1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Copyright (C) International Business Machines Corp., 2000-2004
7 * jfs_imap.c: inode allocation map manager
10 * Each AG has a simple lock which is used to control the serialization of
11 * the AG level lists. This lock should be taken first whenever an AG
12 * level list will be modified or accessed.
14 * Each IAG is locked by obtaining the buffer for the IAG page.
16 * There is also a inode lock for the inode map inode. A read lock needs to
17 * be taken whenever an IAG is read from the map or the global level
18 * information is read. A write lock needs to be taken whenever the global
19 * level information is modified or an atomic operation needs to be used.
21 * If more than one IAG is read at one time, the read lock may not
22 * be given up until all of the IAG's are read. Otherwise, a deadlock
23 * may occur when trying to obtain the read lock while another thread
24 * holding the read lock is waiting on the IAG already being held.
26 * The control page of the inode map is read into memory by diMount().
27 * Thereafter it should only be modified in memory and then it will be
28 * written out when the filesystem is unmounted by diUnmount().
32 #include <linux/buffer_head.h>
33 #include <linux/pagemap.h>
34 #include <linux/quotaops.h>
35 #include <linux/slab.h>
37 #include "jfs_incore.h"
38 #include "jfs_inode.h"
39 #include "jfs_filsys.h"
40 #include "jfs_dinode.h"
43 #include "jfs_metapage.h"
44 #include "jfs_superblock.h"
45 #include "jfs_debug.h"
50 /* iag free list lock */
51 #define IAGFREE_LOCK_INIT(imap) mutex_init(&imap->im_freelock)
52 #define IAGFREE_LOCK(imap) mutex_lock(&imap->im_freelock)
53 #define IAGFREE_UNLOCK(imap) mutex_unlock(&imap->im_freelock)
55 /* per ag iag list locks */
56 #define AG_LOCK_INIT(imap,index) mutex_init(&(imap->im_aglock[index]))
57 #define AG_LOCK(imap,agno) mutex_lock(&imap->im_aglock[agno])
58 #define AG_UNLOCK(imap,agno) mutex_unlock(&imap->im_aglock[agno])
63 static int diAllocAG(struct inomap *, int, bool, struct inode *);
64 static int diAllocAny(struct inomap *, int, bool, struct inode *);
65 static int diAllocBit(struct inomap *, struct iag *, int);
66 static int diAllocExt(struct inomap *, int, struct inode *);
67 static int diAllocIno(struct inomap *, int, struct inode *);
68 static int diFindFree(u32, int);
69 static int diNewExt(struct inomap *, struct iag *, int);
70 static int diNewIAG(struct inomap *, int *, int, struct metapage **);
71 static void duplicateIXtree(struct super_block *, s64, int, s64 *);
73 static int diIAGRead(struct inomap * imap, int, struct metapage **);
74 static int copy_from_dinode(struct dinode *, struct inode *);
75 static void copy_to_dinode(struct dinode *, struct inode *);
80 * FUNCTION: initialize the incore inode map control structures for
81 * a fileset or aggregate init time.
83 * the inode map's control structure (dinomap) is
84 * brought in from disk and placed in virtual memory.
87 * ipimap - pointer to inode map inode for the aggregate or fileset.
91 * -ENOMEM - insufficient free virtual memory.
94 int diMount(struct inode *ipimap)
99 struct dinomap_disk *dinom_le;
102 * allocate/initialize the in-memory inode map control structure
104 /* allocate the in-memory inode map control structure. */
105 imap = kmalloc(sizeof(struct inomap), GFP_KERNEL);
107 jfs_err("diMount: kmalloc returned NULL!");
111 /* read the on-disk inode map control structure. */
113 mp = read_metapage(ipimap,
114 IMAPBLKNO << JFS_SBI(ipimap->i_sb)->l2nbperpage,
121 /* copy the on-disk version to the in-memory version. */
122 dinom_le = (struct dinomap_disk *) mp->data;
123 imap->im_freeiag = le32_to_cpu(dinom_le->in_freeiag);
124 imap->im_nextiag = le32_to_cpu(dinom_le->in_nextiag);
125 atomic_set(&imap->im_numinos, le32_to_cpu(dinom_le->in_numinos));
126 atomic_set(&imap->im_numfree, le32_to_cpu(dinom_le->in_numfree));
127 imap->im_nbperiext = le32_to_cpu(dinom_le->in_nbperiext);
128 imap->im_l2nbperiext = le32_to_cpu(dinom_le->in_l2nbperiext);
129 for (index = 0; index < MAXAG; index++) {
130 imap->im_agctl[index].inofree =
131 le32_to_cpu(dinom_le->in_agctl[index].inofree);
132 imap->im_agctl[index].extfree =
133 le32_to_cpu(dinom_le->in_agctl[index].extfree);
134 imap->im_agctl[index].numinos =
135 le32_to_cpu(dinom_le->in_agctl[index].numinos);
136 imap->im_agctl[index].numfree =
137 le32_to_cpu(dinom_le->in_agctl[index].numfree);
140 /* release the buffer. */
141 release_metapage(mp);
144 * allocate/initialize inode allocation map locks
146 /* allocate and init iag free list lock */
147 IAGFREE_LOCK_INIT(imap);
149 /* allocate and init ag list locks */
150 for (index = 0; index < MAXAG; index++) {
151 AG_LOCK_INIT(imap, index);
154 /* bind the inode map inode and inode map control structure
157 imap->im_ipimap = ipimap;
158 JFS_IP(ipimap)->i_imap = imap;
167 * FUNCTION: write to disk the incore inode map control structures for
168 * a fileset or aggregate at unmount time.
171 * ipimap - pointer to inode map inode for the aggregate or fileset.
175 * -ENOMEM - insufficient free virtual memory.
178 int diUnmount(struct inode *ipimap, int mounterror)
180 struct inomap *imap = JFS_IP(ipimap)->i_imap;
183 * update the on-disk inode map control structure
186 if (!(mounterror || isReadOnly(ipimap)))
190 * Invalidate the page cache buffers
192 truncate_inode_pages(ipimap->i_mapping, 0);
195 * free in-memory control structure
206 int diSync(struct inode *ipimap)
208 struct dinomap_disk *dinom_le;
209 struct inomap *imp = JFS_IP(ipimap)->i_imap;
214 * write imap global conrol page
216 /* read the on-disk inode map control structure */
217 mp = get_metapage(ipimap,
218 IMAPBLKNO << JFS_SBI(ipimap->i_sb)->l2nbperpage,
221 jfs_err("diSync: get_metapage failed!");
225 /* copy the in-memory version to the on-disk version */
226 dinom_le = (struct dinomap_disk *) mp->data;
227 dinom_le->in_freeiag = cpu_to_le32(imp->im_freeiag);
228 dinom_le->in_nextiag = cpu_to_le32(imp->im_nextiag);
229 dinom_le->in_numinos = cpu_to_le32(atomic_read(&imp->im_numinos));
230 dinom_le->in_numfree = cpu_to_le32(atomic_read(&imp->im_numfree));
231 dinom_le->in_nbperiext = cpu_to_le32(imp->im_nbperiext);
232 dinom_le->in_l2nbperiext = cpu_to_le32(imp->im_l2nbperiext);
233 for (index = 0; index < MAXAG; index++) {
234 dinom_le->in_agctl[index].inofree =
235 cpu_to_le32(imp->im_agctl[index].inofree);
236 dinom_le->in_agctl[index].extfree =
237 cpu_to_le32(imp->im_agctl[index].extfree);
238 dinom_le->in_agctl[index].numinos =
239 cpu_to_le32(imp->im_agctl[index].numinos);
240 dinom_le->in_agctl[index].numfree =
241 cpu_to_le32(imp->im_agctl[index].numfree);
244 /* write out the control structure */
248 * write out dirty pages of imap
250 filemap_write_and_wait(ipimap->i_mapping);
252 diWriteSpecial(ipimap, 0);
261 * FUNCTION: initialize an incore inode from disk.
263 * on entry, the specifed incore inode should itself
264 * specify the disk inode number corresponding to the
265 * incore inode (i.e. i_number should be initialized).
267 * this routine handles incore inode initialization for
268 * both "special" and "regular" inodes. special inodes
269 * are those required early in the mount process and
270 * require special handling since much of the file system
271 * is not yet initialized. these "special" inodes are
272 * identified by a NULL inode map inode pointer and are
273 * actually initialized by a call to diReadSpecial().
275 * for regular inodes, the iag describing the disk inode
276 * is read from disk to determine the inode extent address
277 * for the disk inode. with the inode extent address in
278 * hand, the page of the extent that contains the disk
279 * inode is read and the disk inode is copied to the
283 * ip - pointer to incore inode to be initialized from disk.
288 * -ENOMEM - insufficient memory
291 int diRead(struct inode *ip)
293 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
294 int iagno, ino, extno, rc;
295 struct inode *ipimap;
303 unsigned long pageno;
306 jfs_info("diRead: ino = %ld", ip->i_ino);
308 ipimap = sbi->ipimap;
309 JFS_IP(ip)->ipimap = ipimap;
311 /* determine the iag number for this inode (number) */
312 iagno = INOTOIAG(ip->i_ino);
315 imap = JFS_IP(ipimap)->i_imap;
316 IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
317 rc = diIAGRead(imap, iagno, &mp);
318 IREAD_UNLOCK(ipimap);
320 jfs_err("diRead: diIAGRead returned %d", rc);
324 iagp = (struct iag *) mp->data;
326 /* determine inode extent that holds the disk inode */
327 ino = ip->i_ino & (INOSPERIAG - 1);
328 extno = ino >> L2INOSPEREXT;
330 if ((lengthPXD(&iagp->inoext[extno]) != imap->im_nbperiext) ||
331 (addressPXD(&iagp->inoext[extno]) == 0)) {
332 release_metapage(mp);
336 /* get disk block number of the page within the inode extent
337 * that holds the disk inode.
339 blkno = INOPBLK(&iagp->inoext[extno], ino, sbi->l2nbperpage);
341 /* get the ag for the iag */
342 agstart = le64_to_cpu(iagp->agstart);
344 release_metapage(mp);
346 rel_inode = (ino & (INOSPERPAGE - 1));
347 pageno = blkno >> sbi->l2nbperpage;
349 if ((block_offset = ((u32) blkno & (sbi->nbperpage - 1)))) {
351 * OS/2 didn't always align inode extents on page boundaries
354 (sbi->nbperpage - block_offset) << sbi->l2niperblk;
356 if (rel_inode < inodes_left)
357 rel_inode += block_offset << sbi->l2niperblk;
360 rel_inode -= inodes_left;
364 /* read the page of disk inode */
365 mp = read_metapage(ipimap, pageno << sbi->l2nbperpage, PSIZE, 1);
367 jfs_err("diRead: read_metapage failed");
371 /* locate the disk inode requested */
372 dp = (struct dinode *) mp->data;
375 if (ip->i_ino != le32_to_cpu(dp->di_number)) {
376 jfs_error(ip->i_sb, "i_ino != di_number\n");
378 } else if (le32_to_cpu(dp->di_nlink) == 0)
381 /* copy the disk inode to the in-memory inode */
382 rc = copy_from_dinode(dp, ip);
384 release_metapage(mp);
386 /* set the ag for the inode */
387 JFS_IP(ip)->agstart = agstart;
388 JFS_IP(ip)->active_ag = -1;
395 * NAME: diReadSpecial()
397 * FUNCTION: initialize a 'special' inode from disk.
