2 * Copyright (C) International Business Machines Corp., 2000-2003
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.
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.
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
20 * jfs_imap.c: inode allocation map manager
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.
27 * Each IAG is locked by obtaining the buffer for the IAG page.
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.
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.
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().
45 #include <linux/buffer_head.h>
46 #include <linux/pagemap.h>
48 #include "jfs_incore.h"
49 #include "jfs_filsys.h"
50 #include "jfs_dinode.h"
53 #include "jfs_metapage.h"
54 #include "jfs_superblock.h"
55 #include "jfs_debug.h"
60 /* iag free list lock */
61 #define IAGFREE_LOCK_INIT(imap) init_MUTEX(&imap->im_freelock)
62 #define IAGFREE_LOCK(imap) down(&imap->im_freelock)
63 #define IAGFREE_UNLOCK(imap) up(&imap->im_freelock)
65 /* per ag iag list locks */
66 #define AG_LOCK_INIT(imap,index) init_MUTEX(&(imap->im_aglock[index]))
67 #define AG_LOCK(imap,agno) down(&imap->im_aglock[agno])
68 #define AG_UNLOCK(imap,agno) up(&imap->im_aglock[agno])
73 extern struct address_space_operations jfs_aops;
78 static int diAllocAG(struct inomap *, int, boolean_t, struct inode *);
79 static int diAllocAny(struct inomap *, int, boolean_t, struct inode *);
80 static int diAllocBit(struct inomap *, struct iag *, int);
81 static int diAllocExt(struct inomap *, int, struct inode *);
82 static int diAllocIno(struct inomap *, int, struct inode *);
83 static int diFindFree(u32, int);
84 static int diNewExt(struct inomap *, struct iag *, int);
85 static int diNewIAG(struct inomap *, int *, int, struct metapage **);
86 static void duplicateIXtree(struct super_block *, s64, int, s64 *);
88 static int diIAGRead(struct inomap * imap, int, struct metapage **);
89 static int copy_from_dinode(struct dinode *, struct inode *);
90 static void copy_to_dinode(struct dinode *, struct inode *);
93 * debug code for double-checking inode map
95 /* #define _JFS_DEBUG_IMAP 1 */
97 #ifdef _JFS_DEBUG_IMAP
98 #define DBG_DIINIT(imap) DBGdiInit(imap)
99 #define DBG_DIALLOC(imap, ino) DBGdiAlloc(imap, ino)
100 #define DBG_DIFREE(imap, ino) DBGdiFree(imap, ino)
102 static void *DBGdiInit(struct inomap * imap);
103 static void DBGdiAlloc(struct inomap * imap, ino_t ino);
104 static void DBGdiFree(struct inomap * imap, ino_t ino);
106 #define DBG_DIINIT(imap)
107 #define DBG_DIALLOC(imap, ino)
108 #define DBG_DIFREE(imap, ino)
109 #endif /* _JFS_DEBUG_IMAP */
114 * FUNCTION: initialize the incore inode map control structures for
115 * a fileset or aggregate init time.
117 * the inode map's control structure (dinomap) is
118 * brought in from disk and placed in virtual memory.
121 * ipimap - pointer to inode map inode for the aggregate or fileset.
125 * -ENOMEM - insufficient free virtual memory.
128 int diMount(struct inode *ipimap)
133 struct dinomap *dinom_le;
136 * allocate/initialize the in-memory inode map control structure
138 /* allocate the in-memory inode map control structure. */
139 imap = (struct inomap *) kmalloc(sizeof(struct inomap), GFP_KERNEL);
141 jfs_err("diMount: kmalloc returned NULL!");
145 /* read the on-disk inode map control structure. */
147 mp = read_metapage(ipimap,
148 IMAPBLKNO << JFS_SBI(ipimap->i_sb)->l2nbperpage,
155 /* copy the on-disk version to the in-memory version. */
156 dinom_le = (struct dinomap *) mp->data;
157 imap->im_freeiag = le32_to_cpu(dinom_le->in_freeiag);
158 imap->im_nextiag = le32_to_cpu(dinom_le->in_nextiag);
159 atomic_set(&imap->im_numinos, le32_to_cpu(dinom_le->in_numinos));
160 atomic_set(&imap->im_numfree, le32_to_cpu(dinom_le->in_numfree));
161 imap->im_nbperiext = le32_to_cpu(dinom_le->in_nbperiext);
162 imap->im_l2nbperiext = le32_to_cpu(dinom_le->in_l2nbperiext);
163 for (index = 0; index < MAXAG; index++) {
164 imap->im_agctl[index].inofree =
165 le32_to_cpu(dinom_le->in_agctl[index].inofree);
166 imap->im_agctl[index].extfree =
167 le32_to_cpu(dinom_le->in_agctl[index].extfree);
168 imap->im_agctl[index].numinos =
169 le32_to_cpu(dinom_le->in_agctl[index].numinos);
170 imap->im_agctl[index].numfree =
171 le32_to_cpu(dinom_le->in_agctl[index].numfree);
174 /* release the buffer. */
175 release_metapage(mp);
178 * allocate/initialize inode allocation map locks
180 /* allocate and init iag free list lock */
181 IAGFREE_LOCK_INIT(imap);
183 /* allocate and init ag list locks */
184 for (index = 0; index < MAXAG; index++) {
185 AG_LOCK_INIT(imap, index);
188 /* bind the inode map inode and inode map control structure
191 imap->im_ipimap = ipimap;
192 JFS_IP(ipimap)->i_imap = imap;
203 * FUNCTION: write to disk the incore inode map control structures for
204 * a fileset or aggregate at unmount time.
207 * ipimap - pointer to inode map inode for the aggregate or fileset.
211 * -ENOMEM - insufficient free virtual memory.
214 int diUnmount(struct inode *ipimap, int mounterror)
216 struct inomap *imap = JFS_IP(ipimap)->i_imap;
219 * update the on-disk inode map control structure
222 if (!(mounterror || isReadOnly(ipimap)))
226 * Invalidate the page cache buffers
228 truncate_inode_pages(ipimap->i_mapping, 0);
231 * free in-memory control structure
242 int diSync(struct inode *ipimap)
244 struct dinomap *dinom_le;
245 struct inomap *imp = JFS_IP(ipimap)->i_imap;
250 * write imap global conrol page
252 /* read the on-disk inode map control structure */
253 mp = get_metapage(ipimap,
254 IMAPBLKNO << JFS_SBI(ipimap->i_sb)->l2nbperpage,
257 jfs_err("diSync: get_metapage failed!");
261 /* copy the in-memory version to the on-disk version */
262 dinom_le = (struct dinomap *) mp->data;
263 dinom_le->in_freeiag = cpu_to_le32(imp->im_freeiag);
264 dinom_le->in_nextiag = cpu_to_le32(imp->im_nextiag);
265 dinom_le->in_numinos = cpu_to_le32(atomic_read(&imp->im_numinos));
266 dinom_le->in_numfree = cpu_to_le32(atomic_read(&imp->im_numfree));
267 dinom_le->in_nbperiext = cpu_to_le32(imp->im_nbperiext);
268 dinom_le->in_l2nbperiext = cpu_to_le32(imp->im_l2nbperiext);
269 for (index = 0; index < MAXAG; index++) {
270 dinom_le->in_agctl[index].inofree =
271 cpu_to_le32(imp->im_agctl[index].inofree);
272 dinom_le->in_agctl[index].extfree =
273 cpu_to_le32(imp->im_agctl[index].extfree);
274 dinom_le->in_agctl[index].numinos =
275 cpu_to_le32(imp->im_agctl[index].numinos);
276 dinom_le->in_agctl[index].numfree =
277 cpu_to_le32(imp->im_agctl[index].numfree);
280 /* write out the control structure */
284 * write out dirty pages of imap
286 filemap_fdatawrite(ipimap->i_mapping);
287 filemap_fdatawait(ipimap->i_mapping);
289 diWriteSpecial(ipimap, 0);
298 * FUNCTION: initialize an incore inode from disk.
300 * on entry, the specifed incore inode should itself
301 * specify the disk inode number corresponding to the
302 * incore inode (i.e. i_number should be initialized).
304 * this routine handles incore inode initialization for
305 * both "special" and "regular" inodes. special inodes
306 * are those required early in the mount process and
307 * require special handling since much of the file system
308 * is not yet initialized. these "special" inodes are
309 * identified by a NULL inode map inode pointer and are
310 * actually initialized by a call to diReadSpecial().
312 * for regular inodes, the iag describing the disk inode
313 * is read from disk to determine the inode extent address
314 * for the disk inode. with the inode extent address in
315 * hand, the page of the extent that contains the disk
316 * inode is read and the disk inode is copied to the
320 * ip - pointer to incore inode to be initialized from disk.
325 * -ENOMEM - insufficient memory
328 int diRead(struct inode *ip)
330 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
331 int iagno, ino, extno, rc;
332 struct inode *ipimap;
343 jfs_info("diRead: ino = %ld", ip->i_ino);
345 ipimap = sbi->ipimap;
346 JFS_IP(ip)->ipimap = ipimap;
348 /* determine the iag number for this inode (number) */
349 iagno = INOTOIAG(ip->i_ino);
352 imap = JFS_IP(ipimap)->i_imap;
354 rc = diIAGRead(imap, iagno, &mp);
355 IREAD_UNLOCK(ipimap);
357 jfs_err("diRead: diIAGRead returned %d", rc);
361 iagp = (struct iag *) mp->data;
363 /* determine inode extent that holds the disk inode */
364 ino = ip->i_ino & (INOSPERIAG - 1);
365 extno = ino >> L2INOSPEREXT;
367 if ((lengthPXD(&iagp->inoext[extno]) != imap->im_nbperiext) ||
368 (addressPXD(&iagp->inoext[extno]) == 0)) {
369 release_metapage(mp);
373 /* get disk block number of the page within the inode extent
374 * that holds the disk inode.
376 blkno = INOPBLK(&iagp->inoext[extno], ino, sbi->l2nbperpage);
378 /* get the ag for the iag */
379 agstart = le64_to_cpu(iagp->agstart);
381 release_metapage(mp);
383 rel_inode = (ino & (INOSPERPAGE - 1));
384 pageno = blkno >> sbi->l2nbperpage;
386 if ((block_offset = ((u32) blkno & (sbi->nbperpage - 1)))) {
388 * OS/2 didn't always align inode extents on page boundaries
391 (sbi->nbperpage - block_offset) << sbi->l2niperblk;
393 if (rel_inode < inodes_left)
394 rel_inode += block_offset << sbi->l2niperblk;
397 rel_inode -= inodes_left;
401 /* read the page of disk inode */
402 mp = read_metapage(ipimap, pageno << sbi->l2nbperpage, PSIZE, 1);
404 jfs_err("diRead: read_metapage failed");
408 /* locate the the disk inode requested */
409 dp = (struct dinode *) mp->data;
412 if (ip->i_ino != le32_to_cpu(dp->di_number)) {
413 jfs_error(ip->i_sb, "diRead: i_ino != di_number");
415 } else if (le32_to_cpu(dp->di_nlink) == 0)
418 /* copy the disk inode to the in-memory inode */
419 rc = copy_from_dinode(dp, ip);
421 release_metapage(mp);
423 /* set the ag for the inode */
424 JFS_IP(ip)->agno = BLKTOAG(agstart, sbi);
425 JFS_IP(ip)->active_ag = -1;
432 * NAME: diReadSpecial()
434 * FUNCTION: initialize a 'special' inode from disk.
