	le32_add_cpu(&iagp->nfreeinos, -1);
2127 	imap->im_agctl[agno].numfree -= 1;
2128 	atomic_dec(&imap->im_numfree);
2129 
2130 	return (0);
2131 }
2132 
2133 
2134 /*
2135  * NAME:	diNewExt(imap,iagp,extno)
2136  *
2137  * FUNCTION:	initialize a new extent of inodes for an iag, allocating
2138  *		the first inode of the extent for use for the current
2139  *		allocation request.
2140  *
2141  *		disk resources are allocated for the new extent of inodes
2142  *		and the inodes themselves are initialized to reflect their
2143  *		existence within the extent (i.e. their inode numbers and
2144  *		inode extent addresses are set) and their initial state
2145  *		(mode and link count are set to zero).
2146  *
2147  *		if the iag is new, it is not yet on an ag extent free list
2148  *		but will now be placed on this list.
2149  *
2150  *		if the allocation of the new extent causes the iag to
2151  *		have no free extent, the iag will be removed from the
2152  *		ag extent free list.
2153  *
2154  *		if the iag has no free backed inodes, it will be placed
2155  *		on the ag free inode list, since the addition of the new
2156  *		extent will now cause it to have free inodes.
2157  *
2158  *		a careful update approach is used to provide consistency
2159  *		(i.e. list consistency) in the face of updates to multiple
2160  *		buffers.  under this approach, all required buffers are
2161  *		obtained before making any updates and are held until all
2162  *		updates are complete.
2163  *
2164  * PRE CONDITION: Already have buffer lock on iagp.  Already have AG lock on
2165  *	this AG.  Must have read lock on imap inode.
2166  *
2167  * PARAMETERS:
2168  *	imap	- pointer to inode map control structure.
2169  *	iagp	- pointer to iag.
2170  *	extno	- extent number.
2171  *
2172  * RETURN VALUES:
2173  *	0	- success.
2174  *	-ENOSPC	- insufficient disk resources.
2175  *	-EIO	- i/o error.
2176  */
2177 static int diNewExt(struct inomap * imap, struct iag * iagp, int extno)
2178 {
2179 	int agno, iagno, fwd, back, freei = 0, sword, rc;
2180 	struct iag *aiagp = NULL, *biagp = NULL, *ciagp = NULL;
2181 	struct metapage *amp, *bmp, *cmp, *dmp;
2182 	struct inode *ipimap;
2183 	s64 blkno, hint;
2184 	int i, j;
2185 	u32 mask;
2186 	ino_t ino;
2187 	struct dinode *dp;
2188 	struct jfs_sb_info *sbi;
2189 
2190 	/* better have free extents.
2191 	 */
2192 	if (!iagp->nfreeexts) {
2193 		jfs_error(imap->im_ipimap->i_sb, "diNewExt: no free extents");
2194 		return -EIO;
2195 	}
2196 
2197 	/* get the inode map inode.
2198 	 */
2199 	ipimap = imap->im_ipimap;
2200 	sbi = JFS_SBI(ipimap->i_sb);
2201 
2202 	amp = bmp = cmp = NULL;
2203 
2204 	/* get the ag and iag numbers for this iag.
2205 	 */
2206 	agno = BLKTOAG(le64_to_cpu(iagp->agstart), sbi);
2207 	iagno = le32_to_cpu(iagp->iagnum);
2208 
2209 	/* check if this is the last free extent within the
2210 	 * iag.  if so, the iag must be removed from the ag
2211 	 * free extent list, so get the iags preceeding and
2212 	 * following the iag on this list.
2213 	 */
2214 	if (iagp->nfreeexts == cpu_to_le32(1)) {
2215 		if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) {
2216 			if ((rc = diIAGRead(imap, fwd, &amp)))
2217 				return (rc);
2218 			aiagp = (struct iag *) amp->data;
2219 		}
2220 
2221 		if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) {
2222 			if ((rc = diIAGRead(imap, back, &bmp)))
2223 				goto error_out;
2224 			biagp = (struct iag *) bmp->data;
2225 		}
2226 	} else {
2227 		/* the iag has free extents.  if all extents are free
2228 		 * (as is the case for a newly allocated iag), the iag
2229 		 * must be added to the ag free extent list, so get
2230 		 * the iag at the head of the list in preparation for
2231 		 * adding this iag to this list.
2232 		 */
2233 		fwd = back = -1;
2234 		if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2235 			if ((fwd = imap->im_agctl[agno].extfree) >= 0) {
2236 				if ((rc = diIAGRead(imap, fwd, &amp)))
2237 					goto error_out;
2238 				aiagp = (struct iag *) amp->data;
2239 			}
2240 		}
2241 	}
2242 
2243 	/* check if the iag has no free inodes.  if so, the iag
2244 	 * will have to be added to the ag free inode list, so get
2245 	 * the iag at the head of the list in preparation for
2246 	 * adding this iag to this list.  in doing this, we must
2247 	 * check if we already have the iag at the head of
2248 	 * the list in hand.
2249 	 */
2250 	if (iagp->nfreeinos == 0) {
2251 		freei = imap->im_agctl[agno].inofree;
2252 
2253 		if (freei >= 0) {
2254 			if (freei == fwd) {
2255 				ciagp = aiagp;
2256 			} else if (freei == back) {
2257 				ciagp = biagp;
2258 			} else {
2259 				if ((rc = diIAGRead(imap, freei, &cmp)))
2260 					goto error_out;
2261 				ciagp = (struct iag *) cmp->data;
2262 			}
2263 			if (ciagp == NULL) {
2264 				jfs_error(imap->im_ipimap->i_sb,
2265 					  "diNewExt: ciagp == NULL");
2266 				rc = -EIO;
2267 				goto error_out;
2268 			}
2269 		}
2270 	}
2271 
2272 	/* allocate disk space for the inode extent.
