Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) International Business Machines Corp., 2000-2004 | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify | |
5 | * it under the terms of the GNU General Public License as published by | |
6 | * the Free Software Foundation; either version 2 of the License, or | |
7 | * (at your option) any later version. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, | |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See | |
12 | * the GNU General Public License for more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License | |
15 | * along with this program; if not, write to the Free Software | |
16 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
17 | */ | |
18 | ||
19 | /* | |
20 | * jfs_imap.c: inode allocation map manager | |
21 | * | |
22 | * Serialization: | |
23 | * Each AG has a simple lock which is used to control the serialization of | |
24 | * the AG level lists. This lock should be taken first whenever an AG | |
25 | * level list will be modified or accessed. | |
26 | * | |
27 | * Each IAG is locked by obtaining the buffer for the IAG page. | |
28 | * | |
29 | * There is also a inode lock for the inode map inode. A read lock needs to | |
30 | * be taken whenever an IAG is read from the map or the global level | |
31 | * information is read. A write lock needs to be taken whenever the global | |
32 | * level information is modified or an atomic operation needs to be used. | |
33 | * | |
34 | * If more than one IAG is read at one time, the read lock may not | |
35 | * be given up until all of the IAG's are read. Otherwise, a deadlock | |
36 | * may occur when trying to obtain the read lock while another thread | |
37 | * holding the read lock is waiting on the IAG already being held. | |
38 | * | |
39 | * The control page of the inode map is read into memory by diMount(). | |
40 | * Thereafter it should only be modified in memory and then it will be | |
41 | * written out when the filesystem is unmounted by diUnmount(). | |
42 | */ | |
43 | ||
44 | #include <linux/fs.h> | |
45 | #include <linux/buffer_head.h> | |
46 | #include <linux/pagemap.h> | |
47 | #include <linux/quotaops.h> | |
48 | ||
49 | #include "jfs_incore.h" | |
1868f4aa | 50 | #include "jfs_inode.h" |
1da177e4 LT |
51 | #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 | /* | |
60 | * imap locks | |
61 | */ | |
62 | /* iag free list lock */ | |
63 | #define IAGFREE_LOCK_INIT(imap) init_MUTEX(&imap->im_freelock) | |
64 | #define IAGFREE_LOCK(imap) down(&imap->im_freelock) | |
65 | #define IAGFREE_UNLOCK(imap) up(&imap->im_freelock) | |
66 | ||
67 | /* per ag iag list locks */ | |
68 | #define AG_LOCK_INIT(imap,index) init_MUTEX(&(imap->im_aglock[index])) | |
69 | #define AG_LOCK(imap,agno) down(&imap->im_aglock[agno]) | |
70 | #define AG_UNLOCK(imap,agno) up(&imap->im_aglock[agno]) | |
71 | ||
1da177e4 LT |
72 | /* |
73 | * forward references | |
74 | */ | |
75 | static int diAllocAG(struct inomap *, int, boolean_t, struct inode *); | |
76 | static int diAllocAny(struct inomap *, int, boolean_t, struct inode *); | |
77 | static int diAllocBit(struct inomap *, struct iag *, int); | |
78 | static int diAllocExt(struct inomap *, int, struct inode *); | |
79 | static int diAllocIno(struct inomap *, int, struct inode *); | |
80 | static int diFindFree(u32, int); | |
81 | static int diNewExt(struct inomap *, struct iag *, int); | |
82 | static int diNewIAG(struct inomap *, int *, int, struct metapage **); | |
83 | static void duplicateIXtree(struct super_block *, s64, int, s64 *); | |
84 | ||
85 | static int diIAGRead(struct inomap * imap, int, struct metapage **); | |
86 | static int copy_from_dinode(struct dinode *, struct inode *); | |
87 | static void copy_to_dinode(struct dinode *, struct inode *); | |
88 | ||
1da177e4 LT |
89 | /* |
90 | * NAME: diMount() | |
91 | * | |
92 | * FUNCTION: initialize the incore inode map control structures for | |
93 | * a fileset or aggregate init time. | |
94 | * | |
95 | * the inode map's control structure (dinomap) is | |
96 | * brought in from disk and placed in virtual memory. | |
97 | * | |
98 | * PARAMETERS: | |
99 | * ipimap - pointer to inode map inode for the aggregate or fileset. | |
100 | * | |
101 | * RETURN VALUES: | |
102 | * 0 - success | |
103 | * -ENOMEM - insufficient free virtual memory. | |
104 | * -EIO - i/o error. | |
105 | */ | |
106 | int diMount(struct inode *ipimap) | |
107 | { | |
108 | struct inomap *imap; | |
109 | struct metapage *mp; | |
110 | int index; | |
111 | struct dinomap_disk *dinom_le; | |
112 | ||
113 | /* | |
114 | * allocate/initialize the in-memory inode map control structure | |
115 | */ | |
116 | /* allocate the in-memory inode map control structure. */ | |
117 | imap = (struct inomap *) kmalloc(sizeof(struct inomap), GFP_KERNEL); | |
118 | if (imap == NULL) { | |
119 | jfs_err("diMount: kmalloc returned NULL!"); | |
120 | return -ENOMEM; | |
121 | } | |
122 | ||
123 | /* read the on-disk inode map control structure. */ | |
124 | ||
125 | mp = read_metapage(ipimap, | |
126 | IMAPBLKNO << JFS_SBI(ipimap->i_sb)->l2nbperpage, | |
127 | PSIZE, 0); | |
128 | if (mp == NULL) { | |
129 | kfree(imap); | |
130 | return -EIO; | |
131 | } | |
132 | ||
133 | /* copy the on-disk version to the in-memory version. */ | |
134 | dinom_le = (struct dinomap_disk *) mp->data; | |
135 | imap->im_freeiag = le32_to_cpu(dinom_le->in_freeiag); | |
136 | imap->im_nextiag = le32_to_cpu(dinom_le->in_nextiag); | |
137 | atomic_set(&imap->im_numinos, le32_to_cpu(dinom_le->in_numinos)); | |
138 | atomic_set(&imap->im_numfree, le32_to_cpu(dinom_le->in_numfree)); | |
139 | imap->im_nbperiext = le32_to_cpu(dinom_le->in_nbperiext); | |
140 | imap->im_l2nbperiext = le32_to_cpu(dinom_le->in_l2nbperiext); | |
141 | for (index = 0; index < MAXAG; index++) { | |
142 | imap->im_agctl[index].inofree = | |
143 | le32_to_cpu(dinom_le->in_agctl[index].inofree); | |
144 | imap->im_agctl[index].extfree = | |
145 | le32_to_cpu(dinom_le->in_agctl[index].extfree); | |
146 | imap->im_agctl[index].numinos = | |
147 | le32_to_cpu(dinom_le->in_agctl[index].numinos); | |
148 | imap->im_agctl[index].numfree = | |
149 | le32_to_cpu(dinom_le->in_agctl[index].numfree); | |
150 | } | |
151 | ||
152 | /* release the buffer. */ | |
153 | release_metapage(mp); | |
154 | ||
155 | /* | |
156 | * allocate/initialize inode allocation map locks | |
157 | */ | |
158 | /* allocate and init iag free list lock */ | |
159 | IAGFREE_LOCK_INIT(imap); | |
160 | ||
161 | /* allocate and init ag list locks */ | |
162 | for (index = 0; index < MAXAG; index++) { | |
163 | AG_LOCK_INIT(imap, index); | |
164 | } | |
165 | ||
166 | /* bind the inode map inode and inode map control structure | |
167 | * to each other. | |
168 | */ | |
169 | imap->im_ipimap = ipimap; | |
170 | JFS_IP(ipimap)->i_imap = imap; | |
171 | ||
1da177e4 LT |
172 | return (0); |
173 | } | |
174 | ||
175 | ||
176 | /* | |
177 | * NAME: diUnmount() | |
178 | * | |
179 | * FUNCTION: write to disk the incore inode map control structures for | |
180 | * a fileset or aggregate at unmount time. | |
181 | * | |
182 | * PARAMETERS: | |
183 | * ipimap - pointer to inode map inode for the aggregate or fileset. | |
184 | * | |
185 | * RETURN VALUES: | |
186 | * 0 - success | |
187 | * -ENOMEM - insufficient free virtual memory. | |
188 | * -EIO - i/o error. | |
189 | */ | |
190 | int diUnmount(struct inode *ipimap, int mounterror) | |
191 | { | |
192 | struct inomap *imap = JFS_IP(ipimap)->i_imap; | |
193 | ||
194 | /* | |
195 | * update the on-disk inode map control structure | |
196 | */ | |
197 | ||
198 | if (!(mounterror || isReadOnly(ipimap))) | |
199 | diSync(ipimap); | |
200 | ||
201 | /* | |
202 | * Invalidate the page cache buffers | |
203 | */ | |
204 | truncate_inode_pages(ipimap->i_mapping, 0); | |
205 | ||
206 | /* | |
207 | * free in-memory control structure | |
208 | */ | |
209 | kfree(imap); | |
210 | ||
211 | return (0); | |
212 | } | |
213 | ||
214 | ||
215 | /* | |
216 | * diSync() | |
217 | */ | |
218 | int diSync(struct inode *ipimap) | |
219 | { | |
220 | struct dinomap_disk *dinom_le; | |
221 | struct inomap *imp = JFS_IP(ipimap)->i_imap; | |
222 | struct metapage *mp; | |
223 | int index; | |
224 | ||
225 | /* | |
226 | * write imap global conrol page | |
227 | */ | |
228 | /* read the on-disk inode map control structure */ | |
229 | mp = get_metapage(ipimap, | |
230 | IMAPBLKNO << JFS_SBI(ipimap->i_sb)->l2nbperpage, | |
231 | PSIZE, 0); | |
232 | if (mp == NULL) { | |
233 | jfs_err("diSync: get_metapage failed!"); | |
234 | return -EIO; | |
235 | } | |
236 | ||
237 | /* copy the in-memory version to the on-disk version */ | |
238 | dinom_le = (struct dinomap_disk *) mp->data; | |
239 | dinom_le->in_freeiag = cpu_to_le32(imp->im_freeiag); | |
240 | dinom_le->in_nextiag = cpu_to_le32(imp->im_nextiag); | |
241 | dinom_le->in_numinos = cpu_to_le32(atomic_read(&imp->im_numinos)); | |
242 | dinom_le->in_numfree = cpu_to_le32(atomic_read(&imp->im_numfree)); | |
243 | dinom_le->in_nbperiext = cpu_to_le32(imp->im_nbperiext); | |
244 | dinom_le->in_l2nbperiext = cpu_to_le32(imp->im_l2nbperiext); | |
245 | for (index = 0; index < MAXAG; index++) { | |
246 | dinom_le->in_agctl[index].inofree = | |
247 | cpu_to_le32(imp->im_agctl[index].inofree); | |
248 | dinom_le->in_agctl[index].extfree = | |
249 | cpu_to_le32(imp->im_agctl[index].extfree); | |
250 | dinom_le->in_agctl[index].numinos = | |
251 | cpu_to_le32(imp->im_agctl[index].numinos); | |
252 | dinom_le->in_agctl[index].numfree = | |
253 | cpu_to_le32(imp->im_agctl[index].numfree); | |
254 | } | |
255 | ||
256 | /* write out the control structure */ | |
257 | write_metapage(mp); | |
258 | ||
259 | /* | |
260 | * write out dirty pages of imap | |
261 | */ | |
262 | filemap_fdatawrite(ipimap->i_mapping); | |
263 | filemap_fdatawait(ipimap->i_mapping); | |
264 | ||
265 | diWriteSpecial(ipimap, 0); | |
266 | ||
267 | return (0); | |
268 | } | |
269 | ||
270 | ||
271 | /* | |
272 | * NAME: diRead() | |
273 | * | |
274 | * FUNCTION: initialize an incore inode from disk. | |
275 | * | |
276 | * on entry, the specifed incore inode should itself | |
277 | * specify the disk inode number corresponding to the | |
278 | * incore inode (i.e. i_number should be initialized). | |
279 | * | |
280 | * this routine handles incore inode initialization for | |
281 | * both "special" and "regular" inodes. special inodes | |
282 | * are those required early in the mount process and | |
283 | * require special handling since much of the file system | |
284 | * is not yet initialized. these "special" inodes are | |
285 | * identified by a NULL inode map inode pointer and are | |
286 | * actually initialized by a call to diReadSpecial(). | |
287 | * | |
288 | * for regular inodes, the iag describing the disk inode | |
289 | * is read from disk to determine the inode extent address | |
290 | * for the disk inode. with the inode extent address in | |
291 | * hand, the page of the extent that contains the disk | |
292 | * inode is read and the disk inode is copied to the | |
293 | * incore inode. | |
294 | * | |
295 | * PARAMETERS: | |
296 | * ip - pointer to incore inode to be initialized from disk. | |
297 | * | |
298 | * RETURN VALUES: | |
299 | * 0 - success | |
300 | * -EIO - i/o error. | |
301 | * -ENOMEM - insufficient memory | |
302 | * | |
303 | */ | |
304 | int diRead(struct inode *ip) | |
305 | { | |
306 | struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb); | |
307 | int iagno, ino, extno, rc; | |
308 | struct inode *ipimap; | |
309 | struct dinode *dp; | |
310 | struct iag *iagp; | |
311 | struct metapage *mp; | |
312 | s64 blkno, agstart; | |
313 | struct inomap *imap; | |
314 | int block_offset; | |
315 | int inodes_left; | |
316 | uint pageno; | |
317 | int rel_inode; | |
318 | ||
319 | jfs_info("diRead: ino = %ld", ip->i_ino); | |
320 | ||
321 | ipimap = sbi->ipimap; | |
322 | JFS_IP(ip)->ipimap = ipimap; | |
323 | ||
324 | /* determine the iag number for this inode (number) */ | |
325 | iagno = INOTOIAG(ip->i_ino); | |
326 | ||
327 | /* read the iag */ | |
328 | imap = JFS_IP(ipimap)->i_imap; | |
329 | IREAD_LOCK(ipimap); | |
330 | rc = diIAGRead(imap, iagno, &mp); | |
331 | IREAD_UNLOCK(ipimap); | |
332 | if (rc) { | |
333 | jfs_err("diRead: diIAGRead returned %d", rc); | |
334 | return (rc); | |
335 | } | |
336 | ||
337 | iagp = (struct iag *) mp->data; | |
338 | ||
339 | /* determine inode extent that holds the disk inode */ | |
340 | ino = ip->i_ino & (INOSPERIAG - 1); | |
341 | extno = ino >> L2INOSPEREXT; | |
342 | ||
343 | if ((lengthPXD(&iagp->inoext[extno]) != imap->im_nbperiext) || | |
344 | (addressPXD(&iagp->inoext[extno]) == 0)) { | |
345 | release_metapage(mp); | |
346 | return -ESTALE; | |
347 | } | |
348 | ||
349 | /* get disk block number of the page within the inode extent | |
350 | * that holds the disk inode. | |
351 | */ | |
352 | blkno = INOPBLK(&iagp->inoext[extno], ino, sbi->l2nbperpage); | |
353 | ||
354 | /* get the ag for the iag */ | |
355 | agstart = le64_to_cpu(iagp->agstart); | |
356 | ||
357 | release_metapage(mp); | |
358 | ||
359 | rel_inode = (ino & (INOSPERPAGE - 1)); | |
360 | pageno = blkno >> sbi->l2nbperpage; | |
361 | ||
362 | if ((block_offset = ((u32) blkno & (sbi->nbperpage - 1)))) { | |
363 | /* | |
364 | * OS/2 didn't always align inode extents on page boundaries | |
365 | */ | |
366 | inodes_left = | |
367 | (sbi->nbperpage - block_offset) << sbi->l2niperblk; | |
368 | ||
369 | if (rel_inode < inodes_left) | |
370 | rel_inode += block_offset << sbi->l2niperblk; | |
371 | else { | |
372 | pageno += 1; | |
373 | rel_inode -= inodes_left; | |
374 | } | |
375 | } | |
376 | ||
377 | /* read the page of disk inode */ | |
378 | mp = read_metapage(ipimap, pageno << sbi->l2nbperpage, PSIZE, 1); | |
379 | if (mp == 0) { | |
380 | jfs_err("diRead: read_metapage failed"); | |
381 | return -EIO; | |
382 | } | |
383 | ||
384 | /* locate the the disk inode requested */ | |
385 | dp = (struct dinode *) mp->data; | |
386 | dp += rel_inode; | |
387 | ||
388 | if (ip->i_ino != le32_to_cpu(dp->di_number)) { | |