436 * this routines handles aggregate level inodes. The
437 * inode cache cannot differentiate between the
438 * aggregate inodes and the filesystem inodes, so we
439 * handle these here. We don't actually use the aggregate
440 * inode map, since these inodes are at a fixed location
441 * and in some cases the aggregate inode map isn't initialized
445 * sb - filesystem superblock
446 * inum - aggregate inode number
447 * secondary - 1 if secondary aggregate inode table
450 * new inode - success
453 struct inode *diReadSpecial(struct super_block *sb, ino_t inum, int secondary)
455 struct jfs_sb_info *sbi = JFS_SBI(sb);
463 jfs_err("diReadSpecial: new_inode returned NULL!");
468 address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage;
469 JFS_IP(ip)->ipimap = sbi->ipaimap2;
471 address = AITBL_OFF >> L2PSIZE;
472 JFS_IP(ip)->ipimap = sbi->ipaimap;
475 ASSERT(inum < INOSPEREXT);
479 address += inum >> 3; /* 8 inodes per 4K page */
481 /* read the page of fixed disk inode (AIT) in raw mode */
482 mp = read_metapage(ip, address << sbi->l2nbperpage, PSIZE, 1);
484 ip->i_nlink = 1; /* Don't want iput() deleting it */
489 /* get the pointer to the disk inode of interest */
490 dp = (struct dinode *) (mp->data);
491 dp += inum % 8; /* 8 inodes per 4K page */
493 /* copy on-disk inode to in-memory inode */
494 if ((copy_from_dinode(dp, ip)) != 0) {
495 /* handle bad return by returning NULL for ip */
496 ip->i_nlink = 1; /* Don't want iput() deleting it */
498 /* release the page */
499 release_metapage(mp);
504 ip->i_mapping->a_ops = &jfs_aops;
505 mapping_set_gfp_mask(ip->i_mapping, GFP_NOFS);
507 if ((inum == FILESYSTEM_I) && (JFS_IP(ip)->ipimap == sbi->ipaimap)) {
508 sbi->gengen = le32_to_cpu(dp->di_gengen);
509 sbi->inostamp = le32_to_cpu(dp->di_inostamp);
512 /* release the page */
513 release_metapage(mp);
519 * NAME: diWriteSpecial()
521 * FUNCTION: Write the special inode to disk
525 * secondary - 1 if secondary aggregate inode table
527 * RETURN VALUES: none
530 void diWriteSpecial(struct inode *ip, int secondary)
532 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
535 ino_t inum = ip->i_ino;
538 ip->i_state &= ~I_DIRTY;
541 address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage;
543 address = AITBL_OFF >> L2PSIZE;
545 ASSERT(inum < INOSPEREXT);
547 address += inum >> 3; /* 8 inodes per 4K page */
549 /* read the page of fixed disk inode (AIT) in raw mode */
550 mp = read_metapage(ip, address << sbi->l2nbperpage, PSIZE, 1);
552 jfs_err("diWriteSpecial: failed to read aggregate inode "
557 /* get the pointer to the disk inode of interest */
558 dp = (struct dinode *) (mp->data);
559 dp += inum % 8; /* 8 inodes per 4K page */
561 /* copy on-disk inode to in-memory inode */
562 copy_to_dinode(dp, ip);
563 memcpy(&dp->di_xtroot, &JFS_IP(ip)->i_xtroot, 288);
565 if (inum == FILESYSTEM_I)
566 dp->di_gengen = cpu_to_le32(sbi->gengen);
573 * NAME: diFreeSpecial()
575 * FUNCTION: Free allocated space for special inode
577 void diFreeSpecial(struct inode *ip)
580 jfs_err("diFreeSpecial called with NULL ip!");
583 filemap_fdatawrite(ip->i_mapping);
584 filemap_fdatawait(ip->i_mapping);
585 truncate_inode_pages(ip->i_mapping, 0);
594 * FUNCTION: write the on-disk inode portion of the in-memory inode
595 * to its corresponding on-disk inode.
597 * on entry, the specifed incore inode should itself
598 * specify the disk inode number corresponding to the
599 * incore inode (i.e. i_number should be initialized).
601 * the inode contains the inode extent address for the disk
602 * inode. with the inode extent address in hand, the
603 * page of the extent that contains the disk inode is
604 * read and the disk inode portion of the incore inode
605 * is copied to the disk inode.
608 * tid - transacation id
609 * ip - pointer to incore inode to be written to the inode extent.
615 int diWrite(tid_t tid, struct inode *ip)
617 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
618 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
629 struct inode *ipimap;
632 struct tlock *ditlck, *tlck;
633 struct linelock *dilinelock, *ilinelock;
637 ipimap = jfs_ip->ipimap;
639 ino = ip->i_ino & (INOSPERIAG - 1);
641 if (!addressPXD(&(jfs_ip->ixpxd)) ||
642 (lengthPXD(&(jfs_ip->ixpxd)) !=
643 JFS_IP(ipimap)->i_imap->im_nbperiext)) {
644 jfs_error(ip->i_sb, "diWrite: ixpxd invalid");
649 * read the page of disk inode containing the specified inode:
651 /* compute the block address of the page */
652 blkno = INOPBLK(&(jfs_ip->ixpxd), ino, sbi->l2nbperpage);
654 rel_inode = (ino & (INOSPERPAGE - 1));
655 pageno = blkno >> sbi->l2nbperpage;
657 if ((block_offset = ((u32) blkno & (sbi->nbperpage - 1)))) {
659 * OS/2 didn't always align inode extents on page boundaries
662 (sbi->nbperpage - block_offset) << sbi->l2niperblk;
664 if (rel_inode < inodes_left)
665 rel_inode += block_offset << sbi->l2niperblk;
668 rel_inode -= inodes_left;
671 /* read the page of disk inode */
673 mp = read_metapage(ipimap, pageno << sbi->l2nbperpage, PSIZE, 1);
677 /* get the pointer to the disk inode */
678 dp = (struct dinode *) mp->data;
681 dioffset = (ino & (INOSPERPAGE - 1)) << L2DISIZE;
684 * acquire transaction lock on the on-disk inode;
685 * N.B. tlock is acquired on ipimap not ip;
688 txLock(tid, ipimap, mp, tlckINODE | tlckENTRY)) == NULL)
690 dilinelock = (struct linelock *) & ditlck->lock;
693 * copy btree root from in-memory inode to on-disk inode
695 * (tlock is taken from inline B+-tree root in in-memory
696 * inode when the B+-tree root is updated, which is pointed
697 * by jfs_ip->blid as well as being on tx tlock list)
699 * further processing of btree root is based on the copy
700 * in in-memory inode, where txLog() will log from, and,
701 * for xtree root, txUpdateMap() will update map and reset
705 if (S_ISDIR(ip->i_mode) && (lid = jfs_ip->xtlid)) {
707 * This is the special xtree inside the directory for storing
708 * the directory table
714 tlck = lid_to_tlock(lid);
715 assert(tlck->type & tlckXTREE);
716 tlck->type |= tlckBTROOT;
718 ilinelock = (struct linelock *) & tlck->lock;
721 * copy xtree root from inode to dinode:
723 p = &jfs_ip->i_xtroot;
724 xp = (xtpage_t *) &dp->di_dirtable;
726 for (n = 0; n < ilinelock->index; n++, lv++) {
727 memcpy(&xp->xad[lv->offset], &p->xad[lv->offset],
728 lv->length << L2XTSLOTSIZE);
731 /* reset on-disk (metadata page) xtree XAD_NEW bit */
732 xad = &xp->xad[XTENTRYSTART];
733 for (n = XTENTRYSTART;
734 n < le16_to_cpu(xp->header.nextindex); n++, xad++)
735 if (xad->flag & (XAD_NEW | XAD_EXTENDED))
736 xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
739 if ((lid = jfs_ip->blid) == 0)
743 tlck = lid_to_tlock(lid);
745 tlck->type |= tlckBTROOT;
747 ilinelock = (struct linelock *) & tlck->lock;
750 * regular file: 16 byte (XAD slot) granularity
752 if (type & tlckXTREE) {
757 * copy xtree root from inode to dinode:
759 p = &jfs_ip->i_xtroot;
762 for (n = 0; n < ilinelock->index; n++, lv++) {
763 memcpy(&xp->xad[lv->offset], &p->xad[lv->offset],
764 lv->length << L2XTSLOTSIZE);
767 /* reset on-disk (metadata page) xtree XAD_NEW bit */
768 xad = &xp->xad[XTENTRYSTART];
769 for (n = XTENTRYSTART;
770 n < le16_to_cpu(xp->header.nextindex); n++, xad++)
771 if (xad->flag & (XAD_NEW | XAD_EXTENDED))
772 xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
775 * directory: 32 byte (directory entry slot) granularity
777 else if (type & tlckDTREE) {
781 * copy dtree root from inode to dinode:
783 p = (dtpage_t *) &jfs_ip->i_dtroot;
784 xp = (dtpage_t *) & dp->di_dtroot;
786 for (n = 0; n < ilinelock->index; n++, lv++) {
787 memcpy(&xp->slot[lv->offset], &p->slot[lv->offset],
788 lv->length << L2DTSLOTSIZE);
791 jfs_err("diWrite: UFO tlock");
796 * copy inline symlink from in-memory inode to on-disk inode
798 if (S_ISLNK(ip->i_mode) && ip->i_size < IDATASIZE) {
799 lv = & dilinelock->lv[dilinelock->index];
800 lv->offset = (dioffset + 2 * 128) >> L2INODESLOTSIZE;
802 memcpy(&dp->di_fastsymlink, jfs_ip->i_inline, IDATASIZE);
806 * copy inline data from in-memory inode to on-disk inode:
807 * 128 byte slot granularity
809 if (test_cflag(COMMIT_Inlineea, ip)) {
810 lv = & dilinelock->lv[dilinelock->index];
811 lv->offset = (dioffset + 3 * 128) >> L2INODESLOTSIZE;
813 memcpy(&dp->di_inlineea, jfs_ip->i_inline_ea, INODESLOTSIZE);
816 clear_cflag(COMMIT_Inlineea, ip);
820 * lock/copy inode base: 128 byte slot granularity
823 lv = & dilinelock->lv[dilinelock->index];
824 lv->offset = dioffset >> L2INODESLOTSIZE;
825 copy_to_dinode(dp, ip);
826 if (test_and_clear_cflag(COMMIT_Dirtable, ip)) {
828 memcpy(&dp->di_dirtable, &jfs_ip->i_dirtable, 96);
835 * We aren't logging changes to the DASD used in directory inodes,
836 * but we need to write them to disk. If we don't unmount cleanly,
837 * mount will recalculate the DASD used.
839 if (S_ISDIR(ip->i_mode)
840 && (ip->i_ipmnt->i_mntflag & JFS_DASD_ENABLED))
841 memcpy(&dp->di_DASD, &ip->i_DASD, sizeof(struct dasd));
842 #endif /* _JFS_FASTDASD */
844 /* release the buffer holding the updated on-disk inode.
845 * the buffer will be later written by commit processing.
856 * FUNCTION: free a specified inode from the inode working map
857 * for a fileset or aggregate.
859 * if the inode to be freed represents the first (only)
860 * free inode within the iag, the iag will be placed on
861 * the ag free inode list.
863 * freeing the inode will cause the inode extent to be
864 * freed if the inode is the only allocated inode within
865 * the extent. in this case all the disk resource backing
866 * up the inode extent will be freed. in addition, the iag
867 * will be placed on the ag extent free list if the extent
868 * is the first free extent in the iag. if freeing the
869 * extent also means that no free inodes will exist for
870 * the iag, the iag will also be removed from the ag free
873 * the iag describing the inode will be freed if the extent
874 * is to be freed and it is the only backed extent within
875 * the iag. in this case, the iag will be removed from the
876 * ag free extent list and ag free inode list and placed on
877 * the inode map's free iag list.
