2 * Copyright (C) International Business Machines Corp., 2000-2004
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>
47 #include <linux/quotaops.h>
48 #include <linux/slab.h>
50 #include "jfs_incore.h"
51 #include "jfs_inode.h"
52 #include "jfs_filsys.h"
53 #include "jfs_dinode.h"
56 #include "jfs_metapage.h"
57 #include "jfs_superblock.h"
58 #include "jfs_debug.h"
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)
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])
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 *);
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 *);
93 * FUNCTION: initialize the incore inode map control structures for
94 * a fileset or aggregate init time.
96 * the inode map's control structure (dinomap) is
97 * brought in from disk and placed in virtual memory.
100 * ipimap - pointer to inode map inode for the aggregate or fileset.
104 * -ENOMEM - insufficient free virtual memory.
107 int diMount(struct inode *ipimap)
112 struct dinomap_disk *dinom_le;
115 * allocate/initialize the in-memory inode map control structure
117 /* allocate the in-memory inode map control structure. */
118 imap = kmalloc(sizeof(struct inomap), GFP_KERNEL);
120 jfs_err("diMount: kmalloc returned NULL!");
124 /* read the on-disk inode map control structure. */
126 mp = read_metapage(ipimap,
127 IMAPBLKNO << JFS_SBI(ipimap->i_sb)->l2nbperpage,
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);
153 /* release the buffer. */
154 release_metapage(mp);
157 * allocate/initialize inode allocation map locks
159 /* allocate and init iag free list lock */
160 IAGFREE_LOCK_INIT(imap);
162 /* allocate and init ag list locks */
163 for (index = 0; index < MAXAG; index++) {
164 AG_LOCK_INIT(imap, index);
167 /* bind the inode map inode and inode map control structure
170 imap->im_ipimap = ipimap;
171 JFS_IP(ipimap)->i_imap = imap;
180 * FUNCTION: write to disk the incore inode map control structures for
181 * a fileset or aggregate at unmount time.
184 * ipimap - pointer to inode map inode for the aggregate or fileset.
188 * -ENOMEM - insufficient free virtual memory.
191 int diUnmount(struct inode *ipimap, int mounterror)
193 struct inomap *imap = JFS_IP(ipimap)->i_imap;
196 * update the on-disk inode map control structure
199 if (!(mounterror || isReadOnly(ipimap)))
203 * Invalidate the page cache buffers
205 truncate_inode_pages(ipimap->i_mapping, 0);
208 * free in-memory control structure
211 JFS_IP(ipimap)->i_imap = NULL;
220 int diSync(struct inode *ipimap)
222 struct dinomap_disk *dinom_le;
223 struct inomap *imp = JFS_IP(ipimap)->i_imap;
228 * write imap global conrol page
230 /* read the on-disk inode map control structure */
231 mp = get_metapage(ipimap,
232 IMAPBLKNO << JFS_SBI(ipimap->i_sb)->l2nbperpage,
235 jfs_err("diSync: get_metapage failed!");
239 /* copy the in-memory version to the on-disk version */
240 dinom_le = (struct dinomap_disk *) mp->data;
241 dinom_le->in_freeiag = cpu_to_le32(imp->im_freeiag);
242 dinom_le->in_nextiag = cpu_to_le32(imp->im_nextiag);
243 dinom_le->in_numinos = cpu_to_le32(atomic_read(&imp->im_numinos));
244 dinom_le->in_numfree = cpu_to_le32(atomic_read(&imp->im_numfree));
245 dinom_le->in_nbperiext = cpu_to_le32(imp->im_nbperiext);
246 dinom_le->in_l2nbperiext = cpu_to_le32(imp->im_l2nbperiext);
247 for (index = 0; index < MAXAG; index++) {
248 dinom_le->in_agctl[index].inofree =
249 cpu_to_le32(imp->im_agctl[index].inofree);
250 dinom_le->in_agctl[index].extfree =
251 cpu_to_le32(imp->im_agctl[index].extfree);
252 dinom_le->in_agctl[index].numinos =
253 cpu_to_le32(imp->im_agctl[index].numinos);
254 dinom_le->in_agctl[index].numfree =
255 cpu_to_le32(imp->im_agctl[index].numfree);
258 /* write out the control structure */
262 * write out dirty pages of imap
264 filemap_write_and_wait(ipimap->i_mapping);
266 diWriteSpecial(ipimap, 0);
275 * FUNCTION: initialize an incore inode from disk.
277 * on entry, the specifed incore inode should itself
278 * specify the disk inode number corresponding to the
279 * incore inode (i.e. i_number should be initialized).
281 * this routine handles incore inode initialization for
282 * both "special" and "regular" inodes. special inodes
283 * are those required early in the mount process and
284 * require special handling since much of the file system
285 * is not yet initialized. these "special" inodes are
286 * identified by a NULL inode map inode pointer and are
287 * actually initialized by a call to diReadSpecial().
289 * for regular inodes, the iag describing the disk inode
290 * is read from disk to determine the inode extent address
291 * for the disk inode. with the inode extent address in
292 * hand, the page of the extent that contains the disk
293 * inode is read and the disk inode is copied to the
297 * ip - pointer to incore inode to be initialized from disk.
302 * -ENOMEM - insufficient memory
305 int diRead(struct inode *ip)
307 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
308 int iagno, ino, extno, rc;
309 struct inode *ipimap;
317 unsigned long pageno;
320 jfs_info("diRead: ino = %ld", ip->i_ino);
322 ipimap = sbi->ipimap;
323 JFS_IP(ip)->ipimap = ipimap;
325 /* determine the iag number for this inode (number) */
326 iagno = INOTOIAG(ip->i_ino);
329 imap = JFS_IP(ipimap)->i_imap;
330 IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
331 rc = diIAGRead(imap, iagno, &mp);
332 IREAD_UNLOCK(ipimap);
334 jfs_err("diRead: diIAGRead returned %d", rc);
338 iagp = (struct iag *) mp->data;
340 /* determine inode extent that holds the disk inode */
341 ino = ip->i_ino & (INOSPERIAG - 1);
342 extno = ino >> L2INOSPEREXT;
344 if ((lengthPXD(&iagp->inoext[extno]) != imap->im_nbperiext) ||
345 (addressPXD(&iagp->inoext[extno]) == 0)) {
346 release_metapage(mp);
350 /* get disk block number of the page within the inode extent
351 * that holds the disk inode.
353 blkno = INOPBLK(&iagp->inoext[extno], ino, sbi->l2nbperpage);
355 /* get the ag for the iag */
356 agstart = le64_to_cpu(iagp->agstart);
358 release_metapage(mp);
360 rel_inode = (ino & (INOSPERPAGE - 1));
361 pageno = blkno >> sbi->l2nbperpage;
363 if ((block_offset = ((u32) blkno & (sbi->nbperpage - 1)))) {
365 * OS/2 didn't always align inode extents on page boundaries
368 (sbi->nbperpage - block_offset) << sbi->l2niperblk;
370 if (rel_inode < inodes_left)
371 rel_inode += block_offset << sbi->l2niperblk;
374 rel_inode -= inodes_left;
378 /* read the page of disk inode */
379 mp = read_metapage(ipimap, pageno << sbi->l2nbperpage, PSIZE, 1);
381 jfs_err("diRead: read_metapage failed");
385 /* locate the disk inode requested */
386 dp = (struct dinode *) mp->data;
389 if (ip->i_ino != le32_to_cpu(dp->di_number)) {
390 jfs_error(ip->i_sb, "i_ino != di_number\n");
392 } else if (le32_to_cpu(dp->di_nlink) == 0)
395 /* copy the disk inode to the in-memory inode */
396 rc = copy_from_dinode(dp, ip);
398 release_metapage(mp);
400 /* set the ag for the inode */
401 JFS_IP(ip)->agstart = agstart;
402 JFS_IP(ip)->active_ag = -1;
409 * NAME: diReadSpecial()
411 * FUNCTION: initialize a 'special' inode from disk.
413 * this routines handles aggregate level inodes. The
414 * inode cache cannot differentiate between the
415 * aggregate inodes and the filesystem inodes, so we
416 * handle these here. We don't actually use the aggregate
417 * inode map, since these inodes are at a fixed location
418 * and in some cases the aggregate inode map isn't initialized
422 * sb - filesystem superblock
423 * inum - aggregate inode number
424 * secondary - 1 if secondary aggregate inode table
427 * new inode - success
430 struct inode *diReadSpecial(struct super_block *sb, ino_t inum, int secondary)
432 struct jfs_sb_info *sbi = JFS_SBI(sb);
440 jfs_err("diReadSpecial: new_inode returned NULL!");
445 address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage;
446 JFS_IP(ip)->ipimap = sbi->ipaimap2;
448 address = AITBL_OFF >> L2PSIZE;
449 JFS_IP(ip)->ipimap = sbi->ipaimap;
452 ASSERT(inum < INOSPEREXT);
456 address += inum >> 3; /* 8 inodes per 4K page */
458 /* read the page of fixed disk inode (AIT) in raw mode */
459 mp = read_metapage(ip, address << sbi->l2nbperpage, PSIZE, 1);
461 set_nlink(ip, 1); /* Don't want iput() deleting it */
466 /* get the pointer to the disk inode of interest */
467 dp = (struct dinode *) (mp->data);
468 dp += inum % 8; /* 8 inodes per 4K page */
470 /* copy on-disk inode to in-memory inode */
471 if ((copy_from_dinode(dp, ip)) != 0) {
472 /* handle bad return by returning NULL for ip */
473 set_nlink(ip, 1); /* Don't want iput() deleting it */
475 /* release the page */
476 release_metapage(mp);
481 ip->i_mapping->a_ops = &jfs_metapage_aops;
482 mapping_set_gfp_mask(ip->i_mapping, GFP_NOFS);
484 /* Allocations to metadata inodes should not affect quotas */
485 ip->i_flags |= S_NOQUOTA;
487 if ((inum == FILESYSTEM_I) && (JFS_IP(ip)->ipimap == sbi->ipaimap)) {
488 sbi->gengen = le32_to_cpu(dp->di_gengen);
489 sbi->inostamp = le32_to_cpu(dp->di_inostamp);
492 /* release the page */
493 release_metapage(mp);
501 * NAME: diWriteSpecial()
503 * FUNCTION: Write the special inode to disk
507 * secondary - 1 if secondary aggregate inode table
509 * RETURN VALUES: none
512 void diWriteSpecial(struct inode *ip, int secondary)
514 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
517 ino_t inum = ip->i_ino;
521 address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage;
523 address = AITBL_OFF >> L2PSIZE;
525 ASSERT(inum < INOSPEREXT);
527 address += inum >> 3; /* 8 inodes per 4K page */
529 /* read the page of fixed disk inode (AIT) in raw mode */
530 mp = read_metapage(ip, address << sbi->l2nbperpage, PSIZE, 1);
532 jfs_err("diWriteSpecial: failed to read aggregate inode extent!");
536 /* get the pointer to the disk inode of interest */
537 dp = (struct dinode *) (mp->data);
538 dp += inum % 8; /* 8 inodes per 4K page */
540 /* copy on-disk inode to in-memory inode */
541 copy_to_dinode(dp, ip);
542 memcpy(&dp->di_xtroot, &JFS_IP(ip)->i_xtroot, 288);
544 if (inum == FILESYSTEM_I)
545 dp->di_gengen = cpu_to_le32(sbi->gengen);
552 * NAME: diFreeSpecial()
554 * FUNCTION: Free allocated space for special inode
556 void diFreeSpecial(struct inode *ip)
559 jfs_err("diFreeSpecial called with NULL ip!");
562 filemap_write_and_wait(ip->i_mapping);
563 truncate_inode_pages(ip->i_mapping, 0);
572 * FUNCTION: write the on-disk inode portion of the in-memory inode
573 * to its corresponding on-disk inode.