2273 	 */
2274 	if ((extno == 0) || (addressPXD(&iagp->inoext[extno - 1]) == 0))
2275 		hint = ((s64) agno << sbi->bmap->db_agl2size) - 1;
2276 	else
2277 		hint = addressPXD(&iagp->inoext[extno - 1]) +
2278 		    lengthPXD(&iagp->inoext[extno - 1]) - 1;
2279 
2280 	if ((rc = dbAlloc(ipimap, hint, (s64) imap->im_nbperiext, &blkno)))
2281 		goto error_out;
2282 
2283 	/* compute the inode number of the first inode within the
2284 	 * extent.
2285 	 */
2286 	ino = (iagno << L2INOSPERIAG) + (extno << L2INOSPEREXT);
2287 
2288 	/* initialize the inodes within the newly allocated extent a
2289 	 * page at a time.
2290 	 */
2291 	for (i = 0; i < imap->im_nbperiext; i += sbi->nbperpage) {
2292 		/* get a buffer for this page of disk inodes.
2293 		 */
2294 		dmp = get_metapage(ipimap, blkno + i, PSIZE, 1);
2295 		if (dmp == NULL) {
2296 			rc = -EIO;
2297 			goto error_out;
2298 		}
2299 		dp = (struct dinode *) dmp->data;
2300 
2301 		/* initialize the inode number, mode, link count and
2302 		 * inode extent address.
2303 		 */
2304 		for (j = 0; j < INOSPERPAGE; j++, dp++, ino++) {
2305 			dp->di_inostamp = cpu_to_le32(sbi->inostamp);
2306 			dp->di_number = cpu_to_le32(ino);
2307 			dp->di_fileset = cpu_to_le32(FILESYSTEM_I);
2308 			dp->di_mode = 0;
2309 			dp->di_nlink = 0;
2310 			PXDaddress(&(dp->di_ixpxd), blkno);
2311 			PXDlength(&(dp->di_ixpxd), imap->im_nbperiext);
2312 		}
2313 		write_metapage(dmp);
2314 	}
2315 
2316 	/* if this is the last free extent within the iag, remove the
2317 	 * iag from the ag free extent list.
2318 	 */
2319 	if (iagp->nfreeexts == cpu_to_le32(1)) {
2320 		if (fwd >= 0)
2321 			aiagp->extfreeback = iagp->extfreeback;
2322 
2323 		if (back >= 0)
2324 			biagp->extfreefwd = iagp->extfreefwd;
2325 		else
2326 			imap->im_agctl[agno].extfree =
2327 			    le32_to_cpu(iagp->extfreefwd);
2328 
2329 		iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1);
2330 	} else {
2331 		/* if the iag has all free extents (newly allocated iag),
2332 		 * add the iag to the ag free extent list.
2333 		 */
2334 		if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2335 			if (fwd >= 0)
2336 				aiagp->extfreeback = cpu_to_le32(iagno);
2337 
2338 			iagp->extfreefwd = cpu_to_le32(fwd);
2339 			iagp->extfreeback = cpu_to_le32(-1);
2340 			imap->im_agctl[agno].extfree = iagno;
2341 		}
2342 	}
2343 
2344 	/* if the iag has no free inodes, add the iag to the
2345 	 * ag free inode list.
2346 	 */
2347 	if (iagp->nfreeinos == 0) {
2348 		if (freei >= 0)
2349 			ciagp->inofreeback = cpu_to_le32(iagno);
2350 
2351 		iagp->inofreefwd =
2352 		    cpu_to_le32(imap->im_agctl[agno].inofree);
2353 		iagp->inofreeback = cpu_to_le32(-1);
2354 		imap->im_agctl[agno].inofree = iagno;
2355 	}
2356 
2357 	/* initialize the extent descriptor of the extent. */
2358 	PXDlength(&iagp->inoext[extno], imap->im_nbperiext);
2359 	PXDaddress(&iagp->inoext[extno], blkno);
2360 
2361 	/* initialize the working and persistent map of the extent.
2362 	 * the working map will be initialized such that
2363 	 * it indicates the first inode of the extent is allocated.
2364 	 */
2365 	iagp->wmap[extno] = cpu_to_le32(HIGHORDER);
2366 	iagp->pmap[extno] = 0;
2367 
2368 	/* update the free inode and free extent summary maps
2369 	 * for the extent to indicate the extent has free inodes
2370 	 * and no longer represents a free extent.
2371 	 */
2372 	sword = extno >> L2EXTSPERSUM;
2373 	mask = HIGHORDER >> (extno & (EXTSPERSUM - 1));
2374 	iagp->extsmap[sword] |= cpu_to_le32(mask);
2375 	iagp->inosmap[sword] &= cpu_to_le32(~mask);
2376 
2377 	/* update the free inode and free extent counts for the
2378 	 * iag.
2379 	 */
2380 	le32_add_cpu(&iagp->nfreeinos, (INOSPEREXT - 1));
2381 	le32_add_cpu(&iagp->nfreeexts, -1);
2382 
2383 	/* update the free and backed inode counts for the ag.
2384 	 */
2385 	imap->im_agctl[agno].numfree += (INOSPEREXT - 1);
2386 	imap->im_agctl[agno].numinos += INOSPEREXT;
2387 
2388 	/* update the free and backed inode counts for the inode map.