879 * a careful update approach is used to provide consistency
880 * in the face of updates to multiple buffers. under this
881 * approach, all required buffers are obtained before making
882 * any updates and are held until all updates are complete.
885 * ip - inode to be freed.
891 int diFree(struct inode *ip)
894 ino_t inum = ip->i_ino;
895 struct iag *iagp, *aiagp, *biagp, *ciagp, *diagp;
896 struct metapage *mp, *amp, *bmp, *cmp, *dmp;
897 int iagno, ino, extno, bitno, sword, agno;
900 struct inode *ipimap = JFS_SBI(ip->i_sb)->ipimap;
901 struct inomap *imap = JFS_IP(ipimap)->i_imap;
904 struct inode *iplist[3];
906 struct pxd_lock *pxdlock;
909 * This is just to suppress compiler warnings. The same logic that
910 * references these variables is used to initialize them.
912 aiagp = biagp = ciagp = diagp = NULL;
914 /* get the iag number containing the inode.
916 iagno = INOTOIAG(inum);
918 /* make sure that the iag is contained within
921 if (iagno >= imap->im_nextiag) {
922 dump_mem("imap", imap, 32);
924 "diFree: inum = %d, iagno = %d, nextiag = %d",
925 (uint) inum, iagno, imap->im_nextiag);
929 /* get the allocation group for this ino.
931 agno = JFS_IP(ip)->agno;
933 /* Lock the AG specific inode map information
937 /* Obtain read lock in imap inode. Don't release it until we have
938 * read all of the IAG's that we are going to.
944 if ((rc = diIAGRead(imap, iagno, &mp))) {
945 IREAD_UNLOCK(ipimap);
946 AG_UNLOCK(imap, agno);
949 iagp = (struct iag *) mp->data;
951 /* get the inode number and extent number of the inode within
952 * the iag and the inode number within the extent.
954 ino = inum & (INOSPERIAG - 1);
955 extno = ino >> L2INOSPEREXT;
956 bitno = ino & (INOSPEREXT - 1);
957 mask = HIGHORDER >> bitno;
959 if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
961 "diFree: wmap shows inode already free");
964 if (!addressPXD(&iagp->inoext[extno])) {
965 release_metapage(mp);
966 IREAD_UNLOCK(ipimap);
967 AG_UNLOCK(imap, agno);
968 jfs_error(ip->i_sb, "diFree: invalid inoext");
972 /* compute the bitmap for the extent reflecting the freed inode.
974 bitmap = le32_to_cpu(iagp->wmap[extno]) & ~mask;
976 if (imap->im_agctl[agno].numfree > imap->im_agctl[agno].numinos) {
977 release_metapage(mp);
978 IREAD_UNLOCK(ipimap);
979 AG_UNLOCK(imap, agno);
980 jfs_error(ip->i_sb, "diFree: numfree > numinos");
984 * inode extent still has some inodes or below low water mark:
985 * keep the inode extent;
988 imap->im_agctl[agno].numfree < 96 ||
989 (imap->im_agctl[agno].numfree < 288 &&
990 (((imap->im_agctl[agno].numfree * 100) /
991 imap->im_agctl[agno].numinos) <= 25))) {
992 /* if the iag currently has no free inodes (i.e.,
993 * the inode being freed is the first free inode of iag),
994 * insert the iag at head of the inode free list for the ag.
996 if (iagp->nfreeinos == 0) {
997 /* check if there are any iags on the ag inode
998 * free list. if so, read the first one so that
999 * we can link the current iag onto the list at
1002 if ((fwd = imap->im_agctl[agno].inofree) >= 0) {
1003 /* read the iag that currently is the head
1006 if ((rc = diIAGRead(imap, fwd, &))) {
1007 IREAD_UNLOCK(ipimap);
1008 AG_UNLOCK(imap, agno);
1009 release_metapage(mp);
1012 aiagp = (struct iag *) amp->data;
1014 /* make current head point back to the iag.
1016 aiagp->inofreeback = cpu_to_le32(iagno);
1018 write_metapage(amp);
1021 /* iag points forward to current head and iag
1022 * becomes the new head of the list.
1025 cpu_to_le32(imap->im_agctl[agno].inofree);
1026 iagp->inofreeback = -1;
1027 imap->im_agctl[agno].inofree = iagno;
1029 IREAD_UNLOCK(ipimap);
1031 /* update the free inode summary map for the extent if
1032 * freeing the inode means the extent will now have free
1033 * inodes (i.e., the inode being freed is the first free
1036 if (iagp->wmap[extno] == ONES) {
1037 sword = extno >> L2EXTSPERSUM;
1038 bitno = extno & (EXTSPERSUM - 1);
1039 iagp->inosmap[sword] &=
1040 cpu_to_le32(~(HIGHORDER >> bitno));
1043 /* update the bitmap.
1045 iagp->wmap[extno] = cpu_to_le32(bitmap);
1046 DBG_DIFREE(imap, inum);
1048 /* update the free inode counts at the iag, ag and
1052 cpu_to_le32(le32_to_cpu(iagp->nfreeinos) + 1);
1053 imap->im_agctl[agno].numfree += 1;
1054 atomic_inc(&imap->im_numfree);
1056 /* release the AG inode map lock
1058 AG_UNLOCK(imap, agno);
1068 * inode extent has become free and above low water mark:
1069 * free the inode extent;
1073 * prepare to update iag list(s) (careful update step 1)
1075 amp = bmp = cmp = dmp = NULL;
1078 /* check if the iag currently has no free extents. if so,
1079 * it will be placed on the head of the ag extent free list.
1081 if (iagp->nfreeexts == 0) {
1082 /* check if the ag extent free list has any iags.
1083 * if so, read the iag at the head of the list now.
1084 * this (head) iag will be updated later to reflect
1085 * the addition of the current iag at the head of
1088 if ((fwd = imap->im_agctl[agno].extfree) >= 0) {
1089 if ((rc = diIAGRead(imap, fwd, &)))
1091 aiagp = (struct iag *) amp->data;
1094 /* iag has free extents. check if the addition of a free
1095 * extent will cause all extents to be free within this
1096 * iag. if so, the iag will be removed from the ag extent
1097 * free list and placed on the inode map's free iag list.
1099 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) {
1100 /* in preparation for removing the iag from the
1101 * ag extent free list, read the iags preceeding
1102 * and following the iag on the ag extent free
1105 if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) {
1106 if ((rc = diIAGRead(imap, fwd, &)))
1108 aiagp = (struct iag *) amp->data;
1111 if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) {
1112 if ((rc = diIAGRead(imap, back, &bmp)))
1114 biagp = (struct iag *) bmp->data;
1119 /* remove the iag from the ag inode free list if freeing
1120 * this extent cause the iag to have no free inodes.
1122 if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) {
1123 int inofreeback = le32_to_cpu(iagp->inofreeback);
1124 int inofreefwd = le32_to_cpu(iagp->inofreefwd);
1126 /* in preparation for removing the iag from the
1127 * ag inode free list, read the iags preceeding
1128 * and following the iag on the ag inode free
1129 * list. before reading these iags, we must make
1130 * sure that we already don't have them in hand
1131 * from up above, since re-reading an iag (buffer)
1132 * we are currently holding would cause a deadlock.
1134 if (inofreefwd >= 0) {
1136 if (inofreefwd == fwd)
1137 ciagp = (struct iag *) amp->data;
1138 else if (inofreefwd == back)
1139 ciagp = (struct iag *) bmp->data;
1142 diIAGRead(imap, inofreefwd, &cmp)))
1144 ciagp = (struct iag *) cmp->data;
1146 assert(ciagp != NULL);
1149 if (inofreeback >= 0) {
1150 if (inofreeback == fwd)
1151 diagp = (struct iag *) amp->data;
1152 else if (inofreeback == back)
1153 diagp = (struct iag *) bmp->data;
1156 diIAGRead(imap, inofreeback, &dmp)))
1158 diagp = (struct iag *) dmp->data;
1160 assert(diagp != NULL);
1164 IREAD_UNLOCK(ipimap);
1167 * invalidate any page of the inode extent freed from buffer cache;
1169 freepxd = iagp->inoext[extno];
1170 invalidate_pxd_metapages(ip, freepxd);
1173 * update iag list(s) (careful update step 2)
1175 /* add the iag to the ag extent free list if this is the
1176 * first free extent for the iag.
1178 if (iagp->nfreeexts == 0) {
1180 aiagp->extfreeback = cpu_to_le32(iagno);
1183 cpu_to_le32(imap->im_agctl[agno].extfree);
1184 iagp->extfreeback = -1;
1185 imap->im_agctl[agno].extfree = iagno;
1187 /* remove the iag from the ag extent list if all extents
1188 * are now free and place it on the inode map iag free list.
1190 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) {
1192 aiagp->extfreeback = iagp->extfreeback;
1195 biagp->extfreefwd = iagp->extfreefwd;
1197 imap->im_agctl[agno].extfree =
1198 le32_to_cpu(iagp->extfreefwd);
1200 iagp->extfreefwd = iagp->extfreeback = -1;
1203 iagp->iagfree = cpu_to_le32(imap->im_freeiag);
1204 imap->im_freeiag = iagno;
1205 IAGFREE_UNLOCK(imap);
1209 /* remove the iag from the ag inode free list if freeing
1210 * this extent causes the iag to have no free inodes.
1212 if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) {
1213 if ((int) le32_to_cpu(iagp->inofreefwd) >= 0)
1214 ciagp->inofreeback = iagp->inofreeback;
1216 if ((int) le32_to_cpu(iagp->inofreeback) >= 0)
1217 diagp->inofreefwd = iagp->inofreefwd;
1219 imap->im_agctl[agno].inofree =
1220 le32_to_cpu(iagp->inofreefwd);
1222 iagp->inofreefwd = iagp->inofreeback = -1;
1225 /* update the inode extent address and working map
1226 * to reflect the free extent.
1227 * the permanent map should have been updated already
1228 * for the inode being freed.
1230 if (iagp->pmap[extno] != 0) {
1231 jfs_error(ip->i_sb, "diFree: the pmap does not show inode free");
1233 iagp->wmap[extno] = 0;
1234 DBG_DIFREE(imap, inum);
1235 PXDlength(&iagp->inoext[extno], 0);
1236 PXDaddress(&iagp->inoext[extno], 0);
1238 /* update the free extent and free inode summary maps
1239 * to reflect the freed extent.
1240 * the inode summary map is marked to indicate no inodes
1241 * available for the freed extent.
1243 sword = extno >> L2EXTSPERSUM;
1244 bitno = extno & (EXTSPERSUM - 1);
1245 mask = HIGHORDER >> bitno;
1246 iagp->inosmap[sword] |= cpu_to_le32(mask);
1247 iagp->extsmap[sword] &= cpu_to_le32(~mask);
1249 /* update the number of free inodes and number of free extents
1252 iagp->nfreeinos = cpu_to_le32(le32_to_cpu(iagp->nfreeinos) -
1254 iagp->nfreeexts = cpu_to_le32(le32_to_cpu(iagp->nfreeexts) + 1);
1256 /* update the number of free inodes and backed inodes
1257 * at the ag and inode map level.
