Commit | Line | Data |
---|---|---|
84c30241 KB |
1 | /* |
2 | * Copyright (c) 1991 Regents of the University of California. | |
3 | * All rights reserved. | |
4 | * | |
5 | * %sccs.include.redist.c% | |
6 | * | |
fa870596 | 7 | * @(#)lfs_segment.c 7.28 (Berkeley) %G% |
84c30241 KB |
8 | */ |
9 | ||
34a084a9 KB |
10 | #include <sys/param.h> |
11 | #include <sys/systm.h> | |
12 | #include <sys/namei.h> | |
34a084a9 | 13 | #include <sys/kernel.h> |
a1b8db53 | 14 | #include <sys/resourcevar.h> |
34a084a9 KB |
15 | #include <sys/file.h> |
16 | #include <sys/stat.h> | |
17 | #include <sys/buf.h> | |
18 | #include <sys/proc.h> | |
19 | #include <sys/conf.h> | |
20 | #include <sys/vnode.h> | |
34a084a9 KB |
21 | #include <sys/malloc.h> |
22 | #include <sys/mount.h> | |
23 | ||
33d38333 KM |
24 | #include <miscfs/specfs/specdev.h> |
25 | #include <miscfs/fifofs/fifo.h> | |
26 | ||
0a011bb1 KB |
27 | #include <ufs/ufs/quota.h> |
28 | #include <ufs/ufs/inode.h> | |
29 | #include <ufs/ufs/dir.h> | |
30 | #include <ufs/ufs/ufsmount.h> | |
34a084a9 | 31 | |
0a011bb1 KB |
32 | #include <ufs/lfs/lfs.h> |
33 | #include <ufs/lfs/lfs_extern.h> | |
84c30241 | 34 | |
dc7e45d3 | 35 | /* In-memory description of a segment about to be written. */ |
dc7e45d3 | 36 | struct segment { |
a1b8db53 KB |
37 | struct buf **bpp; /* pointer to buffer array */ |
38 | struct buf **cbpp; /* pointer to next available bp */ | |
39 | struct buf *ibp; /* buffer pointer to inode page */ | |
40 | struct finfo *fip; /* current fileinfo pointer */ | |
dc7e45d3 KB |
41 | void *segsum; /* segment summary info */ |
42 | u_long ninodes; /* number of inodes in this segment */ | |
43 | u_long seg_bytes_left; /* bytes left in segment */ | |
44 | u_long sum_bytes_left; /* bytes left in summary block */ | |
45 | u_long seg_number; /* number of this segment */ | |
46 | #define SEGM_CKP 0x01 /* doing a checkpoint */ | |
47 | u_long seg_flags; /* run-time flags for this segment */ | |
dc7e45d3 KB |
48 | }; |
49 | ||
84c30241 | 50 | /* |
dc7e45d3 KB |
51 | * Determine if it's OK to start a partial in this segment, or if we need |
52 | * to go on to a new segment. | |
8954e52c | 53 | */ |
dc7e45d3 KB |
54 | #define LFS_PARTIAL_FITS(fs) \ |
55 | ((fs)->lfs_dbpseg - ((fs)->lfs_offset - (fs)->lfs_curseg) > \ | |
56 | 1 << (fs)->lfs_fsbtodb) | |
57 | ||
80443139 | 58 | void lfs_callback __P((struct buf *)); |
a1b8db53 KB |
59 | void lfs_gather __P((struct lfs *, struct segment *, |
60 | struct vnode *, int (*) __P((struct lfs *, struct buf *)))); | |
61 | void lfs_initseg __P((struct lfs *, struct segment *)); | |
62 | void lfs_iset __P((struct inode *, daddr_t, time_t)); | |
63 | int lfs_match_data __P((struct lfs *, struct buf *)); | |
64 | int lfs_match_dindir __P((struct lfs *, struct buf *)); | |
65 | int lfs_match_indir __P((struct lfs *, struct buf *)); | |
66 | int lfs_match_tindir __P((struct lfs *, struct buf *)); | |
67 | struct buf * | |
4c60cc5b | 68 | lfs_newbuf __P((struct lfs *, daddr_t, size_t)); |
87804018 | 69 | void lfs_newseg __P((struct lfs *)); |
a1b8db53 | 70 | void lfs_shellsort __P((struct buf **, daddr_t *, register int)); |
87804018 | 71 | void lfs_updatemeta __P((struct lfs *, |
a1b8db53 KB |
72 | struct segment *, struct vnode *, daddr_t *, struct buf **, int)); |
73 | void lfs_writefile __P((struct lfs *, struct segment *, struct vnode *)); | |
3ce71481 KB |
74 | int lfs_writeinode __P((struct lfs *, struct segment *, struct inode *)); |
75 | int lfs_writeseg __P((struct lfs *, struct segment *)); | |
a1b8db53 | 76 | void lfs_writesuper __P((struct lfs *, struct segment *)); |
3ce71481 KB |
77 | void lfs_writevnodes __P((struct lfs *fs, struct mount *mp, |
78 | struct segment *sp, int dirops)); | |
84c30241 | 79 | |
dc7e45d3 KB |
80 | int lfs_allclean_wakeup; /* Cleaner wakeup address. */ |
81 | ||
44819c56 KB |
82 | /* |
83 | * Ifile and meta data blocks are not marked busy, so segment writes MUST be | |
84 | * single threaded. Currently, there are two paths into lfs_segwrite, sync() | |
85 | * and getnewbuf(). They both mark the file system busy. Lfs_vflush() | |
86 | * explicitly marks the file system busy. So lfs_segwrite is safe. I think. | |
87 | */ | |
88 | ||
89 | int | |
90 | lfs_vflush(vp) | |
91 | struct vnode *vp; | |
92 | { | |
93 | struct inode *ip; | |
94 | struct lfs *fs; | |
44819c56 KB |
95 | struct segment *sp; |
96 | int error, s; | |
97 | ||
200cb75d KB |
98 | fs = VFSTOUFS(vp->v_mount)->um_lfs; |
99 | lfs_seglock(fs); | |
44819c56 KB |
100 | |
101 | /* | |
102 | * Allocate a segment structure and enough space to hold pointers to | |
103 | * the maximum possible number of buffers which can be described in a | |
104 | * single summary block. | |
105 | */ | |
106 | sp = malloc(sizeof(struct segment), M_SEGMENT, M_WAITOK); | |
107 | sp->bpp = malloc(((LFS_SUMMARY_SIZE - sizeof(SEGSUM)) / | |
