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 | * | |
9a46ddb2 | 7 | * @(#)lfs_segment.c 7.12 (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> | |
21 | #include <sys/specdev.h> | |
22 | #include <sys/fifo.h> | |
23 | #include <sys/malloc.h> | |
24 | #include <sys/mount.h> | |
25 | ||
0a011bb1 KB |
26 | #include <ufs/ufs/quota.h> |
27 | #include <ufs/ufs/inode.h> | |
28 | #include <ufs/ufs/dir.h> | |
29 | #include <ufs/ufs/ufsmount.h> | |
34a084a9 | 30 | |
0a011bb1 KB |
31 | #include <ufs/lfs/lfs.h> |
32 | #include <ufs/lfs/lfs_extern.h> | |
84c30241 | 33 | |
dc7e45d3 | 34 | /* In-memory description of a segment about to be written. */ |
dc7e45d3 | 35 | struct segment { |
a1b8db53 KB |
36 | struct buf **bpp; /* pointer to buffer array */ |
37 | struct buf **cbpp; /* pointer to next available bp */ | |
38 | struct buf *ibp; /* buffer pointer to inode page */ | |
39 | struct finfo *fip; /* current fileinfo pointer */ | |
dc7e45d3 KB |
40 | void *segsum; /* segment summary info */ |
41 | u_long ninodes; /* number of inodes in this segment */ | |
42 | u_long seg_bytes_left; /* bytes left in segment */ | |
43 | u_long sum_bytes_left; /* bytes left in summary block */ | |
44 | u_long seg_number; /* number of this segment */ | |
45 | #define SEGM_CKP 0x01 /* doing a checkpoint */ | |
46 | u_long seg_flags; /* run-time flags for this segment */ | |
dc7e45d3 KB |
47 | }; |
48 | ||
84c30241 | 49 | /* |
dc7e45d3 KB |
50 | * Determine if it's OK to start a partial in this segment, or if we need |
51 | * to go on to a new segment. | |
8954e52c | 52 | */ |
dc7e45d3 KB |
53 | #define LFS_PARTIAL_FITS(fs) \ |
54 | ((fs)->lfs_dbpseg - ((fs)->lfs_offset - (fs)->lfs_curseg) > \ | |
55 | 1 << (fs)->lfs_fsbtodb) | |
56 | ||
a1b8db53 KB |
57 | int lfs_callback __P((struct buf *)); |
58 | void lfs_gather __P((struct lfs *, struct segment *, | |
59 | struct vnode *, int (*) __P((struct lfs *, struct buf *)))); | |
60 | void lfs_initseg __P((struct lfs *, struct segment *)); | |
61 | void lfs_iset __P((struct inode *, daddr_t, time_t)); | |
62 | int lfs_match_data __P((struct lfs *, struct buf *)); | |
63 | int lfs_match_dindir __P((struct lfs *, struct buf *)); | |
64 | int lfs_match_indir __P((struct lfs *, struct buf *)); | |
65 | int lfs_match_tindir __P((struct lfs *, struct buf *)); | |
66 | struct buf * | |
67 | lfs_newbuf __P((struct lfs *, struct segment *, daddr_t, size_t)); | |
87804018 | 68 | void lfs_newseg __P((struct lfs *)); |
a1b8db53 | 69 | void lfs_shellsort __P((struct buf **, daddr_t *, register int)); |
87804018 | 70 | void lfs_updatemeta __P((struct lfs *, |
a1b8db53 KB |
71 | struct segment *, struct vnode *, daddr_t *, struct buf **, int)); |
72 | void lfs_writefile __P((struct lfs *, struct segment *, struct vnode *)); | |
73 | void lfs_writeinode __P((struct lfs *, struct segment *, struct inode *)); | |
74 | void lfs_writeseg __P((struct lfs *, struct segment *)); | |
75 | void lfs_writesuper __P((struct lfs *, struct segment *)); | |
84c30241 | 76 | |
dc7e45d3 KB |
77 | int lfs_allclean_wakeup; /* Cleaner wakeup address. */ |
78 | ||
44819c56 KB |
79 | /* |
80 | * Ifile and meta data blocks are not marked busy, so segment writes MUST be | |
81 | * single threaded. Currently, there are two paths into lfs_segwrite, sync() | |
82 | * and getnewbuf(). They both mark the file system busy. Lfs_vflush() | |
83 | * explicitly marks the file system busy. So lfs_segwrite is safe. I think. | |
84 | */ | |
85 | ||
86 | int | |
87 | lfs_vflush(vp) | |
88 | struct vnode *vp; | |
89 | { | |
90 | struct inode *ip; | |
91 | struct lfs *fs; | |
92 | struct mount *mp; | |
93 | struct segment *sp; | |
94 | int error, s; | |
95 | ||
96 | #ifdef VERBOSE | |
97 | printf("lfs_vflush\n"); | |
98 | #endif | |
99 | mp = vp->v_mount; | |
100 | fs = VFSTOUFS(mp)->um_lfs; | |
101 | ||
102 | /* | |
103 | * XXX | |
104 | * check flags? | |
105 | * mp->mnt_flag & (MNT_MLOCK|MNT_RDONLY|MNT_MPBUSY) || | |
106 | */ | |
107 | if (vfs_busy(mp)) | |
108 | return (0); | |
109 | ||
110 | /* | |
111 | * Allocate a segment structure and enough space to hold pointers to | |
112 | * the maximum possible number of buffers which can be described in a | |
113 | * single summary block. | |