1259 imap->im_agctl[agno].numfree -= (INOSPEREXT - 1);
1260 imap->im_agctl[agno].numinos -= INOSPEREXT;
1261 atomic_sub(INOSPEREXT - 1, &imap->im_numfree);
1262 atomic_sub(INOSPEREXT, &imap->im_numinos);
1265 write_metapage(amp);
1267 write_metapage(bmp);
1269 write_metapage(cmp);
1271 write_metapage(dmp);
1274 * start transaction to update block allocation map
1275 * for the inode extent freed;
1277 * N.B. AG_LOCK is released and iag will be released below, and
1278 * other thread may allocate inode from/reusing the ixad freed
1279 * BUT with new/different backing inode extent from the extent
1280 * to be freed by the transaction;
1282 tid = txBegin(ipimap->i_sb, COMMIT_FORCE);
1284 /* acquire tlock of the iag page of the freed ixad
1285 * to force the page NOHOMEOK (even though no data is
1286 * logged from the iag page) until NOREDOPAGE|FREEXTENT log
1287 * for the free of the extent is committed;
1288 * write FREEXTENT|NOREDOPAGE log record
1289 * N.B. linelock is overlaid as freed extent descriptor;
1291 tlck = txLock(tid, ipimap, mp, tlckINODE | tlckFREE);
1292 pxdlock = (struct pxd_lock *) & tlck->lock;
1293 pxdlock->flag = mlckFREEPXD;
1294 pxdlock->pxd = freepxd;
1302 * logredo needs the IAG number and IAG extent index in order
1303 * to ensure that the IMap is consistent. The least disruptive
1304 * way to pass these values through to the transaction manager
1305 * is in the iplist array.
1307 * It's not pretty, but it works.
1309 iplist[1] = (struct inode *) (size_t)iagno;
1310 iplist[2] = (struct inode *) (size_t)extno;
1312 rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE);
1316 /* unlock the AG inode map information */
1317 AG_UNLOCK(imap, agno);
1322 IREAD_UNLOCK(ipimap);
1325 release_metapage(amp);
1327 release_metapage(bmp);
1329 release_metapage(cmp);
1331 release_metapage(dmp);
1333 AG_UNLOCK(imap, agno);
1335 release_metapage(mp);
1341 * There are several places in the diAlloc* routines where we initialize
1345 diInitInode(struct inode *ip, int iagno, int ino, int extno, struct iag * iagp)
1347 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
1348 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
1350 ip->i_ino = (iagno << L2INOSPERIAG) + ino;
1351 DBG_DIALLOC(JFS_IP(ipimap)->i_imap, ip->i_ino);
1352 jfs_ip->ixpxd = iagp->inoext[extno];
1353 jfs_ip->agno = BLKTOAG(le64_to_cpu(iagp->agstart), sbi);
1354 jfs_ip->active_ag = -1;
1359 * NAME: diAlloc(pip,dir,ip)
1361 * FUNCTION: allocate a disk inode from the inode working map
1362 * for a fileset or aggregate.
1365 * pip - pointer to incore inode for the parent inode.
1366 * dir - TRUE if the new disk inode is for a directory.
1367 * ip - pointer to a new inode
1371 * -ENOSPC - insufficient disk resources.
1374 int diAlloc(struct inode *pip, boolean_t dir, struct inode *ip)
1376 int rc, ino, iagno, addext, extno, bitno, sword;
1377 int nwords, rem, i, agno;
1378 u32 mask, inosmap, extsmap;
1379 struct inode *ipimap;
1380 struct metapage *mp;
1383 struct inomap *imap;
1385 /* get the pointers to the inode map inode and the
1386 * corresponding imap control structure.
1388 ipimap = JFS_SBI(pip->i_sb)->ipimap;
1389 imap = JFS_IP(ipimap)->i_imap;
1390 JFS_IP(ip)->ipimap = ipimap;
1391 JFS_IP(ip)->fileset = FILESYSTEM_I;
1393 /* for a directory, the allocation policy is to start
1394 * at the ag level using the preferred ag.
1397 agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap);
1398 AG_LOCK(imap, agno);
1402 /* for files, the policy starts off by trying to allocate from
1403 * the same iag containing the parent disk inode:
1404 * try to allocate the new disk inode close to the parent disk
1405 * inode, using parent disk inode number + 1 as the allocation
1406 * hint. (we use a left-to-right policy to attempt to avoid
1407 * moving backward on the disk.) compute the hint within the
1408 * file system and the iag.
1411 /* get the ag number of this iag */
1412 agno = JFS_IP(pip)->agno;
1414 if (atomic_read(&JFS_SBI(pip->i_sb)->bmap->db_active[agno])) {
1416 * There is an open file actively growing. We want to
1417 * allocate new inodes from a different ag to avoid
1418 * fragmentation problems.
1420 agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap);
1421 AG_LOCK(imap, agno);
1425 inum = pip->i_ino + 1;
1426 ino = inum & (INOSPERIAG - 1);
1428 /* back off the the hint if it is outside of the iag */
1432 /* lock the AG inode map information */
1433 AG_LOCK(imap, agno);
1435 /* Get read lock on imap inode */
1438 /* get the iag number and read the iag */
1439 iagno = INOTOIAG(inum);
1440 if ((rc = diIAGRead(imap, iagno, &mp))) {
1441 IREAD_UNLOCK(ipimap);
1442 AG_UNLOCK(imap, agno);
1445 iagp = (struct iag *) mp->data;
1447 /* determine if new inode extent is allowed to be added to the iag.
1448 * new inode extent can be added to the iag if the ag
1449 * has less than 32 free disk inodes and the iag has free extents.
1451 addext = (imap->im_agctl[agno].numfree < 32 && iagp->nfreeexts);
1454 * try to allocate from the IAG
1456 /* check if the inode may be allocated from the iag
1457 * (i.e. the inode has free inodes or new extent can be added).
1459 if (iagp->nfreeinos || addext) {
1460 /* determine the extent number of the hint.
1462 extno = ino >> L2INOSPEREXT;
1464 /* check if the extent containing the hint has backed
1465 * inodes. if so, try to allocate within this extent.
1467 if (addressPXD(&iagp->inoext[extno])) {
1468 bitno = ino & (INOSPEREXT - 1);
1470 diFindFree(le32_to_cpu(iagp->wmap[extno]),
1473 ino = (extno << L2INOSPEREXT) + bitno;
1475 /* a free inode (bit) was found within this
1476 * extent, so allocate it.
1478 rc = diAllocBit(imap, iagp, ino);
1479 IREAD_UNLOCK(ipimap);
1483 /* set the results of the allocation
1484 * and write the iag.
1486 diInitInode(ip, iagno, ino, extno,
1488 mark_metapage_dirty(mp);
1490 release_metapage(mp);
1492 /* free the AG lock and return.
1494 AG_UNLOCK(imap, agno);
1501 EXTSPERIAG - 1) ? 0 : extno + 1;
1505 * no free inodes within the extent containing the hint.
1507 * try to allocate from the backed extents following
1508 * hint or, if appropriate (i.e. addext is true), allocate
1509 * an extent of free inodes at or following the extent
1510 * containing the hint.
1512 * the free inode and free extent summary maps are used
1513 * here, so determine the starting summary map position
1514 * and the number of words we'll have to examine. again,
1515 * the approach is to allocate following the hint, so we
1516 * might have to initially ignore prior bits of the summary
1517 * map that represent extents prior to the extent containing
1518 * the hint and later revisit these bits.
1520 bitno = extno & (EXTSPERSUM - 1);
1521 nwords = (bitno == 0) ? SMAPSZ : SMAPSZ + 1;
1522 sword = extno >> L2EXTSPERSUM;
1524 /* mask any prior bits for the starting words of the
1527 mask = ONES << (EXTSPERSUM - bitno);
1528 inosmap = le32_to_cpu(iagp->inosmap[sword]) | mask;
1529 extsmap = le32_to_cpu(iagp->extsmap[sword]) | mask;
1531 /* scan the free inode and free extent summary maps for
1534 for (i = 0; i < nwords; i++) {
1535 /* check if this word of the free inode summary
1536 * map describes an extent with free inodes.
1539 /* an extent with free inodes has been
1540 * found. determine the extent number
1541 * and the inode number within the extent.
1543 rem = diFindFree(inosmap, 0);
1544 extno = (sword << L2EXTSPERSUM) + rem;
1545 rem = diFindFree(le32_to_cpu(iagp->wmap[extno]),
1547 if (rem >= INOSPEREXT) {
1548 IREAD_UNLOCK(ipimap);
1549 release_metapage(mp);
1550 AG_UNLOCK(imap, agno);
1552 "diAlloc: can't find free bit "
1557 /* determine the inode number within the
1558 * iag and allocate the inode from the
1561 ino = (extno << L2INOSPEREXT) + rem;
1562 rc = diAllocBit(imap, iagp, ino);
1563 IREAD_UNLOCK(ipimap);
1567 /* set the results of the allocation
1568 * and write the iag.
1570 diInitInode(ip, iagno, ino, extno,
1572 mark_metapage_dirty(mp);
1574 release_metapage(mp);
1576 /* free the AG lock and return.
1578 AG_UNLOCK(imap, agno);
1583 /* check if we may allocate an extent of free
1584 * inodes and whether this word of the free
1585 * extents summary map describes a free extent.
1587 if (addext && ~extsmap) {
1588 /* a free extent has been found. determine
1589 * the extent number.
1591 rem = diFindFree(extsmap, 0);
1592 extno = (sword << L2EXTSPERSUM) + rem;
1594 /* allocate an extent of free inodes.
1596 if ((rc = diNewExt(imap, iagp, extno))) {
1597 /* if there is no disk space for a
1598 * new extent, try to allocate the
1599 * disk inode from somewhere else.
1606 /* set the results of the allocation
1607 * and write the iag.
1609 diInitInode(ip, iagno,
1610 extno << L2INOSPEREXT,
1612 mark_metapage_dirty(mp);
1614 release_metapage(mp);
1615 /* free the imap inode & the AG lock & return.
1617 IREAD_UNLOCK(ipimap);
1618 AG_UNLOCK(imap, agno);
1622 /* move on to the next set of summary map words.
1624 sword = (sword == SMAPSZ - 1) ? 0 : sword + 1;
1625 inosmap = le32_to_cpu(iagp->inosmap[sword]);
1626 extsmap = le32_to_cpu(iagp->extsmap[sword]);
1629 /* unlock imap inode */
1630 IREAD_UNLOCK(ipimap);
1632 /* nothing doing in this iag, so release it. */
1633 release_metapage(mp);
1637 * try to allocate anywhere within the same AG as the parent inode.
1639 rc = diAllocAG(imap, agno, dir, ip);
1641 AG_UNLOCK(imap, agno);
1647 * try to allocate in any AG.
1649 return (diAllocAny(imap, agno, dir, ip));
1654 * NAME: diAllocAG(imap,agno,dir,ip)
1656 * FUNCTION: allocate a disk inode from the allocation group.
1658 * this routine first determines if a new extent of free
1659 * inodes should be added for the allocation group, with
1660 * the current request satisfied from this extent. if this
1661 * is the case, an attempt will be made to do just that. if
1662 * this attempt fails or it has been determined that a new
1663 * extent should not be added, an attempt is made to satisfy
1664 * the request by allocating an existing (backed) free inode
1665 * from the allocation group.
1667 * PRE CONDITION: Already have the AG lock for this AG.
1670 * imap - pointer to inode map control structure.
1671 * agno - allocation group to allocate from.
1672 * dir - TRUE if the new disk inode is for a directory.