108 | sizeof(daddr_t) + 1) * sizeof(struct buf *), M_SEGMENT, M_WAITOK); | |
109 | sp->seg_flags = SEGM_CKP; | |
44819c56 KB |
110 | |
111 | /* | |
112 | * Keep a cumulative count of the outstanding I/O operations. If the | |
113 | * disk drive catches up with us it could go to zero before we finish, | |
114 | * so we artificially increment it by one until we've scheduled all of | |
115 | * the writes we intend to do. | |
116 | */ | |
117 | s = splbio(); | |
4c60cc5b | 118 | ++fs->lfs_iocount; |
44819c56 KB |
119 | splx(s); |
120 | ||
44819c56 | 121 | ip = VTOI(vp); |
290d4594 | 122 | do { |
685d5160 | 123 | lfs_initseg(fs, sp); |
290d4594 KB |
124 | do { |
125 | if (vp->v_dirtyblkhd != NULL) | |
126 | lfs_writefile(fs, sp, vp); | |
127 | } while (lfs_writeinode(fs, sp, ip)); | |
128 | ip->i_flags &= ~(IMOD | IACC | IUPD | ICHG); | |
44819c56 | 129 | |
290d4594 | 130 | } while (lfs_writeseg(fs, sp) && ip->i_number == LFS_IFILE_INUM); |
44819c56 KB |
131 | |
132 | /* | |
133 | * If the I/O count is non-zero, sleep until it reaches zero. At the | |
134 | * moment, the user's process hangs around so we can sleep. | |
135 | */ | |
136 | s = splbio(); | |
137 | if (--fs->lfs_iocount && (error = | |
14df3d9f KB |
138 | tsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs vflush", 0))) { |
139 | free(sp->bpp, M_SEGMENT); | |
140 | free(sp, M_SEGMENT); | |
44819c56 | 141 | return (error); |
14df3d9f | 142 | } |
44819c56 | 143 | splx(s); |
200cb75d | 144 | lfs_segunlock(fs); |
44819c56 | 145 | |
14df3d9f KB |
146 | /* |
147 | * XXX | |
148 | * Should be writing a checkpoint? | |
149 | */ | |
44819c56 KB |
150 | free(sp->bpp, M_SEGMENT); |
151 | free(sp, M_SEGMENT); | |
152 | ||
153 | return (0); | |
154 | } | |
155 | ||
3ce71481 KB |
156 | void |
157 | lfs_writevnodes(fs, mp, sp, dirops) | |
158 | struct lfs *fs; | |
159 | struct mount *mp; | |
160 | struct segment *sp; | |
161 | int dirops; | |
162 | { | |
163 | struct inode *ip; | |
164 | struct vnode *vp; | |
165 | int error, s; | |
166 | ||
167 | loop: for (vp = mp->mnt_mounth; vp; vp = vp->v_mountf) { | |
168 | /* | |
169 | * If the vnode that we are about to sync is no longer | |
170 | * associated with this mount point, start over. | |
171 | */ | |
172 | if (vp->v_mount != mp) | |
173 | goto loop; | |
174 | ||
175 | if (dirops && !(vp->v_flag & VDIROP) || | |
176 | !dirops && (vp->v_flag & VDIROP)) | |
177 | continue; | |
178 | /* | |
179 | * XXX | |
180 | * Up the ref count so we don't get tossed out of | |
181 | * memory. | |
182 | */ | |
183 | VREF(vp); | |
184 | ||
185 | /* | |
186 | * Write the inode/file if dirty and it's not the | |
187 | * the IFILE. | |
188 | */ | |
189 | ip = VTOI(vp); | |
190 | if ((ip->i_flag & (IMOD | IACC | IUPD | ICHG) || | |
191 | vp->v_dirtyblkhd != NULL) && | |
192 | ip->i_number != LFS_IFILE_INUM) { | |
193 | if (vp->v_dirtyblkhd != NULL) | |
194 | lfs_writefile(fs, sp, vp); | |
195 | (void) lfs_writeinode(fs, sp, ip); | |
196 | ip->i_flags &= ~(IMOD | IACC | IUPD | ICHG); | |
197 | } | |
198 | vp->v_flag &= ~VDIROP; | |
199 | vrele(vp); | |
200 | } | |
201 | } | |
202 | ||
84c30241 | 203 | int |
275ca4f0 | 204 | lfs_segwrite(mp, do_ckp) |
a1b8db53 | 205 | struct mount *mp; |
dc7e45d3 | 206 | int do_ckp; /* Do a checkpoint. */ |
84c30241 | 207 | { |
5b4e3ef5 | 208 | struct buf *bp; |
a1b8db53 | 209 | struct inode *ip; |
0a011bb1 | 210 | struct lfs *fs; |
a1b8db53 KB |
211 | struct segment *sp; |
212 | struct vnode *vp; | |
5b4e3ef5 KB |
213 | SEGUSE *segusep; |
214 | daddr_t ibno; | |
215 | int error, i, s; | |
84c30241 | 216 | |
44819c56 | 217 | fs = VFSTOUFS(mp)->um_lfs; |
200cb75d | 218 | lfs_seglock(fs); |
44819c56 KB |
219 | |
220 | /* | |
221 | * Allocate a segment structure and enough space to hold pointers to | |
222 | * the maximum possible number of buffers which can be described in a | |
223 | * single summary block. | |
224 | */ | |
225 | sp = malloc(sizeof(struct segment), M_SEGMENT, M_WAITOK); | |
226 | sp->bpp = malloc(((LFS_SUMMARY_SIZE - sizeof(SEGSUM)) / | |
227 | sizeof(daddr_t) + 1) * sizeof(struct buf *), M_SEGMENT, M_WAITOK); | |
228 | sp->seg_flags = do_ckp ? SEGM_CKP : 0; | |
229 | lfs_initseg(fs, sp); | |
a1b8db53 | 230 | |
8954e52c | 231 | /* |
4c60cc5b KB |
232 | * Keep a cumulative count of the outstanding I/O operations. If the |
233 | * disk drive catches up with us it could go to zero before we finish, | |
234 | * so we artificially increment it by one until we've scheduled all of | |
235 | * the writes we intend to do. If not a checkpoint, we never do the | |
236 | * final decrement, avoiding the wakeup in the callback routine. | |
8954e52c | 237 | */ |
4c60cc5b | 238 | s = splbio(); |
290d4594 | 239 | ++fs->lfs_iocount; |
4c60cc5b | 240 | splx(s); |
aa4dc149 | 241 | |
3ce71481 | 242 | lfs_writevnodes(fs, mp, sp, 0); |
3ce71481 KB |
