Commit | Line | Data |
---|---|---|
973ecc4f | 1 | /* vfs_cluster.c 4.4 %G% */ |
663dbc72 BJ |
2 | |
3 | #include "../h/param.h" | |
4 | #include "../h/systm.h" | |
5 | #include "../h/dir.h" | |
6 | #include "../h/user.h" | |
7 | #include "../h/buf.h" | |
8 | #include "../h/conf.h" | |
9 | #include "../h/proc.h" | |
10 | #include "../h/seg.h" | |
11 | #include "../h/pte.h" | |
12 | #include "../h/vm.h" | |
973ecc4f | 13 | #include "../h/trace.h" |
663dbc72 | 14 | |
5603d07d BJ |
15 | /* |
16 | * The following several routines allocate and free | |
17 | * buffers with various side effects. In general the | |
18 | * arguments to an allocate routine are a device and | |
19 | * a block number, and the value is a pointer to | |
20 | * to the buffer header; the buffer is marked "busy" | |
21 | * so that no one else can touch it. If the block was | |
22 | * already in core, no I/O need be done; if it is | |
23 | * already busy, the process waits until it becomes free. | |
24 | * The following routines allocate a buffer: | |
25 | * getblk | |
26 | * bread | |
27 | * breada | |
28 | * baddr (if it is incore) | |
29 | * Eventually the buffer must be released, possibly with the | |
30 | * side effect of writing it out, by using one of | |
31 | * bwrite | |
32 | * bdwrite | |
33 | * bawrite | |
34 | * brelse | |
35 | */ | |
36 | ||
37 | #define BUFHSZ 63 | |
38 | #define BUFHASH(blkno) (blkno % BUFHSZ) | |
39 | short bufhash[BUFHSZ]; | |
40 | ||
41 | /* | |
42 | * Initialize hash links for buffers. | |
43 | */ | |
44 | bhinit() | |
45 | { | |
46 | register int i; | |
47 | ||
48 | for (i = 0; i < BUFHSZ; i++) | |
49 | bufhash[i] = -1; | |
50 | } | |
51 | ||
663dbc72 BJ |
52 | /* #define DISKMON 1 */ |
53 | ||
54 | #ifdef DISKMON | |
55 | struct { | |
56 | int nbuf; | |
57 | long nread; | |
58 | long nreada; | |
59 | long ncache; | |
60 | long nwrite; | |
61 | long bufcount[NBUF]; | |
62 | } io_info; | |
63 | #endif | |
64 | ||
65 | /* | |
66 | * Swap IO headers - | |
67 | * They contain the necessary information for the swap I/O. | |
68 | * At any given time, a swap header can be in three | |
69 | * different lists. When free it is in the free list, | |
70 | * when allocated and the I/O queued, it is on the swap | |
71 | * device list, and finally, if the operation was a dirty | |
72 | * page push, when the I/O completes, it is inserted | |
73 | * in a list of cleaned pages to be processed by the pageout daemon. | |
74 | */ | |
75 | struct buf swbuf[NSWBUF]; | |
76 | short swsize[NSWBUF]; /* CAN WE JUST USE B_BCOUNT? */ | |
77 | int swpf[NSWBUF]; | |
78 | ||
663dbc72 BJ |
79 | |
80 | #ifdef FASTVAX | |
81 | #define notavail(bp) \ | |
82 | { \ | |
83 | int s = spl6(); \ | |
84 | (bp)->av_back->av_forw = (bp)->av_forw; \ | |
85 | (bp)->av_forw->av_back = (bp)->av_back; \ | |
86 | (bp)->b_flags |= B_BUSY; \ | |
87 | splx(s); \ | |
88 | } | |
89 | #endif | |
90 | ||
91 | /* | |
92 | * Read in (if necessary) the block and return a buffer pointer. | |
93 | */ | |
94 | struct buf * | |
95 | bread(dev, blkno) | |
96 | dev_t dev; | |
97 | daddr_t blkno; | |
98 | { | |
99 | register struct buf *bp; | |
100 | ||
