]>
Commit | Line | Data |
---|---|---|
1227625a SP |
1 | /* |
2 | * Ported to Linux's Second Extended File System as part of the | |
3 | * dump and restore backup suit | |
b45f51d6 | 4 | * Remy Card <card@Linux.EU.Org>, 1994-1997 |
ebcbe7f6 | 5 | * Stelian Pop <pop@cybercable.fr>, 1999-2000 |
1227625a SP |
6 | */ |
7 | ||
8 | /*- | |
9 | * Copyright (c) 1980, 1991, 1993 | |
10 | * The Regents of the University of California. All rights reserved. | |
11 | * | |
12 | * Redistribution and use in source and binary forms, with or without | |
13 | * modification, are permitted provided that the following conditions | |
14 | * are met: | |
15 | * 1. Redistributions of source code must retain the above copyright | |
16 | * notice, this list of conditions and the following disclaimer. | |
17 | * 2. Redistributions in binary form must reproduce the above copyright | |
18 | * notice, this list of conditions and the following disclaimer in the | |
19 | * documentation and/or other materials provided with the distribution. | |
20 | * 3. All advertising materials mentioning features or use of this software | |
21 | * must display the following acknowledgement: | |
22 | * This product includes software developed by the University of | |
23 | * California, Berkeley and its contributors. | |
24 | * 4. Neither the name of the University nor the names of its contributors | |
25 | * may be used to endorse or promote products derived from this software | |
26 | * without specific prior written permission. | |
27 | * | |
28 | * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND | |
29 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE | |
30 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE | |
31 | * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE | |
32 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL | |
33 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS | |
34 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) | |
35 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT | |
36 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY | |
37 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF | |
38 | * SUCH DAMAGE. | |
39 | */ | |
40 | ||
df9ae507 SP |
41 | #ifndef lint |
42 | static const char rcsid[] = | |
92a9bf12 | 43 | "$Id: tape.c,v 1.14 2000/03/01 10:16:05 stelian Exp $"; |
df9ae507 SP |
44 | #endif /* not lint */ |
45 | ||
b45f51d6 SP |
46 | #ifdef __linux__ |
47 | #include <sys/types.h> | |
48 | #include <linux/types.h> | |
49 | #endif | |
1227625a SP |
50 | #include <sys/param.h> |
51 | #include <sys/socket.h> | |
52 | #include <sys/time.h> | |
53 | #include <sys/wait.h> | |
54 | #ifdef __linux__ | |
55 | #include <linux/ext2_fs.h> | |
56 | #include <bsdcompat.h> | |
57 | #else /* __linux__ */ | |
58 | #ifdef sunos | |
59 | #include <sys/vnode.h> | |
60 | ||
61 | #include <ufs/fs.h> | |
62 | #include <ufs/inode.h> | |
63 | #else | |
64 | #include <ufs/ufs/dinode.h> | |
65 | #include <ufs/ffs/fs.h> | |
66 | #endif | |
67 | #endif /* __linux__ */ | |
68 | ||
69 | #include <protocols/dumprestore.h> | |
