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