389 | jfs_error(ip->i_sb, "diRead: i_ino != di_number"); | |
390 | rc = -EIO; | |
391 | } else if (le32_to_cpu(dp->di_nlink) == 0) | |
392 | rc = -ESTALE; | |
393 | else | |
394 | /* copy the disk inode to the in-memory inode */ | |
395 | rc = copy_from_dinode(dp, ip); | |
396 | ||
397 | release_metapage(mp); | |
398 | ||
399 | /* set the ag for the inode */ | |
400 | JFS_IP(ip)->agno = BLKTOAG(agstart, sbi); | |
401 | JFS_IP(ip)->active_ag = -1; | |
402 | ||
403 | return (rc); | |
404 | } | |
405 | ||
406 | ||
407 | /* | |
408 | * NAME: diReadSpecial() | |
409 | * | |
410 | * FUNCTION: initialize a 'special' inode from disk. | |
411 | * | |
412 | * this routines handles aggregate level inodes. The | |
413 | * inode cache cannot differentiate between the | |
414 | * aggregate inodes and the filesystem inodes, so we | |
415 | * handle these here. We don't actually use the aggregate | |
416 | * inode map, since these inodes are at a fixed location | |
417 | * and in some cases the aggregate inode map isn't initialized | |
418 | * yet. | |
419 | * | |
420 | * PARAMETERS: | |
421 | * sb - filesystem superblock | |
422 | * inum - aggregate inode number | |
423 | * secondary - 1 if secondary aggregate inode table | |
424 | * | |
425 | * RETURN VALUES: | |
426 | * new inode - success | |
427 | * NULL - i/o error. | |
428 | */ | |
429 | struct inode *diReadSpecial(struct super_block *sb, ino_t inum, int secondary) | |
430 | { | |
431 | struct jfs_sb_info *sbi = JFS_SBI(sb); | |
432 | uint address; | |
433 | struct dinode *dp; | |
434 | struct inode *ip; | |
435 | struct metapage *mp; | |
436 | ||
437 | ip = new_inode(sb); | |
438 | if (ip == NULL) { | |
439 | jfs_err("diReadSpecial: new_inode returned NULL!"); | |
440 | return ip; | |
441 | } | |
442 | ||
443 | if (secondary) { | |
444 | address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage; | |
445 | JFS_IP(ip)->ipimap = sbi->ipaimap2; | |
446 | } else { | |
447 | address = AITBL_OFF >> L2PSIZE; | |
448 | JFS_IP(ip)->ipimap = sbi->ipaimap; | |
449 | } | |
450 | ||
451 | ASSERT(inum < INOSPEREXT); | |
452 | ||
453 | ip->i_ino = inum; | |
454 | ||
455 | address += inum >> 3; /* 8 inodes per 4K page */ | |
456 | ||
457 | /* read the page of fixed disk inode (AIT) in raw mode */ | |
458 | mp = read_metapage(ip, address << sbi->l2nbperpage, PSIZE, 1); | |
459 | if (mp == NULL) { | |
460 | ip->i_nlink = 1; /* Don't want iput() deleting it */ | |
461 | iput(ip); | |
462 | return (NULL); | |
463 | } | |
464 | ||
465 | /* get the pointer to the disk inode of interest */ | |
466 | dp = (struct dinode *) (mp->data); | |
467 | dp += inum % 8; /* 8 inodes per 4K page */ | |
468 | ||
469 | /* copy on-disk inode to in-memory inode */ | |
470 | if ((copy_from_dinode(dp, ip)) != 0) { | |
471 | /* handle bad return by returning NULL for ip */ | |
472 | ip->i_nlink = 1; /* Don't want iput() deleting it */ | |
473 | iput(ip); | |
474 | /* release the page */ | |
475 | release_metapage(mp); | |
476 | return (NULL); | |
477 | ||
478 | } | |
479 | ||
7fab479b | 480 | ip->i_mapping->a_ops = &jfs_metapage_aops; |
1da177e4 LT |
481 | mapping_set_gfp_mask(ip->i_mapping, GFP_NOFS); |
482 | ||
483 | /* Allocations to metadata inodes should not affect quotas */ | |
484 | ip->i_flags |= S_NOQUOTA; | |
485 | ||
486 | if ((inum == FILESYSTEM_I) && (JFS_IP(ip)->ipimap == sbi->ipaimap)) { | |
487 | sbi->gengen = le32_to_cpu(dp->di_gengen); | |
488 | sbi->inostamp = le32_to_cpu(dp->di_inostamp); | |
489 | } | |
490 | ||
491 | /* release the page */ | |
492 | release_metapage(mp); | |
493 | ||
494 | return (ip); | |
495 | } | |
496 | ||
497 | /* | |
498 | * NAME: diWriteSpecial() | |
499 | * | |
500 | * FUNCTION: Write the special inode to disk | |
501 | * | |
502 | * PARAMETERS: | |
503 | * ip - special inode | |
504 | * secondary - 1 if secondary aggregate inode table | |
505 | * | |
506 | * RETURN VALUES: none | |
507 | */ | |
508 | ||
509 | void diWriteSpecial(struct inode *ip, int secondary) | |
510 | { | |
511 | struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb); | |
512 | uint address; | |
513 | struct dinode *dp; | |
514 | ino_t inum = ip->i_ino; | |
515 | struct metapage *mp; | |
516 | ||
517 | ip->i_state &= ~I_DIRTY; | |
518 | ||
519 | if (secondary) | |
520 | address = addressPXD(&sbi->ait2) >> sbi->l2nbperpage; | |
521 | else | |
522 | address = AITBL_OFF >> L2PSIZE; | |
523 | ||
524 | ASSERT(inum < INOSPEREXT); | |
525 | ||
526 | address += inum >> 3; /* 8 inodes per 4K page */ | |
527 | ||
528 | /* read the page of fixed disk inode (AIT) in raw mode */ | |
529 | mp = read_metapage(ip, address << sbi->l2nbperpage, PSIZE, 1); | |
530 | if (mp == NULL) { | |
531 | jfs_err("diWriteSpecial: failed to read aggregate inode " | |
532 | "extent!"); | |
533 | return; | |
534 | } | |
535 | ||
536 | /* get the pointer to the disk inode of interest */ | |
537 | dp = (struct dinode *) (mp->data); | |
538 | dp += inum % 8; /* 8 inodes per 4K page */ | |
539 | ||
540 | /* copy on-disk inode to in-memory inode */ | |
541 | copy_to_dinode(dp, ip); | |
542 | memcpy(&dp->di_xtroot, &JFS_IP(ip)->i_xtroot, 288); | |
543 | ||
544 | if (inum == FILESYSTEM_I) | |
545 | dp->di_gengen = cpu_to_le32(sbi->gengen); | |
546 | ||
547 | /* write the page */ | |
548 | write_metapage(mp); | |
549 | } | |
550 | ||
551 | /* | |
552 | * NAME: diFreeSpecial() | |
553 | * | |
554 | * FUNCTION: Free allocated space for special inode | |
555 | */ | |
556 | void diFreeSpecial(struct inode *ip) | |
557 | { | |
558 | if (ip == NULL) { | |
559 | jfs_err("diFreeSpecial called with NULL ip!"); | |
560 | return; | |
561 | } | |
562 | filemap_fdatawrite(ip->i_mapping); | |
563 | filemap_fdatawait(ip->i_mapping); | |
564 | truncate_inode_pages(ip->i_mapping, 0); | |
565 | iput(ip); | |
566 | } | |
567 | ||
568 | ||
569 | ||
570 | /* | |
571 | * NAME: diWrite() | |
572 | * | |
573 | * FUNCTION: write the on-disk inode portion of the in-memory inode | |
574 | * to its corresponding on-disk inode. | |
575 | * | |
576 | * on entry, the specifed incore inode should itself | |
577 | * specify the disk inode number corresponding to the | |
578 | * incore inode (i.e. i_number should be initialized). | |
579 | * | |
580 | * the inode contains the inode extent address for the disk | |
581 | * inode. with the inode extent address in hand, the | |
582 | * page of the extent that contains the disk inode is | |
583 | * read and the disk inode portion of the incore inode | |
584 | * is copied to the disk inode. | |
585 | * | |
586 | * PARAMETERS: | |
587 | * tid - transacation id | |
588 | * ip - pointer to incore inode to be written to the inode extent. | |
589 | * | |
590 | * RETURN VALUES: | |
591 | * 0 - success | |
592 | * -EIO - i/o error. | |
593 | */ | |
594 | int diWrite(tid_t tid, struct inode *ip) | |
595 | { | |
596 | struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb); | |
597 | struct jfs_inode_info *jfs_ip = JFS_IP(ip); | |
598 | int rc = 0; | |
599 | s32 ino; | |
600 | struct dinode *dp; | |
601 | s64 blkno; | |
602 | int block_offset; | |
603 | int inodes_left; | |
604 | struct metapage *mp; | |
605 | uint pageno; | |
606 | int rel_inode; | |
607 | int dioffset; | |
608 | struct inode *ipimap; | |
609 | uint type; | |
610 | lid_t lid; | |
611 | struct tlock *ditlck, *tlck; | |
612 | struct linelock *dilinelock, *ilinelock; | |
613 | struct lv *lv; | |
614 | int n; | |
615 | ||
616 | ipimap = jfs_ip->ipimap; | |
617 | ||
618 | ino = ip->i_ino & (INOSPERIAG - 1); | |
619 | ||
620 | if (!addressPXD(&(jfs_ip->ixpxd)) || | |
621 | (lengthPXD(&(jfs_ip->ixpxd)) != | |
622 | JFS_IP(ipimap)->i_imap->im_nbperiext)) { | |
623 | jfs_error(ip->i_sb, "diWrite: ixpxd invalid"); | |
624 | return -EIO; | |
625 | } | |
626 | ||
627 | /* | |
628 | * read the page of disk inode containing the specified inode: | |
629 | */ | |
630 | /* compute the block address of the page */ | |
631 | blkno = INOPBLK(&(jfs_ip->ixpxd), ino, sbi->l2nbperpage); | |
632 | ||
633 | rel_inode = (ino & (INOSPERPAGE - 1)); | |
634 | pageno = blkno >> sbi->l2nbperpage; | |
635 | ||
636 | if ((block_offset = ((u32) blkno & (sbi->nbperpage - 1)))) { | |
637 | /* | |
638 | * OS/2 didn't always align inode extents on page boundaries | |
639 | */ | |
640 | inodes_left = | |
641 | (sbi->nbperpage - block_offset) << sbi->l2niperblk; | |
642 | ||
643 | if (rel_inode < inodes_left) | |
644 | rel_inode += block_offset << sbi->l2niperblk; | |
645 | else { | |
646 | pageno += 1; | |
647 | rel_inode -= inodes_left; | |
648 | } | |
649 | } | |
650 | /* read the page of disk inode */ | |
651 | retry: | |
652 | mp = read_metapage(ipimap, pageno << sbi->l2nbperpage, PSIZE, 1); | |
653 | if (mp == 0) | |
654 | return -EIO; | |
655 | ||
656 | /* get the pointer to the disk inode */ | |
657 | dp = (struct dinode *) mp->data; | |
658 | dp += rel_inode; | |
659 | ||
660 | dioffset = (ino & (INOSPERPAGE - 1)) << L2DISIZE; | |
661 | ||
662 | /* | |
663 | * acquire transaction lock on the on-disk inode; | |
664 | * N.B. tlock is acquired on ipimap not ip; | |
665 | */ | |
666 | if ((ditlck = | |
667 | txLock(tid, ipimap, mp, tlckINODE | tlckENTRY)) == NULL) | |
668 | goto retry; | |
669 | dilinelock = (struct linelock *) & ditlck->lock; | |
670 | ||
671 | /* | |
672 | * copy btree root from in-memory inode to on-disk inode | |
673 | * | |
674 | * (tlock is taken from inline B+-tree root in in-memory | |
675 | * inode when the B+-tree root is updated, which is pointed | |
676 | * by jfs_ip->blid as well as being on tx tlock list) | |
677 | * | |
678 | * further processing of btree root is based on the copy | |
679 | * in in-memory inode, where txLog() will log from, and, | |
680 | * for xtree root, txUpdateMap() will update map and reset | |
681 | * XAD_NEW bit; | |
682 | */ | |
683 | ||
684 | if (S_ISDIR(ip->i_mode) && (lid = jfs_ip->xtlid)) { | |
685 | /* | |
686 | * This is the special xtree inside the directory for storing | |
687 | * the directory table | |
688 | */ | |
689 | xtpage_t *p, *xp; | |
690 | xad_t *xad; | |
691 | ||
692 | jfs_ip->xtlid = 0; | |
693 | tlck = lid_to_tlock(lid); | |
694 | assert(tlck->type & tlckXTREE); | |
695 | tlck->type |= tlckBTROOT; | |
696 | tlck->mp = mp; | |
697 | ilinelock = (struct linelock *) & tlck->lock; | |
698 | ||
699 | /* | |
700 | * copy xtree root from inode to dinode: | |
701 | */ | |
702 | p = &jfs_ip->i_xtroot; | |
703 | xp = (xtpage_t *) &dp->di_dirtable; | |
704 | lv = ilinelock->lv; | |
705 | for (n = 0; n < ilinelock->index; n++, lv++) { | |
706 | memcpy(&xp->xad[lv->offset], &p->xad[lv->offset], | |
707 | lv->length << L2XTSLOTSIZE); | |
708 | } | |
709 | ||
710 | /* reset on-disk (metadata page) xtree XAD_NEW bit */ | |
711 | xad = &xp->xad[XTENTRYSTART]; | |
712 | for (n = XTENTRYSTART; | |
713 | n < le16_to_cpu(xp->header.nextindex); n++, xad++) | |
714 | if (xad->flag & (XAD_NEW | XAD_EXTENDED)) | |
715 | xad->flag &= ~(XAD_NEW | XAD_EXTENDED); | |
716 | } | |
717 | ||
718 | if ((lid = jfs_ip->blid) == 0) | |
719 | goto inlineData; | |
720 | jfs_ip->blid = 0; | |
721 | ||
722 | tlck = lid_to_tlock(lid); | |
723 | type = tlck->type; | |
724 | tlck->type |= tlckBTROOT; | |
725 | tlck->mp = mp; | |
726 | ilinelock = (struct linelock *) & tlck->lock; | |
727 | ||
728 | /* | |
729 | * regular file: 16 byte (XAD slot) granularity | |
730 | */ | |
731 | if (type & tlckXTREE) { | |
732 | xtpage_t *p, *xp; | |
733 | xad_t *xad; | |
734 | ||
735 | /* | |
736 | * copy xtree root from inode to dinode: | |
737 | */ | |
738 | p = &jfs_ip->i_xtroot; | |
739 | xp = &dp->di_xtroot; | |
740 | lv = ilinelock->lv; | |
741 | for (n = 0; n < ilinelock->index; n++, lv++) { | |
742 | memcpy(&xp->xad[lv->offset], &p->xad[lv->offset], | |
743 | lv->length << L2XTSLOTSIZE); | |
744 | } | |
745 | ||
746 | /* reset on-disk (metadata page) xtree XAD_NEW bit */ | |
747 | xad = &xp->xad[XTENTRYSTART]; | |
748 | for (n = XTENTRYSTART; | |
749 | n < le16_to_cpu(xp->header.nextindex); n++, xad++) | |
750 | if (xad->flag & (XAD_NEW | XAD_EXTENDED)) | |
751 | xad->flag &= ~(XAD_NEW | XAD_EXTENDED); | |
752 | } | |
753 | /* | |
754 | * directory: 32 byte (directory entry slot) granularity | |
755 | */ | |
756 | else if (type & tlckDTREE) { | |
757 | dtpage_t *p, *xp; | |
758 | ||
759 | /* | |
760 | * copy dtree root from inode to dinode: | |
761 | */ | |
762 | p = (dtpage_t *) &jfs_ip->i_dtroot; | |
763 | xp = (dtpage_t *) & dp->di_dtroot; | |
764 | lv = ilinelock->lv; | |
765 | for (n = 0; n < ilinelock->index; n++, lv++) { | |
766 | memcpy(&xp->slot[lv->offset], &p->slot[lv->offset], | |
767 | lv->length << L2DTSLOTSIZE); | |
768 | } | |
769 | } else { | |
770 | jfs_err("diWrite: UFO tlock"); | |
771 | } | |
772 | ||
773 | inlineData: | |
774 | /* | |
775 | * copy inline symlink from in-memory inode to on-disk inode | |
776 | */ | |
777 | if (S_ISLNK(ip->i_mode) && ip->i_size < IDATASIZE) { | |
778 | lv = & dilinelock->lv[dilinelock->index]; | |
779 | lv->offset = (dioffset + 2 * 128) >> L2INODESLOTSIZE; | |
780 | lv->length = 2; | |
781 | memcpy(&dp->di_fastsymlink, jfs_ip->i_inline, IDATASIZE); | |
782 | dilinelock->index++; | |
783 | } | |
784 | /* | |
785 | * copy inline data from in-memory inode to on-disk inode: | |
786 | * 128 byte slot granularity | |
787 | */ | |
788 | if (test_cflag(COMMIT_Inlineea, ip)) { | |
789 | lv = & dilinelock->lv[dilinelock->index]; | |
790 | lv->offset = (dioffset + 3 * 128) >> L2INODESLOTSIZE; | |
791 | lv->length = 1; | |
792 | memcpy(&dp->di_inlineea, jfs_ip->i_inline_ea, INODESLOTSIZE); | |
793 | dilinelock->index++; | |
794 | ||
795 | clear_cflag(COMMIT_Inlineea, ip); | |
796 | } | |
797 | ||
798 | /* | |
799 | * lock/copy inode base: 128 byte slot granularity | |
800 | */ | |
801 | // baseDinode: | |
802 | lv = & dilinelock->lv[dilinelock->index]; | |
803 | lv->offset = dioffset >> L2INODESLOTSIZE; | |
804 | copy_to_dinode(dp, ip); | |
805 | if (test_and_clear_cflag(COMMIT_Dirtable, ip)) { | |
806 | lv->length = 2; | |
807 | memcpy(&dp->di_dirtable, &jfs_ip->i_dirtable, 96); | |
808 | } else | |
809 | lv->length = 1; | |
810 | dilinelock->index++; | |
811 | ||
812 | #ifdef _JFS_FASTDASD | |
813 | /* | |