2389 	 */
2390 	atomic_add(INOSPEREXT - 1, &imap->im_numfree);
2391 	atomic_add(INOSPEREXT, &imap->im_numinos);
2392 
2393 	/* write the iags.
2394 	 */
2395 	if (amp)
2396 		write_metapage(amp);
2397 	if (bmp)
2398 		write_metapage(bmp);
2399 	if (cmp)
2400 		write_metapage(cmp);
2401 
2402 	return (0);
2403 
2404       error_out:
2405 
2406 	/* release the iags.
2407 	 */
2408 	if (amp)
2409 		release_metapage(amp);
2410 	if (bmp)
2411 		release_metapage(bmp);
2412 	if (cmp)
2413 		release_metapage(cmp);
2414 
2415 	return (rc);
2416 }
2417 
2418 
2419 /*
2420  * NAME:	diNewIAG(imap,iagnop,agno)
2421  *
2422  * FUNCTION:	allocate a new iag for an allocation group.
2423  *
2424  *		first tries to allocate the iag from the inode map
2425  *		iagfree list:
2426  *		if the list has free iags, the head of the list is removed
2427  *		and returned to satisfy the request.
2428  *		if the inode map's iag free list is empty, the inode map
2429  *		is extended to hold a new iag. this new iag is initialized
2430  *		and returned to satisfy the request.
2431  *
2432  * PARAMETERS:
2433  *	imap	- pointer to inode map control structure.
2434  *	iagnop	- pointer to an iag number set with the number of the
2435  *		  newly allocated iag upon successful return.
2436  *	agno	- allocation group number.
2437  *	bpp	- Buffer pointer to be filled in with new IAG's buffer
2438  *
2439  * RETURN VALUES:
2440  *	0	- success.
2441  *	-ENOSPC	- insufficient disk resources.
2442  *	-EIO	- i/o error.
2443  *
2444  * serialization:
2445  *	AG lock held on entry/exit;
2446  *	write lock on the map is held inside;
2447  *	read lock on the map is held on successful completion;
2448  *
2449  * note: new iag transaction:
2450  * . synchronously write iag;
2451  * . write log of xtree and inode of imap;
2452  * . commit;
2453  * . synchronous write of xtree (right to left, bottom to top);
2454  * . at start of logredo(): init in-memory imap with one additional iag page;
2455  * . at end of logredo(): re-read imap inode to determine
2456  *   new imap size;
2457  */
2458 static int
2459 diNewIAG(struct inomap * imap, int *iagnop, int agno, struct metapage ** mpp)
2460 {
2461 	int rc;
2462 	int iagno, i, xlen;
2463 	struct inode *ipimap;
2464 	struct super_block *sb;
2465 	struct jfs_sb_info *sbi;
2466 	struct metapage *mp;
2467 	struct iag *iagp;
2468 	s64 xaddr = 0;
2469 	s64 blkno;
2470 	tid_t tid;
2471 	struct inode *iplist[1];
2472 
2473 	/* pick up pointers to the inode map and mount inodes */
2474 	ipimap = imap->im_ipimap;
2475 	sb = ipimap->i_sb;
2476 	sbi = JFS_SBI(sb);
2477 
2478 	/* acquire the free iag lock */
2479 	IAGFREE_LOCK(imap);
2480 
2481 	/* if there are any iags on the inode map free iag list,
2482 	 * allocate the iag from the head of the list.
2483 	 */
2484 	if (imap->im_freeiag >= 0) {
2485 		/* pick up the iag number at the head of the list */
2486 		iagno = imap->im_freeiag;
2487 
2488 		/* determine the logical block number of the iag */
2489 		blkno = IAGTOLBLK(iagno, sbi->l2nbperpage);
2490 	} else {
2491 		/* no free iags. the inode map will have to be extented
2492 		 * to include a new iag.
2493 		 */
2494 
2495 		/* acquire inode map lock */
2496 		IWRITE_LOCK(ipimap, RDWRLOCK_IMAP);
2497 
2498 		if (ipimap->i_size >> L2PSIZE != imap->im_nextiag + 1) {
2499 			IWRITE_UNLOCK(ipimap);
2500 			IAGFREE_UNLOCK(imap);
2501 			jfs_error(imap->im_ipimap->i_sb,
2502 				  "diNewIAG: ipimap->i_size is wrong");
2503 			return -EIO;
2504 		}
2505 
2506 
2507 		/* get the next avaliable iag number */
2508 		iagno = imap->im_nextiag;
2509 
2510 		/* make sure that we have not exceeded the maximum inode
2511 		 * number limit.
2512 		 */
2513 		if (iagno > (MAXIAGS - 1)) {
2514 			/* release the inode map lock */
2515 			IWRITE_UNLOCK(ipimap);
2516 
2517 			rc = -ENOSPC;
2518 			goto out;
2519 		}
2520 
2521 		/*
2522 		 * synchronously append new iag page.