101 | bp = getblk(dev, blkno); | |
102 | if (bp->b_flags&B_DONE) { | |
973ecc4f BJ |
103 | #ifdef EPAWNJ |
104 | trace(TR_BREAD|TR_HIT, dev, blkno); | |
105 | #endif | |
663dbc72 BJ |
106 | #ifdef DISKMON |
107 | io_info.ncache++; | |
108 | #endif | |
109 | return(bp); | |
110 | } | |
111 | bp->b_flags |= B_READ; | |
112 | bp->b_bcount = BSIZE; | |
113 | (*bdevsw[major(dev)].d_strategy)(bp); | |
973ecc4f BJ |
114 | #ifdef EPAWNJ |
115 | trace(TR_BREAD|TR_MISS, dev, blkno); | |
116 | #endif | |
663dbc72 BJ |
117 | #ifdef DISKMON |
118 | io_info.nread++; | |
119 | #endif | |
120 | u.u_vm.vm_inblk++; /* pay for read */ | |
121 | iowait(bp); | |
122 | return(bp); | |
123 | } | |
124 | ||
125 | /* | |
126 | * Read in the block, like bread, but also start I/O on the | |
127 | * read-ahead block (which is not allocated to the caller) | |
128 | */ | |
129 | struct buf * | |
130 | breada(dev, blkno, rablkno) | |
131 | dev_t dev; | |
132 | daddr_t blkno, rablkno; | |
133 | { | |
134 | register struct buf *bp, *rabp; | |
135 | ||
136 | bp = NULL; | |
137 | if (!incore(dev, blkno)) { | |
138 | bp = getblk(dev, blkno); | |
139 | if ((bp->b_flags&B_DONE) == 0) { | |
140 | bp->b_flags |= B_READ; | |
141 | bp->b_bcount = BSIZE; | |
142 | (*bdevsw[major(dev)].d_strategy)(bp); | |
973ecc4f BJ |
143 | #ifdef EPAWNJ |
144 | trace(TR_BREAD|TR_MISS, dev, blkno); | |
145 | #endif | |
663dbc72 BJ |
146 | #ifdef DISKMON |
147 | io_info.nread++; | |
148 | #endif | |
149 | u.u_vm.vm_inblk++; /* pay for read */ | |
150 | } | |
973ecc4f BJ |
151 | #ifdef EPAWNJ |
152 | else | |
153 | trace(TR_BREAD|TR_HIT, dev, blkno); | |
154 | #endif | |
663dbc72 BJ |
155 | } |
156 | if (rablkno && !incore(dev, rablkno)) { | |
157 | rabp = getblk(dev, rablkno); | |
973ecc4f | 158 | if (rabp->b_flags & B_DONE) { |
663dbc72 | 159 | brelse(rabp); |
973ecc4f BJ |
160 | #ifdef EPAWNJ |
161 | trace(TR_BREAD|TR_HIT|TR_RA, dev, blkno); | |
162 | #endif | |
163 | } else { | |
663dbc72 BJ |
164 | rabp->b_flags |= B_READ|B_ASYNC; |
165 | rabp->b_bcount = BSIZE; | |
166 | (*bdevsw[major(dev)].d_strategy)(rabp); | |
973ecc4f BJ |
167 | #ifdef EPAWNJ |
168 | trace(TR_BREAD|TR_MISS|TR_RA, dev, rablock); | |
169 | #endif | |
663dbc72 BJ |
170 | #ifdef DISKMON |
171 | io_info.nreada++; | |
172 | #endif | |
173 | u.u_vm.vm_inblk++; /* pay in advance */ | |
174 | } | |
175 | } | |
176 | if(bp == NULL) | |
177 | return(bread(dev, blkno)); | |
178 | iowait(bp); | |
179 | return(bp); | |
180 | } | |
181 | ||
182 | /* | |
183 | * Write the buffer, waiting for completion. | |
184 | * Then release the buffer. | |
185 | */ | |
186 | bwrite(bp) | |
187 | register struct buf *bp; | |
188 | { | |
189 | register flag; | |
190 | ||
191 | flag = bp->b_flags; | |
192 | bp->b_flags &= ~(B_READ | B_DONE | B_ERROR | B_DELWRI | B_AGE); | |
193 | bp->b_bcount = BSIZE; | |
194 | #ifdef DISKMON | |
195 | io_info.nwrite++; | |
196 | #endif | |
197 | if ((flag&B_DELWRI) == 0) | |
198 | u.u_vm.vm_oublk++; /* noone paid yet */ | |
973ecc4f BJ |
199 | #ifdef EPAWNJ |
200 | trace(TR_BWRITE, bp->b_dev, dbtofsb(bp->b_blkno)); | |
201 | #endif | |
663dbc72 BJ |