399 * this routines handles aggregate level inodes. The
400 * inode cache cannot differentiate between the
401 * aggregate inodes and the filesystem inodes, so we
402 * handle these here. We don't actually use the aggregate
403 * inode map, since these inodes are at a fixed location
404 * and in some cases the aggregate inode map isn't initialized
408 * sb - filesystem superblock
409 * inum - aggregate inode number
410 * secondary - 1 if secondary aggregate inode table
413 * new inode - success
416 struct inode *diReadSpecial(struct super_block *sb, ino_t inum, int secondary)
418 struct jfs_sb_info *sbi = JFS_SBI(sb);
426 jfs_err("diReadSpecial: new_inode returned NULL!");
431 address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage;
432 JFS_IP(ip)->ipimap = sbi->ipaimap2;
434 address = AITBL_OFF >> L2PSIZE;
435 JFS_IP(ip)->ipimap = sbi->ipaimap;
438 ASSERT(inum < INOSPEREXT);
442 address += inum >> 3; /* 8 inodes per 4K page */
444 /* read the page of fixed disk inode (AIT) in raw mode */
445 mp = read_metapage(ip, address << sbi->l2nbperpage, PSIZE, 1);
447 set_nlink(ip, 1); /* Don't want iput() deleting it */
452 /* get the pointer to the disk inode of interest */
453 dp = (struct dinode *) (mp->data);
454 dp += inum % 8; /* 8 inodes per 4K page */
456 /* copy on-disk inode to in-memory inode */
457 if ((copy_from_dinode(dp, ip)) != 0) {
458 /* handle bad return by returning NULL for ip */
459 set_nlink(ip, 1); /* Don't want iput() deleting it */
461 /* release the page */
462 release_metapage(mp);
467 ip->i_mapping->a_ops = &jfs_metapage_aops;
468 mapping_set_gfp_mask(ip->i_mapping, GFP_NOFS);
470 /* Allocations to metadata inodes should not affect quotas */
471 ip->i_flags |= S_NOQUOTA;
473 if ((inum == FILESYSTEM_I) && (JFS_IP(ip)->ipimap == sbi->ipaimap)) {
474 sbi->gengen = le32_to_cpu(dp->di_gengen);
475 sbi->inostamp = le32_to_cpu(dp->di_inostamp);
478 /* release the page */
479 release_metapage(mp);
487 * NAME: diWriteSpecial()
489 * FUNCTION: Write the special inode to disk
493 * secondary - 1 if secondary aggregate inode table
495 * RETURN VALUES: none
498 void diWriteSpecial(struct inode *ip, int secondary)
500 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
503 ino_t inum = ip->i_ino;
507 address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage;
509 address = AITBL_OFF >> L2PSIZE;
511 ASSERT(inum < INOSPEREXT);
513 address += inum >> 3; /* 8 inodes per 4K page */
515 /* read the page of fixed disk inode (AIT) in raw mode */
516 mp = read_metapage(ip, address << sbi->l2nbperpage, PSIZE, 1);
518 jfs_err("diWriteSpecial: failed to read aggregate inode extent!");
522 /* get the pointer to the disk inode of interest */
523 dp = (struct dinode *) (mp->data);
524 dp += inum % 8; /* 8 inodes per 4K page */
526 /* copy on-disk inode to in-memory inode */
527 copy_to_dinode(dp, ip);
528 memcpy(&dp->di_xtroot, &JFS_IP(ip)->i_xtroot, 288);
530 if (inum == FILESYSTEM_I)
531 dp->di_gengen = cpu_to_le32(sbi->gengen);
538 * NAME: diFreeSpecial()
540 * FUNCTION: Free allocated space for special inode
542 void diFreeSpecial(struct inode *ip)
545 jfs_err("diFreeSpecial called with NULL ip!");
548 filemap_write_and_wait(ip->i_mapping);
549 truncate_inode_pages(ip->i_mapping, 0);
558 * FUNCTION: write the on-disk inode portion of the in-memory inode
559 * to its corresponding on-disk inode.
561 * on entry, the specifed incore inode should itself
562 * specify the disk inode number corresponding to the
563 * incore inode (i.e. i_number should be initialized).
565 * the inode contains the inode extent address for the disk
566 * inode. with the inode extent address in hand, the
567 * page of the extent that contains the disk inode is
568 * read and the disk inode portion of the incore inode
569 * is copied to the disk inode.
572 * tid - transacation id
573 * ip - pointer to incore inode to be written to the inode extent.
579 int diWrite(tid_t tid, struct inode *ip)
581 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
582 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
590 unsigned long pageno;
593 struct inode *ipimap;
596 struct tlock *ditlck, *tlck;
597 struct linelock *dilinelock, *ilinelock;
601 ipimap = jfs_ip->ipimap;
603 ino = ip->i_ino & (INOSPERIAG - 1);
605 if (!addressPXD(&(jfs_ip->ixpxd)) ||
606 (lengthPXD(&(jfs_ip->ixpxd)) !=
607 JFS_IP(ipimap)->i_imap->im_nbperiext)) {
608 jfs_error(ip->i_sb, "ixpxd invalid\n");
613 * read the page of disk inode containing the specified inode:
615 /* compute the block address of the page */
616 blkno = INOPBLK(&(jfs_ip->ixpxd), ino, sbi->l2nbperpage);
618 rel_inode = (ino & (INOSPERPAGE - 1));
619 pageno = blkno >> sbi->l2nbperpage;
621 if ((block_offset = ((u32) blkno & (sbi->nbperpage - 1)))) {
623 * OS/2 didn't always align inode extents on page boundaries
626 (sbi->nbperpage - block_offset) << sbi->l2niperblk;
628 if (rel_inode < inodes_left)
629 rel_inode += block_offset << sbi->l2niperblk;
632 rel_inode -= inodes_left;
635 /* read the page of disk inode */
637 mp = read_metapage(ipimap, pageno << sbi->l2nbperpage, PSIZE, 1);
641 /* get the pointer to the disk inode */
642 dp = (struct dinode *) mp->data;
645 dioffset = (ino & (INOSPERPAGE - 1)) << L2DISIZE;
648 * acquire transaction lock on the on-disk inode;
649 * N.B. tlock is acquired on ipimap not ip;
652 txLock(tid, ipimap, mp, tlckINODE | tlckENTRY)) == NULL)
654 dilinelock = (struct linelock *) & ditlck->lock;
657 * copy btree root from in-memory inode to on-disk inode
659 * (tlock is taken from inline B+-tree root in in-memory
660 * inode when the B+-tree root is updated, which is pointed
661 * by jfs_ip->blid as well as being on tx tlock list)
663 * further processing of btree root is based on the copy
664 * in in-memory inode, where txLog() will log from, and,
665 * for xtree root, txUpdateMap() will update map and reset
669 if (S_ISDIR(ip->i_mode) && (lid = jfs_ip->xtlid)) {
671 * This is the special xtree inside the directory for storing
672 * the directory table
678 tlck = lid_to_tlock(lid);
679 assert(tlck->type & tlckXTREE);
680 tlck->type |= tlckBTROOT;
682 ilinelock = (struct linelock *) & tlck->lock;
685 * copy xtree root from inode to dinode:
687 p = &jfs_ip->i_xtroot;
688 xp = (xtpage_t *) &dp->di_dirtable;
690 for (n = 0; n < ilinelock->index; n++, lv++) {
691 memcpy(&xp->xad[lv->offset], &p->xad[lv->offset],
692 lv->length << L2XTSLOTSIZE);
695 /* reset on-disk (metadata page) xtree XAD_NEW bit */
696 xad = &xp->xad[XTENTRYSTART];
697 for (n = XTENTRYSTART;
698 n < le16_to_cpu(xp->header.nextindex); n++, xad++)
699 if (xad->flag & (XAD_NEW | XAD_EXTENDED))
700 xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
703 if ((lid = jfs_ip->blid) == 0)
707 tlck = lid_to_tlock(lid);
709 tlck->type |= tlckBTROOT;
711 ilinelock = (struct linelock *) & tlck->lock;
714 * regular file: 16 byte (XAD slot) granularity
716 if (type & tlckXTREE) {
721 * copy xtree root from inode to dinode:
723 p = &jfs_ip->i_xtroot;
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 * directory: 32 byte (directory entry slot) granularity
741 else if (type & tlckDTREE) {
745 * copy dtree root from inode to dinode:
747 p = (dtpage_t *) &jfs_ip->i_dtroot;
748 xp = (dtpage_t *) & dp->di_dtroot;
750 for (n = 0; n < ilinelock->index; n++, lv++) {
751 memcpy(&xp->slot[lv->offset], &p->slot[lv->offset],
752 lv->length << L2DTSLOTSIZE);
755 jfs_err("diWrite: UFO tlock");
760 * copy inline symlink from in-memory inode to on-disk inode
762 if (S_ISLNK(ip->i_mode) && ip->i_size < IDATASIZE) {
763 lv = & dilinelock->lv[dilinelock->index];
764 lv->offset = (dioffset + 2 * 128) >> L2INODESLOTSIZE;
766 memcpy(&dp->di_fastsymlink, jfs_ip->i_inline, IDATASIZE);
770 * copy inline data from in-memory inode to on-disk inode:
771 * 128 byte slot granularity
773 if (test_cflag(COMMIT_Inlineea, ip)) {
774 lv = & dilinelock->lv[dilinelock->index];
775 lv->offset = (dioffset + 3 * 128) >> L2INODESLOTSIZE;
777 memcpy(&dp->di_inlineea, jfs_ip->i_inline_ea, INODESLOTSIZE);
780 clear_cflag(COMMIT_Inlineea, ip);
784 * lock/copy inode base: 128 byte slot granularity
786 lv = & dilinelock->lv[dilinelock->index];
787 lv->offset = dioffset >> L2INODESLOTSIZE;
788 copy_to_dinode(dp, ip);
789 if (test_and_clear_cflag(COMMIT_Dirtable, ip)) {
791 memcpy(&dp->di_dirtable, &jfs_ip->i_dirtable, 96);
796 /* release the buffer holding the updated on-disk inode.
797 * the buffer will be later written by commit processing.
808 * FUNCTION: free a specified inode from the inode working map
809 * for a fileset or aggregate.
811 * if the inode to be freed represents the first (only)
812 * free inode within the iag, the iag will be placed on
813 * the ag free inode list.
815 * freeing the inode will cause the inode extent to be
816 * freed if the inode is the only allocated inode within
817 * the extent. in this case all the disk resource backing
818 * up the inode extent will be freed. in addition, the iag
819 * will be placed on the ag extent free list if the extent
820 * is the first free extent in the iag. if freeing the
821 * extent also means that no free inodes will exist for
822 * the iag, the iag will also be removed from the ag free
825 * the iag describing the inode will be freed if the extent
826 * is to be freed and it is the only backed extent within
827 * the iag. in this case, the iag will be removed from the
828 * ag free extent list and ag free inode list and placed on
829 * the inode map's free iag list.
831 * a careful update approach is used to provide consistency
832 * in the face of updates to multiple buffers. under this
833 * approach, all required buffers are obtained before making
834 * any updates and are held until all updates are complete.
837 * ip - inode to be freed.
