blob: ccbe60aff83de64edddc27381e020c83240adb1b [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 */
Ingo Molnar1de87442006-01-24 15:22:50 -060069#define IAGFREE_LOCK_INIT(imap) mutex_init(&imap->im_freelock)
70#define IAGFREE_LOCK(imap) mutex_lock(&imap->im_freelock)
71#define IAGFREE_UNLOCK(imap) mutex_unlock(&imap->im_freelock)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
73/* per ag iag list locks */
Ingo Molnar1de87442006-01-24 15:22:50 -060074#define AG_LOCK_INIT(imap,index) mutex_init(&(imap->im_aglock[index]))
75#define AG_LOCK(imap,agno) mutex_lock(&imap->im_aglock[agno])
76#define AG_UNLOCK(imap,agno) mutex_unlock(&imap->im_aglock[agno])
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
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 */
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800268 filemap_write_and_wait(ipimap->i_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269
270 diWriteSpecial(ipimap, 0);
271
272 return (0);
273}
274
275
276/*
277 * NAME: diRead()
278 *
279 * FUNCTION: initialize an incore inode from disk.
280 *
281 * on entry, the specifed incore inode should itself
282 * specify the disk inode number corresponding to the
283 * incore inode (i.e. i_number should be initialized).
284 *
285 * this routine handles incore inode initialization for
286 * both "special" and "regular" inodes. special inodes
287 * are those required early in the mount process and
288 * require special handling since much of the file system
289 * is not yet initialized. these "special" inodes are
290 * identified by a NULL inode map inode pointer and are
291 * actually initialized by a call to diReadSpecial().
292 *
293 * for regular inodes, the iag describing the disk inode
294 * is read from disk to determine the inode extent address
295 * for the disk inode. with the inode extent address in
296 * hand, the page of the extent that contains the disk
297 * inode is read and the disk inode is copied to the
298 * incore inode.
299 *
300 * PARAMETERS:
301 * ip - pointer to incore inode to be initialized from disk.
302 *
303 * RETURN VALUES:
304 * 0 - success
305 * -EIO - i/o error.
306 * -ENOMEM - insufficient memory
307 *
308 */
309int diRead(struct inode *ip)
310{
311 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
312 int iagno, ino, extno, rc;
313 struct inode *ipimap;
314 struct dinode *dp;
315 struct iag *iagp;
316 struct metapage *mp;
317 s64 blkno, agstart;
318 struct inomap *imap;
319 int block_offset;
320 int inodes_left;
321 uint pageno;
322 int rel_inode;
323
324 jfs_info("diRead: ino = %ld", ip->i_ino);
325
326 ipimap = sbi->ipimap;
327 JFS_IP(ip)->ipimap = ipimap;
328
329 /* determine the iag number for this inode (number) */
330 iagno = INOTOIAG(ip->i_ino);
331
332 /* read the iag */
333 imap = JFS_IP(ipimap)->i_imap;
334 IREAD_LOCK(ipimap);
335 rc = diIAGRead(imap, iagno, &mp);
336 IREAD_UNLOCK(ipimap);
337 if (rc) {
338 jfs_err("diRead: diIAGRead returned %d", rc);
339 return (rc);
340 }
341
342 iagp = (struct iag *) mp->data;
343
344 /* determine inode extent that holds the disk inode */
345 ino = ip->i_ino & (INOSPERIAG - 1);
346 extno = ino >> L2INOSPEREXT;
347
348 if ((lengthPXD(&iagp->inoext[extno]) != imap->im_nbperiext) ||
349 (addressPXD(&iagp->inoext[extno]) == 0)) {
350 release_metapage(mp);
351 return -ESTALE;
352 }
353
354 /* get disk block number of the page within the inode extent
355 * that holds the disk inode.
356 */
357 blkno = INOPBLK(&iagp->inoext[extno], ino, sbi->l2nbperpage);
358
359 /* get the ag for the iag */
360 agstart = le64_to_cpu(iagp->agstart);
361
362 release_metapage(mp);
363
364 rel_inode = (ino & (INOSPERPAGE - 1));
365 pageno = blkno >> sbi->l2nbperpage;
366
367 if ((block_offset = ((u32) blkno & (sbi->nbperpage - 1)))) {
368 /*
369 * OS/2 didn't always align inode extents on page boundaries
370 */
371 inodes_left =
372 (sbi->nbperpage - block_offset) << sbi->l2niperblk;
373
374 if (rel_inode < inodes_left)
375 rel_inode += block_offset << sbi->l2niperblk;
376 else {
377 pageno += 1;
378 rel_inode -= inodes_left;
379 }
380 }
381
382 /* read the page of disk inode */
383 mp = read_metapage(ipimap, pageno << sbi->l2nbperpage, PSIZE, 1);
384 if (mp == 0) {
385 jfs_err("diRead: read_metapage failed");
386 return -EIO;
387 }
388
389 /* locate the the disk inode requested */
390 dp = (struct dinode *) mp->data;
391 dp += rel_inode;
392
393 if (ip->i_ino != le32_to_cpu(dp->di_number)) {
394 jfs_error(ip->i_sb, "diRead: i_ino != di_number");
395 rc = -EIO;
396 } else if (le32_to_cpu(dp->di_nlink) == 0)
397 rc = -ESTALE;
398 else
399 /* copy the disk inode to the in-memory inode */
400 rc = copy_from_dinode(dp, ip);
401
402 release_metapage(mp);
403
404 /* set the ag for the inode */
405 JFS_IP(ip)->agno = BLKTOAG(agstart, sbi);
406 JFS_IP(ip)->active_ag = -1;
407
408 return (rc);
409}
410
411
412/*
413 * NAME: diReadSpecial()
414 *
415 * FUNCTION: initialize a 'special' inode from disk.
416 *
417 * this routines handles aggregate level inodes. The
418 * inode cache cannot differentiate between the
419 * aggregate inodes and the filesystem inodes, so we
420 * handle these here. We don't actually use the aggregate
421 * inode map, since these inodes are at a fixed location
422 * and in some cases the aggregate inode map isn't initialized
423 * yet.
424 *
425 * PARAMETERS:
426 * sb - filesystem superblock
427 * inum - aggregate inode number
428 * secondary - 1 if secondary aggregate inode table
429 *
430 * RETURN VALUES:
431 * new inode - success
432 * NULL - i/o error.
433 */
434struct inode *diReadSpecial(struct super_block *sb, ino_t inum, int secondary)
435{
436 struct jfs_sb_info *sbi = JFS_SBI(sb);
437 uint address;
438 struct dinode *dp;
439 struct inode *ip;
440 struct metapage *mp;
441
442 ip = new_inode(sb);
443 if (ip == NULL) {
444 jfs_err("diReadSpecial: new_inode returned NULL!");
445 return ip;
446 }
447
448 if (secondary) {
449 address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage;
450 JFS_IP(ip)->ipimap = sbi->ipaimap2;
451 } else {
452 address = AITBL_OFF >> L2PSIZE;
453 JFS_IP(ip)->ipimap = sbi->ipaimap;
454 }
455
456 ASSERT(inum < INOSPEREXT);
457
458 ip->i_ino = inum;
459
460 address += inum >> 3; /* 8 inodes per 4K page */
461
462 /* read the page of fixed disk inode (AIT) in raw mode */
463 mp = read_metapage(ip, address << sbi->l2nbperpage, PSIZE, 1);
464 if (mp == NULL) {
465 ip->i_nlink = 1; /* Don't want iput() deleting it */
466 iput(ip);
467 return (NULL);
468 }
469
470 /* get the pointer to the disk inode of interest */
471 dp = (struct dinode *) (mp->data);
472 dp += inum % 8; /* 8 inodes per 4K page */
473
474 /* copy on-disk inode to in-memory inode */
475 if ((copy_from_dinode(dp, ip)) != 0) {
476 /* handle bad return by returning NULL for ip */
477 ip->i_nlink = 1; /* Don't want iput() deleting it */
478 iput(ip);
479 /* release the page */
480 release_metapage(mp);
481 return (NULL);
482
483 }
484
Dave Kleikamp7fab4792005-05-02 12:25:02 -0600485 ip->i_mapping->a_ops = &jfs_metapage_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 mapping_set_gfp_mask(ip->i_mapping, GFP_NOFS);
487
488 /* Allocations to metadata inodes should not affect quotas */
489 ip->i_flags |= S_NOQUOTA;
490
491 if ((inum == FILESYSTEM_I) && (JFS_IP(ip)->ipimap == sbi->ipaimap)) {
492 sbi->gengen = le32_to_cpu(dp->di_gengen);
493 sbi->inostamp = le32_to_cpu(dp->di_inostamp);
494 }
495
496 /* release the page */
497 release_metapage(mp);
498
Dave Kleikampac17b8b2005-10-03 15:32:11 -0500499 hlist_add_head(&ip->i_hash, &aggregate_hash);
500
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 return (ip);
502}
503
504/*
505 * NAME: diWriteSpecial()
506 *
507 * FUNCTION: Write the special inode to disk
508 *
509 * PARAMETERS:
510 * ip - special inode
511 * secondary - 1 if secondary aggregate inode table
512 *
513 * RETURN VALUES: none
514 */
515
516void diWriteSpecial(struct inode *ip, int secondary)
517{
518 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
519 uint address;
520 struct dinode *dp;
521 ino_t inum = ip->i_ino;
522 struct metapage *mp;
523
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 if (secondary)
525 address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage;
526 else
527 address = AITBL_OFF >> L2PSIZE;
528
529 ASSERT(inum < INOSPEREXT);
530
531 address += inum >> 3; /* 8 inodes per 4K page */
532
533 /* read the page of fixed disk inode (AIT) in raw mode */
534 mp = read_metapage(ip, address << sbi->l2nbperpage, PSIZE, 1);
535 if (mp == NULL) {
536 jfs_err("diWriteSpecial: failed to read aggregate inode "
537 "extent!");
538 return;
539 }
540
541 /* get the pointer to the disk inode of interest */
542 dp = (struct dinode *) (mp->data);
543 dp += inum % 8; /* 8 inodes per 4K page */
544
545 /* copy on-disk inode to in-memory inode */
546 copy_to_dinode(dp, ip);
547 memcpy(&dp->di_xtroot, &JFS_IP(ip)->i_xtroot, 288);
548
549 if (inum == FILESYSTEM_I)
550 dp->di_gengen = cpu_to_le32(sbi->gengen);
551
552 /* write the page */
553 write_metapage(mp);
554}
555
556/*
557 * NAME: diFreeSpecial()
558 *
559 * FUNCTION: Free allocated space for special inode
560 */
561void diFreeSpecial(struct inode *ip)
562{
563 if (ip == NULL) {
564 jfs_err("diFreeSpecial called with NULL ip!");
565 return;
566 }
OGAWA Hirofumi28fd1292006-01-08 01:02:14 -0800567 filemap_write_and_wait(ip->i_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 truncate_inode_pages(ip->i_mapping, 0);
569 iput(ip);
570}
571
572
573
574/*
575 * NAME: diWrite()
576 *
577 * FUNCTION: write the on-disk inode portion of the in-memory inode
578 * to its corresponding on-disk inode.
579 *
580 * on entry, the specifed incore inode should itself
581 * specify the disk inode number corresponding to the
582 * incore inode (i.e. i_number should be initialized).
583 *
584 * the inode contains the inode extent address for the disk
585 * inode. with the inode extent address in hand, the
586 * page of the extent that contains the disk inode is
587 * read and the disk inode portion of the incore inode
588 * is copied to the disk inode.
589 *
590 * PARAMETERS:
591 * tid - transacation id
592 * ip - pointer to incore inode to be written to the inode extent.
593 *
594 * RETURN VALUES:
595 * 0 - success
596 * -EIO - i/o error.