202 | (*bdevsw[major(bp->b_dev)].d_strategy)(bp); |
203 | if ((flag&B_ASYNC) == 0) { | |
204 | iowait(bp); | |
205 | brelse(bp); | |
206 | } else if (flag & B_DELWRI) | |
207 | bp->b_flags |= B_AGE; | |
208 | else | |
209 | geterror(bp); | |
210 | } | |
211 | ||
212 | /* | |
213 | * Release the buffer, marking it so that if it is grabbed | |
214 | * for another purpose it will be written out before being | |
215 | * given up (e.g. when writing a partial block where it is | |
216 | * assumed that another write for the same block will soon follow). | |
217 | * This can't be done for magtape, since writes must be done | |
218 | * in the same order as requested. | |
219 | */ | |
220 | bdwrite(bp) | |
221 | register struct buf *bp; | |
222 | { | |
223 | register struct buf *dp; | |
224 | ||
225 | if ((bp->b_flags&B_DELWRI) == 0) | |
226 | u.u_vm.vm_oublk++; /* noone paid yet */ | |
227 | dp = bdevsw[major(bp->b_dev)].d_tab; | |
228 | if(dp->b_flags & B_TAPE) | |
229 | bawrite(bp); | |
230 | else { | |
231 | bp->b_flags |= B_DELWRI | B_DONE; | |
232 | brelse(bp); | |
233 | } | |
234 | } | |
235 | ||
236 | /* | |
237 | * Release the buffer, start I/O on it, but don't wait for completion. | |
238 | */ | |
239 | bawrite(bp) | |
240 | register struct buf *bp; | |
241 | { | |
242 | ||
243 | bp->b_flags |= B_ASYNC; | |
244 | bwrite(bp); | |
245 | } | |
246 | ||
247 | /* | |
248 | * release the buffer, with no I/O implied. | |
249 | */ | |
250 | brelse(bp) | |
251 | register struct buf *bp; | |
252 | { | |
253 | register struct buf **backp; | |
254 | register s; | |
255 | ||
256 | if (bp->b_flags&B_WANTED) | |
257 | wakeup((caddr_t)bp); | |
258 | if (bfreelist.b_flags&B_WANTED) { | |
259 | bfreelist.b_flags &= ~B_WANTED; | |
260 | wakeup((caddr_t)&bfreelist); | |
261 | } | |
5603d07d BJ |
262 | if ((bp->b_flags&B_ERROR) && bp->b_dev != NODEV) { |
263 | bunhash(bp); | |
663dbc72 | 264 | bp->b_dev = NODEV; /* no assoc. on error */ |
5603d07d | 265 | } |
663dbc72 BJ |
266 | s = spl6(); |
267 | if(bp->b_flags & (B_AGE|B_ERROR)) { | |
268 | backp = &bfreelist.av_forw; | |
269 | (*backp)->av_back = bp; | |
270 | bp->av_forw = *backp; | |
271 | *backp = bp; | |
272 | bp->av_back = &bfreelist; | |
273 | } else { | |
274 | backp = &bfreelist.av_back; | |
275 | (*backp)->av_forw = bp; | |
276 | bp->av_back = *backp; | |
277 | *backp = bp; | |
278 | bp->av_forw = &bfreelist; | |
279 | } | |
280 | bp->b_flags &= ~(B_WANTED|B_BUSY|B_ASYNC|B_AGE); | |
281 | splx(s); | |
282 | } | |
283 | ||
284 | /* | |
285 | * See if the block is associated with some buffer | |
286 | * (mainly to avoid getting hung up on a wait in breada) | |
287 | */ | |
288 | incore(dev, blkno) | |
289 | dev_t dev; | |
290 | daddr_t blkno; | |
291 | { | |
292 | register struct buf *bp; | |
663dbc72 BJ |
293 | register int dblkno = fsbtodb(blkno); |
294 | ||
5603d07d BJ |
295 | for (bp = &buf[bufhash[BUFHASH(blkno)]]; bp != &buf[-1]; |
296 | bp = &buf[bp->b_hlink]) | |
fe8987fb | 297 | if (bp->b_blkno == dblkno && bp->b_dev == dev) |
5603d07d | 298 | return (1); |
5603d07d | 299 | return (0); |
663dbc72 BJ |
300 | } |
301 | ||
302 | struct buf * | |
