blob: 28201b194f531ba10aa819abccb030bc7ea88cd0 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
49#include "jfs_incore.h"
Dave Kleikamp1868f4a2005-05-04 15:29:35 -050050#include "jfs_inode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include "jfs_filsys.h"
52#include "jfs_dinode.h"
53#include "jfs_dmap.h"
54#include "jfs_imap.h"
55#include "jfs_metapage.h"
56#include "jfs_superblock.h"
57#include "jfs_debug.h"
58
59/*
Dave Kleikampac17b8b2005-10-03 15:32:11 -050060 * __mark_inode_dirty expects inodes to be hashed. Since we don't want
61 * special inodes in the fileset inode space, we hash them to a dummy head
62 */
63static HLIST_HEAD(aggregate_hash);
64
65/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 * imap locks
67 */
68/* iag free list lock */
69#define IAGFREE_LOCK_INIT(imap) init_MUTEX(&imap->im_freelock)
70#define IAGFREE_LOCK(imap) down(&imap->im_freelock)
71#define IAGFREE_UNLOCK(imap) up(&imap->im_freelock)
72
73/* per ag iag list locks */
74#define AG_LOCK_INIT(imap,index) init_MUTEX(&(imap->im_aglock[index]))
75#define AG_LOCK(imap,agno) down(&imap->im_aglock[agno])
76#define AG_UNLOCK(imap,agno) up(&imap->im_aglock[agno])
77
78/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 * forward references
80 */
81static int diAllocAG(struct inomap *, int, boolean_t, struct inode *);
82static int diAllocAny(struct inomap *, int, boolean_t, struct inode *);
83static int diAllocBit(struct inomap *, struct iag *, int);
84static int diAllocExt(struct inomap *, int, struct inode *);
85static int diAllocIno(struct inomap *, int, struct inode *);
86static int diFindFree(u32, int);
87static int diNewExt(struct inomap *, struct iag *, int);
88static int diNewIAG(struct inomap *, int *, int, struct metapage **);
89static void duplicateIXtree(struct super_block *, s64, int, s64 *);
90
91static int diIAGRead(struct inomap * imap, int, struct metapage **);
92static int copy_from_dinode(struct dinode *, struct inode *);
93static void copy_to_dinode(struct dinode *, struct inode *);
94
95/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 * NAME: diMount()
97 *
98 * FUNCTION: initialize the incore inode map control structures for
99 * a fileset or aggregate init time.
100 *
101 * the inode map's control structure (dinomap) is
102 * brought in from disk and placed in virtual memory.
103 *
104 * PARAMETERS:
105 * ipimap - pointer to inode map inode for the aggregate or fileset.
106 *
107 * RETURN VALUES:
108 * 0 - success
109 * -ENOMEM - insufficient free virtual memory.
110 * -EIO - i/o error.
111 */
112int diMount(struct inode *ipimap)
113{
114 struct inomap *imap;
115 struct metapage *mp;
116 int index;
117 struct dinomap_disk *dinom_le;
118
119 /*
120 * allocate/initialize the in-memory inode map control structure
121 */
122 /* allocate the in-memory inode map control structure. */
123 imap = (struct inomap *) kmalloc(sizeof(struct inomap), GFP_KERNEL);
124 if (imap == NULL) {
125 jfs_err("diMount: kmalloc returned NULL!");
126 return -ENOMEM;
127 }
128
129 /* read the on-disk inode map control structure. */
130
131 mp = read_metapage(ipimap,
132 IMAPBLKNO << JFS_SBI(ipimap->i_sb)->l2nbperpage,
133 PSIZE, 0);
134 if (mp == NULL) {
135 kfree(imap);
136 return -EIO;
137 }
138
139 /* copy the on-disk version to the in-memory version. */
140 dinom_le = (struct dinomap_disk *) mp->data;
141 imap->im_freeiag = le32_to_cpu(dinom_le->in_freeiag);
142 imap->im_nextiag = le32_to_cpu(dinom_le->in_nextiag);
143 atomic_set(&imap->im_numinos, le32_to_cpu(dinom_le->in_numinos));
144 atomic_set(&imap->im_numfree, le32_to_cpu(dinom_le->in_numfree));
145 imap->im_nbperiext = le32_to_cpu(dinom_le->in_nbperiext);
146 imap->im_l2nbperiext = le32_to_cpu(dinom_le->in_l2nbperiext);
147 for (index = 0; index < MAXAG; index++) {
148 imap->im_agctl[index].inofree =
149 le32_to_cpu(dinom_le->in_agctl[index].inofree);
150 imap->im_agctl[index].extfree =
151 le32_to_cpu(dinom_le->in_agctl[index].extfree);
152 imap->im_agctl[index].numinos =
153 le32_to_cpu(dinom_le->in_agctl[index].numinos);
154 imap->im_agctl[index].numfree =
155 le32_to_cpu(dinom_le->in_agctl[index].numfree);
156 }
157
158 /* release the buffer. */
159 release_metapage(mp);
160
161 /*
162 * allocate/initialize inode allocation map locks
163 */
164 /* allocate and init iag free list lock */
165 IAGFREE_LOCK_INIT(imap);
166
167 /* allocate and init ag list locks */
168 for (index = 0; index < MAXAG; index++) {
169 AG_LOCK_INIT(imap, index);
170 }
171
172 /* bind the inode map inode and inode map control structure
173 * to each other.
174 */
175 imap->im_ipimap = ipimap;
176 JFS_IP(ipimap)->i_imap = imap;
177
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 return (0);
179}
180
181
182/*
183 * NAME: diUnmount()
184 *
185 * FUNCTION: write to disk the incore inode map control structures for
186 * a fileset or aggregate at unmount time.
187 *
188 * PARAMETERS:
189 * ipimap - pointer to inode map inode for the aggregate or fileset.
190 *
191 * RETURN VALUES:
192 * 0 - success
193 * -ENOMEM - insufficient free virtual memory.
194 * -EIO - i/o error.
195 */
196int diUnmount(struct inode *ipimap, int mounterror)
197{
198 struct inomap *imap = JFS_IP(ipimap)->i_imap;
199
200 /*
201 * update the on-disk inode map control structure
202 */
203
204 if (!(mounterror || isReadOnly(ipimap)))
205 diSync(ipimap);
206
207 /*
208 * Invalidate the page cache buffers
209 */
210 truncate_inode_pages(ipimap->i_mapping, 0);
211
212 /*
213 * free in-memory control structure
214 */
215 kfree(imap);
216
217 return (0);
218}
219
220
221/*
222 * diSync()
223 */
224int diSync(struct inode *ipimap)
225{
226 struct dinomap_disk *dinom_le;
227 struct inomap *imp = JFS_IP(ipimap)->i_imap;
228 struct metapage *mp;
229 int index;
230
231 /*
232 * write imap global conrol page
233 */
234 /* read the on-disk inode map control structure */
235 mp = get_metapage(ipimap,
236 IMAPBLKNO << JFS_SBI(ipimap->i_sb)->l2nbperpage,
237 PSIZE, 0);
238 if (mp == NULL) {
239 jfs_err("diSync: get_metapage failed!");
240 return -EIO;
241 }
242
243 /* copy the in-memory version to the on-disk version */
244 dinom_le = (struct dinomap_disk *) mp->data;
245 dinom_le->in_freeiag = cpu_to_le32(imp->im_freeiag);
246 dinom_le->in_nextiag = cpu_to_le32(imp->im_nextiag);
247 dinom_le->in_numinos = cpu_to_le32(atomic_read(&imp->im_numinos));
248 dinom_le->in_numfree = cpu_to_le32(atomic_read(&imp->im_numfree));
249 dinom_le->in_nbperiext = cpu_to_le32(imp->im_nbperiext);
250 dinom_le->in_l2nbperiext = cpu_to_le32(imp->im_l2nbperiext);
251 for (index = 0; index < MAXAG; index++) {
252 dinom_le->in_agctl[index].inofree =
253 cpu_to_le32(imp->im_agctl[index].inofree);
254 dinom_le->in_agctl[index].extfree =
255 cpu_to_le32(imp->im_agctl[index].extfree);
256 dinom_le->in_agctl[index].numinos =
257 cpu_to_le32(imp->im_agctl[index].numinos);
258 dinom_le->in_agctl[index].numfree =
259 cpu_to_le32(imp->im_agctl[index].numfree);
260 }
261
262 /* write out the control structure */
263 write_metapage(mp);
264
265 /*
266 * write out dirty pages of imap
267 */
268 filemap_fdatawrite(ipimap->i_mapping);
269 filemap_fdatawait(ipimap->i_mapping);
270
271 diWriteSpecial(ipimap, 0);
272
273 return (0);
274}
275
276
277/*
278 * NAME: diRead()
279 *
280 * FUNCTION: initialize an incore inode from disk.
281 *
282 * on entry, the specifed incore inode should itself
283 * specify the disk inode number corresponding to the
284 * incore inode (i.e. i_number should be initialized).
285 *
286 * this routine handles incore inode initialization for
287 * both "special" and "regular" inodes. special inodes
288 * are those required early in the mount process and
289 * require special handling since much of the file system
290 * is not yet initialized. these "special" inodes are
291 * identified by a NULL inode map inode pointer and are
292 * actually initialized by a call to diReadSpecial().
293 *
294 * for regular inodes, the iag describing the disk inode
295 * is read from disk to determine the inode extent address
296 * for the disk inode. with the inode extent address in
297 * hand, the page of the extent that contains the disk
298 * inode is read and the disk inode is copied to the
299 * incore inode.
300 *
301 * PARAMETERS:
302 * ip - pointer to incore inode to be initialized from disk.
303 *
304 * RETURN VALUES:
305 * 0 - success
306 * -EIO - i/o error.
307 * -ENOMEM - insufficient memory
308 *
309 */
310int diRead(struct inode *ip)
311{
312 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
313 int iagno, ino, extno, rc;
314 struct inode *ipimap;
315 struct dinode *dp;
316 struct iag *iagp;
317 struct metapage *mp;
318 s64 blkno, agstart;
319 struct inomap *imap;
320 int block_offset;
321 int inodes_left;
322 uint pageno;
323 int rel_inode;
324
325 jfs_info("diRead: ino = %ld", ip->i_ino);
326
327 ipimap = sbi->ipimap;
328 JFS_IP(ip)->ipimap = ipimap;
329
330 /* determine the iag number for this inode (number) */
331 iagno = INOTOIAG(ip->i_ino);
332
333 /* read the iag */
334 imap = JFS_IP(ipimap)->i_imap;
335 IREAD_LOCK(ipimap);
336 rc = diIAGRead(imap, iagno, &mp);
337 IREAD_UNLOCK(ipimap);
338 if (rc) {
339 jfs_err("diRead: diIAGRead returned %d", rc);
340 return (rc);
341 }
342
343 iagp = (struct iag *) mp->data;
344
345 /* determine inode extent that holds the disk inode */
346 ino = ip->i_ino & (INOSPERIAG - 1);
347 extno = ino >> L2INOSPEREXT;
348
349 if ((lengthPXD(&iagp->inoext[extno]) != imap->im_nbperiext) ||
350 (addressPXD(&iagp->inoext[extno]) == 0)) {
351 release_metapage(mp);
352 return -ESTALE;
353 }
354
355 /* get disk block number of the page within the inode extent
356 * that holds the disk inode.
357 */
358 blkno = INOPBLK(&iagp->inoext[extno], ino, sbi->l2nbperpage);
359
360 /* get the ag for the iag */
361 agstart = le64_to_cpu(iagp->agstart);
362
363 release_metapage(mp);
364
365 rel_inode = (ino & (INOSPERPAGE - 1));
366 pageno = blkno >> sbi->l2nbperpage;
367
368 if ((block_offset = ((u32) blkno & (sbi->nbperpage - 1)))) {
369 /*
370 * OS/2 didn't always align inode extents on page boundaries
371 */
372 inodes_left =
373 (sbi->nbperpage - block_offset) << sbi->l2niperblk;
374
375 if (rel_inode < inodes_left)
376 rel_inode += block_offset << sbi->l2niperblk;
377 else {
378 pageno += 1;
379 rel_inode -= inodes_left;
380 }
381 }
382
383 /* read the page of disk inode */
384 mp = read_metapage(ipimap, pageno << sbi->l2nbperpage, PSIZE, 1);
385 if (mp == 0) {
386 jfs_err("diRead: read_metapage failed");
387 return -EIO;
388 }
389
390 /* locate the the disk inode requested */
391 dp = (struct dinode *) mp->data;
392 dp += rel_inode;
393
394 if (ip->i_ino != le32_to_cpu(dp->di_number)) {
395 jfs_error(ip->i_sb, "diRead: i_ino != di_number");
396 rc = -EIO;
397 } else if (le32_to_cpu(dp->di_nlink) == 0)
398 rc = -ESTALE;
399 else
400 /* copy the disk inode to the in-memory inode */
401 rc = copy_from_dinode(dp, ip);
402
403 release_metapage(mp);
404
405 /* set the ag for the inode */
406 JFS_IP(ip)->agno = BLKTOAG(agstart, sbi);
407 JFS_IP(ip)->active_ag = -1;
408
409 return (rc);
410}
411
412
413/*
414 * NAME: diReadSpecial()
415 *
416 * FUNCTION: initialize a 'special' inode from disk.
417 *
418 * this routines handles aggregate level inodes. The
419 * inode cache cannot differentiate between the
420 * aggregate inodes and the filesystem inodes, so we
421 * handle these here. We don't actually use the aggregate
422 * inode map, since these inodes are at a fixed location
423 * and in some cases the aggregate inode map isn't initialized
424 * yet.
425 *
426 * PARAMETERS:
427 * sb - filesystem superblock
428 * inum - aggregate inode number
429 * secondary - 1 if secondary aggregate inode table
430 *
431 * RETURN VALUES:
432 * new inode - success
433 * NULL - i/o error.