597 */
598int diWrite(tid_t tid, struct inode *ip)
599{
600 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
601 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
602 int rc = 0;
603 s32 ino;
604 struct dinode *dp;
605 s64 blkno;
606 int block_offset;
607 int inodes_left;
608 struct metapage *mp;
609 uint pageno;
610 int rel_inode;
611 int dioffset;
612 struct inode *ipimap;
613 uint type;
614 lid_t lid;
615 struct tlock *ditlck, *tlck;
616 struct linelock *dilinelock, *ilinelock;
617 struct lv *lv;
618 int n;
619
620 ipimap = jfs_ip->ipimap;
621
622 ino = ip->i_ino & (INOSPERIAG - 1);
623
624 if (!addressPXD(&(jfs_ip->ixpxd)) ||
625 (lengthPXD(&(jfs_ip->ixpxd)) !=
626 JFS_IP(ipimap)->i_imap->im_nbperiext)) {
627 jfs_error(ip->i_sb, "diWrite: ixpxd invalid");
628 return -EIO;
629 }
630
631 /*
632 * read the page of disk inode containing the specified inode:
633 */
634 /* compute the block address of the page */
635 blkno = INOPBLK(&(jfs_ip->ixpxd), ino, sbi->l2nbperpage);
636
637 rel_inode = (ino & (INOSPERPAGE - 1));
638 pageno = blkno >> sbi->l2nbperpage;
639
640 if ((block_offset = ((u32) blkno & (sbi->nbperpage - 1)))) {
641 /*
642 * OS/2 didn't always align inode extents on page boundaries
643 */
644 inodes_left =
645 (sbi->nbperpage - block_offset) << sbi->l2niperblk;
646
647 if (rel_inode < inodes_left)
648 rel_inode += block_offset << sbi->l2niperblk;
649 else {
650 pageno += 1;
651 rel_inode -= inodes_left;
652 }
653 }
654 /* read the page of disk inode */
655 retry:
656 mp = read_metapage(ipimap, pageno << sbi->l2nbperpage, PSIZE, 1);
657 if (mp == 0)
658 return -EIO;
659
660 /* get the pointer to the disk inode */
661 dp = (struct dinode *) mp->data;
662 dp += rel_inode;
663
664 dioffset = (ino & (INOSPERPAGE - 1)) << L2DISIZE;
665
666 /*
667 * acquire transaction lock on the on-disk inode;
668 * N.B. tlock is acquired on ipimap not ip;
669 */
670 if ((ditlck =
671 txLock(tid, ipimap, mp, tlckINODE | tlckENTRY)) == NULL)
672 goto retry;
673 dilinelock = (struct linelock *) & ditlck->lock;
674
675 /*
676 * copy btree root from in-memory inode to on-disk inode
677 *
678 * (tlock is taken from inline B+-tree root in in-memory
679 * inode when the B+-tree root is updated, which is pointed
680 * by jfs_ip->blid as well as being on tx tlock list)
681 *
682 * further processing of btree root is based on the copy
683 * in in-memory inode, where txLog() will log from, and,
684 * for xtree root, txUpdateMap() will update map and reset
685 * XAD_NEW bit;
686 */
687
688 if (S_ISDIR(ip->i_mode) && (lid = jfs_ip->xtlid)) {
689 /*
690 * This is the special xtree inside the directory for storing
691 * the directory table
692 */
693 xtpage_t *p, *xp;
694 xad_t *xad;
695
696 jfs_ip->xtlid = 0;
697 tlck = lid_to_tlock(lid);
698 assert(tlck->type & tlckXTREE);
699 tlck->type |= tlckBTROOT;
700 tlck->mp = mp;
701 ilinelock = (struct linelock *) & tlck->lock;
702
703 /*
704 * copy xtree root from inode to dinode:
705 */
706 p = &jfs_ip->i_xtroot;
707 xp = (xtpage_t *) &dp->di_dirtable;
708 lv = ilinelock->lv;
709 for (n = 0; n < ilinelock->index; n++, lv++) {
710 memcpy(&xp->xad[lv->offset], &p->xad[lv->offset],
711 lv->length << L2XTSLOTSIZE);
712 }
713
714 /* reset on-disk (metadata page) xtree XAD_NEW bit */
715 xad = &xp->xad[XTENTRYSTART];
716 for (n = XTENTRYSTART;
717 n < le16_to_cpu(xp->header.nextindex); n++, xad++)
718 if (xad->flag & (XAD_NEW | XAD_EXTENDED))
719 xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
720 }
721
722 if ((lid = jfs_ip->blid) == 0)
723 goto inlineData;
724 jfs_ip->blid = 0;
725
726 tlck = lid_to_tlock(lid);
727 type = tlck->type;
728 tlck->type |= tlckBTROOT;
729 tlck->mp = mp;
730 ilinelock = (struct linelock *) & tlck->lock;
731
732 /*
733 * regular file: 16 byte (XAD slot) granularity
734 */
735 if (type & tlckXTREE) {
736 xtpage_t *p, *xp;
737 xad_t *xad;
738
739 /*
740 * copy xtree root from inode to dinode:
741 */
742 p = &jfs_ip->i_xtroot;
743 xp = &dp->di_xtroot;
744 lv = ilinelock->lv;
745 for (n = 0; n < ilinelock->index; n++, lv++) {
746 memcpy(&xp->xad[lv->offset], &p->xad[lv->offset],
747 lv->length << L2XTSLOTSIZE);
748 }
749
750 /* reset on-disk (metadata page) xtree XAD_NEW bit */
751 xad = &xp->xad[XTENTRYSTART];
752 for (n = XTENTRYSTART;
753 n < le16_to_cpu(xp->header.nextindex); n++, xad++)
754 if (xad->flag & (XAD_NEW | XAD_EXTENDED))
755 xad->flag &= ~(XAD_NEW | XAD_EXTENDED);
756 }
757 /*
758 * directory: 32 byte (directory entry slot) granularity
759 */
760 else if (type & tlckDTREE) {
761 dtpage_t *p, *xp;
762
763 /*
764 * copy dtree root from inode to dinode:
765 */
766 p = (dtpage_t *) &jfs_ip->i_dtroot;
767 xp = (dtpage_t *) & dp->di_dtroot;
768 lv = ilinelock->lv;
769 for (n = 0; n < ilinelock->index; n++, lv++) {
770 memcpy(&xp->slot[lv->offset], &p->slot[lv->offset],
771 lv->length << L2DTSLOTSIZE);
772 }
773 } else {
774 jfs_err("diWrite: UFO tlock");
775 }
776
777 inlineData:
778 /*
779 * copy inline symlink from in-memory inode to on-disk inode
780 */
781 if (S_ISLNK(ip->i_mode) && ip->i_size < IDATASIZE) {
782 lv = & dilinelock->lv[dilinelock->index];
783 lv->offset = (dioffset + 2 * 128) >> L2INODESLOTSIZE;
784 lv->length = 2;
785 memcpy(&dp->di_fastsymlink, jfs_ip->i_inline, IDATASIZE);
786 dilinelock->index++;
787 }
788 /*
789 * copy inline data from in-memory inode to on-disk inode:
790 * 128 byte slot granularity
791 */
792 if (test_cflag(COMMIT_Inlineea, ip)) {
793 lv = & dilinelock->lv[dilinelock->index];
794 lv->offset = (dioffset + 3 * 128) >> L2INODESLOTSIZE;
795 lv->length = 1;
796 memcpy(&dp->di_inlineea, jfs_ip->i_inline_ea, INODESLOTSIZE);
797 dilinelock->index++;
798
799 clear_cflag(COMMIT_Inlineea, ip);
800 }
801
802 /*
803 * lock/copy inode base: 128 byte slot granularity
804 */
805// baseDinode:
806 lv = & dilinelock->lv[dilinelock->index];
807 lv->offset = dioffset >> L2INODESLOTSIZE;
808 copy_to_dinode(dp, ip);
809 if (test_and_clear_cflag(COMMIT_Dirtable, ip)) {
810 lv->length = 2;
811 memcpy(&dp->di_dirtable, &jfs_ip->i_dirtable, 96);
812 } else
813 lv->length = 1;
814 dilinelock->index++;
815
816#ifdef _JFS_FASTDASD
817 /*
818 * We aren't logging changes to the DASD used in directory inodes,
819 * but we need to write them to disk. If we don't unmount cleanly,
820 * mount will recalculate the DASD used.
821 */
822 if (S_ISDIR(ip->i_mode)
823 && (ip->i_ipmnt->i_mntflag & JFS_DASD_ENABLED))
824 memcpy(&dp->di_DASD, &ip->i_DASD, sizeof(struct dasd));
825#endif /* _JFS_FASTDASD */
826
827 /* release the buffer holding the updated on-disk inode.
828 * the buffer will be later written by commit processing.
829 */
830 write_metapage(mp);
831
832 return (rc);
833}
834
835
836/*
837 * NAME: diFree(ip)
838 *
839 * FUNCTION: free a specified inode from the inode working map
840 * for a fileset or aggregate.
841 *
842 * if the inode to be freed represents the first (only)
843 * free inode within the iag, the iag will be placed on
844 * the ag free inode list.
845 *
846 * freeing the inode will cause the inode extent to be
847 * freed if the inode is the only allocated inode within
848 * the extent. in this case all the disk resource backing
849 * up the inode extent will be freed. in addition, the iag
850 * will be placed on the ag extent free list if the extent
851 * is the first free extent in the iag. if freeing the
852 * extent also means that no free inodes will exist for
853 * the iag, the iag will also be removed from the ag free
854 * inode list.
855 *
856 * the iag describing the inode will be freed if the extent
857 * is to be freed and it is the only backed extent within
858 * the iag. in this case, the iag will be removed from the
859 * ag free extent list and ag free inode list and placed on
860 * the inode map's free iag list.
861 *
862 * a careful update approach is used to provide consistency
863 * in the face of updates to multiple buffers. under this
864 * approach, all required buffers are obtained before making
865 * any updates and are held until all updates are complete.
866 *
867 * PARAMETERS:
868 * ip - inode to be freed.
869 *
870 * RETURN VALUES:
871 * 0 - success
872 * -EIO - i/o error.
873 */
874int diFree(struct inode *ip)
875{
876 int rc;
877 ino_t inum = ip->i_ino;
878 struct iag *iagp, *aiagp, *biagp, *ciagp, *diagp;
879 struct metapage *mp, *amp, *bmp, *cmp, *dmp;
880 int iagno, ino, extno, bitno, sword, agno;
881 int back, fwd;
882 u32 bitmap, mask;
883 struct inode *ipimap = JFS_SBI(ip->i_sb)->ipimap;
884 struct inomap *imap = JFS_IP(ipimap)->i_imap;
885 pxd_t freepxd;
886 tid_t tid;
887 struct inode *iplist[3];
888 struct tlock *tlck;
889 struct pxd_lock *pxdlock;
890
891 /*
892 * This is just to suppress compiler warnings. The same logic that
893 * references these variables is used to initialize them.
894 */
895 aiagp = biagp = ciagp = diagp = NULL;
896
897 /* get the iag number containing the inode.
898 */
899 iagno = INOTOIAG(inum);
900
901 /* make sure that the iag is contained within
902 * the map.
903 */
904 if (iagno >= imap->im_nextiag) {
905 dump_mem("imap", imap, 32);
906 jfs_error(ip->i_sb,
907 "diFree: inum = %d, iagno = %d, nextiag = %d",
908 (uint) inum, iagno, imap->im_nextiag);
909 return -EIO;
910 }
911
912 /* get the allocation group for this ino.
913 */
914 agno = JFS_IP(ip)->agno;
915
916 /* Lock the AG specific inode map information
917 */
918 AG_LOCK(imap, agno);
919
920 /* Obtain read lock in imap inode. Don't release it until we have
921 * read all of the IAG's that we are going to.
922 */
923 IREAD_LOCK(ipimap);
924
925 /* read the iag.
926 */
927 if ((rc = diIAGRead(imap, iagno, &mp))) {
928 IREAD_UNLOCK(ipimap);
929 AG_UNLOCK(imap, agno);
930 return (rc);
931 }
932 iagp = (struct iag *) mp->data;
933
934 /* get the inode number and extent number of the inode within
935 * the iag and the inode number within the extent.
936 */
937 ino = inum & (INOSPERIAG - 1);
938 extno = ino >> L2INOSPEREXT;
939 bitno = ino & (INOSPEREXT - 1);
940 mask = HIGHORDER >> bitno;
941
942 if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
943 jfs_error(ip->i_sb,
944 "diFree: wmap shows inode already free");
945 }
946
947 if (!addressPXD(&iagp->inoext[extno])) {
948 release_metapage(mp);
949 IREAD_UNLOCK(ipimap);
950 AG_UNLOCK(imap, agno);
951 jfs_error(ip->i_sb, "diFree: invalid inoext");
952 return -EIO;
953 }
954
955 /* compute the bitmap for the extent reflecting the freed inode.
956 */
957 bitmap = le32_to_cpu(iagp->wmap[extno]) & ~mask;
958
959 if (imap->im_agctl[agno].numfree > imap->im_agctl[agno].numinos) {
960 release_metapage(mp);
961 IREAD_UNLOCK(ipimap);
962 AG_UNLOCK(imap, agno);
963 jfs_error(ip->i_sb, "diFree: numfree > numinos");
964 return -EIO;
965 }
966 /*
967 * inode extent still has some inodes or below low water mark:
968 * keep the inode extent;
969 */
970 if (bitmap ||
971 imap->im_agctl[agno].numfree < 96 ||
972 (imap->im_agctl[agno].numfree < 288 &&
973 (((imap->im_agctl[agno].numfree * 100) /
974 imap->im_agctl[agno].numinos) <= 25))) {
975 /* if the iag currently has no free inodes (i.e.,
976 * the inode being freed is the first free inode of iag),
977 * insert the iag at head of the inode free list for the ag.
978 */
979 if (iagp->nfreeinos == 0) {
980 /* check if there are any iags on the ag inode
981 * free list. if so, read the first one so that
982 * we can link the current iag onto the list at
983 * the head.
984 */
985 if ((fwd = imap->im_agctl[agno].inofree) >= 0) {
986 /* read the iag that currently is the head
987 * of the list.
988 */
989 if ((rc = diIAGRead(imap, fwd, &amp))) {
990 IREAD_UNLOCK(ipimap);
991 AG_UNLOCK(imap, agno);
992 release_metapage(mp);
993 return (rc);
994 }
995 aiagp = (struct iag *) amp->data;
996
997 /* make current head point back to the iag.
998 */
999 aiagp->inofreeback = cpu_to_le32(iagno);
1000
1001 write_metapage(amp);
1002 }
1003
1004 /* iag points forward to current head and iag
1005 * becomes the new head of the list.
1006 */
1007 iagp->inofreefwd =
1008 cpu_to_le32(imap->im_agctl[agno].inofree);
1009 iagp->inofreeback = cpu_to_le32(-1);
1010 imap->im_agctl[agno].inofree = iagno;
1011 }
1012 IREAD_UNLOCK(ipimap);
1013
1014 /* update the free inode summary map for the extent if
1015 * freeing the inode means the extent will now have free
1016 * inodes (i.e., the inode being freed is the first free
1017 * inode of extent),
1018 */
1019 if (iagp->wmap[extno] == cpu_to_le32(ONES)) {
1020 sword = extno >> L2EXTSPERSUM;
1021 bitno = extno & (EXTSPERSUM - 1);
1022 iagp->inosmap[sword] &=
1023 cpu_to_le32(~(HIGHORDER >> bitno));
1024 }
1025
1026 /* update the bitmap.