70 | ||
71 | #include <errno.h> | |
72 | #include <fcntl.h> | |
73 | #include <setjmp.h> | |
74 | #include <signal.h> | |
75 | #include <stdio.h> | |
76 | #ifdef __STDC__ | |
77 | #include <stdlib.h> | |
78 | #include <string.h> | |
79 | #include <unistd.h> | |
80 | #else | |
81 | int write(), read(); | |
82 | #endif | |
83 | ||
c930abff | 84 | #ifdef __linux__ |
1227625a SP |
85 | #include <ext2fs/ext2fs.h> |
86 | #endif | |
c930abff | 87 | |
1227625a | 88 | #include "dump.h" |
1227625a SP |
89 | |
90 | int writesize; /* size of malloc()ed buffer for tape */ | |
91 | long lastspclrec = -1; /* tape block number of last written header */ | |
92 | int trecno = 0; /* next record to write in current block */ | |
93 | extern long blocksperfile; /* number of blocks per output file */ | |
94 | long blocksthisvol; /* number of blocks on current output file */ | |
95 | extern int ntrec; /* blocking factor on tape */ | |
96 | extern int cartridge; | |
97 | extern char *host; | |
98 | char *nexttape; | |
f34aca37 | 99 | extern pid_t rshpid; |
1227625a | 100 | |
ddd2ef55 SP |
101 | static ssize_t atomic_read __P((int, void *, size_t)); |
102 | static ssize_t atomic_write __P((int, const void *, size_t)); | |
1227625a SP |
103 | static void doslave __P((int, int)); |
104 | static void enslave __P((void)); | |
105 | static void flushtape __P((void)); | |
106 | static void killall __P((void)); | |
107 | static void rollforward __P((void)); | |
108 | ||
109 | /* | |
110 | * Concurrent dump mods (Caltech) - disk block reading and tape writing | |
111 | * are exported to several slave processes. While one slave writes the | |
112 | * tape, the others read disk blocks; they pass control of the tape in | |
113 | * a ring via signals. The parent process traverses the filesystem and | |
114 | * sends writeheader()'s and lists of daddr's to the slaves via pipes. | |
115 | * The following structure defines the instruction packets sent to slaves. | |
116 | */ | |
117 | struct req { | |
118 | daddr_t dblk; | |
119 | int count; | |
120 | }; | |
121 | int reqsiz; | |
122 | ||
123 | #define SLAVES 3 /* 1 slave writing, 1 reading, 1 for slack */ | |
124 | struct slave { | |
125 | int tapea; /* header number at start of this chunk */ | |
126 | int count; /* count to next header (used for TS_TAPE */ | |
127 | /* after EOT) */ | |
128 | int inode; /* inode that we are currently dealing with */ | |
129 | int fd; /* FD for this slave */ | |
130 | int pid; /* PID for this slave */ | |
131 | int sent; /* 1 == we've sent this slave requests */ | |
132 | int firstrec; /* record number of this block */ | |
133 | char (*tblock)[TP_BSIZE]; /* buffer for data blocks */ | |
134 | struct req *req; /* buffer for requests */ | |
135 | } slaves[SLAVES+1]; | |
136 | struct slave *slp; | |
137 | ||
138 | char (*nextblock)[TP_BSIZE]; | |
139 | ||
ddd2ef55 SP |
140 | static time_t tstart_volume; /* time of volume start */ |
141 | static int tapea_volume; /* value of spcl.c_tapea at volume start */ | |
142 | ||
1227625a SP |
143 | int master; /* pid of master, for sending error signals */ |