303 | baddr(dev, blkno) | |
304 | dev_t dev; | |
305 | daddr_t blkno; | |
306 | { | |
307 | ||
308 | if (incore(dev, blkno)) | |
309 | return (bread(dev, blkno)); | |
310 | return (0); | |
311 | } | |
312 | ||
313 | /* | |
314 | * Assign a buffer for the given block. If the appropriate | |
315 | * block is already associated, return it; otherwise search | |
316 | * for the oldest non-busy buffer and reassign it. | |
317 | */ | |
318 | struct buf * | |
319 | getblk(dev, blkno) | |
320 | dev_t dev; | |
321 | daddr_t blkno; | |
322 | { | |
5603d07d | 323 | register struct buf *bp, *dp, *ep; |
01659974 | 324 | register int i, x, dblkno; |
663dbc72 | 325 | |
01659974 BJ |
326 | if ((unsigned)blkno >= 1 << (sizeof(int)*NBBY-PGSHIFT)) |
327 | blkno = 1 << ((sizeof(int)*NBBY-PGSHIFT) + 1); | |
328 | dblkno = fsbtodb(blkno); | |
663dbc72 | 329 | loop: |
81263dba | 330 | (void) spl0(); |
5603d07d BJ |
331 | for (bp = &buf[bufhash[BUFHASH(blkno)]]; bp != &buf[-1]; |
332 | bp = &buf[bp->b_hlink]) { | |
333 | if (bp->b_blkno != dblkno || bp->b_dev != dev) | |
663dbc72 | 334 | continue; |
81263dba | 335 | (void) spl6(); |
663dbc72 BJ |
336 | if (bp->b_flags&B_BUSY) { |
337 | bp->b_flags |= B_WANTED; | |
338 | sleep((caddr_t)bp, PRIBIO+1); | |
339 | goto loop; | |
340 | } | |
81263dba | 341 | (void) spl0(); |
663dbc72 BJ |
342 | #ifdef DISKMON |
343 | i = 0; | |
344 | dp = bp->av_forw; | |
345 | while (dp != &bfreelist) { | |
346 | i++; | |
347 | dp = dp->av_forw; | |
348 | } | |
349 | if (i<NBUF) | |
350 | io_info.bufcount[i]++; | |
351 | #endif | |
352 | notavail(bp); | |
353 | bp->b_flags |= B_CACHE; | |
354 | return(bp); | |
355 | } | |
5603d07d BJ |
356 | if (major(dev) >= nblkdev) |
357 | panic("blkdev"); | |
358 | dp = bdevsw[major(dev)].d_tab; | |
359 | if (dp == NULL) | |
360 | panic("devtab"); | |
81263dba | 361 | (void) spl6(); |
663dbc72 BJ |
362 | if (bfreelist.av_forw == &bfreelist) { |
363 | bfreelist.b_flags |= B_WANTED; | |
364 | sleep((caddr_t)&bfreelist, PRIBIO+1); | |
365 | goto loop; | |
366 | } | |
283cac0a | 367 | (void) spl0(); |
663dbc72 BJ |
368 | bp = bfreelist.av_forw; |
369 | notavail(bp); | |
370 | if (bp->b_flags & B_DELWRI) { | |
371 | bp->b_flags |= B_ASYNC; | |
372 | bwrite(bp); | |
373 | goto loop; | |
374 | } | |
5603d07d BJ |
375 | if (bp->b_dev == NODEV) |
376 | goto done; | |
377 | /* INLINE EXPANSION OF bunhash(bp) */ | |
973ecc4f BJ |
378 | #ifdef EPAWNJ |
379 | trace(TR_BRELSE, bp->b_dev, dbtofsb(bp->b_blkno)); | |
380 | #endif | |
99fae0e8 | 381 | (void) spl6(); |
5603d07d BJ |
382 | i = BUFHASH(dbtofsb(bp->b_blkno)); |
383 | x = bp - buf; | |
384 | if (bufhash[i] == x) { | |
385 | bufhash[i] = bp->b_hlink; | |
386 | } else { | |
387 | for (ep = &buf[bufhash[i]]; ep != &buf[-1]; | |
388 | ep = &buf[ep->b_hlink]) | |
389 | if (ep->b_hlink == x) { | |
390 | ep->b_hlink = bp->b_hlink; | |
391 | goto done; | |
392 | } | |
393 | panic("getblk"); | |
394 | } | |
395 | done: | |
99fae0e8 | 396 | (void) spl0(); |
5603d07d | 397 | /* END INLINE EXPANSION */ |
663dbc72 BJ |
398 | bp->b_flags = B_BUSY; |
399 | bp->b_back->b_forw = bp->b_forw; | |
400 | bp->b_forw->b_back = bp->b_back; | |