575 * on entry, the specifed incore inode should itself
576 * specify the disk inode number corresponding to the
577 * incore inode (i.e. i_number should be initialized).
579 * the inode contains the inode extent address for the disk
580 * inode. with the inode extent address in hand, the
581 * page of the extent that contains the disk inode is
582 * read and the disk inode portion of the incore inode
583 * is copied to the disk inode.
586 * tid - transacation id
587 * ip - pointer to incore inode to be written to the inode extent.
593 int diWrite(tid_t tid, struct inode *ip)
595 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
596 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
604 unsigned long pageno;
607 struct inode *ipimap;
610 struct tlock *ditlck, *tlck;
611 struct linelock *dilinelock, *ilinelock;
615 ipimap = jfs_ip->ipimap;
617 ino = ip->i_ino & (INOSPERIAG - 1);
619 if (!addressPXD(&(jfs_ip->ixpxd)) ||
620 (lengthPXD(&(jfs_ip->ixpxd)) !=
621 JFS_IP(ipimap)->i_imap->im_nbperiext)) {
622 jfs_error(ip->i_sb, "ixpxd invalid\n");
627 * read the page of disk inode containing the specified inode:
629 /* compute the block address of the page */
630 blkno = INOPBLK(&(jfs_ip->ixpxd), ino, sbi->l2nbperpage);
632 rel_inode = (ino & (INOSPERPAGE - 1));
633 pageno = blkno >> sbi->l2nbperpage;
635 if ((block_offset = ((u32) blkno & (sbi->nbperpage - 1)))) {
637 * OS/2 didn't always align inode extents on page boundaries
640 (sbi->nbperpage - block_offset) << sbi->l2niperblk;
642 if (rel_inode < inodes_left)
643 rel_inode += block_offset << sbi->l2niperblk;
646 rel_inode -= inodes_left;
649 /* read the page of disk inode */
651 mp = read_metapage(ipimap, pageno << sbi->l2nbperpage, PSIZE, 1);
655 /* get the pointer to the disk inode */
656 dp = (struct dinode *) mp->data;
659 dioffset = (ino & (INOSPERPAGE - 1)) << L2DISIZE;
662 * acquire transaction lock on the on-disk inode;
663 * N.B. tlock is acquired on ipimap not ip;
666 txLock(tid, ipimap, mp, tlckINODE | tlckENTRY)) == NULL)
668 dilinelock = (struct linelock *) & ditlck->lock;
671 * copy btree root from in-memory inode to on-disk inode
673 * (tlock is taken from inline B+-tree root in in-memory
674 * inode when the B+-tree root is updated, which is pointed
675 * by jfs_ip->blid as well as being on tx tlock list)
677 * further processing of btree root is based on the copy
678 * in in-memory inode, where txLog() will log from, and,
679 * for xtree root, txUpdateMap() will update map and reset
683 if (S_ISDIR(ip->i_mode) && (lid = jfs_ip->xtlid)) {
685 * This is the special xtree inside the directory for storing
686 * the directory table
692 tlck = lid_to_tlock(lid);
693 assert(tlck->type & tlckXTREE);
694 tlck->type |= tlckBTROOT;
696 ilinelock = (struct linelock *) & tlck->lock;
699 * copy xtree root from inode to dinode:
701 p = &jfs_ip->i_xtroot;
702 xp = (xtpage_t *) &dp->di_dirtable;
704 for (n = 0; n < ilinelock->index; n++, lv++) {
705 memcpy(&xp->xad[lv->offset], &p->xad[lv->offset],
706 lv->length << L2XTSLOTSIZE);
709 /* reset on-disk (metadata page) xtree XAD_NEW bit */
710 xad = &xp->xad[XTENTRYSTART];
711 for (n = XTENTRYSTART;
712 n < le16_to_cpu(xp->header.nextindex); n++, xad++)
713 if (xad->flag & (XAD_NEW | XAD_EXTENDED))
714 xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
717 if ((lid = jfs_ip->blid) == 0)
721 tlck = lid_to_tlock(lid);
723 tlck->type |= tlckBTROOT;
725 ilinelock = (struct linelock *) & tlck->lock;
728 * regular file: 16 byte (XAD slot) granularity
730 if (type & tlckXTREE) {
735 * copy xtree root from inode to dinode:
737 p = &jfs_ip->i_xtroot;
740 for (n = 0; n < ilinelock->index; n++, lv++) {
741 memcpy(&xp->xad[lv->offset], &p->xad[lv->offset],
742 lv->length << L2XTSLOTSIZE);
745 /* reset on-disk (metadata page) xtree XAD_NEW bit */
746 xad = &xp->xad[XTENTRYSTART];
747 for (n = XTENTRYSTART;
748 n < le16_to_cpu(xp->header.nextindex); n++, xad++)
749 if (xad->flag & (XAD_NEW | XAD_EXTENDED))
750 xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
753 * directory: 32 byte (directory entry slot) granularity
755 else if (type & tlckDTREE) {
759 * copy dtree root from inode to dinode:
761 p = (dtpage_t *) &jfs_ip->i_dtroot;
762 xp = (dtpage_t *) & dp->di_dtroot;
764 for (n = 0; n < ilinelock->index; n++, lv++) {
765 memcpy(&xp->slot[lv->offset], &p->slot[lv->offset],
766 lv->length << L2DTSLOTSIZE);
769 jfs_err("diWrite: UFO tlock");
774 * copy inline symlink from in-memory inode to on-disk inode
776 if (S_ISLNK(ip->i_mode) && ip->i_size < IDATASIZE) {
777 lv = & dilinelock->lv[dilinelock->index];
778 lv->offset = (dioffset + 2 * 128) >> L2INODESLOTSIZE;
780 memcpy(&dp->di_fastsymlink, jfs_ip->i_inline, IDATASIZE);
784 * copy inline data from in-memory inode to on-disk inode:
785 * 128 byte slot granularity
787 if (test_cflag(COMMIT_Inlineea, ip)) {
788 lv = & dilinelock->lv[dilinelock->index];
789 lv->offset = (dioffset + 3 * 128) >> L2INODESLOTSIZE;
791 memcpy(&dp->di_inlineea, jfs_ip->i_inline_ea, INODESLOTSIZE);
794 clear_cflag(COMMIT_Inlineea, ip);
798 * lock/copy inode base: 128 byte slot granularity
800 lv = & dilinelock->lv[dilinelock->index];
801 lv->offset = dioffset >> L2INODESLOTSIZE;
802 copy_to_dinode(dp, ip);
803 if (test_and_clear_cflag(COMMIT_Dirtable, ip)) {
805 memcpy(&dp->di_dirtable, &jfs_ip->i_dirtable, 96);
810 /* release the buffer holding the updated on-disk inode.
811 * the buffer will be later written by commit processing.
822 * FUNCTION: free a specified inode from the inode working map
823 * for a fileset or aggregate.
825 * if the inode to be freed represents the first (only)
826 * free inode within the iag, the iag will be placed on
827 * the ag free inode list.
829 * freeing the inode will cause the inode extent to be
830 * freed if the inode is the only allocated inode within
831 * the extent. in this case all the disk resource backing
832 * up the inode extent will be freed. in addition, the iag
833 * will be placed on the ag extent free list if the extent
834 * is the first free extent in the iag. if freeing the
835 * extent also means that no free inodes will exist for
836 * the iag, the iag will also be removed from the ag free
839 * the iag describing the inode will be freed if the extent
840 * is to be freed and it is the only backed extent within
841 * the iag. in this case, the iag will be removed from the
842 * ag free extent list and ag free inode list and placed on
843 * the inode map's free iag list.
845 * a careful update approach is used to provide consistency
846 * in the face of updates to multiple buffers. under this
847 * approach, all required buffers are obtained before making
848 * any updates and are held until all updates are complete.
851 * ip - inode to be freed.
857 int diFree(struct inode *ip)
860 ino_t inum = ip->i_ino;
861 struct iag *iagp, *aiagp, *biagp, *ciagp, *diagp;
862 struct metapage *mp, *amp, *bmp, *cmp, *dmp;
863 int iagno, ino, extno, bitno, sword, agno;
866 struct inode *ipimap = JFS_SBI(ip->i_sb)->ipimap;
867 struct inomap *imap = JFS_IP(ipimap)->i_imap;
870 struct inode *iplist[3];
872 struct pxd_lock *pxdlock;
875 * This is just to suppress compiler warnings. The same logic that
876 * references these variables is used to initialize them.
878 aiagp = biagp = ciagp = diagp = NULL;
880 /* get the iag number containing the inode.
882 iagno = INOTOIAG(inum);
884 /* make sure that the iag is contained within
887 if (iagno >= imap->im_nextiag) {
888 print_hex_dump(KERN_ERR, "imap: ", DUMP_PREFIX_ADDRESS, 16, 4,
890 jfs_error(ip->i_sb, "inum = %d, iagno = %d, nextiag = %d\n",
891 (uint) inum, iagno, imap->im_nextiag);
895 /* get the allocation group for this ino.
897 agno = BLKTOAG(JFS_IP(ip)->agstart, JFS_SBI(ip->i_sb));
899 /* Lock the AG specific inode map information
903 /* Obtain read lock in imap inode. Don't release it until we have
904 * read all of the IAG's that we are going to.
906 IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
910 if ((rc = diIAGRead(imap, iagno, &mp))) {
911 IREAD_UNLOCK(ipimap);
912 AG_UNLOCK(imap, agno);
915 iagp = (struct iag *) mp->data;
917 /* get the inode number and extent number of the inode within
918 * the iag and the inode number within the extent.
920 ino = inum & (INOSPERIAG - 1);
921 extno = ino >> L2INOSPEREXT;
922 bitno = ino & (INOSPEREXT - 1);
923 mask = HIGHORDER >> bitno;
925 if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
926 jfs_error(ip->i_sb, "wmap shows inode already free\n");
929 if (!addressPXD(&iagp->inoext[extno])) {
930 release_metapage(mp);
931 IREAD_UNLOCK(ipimap);
932 AG_UNLOCK(imap, agno);
933 jfs_error(ip->i_sb, "invalid inoext\n");
937 /* compute the bitmap for the extent reflecting the freed inode.