2523 		 */
2524 		/* determine the logical address of iag page to append */
2525 		blkno = IAGTOLBLK(iagno, sbi->l2nbperpage);
2526 
2527 		/* Allocate extent for new iag page */
2528 		xlen = sbi->nbperpage;
2529 		if ((rc = dbAlloc(ipimap, 0, (s64) xlen, &xaddr))) {
2530 			/* release the inode map lock */
2531 			IWRITE_UNLOCK(ipimap);
2532 
2533 			goto out;
2534 		}
2535 
2536 		/*
2537 		 * start transaction of update of the inode map
2538 		 * addressing structure pointing to the new iag page;
2539 		 */
2540 		tid = txBegin(sb, COMMIT_FORCE);
2541 		mutex_lock(&JFS_IP(ipimap)->commit_mutex);
2542 
2543 		/* update the inode map addressing structure to point to it */
2544 		if ((rc =
2545 		     xtInsert(tid, ipimap, 0, blkno, xlen, &xaddr, 0))) {
2546 			txEnd(tid);
2547 			mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
2548 			/* Free the blocks allocated for the iag since it was
2549 			 * not successfully added to the inode map
2550 			 */
2551 			dbFree(ipimap, xaddr, (s64) xlen);
2552 
2553 			/* release the inode map lock */
2554 			IWRITE_UNLOCK(ipimap);
2555 
2556 			goto out;
2557 		}
2558 
2559 		/* update the inode map's inode to reflect the extension */
2560 		ipimap->i_size += PSIZE;
2561 		inode_add_bytes(ipimap, PSIZE);
2562 
2563 		/* assign a buffer for the page */
2564 		mp = get_metapage(ipimap, blkno, PSIZE, 0);
2565 		if (!mp) {
2566 			/*
2567 			 * This is very unlikely since we just created the
2568 			 * extent, but let's try to handle it correctly
2569 			 */
2570 			xtTruncate(tid, ipimap, ipimap->i_size - PSIZE,
2571 				   COMMIT_PWMAP);
2572 
2573 			txAbort(tid, 0);
2574 			txEnd(tid);
2575 			mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
2576 
2577 			/* release the inode map lock */
2578 			IWRITE_UNLOCK(ipimap);
2579 
2580 			rc = -EIO;
2581 			goto out;
2582 		}
2583 		iagp = (struct iag *) mp->data;
2584 
2585 		/* init the iag */
2586 		memset(iagp, 0, sizeof(struct iag));
2587 		iagp->iagnum = cpu_to_le32(iagno);
2588 		iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1);
2589 		iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1);
2590 		iagp->iagfree = cpu_to_le32(-1);
2591 		iagp->nfreeinos = 0;
2592 		iagp->nfreeexts = cpu_to_le32(EXTSPERIAG);
2593 
2594 		/* initialize the free inode summary map (free extent
2595 		 * summary map initialization handled by bzero).
2596 		 */
2597 		for (i = 0; i < SMAPSZ; i++)
2598 			iagp->inosmap[i] = cpu_to_le32(ONES);
2599 
2600 		/*
2601 		 * Write and sync the metapage
2602 		 */
2603 		flush_metapage(mp);
2604 
2605 		/*
2606 		 * txCommit(COMMIT_FORCE) will synchronously write address
2607 		 * index pages and inode after commit in careful update order
2608 		 * of address index pages (right to left, bottom up);
2609 		 */
2610 		iplist[0] = ipimap;
2611 		rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE);
2612 
2613 		txEnd(tid);
2614 		mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
2615 
2616 		duplicateIXtree(sb, blkno, xlen, &xaddr);
2617 
2618 		/* update the next avaliable iag number */
2619 		imap->im_nextiag += 1;
2620 
2621 		/* Add the iag to the iag free list so we don't lose the iag
2622 		 * if a failure happens now.
2623 		 */
2624 		imap->im_freeiag = iagno;
2625 
2626 		/* Until we have logredo working, we want the imap inode &
2627 		 * control page to be up to date.
2628 		 */
2629 		diSync(ipimap);
2630 
2631 		/* release the inode map lock */
2632 		IWRITE_UNLOCK(ipimap);
2633 	}
2634 
2635 	/* obtain read lock on map */
2636 	IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
2637 
2638 	/* read the iag */
2639 	if ((rc = diIAGRead(imap, iagno, &mp))) {
2640 		IREAD_UNLOCK(ipimap);
2641 		rc = -EIO;
2642 		goto out;
2643 	}
2644 	iagp = (struct iag *) mp->data;
2645 
2646 	/* remove the iag from the iag free list */
2647 	imap->im_freeiag = le32_to_cpu(iagp->iagfree);
2648 	iagp->iagfree = cpu_to_le32(-1);
2649 
2650 	/* set the return iag number and buffer pointer */
2651 	*iagnop = iagno;
2652 	*mpp = mp;
2653 
2654       out:
2655 	/* release the iag free lock */
2656 	IAGFREE_UNLOCK(imap);
2657 
2658 	return (rc);
2659 }
2660 
2661 /*
2662  * NAME:	diIAGRead()
2663  *
2664  * FUNCTION:	get the buffer for the specified iag within a fileset
2665  *		or aggregate inode map.
2666  *
2667  * PARAMETERS:
2668  *	imap	- pointer to inode map control structure.
2669  *	iagno	- iag number.
2670  *	bpp	- point to buffer pointer to be filled in on successful
2671  *		  exit.
2672  *
2673  * SERIALIZATION:
2674  *	must have read lock on imap inode
2675  *	(When called by diExtendFS, the filesystem is quiesced, therefore
2676  *	 the read lock is unnecessary.)
2677  *
2678  * RETURN VALUES:
2679  *	0	- success.
2680  *	-EIO	- i/o error.
2681  */
2682 static int diIAGRead(struct inomap * imap, int iagno, struct metapage ** mpp)
2683 {
2684 	struct inode *ipimap = imap->im_ipimap;
2685 	s64 blkno;
2686 
2687 	/* compute the logical block number of the iag. */
2688 	blkno = IAGTOLBLK(iagno, JFS_SBI(ipimap->i_sb)->l2nbperpage);
2689 
2690 	/* read the iag. */
2691 	*mpp = read_metapage(ipimap, blkno, PSIZE, 0);
2692 	if (*mpp == NULL) {
2693 		return -EIO;
2694 	}
2695 
2696 	return (0);
2697 }
2698 
2699 /*
2700  * NAME:	diFindFree()
2701  *
2702  * FUNCTION:	find the first free bit in a word starting at
2703  *		the specified bit position.