1027 */
1028 iagp->wmap[extno] = cpu_to_le32(bitmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029
1030 /* update the free inode counts at the iag, ag and
1031 * map level.
1032 */
1033 iagp->nfreeinos =
1034 cpu_to_le32(le32_to_cpu(iagp->nfreeinos) + 1);
1035 imap->im_agctl[agno].numfree += 1;
1036 atomic_inc(&imap->im_numfree);
1037
1038 /* release the AG inode map lock
1039 */
1040 AG_UNLOCK(imap, agno);
1041
1042 /* write the iag */
1043 write_metapage(mp);
1044
1045 return (0);
1046 }
1047
1048
1049 /*
1050 * inode extent has become free and above low water mark:
1051 * free the inode extent;
1052 */
1053
1054 /*
1055 * prepare to update iag list(s) (careful update step 1)
1056 */
1057 amp = bmp = cmp = dmp = NULL;
1058 fwd = back = -1;
1059
1060 /* check if the iag currently has no free extents. if so,
1061 * it will be placed on the head of the ag extent free list.
1062 */
1063 if (iagp->nfreeexts == 0) {
1064 /* check if the ag extent free list has any iags.
1065 * if so, read the iag at the head of the list now.
1066 * this (head) iag will be updated later to reflect
1067 * the addition of the current iag at the head of
1068 * the list.
1069 */
1070 if ((fwd = imap->im_agctl[agno].extfree) >= 0) {
1071 if ((rc = diIAGRead(imap, fwd, &amp)))
1072 goto error_out;
1073 aiagp = (struct iag *) amp->data;
1074 }
1075 } else {
1076 /* iag has free extents. check if the addition of a free
1077 * extent will cause all extents to be free within this
1078 * iag. if so, the iag will be removed from the ag extent
1079 * free list and placed on the inode map's free iag list.
1080 */
1081 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) {
1082 /* in preparation for removing the iag from the
1083 * ag extent free list, read the iags preceeding
1084 * and following the iag on the ag extent free
1085 * list.
1086 */
1087 if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) {
1088 if ((rc = diIAGRead(imap, fwd, &amp)))
1089 goto error_out;
1090 aiagp = (struct iag *) amp->data;
1091 }
1092
1093 if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) {
1094 if ((rc = diIAGRead(imap, back, &bmp)))
1095 goto error_out;
1096 biagp = (struct iag *) bmp->data;
1097 }
1098 }
1099 }
1100
1101 /* remove the iag from the ag inode free list if freeing
1102 * this extent cause the iag to have no free inodes.
1103 */
1104 if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) {
1105 int inofreeback = le32_to_cpu(iagp->inofreeback);
1106 int inofreefwd = le32_to_cpu(iagp->inofreefwd);
1107
1108 /* in preparation for removing the iag from the
1109 * ag inode free list, read the iags preceeding
1110 * and following the iag on the ag inode free
1111 * list. before reading these iags, we must make
1112 * sure that we already don't have them in hand
1113 * from up above, since re-reading an iag (buffer)
1114 * we are currently holding would cause a deadlock.
1115 */
1116 if (inofreefwd >= 0) {
1117
1118 if (inofreefwd == fwd)
1119 ciagp = (struct iag *) amp->data;
1120 else if (inofreefwd == back)
1121 ciagp = (struct iag *) bmp->data;
1122 else {
1123 if ((rc =
1124 diIAGRead(imap, inofreefwd, &cmp)))
1125 goto error_out;
1126 ciagp = (struct iag *) cmp->data;
1127 }
1128 assert(ciagp != NULL);
1129 }
1130
1131 if (inofreeback >= 0) {
1132 if (inofreeback == fwd)
1133 diagp = (struct iag *) amp->data;
1134 else if (inofreeback == back)
1135 diagp = (struct iag *) bmp->data;
1136 else {
1137 if ((rc =
1138 diIAGRead(imap, inofreeback, &dmp)))
1139 goto error_out;
1140 diagp = (struct iag *) dmp->data;
1141 }
1142 assert(diagp != NULL);
1143 }
1144 }
1145
1146 IREAD_UNLOCK(ipimap);
1147
1148 /*
1149 * invalidate any page of the inode extent freed from buffer cache;
1150 */
1151 freepxd = iagp->inoext[extno];
1152 invalidate_pxd_metapages(ip, freepxd);
1153
1154 /*
1155 * update iag list(s) (careful update step 2)
1156 */
1157 /* add the iag to the ag extent free list if this is the
1158 * first free extent for the iag.
1159 */
1160 if (iagp->nfreeexts == 0) {
1161 if (fwd >= 0)
1162 aiagp->extfreeback = cpu_to_le32(iagno);
1163
1164 iagp->extfreefwd =
1165 cpu_to_le32(imap->im_agctl[agno].extfree);
1166 iagp->extfreeback = cpu_to_le32(-1);
1167 imap->im_agctl[agno].extfree = iagno;
1168 } else {
1169 /* remove the iag from the ag extent list if all extents
1170 * are now free and place it on the inode map iag free list.
1171 */
1172 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) {
1173 if (fwd >= 0)
1174 aiagp->extfreeback = iagp->extfreeback;
1175
1176 if (back >= 0)
1177 biagp->extfreefwd = iagp->extfreefwd;
1178 else
1179 imap->im_agctl[agno].extfree =
1180 le32_to_cpu(iagp->extfreefwd);
1181
1182 iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1);
1183
1184 IAGFREE_LOCK(imap);
1185 iagp->iagfree = cpu_to_le32(imap->im_freeiag);
1186 imap->im_freeiag = iagno;
1187 IAGFREE_UNLOCK(imap);
1188 }
1189 }
1190
1191 /* remove the iag from the ag inode free list if freeing
1192 * this extent causes the iag to have no free inodes.
1193 */
1194 if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) {
1195 if ((int) le32_to_cpu(iagp->inofreefwd) >= 0)
1196 ciagp->inofreeback = iagp->inofreeback;
1197
1198 if ((int) le32_to_cpu(iagp->inofreeback) >= 0)
1199 diagp->inofreefwd = iagp->inofreefwd;
1200 else
1201 imap->im_agctl[agno].inofree =
1202 le32_to_cpu(iagp->inofreefwd);
1203
1204 iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1);
1205 }
1206
1207 /* update the inode extent address and working map
1208 * to reflect the free extent.
1209 * the permanent map should have been updated already
1210 * for the inode being freed.
1211 */
1212 if (iagp->pmap[extno] != 0) {
1213 jfs_error(ip->i_sb, "diFree: the pmap does not show inode free");
1214 }
1215 iagp->wmap[extno] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 PXDlength(&iagp->inoext[extno], 0);
1217 PXDaddress(&iagp->inoext[extno], 0);
1218
1219 /* update the free extent and free inode summary maps
1220 * to reflect the freed extent.
1221 * the inode summary map is marked to indicate no inodes
1222 * available for the freed extent.
1223 */
1224 sword = extno >> L2EXTSPERSUM;
1225 bitno = extno & (EXTSPERSUM - 1);
1226 mask = HIGHORDER >> bitno;
1227 iagp->inosmap[sword] |= cpu_to_le32(mask);
1228 iagp->extsmap[sword] &= cpu_to_le32(~mask);
1229
1230 /* update the number of free inodes and number of free extents
1231 * for the iag.
1232 */
1233 iagp->nfreeinos = cpu_to_le32(le32_to_cpu(iagp->nfreeinos) -
1234 (INOSPEREXT - 1));
1235 iagp->nfreeexts = cpu_to_le32(le32_to_cpu(iagp->nfreeexts) + 1);
1236
1237 /* update the number of free inodes and backed inodes
1238 * at the ag and inode map level.
1239 */
1240 imap->im_agctl[agno].numfree -= (INOSPEREXT - 1);
1241 imap->im_agctl[agno].numinos -= INOSPEREXT;
1242 atomic_sub(INOSPEREXT - 1, &imap->im_numfree);
1243 atomic_sub(INOSPEREXT, &imap->im_numinos);
1244
1245 if (amp)
1246 write_metapage(amp);
1247 if (bmp)
1248 write_metapage(bmp);
1249 if (cmp)
1250 write_metapage(cmp);
1251 if (dmp)
1252 write_metapage(dmp);
1253
1254 /*
1255 * start transaction to update block allocation map
1256 * for the inode extent freed;
1257 *
1258 * N.B. AG_LOCK is released and iag will be released below, and
1259 * other thread may allocate inode from/reusing the ixad freed
1260 * BUT with new/different backing inode extent from the extent
1261 * to be freed by the transaction;
1262 */
1263 tid = txBegin(ipimap->i_sb, COMMIT_FORCE);
Ingo Molnar1de87442006-01-24 15:22:50 -06001264 mutex_lock(&JFS_IP(ipimap)->commit_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265
1266 /* acquire tlock of the iag page of the freed ixad
1267 * to force the page NOHOMEOK (even though no data is
1268 * logged from the iag page) until NOREDOPAGE|FREEXTENT log
1269 * for the free of the extent is committed;
1270 * write FREEXTENT|NOREDOPAGE log record
1271 * N.B. linelock is overlaid as freed extent descriptor;
1272 */
1273 tlck = txLock(tid, ipimap, mp, tlckINODE | tlckFREE);
1274 pxdlock = (struct pxd_lock *) & tlck->lock;
1275 pxdlock->flag = mlckFREEPXD;
1276 pxdlock->pxd = freepxd;
1277 pxdlock->index = 1;
1278
1279 write_metapage(mp);
1280
1281 iplist[0] = ipimap;
1282
1283 /*
1284 * logredo needs the IAG number and IAG extent index in order
1285 * to ensure that the IMap is consistent. The least disruptive
1286 * way to pass these values through to the transaction manager
1287 * is in the iplist array.
1288 *
1289 * It's not pretty, but it works.
1290 */
1291 iplist[1] = (struct inode *) (size_t)iagno;
1292 iplist[2] = (struct inode *) (size_t)extno;
1293
1294 rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE);
1295
1296 txEnd(tid);
Ingo Molnar1de87442006-01-24 15:22:50 -06001297 mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298
1299 /* unlock the AG inode map information */
1300 AG_UNLOCK(imap, agno);
1301
1302 return (0);
1303
1304 error_out:
1305 IREAD_UNLOCK(ipimap);
1306
1307 if (amp)
1308 release_metapage(amp);
1309 if (bmp)
1310 release_metapage(bmp);
1311 if (cmp)
1312 release_metapage(cmp);
1313 if (dmp)
1314 release_metapage(dmp);
1315
1316 AG_UNLOCK(imap, agno);
1317
1318 release_metapage(mp);
1319
1320 return (rc);
1321}
1322
1323/*
1324 * There are several places in the diAlloc* routines where we initialize
1325 * the inode.
1326 */
1327static inline void
1328diInitInode(struct inode *ip, int iagno, int ino, int extno, struct iag * iagp)
1329{
1330 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
1331 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
1332
1333 ip->i_ino = (iagno << L2INOSPERIAG) + ino;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 jfs_ip->ixpxd = iagp->inoext[extno];
1335 jfs_ip->agno = BLKTOAG(le64_to_cpu(iagp->agstart), sbi);
1336 jfs_ip->active_ag = -1;
1337}
1338
1339
1340/*
1341 * NAME: diAlloc(pip,dir,ip)
1342 *
1343 * FUNCTION: allocate a disk inode from the inode working map
1344 * for a fileset or aggregate.
1345 *
1346 * PARAMETERS:
1347 * pip - pointer to incore inode for the parent inode.
1348 * dir - TRUE if the new disk inode is for a directory.
1349 * ip - pointer to a new inode
1350 *
1351 * RETURN VALUES:
1352 * 0 - success.
1353 * -ENOSPC - insufficient disk resources.
1354 * -EIO - i/o error.
1355 */
1356int diAlloc(struct inode *pip, boolean_t dir, struct inode *ip)
1357{
1358 int rc, ino, iagno, addext, extno, bitno, sword;
1359 int nwords, rem, i, agno;
1360 u32 mask, inosmap, extsmap;
1361 struct inode *ipimap;
1362 struct metapage *mp;
1363 ino_t inum;
1364 struct iag *iagp;
1365 struct inomap *imap;
1366
1367 /* get the pointers to the inode map inode and the
1368 * corresponding imap control structure.
1369 */
1370 ipimap = JFS_SBI(pip->i_sb)->ipimap;
1371 imap = JFS_IP(ipimap)->i_imap;
1372 JFS_IP(ip)->ipimap = ipimap;
1373 JFS_IP(ip)->fileset = FILESYSTEM_I;
1374
1375 /* for a directory, the allocation policy is to start
1376 * at the ag level using the preferred ag.
1377 */
1378 if (dir == TRUE) {
1379 agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap);
1380 AG_LOCK(imap, agno);
1381 goto tryag;
1382 }
1383
1384 /* for files, the policy starts off by trying to allocate from
1385 * the same iag containing the parent disk inode:
1386 * try to allocate the new disk inode close to the parent disk
1387 * inode, using parent disk inode number + 1 as the allocation
1388 * hint. (we use a left-to-right policy to attempt to avoid
1389 * moving backward on the disk.) compute the hint within the
1390 * file system and the iag.
1391 */
1392
1393 /* get the ag number of this iag */
1394 agno = JFS_IP(pip)->agno;
1395
1396 if (atomic_read(&JFS_SBI(pip->i_sb)->bmap->db_active[agno])) {
1397 /*
1398 * There is an open file actively growing. We want to
1399 * allocate new inodes from a different ag to avoid
1400 * fragmentation problems.