114 | */ | |
115 | sp = malloc(sizeof(struct segment), M_SEGMENT, M_WAITOK); | |
116 | sp->bpp = malloc(((LFS_SUMMARY_SIZE - sizeof(SEGSUM)) / | |
117 | sizeof(daddr_t) + 1) * sizeof(struct buf *), M_SEGMENT, M_WAITOK); | |
118 | sp->seg_flags = SEGM_CKP; | |
119 | lfs_initseg(fs, sp); | |
120 | ||
121 | /* | |
122 | * Keep a cumulative count of the outstanding I/O operations. If the | |
123 | * disk drive catches up with us it could go to zero before we finish, | |
124 | * so we artificially increment it by one until we've scheduled all of | |
125 | * the writes we intend to do. | |
126 | */ | |
127 | s = splbio(); | |
128 | fs->lfs_iocount = 1; | |
129 | splx(s); | |
130 | ||
131 | if (vp->v_dirtyblkhd != NULL) | |
132 | lfs_writefile(fs, sp, vp); | |
133 | ip = VTOI(vp); | |
134 | lfs_writeinode(fs, sp, ip); | |
135 | ip->i_flags &= ~(IMOD | IACC | IUPD | ICHG); | |
136 | ||
137 | lfs_writeseg(fs, sp); | |
138 | ||
139 | /* | |
140 | * If the I/O count is non-zero, sleep until it reaches zero. At the | |
141 | * moment, the user's process hangs around so we can sleep. | |
142 | */ | |
143 | s = splbio(); | |
144 | if (--fs->lfs_iocount && (error = | |
b68b931c | 145 | tsleep((caddr_t)&fs->lfs_iocount, PRIBIO + 1, "lfs vflush", 0))) |
44819c56 KB |
146 | return (error); |
147 | splx(s); | |
148 | vfs_unbusy(mp); | |
149 | ||
150 | free(sp->bpp, M_SEGMENT); | |
151 | free(sp, M_SEGMENT); | |
152 | ||
153 | return (0); | |
154 | } | |
155 | ||
84c30241 | 156 | int |
275ca4f0 | 157 | lfs_segwrite(mp, do_ckp) |
a1b8db53 | 158 | struct mount *mp; |
dc7e45d3 | 159 | int do_ckp; /* Do a checkpoint. */ |
84c30241 | 160 | { |
a1b8db53 | 161 | struct inode *ip; |
0a011bb1 | 162 | struct lfs *fs; |
a1b8db53 KB |
163 | struct segment *sp; |
164 | struct vnode *vp; | |
44819c56 | 165 | int error, islocked, s; |
84c30241 | 166 | |
dc7e45d3 KB |
167 | #ifdef VERBOSE |
168 | printf("lfs_segwrite\n"); | |
aa4dc149 | 169 | #endif |
44819c56 KB |
170 | fs = VFSTOUFS(mp)->um_lfs; |
171 | ||
172 | /* | |
173 | * Allocate a segment structure and enough space to hold pointers to | |
174 | * the maximum possible number of buffers which can be described in a | |
175 | * single summary block. | |
176 | */ | |
177 | sp = malloc(sizeof(struct segment), M_SEGMENT, M_WAITOK); | |
178 | sp->bpp = malloc(((LFS_SUMMARY_SIZE - sizeof(SEGSUM)) / | |
179 | sizeof(daddr_t) + 1) * sizeof(struct buf *), M_SEGMENT, M_WAITOK); | |
180 | sp->seg_flags = do_ckp ? SEGM_CKP : 0; | |
181 | lfs_initseg(fs, sp); | |
a1b8db53 | 182 | |
8954e52c | 183 | /* |
dc7e45d3 KB |
184 | * If doing a checkpoint, we keep a cumulative count of the outstanding |
185 | * I/O operations. If the disk drive catches up with us it could go to | |
186 | * zero before we finish, so we artificially increment it by one until | |
187 | * we've scheduled all of the writes we intend to do. | |
8954e52c | 188 | */ |
dc7e45d3 KB |
189 | if (do_ckp) { |
190 | s = splbio(); | |
191 | fs->lfs_iocount = 1; | |
192 | splx(s); | |
193 | } | |
aa4dc149 | 194 | |
44819c56 | 195 | loop: for (vp = mp->mnt_mounth; vp; vp = vp->v_mountf) { |
84c30241 KB |
196 | /* |
197 | * If the vnode that we are about to sync is no longer | |
198 | * associated with this mount point, start over. | |
199 | */ | |
84c30241 KB |
200 | if (vp->v_mount != mp) |
201 | goto loop; | |
dc7e45d3 | 202 | |
44819c56 | 203 | islocked = VOP_ISLOCKED(vp); |
dc7e45d3 | 204 | |
87804018 KB |
205 | /* |
206 | * XXX | |
207 | * This is wrong, I think -- we should just wait until we | |
208 | * get the vnode and go on. Probably going to reschedule | |
209 | * all of the writes we already scheduled... | |
210 | */ | |
44819c56 KB |
211 | if (islocked) |
212 | VREF(vp); | |
213 | else if (vget(vp)) | |
a1b8db53 KB |
214 | { |
215 | printf("lfs_segment: failed to get vnode (tell Keith)!\n"); | |
84c30241 | 216 | goto loop; |
a1b8db53 | 217 | } |
44819c56 KB |
218 | /* |
219 | * Write the inode/file if dirty and it's not the | |
220 | * the IFILE. | |
221 | */ | |
222 | ip = VTOI(vp); | |
223 | if ((ip->i_flag & (IMOD | IACC | IUPD | ICHG) || | |
224 | vp->v_dirtyblkhd != NULL) && | |
225 | ip->i_number != LFS_IFILE_INUM) { | |
226 | if (vp->v_dirtyblkhd != NULL) | |
227 | lfs_writefile(fs, sp, vp); | |