243 | fs->lfs_writer = 1; |
244 | if (fs->lfs_dirops && (error = | |
245 | tsleep(&fs->lfs_writer, PRIBIO + 1, "lfs writer", 0))) { | |
246 | free(sp->bpp, M_SEGMENT); | |
247 | free(sp, M_SEGMENT); | |
248 | fs->lfs_writer = 0; | |
290d4594 | 249 | return (error); |
3ce71481 | 250 | } |
dc7e45d3 | 251 | |
3ce71481 | 252 | lfs_writevnodes(fs, mp, sp, 1); |
dc7e45d3 | 253 | |
3ce71481 | 254 | /* |
5b4e3ef5 KB |
255 | * If we are doing a checkpoint, mark everything since the |
256 | * last checkpoint as no longer ACTIVE. | |
3ce71481 | 257 | */ |
5b4e3ef5 KB |
258 | if (do_ckp) |
259 | for (ibno = fs->lfs_cleansz + fs->lfs_segtabsz; | |
260 | --ibno >= fs->lfs_cleansz; ) { | |
261 | if (bread(fs->lfs_ivnode, ibno, fs->lfs_bsize, | |
262 | NOCRED, &bp)) | |
263 | ||
264 | panic("lfs: ifile read"); | |
265 | segusep = (SEGUSE *)bp->b_un.b_addr; | |
266 | for (i = fs->lfs_sepb; i--; segusep++) | |
267 | segusep->su_flags &= ~SEGUSE_ACTIVE; | |
268 | ||
269 | LFS_UBWRITE(bp); | |
270 | } | |
271 | ||
3ce71481 | 272 | if (do_ckp || fs->lfs_doifile) { |
87804018 KB |
273 | vp = fs->lfs_ivnode; |
274 | while (vget(vp)); | |
275 | ip = VTOI(vp); | |
5b4e3ef5 KB |
276 | if (vp->v_dirtyblkhd != NULL) |
277 | lfs_writefile(fs, sp, vp); | |
278 | (void)lfs_writeinode(fs, sp, ip); | |
87804018 KB |
279 | ip->i_flags &= ~(IMOD | IACC | IUPD | ICHG); |
280 | vput(vp); | |
5b4e3ef5 KB |
281 | /* |
282 | * This should never happen because we just guaranteed | |
283 | * that all the segment usage table blocks are dirty, so | |
284 | * no new ones should get written. | |
285 | */ | |
286 | if (lfs_writeseg(fs, sp) && do_ckp) | |
287 | panic("lfs_segwrite: created dirty blocks on ckp"); | |
3ce71481 KB |
288 | } else |
289 | (void) lfs_writeseg(fs, sp); | |
aa4dc149 | 290 | |
275ca4f0 | 291 | /* |
dc7e45d3 KB |
292 | * If the I/O count is non-zero, sleep until it reaches zero. At the |
293 | * moment, the user's process hangs around so we can sleep. | |
275ca4f0 | 294 | */ |
3ce71481 KB |
295 | fs->lfs_writer = 0; |
296 | fs->lfs_doifile = 0; | |
297 | wakeup(&fs->lfs_dirops); | |
298 | ||
290d4594 KB |
299 | s = splbio(); |
300 | --fs->lfs_iocount; | |
dc7e45d3 | 301 | if (do_ckp) { |
4c60cc5b | 302 | if (fs->lfs_iocount && (error = |
14df3d9f KB |
303 | tsleep(&fs->lfs_iocount, PRIBIO + 1, "lfs sync", 0))) { |
304 | free(sp->bpp, M_SEGMENT); | |
305 | free(sp, M_SEGMENT); | |
12304d41 | 306 | return (error); |
14df3d9f | 307 | } |
dc7e45d3 KB |
308 | splx(s); |
309 | lfs_writesuper(fs, sp); | |
4c60cc5b KB |
310 | } else |
311 | splx(s); | |
275ca4f0 | 312 | |
200cb75d KB |
313 | lfs_segunlock(fs); |
314 | ||
c222b129 KB |
315 | free(sp->bpp, M_SEGMENT); |
316 | free(sp, M_SEGMENT); | |
275ca4f0 | 317 | |
dc7e45d3 | 318 | return (0); |
84c30241 KB |
319 | } |
320 | ||
dc7e45d3 KB |
321 | /* |
322 | * Write the dirty blocks associated with a vnode. | |
323 | */ | |
87804018 | 324 | void |
dc7e45d3 | 325 | lfs_writefile(fs, sp, vp) |
0a011bb1 | 326 | struct lfs *fs; |
a1b8db53 KB |
327 | struct segment *sp; |
328 | struct vnode *vp; | |
84c30241 | 329 | { |
dc7e45d3 | 330 | struct buf *bp; |
a1b8db53 | 331 | struct finfo *fip; |
dc7e45d3 | 332 | IFILE *ifp; |
275ca4f0 | 333 | |
a1b8db53 KB |
334 | if (sp->seg_bytes_left < fs->lfs_bsize || |
335 | sp->sum_bytes_left < sizeof(struct finfo)) { | |
3ce71481 | 336 | (void) lfs_writeseg(fs, sp); |
a1b8db53 KB |
337 | lfs_initseg(fs, sp); |
338 | } | |
339 | sp->sum_bytes_left -= sizeof(struct finfo) - sizeof(daddr_t); | |
84c30241 | 340 | |
a1b8db53 KB |
341 | fip = sp->fip; |
342 | fip->fi_nblocks = 0; | |
343 | fip->fi_ino = VTOI(vp)->i_number; | |
344 | LFS_IENTRY(ifp, fs, fip->fi_ino, bp); | |
345 | fip->fi_version = ifp->if_version; | |
346 | brelse(bp); | |
347 | ||
348 | /* | |
349 | * It may not be necessary to write the meta-data blocks at this point, | |
350 | * as the roll-forward recovery code should be able to reconstruct the | |
351 | * list. | |
352 | */ | |
353 | lfs_gather(fs, sp, vp, lfs_match_data); | |
354 | lfs_gather(fs, sp, vp, lfs_match_indir); | |
355 | lfs_gather(fs, sp, vp, lfs_match_dindir); | |
dc7e45d3 | 356 | #ifdef TRIPLE |
a1b8db53 | 357 | lfs_gather(fs, sp, vp, lfs_match_tindir); |
dc7e45d3 | 358 | #endif |
aa4dc149 | 359 | |
a1b8db53 | 360 | fip = sp->fip; |
dc7e45d3 | 361 | #ifdef META |
a1b8db53 | 362 | printf("lfs_writefile: adding %d blocks\n", fip->fi_nblocks); |
dc7e45d3 | 363 | #endif |
a1b8db53 KB |
364 | if (fip->fi_nblocks != 0) { |
365 | ++((SEGSUM *)(sp->segsum))->ss_nfinfo; | |
366 | sp->fip = | |
367 | (struct finfo *)((caddr_t)fip + sizeof(struct finfo) + | |
368 | sizeof(daddr_t) * (fip->fi_nblocks - 1)); | |
9a46ddb2 CS |
369 | } else |
370 | sp->sum_bytes_left += sizeof(struct finfo) - sizeof(daddr_t); | |
12304d41 KB |
371 | } |
372 | ||
3ce71481 | 373 | int |
12304d41 KB |