434 */
435struct inode *diReadSpecial(struct super_block *sb, ino_t inum, int secondary)
436{
437 struct jfs_sb_info *sbi = JFS_SBI(sb);
438 uint address;
439 struct dinode *dp;
440 struct inode *ip;
441 struct metapage *mp;
442
443 ip = new_inode(sb);
444 if (ip == NULL) {
445 jfs_err("diReadSpecial: new_inode returned NULL!");
446 return ip;
447 }
448
449 if (secondary) {
450 address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage;
451 JFS_IP(ip)->ipimap = sbi->ipaimap2;
452 } else {
453 address = AITBL_OFF >> L2PSIZE;
454 JFS_IP(ip)->ipimap = sbi->ipaimap;
455 }
456
457 ASSERT(inum < INOSPEREXT);
458
459 ip->i_ino = inum;
460
461 address += inum >> 3; /* 8 inodes per 4K page */
462
463 /* read the page of fixed disk inode (AIT) in raw mode */
464 mp = read_metapage(ip, address << sbi->l2nbperpage, PSIZE, 1);
465 if (mp == NULL) {
466 ip->i_nlink = 1; /* Don't want iput() deleting it */
467 iput(ip);
468 return (NULL);
469 }
470
471 /* get the pointer to the disk inode of interest */
472 dp = (struct dinode *) (mp->data);
473 dp += inum % 8; /* 8 inodes per 4K page */
474
475 /* copy on-disk inode to in-memory inode */
476 if ((copy_from_dinode(dp, ip)) != 0) {
477 /* handle bad return by returning NULL for ip */
478 ip->i_nlink = 1; /* Don't want iput() deleting it */
479 iput(ip);
480 /* release the page */
481 release_metapage(mp);
482 return (NULL);
483
484 }
485
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600486 ip->i_mapping->a_ops = &jfs_metapage_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 mapping_set_gfp_mask(ip->i_mapping, GFP_NOFS);
488
489 /* Allocations to metadata inodes should not affect quotas */
490 ip->i_flags |= S_NOQUOTA;
491
492 if ((inum == FILESYSTEM_I) && (JFS_IP(ip)->ipimap == sbi->ipaimap)) {
493 sbi->gengen = le32_to_cpu(dp->di_gengen);
494 sbi->inostamp = le32_to_cpu(dp->di_inostamp);
495 }
496
497 /* release the page */
498 release_metapage(mp);
499
Dave Kleikampac17b8b2005-10-03 15:32:11 -0500500 hlist_add_head(&ip->i_hash, &aggregate_hash);
501
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 if (secondary)
526 address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage;
527 else
528 address = AITBL_OFF >> L2PSIZE;
529
530 ASSERT(inum < INOSPEREXT);
531
532 address += inum >> 3; /* 8 inodes per 4K page */
533
534 /* read the page of fixed disk inode (AIT) in raw mode */
535 mp = read_metapage(ip, address << sbi->l2nbperpage, PSIZE, 1);
536 if (mp == NULL) {
537 jfs_err("diWriteSpecial: failed to read aggregate inode "
538 "extent!");
539 return;
540 }
541
542 /* get the pointer to the disk inode of interest */
543 dp = (struct dinode *) (mp->data);
544 dp += inum % 8; /* 8 inodes per 4K page */
545
546 /* copy on-disk inode to in-memory inode */
547 copy_to_dinode(dp, ip);
548 memcpy(&dp->di_xtroot, &JFS_IP(ip)->i_xtroot, 288);
549
550 if (inum == FILESYSTEM_I)
551 dp->di_gengen = cpu_to_le32(sbi->gengen);
552
553 /* write the page */
554 write_metapage(mp);
555}
556
557/*
558 * NAME: diFreeSpecial()
559 *
560 * FUNCTION: Free allocated space for special inode
561 */
562void diFreeSpecial(struct inode *ip)
563{
564 if (ip == NULL) {
565 jfs_err("diFreeSpecial called with NULL ip!");
566 return;
567 }
568 filemap_fdatawrite(ip->i_mapping);
569 filemap_fdatawait(ip->i_mapping);
570 truncate_inode_pages(ip->i_mapping, 0);
571 iput(ip);
572}
573
574
575
576/*
577 * NAME: diWrite()
578 *
579 * FUNCTION: write the on-disk inode portion of the in-memory inode
580 * to its corresponding on-disk inode.
581 *
582 * on entry, the specifed incore inode should itself
583 * specify the disk inode number corresponding to the
584 * incore inode (i.e. i_number should be initialized).
585 *
586 * the inode contains the inode extent address for the disk
587 * inode. with the inode extent address in hand, the
588 * page of the extent that contains the disk inode is
589 * read and the disk inode portion of the incore inode
590 * is copied to the disk inode.
591 *
592 * PARAMETERS:
593 * tid - transacation id
594 * ip - pointer to incore inode to be written to the inode extent.
595 *
596 * RETURN VALUES:
597 * 0 - success
598 * -EIO - i/o error.
599 */
600int diWrite(tid_t tid, struct inode *ip)
601{
602 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
603 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
604 int rc = 0;
605 s32 ino;
606 struct dinode *dp;
607 s64 blkno;
608 int block_offset;
609 int inodes_left;
610 struct metapage *mp;
611 uint pageno;
612 int rel_inode;
613 int dioffset;
614 struct inode *ipimap;
615 uint type;
616 lid_t lid;
617 struct tlock *ditlck, *tlck;
618 struct linelock *dilinelock, *ilinelock;
619 struct lv *lv;
620 int n;
621
622 ipimap = jfs_ip->ipimap;
623
624 ino = ip->i_ino & (INOSPERIAG - 1);
625
626 if (!addressPXD(&(jfs_ip->ixpxd)) ||
627 (lengthPXD(&(jfs_ip->ixpxd)) !=
628 JFS_IP(ipimap)->i_imap->im_nbperiext)) {
629 jfs_error(ip->i_sb, "diWrite: ixpxd invalid");
630 return -EIO;
631 }
632
633 /*
634 * read the page of disk inode containing the specified inode:
635 */
636 /* compute the block address of the page */
637 blkno = INOPBLK(&(jfs_ip->ixpxd), ino, sbi->l2nbperpage);
638
639 rel_inode = (ino & (INOSPERPAGE - 1));
640 pageno = blkno >> sbi->l2nbperpage;
641
642 if ((block_offset = ((u32) blkno & (sbi->nbperpage - 1)))) {
643 /*
644 * OS/2 didn't always align inode extents on page boundaries
645 */
646 inodes_left =
647 (sbi->nbperpage - block_offset) << sbi->l2niperblk;
648
649 if (rel_inode < inodes_left)
650 rel_inode += block_offset << sbi->l2niperblk;
651 else {
652 pageno += 1;
653 rel_inode -= inodes_left;
654 }
655 }
656 /* read the page of disk inode */
657 retry:
658 mp = read_metapage(ipimap, pageno << sbi->l2nbperpage, PSIZE, 1);
659 if (mp == 0)
660 return -EIO;
661
662 /* get the pointer to the disk inode */
663 dp = (struct dinode *) mp->data;
664 dp += rel_inode;
665
666 dioffset = (ino & (INOSPERPAGE - 1)) << L2DISIZE;
667
668 /*
669 * acquire transaction lock on the on-disk inode;
670 * N.B. tlock is acquired on ipimap not ip;
671 */
672 if ((ditlck =
673 txLock(tid, ipimap, mp, tlckINODE | tlckENTRY)) == NULL)
674 goto retry;
675 dilinelock = (struct linelock *) & ditlck->lock;
676
677 /*
678 * copy btree root from in-memory inode to on-disk inode
679 *
680 * (tlock is taken from inline B+-tree root in in-memory
681 * inode when the B+-tree root is updated, which is pointed
682 * by jfs_ip->blid as well as being on tx tlock list)
683 *
684 * further processing of btree root is based on the copy
685 * in in-memory inode, where txLog() will log from, and,
686 * for xtree root, txUpdateMap() will update map and reset
687 * XAD_NEW bit;
688 */
689
690 if (S_ISDIR(ip->i_mode) && (lid = jfs_ip->xtlid)) {
691 /*
692 * This is the special xtree inside the directory for storing
693 * the directory table
694 */
695 xtpage_t *p, *xp;
696 xad_t *xad;
697
698 jfs_ip->xtlid = 0;
699 tlck = lid_to_tlock(lid);
700 assert(tlck->type & tlckXTREE);
701 tlck->type |= tlckBTROOT;
702 tlck->mp = mp;
703 ilinelock = (struct linelock *) & tlck->lock;
704
705 /*
706 * copy xtree root from inode to dinode:
707 */
708 p = &jfs_ip->i_xtroot;
709 xp = (xtpage_t *) &dp->di_dirtable;
710 lv = ilinelock->lv;
711 for (n = 0; n < ilinelock->index; n++, lv++) {
712 memcpy(&xp->xad[lv->offset], &p->xad[lv->offset],
713 lv->length << L2XTSLOTSIZE);
714 }
715
716 /* reset on-disk (metadata page) xtree XAD_NEW bit */
717 xad = &xp->xad[XTENTRYSTART];
718 for (n = XTENTRYSTART;
719 n < le16_to_cpu(xp->header.nextindex); n++, xad++)
720 if (xad->flag & (XAD_NEW | XAD_EXTENDED))
721 xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
722 }
723
724 if ((lid = jfs_ip->blid) == 0)
725 goto inlineData;
726 jfs_ip->blid = 0;
727
728 tlck = lid_to_tlock(lid);
729 type = tlck->type;
730 tlck->type |= tlckBTROOT;
731 tlck->mp = mp;
732 ilinelock = (struct linelock *) & tlck->lock;
733
734 /*
735 * regular file: 16 byte (XAD slot) granularity
736 */
737 if (type & tlckXTREE) {
738 xtpage_t *p, *xp;
739 xad_t *xad;
740
741 /*
742 * copy xtree root from inode to dinode:
743 */
744 p = &jfs_ip->i_xtroot;
745 xp = &dp->di_xtroot;
746 lv = ilinelock->lv;
747 for (n = 0; n < ilinelock->index; n++, lv++) {
748 memcpy(&xp->xad[lv->offset], &p->xad[lv->offset],
749 lv->length << L2XTSLOTSIZE);
750 }
751
752 /* reset on-disk (metadata page) xtree XAD_NEW bit */
753 xad = &xp->xad[XTENTRYSTART];
754 for (n = XTENTRYSTART;
755 n < le16_to_cpu(xp->header.nextindex); n++, xad++)
756 if (xad->flag & (XAD_NEW | XAD_EXTENDED))
757 xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
758 }
759 /*
760 * directory: 32 byte (directory entry slot) granularity
761 */
762 else if (type & tlckDTREE) {
763 dtpage_t *p, *xp;
764
765 /*
766 * copy dtree root from inode to dinode:
767 */
768 p = (dtpage_t *) &jfs_ip->i_dtroot;
769 xp = (dtpage_t *) & dp->di_dtroot;
770 lv = ilinelock->lv;
771 for (n = 0; n < ilinelock->index; n++, lv++) {
772 memcpy(&xp->slot[lv->offset], &p->slot[lv->offset],
773 lv->length << L2DTSLOTSIZE);
774 }
775 } else {
776 jfs_err("diWrite: UFO tlock");
777 }
778
779 inlineData:
780 /*
781 * copy inline symlink from in-memory inode to on-disk inode
782 */
783 if (S_ISLNK(ip->i_mode) && ip->i_size < IDATASIZE) {
784 lv = & dilinelock->lv[dilinelock->index];
785 lv->offset = (dioffset + 2 * 128) >> L2INODESLOTSIZE;
786 lv->length = 2;
787 memcpy(&dp->di_fastsymlink, jfs_ip->i_inline, IDATASIZE);
788 dilinelock->index++;
789 }
790 /*
791 * copy inline data from in-memory inode to on-disk inode:
792 * 128 byte slot granularity
793 */
794 if (test_cflag(COMMIT_Inlineea, ip)) {
795 lv = & dilinelock->lv[dilinelock->index];
796 lv->offset = (dioffset + 3 * 128) >> L2INODESLOTSIZE;
797 lv->length = 1;
798 memcpy(&dp->di_inlineea, jfs_ip->i_inline_ea, INODESLOTSIZE);
799 dilinelock->index++;
800
801 clear_cflag(COMMIT_Inlineea, ip);
802 }
803
804 /*
805 * lock/copy inode base: 128 byte slot granularity
806 */
807// baseDinode:
808 lv = & dilinelock->lv[dilinelock->index];
809 lv->offset = dioffset >> L2INODESLOTSIZE;
810 copy_to_dinode(dp, ip);
811 if (test_and_clear_cflag(COMMIT_Dirtable, ip)) {
812 lv->length = 2;
813 memcpy(&dp->di_dirtable, &jfs_ip->i_dirtable, 96);
814 } else
815 lv->length = 1;
816 dilinelock->index++;
817
818#ifdef _JFS_FASTDASD
819 /*
820 * We aren't logging changes to the DASD used in directory inodes,
821 * but we need to write them to disk. If we don't unmount cleanly,
822 * mount will recalculate the DASD used.
823 */
824 if (S_ISDIR(ip->i_mode)
825 && (ip->i_ipmnt->i_mntflag & JFS_DASD_ENABLED))
826 memcpy(&dp->di_DASD, &ip->i_DASD, sizeof(struct dasd));
827#endif /* _JFS_FASTDASD */
828
829 /* release the buffer holding the updated on-disk inode.
830 * the buffer will be later written by commit processing.
831 */
832 write_metapage(mp);
833
834 return (rc);
835}
836
837
838/*
839 * NAME: diFree(ip)
840 *
841 * FUNCTION: free a specified inode from the inode working map
842 * for a fileset or aggregate.
843 *
844 * if the inode to be freed represents the first (only)
845 * free inode within the iag, the iag will be placed on
846 * the ag free inode list.
847 *
848 * freeing the inode will cause the inode extent to be
849 * freed if the inode is the only allocated inode within
850 * the extent. in this case all the disk resource backing
851 * up the inode extent will be freed. in addition, the iag
852 * will be placed on the ag extent free list if the extent
853 * is the first free extent in the iag. if freeing the
854 * extent also means that no free inodes will exist for
855 * the iag, the iag will also be removed from the ag free
856 * inode list.
857 *
858 * the iag describing the inode will be freed if the extent
859 * is to be freed and it is the only backed extent within
860 * the iag. in this case, the iag will be removed from the
861 * ag free extent list and ag free inode list and placed on
862 * the inode map's free iag list.