228 | lfs_writeinode(fs, sp, ip); | |
229 | ip->i_flags &= ~(IMOD | IACC | IUPD | ICHG); | |
230 | } | |
231 | if (islocked) | |
232 | vrele(vp); | |
233 | else | |
234 | vput(vp); | |
84c30241 | 235 | } |
dc7e45d3 | 236 | if (do_ckp) { |
87804018 KB |
237 | vp = fs->lfs_ivnode; |
238 | while (vget(vp)); | |
239 | ip = VTOI(vp); | |
240 | if (vp->v_dirtyblkhd != NULL) | |
241 | lfs_writefile(fs, sp, vp); | |
242 | lfs_writeinode(fs, sp, ip); | |
243 | ip->i_flags &= ~(IMOD | IACC | IUPD | ICHG); | |
244 | vput(vp); | |
dc7e45d3 | 245 | } |
aa4dc149 KB |
246 | lfs_writeseg(fs, sp); |
247 | ||
275ca4f0 | 248 | /* |
dc7e45d3 KB |
249 | * If the I/O count is non-zero, sleep until it reaches zero. At the |
250 | * moment, the user's process hangs around so we can sleep. | |
275ca4f0 | 251 | */ |
dc7e45d3 KB |
252 | if (do_ckp) { |
253 | s = splbio(); | |
b68b931c KM |
254 | if (--fs->lfs_iocount && |
255 | (error = tsleep((caddr_t)&fs->lfs_iocount, PRIBIO + 1, | |
256 | "lfs sync", 0))) | |
12304d41 | 257 | return (error); |
dc7e45d3 KB |
258 | splx(s); |
259 | lfs_writesuper(fs, sp); | |
260 | } | |
275ca4f0 | 261 | |
c222b129 KB |
262 | free(sp->bpp, M_SEGMENT); |
263 | free(sp, M_SEGMENT); | |
275ca4f0 | 264 | |
dc7e45d3 | 265 | return (0); |
84c30241 KB |
266 | } |
267 | ||
dc7e45d3 KB |
268 | /* |
269 | * Write the dirty blocks associated with a vnode. | |
270 | */ | |
87804018 | 271 | void |
dc7e45d3 | 272 | lfs_writefile(fs, sp, vp) |
0a011bb1 | 273 | struct lfs *fs; |
a1b8db53 KB |
274 | struct segment *sp; |
275 | struct vnode *vp; | |
84c30241 | 276 | { |
dc7e45d3 | 277 | struct buf *bp; |
a1b8db53 | 278 | struct finfo *fip; |
dc7e45d3 | 279 | IFILE *ifp; |
275ca4f0 | 280 | |
dc7e45d3 KB |
281 | #ifdef VERBOSE |
282 | printf("lfs_writefile\n"); | |
283 | #endif | |
a1b8db53 KB |
284 | if (sp->seg_bytes_left < fs->lfs_bsize || |
285 | sp->sum_bytes_left < sizeof(struct finfo)) { | |
286 | lfs_writeseg(fs, sp); | |
287 | lfs_initseg(fs, sp); | |
288 | } | |
289 | sp->sum_bytes_left -= sizeof(struct finfo) - sizeof(daddr_t); | |
84c30241 | 290 | |
a1b8db53 KB |
291 | fip = sp->fip; |
292 | fip->fi_nblocks = 0; | |
293 | fip->fi_ino = VTOI(vp)->i_number; | |
294 | LFS_IENTRY(ifp, fs, fip->fi_ino, bp); | |
295 | fip->fi_version = ifp->if_version; | |
296 | brelse(bp); | |
297 | ||
298 | /* | |
299 | * It may not be necessary to write the meta-data blocks at this point, | |
300 | * as the roll-forward recovery code should be able to reconstruct the | |
301 | * list. | |
302 | */ | |
303 | lfs_gather(fs, sp, vp, lfs_match_data); | |
304 | lfs_gather(fs, sp, vp, lfs_match_indir); | |
305 | lfs_gather(fs, sp, vp, lfs_match_dindir); | |
dc7e45d3 | 306 | #ifdef TRIPLE |
a1b8db53 | 307 | lfs_gather(fs, sp, vp, lfs_match_tindir); |
dc7e45d3 | 308 | #endif |
aa4dc149 | 309 | |
a1b8db53 | 310 | fip = sp->fip; |
dc7e45d3 | 311 | #ifdef META |
a1b8db53 | 312 | printf("lfs_writefile: adding %d blocks\n", fip->fi_nblocks); |
dc7e45d3 | 313 | #endif |
a1b8db53 KB |
314 | if (fip->fi_nblocks != 0) { |
315 | ++((SEGSUM *)(sp->segsum))->ss_nfinfo; | |
316 | sp->fip = | |
317 | (struct finfo *)((caddr_t)fip + sizeof(struct finfo) + | |
318 | sizeof(daddr_t) * (fip->fi_nblocks - 1)); | |
9a46ddb2 CS |
319 | } else |
320 | sp->sum_bytes_left += sizeof(struct finfo) - sizeof(daddr_t); | |
12304d41 KB |
321 | } |
322 | ||
87804018 | 323 | void |
12304d41 KB |
324 | lfs_writeinode(fs, sp, ip) |
325 | struct lfs *fs; | |
a1b8db53 KB |
326 | struct segment *sp; |
327 | struct inode *ip; | |
12304d41 | 328 | { |
a1b8db53 | 329 | struct buf *bp, *ibp; |
87804018 | 330 | IFILE *ifp; |
9a46ddb2 CS |
331 | SEGUSE *sup; |
332 | daddr_t daddr; | |
87804018 | 333 | ino_t ino; |
12304d41 KB |
334 | int ndx; |
335 | ||
336 | #ifdef VERBOSE | |
337 | printf("lfs_writeinode\n"); | |
338 | #endif | |
339 | /* Allocate a new inode block if necessary. */ | |
340 | if (sp->ibp == NULL) { | |
341 | /* Allocate a new segment if necessary. */ | |
342 | if (sp->seg_bytes_left < fs->lfs_bsize || | |
343 | sp->sum_bytes_left < sizeof(daddr_t)) { | |
344 | lfs_writeseg(fs, sp); | |
345 | lfs_initseg(fs, sp); | |
346 | } | |
347 | ||
348 | /* Get next inode block. */ | |
9a46ddb2 | 349 | daddr = fs->lfs_offset; |