843 int diFree(struct inode *ip)
846 ino_t inum = ip->i_ino;
847 struct iag *iagp, *aiagp, *biagp, *ciagp, *diagp;
848 struct metapage *mp, *amp, *bmp, *cmp, *dmp;
849 int iagno, ino, extno, bitno, sword, agno;
852 struct inode *ipimap = JFS_SBI(ip->i_sb)->ipimap;
853 struct inomap *imap = JFS_IP(ipimap)->i_imap;
856 struct inode *iplist[3];
858 struct pxd_lock *pxdlock;
861 * This is just to suppress compiler warnings. The same logic that
862 * references these variables is used to initialize them.
864 aiagp = biagp = ciagp = diagp = NULL;
866 /* get the iag number containing the inode.
868 iagno = INOTOIAG(inum);
870 /* make sure that the iag is contained within
873 if (iagno >= imap->im_nextiag) {
874 print_hex_dump(KERN_ERR, "imap: ", DUMP_PREFIX_ADDRESS, 16, 4,
876 jfs_error(ip->i_sb, "inum = %d, iagno = %d, nextiag = %d\n",
877 (uint) inum, iagno, imap->im_nextiag);
881 /* get the allocation group for this ino.
883 agno = BLKTOAG(JFS_IP(ip)->agstart, JFS_SBI(ip->i_sb));
885 /* Lock the AG specific inode map information
889 /* Obtain read lock in imap inode. Don't release it until we have
890 * read all of the IAG's that we are going to.
892 IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
896 if ((rc = diIAGRead(imap, iagno, &mp))) {
897 IREAD_UNLOCK(ipimap);
898 AG_UNLOCK(imap, agno);
901 iagp = (struct iag *) mp->data;
903 /* get the inode number and extent number of the inode within
904 * the iag and the inode number within the extent.
906 ino = inum & (INOSPERIAG - 1);
907 extno = ino >> L2INOSPEREXT;
908 bitno = ino & (INOSPEREXT - 1);
909 mask = HIGHORDER >> bitno;
911 if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
912 jfs_error(ip->i_sb, "wmap shows inode already free\n");
915 if (!addressPXD(&iagp->inoext[extno])) {
916 release_metapage(mp);
917 IREAD_UNLOCK(ipimap);
918 AG_UNLOCK(imap, agno);
919 jfs_error(ip->i_sb, "invalid inoext\n");
923 /* compute the bitmap for the extent reflecting the freed inode.
925 bitmap = le32_to_cpu(iagp->wmap[extno]) & ~mask;
927 if (imap->im_agctl[agno].numfree > imap->im_agctl[agno].numinos) {
928 release_metapage(mp);
929 IREAD_UNLOCK(ipimap);
930 AG_UNLOCK(imap, agno);
931 jfs_error(ip->i_sb, "numfree > numinos\n");
935 * inode extent still has some inodes or below low water mark:
936 * keep the inode extent;
939 imap->im_agctl[agno].numfree < 96 ||
940 (imap->im_agctl[agno].numfree < 288 &&
941 (((imap->im_agctl[agno].numfree * 100) /
942 imap->im_agctl[agno].numinos) <= 25))) {
943 /* if the iag currently has no free inodes (i.e.,
944 * the inode being freed is the first free inode of iag),
945 * insert the iag at head of the inode free list for the ag.
947 if (iagp->nfreeinos == 0) {
948 /* check if there are any iags on the ag inode
949 * free list. if so, read the first one so that
950 * we can link the current iag onto the list at
953 if ((fwd = imap->im_agctl[agno].inofree) >= 0) {
954 /* read the iag that currently is the head
957 if ((rc = diIAGRead(imap, fwd, &))) {
958 IREAD_UNLOCK(ipimap);
959 AG_UNLOCK(imap, agno);
960 release_metapage(mp);
963 aiagp = (struct iag *) amp->data;
965 /* make current head point back to the iag.
967 aiagp->inofreeback = cpu_to_le32(iagno);
972 /* iag points forward to current head and iag
973 * becomes the new head of the list.
976 cpu_to_le32(imap->im_agctl[agno].inofree);
977 iagp->inofreeback = cpu_to_le32(-1);
978 imap->im_agctl[agno].inofree = iagno;
980 IREAD_UNLOCK(ipimap);
982 /* update the free inode summary map for the extent if
983 * freeing the inode means the extent will now have free
984 * inodes (i.e., the inode being freed is the first free
987 if (iagp->wmap[extno] == cpu_to_le32(ONES)) {
988 sword = extno >> L2EXTSPERSUM;
989 bitno = extno & (EXTSPERSUM - 1);
990 iagp->inosmap[sword] &=
991 cpu_to_le32(~(HIGHORDER >> bitno));
994 /* update the bitmap.
996 iagp->wmap[extno] = cpu_to_le32(bitmap);
998 /* update the free inode counts at the iag, ag and
1001 le32_add_cpu(&iagp->nfreeinos, 1);
1002 imap->im_agctl[agno].numfree += 1;
1003 atomic_inc(&imap->im_numfree);
1005 /* release the AG inode map lock
1007 AG_UNLOCK(imap, agno);
1017 * inode extent has become free and above low water mark:
1018 * free the inode extent;
1022 * prepare to update iag list(s) (careful update step 1)
1024 amp = bmp = cmp = dmp = NULL;
1027 /* check if the iag currently has no free extents. if so,
1028 * it will be placed on the head of the ag extent free list.
1030 if (iagp->nfreeexts == 0) {
1031 /* check if the ag extent free list has any iags.
1032 * if so, read the iag at the head of the list now.
1033 * this (head) iag will be updated later to reflect
1034 * the addition of the current iag at the head of
1037 if ((fwd = imap->im_agctl[agno].extfree) >= 0) {
1038 if ((rc = diIAGRead(imap, fwd, &)))
1040 aiagp = (struct iag *) amp->data;
1043 /* iag has free extents. check if the addition of a free
1044 * extent will cause all extents to be free within this
1045 * iag. if so, the iag will be removed from the ag extent
1046 * free list and placed on the inode map's free iag list.
1048 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) {
1049 /* in preparation for removing the iag from the
1050 * ag extent free list, read the iags preceding
1051 * and following the iag on the ag extent free
1054 if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) {
1055 if ((rc = diIAGRead(imap, fwd, &)))
1057 aiagp = (struct iag *) amp->data;
1060 if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) {
1061 if ((rc = diIAGRead(imap, back, &bmp)))
1063 biagp = (struct iag *) bmp->data;
1068 /* remove the iag from the ag inode free list if freeing
1069 * this extent cause the iag to have no free inodes.
1071 if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) {
1072 int inofreeback = le32_to_cpu(iagp->inofreeback);
1073 int inofreefwd = le32_to_cpu(iagp->inofreefwd);
1075 /* in preparation for removing the iag from the
1076 * ag inode free list, read the iags preceding
1077 * and following the iag on the ag inode free
1078 * list. before reading these iags, we must make
1079 * sure that we already don't have them in hand
1080 * from up above, since re-reading an iag (buffer)
1081 * we are currently holding would cause a deadlock.
1083 if (inofreefwd >= 0) {
1085 if (inofreefwd == fwd)
1086 ciagp = (struct iag *) amp->data;
1087 else if (inofreefwd == back)
1088 ciagp = (struct iag *) bmp->data;
1091 diIAGRead(imap, inofreefwd, &cmp)))
1093 ciagp = (struct iag *) cmp->data;
1095 assert(ciagp != NULL);
1098 if (inofreeback >= 0) {
1099 if (inofreeback == fwd)
1100 diagp = (struct iag *) amp->data;
1101 else if (inofreeback == back)
1102 diagp = (struct iag *) bmp->data;
1105 diIAGRead(imap, inofreeback, &dmp)))
1107 diagp = (struct iag *) dmp->data;
1109 assert(diagp != NULL);
1113 IREAD_UNLOCK(ipimap);
1116 * invalidate any page of the inode extent freed from buffer cache;
1118 freepxd = iagp->inoext[extno];
1119 invalidate_pxd_metapages(ip, freepxd);
1122 * update iag list(s) (careful update step 2)
1124 /* add the iag to the ag extent free list if this is the
1125 * first free extent for the iag.
1127 if (iagp->nfreeexts == 0) {
1129 aiagp->extfreeback = cpu_to_le32(iagno);
1132 cpu_to_le32(imap->im_agctl[agno].extfree);
1133 iagp->extfreeback = cpu_to_le32(-1);
1134 imap->im_agctl[agno].extfree = iagno;
1136 /* remove the iag from the ag extent list if all extents
1137 * are now free and place it on the inode map iag free list.
1139 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) {
1141 aiagp->extfreeback = iagp->extfreeback;
1144 biagp->extfreefwd = iagp->extfreefwd;
1146 imap->im_agctl[agno].extfree =
1147 le32_to_cpu(iagp->extfreefwd);
1149 iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1);
1152 iagp->iagfree = cpu_to_le32(imap->im_freeiag);
1153 imap->im_freeiag = iagno;
1154 IAGFREE_UNLOCK(imap);
1158 /* remove the iag from the ag inode free list if freeing
1159 * this extent causes the iag to have no free inodes.
1161 if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) {
1162 if ((int) le32_to_cpu(iagp->inofreefwd) >= 0)
1163 ciagp->inofreeback = iagp->inofreeback;
1165 if ((int) le32_to_cpu(iagp->inofreeback) >= 0)
1166 diagp->inofreefwd = iagp->inofreefwd;
1168 imap->im_agctl[agno].inofree =
1169 le32_to_cpu(iagp->inofreefwd);
1171 iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1);
1174 /* update the inode extent address and working map
1175 * to reflect the free extent.
1176 * the permanent map should have been updated already
1177 * for the inode being freed.
1179 if (iagp->pmap[extno] != 0) {
1180 jfs_error(ip->i_sb, "the pmap does not show inode free\n");
1182 iagp->wmap[extno] = 0;
1183 PXDlength(&iagp->inoext[extno], 0);
1184 PXDaddress(&iagp->inoext[extno], 0);
1186 /* update the free extent and free inode summary maps
1187 * to reflect the freed extent.
1188 * the inode summary map is marked to indicate no inodes
1189 * available for the freed extent.
1191 sword = extno >> L2EXTSPERSUM;
1192 bitno = extno & (EXTSPERSUM - 1);
1193 mask = HIGHORDER >> bitno;
1194 iagp->inosmap[sword] |= cpu_to_le32(mask);
1195 iagp->extsmap[sword] &= cpu_to_le32(~mask);
1197 /* update the number of free inodes and number of free extents
1200 le32_add_cpu(&iagp->nfreeinos, -(INOSPEREXT - 1));
1201 le32_add_cpu(&iagp->nfreeexts, 1);
1203 /* update the number of free inodes and backed inodes
1204 * at the ag and inode map level.