939 bitmap = le32_to_cpu(iagp->wmap[extno]) & ~mask;
941 if (imap->im_agctl[agno].numfree > imap->im_agctl[agno].numinos) {
942 release_metapage(mp);
943 IREAD_UNLOCK(ipimap);
944 AG_UNLOCK(imap, agno);
945 jfs_error(ip->i_sb, "numfree > numinos\n");
949 * inode extent still has some inodes or below low water mark:
950 * keep the inode extent;
953 imap->im_agctl[agno].numfree < 96 ||
954 (imap->im_agctl[agno].numfree < 288 &&
955 (((imap->im_agctl[agno].numfree * 100) /
956 imap->im_agctl[agno].numinos) <= 25))) {
957 /* if the iag currently has no free inodes (i.e.,
958 * the inode being freed is the first free inode of iag),
959 * insert the iag at head of the inode free list for the ag.
961 if (iagp->nfreeinos == 0) {
962 /* check if there are any iags on the ag inode
963 * free list. if so, read the first one so that
964 * we can link the current iag onto the list at
967 if ((fwd = imap->im_agctl[agno].inofree) >= 0) {
968 /* read the iag that currently is the head
971 if ((rc = diIAGRead(imap, fwd, &))) {
972 IREAD_UNLOCK(ipimap);
973 AG_UNLOCK(imap, agno);
974 release_metapage(mp);
977 aiagp = (struct iag *) amp->data;
979 /* make current head point back to the iag.
981 aiagp->inofreeback = cpu_to_le32(iagno);
986 /* iag points forward to current head and iag
987 * becomes the new head of the list.
990 cpu_to_le32(imap->im_agctl[agno].inofree);
991 iagp->inofreeback = cpu_to_le32(-1);
992 imap->im_agctl[agno].inofree = iagno;
994 IREAD_UNLOCK(ipimap);
996 /* update the free inode summary map for the extent if
997 * freeing the inode means the extent will now have free
998 * inodes (i.e., the inode being freed is the first free
1001 if (iagp->wmap[extno] == cpu_to_le32(ONES)) {
1002 sword = extno >> L2EXTSPERSUM;
1003 bitno = extno & (EXTSPERSUM - 1);
1004 iagp->inosmap[sword] &=
1005 cpu_to_le32(~(HIGHORDER >> bitno));
1008 /* update the bitmap.
1010 iagp->wmap[extno] = cpu_to_le32(bitmap);
1012 /* update the free inode counts at the iag, ag and
1015 le32_add_cpu(&iagp->nfreeinos, 1);
1016 imap->im_agctl[agno].numfree += 1;
1017 atomic_inc(&imap->im_numfree);
1019 /* release the AG inode map lock
1021 AG_UNLOCK(imap, agno);
1031 * inode extent has become free and above low water mark:
1032 * free the inode extent;
1036 * prepare to update iag list(s) (careful update step 1)
1038 amp = bmp = cmp = dmp = NULL;
1041 /* check if the iag currently has no free extents. if so,
1042 * it will be placed on the head of the ag extent free list.
1044 if (iagp->nfreeexts == 0) {
1045 /* check if the ag extent free list has any iags.
1046 * if so, read the iag at the head of the list now.
1047 * this (head) iag will be updated later to reflect
1048 * the addition of the current iag at the head of
1051 if ((fwd = imap->im_agctl[agno].extfree) >= 0) {
1052 if ((rc = diIAGRead(imap, fwd, &)))
1054 aiagp = (struct iag *) amp->data;
1057 /* iag has free extents. check if the addition of a free
1058 * extent will cause all extents to be free within this
1059 * iag. if so, the iag will be removed from the ag extent
1060 * free list and placed on the inode map's free iag list.
1062 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) {
1063 /* in preparation for removing the iag from the
1064 * ag extent free list, read the iags preceding
1065 * and following the iag on the ag extent free
1068 if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) {
1069 if ((rc = diIAGRead(imap, fwd, &)))
1071 aiagp = (struct iag *) amp->data;
1074 if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) {
1075 if ((rc = diIAGRead(imap, back, &bmp)))
1077 biagp = (struct iag *) bmp->data;
1082 /* remove the iag from the ag inode free list if freeing
1083 * this extent cause the iag to have no free inodes.
1085 if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) {
1086 int inofreeback = le32_to_cpu(iagp->inofreeback);
1087 int inofreefwd = le32_to_cpu(iagp->inofreefwd);
1089 /* in preparation for removing the iag from the
1090 * ag inode free list, read the iags preceding
1091 * and following the iag on the ag inode free
1092 * list. before reading these iags, we must make
1093 * sure that we already don't have them in hand
1094 * from up above, since re-reading an iag (buffer)
1095 * we are currently holding would cause a deadlock.
1097 if (inofreefwd >= 0) {
1099 if (inofreefwd == fwd)
1100 ciagp = (struct iag *) amp->data;
1101 else if (inofreefwd == back)
1102 ciagp = (struct iag *) bmp->data;
1105 diIAGRead(imap, inofreefwd, &cmp)))
1107 ciagp = (struct iag *) cmp->data;
1109 assert(ciagp != NULL);
1112 if (inofreeback >= 0) {
1113 if (inofreeback == fwd)
1114 diagp = (struct iag *) amp->data;
1115 else if (inofreeback == back)
1116 diagp = (struct iag *) bmp->data;
1119 diIAGRead(imap, inofreeback, &dmp)))
1121 diagp = (struct iag *) dmp->data;
1123 assert(diagp != NULL);
1127 IREAD_UNLOCK(ipimap);
1130 * invalidate any page of the inode extent freed from buffer cache;
1132 freepxd = iagp->inoext[extno];
1133 invalidate_pxd_metapages(ip, freepxd);
1136 * update iag list(s) (careful update step 2)
1138 /* add the iag to the ag extent free list if this is the
1139 * first free extent for the iag.
1141 if (iagp->nfreeexts == 0) {
1143 aiagp->extfreeback = cpu_to_le32(iagno);
1146 cpu_to_le32(imap->im_agctl[agno].extfree);
1147 iagp->extfreeback = cpu_to_le32(-1);
1148 imap->im_agctl[agno].extfree = iagno;
1150 /* remove the iag from the ag extent list if all extents
1151 * are now free and place it on the inode map iag free list.
1153 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) {
1155 aiagp->extfreeback = iagp->extfreeback;
1158 biagp->extfreefwd = iagp->extfreefwd;
1160 imap->im_agctl[agno].extfree =
1161 le32_to_cpu(iagp->extfreefwd);
1163 iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1);
1166 iagp->iagfree = cpu_to_le32(imap->im_freeiag);
1167 imap->im_freeiag = iagno;
1168 IAGFREE_UNLOCK(imap);
1172 /* remove the iag from the ag inode free list if freeing
1173 * this extent causes the iag to have no free inodes.
1175 if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) {
1176 if ((int) le32_to_cpu(iagp->inofreefwd) >= 0)
1177 ciagp->inofreeback = iagp->inofreeback;
1179 if ((int) le32_to_cpu(iagp->inofreeback) >= 0)
1180 diagp->inofreefwd = iagp->inofreefwd;
1182 imap->im_agctl[agno].inofree =
1183 le32_to_cpu(iagp->inofreefwd);
1185 iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1);
1188 /* update the inode extent address and working map
1189 * to reflect the free extent.
1190 * the permanent map should have been updated already
1191 * for the inode being freed.
1193 if (iagp->pmap[extno] != 0) {
1194 jfs_error(ip->i_sb, "the pmap does not show inode free\n");
1196 iagp->wmap[extno] = 0;
1197 PXDlength(&iagp->inoext[extno], 0);
1198 PXDaddress(&iagp->inoext[extno], 0);
1200 /* update the free extent and free inode summary maps
1201 * to reflect the freed extent.
1202 * the inode summary map is marked to indicate no inodes
1203 * available for the freed extent.
1205 sword = extno >> L2EXTSPERSUM;
1206 bitno = extno & (EXTSPERSUM - 1);
1207 mask = HIGHORDER >> bitno;
1208 iagp->inosmap[sword] |= cpu_to_le32(mask);
1209 iagp->extsmap[sword] &= cpu_to_le32(~mask);
1211 /* update the number of free inodes and number of free extents
1214 le32_add_cpu(&iagp->nfreeinos, -(INOSPEREXT - 1));
1215 le32_add_cpu(&iagp->nfreeexts, 1);
1217 /* update the number of free inodes and backed inodes
1218 * at the ag and inode map level.
1220 imap->im_agctl[agno].numfree -= (INOSPEREXT - 1);
1221 imap->im_agctl[agno].numinos -= INOSPEREXT;
1222 atomic_sub(INOSPEREXT - 1, &imap->im_numfree);
1223 atomic_sub(INOSPEREXT, &imap->im_numinos);
1226 write_metapage(amp);
1228 write_metapage(bmp);
1230 write_metapage(cmp);
1232 write_metapage(dmp);
1235 * start transaction to update block allocation map
1236 * for the inode extent freed;
1238 * N.B. AG_LOCK is released and iag will be released below, and
1239 * other thread may allocate inode from/reusing the ixad freed
1240 * BUT with new/different backing inode extent from the extent
1241 * to be freed by the transaction;
1243 tid = txBegin(ipimap->i_sb, COMMIT_FORCE);
1244 mutex_lock(&JFS_IP(ipimap)->commit_mutex);
1246 /* acquire tlock of the iag page of the freed ixad
1247 * to force the page NOHOMEOK (even though no data is
1248 * logged from the iag page) until NOREDOPAGE|FREEXTENT log
1249 * for the free of the extent is committed;
1250 * write FREEXTENT|NOREDOPAGE log record
1251 * N.B. linelock is overlaid as freed extent descriptor;
1253 tlck = txLock(tid, ipimap, mp, tlckINODE | tlckFREE);
1254 pxdlock = (struct pxd_lock *) & tlck->lock;
1255 pxdlock->flag = mlckFREEPXD;
1256 pxdlock->pxd = freepxd;
1264 * logredo needs the IAG number and IAG extent index in order
1265 * to ensure that the IMap is consistent. The least disruptive
1266 * way to pass these values through to the transaction manager
1267 * is in the iplist array.
1269 * It's not pretty, but it works.
1271 iplist[1] = (struct inode *) (size_t)iagno;
1272 iplist[2] = (struct inode *) (size_t)extno;
1274 rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE);
1277 mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
1279 /* unlock the AG inode map information */
1280 AG_UNLOCK(imap, agno);
1285 IREAD_UNLOCK(ipimap);
1288 release_metapage(amp);
1290 release_metapage(bmp);
1292 release_metapage(cmp);
1294 release_metapage(dmp);
1296 AG_UNLOCK(imap, agno);
1298 release_metapage(mp);
1304 * There are several places in the diAlloc* routines where we initialize
1308 diInitInode(struct inode *ip, int iagno, int ino, int extno, struct iag * iagp)
1310 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
1312 ip->i_ino = (iagno << L2INOSPERIAG) + ino;
1313 jfs_ip->ixpxd = iagp->inoext[extno];
1314 jfs_ip->agstart = le64_to_cpu(iagp->agstart);
1315 jfs_ip->active_ag = -1;
1320 * NAME: diAlloc(pip,dir,ip)
1322 * FUNCTION: allocate a disk inode from the inode working map
1323 * for a fileset or aggregate.