374 | lfs_writeinode(fs, sp, ip) |
375 | struct lfs *fs; | |
a1b8db53 KB |
376 | struct segment *sp; |
377 | struct inode *ip; | |
12304d41 | 378 | { |
a1b8db53 | 379 | struct buf *bp, *ibp; |
87804018 | 380 | IFILE *ifp; |
9a46ddb2 CS |
381 | SEGUSE *sup; |
382 | daddr_t daddr; | |
87804018 | 383 | ino_t ino; |
12304d41 | 384 | int ndx; |
3ce71481 | 385 | int redo_ifile = 0; |
12304d41 | 386 | |
12304d41 KB |
387 | /* Allocate a new inode block if necessary. */ |
388 | if (sp->ibp == NULL) { | |
389 | /* Allocate a new segment if necessary. */ | |
390 | if (sp->seg_bytes_left < fs->lfs_bsize || | |
391 | sp->sum_bytes_left < sizeof(daddr_t)) { | |
3ce71481 | 392 | (void) lfs_writeseg(fs, sp); |
12304d41 KB |
393 | lfs_initseg(fs, sp); |
394 | } | |
395 | ||
396 | /* Get next inode block. */ | |
9a46ddb2 | 397 | daddr = fs->lfs_offset; |
12304d41 KB |
398 | fs->lfs_offset += fsbtodb(fs, 1); |
399 | sp->ibp = *sp->cbpp++ = | |
4c60cc5b | 400 | lfs_newbuf(fs, daddr, fs->lfs_bsize); |
12304d41 | 401 | |
4c60cc5b | 402 | /* Set remaining space counters. */ |
12304d41 KB |
403 | sp->seg_bytes_left -= fs->lfs_bsize; |
404 | sp->sum_bytes_left -= sizeof(daddr_t); | |
87804018 | 405 | ndx = LFS_SUMMARY_SIZE / sizeof(daddr_t) - |
12304d41 | 406 | sp->ninodes / INOPB(fs) - 1; |
9a46ddb2 | 407 | ((daddr_t *)(sp->segsum))[ndx] = daddr; |
12304d41 KB |
408 | } |
409 | ||
a1b8db53 | 410 | /* Update the inode times and copy the inode onto the inode page. */ |
87804018 | 411 | ITIMES(ip, &time, &time); |
12304d41 | 412 | bp = sp->ibp; |
a1b8db53 | 413 | bp->b_un.b_dino[sp->ninodes % INOPB(fs)] = ip->i_din; |
12304d41 KB |
414 | |
415 | /* Increment inode count in segment summary block. */ | |
416 | ++((SEGSUM *)(sp->segsum))->ss_ninos; | |
417 | ||
418 | /* If this page is full, set flag to allocate a new page. */ | |
419 | if (++sp->ninodes % INOPB(fs) == 0) | |
420 | sp->ibp = NULL; | |
421 | ||
422 | /* | |
87804018 KB |
423 | * If updating the ifile, update the super-block. Update the disk |
424 | * address and access times for this inode in the ifile. | |
12304d41 | 425 | */ |
87804018 | 426 | ino = ip->i_number; |
4645c316 KB |
427 | if (ino == LFS_IFILE_INUM) { |
428 | daddr = fs->lfs_idaddr; | |
12304d41 | 429 | fs->lfs_idaddr = bp->b_blkno; |
4645c316 KB |
430 | } else { |
431 | LFS_IENTRY(ifp, fs, ino, ibp); | |
432 | daddr = ifp->if_daddr; | |
433 | ifp->if_daddr = bp->b_blkno; | |
434 | LFS_UBWRITE(ibp); | |
435 | } | |
9a46ddb2 | 436 | |
3ce71481 KB |
437 | /* |
438 | * No need to update segment usage if there was no former inode address | |
439 | * or if the last inode address is in the current partial segment. | |
440 | */ | |
441 | if (daddr != LFS_UNUSED_DADDR && | |
685d5160 | 442 | !(daddr >= fs->lfs_lastpseg && daddr <= bp->b_blkno)) { |
9a46ddb2 CS |
443 | LFS_SEGENTRY(sup, fs, datosn(fs, daddr), bp); |
444 | #ifdef DIAGNOSTIC | |
3ce71481 | 445 | if (sup->su_nbytes < sizeof(struct dinode)) { |
2b3ba73a KB |
446 | /* XXX -- Change to a panic. */ |
447 | printf("lfs: negative bytes (segment %d)\n", | |
9a46ddb2 | 448 | datosn(fs, daddr)); |
3ce71481 KB |
449 | panic("negative bytes"); |
450 | } | |
9a46ddb2 CS |
451 | #endif |
452 | sup->su_nbytes -= sizeof(struct dinode); | |
453 | LFS_UBWRITE(bp); | |
4645c316 | 454 | redo_ifile = (ino == LFS_IFILE_INUM && !(bp->b_flags & B_GATHERED)); |
9a46ddb2 | 455 | } |
290d4594 | 456 | return (redo_ifile); |
275ca4f0 KB |
457 | } |
458 | ||
87804018 | 459 | void |
275ca4f0 | 460 | lfs_gather(fs, sp, vp, match) |
0a011bb1 | 461 | struct lfs *fs; |
a1b8db53 KB |
462 | struct segment *sp; |
463 | struct vnode *vp; | |
464 | int (*match) __P((struct lfs *, struct buf *)); | |
275ca4f0 | 465 | { |
3ce71481 KB |
466 | struct buf **bpp, *bp; |
467 | struct buf *lastbp; | |
a1b8db53 KB |
468 | struct finfo *fip; |
469 | struct inode *ip; | |
275ca4f0 | 470 | daddr_t *lbp, *start_lbp; |
aa4dc149 KB |
471 | u_long version; |
472 | int s; | |
275ca4f0 KB |
473 | |
474 | ip = VTOI(vp); | |
475 | bpp = sp->cbpp; | |
476 | fip = sp->fip; | |
275ca4f0 | 477 | start_lbp = lbp = &fip->fi_blocks[fip->fi_nblocks]; |
275ca4f0 | 478 | |
92f4ed04 | 479 | loop: s = splbio(); |
3ce71481 KB |
480 | lastbp = NULL; |
481 | for (bp = vp->v_dirtyblkhd; bp; lastbp = bp, bp = bp->b_blockf) { | |
482 | if (bp->b_flags & B_BUSY || !match(fs, bp) || | |
483 | bp->b_flags & B_GATHERED) | |
275ca4f0 | 484 | continue; |
aa4dc149 | 485 | #ifdef DIAGNOSTIC |
dc7e45d3 | 486 | if (!(bp->b_flags & B_DELWRI)) |
12304d41 | 487 | panic("lfs_gather: bp not B_DELWRI"); |
dc7e45d3 | 488 | if (!(bp->b_flags & B_LOCKED)) |
12304d41 | 489 | panic("lfs_gather: bp not B_LOCKED"); |
aa4dc149 | 490 | #endif |
aa4dc149 | 491 | /* |
dc7e45d3 KB |
492 | * If full, finish this segment. We may be doing I/O, so |
493 | * release and reacquire the splbio(). | |
aa4dc149 KB |
494 | */ |