1673 * ip - pointer to the new inode to be filled in on successful return
1674 * with the disk inode number allocated, its extent address
1675 * and the start of the ag.
1679 * -ENOSPC - insufficient disk resources.
1683 diAllocAG(struct inomap * imap, int agno, boolean_t dir, struct inode *ip)
1685 int rc, addext, numfree, numinos;
1687 /* get the number of free and the number of backed disk
1688 * inodes currently within the ag.
1690 numfree = imap->im_agctl[agno].numfree;
1691 numinos = imap->im_agctl[agno].numinos;
1693 if (numfree > numinos) {
1694 jfs_error(ip->i_sb, "diAllocAG: numfree > numinos");
1698 /* determine if we should allocate a new extent of free inodes
1699 * within the ag: for directory inodes, add a new extent
1700 * if there are a small number of free inodes or number of free
1701 * inodes is a small percentage of the number of backed inodes.
1704 addext = (numfree < 64 ||
1706 && ((numfree * 100) / numinos) <= 20));
1708 addext = (numfree == 0);
1711 * try to allocate a new extent of free inodes.
1714 /* if free space is not avaliable for this new extent, try
1715 * below to allocate a free and existing (already backed)
1716 * inode from the ag.
1718 if ((rc = diAllocExt(imap, agno, ip)) != -ENOSPC)
1723 * try to allocate an existing free inode from the ag.
1725 return (diAllocIno(imap, agno, ip));
1730 * NAME: diAllocAny(imap,agno,dir,iap)
1732 * FUNCTION: allocate a disk inode from any other allocation group.
1734 * this routine is called when an allocation attempt within
1735 * the primary allocation group has failed. if attempts to
1736 * allocate an inode from any allocation group other than the
1737 * specified primary group.
1740 * imap - pointer to inode map control structure.
1741 * agno - primary allocation group (to avoid).
1742 * dir - TRUE if the new disk inode is for a directory.
1743 * ip - pointer to a new inode to be filled in on successful return
1744 * with the disk inode number allocated, its extent address
1745 * and the start of the ag.
1749 * -ENOSPC - insufficient disk resources.
1753 diAllocAny(struct inomap * imap, int agno, boolean_t dir, struct inode *ip)
1756 int maxag = JFS_SBI(imap->im_ipimap->i_sb)->bmap->db_maxag;
1759 /* try to allocate from the ags following agno up to
1760 * the maximum ag number.
1762 for (ag = agno + 1; ag <= maxag; ag++) {
1765 rc = diAllocAG(imap, ag, dir, ip);
1767 AG_UNLOCK(imap, ag);
1773 /* try to allocate from the ags in front of agno.
1775 for (ag = 0; ag < agno; ag++) {
1778 rc = diAllocAG(imap, ag, dir, ip);
1780 AG_UNLOCK(imap, ag);
1786 /* no free disk inodes.
1793 * NAME: diAllocIno(imap,agno,ip)
1795 * FUNCTION: allocate a disk inode from the allocation group's free
1796 * inode list, returning an error if this free list is
1797 * empty (i.e. no iags on the list).
1799 * allocation occurs from the first iag on the list using
1800 * the iag's free inode summary map to find the leftmost
1801 * free inode in the iag.
1803 * PRE CONDITION: Already have AG lock for this AG.
1806 * imap - pointer to inode map control structure.
1807 * agno - allocation group.
1808 * ip - pointer to new inode to be filled in on successful return
1809 * with the disk inode number allocated, its extent address
1810 * and the start of the ag.
1814 * -ENOSPC - insufficient disk resources.
1817 static int diAllocIno(struct inomap * imap, int agno, struct inode *ip)
1819 int iagno, ino, rc, rem, extno, sword;
1820 struct metapage *mp;
1823 /* check if there are iags on the ag's free inode list.
1825 if ((iagno = imap->im_agctl[agno].inofree) < 0)
1828 /* obtain read lock on imap inode */
1829 IREAD_LOCK(imap->im_ipimap);
1831 /* read the iag at the head of the list.
1833 if ((rc = diIAGRead(imap, iagno, &mp))) {
1834 IREAD_UNLOCK(imap->im_ipimap);
1837 iagp = (struct iag *) mp->data;
1839 /* better be free inodes in this iag if it is on the
1842 if (!iagp->nfreeinos) {
1843 IREAD_UNLOCK(imap->im_ipimap);
1844 release_metapage(mp);
1846 "diAllocIno: nfreeinos = 0, but iag on freelist");
1850 /* scan the free inode summary map to find an extent
1853 for (sword = 0;; sword++) {
1854 if (sword >= SMAPSZ) {
1855 IREAD_UNLOCK(imap->im_ipimap);
1856 release_metapage(mp);
1858 "diAllocIno: free inode not found in summary map");
1862 if (~iagp->inosmap[sword])
1866 /* found a extent with free inodes. determine
1867 * the extent number.
1869 rem = diFindFree(le32_to_cpu(iagp->inosmap[sword]), 0);
1870 if (rem >= EXTSPERSUM) {
1871 IREAD_UNLOCK(imap->im_ipimap);
1872 release_metapage(mp);
1873 jfs_error(ip->i_sb, "diAllocIno: no free extent found");
1876 extno = (sword << L2EXTSPERSUM) + rem;
1878 /* find the first free inode in the extent.
1880 rem = diFindFree(le32_to_cpu(iagp->wmap[extno]), 0);
1881 if (rem >= INOSPEREXT) {
1882 IREAD_UNLOCK(imap->im_ipimap);
1883 release_metapage(mp);
1884 jfs_error(ip->i_sb, "diAllocIno: free inode not found");
1888 /* compute the inode number within the iag.
1890 ino = (extno << L2INOSPEREXT) + rem;
1892 /* allocate the inode.
1894 rc = diAllocBit(imap, iagp, ino);
1895 IREAD_UNLOCK(imap->im_ipimap);
1897 release_metapage(mp);
1901 /* set the results of the allocation and write the iag.
1903 diInitInode(ip, iagno, ino, extno, iagp);
1911 * NAME: diAllocExt(imap,agno,ip)
1913 * FUNCTION: add a new extent of free inodes to an iag, allocating
1914 * an inode from this extent to satisfy the current allocation
1917 * this routine first tries to find an existing iag with free
1918 * extents through the ag free extent list. if list is not
1919 * empty, the head of the list will be selected as the home
1920 * of the new extent of free inodes. otherwise (the list is
1921 * empty), a new iag will be allocated for the ag to contain
1924 * once an iag has been selected, the free extent summary map
1925 * is used to locate a free extent within the iag and diNewExt()
1926 * is called to initialize the extent, with initialization
1927 * including the allocation of the first inode of the extent
1928 * for the purpose of satisfying this request.
1931 * imap - pointer to inode map control structure.
1932 * agno - allocation group number.
1933 * ip - pointer to new inode to be filled in on successful return
1934 * with the disk inode number allocated, its extent address
1935 * and the start of the ag.
1939 * -ENOSPC - insufficient disk resources.
1942 static int diAllocExt(struct inomap * imap, int agno, struct inode *ip)
1944 int rem, iagno, sword, extno, rc;
1945 struct metapage *mp;
1948 /* check if the ag has any iags with free extents. if not,
1949 * allocate a new iag for the ag.
1951 if ((iagno = imap->im_agctl[agno].extfree) < 0) {
1952 /* If successful, diNewIAG will obtain the read lock on the
1955 if ((rc = diNewIAG(imap, &iagno, agno, &mp))) {
1958 iagp = (struct iag *) mp->data;
1960 /* set the ag number if this a brand new iag
1963 cpu_to_le64(AGTOBLK(agno, imap->im_ipimap));
1967 IREAD_LOCK(imap->im_ipimap);
1968 if ((rc = diIAGRead(imap, iagno, &mp))) {
1969 IREAD_UNLOCK(imap->im_ipimap);
1970 jfs_error(ip->i_sb, "diAllocExt: error reading iag");
1973 iagp = (struct iag *) mp->data;
1976 /* using the free extent summary map, find a free extent.
1978 for (sword = 0;; sword++) {
1979 if (sword >= SMAPSZ) {
1980 release_metapage(mp);
1981 IREAD_UNLOCK(imap->im_ipimap);
1983 "diAllocExt: free ext summary map not found");
1986 if (~iagp->extsmap[sword])
1990 /* determine the extent number of the free extent.
1992 rem = diFindFree(le32_to_cpu(iagp->extsmap[sword]), 0);
1993 if (rem >= EXTSPERSUM) {
1994 release_metapage(mp);
1995 IREAD_UNLOCK(imap->im_ipimap);
1996 jfs_error(ip->i_sb, "diAllocExt: free extent not found");
1999 extno = (sword << L2EXTSPERSUM) + rem;
2001 /* initialize the new extent.
2003 rc = diNewExt(imap, iagp, extno);
2004 IREAD_UNLOCK(imap->im_ipimap);
2006 /* something bad happened. if a new iag was allocated,
2007 * place it back on the inode map's iag free list, and
2008 * clear the ag number information.
2010 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2012 iagp->iagfree = cpu_to_le32(imap->im_freeiag);
2013 imap->im_freeiag = iagno;
2014 IAGFREE_UNLOCK(imap);
2020 /* set the results of the allocation and write the iag.
2022 diInitInode(ip, iagno, extno << L2INOSPEREXT, extno, iagp);
2031 * NAME: diAllocBit(imap,iagp,ino)
2033 * FUNCTION: allocate a backed inode from an iag.
2035 * this routine performs the mechanics of allocating a
2036 * specified inode from a backed extent.
2038 * if the inode to be allocated represents the last free
2039 * inode within the iag, the iag will be removed from the
2040 * ag free inode list.
2042 * a careful update approach is used to provide consistency
2043 * in the face of updates to multiple buffers. under this
2044 * approach, all required buffers are obtained before making
2045 * any updates and are held all are updates are complete.
2047 * PRE CONDITION: Already have buffer lock on iagp. Already have AG lock on
2048 * this AG. Must have read lock on imap inode.
2051 * imap - pointer to inode map control structure.
2052 * iagp - pointer to iag.
2053 * ino - inode number to be allocated within the iag.
2057 * -ENOSPC - insufficient disk resources.
2060 static int diAllocBit(struct inomap * imap, struct iag * iagp, int ino)
2062 int extno, bitno, agno, sword, rc;
2063 struct metapage *amp = NULL, *bmp = NULL;
2064 struct iag *aiagp = 0, *biagp = 0;
2067 /* check if this is the last free inode within the iag.
2068 * if so, it will have to be removed from the ag free
2069 * inode list, so get the iags preceeding and following
2072 if (iagp->nfreeinos == cpu_to_le32(1)) {
2073 if ((int) le32_to_cpu(iagp->inofreefwd) >= 0) {
2075 diIAGRead(imap, le32_to_cpu(iagp->inofreefwd),
2078 aiagp = (struct iag *) amp->data;
2081 if ((int) le32_to_cpu(iagp->inofreeback) >= 0) {
2084 le32_to_cpu(iagp->inofreeback),
2087 release_metapage(amp);
2090 biagp = (struct iag *) bmp->data;
2094 /* get the ag number, extent number, inode number within
2097 agno = BLKTOAG(le64_to_cpu(iagp->agstart), JFS_SBI(imap->im_ipimap->i_sb));
2098 extno = ino >> L2INOSPEREXT;
2099 bitno = ino & (INOSPEREXT - 1);
2101 /* compute the mask for setting the map.
2103 mask = HIGHORDER >> bitno;
2105 /* the inode should be free and backed.