2704  *
2705  * PARAMETERS:
2706  *	word	- word to be examined.
2707  *	start	- starting bit position.
2708  *
2709  * RETURN VALUES:
2710  *	bit position of first free bit in the word or 32 if
2711  *	no free bits were found.
2712  */
2713 static int diFindFree(u32 word, int start)
2714 {
2715 	int bitno;
2716 	assert(start < 32);
2717 	/* scan the word for the first free bit. */
2718 	for (word <<= start, bitno = start; bitno < 32;
2719 	     bitno++, word <<= 1) {
2720 		if ((word & HIGHORDER) == 0)
2721 			break;
2722 	}
2723 	return (bitno);
2724 }
2725 
2726 /*
2727  * NAME:	diUpdatePMap()
2728  *
2729  * FUNCTION: Update the persistent map in an IAG for the allocation or
2730  *	freeing of the specified inode.
2731  *
2732  * PRE CONDITIONS: Working map has already been updated for allocate.
2733  *
2734  * PARAMETERS:
2735  *	ipimap	- Incore inode map inode
2736  *	inum	- Number of inode to mark in permanent map
2737  *	is_free	- If 'true' indicates inode should be marked freed, otherwise
2738  *		  indicates inode should be marked allocated.
2739  *
2740  * RETURN VALUES:
2741  *		0 for success
2742  */
2743 int
2744 diUpdatePMap(struct inode *ipimap,
2745 	     unsigned long inum, bool is_free, struct tblock * tblk)
2746 {
2747 	int rc;
2748 	struct iag *iagp;
2749 	struct metapage *mp;
2750 	int iagno, ino, extno, bitno;
2751 	struct inomap *imap;
2752 	u32 mask;
2753 	struct jfs_log *log;
2754 	int lsn, difft, diffp;
2755 	unsigned long flags;
2756 
2757 	imap = JFS_IP(ipimap)->i_imap;
2758 	/* get the iag number containing the inode */
2759 	iagno = INOTOIAG(inum);
2760 	/* make sure that the iag is contained within the map */
2761 	if (iagno >= imap->im_nextiag) {
2762 		jfs_error(ipimap->i_sb,
2763 			  "diUpdatePMap: the iag is outside the map");
2764 		return -EIO;
2765 	}
2766 	/* read the iag */
2767 	IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
2768 	rc = diIAGRead(imap, iagno, &mp);
2769 	IREAD_UNLOCK(ipimap);
2770 	if (rc)
2771 		return (rc);
2772 	metapage_wait_for_io(mp);
2773 	iagp = (struct iag *) mp->data;
2774 	/* get the inode number and extent number of the inode within
2775 	 * the iag and the inode number within the extent.
2776 	 */
2777 	ino = inum & (INOSPERIAG - 1);
2778 	extno = ino >> L2INOSPEREXT;
2779 	bitno = ino & (INOSPEREXT - 1);
2780 	mask = HIGHORDER >> bitno;
2781 	/*
2782 	 * mark the inode free in persistent map:
2783 	 */
2784 	if (is_free) {
2785 		/* The inode should have been allocated both in working
2786 		 * map and in persistent map;
2787 		 * the inode will be freed from working map at the release
2788 		 * of last reference release;
2789 		 */
2790 		if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
2791 			jfs_error(ipimap->i_sb,
2792 				  "diUpdatePMap: inode %ld not marked as "
2793 				  "allocated in wmap!", inum);
2794 		}
2795 		if (!(le32_to_cpu(iagp->pmap[extno]) & mask)) {
2796 			jfs_error(ipimap->i_sb,
2797 				  "diUpdatePMap: inode %ld not marked as "
2798 				  "allocated in pmap!", inum);
2799 		}
2800 		/* update the bitmap for the extent of the freed inode */
2801 		iagp->pmap[extno] &= cpu_to_le32(~mask);
2802 	}
2803 	/*
2804 	 * mark the inode allocated in persistent map:
2805 	 */
2806 	else {
2807 		/* The inode should be already allocated in the working map
2808 		 * and should be free in persistent map;
2809 		 */
2810 		if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
2811 			release_metapage(mp);
2812 			jfs_error(ipimap->i_sb,
2813 				  "diUpdatePMap: the inode is not allocated in "
2814 				  "the working map");
2815 			return -EIO;
2816 		}
2817 		if ((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) {
2818 			release_metapage(mp);
2819 			jfs_error(ipimap->i_sb,
2820 				  "diUpdatePMap: the inode is not free in the "
2821 				  "persistent map");
2822 			return -EIO;
2823 		}
2824 		/* update the bitmap for the extent of the allocated inode */
2825 		iagp->pmap[extno] |= cpu_to_le32(mask);
2826 	}
2827 	/*
2828 	 * update iag lsn