1326 * pip - pointer to incore inode for the parent inode.
1327 * dir - 'true' if the new disk inode is for a directory.
1328 * ip - pointer to a new inode
1332 * -ENOSPC - insufficient disk resources.
1335 int diAlloc(struct inode *pip, bool dir, struct inode *ip)
1337 int rc, ino, iagno, addext, extno, bitno, sword;
1338 int nwords, rem, i, agno, dn_numag;
1339 u32 mask, inosmap, extsmap;
1340 struct inode *ipimap;
1341 struct metapage *mp;
1344 struct inomap *imap;
1346 /* get the pointers to the inode map inode and the
1347 * corresponding imap control structure.
1349 ipimap = JFS_SBI(pip->i_sb)->ipimap;
1350 imap = JFS_IP(ipimap)->i_imap;
1351 JFS_IP(ip)->ipimap = ipimap;
1352 JFS_IP(ip)->fileset = FILESYSTEM_I;
1354 /* for a directory, the allocation policy is to start
1355 * at the ag level using the preferred ag.
1358 agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap);
1359 AG_LOCK(imap, agno);
1363 /* for files, the policy starts off by trying to allocate from
1364 * the same iag containing the parent disk inode:
1365 * try to allocate the new disk inode close to the parent disk
1366 * inode, using parent disk inode number + 1 as the allocation
1367 * hint. (we use a left-to-right policy to attempt to avoid
1368 * moving backward on the disk.) compute the hint within the
1369 * file system and the iag.
1372 /* get the ag number of this iag */
1373 agno = BLKTOAG(JFS_IP(pip)->agstart, JFS_SBI(pip->i_sb));
1374 dn_numag = JFS_SBI(pip->i_sb)->bmap->db_numag;
1375 if (agno < 0 || agno > dn_numag)
1378 if (atomic_read(&JFS_SBI(pip->i_sb)->bmap->db_active[agno])) {
1380 * There is an open file actively growing. We want to
1381 * allocate new inodes from a different ag to avoid
1382 * fragmentation problems.
1384 agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap);
1385 AG_LOCK(imap, agno);
1389 inum = pip->i_ino + 1;
1390 ino = inum & (INOSPERIAG - 1);
1392 /* back off the hint if it is outside of the iag */
1396 /* lock the AG inode map information */
1397 AG_LOCK(imap, agno);
1399 /* Get read lock on imap inode */
1400 IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
1402 /* get the iag number and read the iag */
1403 iagno = INOTOIAG(inum);
1404 if ((rc = diIAGRead(imap, iagno, &mp))) {
1405 IREAD_UNLOCK(ipimap);
1406 AG_UNLOCK(imap, agno);
1409 iagp = (struct iag *) mp->data;
1411 /* determine if new inode extent is allowed to be added to the iag.
1412 * new inode extent can be added to the iag if the ag
1413 * has less than 32 free disk inodes and the iag has free extents.
1415 addext = (imap->im_agctl[agno].numfree < 32 && iagp->nfreeexts);
1418 * try to allocate from the IAG
1420 /* check if the inode may be allocated from the iag
1421 * (i.e. the inode has free inodes or new extent can be added).
1423 if (iagp->nfreeinos || addext) {
1424 /* determine the extent number of the hint.
1426 extno = ino >> L2INOSPEREXT;
1428 /* check if the extent containing the hint has backed
1429 * inodes. if so, try to allocate within this extent.
1431 if (addressPXD(&iagp->inoext[extno])) {
1432 bitno = ino & (INOSPEREXT - 1);
1434 diFindFree(le32_to_cpu(iagp->wmap[extno]),
1437 ino = (extno << L2INOSPEREXT) + bitno;
1439 /* a free inode (bit) was found within this
1440 * extent, so allocate it.
1442 rc = diAllocBit(imap, iagp, ino);
1443 IREAD_UNLOCK(ipimap);
1447 /* set the results of the allocation
1448 * and write the iag.
1450 diInitInode(ip, iagno, ino, extno,
1452 mark_metapage_dirty(mp);
1454 release_metapage(mp);
1456 /* free the AG lock and return.
1458 AG_UNLOCK(imap, agno);
1465 EXTSPERIAG - 1) ? 0 : extno + 1;
1469 * no free inodes within the extent containing the hint.
1471 * try to allocate from the backed extents following
1472 * hint or, if appropriate (i.e. addext is true), allocate
1473 * an extent of free inodes at or following the extent
1474 * containing the hint.
1476 * the free inode and free extent summary maps are used
1477 * here, so determine the starting summary map position
1478 * and the number of words we'll have to examine. again,
1479 * the approach is to allocate following the hint, so we
1480 * might have to initially ignore prior bits of the summary
1481 * map that represent extents prior to the extent containing
1482 * the hint and later revisit these bits.
1484 bitno = extno & (EXTSPERSUM - 1);
1485 nwords = (bitno == 0) ? SMAPSZ : SMAPSZ + 1;
1486 sword = extno >> L2EXTSPERSUM;
1488 /* mask any prior bits for the starting words of the
1491 mask = (bitno == 0) ? 0 : (ONES << (EXTSPERSUM - bitno));
1492 inosmap = le32_to_cpu(iagp->inosmap[sword]) | mask;
1493 extsmap = le32_to_cpu(iagp->extsmap[sword]) | mask;
1495 /* scan the free inode and free extent summary maps for
1498 for (i = 0; i < nwords; i++) {
1499 /* check if this word of the free inode summary
1500 * map describes an extent with free inodes.
1503 /* an extent with free inodes has been
1504 * found. determine the extent number
1505 * and the inode number within the extent.
1507 rem = diFindFree(inosmap, 0);
1508 extno = (sword << L2EXTSPERSUM) + rem;
1509 rem = diFindFree(le32_to_cpu(iagp->wmap[extno]),
1511 if (rem >= INOSPEREXT) {
1512 IREAD_UNLOCK(ipimap);
1513 release_metapage(mp);
1514 AG_UNLOCK(imap, agno);
1516 "can't find free bit in wmap\n");
1520 /* determine the inode number within the
1521 * iag and allocate the inode from the
1524 ino = (extno << L2INOSPEREXT) + rem;
1525 rc = diAllocBit(imap, iagp, ino);
1526 IREAD_UNLOCK(ipimap);
1530 /* set the results of the allocation
1531 * and write the iag.
1533 diInitInode(ip, iagno, ino, extno,
1535 mark_metapage_dirty(mp);
1537 release_metapage(mp);
1539 /* free the AG lock and return.
1541 AG_UNLOCK(imap, agno);
1546 /* check if we may allocate an extent of free
1547 * inodes and whether this word of the free
1548 * extents summary map describes a free extent.
1550 if (addext && ~extsmap) {
1551 /* a free extent has been found. determine
1552 * the extent number.
1554 rem = diFindFree(extsmap, 0);
1555 extno = (sword << L2EXTSPERSUM) + rem;
1557 /* allocate an extent of free inodes.
1559 if ((rc = diNewExt(imap, iagp, extno))) {
1560 /* if there is no disk space for a
1561 * new extent, try to allocate the
1562 * disk inode from somewhere else.
1569 /* set the results of the allocation
1570 * and write the iag.
1572 diInitInode(ip, iagno,
1573 extno << L2INOSPEREXT,
1575 mark_metapage_dirty(mp);
1577 release_metapage(mp);
1578 /* free the imap inode & the AG lock & return.
1580 IREAD_UNLOCK(ipimap);
1581 AG_UNLOCK(imap, agno);
1585 /* move on to the next set of summary map words.
1587 sword = (sword == SMAPSZ - 1) ? 0 : sword + 1;
1588 inosmap = le32_to_cpu(iagp->inosmap[sword]);
1589 extsmap = le32_to_cpu(iagp->extsmap[sword]);
1592 /* unlock imap inode */
1593 IREAD_UNLOCK(ipimap);
1595 /* nothing doing in this iag, so release it. */
1596 release_metapage(mp);
1600 * try to allocate anywhere within the same AG as the parent inode.
1602 rc = diAllocAG(imap, agno, dir, ip);
1604 AG_UNLOCK(imap, agno);
1610 * try to allocate in any AG.
1612 return (diAllocAny(imap, agno, dir, ip));
1617 * NAME: diAllocAG(imap,agno,dir,ip)
1619 * FUNCTION: allocate a disk inode from the allocation group.
1621 * this routine first determines if a new extent of free
1622 * inodes should be added for the allocation group, with
1623 * the current request satisfied from this extent. if this
1624 * is the case, an attempt will be made to do just that. if
1625 * this attempt fails or it has been determined that a new
1626 * extent should not be added, an attempt is made to satisfy
1627 * the request by allocating an existing (backed) free inode
1628 * from the allocation group.
1630 * PRE CONDITION: Already have the AG lock for this AG.
1633 * imap - pointer to inode map control structure.
1634 * agno - allocation group to allocate from.
1635 * dir - 'true' if the new disk inode is for a directory.
1636 * ip - pointer to the new inode to be filled in on successful return
1637 * with the disk inode number allocated, its extent address
1638 * and the start of the ag.
1642 * -ENOSPC - insufficient disk resources.
1646 diAllocAG(struct inomap * imap, int agno, bool dir, struct inode *ip)
1648 int rc, addext, numfree, numinos;
1650 /* get the number of free and the number of backed disk
1651 * inodes currently within the ag.
1653 numfree = imap->im_agctl[agno].numfree;
1654 numinos = imap->im_agctl[agno].numinos;
1656 if (numfree > numinos) {
1657 jfs_error(ip->i_sb, "numfree > numinos\n");
1661 /* determine if we should allocate a new extent of free inodes
1662 * within the ag: for directory inodes, add a new extent
1663 * if there are a small number of free inodes or number of free
1664 * inodes is a small percentage of the number of backed inodes.
1667 addext = (numfree < 64 ||
1669 && ((numfree * 100) / numinos) <= 20));
1671 addext = (numfree == 0);
1674 * try to allocate a new extent of free inodes.
1677 /* if free space is not available for this new extent, try
1678 * below to allocate a free and existing (already backed)
1679 * inode from the ag.
1681 if ((rc = diAllocExt(imap, agno, ip)) != -ENOSPC)
1686 * try to allocate an existing free inode from the ag.
1688 return (diAllocIno(imap, agno, ip));
1693 * NAME: diAllocAny(imap,agno,dir,iap)
1695 * FUNCTION: allocate a disk inode from any other allocation group.
1697 * this routine is called when an allocation attempt within
1698 * the primary allocation group has failed. if attempts to
1699 * allocate an inode from any allocation group other than the
1700 * specified primary group.