1206 imap->im_agctl[agno].numfree -= (INOSPEREXT - 1);
1207 imap->im_agctl[agno].numinos -= INOSPEREXT;
1208 atomic_sub(INOSPEREXT - 1, &imap->im_numfree);
1209 atomic_sub(INOSPEREXT, &imap->im_numinos);
1212 write_metapage(amp);
1214 write_metapage(bmp);
1216 write_metapage(cmp);
1218 write_metapage(dmp);
1221 * start transaction to update block allocation map
1222 * for the inode extent freed;
1224 * N.B. AG_LOCK is released and iag will be released below, and
1225 * other thread may allocate inode from/reusing the ixad freed
1226 * BUT with new/different backing inode extent from the extent
1227 * to be freed by the transaction;
1229 tid = txBegin(ipimap->i_sb, COMMIT_FORCE);
1230 mutex_lock(&JFS_IP(ipimap)->commit_mutex);
1232 /* acquire tlock of the iag page of the freed ixad
1233 * to force the page NOHOMEOK (even though no data is
1234 * logged from the iag page) until NOREDOPAGE|FREEXTENT log
1235 * for the free of the extent is committed;
1236 * write FREEXTENT|NOREDOPAGE log record
1237 * N.B. linelock is overlaid as freed extent descriptor;
1239 tlck = txLock(tid, ipimap, mp, tlckINODE | tlckFREE);
1240 pxdlock = (struct pxd_lock *) & tlck->lock;
1241 pxdlock->flag = mlckFREEPXD;
1242 pxdlock->pxd = freepxd;
1250 * logredo needs the IAG number and IAG extent index in order
1251 * to ensure that the IMap is consistent. The least disruptive
1252 * way to pass these values through to the transaction manager
1253 * is in the iplist array.
1255 * It's not pretty, but it works.
1257 iplist[1] = (struct inode *) (size_t)iagno;
1258 iplist[2] = (struct inode *) (size_t)extno;
1260 rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE);
1263 mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
1265 /* unlock the AG inode map information */
1266 AG_UNLOCK(imap, agno);
1271 IREAD_UNLOCK(ipimap);
1274 release_metapage(amp);
1276 release_metapage(bmp);
1278 release_metapage(cmp);
1280 release_metapage(dmp);
1282 AG_UNLOCK(imap, agno);
1284 release_metapage(mp);
1290 * There are several places in the diAlloc* routines where we initialize
1294 diInitInode(struct inode *ip, int iagno, int ino, int extno, struct iag * iagp)
1296 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
1298 ip->i_ino = (iagno << L2INOSPERIAG) + ino;
1299 jfs_ip->ixpxd = iagp->inoext[extno];
1300 jfs_ip->agstart = le64_to_cpu(iagp->agstart);
1301 jfs_ip->active_ag = -1;
1306 * NAME: diAlloc(pip,dir,ip)
1308 * FUNCTION: allocate a disk inode from the inode working map
1309 * for a fileset or aggregate.
1312 * pip - pointer to incore inode for the parent inode.
1313 * dir - 'true' if the new disk inode is for a directory.
1314 * ip - pointer to a new inode
1318 * -ENOSPC - insufficient disk resources.
1321 int diAlloc(struct inode *pip, bool dir, struct inode *ip)
1323 int rc, ino, iagno, addext, extno, bitno, sword;
1324 int nwords, rem, i, agno;
1325 u32 mask, inosmap, extsmap;
1326 struct inode *ipimap;
1327 struct metapage *mp;
1330 struct inomap *imap;
1332 /* get the pointers to the inode map inode and the
1333 * corresponding imap control structure.
1335 ipimap = JFS_SBI(pip->i_sb)->ipimap;
1336 imap = JFS_IP(ipimap)->i_imap;
1337 JFS_IP(ip)->ipimap = ipimap;
1338 JFS_IP(ip)->fileset = FILESYSTEM_I;
1340 /* for a directory, the allocation policy is to start
1341 * at the ag level using the preferred ag.
1344 agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap);
1345 AG_LOCK(imap, agno);
1349 /* for files, the policy starts off by trying to allocate from
1350 * the same iag containing the parent disk inode:
1351 * try to allocate the new disk inode close to the parent disk
1352 * inode, using parent disk inode number + 1 as the allocation
1353 * hint. (we use a left-to-right policy to attempt to avoid
1354 * moving backward on the disk.) compute the hint within the
1355 * file system and the iag.
1358 /* get the ag number of this iag */
1359 agno = BLKTOAG(JFS_IP(pip)->agstart, JFS_SBI(pip->i_sb));
1361 if (atomic_read(&JFS_SBI(pip->i_sb)->bmap->db_active[agno])) {
1363 * There is an open file actively growing. We want to
1364 * allocate new inodes from a different ag to avoid
1365 * fragmentation problems.
1367 agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap);
1368 AG_LOCK(imap, agno);
1372 inum = pip->i_ino + 1;
1373 ino = inum & (INOSPERIAG - 1);
1375 /* back off the hint if it is outside of the iag */
1379 /* lock the AG inode map information */
1380 AG_LOCK(imap, agno);
1382 /* Get read lock on imap inode */
1383 IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
1385 /* get the iag number and read the iag */
1386 iagno = INOTOIAG(inum);
1387 if ((rc = diIAGRead(imap, iagno, &mp))) {
1388 IREAD_UNLOCK(ipimap);
1389 AG_UNLOCK(imap, agno);
1392 iagp = (struct iag *) mp->data;
1394 /* determine if new inode extent is allowed to be added to the iag.
1395 * new inode extent can be added to the iag if the ag
1396 * has less than 32 free disk inodes and the iag has free extents.
1398 addext = (imap->im_agctl[agno].numfree < 32 && iagp->nfreeexts);
1401 * try to allocate from the IAG
1403 /* check if the inode may be allocated from the iag
1404 * (i.e. the inode has free inodes or new extent can be added).
1406 if (iagp->nfreeinos || addext) {
1407 /* determine the extent number of the hint.
1409 extno = ino >> L2INOSPEREXT;
1411 /* check if the extent containing the hint has backed
1412 * inodes. if so, try to allocate within this extent.
1414 if (addressPXD(&iagp->inoext[extno])) {
1415 bitno = ino & (INOSPEREXT - 1);
1417 diFindFree(le32_to_cpu(iagp->wmap[extno]),
1420 ino = (extno << L2INOSPEREXT) + bitno;
1422 /* a free inode (bit) was found within this
1423 * extent, so allocate it.
1425 rc = diAllocBit(imap, iagp, ino);
1426 IREAD_UNLOCK(ipimap);
1430 /* set the results of the allocation
1431 * and write the iag.
1433 diInitInode(ip, iagno, ino, extno,
1435 mark_metapage_dirty(mp);
1437 release_metapage(mp);
1439 /* free the AG lock and return.
1441 AG_UNLOCK(imap, agno);
1448 EXTSPERIAG - 1) ? 0 : extno + 1;
1452 * no free inodes within the extent containing the hint.
1454 * try to allocate from the backed extents following
1455 * hint or, if appropriate (i.e. addext is true), allocate
1456 * an extent of free inodes at or following the extent
1457 * containing the hint.
1459 * the free inode and free extent summary maps are used
1460 * here, so determine the starting summary map position
1461 * and the number of words we'll have to examine. again,
1462 * the approach is to allocate following the hint, so we
1463 * might have to initially ignore prior bits of the summary
1464 * map that represent extents prior to the extent containing
1465 * the hint and later revisit these bits.
1467 bitno = extno & (EXTSPERSUM - 1);
1468 nwords = (bitno == 0) ? SMAPSZ : SMAPSZ + 1;
1469 sword = extno >> L2EXTSPERSUM;
1471 /* mask any prior bits for the starting words of the
1474 mask = (bitno == 0) ? 0 : (ONES << (EXTSPERSUM - bitno));
1475 inosmap = le32_to_cpu(iagp->inosmap[sword]) | mask;
1476 extsmap = le32_to_cpu(iagp->extsmap[sword]) | mask;
1478 /* scan the free inode and free extent summary maps for
1481 for (i = 0; i < nwords; i++) {
1482 /* check if this word of the free inode summary
1483 * map describes an extent with free inodes.
1486 /* an extent with free inodes has been
1487 * found. determine the extent number
1488 * and the inode number within the extent.
1490 rem = diFindFree(inosmap, 0);
1491 extno = (sword << L2EXTSPERSUM) + rem;
1492 rem = diFindFree(le32_to_cpu(iagp->wmap[extno]),
1494 if (rem >= INOSPEREXT) {
1495 IREAD_UNLOCK(ipimap);
1496 release_metapage(mp);
1497 AG_UNLOCK(imap, agno);
1499 "can't find free bit in wmap\n");
1503 /* determine the inode number within the
1504 * iag and allocate the inode from the
1507 ino = (extno << L2INOSPEREXT) + rem;
1508 rc = diAllocBit(imap, iagp, ino);
1509 IREAD_UNLOCK(ipimap);
1513 /* set the results of the allocation
1514 * and write the iag.
1516 diInitInode(ip, iagno, ino, extno,
1518 mark_metapage_dirty(mp);
1520 release_metapage(mp);
1522 /* free the AG lock and return.
1524 AG_UNLOCK(imap, agno);
1529 /* check if we may allocate an extent of free
1530 * inodes and whether this word of the free
1531 * extents summary map describes a free extent.
1533 if (addext && ~extsmap) {
1534 /* a free extent has been found. determine
1535 * the extent number.
1537 rem = diFindFree(extsmap, 0);
1538 extno = (sword << L2EXTSPERSUM) + rem;
1540 /* allocate an extent of free inodes.
1542 if ((rc = diNewExt(imap, iagp, extno))) {
1543 /* if there is no disk space for a
1544 * new extent, try to allocate the
1545 * disk inode from somewhere else.
1552 /* set the results of the allocation
1553 * and write the iag.
1555 diInitInode(ip, iagno,
1556 extno << L2INOSPEREXT,
1558 mark_metapage_dirty(mp);
1560 release_metapage(mp);
1561 /* free the imap inode & the AG lock & return.
1563 IREAD_UNLOCK(ipimap);
1564 AG_UNLOCK(imap, agno);
1568 /* move on to the next set of summary map words.
1570 sword = (sword == SMAPSZ - 1) ? 0 : sword + 1;
1571 inosmap = le32_to_cpu(iagp->inosmap[sword]);
1572 extsmap = le32_to_cpu(iagp->extsmap[sword]);
1575 /* unlock imap inode */
1576 IREAD_UNLOCK(ipimap);
1578 /* nothing doing in this iag, so release it. */
1579 release_metapage(mp);
1583 * try to allocate anywhere within the same AG as the parent inode.
1585 rc = diAllocAG(imap, agno, dir, ip);
1587 AG_UNLOCK(imap, agno);
1593 * try to allocate in any AG.
1595 return (diAllocAny(imap, agno, dir, ip));
1600 * NAME: diAllocAG(imap,agno,dir,ip)
1602 * FUNCTION: allocate a disk inode from the allocation group.
1604 * this routine first determines if a new extent of free
1605 * inodes should be added for the allocation group, with
1606 * the current request satisfied from this extent. if this
1607 * is the case, an attempt will be made to do just that. if
1608 * this attempt fails or it has been determined that a new
1609 * extent should not be added, an attempt is made to satisfy
1610 * the request by allocating an existing (backed) free inode
1611 * from the allocation group.
1613 * PRE CONDITION: Already have the AG lock for this AG.
1616 * imap - pointer to inode map control structure.
1617 * agno - allocation group to allocate from.
1618 * dir - 'true' if the new disk inode is for a directory.
1619 * ip - pointer to the new inode to be filled in on successful return
1620 * with the disk inode number allocated, its extent address
1621 * and the start of the ag.
1625 * -ENOSPC - insufficient disk resources.