863 *
864 * a careful update approach is used to provide consistency
865 * in the face of updates to multiple buffers. under this
866 * approach, all required buffers are obtained before making
867 * any updates and are held until all updates are complete.
868 *
869 * PARAMETERS:
870 * ip - inode to be freed.
871 *
872 * RETURN VALUES:
873 * 0 - success
874 * -EIO - i/o error.
875 */
876int diFree(struct inode *ip)
877{
878 int rc;
879 ino_t inum = ip->i_ino;
880 struct iag *iagp, *aiagp, *biagp, *ciagp, *diagp;
881 struct metapage *mp, *amp, *bmp, *cmp, *dmp;
882 int iagno, ino, extno, bitno, sword, agno;
883 int back, fwd;
884 u32 bitmap, mask;
885 struct inode *ipimap = JFS_SBI(ip->i_sb)->ipimap;
886 struct inomap *imap = JFS_IP(ipimap)->i_imap;
887 pxd_t freepxd;
888 tid_t tid;
889 struct inode *iplist[3];
890 struct tlock *tlck;
891 struct pxd_lock *pxdlock;
892
893 /*
894 * This is just to suppress compiler warnings. The same logic that
895 * references these variables is used to initialize them.
896 */
897 aiagp = biagp = ciagp = diagp = NULL;
898
899 /* get the iag number containing the inode.
900 */
901 iagno = INOTOIAG(inum);
902
903 /* make sure that the iag is contained within
904 * the map.
905 */
906 if (iagno >= imap->im_nextiag) {
907 dump_mem("imap", imap, 32);
908 jfs_error(ip->i_sb,
909 "diFree: inum = %d, iagno = %d, nextiag = %d",
910 (uint) inum, iagno, imap->im_nextiag);
911 return -EIO;
912 }
913
914 /* get the allocation group for this ino.
915 */
916 agno = JFS_IP(ip)->agno;
917
918 /* Lock the AG specific inode map information
919 */
920 AG_LOCK(imap, agno);
921
922 /* Obtain read lock in imap inode. Don't release it until we have
923 * read all of the IAG's that we are going to.
924 */
925 IREAD_LOCK(ipimap);
926
927 /* read the iag.
928 */
929 if ((rc = diIAGRead(imap, iagno, &mp))) {
930 IREAD_UNLOCK(ipimap);
931 AG_UNLOCK(imap, agno);
932 return (rc);
933 }
934 iagp = (struct iag *) mp->data;
935
936 /* get the inode number and extent number of the inode within
937 * the iag and the inode number within the extent.
938 */
939 ino = inum & (INOSPERIAG - 1);
940 extno = ino >> L2INOSPEREXT;
941 bitno = ino & (INOSPEREXT - 1);
942 mask = HIGHORDER >> bitno;
943
944 if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
945 jfs_error(ip->i_sb,
946 "diFree: wmap shows inode already free");
947 }
948
949 if (!addressPXD(&iagp->inoext[extno])) {
950 release_metapage(mp);
951 IREAD_UNLOCK(ipimap);
952 AG_UNLOCK(imap, agno);
953 jfs_error(ip->i_sb, "diFree: invalid inoext");
954 return -EIO;
955 }
956
957 /* compute the bitmap for the extent reflecting the freed inode.
958 */
959 bitmap = le32_to_cpu(iagp->wmap[extno]) & ~mask;
960
961 if (imap->im_agctl[agno].numfree > imap->im_agctl[agno].numinos) {
962 release_metapage(mp);
963 IREAD_UNLOCK(ipimap);
964 AG_UNLOCK(imap, agno);
965 jfs_error(ip->i_sb, "diFree: numfree > numinos");
966 return -EIO;
967 }
968 /*
969 * inode extent still has some inodes or below low water mark:
970 * keep the inode extent;
971 */
972 if (bitmap ||
973 imap->im_agctl[agno].numfree < 96 ||
974 (imap->im_agctl[agno].numfree < 288 &&
975 (((imap->im_agctl[agno].numfree * 100) /
976 imap->im_agctl[agno].numinos) <= 25))) {
977 /* if the iag currently has no free inodes (i.e.,
978 * the inode being freed is the first free inode of iag),
979 * insert the iag at head of the inode free list for the ag.
980 */
981 if (iagp->nfreeinos == 0) {
982 /* check if there are any iags on the ag inode
983 * free list. if so, read the first one so that
984 * we can link the current iag onto the list at
985 * the head.
986 */
987 if ((fwd = imap->im_agctl[agno].inofree) >= 0) {
988 /* read the iag that currently is the head
989 * of the list.
990 */
991 if ((rc = diIAGRead(imap, fwd, &amp))) {
992 IREAD_UNLOCK(ipimap);
993 AG_UNLOCK(imap, agno);
994 release_metapage(mp);
995 return (rc);
996 }
997 aiagp = (struct iag *) amp->data;
998
999 /* make current head point back to the iag.
1000 */
1001 aiagp->inofreeback = cpu_to_le32(iagno);
1002
1003 write_metapage(amp);
1004 }
1005
1006 /* iag points forward to current head and iag
1007 * becomes the new head of the list.
1008 */
1009 iagp->inofreefwd =
1010 cpu_to_le32(imap->im_agctl[agno].inofree);
1011 iagp->inofreeback = cpu_to_le32(-1);
1012 imap->im_agctl[agno].inofree = iagno;
1013 }
1014 IREAD_UNLOCK(ipimap);
1015
1016 /* update the free inode summary map for the extent if
1017 * freeing the inode means the extent will now have free
1018 * inodes (i.e., the inode being freed is the first free
1019 * inode of extent),
1020 */
1021 if (iagp->wmap[extno] == cpu_to_le32(ONES)) {
1022 sword = extno >> L2EXTSPERSUM;
1023 bitno = extno & (EXTSPERSUM - 1);
1024 iagp->inosmap[sword] &=
1025 cpu_to_le32(~(HIGHORDER >> bitno));
1026 }
1027
1028 /* update the bitmap.
1029 */
1030 iagp->wmap[extno] = cpu_to_le32(bitmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031
1032 /* update the free inode counts at the iag, ag and
1033 * map level.
1034 */
1035 iagp->nfreeinos =
1036 cpu_to_le32(le32_to_cpu(iagp->nfreeinos) + 1);
1037 imap->im_agctl[agno].numfree += 1;
1038 atomic_inc(&imap->im_numfree);
1039
1040 /* release the AG inode map lock
1041 */
1042 AG_UNLOCK(imap, agno);
1043
1044 /* write the iag */
1045 write_metapage(mp);
1046
1047 return (0);
1048 }
1049
1050
1051 /*
1052 * inode extent has become free and above low water mark:
1053 * free the inode extent;
1054 */
1055
1056 /*
1057 * prepare to update iag list(s) (careful update step 1)
1058 */
1059 amp = bmp = cmp = dmp = NULL;
1060 fwd = back = -1;
1061
1062 /* check if the iag currently has no free extents. if so,
1063 * it will be placed on the head of the ag extent free list.
1064 */
1065 if (iagp->nfreeexts == 0) {
1066 /* check if the ag extent free list has any iags.
1067 * if so, read the iag at the head of the list now.
1068 * this (head) iag will be updated later to reflect
1069 * the addition of the current iag at the head of
1070 * the list.
1071 */
1072 if ((fwd = imap->im_agctl[agno].extfree) >= 0) {
1073 if ((rc = diIAGRead(imap, fwd, &amp)))
1074 goto error_out;
1075 aiagp = (struct iag *) amp->data;
1076 }
1077 } else {
1078 /* iag has free extents. check if the addition of a free
1079 * extent will cause all extents to be free within this
1080 * iag. if so, the iag will be removed from the ag extent
1081 * free list and placed on the inode map's free iag list.
1082 */
1083 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) {
1084 /* in preparation for removing the iag from the
1085 * ag extent free list, read the iags preceeding
1086 * and following the iag on the ag extent free
1087 * list.
1088 */
1089 if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) {
1090 if ((rc = diIAGRead(imap, fwd, &amp)))
1091 goto error_out;
1092 aiagp = (struct iag *) amp->data;
1093 }
1094
1095 if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) {
1096 if ((rc = diIAGRead(imap, back, &bmp)))
1097 goto error_out;
1098 biagp = (struct iag *) bmp->data;
1099 }
1100 }
1101 }
1102
1103 /* remove the iag from the ag inode free list if freeing
1104 * this extent cause the iag to have no free inodes.
1105 */
1106 if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) {
1107 int inofreeback = le32_to_cpu(iagp->inofreeback);
1108 int inofreefwd = le32_to_cpu(iagp->inofreefwd);
1109
1110 /* in preparation for removing the iag from the
1111 * ag inode free list, read the iags preceeding
1112 * and following the iag on the ag inode free
1113 * list. before reading these iags, we must make
1114 * sure that we already don't have them in hand
1115 * from up above, since re-reading an iag (buffer)
1116 * we are currently holding would cause a deadlock.
1117 */
1118 if (inofreefwd >= 0) {
1119
1120 if (inofreefwd == fwd)
1121 ciagp = (struct iag *) amp->data;
1122 else if (inofreefwd == back)
1123 ciagp = (struct iag *) bmp->data;
1124 else {
1125 if ((rc =
1126 diIAGRead(imap, inofreefwd, &cmp)))
1127 goto error_out;
1128 ciagp = (struct iag *) cmp->data;
1129 }
1130 assert(ciagp != NULL);
1131 }
1132
1133 if (inofreeback >= 0) {
1134 if (inofreeback == fwd)
1135 diagp = (struct iag *) amp->data;
1136 else if (inofreeback == back)
1137 diagp = (struct iag *) bmp->data;
1138 else {
1139 if ((rc =
1140 diIAGRead(imap, inofreeback, &dmp)))
1141 goto error_out;
1142 diagp = (struct iag *) dmp->data;
1143 }
1144 assert(diagp != NULL);
1145 }
1146 }
1147
1148 IREAD_UNLOCK(ipimap);
1149
1150 /*
1151 * invalidate any page of the inode extent freed from buffer cache;
1152 */
1153 freepxd = iagp->inoext[extno];
1154 invalidate_pxd_metapages(ip, freepxd);
1155
1156 /*
1157 * update iag list(s) (careful update step 2)
1158 */
1159 /* add the iag to the ag extent free list if this is the
1160 * first free extent for the iag.
1161 */
1162 if (iagp->nfreeexts == 0) {
1163 if (fwd >= 0)
1164 aiagp->extfreeback = cpu_to_le32(iagno);
1165
1166 iagp->extfreefwd =
1167 cpu_to_le32(imap->im_agctl[agno].extfree);
1168 iagp->extfreeback = cpu_to_le32(-1);
1169 imap->im_agctl[agno].extfree = iagno;
1170 } else {
1171 /* remove the iag from the ag extent list if all extents
1172 * are now free and place it on the inode map iag free list.
1173 */
1174 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) {
1175 if (fwd >= 0)
1176 aiagp->extfreeback = iagp->extfreeback;
1177
1178 if (back >= 0)
1179 biagp->extfreefwd = iagp->extfreefwd;
1180 else
1181 imap->im_agctl[agno].extfree =
1182 le32_to_cpu(iagp->extfreefwd);
1183
1184 iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1);
1185
1186 IAGFREE_LOCK(imap);
1187 iagp->iagfree = cpu_to_le32(imap->im_freeiag);
1188 imap->im_freeiag = iagno;
1189 IAGFREE_UNLOCK(imap);
1190 }
1191 }
1192
1193 /* remove the iag from the ag inode free list if freeing
1194 * this extent causes the iag to have no free inodes.
1195 */
1196 if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) {
1197 if ((int) le32_to_cpu(iagp->inofreefwd) >= 0)
1198 ciagp->inofreeback = iagp->inofreeback;
1199
1200 if ((int) le32_to_cpu(iagp->inofreeback) >= 0)
1201 diagp->inofreefwd = iagp->inofreefwd;
1202 else
1203 imap->im_agctl[agno].inofree =
1204 le32_to_cpu(iagp->inofreefwd);
1205
1206 iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1);
1207 }
1208
1209 /* update the inode extent address and working map
1210 * to reflect the free extent.
1211 * the permanent map should have been updated already
1212 * for the inode being freed.
1213 */
1214 if (iagp->pmap[extno] != 0) {
1215 jfs_error(ip->i_sb, "diFree: the pmap does not show inode free");
1216 }
1217 iagp->wmap[extno] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 PXDlength(&iagp->inoext[extno], 0);
1219 PXDaddress(&iagp->inoext[extno], 0);
1220
1221 /* update the free extent and free inode summary maps
1222 * to reflect the freed extent.
1223 * the inode summary map is marked to indicate no inodes
1224 * available for the freed extent.
1225 */
1226 sword = extno >> L2EXTSPERSUM;
1227 bitno = extno & (EXTSPERSUM - 1);
1228 mask = HIGHORDER >> bitno;
1229 iagp->inosmap[sword] |= cpu_to_le32(mask);
1230 iagp->extsmap[sword] &= cpu_to_le32(~mask);
1231
1232 /* update the number of free inodes and number of free extents
1233 * for the iag.
1234 */
1235 iagp->nfreeinos = cpu_to_le32(le32_to_cpu(iagp->nfreeinos) -
1236 (INOSPEREXT - 1));
1237 iagp->nfreeexts = cpu_to_le32(le32_to_cpu(iagp->nfreeexts) + 1);
1238
1239 /* update the number of free inodes and backed inodes
1240 * at the ag and inode map level.