1703 * imap - pointer to inode map control structure.
1704 * agno - primary allocation group (to avoid).
1705 * dir - 'true' if the new disk inode is for a directory.
1706 * ip - pointer to a new inode to be filled in on successful return
1707 * with the disk inode number allocated, its extent address
1708 * and the start of the ag.
1712 * -ENOSPC - insufficient disk resources.
1716 diAllocAny(struct inomap * imap, int agno, bool dir, struct inode *ip)
1719 int maxag = JFS_SBI(imap->im_ipimap->i_sb)->bmap->db_maxag;
1722 /* try to allocate from the ags following agno up to
1723 * the maximum ag number.
1725 for (ag = agno + 1; ag <= maxag; ag++) {
1728 rc = diAllocAG(imap, ag, dir, ip);
1730 AG_UNLOCK(imap, ag);
1736 /* try to allocate from the ags in front of agno.
1738 for (ag = 0; ag < agno; ag++) {
1741 rc = diAllocAG(imap, ag, dir, ip);
1743 AG_UNLOCK(imap, ag);
1749 /* no free disk inodes.
1756 * NAME: diAllocIno(imap,agno,ip)
1758 * FUNCTION: allocate a disk inode from the allocation group's free
1759 * inode list, returning an error if this free list is
1760 * empty (i.e. no iags on the list).
1762 * allocation occurs from the first iag on the list using
1763 * the iag's free inode summary map to find the leftmost
1764 * free inode in the iag.
1766 * PRE CONDITION: Already have AG lock for this AG.
1769 * imap - pointer to inode map control structure.
1770 * agno - allocation group.
1771 * ip - pointer to new inode to be filled in on successful return
1772 * with the disk inode number allocated, its extent address
1773 * and the start of the ag.
1777 * -ENOSPC - insufficient disk resources.
1780 static int diAllocIno(struct inomap * imap, int agno, struct inode *ip)
1782 int iagno, ino, rc, rem, extno, sword;
1783 struct metapage *mp;
1786 /* check if there are iags on the ag's free inode list.
1788 if ((iagno = imap->im_agctl[agno].inofree) < 0)
1791 /* obtain read lock on imap inode */
1792 IREAD_LOCK(imap->im_ipimap, RDWRLOCK_IMAP);
1794 /* read the iag at the head of the list.
1796 if ((rc = diIAGRead(imap, iagno, &mp))) {
1797 IREAD_UNLOCK(imap->im_ipimap);
1800 iagp = (struct iag *) mp->data;
1802 /* better be free inodes in this iag if it is on the
1805 if (!iagp->nfreeinos) {
1806 IREAD_UNLOCK(imap->im_ipimap);
1807 release_metapage(mp);
1808 jfs_error(ip->i_sb, "nfreeinos = 0, but iag on freelist\n");
1812 /* scan the free inode summary map to find an extent
1815 for (sword = 0;; sword++) {
1816 if (sword >= SMAPSZ) {
1817 IREAD_UNLOCK(imap->im_ipimap);
1818 release_metapage(mp);
1820 "free inode not found in summary map\n");
1824 if (~iagp->inosmap[sword])
1828 /* found a extent with free inodes. determine
1829 * the extent number.
1831 rem = diFindFree(le32_to_cpu(iagp->inosmap[sword]), 0);
1832 if (rem >= EXTSPERSUM) {
1833 IREAD_UNLOCK(imap->im_ipimap);
1834 release_metapage(mp);
1835 jfs_error(ip->i_sb, "no free extent found\n");
1838 extno = (sword << L2EXTSPERSUM) + rem;
1840 /* find the first free inode in the extent.
1842 rem = diFindFree(le32_to_cpu(iagp->wmap[extno]), 0);
1843 if (rem >= INOSPEREXT) {
1844 IREAD_UNLOCK(imap->im_ipimap);
1845 release_metapage(mp);
1846 jfs_error(ip->i_sb, "free inode not found\n");
1850 /* compute the inode number within the iag.
1852 ino = (extno << L2INOSPEREXT) + rem;
1854 /* allocate the inode.
1856 rc = diAllocBit(imap, iagp, ino);
1857 IREAD_UNLOCK(imap->im_ipimap);
1859 release_metapage(mp);
1863 /* set the results of the allocation and write the iag.
1865 diInitInode(ip, iagno, ino, extno, iagp);
1873 * NAME: diAllocExt(imap,agno,ip)
1875 * FUNCTION: add a new extent of free inodes to an iag, allocating
1876 * an inode from this extent to satisfy the current allocation
1879 * this routine first tries to find an existing iag with free
1880 * extents through the ag free extent list. if list is not
1881 * empty, the head of the list will be selected as the home
1882 * of the new extent of free inodes. otherwise (the list is
1883 * empty), a new iag will be allocated for the ag to contain
1886 * once an iag has been selected, the free extent summary map
1887 * is used to locate a free extent within the iag and diNewExt()
1888 * is called to initialize the extent, with initialization
1889 * including the allocation of the first inode of the extent
1890 * for the purpose of satisfying this request.
1893 * imap - pointer to inode map control structure.
1894 * agno - allocation group number.
1895 * ip - pointer to new inode to be filled in on successful return
1896 * with the disk inode number allocated, its extent address
1897 * and the start of the ag.
1901 * -ENOSPC - insufficient disk resources.
1904 static int diAllocExt(struct inomap * imap, int agno, struct inode *ip)
1906 int rem, iagno, sword, extno, rc;
1907 struct metapage *mp;
1910 /* check if the ag has any iags with free extents. if not,
1911 * allocate a new iag for the ag.
1913 if ((iagno = imap->im_agctl[agno].extfree) < 0) {
1914 /* If successful, diNewIAG will obtain the read lock on the
1917 if ((rc = diNewIAG(imap, &iagno, agno, &mp))) {
1920 iagp = (struct iag *) mp->data;
1922 /* set the ag number if this a brand new iag
1925 cpu_to_le64(AGTOBLK(agno, imap->im_ipimap));
1929 IREAD_LOCK(imap->im_ipimap, RDWRLOCK_IMAP);
1930 if ((rc = diIAGRead(imap, iagno, &mp))) {
1931 IREAD_UNLOCK(imap->im_ipimap);
1932 jfs_error(ip->i_sb, "error reading iag\n");
1935 iagp = (struct iag *) mp->data;
1938 /* using the free extent summary map, find a free extent.
1940 for (sword = 0;; sword++) {
1941 if (sword >= SMAPSZ) {
1942 release_metapage(mp);
1943 IREAD_UNLOCK(imap->im_ipimap);
1944 jfs_error(ip->i_sb, "free ext summary map not found\n");
1947 if (~iagp->extsmap[sword])
1951 /* determine the extent number of the free extent.
1953 rem = diFindFree(le32_to_cpu(iagp->extsmap[sword]), 0);
1954 if (rem >= EXTSPERSUM) {
1955 release_metapage(mp);
1956 IREAD_UNLOCK(imap->im_ipimap);
1957 jfs_error(ip->i_sb, "free extent not found\n");
1960 extno = (sword << L2EXTSPERSUM) + rem;
1962 /* initialize the new extent.
1964 rc = diNewExt(imap, iagp, extno);
1965 IREAD_UNLOCK(imap->im_ipimap);
1967 /* something bad happened. if a new iag was allocated,
1968 * place it back on the inode map's iag free list, and
1969 * clear the ag number information.
1971 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
1973 iagp->iagfree = cpu_to_le32(imap->im_freeiag);
1974 imap->im_freeiag = iagno;
1975 IAGFREE_UNLOCK(imap);
1981 /* set the results of the allocation and write the iag.
1983 diInitInode(ip, iagno, extno << L2INOSPEREXT, extno, iagp);
1992 * NAME: diAllocBit(imap,iagp,ino)
1994 * FUNCTION: allocate a backed inode from an iag.
1996 * this routine performs the mechanics of allocating a
1997 * specified inode from a backed extent.
1999 * if the inode to be allocated represents the last free
2000 * inode within the iag, the iag will be removed from the
2001 * ag free inode list.
2003 * a careful update approach is used to provide consistency
2004 * in the face of updates to multiple buffers. under this
2005 * approach, all required buffers are obtained before making
2006 * any updates and are held all are updates are complete.
2008 * PRE CONDITION: Already have buffer lock on iagp. Already have AG lock on
2009 * this AG. Must have read lock on imap inode.
2012 * imap - pointer to inode map control structure.
2013 * iagp - pointer to iag.
2014 * ino - inode number to be allocated within the iag.
2018 * -ENOSPC - insufficient disk resources.
2021 static int diAllocBit(struct inomap * imap, struct iag * iagp, int ino)
2023 int extno, bitno, agno, sword, rc;
2024 struct metapage *amp = NULL, *bmp = NULL;
2025 struct iag *aiagp = NULL, *biagp = NULL;
2028 /* check if this is the last free inode within the iag.
2029 * if so, it will have to be removed from the ag free
2030 * inode list, so get the iags preceding and following
2033 if (iagp->nfreeinos == cpu_to_le32(1)) {
2034 if ((int) le32_to_cpu(iagp->inofreefwd) >= 0) {
2036 diIAGRead(imap, le32_to_cpu(iagp->inofreefwd),
2039 aiagp = (struct iag *) amp->data;
2042 if ((int) le32_to_cpu(iagp->inofreeback) >= 0) {
2045 le32_to_cpu(iagp->inofreeback),
2048 release_metapage(amp);
2051 biagp = (struct iag *) bmp->data;
2055 /* get the ag number, extent number, inode number within
2058 agno = BLKTOAG(le64_to_cpu(iagp->agstart), JFS_SBI(imap->im_ipimap->i_sb));
2059 extno = ino >> L2INOSPEREXT;
2060 bitno = ino & (INOSPEREXT - 1);
2062 /* compute the mask for setting the map.
2064 mask = HIGHORDER >> bitno;
2066 /* the inode should be free and backed.
2068 if (((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) ||
2069 ((le32_to_cpu(iagp->wmap[extno]) & mask) != 0) ||
2070 (addressPXD(&iagp->inoext[extno]) == 0)) {
2072 release_metapage(amp);
2074 release_metapage(bmp);
2076 jfs_error(imap->im_ipimap->i_sb, "iag inconsistent\n");
2080 /* mark the inode as allocated in the working map.
2082 iagp->wmap[extno] |= cpu_to_le32(mask);
2084 /* check if all inodes within the extent are now
2085 * allocated. if so, update the free inode summary
2086 * map to reflect this.
2088 if (iagp->wmap[extno] == cpu_to_le32(ONES)) {
2089 sword = extno >> L2EXTSPERSUM;
2090 bitno = extno & (EXTSPERSUM - 1);
2091 iagp->inosmap[sword] |= cpu_to_le32(HIGHORDER >> bitno);
2094 /* if this was the last free inode in the iag, remove the
2095 * iag from the ag free inode list.