495 | if (sp->sum_bytes_left < sizeof(daddr_t) || | |
275ca4f0 | 496 | sp->seg_bytes_left < fs->lfs_bsize) { |
275ca4f0 | 497 | splx(s); |
aa4dc149 | 498 | lfs_updatemeta(fs, |
dc7e45d3 | 499 | sp, vp, start_lbp, bpp, lbp - start_lbp); |
275ca4f0 | 500 | |
aa4dc149 KB |
501 | /* Add the current file to the segment summary. */ |
502 | ++((SEGSUM *)(sp->segsum))->ss_nfinfo; | |
275ca4f0 | 503 | |
aa4dc149 | 504 | version = fip->fi_version; |
3ce71481 | 505 | (void) lfs_writeseg(fs, sp); |
12304d41 | 506 | lfs_initseg(fs, sp); |
aa4dc149 | 507 | |
275ca4f0 | 508 | fip = sp->fip; |
275ca4f0 | 509 | fip->fi_version = version; |
aa4dc149 | 510 | fip->fi_ino = ip->i_number; |
275ca4f0 | 511 | start_lbp = lbp = fip->fi_blocks; |
aa4dc149 | 512 | |
9a46ddb2 CS |
513 | sp->sum_bytes_left -= |
514 | sizeof(struct finfo) - sizeof(daddr_t); | |
515 | ||
aa4dc149 | 516 | bpp = sp->cbpp; |
92f4ed04 | 517 | goto loop; |
275ca4f0 | 518 | } |
9a46ddb2 CS |
519 | |
520 | /* Insert into the buffer list, update the FINFO block. */ | |
3ce71481 | 521 | bp->b_flags |= B_GATHERED; |
9a46ddb2 CS |
522 | *sp->cbpp++ = bp; |
523 | ++fip->fi_nblocks; | |
524 | *lbp++ = bp->b_lblkno; | |
525 | ||
526 | sp->sum_bytes_left -= sizeof(daddr_t); | |
527 | sp->seg_bytes_left -= bp->b_bufsize; | |
84c30241 | 528 | } |
275ca4f0 | 529 | splx(s); |
dc7e45d3 | 530 | lfs_updatemeta(fs, sp, vp, start_lbp, bpp, lbp - start_lbp); |
84c30241 KB |
531 | } |
532 | ||
84c30241 | 533 | /* |
aa4dc149 | 534 | * Update the metadata that points to the blocks listed in the FINFO |
84c30241 KB |
535 | * array. |
536 | */ | |
87804018 | 537 | void |
dc7e45d3 | 538 | lfs_updatemeta(fs, sp, vp, lbp, bpp, nblocks) |
0a011bb1 | 539 | struct lfs *fs; |
a1b8db53 KB |
540 | struct segment *sp; |
541 | struct vnode *vp; | |
275ca4f0 | 542 | daddr_t *lbp; |
a1b8db53 | 543 | struct buf **bpp; |
275ca4f0 | 544 | int nblocks; |
84c30241 | 545 | { |
12304d41 | 546 | SEGUSE *sup; |
a1b8db53 | 547 | struct buf *bp; |
dc7e45d3 | 548 | INDIR a[NIADDR], *ap; |
a1b8db53 | 549 | struct inode *ip; |
12304d41 | 550 | daddr_t daddr, lbn, off; |
dc7e45d3 | 551 | int db_per_fsb, error, i, num; |
84c30241 | 552 | |
aa4dc149 | 553 | if (nblocks == 0) |
275ca4f0 KB |
554 | return; |
555 | ||
12304d41 | 556 | /* Sort the blocks. */ |
87804018 | 557 | lfs_shellsort(bpp, lbp, nblocks); |
275ca4f0 | 558 | |
12304d41 KB |
559 | /* |
560 | * Assign disk addresses, and update references to the logical | |
561 | * block and the segment usage information. | |
562 | */ | |
dc7e45d3 | 563 | db_per_fsb = fsbtodb(fs, 1); |
12304d41 KB |
564 | for (i = nblocks; i--; ++bpp) { |
565 | lbn = *lbp++; | |
566 | (*bpp)->b_blkno = off = fs->lfs_offset; | |
dc7e45d3 | 567 | fs->lfs_offset += db_per_fsb; |
275ca4f0 | 568 | |
dc7e45d3 | 569 | if (error = lfs_bmaparray(vp, lbn, &daddr, a, &num)) |
a1b8db53 | 570 | panic("lfs_updatemeta: lfs_bmaparray %d", error); |
dc7e45d3 KB |
571 | ip = VTOI(vp); |
572 | switch (num) { | |
573 | case 0: | |
12304d41 | 574 | ip->i_db[lbn] = off; |
dc7e45d3 KB |
575 | break; |
576 | case 1: | |
12304d41 | 577 | ip->i_ib[a[0].in_off] = off; |
dc7e45d3 KB |
578 | break; |
579 | default: | |
580 | ap = &a[num - 1]; | |
dc7e45d3 KB |
581 | if (bread(vp, ap->in_lbn, fs->lfs_bsize, NOCRED, &bp)) |
582 | panic("lfs_updatemeta: bread bno %d", | |
583 | ap->in_lbn); | |
4256edb9 KB |
584 | /* |
585 | * Bread may create a new indirect block which needs | |
586 | * to get counted for the inode. | |
587 | */ | |
5b4e3ef5 | 588 | if (bp->b_blkno == -1 && !(bp->b_flags & B_CACHE)) { |
4256edb9 | 589 | ip->i_blocks += btodb(fs->lfs_bsize); |
5b4e3ef5 KB |
590 | fs->lfs_bfree -= btodb(fs->lfs_bsize); |
591 | } | |
12304d41 | 592 | bp->b_un.b_daddr[ap->in_off] = off; |
9342689a | 593 | VOP_BWRITE(bp); |
12304d41 KB |
594 | } |
595 | ||
596 | /* Update segment usage information. */ | |
597 | if (daddr != UNASSIGNED) { | |
598 | LFS_SEGENTRY(sup, fs, datosn(fs, daddr), bp); | |
12304d41 | 599 | #ifdef DIAGNOSTIC |
3ce71481 | 600 | if (sup->su_nbytes < fs->lfs_bsize) { |
2b3ba73a KB |
601 | /* XXX -- Change to a panic. */ |
602 | printf("lfs: negative bytes (segment %d)\n", | |
12304d41 | 603 | datosn(fs, daddr)); |
3ce71481 KB |
604 | panic ("Negative Bytes"); |
605 | } | |
12304d41 KB |
606 | #endif |
607 | sup->su_nbytes -= fs->lfs_bsize; | |
a1b8db53 | 608 | LFS_UBWRITE(bp); |
84c30241 | 609 | } |
84c30241 | 610 | } |
84c30241 KB |
611 | } |
612 | ||
12304d41 KB |
613 | /* |
614 | * Start a new segment. | |
615 | */ | |
87804018 | 616 | void |
12304d41 | 617 | lfs_initseg(fs, sp) |
0a011bb1 | 618 | struct lfs *fs; |
a1b8db53 | 619 | struct segment *sp; |
84c30241 | 620 | { |
12304d41 KB |
621 | SEGUSE *sup; |
622 | SEGSUM *ssp; | |
623 | struct buf *bp; | |
624 | daddr_t lbn, *lbnp; | |
275ca4f0 | 625 | |
12304d41 | 626 | /* Advance to the next segment. */ |