401 | bp->b_forw = dp->b_forw; | |
402 | bp->b_back = dp; | |
403 | dp->b_forw->b_back = bp; | |
404 | dp->b_forw = bp; | |
405 | bp->b_dev = dev; | |
406 | bp->b_blkno = dblkno; | |
5603d07d BJ |
407 | i = BUFHASH(blkno); |
408 | bp->b_hlink = bufhash[i]; | |
409 | bufhash[i] = bp - buf; | |
663dbc72 BJ |
410 | return(bp); |
411 | } | |
412 | ||
413 | /* | |
414 | * get an empty block, | |
415 | * not assigned to any particular device | |
416 | */ | |
417 | struct buf * | |
418 | geteblk() | |
419 | { | |
436518b9 | 420 | register struct buf *bp, *dp; |
663dbc72 BJ |
421 | |
422 | loop: | |
81263dba | 423 | (void) spl6(); |
663dbc72 BJ |
424 | while (bfreelist.av_forw == &bfreelist) { |
425 | bfreelist.b_flags |= B_WANTED; | |
426 | sleep((caddr_t)&bfreelist, PRIBIO+1); | |
427 | } | |
81263dba | 428 | (void) spl0(); |
663dbc72 BJ |
429 | dp = &bfreelist; |
430 | bp = bfreelist.av_forw; | |
431 | notavail(bp); | |
432 | if (bp->b_flags & B_DELWRI) { | |
433 | bp->b_flags |= B_ASYNC; | |
434 | bwrite(bp); | |
435 | goto loop; | |
436 | } | |
973ecc4f BJ |
437 | if (bp->b_dev != NODEV) { |
438 | #ifdef EPAWNJ | |
439 | trace(TR_BRELSE, bp->b_dev, dbtofsb(bp->b_blkno)); | |
440 | #endif | |
5603d07d | 441 | bunhash(bp); |
973ecc4f | 442 | } |
663dbc72 BJ |
443 | bp->b_flags = B_BUSY; |
444 | bp->b_back->b_forw = bp->b_forw; | |
445 | bp->b_forw->b_back = bp->b_back; | |
446 | bp->b_forw = dp->b_forw; | |
447 | bp->b_back = dp; | |
448 | dp->b_forw->b_back = bp; | |
449 | dp->b_forw = bp; | |
450 | bp->b_dev = (dev_t)NODEV; | |
5603d07d | 451 | bp->b_hlink = -1; |
663dbc72 BJ |
452 | return(bp); |
453 | } | |
454 | ||
5603d07d BJ |
455 | bunhash(bp) |
456 | register struct buf *bp; | |
457 | { | |
458 | register struct buf *ep; | |
99fae0e8 | 459 | register int i, x, s; |
5603d07d BJ |
460 | |
461 | if (bp->b_dev == NODEV) | |
462 | return; | |
99fae0e8 | 463 | s = spl6(); |
5603d07d BJ |
464 | i = BUFHASH(dbtofsb(bp->b_blkno)); |
465 | x = bp - buf; | |
466 | if (bufhash[i] == x) { | |
467 | bufhash[i] = bp->b_hlink; | |
99fae0e8 | 468 | goto ret; |
5603d07d BJ |
469 | } |
470 | for (ep = &buf[bufhash[i]]; ep != &buf[-1]; | |
471 | ep = &buf[ep->b_hlink]) | |
472 | if (ep->b_hlink == x) { | |
473 | ep->b_hlink = bp->b_hlink; | |
99fae0e8 | 474 | goto ret; |
5603d07d BJ |
475 | } |
476 | panic("bunhash"); | |
99fae0e8 BJ |
477 | ret: |
478 | splx(s); | |
5603d07d BJ |
479 | } |
480 | ||
663dbc72 BJ |
481 | /* |
482 | * Wait for I/O completion on the buffer; return errors | |
483 | * to the user. | |
484 | */ | |
485 | iowait(bp) | |
486 | register struct buf *bp; | |
487 | { | |
488 | ||
81263dba | 489 | (void) spl6(); |
663dbc72 BJ |
490 | while ((bp->b_flags&B_DONE)==0) |
491 | sleep((caddr_t)bp, PRIBIO); | |
81263dba | 492 | (void) spl0(); |
663dbc72 BJ |
493 | geterror(bp); |
494 | } | |
495 | ||
496 | #ifndef FASTVAX | |
497 | /* | |
498 | * Unlink a buffer from the available list and mark it busy. | |
499 | * (internal interface) | |
500 | */ | |
501 | notavail(bp) | |
502 | register struct buf *bp; | |
503 | { | |
504 | register s; | |
505 | ||
506 | s = spl6(); | |