2097 if (iagp->nfreeinos == cpu_to_le32(1)) {
2099 aiagp->inofreeback = iagp->inofreeback;
2100 write_metapage(amp);
2104 biagp->inofreefwd = iagp->inofreefwd;
2105 write_metapage(bmp);
2107 imap->im_agctl[agno].inofree =
2108 le32_to_cpu(iagp->inofreefwd);
2110 iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1);
2113 /* update the free inode count at the iag, ag, inode
2116 le32_add_cpu(&iagp->nfreeinos, -1);
2117 imap->im_agctl[agno].numfree -= 1;
2118 atomic_dec(&imap->im_numfree);
2125 * NAME: diNewExt(imap,iagp,extno)
2127 * FUNCTION: initialize a new extent of inodes for an iag, allocating
2128 * the first inode of the extent for use for the current
2129 * allocation request.
2131 * disk resources are allocated for the new extent of inodes
2132 * and the inodes themselves are initialized to reflect their
2133 * existence within the extent (i.e. their inode numbers and
2134 * inode extent addresses are set) and their initial state
2135 * (mode and link count are set to zero).
2137 * if the iag is new, it is not yet on an ag extent free list
2138 * but will now be placed on this list.
2140 * if the allocation of the new extent causes the iag to
2141 * have no free extent, the iag will be removed from the
2142 * ag extent free list.
2144 * if the iag has no free backed inodes, it will be placed
2145 * on the ag free inode list, since the addition of the new
2146 * extent will now cause it to have free inodes.
2148 * a careful update approach is used to provide consistency
2149 * (i.e. list consistency) in the face of updates to multiple
2150 * buffers. under this approach, all required buffers are
2151 * obtained before making any updates and are held until all
2152 * updates are complete.
2154 * PRE CONDITION: Already have buffer lock on iagp. Already have AG lock on
2155 * this AG. Must have read lock on imap inode.
2158 * imap - pointer to inode map control structure.
2159 * iagp - pointer to iag.
2160 * extno - extent number.
2164 * -ENOSPC - insufficient disk resources.
2167 static int diNewExt(struct inomap * imap, struct iag * iagp, int extno)
2169 int agno, iagno, fwd, back, freei = 0, sword, rc;
2170 struct iag *aiagp = NULL, *biagp = NULL, *ciagp = NULL;
2171 struct metapage *amp, *bmp, *cmp, *dmp;
2172 struct inode *ipimap;
2178 struct jfs_sb_info *sbi;
2180 /* better have free extents.
2182 if (!iagp->nfreeexts) {
2183 jfs_error(imap->im_ipimap->i_sb, "no free extents\n");
2187 /* get the inode map inode.
2189 ipimap = imap->im_ipimap;
2190 sbi = JFS_SBI(ipimap->i_sb);
2192 amp = bmp = cmp = NULL;
2194 /* get the ag and iag numbers for this iag.
2196 agno = BLKTOAG(le64_to_cpu(iagp->agstart), sbi);
2197 if (agno >= MAXAG || agno < 0)
2200 iagno = le32_to_cpu(iagp->iagnum);
2202 /* check if this is the last free extent within the
2203 * iag. if so, the iag must be removed from the ag
2204 * free extent list, so get the iags preceding and
2205 * following the iag on this list.
2207 if (iagp->nfreeexts == cpu_to_le32(1)) {
2208 if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) {
2209 if ((rc = diIAGRead(imap, fwd, &)))
2211 aiagp = (struct iag *) amp->data;
2214 if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) {
2215 if ((rc = diIAGRead(imap, back, &bmp)))
2217 biagp = (struct iag *) bmp->data;
2220 /* the iag has free extents. if all extents are free
2221 * (as is the case for a newly allocated iag), the iag
2222 * must be added to the ag free extent list, so get
2223 * the iag at the head of the list in preparation for
2224 * adding this iag to this list.
2227 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2228 if ((fwd = imap->im_agctl[agno].extfree) >= 0) {
2229 if ((rc = diIAGRead(imap, fwd, &)))
2231 aiagp = (struct iag *) amp->data;
2236 /* check if the iag has no free inodes. if so, the iag
2237 * will have to be added to the ag free inode list, so get
2238 * the iag at the head of the list in preparation for
2239 * adding this iag to this list. in doing this, we must
2240 * check if we already have the iag at the head of
2243 if (iagp->nfreeinos == 0) {
2244 freei = imap->im_agctl[agno].inofree;
2249 } else if (freei == back) {
2252 if ((rc = diIAGRead(imap, freei, &cmp)))
2254 ciagp = (struct iag *) cmp->data;
2256 if (ciagp == NULL) {
2257 jfs_error(imap->im_ipimap->i_sb,
2265 /* allocate disk space for the inode extent.
2267 if ((extno == 0) || (addressPXD(&iagp->inoext[extno - 1]) == 0))
2268 hint = ((s64) agno << sbi->bmap->db_agl2size) - 1;
2270 hint = addressPXD(&iagp->inoext[extno - 1]) +
2271 lengthPXD(&iagp->inoext[extno - 1]) - 1;
2273 if ((rc = dbAlloc(ipimap, hint, (s64) imap->im_nbperiext, &blkno)))
2276 /* compute the inode number of the first inode within the
2279 ino = (iagno << L2INOSPERIAG) + (extno << L2INOSPEREXT);
2281 /* initialize the inodes within the newly allocated extent a
2284 for (i = 0; i < imap->im_nbperiext; i += sbi->nbperpage) {
2285 /* get a buffer for this page of disk inodes.
2287 dmp = get_metapage(ipimap, blkno + i, PSIZE, 1);
2292 dp = (struct dinode *) dmp->data;
2294 /* initialize the inode number, mode, link count and
2295 * inode extent address.
2297 for (j = 0; j < INOSPERPAGE; j++, dp++, ino++) {
2298 dp->di_inostamp = cpu_to_le32(sbi->inostamp);
2299 dp->di_number = cpu_to_le32(ino);
2300 dp->di_fileset = cpu_to_le32(FILESYSTEM_I);
2303 PXDaddress(&(dp->di_ixpxd), blkno);
2304 PXDlength(&(dp->di_ixpxd), imap->im_nbperiext);
2306 write_metapage(dmp);
2309 /* if this is the last free extent within the iag, remove the
2310 * iag from the ag free extent list.
2312 if (iagp->nfreeexts == cpu_to_le32(1)) {
2314 aiagp->extfreeback = iagp->extfreeback;
2317 biagp->extfreefwd = iagp->extfreefwd;
2319 imap->im_agctl[agno].extfree =
2320 le32_to_cpu(iagp->extfreefwd);
2322 iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1);
2324 /* if the iag has all free extents (newly allocated iag),
2325 * add the iag to the ag free extent list.
2327 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2329 aiagp->extfreeback = cpu_to_le32(iagno);
2331 iagp->extfreefwd = cpu_to_le32(fwd);
2332 iagp->extfreeback = cpu_to_le32(-1);
2333 imap->im_agctl[agno].extfree = iagno;
2337 /* if the iag has no free inodes, add the iag to the
2338 * ag free inode list.
2340 if (iagp->nfreeinos == 0) {
2342 ciagp->inofreeback = cpu_to_le32(iagno);
2345 cpu_to_le32(imap->im_agctl[agno].inofree);
2346 iagp->inofreeback = cpu_to_le32(-1);
2347 imap->im_agctl[agno].inofree = iagno;
2350 /* initialize the extent descriptor of the extent. */
2351 PXDlength(&iagp->inoext[extno], imap->im_nbperiext);
2352 PXDaddress(&iagp->inoext[extno], blkno);
2354 /* initialize the working and persistent map of the extent.
2355 * the working map will be initialized such that
2356 * it indicates the first inode of the extent is allocated.
2358 iagp->wmap[extno] = cpu_to_le32(HIGHORDER);
2359 iagp->pmap[extno] = 0;
2361 /* update the free inode and free extent summary maps
2362 * for the extent to indicate the extent has free inodes
2363 * and no longer represents a free extent.
2365 sword = extno >> L2EXTSPERSUM;
2366 mask = HIGHORDER >> (extno & (EXTSPERSUM - 1));
2367 iagp->extsmap[sword] |= cpu_to_le32(mask);
2368 iagp->inosmap[sword] &= cpu_to_le32(~mask);
2370 /* update the free inode and free extent counts for the
2373 le32_add_cpu(&iagp->nfreeinos, (INOSPEREXT - 1));
2374 le32_add_cpu(&iagp->nfreeexts, -1);
2376 /* update the free and backed inode counts for the ag.
2378 imap->im_agctl[agno].numfree += (INOSPEREXT - 1);
2379 imap->im_agctl[agno].numinos += INOSPEREXT;
2381 /* update the free and backed inode counts for the inode map.
2383 atomic_add(INOSPEREXT - 1, &imap->im_numfree);
2384 atomic_add(INOSPEREXT, &imap->im_numinos);
2389 write_metapage(amp);
2391 write_metapage(bmp);
2393 write_metapage(cmp);
2399 /* release the iags.
2402 release_metapage(amp);
2404 release_metapage(bmp);
2406 release_metapage(cmp);
2413 * NAME: diNewIAG(imap,iagnop,agno)
2415 * FUNCTION: allocate a new iag for an allocation group.
2417 * first tries to allocate the iag from the inode map
2419 * if the list has free iags, the head of the list is removed
2420 * and returned to satisfy the request.
2421 * if the inode map's iag free list is empty, the inode map
2422 * is extended to hold a new iag. this new iag is initialized
2423 * and returned to satisfy the request.
2426 * imap - pointer to inode map control structure.
2427 * iagnop - pointer to an iag number set with the number of the
2428 * newly allocated iag upon successful return.
2429 * agno - allocation group number.
2430 * bpp - Buffer pointer to be filled in with new IAG's buffer
2434 * -ENOSPC - insufficient disk resources.
2438 * AG lock held on entry/exit;
2439 * write lock on the map is held inside;
2440 * read lock on the map is held on successful completion;
2442 * note: new iag transaction:
2443 * . synchronously write iag;
2444 * . write log of xtree and inode of imap;
2446 * . synchronous write of xtree (right to left, bottom to top);
2447 * . at start of logredo(): init in-memory imap with one additional iag page;
2448 * . at end of logredo(): re-read imap inode to determine
2452 diNewIAG(struct inomap * imap, int *iagnop, int agno, struct metapage ** mpp)
2456 struct inode *ipimap;
2457 struct super_block *sb;
2458 struct jfs_sb_info *sbi;
2459 struct metapage *mp;
2464 struct inode *iplist[1];
2466 /* pick up pointers to the inode map and mount inodes */
2467 ipimap = imap->im_ipimap;
2471 /* acquire the free iag lock */
2474 /* if there are any iags on the inode map free iag list,
2475 * allocate the iag from the head of the list.
2477 if (imap->im_freeiag >= 0) {
2478 /* pick up the iag number at the head of the list */
2479 iagno = imap->im_freeiag;
2481 /* determine the logical block number of the iag */
2482 blkno = IAGTOLBLK(iagno, sbi->l2nbperpage);
2484 /* no free iags. the inode map will have to be extented
2485 * to include a new iag.