1401 */
1402 agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap);
1403 AG_LOCK(imap, agno);
1404 goto tryag;
1405 }
1406
1407 inum = pip->i_ino + 1;
1408 ino = inum & (INOSPERIAG - 1);
1409
1410 /* back off the the hint if it is outside of the iag */
1411 if (ino == 0)
1412 inum = pip->i_ino;
1413
1414 /* lock the AG inode map information */
1415 AG_LOCK(imap, agno);
1416
1417 /* Get read lock on imap inode */
1418 IREAD_LOCK(ipimap);
1419
1420 /* get the iag number and read the iag */
1421 iagno = INOTOIAG(inum);
1422 if ((rc = diIAGRead(imap, iagno, &mp))) {
1423 IREAD_UNLOCK(ipimap);
1424 AG_UNLOCK(imap, agno);
1425 return (rc);
1426 }
1427 iagp = (struct iag *) mp->data;
1428
1429 /* determine if new inode extent is allowed to be added to the iag.
1430 * new inode extent can be added to the iag if the ag
1431 * has less than 32 free disk inodes and the iag has free extents.
1432 */
1433 addext = (imap->im_agctl[agno].numfree < 32 && iagp->nfreeexts);
1434
1435 /*
1436 * try to allocate from the IAG
1437 */
1438 /* check if the inode may be allocated from the iag
1439 * (i.e. the inode has free inodes or new extent can be added).
1440 */
1441 if (iagp->nfreeinos || addext) {
1442 /* determine the extent number of the hint.
1443 */
1444 extno = ino >> L2INOSPEREXT;
1445
1446 /* check if the extent containing the hint has backed
1447 * inodes. if so, try to allocate within this extent.
1448 */
1449 if (addressPXD(&iagp->inoext[extno])) {
1450 bitno = ino & (INOSPEREXT - 1);
1451 if ((bitno =
1452 diFindFree(le32_to_cpu(iagp->wmap[extno]),
1453 bitno))
1454 < INOSPEREXT) {
1455 ino = (extno << L2INOSPEREXT) + bitno;
1456
1457 /* a free inode (bit) was found within this
1458 * extent, so allocate it.
1459 */
1460 rc = diAllocBit(imap, iagp, ino);
1461 IREAD_UNLOCK(ipimap);
1462 if (rc) {
1463 assert(rc == -EIO);
1464 } else {
1465 /* set the results of the allocation
1466 * and write the iag.
1467 */
1468 diInitInode(ip, iagno, ino, extno,
1469 iagp);
1470 mark_metapage_dirty(mp);
1471 }
1472 release_metapage(mp);
1473
1474 /* free the AG lock and return.
1475 */
1476 AG_UNLOCK(imap, agno);
1477 return (rc);
1478 }
1479
1480 if (!addext)
1481 extno =
1482 (extno ==
1483 EXTSPERIAG - 1) ? 0 : extno + 1;
1484 }
1485
1486 /*
1487 * no free inodes within the extent containing the hint.
1488 *
1489 * try to allocate from the backed extents following
1490 * hint or, if appropriate (i.e. addext is true), allocate
1491 * an extent of free inodes at or following the extent
1492 * containing the hint.
1493 *
1494 * the free inode and free extent summary maps are used
1495 * here, so determine the starting summary map position
1496 * and the number of words we'll have to examine. again,
1497 * the approach is to allocate following the hint, so we
1498 * might have to initially ignore prior bits of the summary
1499 * map that represent extents prior to the extent containing
1500 * the hint and later revisit these bits.
1501 */
1502 bitno = extno & (EXTSPERSUM - 1);
1503 nwords = (bitno == 0) ? SMAPSZ : SMAPSZ + 1;
1504 sword = extno >> L2EXTSPERSUM;
1505
1506 /* mask any prior bits for the starting words of the
1507 * summary map.
1508 */
1509 mask = ONES << (EXTSPERSUM - bitno);
1510 inosmap = le32_to_cpu(iagp->inosmap[sword]) | mask;
1511 extsmap = le32_to_cpu(iagp->extsmap[sword]) | mask;
1512
1513 /* scan the free inode and free extent summary maps for
1514 * free resources.
1515 */
1516 for (i = 0; i < nwords; i++) {
1517 /* check if this word of the free inode summary
1518 * map describes an extent with free inodes.
1519 */
1520 if (~inosmap) {
1521 /* an extent with free inodes has been
1522 * found. determine the extent number
1523 * and the inode number within the extent.
1524 */
1525 rem = diFindFree(inosmap, 0);
1526 extno = (sword << L2EXTSPERSUM) + rem;
1527 rem = diFindFree(le32_to_cpu(iagp->wmap[extno]),
1528 0);
1529 if (rem >= INOSPEREXT) {
1530 IREAD_UNLOCK(ipimap);
1531 release_metapage(mp);
1532 AG_UNLOCK(imap, agno);
1533 jfs_error(ip->i_sb,
1534 "diAlloc: can't find free bit "
1535 "in wmap");
1536 return EIO;
1537 }
1538
1539 /* determine the inode number within the
1540 * iag and allocate the inode from the
1541 * map.
1542 */
1543 ino = (extno << L2INOSPEREXT) + rem;
1544 rc = diAllocBit(imap, iagp, ino);
1545 IREAD_UNLOCK(ipimap);
1546 if (rc)
1547 assert(rc == -EIO);
1548 else {
1549 /* set the results of the allocation
1550 * and write the iag.
1551 */
1552 diInitInode(ip, iagno, ino, extno,
1553 iagp);
1554 mark_metapage_dirty(mp);
1555 }
1556 release_metapage(mp);
1557
1558 /* free the AG lock and return.
1559 */
1560 AG_UNLOCK(imap, agno);
1561 return (rc);
1562
1563 }
1564
1565 /* check if we may allocate an extent of free
1566 * inodes and whether this word of the free
1567 * extents summary map describes a free extent.
1568 */
1569 if (addext && ~extsmap) {
1570 /* a free extent has been found. determine
1571 * the extent number.
1572 */
1573 rem = diFindFree(extsmap, 0);
1574 extno = (sword << L2EXTSPERSUM) + rem;
1575
1576 /* allocate an extent of free inodes.
1577 */
1578 if ((rc = diNewExt(imap, iagp, extno))) {
1579 /* if there is no disk space for a
1580 * new extent, try to allocate the
1581 * disk inode from somewhere else.
1582 */
1583 if (rc == -ENOSPC)
1584 break;
1585
1586 assert(rc == -EIO);
1587 } else {
1588 /* set the results of the allocation
1589 * and write the iag.
1590 */
1591 diInitInode(ip, iagno,
1592 extno << L2INOSPEREXT,
1593 extno, iagp);
1594 mark_metapage_dirty(mp);
1595 }
1596 release_metapage(mp);
1597 /* free the imap inode & the AG lock & return.
1598 */
1599 IREAD_UNLOCK(ipimap);
1600 AG_UNLOCK(imap, agno);
1601 return (rc);
1602 }
1603
1604 /* move on to the next set of summary map words.
1605 */
1606 sword = (sword == SMAPSZ - 1) ? 0 : sword + 1;
1607 inosmap = le32_to_cpu(iagp->inosmap[sword]);
1608 extsmap = le32_to_cpu(iagp->extsmap[sword]);
1609 }
1610 }
1611 /* unlock imap inode */
1612 IREAD_UNLOCK(ipimap);
1613
1614 /* nothing doing in this iag, so release it. */
1615 release_metapage(mp);
1616
1617 tryag:
1618 /*
1619 * try to allocate anywhere within the same AG as the parent inode.
1620 */
1621 rc = diAllocAG(imap, agno, dir, ip);
1622
1623 AG_UNLOCK(imap, agno);
1624
1625 if (rc != -ENOSPC)
1626 return (rc);
1627
1628 /*
1629 * try to allocate in any AG.
1630 */
1631 return (diAllocAny(imap, agno, dir, ip));
1632}
1633
1634
1635/*
1636 * NAME: diAllocAG(imap,agno,dir,ip)
1637 *
1638 * FUNCTION: allocate a disk inode from the allocation group.
1639 *
1640 * this routine first determines if a new extent of free
1641 * inodes should be added for the allocation group, with
1642 * the current request satisfied from this extent. if this
1643 * is the case, an attempt will be made to do just that. if
1644 * this attempt fails or it has been determined that a new
1645 * extent should not be added, an attempt is made to satisfy
1646 * the request by allocating an existing (backed) free inode
1647 * from the allocation group.
1648 *
1649 * PRE CONDITION: Already have the AG lock for this AG.
1650 *
1651 * PARAMETERS:
1652 * imap - pointer to inode map control structure.
1653 * agno - allocation group to allocate from.
1654 * dir - TRUE if the new disk inode is for a directory.
1655 * ip - pointer to the new inode to be filled in on successful return
1656 * with the disk inode number allocated, its extent address
1657 * and the start of the ag.
1658 *
1659 * RETURN VALUES:
1660 * 0 - success.
1661 * -ENOSPC - insufficient disk resources.
1662 * -EIO - i/o error.
1663 */
1664static int
1665diAllocAG(struct inomap * imap, int agno, boolean_t dir, struct inode *ip)
1666{
1667 int rc, addext, numfree, numinos;
1668
1669 /* get the number of free and the number of backed disk
1670 * inodes currently within the ag.
1671 */
1672 numfree = imap->im_agctl[agno].numfree;
1673 numinos = imap->im_agctl[agno].numinos;
1674
1675 if (numfree > numinos) {
1676 jfs_error(ip->i_sb, "diAllocAG: numfree > numinos");
1677 return -EIO;
1678 }
1679
1680 /* determine if we should allocate a new extent of free inodes
1681 * within the ag: for directory inodes, add a new extent
1682 * if there are a small number of free inodes or number of free
1683 * inodes is a small percentage of the number of backed inodes.
1684 */
1685 if (dir == TRUE)
1686 addext = (numfree < 64 ||
1687 (numfree < 256
1688 && ((numfree * 100) / numinos) <= 20));
1689 else
1690 addext = (numfree == 0);
1691
1692 /*
1693 * try to allocate a new extent of free inodes.
1694 */
1695 if (addext) {
1696 /* if free space is not avaliable for this new extent, try
1697 * below to allocate a free and existing (already backed)
1698 * inode from the ag.
1699 */
1700 if ((rc = diAllocExt(imap, agno, ip)) != -ENOSPC)
1701 return (rc);
1702 }
1703
1704 /*
1705 * try to allocate an existing free inode from the ag.
1706 */
1707 return (diAllocIno(imap, agno, ip));
1708}
1709
1710
1711/*
1712 * NAME: diAllocAny(imap,agno,dir,iap)
1713 *
1714 * FUNCTION: allocate a disk inode from any other allocation group.
1715 *
1716 * this routine is called when an allocation attempt within
1717 * the primary allocation group has failed. if attempts to
1718 * allocate an inode from any allocation group other than the
1719 * specified primary group.
1720 *
1721 * PARAMETERS:
1722 * imap - pointer to inode map control structure.
1723 * agno - primary allocation group (to avoid).
1724 * dir - TRUE if the new disk inode is for a directory.
1725 * ip - pointer to a new inode to be filled in on successful return
1726 * with the disk inode number allocated, its extent address
1727 * and the start of the ag.
1728 *
1729 * RETURN VALUES:
1730 * 0 - success.
1731 * -ENOSPC - insufficient disk resources.
1732 * -EIO - i/o error.
1733 */
1734static int
1735diAllocAny(struct inomap * imap, int agno, boolean_t dir, struct inode *ip)
1736{
1737 int ag, rc;
1738 int maxag = JFS_SBI(imap->im_ipimap->i_sb)->bmap->db_maxag;
1739
1740
1741 /* try to allocate from the ags following agno up to
1742 * the maximum ag number.
1743 */
1744 for (ag = agno + 1; ag <= maxag; ag++) {
1745 AG_LOCK(imap, ag);
1746
1747 rc = diAllocAG(imap, ag, dir, ip);
1748
1749 AG_UNLOCK(imap, ag);
1750
1751 if (rc != -ENOSPC)
1752 return (rc);
1753 }
1754
1755 /* try to allocate from the ags in front of agno.
1756 */
1757 for (ag = 0; ag < agno; ag++) {
1758 AG_LOCK(imap, ag);
1759
1760 rc = diAllocAG(imap, ag, dir, ip);
1761
1762 AG_UNLOCK(imap, ag);
1763
1764 if (rc != -ENOSPC)
1765 return (rc);
1766 }
1767
1768 /* no free disk inodes.
1769 */
1770 return -ENOSPC;
1771}
1772
1773
1774/*
1775 * NAME: diAllocIno(imap,agno,ip)
1776 *
1777 * FUNCTION: allocate a disk inode from the allocation group's free
1778 * inode list, returning an error if this free list is
1779 * empty (i.e. no iags on the list).
1780 *
1781 * allocation occurs from the first iag on the list using
1782 * the iag's free inode summary map to find the leftmost
1783 * free inode in the iag.
1784 *
1785 * PRE CONDITION: Already have AG lock for this AG.
1786 *
1787 * PARAMETERS:
1788 * imap - pointer to inode map control structure.
1789 * agno - allocation group.
1790 * ip - pointer to new inode to be filled in on successful return
1791 * with the disk inode number allocated, its extent address
1792 * and the start of the ag.
1793 *
1794 * RETURN VALUES:
1795 * 0 - success.
1796 * -ENOSPC - insufficient disk resources.
1797 * -EIO - i/o error.
1798 */
1799static int diAllocIno(struct inomap * imap, int agno, struct inode *ip)
1800{
1801 int iagno, ino, rc, rem, extno, sword;
1802 struct metapage *mp;
1803 struct iag *iagp;
1804
1805 /* check if there are iags on the ag's free inode list.