507 | bp->av_back->av_forw = bp->av_forw; | |
508 | bp->av_forw->av_back = bp->av_back; | |
509 | bp->b_flags |= B_BUSY; | |
510 | splx(s); | |
511 | } | |
512 | #endif | |
513 | ||
514 | /* | |
515 | * Mark I/O complete on a buffer. If the header | |
516 | * indicates a dirty page push completion, the | |
517 | * header is inserted into the ``cleaned'' list | |
518 | * to be processed by the pageout daemon. Otherwise | |
519 | * release it if I/O is asynchronous, and wake | |
520 | * up anyone waiting for it. | |
521 | */ | |
522 | iodone(bp) | |
523 | register struct buf *bp; | |
524 | { | |
525 | register int s; | |
526 | ||
80e7c811 BJ |
527 | if (bp->b_flags & B_DONE) |
528 | panic("dup iodone"); | |
663dbc72 BJ |
529 | bp->b_flags |= B_DONE; |
530 | if (bp->b_flags & B_DIRTY) { | |
531 | if (bp->b_flags & B_ERROR) | |
532 | panic("IO err in push"); | |
533 | s = spl6(); | |
534 | cnt.v_pgout++; | |
535 | bp->av_forw = bclnlist; | |
536 | bp->b_bcount = swsize[bp - swbuf]; | |
537 | bp->b_pfcent = swpf[bp - swbuf]; | |
538 | bclnlist = bp; | |
539 | if (bswlist.b_flags & B_WANTED) | |
540 | wakeup((caddr_t)&proc[2]); | |
541 | splx(s); | |
a3ee1d55 | 542 | return; |
663dbc72 BJ |
543 | } |
544 | if (bp->b_flags&B_ASYNC) | |
545 | brelse(bp); | |
546 | else { | |
547 | bp->b_flags &= ~B_WANTED; | |
548 | wakeup((caddr_t)bp); | |
549 | } | |
550 | } | |
551 | ||
552 | /* | |
553 | * Zero the core associated with a buffer. | |
554 | */ | |
555 | clrbuf(bp) | |
556 | struct buf *bp; | |
557 | { | |
558 | register *p; | |
559 | register c; | |
560 | ||
561 | p = bp->b_un.b_words; | |
562 | c = BSIZE/sizeof(int); | |
563 | do | |
564 | *p++ = 0; | |
565 | while (--c); | |
566 | bp->b_resid = 0; | |
567 | } | |
568 | ||
569 | /* | |
570 | * swap I/O - | |
571 | * | |
572 | * If the flag indicates a dirty page push initiated | |
573 | * by the pageout daemon, we map the page into the i th | |
574 | * virtual page of process 2 (the daemon itself) where i is | |
575 | * the index of the swap header that has been allocated. | |
576 | * We simply initialize the header and queue the I/O but | |
577 | * do not wait for completion. When the I/O completes, | |
578 | * iodone() will link the header to a list of cleaned | |
579 | * pages to be processed by the pageout daemon. | |
580 | */ | |
581 | swap(p, dblkno, addr, nbytes, rdflg, flag, dev, pfcent) | |
582 | struct proc *p; | |
583 | swblk_t dblkno; | |
584 | caddr_t addr; | |
585 | int flag, nbytes; | |
586 | dev_t dev; | |
587 | unsigned pfcent; | |
588 | { | |
589 | register struct buf *bp; | |
590 | register int c; | |
591 | int p2dp; | |
592 | register struct pte *dpte, *vpte; | |
593 | ||
81263dba | 594 | (void) spl6(); |
663dbc72 BJ |
595 | while (bswlist.av_forw == NULL) { |
596 | bswlist.b_flags |= B_WANTED; | |
597 | sleep((caddr_t)&bswlist, PSWP+1); | |
598 | } | |
599 | bp = bswlist.av_forw; | |
600 | bswlist.av_forw = bp->av_forw; | |
81263dba | 601 | (void) spl0(); |
663dbc72 BJ |
602 | |
603 | bp->b_flags = B_BUSY | B_PHYS | rdflg | flag; | |
604 | if ((bp->b_flags & (B_DIRTY|B_PGIN)) == 0) | |
605 | if (rdflg == B_READ) | |