1241 */
1242 imap->im_agctl[agno].numfree -= (INOSPEREXT - 1);
1243 imap->im_agctl[agno].numinos -= INOSPEREXT;
1244 atomic_sub(INOSPEREXT - 1, &imap->im_numfree);
1245 atomic_sub(INOSPEREXT, &imap->im_numinos);
1246
1247 if (amp)
1248 write_metapage(amp);
1249 if (bmp)
1250 write_metapage(bmp);
1251 if (cmp)
1252 write_metapage(cmp);
1253 if (dmp)
1254 write_metapage(dmp);
1255
1256 /*
1257 * start transaction to update block allocation map
1258 * for the inode extent freed;
1259 *
1260 * N.B. AG_LOCK is released and iag will be released below, and
1261 * other thread may allocate inode from/reusing the ixad freed
1262 * BUT with new/different backing inode extent from the extent
1263 * to be freed by the transaction;
1264 */
1265 tid = txBegin(ipimap->i_sb, COMMIT_FORCE);
1266 down(&JFS_IP(ipimap)->commit_sem);
1267
1268 /* acquire tlock of the iag page of the freed ixad
1269 * to force the page NOHOMEOK (even though no data is
1270 * logged from the iag page) until NOREDOPAGE|FREEXTENT log
1271 * for the free of the extent is committed;
1272 * write FREEXTENT|NOREDOPAGE log record
1273 * N.B. linelock is overlaid as freed extent descriptor;
1274 */
1275 tlck = txLock(tid, ipimap, mp, tlckINODE | tlckFREE);
1276 pxdlock = (struct pxd_lock *) & tlck->lock;
1277 pxdlock->flag = mlckFREEPXD;
1278 pxdlock->pxd = freepxd;
1279 pxdlock->index = 1;
1280
1281 write_metapage(mp);
1282
1283 iplist[0] = ipimap;
1284
1285 /*
1286 * logredo needs the IAG number and IAG extent index in order
1287 * to ensure that the IMap is consistent. The least disruptive
1288 * way to pass these values through to the transaction manager
1289 * is in the iplist array.
1290 *
1291 * It's not pretty, but it works.
1292 */
1293 iplist[1] = (struct inode *) (size_t)iagno;
1294 iplist[2] = (struct inode *) (size_t)extno;
1295
1296 rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE);
1297
1298 txEnd(tid);
1299 up(&JFS_IP(ipimap)->commit_sem);
1300
1301 /* unlock the AG inode map information */
1302 AG_UNLOCK(imap, agno);
1303
1304 return (0);
1305
1306 error_out:
1307 IREAD_UNLOCK(ipimap);
1308
1309 if (amp)
1310 release_metapage(amp);
1311 if (bmp)
1312 release_metapage(bmp);
1313 if (cmp)
1314 release_metapage(cmp);
1315 if (dmp)
1316 release_metapage(dmp);
1317
1318 AG_UNLOCK(imap, agno);
1319
1320 release_metapage(mp);
1321
1322 return (rc);
1323}
1324
1325/*
1326 * There are several places in the diAlloc* routines where we initialize
1327 * the inode.
1328 */
1329static inline void
1330diInitInode(struct inode *ip, int iagno, int ino, int extno, struct iag * iagp)
1331{
1332 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
1333 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
1334
1335 ip->i_ino = (iagno << L2INOSPERIAG) + ino;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 jfs_ip->ixpxd = iagp->inoext[extno];
1337 jfs_ip->agno = BLKTOAG(le64_to_cpu(iagp->agstart), sbi);
1338 jfs_ip->active_ag = -1;
1339}
1340
1341
1342/*
1343 * NAME: diAlloc(pip,dir,ip)
1344 *
1345 * FUNCTION: allocate a disk inode from the inode working map
1346 * for a fileset or aggregate.
1347 *
1348 * PARAMETERS:
1349 * pip - pointer to incore inode for the parent inode.
1350 * dir - TRUE if the new disk inode is for a directory.
1351 * ip - pointer to a new inode
1352 *
1353 * RETURN VALUES:
1354 * 0 - success.
1355 * -ENOSPC - insufficient disk resources.
1356 * -EIO - i/o error.
1357 */
1358int diAlloc(struct inode *pip, boolean_t dir, struct inode *ip)
1359{
1360 int rc, ino, iagno, addext, extno, bitno, sword;
1361 int nwords, rem, i, agno;
1362 u32 mask, inosmap, extsmap;
1363 struct inode *ipimap;
1364 struct metapage *mp;
1365 ino_t inum;
1366 struct iag *iagp;
1367 struct inomap *imap;
1368
1369 /* get the pointers to the inode map inode and the
1370 * corresponding imap control structure.
1371 */
1372 ipimap = JFS_SBI(pip->i_sb)->ipimap;
1373 imap = JFS_IP(ipimap)->i_imap;
1374 JFS_IP(ip)->ipimap = ipimap;
1375 JFS_IP(ip)->fileset = FILESYSTEM_I;
1376
1377 /* for a directory, the allocation policy is to start
1378 * at the ag level using the preferred ag.
1379 */
1380 if (dir == TRUE) {
1381 agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap);
1382 AG_LOCK(imap, agno);
1383 goto tryag;
1384 }
1385
1386 /* for files, the policy starts off by trying to allocate from
1387 * the same iag containing the parent disk inode:
1388 * try to allocate the new disk inode close to the parent disk
1389 * inode, using parent disk inode number + 1 as the allocation
1390 * hint. (we use a left-to-right policy to attempt to avoid
1391 * moving backward on the disk.) compute the hint within the
1392 * file system and the iag.
1393 */
1394
1395 /* get the ag number of this iag */
1396 agno = JFS_IP(pip)->agno;
1397
1398 if (atomic_read(&JFS_SBI(pip->i_sb)->bmap->db_active[agno])) {
1399 /*
1400 * There is an open file actively growing. We want to
1401 * allocate new inodes from a different ag to avoid
1402 * fragmentation problems.
1403 */
1404 agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap);
1405 AG_LOCK(imap, agno);
1406 goto tryag;
1407 }
1408
1409 inum = pip->i_ino + 1;
1410 ino = inum & (INOSPERIAG - 1);
1411
1412 /* back off the the hint if it is outside of the iag */
1413 if (ino == 0)
1414 inum = pip->i_ino;
1415
1416 /* lock the AG inode map information */
1417 AG_LOCK(imap, agno);
1418
1419 /* Get read lock on imap inode */
1420 IREAD_LOCK(ipimap);
1421
1422 /* get the iag number and read the iag */
1423 iagno = INOTOIAG(inum);
1424 if ((rc = diIAGRead(imap, iagno, &mp))) {
1425 IREAD_UNLOCK(ipimap);
1426 AG_UNLOCK(imap, agno);
1427 return (rc);
1428 }
1429 iagp = (struct iag *) mp->data;
1430
1431 /* determine if new inode extent is allowed to be added to the iag.
1432 * new inode extent can be added to the iag if the ag
1433 * has less than 32 free disk inodes and the iag has free extents.
1434 */
1435 addext = (imap->im_agctl[agno].numfree < 32 && iagp->nfreeexts);
1436
1437 /*
1438 * try to allocate from the IAG
1439 */
1440 /* check if the inode may be allocated from the iag
1441 * (i.e. the inode has free inodes or new extent can be added).
1442 */
1443 if (iagp->nfreeinos || addext) {
1444 /* determine the extent number of the hint.
1445 */
1446 extno = ino >> L2INOSPEREXT;
1447
1448 /* check if the extent containing the hint has backed
1449 * inodes. if so, try to allocate within this extent.
1450 */
1451 if (addressPXD(&iagp->inoext[extno])) {
1452 bitno = ino & (INOSPEREXT - 1);
1453 if ((bitno =
1454 diFindFree(le32_to_cpu(iagp->wmap[extno]),
1455 bitno))
1456 < INOSPEREXT) {
1457 ino = (extno << L2INOSPEREXT) + bitno;
1458
1459 /* a free inode (bit) was found within this
1460 * extent, so allocate it.
1461 */
1462 rc = diAllocBit(imap, iagp, ino);
1463 IREAD_UNLOCK(ipimap);
1464 if (rc) {
1465 assert(rc == -EIO);
1466 } else {
1467 /* set the results of the allocation
1468 * and write the iag.
1469 */
1470 diInitInode(ip, iagno, ino, extno,
1471 iagp);
1472 mark_metapage_dirty(mp);
1473 }
1474 release_metapage(mp);
1475
1476 /* free the AG lock and return.
1477 */
1478 AG_UNLOCK(imap, agno);
1479 return (rc);
1480 }
1481
1482 if (!addext)
1483 extno =
1484 (extno ==
1485 EXTSPERIAG - 1) ? 0 : extno + 1;
1486 }
1487
1488 /*
1489 * no free inodes within the extent containing the hint.
1490 *
1491 * try to allocate from the backed extents following
1492 * hint or, if appropriate (i.e. addext is true), allocate
1493 * an extent of free inodes at or following the extent
1494 * containing the hint.
1495 *
1496 * the free inode and free extent summary maps are used
1497 * here, so determine the starting summary map position
1498 * and the number of words we'll have to examine. again,
1499 * the approach is to allocate following the hint, so we
1500 * might have to initially ignore prior bits of the summary
1501 * map that represent extents prior to the extent containing
1502 * the hint and later revisit these bits.
1503 */
1504 bitno = extno & (EXTSPERSUM - 1);
1505 nwords = (bitno == 0) ? SMAPSZ : SMAPSZ + 1;
1506 sword = extno >> L2EXTSPERSUM;
1507
1508 /* mask any prior bits for the starting words of the
1509 * summary map.
1510 */
1511 mask = ONES << (EXTSPERSUM - bitno);
1512 inosmap = le32_to_cpu(iagp->inosmap[sword]) | mask;
1513 extsmap = le32_to_cpu(iagp->extsmap[sword]) | mask;
1514
1515 /* scan the free inode and free extent summary maps for
1516 * free resources.
1517 */
1518 for (i = 0; i < nwords; i++) {
1519 /* check if this word of the free inode summary
1520 * map describes an extent with free inodes.
1521 */
1522 if (~inosmap) {
1523 /* an extent with free inodes has been
1524 * found. determine the extent number
1525 * and the inode number within the extent.
1526 */
1527 rem = diFindFree(inosmap, 0);
1528 extno = (sword << L2EXTSPERSUM) + rem;
1529 rem = diFindFree(le32_to_cpu(iagp->wmap[extno]),
1530 0);
1531 if (rem >= INOSPEREXT) {
1532 IREAD_UNLOCK(ipimap);
1533 release_metapage(mp);
1534 AG_UNLOCK(imap, agno);
1535 jfs_error(ip->i_sb,
1536 "diAlloc: can't find free bit "
1537 "in wmap");
1538 return EIO;
1539 }
1540
1541 /* determine the inode number within the
1542 * iag and allocate the inode from the
1543 * map.
1544 */
1545 ino = (extno << L2INOSPEREXT) + rem;
1546 rc = diAllocBit(imap, iagp, ino);
1547 IREAD_UNLOCK(ipimap);
1548 if (rc)
1549 assert(rc == -EIO);
1550 else {
1551 /* set the results of the allocation
1552 * and write the iag.
1553 */
1554 diInitInode(ip, iagno, ino, extno,
1555 iagp);
1556 mark_metapage_dirty(mp);
1557 }
1558 release_metapage(mp);
1559
1560 /* free the AG lock and return.
1561 */
1562 AG_UNLOCK(imap, agno);
1563 return (rc);
1564
1565 }
1566
1567 /* check if we may allocate an extent of free
1568 * inodes and whether this word of the free
1569 * extents summary map describes a free extent.
1570 */
1571 if (addext && ~extsmap) {
1572 /* a free extent has been found. determine
1573 * the extent number.
1574 */
1575 rem = diFindFree(extsmap, 0);
1576 extno = (sword << L2EXTSPERSUM) + rem;
1577
1578 /* allocate an extent of free inodes.
1579 */
1580 if ((rc = diNewExt(imap, iagp, extno))) {
1581 /* if there is no disk space for a
1582 * new extent, try to allocate the
1583 * disk inode from somewhere else.
1584 */
1585 if (rc == -ENOSPC)
1586 break;
1587
1588 assert(rc == -EIO);
1589 } else {
1590 /* set the results of the allocation
1591 * and write the iag.
1592 */
1593 diInitInode(ip, iagno,
1594 extno << L2INOSPEREXT,
1595 extno, iagp);
1596 mark_metapage_dirty(mp);
1597 }
1598 release_metapage(mp);
1599 /* free the imap inode & the AG lock & return.
1600 */
1601 IREAD_UNLOCK(ipimap);
1602 AG_UNLOCK(imap, agno);
1603 return (rc);
1604 }
1605
1606 /* move on to the next set of summary map words.
1607 */
1608 sword = (sword == SMAPSZ - 1) ? 0 : sword + 1;
1609 inosmap = le32_to_cpu(iagp->inosmap[sword]);
1610 extsmap = le32_to_cpu(iagp->extsmap[sword]);
1611 }
1612 }
1613 /* unlock imap inode */
1614 IREAD_UNLOCK(ipimap);
1615
1616 /* nothing doing in this iag, so release it. */
1617 release_metapage(mp);
1618
1619 tryag:
1620 /*
1621 * try to allocate anywhere within the same AG as the parent inode.
1622 */
1623 rc = diAllocAG(imap, agno, dir, ip);
1624
1625 AG_UNLOCK(imap, agno);
1626
1627 if (rc != -ENOSPC)
1628 return (rc);
1629
1630 /*
1631 * try to allocate in any AG.
1632 */
1633 return (diAllocAny(imap, agno, dir, ip));
1634}
1635
1636
1637/*
1638 * NAME: diAllocAG(imap,agno,dir,ip)
1639 *
1640 * FUNCTION: allocate a disk inode from the allocation group.
1641 *
1642 * this routine first determines if a new extent of free
1643 * inodes should be added for the allocation group, with
1644 * the current request satisfied from this extent. if this
1645 * is the case, an attempt will be made to do just that. if
1646 * this attempt fails or it has been determined that a new
1647 * extent should not be added, an attempt is made to satisfy
1648 * the request by allocating an existing (backed) free inode
1649 * from the allocation group.
1650 *
1651 * PRE CONDITION: Already have the AG lock for this AG.
1652 *
1653 * PARAMETERS:
1654 * imap - pointer to inode map control structure.
1655 * agno - allocation group to allocate from.
1656 * dir - TRUE if the new disk inode is for a directory.
1657 * ip - pointer to the new inode to be filled in on successful return
1658 * with the disk inode number allocated, its extent address
1659 * and the start of the ag.
1660 *
1661 * RETURN VALUES:
1662 * 0 - success.
1663 * -ENOSPC - insufficient disk resources.
1664 * -EIO - i/o error.
1665 */
1666static int
1667diAllocAG(struct inomap * imap, int agno, boolean_t dir, struct inode *ip)
1668{
1669 int rc, addext, numfree, numinos;
1670
1671 /* get the number of free and the number of backed disk
1672 * inodes currently within the ag.