1806 */
1807 if ((iagno = imap->im_agctl[agno].inofree) < 0)
1808 return -ENOSPC;
1809
1810 /* obtain read lock on imap inode */
1811 IREAD_LOCK(imap->im_ipimap);
1812
1813 /* read the iag at the head of the list.
1814 */
1815 if ((rc = diIAGRead(imap, iagno, &mp))) {
1816 IREAD_UNLOCK(imap->im_ipimap);
1817 return (rc);
1818 }
1819 iagp = (struct iag *) mp->data;
1820
1821 /* better be free inodes in this iag if it is on the
1822 * list.
1823 */
1824 if (!iagp->nfreeinos) {
1825 IREAD_UNLOCK(imap->im_ipimap);
1826 release_metapage(mp);
1827 jfs_error(ip->i_sb,
1828 "diAllocIno: nfreeinos = 0, but iag on freelist");
1829 return -EIO;
1830 }
1831
1832 /* scan the free inode summary map to find an extent
1833 * with free inodes.
1834 */
1835 for (sword = 0;; sword++) {
1836 if (sword >= SMAPSZ) {
1837 IREAD_UNLOCK(imap->im_ipimap);
1838 release_metapage(mp);
1839 jfs_error(ip->i_sb,
1840 "diAllocIno: free inode not found in summary map");
1841 return -EIO;
1842 }
1843
1844 if (~iagp->inosmap[sword])
1845 break;
1846 }
1847
1848 /* found a extent with free inodes. determine
1849 * the extent number.
1850 */
1851 rem = diFindFree(le32_to_cpu(iagp->inosmap[sword]), 0);
1852 if (rem >= EXTSPERSUM) {
1853 IREAD_UNLOCK(imap->im_ipimap);
1854 release_metapage(mp);
1855 jfs_error(ip->i_sb, "diAllocIno: no free extent found");
1856 return -EIO;
1857 }
1858 extno = (sword << L2EXTSPERSUM) + rem;
1859
1860 /* find the first free inode in the extent.
1861 */
1862 rem = diFindFree(le32_to_cpu(iagp->wmap[extno]), 0);
1863 if (rem >= INOSPEREXT) {
1864 IREAD_UNLOCK(imap->im_ipimap);
1865 release_metapage(mp);
1866 jfs_error(ip->i_sb, "diAllocIno: free inode not found");
1867 return -EIO;
1868 }
1869
1870 /* compute the inode number within the iag.
1871 */
1872 ino = (extno << L2INOSPEREXT) + rem;
1873
1874 /* allocate the inode.
1875 */
1876 rc = diAllocBit(imap, iagp, ino);
1877 IREAD_UNLOCK(imap->im_ipimap);
1878 if (rc) {
1879 release_metapage(mp);
1880 return (rc);
1881 }
1882
1883 /* set the results of the allocation and write the iag.
1884 */
1885 diInitInode(ip, iagno, ino, extno, iagp);
1886 write_metapage(mp);
1887
1888 return (0);
1889}
1890
1891
1892/*
1893 * NAME: diAllocExt(imap,agno,ip)
1894 *
1895 * FUNCTION: add a new extent of free inodes to an iag, allocating
1896 * an inode from this extent to satisfy the current allocation
1897 * request.
1898 *
1899 * this routine first tries to find an existing iag with free
1900 * extents through the ag free extent list. if list is not
1901 * empty, the head of the list will be selected as the home
1902 * of the new extent of free inodes. otherwise (the list is
1903 * empty), a new iag will be allocated for the ag to contain
1904 * the extent.
1905 *
1906 * once an iag has been selected, the free extent summary map
1907 * is used to locate a free extent within the iag and diNewExt()
1908 * is called to initialize the extent, with initialization
1909 * including the allocation of the first inode of the extent
1910 * for the purpose of satisfying this request.
1911 *
1912 * PARAMETERS:
1913 * imap - pointer to inode map control structure.
1914 * agno - allocation group number.
1915 * ip - pointer to new inode to be filled in on successful return
1916 * with the disk inode number allocated, its extent address
1917 * and the start of the ag.
1918 *
1919 * RETURN VALUES:
1920 * 0 - success.
1921 * -ENOSPC - insufficient disk resources.
1922 * -EIO - i/o error.
1923 */
1924static int diAllocExt(struct inomap * imap, int agno, struct inode *ip)
1925{
1926 int rem, iagno, sword, extno, rc;
1927 struct metapage *mp;
1928 struct iag *iagp;
1929
1930 /* check if the ag has any iags with free extents. if not,
1931 * allocate a new iag for the ag.
1932 */
1933 if ((iagno = imap->im_agctl[agno].extfree) < 0) {
1934 /* If successful, diNewIAG will obtain the read lock on the
1935 * imap inode.
1936 */
1937 if ((rc = diNewIAG(imap, &iagno, agno, &mp))) {
1938 return (rc);
1939 }
1940 iagp = (struct iag *) mp->data;
1941
1942 /* set the ag number if this a brand new iag
1943 */
1944 iagp->agstart =
1945 cpu_to_le64(AGTOBLK(agno, imap->im_ipimap));
1946 } else {
1947 /* read the iag.
1948 */
1949 IREAD_LOCK(imap->im_ipimap);
1950 if ((rc = diIAGRead(imap, iagno, &mp))) {
1951 IREAD_UNLOCK(imap->im_ipimap);
1952 jfs_error(ip->i_sb, "diAllocExt: error reading iag");
1953 return rc;
1954 }
1955 iagp = (struct iag *) mp->data;
1956 }
1957
1958 /* using the free extent summary map, find a free extent.
1959 */
1960 for (sword = 0;; sword++) {
1961 if (sword >= SMAPSZ) {
1962 release_metapage(mp);
1963 IREAD_UNLOCK(imap->im_ipimap);
1964 jfs_error(ip->i_sb,
1965 "diAllocExt: free ext summary map not found");
1966 return -EIO;
1967 }
1968 if (~iagp->extsmap[sword])
1969 break;
1970 }
1971
1972 /* determine the extent number of the free extent.
1973 */
1974 rem = diFindFree(le32_to_cpu(iagp->extsmap[sword]), 0);
1975 if (rem >= EXTSPERSUM) {
1976 release_metapage(mp);
1977 IREAD_UNLOCK(imap->im_ipimap);
1978 jfs_error(ip->i_sb, "diAllocExt: free extent not found");
1979 return -EIO;
1980 }
1981 extno = (sword << L2EXTSPERSUM) + rem;
1982
1983 /* initialize the new extent.
1984 */
1985 rc = diNewExt(imap, iagp, extno);
1986 IREAD_UNLOCK(imap->im_ipimap);
1987 if (rc) {
1988 /* something bad happened. if a new iag was allocated,
1989 * place it back on the inode map's iag free list, and
1990 * clear the ag number information.
1991 */
1992 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
1993 IAGFREE_LOCK(imap);
1994 iagp->iagfree = cpu_to_le32(imap->im_freeiag);
1995 imap->im_freeiag = iagno;
1996 IAGFREE_UNLOCK(imap);
1997 }
1998 write_metapage(mp);
1999 return (rc);
2000 }
2001
2002 /* set the results of the allocation and write the iag.
2003 */
2004 diInitInode(ip, iagno, extno << L2INOSPEREXT, extno, iagp);
2005
2006 write_metapage(mp);
2007
2008 return (0);
2009}
2010
2011
2012/*
2013 * NAME: diAllocBit(imap,iagp,ino)
2014 *
2015 * FUNCTION: allocate a backed inode from an iag.
2016 *
2017 * this routine performs the mechanics of allocating a
2018 * specified inode from a backed extent.
2019 *
2020 * if the inode to be allocated represents the last free
2021 * inode within the iag, the iag will be removed from the
2022 * ag free inode list.
2023 *
2024 * a careful update approach is used to provide consistency
2025 * in the face of updates to multiple buffers. under this
2026 * approach, all required buffers are obtained before making
2027 * any updates and are held all are updates are complete.
2028 *
2029 * PRE CONDITION: Already have buffer lock on iagp. Already have AG lock on
2030 * this AG. Must have read lock on imap inode.
2031 *
2032 * PARAMETERS:
2033 * imap - pointer to inode map control structure.
2034 * iagp - pointer to iag.
2035 * ino - inode number to be allocated within the iag.
2036 *
2037 * RETURN VALUES:
2038 * 0 - success.
2039 * -ENOSPC - insufficient disk resources.
2040 * -EIO - i/o error.
2041 */
2042static int diAllocBit(struct inomap * imap, struct iag * iagp, int ino)
2043{
2044 int extno, bitno, agno, sword, rc;
2045 struct metapage *amp = NULL, *bmp = NULL;
2046 struct iag *aiagp = NULL, *biagp = NULL;
2047 u32 mask;
2048
2049 /* check if this is the last free inode within the iag.
2050 * if so, it will have to be removed from the ag free
2051 * inode list, so get the iags preceeding and following
2052 * it on the list.
2053 */
2054 if (iagp->nfreeinos == cpu_to_le32(1)) {
2055 if ((int) le32_to_cpu(iagp->inofreefwd) >= 0) {
2056 if ((rc =
2057 diIAGRead(imap, le32_to_cpu(iagp->inofreefwd),
2058 &amp)))
2059 return (rc);
2060 aiagp = (struct iag *) amp->data;
2061 }
2062
2063 if ((int) le32_to_cpu(iagp->inofreeback) >= 0) {
2064 if ((rc =
2065 diIAGRead(imap,
2066 le32_to_cpu(iagp->inofreeback),
2067 &bmp))) {
2068 if (amp)
2069 release_metapage(amp);
2070 return (rc);
2071 }
2072 biagp = (struct iag *) bmp->data;
2073 }
2074 }
2075
2076 /* get the ag number, extent number, inode number within
2077 * the extent.
2078 */
2079 agno = BLKTOAG(le64_to_cpu(iagp->agstart), JFS_SBI(imap->im_ipimap->i_sb));
2080 extno = ino >> L2INOSPEREXT;
2081 bitno = ino & (INOSPEREXT - 1);
2082
2083 /* compute the mask for setting the map.
2084 */
2085 mask = HIGHORDER >> bitno;
2086
2087 /* the inode should be free and backed.
2088 */
2089 if (((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) ||
2090 ((le32_to_cpu(iagp->wmap[extno]) & mask) != 0) ||
2091 (addressPXD(&iagp->inoext[extno]) == 0)) {
2092 if (amp)
2093 release_metapage(amp);
2094 if (bmp)
2095 release_metapage(bmp);
2096
2097 jfs_error(imap->im_ipimap->i_sb,
2098 "diAllocBit: iag inconsistent");
2099 return -EIO;
2100 }
2101
2102 /* mark the inode as allocated in the working map.
2103 */
2104 iagp->wmap[extno] |= cpu_to_le32(mask);
2105
2106 /* check if all inodes within the extent are now
2107 * allocated. if so, update the free inode summary
2108 * map to reflect this.
2109 */
2110 if (iagp->wmap[extno] == cpu_to_le32(ONES)) {
2111 sword = extno >> L2EXTSPERSUM;
2112 bitno = extno & (EXTSPERSUM - 1);
2113 iagp->inosmap[sword] |= cpu_to_le32(HIGHORDER >> bitno);
2114 }
2115
2116 /* if this was the last free inode in the iag, remove the
2117 * iag from the ag free inode list.
2118 */
2119 if (iagp->nfreeinos == cpu_to_le32(1)) {
2120 if (amp) {
2121 aiagp->inofreeback = iagp->inofreeback;
2122 write_metapage(amp);
2123 }
2124
2125 if (bmp) {
2126 biagp->inofreefwd = iagp->inofreefwd;
2127 write_metapage(bmp);
2128 } else {
2129 imap->im_agctl[agno].inofree =
2130 le32_to_cpu(iagp->inofreefwd);
2131 }
2132 iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1);
2133 }
2134
2135 /* update the free inode count at the iag, ag, inode
2136 * map levels.
2137 */
2138 iagp->nfreeinos = cpu_to_le32(le32_to_cpu(iagp->nfreeinos) - 1);
2139 imap->im_agctl[agno].numfree -= 1;
2140 atomic_dec(&imap->im_numfree);
2141
2142 return (0);
2143}
2144
2145
2146/*
2147 * NAME: diNewExt(imap,iagp,extno)
2148 *
2149 * FUNCTION: initialize a new extent of inodes for an iag, allocating
2150 * the first inode of the extent for use for the current
2151 * allocation request.
2152 *
2153 * disk resources are allocated for the new extent of inodes
2154 * and the inodes themselves are initialized to reflect their
2155 * existence within the extent (i.e. their inode numbers and
2156 * inode extent addresses are set) and their initial state
2157 * (mode and link count are set to zero).
2158 *
2159 * if the iag is new, it is not yet on an ag extent free list
2160 * but will now be placed on this list.
2161 *
2162 * if the allocation of the new extent causes the iag to
2163 * have no free extent, the iag will be removed from the
2164 * ag extent free list.
2165 *
2166 * if the iag has no free backed inodes, it will be placed
2167 * on the ag free inode list, since the addition of the new
2168 * extent will now cause it to have free inodes.
2169 *
2170 * a careful update approach is used to provide consistency
2171 * (i.e. list consistency) in the face of updates to multiple
2172 * buffers. under this approach, all required buffers are
2173 * obtained before making any updates and are held until all
2174 * updates are complete.
2175 *
2176 * PRE CONDITION: Already have buffer lock on iagp. Already have AG lock on
2177 * this AG. Must have read lock on imap inode.