1629 diAllocAG(struct inomap * imap, int agno, bool dir, struct inode *ip)
1631 int rc, addext, numfree, numinos;
1633 /* get the number of free and the number of backed disk
1634 * inodes currently within the ag.
1636 numfree = imap->im_agctl[agno].numfree;
1637 numinos = imap->im_agctl[agno].numinos;
1639 if (numfree > numinos) {
1640 jfs_error(ip->i_sb, "numfree > numinos\n");
1644 /* determine if we should allocate a new extent of free inodes
1645 * within the ag: for directory inodes, add a new extent
1646 * if there are a small number of free inodes or number of free
1647 * inodes is a small percentage of the number of backed inodes.
1650 addext = (numfree < 64 ||
1652 && ((numfree * 100) / numinos) <= 20));
1654 addext = (numfree == 0);
1657 * try to allocate a new extent of free inodes.
1660 /* if free space is not available for this new extent, try
1661 * below to allocate a free and existing (already backed)
1662 * inode from the ag.
1664 if ((rc = diAllocExt(imap, agno, ip)) != -ENOSPC)
1669 * try to allocate an existing free inode from the ag.
1671 return (diAllocIno(imap, agno, ip));
1676 * NAME: diAllocAny(imap,agno,dir,iap)
1678 * FUNCTION: allocate a disk inode from any other allocation group.
1680 * this routine is called when an allocation attempt within
1681 * the primary allocation group has failed. if attempts to
1682 * allocate an inode from any allocation group other than the
1683 * specified primary group.
1686 * imap - pointer to inode map control structure.
1687 * agno - primary allocation group (to avoid).
1688 * dir - 'true' if the new disk inode is for a directory.
1689 * ip - pointer to a new inode to be filled in on successful return
1690 * with the disk inode number allocated, its extent address
1691 * and the start of the ag.
1695 * -ENOSPC - insufficient disk resources.
1699 diAllocAny(struct inomap * imap, int agno, bool dir, struct inode *ip)
1702 int maxag = JFS_SBI(imap->im_ipimap->i_sb)->bmap->db_maxag;
1705 /* try to allocate from the ags following agno up to
1706 * the maximum ag number.
1708 for (ag = agno + 1; ag <= maxag; ag++) {
1711 rc = diAllocAG(imap, ag, dir, ip);
1713 AG_UNLOCK(imap, ag);
1719 /* try to allocate from the ags in front of agno.
1721 for (ag = 0; ag < agno; ag++) {
1724 rc = diAllocAG(imap, ag, dir, ip);
1726 AG_UNLOCK(imap, ag);
1732 /* no free disk inodes.
1739 * NAME: diAllocIno(imap,agno,ip)
1741 * FUNCTION: allocate a disk inode from the allocation group's free
1742 * inode list, returning an error if this free list is
1743 * empty (i.e. no iags on the list).
1745 * allocation occurs from the first iag on the list using
1746 * the iag's free inode summary map to find the leftmost
1747 * free inode in the iag.
1749 * PRE CONDITION: Already have AG lock for this AG.
1752 * imap - pointer to inode map control structure.
1753 * agno - allocation group.
1754 * ip - pointer to new inode to be filled in on successful return
1755 * with the disk inode number allocated, its extent address
1756 * and the start of the ag.
1760 * -ENOSPC - insufficient disk resources.
1763 static int diAllocIno(struct inomap * imap, int agno, struct inode *ip)
1765 int iagno, ino, rc, rem, extno, sword;
1766 struct metapage *mp;
1769 /* check if there are iags on the ag's free inode list.
1771 if ((iagno = imap->im_agctl[agno].inofree) < 0)
1774 /* obtain read lock on imap inode */
1775 IREAD_LOCK(imap->im_ipimap, RDWRLOCK_IMAP);
1777 /* read the iag at the head of the list.
1779 if ((rc = diIAGRead(imap, iagno, &mp))) {
1780 IREAD_UNLOCK(imap->im_ipimap);
1783 iagp = (struct iag *) mp->data;
1785 /* better be free inodes in this iag if it is on the
1788 if (!iagp->nfreeinos) {
1789 IREAD_UNLOCK(imap->im_ipimap);
1790 release_metapage(mp);
1791 jfs_error(ip->i_sb, "nfreeinos = 0, but iag on freelist\n");
1795 /* scan the free inode summary map to find an extent
1798 for (sword = 0;; sword++) {
1799 if (sword >= SMAPSZ) {
1800 IREAD_UNLOCK(imap->im_ipimap);
1801 release_metapage(mp);
1803 "free inode not found in summary map\n");
1807 if (~iagp->inosmap[sword])
1811 /* found a extent with free inodes. determine
1812 * the extent number.
1814 rem = diFindFree(le32_to_cpu(iagp->inosmap[sword]), 0);
1815 if (rem >= EXTSPERSUM) {
1816 IREAD_UNLOCK(imap->im_ipimap);
1817 release_metapage(mp);
1818 jfs_error(ip->i_sb, "no free extent found\n");
1821 extno = (sword << L2EXTSPERSUM) + rem;
1823 /* find the first free inode in the extent.
1825 rem = diFindFree(le32_to_cpu(iagp->wmap[extno]), 0);
1826 if (rem >= INOSPEREXT) {
1827 IREAD_UNLOCK(imap->im_ipimap);
1828 release_metapage(mp);
1829 jfs_error(ip->i_sb, "free inode not found\n");
1833 /* compute the inode number within the iag.
1835 ino = (extno << L2INOSPEREXT) + rem;
1837 /* allocate the inode.
1839 rc = diAllocBit(imap, iagp, ino);
1840 IREAD_UNLOCK(imap->im_ipimap);
1842 release_metapage(mp);
1846 /* set the results of the allocation and write the iag.
1848 diInitInode(ip, iagno, ino, extno, iagp);
1856 * NAME: diAllocExt(imap,agno,ip)
1858 * FUNCTION: add a new extent of free inodes to an iag, allocating
1859 * an inode from this extent to satisfy the current allocation
1862 * this routine first tries to find an existing iag with free
1863 * extents through the ag free extent list. if list is not
1864 * empty, the head of the list will be selected as the home
1865 * of the new extent of free inodes. otherwise (the list is
1866 * empty), a new iag will be allocated for the ag to contain
1869 * once an iag has been selected, the free extent summary map
1870 * is used to locate a free extent within the iag and diNewExt()
1871 * is called to initialize the extent, with initialization
1872 * including the allocation of the first inode of the extent
1873 * for the purpose of satisfying this request.
1876 * imap - pointer to inode map control structure.
1877 * agno - allocation group number.
1878 * ip - pointer to new inode to be filled in on successful return
1879 * with the disk inode number allocated, its extent address
1880 * and the start of the ag.
1884 * -ENOSPC - insufficient disk resources.
1887 static int diAllocExt(struct inomap * imap, int agno, struct inode *ip)
1889 int rem, iagno, sword, extno, rc;
1890 struct metapage *mp;
1893 /* check if the ag has any iags with free extents. if not,
1894 * allocate a new iag for the ag.
1896 if ((iagno = imap->im_agctl[agno].extfree) < 0) {
1897 /* If successful, diNewIAG will obtain the read lock on the
1900 if ((rc = diNewIAG(imap, &iagno, agno, &mp))) {
1903 iagp = (struct iag *) mp->data;
1905 /* set the ag number if this a brand new iag
1908 cpu_to_le64(AGTOBLK(agno, imap->im_ipimap));
1912 IREAD_LOCK(imap->im_ipimap, RDWRLOCK_IMAP);
1913 if ((rc = diIAGRead(imap, iagno, &mp))) {
1914 IREAD_UNLOCK(imap->im_ipimap);
1915 jfs_error(ip->i_sb, "error reading iag\n");
1918 iagp = (struct iag *) mp->data;
1921 /* using the free extent summary map, find a free extent.
1923 for (sword = 0;; sword++) {
1924 if (sword >= SMAPSZ) {
1925 release_metapage(mp);
1926 IREAD_UNLOCK(imap->im_ipimap);
1927 jfs_error(ip->i_sb, "free ext summary map not found\n");
1930 if (~iagp->extsmap[sword])
1934 /* determine the extent number of the free extent.
1936 rem = diFindFree(le32_to_cpu(iagp->extsmap[sword]), 0);
1937 if (rem >= EXTSPERSUM) {
1938 release_metapage(mp);
1939 IREAD_UNLOCK(imap->im_ipimap);
1940 jfs_error(ip->i_sb, "free extent not found\n");
1943 extno = (sword << L2EXTSPERSUM) + rem;
1945 /* initialize the new extent.
1947 rc = diNewExt(imap, iagp, extno);
1948 IREAD_UNLOCK(imap->im_ipimap);
1950 /* something bad happened. if a new iag was allocated,
1951 * place it back on the inode map's iag free list, and
1952 * clear the ag number information.
1954 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
1956 iagp->iagfree = cpu_to_le32(imap->im_freeiag);
1957 imap->im_freeiag = iagno;
1958 IAGFREE_UNLOCK(imap);
1964 /* set the results of the allocation and write the iag.
1966 diInitInode(ip, iagno, extno << L2INOSPEREXT, extno, iagp);
1975 * NAME: diAllocBit(imap,iagp,ino)
1977 * FUNCTION: allocate a backed inode from an iag.
1979 * this routine performs the mechanics of allocating a
1980 * specified inode from a backed extent.
1982 * if the inode to be allocated represents the last free
1983 * inode within the iag, the iag will be removed from the
1984 * ag free inode list.
1986 * a careful update approach is used to provide consistency
1987 * in the face of updates to multiple buffers. under this
1988 * approach, all required buffers are obtained before making
1989 * any updates and are held all are updates are complete.
1991 * PRE CONDITION: Already have buffer lock on iagp. Already have AG lock on
1992 * this AG. Must have read lock on imap inode.
1995 * imap - pointer to inode map control structure.
1996 * iagp - pointer to iag.
1997 * ino - inode number to be allocated within the iag.
2001 * -ENOSPC - insufficient disk resources.
2004 static int diAllocBit(struct inomap * imap, struct iag * iagp, int ino)
2006 int extno, bitno, agno, sword, rc;
2007 struct metapage *amp = NULL, *bmp = NULL;
2008 struct iag *aiagp = NULL, *biagp = NULL;
2011 /* check if this is the last free inode within the iag.
2012 * if so, it will have to be removed from the ag free
2013 * inode list, so get the iags preceding and following
2016 if (iagp->nfreeinos == cpu_to_le32(1)) {
2017 if ((int) le32_to_cpu(iagp->inofreefwd) >= 0) {
2019 diIAGRead(imap, le32_to_cpu(iagp->inofreefwd),
2022 aiagp = (struct iag *) amp->data;
2025 if ((int) le32_to_cpu(iagp->inofreeback) >= 0) {
2028 le32_to_cpu(iagp->inofreeback),
2031 release_metapage(amp);
2034 biagp = (struct iag *) bmp->data;
2038 /* get the ag number, extent number, inode number within
2041 agno = BLKTOAG(le64_to_cpu(iagp->agstart), JFS_SBI(imap->im_ipimap->i_sb));
2042 extno = ino >> L2INOSPEREXT;
2043 bitno = ino & (INOSPEREXT - 1);
2045 /* compute the mask for setting the map.
2047 mask = HIGHORDER >> bitno;
2049 /* the inode should be free and backed.