144 | int tenths; /* length of tape used per block written */ | |
145 | static int caught; /* have we caught the signal to proceed? */ | |
146 | static int ready; /* have we reached the lock point without having */ | |
147 | /* received the SIGUSR2 signal from the prev slave? */ | |
ddd2ef55 | 148 | static sigjmp_buf jmpbuf; /* where to jump to if we are ready when the */ |
1227625a SP |
149 | /* SIGUSR2 arrives from the previous slave */ |
150 | ||
151 | int | |
ddd2ef55 | 152 | alloctape(void) |
1227625a SP |
153 | { |
154 | int pgoff = getpagesize() - 1; | |
155 | char *buf; | |
156 | int i; | |
157 | ||
158 | writesize = ntrec * TP_BSIZE; | |
159 | reqsiz = (ntrec + 1) * sizeof(struct req); | |
160 | /* | |
161 | * CDC 92181's and 92185's make 0.8" gaps in 1600-bpi start/stop mode | |
162 | * (see DEC TU80 User's Guide). The shorter gaps of 6250-bpi require | |
163 | * repositioning after stopping, i.e, streaming mode, where the gap is | |
164 | * variable, 0.30" to 0.45". The gap is maximal when the tape stops. | |
165 | */ | |
b45f51d6 | 166 | if (blocksperfile == 0 && !unlimited) |
1227625a SP |
167 | tenths = writesize / density + |
168 | (cartridge ? 16 : density == 625 ? 5 : 8); | |
169 | /* | |
170 | * Allocate tape buffer contiguous with the array of instruction | |
171 | * packets, so flushtape() can write them together with one write(). | |
172 | * Align tape buffer on page boundary to speed up tape write(). | |
173 | */ | |
174 | for (i = 0; i <= SLAVES; i++) { | |
175 | buf = (char *) | |
176 | malloc((unsigned)(reqsiz + writesize + pgoff + TP_BSIZE)); | |
177 | if (buf == NULL) | |
178 | return(0); | |
179 | slaves[i].tblock = (char (*)[TP_BSIZE]) | |
180 | #ifdef __linux__ | |
181 | (((long)&buf[reqsiz] + pgoff) &~ pgoff); | |
182 | #else | |
183 | (((long)&buf[ntrec + 1] + pgoff) &~ pgoff); | |
184 | #endif | |
185 | slaves[i].req = (struct req *)slaves[i].tblock - ntrec - 1; | |
186 | } | |
187 | slp = &slaves[0]; | |
188 | slp->count = 1; | |
189 | slp->tapea = 0; | |
190 | slp->firstrec = 0; | |
191 | nextblock = slp->tblock; | |
192 | return(1); | |
193 | } | |
194 | ||
195 | void | |
ddd2ef55 | 196 | writerec(const void *dp, int isspcl) |
1227625a SP |
197 | { |
198 | ||
199 | slp->req[trecno].dblk = (daddr_t)0; | |
200 | slp->req[trecno].count = 1; | |
ddd2ef55 SP |
201 | /* XXX post increment triggers an egcs-1.1.2-12 bug on alpha/sparc */ |
202 | *(union u_spcl *)(*(nextblock)) = *(union u_spcl *)dp; | |
203 | nextblock++; | |
1227625a SP |
204 | if (isspcl) |
205 | lastspclrec = spcl.c_tapea; | |
206 | trecno++; | |
207 | spcl.c_tapea++; | |
208 | if (trecno >= ntrec) | |
209 | flushtape(); | |
210 | } | |
211 | ||
212 | void | |
ddd2ef55 | 213 | dumpblock(daddr_t blkno, int size) |
1227625a SP |
214 | { |
215 | int avail, tpblks, dblkno; | |
216 | ||
217 | dblkno = fsbtodb(sblock, blkno); | |
218 | tpblks = size >> tp_bshift; | |
219 | while ((avail = MIN(tpblks, ntrec - trecno)) > 0) { | |
220 | slp->req[trecno].dblk = dblkno; | |
221 | slp->req[trecno].count = avail; | |