2829 	 */
2830 	lsn = tblk->lsn;
2831 	log = JFS_SBI(tblk->sb)->log;
2832 	LOGSYNC_LOCK(log, flags);
2833 	if (mp->lsn != 0) {
2834 		/* inherit older/smaller lsn */
2835 		logdiff(difft, lsn, log);
2836 		logdiff(diffp, mp->lsn, log);
2837 		if (difft < diffp) {
2838 			mp->lsn = lsn;
2839 			/* move mp after tblock in logsync list */
2840 			list_move(&mp->synclist, &tblk->synclist);
2841 		}
2842 		/* inherit younger/larger clsn */
2843 		assert(mp->clsn);
2844 		logdiff(difft, tblk->clsn, log);
2845 		logdiff(diffp, mp->clsn, log);
2846 		if (difft > diffp)
2847 			mp->clsn = tblk->clsn;
2848 	} else {
2849 		mp->log = log;
2850 		mp->lsn = lsn;
2851 		/* insert mp after tblock in logsync list */
2852 		log->count++;
2853 		list_add(&mp->synclist, &tblk->synclist);
2854 		mp->clsn = tblk->clsn;
2855 	}
2856 	LOGSYNC_UNLOCK(log, flags);
2857 	write_metapage(mp);
2858 	return (0);
2859 }
2860 
2861 /*
2862  *	diExtendFS()
2863  *
2864  * function: update imap for extendfs();
2865  *
2866  * note: AG size has been increased s.t. each k old contiguous AGs are
2867  * coalesced into a new AG;
2868  */
2869 int diExtendFS(struct inode *ipimap, struct inode *ipbmap)
2870 {
2871 	int rc, rcx = 0;
2872 	struct inomap *imap = JFS_IP(ipimap)->i_imap;
2873 	struct iag *iagp = NULL, *hiagp = NULL;
2874 	struct bmap *mp = JFS_SBI(ipbmap->i_sb)->bmap;
2875 	struct metapage *bp, *hbp;
2876 	int i, n, head;
2877 	int numinos, xnuminos = 0, xnumfree = 0;
2878 	s64 agstart;
2879 
2880 	jfs_info("diExtendFS: nextiag:%d numinos:%d numfree:%d",
2881 		   imap->im_nextiag, atomic_read(&imap->im_numinos),
2882 		   atomic_read(&imap->im_numfree));
2883 
2884 	/*
2885 	 *	reconstruct imap
2886 	 *
2887 	 * coalesce contiguous k (newAGSize/oldAGSize) AGs;
2888 	 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn;
2889 	 * note: new AG size = old AG size * (2**x).
2890 	 */
2891 
2892 	/* init per AG control information im_agctl[] */
2893 	for (i = 0; i < MAXAG; i++) {
2894 		imap->im_agctl[i].inofree = -1;
2895 		imap->im_agctl[i].extfree = -1;
2896 		imap->im_agctl[i].numinos = 0;	/* number of backed inodes */
2897 		imap->im_agctl[i].numfree = 0;	/* number of free backed inodes */
2898 	}
2899 
2900 	/*
2901 	 *	process each iag page of the map.
2902 	 *
2903 	 * rebuild AG Free Inode List, AG Free Inode Extent List;
2904 	 */
2905 	for (i = 0; i < imap->im_nextiag; i++) {
2906 		if ((rc = diIAGRead(imap, i, &bp))) {
2907 			rcx = rc;
2908 			continue;
2909 		}
2910 		iagp = (struct iag *) bp->data;
2911 		if (le32_to_cpu(iagp->iagnum) != i) {
2912 			release_metapage(bp);
2913 			jfs_error(ipimap->i_sb,
2914 				  "diExtendFs: unexpected value of iagnum");
2915 			return -EIO;
2916 		}
2917 
2918 		/* leave free iag in the free iag list */
2919 		if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2920 			release_metapage(bp);
2921 			continue;
2922 		}
2923 
2924 		/* agstart that computes to the same ag is treated as same; */
2925 		agstart = le64_to_cpu(iagp->agstart);
2926 		/* iagp->agstart = agstart & ~(mp->db_agsize - 1); */
2927 		n = agstart >> mp->db_agl2size;
2928 
2929 		/* compute backed inodes */
2930 		numinos = (EXTSPERIAG - le32_to_cpu(iagp->nfreeexts))
2931 		    << L2INOSPEREXT;
2932 		if (numinos > 0) {
2933 			/* merge AG backed inodes */
2934 			imap->im_agctl[n].numinos += numinos;
2935 			xnuminos += numinos;
2936 		}
2937 
2938 		/* if any backed free inodes, insert at AG free inode list */
2939 		if ((int) le32_to_cpu(iagp->nfreeinos) > 0) {
2940 			if ((head = imap->im_agctl[n].inofree) == -1) {
2941 				iagp->inofreefwd = cpu_to_le32(-1);
2942 				iagp->inofreeback = cpu_to_le32(-1);
2943 			} else {
2944 				if ((rc = diIAGRead(imap, head, &hbp))) {
2945 					rcx = rc;
2946 					goto nextiag;
2947 				}
2948 				hiagp = (struct iag *) hbp->data;
2949 				hiagp->inofreeback = iagp->iagnum;
2950 				iagp->inofreefwd = cpu_to_le32(head);
2951 				iagp->inofreeback = cpu_to_le32(-1);
2952 				write_metapage(hbp);
2953 			}
2954 