814 | * We aren't logging changes to the DASD used in directory inodes, | |
815 | * but we need to write them to disk. If we don't unmount cleanly, | |
816 | * mount will recalculate the DASD used. | |
817 | */ | |
818 | if (S_ISDIR(ip->i_mode) | |
819 | && (ip->i_ipmnt->i_mntflag & JFS_DASD_ENABLED)) | |
820 | memcpy(&dp->di_DASD, &ip->i_DASD, sizeof(struct dasd)); | |
821 | #endif /* _JFS_FASTDASD */ | |
822 | ||
823 | /* release the buffer holding the updated on-disk inode. | |
824 | * the buffer will be later written by commit processing. | |
825 | */ | |
826 | write_metapage(mp); | |
827 | ||
828 | return (rc); | |
829 | } | |
830 | ||
831 | ||
832 | /* | |
833 | * NAME: diFree(ip) | |
834 | * | |
835 | * FUNCTION: free a specified inode from the inode working map | |
836 | * for a fileset or aggregate. | |
837 | * | |
838 | * if the inode to be freed represents the first (only) | |
839 | * free inode within the iag, the iag will be placed on | |
840 | * the ag free inode list. | |
841 | * | |
842 | * freeing the inode will cause the inode extent to be | |
843 | * freed if the inode is the only allocated inode within | |
844 | * the extent. in this case all the disk resource backing | |
845 | * up the inode extent will be freed. in addition, the iag | |
846 | * will be placed on the ag extent free list if the extent | |
847 | * is the first free extent in the iag. if freeing the | |
848 | * extent also means that no free inodes will exist for | |
849 | * the iag, the iag will also be removed from the ag free | |
850 | * inode list. | |
851 | * | |
852 | * the iag describing the inode will be freed if the extent | |
853 | * is to be freed and it is the only backed extent within | |
854 | * the iag. in this case, the iag will be removed from the | |
855 | * ag free extent list and ag free inode list and placed on | |
856 | * the inode map's free iag list. | |
857 | * | |
858 | * a careful update approach is used to provide consistency | |
859 | * in the face of updates to multiple buffers. under this | |
860 | * approach, all required buffers are obtained before making | |
861 | * any updates and are held until all updates are complete. | |
862 | * | |
863 | * PARAMETERS: | |
864 | * ip - inode to be freed. | |
865 | * | |
866 | * RETURN VALUES: | |
867 | * 0 - success | |
868 | * -EIO - i/o error. | |
869 | */ | |
870 | int diFree(struct inode *ip) | |
871 | { | |
872 | int rc; | |
873 | ino_t inum = ip->i_ino; | |
874 | struct iag *iagp, *aiagp, *biagp, *ciagp, *diagp; | |
875 | struct metapage *mp, *amp, *bmp, *cmp, *dmp; | |
876 | int iagno, ino, extno, bitno, sword, agno; | |
877 | int back, fwd; | |
878 | u32 bitmap, mask; | |
879 | struct inode *ipimap = JFS_SBI(ip->i_sb)->ipimap; | |
880 | struct inomap *imap = JFS_IP(ipimap)->i_imap; | |
881 | pxd_t freepxd; | |
882 | tid_t tid; | |
883 | struct inode *iplist[3]; | |
884 | struct tlock *tlck; | |
885 | struct pxd_lock *pxdlock; | |
886 | ||
887 | /* | |
888 | * This is just to suppress compiler warnings. The same logic that | |
889 | * references these variables is used to initialize them. | |
890 | */ | |
891 | aiagp = biagp = ciagp = diagp = NULL; | |
892 | ||
893 | /* get the iag number containing the inode. | |
894 | */ | |
895 | iagno = INOTOIAG(inum); | |
896 | ||
897 | /* make sure that the iag is contained within | |
898 | * the map. | |
899 | */ | |
900 | if (iagno >= imap->im_nextiag) { | |
901 | dump_mem("imap", imap, 32); | |
902 | jfs_error(ip->i_sb, | |
903 | "diFree: inum = %d, iagno = %d, nextiag = %d", | |
904 | (uint) inum, iagno, imap->im_nextiag); | |
905 | return -EIO; | |
906 | } | |
907 | ||
908 | /* get the allocation group for this ino. | |
909 | */ | |
910 | agno = JFS_IP(ip)->agno; | |
911 | ||
912 | /* Lock the AG specific inode map information | |
913 | */ | |
914 | AG_LOCK(imap, agno); | |
915 | ||
916 | /* Obtain read lock in imap inode. Don't release it until we have | |
917 | * read all of the IAG's that we are going to. | |
918 | */ | |
919 | IREAD_LOCK(ipimap); | |
920 | ||
921 | /* read the iag. | |
922 | */ | |
923 | if ((rc = diIAGRead(imap, iagno, &mp))) { | |
924 | IREAD_UNLOCK(ipimap); | |
925 | AG_UNLOCK(imap, agno); | |
926 | return (rc); | |
927 | } | |
928 | iagp = (struct iag *) mp->data; | |
929 | ||
930 | /* get the inode number and extent number of the inode within | |
931 | * the iag and the inode number within the extent. | |
932 | */ | |
933 | ino = inum & (INOSPERIAG - 1); | |
934 | extno = ino >> L2INOSPEREXT; | |
935 | bitno = ino & (INOSPEREXT - 1); | |
936 | mask = HIGHORDER >> bitno; | |
937 | ||
938 | if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) { | |
939 | jfs_error(ip->i_sb, | |
940 | "diFree: wmap shows inode already free"); | |
941 | } | |
942 | ||
943 | if (!addressPXD(&iagp->inoext[extno])) { | |
944 | release_metapage(mp); | |
945 | IREAD_UNLOCK(ipimap); | |
946 | AG_UNLOCK(imap, agno); | |
947 | jfs_error(ip->i_sb, "diFree: invalid inoext"); | |
948 | return -EIO; | |
949 | } | |
950 | ||
951 | /* compute the bitmap for the extent reflecting the freed inode. | |
952 | */ | |
953 | bitmap = le32_to_cpu(iagp->wmap[extno]) & ~mask; | |
954 | ||
955 | if (imap->im_agctl[agno].numfree > imap->im_agctl[agno].numinos) { | |
956 | release_metapage(mp); | |
957 | IREAD_UNLOCK(ipimap); | |
958 | AG_UNLOCK(imap, agno); | |
959 | jfs_error(ip->i_sb, "diFree: numfree > numinos"); | |
960 | return -EIO; | |
961 | } | |
962 | /* | |
963 | * inode extent still has some inodes or below low water mark: | |
964 | * keep the inode extent; | |
965 | */ | |
966 | if (bitmap || | |
967 | imap->im_agctl[agno].numfree < 96 || | |
968 | (imap->im_agctl[agno].numfree < 288 && | |
969 | (((imap->im_agctl[agno].numfree * 100) / | |
970 | imap->im_agctl[agno].numinos) <= 25))) { | |
971 | /* if the iag currently has no free inodes (i.e., | |
972 | * the inode being freed is the first free inode of iag), | |
973 | * insert the iag at head of the inode free list for the ag. | |
974 | */ | |
975 | if (iagp->nfreeinos == 0) { | |
976 | /* check if there are any iags on the ag inode | |
977 | * free list. if so, read the first one so that | |
978 | * we can link the current iag onto the list at | |
979 | * the head. | |
980 | */ | |
981 | if ((fwd = imap->im_agctl[agno].inofree) >= 0) { | |
982 | /* read the iag that currently is the head | |
983 | * of the list. | |
984 | */ | |
985 | if ((rc = diIAGRead(imap, fwd, &))) { | |
986 | IREAD_UNLOCK(ipimap); | |
987 | AG_UNLOCK(imap, agno); | |
988 | release_metapage(mp); | |
989 | return (rc); | |
990 | } | |
991 | aiagp = (struct iag *) amp->data; | |
992 | ||
993 | /* make current head point back to the iag. | |
994 | */ | |
995 | aiagp->inofreeback = cpu_to_le32(iagno); | |
996 | ||
997 | write_metapage(amp); | |
998 | } | |
999 | ||
1000 | /* iag points forward to current head and iag | |
1001 | * becomes the new head of the list. | |
1002 | */ | |
1003 | iagp->inofreefwd = | |
1004 | cpu_to_le32(imap->im_agctl[agno].inofree); | |
1005 | iagp->inofreeback = cpu_to_le32(-1); | |
1006 | imap->im_agctl[agno].inofree = iagno; | |
1007 | } | |
1008 | IREAD_UNLOCK(ipimap); | |
1009 | ||
1010 | /* update the free inode summary map for the extent if | |
1011 | * freeing the inode means the extent will now have free | |
1012 | * inodes (i.e., the inode being freed is the first free | |
1013 | * inode of extent), | |
1014 | */ | |
1015 | if (iagp->wmap[extno] == cpu_to_le32(ONES)) { | |
1016 | sword = extno >> L2EXTSPERSUM; | |
1017 | bitno = extno & (EXTSPERSUM - 1); | |
1018 | iagp->inosmap[sword] &= | |
1019 | cpu_to_le32(~(HIGHORDER >> bitno)); | |
1020 | } | |
1021 | ||
1022 | /* update the bitmap. | |
1023 | */ | |
1024 | iagp->wmap[extno] = cpu_to_le32(bitmap); | |
1da177e4 LT |
1025 | |
1026 | /* update the free inode counts at the iag, ag and | |
1027 | * map level. | |
1028 | */ | |
1029 | iagp->nfreeinos = | |
1030 | cpu_to_le32(le32_to_cpu(iagp->nfreeinos) + 1); | |
1031 | imap->im_agctl[agno].numfree += 1; | |
1032 | atomic_inc(&imap->im_numfree); | |
1033 | ||
1034 | /* release the AG inode map lock | |
1035 | */ | |
1036 | AG_UNLOCK(imap, agno); | |
1037 | ||
1038 | /* write the iag */ | |
1039 | write_metapage(mp); | |
1040 | ||
1041 | return (0); | |
1042 | } | |
1043 | ||
1044 | ||
1045 | /* | |
1046 | * inode extent has become free and above low water mark: | |
1047 | * free the inode extent; | |
1048 | */ | |
1049 | ||
1050 | /* | |
1051 | * prepare to update iag list(s) (careful update step 1) | |
1052 | */ | |
1053 | amp = bmp = cmp = dmp = NULL; | |
1054 | fwd = back = -1; | |
1055 | ||
1056 | /* check if the iag currently has no free extents. if so, | |
1057 | * it will be placed on the head of the ag extent free list. | |
1058 | */ | |
1059 | if (iagp->nfreeexts == 0) { | |
1060 | /* check if the ag extent free list has any iags. | |
1061 | * if so, read the iag at the head of the list now. | |
1062 | * this (head) iag will be updated later to reflect | |
1063 | * the addition of the current iag at the head of | |
1064 | * the list. | |
1065 | */ | |
1066 | if ((fwd = imap->im_agctl[agno].extfree) >= 0) { | |
1067 | if ((rc = diIAGRead(imap, fwd, &))) | |
1068 | goto error_out; | |
1069 | aiagp = (struct iag *) amp->data; | |
1070 | } | |
1071 | } else { | |
1072 | /* iag has free extents. check if the addition of a free | |
1073 | * extent will cause all extents to be free within this | |
1074 | * iag. if so, the iag will be removed from the ag extent | |
1075 | * free list and placed on the inode map's free iag list. | |
1076 | */ | |
1077 | if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) { | |
1078 | /* in preparation for removing the iag from the | |
1079 | * ag extent free list, read the iags preceeding | |
1080 | * and following the iag on the ag extent free | |
1081 | * list. | |
1082 | */ | |
1083 | if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) { | |
1084 | if ((rc = diIAGRead(imap, fwd, &))) | |
1085 | goto error_out; | |
1086 | aiagp = (struct iag *) amp->data; | |
1087 | } | |
1088 | ||
1089 | if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) { | |
1090 | if ((rc = diIAGRead(imap, back, &bmp))) | |
1091 | goto error_out; | |
1092 | biagp = (struct iag *) bmp->data; | |
1093 | } | |
1094 | } | |
1095 | } | |
1096 | ||
1097 | /* remove the iag from the ag inode free list if freeing | |
1098 | * this extent cause the iag to have no free inodes. | |
1099 | */ | |
1100 | if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) { | |
1101 | int inofreeback = le32_to_cpu(iagp->inofreeback); | |
1102 | int inofreefwd = le32_to_cpu(iagp->inofreefwd); | |
1103 | ||
1104 | /* in preparation for removing the iag from the | |
1105 | * ag inode free list, read the iags preceeding | |
1106 | * and following the iag on the ag inode free | |
1107 | * list. before reading these iags, we must make | |
1108 | * sure that we already don't have them in hand | |
1109 | * from up above, since re-reading an iag (buffer) | |
1110 | * we are currently holding would cause a deadlock. | |
1111 | */ | |
1112 | if (inofreefwd >= 0) { | |
1113 | ||
1114 | if (inofreefwd == fwd) | |
1115 | ciagp = (struct iag *) amp->data; | |
1116 | else if (inofreefwd == back) | |
1117 | ciagp = (struct iag *) bmp->data; | |
1118 | else { | |
1119 | if ((rc = | |
1120 | diIAGRead(imap, inofreefwd, &cmp))) | |
1121 | goto error_out; | |
1122 | ciagp = (struct iag *) cmp->data; | |
1123 | } | |
1124 | assert(ciagp != NULL); | |
1125 | } | |
1126 | ||
1127 | if (inofreeback >= 0) { | |
1128 | if (inofreeback == fwd) | |
1129 | diagp = (struct iag *) amp->data; | |
1130 | else if (inofreeback == back) | |
1131 | diagp = (struct iag *) bmp->data; | |
1132 | else { | |
1133 | if ((rc = | |
1134 | diIAGRead(imap, inofreeback, &dmp))) | |
1135 | goto error_out; | |
1136 | diagp = (struct iag *) dmp->data; | |
1137 | } | |
1138 | assert(diagp != NULL); | |
1139 | } | |
1140 | } | |
1141 | ||
1142 | IREAD_UNLOCK(ipimap); | |
1143 | ||
1144 | /* | |
1145 | * invalidate any page of the inode extent freed from buffer cache; | |
1146 | */ | |
1147 | freepxd = iagp->inoext[extno]; | |
1148 | invalidate_pxd_metapages(ip, freepxd); | |
1149 | ||
1150 | /* | |
1151 | * update iag list(s) (careful update step 2) | |
1152 | */ | |
1153 | /* add the iag to the ag extent free list if this is the | |
1154 | * first free extent for the iag. | |
1155 | */ | |
1156 | if (iagp->nfreeexts == 0) { | |
1157 | if (fwd >= 0) | |
1158 | aiagp->extfreeback = cpu_to_le32(iagno); | |
1159 | ||
1160 | iagp->extfreefwd = | |
1161 | cpu_to_le32(imap->im_agctl[agno].extfree); | |
1162 | iagp->extfreeback = cpu_to_le32(-1); | |
1163 | imap->im_agctl[agno].extfree = iagno; | |
1164 | } else { | |
1165 | /* remove the iag from the ag extent list if all extents | |
1166 | * are now free and place it on the inode map iag free list. | |
1167 | */ | |
1168 | if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG - 1)) { | |
1169 | if (fwd >= 0) | |
1170 | aiagp->extfreeback = iagp->extfreeback; | |
1171 | ||
1172 | if (back >= 0) | |
1173 | biagp->extfreefwd = iagp->extfreefwd; | |
1174 | else | |
1175 | imap->im_agctl[agno].extfree = | |
1176 | le32_to_cpu(iagp->extfreefwd); | |
1177 | ||
1178 | iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1); | |
1179 | ||
1180 | IAGFREE_LOCK(imap); | |
1181 | iagp->iagfree = cpu_to_le32(imap->im_freeiag); | |
1182 | imap->im_freeiag = iagno; | |
1183 | IAGFREE_UNLOCK(imap); | |
1184 | } | |
1185 | } | |
1186 | ||
1187 | /* remove the iag from the ag inode free list if freeing | |
1188 | * this extent causes the iag to have no free inodes. | |
1189 | */ | |
1190 | if (iagp->nfreeinos == cpu_to_le32(INOSPEREXT - 1)) { | |
1191 | if ((int) le32_to_cpu(iagp->inofreefwd) >= 0) | |
1192 | ciagp->inofreeback = iagp->inofreeback; | |
1193 | ||
1194 | if ((int) le32_to_cpu(iagp->inofreeback) >= 0) | |
1195 | diagp->inofreefwd = iagp->inofreefwd; | |
1196 | else | |
1197 | imap->im_agctl[agno].inofree = | |
1198 | le32_to_cpu(iagp->inofreefwd); | |
1199 | ||
1200 | iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1); | |
1201 | } | |