2178 *
2179 * PARAMETERS:
2180 * imap - pointer to inode map control structure.
2181 * iagp - pointer to iag.
2182 * extno - extent number.
2183 *
2184 * RETURN VALUES:
2185 * 0 - success.
2186 * -ENOSPC - insufficient disk resources.
2187 * -EIO - i/o error.
2188 */
2189static int diNewExt(struct inomap * imap, struct iag * iagp, int extno)
2190{
2191 int agno, iagno, fwd, back, freei = 0, sword, rc;
2192 struct iag *aiagp = NULL, *biagp = NULL, *ciagp = NULL;
2193 struct metapage *amp, *bmp, *cmp, *dmp;
2194 struct inode *ipimap;
2195 s64 blkno, hint;
2196 int i, j;
2197 u32 mask;
2198 ino_t ino;
2199 struct dinode *dp;
2200 struct jfs_sb_info *sbi;
2201
2202 /* better have free extents.
2203 */
2204 if (!iagp->nfreeexts) {
2205 jfs_error(imap->im_ipimap->i_sb, "diNewExt: no free extents");
2206 return -EIO;
2207 }
2208
2209 /* get the inode map inode.
2210 */
2211 ipimap = imap->im_ipimap;
2212 sbi = JFS_SBI(ipimap->i_sb);
2213
2214 amp = bmp = cmp = NULL;
2215
2216 /* get the ag and iag numbers for this iag.
2217 */
2218 agno = BLKTOAG(le64_to_cpu(iagp->agstart), sbi);
2219 iagno = le32_to_cpu(iagp->iagnum);
2220
2221 /* check if this is the last free extent within the
2222 * iag. if so, the iag must be removed from the ag
2223 * free extent list, so get the iags preceeding and
2224 * following the iag on this list.
2225 */
2226 if (iagp->nfreeexts == cpu_to_le32(1)) {
2227 if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) {
2228 if ((rc = diIAGRead(imap, fwd, &amp)))
2229 return (rc);
2230 aiagp = (struct iag *) amp->data;
2231 }
2232
2233 if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) {
2234 if ((rc = diIAGRead(imap, back, &bmp)))
2235 goto error_out;
2236 biagp = (struct iag *) bmp->data;
2237 }
2238 } else {
2239 /* the iag has free extents. if all extents are free
2240 * (as is the case for a newly allocated iag), the iag
2241 * must be added to the ag free extent list, so get
2242 * the iag at the head of the list in preparation for
2243 * adding this iag to this list.
2244 */
2245 fwd = back = -1;
2246 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2247 if ((fwd = imap->im_agctl[agno].extfree) >= 0) {
2248 if ((rc = diIAGRead(imap, fwd, &amp)))
2249 goto error_out;
2250 aiagp = (struct iag *) amp->data;
2251 }
2252 }
2253 }
2254
2255 /* check if the iag has no free inodes. if so, the iag
2256 * will have to be added to the ag free inode list, so get
2257 * the iag at the head of the list in preparation for
2258 * adding this iag to this list. in doing this, we must
2259 * check if we already have the iag at the head of
2260 * the list in hand.
2261 */
2262 if (iagp->nfreeinos == 0) {
2263 freei = imap->im_agctl[agno].inofree;
2264
2265 if (freei >= 0) {
2266 if (freei == fwd) {
2267 ciagp = aiagp;
2268 } else if (freei == back) {
2269 ciagp = biagp;
2270 } else {
2271 if ((rc = diIAGRead(imap, freei, &cmp)))
2272 goto error_out;
2273 ciagp = (struct iag *) cmp->data;
2274 }
2275 if (ciagp == NULL) {
2276 jfs_error(imap->im_ipimap->i_sb,
2277 "diNewExt: ciagp == NULL");
2278 rc = -EIO;
2279 goto error_out;
2280 }
2281 }
2282 }
2283
2284 /* allocate disk space for the inode extent.
2285 */
2286 if ((extno == 0) || (addressPXD(&iagp->inoext[extno - 1]) == 0))
2287 hint = ((s64) agno << sbi->bmap->db_agl2size) - 1;
2288 else
2289 hint = addressPXD(&iagp->inoext[extno - 1]) +
2290 lengthPXD(&iagp->inoext[extno - 1]) - 1;
2291
2292 if ((rc = dbAlloc(ipimap, hint, (s64) imap->im_nbperiext, &blkno)))
2293 goto error_out;
2294
2295 /* compute the inode number of the first inode within the
2296 * extent.
2297 */
2298 ino = (iagno << L2INOSPERIAG) + (extno << L2INOSPEREXT);
2299
2300 /* initialize the inodes within the newly allocated extent a
2301 * page at a time.
2302 */
2303 for (i = 0; i < imap->im_nbperiext; i += sbi->nbperpage) {
2304 /* get a buffer for this page of disk inodes.
2305 */
2306 dmp = get_metapage(ipimap, blkno + i, PSIZE, 1);
2307 if (dmp == NULL) {
2308 rc = -EIO;
2309 goto error_out;
2310 }
2311 dp = (struct dinode *) dmp->data;
2312
2313 /* initialize the inode number, mode, link count and
2314 * inode extent address.
2315 */
2316 for (j = 0; j < INOSPERPAGE; j++, dp++, ino++) {
2317 dp->di_inostamp = cpu_to_le32(sbi->inostamp);
2318 dp->di_number = cpu_to_le32(ino);
2319 dp->di_fileset = cpu_to_le32(FILESYSTEM_I);
2320 dp->di_mode = 0;
2321 dp->di_nlink = 0;
2322 PXDaddress(&(dp->di_ixpxd), blkno);
2323 PXDlength(&(dp->di_ixpxd), imap->im_nbperiext);
2324 }
2325 write_metapage(dmp);
2326 }
2327
2328 /* if this is the last free extent within the iag, remove the
2329 * iag from the ag free extent list.
2330 */
2331 if (iagp->nfreeexts == cpu_to_le32(1)) {
2332 if (fwd >= 0)
2333 aiagp->extfreeback = iagp->extfreeback;
2334
2335 if (back >= 0)
2336 biagp->extfreefwd = iagp->extfreefwd;
2337 else
2338 imap->im_agctl[agno].extfree =
2339 le32_to_cpu(iagp->extfreefwd);
2340
2341 iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1);
2342 } else {
2343 /* if the iag has all free extents (newly allocated iag),
2344 * add the iag to the ag free extent list.
2345 */
2346 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2347 if (fwd >= 0)
2348 aiagp->extfreeback = cpu_to_le32(iagno);
2349
2350 iagp->extfreefwd = cpu_to_le32(fwd);
2351 iagp->extfreeback = cpu_to_le32(-1);
2352 imap->im_agctl[agno].extfree = iagno;
2353 }
2354 }
2355
2356 /* if the iag has no free inodes, add the iag to the
2357 * ag free inode list.
2358 */
2359 if (iagp->nfreeinos == 0) {
2360 if (freei >= 0)
2361 ciagp->inofreeback = cpu_to_le32(iagno);
2362
2363 iagp->inofreefwd =
2364 cpu_to_le32(imap->im_agctl[agno].inofree);
2365 iagp->inofreeback = cpu_to_le32(-1);
2366 imap->im_agctl[agno].inofree = iagno;
2367 }
2368
2369 /* initialize the extent descriptor of the extent. */
2370 PXDlength(&iagp->inoext[extno], imap->im_nbperiext);
2371 PXDaddress(&iagp->inoext[extno], blkno);
2372
2373 /* initialize the working and persistent map of the extent.
2374 * the working map will be initialized such that
2375 * it indicates the first inode of the extent is allocated.
2376 */
2377 iagp->wmap[extno] = cpu_to_le32(HIGHORDER);
2378 iagp->pmap[extno] = 0;
2379
2380 /* update the free inode and free extent summary maps
2381 * for the extent to indicate the extent has free inodes
2382 * and no longer represents a free extent.
2383 */
2384 sword = extno >> L2EXTSPERSUM;
2385 mask = HIGHORDER >> (extno & (EXTSPERSUM - 1));
2386 iagp->extsmap[sword] |= cpu_to_le32(mask);
2387 iagp->inosmap[sword] &= cpu_to_le32(~mask);
2388
2389 /* update the free inode and free extent counts for the
2390 * iag.
2391 */
2392 iagp->nfreeinos = cpu_to_le32(le32_to_cpu(iagp->nfreeinos) +
2393 (INOSPEREXT - 1));
2394 iagp->nfreeexts = cpu_to_le32(le32_to_cpu(iagp->nfreeexts) - 1);
2395
2396 /* update the free and backed inode counts for the ag.
2397 */
2398 imap->im_agctl[agno].numfree += (INOSPEREXT - 1);
2399 imap->im_agctl[agno].numinos += INOSPEREXT;
2400
2401 /* update the free and backed inode counts for the inode map.
2402 */
2403 atomic_add(INOSPEREXT - 1, &imap->im_numfree);
2404 atomic_add(INOSPEREXT, &imap->im_numinos);
2405
2406 /* write the iags.
2407 */
2408 if (amp)
2409 write_metapage(amp);
2410 if (bmp)
2411 write_metapage(bmp);
2412 if (cmp)
2413 write_metapage(cmp);
2414
2415 return (0);
2416
2417 error_out:
2418
2419 /* release the iags.
2420 */
2421 if (amp)
2422 release_metapage(amp);
2423 if (bmp)
2424 release_metapage(bmp);
2425 if (cmp)
2426 release_metapage(cmp);
2427
2428 return (rc);
2429}
2430
2431
2432/*
2433 * NAME: diNewIAG(imap,iagnop,agno)
2434 *
2435 * FUNCTION: allocate a new iag for an allocation group.
2436 *
2437 * first tries to allocate the iag from the inode map
2438 * iagfree list:
2439 * if the list has free iags, the head of the list is removed
2440 * and returned to satisfy the request.
2441 * if the inode map's iag free list is empty, the inode map
2442 * is extended to hold a new iag. this new iag is initialized
2443 * and returned to satisfy the request.
2444 *
2445 * PARAMETERS:
2446 * imap - pointer to inode map control structure.
2447 * iagnop - pointer to an iag number set with the number of the
2448 * newly allocated iag upon successful return.
2449 * agno - allocation group number.
2450 * bpp - Buffer pointer to be filled in with new IAG's buffer
2451 *
2452 * RETURN VALUES:
2453 * 0 - success.
2454 * -ENOSPC - insufficient disk resources.
2455 * -EIO - i/o error.
2456 *
2457 * serialization:
2458 * AG lock held on entry/exit;
2459 * write lock on the map is held inside;
2460 * read lock on the map is held on successful completion;
2461 *
2462 * note: new iag transaction:
2463 * . synchronously write iag;
2464 * . write log of xtree and inode of imap;
2465 * . commit;
2466 * . synchronous write of xtree (right to left, bottom to top);
2467 * . at start of logredo(): init in-memory imap with one additional iag page;
2468 * . at end of logredo(): re-read imap inode to determine
2469 * new imap size;
2470 */
2471static int
2472diNewIAG(struct inomap * imap, int *iagnop, int agno, struct metapage ** mpp)
2473{
2474 int rc;
2475 int iagno, i, xlen;
2476 struct inode *ipimap;
2477 struct super_block *sb;
2478 struct jfs_sb_info *sbi;
2479 struct metapage *mp;
2480 struct iag *iagp;
2481 s64 xaddr = 0;
2482 s64 blkno;
2483 tid_t tid;
2484#ifdef _STILL_TO_PORT
2485 xad_t xad;
2486#endif /* _STILL_TO_PORT */
2487 struct inode *iplist[1];
2488
2489 /* pick up pointers to the inode map and mount inodes */
2490 ipimap = imap->im_ipimap;
2491 sb = ipimap->i_sb;
2492 sbi = JFS_SBI(sb);
2493
2494 /* acquire the free iag lock */
2495 IAGFREE_LOCK(imap);
2496
2497 /* if there are any iags on the inode map free iag list,
2498 * allocate the iag from the head of the list.
2499 */
2500 if (imap->im_freeiag >= 0) {
2501 /* pick up the iag number at the head of the list */
2502 iagno = imap->im_freeiag;
2503
2504 /* determine the logical block number of the iag */
2505 blkno = IAGTOLBLK(iagno, sbi->l2nbperpage);
2506 } else {
2507 /* no free iags. the inode map will have to be extented
2508 * to include a new iag.
2509 */
2510
2511 /* acquire inode map lock */
2512 IWRITE_LOCK(ipimap);
2513
2514 if (ipimap->i_size >> L2PSIZE != imap->im_nextiag + 1) {
2515 IWRITE_UNLOCK(ipimap);
2516 IAGFREE_UNLOCK(imap);
2517 jfs_error(imap->im_ipimap->i_sb,
2518 "diNewIAG: ipimap->i_size is wrong");
2519 return -EIO;
2520 }
2521
2522
2523 /* get the next avaliable iag number */
2524 iagno = imap->im_nextiag;
2525
2526 /* make sure that we have not exceeded the maximum inode
2527 * number limit.
2528 */
2529 if (iagno > (MAXIAGS - 1)) {
2530 /* release the inode map lock */
2531 IWRITE_UNLOCK(ipimap);
2532
2533 rc = -ENOSPC;
2534 goto out;
2535 }
2536
2537 /*
2538 * synchronously append new iag page.