2955 			imap->im_agctl[n].inofree =
2956 			    le32_to_cpu(iagp->iagnum);
2957 
2958 			/* merge AG backed free inodes */
2959 			imap->im_agctl[n].numfree +=
2960 			    le32_to_cpu(iagp->nfreeinos);
2961 			xnumfree += le32_to_cpu(iagp->nfreeinos);
2962 		}
2963 
2964 		/* if any free extents, insert at AG free extent list */
2965 		if (le32_to_cpu(iagp->nfreeexts) > 0) {
2966 			if ((head = imap->im_agctl[n].extfree) == -1) {
2967 				iagp->extfreefwd = cpu_to_le32(-1);
2968 				iagp->extfreeback = cpu_to_le32(-1);
2969 			} else {
2970 				if ((rc = diIAGRead(imap, head, &hbp))) {
2971 					rcx = rc;
2972 					goto nextiag;
2973 				}
2974 				hiagp = (struct iag *) hbp->data;
2975 				hiagp->extfreeback = iagp->iagnum;
2976 				iagp->extfreefwd = cpu_to_le32(head);
2977 				iagp->extfreeback = cpu_to_le32(-1);
2978 				write_metapage(hbp);
2979 			}
2980 
2981 			imap->im_agctl[n].extfree =
2982 			    le32_to_cpu(iagp->iagnum);
2983 		}
2984 
2985 	      nextiag:
2986 		write_metapage(bp);
2987 	}
2988 
2989 	if (xnuminos != atomic_read(&imap->im_numinos) ||
2990 	    xnumfree != atomic_read(&imap->im_numfree)) {
2991 		jfs_error(ipimap->i_sb,
2992 			  "diExtendFs: numinos or numfree incorrect");
2993 		return -EIO;
2994 	}
2995 
2996 	return rcx;
2997 }
2998 
2999 
3000 /*
3001  *	duplicateIXtree()
3002  *
3003  * serialization: IWRITE_LOCK held on entry/exit
3004  *
3005  * note: shadow page with regular inode (rel.2);
3006  */
3007 static void duplicateIXtree(struct super_block *sb, s64 blkno,
3008 			    int xlen, s64 *xaddr)
3009 {
3010 	struct jfs_superblock *j_sb;
3011 	struct buffer_head *bh;
3012 	struct inode *ip;
3013 	tid_t tid;
3014 
3015 	/* if AIT2 ipmap2 is bad, do not try to update it */
3016 	if (JFS_SBI(sb)->mntflag & JFS_BAD_SAIT)	/* s_flag */
3017 		return;
3018 	ip = diReadSpecial(sb, FILESYSTEM_I, 1);
3019 	if (ip == NULL) {
3020 		JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT;
3021 		if (readSuper(sb, &bh))
3022 			return;
3023 		j_sb = (struct jfs_superblock *)bh->b_data;
3024 		j_sb->s_flag |= cpu_to_le32(JFS_BAD_SAIT);
3025 
3026 		mark_buffer_dirty(bh);
3027 		sync_dirty_buffer(bh);
3028 		brelse(bh);
3029 		return;
3030 	}
3031 
3032 	/* start transaction */
3033 	tid = txBegin(sb, COMMIT_FORCE);
3034 	/* update the inode map addressing structure to point to it */
3035 	if (xtInsert(tid, ip, 0, blkno, xlen, xaddr, 0)) {
3036 		JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT;
3037 		txAbort(tid, 1);
3038 		goto cleanup;
3039 
3040 	}
3041 	/* update the inode map's inode to reflect the extension */
3042 	ip->i_size += PSIZE;
3043 	inode_add_bytes(ip, PSIZE);
3044 	txCommit(tid, 1, &ip, COMMIT_FORCE);
3045       cleanup:
3046 	txEnd(tid);
3047 	diFreeSpecial(ip);
3048 }
3049 
3050 /*
3051  * NAME:	copy_from_dinode()
3052  *
3053  * FUNCTION:	Copies inode info from disk inode to in-memory inode
3054  *
3055  * RETURN VALUES:
3056  *	0	- success
3057  *	-ENOMEM	- insufficient memory
3058  */
3059 static int copy_from_dinode(struct dinode * dip, struct inode *ip)
3060 {
3061 	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
3062 	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
3063 
3064 	jfs_ip->fileset = le32_to_cpu(dip->di_fileset);
3065 	jfs_ip->mode2 = le32_to_cpu(dip->di_mode);
3066 	jfs_set_inode_flags(ip);
3067 
3068 	ip->i_mode = le32_to_cpu(dip->di_mode) & 0xffff;
3069 	if (sbi->umask != -1) {
3070 		ip->i_mode = (ip->i_mode & ~0777) | (0777 & ~sbi->umask);
3071 		/* For directories, add x permission if r is allowed by umask */
3072 		if (S_ISDIR(ip->i_mode)) {
3073 			if (ip->i_mode & 0400)
3074 				ip->i_mode |= 0100;
3075 			if (ip->i_mode & 0040)
3076 				ip->i_mode |= 0010;
3077 			if (ip->i_mode & 0004)
3078 				ip->i_mode |= 0001;
3079 		}
3080 	}
3081 	ip->i_nlink = le32_to_cpu(dip->di_nlink);
3082 
3083 	jfs_ip->saved_uid = le32_to_cpu(dip->di_uid);
3084 	if (sbi->uid == -1)
3085 		ip->i_uid = jfs_ip->saved_uid;
3086 	else {
3087 		ip->i_uid = sbi->uid;
3088 	}
3089 