606 | sum.v_pswpin += btoc(nbytes); | |
607 | else | |
608 | sum.v_pswpout += btoc(nbytes); | |
609 | bp->b_proc = p; | |
610 | if (flag & B_DIRTY) { | |
611 | p2dp = ((bp - swbuf) * CLSIZE) * KLMAX; | |
612 | dpte = dptopte(&proc[2], p2dp); | |
613 | vpte = vtopte(p, btop(addr)); | |
614 | for (c = 0; c < nbytes; c += NBPG) { | |
615 | if (vpte->pg_pfnum == 0 || vpte->pg_fod) | |
616 | panic("swap bad pte"); | |
617 | *dpte++ = *vpte++; | |
618 | } | |
619 | bp->b_un.b_addr = (caddr_t)ctob(p2dp); | |
620 | } else | |
621 | bp->b_un.b_addr = addr; | |
622 | while (nbytes > 0) { | |
623 | c = imin(ctob(120), nbytes); | |
624 | bp->b_bcount = c; | |
625 | bp->b_blkno = dblkno; | |
626 | bp->b_dev = dev; | |
d2f87136 BJ |
627 | if (flag & B_DIRTY) { |
628 | swpf[bp - swbuf] = pfcent; | |
629 | swsize[bp - swbuf] = nbytes; | |
630 | } | |
663dbc72 BJ |
631 | (*bdevsw[major(dev)].d_strategy)(bp); |
632 | if (flag & B_DIRTY) { | |
633 | if (c < nbytes) | |
634 | panic("big push"); | |
663dbc72 BJ |
635 | return; |
636 | } | |
81263dba | 637 | (void) spl6(); |
663dbc72 BJ |
638 | while((bp->b_flags&B_DONE)==0) |
639 | sleep((caddr_t)bp, PSWP); | |
81263dba | 640 | (void) spl0(); |
663dbc72 BJ |
641 | bp->b_un.b_addr += c; |
642 | bp->b_flags &= ~B_DONE; | |
643 | if (bp->b_flags & B_ERROR) { | |
644 | if ((flag & (B_UAREA|B_PAGET)) || rdflg == B_WRITE) | |
645 | panic("hard IO err in swap"); | |
646 | swkill(p, (char *)0); | |
647 | } | |
648 | nbytes -= c; | |
649 | dblkno += btoc(c); | |
650 | } | |
81263dba | 651 | (void) spl6(); |
663dbc72 BJ |
652 | bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS|B_PAGET|B_UAREA|B_DIRTY); |
653 | bp->av_forw = bswlist.av_forw; | |
654 | bswlist.av_forw = bp; | |
655 | if (bswlist.b_flags & B_WANTED) { | |
656 | bswlist.b_flags &= ~B_WANTED; | |
657 | wakeup((caddr_t)&bswlist); | |
658 | wakeup((caddr_t)&proc[2]); | |
659 | } | |
81263dba | 660 | (void) spl0(); |
663dbc72 BJ |
661 | } |
662 | ||
663 | /* | |
664 | * If rout == 0 then killed on swap error, else | |
665 | * rout is the name of the routine where we ran out of | |
666 | * swap space. | |
667 | */ | |
668 | swkill(p, rout) | |
669 | struct proc *p; | |
670 | char *rout; | |
671 | { | |
672 | ||
673 | printf("%d: ", p->p_pid); | |
674 | if (rout) | |
675 | printf("out of swap space in %s\n", rout); | |
676 | else | |
677 | printf("killed on swap error\n"); | |
678 | /* | |
679 | * To be sure no looping (e.g. in vmsched trying to | |
680 | * swap out) mark process locked in core (as though | |
681 | * done by user) after killing it so noone will try | |
682 | * to swap it out. | |
683 | */ | |
a30d2e97 | 684 | psignal(p, SIGKILL); |
663dbc72 BJ |
685 | p->p_flag |= SULOCK; |
686 | } | |
687 | ||
688 | /* | |
689 | * make sure all write-behind blocks | |
690 | * on dev (or NODEV for all) | |
691 | * are flushed out. | |
692 | * (from umount and update) | |
693 | */ | |
694 | bflush(dev) | |
695 | dev_t dev; | |
696 | { | |
697 | register struct buf *bp; | |
698 | ||
699 | loop: | |
81263dba | 700 | (void) spl6(); |
663dbc72 BJ |
701 | for (bp = bfreelist.av_forw; bp != &bfreelist; bp = bp->av_forw) { |