1673 */
1674 numfree = imap->im_agctl[agno].numfree;
1675 numinos = imap->im_agctl[agno].numinos;
1676
1677 if (numfree > numinos) {
1678 jfs_error(ip->i_sb, "diAllocAG: numfree > numinos");
1679 return -EIO;
1680 }
1681
1682 /* determine if we should allocate a new extent of free inodes
1683 * within the ag: for directory inodes, add a new extent
1684 * if there are a small number of free inodes or number of free
1685 * inodes is a small percentage of the number of backed inodes.
1686 */
1687 if (dir == TRUE)
1688 addext = (numfree < 64 ||
1689 (numfree < 256
1690 && ((numfree * 100) / numinos) <= 20));
1691 else
1692 addext = (numfree == 0);
1693
1694 /*
1695 * try to allocate a new extent of free inodes.
1696 */
1697 if (addext) {
1698 /* if free space is not avaliable for this new extent, try
1699 * below to allocate a free and existing (already backed)
1700 * inode from the ag.
1701 */
1702 if ((rc = diAllocExt(imap, agno, ip)) != -ENOSPC)
1703 return (rc);
1704 }
1705
1706 /*
1707 * try to allocate an existing free inode from the ag.
1708 */
1709 return (diAllocIno(imap, agno, ip));
1710}
1711
1712
1713/*
1714 * NAME: diAllocAny(imap,agno,dir,iap)
1715 *
1716 * FUNCTION: allocate a disk inode from any other allocation group.
1717 *
1718 * this routine is called when an allocation attempt within
1719 * the primary allocation group has failed. if attempts to
1720 * allocate an inode from any allocation group other than the
1721 * specified primary group.
1722 *
1723 * PARAMETERS:
1724 * imap - pointer to inode map control structure.
1725 * agno - primary allocation group (to avoid).
1726 * dir - TRUE if the new disk inode is for a directory.
1727 * ip - pointer to a new inode to be filled in on successful return
1728 * with the disk inode number allocated, its extent address
1729 * and the start of the ag.
1730 *
1731 * RETURN VALUES:
1732 * 0 - success.
1733 * -ENOSPC - insufficient disk resources.
1734 * -EIO - i/o error.
1735 */
1736static int
1737diAllocAny(struct inomap * imap, int agno, boolean_t dir, struct inode *ip)
1738{
1739 int ag, rc;
1740 int maxag = JFS_SBI(imap->im_ipimap->i_sb)->bmap->db_maxag;
1741
1742
1743 /* try to allocate from the ags following agno up to
1744 * the maximum ag number.
1745 */
1746 for (ag = agno + 1; ag <= maxag; ag++) {
1747 AG_LOCK(imap, ag);
1748
1749 rc = diAllocAG(imap, ag, dir, ip);
1750
1751 AG_UNLOCK(imap, ag);
1752
1753 if (rc != -ENOSPC)
1754 return (rc);
1755 }
1756
1757 /* try to allocate from the ags in front of agno.
1758 */
1759 for (ag = 0; ag < agno; ag++) {
1760 AG_LOCK(imap, ag);
1761
1762 rc = diAllocAG(imap, ag, dir, ip);
1763
1764 AG_UNLOCK(imap, ag);
1765
1766 if (rc != -ENOSPC)
1767 return (rc);
1768 }
1769
1770 /* no free disk inodes.
1771 */
1772 return -ENOSPC;
1773}
1774
1775
1776/*
1777 * NAME: diAllocIno(imap,agno,ip)
1778 *
1779 * FUNCTION: allocate a disk inode from the allocation group's free
1780 * inode list, returning an error if this free list is
1781 * empty (i.e. no iags on the list).
1782 *
1783 * allocation occurs from the first iag on the list using
1784 * the iag's free inode summary map to find the leftmost
1785 * free inode in the iag.
1786 *
1787 * PRE CONDITION: Already have AG lock for this AG.
1788 *
1789 * PARAMETERS:
1790 * imap - pointer to inode map control structure.
1791 * agno - allocation group.
1792 * ip - pointer to new inode to be filled in on successful return
1793 * with the disk inode number allocated, its extent address
1794 * and the start of the ag.
1795 *
1796 * RETURN VALUES:
1797 * 0 - success.
1798 * -ENOSPC - insufficient disk resources.
1799 * -EIO - i/o error.
1800 */
1801static int diAllocIno(struct inomap * imap, int agno, struct inode *ip)
1802{
1803 int iagno, ino, rc, rem, extno, sword;
1804 struct metapage *mp;
1805 struct iag *iagp;
1806
1807 /* check if there are iags on the ag's free inode list.
1808 */
1809 if ((iagno = imap->im_agctl[agno].inofree) < 0)
1810 return -ENOSPC;
1811
1812 /* obtain read lock on imap inode */
1813 IREAD_LOCK(imap->im_ipimap);
1814
1815 /* read the iag at the head of the list.
1816 */
1817 if ((rc = diIAGRead(imap, iagno, &mp))) {
1818 IREAD_UNLOCK(imap->im_ipimap);
1819 return (rc);
1820 }
1821 iagp = (struct iag *) mp->data;
1822
1823 /* better be free inodes in this iag if it is on the
1824 * list.
1825 */
1826 if (!iagp->nfreeinos) {
1827 IREAD_UNLOCK(imap->im_ipimap);
1828 release_metapage(mp);
1829 jfs_error(ip->i_sb,
1830 "diAllocIno: nfreeinos = 0, but iag on freelist");
1831 return -EIO;
1832 }
1833
1834 /* scan the free inode summary map to find an extent
1835 * with free inodes.
1836 */
1837 for (sword = 0;; sword++) {
1838 if (sword >= SMAPSZ) {
1839 IREAD_UNLOCK(imap->im_ipimap);
1840 release_metapage(mp);
1841 jfs_error(ip->i_sb,
1842 "diAllocIno: free inode not found in summary map");
1843 return -EIO;
1844 }
1845
1846 if (~iagp->inosmap[sword])
1847 break;
1848 }
1849
1850 /* found a extent with free inodes. determine
1851 * the extent number.
1852 */
1853 rem = diFindFree(le32_to_cpu(iagp->inosmap[sword]), 0);
1854 if (rem >= EXTSPERSUM) {
1855 IREAD_UNLOCK(imap->im_ipimap);
1856 release_metapage(mp);
1857 jfs_error(ip->i_sb, "diAllocIno: no free extent found");
1858 return -EIO;
1859 }
1860 extno = (sword << L2EXTSPERSUM) + rem;
1861
1862 /* find the first free inode in the extent.
1863 */
1864 rem = diFindFree(le32_to_cpu(iagp->wmap[extno]), 0);
1865 if (rem >= INOSPEREXT) {
1866 IREAD_UNLOCK(imap->im_ipimap);
1867 release_metapage(mp);
1868 jfs_error(ip->i_sb, "diAllocIno: free inode not found");
1869 return -EIO;
1870 }
1871
1872 /* compute the inode number within the iag.
1873 */
1874 ino = (extno << L2INOSPEREXT) + rem;
1875
1876 /* allocate the inode.
1877 */
1878 rc = diAllocBit(imap, iagp, ino);
1879 IREAD_UNLOCK(imap->im_ipimap);
1880 if (rc) {
1881 release_metapage(mp);
1882 return (rc);
1883 }
1884
1885 /* set the results of the allocation and write the iag.
1886 */
1887 diInitInode(ip, iagno, ino, extno, iagp);
1888 write_metapage(mp);
1889
1890 return (0);
1891}
1892
1893
1894/*
1895 * NAME: diAllocExt(imap,agno,ip)
1896 *
1897 * FUNCTION: add a new extent of free inodes to an iag, allocating
1898 * an inode from this extent to satisfy the current allocation
1899 * request.
1900 *
1901 * this routine first tries to find an existing iag with free
1902 * extents through the ag free extent list. if list is not
1903 * empty, the head of the list will be selected as the home
1904 * of the new extent of free inodes. otherwise (the list is
1905 * empty), a new iag will be allocated for the ag to contain
1906 * the extent.
1907 *
1908 * once an iag has been selected, the free extent summary map
1909 * is used to locate a free extent within the iag and diNewExt()
1910 * is called to initialize the extent, with initialization
1911 * including the allocation of the first inode of the extent
1912 * for the purpose of satisfying this request.
1913 *
1914 * PARAMETERS:
1915 * imap - pointer to inode map control structure.
1916 * agno - allocation group number.
1917 * ip - pointer to new inode to be filled in on successful return
1918 * with the disk inode number allocated, its extent address
1919 * and the start of the ag.
1920 *
1921 * RETURN VALUES:
1922 * 0 - success.
1923 * -ENOSPC - insufficient disk resources.
1924 * -EIO - i/o error.
1925 */
1926static int diAllocExt(struct inomap * imap, int agno, struct inode *ip)
1927{
1928 int rem, iagno, sword, extno, rc;
1929 struct metapage *mp;
1930 struct iag *iagp;
1931
1932 /* check if the ag has any iags with free extents. if not,
1933 * allocate a new iag for the ag.
1934 */
1935 if ((iagno = imap->im_agctl[agno].extfree) < 0) {
1936 /* If successful, diNewIAG will obtain the read lock on the
1937 * imap inode.
1938 */
1939 if ((rc = diNewIAG(imap, &iagno, agno, &mp))) {
1940 return (rc);
1941 }
1942 iagp = (struct iag *) mp->data;
1943
1944 /* set the ag number if this a brand new iag
1945 */
1946 iagp->agstart =
1947 cpu_to_le64(AGTOBLK(agno, imap->im_ipimap));
1948 } else {
1949 /* read the iag.
1950 */
1951 IREAD_LOCK(imap->im_ipimap);
1952 if ((rc = diIAGRead(imap, iagno, &mp))) {
1953 IREAD_UNLOCK(imap->im_ipimap);
1954 jfs_error(ip->i_sb, "diAllocExt: error reading iag");
1955 return rc;
1956 }
1957 iagp = (struct iag *) mp->data;
1958 }
1959
1960 /* using the free extent summary map, find a free extent.
1961 */
1962 for (sword = 0;; sword++) {
1963 if (sword >= SMAPSZ) {
1964 release_metapage(mp);
1965 IREAD_UNLOCK(imap->im_ipimap);
1966 jfs_error(ip->i_sb,
1967 "diAllocExt: free ext summary map not found");
1968 return -EIO;
1969 }
1970 if (~iagp->extsmap[sword])
1971 break;
1972 }
1973
1974 /* determine the extent number of the free extent.
1975 */
1976 rem = diFindFree(le32_to_cpu(iagp->extsmap[sword]), 0);
1977 if (rem >= EXTSPERSUM) {
1978 release_metapage(mp);
1979 IREAD_UNLOCK(imap->im_ipimap);
1980 jfs_error(ip->i_sb, "diAllocExt: free extent not found");
1981 return -EIO;
1982 }
1983 extno = (sword << L2EXTSPERSUM) + rem;
1984
1985 /* initialize the new extent.
1986 */
1987 rc = diNewExt(imap, iagp, extno);
1988 IREAD_UNLOCK(imap->im_ipimap);
1989 if (rc) {
1990 /* something bad happened. if a new iag was allocated,
1991 * place it back on the inode map's iag free list, and
1992 * clear the ag number information.
1993 */
1994 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
1995 IAGFREE_LOCK(imap);
1996 iagp->iagfree = cpu_to_le32(imap->im_freeiag);
1997 imap->im_freeiag = iagno;
1998 IAGFREE_UNLOCK(imap);
1999 }
2000 write_metapage(mp);
2001 return (rc);
2002 }
2003
2004 /* set the results of the allocation and write the iag.
2005 */
2006 diInitInode(ip, iagno, extno << L2INOSPEREXT, extno, iagp);
2007
2008 write_metapage(mp);
2009
2010 return (0);
2011}
2012
2013
2014/*
2015 * NAME: diAllocBit(imap,iagp,ino)
2016 *
2017 * FUNCTION: allocate a backed inode from an iag.
2018 *
2019 * this routine performs the mechanics of allocating a
2020 * specified inode from a backed extent.
2021 *
2022 * if the inode to be allocated represents the last free
2023 * inode within the iag, the iag will be removed from the
2024 * ag free inode list.
2025 *
2026 * a careful update approach is used to provide consistency
2027 * in the face of updates to multiple buffers. under this
2028 * approach, all required buffers are obtained before making
2029 * any updates and are held all are updates are complete.
2030 *
2031 * PRE CONDITION: Already have buffer lock on iagp. Already have AG lock on
2032 * this AG. Must have read lock on imap inode.
2033 *
2034 * PARAMETERS:
2035 * imap - pointer to inode map control structure.
2036 * iagp - pointer to iag.
2037 * ino - inode number to be allocated within the iag.
2038 *
2039 * RETURN VALUES:
2040 * 0 - success.
2041 * -ENOSPC - insufficient disk resources.
2042 * -EIO - i/o error.
2043 */
2044static int diAllocBit(struct inomap * imap, struct iag * iagp, int ino)
2045{
2046 int extno, bitno, agno, sword, rc;
2047 struct metapage *amp = NULL, *bmp = NULL;
2048 struct iag *aiagp = NULL, *biagp = NULL;
2049 u32 mask;
2050
2051 /* check if this is the last free inode within the iag.
2052 * if so, it will have to be removed from the ag free
2053 * inode list, so get the iags preceeding and following
2054 * it on the list.
2055 */
2056 if (iagp->nfreeinos == cpu_to_le32(1)) {
2057 if ((int) le32_to_cpu(iagp->inofreefwd) >= 0) {
2058 if ((rc =
2059 diIAGRead(imap, le32_to_cpu(iagp->inofreefwd),
2060 &amp)))
2061 return (rc);
2062 aiagp = (struct iag *) amp->data;
2063 }
2064
2065 if ((int) le32_to_cpu(iagp->inofreeback) >= 0) {
2066 if ((rc =
2067 diIAGRead(imap,
2068 le32_to_cpu(iagp->inofreeback),
2069 &bmp))) {
2070 if (amp)
2071 release_metapage(amp);
2072 return (rc);
2073 }
2074 biagp = (struct iag *) bmp->data;
2075 }
2076 }
2077
2078 /* get the ag number, extent number, inode number within
2079 * the extent.