222 | trecno += avail; | |
223 | spcl.c_tapea += avail; | |
224 | if (trecno >= ntrec) | |
225 | flushtape(); | |
226 | dblkno += avail << (tp_bshift - dev_bshift); | |
227 | tpblks -= avail; | |
228 | } | |
229 | } | |
230 | ||
231 | int nogripe = 0; | |
232 | ||
ddd2ef55 SP |
233 | static void |
234 | tperror(int signo) | |
1227625a SP |
235 | { |
236 | ||
237 | if (pipeout) { | |
238 | msg("write error on %s\n", tape); | |
239 | quit("Cannot recover\n"); | |
240 | /* NOTREACHED */ | |
241 | } | |
242 | msg("write error %d blocks into volume %d\n", blocksthisvol, tapeno); | |
243 | broadcast("DUMP WRITE ERROR!\n"); | |
244 | if (!query("Do you want to restart?")) | |
245 | dumpabort(0); | |
246 | msg("Closing this volume. Prepare to restart with new media;\n"); | |
247 | msg("this dump volume will be rewritten.\n"); | |
248 | killall(); | |
249 | nogripe = 1; | |
250 | close_rewind(); | |
251 | Exit(X_REWRITE); | |
252 | } | |
253 | ||
ddd2ef55 SP |
254 | static void |
255 | sigpipe(int signo) | |
1227625a SP |
256 | { |
257 | ||
258 | quit("Broken pipe\n"); | |
259 | } | |
260 | ||
ddd2ef55 SP |
261 | /* |
262 | * do_stats -- | |
263 | * Update xferrate stats | |
264 | */ | |
265 | time_t | |
266 | do_stats(void) | |
267 | { | |
268 | time_t tnow, ttaken; | |
269 | int blocks; | |
270 | ||
8d4197bb SP |
271 | #ifdef __linux__ |
272 | (void)time4(&tnow); | |
273 | #else | |
ddd2ef55 | 274 | (void)time(&tnow); |
8d4197bb | 275 | #endif |
ddd2ef55 SP |
276 | ttaken = tnow - tstart_volume; |
277 | blocks = spcl.c_tapea - tapea_volume; | |
8d4197bb SP |
278 | msg("Volume %d completed at: %s", tapeno, |
279 | #ifdef __linux__ | |
280 | ctime4(&tnow)); | |
281 | #else | |
282 | ctime(&tnow)); | |
283 | #endif | |
ddd2ef55 SP |
284 | if (ttaken > 0) { |
285 | msg("Volume %d took %d:%02d:%02d\n", tapeno, | |
286 | ttaken / 3600, (ttaken % 3600) / 60, ttaken % 60); | |
287 | msg("Volume %d transfer rate: %ld KB/s\n", tapeno, | |
288 | blocks / ttaken); | |
289 | xferrate += blocks / ttaken; | |
290 | } | |
291 | return(tnow); | |
292 | } | |
293 | ||
294 | #if defined(SIGINFO) | |
295 | /* | |
296 | * statussig -- | |
297 | * information message upon receipt of SIGINFO | |
298 | * (derived from optr.c::timeest()) | |
299 | */ | |
300 | void | |
301 | statussig(int notused) | |
302 | { | |
303 | time_t tnow, deltat; | |
304 | char msgbuf[128]; | |
305 | int save_errno = errno; | |
306 | ||
307 | if (blockswritten < 500) | |
308 | return; | |
8d4197bb SP |
309 | #ifdef __linux__ |
310 | (void) time4(&tnow); | |
311 | #else | |
ddd2ef55 | 312 | (void) time((time_t *) &tnow); |
8d4197bb | 313 | #endif |
8eb05297 SP |
314 | if (blockswritten > tapesize) |
315 | tapesize = blockswritten; | |
ddd2ef55 SP |
316 | deltat = tstart_writing - tnow + (1.0 * (tnow - tstart_writing)) |
317 | / blockswritten * tapesize; | |
318 | (void)snprintf(msgbuf, sizeof(msgbuf), | |
319 | "%3.2f%% done at %ld KB/s, finished in %d:%02d\n", | |
320 | (blockswritten * 100.0) / tapesize, | |
321 | (spcl.c_tapea - tapea_volume) / (tnow - tstart_volume), | |