2107 if (((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) ||
2108 ((le32_to_cpu(iagp->wmap[extno]) & mask) != 0) ||
2109 (addressPXD(&iagp->inoext[extno]) == 0)) {
2111 release_metapage(amp);
2113 release_metapage(bmp);
2115 jfs_error(imap->im_ipimap->i_sb,
2116 "diAllocBit: iag inconsistent");
2120 /* mark the inode as allocated in the working map.
2122 iagp->wmap[extno] |= cpu_to_le32(mask);
2124 /* check if all inodes within the extent are now
2125 * allocated. if so, update the free inode summary
2126 * map to reflect this.
2128 if (iagp->wmap[extno] == ONES) {
2129 sword = extno >> L2EXTSPERSUM;
2130 bitno = extno & (EXTSPERSUM - 1);
2131 iagp->inosmap[sword] |= cpu_to_le32(HIGHORDER >> bitno);
2134 /* if this was the last free inode in the iag, remove the
2135 * iag from the ag free inode list.
2137 if (iagp->nfreeinos == cpu_to_le32(1)) {
2139 aiagp->inofreeback = iagp->inofreeback;
2140 write_metapage(amp);
2144 biagp->inofreefwd = iagp->inofreefwd;
2145 write_metapage(bmp);
2147 imap->im_agctl[agno].inofree =
2148 le32_to_cpu(iagp->inofreefwd);
2150 iagp->inofreefwd = iagp->inofreeback = -1;
2153 /* update the free inode count at the iag, ag, inode
2156 iagp->nfreeinos = cpu_to_le32(le32_to_cpu(iagp->nfreeinos) - 1);
2157 imap->im_agctl[agno].numfree -= 1;
2158 atomic_dec(&imap->im_numfree);
2165 * NAME: diNewExt(imap,iagp,extno)
2167 * FUNCTION: initialize a new extent of inodes for an iag, allocating
2168 * the first inode of the extent for use for the current
2169 * allocation request.
2171 * disk resources are allocated for the new extent of inodes
2172 * and the inodes themselves are initialized to reflect their
2173 * existence within the extent (i.e. their inode numbers and
2174 * inode extent addresses are set) and their initial state
2175 * (mode and link count are set to zero).
2177 * if the iag is new, it is not yet on an ag extent free list
2178 * but will now be placed on this list.
2180 * if the allocation of the new extent causes the iag to
2181 * have no free extent, the iag will be removed from the
2182 * ag extent free list.
2184 * if the iag has no free backed inodes, it will be placed
2185 * on the ag free inode list, since the addition of the new
2186 * extent will now cause it to have free inodes.
2188 * a careful update approach is used to provide consistency
2189 * (i.e. list consistency) in the face of updates to multiple
2190 * buffers. under this approach, all required buffers are
2191 * obtained before making any updates and are held until all
2192 * updates are complete.
2194 * PRE CONDITION: Already have buffer lock on iagp. Already have AG lock on
2195 * this AG. Must have read lock on imap inode.
2198 * imap - pointer to inode map control structure.
2199 * iagp - pointer to iag.
2200 * extno - extent number.
2204 * -ENOSPC - insufficient disk resources.
2207 static int diNewExt(struct inomap * imap, struct iag * iagp, int extno)
2209 int agno, iagno, fwd, back, freei = 0, sword, rc;
2210 struct iag *aiagp = 0, *biagp = 0, *ciagp = 0;
2211 struct metapage *amp, *bmp, *cmp, *dmp;
2212 struct inode *ipimap;
2218 struct jfs_sb_info *sbi;
2220 /* better have free extents.
2222 if (!iagp->nfreeexts) {
2223 jfs_error(imap->im_ipimap->i_sb, "diNewExt: no free extents");
2227 /* get the inode map inode.
2229 ipimap = imap->im_ipimap;
2230 sbi = JFS_SBI(ipimap->i_sb);
2232 amp = bmp = cmp = NULL;
2234 /* get the ag and iag numbers for this iag.
2236 agno = BLKTOAG(le64_to_cpu(iagp->agstart), sbi);
2237 iagno = le32_to_cpu(iagp->iagnum);
2239 /* check if this is the last free extent within the
2240 * iag. if so, the iag must be removed from the ag
2241 * free extent list, so get the iags preceeding and
2242 * following the iag on this list.
2244 if (iagp->nfreeexts == cpu_to_le32(1)) {
2245 if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) {
2246 if ((rc = diIAGRead(imap, fwd, &)))
2248 aiagp = (struct iag *) amp->data;
2251 if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) {
2252 if ((rc = diIAGRead(imap, back, &bmp)))
2254 biagp = (struct iag *) bmp->data;
2257 /* the iag has free extents. if all extents are free
2258 * (as is the case for a newly allocated iag), the iag
2259 * must be added to the ag free extent list, so get
2260 * the iag at the head of the list in preparation for
2261 * adding this iag to this list.
2264 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2265 if ((fwd = imap->im_agctl[agno].extfree) >= 0) {
2266 if ((rc = diIAGRead(imap, fwd, &)))
2268 aiagp = (struct iag *) amp->data;
2273 /* check if the iag has no free inodes. if so, the iag
2274 * will have to be added to the ag free inode list, so get
2275 * the iag at the head of the list in preparation for
2276 * adding this iag to this list. in doing this, we must
2277 * check if we already have the iag at the head of
2280 if (iagp->nfreeinos == 0) {
2281 freei = imap->im_agctl[agno].inofree;
2286 } else if (freei == back) {
2289 if ((rc = diIAGRead(imap, freei, &cmp)))
2291 ciagp = (struct iag *) cmp->data;
2293 if (ciagp == NULL) {
2294 jfs_error(imap->im_ipimap->i_sb,
2295 "diNewExt: ciagp == NULL");
2302 /* allocate disk space for the inode extent.
2304 if ((extno == 0) || (addressPXD(&iagp->inoext[extno - 1]) == 0))
2305 hint = ((s64) agno << sbi->bmap->db_agl2size) - 1;
2307 hint = addressPXD(&iagp->inoext[extno - 1]) +
2308 lengthPXD(&iagp->inoext[extno - 1]) - 1;
2310 if ((rc = dbAlloc(ipimap, hint, (s64) imap->im_nbperiext, &blkno)))
2313 /* compute the inode number of the first inode within the
2316 ino = (iagno << L2INOSPERIAG) + (extno << L2INOSPEREXT);
2318 /* initialize the inodes within the newly allocated extent a
2321 for (i = 0; i < imap->im_nbperiext; i += sbi->nbperpage) {
2322 /* get a buffer for this page of disk inodes.
2324 dmp = get_metapage(ipimap, blkno + i, PSIZE, 1);
2329 dp = (struct dinode *) dmp->data;
2331 /* initialize the inode number, mode, link count and
2332 * inode extent address.
2334 for (j = 0; j < INOSPERPAGE; j++, dp++, ino++) {
2335 dp->di_inostamp = cpu_to_le32(sbi->inostamp);
2336 dp->di_number = cpu_to_le32(ino);
2337 dp->di_fileset = cpu_to_le32(FILESYSTEM_I);
2340 PXDaddress(&(dp->di_ixpxd), blkno);
2341 PXDlength(&(dp->di_ixpxd), imap->im_nbperiext);
2343 write_metapage(dmp);
2346 /* if this is the last free extent within the iag, remove the
2347 * iag from the ag free extent list.
2349 if (iagp->nfreeexts == cpu_to_le32(1)) {
2351 aiagp->extfreeback = iagp->extfreeback;
2354 biagp->extfreefwd = iagp->extfreefwd;
2356 imap->im_agctl[agno].extfree =
2357 le32_to_cpu(iagp->extfreefwd);
2359 iagp->extfreefwd = iagp->extfreeback = -1;
2361 /* if the iag has all free extents (newly allocated iag),
2362 * add the iag to the ag free extent list.
2364 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2366 aiagp->extfreeback = cpu_to_le32(iagno);
2368 iagp->extfreefwd = cpu_to_le32(fwd);
2369 iagp->extfreeback = -1;
2370 imap->im_agctl[agno].extfree = iagno;
2374 /* if the iag has no free inodes, add the iag to the
2375 * ag free inode list.
2377 if (iagp->nfreeinos == 0) {
2379 ciagp->inofreeback = cpu_to_le32(iagno);
2382 cpu_to_le32(imap->im_agctl[agno].inofree);
2383 iagp->inofreeback = -1;
2384 imap->im_agctl[agno].inofree = iagno;
2387 /* initialize the extent descriptor of the extent. */
2388 PXDlength(&iagp->inoext[extno], imap->im_nbperiext);
2389 PXDaddress(&iagp->inoext[extno], blkno);
2391 /* initialize the working and persistent map of the extent.
2392 * the working map will be initialized such that
2393 * it indicates the first inode of the extent is allocated.
2395 iagp->wmap[extno] = cpu_to_le32(HIGHORDER);
2396 iagp->pmap[extno] = 0;
2398 /* update the free inode and free extent summary maps
2399 * for the extent to indicate the extent has free inodes
2400 * and no longer represents a free extent.
2402 sword = extno >> L2EXTSPERSUM;
2403 mask = HIGHORDER >> (extno & (EXTSPERSUM - 1));
2404 iagp->extsmap[sword] |= cpu_to_le32(mask);
2405 iagp->inosmap[sword] &= cpu_to_le32(~mask);
2407 /* update the free inode and free extent counts for the
2410 iagp->nfreeinos = cpu_to_le32(le32_to_cpu(iagp->nfreeinos) +
2412 iagp->nfreeexts = cpu_to_le32(le32_to_cpu(iagp->nfreeexts) - 1);
2414 /* update the free and backed inode counts for the ag.
2416 imap->im_agctl[agno].numfree += (INOSPEREXT - 1);
2417 imap->im_agctl[agno].numinos += INOSPEREXT;
2419 /* update the free and backed inode counts for the inode map.
2421 atomic_add(INOSPEREXT - 1, &imap->im_numfree);
2422 atomic_add(INOSPEREXT, &imap->im_numinos);
2427 write_metapage(amp);
2429 write_metapage(bmp);
2431 write_metapage(cmp);
2437 /* release the iags.
2440 release_metapage(amp);
2442 release_metapage(bmp);
2444 release_metapage(cmp);
2451 * NAME: diNewIAG(imap,iagnop,agno)
2453 * FUNCTION: allocate a new iag for an allocation group.
2455 * first tries to allocate the iag from the inode map
2457 * if the list has free iags, the head of the list is removed
2458 * and returned to satisfy the request.
2459 * if the inode map's iag free list is empty, the inode map
2460 * is extended to hold a new iag. this new iag is initialized
2461 * and returned to satisfy the request.
2464 * imap - pointer to inode map control structure.
2465 * iagnop - pointer to an iag number set with the number of the
2466 * newly allocated iag upon successful return.
2467 * agno - allocation group number.
2468 * bpp - Buffer pointer to be filled in with new IAG's buffer
2472 * -ENOSPC - insufficient disk resources.
2476 * AG lock held on entry/exit;
2477 * write lock on the map is held inside;
2478 * read lock on the map is held on successful completion;
2480 * note: new iag transaction:
2481 * . synchronously write iag;
2482 * . write log of xtree and inode of imap;
2484 * . synchronous write of xtree (right to left, bottom to top);
2485 * . at start of logredo(): init in-memory imap with one additional iag page;
2486 * . at end of logredo(): re-read imap inode to determine
2490 diNewIAG(struct inomap * imap, int *iagnop, int agno, struct metapage ** mpp)
2494 struct inode *ipimap;
2495 struct super_block *sb;
2496 struct jfs_sb_info *sbi;
2497 struct metapage *mp;
2502 #ifdef _STILL_TO_PORT
2504 #endif /* _STILL_TO_PORT */
2505 struct inode *iplist[1];
2507 /* pick up pointers to the inode map and mount inodes */
2508 ipimap = imap->im_ipimap;
2512 /* acquire the free iag lock */
2515 /* if there are any iags on the inode map free iag list,
2516 * allocate the iag from the head of the list.