1202 | ||
1203 | /* update the inode extent address and working map | |
1204 | * to reflect the free extent. | |
1205 | * the permanent map should have been updated already | |
1206 | * for the inode being freed. | |
1207 | */ | |
1208 | if (iagp->pmap[extno] != 0) { | |
1209 | jfs_error(ip->i_sb, "diFree: the pmap does not show inode free"); | |
1210 | } | |
1211 | iagp->wmap[extno] = 0; | |
1da177e4 LT |
1212 | PXDlength(&iagp->inoext[extno], 0); |
1213 | PXDaddress(&iagp->inoext[extno], 0); | |
1214 | ||
1215 | /* update the free extent and free inode summary maps | |
1216 | * to reflect the freed extent. | |
1217 | * the inode summary map is marked to indicate no inodes | |
1218 | * available for the freed extent. | |
1219 | */ | |
1220 | sword = extno >> L2EXTSPERSUM; | |
1221 | bitno = extno & (EXTSPERSUM - 1); | |
1222 | mask = HIGHORDER >> bitno; | |
1223 | iagp->inosmap[sword] |= cpu_to_le32(mask); | |
1224 | iagp->extsmap[sword] &= cpu_to_le32(~mask); | |
1225 | ||
1226 | /* update the number of free inodes and number of free extents | |
1227 | * for the iag. | |
1228 | */ | |
1229 | iagp->nfreeinos = cpu_to_le32(le32_to_cpu(iagp->nfreeinos) - | |
1230 | (INOSPEREXT - 1)); | |
1231 | iagp->nfreeexts = cpu_to_le32(le32_to_cpu(iagp->nfreeexts) + 1); | |
1232 | ||
1233 | /* update the number of free inodes and backed inodes | |
1234 | * at the ag and inode map level. | |
1235 | */ | |
1236 | imap->im_agctl[agno].numfree -= (INOSPEREXT - 1); | |
1237 | imap->im_agctl[agno].numinos -= INOSPEREXT; | |
1238 | atomic_sub(INOSPEREXT - 1, &imap->im_numfree); | |
1239 | atomic_sub(INOSPEREXT, &imap->im_numinos); | |
1240 | ||
1241 | if (amp) | |
1242 | write_metapage(amp); | |
1243 | if (bmp) | |
1244 | write_metapage(bmp); | |
1245 | if (cmp) | |
1246 | write_metapage(cmp); | |
1247 | if (dmp) | |
1248 | write_metapage(dmp); | |
1249 | ||
1250 | /* | |
1251 | * start transaction to update block allocation map | |
1252 | * for the inode extent freed; | |
1253 | * | |
1254 | * N.B. AG_LOCK is released and iag will be released below, and | |
1255 | * other thread may allocate inode from/reusing the ixad freed | |
1256 | * BUT with new/different backing inode extent from the extent | |
1257 | * to be freed by the transaction; | |
1258 | */ | |
1259 | tid = txBegin(ipimap->i_sb, COMMIT_FORCE); | |
1260 | down(&JFS_IP(ipimap)->commit_sem); | |
1261 | ||
1262 | /* acquire tlock of the iag page of the freed ixad | |
1263 | * to force the page NOHOMEOK (even though no data is | |
1264 | * logged from the iag page) until NOREDOPAGE|FREEXTENT log | |
1265 | * for the free of the extent is committed; | |
1266 | * write FREEXTENT|NOREDOPAGE log record | |
1267 | * N.B. linelock is overlaid as freed extent descriptor; | |
1268 | */ | |
1269 | tlck = txLock(tid, ipimap, mp, tlckINODE | tlckFREE); | |
1270 | pxdlock = (struct pxd_lock *) & tlck->lock; | |
1271 | pxdlock->flag = mlckFREEPXD; | |
1272 | pxdlock->pxd = freepxd; | |
1273 | pxdlock->index = 1; | |
1274 | ||
1275 | write_metapage(mp); | |
1276 | ||
1277 | iplist[0] = ipimap; | |
1278 | ||
1279 | /* | |
1280 | * logredo needs the IAG number and IAG extent index in order | |
1281 | * to ensure that the IMap is consistent. The least disruptive | |
1282 | * way to pass these values through to the transaction manager | |
1283 | * is in the iplist array. | |
1284 | * | |
1285 | * It's not pretty, but it works. | |
1286 | */ | |
1287 | iplist[1] = (struct inode *) (size_t)iagno; | |
1288 | iplist[2] = (struct inode *) (size_t)extno; | |
1289 | ||
1290 | rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE); | |
1291 | ||
1292 | txEnd(tid); | |
1293 | up(&JFS_IP(ipimap)->commit_sem); | |
1294 | ||
1295 | /* unlock the AG inode map information */ | |
1296 | AG_UNLOCK(imap, agno); | |
1297 | ||
1298 | return (0); | |
1299 | ||
1300 | error_out: | |
1301 | IREAD_UNLOCK(ipimap); | |
1302 | ||
1303 | if (amp) | |
1304 | release_metapage(amp); | |
1305 | if (bmp) | |
1306 | release_metapage(bmp); | |
1307 | if (cmp) | |
1308 | release_metapage(cmp); | |
1309 | if (dmp) | |
1310 | release_metapage(dmp); | |
1311 | ||
1312 | AG_UNLOCK(imap, agno); | |
1313 | ||
1314 | release_metapage(mp); | |
1315 | ||
1316 | return (rc); | |
1317 | } | |
1318 | ||
1319 | /* | |
1320 | * There are several places in the diAlloc* routines where we initialize | |
1321 | * the inode. | |
1322 | */ | |
1323 | static inline void | |
1324 | diInitInode(struct inode *ip, int iagno, int ino, int extno, struct iag * iagp) | |
1325 | { | |
1326 | struct jfs_sb_info *sbi = JFS_SBI(ip->i_sb); | |
1327 | struct jfs_inode_info *jfs_ip = JFS_IP(ip); | |
1328 | ||
1329 | ip->i_ino = (iagno << L2INOSPERIAG) + ino; | |
1da177e4 LT |
1330 | jfs_ip->ixpxd = iagp->inoext[extno]; |
1331 | jfs_ip->agno = BLKTOAG(le64_to_cpu(iagp->agstart), sbi); | |
1332 | jfs_ip->active_ag = -1; | |
1333 | } | |
1334 | ||
1335 | ||
1336 | /* | |
1337 | * NAME: diAlloc(pip,dir,ip) | |
1338 | * | |
1339 | * FUNCTION: allocate a disk inode from the inode working map | |
1340 | * for a fileset or aggregate. | |
1341 | * | |
1342 | * PARAMETERS: | |
1343 | * pip - pointer to incore inode for the parent inode. | |
1344 | * dir - TRUE if the new disk inode is for a directory. | |
1345 | * ip - pointer to a new inode | |
1346 | * | |
1347 | * RETURN VALUES: | |
1348 | * 0 - success. | |
1349 | * -ENOSPC - insufficient disk resources. | |
1350 | * -EIO - i/o error. | |
1351 | */ | |
1352 | int diAlloc(struct inode *pip, boolean_t dir, struct inode *ip) | |
1353 | { | |
1354 | int rc, ino, iagno, addext, extno, bitno, sword; | |
1355 | int nwords, rem, i, agno; | |
1356 | u32 mask, inosmap, extsmap; | |
1357 | struct inode *ipimap; | |
1358 | struct metapage *mp; | |
1359 | ino_t inum; | |
1360 | struct iag *iagp; | |
1361 | struct inomap *imap; | |
1362 | ||
1363 | /* get the pointers to the inode map inode and the | |
1364 | * corresponding imap control structure. | |
1365 | */ | |
1366 | ipimap = JFS_SBI(pip->i_sb)->ipimap; | |
1367 | imap = JFS_IP(ipimap)->i_imap; | |
1368 | JFS_IP(ip)->ipimap = ipimap; | |
1369 | JFS_IP(ip)->fileset = FILESYSTEM_I; | |
1370 | ||
1371 | /* for a directory, the allocation policy is to start | |
1372 | * at the ag level using the preferred ag. | |
1373 | */ | |
1374 | if (dir == TRUE) { | |
1375 | agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap); | |
1376 | AG_LOCK(imap, agno); | |
1377 | goto tryag; | |
1378 | } | |
1379 | ||
1380 | /* for files, the policy starts off by trying to allocate from | |
1381 | * the same iag containing the parent disk inode: | |
1382 | * try to allocate the new disk inode close to the parent disk | |
1383 | * inode, using parent disk inode number + 1 as the allocation | |
1384 | * hint. (we use a left-to-right policy to attempt to avoid | |
1385 | * moving backward on the disk.) compute the hint within the | |
1386 | * file system and the iag. | |
1387 | */ | |
1388 | ||
1389 | /* get the ag number of this iag */ | |
1390 | agno = JFS_IP(pip)->agno; | |
1391 | ||
1392 | if (atomic_read(&JFS_SBI(pip->i_sb)->bmap->db_active[agno])) { | |
1393 | /* | |
1394 | * There is an open file actively growing. We want to | |
1395 | * allocate new inodes from a different ag to avoid | |
1396 | * fragmentation problems. | |
1397 | */ | |
1398 | agno = dbNextAG(JFS_SBI(pip->i_sb)->ipbmap); | |
1399 | AG_LOCK(imap, agno); | |
1400 | goto tryag; | |
1401 | } | |
1402 | ||
1403 | inum = pip->i_ino + 1; | |
1404 | ino = inum & (INOSPERIAG - 1); | |
1405 | ||
1406 | /* back off the the hint if it is outside of the iag */ | |
1407 | if (ino == 0) | |
1408 | inum = pip->i_ino; | |
1409 | ||
1410 | /* lock the AG inode map information */ | |
1411 | AG_LOCK(imap, agno); | |
1412 | ||
1413 | /* Get read lock on imap inode */ | |
1414 | IREAD_LOCK(ipimap); | |
1415 | ||
1416 | /* get the iag number and read the iag */ | |
1417 | iagno = INOTOIAG(inum); | |
1418 | if ((rc = diIAGRead(imap, iagno, &mp))) { | |
1419 | IREAD_UNLOCK(ipimap); | |
1420 | AG_UNLOCK(imap, agno); | |
1421 | return (rc); | |
1422 | } | |
1423 | iagp = (struct iag *) mp->data; | |
1424 | ||
1425 | /* determine if new inode extent is allowed to be added to the iag. | |
1426 | * new inode extent can be added to the iag if the ag | |
1427 | * has less than 32 free disk inodes and the iag has free extents. | |
1428 | */ | |
1429 | addext = (imap->im_agctl[agno].numfree < 32 && iagp->nfreeexts); | |
1430 | ||
1431 | /* | |
1432 | * try to allocate from the IAG | |
1433 | */ | |
1434 | /* check if the inode may be allocated from the iag | |
1435 | * (i.e. the inode has free inodes or new extent can be added). | |
1436 | */ | |
1437 | if (iagp->nfreeinos || addext) { | |
1438 | /* determine the extent number of the hint. | |
1439 | */ | |
1440 | extno = ino >> L2INOSPEREXT; | |
1441 | ||
1442 | /* check if the extent containing the hint has backed | |
1443 | * inodes. if so, try to allocate within this extent. | |
1444 | */ | |
1445 | if (addressPXD(&iagp->inoext[extno])) { | |
1446 | bitno = ino & (INOSPEREXT - 1); | |
1447 | if ((bitno = | |
1448 | diFindFree(le32_to_cpu(iagp->wmap[extno]), | |
1449 | bitno)) | |
1450 | < INOSPEREXT) { | |
1451 | ino = (extno << L2INOSPEREXT) + bitno; | |
1452 | ||
1453 | /* a free inode (bit) was found within this | |
1454 | * extent, so allocate it. | |
1455 | */ | |
1456 | rc = diAllocBit(imap, iagp, ino); | |
1457 | IREAD_UNLOCK(ipimap); | |
1458 | if (rc) { | |
1459 | assert(rc == -EIO); | |
1460 | } else { | |
1461 | /* set the results of the allocation | |
1462 | * and write the iag. | |
1463 | */ | |
1464 | diInitInode(ip, iagno, ino, extno, | |
1465 | iagp); | |
1466 | mark_metapage_dirty(mp); | |
1467 | } | |
1468 | release_metapage(mp); | |
1469 | ||
1470 | /* free the AG lock and return. | |
1471 | */ | |
1472 | AG_UNLOCK(imap, agno); | |
1473 | return (rc); | |
1474 | } | |
1475 | ||
1476 | if (!addext) | |
1477 | extno = | |
1478 | (extno == | |
1479 | EXTSPERIAG - 1) ? 0 : extno + 1; | |
1480 | } | |
1481 | ||
1482 | /* | |
1483 | * no free inodes within the extent containing the hint. | |
1484 | * | |
1485 | * try to allocate from the backed extents following | |
1486 | * hint or, if appropriate (i.e. addext is true), allocate | |
1487 | * an extent of free inodes at or following the extent | |
1488 | * containing the hint. | |
1489 | * | |
1490 | * the free inode and free extent summary maps are used | |
1491 | * here, so determine the starting summary map position | |
1492 | * and the number of words we'll have to examine. again, | |
1493 | * the approach is to allocate following the hint, so we | |
1494 | * might have to initially ignore prior bits of the summary | |
1495 | * map that represent extents prior to the extent containing | |
1496 | * the hint and later revisit these bits. | |
1497 | */ | |
1498 | bitno = extno & (EXTSPERSUM - 1); | |
1499 | nwords = (bitno == 0) ? SMAPSZ : SMAPSZ + 1; | |
1500 | sword = extno >> L2EXTSPERSUM; | |
1501 | ||
1502 | /* mask any prior bits for the starting words of the | |
1503 | * summary map. | |
1504 | */ | |
1505 | mask = ONES << (EXTSPERSUM - bitno); | |
1506 | inosmap = le32_to_cpu(iagp->inosmap[sword]) | mask; | |
1507 | extsmap = le32_to_cpu(iagp->extsmap[sword]) | mask; | |
1508 | ||
1509 | /* scan the free inode and free extent summary maps for | |
1510 | * free resources. | |
1511 | */ | |
1512 | for (i = 0; i < nwords; i++) { | |
1513 | /* check if this word of the free inode summary | |
1514 | * map describes an extent with free inodes. | |
1515 | */ | |
1516 | if (~inosmap) { | |
1517 | /* an extent with free inodes has been | |
1518 | * found. determine the extent number | |
1519 | * and the inode number within the extent. | |
1520 | */ | |
1521 | rem = diFindFree(inosmap, 0); | |
1522 | extno = (sword << L2EXTSPERSUM) + rem; | |
1523 | rem = diFindFree(le32_to_cpu(iagp->wmap[extno]), | |
1524 | 0); | |
1525 | if (rem >= INOSPEREXT) { | |
1526 | IREAD_UNLOCK(ipimap); | |
1527 | release_metapage(mp); | |
1528 | AG_UNLOCK(imap, agno); | |
1529 | jfs_error(ip->i_sb, | |
1530 | "diAlloc: can't find free bit " | |
1531 | "in wmap"); | |
1532 | return EIO; | |
1533 | } | |
1534 | ||
1535 | /* determine the inode number within the | |
1536 | * iag and allocate the inode from the | |
1537 | * map. | |
1538 | */ | |
1539 | ino = (extno << L2INOSPEREXT) + rem; | |
1540 | rc = diAllocBit(imap, iagp, ino); | |
1541 | IREAD_UNLOCK(ipimap); | |
1542 | if (rc) | |
1543 | assert(rc == -EIO); | |
1544 | else { | |
1545 | /* set the results of the allocation | |
1546 | * and write the iag. | |
1547 | */ | |
1548 | diInitInode(ip, iagno, ino, extno, | |
1549 | iagp); | |
1550 | mark_metapage_dirty(mp); | |
1551 | } | |
1552 | release_metapage(mp); | |
1553 | ||
1554 | /* free the AG lock and return. | |
1555 | */ | |
1556 | AG_UNLOCK(imap, agno); | |
1557 | return (rc); | |
1558 | ||
1559 | } | |
1560 | ||
1561 | /* check if we may allocate an extent of free | |
1562 | * inodes and whether this word of the free | |
1563 | * extents summary map describes a free extent. | |
1564 | */ | |
1565 | if (addext && ~extsmap) { | |
1566 | /* a free extent has been found. determine | |
1567 | * the extent number. | |
1568 | */ | |
1569 | rem = diFindFree(extsmap, 0); | |
1570 | extno = (sword << L2EXTSPERSUM) + rem; | |
1571 | ||
1572 | /* allocate an extent of free inodes. | |
1573 | */ | |
1574 | if ((rc = diNewExt(imap, iagp, extno))) { | |
1575 | /* if there is no disk space for a | |
1576 | * new extent, try to allocate the | |
1577 | * disk inode from somewhere else. | |
1578 | */ | |
1579 | if (rc == -ENOSPC) | |
1580 | break; | |
1581 | ||
1582 | assert(rc == -EIO); | |
1583 | } else { | |
1584 | /* set the results of the allocation | |
1585 | * and write the iag. | |
1586 | */ | |
1587 | diInitInode(ip, iagno, | |
1588 | extno << L2INOSPEREXT, | |
1589 | extno, iagp); | |
1590 | mark_metapage_dirty(mp); | |
1591 | } | |
1592 | release_metapage(mp); | |
1593 | /* free the imap inode & the AG lock & return. | |
1594 | */ | |
1595 | IREAD_UNLOCK(ipimap); | |
1596 | AG_UNLOCK(imap, agno); | |
1597 | return (rc); | |
1598 | } | |
1599 | ||