c222b129 | 627 | if (!LFS_PARTIAL_FITS(fs)) { |
9a46ddb2 | 628 | /* Wake up any cleaning procs waiting on this file system. */ |
4c60cc5b KB |
629 | wakeup(&fs->lfs_nextseg); |
630 | wakeup(&lfs_allclean_wakeup); | |
9a46ddb2 | 631 | |
c222b129 KB |
632 | lfs_newseg(fs); |
633 | fs->lfs_offset = fs->lfs_curseg; | |
12304d41 KB |
634 | sp->seg_number = datosn(fs, fs->lfs_curseg); |
635 | sp->seg_bytes_left = fs->lfs_dbpseg * DEV_BSIZE; | |
636 | ||
637 | /* | |
c222b129 KB |
638 | * If the segment contains a superblock, update the offset |
639 | * and summary address to skip over it. | |
12304d41 | 640 | */ |
87804018 | 641 | LFS_SEGENTRY(sup, fs, sp->seg_number, bp); |
c222b129 | 642 | if (sup->su_flags & SEGUSE_SUPERBLOCK) { |
12304d41 KB |
643 | fs->lfs_offset += LFS_SBPAD / DEV_BSIZE; |
644 | sp->seg_bytes_left -= LFS_SBPAD; | |
275ca4f0 | 645 | } |
a1b8db53 | 646 | brelse(bp); |
12304d41 KB |
647 | } else { |
648 | sp->seg_number = datosn(fs, fs->lfs_curseg); | |
649 | sp->seg_bytes_left = (fs->lfs_dbpseg - | |
650 | (fs->lfs_offset - fs->lfs_curseg)) * DEV_BSIZE; | |
651 | } | |
3ce71481 | 652 | fs->lfs_lastpseg = fs->lfs_offset; |
aa4dc149 | 653 | |
12304d41 KB |
654 | sp->ibp = NULL; |
655 | sp->ninodes = 0; | |
aa4dc149 | 656 | |
12304d41 KB |
657 | /* Get a new buffer for SEGSUM and enter it into the buffer list. */ |
658 | sp->cbpp = sp->bpp; | |
4c60cc5b | 659 | *sp->cbpp = lfs_newbuf(fs, fs->lfs_offset, LFS_SUMMARY_SIZE); |
12304d41 KB |
660 | sp->segsum = (*sp->cbpp)->b_un.b_addr; |
661 | ++sp->cbpp; | |
662 | fs->lfs_offset += LFS_SUMMARY_SIZE / DEV_BSIZE; | |
aa4dc149 | 663 | |
12304d41 KB |
664 | /* Set point to SEGSUM, initialize it. */ |
665 | ssp = sp->segsum; | |
666 | ssp->ss_next = fs->lfs_nextseg; | |
12304d41 | 667 | ssp->ss_nfinfo = ssp->ss_ninos = 0; |
aa4dc149 | 668 | |
12304d41 | 669 | /* Set pointer to first FINFO, initialize it. */ |
a1b8db53 | 670 | sp->fip = (struct finfo *)(sp->segsum + sizeof(SEGSUM)); |
12304d41 | 671 | sp->fip->fi_nblocks = 0; |
aa4dc149 | 672 | |
12304d41 KB |
673 | sp->seg_bytes_left -= LFS_SUMMARY_SIZE; |
674 | sp->sum_bytes_left = LFS_SUMMARY_SIZE - sizeof(SEGSUM); | |
675 | } | |
aa4dc149 | 676 | |
12304d41 KB |
677 | /* |
678 | * Return the next segment to write. | |
679 | */ | |
87804018 | 680 | void |
12304d41 KB |
681 | lfs_newseg(fs) |
682 | struct lfs *fs; | |
683 | { | |
c222b129 | 684 | CLEANERINFO *cip; |
12304d41 KB |
685 | SEGUSE *sup; |
686 | struct buf *bp; | |
c222b129 | 687 | int curseg, isdirty, sn; |
12304d41 | 688 | |
5b4e3ef5 KB |
689 | LFS_SEGENTRY(sup, fs, datosn(fs, fs->lfs_nextseg), bp); |
690 | sup->su_flags |= SEGUSE_DIRTY; | |
691 | LFS_UBWRITE(bp); | |
c222b129 KB |
692 | |
693 | LFS_CLEANERINFO(cip, fs, bp); | |
694 | --cip->clean; | |
695 | ++cip->dirty; | |
a1b8db53 | 696 | LFS_UBWRITE(bp); |
c222b129 KB |
697 | |
698 | fs->lfs_lastseg = fs->lfs_curseg; | |
699 | fs->lfs_curseg = fs->lfs_nextseg; | |
700 | for (sn = curseg = datosn(fs, fs->lfs_curseg);;) { | |
12304d41 | 701 | sn = (sn + 1) % fs->lfs_nseg; |
c222b129 | 702 | if (sn == curseg) |
12304d41 KB |
703 | panic("lfs_nextseg: no clean segments"); |
704 | LFS_SEGENTRY(sup, fs, sn, bp); | |
705 | isdirty = sup->su_flags & SEGUSE_DIRTY; | |
a1b8db53 | 706 | brelse(bp); |
12304d41 KB |
707 | if (!isdirty) |
708 | break; | |
709 | } | |
5b4e3ef5 | 710 | |
c222b129 | 711 | fs->lfs_nextseg = sntoda(fs, sn); |
84c30241 KB |
712 | } |
713 | ||
3ce71481 | 714 | int |
84c30241 | 715 | lfs_writeseg(fs, sp) |
0a011bb1 | 716 | struct lfs *fs; |
a1b8db53 | 717 | struct segment *sp; |
84c30241 | 718 | { |
4c60cc5b | 719 | struct buf **bpp, *bp, *cbp; |
84c30241 | 720 | SEGUSE *sup; |
a1b8db53 | 721 | SEGSUM *ssp; |
dc7e45d3 | 722 | dev_t i_dev; |
4c60cc5b | 723 | size_t size; |
3ce71481 KB |
724 | u_long *datap, *dp; |
725 | int ch_per_blk, do_again, i, nblocks, num, s; | |
726 | int (*strategy)__P((struct vop_strategy_args *)); | |
200cb75d | 727 | struct vop_strategy_args vop_strategy_a; |
5b4e3ef5 | 728 | u_short ninos; |
4c60cc5b | 729 | char *p; |
84c30241 | 730 | |
3ce71481 KB |
731 | /* Checkpoint always writes superblock, even if no data blocks. */ |
732 | if ((nblocks = sp->cbpp - sp->bpp) == 0 && !(sp->seg_flags & SEGM_CKP)) | |
290d4594 | 733 | return (0); |
a1b8db53 | 734 | |
84c30241 | 735 | /* |
a1b8db53 KB |
736 | * Compute checksum across data and then across summary; the first |
737 | * block (the summary block) is skipped. Set the create time here | |
738 | * so that it's guaranteed to be later than the inode mod times. | |
dc7e45d3 KB |
739 | * |
740 | * XXX | |
741 | * Fix this to do it inline, instead of malloc/copy. | |
84c30241 | 742 | */ |
dc7e45d3 | 743 | datap = dp = malloc(nblocks * sizeof(u_long), M_SEGMENT, M_WAITOK); |
12304d41 KB |
744 | for (bpp = sp->bpp, i = nblocks - 1; i--;) |
745 | *dp++ = (*++bpp)->b_un.b_words[0]; | |