2539 */
2540 /* determine the logical address of iag page to append */
2541 blkno = IAGTOLBLK(iagno, sbi->l2nbperpage);
2542
2543 /* Allocate extent for new iag page */
2544 xlen = sbi->nbperpage;
2545 if ((rc = dbAlloc(ipimap, 0, (s64) xlen, &xaddr))) {
2546 /* release the inode map lock */
2547 IWRITE_UNLOCK(ipimap);
2548
2549 goto out;
2550 }
2551
Dave Kleikampd2e83702005-05-02 12:24:51 -06002552 /*
2553 * start transaction of update of the inode map
2554 * addressing structure pointing to the new iag page;
2555 */
2556 tid = txBegin(sb, COMMIT_FORCE);
Ingo Molnar1de87442006-01-24 15:22:50 -06002557 mutex_lock(&JFS_IP(ipimap)->commit_mutex);
Dave Kleikampd2e83702005-05-02 12:24:51 -06002558
2559 /* update the inode map addressing structure to point to it */
2560 if ((rc =
2561 xtInsert(tid, ipimap, 0, blkno, xlen, &xaddr, 0))) {
2562 txEnd(tid);
Ingo Molnar1de87442006-01-24 15:22:50 -06002563 mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 /* Free the blocks allocated for the iag since it was
2565 * not successfully added to the inode map
2566 */
2567 dbFree(ipimap, xaddr, (s64) xlen);
2568
2569 /* release the inode map lock */
2570 IWRITE_UNLOCK(ipimap);
2571
Dave Kleikampd2e83702005-05-02 12:24:51 -06002572 goto out;
2573 }
2574
2575 /* update the inode map's inode to reflect the extension */
2576 ipimap->i_size += PSIZE;
2577 inode_add_bytes(ipimap, PSIZE);
2578
2579 /* assign a buffer for the page */
2580 mp = get_metapage(ipimap, blkno, PSIZE, 0);
2581 if (!mp) {
2582 /*
2583 * This is very unlikely since we just created the
2584 * extent, but let's try to handle it correctly
2585 */
2586 xtTruncate(tid, ipimap, ipimap->i_size - PSIZE,
2587 COMMIT_PWMAP);
2588
2589 txAbort(tid, 0);
2590 txEnd(tid);
2591
2592 /* release the inode map lock */
2593 IWRITE_UNLOCK(ipimap);
2594
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595 rc = -EIO;
2596 goto out;
2597 }
2598 iagp = (struct iag *) mp->data;
2599
2600 /* init the iag */
2601 memset(iagp, 0, sizeof(struct iag));
2602 iagp->iagnum = cpu_to_le32(iagno);
2603 iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1);
2604 iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1);
2605 iagp->iagfree = cpu_to_le32(-1);
2606 iagp->nfreeinos = 0;
2607 iagp->nfreeexts = cpu_to_le32(EXTSPERIAG);
2608
2609 /* initialize the free inode summary map (free extent
2610 * summary map initialization handled by bzero).
2611 */
2612 for (i = 0; i < SMAPSZ; i++)
2613 iagp->inosmap[i] = cpu_to_le32(ONES);
2614
2615 /*
Dave Kleikampd2e83702005-05-02 12:24:51 -06002616 * Write and sync the metapage
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 flush_metapage(mp);
2619
2620 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002621 * txCommit(COMMIT_FORCE) will synchronously write address
2622 * index pages and inode after commit in careful update order
2623 * of address index pages (right to left, bottom up);
2624 */
2625 iplist[0] = ipimap;
2626 rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE);
2627
2628 txEnd(tid);
Ingo Molnar1de87442006-01-24 15:22:50 -06002629 mutex_unlock(&JFS_IP(ipimap)->commit_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630
2631 duplicateIXtree(sb, blkno, xlen, &xaddr);
2632
2633 /* update the next avaliable iag number */
2634 imap->im_nextiag += 1;
2635
2636 /* Add the iag to the iag free list so we don't lose the iag
2637 * if a failure happens now.
2638 */
2639 imap->im_freeiag = iagno;
2640
2641 /* Until we have logredo working, we want the imap inode &
2642 * control page to be up to date.
2643 */
2644 diSync(ipimap);
2645
2646 /* release the inode map lock */
2647 IWRITE_UNLOCK(ipimap);
2648 }
2649
2650 /* obtain read lock on map */
2651 IREAD_LOCK(ipimap);
2652
2653 /* read the iag */
2654 if ((rc = diIAGRead(imap, iagno, &mp))) {
2655 IREAD_UNLOCK(ipimap);
2656 rc = -EIO;
2657 goto out;
2658 }
2659 iagp = (struct iag *) mp->data;
2660
2661 /* remove the iag from the iag free list */
2662 imap->im_freeiag = le32_to_cpu(iagp->iagfree);
2663 iagp->iagfree = cpu_to_le32(-1);
2664
2665 /* set the return iag number and buffer pointer */
2666 *iagnop = iagno;
2667 *mpp = mp;
2668
2669 out:
2670 /* release the iag free lock */
2671 IAGFREE_UNLOCK(imap);
2672
2673 return (rc);
2674}
2675
2676/*
2677 * NAME: diIAGRead()
2678 *
2679 * FUNCTION: get the buffer for the specified iag within a fileset
2680 * or aggregate inode map.
2681 *
2682 * PARAMETERS:
2683 * imap - pointer to inode map control structure.
2684 * iagno - iag number.
2685 * bpp - point to buffer pointer to be filled in on successful
2686 * exit.
2687 *
2688 * SERIALIZATION:
2689 * must have read lock on imap inode
2690 * (When called by diExtendFS, the filesystem is quiesced, therefore
2691 * the read lock is unnecessary.)
2692 *
2693 * RETURN VALUES:
2694 * 0 - success.
2695 * -EIO - i/o error.
2696 */
2697static int diIAGRead(struct inomap * imap, int iagno, struct metapage ** mpp)
2698{
2699 struct inode *ipimap = imap->im_ipimap;
2700 s64 blkno;
2701
2702 /* compute the logical block number of the iag. */
2703 blkno = IAGTOLBLK(iagno, JFS_SBI(ipimap->i_sb)->l2nbperpage);
2704
2705 /* read the iag. */
2706 *mpp = read_metapage(ipimap, blkno, PSIZE, 0);
2707 if (*mpp == NULL) {
2708 return -EIO;
2709 }
2710
2711 return (0);
2712}
2713
2714/*
2715 * NAME: diFindFree()
2716 *
2717 * FUNCTION: find the first free bit in a word starting at
2718 * the specified bit position.
2719 *
2720 * PARAMETERS:
2721 * word - word to be examined.
2722 * start - starting bit position.
2723 *
2724 * RETURN VALUES:
2725 * bit position of first free bit in the word or 32 if
2726 * no free bits were found.
2727 */
2728static int diFindFree(u32 word, int start)
2729{
2730 int bitno;
2731 assert(start < 32);
2732 /* scan the word for the first free bit. */
2733 for (word <<= start, bitno = start; bitno < 32;
2734 bitno++, word <<= 1) {
2735 if ((word & HIGHORDER) == 0)
2736 break;
2737 }
2738 return (bitno);
2739}
2740
2741/*
2742 * NAME: diUpdatePMap()
2743 *
2744 * FUNCTION: Update the persistent map in an IAG for the allocation or
2745 * freeing of the specified inode.
2746 *
2747 * PRE CONDITIONS: Working map has already been updated for allocate.
2748 *
2749 * PARAMETERS:
2750 * ipimap - Incore inode map inode
2751 * inum - Number of inode to mark in permanent map
2752 * is_free - If TRUE indicates inode should be marked freed, otherwise
2753 * indicates inode should be marked allocated.
2754 *
2755 * RETURN VALUES:
2756 * 0 for success
2757 */
2758int
2759diUpdatePMap(struct inode *ipimap,
2760 unsigned long inum, boolean_t is_free, struct tblock * tblk)
2761{
2762 int rc;
2763 struct iag *iagp;
2764 struct metapage *mp;
2765 int iagno, ino, extno, bitno;
2766 struct inomap *imap;
2767 u32 mask;
2768 struct jfs_log *log;
2769 int lsn, difft, diffp;
Dave Kleikamp7fab4792005-05-02 12:25:02 -06002770 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771
2772 imap = JFS_IP(ipimap)->i_imap;
2773 /* get the iag number containing the inode */
2774 iagno = INOTOIAG(inum);
2775 /* make sure that the iag is contained within the map */
2776 if (iagno >= imap->im_nextiag) {
2777 jfs_error(ipimap->i_sb,
2778 "diUpdatePMap: the iag is outside the map");
2779 return -EIO;
2780 }
2781 /* read the iag */
2782 IREAD_LOCK(ipimap);
2783 rc = diIAGRead(imap, iagno, &mp);
2784 IREAD_UNLOCK(ipimap);
2785 if (rc)
2786 return (rc);
Dave Kleikamp7fab4792005-05-02 12:25:02 -06002787 metapage_wait_for_io(mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788 iagp = (struct iag *) mp->data;
2789 /* get the inode number and extent number of the inode within
2790 * the iag and the inode number within the extent.
2791 */
2792 ino = inum & (INOSPERIAG - 1);
2793 extno = ino >> L2INOSPEREXT;
2794 bitno = ino & (INOSPEREXT - 1);
2795 mask = HIGHORDER >> bitno;
2796 /*
2797 * mark the inode free in persistent map:
2798 */
2799 if (is_free == TRUE) {
2800 /* The inode should have been allocated both in working
2801 * map and in persistent map;
2802 * the inode will be freed from working map at the release
2803 * of last reference release;
2804 */
2805 if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
2806 jfs_error(ipimap->i_sb,
2807 "diUpdatePMap: inode %ld not marked as "
2808 "allocated in wmap!", inum);
2809 }
2810 if (!(le32_to_cpu(iagp->pmap[extno]) & mask)) {
2811 jfs_error(ipimap->i_sb,
2812 "diUpdatePMap: inode %ld not marked as "
2813 "allocated in pmap!", inum);
2814 }
2815 /* update the bitmap for the extent of the freed inode */
2816 iagp->pmap[extno] &= cpu_to_le32(~mask);
2817 }
2818 /*
2819 * mark the inode allocated in persistent map:
2820 */
2821 else {
2822 /* The inode should be already allocated in the working map
2823 * and should be free in persistent map;
2824 */
2825 if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) {
2826 release_metapage(mp);
2827 jfs_error(ipimap->i_sb,
2828 "diUpdatePMap: the inode is not allocated in "
2829 "the working map");
2830 return -EIO;
2831 }
2832 if ((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) {
2833 release_metapage(mp);
2834 jfs_error(ipimap->i_sb,
2835 "diUpdatePMap: the inode is not free in the "
2836 "persistent map");
2837 return -EIO;
2838 }
2839 /* update the bitmap for the extent of the allocated inode */
2840 iagp->pmap[extno] |= cpu_to_le32(mask);
2841 }
2842 /*
2843 * update iag lsn
2844 */
2845 lsn = tblk->lsn;
2846 log = JFS_SBI(tblk->sb)->log;
Dave Kleikampbe0bf7d2006-03-08 10:59:15 -06002847 LOGSYNC_LOCK(log, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848 if (mp->lsn != 0) {
2849 /* inherit older/smaller lsn */
2850 logdiff(difft, lsn, log);
2851 logdiff(diffp, mp->lsn, log);
2852 if (difft < diffp) {
2853 mp->lsn = lsn;
2854 /* move mp after tblock in logsync list */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855 list_move(&mp->synclist, &tblk->synclist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856 }
2857 /* inherit younger/larger clsn */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858 assert(mp->clsn);
2859 logdiff(difft, tblk->clsn, log);
2860 logdiff(diffp, mp->clsn, log);
2861 if (difft > diffp)
2862 mp->clsn = tblk->clsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863 } else {
2864 mp->log = log;
2865 mp->lsn = lsn;
2866 /* insert mp after tblock in logsync list */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 log->count++;
2868 list_add(&mp->synclist, &tblk->synclist);
2869 mp->clsn = tblk->clsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870 }
Dave Kleikampbe0bf7d2006-03-08 10:59:15 -06002871 LOGSYNC_UNLOCK(log, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 write_metapage(mp);
2873 return (0);
2874}
2875
2876/*
2877 * diExtendFS()
2878 *
2879 * function: update imap for extendfs();
2880 *
2881 * note: AG size has been increased s.t. each k old contiguous AGs are
2882 * coalesced into a new AG;
2883 */
2884int diExtendFS(struct inode *ipimap, struct inode *ipbmap)
2885{
2886 int rc, rcx = 0;
2887 struct inomap *imap = JFS_IP(ipimap)->i_imap;
2888 struct iag *iagp = NULL, *hiagp = NULL;
2889 struct bmap *mp = JFS_SBI(ipbmap->i_sb)->bmap;
2890 struct metapage *bp, *hbp;
2891 int i, n, head;
2892 int numinos, xnuminos = 0, xnumfree = 0;
2893 s64 agstart;
2894
2895 jfs_info("diExtendFS: nextiag:%d numinos:%d numfree:%d",
2896 imap->im_nextiag, atomic_read(&imap->im_numinos),
2897 atomic_read(&imap->im_numfree));
2898
2899 /*
2900 * reconstruct imap
2901 *
2902 * coalesce contiguous k (newAGSize/oldAGSize) AGs;
2903 * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn;
2904 * note: new AG size = old AG size * (2**x).
2905 */
2906
2907 /* init per AG control information im_agctl[] */
2908 for (i = 0; i < MAXAG; i++) {
2909 imap->im_agctl[i].inofree = -1;
2910 imap->im_agctl[i].extfree = -1;
2911 imap->im_agctl[i].numinos = 0; /* number of backed inodes */
2912 imap->im_agctl[i].numfree = 0; /* number of free backed inodes */
2913 }
2914
2915 /*
2916 * process each iag page of the map.