12304d41 KB |
350 | fs->lfs_offset += fsbtodb(fs, 1); |
351 | sp->ibp = *sp->cbpp++ = | |
9a46ddb2 | 352 | lfs_newbuf(fs, sp, daddr, fs->lfs_bsize); |
12304d41 KB |
353 | |
354 | /* Set remaining space counter. */ | |
355 | sp->seg_bytes_left -= fs->lfs_bsize; | |
356 | sp->sum_bytes_left -= sizeof(daddr_t); | |
87804018 | 357 | ndx = LFS_SUMMARY_SIZE / sizeof(daddr_t) - |
12304d41 | 358 | sp->ninodes / INOPB(fs) - 1; |
9a46ddb2 | 359 | ((daddr_t *)(sp->segsum))[ndx] = daddr; |
12304d41 KB |
360 | } |
361 | ||
a1b8db53 | 362 | /* Update the inode times and copy the inode onto the inode page. */ |
87804018 | 363 | ITIMES(ip, &time, &time); |
12304d41 | 364 | bp = sp->ibp; |
a1b8db53 | 365 | bp->b_un.b_dino[sp->ninodes % INOPB(fs)] = ip->i_din; |
12304d41 KB |
366 | |
367 | /* Increment inode count in segment summary block. */ | |
368 | ++((SEGSUM *)(sp->segsum))->ss_ninos; | |
369 | ||
370 | /* If this page is full, set flag to allocate a new page. */ | |
371 | if (++sp->ninodes % INOPB(fs) == 0) | |
372 | sp->ibp = NULL; | |
373 | ||
374 | /* | |
87804018 KB |
375 | * If updating the ifile, update the super-block. Update the disk |
376 | * address and access times for this inode in the ifile. | |
12304d41 | 377 | */ |
87804018 KB |
378 | ino = ip->i_number; |
379 | if (ino == LFS_IFILE_INUM) | |
12304d41 | 380 | fs->lfs_idaddr = bp->b_blkno; |
87804018 KB |
381 | |
382 | LFS_IENTRY(ifp, fs, ino, ibp); | |
9a46ddb2 | 383 | daddr = ifp->if_daddr; |
87804018 | 384 | ifp->if_daddr = bp->b_blkno; |
a1b8db53 | 385 | LFS_UBWRITE(ibp); |
9a46ddb2 CS |
386 | |
387 | if (daddr != LFS_UNUSED_DADDR) { | |
388 | LFS_SEGENTRY(sup, fs, datosn(fs, daddr), bp); | |
389 | #ifdef DIAGNOSTIC | |
390 | if (sup->su_nbytes < sizeof(struct dinode)) | |
391 | panic("lfs: negative bytes (segment %d)\n", | |
392 | datosn(fs, daddr)); | |
393 | #endif | |
394 | sup->su_nbytes -= sizeof(struct dinode); | |
395 | LFS_UBWRITE(bp); | |
396 | } | |
275ca4f0 KB |
397 | } |
398 | ||
87804018 | 399 | void |
275ca4f0 | 400 | lfs_gather(fs, sp, vp, match) |
0a011bb1 | 401 | struct lfs *fs; |
a1b8db53 KB |
402 | struct segment *sp; |
403 | struct vnode *vp; | |
404 | int (*match) __P((struct lfs *, struct buf *)); | |
275ca4f0 | 405 | { |
a1b8db53 KB |
406 | struct buf **bpp, *bp, *nbp; |
407 | struct finfo *fip; | |
408 | struct inode *ip; | |
275ca4f0 | 409 | daddr_t *lbp, *start_lbp; |
aa4dc149 KB |
410 | u_long version; |
411 | int s; | |
275ca4f0 | 412 | |
dc7e45d3 KB |
413 | #ifdef VERBOSE |
414 | printf("lfs_gather\n"); | |
415 | #endif | |
275ca4f0 KB |
416 | ip = VTOI(vp); |
417 | bpp = sp->cbpp; | |
418 | fip = sp->fip; | |
275ca4f0 | 419 | start_lbp = lbp = &fip->fi_blocks[fip->fi_nblocks]; |
275ca4f0 KB |
420 | |
421 | s = splbio(); | |
422 | for (bp = vp->v_dirtyblkhd; bp; bp = nbp) { | |
423 | nbp = bp->b_blockf; | |
12304d41 KB |
424 | /* |
425 | * XXX | |
9a46ddb2 | 426 | * Should sleep on any BUSY buffer if doing an fsync? |
12304d41 | 427 | */ |
44819c56 | 428 | if (bp->b_flags & B_BUSY || !match(fs, bp)) |
275ca4f0 | 429 | continue; |
aa4dc149 | 430 | #ifdef DIAGNOSTIC |
dc7e45d3 | 431 | if (!(bp->b_flags & B_DELWRI)) |
12304d41 | 432 | panic("lfs_gather: bp not B_DELWRI"); |
dc7e45d3 | 433 | if (!(bp->b_flags & B_LOCKED)) |
12304d41 | 434 | panic("lfs_gather: bp not B_LOCKED"); |
aa4dc149 | 435 | #endif |
aa4dc149 | 436 | /* |
dc7e45d3 KB |
437 | * If full, finish this segment. We may be doing I/O, so |
438 | * release and reacquire the splbio(). | |
aa4dc149 KB |
439 | */ |
440 | if (sp->sum_bytes_left < sizeof(daddr_t) || | |
275ca4f0 | 441 | sp->seg_bytes_left < fs->lfs_bsize) { |
275ca4f0 | 442 | splx(s); |
aa4dc149 | 443 | lfs_updatemeta(fs, |
dc7e45d3 | 444 | sp, vp, start_lbp, bpp, lbp - start_lbp); |
275ca4f0 | 445 | |
aa4dc149 KB |
446 | /* Add the current file to the segment summary. */ |
447 | ++((SEGSUM *)(sp->segsum))->ss_nfinfo; | |
275ca4f0 | 448 | |
aa4dc149 | 449 | version = fip->fi_version; |
dc7e45d3 | 450 | lfs_writeseg(fs, sp); |
12304d41 | 451 | lfs_initseg(fs, sp); |
aa4dc149 | 452 | |
275ca4f0 | 453 | fip = sp->fip; |
275ca4f0 | 454 | fip->fi_version = version; |
aa4dc149 | 455 | fip->fi_ino = ip->i_number; |
275ca4f0 | 456 | start_lbp = lbp = fip->fi_blocks; |