1600 | /* move on to the next set of summary map words. | |
1601 | */ | |
1602 | sword = (sword == SMAPSZ - 1) ? 0 : sword + 1; | |
1603 | inosmap = le32_to_cpu(iagp->inosmap[sword]); | |
1604 | extsmap = le32_to_cpu(iagp->extsmap[sword]); | |
1605 | } | |
1606 | } | |
1607 | /* unlock imap inode */ | |
1608 | IREAD_UNLOCK(ipimap); | |
1609 | ||
1610 | /* nothing doing in this iag, so release it. */ | |
1611 | release_metapage(mp); | |
1612 | ||
1613 | tryag: | |
1614 | /* | |
1615 | * try to allocate anywhere within the same AG as the parent inode. | |
1616 | */ | |
1617 | rc = diAllocAG(imap, agno, dir, ip); | |
1618 | ||
1619 | AG_UNLOCK(imap, agno); | |
1620 | ||
1621 | if (rc != -ENOSPC) | |
1622 | return (rc); | |
1623 | ||
1624 | /* | |
1625 | * try to allocate in any AG. | |
1626 | */ | |
1627 | return (diAllocAny(imap, agno, dir, ip)); | |
1628 | } | |
1629 | ||
1630 | ||
1631 | /* | |
1632 | * NAME: diAllocAG(imap,agno,dir,ip) | |
1633 | * | |
1634 | * FUNCTION: allocate a disk inode from the allocation group. | |
1635 | * | |
1636 | * this routine first determines if a new extent of free | |
1637 | * inodes should be added for the allocation group, with | |
1638 | * the current request satisfied from this extent. if this | |
1639 | * is the case, an attempt will be made to do just that. if | |
1640 | * this attempt fails or it has been determined that a new | |
1641 | * extent should not be added, an attempt is made to satisfy | |
1642 | * the request by allocating an existing (backed) free inode | |
1643 | * from the allocation group. | |
1644 | * | |
1645 | * PRE CONDITION: Already have the AG lock for this AG. | |
1646 | * | |
1647 | * PARAMETERS: | |
1648 | * imap - pointer to inode map control structure. | |
1649 | * agno - allocation group to allocate from. | |
1650 | * dir - TRUE if the new disk inode is for a directory. | |
1651 | * ip - pointer to the new inode to be filled in on successful return | |
1652 | * with the disk inode number allocated, its extent address | |
1653 | * and the start of the ag. | |
1654 | * | |
1655 | * RETURN VALUES: | |
1656 | * 0 - success. | |
1657 | * -ENOSPC - insufficient disk resources. | |
1658 | * -EIO - i/o error. | |
1659 | */ | |
1660 | static int | |
1661 | diAllocAG(struct inomap * imap, int agno, boolean_t dir, struct inode *ip) | |
1662 | { | |
1663 | int rc, addext, numfree, numinos; | |
1664 | ||
1665 | /* get the number of free and the number of backed disk | |
1666 | * inodes currently within the ag. | |
1667 | */ | |
1668 | numfree = imap->im_agctl[agno].numfree; | |
1669 | numinos = imap->im_agctl[agno].numinos; | |
1670 | ||
1671 | if (numfree > numinos) { | |
1672 | jfs_error(ip->i_sb, "diAllocAG: numfree > numinos"); | |
1673 | return -EIO; | |
1674 | } | |
1675 | ||
1676 | /* determine if we should allocate a new extent of free inodes | |
1677 | * within the ag: for directory inodes, add a new extent | |
1678 | * if there are a small number of free inodes or number of free | |
1679 | * inodes is a small percentage of the number of backed inodes. | |
1680 | */ | |
1681 | if (dir == TRUE) | |
1682 | addext = (numfree < 64 || | |
1683 | (numfree < 256 | |
1684 | && ((numfree * 100) / numinos) <= 20)); | |
1685 | else | |
1686 | addext = (numfree == 0); | |
1687 | ||
1688 | /* | |
1689 | * try to allocate a new extent of free inodes. | |
1690 | */ | |
1691 | if (addext) { | |
1692 | /* if free space is not avaliable for this new extent, try | |
1693 | * below to allocate a free and existing (already backed) | |
1694 | * inode from the ag. | |
1695 | */ | |
1696 | if ((rc = diAllocExt(imap, agno, ip)) != -ENOSPC) | |
1697 | return (rc); | |
1698 | } | |
1699 | ||
1700 | /* | |
1701 | * try to allocate an existing free inode from the ag. | |
1702 | */ | |
1703 | return (diAllocIno(imap, agno, ip)); | |
1704 | } | |
1705 | ||
1706 | ||
1707 | /* | |
1708 | * NAME: diAllocAny(imap,agno,dir,iap) | |
1709 | * | |
1710 | * FUNCTION: allocate a disk inode from any other allocation group. | |
1711 | * | |
1712 | * this routine is called when an allocation attempt within | |
1713 | * the primary allocation group has failed. if attempts to | |
1714 | * allocate an inode from any allocation group other than the | |
1715 | * specified primary group. | |
1716 | * | |
1717 | * PARAMETERS: | |
1718 | * imap - pointer to inode map control structure. | |
1719 | * agno - primary allocation group (to avoid). | |
1720 | * dir - TRUE if the new disk inode is for a directory. | |
1721 | * ip - pointer to a new inode to be filled in on successful return | |
1722 | * with the disk inode number allocated, its extent address | |
1723 | * and the start of the ag. | |
1724 | * | |
1725 | * RETURN VALUES: | |
1726 | * 0 - success. | |
1727 | * -ENOSPC - insufficient disk resources. | |
1728 | * -EIO - i/o error. | |
1729 | */ | |
1730 | static int | |
1731 | diAllocAny(struct inomap * imap, int agno, boolean_t dir, struct inode *ip) | |
1732 | { | |
1733 | int ag, rc; | |
1734 | int maxag = JFS_SBI(imap->im_ipimap->i_sb)->bmap->db_maxag; | |
1735 | ||
1736 | ||
1737 | /* try to allocate from the ags following agno up to | |
1738 | * the maximum ag number. | |
1739 | */ | |
1740 | for (ag = agno + 1; ag <= maxag; ag++) { | |
1741 | AG_LOCK(imap, ag); | |
1742 | ||
1743 | rc = diAllocAG(imap, ag, dir, ip); | |
1744 | ||
1745 | AG_UNLOCK(imap, ag); | |
1746 | ||
1747 | if (rc != -ENOSPC) | |
1748 | return (rc); | |
1749 | } | |
1750 | ||
1751 | /* try to allocate from the ags in front of agno. | |
1752 | */ | |
1753 | for (ag = 0; ag < agno; ag++) { | |
1754 | AG_LOCK(imap, ag); | |
1755 | ||
1756 | rc = diAllocAG(imap, ag, dir, ip); | |
1757 | ||
1758 | AG_UNLOCK(imap, ag); | |
1759 | ||
1760 | if (rc != -ENOSPC) | |
1761 | return (rc); | |
1762 | } | |
1763 | ||
1764 | /* no free disk inodes. | |
1765 | */ | |
1766 | return -ENOSPC; | |
1767 | } | |
1768 | ||
1769 | ||
1770 | /* | |
1771 | * NAME: diAllocIno(imap,agno,ip) | |
1772 | * | |
1773 | * FUNCTION: allocate a disk inode from the allocation group's free | |
1774 | * inode list, returning an error if this free list is | |
1775 | * empty (i.e. no iags on the list). | |
1776 | * | |
1777 | * allocation occurs from the first iag on the list using | |
1778 | * the iag's free inode summary map to find the leftmost | |
1779 | * free inode in the iag. | |
1780 | * | |
1781 | * PRE CONDITION: Already have AG lock for this AG. | |
1782 | * | |
1783 | * PARAMETERS: | |
1784 | * imap - pointer to inode map control structure. | |
1785 | * agno - allocation group. | |
1786 | * ip - pointer to new inode to be filled in on successful return | |
1787 | * with the disk inode number allocated, its extent address | |
1788 | * and the start of the ag. | |
1789 | * | |
1790 | * RETURN VALUES: | |
1791 | * 0 - success. | |
1792 | * -ENOSPC - insufficient disk resources. | |
1793 | * -EIO - i/o error. | |
1794 | */ | |
1795 | static int diAllocIno(struct inomap * imap, int agno, struct inode *ip) | |
1796 | { | |
1797 | int iagno, ino, rc, rem, extno, sword; | |
1798 | struct metapage *mp; | |
1799 | struct iag *iagp; | |
1800 | ||
1801 | /* check if there are iags on the ag's free inode list. | |
1802 | */ | |
1803 | if ((iagno = imap->im_agctl[agno].inofree) < 0) | |
1804 | return -ENOSPC; | |
1805 | ||
1806 | /* obtain read lock on imap inode */ | |
1807 | IREAD_LOCK(imap->im_ipimap); | |
1808 | ||
1809 | /* read the iag at the head of the list. | |
1810 | */ | |
1811 | if ((rc = diIAGRead(imap, iagno, &mp))) { | |
1812 | IREAD_UNLOCK(imap->im_ipimap); | |
1813 | return (rc); | |
1814 | } | |
1815 | iagp = (struct iag *) mp->data; | |
1816 | ||
1817 | /* better be free inodes in this iag if it is on the | |
1818 | * list. | |
1819 | */ | |
1820 | if (!iagp->nfreeinos) { | |
1821 | IREAD_UNLOCK(imap->im_ipimap); | |
1822 | release_metapage(mp); | |
1823 | jfs_error(ip->i_sb, | |
1824 | "diAllocIno: nfreeinos = 0, but iag on freelist"); | |
1825 | return -EIO; | |
1826 | } | |
1827 | ||
1828 | /* scan the free inode summary map to find an extent | |
1829 | * with free inodes. | |
1830 | */ | |
1831 | for (sword = 0;; sword++) { | |
1832 | if (sword >= SMAPSZ) { | |
1833 | IREAD_UNLOCK(imap->im_ipimap); | |
1834 | release_metapage(mp); | |
1835 | jfs_error(ip->i_sb, | |
1836 | "diAllocIno: free inode not found in summary map"); | |
1837 | return -EIO; | |
1838 | } | |
1839 | ||
1840 | if (~iagp->inosmap[sword]) | |
1841 | break; | |
1842 | } | |
1843 | ||
1844 | /* found a extent with free inodes. determine | |
1845 | * the extent number. | |
1846 | */ | |
1847 | rem = diFindFree(le32_to_cpu(iagp->inosmap[sword]), 0); | |
1848 | if (rem >= EXTSPERSUM) { | |
1849 | IREAD_UNLOCK(imap->im_ipimap); | |
1850 | release_metapage(mp); | |
1851 | jfs_error(ip->i_sb, "diAllocIno: no free extent found"); | |
1852 | return -EIO; | |
1853 | } | |
1854 | extno = (sword << L2EXTSPERSUM) + rem; | |
1855 | ||
1856 | /* find the first free inode in the extent. | |
1857 | */ | |
1858 | rem = diFindFree(le32_to_cpu(iagp->wmap[extno]), 0); | |
1859 | if (rem >= INOSPEREXT) { | |
1860 | IREAD_UNLOCK(imap->im_ipimap); | |
1861 | release_metapage(mp); | |
1862 | jfs_error(ip->i_sb, "diAllocIno: free inode not found"); | |
1863 | return -EIO; | |
1864 | } | |
1865 | ||
1866 | /* compute the inode number within the iag. | |
1867 | */ | |
1868 | ino = (extno << L2INOSPEREXT) + rem; | |
1869 | ||
1870 | /* allocate the inode. | |
1871 | */ | |
1872 | rc = diAllocBit(imap, iagp, ino); | |
1873 | IREAD_UNLOCK(imap->im_ipimap); | |
1874 | if (rc) { | |
1875 | release_metapage(mp); | |
1876 | return (rc); | |
1877 | } | |
1878 | ||
1879 | /* set the results of the allocation and write the iag. | |
1880 | */ | |
1881 | diInitInode(ip, iagno, ino, extno, iagp); | |
1882 | write_metapage(mp); | |
1883 | ||
1884 | return (0); | |
1885 | } | |
1886 | ||
1887 | ||
1888 | /* | |
1889 | * NAME: diAllocExt(imap,agno,ip) | |
1890 | * | |
1891 | * FUNCTION: add a new extent of free inodes to an iag, allocating | |
1892 | * an inode from this extent to satisfy the current allocation | |
1893 | * request. | |
1894 | * | |
1895 | * this routine first tries to find an existing iag with free | |
1896 | * extents through the ag free extent list. if list is not | |
1897 | * empty, the head of the list will be selected as the home | |
1898 | * of the new extent of free inodes. otherwise (the list is | |
1899 | * empty), a new iag will be allocated for the ag to contain | |
1900 | * the extent. | |
1901 | * | |
1902 | * once an iag has been selected, the free extent summary map | |
1903 | * is used to locate a free extent within the iag and diNewExt() | |
1904 | * is called to initialize the extent, with initialization | |
1905 | * including the allocation of the first inode of the extent | |
1906 | * for the purpose of satisfying this request. | |
1907 | * | |
1908 | * PARAMETERS: | |
1909 | * imap - pointer to inode map control structure. | |
1910 | * agno - allocation group number. | |
1911 | * ip - pointer to new inode to be filled in on successful return | |
1912 | * with the disk inode number allocated, its extent address | |
1913 | * and the start of the ag. | |
1914 | * | |
1915 | * RETURN VALUES: | |
1916 | * 0 - success. | |
1917 | * -ENOSPC - insufficient disk resources. | |
1918 | * -EIO - i/o error. | |
1919 | */ | |
1920 | static int diAllocExt(struct inomap * imap, int agno, struct inode *ip) | |
1921 | { | |
1922 | int rem, iagno, sword, extno, rc; | |
1923 | struct metapage *mp; | |
1924 | struct iag *iagp; | |
1925 | ||
1926 | /* check if the ag has any iags with free extents. if not, | |
1927 | * allocate a new iag for the ag. | |
1928 | */ | |
1929 | if ((iagno = imap->im_agctl[agno].extfree) < 0) { | |
1930 | /* If successful, diNewIAG will obtain the read lock on the | |
1931 | * imap inode. | |
1932 | */ | |
1933 | if ((rc = diNewIAG(imap, &iagno, agno, &mp))) { | |
1934 | return (rc); | |
1935 | } | |
1936 | iagp = (struct iag *) mp->data; | |
1937 | ||
1938 | /* set the ag number if this a brand new iag | |
1939 | */ | |
1940 | iagp->agstart = | |
1941 | cpu_to_le64(AGTOBLK(agno, imap->im_ipimap)); | |
1942 | } else { | |
1943 | /* read the iag. | |
1944 | */ | |
1945 | IREAD_LOCK(imap->im_ipimap); | |
1946 | if ((rc = diIAGRead(imap, iagno, &mp))) { | |
1947 | IREAD_UNLOCK(imap->im_ipimap); | |
1948 | jfs_error(ip->i_sb, "diAllocExt: error reading iag"); | |
1949 | return rc; | |
1950 | } | |
1951 | iagp = (struct iag *) mp->data; | |
1952 | } | |
1953 | ||
1954 | /* using the free extent summary map, find a free extent. | |
1955 | */ | |
1956 | for (sword = 0;; sword++) { | |
1957 | if (sword >= SMAPSZ) { | |
1958 | release_metapage(mp); | |
1959 | IREAD_UNLOCK(imap->im_ipimap); | |
1960 | jfs_error(ip->i_sb, | |
1961 | "diAllocExt: free ext summary map not found"); | |
1962 | return -EIO; | |
1963 | } | |
1964 | if (~iagp->extsmap[sword]) | |
1965 | break; | |
1966 | } | |
1967 | ||
1968 | /* determine the extent number of the free extent. | |
1969 | */ | |
1970 | rem = diFindFree(le32_to_cpu(iagp->extsmap[sword]), 0); | |
1971 | if (rem >= EXTSPERSUM) { | |
1972 | release_metapage(mp); | |
1973 | IREAD_UNLOCK(imap->im_ipimap); | |
1974 | jfs_error(ip->i_sb, "diAllocExt: free extent not found"); | |
1975 | return -EIO; | |
1976 | } | |
1977 | extno = (sword << L2EXTSPERSUM) + rem; | |
1978 | ||
1979 | /* initialize the new extent. | |
1980 | */ | |
1981 | rc = diNewExt(imap, iagp, extno); | |
1982 | IREAD_UNLOCK(imap->im_ipimap); | |
1983 | if (rc) { | |
1984 | /* something bad happened. if a new iag was allocated, | |
1985 | * place it back on the inode map's iag free list, and | |
1986 | * clear the ag number information. | |
1987 | */ | |
1988 | if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) { | |
1989 | IAGFREE_LOCK(imap); | |
1990 | iagp->iagfree = cpu_to_le32(imap->im_freeiag); | |
1991 | imap->im_freeiag = iagno; | |
1992 | IAGFREE_UNLOCK(imap); | |
1993 | } | |
1994 | write_metapage(mp); | |
1995 | return (rc); | |
1996 | } | |
1997 | ||
1998 | /* set the results of the allocation and write the iag. | |
1999 | */ | |