2051 if (((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) ||
2052 ((le32_to_cpu(iagp->wmap[extno]) & mask) != 0) ||
2053 (addressPXD(&iagp->inoext[extno]) == 0)) {
2055 release_metapage(amp);
2057 release_metapage(bmp);
2059 jfs_error(imap->im_ipimap->i_sb, "iag inconsistent\n");
2063 /* mark the inode as allocated in the working map.
2065 iagp->wmap[extno] |= cpu_to_le32(mask);
2067 /* check if all inodes within the extent are now
2068 * allocated. if so, update the free inode summary
2069 * map to reflect this.
2071 if (iagp->wmap[extno] == cpu_to_le32(ONES)) {
2072 sword = extno >> L2EXTSPERSUM;
2073 bitno = extno & (EXTSPERSUM - 1);
2074 iagp->inosmap[sword] |= cpu_to_le32(HIGHORDER >> bitno);
2077 /* if this was the last free inode in the iag, remove the
2078 * iag from the ag free inode list.
2080 if (iagp->nfreeinos == cpu_to_le32(1)) {
2082 aiagp->inofreeback = iagp->inofreeback;
2083 write_metapage(amp);
2087 biagp->inofreefwd = iagp->inofreefwd;
2088 write_metapage(bmp);
2090 imap->im_agctl[agno].inofree =
2091 le32_to_cpu(iagp->inofreefwd);
2093 iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1);
2096 /* update the free inode count at the iag, ag, inode
2099 le32_add_cpu(&iagp->nfreeinos, -1);
2100 imap->im_agctl[agno].numfree -= 1;
2101 atomic_dec(&imap->im_numfree);
2108 * NAME: diNewExt(imap,iagp,extno)
2110 * FUNCTION: initialize a new extent of inodes for an iag, allocating
2111 * the first inode of the extent for use for the current
2112 * allocation request.
2114 * disk resources are allocated for the new extent of inodes
2115 * and the inodes themselves are initialized to reflect their
2116 * existence within the extent (i.e. their inode numbers and
2117 * inode extent addresses are set) and their initial state
2118 * (mode and link count are set to zero).
2120 * if the iag is new, it is not yet on an ag extent free list
2121 * but will now be placed on this list.
2123 * if the allocation of the new extent causes the iag to
2124 * have no free extent, the iag will be removed from the
2125 * ag extent free list.
2127 * if the iag has no free backed inodes, it will be placed
2128 * on the ag free inode list, since the addition of the new
2129 * extent will now cause it to have free inodes.
2131 * a careful update approach is used to provide consistency
2132 * (i.e. list consistency) in the face of updates to multiple
2133 * buffers. under this approach, all required buffers are
2134 * obtained before making any updates and are held until all
2135 * updates are complete.
2137 * PRE CONDITION: Already have buffer lock on iagp. Already have AG lock on
2138 * this AG. Must have read lock on imap inode.
2141 * imap - pointer to inode map control structure.
2142 * iagp - pointer to iag.
2143 * extno - extent number.
2147 * -ENOSPC - insufficient disk resources.
2150 static int diNewExt(struct inomap * imap, struct iag * iagp, int extno)
2152 int agno, iagno, fwd, back, freei = 0, sword, rc;
2153 struct iag *aiagp = NULL, *biagp = NULL, *ciagp = NULL;
2154 struct metapage *amp, *bmp, *cmp, *dmp;
2155 struct inode *ipimap;
2161 struct jfs_sb_info *sbi;
2163 /* better have free extents.
2165 if (!iagp->nfreeexts) {
2166 jfs_error(imap->im_ipimap->i_sb, "no free extents\n");
2170 /* get the inode map inode.
2172 ipimap = imap->im_ipimap;
2173 sbi = JFS_SBI(ipimap->i_sb);
2175 amp = bmp = cmp = NULL;
2177 /* get the ag and iag numbers for this iag.
2179 agno = BLKTOAG(le64_to_cpu(iagp->agstart), sbi);
2180 iagno = le32_to_cpu(iagp->iagnum);
2182 /* check if this is the last free extent within the
2183 * iag. if so, the iag must be removed from the ag
2184 * free extent list, so get the iags preceding and
2185 * following the iag on this list.
2187 if (iagp->nfreeexts == cpu_to_le32(1)) {
2188 if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) {
2189 if ((rc = diIAGRead(imap, fwd, &)))
2191 aiagp = (struct iag *) amp->data;
2194 if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) {
2195 if ((rc = diIAGRead(imap, back, &bmp)))
2197 biagp = (struct iag *) bmp->data;
2200 /* the iag has free extents. if all extents are free
2201 * (as is the case for a newly allocated iag), the iag
2202 * must be added to the ag free extent list, so get
2203 * the iag at the head of the list in preparation for
2204 * adding this iag to this list.
2207 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2208 if ((fwd = imap->im_agctl[agno].extfree) >= 0) {
2209 if ((rc = diIAGRead(imap, fwd, &)))
2211 aiagp = (struct iag *) amp->data;
2216 /* check if the iag has no free inodes. if so, the iag
2217 * will have to be added to the ag free inode list, so get
2218 * the iag at the head of the list in preparation for
2219 * adding this iag to this list. in doing this, we must
2220 * check if we already have the iag at the head of
2223 if (iagp->nfreeinos == 0) {
2224 freei = imap->im_agctl[agno].inofree;
2229 } else if (freei == back) {
2232 if ((rc = diIAGRead(imap, freei, &cmp)))
2234 ciagp = (struct iag *) cmp->data;
2236 if (ciagp == NULL) {
2237 jfs_error(imap->im_ipimap->i_sb,
2245 /* allocate disk space for the inode extent.
2247 if ((extno == 0) || (addressPXD(&iagp->inoext[extno - 1]) == 0))
2248 hint = ((s64) agno << sbi->bmap->db_agl2size) - 1;
2250 hint = addressPXD(&iagp->inoext[extno - 1]) +
2251 lengthPXD(&iagp->inoext[extno - 1]) - 1;
2253 if ((rc = dbAlloc(ipimap, hint, (s64) imap->im_nbperiext, &blkno)))
2256 /* compute the inode number of the first inode within the
2259 ino = (iagno << L2INOSPERIAG) + (extno << L2INOSPEREXT);
2261 /* initialize the inodes within the newly allocated extent a
2264 for (i = 0; i < imap->im_nbperiext; i += sbi->nbperpage) {
2265 /* get a buffer for this page of disk inodes.
2267 dmp = get_metapage(ipimap, blkno + i, PSIZE, 1);
2272 dp = (struct dinode *) dmp->data;
2274 /* initialize the inode number, mode, link count and
2275 * inode extent address.
2277 for (j = 0; j < INOSPERPAGE; j++, dp++, ino++) {
2278 dp->di_inostamp = cpu_to_le32(sbi->inostamp);
2279 dp->di_number = cpu_to_le32(ino);
2280 dp->di_fileset = cpu_to_le32(FILESYSTEM_I);
2283 PXDaddress(&(dp->di_ixpxd), blkno);
2284 PXDlength(&(dp->di_ixpxd), imap->im_nbperiext);
2286 write_metapage(dmp);
2289 /* if this is the last free extent within the iag, remove the
2290 * iag from the ag free extent list.
2292 if (iagp->nfreeexts == cpu_to_le32(1)) {
2294 aiagp->extfreeback = iagp->extfreeback;
2297 biagp->extfreefwd = iagp->extfreefwd;
2299 imap->im_agctl[agno].extfree =
2300 le32_to_cpu(iagp->extfreefwd);
2302 iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1);
2304 /* if the iag has all free extents (newly allocated iag),
2305 * add the iag to the ag free extent list.
2307 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2309 aiagp->extfreeback = cpu_to_le32(iagno);
2311 iagp->extfreefwd = cpu_to_le32(fwd);
2312 iagp->extfreeback = cpu_to_le32(-1);
2313 imap->im_agctl[agno].extfree = iagno;
2317 /* if the iag has no free inodes, add the iag to the
2318 * ag free inode list.
2320 if (iagp->nfreeinos == 0) {
2322 ciagp->inofreeback = cpu_to_le32(iagno);
2325 cpu_to_le32(imap->im_agctl[agno].inofree);
2326 iagp->inofreeback = cpu_to_le32(-1);
2327 imap->im_agctl[agno].inofree = iagno;
2330 /* initialize the extent descriptor of the extent. */
2331 PXDlength(&iagp->inoext[extno], imap->im_nbperiext);
2332 PXDaddress(&iagp->inoext[extno], blkno);
2334 /* initialize the working and persistent map of the extent.
2335 * the working map will be initialized such that
2336 * it indicates the first inode of the extent is allocated.
2338 iagp->wmap[extno] = cpu_to_le32(HIGHORDER);
2339 iagp->pmap[extno] = 0;
2341 /* update the free inode and free extent summary maps
2342 * for the extent to indicate the extent has free inodes
2343 * and no longer represents a free extent.
2345 sword = extno >> L2EXTSPERSUM;
2346 mask = HIGHORDER >> (extno & (EXTSPERSUM - 1));
2347 iagp->extsmap[sword] |= cpu_to_le32(mask);
2348 iagp->inosmap[sword] &= cpu_to_le32(~mask);
2350 /* update the free inode and free extent counts for the
2353 le32_add_cpu(&iagp->nfreeinos, (INOSPEREXT - 1));
2354 le32_add_cpu(&iagp->nfreeexts, -1);
2356 /* update the free and backed inode counts for the ag.
2358 imap->im_agctl[agno].numfree += (INOSPEREXT - 1);
2359 imap->im_agctl[agno].numinos += INOSPEREXT;
2361 /* update the free and backed inode counts for the inode map.
2363 atomic_add(INOSPEREXT - 1, &imap->im_numfree);
2364 atomic_add(INOSPEREXT, &imap->im_numinos);
2369 write_metapage(amp);
2371 write_metapage(bmp);
2373 write_metapage(cmp);
2379 /* release the iags.
2382 release_metapage(amp);
2384 release_metapage(bmp);
2386 release_metapage(cmp);
2393 * NAME: diNewIAG(imap,iagnop,agno)
2395 * FUNCTION: allocate a new iag for an allocation group.
2397 * first tries to allocate the iag from the inode map
2399 * if the list has free iags, the head of the list is removed
2400 * and returned to satisfy the request.
2401 * if the inode map's iag free list is empty, the inode map
2402 * is extended to hold a new iag. this new iag is initialized
2403 * and returned to satisfy the request.
2406 * imap - pointer to inode map control structure.
2407 * iagnop - pointer to an iag number set with the number of the
2408 * newly allocated iag upon successful return.
2409 * agno - allocation group number.
2410 * bpp - Buffer pointer to be filled in with new IAG's buffer
2414 * -ENOSPC - insufficient disk resources.
2418 * AG lock held on entry/exit;
2419 * write lock on the map is held inside;
2420 * read lock on the map is held on successful completion;
2422 * note: new iag transaction:
2423 * . synchronously write iag;
2424 * . write log of xtree and inode of imap;
2426 * . synchronous write of xtree (right to left, bottom to top);
2427 * . at start of logredo(): init in-memory imap with one additional iag page;
2428 * . at end of logredo(): re-read imap inode to determine
2432 diNewIAG(struct inomap * imap, int *iagnop, int agno, struct metapage ** mpp)
2436 struct inode *ipimap;
2437 struct super_block *sb;
2438 struct jfs_sb_info *sbi;
2439 struct metapage *mp;
2444 struct inode *iplist[1];
2446 /* pick up pointers to the inode map and mount inodes */
2447 ipimap = imap->im_ipimap;
2451 /* acquire the free iag lock */
2454 /* if there are any iags on the inode map free iag list,
2455 * allocate the iag from the head of the list.