2080 */
2081 agno = BLKTOAG(le64_to_cpu(iagp->agstart), JFS_SBI(imap->im_ipimap->i_sb));
2082 extno = ino >> L2INOSPEREXT;
2083 bitno = ino & (INOSPEREXT - 1);
2084
2085 /* compute the mask for setting the map.
2086 */
2087 mask = HIGHORDER >> bitno;
2088
2089 /* the inode should be free and backed.
2090 */
2091 if (((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) ||
2092 ((le32_to_cpu(iagp->wmap[extno]) & mask) != 0) ||
2093 (addressPXD(&iagp->inoext[extno]) == 0)) {
2094 if (amp)
2095 release_metapage(amp);
2096 if (bmp)
2097 release_metapage(bmp);
2098
2099 jfs_error(imap->im_ipimap->i_sb,
2100 "diAllocBit: iag inconsistent");
2101 return -EIO;
2102 }
2103
2104 /* mark the inode as allocated in the working map.
2105 */
2106 iagp->wmap[extno] |= cpu_to_le32(mask);
2107
2108 /* check if all inodes within the extent are now
2109 * allocated. if so, update the free inode summary
2110 * map to reflect this.
2111 */
2112 if (iagp->wmap[extno] == cpu_to_le32(ONES)) {
2113 sword = extno >> L2EXTSPERSUM;
2114 bitno = extno & (EXTSPERSUM - 1);
2115 iagp->inosmap[sword] |= cpu_to_le32(HIGHORDER >> bitno);
2116 }
2117
2118 /* if this was the last free inode in the iag, remove the
2119 * iag from the ag free inode list.
2120 */
2121 if (iagp->nfreeinos == cpu_to_le32(1)) {
2122 if (amp) {
2123 aiagp->inofreeback = iagp->inofreeback;
2124 write_metapage(amp);
2125 }
2126
2127 if (bmp) {
2128 biagp->inofreefwd = iagp->inofreefwd;
2129 write_metapage(bmp);
2130 } else {
2131 imap->im_agctl[agno].inofree =
2132 le32_to_cpu(iagp->inofreefwd);
2133 }
2134 iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1);
2135 }
2136
2137 /* update the free inode count at the iag, ag, inode
2138 * map levels.
2139 */
2140 iagp->nfreeinos = cpu_to_le32(le32_to_cpu(iagp->nfreeinos) - 1);
2141 imap->im_agctl[agno].numfree -= 1;
2142 atomic_dec(&imap->im_numfree);
2143
2144 return (0);
2145}
2146
2147
2148/*
2149 * NAME: diNewExt(imap,iagp,extno)
2150 *
2151 * FUNCTION: initialize a new extent of inodes for an iag, allocating
2152 * the first inode of the extent for use for the current
2153 * allocation request.
2154 *
2155 * disk resources are allocated for the new extent of inodes
2156 * and the inodes themselves are initialized to reflect their
2157 * existence within the extent (i.e. their inode numbers and
2158 * inode extent addresses are set) and their initial state
2159 * (mode and link count are set to zero).
2160 *
2161 * if the iag is new, it is not yet on an ag extent free list
2162 * but will now be placed on this list.
2163 *
2164 * if the allocation of the new extent causes the iag to
2165 * have no free extent, the iag will be removed from the
2166 * ag extent free list.
2167 *
2168 * if the iag has no free backed inodes, it will be placed
2169 * on the ag free inode list, since the addition of the new
2170 * extent will now cause it to have free inodes.
2171 *
2172 * a careful update approach is used to provide consistency
2173 * (i.e. list consistency) in the face of updates to multiple
2174 * buffers. under this approach, all required buffers are
2175 * obtained before making any updates and are held until all
2176 * updates are complete.
2177 *
2178 * PRE CONDITION: Already have buffer lock on iagp. Already have AG lock on
2179 * this AG. Must have read lock on imap inode.
2180 *
2181 * PARAMETERS:
2182 * imap - pointer to inode map control structure.
2183 * iagp - pointer to iag.
2184 * extno - extent number.
2185 *
2186 * RETURN VALUES:
2187 * 0 - success.
2188 * -ENOSPC - insufficient disk resources.
2189 * -EIO - i/o error.
2190 */
2191static int diNewExt(struct inomap * imap, struct iag * iagp, int extno)
2192{
2193 int agno, iagno, fwd, back, freei = 0, sword, rc;
2194 struct iag *aiagp = NULL, *biagp = NULL, *ciagp = NULL;
2195 struct metapage *amp, *bmp, *cmp, *dmp;
2196 struct inode *ipimap;
2197 s64 blkno, hint;
2198 int i, j;
2199 u32 mask;
2200 ino_t ino;
2201 struct dinode *dp;
2202 struct jfs_sb_info *sbi;
2203
2204 /* better have free extents.
2205 */
2206 if (!iagp->nfreeexts) {
2207 jfs_error(imap->im_ipimap->i_sb, "diNewExt: no free extents");
2208 return -EIO;
2209 }
2210
2211 /* get the inode map inode.
2212 */
2213 ipimap = imap->im_ipimap;
2214 sbi = JFS_SBI(ipimap->i_sb);
2215
2216 amp = bmp = cmp = NULL;
2217
2218 /* get the ag and iag numbers for this iag.
2219 */
2220 agno = BLKTOAG(le64_to_cpu(iagp->agstart), sbi);
2221 iagno = le32_to_cpu(iagp->iagnum);
2222
2223 /* check if this is the last free extent within the
2224 * iag. if so, the iag must be removed from the ag
2225 * free extent list, so get the iags preceeding and
2226 * following the iag on this list.
2227 */
2228 if (iagp->nfreeexts == cpu_to_le32(1)) {
2229 if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) {
2230 if ((rc = diIAGRead(imap, fwd, &amp)))
2231 return (rc);
2232 aiagp = (struct iag *) amp->data;
2233 }
2234
2235 if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) {
2236 if ((rc = diIAGRead(imap, back, &bmp)))
2237 goto error_out;
2238 biagp = (struct iag *) bmp->data;
2239 }
2240 } else {
2241 /* the iag has free extents. if all extents are free
2242 * (as is the case for a newly allocated iag), the iag
2243 * must be added to the ag free extent list, so get
2244 * the iag at the head of the list in preparation for
2245 * adding this iag to this list.
2246 */
2247 fwd = back = -1;
2248 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2249 if ((fwd = imap->im_agctl[agno].extfree) >= 0) {
2250 if ((rc = diIAGRead(imap, fwd, &amp)))
2251 goto error_out;
2252 aiagp = (struct iag *) amp->data;
2253 }
2254 }
2255 }
2256
2257 /* check if the iag has no free inodes. if so, the iag
2258 * will have to be added to the ag free inode list, so get
2259 * the iag at the head of the list in preparation for
2260 * adding this iag to this list. in doing this, we must
2261 * check if we already have the iag at the head of
2262 * the list in hand.
2263 */
2264 if (iagp->nfreeinos == 0) {
2265 freei = imap->im_agctl[agno].inofree;
2266
2267 if (freei >= 0) {
2268 if (freei == fwd) {
2269 ciagp = aiagp;
2270 } else if (freei == back) {
2271 ciagp = biagp;
2272 } else {
2273 if ((rc = diIAGRead(imap, freei, &cmp)))
2274 goto error_out;
2275 ciagp = (struct iag *) cmp->data;
2276 }
2277 if (ciagp == NULL) {
2278 jfs_error(imap->im_ipimap->i_sb,
2279 "diNewExt: ciagp == NULL");
2280 rc = -EIO;
2281 goto error_out;
2282 }
2283 }
2284 }
2285
2286 /* allocate disk space for the inode extent.
2287 */
2288 if ((extno == 0) || (addressPXD(&iagp->inoext[extno - 1]) == 0))
2289 hint = ((s64) agno << sbi->bmap->db_agl2size) - 1;
2290 else
2291 hint = addressPXD(&iagp->inoext[extno - 1]) +
2292 lengthPXD(&iagp->inoext[extno - 1]) - 1;
2293
2294 if ((rc = dbAlloc(ipimap, hint, (s64) imap->im_nbperiext, &blkno)))
2295 goto error_out;
2296
2297 /* compute the inode number of the first inode within the
2298 * extent.
2299 */
2300 ino = (iagno << L2INOSPERIAG) + (extno << L2INOSPEREXT);
2301
2302 /* initialize the inodes within the newly allocated extent a
2303 * page at a time.
2304 */
2305 for (i = 0; i < imap->im_nbperiext; i += sbi->nbperpage) {
2306 /* get a buffer for this page of disk inodes.
2307 */
2308 dmp = get_metapage(ipimap, blkno + i, PSIZE, 1);
2309 if (dmp == NULL) {
2310 rc = -EIO;
2311 goto error_out;
2312 }
2313 dp = (struct dinode *) dmp->data;
2314
2315 /* initialize the inode number, mode, link count and
2316 * inode extent address.
2317 */
2318 for (j = 0; j < INOSPERPAGE; j++, dp++, ino++) {
2319 dp->di_inostamp = cpu_to_le32(sbi->inostamp);
2320 dp->di_number = cpu_to_le32(ino);
2321 dp->di_fileset = cpu_to_le32(FILESYSTEM_I);
2322 dp->di_mode = 0;
2323 dp->di_nlink = 0;
2324 PXDaddress(&(dp->di_ixpxd), blkno);
2325 PXDlength(&(dp->di_ixpxd), imap->im_nbperiext);
2326 }
2327 write_metapage(dmp);
2328 }
2329
2330 /* if this is the last free extent within the iag, remove the
2331 * iag from the ag free extent list.
2332 */
2333 if (iagp->nfreeexts == cpu_to_le32(1)) {
2334 if (fwd >= 0)
2335 aiagp->extfreeback = iagp->extfreeback;
2336
2337 if (back >= 0)
2338 biagp->extfreefwd = iagp->extfreefwd;
2339 else
2340 imap->im_agctl[agno].extfree =
2341 le32_to_cpu(iagp->extfreefwd);
2342
2343 iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1);
2344 } else {
2345 /* if the iag has all free extents (newly allocated iag),
2346 * add the iag to the ag free extent list.
2347 */
2348 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2349 if (fwd >= 0)
2350 aiagp->extfreeback = cpu_to_le32(iagno);
2351
2352 iagp->extfreefwd = cpu_to_le32(fwd);
2353 iagp->extfreeback = cpu_to_le32(-1);
2354 imap->im_agctl[agno].extfree = iagno;
2355 }
2356 }
2357
2358 /* if the iag has no free inodes, add the iag to the
2359 * ag free inode list.
2360 */
2361 if (iagp->nfreeinos == 0) {
2362 if (freei >= 0)
2363 ciagp->inofreeback = cpu_to_le32(iagno);
2364
2365 iagp->inofreefwd =
2366 cpu_to_le32(imap->im_agctl[agno].inofree);
2367 iagp->inofreeback = cpu_to_le32(-1);
2368 imap->im_agctl[agno].inofree = iagno;
2369 }
2370
2371 /* initialize the extent descriptor of the extent. */
2372 PXDlength(&iagp->inoext[extno], imap->im_nbperiext);
2373 PXDaddress(&iagp->inoext[extno], blkno);
2374
2375 /* initialize the working and persistent map of the extent.
2376 * the working map will be initialized such that
2377 * it indicates the first inode of the extent is allocated.
2378 */
2379 iagp->wmap[extno] = cpu_to_le32(HIGHORDER);
2380 iagp->pmap[extno] = 0;
2381
2382 /* update the free inode and free extent summary maps
2383 * for the extent to indicate the extent has free inodes
2384 * and no longer represents a free extent.
2385 */
2386 sword = extno >> L2EXTSPERSUM;
2387 mask = HIGHORDER >> (extno & (EXTSPERSUM - 1));
2388 iagp->extsmap[sword] |= cpu_to_le32(mask);
2389 iagp->inosmap[sword] &= cpu_to_le32(~mask);
2390
2391 /* update the free inode and free extent counts for the
2392 * iag.
2393 */
2394 iagp->nfreeinos = cpu_to_le32(le32_to_cpu(iagp->nfreeinos) +
2395 (INOSPEREXT - 1));
2396 iagp->nfreeexts = cpu_to_le32(le32_to_cpu(iagp->nfreeexts) - 1);
2397
2398 /* update the free and backed inode counts for the ag.
2399 */
2400 imap->im_agctl[agno].numfree += (INOSPEREXT - 1);
2401 imap->im_agctl[agno].numinos += INOSPEREXT;
2402
2403 /* update the free and backed inode counts for the inode map.
2404 */
2405 atomic_add(INOSPEREXT - 1, &imap->im_numfree);
2406 atomic_add(INOSPEREXT, &imap->im_numinos);
2407
2408 /* write the iags.
2409 */
2410 if (amp)
2411 write_metapage(amp);
2412 if (bmp)
2413 write_metapage(bmp);
2414 if (cmp)
2415 write_metapage(cmp);
2416
2417 return (0);
2418
2419 error_out:
2420
2421 /* release the iags.
2422 */
2423 if (amp)
2424 release_metapage(amp);
2425 if (bmp)
2426 release_metapage(bmp);
2427 if (cmp)
2428 release_metapage(cmp);
2429
2430 return (rc);
2431}
2432
2433
2434/*
2435 * NAME: diNewIAG(imap,iagnop,agno)
2436 *
2437 * FUNCTION: allocate a new iag for an allocation group.
2438 *
2439 * first tries to allocate the iag from the inode map
2440 * iagfree list:
2441 * if the list has free iags, the head of the list is removed
2442 * and returned to satisfy the request.
2443 * if the inode map's iag free list is empty, the inode map
2444 * is extended to hold a new iag. this new iag is initialized
2445 * and returned to satisfy the request.
2446 *
2447 * PARAMETERS:
2448 * imap - pointer to inode map control structure.
2449 * iagnop - pointer to an iag number set with the number of the
2450 * newly allocated iag upon successful return.
2451 * agno - allocation group number.
2452 * bpp - Buffer pointer to be filled in with new IAG's buffer
2453 *
2454 * RETURN VALUES:
2455 * 0 - success.
2456 * -ENOSPC - insufficient disk resources.
2457 * -EIO - i/o error.