322 | (int)(deltat / 3600), (int)((deltat % 3600) / 60)); | |
323 | write(STDERR_FILENO, msgbuf, strlen(msgbuf)); | |
324 | errno = save_errno; | |
325 | } | |
326 | #endif | |
327 | ||
1227625a | 328 | static void |
ddd2ef55 | 329 | flushtape(void) |
1227625a SP |
330 | { |
331 | int i, blks, got; | |
332 | long lastfirstrec; | |
333 | ||
334 | int siz = (char *)nextblock - (char *)slp->req; | |
335 | ||
336 | slp->req[trecno].count = 0; /* Sentinel */ | |
337 | ||
ddd2ef55 | 338 | if (atomic_write( slp->fd, (char *)slp->req, siz) != siz) |
1227625a SP |
339 | quit("error writing command pipe: %s\n", strerror(errno)); |
340 | slp->sent = 1; /* we sent a request, read the response later */ | |
341 | ||
342 | lastfirstrec = slp->firstrec; | |
343 | ||
344 | if (++slp >= &slaves[SLAVES]) | |
345 | slp = &slaves[0]; | |
346 | ||
347 | /* Read results back from next slave */ | |
348 | if (slp->sent) { | |
ddd2ef55 | 349 | if (atomic_read( slp->fd, (char *)&got, sizeof got) |
1227625a SP |
350 | != sizeof got) { |
351 | perror(" DUMP: error reading command pipe in master"); | |
352 | dumpabort(0); | |
353 | } | |
354 | slp->sent = 0; | |
355 | ||
356 | /* Check for end of tape */ | |
357 | if (got < writesize) { | |
358 | msg("End of tape detected\n"); | |
359 | ||
360 | /* | |
361 | * Drain the results, don't care what the values were. | |
362 | * If we read them here then trewind won't... | |
363 | */ | |
364 | for (i = 0; i < SLAVES; i++) { | |
365 | if (slaves[i].sent) { | |
ddd2ef55 | 366 | if (atomic_read( slaves[i].fd, |
1227625a SP |
367 | (char *)&got, sizeof got) |
368 | != sizeof got) { | |
369 | perror(" DUMP: error reading command pipe in master"); | |
370 | dumpabort(0); | |
371 | } | |
372 | slaves[i].sent = 0; | |
373 | } | |
374 | } | |
375 | ||
376 | close_rewind(); | |
377 | rollforward(); | |
378 | return; | |
379 | } | |
380 | } | |
381 | ||
382 | blks = 0; | |
383 | if (spcl.c_type != TS_END) { | |
384 | for (i = 0; i < spcl.c_count; i++) | |
385 | if (spcl.c_addr[i] != 0) | |
386 | blks++; | |
387 | } | |
388 | slp->count = lastspclrec + blks + 1 - spcl.c_tapea; | |
389 | slp->tapea = spcl.c_tapea; | |
390 | slp->firstrec = lastfirstrec + ntrec; | |
391 | slp->inode = curino; | |
392 | nextblock = slp->tblock; | |
393 | trecno = 0; | |
394 | asize += tenths; | |
395 | blockswritten += ntrec; | |
396 | blocksthisvol += ntrec; | |
b45f51d6 | 397 | if (!pipeout && !unlimited && (blocksperfile ? |
1227625a SP |
398 | (blocksthisvol >= blocksperfile) : (asize > tsize))) { |
399 | close_rewind(); | |
400 | startnewtape(0); | |
401 | } | |
402 | timeest(); | |
403 | } | |
404 | ||
405 | void | |
ddd2ef55 | 406 | trewind(void) |
1227625a SP |
407 | { |
408 | int f; | |
409 | int got; | |
410 | ||
411 | for (f = 0; f < SLAVES; f++) { | |
412 | /* | |
b45f51d6 SP |
413 | * Drain the results, but unlike EOT we DO (or should) care |
414 | * what the return values were, since if we detect EOT after | |
415 | * we think we've written the last blocks to the tape anyway, | |
1227625a SP |