a1b8db53 | 746 | ssp = (SEGSUM *)sp->segsum; |
89bed312 | 747 | ssp->ss_create = time.tv_sec; |
685d5160 | 748 | ssp->ss_datasum = cksum(datap, (nblocks - 1) * sizeof(u_long)); |
a1b8db53 KB |
749 | ssp->ss_sumsum = |
750 | cksum(&ssp->ss_datasum, LFS_SUMMARY_SIZE - sizeof(ssp->ss_sumsum)); | |
c222b129 | 751 | free(datap, M_SEGMENT); |
8954e52c | 752 | |
3ce71481 KB |
753 | /* Update the segment usage information. */ |
754 | LFS_SEGENTRY(sup, fs, sp->seg_number, bp); | |
5b4e3ef5 KB |
755 | ninos = (ssp->ss_ninos + INOPB(fs) - 1) / INOPB(fs); |
756 | sup->su_nbytes += nblocks - 1 - ninos << fs->lfs_bshift; | |
3ce71481 | 757 | sup->su_nbytes += ssp->ss_ninos * sizeof(struct dinode); |
685d5160 | 758 | sup->su_nbytes += LFS_SUMMARY_SIZE; |
3ce71481 | 759 | sup->su_lastmod = time.tv_sec; |
5b4e3ef5 KB |
760 | sup->su_flags |= SEGUSE_ACTIVE; |
761 | sup->su_ninos += ninos; | |
762 | ++sup->su_nsums; | |
3ce71481 | 763 | LFS_UBWRITE(bp); |
5b4e3ef5 | 764 | fs->lfs_bfree -= (fsbtodb(fs, ninos) + LFS_SUMMARY_SIZE / DEV_BSIZE); |
3ce71481 KB |
765 | do_again = !(bp->b_flags & B_GATHERED); |
766 | ||
dc7e45d3 | 767 | i_dev = VTOI(fs->lfs_ivnode)->i_dev; |
8cfb9f42 | 768 | strategy = VTOI(fs->lfs_ivnode)->i_devvp->v_op[VOFFSET(vop_strategy)]; |
275ca4f0 | 769 | |
4c60cc5b KB |
770 | /* |
771 | * When we simply write the blocks we lose a rotation for every block | |
772 | * written. To avoid this problem, we allocate memory in chunks, copy | |
773 | * the buffers into the chunk and write the chunk. 56K was chosen as | |
774 | * some driver/controllers can't handle unsigned 16 bit transfers. | |
775 | * When the data is copied to the chunk, turn off the the B_LOCKED bit | |
776 | * and brelse the buffer (which will move them to the LRU list). Add | |
777 | * the B_CALL flag to the buffer header so we can count I/O's for the | |
778 | * checkpoints and so we can release the allocated memory. | |
779 | * | |
780 | * XXX | |
781 | * This should be removed if the new virtual memory system allows us to | |
782 | * easily make the buffers contiguous in kernel memory and if that's | |
783 | * fast enough. | |
784 | */ | |
785 | #define LFS_CHUNKSIZE (56 * 1024) | |
786 | ch_per_blk = LFS_CHUNKSIZE / fs->lfs_bsize; | |
787 | for (bpp = sp->bpp, i = nblocks; i;) { | |
788 | num = ch_per_blk; | |
789 | if (num > i) | |
790 | num = i; | |
791 | i -= num; | |
792 | size = num * fs->lfs_bsize; | |
793 | ||
794 | cbp = lfs_newbuf(fs, (*bpp)->b_blkno, 0); | |
795 | cbp->b_dev = i_dev; | |
796 | cbp->b_flags = B_ASYNC | B_BUSY | B_CALL; | |
797 | cbp->b_iodone = lfs_callback; | |
798 | cbp->b_saveaddr = cbp->b_un.b_addr; | |
799 | cbp->b_un.b_addr = malloc(size, M_SEGMENT, M_WAITOK); | |
800 | ||
801 | s = splbio(); | |
802 | ++fs->lfs_iocount; | |
803 | for (p = cbp->b_un.b_addr; num--;) { | |
804 | bp = *bpp++; | |
805 | bcopy(bp->b_un.b_addr, p, bp->b_bcount); | |
806 | p += bp->b_bcount; | |
3ce71481 KB |
807 | bp->b_flags &= ~(B_DONE | B_ERROR | B_READ | B_DELWRI | |
808 | B_LOCKED | B_GATHERED); | |
809 | if (!(bp->b_flags & (B_NOCACHE | B_INVAL))) { | |
4c60cc5b KB |
810 | bremfree(bp); |
811 | reassignbuf(bp, bp->b_vp); | |
812 | } | |
813 | brelse(bp); | |
dc7e45d3 | 814 | } |
4c60cc5b KB |
815 | splx(s); |
816 | cbp->b_bcount = p - cbp->b_un.b_addr; | |
8cfb9f42 JH |
817 | vop_strategy_a.a_desc = VDESC(vop_strategy); |
818 | vop_strategy_a.a_bp = cbp; | |
819 | (strategy)(&vop_strategy_a); | |
8954e52c | 820 | } |
290d4594 | 821 | return (do_again); |
275ca4f0 KB |
822 | } |
823 | ||
87804018 | 824 | void |
dc7e45d3 | 825 | lfs_writesuper(fs, sp) |
0a011bb1 | 826 | struct lfs *fs; |
a1b8db53 | 827 | struct segment *sp; |
275ca4f0 | 828 | { |
a1b8db53 | 829 | struct buf *bp; |
dc7e45d3 | 830 | dev_t i_dev; |
8cfb9f42 | 831 | int (*strategy) __P((struct vop_strategy_args *)); |
200cb75d | 832 | struct vop_strategy_args vop_strategy_a; |
275ca4f0 | 833 | |
dc7e45d3 | 834 | i_dev = VTOI(fs->lfs_ivnode)->i_dev; |
8cfb9f42 | 835 | strategy = VTOI(fs->lfs_ivnode)->i_devvp->v_op[VOFFSET(vop_strategy)]; |
14712628 | 836 | |
aa4dc149 | 837 | /* Checksum the superblock and copy it into a buffer. */ |
0a011bb1 | 838 | fs->lfs_cksum = cksum(fs, sizeof(struct lfs) - sizeof(fs->lfs_cksum)); |
4c60cc5b | 839 | bp = lfs_newbuf(fs, fs->lfs_sboffs[0], LFS_SBPAD); |
dc7e45d3 | 840 | *bp->b_un.b_lfs = *fs; |
275ca4f0 | 841 | |
14712628 | 842 | /* Write the first superblock (wait). */ |
dc7e45d3 | 843 | bp->b_dev = i_dev; |
dc7e45d3 | 844 | bp->b_flags |= B_BUSY; |
12304d41 | 845 | bp->b_flags &= ~(B_DONE | B_ERROR | B_READ | B_DELWRI); |
8cfb9f42 JH |
846 | vop_strategy_a.a_desc = VDESC(vop_strategy); |
847 | vop_strategy_a.a_bp = bp; | |
848 | (strategy)(&vop_strategy_a); | |
275ca4f0 | 849 | biowait(bp); |
aa4dc149 | 850 | |
14712628 | 851 | /* Write the second superblock (don't wait). */ |