2917 *
2918 * rebuild AG Free Inode List, AG Free Inode Extent List;
2919 */
2920 for (i = 0; i < imap->im_nextiag; i++) {
2921 if ((rc = diIAGRead(imap, i, &bp))) {
2922 rcx = rc;
2923 continue;
2924 }
2925 iagp = (struct iag *) bp->data;
2926 if (le32_to_cpu(iagp->iagnum) != i) {
2927 release_metapage(bp);
2928 jfs_error(ipimap->i_sb,
2929 "diExtendFs: unexpected value of iagnum");
2930 return -EIO;
2931 }
2932
2933 /* leave free iag in the free iag list */
2934 if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) {
2935 release_metapage(bp);
2936 continue;
2937 }
2938
2939 /* agstart that computes to the same ag is treated as same; */
2940 agstart = le64_to_cpu(iagp->agstart);
2941 /* iagp->agstart = agstart & ~(mp->db_agsize - 1); */
2942 n = agstart >> mp->db_agl2size;
2943
2944 /* compute backed inodes */
2945 numinos = (EXTSPERIAG - le32_to_cpu(iagp->nfreeexts))
2946 << L2INOSPEREXT;
2947 if (numinos > 0) {
2948 /* merge AG backed inodes */
2949 imap->im_agctl[n].numinos += numinos;
2950 xnuminos += numinos;
2951 }
2952
2953 /* if any backed free inodes, insert at AG free inode list */
2954 if ((int) le32_to_cpu(iagp->nfreeinos) > 0) {
2955 if ((head = imap->im_agctl[n].inofree) == -1) {
2956 iagp->inofreefwd = cpu_to_le32(-1);
2957 iagp->inofreeback = cpu_to_le32(-1);
2958 } else {
2959 if ((rc = diIAGRead(imap, head, &hbp))) {
2960 rcx = rc;
2961 goto nextiag;
2962 }
2963 hiagp = (struct iag *) hbp->data;
2964 hiagp->inofreeback = iagp->iagnum;
2965 iagp->inofreefwd = cpu_to_le32(head);
2966 iagp->inofreeback = cpu_to_le32(-1);
2967 write_metapage(hbp);
2968 }
2969
2970 imap->im_agctl[n].inofree =
2971 le32_to_cpu(iagp->iagnum);
2972
2973 /* merge AG backed free inodes */
2974 imap->im_agctl[n].numfree +=
2975 le32_to_cpu(iagp->nfreeinos);
2976 xnumfree += le32_to_cpu(iagp->nfreeinos);
2977 }
2978
2979 /* if any free extents, insert at AG free extent list */
2980 if (le32_to_cpu(iagp->nfreeexts) > 0) {
2981 if ((head = imap->im_agctl[n].extfree) == -1) {
2982 iagp->extfreefwd = cpu_to_le32(-1);
2983 iagp->extfreeback = cpu_to_le32(-1);
2984 } else {
2985 if ((rc = diIAGRead(imap, head, &hbp))) {
2986 rcx = rc;
2987 goto nextiag;
2988 }
2989 hiagp = (struct iag *) hbp->data;
2990 hiagp->extfreeback = iagp->iagnum;
2991 iagp->extfreefwd = cpu_to_le32(head);
2992 iagp->extfreeback = cpu_to_le32(-1);
2993 write_metapage(hbp);
2994 }
2995
2996 imap->im_agctl[n].extfree =
2997 le32_to_cpu(iagp->iagnum);
2998 }
2999
3000 nextiag:
3001 write_metapage(bp);
3002 }
3003
3004 if (xnuminos != atomic_read(&imap->im_numinos) ||
3005 xnumfree != atomic_read(&imap->im_numfree)) {
3006 jfs_error(ipimap->i_sb,
3007 "diExtendFs: numinos or numfree incorrect");
3008 return -EIO;
3009 }
3010
3011 return rcx;
3012}
3013
3014
3015/*
3016 * duplicateIXtree()
3017 *
3018 * serialization: IWRITE_LOCK held on entry/exit
3019 *
3020 * note: shadow page with regular inode (rel.2);
3021 */
3022static void duplicateIXtree(struct super_block *sb, s64 blkno,
3023 int xlen, s64 *xaddr)
3024{
3025 struct jfs_superblock *j_sb;
3026 struct buffer_head *bh;
3027 struct inode *ip;
3028 tid_t tid;
3029
3030 /* if AIT2 ipmap2 is bad, do not try to update it */
3031 if (JFS_SBI(sb)->mntflag & JFS_BAD_SAIT) /* s_flag */
3032 return;
3033 ip = diReadSpecial(sb, FILESYSTEM_I, 1);
3034 if (ip == NULL) {
3035 JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT;
3036 if (readSuper(sb, &bh))
3037 return;
3038 j_sb = (struct jfs_superblock *)bh->b_data;
3039 j_sb->s_flag |= cpu_to_le32(JFS_BAD_SAIT);
3040
3041 mark_buffer_dirty(bh);
3042 sync_dirty_buffer(bh);
3043 brelse(bh);
3044 return;
3045 }
3046
3047 /* start transaction */
3048 tid = txBegin(sb, COMMIT_FORCE);
3049 /* update the inode map addressing structure to point to it */
3050 if (xtInsert(tid, ip, 0, blkno, xlen, xaddr, 0)) {
3051 JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT;
3052 txAbort(tid, 1);
3053 goto cleanup;
3054
3055 }
3056 /* update the inode map's inode to reflect the extension */
3057 ip->i_size += PSIZE;
3058 inode_add_bytes(ip, PSIZE);
3059 txCommit(tid, 1, &ip, COMMIT_FORCE);
3060 cleanup:
3061 txEnd(tid);
3062 diFreeSpecial(ip);
3063}
3064
3065/*
3066 * NAME: copy_from_dinode()
3067 *
3068 * FUNCTION: Copies inode info from disk inode to in-memory inode
3069 *
3070 * RETURN VALUES:
3071 * 0 - success
3072 * -ENOMEM - insufficient memory
3073 */
3074static int copy_from_dinode(struct dinode * dip, struct inode *ip)
3075{
3076 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
Dave Kleikamp69eb66d2006-03-09 13:59:30 -06003077 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003078
3079 jfs_ip->fileset = le32_to_cpu(dip->di_fileset);
3080 jfs_ip->mode2 = le32_to_cpu(dip->di_mode);
3081
3082 ip->i_mode = le32_to_cpu(dip->di_mode) & 0xffff;
Dave Kleikamp69eb66d2006-03-09 13:59:30 -06003083 if (sbi->umask != -1) {
3084 ip->i_mode = (ip->i_mode & ~0777) | (0777 & ~sbi->umask);
3085 /* For directories, add x permission if r is allowed by umask */
3086 if (S_ISDIR(ip->i_mode)) {
3087 if (ip->i_mode & 0400)
3088 ip->i_mode |= 0100;
3089 if (ip->i_mode & 0040)
3090 ip->i_mode |= 0010;
3091 if (ip->i_mode & 0004)
3092 ip->i_mode |= 0001;
3093 }
3094 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095 ip->i_nlink = le32_to_cpu(dip->di_nlink);
Dave Kleikamp69eb66d2006-03-09 13:59:30 -06003096
3097 jfs_ip->saved_uid = le32_to_cpu(dip->di_uid);
3098 if (sbi->uid == -1)
3099 ip->i_uid = jfs_ip->saved_uid;
3100 else {
3101 ip->i_uid = sbi->uid;
3102 }
3103
3104 jfs_ip->saved_gid = le32_to_cpu(dip->di_gid);
3105 if (sbi->gid == -1)
3106 ip->i_gid = jfs_ip->saved_gid;
3107 else {
3108 ip->i_gid = sbi->gid;
3109 }
3110
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111 ip->i_size = le64_to_cpu(dip->di_size);
3112 ip->i_atime.tv_sec = le32_to_cpu(dip->di_atime.tv_sec);
3113 ip->i_atime.tv_nsec = le32_to_cpu(dip->di_atime.tv_nsec);
3114 ip->i_mtime.tv_sec = le32_to_cpu(dip->di_mtime.tv_sec);
3115 ip->i_mtime.tv_nsec = le32_to_cpu(dip->di_mtime.tv_nsec);
3116 ip->i_ctime.tv_sec = le32_to_cpu(dip->di_ctime.tv_sec);
3117 ip->i_ctime.tv_nsec = le32_to_cpu(dip->di_ctime.tv_nsec);
3118 ip->i_blksize = ip->i_sb->s_blocksize;
3119 ip->i_blocks = LBLK2PBLK(ip->i_sb, le64_to_cpu(dip->di_nblocks));
3120 ip->i_generation = le32_to_cpu(dip->di_gen);
3121
3122 jfs_ip->ixpxd = dip->di_ixpxd; /* in-memory pxd's are little-endian */
3123 jfs_ip->acl = dip->di_acl; /* as are dxd's */
3124 jfs_ip->ea = dip->di_ea;
3125 jfs_ip->next_index = le32_to_cpu(dip->di_next_index);
3126 jfs_ip->otime = le32_to_cpu(dip->di_otime.tv_sec);
3127 jfs_ip->acltype = le32_to_cpu(dip->di_acltype);
3128
3129 if (S_ISCHR(ip->i_mode) || S_ISBLK(ip->i_mode)) {
3130 jfs_ip->dev = le32_to_cpu(dip->di_rdev);
3131 ip->i_rdev = new_decode_dev(jfs_ip->dev);
3132 }
3133
3134 if (S_ISDIR(ip->i_mode)) {
3135 memcpy(&jfs_ip->i_dirtable, &dip->di_dirtable, 384);
3136 } else if (S_ISREG(ip->i_mode) || S_ISLNK(ip->i_mode)) {
3137 memcpy(&jfs_ip->i_xtroot, &dip->di_xtroot, 288);
3138 } else
3139 memcpy(&jfs_ip->i_inline_ea, &dip->di_inlineea, 128);
3140
3141 /* Zero the in-memory-only stuff */
3142 jfs_ip->cflag = 0;
3143 jfs_ip->btindex = 0;
3144 jfs_ip->btorder = 0;
3145 jfs_ip->bxflag = 0;
3146 jfs_ip->blid = 0;
3147 jfs_ip->atlhead = 0;
3148 jfs_ip->atltail = 0;
3149 jfs_ip->xtlid = 0;
3150 return (0);
3151}
3152
3153/*
3154 * NAME: copy_to_dinode()
3155 *
3156 * FUNCTION: Copies inode info from in-memory inode to disk inode
3157 */
3158static void copy_to_dinode(struct dinode * dip, struct inode *ip)
3159{
3160 struct jfs_inode_info *jfs_ip = JFS_IP(ip);
Dave Kleikamp69eb66d2006-03-09 13:59:30 -06003161 struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162
3163 dip->di_fileset = cpu_to_le32(jfs_ip->fileset);
Dave Kleikamp69eb66d2006-03-09 13:59:30 -06003164 dip->di_inostamp = cpu_to_le32(sbi->inostamp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003165 dip->di_number = cpu_to_le32(ip->i_ino);
3166 dip->di_gen = cpu_to_le32(ip->i_generation);
3167 dip->di_size = cpu_to_le64(ip->i_size);
3168 dip->di_nblocks = cpu_to_le64(PBLK2LBLK(ip->i_sb, ip->i_blocks));
3169 dip->di_nlink = cpu_to_le32(ip->i_nlink);
Dave Kleikamp69eb66d2006-03-09 13:59:30 -06003170 if (sbi->uid == -1)
3171 dip->di_uid = cpu_to_le32(ip->i_uid);
3172 else
3173 dip->di_uid = cpu_to_le32(jfs_ip->saved_uid);
3174 if (sbi->gid == -1)
3175 dip->di_gid = cpu_to_le32(ip->i_gid);
3176 else
3177 dip->di_gid = cpu_to_le32(jfs_ip->saved_gid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003178 /*
3179 * mode2 is only needed for storing the higher order bits.
3180 * Trust i_mode for the lower order ones
3181 */
Dave Kleikamp69eb66d2006-03-09 13:59:30 -06003182 if (sbi->umask == -1)
3183 dip->di_mode = cpu_to_le32((jfs_ip->mode2 & 0xffff0000) |
3184 ip->i_mode);
3185 else /* Leave the original permissions alone */
3186 dip->di_mode = cpu_to_le32(jfs_ip->mode2);
3187
Linus Torvalds1da177e2005-04-16 15:20:36 -07003188 dip->di_atime.tv_sec = cpu_to_le32(ip->i_atime.tv_sec);
3189 dip->di_atime.tv_nsec = cpu_to_le32(ip->i_atime.tv_nsec);
3190 dip->di_ctime.tv_sec = cpu_to_le32(ip->i_ctime.tv_sec);
3191 dip->di_ctime.tv_nsec = cpu_to_le32(ip->i_ctime.tv_nsec);
3192 dip->di_mtime.tv_sec = cpu_to_le32(ip->i_mtime.tv_sec);
3193 dip->di_mtime.tv_nsec = cpu_to_le32(ip->i_mtime.tv_nsec);
3194 dip->di_ixpxd = jfs_ip->ixpxd; /* in-memory pxd's are little-endian */
3195 dip->di_acl = jfs_ip->acl; /* as are dxd's */
3196 dip->di_ea = jfs_ip->ea;
3197 dip->di_next_index = cpu_to_le32(jfs_ip->next_index);
3198 dip->di_otime.tv_sec = cpu_to_le32(jfs_ip->otime);
3199 dip->di_otime.tv_nsec = 0;
3200 dip->di_acltype = cpu_to_le32(jfs_ip->acltype);
3201 if (S_ISCHR(ip->i_mode) || S_ISBLK(ip->i_mode))
3202 dip->di_rdev = cpu_to_le32(jfs_ip->dev);
3203}