aa4dc149 | 457 | |
9a46ddb2 CS |
458 | sp->sum_bytes_left -= |
459 | sizeof(struct finfo) - sizeof(daddr_t); | |
460 | ||
aa4dc149 | 461 | bpp = sp->cbpp; |
275ca4f0 KB |
462 | s = splbio(); |
463 | } | |
9a46ddb2 CS |
464 | |
465 | /* Insert into the buffer list, update the FINFO block. */ | |
466 | *sp->cbpp++ = bp; | |
467 | ++fip->fi_nblocks; | |
468 | *lbp++ = bp->b_lblkno; | |
469 | ||
470 | sp->sum_bytes_left -= sizeof(daddr_t); | |
471 | sp->seg_bytes_left -= bp->b_bufsize; | |
84c30241 | 472 | } |
275ca4f0 | 473 | splx(s); |
dc7e45d3 | 474 | lfs_updatemeta(fs, sp, vp, start_lbp, bpp, lbp - start_lbp); |
84c30241 KB |
475 | } |
476 | ||
84c30241 | 477 | /* |
aa4dc149 | 478 | * Update the metadata that points to the blocks listed in the FINFO |
84c30241 KB |
479 | * array. |
480 | */ | |
87804018 | 481 | void |
dc7e45d3 | 482 | lfs_updatemeta(fs, sp, vp, lbp, bpp, nblocks) |
0a011bb1 | 483 | struct lfs *fs; |
a1b8db53 KB |
484 | struct segment *sp; |
485 | struct vnode *vp; | |
275ca4f0 | 486 | daddr_t *lbp; |
a1b8db53 | 487 | struct buf **bpp; |
275ca4f0 | 488 | int nblocks; |
84c30241 | 489 | { |
12304d41 | 490 | SEGUSE *sup; |
a1b8db53 | 491 | struct buf *bp; |
dc7e45d3 | 492 | INDIR a[NIADDR], *ap; |
a1b8db53 | 493 | struct inode *ip; |
12304d41 | 494 | daddr_t daddr, lbn, off; |
dc7e45d3 | 495 | int db_per_fsb, error, i, num; |
84c30241 | 496 | |
dc7e45d3 KB |
497 | #ifdef VERBOSE |
498 | printf("lfs_updatemeta\n"); | |
499 | #endif | |
aa4dc149 | 500 | if (nblocks == 0) |
275ca4f0 KB |
501 | return; |
502 | ||
12304d41 | 503 | /* Sort the blocks. */ |
87804018 | 504 | lfs_shellsort(bpp, lbp, nblocks); |
275ca4f0 | 505 | |
12304d41 KB |
506 | /* |
507 | * Assign disk addresses, and update references to the logical | |
508 | * block and the segment usage information. | |
509 | */ | |
dc7e45d3 | 510 | db_per_fsb = fsbtodb(fs, 1); |
12304d41 KB |
511 | for (i = nblocks; i--; ++bpp) { |
512 | lbn = *lbp++; | |
513 | (*bpp)->b_blkno = off = fs->lfs_offset; | |
dc7e45d3 | 514 | fs->lfs_offset += db_per_fsb; |
275ca4f0 | 515 | |
dc7e45d3 | 516 | if (error = lfs_bmaparray(vp, lbn, &daddr, a, &num)) |
a1b8db53 | 517 | panic("lfs_updatemeta: lfs_bmaparray %d", error); |
dc7e45d3 KB |
518 | ip = VTOI(vp); |
519 | switch (num) { | |
520 | case 0: | |
12304d41 | 521 | ip->i_db[lbn] = off; |
dc7e45d3 KB |
522 | break; |
523 | case 1: | |
12304d41 | 524 | ip->i_ib[a[0].in_off] = off; |
dc7e45d3 KB |
525 | break; |
526 | default: | |
527 | ap = &a[num - 1]; | |
dc7e45d3 KB |
528 | if (bread(vp, ap->in_lbn, fs->lfs_bsize, NOCRED, &bp)) |
529 | panic("lfs_updatemeta: bread bno %d", | |
530 | ap->in_lbn); | |
12304d41 KB |
531 | bp->b_un.b_daddr[ap->in_off] = off; |
532 | lfs_bwrite(bp); | |
533 | } | |
534 | ||
535 | /* Update segment usage information. */ | |
536 | if (daddr != UNASSIGNED) { | |
537 | LFS_SEGENTRY(sup, fs, datosn(fs, daddr), bp); | |
12304d41 KB |
538 | #ifdef DIAGNOSTIC |
539 | if (sup->su_nbytes < fs->lfs_bsize) | |
540 | panic("lfs: negative bytes (segment %d)\n", | |
541 | datosn(fs, daddr)); | |
542 | #endif | |
543 | sup->su_nbytes -= fs->lfs_bsize; | |
a1b8db53 | 544 | LFS_UBWRITE(bp); |
84c30241 | 545 | } |
84c30241 | 546 | } |
84c30241 KB |
547 | } |
548 | ||
12304d41 KB |
549 | /* |
550 | * Start a new segment. | |
551 | */ | |
87804018 | 552 | void |
12304d41 | 553 | lfs_initseg(fs, sp) |
0a011bb1 | 554 | struct lfs *fs; |
a1b8db53 | 555 | struct segment *sp; |
84c30241 | 556 | { |
12304d41 KB |
557 | SEGUSE *sup; |
558 | SEGSUM *ssp; | |
559 | struct buf *bp; | |
560 | daddr_t lbn, *lbnp; | |
275ca4f0 | 561 | |
dc7e45d3 | 562 | #ifdef VERBOSE |
12304d41 | 563 | printf("lfs_initseg\n"); |
dc7e45d3 | 564 | #endif |
12304d41 | 565 | /* Advance to the next segment. */ |
c222b129 | 566 | if (!LFS_PARTIAL_FITS(fs)) { |
9a46ddb2 CS |
567 | /* Wake up any cleaning procs waiting on this file system. */ |
568 | wakeup((caddr_t)&fs->lfs_nextseg); | |
569 | wakeup((caddr_t)&lfs_allclean_wakeup); | |
570 | ||
c222b129 KB |
571 | lfs_newseg(fs); |
572 | fs->lfs_offset = fs->lfs_curseg; | |
12304d41 KB |
573 | sp->seg_number = datosn(fs, fs->lfs_curseg); |
574 | sp->seg_bytes_left = fs->lfs_dbpseg * DEV_BSIZE; | |
575 | ||
576 | /* | |
c222b129 KB |