416 | * we have to replay those blocks with rollforward. |
417 | * | |
b45f51d6 | 418 | * fixme: punt for now. |
1227625a SP |
419 | */ |
420 | if (slaves[f].sent) { | |
ddd2ef55 | 421 | if (atomic_read( slaves[f].fd, (char *)&got, sizeof got) |
1227625a SP |
422 | != sizeof got) { |
423 | perror(" DUMP: error reading command pipe in master"); | |
424 | dumpabort(0); | |
425 | } | |
426 | slaves[f].sent = 0; | |
427 | if (got != writesize) { | |
428 | msg("EOT detected in last 2 tape records!\n"); | |
429 | msg("Use a longer tape, decrease the size estimate\n"); | |
430 | quit("or use no size estimate at all.\n"); | |
431 | } | |
432 | } | |
433 | (void) close(slaves[f].fd); | |
434 | } | |
435 | while (wait((int *)NULL) >= 0) /* wait for any signals from slaves */ | |
436 | /* void */; | |
437 | ||
438 | if (pipeout) | |
439 | return; | |
440 | ||
441 | msg("Closing %s\n", tape); | |
442 | ||
443 | #ifdef RDUMP | |
444 | if (host) { | |
445 | rmtclose(); | |
446 | while (rmtopen(tape, 0) < 0) | |
447 | sleep(10); | |
448 | rmtclose(); | |
449 | return; | |
450 | } | |
451 | #endif | |
452 | (void) close(tapefd); | |
453 | while ((f = open(tape, 0)) < 0) | |
454 | sleep (10); | |
455 | (void) close(f); | |
456 | } | |
457 | ||
458 | void | |
ddd2ef55 | 459 | close_rewind(void) |
1227625a SP |
460 | { |
461 | trewind(); | |
ddd2ef55 | 462 | (void)do_stats(); |
d1c73b9a | 463 | if (nexttape || Mflag) |
1227625a SP |
464 | return; |
465 | if (!nogripe) { | |
466 | msg("Change Volumes: Mount volume #%d\n", tapeno+1); | |
467 | broadcast("CHANGE DUMP VOLUMES!\7\7\n"); | |
468 | } | |
469 | while (!query("Is the new volume mounted and ready to go?")) | |
470 | if (query("Do you want to abort?")) { | |
471 | dumpabort(0); | |
472 | /*NOTREACHED*/ | |
473 | } | |
474 | } | |
475 | ||
476 | void | |
ddd2ef55 | 477 | rollforward(void) |
1227625a SP |
478 | { |
479 | register struct req *p, *q, *prev; | |
480 | register struct slave *tslp; | |
481 | int i, size, savedtapea, got; | |
482 | union u_spcl *ntb, *otb; | |
b45f51d6 SP |
483 | #ifdef __linux__ |
484 | int blks; | |
485 | long lastfirstrec; | |
486 | #endif | |
1227625a SP |
487 | tslp = &slaves[SLAVES]; |
488 | ntb = (union u_spcl *)tslp->tblock[1]; | |
489 | ||
490 | /* | |
b45f51d6 SP |
491 | * Each of the N slaves should have requests that need to |
492 | * be replayed on the next tape. Use the extra slave buffers | |
493 | * (slaves[SLAVES]) to construct request lists to be sent to | |
1227625a SP |
494 | * each slave in turn. |
495 | */ | |
496 | for (i = 0; i < SLAVES; i++) { | |
497 | q = &tslp->req[1]; | |
498 | otb = (union u_spcl *)slp->tblock; | |
499 | ||
500 | /* | |
b45f51d6 | 501 | * For each request in the current slave, copy it to tslp. |
1227625a SP |
502 | */ |
503 | ||
504 | prev = NULL; | |
505 | for (p = slp->req; p->count > 0; p += p->count) { | |
506 | *q = *p; | |
507 | if (p->dblk == 0) | |
508 | *ntb++ = *otb++; /* copy the datablock also */ | |
509 | prev = q; | |
510 | q += q->count; | |
511 | } | |
512 | if (prev == NULL) | |