2000 | diInitInode(ip, iagno, extno << L2INOSPEREXT, extno, iagp); | |
2001 | ||
2002 | write_metapage(mp); | |
2003 | ||
2004 | return (0); | |
2005 | } | |
2006 | ||
2007 | ||
2008 | /* | |
2009 | * NAME: diAllocBit(imap,iagp,ino) | |
2010 | * | |
2011 | * FUNCTION: allocate a backed inode from an iag. | |
2012 | * | |
2013 | * this routine performs the mechanics of allocating a | |
2014 | * specified inode from a backed extent. | |
2015 | * | |
2016 | * if the inode to be allocated represents the last free | |
2017 | * inode within the iag, the iag will be removed from the | |
2018 | * ag free inode list. | |
2019 | * | |
2020 | * a careful update approach is used to provide consistency | |
2021 | * in the face of updates to multiple buffers. under this | |
2022 | * approach, all required buffers are obtained before making | |
2023 | * any updates and are held all are updates are complete. | |
2024 | * | |
2025 | * PRE CONDITION: Already have buffer lock on iagp. Already have AG lock on | |
2026 | * this AG. Must have read lock on imap inode. | |
2027 | * | |
2028 | * PARAMETERS: | |
2029 | * imap - pointer to inode map control structure. | |
2030 | * iagp - pointer to iag. | |
2031 | * ino - inode number to be allocated within the iag. | |
2032 | * | |
2033 | * RETURN VALUES: | |
2034 | * 0 - success. | |
2035 | * -ENOSPC - insufficient disk resources. | |
2036 | * -EIO - i/o error. | |
2037 | */ | |
2038 | static int diAllocBit(struct inomap * imap, struct iag * iagp, int ino) | |
2039 | { | |
2040 | int extno, bitno, agno, sword, rc; | |
2041 | struct metapage *amp = NULL, *bmp = NULL; | |
2042 | struct iag *aiagp = NULL, *biagp = NULL; | |
2043 | u32 mask; | |
2044 | ||
2045 | /* check if this is the last free inode within the iag. | |
2046 | * if so, it will have to be removed from the ag free | |
2047 | * inode list, so get the iags preceeding and following | |
2048 | * it on the list. | |
2049 | */ | |
2050 | if (iagp->nfreeinos == cpu_to_le32(1)) { | |
2051 | if ((int) le32_to_cpu(iagp->inofreefwd) >= 0) { | |
2052 | if ((rc = | |
2053 | diIAGRead(imap, le32_to_cpu(iagp->inofreefwd), | |
2054 | &))) | |
2055 | return (rc); | |
2056 | aiagp = (struct iag *) amp->data; | |
2057 | } | |
2058 | ||
2059 | if ((int) le32_to_cpu(iagp->inofreeback) >= 0) { | |
2060 | if ((rc = | |
2061 | diIAGRead(imap, | |
2062 | le32_to_cpu(iagp->inofreeback), | |
2063 | &bmp))) { | |
2064 | if (amp) | |
2065 | release_metapage(amp); | |
2066 | return (rc); | |
2067 | } | |
2068 | biagp = (struct iag *) bmp->data; | |
2069 | } | |
2070 | } | |
2071 | ||
2072 | /* get the ag number, extent number, inode number within | |
2073 | * the extent. | |
2074 | */ | |
2075 | agno = BLKTOAG(le64_to_cpu(iagp->agstart), JFS_SBI(imap->im_ipimap->i_sb)); | |
2076 | extno = ino >> L2INOSPEREXT; | |
2077 | bitno = ino & (INOSPEREXT - 1); | |
2078 | ||
2079 | /* compute the mask for setting the map. | |
2080 | */ | |
2081 | mask = HIGHORDER >> bitno; | |
2082 | ||
2083 | /* the inode should be free and backed. | |
2084 | */ | |
2085 | if (((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) || | |
2086 | ((le32_to_cpu(iagp->wmap[extno]) & mask) != 0) || | |
2087 | (addressPXD(&iagp->inoext[extno]) == 0)) { | |
2088 | if (amp) | |
2089 | release_metapage(amp); | |
2090 | if (bmp) | |
2091 | release_metapage(bmp); | |
2092 | ||
2093 | jfs_error(imap->im_ipimap->i_sb, | |
2094 | "diAllocBit: iag inconsistent"); | |
2095 | return -EIO; | |
2096 | } | |
2097 | ||
2098 | /* mark the inode as allocated in the working map. | |
2099 | */ | |
2100 | iagp->wmap[extno] |= cpu_to_le32(mask); | |
2101 | ||
2102 | /* check if all inodes within the extent are now | |
2103 | * allocated. if so, update the free inode summary | |
2104 | * map to reflect this. | |
2105 | */ | |
2106 | if (iagp->wmap[extno] == cpu_to_le32(ONES)) { | |
2107 | sword = extno >> L2EXTSPERSUM; | |
2108 | bitno = extno & (EXTSPERSUM - 1); | |
2109 | iagp->inosmap[sword] |= cpu_to_le32(HIGHORDER >> bitno); | |
2110 | } | |
2111 | ||
2112 | /* if this was the last free inode in the iag, remove the | |
2113 | * iag from the ag free inode list. | |
2114 | */ | |
2115 | if (iagp->nfreeinos == cpu_to_le32(1)) { | |
2116 | if (amp) { | |
2117 | aiagp->inofreeback = iagp->inofreeback; | |
2118 | write_metapage(amp); | |
2119 | } | |
2120 | ||
2121 | if (bmp) { | |
2122 | biagp->inofreefwd = iagp->inofreefwd; | |
2123 | write_metapage(bmp); | |
2124 | } else { | |
2125 | imap->im_agctl[agno].inofree = | |
2126 | le32_to_cpu(iagp->inofreefwd); | |
2127 | } | |
2128 | iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1); | |
2129 | } | |
2130 | ||
2131 | /* update the free inode count at the iag, ag, inode | |
2132 | * map levels. | |
2133 | */ | |
2134 | iagp->nfreeinos = cpu_to_le32(le32_to_cpu(iagp->nfreeinos) - 1); | |
2135 | imap->im_agctl[agno].numfree -= 1; | |
2136 | atomic_dec(&imap->im_numfree); | |
2137 | ||
2138 | return (0); | |
2139 | } | |
2140 | ||
2141 | ||
2142 | /* | |
2143 | * NAME: diNewExt(imap,iagp,extno) | |
2144 | * | |
2145 | * FUNCTION: initialize a new extent of inodes for an iag, allocating | |
2146 | * the first inode of the extent for use for the current | |
2147 | * allocation request. | |
2148 | * | |
2149 | * disk resources are allocated for the new extent of inodes | |
2150 | * and the inodes themselves are initialized to reflect their | |
2151 | * existence within the extent (i.e. their inode numbers and | |
2152 | * inode extent addresses are set) and their initial state | |
2153 | * (mode and link count are set to zero). | |
2154 | * | |
2155 | * if the iag is new, it is not yet on an ag extent free list | |
2156 | * but will now be placed on this list. | |
2157 | * | |
2158 | * if the allocation of the new extent causes the iag to | |
2159 | * have no free extent, the iag will be removed from the | |
2160 | * ag extent free list. | |
2161 | * | |
2162 | * if the iag has no free backed inodes, it will be placed | |
2163 | * on the ag free inode list, since the addition of the new | |
2164 | * extent will now cause it to have free inodes. | |
2165 | * | |
2166 | * a careful update approach is used to provide consistency | |
2167 | * (i.e. list consistency) in the face of updates to multiple | |
2168 | * buffers. under this approach, all required buffers are | |
2169 | * obtained before making any updates and are held until all | |
2170 | * updates are complete. | |
2171 | * | |
2172 | * PRE CONDITION: Already have buffer lock on iagp. Already have AG lock on | |
2173 | * this AG. Must have read lock on imap inode. | |
2174 | * | |
2175 | * PARAMETERS: | |
2176 | * imap - pointer to inode map control structure. | |
2177 | * iagp - pointer to iag. | |
2178 | * extno - extent number. | |
2179 | * | |
2180 | * RETURN VALUES: | |
2181 | * 0 - success. | |
2182 | * -ENOSPC - insufficient disk resources. | |
2183 | * -EIO - i/o error. | |
2184 | */ | |
2185 | static int diNewExt(struct inomap * imap, struct iag * iagp, int extno) | |
2186 | { | |
2187 | int agno, iagno, fwd, back, freei = 0, sword, rc; | |
2188 | struct iag *aiagp = NULL, *biagp = NULL, *ciagp = NULL; | |
2189 | struct metapage *amp, *bmp, *cmp, *dmp; | |
2190 | struct inode *ipimap; | |
2191 | s64 blkno, hint; | |
2192 | int i, j; | |
2193 | u32 mask; | |
2194 | ino_t ino; | |
2195 | struct dinode *dp; | |
2196 | struct jfs_sb_info *sbi; | |
2197 | ||
2198 | /* better have free extents. | |
2199 | */ | |
2200 | if (!iagp->nfreeexts) { | |
2201 | jfs_error(imap->im_ipimap->i_sb, "diNewExt: no free extents"); | |
2202 | return -EIO; | |
2203 | } | |
2204 | ||
2205 | /* get the inode map inode. | |
2206 | */ | |
2207 | ipimap = imap->im_ipimap; | |
2208 | sbi = JFS_SBI(ipimap->i_sb); | |
2209 | ||
2210 | amp = bmp = cmp = NULL; | |
2211 | ||
2212 | /* get the ag and iag numbers for this iag. | |
2213 | */ | |
2214 | agno = BLKTOAG(le64_to_cpu(iagp->agstart), sbi); | |
2215 | iagno = le32_to_cpu(iagp->iagnum); | |
2216 | ||
2217 | /* check if this is the last free extent within the | |
2218 | * iag. if so, the iag must be removed from the ag | |
2219 | * free extent list, so get the iags preceeding and | |
2220 | * following the iag on this list. | |
2221 | */ | |
2222 | if (iagp->nfreeexts == cpu_to_le32(1)) { | |
2223 | if ((fwd = le32_to_cpu(iagp->extfreefwd)) >= 0) { | |
2224 | if ((rc = diIAGRead(imap, fwd, &))) | |
2225 | return (rc); | |
2226 | aiagp = (struct iag *) amp->data; | |
2227 | } | |
2228 | ||
2229 | if ((back = le32_to_cpu(iagp->extfreeback)) >= 0) { | |
2230 | if ((rc = diIAGRead(imap, back, &bmp))) | |
2231 | goto error_out; | |
2232 | biagp = (struct iag *) bmp->data; | |
2233 | } | |
2234 | } else { | |
2235 | /* the iag has free extents. if all extents are free | |
2236 | * (as is the case for a newly allocated iag), the iag | |
2237 | * must be added to the ag free extent list, so get | |
2238 | * the iag at the head of the list in preparation for | |
2239 | * adding this iag to this list. | |
2240 | */ | |
2241 | fwd = back = -1; | |
2242 | if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) { | |
2243 | if ((fwd = imap->im_agctl[agno].extfree) >= 0) { | |
2244 | if ((rc = diIAGRead(imap, fwd, &))) | |
2245 | goto error_out; | |
2246 | aiagp = (struct iag *) amp->data; | |
2247 | } | |
2248 | } | |
2249 | } | |
2250 | ||
2251 | /* check if the iag has no free inodes. if so, the iag | |
2252 | * will have to be added to the ag free inode list, so get | |
2253 | * the iag at the head of the list in preparation for | |
2254 | * adding this iag to this list. in doing this, we must | |
2255 | * check if we already have the iag at the head of | |
2256 | * the list in hand. | |
2257 | */ | |
2258 | if (iagp->nfreeinos == 0) { | |
2259 | freei = imap->im_agctl[agno].inofree; | |
2260 | ||
2261 | if (freei >= 0) { | |
2262 | if (freei == fwd) { | |
2263 | ciagp = aiagp; | |
2264 | } else if (freei == back) { | |
2265 | ciagp = biagp; | |
2266 | } else { | |
2267 | if ((rc = diIAGRead(imap, freei, &cmp))) | |
2268 | goto error_out; | |
2269 | ciagp = (struct iag *) cmp->data; | |
2270 | } | |
2271 | if (ciagp == NULL) { | |
2272 | jfs_error(imap->im_ipimap->i_sb, | |
2273 | "diNewExt: ciagp == NULL"); | |
2274 | rc = -EIO; | |
2275 | goto error_out; | |
2276 | } | |
2277 | } | |
2278 | } | |
2279 | ||
2280 | /* allocate disk space for the inode extent. | |
2281 | */ | |
2282 | if ((extno == 0) || (addressPXD(&iagp->inoext[extno - 1]) == 0)) | |
2283 | hint = ((s64) agno << sbi->bmap->db_agl2size) - 1; | |
2284 | else | |
2285 | hint = addressPXD(&iagp->inoext[extno - 1]) + | |
2286 | lengthPXD(&iagp->inoext[extno - 1]) - 1; | |
2287 | ||
2288 | if ((rc = dbAlloc(ipimap, hint, (s64) imap->im_nbperiext, &blkno))) | |
2289 | goto error_out; | |
2290 | ||
2291 | /* compute the inode number of the first inode within the | |
2292 | * extent. | |
2293 | */ | |
2294 | ino = (iagno << L2INOSPERIAG) + (extno << L2INOSPEREXT); | |
2295 | ||
2296 | /* initialize the inodes within the newly allocated extent a | |
2297 | * page at a time. | |
2298 | */ | |
2299 | for (i = 0; i < imap->im_nbperiext; i += sbi->nbperpage) { | |
2300 | /* get a buffer for this page of disk inodes. | |
2301 | */ | |
2302 | dmp = get_metapage(ipimap, blkno + i, PSIZE, 1); | |
2303 | if (dmp == NULL) { | |
2304 | rc = -EIO; | |
2305 | goto error_out; | |
2306 | } | |
2307 | dp = (struct dinode *) dmp->data; | |
2308 | ||
2309 | /* initialize the inode number, mode, link count and | |
2310 | * inode extent address. | |
2311 | */ | |
2312 | for (j = 0; j < INOSPERPAGE; j++, dp++, ino++) { | |
2313 | dp->di_inostamp = cpu_to_le32(sbi->inostamp); | |
2314 | dp->di_number = cpu_to_le32(ino); | |
2315 | dp->di_fileset = cpu_to_le32(FILESYSTEM_I); | |
2316 | dp->di_mode = 0; | |
2317 | dp->di_nlink = 0; | |
2318 | PXDaddress(&(dp->di_ixpxd), blkno); | |
2319 | PXDlength(&(dp->di_ixpxd), imap->im_nbperiext); | |
2320 | } | |
2321 | write_metapage(dmp); | |
2322 | } | |
2323 | ||
2324 | /* if this is the last free extent within the iag, remove the | |
2325 | * iag from the ag free extent list. | |
2326 | */ | |
2327 | if (iagp->nfreeexts == cpu_to_le32(1)) { | |
2328 | if (fwd >= 0) | |
2329 | aiagp->extfreeback = iagp->extfreeback; | |
2330 | ||
2331 | if (back >= 0) | |
2332 | biagp->extfreefwd = iagp->extfreefwd; | |
2333 | else | |
2334 | imap->im_agctl[agno].extfree = | |
2335 | le32_to_cpu(iagp->extfreefwd); | |
2336 | ||
2337 | iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1); | |
2338 | } else { | |
2339 | /* if the iag has all free extents (newly allocated iag), | |
2340 | * add the iag to the ag free extent list. | |
2341 | */ | |
2342 | if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) { | |
2343 | if (fwd >= 0) | |
2344 | aiagp->extfreeback = cpu_to_le32(iagno); | |
2345 | ||
2346 | iagp->extfreefwd = cpu_to_le32(fwd); | |
2347 | iagp->extfreeback = cpu_to_le32(-1); | |
2348 | imap->im_agctl[agno].extfree = iagno; | |
2349 | } | |
2350 | } | |
2351 | ||
2352 | /* if the iag has no free inodes, add the iag to the | |
2353 | * ag free inode list. | |
2354 | */ | |
2355 | if (iagp->nfreeinos == 0) { | |
2356 | if (freei >= 0) | |
2357 | ciagp->inofreeback = cpu_to_le32(iagno); | |
2358 | ||
2359 | iagp->inofreefwd = | |
2360 | cpu_to_le32(imap->im_agctl[agno].inofree); | |
2361 | iagp->inofreeback = cpu_to_le32(-1); | |
2362 | imap->im_agctl[agno].inofree = iagno; | |
2363 | } | |
2364 | ||
2365 | /* initialize the extent descriptor of the extent. */ | |
2366 | PXDlength(&iagp->inoext[extno], imap->im_nbperiext); | |
2367 | PXDaddress(&iagp->inoext[extno], blkno); | |
2368 | ||
2369 | /* initialize the working and persistent map of the extent. | |
2370 | * the working map will be initialized such that | |
2371 | * it indicates the first inode of the extent is allocated. | |
2372 | */ | |
2373 | iagp->wmap[extno] = cpu_to_le32(HIGHORDER); | |
2374 | iagp->pmap[extno] = 0; | |
2375 | ||
2376 | /* update the free inode and free extent summary maps | |
2377 | * for the extent to indicate the extent has free inodes | |
2378 | * and no longer represents a free extent. | |
2379 | */ | |