577 | * If the segment contains a superblock, update the offset |
578 | * and summary address to skip over it. | |
12304d41 | 579 | */ |
87804018 | 580 | LFS_SEGENTRY(sup, fs, sp->seg_number, bp); |
c222b129 | 581 | if (sup->su_flags & SEGUSE_SUPERBLOCK) { |
12304d41 KB |
582 | fs->lfs_offset += LFS_SBPAD / DEV_BSIZE; |
583 | sp->seg_bytes_left -= LFS_SBPAD; | |
275ca4f0 | 584 | } |
a1b8db53 | 585 | brelse(bp); |
12304d41 KB |
586 | } else { |
587 | sp->seg_number = datosn(fs, fs->lfs_curseg); | |
588 | sp->seg_bytes_left = (fs->lfs_dbpseg - | |
589 | (fs->lfs_offset - fs->lfs_curseg)) * DEV_BSIZE; | |
590 | } | |
aa4dc149 | 591 | |
12304d41 KB |
592 | sp->ibp = NULL; |
593 | sp->ninodes = 0; | |
aa4dc149 | 594 | |
12304d41 KB |
595 | /* Get a new buffer for SEGSUM and enter it into the buffer list. */ |
596 | sp->cbpp = sp->bpp; | |
597 | *sp->cbpp = lfs_newbuf(fs, sp, fs->lfs_offset, LFS_SUMMARY_SIZE); | |
598 | sp->segsum = (*sp->cbpp)->b_un.b_addr; | |
599 | ++sp->cbpp; | |
600 | fs->lfs_offset += LFS_SUMMARY_SIZE / DEV_BSIZE; | |
aa4dc149 | 601 | |
12304d41 KB |
602 | /* Set point to SEGSUM, initialize it. */ |
603 | ssp = sp->segsum; | |
604 | ssp->ss_next = fs->lfs_nextseg; | |
12304d41 | 605 | ssp->ss_nfinfo = ssp->ss_ninos = 0; |
aa4dc149 | 606 | |
12304d41 | 607 | /* Set pointer to first FINFO, initialize it. */ |
a1b8db53 | 608 | sp->fip = (struct finfo *)(sp->segsum + sizeof(SEGSUM)); |
12304d41 | 609 | sp->fip->fi_nblocks = 0; |
aa4dc149 | 610 | |
12304d41 KB |
611 | sp->seg_bytes_left -= LFS_SUMMARY_SIZE; |
612 | sp->sum_bytes_left = LFS_SUMMARY_SIZE - sizeof(SEGSUM); | |
613 | } | |
aa4dc149 | 614 | |
12304d41 KB |
615 | /* |
616 | * Return the next segment to write. | |
617 | */ | |
87804018 | 618 | void |
12304d41 KB |
619 | lfs_newseg(fs) |
620 | struct lfs *fs; | |
621 | { | |
c222b129 | 622 | CLEANERINFO *cip; |
12304d41 KB |
623 | SEGUSE *sup; |
624 | struct buf *bp; | |
c222b129 | 625 | int curseg, isdirty, sn; |
12304d41 KB |
626 | |
627 | #ifdef VERBOSE | |
628 | printf("lfs_newseg\n"); | |
629 | #endif | |
c222b129 KB |
630 | /* |
631 | * Turn off the active bit for the current segment, turn on the | |
632 | * active and dirty bits for the next segment, update the cleaner | |
633 | * info. Set the current segment to the next segment, get a new | |
634 | * next segment. | |
635 | */ | |
636 | LFS_SEGENTRY(sup, fs, datosn(fs, fs->lfs_curseg), bp); | |
637 | sup->su_flags &= ~SEGUSE_ACTIVE; | |
a1b8db53 | 638 | LFS_UBWRITE(bp); |
c222b129 KB |
639 | |
640 | LFS_SEGENTRY(sup, fs, datosn(fs, fs->lfs_nextseg), bp); | |
641 | sup->su_flags |= SEGUSE_ACTIVE | SEGUSE_DIRTY; | |
a1b8db53 | 642 | LFS_UBWRITE(bp); |
c222b129 KB |
643 | |
644 | LFS_CLEANERINFO(cip, fs, bp); | |
645 | --cip->clean; | |
646 | ++cip->dirty; | |
a1b8db53 | 647 | LFS_UBWRITE(bp); |
c222b129 KB |
648 | |
649 | fs->lfs_lastseg = fs->lfs_curseg; | |
650 | fs->lfs_curseg = fs->lfs_nextseg; | |
651 | for (sn = curseg = datosn(fs, fs->lfs_curseg);;) { | |
12304d41 | 652 | sn = (sn + 1) % fs->lfs_nseg; |
c222b129 | 653 | if (sn == curseg) |
12304d41 KB |
654 | panic("lfs_nextseg: no clean segments"); |
655 | LFS_SEGENTRY(sup, fs, sn, bp); | |
656 | isdirty = sup->su_flags & SEGUSE_DIRTY; | |
a1b8db53 | 657 | brelse(bp); |
12304d41 KB |
658 | if (!isdirty) |
659 | break; | |
660 | } | |
c222b129 | 661 | fs->lfs_nextseg = sntoda(fs, sn); |
84c30241 KB |
662 | } |
663 | ||
87804018 | 664 | void |
84c30241 | 665 | lfs_writeseg(fs, sp) |
0a011bb1 | 666 | struct lfs *fs; |
a1b8db53 | 667 | struct segment *sp; |
84c30241 | 668 | { |
a1b8db53 | 669 | struct buf **bpp, *bp; |
84c30241 | 670 | SEGUSE *sup; |
a1b8db53 | 671 | SEGSUM *ssp; |
dc7e45d3 KB |
672 | dev_t i_dev; |
673 | u_long *datap, *dp; | |
aa4dc149 | 674 | void *pmeta; |
a1b8db53 | 675 | int flags, i, nblocks, s, (*strategy)__P((struct buf *)); |
84c30241 | 676 | |
dc7e45d3 KB |
677 | #ifdef VERBOSE |
678 | printf("lfs_writeseg\n"); | |
679 | #endif | |
a1b8db53 KB |
680 | if ((nblocks = sp->cbpp - sp->bpp) == 0) |
681 | return; | |
682 | ||
84c30241 | 683 | /* |
a1b8db53 KB |
684 | * Compute checksum across data and then across summary; the first |
685 | * block (the summary block) is skipped. Set the create time here | |