2518 if (imap->im_freeiag >= 0) {
2519 /* pick up the iag number at the head of the list */
2520 iagno = imap->im_freeiag;
2522 /* determine the logical block number of the iag */
2523 blkno = IAGTOLBLK(iagno, sbi->l2nbperpage);
2525 /* no free iags. the inode map will have to be extented
2526 * to include a new iag.
2529 /* acquire inode map lock */
2530 IWRITE_LOCK(ipimap);
2532 if (ipimap->i_size >> L2PSIZE != imap->im_nextiag + 1) {
2533 IWRITE_UNLOCK(ipimap);
2534 IAGFREE_UNLOCK(imap);
2535 jfs_error(imap->im_ipimap->i_sb,
2536 "diNewIAG: ipimap->i_size is wrong");
2541 /* get the next avaliable iag number */
2542 iagno = imap->im_nextiag;
2544 /* make sure that we have not exceeded the maximum inode
2547 if (iagno > (MAXIAGS - 1)) {
2548 /* release the inode map lock */
2549 IWRITE_UNLOCK(ipimap);
2556 * synchronously append new iag page.
2558 /* determine the logical address of iag page to append */
2559 blkno = IAGTOLBLK(iagno, sbi->l2nbperpage);
2561 /* Allocate extent for new iag page */
2562 xlen = sbi->nbperpage;
2563 if ((rc = dbAlloc(ipimap, 0, (s64) xlen, &xaddr))) {
2564 /* release the inode map lock */
2565 IWRITE_UNLOCK(ipimap);
2570 /* assign a buffer for the page */
2571 mp = get_metapage(ipimap, xaddr, PSIZE, 1);
2573 /* Free the blocks allocated for the iag since it was
2574 * not successfully added to the inode map
2576 dbFree(ipimap, xaddr, (s64) xlen);
2578 /* release the inode map lock */
2579 IWRITE_UNLOCK(ipimap);
2584 iagp = (struct iag *) mp->data;
2587 memset(iagp, 0, sizeof(struct iag));
2588 iagp->iagnum = cpu_to_le32(iagno);
2589 iagp->inofreefwd = iagp->inofreeback = -1;
2590 iagp->extfreefwd = iagp->extfreeback = -1;
2592 iagp->nfreeinos = 0;
2593 iagp->nfreeexts = cpu_to_le32(EXTSPERIAG);
2595 /* initialize the free inode summary map (free extent
2596 * summary map initialization handled by bzero).
2598 for (i = 0; i < SMAPSZ; i++)
2599 iagp->inosmap[i] = ONES;
2602 #ifdef _STILL_TO_PORT
2603 /* synchronously write the iag page */
2605 /* Free the blocks allocated for the iag since it was
2606 * not successfully added to the inode map
2608 dbFree(ipimap, xaddr, (s64) xlen);
2610 /* release the inode map lock */
2611 IWRITE_UNLOCK(ipimap);
2617 /* Now the iag is on disk */
2620 * start tyransaction of update of the inode map
2621 * addressing structure pointing to the new iag page;
2623 #endif /* _STILL_TO_PORT */
2624 tid = txBegin(sb, COMMIT_FORCE);
2626 /* update the inode map addressing structure to point to it */
2628 xtInsert(tid, ipimap, 0, blkno, xlen, &xaddr, 0))) {
2629 /* Free the blocks allocated for the iag since it was
2630 * not successfully added to the inode map
2632 dbFree(ipimap, xaddr, (s64) xlen);
2634 /* release the inode map lock */
2635 IWRITE_UNLOCK(ipimap);
2640 /* update the inode map's inode to reflect the extension */
2641 ipimap->i_size += PSIZE;
2642 ipimap->i_blocks += LBLK2PBLK(sb, xlen);
2645 * txCommit(COMMIT_FORCE) will synchronously write address
2646 * index pages and inode after commit in careful update order
2647 * of address index pages (right to left, bottom up);
2650 rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE);
2654 duplicateIXtree(sb, blkno, xlen, &xaddr);
2656 /* update the next avaliable iag number */
2657 imap->im_nextiag += 1;
2659 /* Add the iag to the iag free list so we don't lose the iag
2660 * if a failure happens now.
2662 imap->im_freeiag = iagno;
2664 /* Until we have logredo working, we want the imap inode &
2665 * control page to be up to date.
2669 /* release the inode map lock */
2670 IWRITE_UNLOCK(ipimap);
2673 /* obtain read lock on map */
2677 if ((rc = diIAGRead(imap, iagno, &mp))) {
2678 IREAD_UNLOCK(ipimap);
2682 iagp = (struct iag *) mp->data;
2684 /* remove the iag from the iag free list */
2685 imap->im_freeiag = le32_to_cpu(iagp->iagfree);
2688 /* set the return iag number and buffer pointer */
2693 /* release the iag free lock */
2694 IAGFREE_UNLOCK(imap);
2702 * FUNCTION: get the buffer for the specified iag within a fileset
2703 * or aggregate inode map.
2706 * imap - pointer to inode map control structure.
2707 * iagno - iag number.
2708 * bpp - point to buffer pointer to be filled in on successful
2712 * must have read lock on imap inode
2713 * (When called by diExtendFS, the filesystem is quiesced, therefore
2714 * the read lock is unnecessary.)
2720 static int diIAGRead(struct inomap * imap, int iagno, struct metapage ** mpp)
2722 struct inode *ipimap = imap->im_ipimap;
2725 /* compute the logical block number of the iag. */
2726 blkno = IAGTOLBLK(iagno, JFS_SBI(ipimap->i_sb)->l2nbperpage);
2729 *mpp = read_metapage(ipimap, blkno, PSIZE, 0);
2738 * NAME: diFindFree()
2740 * FUNCTION: find the first free bit in a word starting at
2741 * the specified bit position.
2744 * word - word to be examined.
2745 * start - starting bit position.
2748 * bit position of first free bit in the word or 32 if
2749 * no free bits were found.
2751 static int diFindFree(u32 word, int start)
2755 /* scan the word for the first free bit. */
2756 for (word <<= start, bitno = start; bitno < 32;
2757 bitno++, word <<= 1) {
2758 if ((word & HIGHORDER) == 0)
2765 * NAME: diUpdatePMap()
2767 * FUNCTION: Update the persistent map in an IAG for the allocation or
2768 * freeing of the specified inode.
2770 * PRE CONDITIONS: Working map has already been updated for allocate.
2773 * ipimap - Incore inode map inode
2774 * inum - Number of inode to mark in permanent map
2775 * is_free - If TRUE indicates inode should be marked freed, otherwise
2776 * indicates inode should be marked allocated.
2782 diUpdatePMap(struct inode *ipimap,
2783 unsigned long inum, boolean_t is_free, struct tblock * tblk)
2787 struct metapage *mp;
2788 int iagno, ino, extno, bitno;
2789 struct inomap *imap;
2791 struct jfs_log *log;
2792 int lsn, difft, diffp;
2794 imap = JFS_IP(ipimap)->i_imap;
2795 /* get the iag number containing the inode */
2796 iagno = INOTOIAG(inum);
2797 /* make sure that the iag is contained within the map */
2798 if (iagno >= imap->im_nextiag) {
2799 jfs_error(ipimap->i_sb,
2800 "diUpdatePMap: the iag is outside the map");
2805 rc = diIAGRead(imap, iagno, &mp);
2806 IREAD_UNLOCK(ipimap);
2809 iagp = (struct iag *) mp->data;
2810 /* get the inode number and extent number of the inode within
2811 * the iag and the inode number within the extent.
2813 ino = inum & (INOSPERIAG - 1);
2814 extno = ino >> L2INOSPEREXT;
2815 bitno = ino & (INOSPEREXT - 1);
2816 mask = HIGHORDER >> bitno;
2818 * mark the inode free in persistent map:
2820 if (is_free == TRUE) {
2821 /* The inode should have been allocated both in working
2822 * map and in persistent map;
2823 * the inode will be freed from working map at the release
2824 * of last reference release;
2826 if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
2827 jfs_error(ipimap->i_sb,
2828 "diUpdatePMap: inode %ld not marked as "
2829 "allocated in wmap!", inum);
2831 if (!(le32_to_cpu(iagp->pmap[extno]) & mask)) {
2832 jfs_error(ipimap->i_sb,
2833 "diUpdatePMap: inode %ld not marked as "
2834 "allocated in pmap!", inum);
2836 /* update the bitmap for the extent of the freed inode */
2837 iagp->pmap[extno] &= cpu_to_le32(~mask);
2840 * mark the inode allocated in persistent map:
2843 /* The inode should be already allocated in the working map
2844 * and should be free in persistent map;
2846 if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
2847 release_metapage(mp);
2848 jfs_error(ipimap->i_sb,
2849 "diUpdatePMap: the inode is not allocated in "
2853 if ((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) {
2854 release_metapage(mp);
2855 jfs_error(ipimap->i_sb,
2856 "diUpdatePMap: the inode is not free in the "
2860 /* update the bitmap for the extent of the allocated inode */
2861 iagp->pmap[extno] |= cpu_to_le32(mask);
2867 log = JFS_SBI(tblk->sb)->log;
2869 /* inherit older/smaller lsn */
2870 logdiff(difft, lsn, log);
2871 logdiff(diffp, mp->lsn, log);
2872 if (difft < diffp) {
2874 /* move mp after tblock in logsync list */
2876 list_move(&mp->synclist, &tblk->synclist);
2877 LOGSYNC_UNLOCK(log);
2879 /* inherit younger/larger clsn */
2882 logdiff(difft, tblk->clsn, log);
2883 logdiff(diffp, mp->clsn, log);
2885 mp->clsn = tblk->clsn;
2886 LOGSYNC_UNLOCK(log);
2890 /* insert mp after tblock in logsync list */
2893 list_add(&mp->synclist, &tblk->synclist);
2894 mp->clsn = tblk->clsn;
2895 LOGSYNC_UNLOCK(log);
2904 * function: update imap for extendfs();
2906 * note: AG size has been increased s.t. each k old contiguous AGs are
2907 * coalesced into a new AG;
2909 int diExtendFS(struct inode *ipimap, struct inode *ipbmap)
2912 struct inomap *imap = JFS_IP(ipimap)->i_imap;
2913 struct iag *iagp = 0, *hiagp = 0;
2914 struct bmap *mp = JFS_SBI(ipbmap->i_sb)->bmap;
2915 struct metapage *bp, *hbp;
2917 int numinos, xnuminos = 0, xnumfree = 0;
2920 jfs_info("diExtendFS: nextiag:%d numinos:%d numfree:%d",
2921 imap->im_nextiag, atomic_read(&imap->im_numinos),
2922 atomic_read(&imap->im_numfree));
2927 * coalesce contiguous k (newAGSize/oldAGSize) AGs;
2928 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn;
2929 * note: new AG size = old AG size * (2**x).
2932 /* init per AG control information im_agctl[] */
2933 for (i = 0; i < MAXAG; i++) {
2934 imap->im_agctl[i].inofree = -1; /* free inode list */
2935 imap->im_agctl[i].extfree = -1; /* free extent list */
2936 imap->im_agctl[i].numinos = 0; /* number of backed inodes */
2937 imap->im_agctl[i].numfree = 0; /* number of free backed inodes */
2941 * process each iag page of the map.