3090 	jfs_ip->saved_gid = le32_to_cpu(dip->di_gid);
3091 	if (sbi->gid == -1)
3092 		ip->i_gid = jfs_ip->saved_gid;
3093 	else {
3094 		ip->i_gid = sbi->gid;
3095 	}
3096 
3097 	ip->i_size = le64_to_cpu(dip->di_size);
3098 	ip->i_atime.tv_sec = le32_to_cpu(dip->di_atime.tv_sec);
3099 	ip->i_atime.tv_nsec = le32_to_cpu(dip->di_atime.tv_nsec);
3100 	ip->i_mtime.tv_sec = le32_to_cpu(dip->di_mtime.tv_sec);
3101 	ip->i_mtime.tv_nsec = le32_to_cpu(dip->di_mtime.tv_nsec);
3102 	ip->i_ctime.tv_sec = le32_to_cpu(dip->di_ctime.tv_sec);
3103 	ip->i_ctime.tv_nsec = le32_to_cpu(dip->di_ctime.tv_nsec);
3104 	ip->i_blocks = LBLK2PBLK(ip->i_sb, le64_to_cpu(dip->di_nblocks));
3105 	ip->i_generation = le32_to_cpu(dip->di_gen);
3106 
3107 	jfs_ip->ixpxd = dip->di_ixpxd;	/* in-memory pxd's are little-endian */
3108 	jfs_ip->acl = dip->di_acl;	/* as are dxd's */
3109 	jfs_ip->ea = dip->di_ea;
3110 	jfs_ip->next_index = le32_to_cpu(dip->di_next_index);
3111 	jfs_ip->otime = le32_to_cpu(dip->di_otime.tv_sec);
3112 	jfs_ip->acltype = le32_to_cpu(dip->di_acltype);
3113 
3114 	if (S_ISCHR(ip->i_mode) || S_ISBLK(ip->i_mode)) {
3115 		jfs_ip->dev = le32_to_cpu(dip->di_rdev);
3116 		ip->i_rdev = new_decode_dev(jfs_ip->dev);
3117 	}
3118 
3119 	if (S_ISDIR(ip->i_mode)) {
3120 		memcpy(&jfs_ip->i_dirtable, &dip->di_dirtable, 384);
3121 	} else if (S_ISREG(ip->i_mode) || S_ISLNK(ip->i_mode)) {
3122 		memcpy(&jfs_ip->i_xtroot, &dip->di_xtroot, 288);
3123 	} else
3124 		memcpy(&jfs_ip->i_inline_ea, &dip->di_inlineea, 128);
3125 
3126 	/* Zero the in-memory-only stuff */
3127 	jfs_ip->cflag = 0;
3128 	jfs_ip->btindex = 0;
3129 	jfs_ip->btorder = 0;
3130 	jfs_ip->bxflag = 0;
3131 	jfs_ip->blid = 0;
3132 	jfs_ip->atlhead = 0;
3133 	jfs_ip->atltail = 0;
3134 	jfs_ip->xtlid = 0;
3135 	return (0);
3136 }
3137 
3138 /*
3139  * NAME:	copy_to_dinode()
3140  *
3141  * FUNCTION:	Copies inode info from in-memory inode to disk inode
3142  */
3143 static void copy_to_dinode(struct dinode * dip, struct inode *ip)
3144 {
3145 	struct jfs_inode_info *jfs_ip = JFS_IP(ip);
3146 	struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
3147 
3148 	dip->di_fileset = cpu_to_le32(jfs_ip->fileset);
3149 	dip->di_inostamp = cpu_to_le32(sbi->inostamp);
3150 	dip->di_number = cpu_to_le32(ip->i_ino);
3151 	dip->di_gen = cpu_to_le32(ip->i_generation);
3152 	dip->di_size = cpu_to_le64(ip->i_size);
3153 	dip->di_nblocks = cpu_to_le64(PBLK2LBLK(ip->i_sb, ip->i_blocks));
3154 	dip->di_nlink = cpu_to_le32(ip->i_nlink);
3155 	if (sbi->uid == -1)
3156 		dip->di_uid = cpu_to_le32(ip->i_uid);
3157 	else
3158 		dip->di_uid = cpu_to_le32(jfs_ip->saved_uid);
3159 	if (sbi->gid == -1)
3160 		dip->di_gid = cpu_to_le32(ip->i_gid);
3161 	else
3162 		dip->di_gid = cpu_to_le32(jfs_ip->saved_gid);
3163 	jfs_get_inode_flags(jfs_ip);
3164 	/*
3165 	 * mode2 is only needed for storing the higher order bits.
3166 	 * Trust i_mode for the lower order ones
3167 	 */
3168 	if (sbi->umask == -1)
3169 		dip->di_mode = cpu_to_le32((jfs_ip->mode2 & 0xffff0000) |
3170 					   ip->i_mode);
3171 	else /* Leave the original permissions alone */
3172 		dip->di_mode = cpu_to_le32(jfs_ip->mode2);
3173 
3174 	dip->di_atime.tv_sec = cpu_to_le32(ip->i_atime.tv_sec);
3175 	dip->di_atime.tv_nsec = cpu_to_le32(ip->i_atime.tv_nsec);
3176 	dip->di_ctime.tv_sec = cpu_to_le32(ip->i_ctime.tv_sec);
3177 	dip->di_ctime.tv_nsec = cpu_to_le32(ip->i_ctime.tv_nsec);
3178 	dip->di_mtime.tv_sec = cpu_to_le32(ip->i_mtime.tv_sec);
3179 	dip->di_mtime.tv_nsec = cpu_to_le32(ip->i_mtime.tv_nsec);
3180 	dip->di_ixpxd = jfs_ip->ixpxd;	/* in-memory pxd's are little-endian */
3181 	dip->di_acl = jfs_ip->acl;	/* as are dxd's */
3182 	dip->di_ea = jfs_ip->ea;
3183 	dip->di_next_index = cpu_to_le32(jfs_ip->next_index);
3184 	dip->di_otime.tv_sec = cpu_to_le32(jfs_ip->otime);
3185 	dip->di_otime.tv_nsec = 0;
3186 	dip->di_acltype = cpu_to_le32(jfs_ip->acltype);
3187 	if (S_ISCHR(ip->i_mode) || S_ISBLK(ip->i_mode))
3188 		dip->di_rdev = cpu_to_le32(jfs_ip->dev);
3189 }
3190