2457 if (imap->im_freeiag >= 0) {
2458 /* pick up the iag number at the head of the list */
2459 iagno = imap->im_freeiag;
2461 /* determine the logical block number of the iag */
2462 blkno = IAGTOLBLK(iagno, sbi->l2nbperpage);
2464 /* no free iags. the inode map will have to be extented
2465 * to include a new iag.
2468 /* acquire inode map lock */
2469 IWRITE_LOCK(ipimap, RDWRLOCK_IMAP);
2471 if (ipimap->i_size >> L2PSIZE != imap->im_nextiag + 1) {
2472 IWRITE_UNLOCK(ipimap);
2473 IAGFREE_UNLOCK(imap);
2474 jfs_error(imap->im_ipimap->i_sb,
2475 "ipimap->i_size is wrong\n");
2480 /* get the next available iag number */
2481 iagno = imap->im_nextiag;
2483 /* make sure that we have not exceeded the maximum inode
2486 if (iagno > (MAXIAGS - 1)) {
2487 /* release the inode map lock */
2488 IWRITE_UNLOCK(ipimap);
2495 * synchronously append new iag page.
2497 /* determine the logical address of iag page to append */
2498 blkno = IAGTOLBLK(iagno, sbi->l2nbperpage);
2500 /* Allocate extent for new iag page */
2501 xlen = sbi->nbperpage;
2502 if ((rc = dbAlloc(ipimap, 0, (s64) xlen, &xaddr))) {
2503 /* release the inode map lock */
2504 IWRITE_UNLOCK(ipimap);
2510 * start transaction of update of the inode map
2511 * addressing structure pointing to the new iag page;
2513 tid = txBegin(sb, COMMIT_FORCE);
2514 mutex_lock(&JFS_IP(ipimap)->commit_mutex);
2516 /* update the inode map addressing structure to point to it */
2518 xtInsert(tid, ipimap, 0, blkno, xlen, &xaddr, 0))) {
2520 mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
2521 /* Free the blocks allocated for the iag since it was
2522 * not successfully added to the inode map
2524 dbFree(ipimap, xaddr, (s64) xlen);
2526 /* release the inode map lock */
2527 IWRITE_UNLOCK(ipimap);
2532 /* update the inode map's inode to reflect the extension */
2533 ipimap->i_size += PSIZE;
2534 inode_add_bytes(ipimap, PSIZE);
2536 /* assign a buffer for the page */
2537 mp = get_metapage(ipimap, blkno, PSIZE, 0);
2540 * This is very unlikely since we just created the
2541 * extent, but let's try to handle it correctly
2543 xtTruncate(tid, ipimap, ipimap->i_size - PSIZE,
2548 mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
2550 /* release the inode map lock */
2551 IWRITE_UNLOCK(ipimap);
2556 iagp = (struct iag *) mp->data;
2559 memset(iagp, 0, sizeof(struct iag));
2560 iagp->iagnum = cpu_to_le32(iagno);
2561 iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1);
2562 iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1);
2563 iagp->iagfree = cpu_to_le32(-1);
2564 iagp->nfreeinos = 0;
2565 iagp->nfreeexts = cpu_to_le32(EXTSPERIAG);
2567 /* initialize the free inode summary map (free extent
2568 * summary map initialization handled by bzero).
2570 for (i = 0; i < SMAPSZ; i++)
2571 iagp->inosmap[i] = cpu_to_le32(ONES);
2574 * Write and sync the metapage
2579 * txCommit(COMMIT_FORCE) will synchronously write address
2580 * index pages and inode after commit in careful update order
2581 * of address index pages (right to left, bottom up);
2584 rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE);
2587 mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
2589 duplicateIXtree(sb, blkno, xlen, &xaddr);
2591 /* update the next available iag number */
2592 imap->im_nextiag += 1;
2594 /* Add the iag to the iag free list so we don't lose the iag
2595 * if a failure happens now.
2597 imap->im_freeiag = iagno;
2599 /* Until we have logredo working, we want the imap inode &
2600 * control page to be up to date.
2604 /* release the inode map lock */
2605 IWRITE_UNLOCK(ipimap);
2608 /* obtain read lock on map */
2609 IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
2612 if ((rc = diIAGRead(imap, iagno, &mp))) {
2613 IREAD_UNLOCK(ipimap);
2617 iagp = (struct iag *) mp->data;
2619 /* remove the iag from the iag free list */
2620 imap->im_freeiag = le32_to_cpu(iagp->iagfree);
2621 iagp->iagfree = cpu_to_le32(-1);
2623 /* set the return iag number and buffer pointer */
2628 /* release the iag free lock */
2629 IAGFREE_UNLOCK(imap);
2637 * FUNCTION: get the buffer for the specified iag within a fileset
2638 * or aggregate inode map.
2641 * imap - pointer to inode map control structure.
2642 * iagno - iag number.
2643 * bpp - point to buffer pointer to be filled in on successful
2647 * must have read lock on imap inode
2648 * (When called by diExtendFS, the filesystem is quiesced, therefore
2649 * the read lock is unnecessary.)
2655 static int diIAGRead(struct inomap * imap, int iagno, struct metapage ** mpp)
2657 struct inode *ipimap = imap->im_ipimap;
2660 /* compute the logical block number of the iag. */
2661 blkno = IAGTOLBLK(iagno, JFS_SBI(ipimap->i_sb)->l2nbperpage);
2664 *mpp = read_metapage(ipimap, blkno, PSIZE, 0);
2673 * NAME: diFindFree()
2675 * FUNCTION: find the first free bit in a word starting at
2676 * the specified bit position.
2679 * word - word to be examined.
2680 * start - starting bit position.
2683 * bit position of first free bit in the word or 32 if
2684 * no free bits were found.
2686 static int diFindFree(u32 word, int start)
2690 /* scan the word for the first free bit. */
2691 for (word <<= start, bitno = start; bitno < 32;
2692 bitno++, word <<= 1) {
2693 if ((word & HIGHORDER) == 0)
2700 * NAME: diUpdatePMap()
2702 * FUNCTION: Update the persistent map in an IAG for the allocation or
2703 * freeing of the specified inode.
2705 * PRE CONDITIONS: Working map has already been updated for allocate.
2708 * ipimap - Incore inode map inode
2709 * inum - Number of inode to mark in permanent map
2710 * is_free - If 'true' indicates inode should be marked freed, otherwise
2711 * indicates inode should be marked allocated.
2717 diUpdatePMap(struct inode *ipimap,
2718 unsigned long inum, bool is_free, struct tblock * tblk)
2722 struct metapage *mp;
2723 int iagno, ino, extno, bitno;
2724 struct inomap *imap;
2726 struct jfs_log *log;
2727 int lsn, difft, diffp;
2728 unsigned long flags;
2730 imap = JFS_IP(ipimap)->i_imap;
2731 /* get the iag number containing the inode */
2732 iagno = INOTOIAG(inum);
2733 /* make sure that the iag is contained within the map */
2734 if (iagno >= imap->im_nextiag) {
2735 jfs_error(ipimap->i_sb, "the iag is outside the map\n");
2739 IREAD_LOCK(ipimap, RDWRLOCK_IMAP);
2740 rc = diIAGRead(imap, iagno, &mp);
2741 IREAD_UNLOCK(ipimap);
2744 metapage_wait_for_io(mp);
2745 iagp = (struct iag *) mp->data;
2746 /* get the inode number and extent number of the inode within
2747 * the iag and the inode number within the extent.
2749 ino = inum & (INOSPERIAG - 1);
2750 extno = ino >> L2INOSPEREXT;
2751 bitno = ino & (INOSPEREXT - 1);
2752 mask = HIGHORDER >> bitno;
2754 * mark the inode free in persistent map:
2757 /* The inode should have been allocated both in working
2758 * map and in persistent map;
2759 * the inode will be freed from working map at the release
2760 * of last reference release;
2762 if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
2763 jfs_error(ipimap->i_sb,
2764 "inode %ld not marked as allocated in wmap!\n",
2767 if (!(le32_to_cpu(iagp->pmap[extno]) & mask)) {
2768 jfs_error(ipimap->i_sb,
2769 "inode %ld not marked as allocated in pmap!\n",
2772 /* update the bitmap for the extent of the freed inode */
2773 iagp->pmap[extno] &= cpu_to_le32(~mask);
2776 * mark the inode allocated in persistent map:
2779 /* The inode should be already allocated in the working map
2780 * and should be free in persistent map;
2782 if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
2783 release_metapage(mp);
2784 jfs_error(ipimap->i_sb,
2785 "the inode is not allocated in the working map\n");
2788 if ((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) {
2789 release_metapage(mp);
2790 jfs_error(ipimap->i_sb,
2791 "the inode is not free in the persistent map\n");
2794 /* update the bitmap for the extent of the allocated inode */
2795 iagp->pmap[extno] |= cpu_to_le32(mask);
2801 log = JFS_SBI(tblk->sb)->log;
2802 LOGSYNC_LOCK(log, flags);
2804 /* inherit older/smaller lsn */
2805 logdiff(difft, lsn, log);
2806 logdiff(diffp, mp->lsn, log);
2807 if (difft < diffp) {
2809 /* move mp after tblock in logsync list */
2810 list_move(&mp->synclist, &tblk->synclist);
2812 /* inherit younger/larger clsn */
2814 logdiff(difft, tblk->clsn, log);
2815 logdiff(diffp, mp->clsn, log);
2817 mp->clsn = tblk->clsn;
2821 /* insert mp after tblock in logsync list */
2823 list_add(&mp->synclist, &tblk->synclist);
2824 mp->clsn = tblk->clsn;
2826 LOGSYNC_UNLOCK(log, flags);
2834 * function: update imap for extendfs();
2836 * note: AG size has been increased s.t. each k old contiguous AGs are
2837 * coalesced into a new AG;
2839 int diExtendFS(struct inode *ipimap, struct inode *ipbmap)
2842 struct inomap *imap = JFS_IP(ipimap)->i_imap;
2843 struct iag *iagp = NULL, *hiagp = NULL;
2844 struct bmap *mp = JFS_SBI(ipbmap->i_sb)->bmap;
2845 struct metapage *bp, *hbp;
2847 int numinos, xnuminos = 0, xnumfree = 0;
2850 jfs_info("diExtendFS: nextiag:%d numinos:%d numfree:%d",
2851 imap->im_nextiag, atomic_read(&imap->im_numinos),
2852 atomic_read(&imap->im_numfree));
2857 * coalesce contiguous k (newAGSize/oldAGSize) AGs;
2858 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn;
2859 * note: new AG size = old AG size * (2**x).
2862 /* init per AG control information im_agctl[] */
2863 for (i = 0; i < MAXAG; i++) {
2864 imap->im_agctl[i].inofree = -1;
2865 imap->im_agctl[i].extfree = -1;
2866 imap->im_agctl[i].numinos = 0; /* number of backed inodes */
2867 imap->im_agctl[i].numfree = 0; /* number of free backed inodes */
2871 * process each iag page of the map.