2943 * rebuild AG Free Inode List, AG Free Inode Extent List;
2945 for (i = 0; i < imap->im_nextiag; i++) {
2946 if ((rc = diIAGRead(imap, i, &bp))) {
2950 iagp = (struct iag *) bp->data;
2951 if (le32_to_cpu(iagp->iagnum) != i) {
2952 release_metapage(bp);
2953 jfs_error(ipimap->i_sb,
2954 "diExtendFs: unexpected value of iagnum");
2958 /* leave free iag in the free iag list */
2959 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2960 release_metapage(bp);
2964 /* agstart that computes to the same ag is treated as same; */
2965 agstart = le64_to_cpu(iagp->agstart);
2966 /* iagp->agstart = agstart & ~(mp->db_agsize - 1); */
2967 n = agstart >> mp->db_agl2size;
2969 /* compute backed inodes */
2970 numinos = (EXTSPERIAG - le32_to_cpu(iagp->nfreeexts))
2973 /* merge AG backed inodes */
2974 imap->im_agctl[n].numinos += numinos;
2975 xnuminos += numinos;
2978 /* if any backed free inodes, insert at AG free inode list */
2979 if ((int) le32_to_cpu(iagp->nfreeinos) > 0) {
2980 if ((head = imap->im_agctl[n].inofree) == -1)
2981 iagp->inofreefwd = iagp->inofreeback = -1;
2983 if ((rc = diIAGRead(imap, head, &hbp))) {
2987 hiagp = (struct iag *) hbp->data;
2988 hiagp->inofreeback =
2989 le32_to_cpu(iagp->iagnum);
2990 iagp->inofreefwd = cpu_to_le32(head);
2991 iagp->inofreeback = -1;
2992 write_metapage(hbp);
2995 imap->im_agctl[n].inofree =
2996 le32_to_cpu(iagp->iagnum);
2998 /* merge AG backed free inodes */
2999 imap->im_agctl[n].numfree +=
3000 le32_to_cpu(iagp->nfreeinos);
3001 xnumfree += le32_to_cpu(iagp->nfreeinos);
3004 /* if any free extents, insert at AG free extent list */
3005 if (le32_to_cpu(iagp->nfreeexts) > 0) {
3006 if ((head = imap->im_agctl[n].extfree) == -1)
3007 iagp->extfreefwd = iagp->extfreeback = -1;
3009 if ((rc = diIAGRead(imap, head, &hbp))) {
3013 hiagp = (struct iag *) hbp->data;
3014 hiagp->extfreeback = iagp->iagnum;
3015 iagp->extfreefwd = cpu_to_le32(head);
3016 iagp->extfreeback = -1;
3017 write_metapage(hbp);
3020 imap->im_agctl[n].extfree =
3021 le32_to_cpu(iagp->iagnum);
3028 if (xnuminos != atomic_read(&imap->im_numinos) ||
3029 xnumfree != atomic_read(&imap->im_numfree)) {
3030 jfs_error(ipimap->i_sb,
3031 "diExtendFs: numinos or numfree incorrect");
3042 * serialization: IWRITE_LOCK held on entry/exit
3044 * note: shadow page with regular inode (rel.2);
3046 static void duplicateIXtree(struct super_block *sb, s64 blkno,
3047 int xlen, s64 *xaddr)
3049 struct jfs_superblock *j_sb;
3050 struct buffer_head *bh;
3054 /* if AIT2 ipmap2 is bad, do not try to update it */
3055 if (JFS_SBI(sb)->mntflag & JFS_BAD_SAIT) /* s_flag */
3057 ip = diReadSpecial(sb, FILESYSTEM_I, 1);
3059 JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT;
3060 if (readSuper(sb, &bh))
3062 j_sb = (struct jfs_superblock *)bh->b_data;
3063 j_sb->s_flag |= JFS_BAD_SAIT;
3065 mark_buffer_dirty(bh);
3066 sync_dirty_buffer(bh);
3071 /* start transaction */
3072 tid = txBegin(sb, COMMIT_FORCE);
3073 /* update the inode map addressing structure to point to it */
3074 if (xtInsert(tid, ip, 0, blkno, xlen, xaddr, 0)) {
3075 JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT;
3080 /* update the inode map's inode to reflect the extension */
3081 ip->i_size += PSIZE;
3082 ip->i_blocks += LBLK2PBLK(sb, xlen);
3083 txCommit(tid, 1, &ip, COMMIT_FORCE);
3090 * NAME: copy_from_dinode()
3092 * FUNCTION: Copies inode info from disk inode to in-memory inode
3096 * -ENOMEM - insufficient memory
3098 static int copy_from_dinode(struct dinode * dip, struct inode *ip)
3100 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
3102 jfs_ip->fileset = le32_to_cpu(dip->di_fileset);
3103 jfs_ip->mode2 = le32_to_cpu(dip->di_mode);
3105 ip->i_mode = le32_to_cpu(dip->di_mode) & 0xffff;
3106 ip->i_nlink = le32_to_cpu(dip->di_nlink);
3107 ip->i_uid = le32_to_cpu(dip->di_uid);
3108 ip->i_gid = le32_to_cpu(dip->di_gid);
3109 ip->i_size = le64_to_cpu(dip->di_size);
3110 ip->i_atime.tv_sec = le32_to_cpu(dip->di_atime.tv_sec);
3111 ip->i_atime.tv_nsec = le32_to_cpu(dip->di_atime.tv_nsec);
3112 ip->i_mtime.tv_sec = le32_to_cpu(dip->di_mtime.tv_sec);
3113 ip->i_mtime.tv_nsec = le32_to_cpu(dip->di_mtime.tv_nsec);
3114 ip->i_ctime.tv_sec = le32_to_cpu(dip->di_ctime.tv_sec);
3115 ip->i_ctime.tv_nsec = le32_to_cpu(dip->di_ctime.tv_nsec);
3116 ip->i_blksize = ip->i_sb->s_blocksize;
3117 ip->i_blocks = LBLK2PBLK(ip->i_sb, le64_to_cpu(dip->di_nblocks));
3118 ip->i_generation = le32_to_cpu(dip->di_gen);
3120 jfs_ip->ixpxd = dip->di_ixpxd; /* in-memory pxd's are little-endian */
3121 jfs_ip->acl = dip->di_acl; /* as are dxd's */
3122 jfs_ip->ea = dip->di_ea;
3123 jfs_ip->next_index = le32_to_cpu(dip->di_next_index);
3124 jfs_ip->otime = le32_to_cpu(dip->di_otime.tv_sec);
3125 jfs_ip->acltype = le32_to_cpu(dip->di_acltype);
3127 if (S_ISCHR(ip->i_mode) || S_ISBLK(ip->i_mode)) {
3128 jfs_ip->dev = le32_to_cpu(dip->di_rdev);
3129 ip->i_rdev = new_decode_dev(jfs_ip->dev);
3132 if (S_ISDIR(ip->i_mode)) {
3133 memcpy(&jfs_ip->i_dirtable, &dip->di_dirtable, 384);
3134 } else if (S_ISREG(ip->i_mode) || S_ISLNK(ip->i_mode)) {
3135 memcpy(&jfs_ip->i_xtroot, &dip->di_xtroot, 288);
3137 memcpy(&jfs_ip->i_inline_ea, &dip->di_inlineea, 128);
3139 /* Zero the in-memory-only stuff */
3141 jfs_ip->btindex = 0;
3142 jfs_ip->btorder = 0;
3145 jfs_ip->atlhead = 0;
3146 jfs_ip->atltail = 0;
3152 * NAME: copy_to_dinode()
3154 * FUNCTION: Copies inode info from in-memory inode to disk inode
3156 static void copy_to_dinode(struct dinode * dip, struct inode *ip)
3158 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
3160 dip->di_fileset = cpu_to_le32(jfs_ip->fileset);
3161 dip->di_inostamp = cpu_to_le32(JFS_SBI(ip->i_sb)->inostamp);
3162 dip->di_number = cpu_to_le32(ip->i_ino);
3163 dip->di_gen = cpu_to_le32(ip->i_generation);
3164 dip->di_size = cpu_to_le64(ip->i_size);
3165 dip->di_nblocks = cpu_to_le64(PBLK2LBLK(ip->i_sb, ip->i_blocks));
3166 dip->di_nlink = cpu_to_le32(ip->i_nlink);
3167 dip->di_uid = cpu_to_le32(ip->i_uid);
3168 dip->di_gid = cpu_to_le32(ip->i_gid);
3170 * mode2 is only needed for storing the higher order bits.
3171 * Trust i_mode for the lower order ones
3173 dip->di_mode = cpu_to_le32((jfs_ip->mode2 & 0xffff0000) | ip->i_mode);
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);
3191 #ifdef _JFS_DEBUG_IMAP
3195 static void *DBGdiInit(struct inomap * imap)
3200 if ((dimap = (u32 *) xmalloc(size, L2PSIZE, kernel_heap)) == NULL)
3202 bzero((void *) dimap, size);
3203 imap->im_DBGdimap = dimap;
3209 static void DBGdiAlloc(struct inomap * imap, ino_t ino)
3211 u32 *dimap = imap->im_DBGdimap;
3216 m = 0x80000000 >> b;
3217 assert(w < 64 * 256);
3219 printk("DEBUG diAlloc: duplicate alloc ino:0x%x\n", ino);
3227 static void DBGdiFree(struct inomap * imap, ino_t ino)
3229 u32 *dimap = imap->im_DBGdimap;
3234 m = 0x80000000 >> b;
3235 assert(w < 64 * 256);
3236 if ((dimap[w] & m) == 0) {
3237 printk("DEBUG diFree: duplicate free ino:0x%x\n", ino);
3242 static void dump_cp(struct inomap * ipimap, char *function, int line)
3244 printk("\n* ********* *\nControl Page %s %d\n", function, line);
3245 printk("FreeIAG %d\tNextIAG %d\n", ipimap->im_freeiag,
3246 ipimap->im_nextiag);
3247 printk("NumInos %d\tNumFree %d\n",
3248 atomic_read(&ipimap->im_numinos),
3249 atomic_read(&ipimap->im_numfree));
3250 printk("AG InoFree %d\tAG ExtFree %d\n",
3251 ipimap->im_agctl[0].inofree, ipimap->im_agctl[0].extfree);
3252 printk("AG NumInos %d\tAG NumFree %d\n",
3253 ipimap->im_agctl[0].numinos, ipimap->im_agctl[0].numfree);
3256 static void dump_iag(struct iag * iag, char *function, int line)
3258 printk("\n* ********* *\nIAG %s %d\n", function, line);
3259 printk("IagNum %d\tIAG Free %d\n", le32_to_cpu(iag->iagnum),
3260 le32_to_cpu(iag->iagfree));
3261 printk("InoFreeFwd %d\tInoFreeBack %d\n",
3262 le32_to_cpu(iag->inofreefwd),
3263 le32_to_cpu(iag->inofreeback));
3264 printk("ExtFreeFwd %d\tExtFreeBack %d\n",
3265 le32_to_cpu(iag->extfreefwd),
3266 le32_to_cpu(iag->extfreeback));
3267 printk("NFreeInos %d\tNFreeExts %d\n", le32_to_cpu(iag->nfreeinos),
3268 le32_to_cpu(iag->nfreeexts));
3270 #endif /* _JFS_DEBUG_IMAP */