686 | * so that it's guaranteed to be later than the inode mod times. | |
dc7e45d3 KB |
687 | * |
688 | * XXX | |
689 | * Fix this to do it inline, instead of malloc/copy. | |
84c30241 | 690 | */ |
dc7e45d3 | 691 | datap = dp = malloc(nblocks * sizeof(u_long), M_SEGMENT, M_WAITOK); |
12304d41 KB |
692 | for (bpp = sp->bpp, i = nblocks - 1; i--;) |
693 | *dp++ = (*++bpp)->b_un.b_words[0]; | |
a1b8db53 | 694 | ssp = (SEGSUM *)sp->segsum; |
89bed312 | 695 | ssp->ss_create = time.tv_sec; |
a1b8db53 KB |
696 | ssp->ss_datasum = cksum(datap, nblocks * sizeof(u_long)); |
697 | ssp->ss_sumsum = | |
698 | cksum(&ssp->ss_datasum, LFS_SUMMARY_SIZE - sizeof(ssp->ss_sumsum)); | |
c222b129 | 699 | free(datap, M_SEGMENT); |
8954e52c | 700 | |
dc7e45d3 KB |
701 | /* |
702 | * When we gathered the blocks for I/O we did not mark them busy or | |
703 | * remove them from the freelist. As we do this, turn off the B_LOCKED | |
704 | * bit so the future brelse will put them on the LRU list, and add the | |
705 | * B_CALL flags if we're doing a checkpoint so we can count I/O's. LFS | |
706 | * requires that the super blocks (on checkpoint) be written after all | |
707 | * the segment data. | |
708 | */ | |
709 | i_dev = VTOI(fs->lfs_ivnode)->i_dev; | |
710 | strategy = VTOI(fs->lfs_ivnode)->i_devvp->v_op->vop_strategy; | |
275ca4f0 | 711 | |
8954e52c | 712 | s = splbio(); |
dc7e45d3 KB |
713 | if (sp->seg_flags & SEGM_CKP) { |
714 | fs->lfs_iocount += nblocks; | |
12304d41 | 715 | flags = B_ASYNC | B_BUSY | B_CALL; |
dc7e45d3 | 716 | } else |
12304d41 | 717 | flags = B_ASYNC | B_BUSY; |
dc7e45d3 KB |
718 | for (bpp = sp->bpp, i = nblocks; i--;) { |
719 | bp = *bpp++; | |
720 | bp->b_flags |= flags; | |
12304d41 KB |
721 | bp->b_flags &= |
722 | ~(B_DONE | B_ERROR | B_READ | B_DELWRI | B_LOCKED); | |
dc7e45d3 KB |
723 | bp->b_dev = i_dev; |
724 | bp->b_iodone = lfs_callback; | |
dc7e45d3 KB |
725 | if (!(bp->b_flags & B_NOCACHE)) { |
726 | bremfree(bp); | |
727 | reassignbuf(bp, bp->b_vp); | |
728 | } | |
8954e52c | 729 | } |
dc7e45d3 | 730 | splx(s); |
87804018 | 731 | |
dc7e45d3 KB |
732 | for (bpp = sp->bpp, i = nblocks; i--;) |
733 | (strategy)(*bpp++); | |
9a46ddb2 CS |
734 | |
735 | /* Update the segment usage information. */ | |
736 | LFS_SEGENTRY(sup, fs, sp->seg_number, bp); | |
737 | sup->su_nbytes += nblocks - 1 - | |
738 | (ssp->ss_ninos + INOPB(fs) - 1) / INOPB(fs) << fs->lfs_bshift; | |
739 | sup->su_nbytes += ssp->ss_ninos * sizeof(struct dinode); | |
740 | sup->su_lastmod = time.tv_sec; | |
741 | LFS_UBWRITE(bp); | |
275ca4f0 KB |
742 | } |
743 | ||
87804018 | 744 | void |
dc7e45d3 | 745 | lfs_writesuper(fs, sp) |
0a011bb1 | 746 | struct lfs *fs; |
a1b8db53 | 747 | struct segment *sp; |
275ca4f0 | 748 | { |
a1b8db53 | 749 | struct buf *bp; |
dc7e45d3 | 750 | dev_t i_dev; |
a1b8db53 | 751 | int (*strategy) __P((struct buf *)); |
275ca4f0 | 752 | |
dc7e45d3 KB |
753 | #ifdef VERBOSE |
754 | printf("lfs_writesuper\n"); | |
755 | #endif | |
dc7e45d3 KB |
756 | i_dev = VTOI(fs->lfs_ivnode)->i_dev; |
757 | strategy = VTOI(fs->lfs_ivnode)->i_devvp->v_op->vop_strategy; | |
14712628 | 758 | |
aa4dc149 | 759 | /* Checksum the superblock and copy it into a buffer. */ |
0a011bb1 | 760 | fs->lfs_cksum = cksum(fs, sizeof(struct lfs) - sizeof(fs->lfs_cksum)); |
dc7e45d3 KB |
761 | bp = lfs_newbuf(fs, sp, fs->lfs_sboffs[0], LFS_SBPAD); |
762 | *bp->b_un.b_lfs = *fs; | |
275ca4f0 | 763 | |
14712628 | 764 | /* Write the first superblock (wait). */ |
dc7e45d3 | 765 | bp->b_dev = i_dev; |
dc7e45d3 | 766 | bp->b_flags |= B_BUSY; |
12304d41 | 767 | bp->b_flags &= ~(B_DONE | B_ERROR | B_READ | B_DELWRI); |
aa4dc149 | 768 | (strategy)(bp); |
275ca4f0 | 769 | biowait(bp); |
aa4dc149 | 770 | |
14712628 | 771 | /* Write the second superblock (don't wait). */ |
275ca4f0 | 772 | bp->b_blkno = bp->b_lblkno = fs->lfs_sboffs[1]; |
12304d41 | 773 | bp->b_flags |= B_ASYNC | B_BUSY; |
dc7e45d3 | 774 | bp->b_flags &= ~(B_DONE | B_ERROR | B_READ | B_DELWRI); |
aa4dc149 | 775 | (strategy)(bp); |
275ca4f0 KB |
776 | } |
777 | ||
aa4dc149 KB |
778 | /* |
779 | * Logical block number match routines used when traversing the dirty block | |
780 | * chain. | |
781 | */ | |
87804018 KB |
782 | int |