275ca4f0 | 852 | bp->b_blkno = bp->b_lblkno = fs->lfs_sboffs[1]; |
12304d41 | 853 | bp->b_flags |= B_ASYNC | B_BUSY; |
dc7e45d3 | 854 | bp->b_flags &= ~(B_DONE | B_ERROR | B_READ | B_DELWRI); |
8cfb9f42 | 855 | (strategy)(&vop_strategy_a); |
275ca4f0 KB |
856 | } |
857 | ||
aa4dc149 KB |
858 | /* |
859 | * Logical block number match routines used when traversing the dirty block | |
860 | * chain. | |
861 | */ | |
87804018 KB |
862 | int |
863 | lfs_match_data(fs, bp) | |
dc7e45d3 | 864 | struct lfs *fs; |
a1b8db53 | 865 | struct buf *bp; |
275ca4f0 | 866 | { |
aa4dc149 | 867 | return (bp->b_lblkno >= 0); |
275ca4f0 KB |
868 | } |
869 | ||
87804018 KB |
870 | int |
871 | lfs_match_indir(fs, bp) | |
dc7e45d3 | 872 | struct lfs *fs; |
a1b8db53 | 873 | struct buf *bp; |
275ca4f0 | 874 | { |
dc7e45d3 KB |
875 | int lbn; |
876 | ||
877 | lbn = bp->b_lblkno; | |
878 | return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 0); | |
275ca4f0 KB |
879 | } |
880 | ||
87804018 KB |
881 | int |
882 | lfs_match_dindir(fs, bp) | |
dc7e45d3 | 883 | struct lfs *fs; |
a1b8db53 | 884 | struct buf *bp; |
275ca4f0 | 885 | { |
dc7e45d3 KB |
886 | int lbn; |
887 | ||
888 | lbn = bp->b_lblkno; | |
889 | return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 1); | |
aa4dc149 KB |
890 | } |
891 | ||
87804018 KB |
892 | int |
893 | lfs_match_tindir(fs, bp) | |
0a011bb1 | 894 | struct lfs *fs; |
a1b8db53 | 895 | struct buf *bp; |
aa4dc149 | 896 | { |
dc7e45d3 | 897 | int lbn; |
aa4dc149 | 898 | |
dc7e45d3 KB |
899 | lbn = bp->b_lblkno; |
900 | return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 2); | |
901 | } | |
aa4dc149 | 902 | |
dc7e45d3 KB |
903 | /* |
904 | * Allocate a new buffer header. | |
905 | */ | |
a1b8db53 | 906 | struct buf * |
4c60cc5b | 907 | lfs_newbuf(fs, daddr, size) |
dc7e45d3 | 908 | struct lfs *fs; |
dc7e45d3 KB |
909 | daddr_t daddr; |
910 | size_t size; | |
911 | { | |
a1b8db53 | 912 | struct buf *bp; |
aa4dc149 | 913 | |
dc7e45d3 KB |
914 | bp = getnewbuf(); |
915 | bremhash(bp); | |
916 | bgetvp(fs->lfs_ivnode, bp); | |
917 | bp->b_bcount = 0; | |
918 | bp->b_lblkno = daddr; | |
919 | bp->b_blkno = daddr; | |
920 | bp->b_error = 0; | |
921 | bp->b_resid = 0; | |
4c60cc5b KB |
922 | if (size) |
923 | allocbuf(bp, size); | |
dc7e45d3 | 924 | bp->b_flags |= B_NOCACHE; |
4c60cc5b | 925 | bp->b_saveaddr = NULL; |
12304d41 | 926 | binshash(bp, &bfreelist[BQ_AGE]); |
dc7e45d3 KB |
927 | return (bp); |
928 | } | |
aa4dc149 | 929 | |
80443139 | 930 | void |
dc7e45d3 | 931 | lfs_callback(bp) |
a1b8db53 | 932 | struct buf *bp; |
dc7e45d3 KB |
933 | { |
934 | struct lfs *fs; | |
aa4dc149 | 935 | |
dc7e45d3 KB |
936 | fs = VFSTOUFS(bp->b_vp->v_mount)->um_lfs; |
937 | #ifdef DIAGNOSTIC | |
938 | if (fs->lfs_iocount == 0) | |
939 | panic("lfs_callback: zero iocount\n"); | |
940 | #endif | |
941 | if (--fs->lfs_iocount == 0) | |
4c60cc5b | 942 | wakeup(&fs->lfs_iocount); |
12304d41 | 943 | |
4c60cc5b KB |
944 | if (bp->b_saveaddr) { |
945 | free(bp->b_un.b_addr, M_SEGMENT); | |
946 | bp->b_un.b_addr = bp->b_saveaddr; | |
2b3ba73a | 947 | bp->b_saveaddr = NULL; |
4c60cc5b | 948 | } |
dc7e45d3 | 949 | brelse(bp); |
84c30241 KB |
950 | } |
951 | ||
952 | /* | |
953 | * Shellsort (diminishing increment sort) from Data Structures and | |
954 | * Algorithms, Aho, Hopcraft and Ullman, 1983 Edition, page 290; | |
955 | * see also Knuth Vol. 3, page 84. The increments are selected from | |
956 | * formula (8), page 95. Roughly O(N^3/2). | |
957 | */ | |
958 | /* | |
959 | * This is our own private copy of shellsort because we want to sort | |
960 | * two parallel arrays (the array of buffer pointers and the array of | |
961 | * logical block numbers) simultaneously. Note that we cast the array | |
962 | * of logical block numbers to a unsigned in this routine so that the | |
963 | * negative block numbers (meta data blocks) sort AFTER the data blocks. | |
964 | */ | |
87804018 KB |
965 | void |
966 | lfs_shellsort(bp_array, lb_array, nmemb) | |
a1b8db53 | 967 | struct buf **bp_array; |
275ca4f0 | 968 | daddr_t *lb_array; |
84c30241 KB |
969 | register int nmemb; |
970 | { | |
971 | static int __rsshell_increments[] = { 4, 1, 0 }; | |
972 | register int incr, *incrp, t1, t2; | |
a1b8db53 | 973 | struct buf *bp_temp; |
84c30241 KB |
974 | u_long lb_temp; |
975 | ||
976 | for (incrp = __rsshell_increments; incr = *incrp++;) | |
977 | for (t1 = incr; t1 < nmemb; ++t1) | |
978 | for (t2 = t1 - incr; t2 >= 0;) | |
979 | if (lb_array[t2] > lb_array[t2 + incr]) { | |
980 | lb_temp = lb_array[t2]; | |
981 | lb_array[t2] = lb_array[t2 + incr]; | |
982 | lb_array[t2 + incr] = lb_temp; | |
983 | bp_temp = bp_array[t2]; | |
984 | bp_array[t2] = bp_array[t2 + incr]; | |
985 | bp_array[t2 + incr] = bp_temp; | |
986 | t2 -= incr; | |
987 | } else | |
988 | break; | |
989 | } |