702 | if (bp->b_flags&B_DELWRI && (dev == NODEV||dev==bp->b_dev)) { | |
703 | bp->b_flags |= B_ASYNC; | |
704 | notavail(bp); | |
705 | bwrite(bp); | |
706 | goto loop; | |
707 | } | |
708 | } | |
81263dba | 709 | (void) spl0(); |
663dbc72 BJ |
710 | } |
711 | ||
712 | /* | |
713 | * Raw I/O. The arguments are | |
714 | * The strategy routine for the device | |
715 | * A buffer, which will always be a special buffer | |
716 | * header owned exclusively by the device for this purpose | |
717 | * The device number | |
718 | * Read/write flag | |
719 | * Essentially all the work is computing physical addresses and | |
720 | * validating them. | |
721 | * If the user has the proper access privilidges, the process is | |
722 | * marked 'delayed unlock' and the pages involved in the I/O are | |
723 | * faulted and locked. After the completion of the I/O, the above pages | |
724 | * are unlocked. | |
725 | */ | |
726 | physio(strat, bp, dev, rw, mincnt) | |
727 | int (*strat)(); | |
728 | register struct buf *bp; | |
729 | unsigned (*mincnt)(); | |
730 | { | |
731 | register int c; | |
732 | char *a; | |
733 | ||
734 | if (useracc(u.u_base,u.u_count,rw==B_READ?B_WRITE:B_READ) == NULL) { | |
735 | u.u_error = EFAULT; | |
736 | return; | |
737 | } | |
81263dba | 738 | (void) spl6(); |
663dbc72 BJ |
739 | while (bp->b_flags&B_BUSY) { |
740 | bp->b_flags |= B_WANTED; | |
741 | sleep((caddr_t)bp, PRIBIO+1); | |
742 | } | |
743 | bp->b_error = 0; | |
744 | bp->b_proc = u.u_procp; | |
745 | bp->b_un.b_addr = u.u_base; | |
746 | while (u.u_count != 0 && bp->b_error==0) { | |
747 | bp->b_flags = B_BUSY | B_PHYS | rw; | |
748 | bp->b_dev = dev; | |
749 | bp->b_blkno = u.u_offset >> PGSHIFT; | |
750 | bp->b_bcount = u.u_count; | |
751 | (*mincnt)(bp); | |
752 | c = bp->b_bcount; | |
753 | u.u_procp->p_flag |= SPHYSIO; | |
754 | vslock(a = bp->b_un.b_addr, c); | |
755 | (*strat)(bp); | |
81263dba | 756 | (void) spl6(); |
663dbc72 BJ |
757 | while ((bp->b_flags&B_DONE) == 0) |
758 | sleep((caddr_t)bp, PRIBIO); | |
759 | vsunlock(a, c, rw); | |
760 | u.u_procp->p_flag &= ~SPHYSIO; | |
761 | if (bp->b_flags&B_WANTED) | |
762 | wakeup((caddr_t)bp); | |
81263dba | 763 | (void) spl0(); |
663dbc72 BJ |
764 | bp->b_un.b_addr += c; |
765 | u.u_count -= c; | |
766 | u.u_offset += c; | |
767 | } | |
768 | bp->b_flags &= ~(B_BUSY|B_WANTED|B_PHYS); | |
769 | u.u_count = bp->b_resid; | |
770 | geterror(bp); | |
771 | } | |
772 | ||
773 | /*ARGSUSED*/ | |
774 | unsigned | |
775 | minphys(bp) | |
776 | struct buf *bp; | |
777 | { | |
778 | ||
779 | if (bp->b_bcount > 60 * 1024) | |
780 | bp->b_bcount = 60 * 1024; | |
781 | } | |
782 | ||
783 | /* | |
784 | * Pick up the device's error number and pass it to the user; | |
785 | * if there is an error but the number is 0 set a generalized | |
786 | * code. Actually the latter is always true because devices | |
787 | * don't yet return specific errors. | |
788 | */ | |
789 | geterror(bp) | |
790 | register struct buf *bp; | |
791 | { | |
792 | ||
793 | if (bp->b_flags&B_ERROR) | |
794 | if ((u.u_error = bp->b_error)==0) | |
795 | u.u_error = EIO; | |
796 | } |