783 | lfs_match_data(fs, bp) | |
dc7e45d3 | 784 | struct lfs *fs; |
a1b8db53 | 785 | struct buf *bp; |
275ca4f0 | 786 | { |
aa4dc149 | 787 | return (bp->b_lblkno >= 0); |
275ca4f0 KB |
788 | } |
789 | ||
87804018 KB |
790 | int |
791 | lfs_match_indir(fs, bp) | |
dc7e45d3 | 792 | struct lfs *fs; |
a1b8db53 | 793 | struct buf *bp; |
275ca4f0 | 794 | { |
dc7e45d3 KB |
795 | int lbn; |
796 | ||
797 | lbn = bp->b_lblkno; | |
798 | return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 0); | |
275ca4f0 KB |
799 | } |
800 | ||
87804018 KB |
801 | int |
802 | lfs_match_dindir(fs, bp) | |
dc7e45d3 | 803 | struct lfs *fs; |
a1b8db53 | 804 | struct buf *bp; |
275ca4f0 | 805 | { |
dc7e45d3 KB |
806 | int lbn; |
807 | ||
808 | lbn = bp->b_lblkno; | |
809 | return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 1); | |
aa4dc149 KB |
810 | } |
811 | ||
87804018 KB |
812 | int |
813 | lfs_match_tindir(fs, bp) | |
0a011bb1 | 814 | struct lfs *fs; |
a1b8db53 | 815 | struct buf *bp; |
aa4dc149 | 816 | { |
dc7e45d3 | 817 | int lbn; |
aa4dc149 | 818 | |
dc7e45d3 KB |
819 | lbn = bp->b_lblkno; |
820 | return (lbn < 0 && (-lbn - NDADDR) % NINDIR(fs) == 2); | |
821 | } | |
aa4dc149 | 822 | |
dc7e45d3 KB |
823 | /* |
824 | * Allocate a new buffer header. | |
825 | */ | |
a1b8db53 | 826 | struct buf * |
dc7e45d3 KB |
827 | lfs_newbuf(fs, sp, daddr, size) |
828 | struct lfs *fs; | |
a1b8db53 | 829 | struct segment *sp; |
dc7e45d3 KB |
830 | daddr_t daddr; |
831 | size_t size; | |
832 | { | |
a1b8db53 | 833 | struct buf *bp; |
aa4dc149 | 834 | |
dc7e45d3 KB |
835 | #ifdef VERBOSE |
836 | printf("lfs_newbuf\n"); | |
837 | #endif | |
838 | bp = getnewbuf(); | |
839 | bremhash(bp); | |
840 | bgetvp(fs->lfs_ivnode, bp); | |
841 | bp->b_bcount = 0; | |
842 | bp->b_lblkno = daddr; | |
843 | bp->b_blkno = daddr; | |
844 | bp->b_error = 0; | |
845 | bp->b_resid = 0; | |
846 | allocbuf(bp, size); | |
847 | bp->b_flags |= B_NOCACHE; | |
12304d41 | 848 | binshash(bp, &bfreelist[BQ_AGE]); |
dc7e45d3 KB |
849 | return (bp); |
850 | } | |
aa4dc149 | 851 | |
87804018 | 852 | int /* XXX should be void */ |
dc7e45d3 | 853 | lfs_callback(bp) |
a1b8db53 | 854 | struct buf *bp; |
dc7e45d3 KB |
855 | { |
856 | struct lfs *fs; | |
aa4dc149 | 857 | |
dc7e45d3 KB |
858 | fs = VFSTOUFS(bp->b_vp->v_mount)->um_lfs; |
859 | #ifdef DIAGNOSTIC | |
860 | if (fs->lfs_iocount == 0) | |
861 | panic("lfs_callback: zero iocount\n"); | |
862 | #endif | |
863 | if (--fs->lfs_iocount == 0) | |
b68b931c | 864 | wakeup((caddr_t)&fs->lfs_iocount); |
12304d41 | 865 | |
dc7e45d3 | 866 | brelse(bp); |
84c30241 KB |
867 | } |
868 | ||
869 | /* | |
870 | * Shellsort (diminishing increment sort) from Data Structures and | |
871 | * Algorithms, Aho, Hopcraft and Ullman, 1983 Edition, page 290; | |
872 | * see also Knuth Vol. 3, page 84. The increments are selected from | |
873 | * formula (8), page 95. Roughly O(N^3/2). | |
874 | */ | |
875 | /* | |
876 | * This is our own private copy of shellsort because we want to sort | |
877 | * two parallel arrays (the array of buffer pointers and the array of | |
878 | * logical block numbers) simultaneously. Note that we cast the array | |
879 | * of logical block numbers to a unsigned in this routine so that the | |
880 | * negative block numbers (meta data blocks) sort AFTER the data blocks. | |
881 | */ | |
87804018 KB |
882 | void |
883 | lfs_shellsort(bp_array, lb_array, nmemb) | |
a1b8db53 | 884 | struct buf **bp_array; |
275ca4f0 | 885 | daddr_t *lb_array; |
84c30241 KB |
886 | register int nmemb; |
887 | { | |
888 | static int __rsshell_increments[] = { 4, 1, 0 }; | |
889 | register int incr, *incrp, t1, t2; | |
a1b8db53 | 890 | struct buf *bp_temp; |
84c30241 KB |
891 | u_long lb_temp; |
892 | ||
893 | for (incrp = __rsshell_increments; incr = *incrp++;) | |
894 | for (t1 = incr; t1 < nmemb; ++t1) | |
895 | for (t2 = t1 - incr; t2 >= 0;) | |
896 | if (lb_array[t2] > lb_array[t2 + incr]) { | |
897 | lb_temp = lb_array[t2]; | |
898 | lb_array[t2] = lb_array[t2 + incr]; | |
899 | lb_array[t2 + incr] = lb_temp; | |
900 | bp_temp = bp_array[t2]; | |
901 | bp_array[t2] = bp_array[t2 + incr]; | |
902 | bp_array[t2 + incr] = bp_temp; | |
903 | t2 -= incr; | |
904 | } else | |
905 | break; | |
906 | } |