2458 *
2459 * serialization:
2460 * AG lock held on entry/exit;
2461 * write lock on the map is held inside;
2462 * read lock on the map is held on successful completion;
2463 *
2464 * note: new iag transaction:
2465 * . synchronously write iag;
2466 * . write log of xtree and inode of imap;
2467 * . commit;
2468 * . synchronous write of xtree (right to left, bottom to top);
2469 * . at start of logredo(): init in-memory imap with one additional iag page;
2470 * . at end of logredo(): re-read imap inode to determine
2471 * new imap size;
2472 */
2473static int
2474diNewIAG(struct inomap * imap, int *iagnop, int agno, struct metapage ** mpp)
2475{
2476 int rc;
2477 int iagno, i, xlen;
2478 struct inode *ipimap;
2479 struct super_block *sb;
2480 struct jfs_sb_info *sbi;
2481 struct metapage *mp;
2482 struct iag *iagp;
2483 s64 xaddr = 0;
2484 s64 blkno;
2485 tid_t tid;
2486#ifdef _STILL_TO_PORT
2487 xad_t xad;
2488#endif /* _STILL_TO_PORT */
2489 struct inode *iplist[1];
2490
2491 /* pick up pointers to the inode map and mount inodes */
2492 ipimap = imap->im_ipimap;
2493 sb = ipimap->i_sb;
2494 sbi = JFS_SBI(sb);
2495
2496 /* acquire the free iag lock */
2497 IAGFREE_LOCK(imap);
2498
2499 /* if there are any iags on the inode map free iag list,
2500 * allocate the iag from the head of the list.
2501 */
2502 if (imap->im_freeiag >= 0) {
2503 /* pick up the iag number at the head of the list */
2504 iagno = imap->im_freeiag;
2505
2506 /* determine the logical block number of the iag */
2507 blkno = IAGTOLBLK(iagno, sbi->l2nbperpage);
2508 } else {
2509 /* no free iags. the inode map will have to be extented
2510 * to include a new iag.
2511 */
2512
2513 /* acquire inode map lock */
2514 IWRITE_LOCK(ipimap);
2515
2516 if (ipimap->i_size >> L2PSIZE != imap->im_nextiag + 1) {
2517 IWRITE_UNLOCK(ipimap);
2518 IAGFREE_UNLOCK(imap);
2519 jfs_error(imap->im_ipimap->i_sb,
2520 "diNewIAG: ipimap->i_size is wrong");
2521 return -EIO;
2522 }
2523
2524
2525 /* get the next avaliable iag number */
2526 iagno = imap->im_nextiag;
2527
2528 /* make sure that we have not exceeded the maximum inode
2529 * number limit.
2530 */
2531 if (iagno > (MAXIAGS - 1)) {
2532 /* release the inode map lock */
2533 IWRITE_UNLOCK(ipimap);
2534
2535 rc = -ENOSPC;
2536 goto out;
2537 }
2538
2539 /*
2540 * synchronously append new iag page.
2541 */
2542 /* determine the logical address of iag page to append */
2543 blkno = IAGTOLBLK(iagno, sbi->l2nbperpage);
2544
2545 /* Allocate extent for new iag page */
2546 xlen = sbi->nbperpage;
2547 if ((rc = dbAlloc(ipimap, 0, (s64) xlen, &xaddr))) {
2548 /* release the inode map lock */
2549 IWRITE_UNLOCK(ipimap);
2550
2551 goto out;
2552 }
2553
Dave Kleikampd2e83702005-05-02 12:24:51 -06002554 /*
2555 * start transaction of update of the inode map
2556 * addressing structure pointing to the new iag page;
2557 */
2558 tid = txBegin(sb, COMMIT_FORCE);
2559 down(&JFS_IP(ipimap)->commit_sem);
2560
2561 /* update the inode map addressing structure to point to it */
2562 if ((rc =
2563 xtInsert(tid, ipimap, 0, blkno, xlen, &xaddr, 0))) {
2564 txEnd(tid);
2565 up(&JFS_IP(ipimap)->commit_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 /* Free the blocks allocated for the iag since it was
2567 * not successfully added to the inode map
2568 */
2569 dbFree(ipimap, xaddr, (s64) xlen);
2570
2571 /* release the inode map lock */
2572 IWRITE_UNLOCK(ipimap);
2573
Dave Kleikampd2e83702005-05-02 12:24:51 -06002574 goto out;
2575 }
2576
2577 /* update the inode map's inode to reflect the extension */
2578 ipimap->i_size += PSIZE;
2579 inode_add_bytes(ipimap, PSIZE);
2580
2581 /* assign a buffer for the page */
2582 mp = get_metapage(ipimap, blkno, PSIZE, 0);
2583 if (!mp) {
2584 /*
2585 * This is very unlikely since we just created the
2586 * extent, but let's try to handle it correctly
2587 */
2588 xtTruncate(tid, ipimap, ipimap->i_size - PSIZE,
2589 COMMIT_PWMAP);
2590
2591 txAbort(tid, 0);
2592 txEnd(tid);
2593
2594 /* release the inode map lock */
2595 IWRITE_UNLOCK(ipimap);
2596
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 rc = -EIO;
2598 goto out;
2599 }
2600 iagp = (struct iag *) mp->data;
2601
2602 /* init the iag */
2603 memset(iagp, 0, sizeof(struct iag));
2604 iagp->iagnum = cpu_to_le32(iagno);
2605 iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1);
2606 iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1);
2607 iagp->iagfree = cpu_to_le32(-1);
2608 iagp->nfreeinos = 0;
2609 iagp->nfreeexts = cpu_to_le32(EXTSPERIAG);
2610
2611 /* initialize the free inode summary map (free extent
2612 * summary map initialization handled by bzero).
2613 */
2614 for (i = 0; i < SMAPSZ; i++)
2615 iagp->inosmap[i] = cpu_to_le32(ONES);
2616
2617 /*
Dave Kleikampd2e83702005-05-02 12:24:51 -06002618 * Write and sync the metapage
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 flush_metapage(mp);
2621
2622 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 * txCommit(COMMIT_FORCE) will synchronously write address
2624 * index pages and inode after commit in careful update order
2625 * of address index pages (right to left, bottom up);
2626 */
2627 iplist[0] = ipimap;
2628 rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE);
2629
2630 txEnd(tid);
2631 up(&JFS_IP(ipimap)->commit_sem);
2632
2633 duplicateIXtree(sb, blkno, xlen, &xaddr);
2634
2635 /* update the next avaliable iag number */
2636 imap->im_nextiag += 1;
2637
2638 /* Add the iag to the iag free list so we don't lose the iag
2639 * if a failure happens now.
2640 */
2641 imap->im_freeiag = iagno;
2642
2643 /* Until we have logredo working, we want the imap inode &
2644 * control page to be up to date.
2645 */
2646 diSync(ipimap);
2647
2648 /* release the inode map lock */
2649 IWRITE_UNLOCK(ipimap);
2650 }
2651
2652 /* obtain read lock on map */
2653 IREAD_LOCK(ipimap);
2654
2655 /* read the iag */
2656 if ((rc = diIAGRead(imap, iagno, &mp))) {
2657 IREAD_UNLOCK(ipimap);
2658 rc = -EIO;
2659 goto out;
2660 }
2661 iagp = (struct iag *) mp->data;
2662
2663 /* remove the iag from the iag free list */
2664 imap->im_freeiag = le32_to_cpu(iagp->iagfree);
2665 iagp->iagfree = cpu_to_le32(-1);
2666
2667 /* set the return iag number and buffer pointer */
2668 *iagnop = iagno;
2669 *mpp = mp;
2670
2671 out:
2672 /* release the iag free lock */
2673 IAGFREE_UNLOCK(imap);
2674
2675 return (rc);
2676}
2677
2678/*
2679 * NAME: diIAGRead()
2680 *
2681 * FUNCTION: get the buffer for the specified iag within a fileset
2682 * or aggregate inode map.
2683 *
2684 * PARAMETERS:
2685 * imap - pointer to inode map control structure.
2686 * iagno - iag number.
2687 * bpp - point to buffer pointer to be filled in on successful
2688 * exit.
2689 *
2690 * SERIALIZATION:
2691 * must have read lock on imap inode
2692 * (When called by diExtendFS, the filesystem is quiesced, therefore
2693 * the read lock is unnecessary.)
2694 *
2695 * RETURN VALUES:
2696 * 0 - success.
2697 * -EIO - i/o error.
2698 */
2699static int diIAGRead(struct inomap * imap, int iagno, struct metapage ** mpp)
2700{
2701 struct inode *ipimap = imap->im_ipimap;
2702 s64 blkno;
2703
2704 /* compute the logical block number of the iag. */
2705 blkno = IAGTOLBLK(iagno, JFS_SBI(ipimap->i_sb)->l2nbperpage);
2706
2707 /* read the iag. */
2708 *mpp = read_metapage(ipimap, blkno, PSIZE, 0);
2709 if (*mpp == NULL) {
2710 return -EIO;
2711 }
2712
2713 return (0);
2714}
2715
2716/*
2717 * NAME: diFindFree()
2718 *
2719 * FUNCTION: find the first free bit in a word starting at
2720 * the specified bit position.
2721 *
2722 * PARAMETERS:
2723 * word - word to be examined.
2724 * start - starting bit position.
2725 *
2726 * RETURN VALUES:
2727 * bit position of first free bit in the word or 32 if
2728 * no free bits were found.
2729 */
2730static int diFindFree(u32 word, int start)
2731{
2732 int bitno;
2733 assert(start < 32);
2734 /* scan the word for the first free bit. */
2735 for (word <<= start, bitno = start; bitno < 32;
2736 bitno++, word <<= 1) {
2737 if ((word & HIGHORDER) == 0)
2738 break;
2739 }
2740 return (bitno);
2741}
2742
2743/*
2744 * NAME: diUpdatePMap()
2745 *
2746 * FUNCTION: Update the persistent map in an IAG for the allocation or
2747 * freeing of the specified inode.
2748 *
2749 * PRE CONDITIONS: Working map has already been updated for allocate.
2750 *
2751 * PARAMETERS:
2752 * ipimap - Incore inode map inode
2753 * inum - Number of inode to mark in permanent map
2754 * is_free - If TRUE indicates inode should be marked freed, otherwise
2755 * indicates inode should be marked allocated.
2756 *
2757 * RETURN VALUES:
2758 * 0 for success
2759 */
2760int
2761diUpdatePMap(struct inode *ipimap,
2762 unsigned long inum, boolean_t is_free, struct tblock * tblk)
2763{
2764 int rc;
2765 struct iag *iagp;
2766 struct metapage *mp;
2767 int iagno, ino, extno, bitno;
2768 struct inomap *imap;
2769 u32 mask;
2770 struct jfs_log *log;
2771 int lsn, difft, diffp;
Dave Kleikamp7fab4792005-05-02 12:25:02 -06002772 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773
2774 imap = JFS_IP(ipimap)->i_imap;
2775 /* get the iag number containing the inode */
2776 iagno = INOTOIAG(inum);
2777 /* make sure that the iag is contained within the map */
2778 if (iagno >= imap->im_nextiag) {
2779 jfs_error(ipimap->i_sb,
2780 "diUpdatePMap: the iag is outside the map");
2781 return -EIO;
2782 }
2783 /* read the iag */
2784 IREAD_LOCK(ipimap);
2785 rc = diIAGRead(imap, iagno, &mp);
2786 IREAD_UNLOCK(ipimap);
2787 if (rc)
2788 return (rc);
Dave Kleikamp7fab4792005-05-02 12:25:02 -06002789 metapage_wait_for_io(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790 iagp = (struct iag *) mp->data;
2791 /* get the inode number and extent number of the inode within
2792 * the iag and the inode number within the extent.
2793 */
2794 ino = inum & (INOSPERIAG - 1);
2795 extno = ino >> L2INOSPEREXT;
2796 bitno = ino & (INOSPEREXT - 1);
2797 mask = HIGHORDER >> bitno;
2798 /*
2799 * mark the inode free in persistent map:
2800 */
2801 if (is_free == TRUE) {
2802 /* The inode should have been allocated both in working
2803 * map and in persistent map;
2804 * the inode will be freed from working map at the release
2805 * of last reference release;
2806 */
2807 if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
2808 jfs_error(ipimap->i_sb,
2809 "diUpdatePMap: inode %ld not marked as "
2810 "allocated in wmap!", inum);
2811 }
2812 if (!(le32_to_cpu(iagp->pmap[extno]) & mask)) {
2813 jfs_error(ipimap->i_sb,
2814 "diUpdatePMap: inode %ld not marked as "
2815 "allocated in pmap!", inum);
2816 }
2817 /* update the bitmap for the extent of the freed inode */
2818 iagp->pmap[extno] &= cpu_to_le32(~mask);
2819 }
2820 /*
2821 * mark the inode allocated in persistent map:
2822 */
2823 else {
2824 /* The inode should be already allocated in the working map
2825 * and should be free in persistent map;
2826 */
2827 if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
2828 release_metapage(mp);
2829 jfs_error(ipimap->i_sb,
2830 "diUpdatePMap: the inode is not allocated in "
2831 "the working map");
2832 return -EIO;
2833 }
2834 if ((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) {
2835 release_metapage(mp);
2836 jfs_error(ipimap->i_sb,
2837 "diUpdatePMap: the inode is not free in the "
2838 "persistent map");
2839 return -EIO;
2840 }
2841 /* update the bitmap for the extent of the allocated inode */
2842 iagp->pmap[extno] |= cpu_to_le32(mask);
2843 }
2844 /*
2845 * update iag lsn
2846 */
2847 lsn = tblk->lsn;
2848 log = JFS_SBI(tblk->sb)->log;
2849 if (mp->lsn != 0) {
2850 /* inherit older/smaller lsn */
2851 logdiff(difft, lsn, log);
2852 logdiff(diffp, mp->lsn, log);
Dave Kleikamp7fab4792005-05-02 12:25:02 -06002853 LOGSYNC_LOCK(log, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854 if (difft < diffp) {
2855 mp->lsn = lsn;
2856 /* move mp after tblock in logsync list */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 list_move(&mp->synclist, &tblk->synclist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858 }
2859 /* inherit younger/larger clsn */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860 assert(mp->clsn);
2861 logdiff(difft, tblk->clsn, log);
2862 logdiff(diffp, mp->clsn, log);
2863 if (difft > diffp)
2864 mp->clsn = tblk->clsn;
Dave Kleikamp7fab4792005-05-02 12:25:02 -06002865 LOGSYNC_UNLOCK(log, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002866 } else {
2867 mp->log = log;
2868 mp->lsn = lsn;
2869 /* insert mp after tblock in logsync list */
Dave Kleikamp7fab4792005-05-02 12:25:02 -06002870 LOGSYNC_LOCK(log, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871 log->count++;
2872 list_add(&mp->synclist, &tblk->synclist);
2873 mp->clsn = tblk->clsn;
Dave Kleikamp7fab4792005-05-02 12:25:02 -06002874 LOGSYNC_UNLOCK(log, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875 }
2876 write_metapage(mp);
2877 return (0);
2878}
2879
2880/*
2881 * diExtendFS()
2882 *
2883 * function: update imap for extendfs();
2884 *
2885 * note: AG size has been increased s.t. each k old contiguous AGs are
2886 * coalesced into a new AG;
2887 */
2888int diExtendFS(struct inode *ipimap, struct inode *ipbmap)
2889{
2890 int rc, rcx = 0;
2891 struct inomap *imap = JFS_IP(ipimap)->i_imap;
2892 struct iag *iagp = NULL, *hiagp = NULL;
2893 struct bmap *mp = JFS_SBI(ipbmap->i_sb)->bmap;
2894 struct metapage *bp, *hbp;
2895 int i, n, head;
2896 int numinos, xnuminos = 0, xnumfree = 0;
2897 s64 agstart;
2898
2899 jfs_info("diExtendFS: nextiag:%d numinos:%d numfree:%d",
2900 imap->im_nextiag, atomic_read(&imap->im_numinos),
2901 atomic_read(&imap->im_numfree));
2902
2903 /*
2904 * reconstruct imap
2905 *
2906 * coalesce contiguous k (newAGSize/oldAGSize) AGs;
2907 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn;
2908 * note: new AG size = old AG size * (2**x).