2380 | sword = extno >> L2EXTSPERSUM; | |
2381 | mask = HIGHORDER >> (extno & (EXTSPERSUM - 1)); | |
2382 | iagp->extsmap[sword] |= cpu_to_le32(mask); | |
2383 | iagp->inosmap[sword] &= cpu_to_le32(~mask); | |
2384 | ||
2385 | /* update the free inode and free extent counts for the | |
2386 | * iag. | |
2387 | */ | |
2388 | iagp->nfreeinos = cpu_to_le32(le32_to_cpu(iagp->nfreeinos) + | |
2389 | (INOSPEREXT - 1)); | |
2390 | iagp->nfreeexts = cpu_to_le32(le32_to_cpu(iagp->nfreeexts) - 1); | |
2391 | ||
2392 | /* update the free and backed inode counts for the ag. | |
2393 | */ | |
2394 | imap->im_agctl[agno].numfree += (INOSPEREXT - 1); | |
2395 | imap->im_agctl[agno].numinos += INOSPEREXT; | |
2396 | ||
2397 | /* update the free and backed inode counts for the inode map. | |
2398 | */ | |
2399 | atomic_add(INOSPEREXT - 1, &imap->im_numfree); | |
2400 | atomic_add(INOSPEREXT, &imap->im_numinos); | |
2401 | ||
2402 | /* write the iags. | |
2403 | */ | |
2404 | if (amp) | |
2405 | write_metapage(amp); | |
2406 | if (bmp) | |
2407 | write_metapage(bmp); | |
2408 | if (cmp) | |
2409 | write_metapage(cmp); | |
2410 | ||
2411 | return (0); | |
2412 | ||
2413 | error_out: | |
2414 | ||
2415 | /* release the iags. | |
2416 | */ | |
2417 | if (amp) | |
2418 | release_metapage(amp); | |
2419 | if (bmp) | |
2420 | release_metapage(bmp); | |
2421 | if (cmp) | |
2422 | release_metapage(cmp); | |
2423 | ||
2424 | return (rc); | |
2425 | } | |
2426 | ||
2427 | ||
2428 | /* | |
2429 | * NAME: diNewIAG(imap,iagnop,agno) | |
2430 | * | |
2431 | * FUNCTION: allocate a new iag for an allocation group. | |
2432 | * | |
2433 | * first tries to allocate the iag from the inode map | |
2434 | * iagfree list: | |
2435 | * if the list has free iags, the head of the list is removed | |
2436 | * and returned to satisfy the request. | |
2437 | * if the inode map's iag free list is empty, the inode map | |
2438 | * is extended to hold a new iag. this new iag is initialized | |
2439 | * and returned to satisfy the request. | |
2440 | * | |
2441 | * PARAMETERS: | |
2442 | * imap - pointer to inode map control structure. | |
2443 | * iagnop - pointer to an iag number set with the number of the | |
2444 | * newly allocated iag upon successful return. | |
2445 | * agno - allocation group number. | |
2446 | * bpp - Buffer pointer to be filled in with new IAG's buffer | |
2447 | * | |
2448 | * RETURN VALUES: | |
2449 | * 0 - success. | |
2450 | * -ENOSPC - insufficient disk resources. | |
2451 | * -EIO - i/o error. | |
2452 | * | |
2453 | * serialization: | |
2454 | * AG lock held on entry/exit; | |
2455 | * write lock on the map is held inside; | |
2456 | * read lock on the map is held on successful completion; | |
2457 | * | |
2458 | * note: new iag transaction: | |
2459 | * . synchronously write iag; | |
2460 | * . write log of xtree and inode of imap; | |
2461 | * . commit; | |
2462 | * . synchronous write of xtree (right to left, bottom to top); | |
2463 | * . at start of logredo(): init in-memory imap with one additional iag page; | |
2464 | * . at end of logredo(): re-read imap inode to determine | |
2465 | * new imap size; | |
2466 | */ | |
2467 | static int | |
2468 | diNewIAG(struct inomap * imap, int *iagnop, int agno, struct metapage ** mpp) | |
2469 | { | |
2470 | int rc; | |
2471 | int iagno, i, xlen; | |
2472 | struct inode *ipimap; | |
2473 | struct super_block *sb; | |
2474 | struct jfs_sb_info *sbi; | |
2475 | struct metapage *mp; | |
2476 | struct iag *iagp; | |
2477 | s64 xaddr = 0; | |
2478 | s64 blkno; | |
2479 | tid_t tid; | |
2480 | #ifdef _STILL_TO_PORT | |
2481 | xad_t xad; | |
2482 | #endif /* _STILL_TO_PORT */ | |
2483 | struct inode *iplist[1]; | |
2484 | ||
2485 | /* pick up pointers to the inode map and mount inodes */ | |
2486 | ipimap = imap->im_ipimap; | |
2487 | sb = ipimap->i_sb; | |
2488 | sbi = JFS_SBI(sb); | |
2489 | ||
2490 | /* acquire the free iag lock */ | |
2491 | IAGFREE_LOCK(imap); | |
2492 | ||
2493 | /* if there are any iags on the inode map free iag list, | |
2494 | * allocate the iag from the head of the list. | |
2495 | */ | |
2496 | if (imap->im_freeiag >= 0) { | |
2497 | /* pick up the iag number at the head of the list */ | |
2498 | iagno = imap->im_freeiag; | |
2499 | ||
2500 | /* determine the logical block number of the iag */ | |
2501 | blkno = IAGTOLBLK(iagno, sbi->l2nbperpage); | |
2502 | } else { | |
2503 | /* no free iags. the inode map will have to be extented | |
2504 | * to include a new iag. | |
2505 | */ | |
2506 | ||
2507 | /* acquire inode map lock */ | |
2508 | IWRITE_LOCK(ipimap); | |
2509 | ||
2510 | if (ipimap->i_size >> L2PSIZE != imap->im_nextiag + 1) { | |
2511 | IWRITE_UNLOCK(ipimap); | |
2512 | IAGFREE_UNLOCK(imap); | |
2513 | jfs_error(imap->im_ipimap->i_sb, | |
2514 | "diNewIAG: ipimap->i_size is wrong"); | |
2515 | return -EIO; | |
2516 | } | |
2517 | ||
2518 | ||
2519 | /* get the next avaliable iag number */ | |
2520 | iagno = imap->im_nextiag; | |
2521 | ||
2522 | /* make sure that we have not exceeded the maximum inode | |
2523 | * number limit. | |
2524 | */ | |
2525 | if (iagno > (MAXIAGS - 1)) { | |
2526 | /* release the inode map lock */ | |
2527 | IWRITE_UNLOCK(ipimap); | |
2528 | ||
2529 | rc = -ENOSPC; | |
2530 | goto out; | |
2531 | } | |
2532 | ||
2533 | /* | |
2534 | * synchronously append new iag page. | |
2535 | */ | |
2536 | /* determine the logical address of iag page to append */ | |
2537 | blkno = IAGTOLBLK(iagno, sbi->l2nbperpage); | |
2538 | ||
2539 | /* Allocate extent for new iag page */ | |
2540 | xlen = sbi->nbperpage; | |
2541 | if ((rc = dbAlloc(ipimap, 0, (s64) xlen, &xaddr))) { | |
2542 | /* release the inode map lock */ | |
2543 | IWRITE_UNLOCK(ipimap); | |
2544 | ||
2545 | goto out; | |
2546 | } | |
2547 | ||
d2e83707 DK |
2548 | /* |
2549 | * start transaction of update of the inode map | |
2550 | * addressing structure pointing to the new iag page; | |
2551 | */ | |
2552 | tid = txBegin(sb, COMMIT_FORCE); | |
2553 | down(&JFS_IP(ipimap)->commit_sem); | |
2554 | ||
2555 | /* update the inode map addressing structure to point to it */ | |
2556 | if ((rc = | |
2557 | xtInsert(tid, ipimap, 0, blkno, xlen, &xaddr, 0))) { | |
2558 | txEnd(tid); | |
2559 | up(&JFS_IP(ipimap)->commit_sem); | |
1da177e4 LT |
2560 | /* Free the blocks allocated for the iag since it was |
2561 | * not successfully added to the inode map | |
2562 | */ | |
2563 | dbFree(ipimap, xaddr, (s64) xlen); | |
2564 | ||
2565 | /* release the inode map lock */ | |
2566 | IWRITE_UNLOCK(ipimap); | |
2567 | ||
d2e83707 DK |
2568 | goto out; |
2569 | } | |
2570 | ||
2571 | /* update the inode map's inode to reflect the extension */ | |
2572 | ipimap->i_size += PSIZE; | |
2573 | inode_add_bytes(ipimap, PSIZE); | |
2574 | ||
2575 | /* assign a buffer for the page */ | |
2576 | mp = get_metapage(ipimap, blkno, PSIZE, 0); | |
2577 | if (!mp) { | |
2578 | /* | |
2579 | * This is very unlikely since we just created the | |
2580 | * extent, but let's try to handle it correctly | |
2581 | */ | |
2582 | xtTruncate(tid, ipimap, ipimap->i_size - PSIZE, | |
2583 | COMMIT_PWMAP); | |
2584 | ||
2585 | txAbort(tid, 0); | |
2586 | txEnd(tid); | |
2587 | ||
2588 | /* release the inode map lock */ | |
2589 | IWRITE_UNLOCK(ipimap); | |
2590 | ||
1da177e4 LT |
2591 | rc = -EIO; |
2592 | goto out; | |
2593 | } | |
2594 | iagp = (struct iag *) mp->data; | |
2595 | ||
2596 | /* init the iag */ | |
2597 | memset(iagp, 0, sizeof(struct iag)); | |
2598 | iagp->iagnum = cpu_to_le32(iagno); | |
2599 | iagp->inofreefwd = iagp->inofreeback = cpu_to_le32(-1); | |
2600 | iagp->extfreefwd = iagp->extfreeback = cpu_to_le32(-1); | |
2601 | iagp->iagfree = cpu_to_le32(-1); | |
2602 | iagp->nfreeinos = 0; | |
2603 | iagp->nfreeexts = cpu_to_le32(EXTSPERIAG); | |
2604 | ||
2605 | /* initialize the free inode summary map (free extent | |
2606 | * summary map initialization handled by bzero). | |
2607 | */ | |
2608 | for (i = 0; i < SMAPSZ; i++) | |
2609 | iagp->inosmap[i] = cpu_to_le32(ONES); | |
2610 | ||
2611 | /* | |
d2e83707 | 2612 | * Write and sync the metapage |
1da177e4 | 2613 | */ |
1da177e4 LT |
2614 | flush_metapage(mp); |
2615 | ||
1da177e4 LT |
2616 | /* |
2617 | * txCommit(COMMIT_FORCE) will synchronously write address | |
2618 | * index pages and inode after commit in careful update order | |
2619 | * of address index pages (right to left, bottom up); | |
2620 | */ | |
2621 | iplist[0] = ipimap; | |
2622 | rc = txCommit(tid, 1, &iplist[0], COMMIT_FORCE); | |
2623 | ||
2624 | txEnd(tid); | |
2625 | up(&JFS_IP(ipimap)->commit_sem); | |
2626 | ||
2627 | duplicateIXtree(sb, blkno, xlen, &xaddr); | |
2628 | ||
2629 | /* update the next avaliable iag number */ | |
2630 | imap->im_nextiag += 1; | |
2631 | ||
2632 | /* Add the iag to the iag free list so we don't lose the iag | |
2633 | * if a failure happens now. | |
2634 | */ | |
2635 | imap->im_freeiag = iagno; | |
2636 | ||
2637 | /* Until we have logredo working, we want the imap inode & | |
2638 | * control page to be up to date. | |
2639 | */ | |
2640 | diSync(ipimap); | |
2641 | ||
2642 | /* release the inode map lock */ | |
2643 | IWRITE_UNLOCK(ipimap); | |
2644 | } | |
2645 | ||
2646 | /* obtain read lock on map */ | |
2647 | IREAD_LOCK(ipimap); | |
2648 | ||
2649 | /* read the iag */ | |
2650 | if ((rc = diIAGRead(imap, iagno, &mp))) { | |
2651 | IREAD_UNLOCK(ipimap); | |
2652 | rc = -EIO; | |
2653 | goto out; | |
2654 | } | |
2655 | iagp = (struct iag *) mp->data; | |
2656 | ||
2657 | /* remove the iag from the iag free list */ | |
2658 | imap->im_freeiag = le32_to_cpu(iagp->iagfree); | |
2659 | iagp->iagfree = cpu_to_le32(-1); | |
2660 | ||
2661 | /* set the return iag number and buffer pointer */ | |
2662 | *iagnop = iagno; | |
2663 | *mpp = mp; | |
2664 | ||
2665 | out: | |
2666 | /* release the iag free lock */ | |
2667 | IAGFREE_UNLOCK(imap); | |
2668 | ||
2669 | return (rc); | |
2670 | } | |
2671 | ||
2672 | /* | |
2673 | * NAME: diIAGRead() | |
2674 | * | |
2675 | * FUNCTION: get the buffer for the specified iag within a fileset | |
2676 | * or aggregate inode map. | |
2677 | * | |
2678 | * PARAMETERS: | |
2679 | * imap - pointer to inode map control structure. | |
2680 | * iagno - iag number. | |
2681 | * bpp - point to buffer pointer to be filled in on successful | |
2682 | * exit. | |
2683 | * | |
2684 | * SERIALIZATION: | |
2685 | * must have read lock on imap inode | |
2686 | * (When called by diExtendFS, the filesystem is quiesced, therefore | |
2687 | * the read lock is unnecessary.) | |
2688 | * | |
2689 | * RETURN VALUES: | |
2690 | * 0 - success. | |
2691 | * -EIO - i/o error. | |
2692 | */ | |
2693 | static int diIAGRead(struct inomap * imap, int iagno, struct metapage ** mpp) | |
2694 | { | |
2695 | struct inode *ipimap = imap->im_ipimap; | |
2696 | s64 blkno; | |
2697 | ||
2698 | /* compute the logical block number of the iag. */ | |
2699 | blkno = IAGTOLBLK(iagno, JFS_SBI(ipimap->i_sb)->l2nbperpage); | |
2700 | ||
2701 | /* read the iag. */ | |
2702 | *mpp = read_metapage(ipimap, blkno, PSIZE, 0); | |
2703 | if (*mpp == NULL) { | |
2704 | return -EIO; | |
2705 | } | |
2706 | ||
2707 | return (0); | |
2708 | } | |
2709 | ||
2710 | /* | |
2711 | * NAME: diFindFree() | |
2712 | * | |
2713 | * FUNCTION: find the first free bit in a word starting at | |
2714 | * the specified bit position. | |
2715 | * | |
2716 | * PARAMETERS: | |
2717 | * word - word to be examined. | |
2718 | * start - starting bit position. | |
2719 | * | |
2720 | * RETURN VALUES: | |
2721 | * bit position of first free bit in the word or 32 if | |
2722 | * no free bits were found. | |
2723 | */ | |
2724 | static int diFindFree(u32 word, int start) | |
2725 | { | |
2726 | int bitno; | |
2727 | assert(start < 32); | |
2728 | /* scan the word for the first free bit. */ | |
2729 | for (word <<= start, bitno = start; bitno < 32; | |
2730 | bitno++, word <<= 1) { | |
2731 | if ((word & HIGHORDER) == 0) | |
2732 | break; | |
2733 | } | |
2734 | return (bitno); | |
2735 | } | |
2736 | ||
2737 | /* | |
2738 | * NAME: diUpdatePMap() | |
2739 | * | |
2740 | * FUNCTION: Update the persistent map in an IAG for the allocation or | |
2741 | * freeing of the specified inode. | |
2742 | * | |
2743 | * PRE CONDITIONS: Working map has already been updated for allocate. | |
2744 | * | |
2745 | * PARAMETERS: | |
2746 | * ipimap - Incore inode map inode | |
2747 | * inum - Number of inode to mark in permanent map | |
2748 | * is_free - If TRUE indicates inode should be marked freed, otherwise | |
2749 | * indicates inode should be marked allocated. | |
2750 | * | |
2751 | * RETURN VALUES: | |
2752 | * 0 for success | |
2753 | */ | |
2754 | int | |
2755 | diUpdatePMap(struct inode *ipimap, | |
2756 | unsigned long inum, boolean_t is_free, struct tblock * tblk) | |
2757 | { | |
2758 | int rc; | |
2759 | struct iag *iagp; | |
2760 | struct metapage *mp; | |
2761 | int iagno, ino, extno, bitno; | |
2762 | struct inomap *imap; | |
2763 | u32 mask; | |
2764 | struct jfs_log *log; | |
2765 | int lsn, difft, diffp; | |
7fab479b | 2766 | unsigned long flags; |
1da177e4 LT |
2767 | |
2768 | imap = JFS_IP(ipimap)->i_imap; | |
2769 | /* get the iag number containing the inode */ | |
2770 | iagno = INOTOIAG(inum); | |
2771 | /* make sure that the iag is contained within the map */ | |
2772 | if (iagno >= imap->im_nextiag) { | |
2773 | jfs_error(ipimap->i_sb, | |
2774 | "diUpdatePMap: the iag is outside the map"); | |
2775 | return -EIO; | |
2776 | } | |
2777 | /* read the iag */ | |
2778 | IREAD_LOCK(ipimap); | |
2779 | rc = diIAGRead(imap, iagno, &mp); | |
2780 | IREAD_UNLOCK(ipimap); | |
2781 | if (rc) | |
2782 | return (rc); | |
7fab479b | 2783 | metapage_wait_for_io(mp); |
1da177e4 LT |
2784 | iagp = (struct iag *) mp->data; |
2785 | /* get the inode number and extent number of the inode within | |
2786 | * the iag and the inode number within the extent. | |
2787 | */ | |
2788 | ino = inum & (INOSPERIAG - 1); | |
2789 | extno = ino >> L2INOSPEREXT; | |
2790 | bitno = ino & (INOSPEREXT - 1); | |
2791 | mask = HIGHORDER >> bitno; | |
2792 | /* | |
2793 | * mark the inode free in persistent map: | |