2488 /* acquire inode map lock */
2489 IWRITE_LOCK(ipimap, RDWRLOCK_IMAP);
2491 if (ipimap->i_size >> L2PSIZE != imap->im_nextiag + 1) {
2492 IWRITE_UNLOCK(ipimap);
2493 IAGFREE_UNLOCK(imap);
2494 jfs_error(imap->im_ipimap->i_sb,
2495 "ipimap->i_size is wrong\n");
2500 /* get the next available iag number */
2501 iagno = imap->im_nextiag;
2503 /* make sure that we have not exceeded the maximum inode
2506 if (iagno > (MAXIAGS - 1)) {
2507 /* release the inode map lock */
2508 IWRITE_UNLOCK(ipimap);
2515 * synchronously append new iag page.
2517 /* determine the logical address of iag page to append */
2518 blkno = IAGTOLBLK(iagno, sbi->l2nbperpage);
2520 /* Allocate extent for new iag page */
2521 xlen = sbi->nbperpage;
2522 if ((rc = dbAlloc(ipimap, 0, (s64) xlen, &xaddr))) {
2523 /* release the inode map lock */
2524 IWRITE_UNLOCK(ipimap);
2530 * start transaction of update of the inode map
2531 * addressing structure pointing to the new iag page;
2533 tid = txBegin(sb, COMMIT_FORCE);
2534 mutex_lock(&JFS_IP(ipimap)->commit_mutex);
2536 /* update the inode map addressing structure to point to it */
2538 xtInsert(tid, ipimap, 0, blkno, xlen, &xaddr, 0))) {
2540 mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
2541 /* Free the blocks allocated for the iag since it was
2542 * not successfully added to the inode map
2544 dbFree(ipimap, xaddr, (s64) xlen);
2546 /* release the inode map lock */
2547 IWRITE_UNLOCK(ipimap);
2552 /* update the inode map's inode to reflect the extension */
2553 ipimap->i_size += PSIZE;
2554 inode_add_bytes(ipimap, PSIZE);
2556 /* assign a buffer for the page */
2557 mp = get_metapage(ipimap, blkno, PSIZE, 0);
2560 * This is very unlikely since we just created the
2561 * extent, but let's try to handle it correctly
2563 xtTruncate(tid, ipimap, ipimap->i_size - PSIZE,
2568 mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
2570 /* release the inode map lock */
2571 IWRITE_UNLOCK(ipimap);
2576 iagp = (struct iag *) mp->data;
2579 memset(iagp, 0, sizeof(struct iag));
2580 iagp->iagnum = cpu_to_le32(iagno);
2581 iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1);
2582 iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1);
2583 iagp->iagfree = cpu_to_le32(-1);
2584 iagp->nfreeinos = 0;
2585 iagp->nfreeexts = cpu_to_le32(EXTSPERIAG);
2587 /* initialize the free inode summary map (free extent
2588 * summary map initialization handled by bzero).
2590 for (i = 0; i < SMAPSZ; i++)
2591 iagp->inosmap[i] = cpu_to_le32(ONES);
2594 * Write and sync the metapage
2599 * txCommit(COMMIT_FORCE) will synchronously write address
2600 * index pages and inode after commit in careful update order
2601 * of address index pages (right to left, bottom up);
2604 rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE);
2607 mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
2609 duplicateIXtree(sb, blkno, xlen, &xaddr);
2611 /* update the next available iag number */
2612 imap->im_nextiag += 1;
2614 /* Add the iag to the iag free list so we don't lose the iag
2615 * if a failure happens now.
2617 imap->im_freeiag = iagno;
2619 /* Until we have logredo working, we want the imap inode &
2620 * control page to be up to date.
2624 /* release the inode map lock */
2625 IWRITE_UNLOCK(ipimap);
2628 /* obtain read lock on map */
2629 IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
2632 if ((rc = diIAGRead(imap, iagno, &mp))) {
2633 IREAD_UNLOCK(ipimap);
2637 iagp = (struct iag *) mp->data;
2639 /* remove the iag from the iag free list */
2640 imap->im_freeiag = le32_to_cpu(iagp->iagfree);
2641 iagp->iagfree = cpu_to_le32(-1);
2643 /* set the return iag number and buffer pointer */
2648 /* release the iag free lock */
2649 IAGFREE_UNLOCK(imap);
2657 * FUNCTION: get the buffer for the specified iag within a fileset
2658 * or aggregate inode map.
2661 * imap - pointer to inode map control structure.
2662 * iagno - iag number.
2663 * bpp - point to buffer pointer to be filled in on successful
2667 * must have read lock on imap inode
2668 * (When called by diExtendFS, the filesystem is quiesced, therefore
2669 * the read lock is unnecessary.)
2675 static int diIAGRead(struct inomap * imap, int iagno, struct metapage ** mpp)
2677 struct inode *ipimap = imap->im_ipimap;
2680 /* compute the logical block number of the iag. */
2681 blkno = IAGTOLBLK(iagno, JFS_SBI(ipimap->i_sb)->l2nbperpage);
2684 *mpp = read_metapage(ipimap, blkno, PSIZE, 0);
2693 * NAME: diFindFree()
2695 * FUNCTION: find the first free bit in a word starting at
2696 * the specified bit position.
2699 * word - word to be examined.
2700 * start - starting bit position.
2703 * bit position of first free bit in the word or 32 if
2704 * no free bits were found.
2706 static int diFindFree(u32 word, int start)
2710 /* scan the word for the first free bit. */
2711 for (word <<= start, bitno = start; bitno < 32;
2712 bitno++, word <<= 1) {
2713 if ((word & HIGHORDER) == 0)
2720 * NAME: diUpdatePMap()
2722 * FUNCTION: Update the persistent map in an IAG for the allocation or
2723 * freeing of the specified inode.
2725 * PRE CONDITIONS: Working map has already been updated for allocate.
2728 * ipimap - Incore inode map inode
2729 * inum - Number of inode to mark in permanent map
2730 * is_free - If 'true' indicates inode should be marked freed, otherwise
2731 * indicates inode should be marked allocated.
2737 diUpdatePMap(struct inode *ipimap,
2738 unsigned long inum, bool is_free, struct tblock * tblk)
2742 struct metapage *mp;
2743 int iagno, ino, extno, bitno;
2744 struct inomap *imap;
2746 struct jfs_log *log;
2747 int lsn, difft, diffp;
2748 unsigned long flags;
2750 imap = JFS_IP(ipimap)->i_imap;
2751 /* get the iag number containing the inode */
2752 iagno = INOTOIAG(inum);
2753 /* make sure that the iag is contained within the map */
2754 if (iagno >= imap->im_nextiag) {
2755 jfs_error(ipimap->i_sb, "the iag is outside the map\n");
2759 IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
2760 rc = diIAGRead(imap, iagno, &mp);
2761 IREAD_UNLOCK(ipimap);
2764 metapage_wait_for_io(mp);
2765 iagp = (struct iag *) mp->data;
2766 /* get the inode number and extent number of the inode within
2767 * the iag and the inode number within the extent.
2769 ino = inum & (INOSPERIAG - 1);
2770 extno = ino >> L2INOSPEREXT;
2771 bitno = ino & (INOSPEREXT - 1);
2772 mask = HIGHORDER >> bitno;
2774 * mark the inode free in persistent map:
2777 /* The inode should have been allocated both in working
2778 * map and in persistent map;
2779 * the inode will be freed from working map at the release
2780 * of last reference release;
2782 if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
2783 jfs_error(ipimap->i_sb,
2784 "inode %ld not marked as allocated in wmap!\n",
2787 if (!(le32_to_cpu(iagp->pmap[extno]) & mask)) {
2788 jfs_error(ipimap->i_sb,
2789 "inode %ld not marked as allocated in pmap!\n",
2792 /* update the bitmap for the extent of the freed inode */
2793 iagp->pmap[extno] &= cpu_to_le32(~mask);
2796 * mark the inode allocated in persistent map:
2799 /* The inode should be already allocated in the working map
2800 * and should be free in persistent map;
2802 if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
2803 release_metapage(mp);
2804 jfs_error(ipimap->i_sb,
2805 "the inode is not allocated in the working map\n");
2808 if ((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) {
2809 release_metapage(mp);
2810 jfs_error(ipimap->i_sb,
2811 "the inode is not free in the persistent map\n");
2814 /* update the bitmap for the extent of the allocated inode */
2815 iagp->pmap[extno] |= cpu_to_le32(mask);
2821 log = JFS_SBI(tblk->sb)->log;
2822 LOGSYNC_LOCK(log, flags);
2824 /* inherit older/smaller lsn */
2825 logdiff(difft, lsn, log);
2826 logdiff(diffp, mp->lsn, log);
2827 if (difft < diffp) {
2829 /* move mp after tblock in logsync list */
2830 list_move(&mp->synclist, &tblk->synclist);
2832 /* inherit younger/larger clsn */
2834 logdiff(difft, tblk->clsn, log);
2835 logdiff(diffp, mp->clsn, log);
2837 mp->clsn = tblk->clsn;
2841 /* insert mp after tblock in logsync list */
2843 list_add(&mp->synclist, &tblk->synclist);
2844 mp->clsn = tblk->clsn;
2846 LOGSYNC_UNLOCK(log, flags);
2854 * function: update imap for extendfs();
2856 * note: AG size has been increased s.t. each k old contiguous AGs are
2857 * coalesced into a new AG;
2859 int diExtendFS(struct inode *ipimap, struct inode *ipbmap)
2862 struct inomap *imap = JFS_IP(ipimap)->i_imap;
2863 struct iag *iagp = NULL, *hiagp = NULL;
2864 struct bmap *mp = JFS_SBI(ipbmap->i_sb)->bmap;
2865 struct metapage *bp, *hbp;
2867 int numinos, xnuminos = 0, xnumfree = 0;
2870 jfs_info("diExtendFS: nextiag:%d numinos:%d numfree:%d",
2871 imap->im_nextiag, atomic_read(&imap->im_numinos),
2872 atomic_read(&imap->im_numfree));
2877 * coalesce contiguous k (newAGSize/oldAGSize) AGs;
2878 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn;
2879 * note: new AG size = old AG size * (2**x).
2882 /* init per AG control information im_agctl[] */
2883 for (i = 0; i < MAXAG; i++) {
2884 imap->im_agctl[i].inofree = -1;
2885 imap->im_agctl[i].extfree = -1;
2886 imap->im_agctl[i].numinos = 0; /* number of backed inodes */
2887 imap->im_agctl[i].numfree = 0; /* number of free backed inodes */
2891 * process each iag page of the map.