2873 * rebuild AG Free Inode List, AG Free Inode Extent List;
2875 for (i = 0; i < imap->im_nextiag; i++) {
2876 if ((rc = diIAGRead(imap, i, &bp))) {
2880 iagp = (struct iag *) bp->data;
2881 if (le32_to_cpu(iagp->iagnum) != i) {
2882 release_metapage(bp);
2883 jfs_error(ipimap->i_sb, "unexpected value of iagnum\n");
2887 /* leave free iag in the free iag list */
2888 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2889 release_metapage(bp);
2893 agstart = le64_to_cpu(iagp->agstart);
2894 n = agstart >> mp->db_agl2size;
2895 iagp->agstart = cpu_to_le64((s64)n << mp->db_agl2size);
2897 /* compute backed inodes */
2898 numinos = (EXTSPERIAG - le32_to_cpu(iagp->nfreeexts))
2901 /* merge AG backed inodes */
2902 imap->im_agctl[n].numinos += numinos;
2903 xnuminos += numinos;
2906 /* if any backed free inodes, insert at AG free inode list */
2907 if ((int) le32_to_cpu(iagp->nfreeinos) > 0) {
2908 if ((head = imap->im_agctl[n].inofree) == -1) {
2909 iagp->inofreefwd = cpu_to_le32(-1);
2910 iagp->inofreeback = cpu_to_le32(-1);
2912 if ((rc = diIAGRead(imap, head, &hbp))) {
2916 hiagp = (struct iag *) hbp->data;
2917 hiagp->inofreeback = iagp->iagnum;
2918 iagp->inofreefwd = cpu_to_le32(head);
2919 iagp->inofreeback = cpu_to_le32(-1);
2920 write_metapage(hbp);
2923 imap->im_agctl[n].inofree =
2924 le32_to_cpu(iagp->iagnum);
2926 /* merge AG backed free inodes */
2927 imap->im_agctl[n].numfree +=
2928 le32_to_cpu(iagp->nfreeinos);
2929 xnumfree += le32_to_cpu(iagp->nfreeinos);
2932 /* if any free extents, insert at AG free extent list */
2933 if (le32_to_cpu(iagp->nfreeexts) > 0) {
2934 if ((head = imap->im_agctl[n].extfree) == -1) {
2935 iagp->extfreefwd = cpu_to_le32(-1);
2936 iagp->extfreeback = cpu_to_le32(-1);
2938 if ((rc = diIAGRead(imap, head, &hbp))) {
2942 hiagp = (struct iag *) hbp->data;
2943 hiagp->extfreeback = iagp->iagnum;
2944 iagp->extfreefwd = cpu_to_le32(head);
2945 iagp->extfreeback = cpu_to_le32(-1);
2946 write_metapage(hbp);
2949 imap->im_agctl[n].extfree =
2950 le32_to_cpu(iagp->iagnum);
2957 if (xnuminos != atomic_read(&imap->im_numinos) ||
2958 xnumfree != atomic_read(&imap->im_numfree)) {
2959 jfs_error(ipimap->i_sb, "numinos or numfree incorrect\n");
2970 * serialization: IWRITE_LOCK held on entry/exit
2972 * note: shadow page with regular inode (rel.2);
2974 static void duplicateIXtree(struct super_block *sb, s64 blkno,
2975 int xlen, s64 *xaddr)
2977 struct jfs_superblock *j_sb;
2978 struct buffer_head *bh;
2982 /* if AIT2 ipmap2 is bad, do not try to update it */
2983 if (JFS_SBI(sb)->mntflag & JFS_BAD_SAIT) /* s_flag */
2985 ip = diReadSpecial(sb, FILESYSTEM_I, 1);
2987 JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT;
2988 if (readSuper(sb, &bh))
2990 j_sb = (struct jfs_superblock *)bh->b_data;
2991 j_sb->s_flag |= cpu_to_le32(JFS_BAD_SAIT);
2993 mark_buffer_dirty(bh);
2994 sync_dirty_buffer(bh);
2999 /* start transaction */
3000 tid = txBegin(sb, COMMIT_FORCE);
3001 /* update the inode map addressing structure to point to it */
3002 if (xtInsert(tid, ip, 0, blkno, xlen, xaddr, 0)) {
3003 JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT;
3008 /* update the inode map's inode to reflect the extension */
3009 ip->i_size += PSIZE;
3010 inode_add_bytes(ip, PSIZE);
3011 txCommit(tid, 1, &ip, COMMIT_FORCE);
3018 * NAME: copy_from_dinode()
3020 * FUNCTION: Copies inode info from disk inode to in-memory inode
3024 * -ENOMEM - insufficient memory
3026 static int copy_from_dinode(struct dinode * dip, struct inode *ip)
3028 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
3029 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
3031 jfs_ip->fileset = le32_to_cpu(dip->di_fileset);
3032 jfs_ip->mode2 = le32_to_cpu(dip->di_mode);
3033 jfs_set_inode_flags(ip);
3035 ip->i_mode = le32_to_cpu(dip->di_mode) & 0xffff;
3036 if (sbi->umask != -1) {
3037 ip->i_mode = (ip->i_mode & ~0777) | (0777 & ~sbi->umask);
3038 /* For directories, add x permission if r is allowed by umask */
3039 if (S_ISDIR(ip->i_mode)) {
3040 if (ip->i_mode & 0400)
3042 if (ip->i_mode & 0040)
3044 if (ip->i_mode & 0004)
3048 set_nlink(ip, le32_to_cpu(dip->di_nlink));
3050 jfs_ip->saved_uid = make_kuid(&init_user_ns, le32_to_cpu(dip->di_uid));
3051 if (!uid_valid(sbi->uid))
3052 ip->i_uid = jfs_ip->saved_uid;
3054 ip->i_uid = sbi->uid;
3057 jfs_ip->saved_gid = make_kgid(&init_user_ns, le32_to_cpu(dip->di_gid));
3058 if (!gid_valid(sbi->gid))
3059 ip->i_gid = jfs_ip->saved_gid;
3061 ip->i_gid = sbi->gid;
3064 ip->i_size = le64_to_cpu(dip->di_size);
3065 ip->i_atime.tv_sec = le32_to_cpu(dip->di_atime.tv_sec);
3066 ip->i_atime.tv_nsec = le32_to_cpu(dip->di_atime.tv_nsec);
3067 ip->i_mtime.tv_sec = le32_to_cpu(dip->di_mtime.tv_sec);
3068 ip->i_mtime.tv_nsec = le32_to_cpu(dip->di_mtime.tv_nsec);
3069 ip->i_ctime.tv_sec = le32_to_cpu(dip->di_ctime.tv_sec);
3070 ip->i_ctime.tv_nsec = le32_to_cpu(dip->di_ctime.tv_nsec);
3071 ip->i_blocks = LBLK2PBLK(ip->i_sb, le64_to_cpu(dip->di_nblocks));
3072 ip->i_generation = le32_to_cpu(dip->di_gen);
3074 jfs_ip->ixpxd = dip->di_ixpxd; /* in-memory pxd's are little-endian */
3075 jfs_ip->acl = dip->di_acl; /* as are dxd's */
3076 jfs_ip->ea = dip->di_ea;
3077 jfs_ip->next_index = le32_to_cpu(dip->di_next_index);
3078 jfs_ip->otime = le32_to_cpu(dip->di_otime.tv_sec);
3079 jfs_ip->acltype = le32_to_cpu(dip->di_acltype);
3081 if (S_ISCHR(ip->i_mode) || S_ISBLK(ip->i_mode)) {
3082 jfs_ip->dev = le32_to_cpu(dip->di_rdev);
3083 ip->i_rdev = new_decode_dev(jfs_ip->dev);
3086 if (S_ISDIR(ip->i_mode)) {
3087 memcpy(&jfs_ip->i_dirtable, &dip->di_dirtable, 384);
3088 } else if (S_ISREG(ip->i_mode) || S_ISLNK(ip->i_mode)) {
3089 memcpy(&jfs_ip->i_xtroot, &dip->di_xtroot, 288);
3091 memcpy(&jfs_ip->i_inline_ea, &dip->di_inlineea, 128);
3093 /* Zero the in-memory-only stuff */
3095 jfs_ip->btindex = 0;
3096 jfs_ip->btorder = 0;
3099 jfs_ip->atlhead = 0;
3100 jfs_ip->atltail = 0;
3106 * NAME: copy_to_dinode()
3108 * FUNCTION: Copies inode info from in-memory inode to disk inode
3110 static void copy_to_dinode(struct dinode * dip, struct inode *ip)
3112 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
3113 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
3115 dip->di_fileset = cpu_to_le32(jfs_ip->fileset);
3116 dip->di_inostamp = cpu_to_le32(sbi->inostamp);
3117 dip->di_number = cpu_to_le32(ip->i_ino);
3118 dip->di_gen = cpu_to_le32(ip->i_generation);
3119 dip->di_size = cpu_to_le64(ip->i_size);
3120 dip->di_nblocks = cpu_to_le64(PBLK2LBLK(ip->i_sb, ip->i_blocks));
3121 dip->di_nlink = cpu_to_le32(ip->i_nlink);
3122 if (!uid_valid(sbi->uid))
3123 dip->di_uid = cpu_to_le32(i_uid_read(ip));
3125 dip->di_uid =cpu_to_le32(from_kuid(&init_user_ns,
3126 jfs_ip->saved_uid));
3127 if (!gid_valid(sbi->gid))
3128 dip->di_gid = cpu_to_le32(i_gid_read(ip));
3130 dip->di_gid = cpu_to_le32(from_kgid(&init_user_ns,
3131 jfs_ip->saved_gid));
3133 * mode2 is only needed for storing the higher order bits.
3134 * Trust i_mode for the lower order ones
3136 if (sbi->umask == -1)
3137 dip->di_mode = cpu_to_le32((jfs_ip->mode2 & 0xffff0000) |
3139 else /* Leave the original permissions alone */
3140 dip->di_mode = cpu_to_le32(jfs_ip->mode2);
3142 dip->di_atime.tv_sec = cpu_to_le32(ip->i_atime.tv_sec);
3143 dip->di_atime.tv_nsec = cpu_to_le32(ip->i_atime.tv_nsec);
3144 dip->di_ctime.tv_sec = cpu_to_le32(ip->i_ctime.tv_sec);
3145 dip->di_ctime.tv_nsec = cpu_to_le32(ip->i_ctime.tv_nsec);
3146 dip->di_mtime.tv_sec = cpu_to_le32(ip->i_mtime.tv_sec);
3147 dip->di_mtime.tv_nsec = cpu_to_le32(ip->i_mtime.tv_nsec);
3148 dip->di_ixpxd = jfs_ip->ixpxd; /* in-memory pxd's are little-endian */
3149 dip->di_acl = jfs_ip->acl; /* as are dxd's */
3150 dip->di_ea = jfs_ip->ea;
3151 dip->di_next_index = cpu_to_le32(jfs_ip->next_index);
3152 dip->di_otime.tv_sec = cpu_to_le32(jfs_ip->otime);
3153 dip->di_otime.tv_nsec = 0;
3154 dip->di_acltype = cpu_to_le32(jfs_ip->acltype);
3155 if (S_ISCHR(ip->i_mode) || S_ISBLK(ip->i_mode))
3156 dip->di_rdev = cpu_to_le32(jfs_ip->dev);