2794 | */ | |
2795 | if (is_free == TRUE) { | |
2796 | /* The inode should have been allocated both in working | |
2797 | * map and in persistent map; | |
2798 | * the inode will be freed from working map at the release | |
2799 | * of last reference release; | |
2800 | */ | |
2801 | if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) { | |
2802 | jfs_error(ipimap->i_sb, | |
2803 | "diUpdatePMap: inode %ld not marked as " | |
2804 | "allocated in wmap!", inum); | |
2805 | } | |
2806 | if (!(le32_to_cpu(iagp->pmap[extno]) & mask)) { | |
2807 | jfs_error(ipimap->i_sb, | |
2808 | "diUpdatePMap: inode %ld not marked as " | |
2809 | "allocated in pmap!", inum); | |
2810 | } | |
2811 | /* update the bitmap for the extent of the freed inode */ | |
2812 | iagp->pmap[extno] &= cpu_to_le32(~mask); | |
2813 | } | |
2814 | /* | |
2815 | * mark the inode allocated in persistent map: | |
2816 | */ | |
2817 | else { | |
2818 | /* The inode should be already allocated in the working map | |
2819 | * and should be free in persistent map; | |
2820 | */ | |
2821 | if (!(le32_to_cpu(iagp->wmap[extno]) & mask)) { | |
2822 | release_metapage(mp); | |
2823 | jfs_error(ipimap->i_sb, | |
2824 | "diUpdatePMap: the inode is not allocated in " | |
2825 | "the working map"); | |
2826 | return -EIO; | |
2827 | } | |
2828 | if ((le32_to_cpu(iagp->pmap[extno]) & mask) != 0) { | |
2829 | release_metapage(mp); | |
2830 | jfs_error(ipimap->i_sb, | |
2831 | "diUpdatePMap: the inode is not free in the " | |
2832 | "persistent map"); | |
2833 | return -EIO; | |
2834 | } | |
2835 | /* update the bitmap for the extent of the allocated inode */ | |
2836 | iagp->pmap[extno] |= cpu_to_le32(mask); | |
2837 | } | |
2838 | /* | |
2839 | * update iag lsn | |
2840 | */ | |
2841 | lsn = tblk->lsn; | |
2842 | log = JFS_SBI(tblk->sb)->log; | |
2843 | if (mp->lsn != 0) { | |
2844 | /* inherit older/smaller lsn */ | |
2845 | logdiff(difft, lsn, log); | |
2846 | logdiff(diffp, mp->lsn, log); | |
7fab479b | 2847 | LOGSYNC_LOCK(log, flags); |
1da177e4 LT |
2848 | if (difft < diffp) { |
2849 | mp->lsn = lsn; | |
2850 | /* move mp after tblock in logsync list */ | |
1da177e4 | 2851 | list_move(&mp->synclist, &tblk->synclist); |
1da177e4 LT |
2852 | } |
2853 | /* inherit younger/larger clsn */ | |
1da177e4 LT |
2854 | assert(mp->clsn); |
2855 | logdiff(difft, tblk->clsn, log); | |
2856 | logdiff(diffp, mp->clsn, log); | |
2857 | if (difft > diffp) | |
2858 | mp->clsn = tblk->clsn; | |
7fab479b | 2859 | LOGSYNC_UNLOCK(log, flags); |
1da177e4 LT |
2860 | } else { |
2861 | mp->log = log; | |
2862 | mp->lsn = lsn; | |
2863 | /* insert mp after tblock in logsync list */ | |
7fab479b | 2864 | LOGSYNC_LOCK(log, flags); |
1da177e4 LT |
2865 | log->count++; |
2866 | list_add(&mp->synclist, &tblk->synclist); | |
2867 | mp->clsn = tblk->clsn; | |
7fab479b | 2868 | LOGSYNC_UNLOCK(log, flags); |
1da177e4 LT |
2869 | } |
2870 | write_metapage(mp); | |
2871 | return (0); | |
2872 | } | |
2873 | ||
2874 | /* | |
2875 | * diExtendFS() | |
2876 | * | |
2877 | * function: update imap for extendfs(); | |
2878 | * | |
2879 | * note: AG size has been increased s.t. each k old contiguous AGs are | |
2880 | * coalesced into a new AG; | |
2881 | */ | |
2882 | int diExtendFS(struct inode *ipimap, struct inode *ipbmap) | |
2883 | { | |
2884 | int rc, rcx = 0; | |
2885 | struct inomap *imap = JFS_IP(ipimap)->i_imap; | |
2886 | struct iag *iagp = NULL, *hiagp = NULL; | |
2887 | struct bmap *mp = JFS_SBI(ipbmap->i_sb)->bmap; | |
2888 | struct metapage *bp, *hbp; | |
2889 | int i, n, head; | |
2890 | int numinos, xnuminos = 0, xnumfree = 0; | |
2891 | s64 agstart; | |
2892 | ||
2893 | jfs_info("diExtendFS: nextiag:%d numinos:%d numfree:%d", | |
2894 | imap->im_nextiag, atomic_read(&imap->im_numinos), | |
2895 | atomic_read(&imap->im_numfree)); | |
2896 | ||
2897 | /* | |
2898 | * reconstruct imap | |
2899 | * | |
2900 | * coalesce contiguous k (newAGSize/oldAGSize) AGs; | |
2901 | * i.e., (AGi, ..., AGj) where i = k*n and j = k*(n+1) - 1 to AGn; | |
2902 | * note: new AG size = old AG size * (2**x). | |
2903 | */ | |
2904 | ||
2905 | /* init per AG control information im_agctl[] */ | |
2906 | for (i = 0; i < MAXAG; i++) { | |
2907 | imap->im_agctl[i].inofree = -1; | |
2908 | imap->im_agctl[i].extfree = -1; | |
2909 | imap->im_agctl[i].numinos = 0; /* number of backed inodes */ | |
2910 | imap->im_agctl[i].numfree = 0; /* number of free backed inodes */ | |
2911 | } | |
2912 | ||
2913 | /* | |
2914 | * process each iag page of the map. | |
2915 | * | |
2916 | * rebuild AG Free Inode List, AG Free Inode Extent List; | |
2917 | */ | |
2918 | for (i = 0; i < imap->im_nextiag; i++) { | |
2919 | if ((rc = diIAGRead(imap, i, &bp))) { | |
2920 | rcx = rc; | |
2921 | continue; | |
2922 | } | |
2923 | iagp = (struct iag *) bp->data; | |
2924 | if (le32_to_cpu(iagp->iagnum) != i) { | |
2925 | release_metapage(bp); | |
2926 | jfs_error(ipimap->i_sb, | |
2927 | "diExtendFs: unexpected value of iagnum"); | |
2928 | return -EIO; | |
2929 | } | |
2930 | ||
2931 | /* leave free iag in the free iag list */ | |
2932 | if (iagp->nfreeexts == cpu_to_le32(EXTSPERIAG)) { | |
2933 | release_metapage(bp); | |
2934 | continue; | |
2935 | } | |
2936 | ||
2937 | /* agstart that computes to the same ag is treated as same; */ | |
2938 | agstart = le64_to_cpu(iagp->agstart); | |
2939 | /* iagp->agstart = agstart & ~(mp->db_agsize - 1); */ | |
2940 | n = agstart >> mp->db_agl2size; | |
2941 | ||
2942 | /* compute backed inodes */ | |
2943 | numinos = (EXTSPERIAG - le32_to_cpu(iagp->nfreeexts)) | |
2944 | << L2INOSPEREXT; | |
2945 | if (numinos > 0) { | |
2946 | /* merge AG backed inodes */ | |
2947 | imap->im_agctl[n].numinos += numinos; | |
2948 | xnuminos += numinos; | |
2949 | } | |
2950 | ||
2951 | /* if any backed free inodes, insert at AG free inode list */ | |
2952 | if ((int) le32_to_cpu(iagp->nfreeinos) > 0) { | |
2953 | if ((head = imap->im_agctl[n].inofree) == -1) { | |
2954 | iagp->inofreefwd = cpu_to_le32(-1); | |
2955 | iagp->inofreeback = cpu_to_le32(-1); | |
2956 | } else { | |
2957 | if ((rc = diIAGRead(imap, head, &hbp))) { | |
2958 | rcx = rc; | |
2959 | goto nextiag; | |
2960 | } | |
2961 | hiagp = (struct iag *) hbp->data; | |
2962 | hiagp->inofreeback = iagp->iagnum; | |
2963 | iagp->inofreefwd = cpu_to_le32(head); | |
2964 | iagp->inofreeback = cpu_to_le32(-1); | |
2965 | write_metapage(hbp); | |
2966 | } | |
2967 | ||
2968 | imap->im_agctl[n].inofree = | |
2969 | le32_to_cpu(iagp->iagnum); | |
2970 | ||
2971 | /* merge AG backed free inodes */ | |
2972 | imap->im_agctl[n].numfree += | |
2973 | le32_to_cpu(iagp->nfreeinos); | |
2974 | xnumfree += le32_to_cpu(iagp->nfreeinos); | |
2975 | } | |
2976 | ||
2977 | /* if any free extents, insert at AG free extent list */ | |
2978 | if (le32_to_cpu(iagp->nfreeexts) > 0) { | |
2979 | if ((head = imap->im_agctl[n].extfree) == -1) { | |
2980 | iagp->extfreefwd = cpu_to_le32(-1); | |
2981 | iagp->extfreeback = cpu_to_le32(-1); | |
2982 | } else { | |
2983 | if ((rc = diIAGRead(imap, head, &hbp))) { | |
2984 | rcx = rc; | |
2985 | goto nextiag; | |
2986 | } | |
2987 | hiagp = (struct iag *) hbp->data; | |
2988 | hiagp->extfreeback = iagp->iagnum; | |
2989 | iagp->extfreefwd = cpu_to_le32(head); | |
2990 | iagp->extfreeback = cpu_to_le32(-1); | |
2991 | write_metapage(hbp); | |
2992 | } | |
2993 | ||
2994 | imap->im_agctl[n].extfree = | |
2995 | le32_to_cpu(iagp->iagnum); | |
2996 | } | |
2997 | ||
2998 | nextiag: | |
2999 | write_metapage(bp); | |
3000 | } | |
3001 | ||
3002 | if (xnuminos != atomic_read(&imap->im_numinos) || | |
3003 | xnumfree != atomic_read(&imap->im_numfree)) { | |
3004 | jfs_error(ipimap->i_sb, | |
3005 | "diExtendFs: numinos or numfree incorrect"); | |
3006 | return -EIO; | |
3007 | } | |
3008 | ||
3009 | return rcx; | |
3010 | } | |
3011 | ||
3012 | ||
3013 | /* | |
3014 | * duplicateIXtree() | |
3015 | * | |
3016 | * serialization: IWRITE_LOCK held on entry/exit | |
3017 | * | |
3018 | * note: shadow page with regular inode (rel.2); | |
3019 | */ | |
3020 | static void duplicateIXtree(struct super_block *sb, s64 blkno, | |
3021 | int xlen, s64 *xaddr) | |
3022 | { | |
3023 | struct jfs_superblock *j_sb; | |
3024 | struct buffer_head *bh; | |
3025 | struct inode *ip; | |
3026 | tid_t tid; | |
3027 | ||
3028 | /* if AIT2 ipmap2 is bad, do not try to update it */ | |
3029 | if (JFS_SBI(sb)->mntflag & JFS_BAD_SAIT) /* s_flag */ | |
3030 | return; | |
3031 | ip = diReadSpecial(sb, FILESYSTEM_I, 1); | |
3032 | if (ip == NULL) { | |
3033 | JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT; | |
3034 | if (readSuper(sb, &bh)) | |
3035 | return; | |
3036 | j_sb = (struct jfs_superblock *)bh->b_data; | |
3037 | j_sb->s_flag |= cpu_to_le32(JFS_BAD_SAIT); | |
3038 | ||
3039 | mark_buffer_dirty(bh); | |
3040 | sync_dirty_buffer(bh); | |
3041 | brelse(bh); | |
3042 | return; | |
3043 | } | |
3044 | ||
3045 | /* start transaction */ | |
3046 | tid = txBegin(sb, COMMIT_FORCE); | |
3047 | /* update the inode map addressing structure to point to it */ | |
3048 | if (xtInsert(tid, ip, 0, blkno, xlen, xaddr, 0)) { | |
3049 | JFS_SBI(sb)->mntflag |= JFS_BAD_SAIT; | |
3050 | txAbort(tid, 1); | |
3051 | goto cleanup; | |
3052 | ||
3053 | } | |
3054 | /* update the inode map's inode to reflect the extension */ | |
3055 | ip->i_size += PSIZE; | |
3056 | inode_add_bytes(ip, PSIZE); | |
3057 | txCommit(tid, 1, &ip, COMMIT_FORCE); | |
3058 | cleanup: | |
3059 | txEnd(tid); | |
3060 | diFreeSpecial(ip); | |
3061 | } | |
3062 | ||
3063 | /* | |
3064 | * NAME: copy_from_dinode() | |
3065 | * | |
3066 | * FUNCTION: Copies inode info from disk inode to in-memory inode | |
3067 | * | |
3068 | * RETURN VALUES: | |
3069 | * 0 - success | |
3070 | * -ENOMEM - insufficient memory | |
3071 | */ | |
3072 | static int copy_from_dinode(struct dinode * dip, struct inode *ip) | |
3073 | { | |
3074 | struct jfs_inode_info *jfs_ip = JFS_IP(ip); | |
3075 | ||
3076 | jfs_ip->fileset = le32_to_cpu(dip->di_fileset); | |
3077 | jfs_ip->mode2 = le32_to_cpu(dip->di_mode); | |
3078 | ||
3079 | ip->i_mode = le32_to_cpu(dip->di_mode) & 0xffff; | |
3080 | ip->i_nlink = le32_to_cpu(dip->di_nlink); | |
3081 | ip->i_uid = le32_to_cpu(dip->di_uid); | |
3082 | ip->i_gid = le32_to_cpu(dip->di_gid); | |
3083 | ip->i_size = le64_to_cpu(dip->di_size); | |
3084 | ip->i_atime.tv_sec = le32_to_cpu(dip->di_atime.tv_sec); | |
3085 | ip->i_atime.tv_nsec = le32_to_cpu(dip->di_atime.tv_nsec); | |
3086 | ip->i_mtime.tv_sec = le32_to_cpu(dip->di_mtime.tv_sec); | |
3087 | ip->i_mtime.tv_nsec = le32_to_cpu(dip->di_mtime.tv_nsec); | |
3088 | ip->i_ctime.tv_sec = le32_to_cpu(dip->di_ctime.tv_sec); | |
3089 | ip->i_ctime.tv_nsec = le32_to_cpu(dip->di_ctime.tv_nsec); | |
3090 | ip->i_blksize = ip->i_sb->s_blocksize; | |
3091 | ip->i_blocks = LBLK2PBLK(ip->i_sb, le64_to_cpu(dip->di_nblocks)); | |
3092 | ip->i_generation = le32_to_cpu(dip->di_gen); | |
3093 | ||
3094 | jfs_ip->ixpxd = dip->di_ixpxd; /* in-memory pxd's are little-endian */ | |
3095 | jfs_ip->acl = dip->di_acl; /* as are dxd's */ | |
3096 | jfs_ip->ea = dip->di_ea; | |
3097 | jfs_ip->next_index = le32_to_cpu(dip->di_next_index); | |
3098 | jfs_ip->otime = le32_to_cpu(dip->di_otime.tv_sec); | |
3099 | jfs_ip->acltype = le32_to_cpu(dip->di_acltype); | |
3100 | ||
3101 | if (S_ISCHR(ip->i_mode) || S_ISBLK(ip->i_mode)) { | |
3102 | jfs_ip->dev = le32_to_cpu(dip->di_rdev); | |
3103 | ip->i_rdev = new_decode_dev(jfs_ip->dev); | |
3104 | } | |
3105 | ||
3106 | if (S_ISDIR(ip->i_mode)) { | |
3107 | memcpy(&jfs_ip->i_dirtable, &dip->di_dirtable, 384); | |
3108 | } else if (S_ISREG(ip->i_mode) || S_ISLNK(ip->i_mode)) { | |
3109 | memcpy(&jfs_ip->i_xtroot, &dip->di_xtroot, 288); | |
3110 | } else | |
3111 | memcpy(&jfs_ip->i_inline_ea, &dip->di_inlineea, 128); | |
3112 | ||
3113 | /* Zero the in-memory-only stuff */ | |
3114 | jfs_ip->cflag = 0; | |
3115 | jfs_ip->btindex = 0; | |
3116 | jfs_ip->btorder = 0; | |
3117 | jfs_ip->bxflag = 0; | |
3118 | jfs_ip->blid = 0; | |
3119 | jfs_ip->atlhead = 0; | |
3120 | jfs_ip->atltail = 0; | |
3121 | jfs_ip->xtlid = 0; | |
3122 | return (0); | |
3123 | } | |
3124 | ||
3125 | /* | |
3126 | * NAME: copy_to_dinode() | |
3127 | * | |
3128 | * FUNCTION: Copies inode info from in-memory inode to disk inode | |
3129 | */ | |
3130 | static void copy_to_dinode(struct dinode * dip, struct inode *ip) | |
3131 | { | |
3132 | struct jfs_inode_info *jfs_ip = JFS_IP(ip); | |
3133 | ||
3134 | dip->di_fileset = cpu_to_le32(jfs_ip->fileset); | |
3135 | dip->di_inostamp = cpu_to_le32(JFS_SBI(ip->i_sb)->inostamp); | |
3136 | dip->di_number = cpu_to_le32(ip->i_ino); | |
3137 | dip->di_gen = cpu_to_le32(ip->i_generation); | |
3138 | dip->di_size = cpu_to_le64(ip->i_size); | |
3139 | dip->di_nblocks = cpu_to_le64(PBLK2LBLK(ip->i_sb, ip->i_blocks)); | |
3140 | dip->di_nlink = cpu_to_le32(ip->i_nlink); | |
3141 | dip->di_uid = cpu_to_le32(ip->i_uid); | |
3142 | dip->di_gid = cpu_to_le32(ip->i_gid); | |
3143 | /* | |
3144 | * mode2 is only needed for storing the higher order bits. | |
3145 | * Trust i_mode for the lower order ones | |
3146 | */ | |
3147 | dip->di_mode = cpu_to_le32((jfs_ip->mode2 & 0xffff0000) | ip->i_mode); | |
3148 | dip->di_atime.tv_sec = cpu_to_le32(ip->i_atime.tv_sec); | |
3149 | dip->di_atime.tv_nsec = cpu_to_le32(ip->i_atime.tv_nsec); | |
3150 | dip->di_ctime.tv_sec = cpu_to_le32(ip->i_ctime.tv_sec); | |
3151 | dip->di_ctime.tv_nsec = cpu_to_le32(ip->i_ctime.tv_nsec); | |
3152 | dip->di_mtime.tv_sec = cpu_to_le32(ip->i_mtime.tv_sec); | |
3153 | dip->di_mtime.tv_nsec = cpu_to_le32(ip->i_mtime.tv_nsec); | |
3154 | dip->di_ixpxd = jfs_ip->ixpxd; /* in-memory pxd's are little-endian */ | |
3155 | dip->di_acl = jfs_ip->acl; /* as are dxd's */ | |
3156 | dip->di_ea = jfs_ip->ea; | |
3157 | dip->di_next_index = cpu_to_le32(jfs_ip->next_index); | |
3158 | dip->di_otime.tv_sec = cpu_to_le32(jfs_ip->otime); | |
3159 | dip->di_otime.tv_nsec = 0; | |
3160 | dip->di_acltype = cpu_to_le32(jfs_ip->acltype); | |
3161 | if (S_ISCHR(ip->i_mode) || S_ISBLK(ip->i_mode)) | |
3162 | dip->di_rdev = cpu_to_le32(jfs_ip->dev); | |
3163 | } |