2893 * rebuild AG Free Inode List, AG Free Inode Extent List;
2895 for (i = 0; i < imap->im_nextiag; i++) {
2896 if ((rc = diIAGRead(imap, i, &bp))) {
2900 iagp = (struct iag *) bp->data;
2901 if (le32_to_cpu(iagp->iagnum) != i) {
2902 release_metapage(bp);
2903 jfs_error(ipimap->i_sb, "unexpected value of iagnum\n");
2907 /* leave free iag in the free iag list */
2908 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2909 release_metapage(bp);
2913 agstart = le64_to_cpu(iagp->agstart);
2914 n = agstart >> mp->db_agl2size;
2915 iagp->agstart = cpu_to_le64((s64)n << mp->db_agl2size);
2917 /* compute backed inodes */
2918 numinos = (EXTSPERIAG - le32_to_cpu(iagp->nfreeexts))
2921 /* merge AG backed inodes */
2922 imap->im_agctl[n].numinos += numinos;
2923 xnuminos += numinos;
2926 /* if any backed free inodes, insert at AG free inode list */
2927 if ((int) le32_to_cpu(iagp->nfreeinos) > 0) {
2928 if ((head = imap->im_agctl[n].inofree) == -1) {
2929 iagp->inofreefwd = cpu_to_le32(-1);
2930 iagp->inofreeback = cpu_to_le32(-1);
2932 if ((rc = diIAGRead(imap, head, &hbp))) {
2936 hiagp = (struct iag *) hbp->data;
2937 hiagp->inofreeback = iagp->iagnum;
2938 iagp->inofreefwd = cpu_to_le32(head);
2939 iagp->inofreeback = cpu_to_le32(-1);
2940 write_metapage(hbp);
2943 imap->im_agctl[n].inofree =
2944 le32_to_cpu(iagp->iagnum);
2946 /* merge AG backed free inodes */
2947 imap->im_agctl[n].numfree +=
2948 le32_to_cpu(iagp->nfreeinos);
2949 xnumfree += le32_to_cpu(iagp->nfreeinos);
2952 /* if any free extents, insert at AG free extent list */
2953 if (le32_to_cpu(iagp->nfreeexts) > 0) {
2954 if ((head = imap->im_agctl[n].extfree) == -1) {
2955 iagp->extfreefwd = cpu_to_le32(-1);
2956 iagp->extfreeback = cpu_to_le32(-1);
2958 if ((rc = diIAGRead(imap, head, &hbp))) {
2962 hiagp = (struct iag *) hbp->data;
2963 hiagp->extfreeback = iagp->iagnum;
2964 iagp->extfreefwd = cpu_to_le32(head);
2965 iagp->extfreeback = cpu_to_le32(-1);
2966 write_metapage(hbp);
2969 imap->im_agctl[n].extfree =
2970 le32_to_cpu(iagp->iagnum);
2977 if (xnuminos != atomic_read(&imap->im_numinos) ||
2978 xnumfree != atomic_read(&imap->im_numfree)) {
2979 jfs_error(ipimap->i_sb, "numinos or numfree incorrect\n");
2990 * serialization: IWRITE_LOCK held on entry/exit
2992 * note: shadow page with regular inode (rel.2);
2994 static void duplicateIXtree(struct super_block *sb, s64 blkno,
2995 int xlen, s64 *xaddr)
2997 struct jfs_superblock *j_sb;
2998 struct buffer_head *bh;
3002 /* if AIT2 ipmap2 is bad, do not try to update it */
3003 if (JFS_SBI(sb)->mntflag & JFS_BAD_SAIT) /* s_flag */
3005 ip = diReadSpecial(sb, FILESYSTEM_I, 1);
3007 JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT;
3008 if (readSuper(sb, &bh))
3010 j_sb = (struct jfs_superblock *)bh->b_data;
3011 j_sb->s_flag |= cpu_to_le32(JFS_BAD_SAIT);
3013 mark_buffer_dirty(bh);
3014 sync_dirty_buffer(bh);
3019 /* start transaction */
3020 tid = txBegin(sb, COMMIT_FORCE);
3021 /* update the inode map addressing structure to point to it */
3022 if (xtInsert(tid, ip, 0, blkno, xlen, xaddr, 0)) {
3023 JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT;
3028 /* update the inode map's inode to reflect the extension */
3029 ip->i_size += PSIZE;
3030 inode_add_bytes(ip, PSIZE);
3031 txCommit(tid, 1, &ip, COMMIT_FORCE);
3038 * NAME: copy_from_dinode()
3040 * FUNCTION: Copies inode info from disk inode to in-memory inode
3044 * -ENOMEM - insufficient memory
3046 static int copy_from_dinode(struct dinode * dip, struct inode *ip)
3048 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
3049 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
3051 jfs_ip->fileset = le32_to_cpu(dip->di_fileset);
3052 jfs_ip->mode2 = le32_to_cpu(dip->di_mode);
3053 jfs_set_inode_flags(ip);
3055 ip->i_mode = le32_to_cpu(dip->di_mode) & 0xffff;
3056 if (sbi->umask != -1) {
3057 ip->i_mode = (ip->i_mode & ~0777) | (0777 & ~sbi->umask);
3058 /* For directories, add x permission if r is allowed by umask */
3059 if (S_ISDIR(ip->i_mode)) {
3060 if (ip->i_mode & 0400)
3062 if (ip->i_mode & 0040)
3064 if (ip->i_mode & 0004)
3068 set_nlink(ip, le32_to_cpu(dip->di_nlink));
3070 jfs_ip->saved_uid = make_kuid(&init_user_ns, le32_to_cpu(dip->di_uid));
3071 if (!uid_valid(sbi->uid))
3072 ip->i_uid = jfs_ip->saved_uid;
3074 ip->i_uid = sbi->uid;
3077 jfs_ip->saved_gid = make_kgid(&init_user_ns, le32_to_cpu(dip->di_gid));
3078 if (!gid_valid(sbi->gid))
3079 ip->i_gid = jfs_ip->saved_gid;
3081 ip->i_gid = sbi->gid;
3084 ip->i_size = le64_to_cpu(dip->di_size);
3085 ip->i_atime.tv_sec = le32_to_cpu(dip->di_atime.tv_sec);
3086 ip->i_atime.tv_nsec = le32_to_cpu(dip->di_atime.tv_nsec);
3087 ip->i_mtime.tv_sec = le32_to_cpu(dip->di_mtime.tv_sec);
3088 ip->i_mtime.tv_nsec = le32_to_cpu(dip->di_mtime.tv_nsec);
3089 ip->i_ctime.tv_sec = le32_to_cpu(dip->di_ctime.tv_sec);
3090 ip->i_ctime.tv_nsec = le32_to_cpu(dip->di_ctime.tv_nsec);
3091 ip->i_blocks = LBLK2PBLK(ip->i_sb, le64_to_cpu(dip->di_nblocks));
3092 ip->i_generation = le32_to_cpu(dip->di_gen);
3094 jfs_ip->ixpxd = dip->di_ixpxd; /* in-memory pxd's are little-endian */
3095 jfs_ip->acl = dip->di_acl; /* as are dxd's */
3096 jfs_ip->ea = dip->di_ea;
3097 jfs_ip->next_index = le32_to_cpu(dip->di_next_index);
3098 jfs_ip->otime = le32_to_cpu(dip->di_otime.tv_sec);
3099 jfs_ip->acltype = le32_to_cpu(dip->di_acltype);
3101 if (S_ISCHR(ip->i_mode) || S_ISBLK(ip->i_mode)) {
3102 jfs_ip->dev = le32_to_cpu(dip->di_rdev);
3103 ip->i_rdev = new_decode_dev(jfs_ip->dev);
3106 if (S_ISDIR(ip->i_mode)) {
3107 memcpy(&jfs_ip->i_dirtable, &dip->di_dirtable, 384);
3108 } else if (S_ISREG(ip->i_mode) || S_ISLNK(ip->i_mode)) {
3109 memcpy(&jfs_ip->i_xtroot, &dip->di_xtroot, 288);
3111 memcpy(&jfs_ip->i_inline_ea, &dip->di_inlineea, 128);
3113 /* Zero the in-memory-only stuff */
3115 jfs_ip->btindex = 0;
3116 jfs_ip->btorder = 0;
3119 jfs_ip->atlhead = 0;
3120 jfs_ip->atltail = 0;
3126 * NAME: copy_to_dinode()
3128 * FUNCTION: Copies inode info from in-memory inode to disk inode
3130 static void copy_to_dinode(struct dinode * dip, struct inode *ip)
3132 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
3133 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
3135 dip->di_fileset = cpu_to_le32(jfs_ip->fileset);
3136 dip->di_inostamp = cpu_to_le32(sbi->inostamp);
3137 dip->di_number = cpu_to_le32(ip->i_ino);
3138 dip->di_gen = cpu_to_le32(ip->i_generation);
3139 dip->di_size = cpu_to_le64(ip->i_size);
3140 dip->di_nblocks = cpu_to_le64(PBLK2LBLK(ip->i_sb, ip->i_blocks));
3141 dip->di_nlink = cpu_to_le32(ip->i_nlink);
3142 if (!uid_valid(sbi->uid))
3143 dip->di_uid = cpu_to_le32(i_uid_read(ip));
3145 dip->di_uid =cpu_to_le32(from_kuid(&init_user_ns,
3146 jfs_ip->saved_uid));
3147 if (!gid_valid(sbi->gid))
3148 dip->di_gid = cpu_to_le32(i_gid_read(ip));
3150 dip->di_gid = cpu_to_le32(from_kgid(&init_user_ns,
3151 jfs_ip->saved_gid));
3153 * mode2 is only needed for storing the higher order bits.
3154 * Trust i_mode for the lower order ones
3156 if (sbi->umask == -1)
3157 dip->di_mode = cpu_to_le32((jfs_ip->mode2 & 0xffff0000) |
3159 else /* Leave the original permissions alone */
3160 dip->di_mode = cpu_to_le32(jfs_ip->mode2);
3162 dip->di_atime.tv_sec = cpu_to_le32(ip->i_atime.tv_sec);
3163 dip->di_atime.tv_nsec = cpu_to_le32(ip->i_atime.tv_nsec);
3164 dip->di_ctime.tv_sec = cpu_to_le32(ip->i_ctime.tv_sec);
3165 dip->di_ctime.tv_nsec = cpu_to_le32(ip->i_ctime.tv_nsec);
3166 dip->di_mtime.tv_sec = cpu_to_le32(ip->i_mtime.tv_sec);
3167 dip->di_mtime.tv_nsec = cpu_to_le32(ip->i_mtime.tv_nsec);
3168 dip->di_ixpxd = jfs_ip->ixpxd; /* in-memory pxd's are little-endian */
3169 dip->di_acl = jfs_ip->acl; /* as are dxd's */
3170 dip->di_ea = jfs_ip->ea;
3171 dip->di_next_index = cpu_to_le32(jfs_ip->next_index);
3172 dip->di_otime.tv_sec = cpu_to_le32(jfs_ip->otime);
3173 dip->di_otime.tv_nsec = 0;
3174 dip->di_acltype = cpu_to_le32(jfs_ip->acltype);
3175 if (S_ISCHR(ip->i_mode) || S_ISBLK(ip->i_mode))
3176 dip->di_rdev = cpu_to_le32(jfs_ip->dev);