2909 */
2910
2911 /* init per AG control information im_agctl[] */
2912 for (i = 0; i < MAXAG; i++) {
2913 imap->im_agctl[i].inofree = -1;
2914 imap->im_agctl[i].extfree = -1;
2915 imap->im_agctl[i].numinos = 0; /* number of backed inodes */
2916 imap->im_agctl[i].numfree = 0; /* number of free backed inodes */
2917 }
2918
2919 /*
2920 * process each iag page of the map.
2921 *
2922 * rebuild AG Free Inode List, AG Free Inode Extent List;
2923 */
2924 for (i = 0; i < imap->im_nextiag; i++) {
2925 if ((rc = diIAGRead(imap, i, &bp))) {
2926 rcx = rc;
2927 continue;
2928 }
2929 iagp = (struct iag *) bp->data;
2930 if (le32_to_cpu(iagp->iagnum) != i) {
2931 release_metapage(bp);
2932 jfs_error(ipimap->i_sb,
2933 "diExtendFs: unexpected value of iagnum");
2934 return -EIO;
2935 }
2936
2937 /* leave free iag in the free iag list */
2938 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2939 release_metapage(bp);
2940 continue;
2941 }
2942
2943 /* agstart that computes to the same ag is treated as same; */
2944 agstart = le64_to_cpu(iagp->agstart);
2945 /* iagp->agstart = agstart & ~(mp->db_agsize - 1); */
2946 n = agstart >> mp->db_agl2size;
2947
2948 /* compute backed inodes */
2949 numinos = (EXTSPERIAG - le32_to_cpu(iagp->nfreeexts))
2950 << L2INOSPEREXT;
2951 if (numinos > 0) {
2952 /* merge AG backed inodes */
2953 imap->im_agctl[n].numinos += numinos;
2954 xnuminos += numinos;
2955 }
2956
2957 /* if any backed free inodes, insert at AG free inode list */
2958 if ((int) le32_to_cpu(iagp->nfreeinos) > 0) {
2959 if ((head = imap->im_agctl[n].inofree) == -1) {
2960 iagp->inofreefwd = cpu_to_le32(-1);
2961 iagp->inofreeback = cpu_to_le32(-1);
2962 } else {
2963 if ((rc = diIAGRead(imap, head, &hbp))) {
2964 rcx = rc;
2965 goto nextiag;
2966 }
2967 hiagp = (struct iag *) hbp->data;
2968 hiagp->inofreeback = iagp->iagnum;
2969 iagp->inofreefwd = cpu_to_le32(head);
2970 iagp->inofreeback = cpu_to_le32(-1);
2971 write_metapage(hbp);
2972 }
2973
2974 imap->im_agctl[n].inofree =
2975 le32_to_cpu(iagp->iagnum);
2976
2977 /* merge AG backed free inodes */
2978 imap->im_agctl[n].numfree +=
2979 le32_to_cpu(iagp->nfreeinos);
2980 xnumfree += le32_to_cpu(iagp->nfreeinos);
2981 }
2982
2983 /* if any free extents, insert at AG free extent list */
2984 if (le32_to_cpu(iagp->nfreeexts) > 0) {
2985 if ((head = imap->im_agctl[n].extfree) == -1) {
2986 iagp->extfreefwd = cpu_to_le32(-1);
2987 iagp->extfreeback = cpu_to_le32(-1);
2988 } else {
2989 if ((rc = diIAGRead(imap, head, &hbp))) {
2990 rcx = rc;
2991 goto nextiag;
2992 }
2993 hiagp = (struct iag *) hbp->data;
2994 hiagp->extfreeback = iagp->iagnum;
2995 iagp->extfreefwd = cpu_to_le32(head);
2996 iagp->extfreeback = cpu_to_le32(-1);
2997 write_metapage(hbp);
2998 }
2999
3000 imap->im_agctl[n].extfree =
3001 le32_to_cpu(iagp->iagnum);
3002 }
3003
3004 nextiag:
3005 write_metapage(bp);
3006 }
3007
3008 if (xnuminos != atomic_read(&imap->im_numinos) ||
3009 xnumfree != atomic_read(&imap->im_numfree)) {
3010 jfs_error(ipimap->i_sb,
3011 "diExtendFs: numinos or numfree incorrect");
3012 return -EIO;
3013 }
3014
3015 return rcx;
3016}
3017
3018
3019/*
3020 * duplicateIXtree()
3021 *
3022 * serialization: IWRITE_LOCK held on entry/exit
3023 *
3024 * note: shadow page with regular inode (rel.2);
3025 */
3026static void duplicateIXtree(struct super_block *sb, s64 blkno,
3027 int xlen, s64 *xaddr)
3028{
3029 struct jfs_superblock *j_sb;
3030 struct buffer_head *bh;
3031 struct inode *ip;
3032 tid_t tid;
3033
3034 /* if AIT2 ipmap2 is bad, do not try to update it */
3035 if (JFS_SBI(sb)->mntflag & JFS_BAD_SAIT) /* s_flag */
3036 return;
3037 ip = diReadSpecial(sb, FILESYSTEM_I, 1);
3038 if (ip == NULL) {
3039 JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT;
3040 if (readSuper(sb, &bh))
3041 return;
3042 j_sb = (struct jfs_superblock *)bh->b_data;
3043 j_sb->s_flag |= cpu_to_le32(JFS_BAD_SAIT);
3044
3045 mark_buffer_dirty(bh);
3046 sync_dirty_buffer(bh);
3047 brelse(bh);
3048 return;
3049 }
3050
3051 /* start transaction */
3052 tid = txBegin(sb, COMMIT_FORCE);
3053 /* update the inode map addressing structure to point to it */
3054 if (xtInsert(tid, ip, 0, blkno, xlen, xaddr, 0)) {
3055 JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT;
3056 txAbort(tid, 1);
3057 goto cleanup;
3058
3059 }
3060 /* update the inode map's inode to reflect the extension */
3061 ip->i_size += PSIZE;
3062 inode_add_bytes(ip, PSIZE);
3063 txCommit(tid, 1, &ip, COMMIT_FORCE);
3064 cleanup:
3065 txEnd(tid);
3066 diFreeSpecial(ip);
3067}
3068
3069/*
3070 * NAME: copy_from_dinode()
3071 *
3072 * FUNCTION: Copies inode info from disk inode to in-memory inode
3073 *
3074 * RETURN VALUES:
3075 * 0 - success
3076 * -ENOMEM - insufficient memory
3077 */
3078static int copy_from_dinode(struct dinode * dip, struct inode *ip)
3079{
3080 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
3081
3082 jfs_ip->fileset = le32_to_cpu(dip->di_fileset);
3083 jfs_ip->mode2 = le32_to_cpu(dip->di_mode);
3084
3085 ip->i_mode = le32_to_cpu(dip->di_mode) & 0xffff;
3086 ip->i_nlink = le32_to_cpu(dip->di_nlink);
3087 ip->i_uid = le32_to_cpu(dip->di_uid);
3088 ip->i_gid = le32_to_cpu(dip->di_gid);
3089 ip->i_size = le64_to_cpu(dip->di_size);
3090 ip->i_atime.tv_sec = le32_to_cpu(dip->di_atime.tv_sec);
3091 ip->i_atime.tv_nsec = le32_to_cpu(dip->di_atime.tv_nsec);
3092 ip->i_mtime.tv_sec = le32_to_cpu(dip->di_mtime.tv_sec);
3093 ip->i_mtime.tv_nsec = le32_to_cpu(dip->di_mtime.tv_nsec);
3094 ip->i_ctime.tv_sec = le32_to_cpu(dip->di_ctime.tv_sec);
3095 ip->i_ctime.tv_nsec = le32_to_cpu(dip->di_ctime.tv_nsec);
3096 ip->i_blksize = ip->i_sb->s_blocksize;
3097 ip->i_blocks = LBLK2PBLK(ip->i_sb, le64_to_cpu(dip->di_nblocks));
3098 ip->i_generation = le32_to_cpu(dip->di_gen);
3099
3100 jfs_ip->ixpxd = dip->di_ixpxd; /* in-memory pxd's are little-endian */
3101 jfs_ip->acl = dip->di_acl; /* as are dxd's */
3102 jfs_ip->ea = dip->di_ea;
3103 jfs_ip->next_index = le32_to_cpu(dip->di_next_index);
3104 jfs_ip->otime = le32_to_cpu(dip->di_otime.tv_sec);
3105 jfs_ip->acltype = le32_to_cpu(dip->di_acltype);
3106
3107 if (S_ISCHR(ip->i_mode) || S_ISBLK(ip->i_mode)) {
3108 jfs_ip->dev = le32_to_cpu(dip->di_rdev);
3109 ip->i_rdev = new_decode_dev(jfs_ip->dev);
3110 }
3111
3112 if (S_ISDIR(ip->i_mode)) {
3113 memcpy(&jfs_ip->i_dirtable, &dip->di_dirtable, 384);
3114 } else if (S_ISREG(ip->i_mode) || S_ISLNK(ip->i_mode)) {
3115 memcpy(&jfs_ip->i_xtroot, &dip->di_xtroot, 288);
3116 } else
3117 memcpy(&jfs_ip->i_inline_ea, &dip->di_inlineea, 128);
3118
3119 /* Zero the in-memory-only stuff */
3120 jfs_ip->cflag = 0;
3121 jfs_ip->btindex = 0;
3122 jfs_ip->btorder = 0;
3123 jfs_ip->bxflag = 0;
3124 jfs_ip->blid = 0;
3125 jfs_ip->atlhead = 0;
3126 jfs_ip->atltail = 0;
3127 jfs_ip->xtlid = 0;
3128 return (0);
3129}
3130
3131/*
3132 * NAME: copy_to_dinode()
3133 *
3134 * FUNCTION: Copies inode info from in-memory inode to disk inode
3135 */
3136static void copy_to_dinode(struct dinode * dip, struct inode *ip)
3137{
3138 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
3139
3140 dip->di_fileset = cpu_to_le32(jfs_ip->fileset);
3141 dip->di_inostamp = cpu_to_le32(JFS_SBI(ip->i_sb)->inostamp);
3142 dip->di_number = cpu_to_le32(ip->i_ino);
3143 dip->di_gen = cpu_to_le32(ip->i_generation);
3144 dip->di_size = cpu_to_le64(ip->i_size);
3145 dip->di_nblocks = cpu_to_le64(PBLK2LBLK(ip->i_sb, ip->i_blocks));
3146 dip->di_nlink = cpu_to_le32(ip->i_nlink);
3147 dip->di_uid = cpu_to_le32(ip->i_uid);
3148 dip->di_gid = cpu_to_le32(ip->i_gid);
3149 /*
3150 * mode2 is only needed for storing the higher order bits.
3151 * Trust i_mode for the lower order ones
3152 */
3153 dip->di_mode = cpu_to_le32((jfs_ip->mode2 & 0xffff0000) | ip->i_mode);
3154 dip->di_atime.tv_sec = cpu_to_le32(ip->i_atime.tv_sec);
3155 dip->di_atime.tv_nsec = cpu_to_le32(ip->i_atime.tv_nsec);
3156 dip->di_ctime.tv_sec = cpu_to_le32(ip->i_ctime.tv_sec);
3157 dip->di_ctime.tv_nsec = cpu_to_le32(ip->i_ctime.tv_nsec);
3158 dip->di_mtime.tv_sec = cpu_to_le32(ip->i_mtime.tv_sec);
3159 dip->di_mtime.tv_nsec = cpu_to_le32(ip->i_mtime.tv_nsec);
3160 dip->di_ixpxd = jfs_ip->ixpxd; /* in-memory pxd's are little-endian */
3161 dip->di_acl = jfs_ip->acl; /* as are dxd's */
3162 dip->di_ea = jfs_ip->ea;
3163 dip->di_next_index = cpu_to_le32(jfs_ip->next_index);
3164 dip->di_otime.tv_sec = cpu_to_le32(jfs_ip->otime);
3165 dip->di_otime.tv_nsec = 0;
3166 dip->di_acltype = cpu_to_le32(jfs_ip->acltype);
3167 if (S_ISCHR(ip->i_mode) || S_ISBLK(ip->i_mode))
3168 dip->di_rdev = cpu_to_le32(jfs_ip->dev);
3169}