513 | quit("rollforward: protocol botch"); | |
514 | if (prev->dblk != 0) | |
515 | prev->count -= 1; | |
516 | else | |
517 | ntb--; | |
518 | q -= 1; | |
519 | q->count = 0; | |
520 | q = &tslp->req[0]; | |
521 | if (i == 0) { | |
522 | q->dblk = 0; | |
523 | q->count = 1; | |
524 | trecno = 0; | |
525 | nextblock = tslp->tblock; | |
526 | savedtapea = spcl.c_tapea; | |
527 | spcl.c_tapea = slp->tapea; | |
528 | startnewtape(0); | |
529 | spcl.c_tapea = savedtapea; | |
530 | lastspclrec = savedtapea - 1; | |
531 | } | |
532 | size = (char *)ntb - (char *)q; | |
ddd2ef55 | 533 | if (atomic_write( slp->fd, (char *)q, size) != size) { |
1227625a SP |
534 | perror(" DUMP: error writing command pipe"); |
535 | dumpabort(0); | |
536 | } | |
537 | slp->sent = 1; | |
b45f51d6 SP |
538 | #ifdef __linux__ |
539 | lastfirstrec = slp->firstrec; | |
540 | #endif | |
1227625a SP |
541 | if (++slp >= &slaves[SLAVES]) |
542 | slp = &slaves[0]; | |
543 | ||
544 | q->count = 1; | |
545 | ||
546 | if (prev->dblk != 0) { | |
547 | /* | |
b45f51d6 SP |
548 | * If the last one was a disk block, make the |
549 | * first of this one be the last bit of that disk | |
1227625a SP |
550 | * block... |
551 | */ | |
552 | q->dblk = prev->dblk + | |
553 | prev->count * (TP_BSIZE / DEV_BSIZE); | |
554 | ntb = (union u_spcl *)tslp->tblock; | |
555 | } else { | |
556 | /* | |
b45f51d6 | 557 | * It wasn't a disk block. Copy the data to its |
1227625a SP |
558 | * new location in the buffer. |
559 | */ | |
560 | q->dblk = 0; | |
561 | *((union u_spcl *)tslp->tblock) = *ntb; | |
562 | ntb = (union u_spcl *)tslp->tblock[1]; | |
563 | } | |
564 | } | |
565 | slp->req[0] = *q; | |
566 | nextblock = slp->tblock; | |
567 | if (q->dblk == 0) | |
b45f51d6 SP |
568 | #ifdef __linux__ |
569 | *(union u_spcl *)(*(nextblock)++) = *(union u_spcl *)tslp->tblock; | |
570 | #else | |
1227625a | 571 | nextblock++; |
b45f51d6 | 572 | #endif |
1227625a SP |
573 | trecno = 1; |
574 | ||
575 | /* | |
576 | * Clear the first slaves' response. One hopes that it | |
577 | * worked ok, otherwise the tape is much too short! | |
578 | */ | |
579 | if (slp->sent) { | |
ddd2ef55 | 580 | if (atomic_read( slp->fd, (char *)&got, sizeof got) |
1227625a SP |
581 | != sizeof got) { |
582 | perror(" DUMP: error reading command pipe in master"); | |
583 | dumpabort(0); | |
584 | } | |
585 | slp->sent = 0; | |
586 | ||
587 | if (got != writesize) { | |
588 | quit("EOT detected at start of the tape!\n"); | |
589 | } | |
590 | } | |
b45f51d6 SP |
591 | |
592 | #ifdef __linux__ | |
593 | blks = 0; | |
594 | if (spcl.c_type != TS_END) { | |
595 | for (i = 0; i < spcl.c_count; i++) | |
596 | if (spcl.c_addr[i] != 0) | |
597 | blks++; | |
598 | } | |
599 | ||
600 | slp->firstrec = lastfirstrec + ntrec; | |
601 | slp->count = lastspclrec + blks + 1 - spcl.c_tapea; | |
602 | slp->inode = curino; | |
603 | asize += tenths; | |
604 | blockswritten += ntrec; | |
605 | blocksthisvol += ntrec; | |
606 | #endif | |
1227625a SP |
607 | } |
608 | ||
609 | /* | |
610 | * We implement taking and restoring checkpoints on the tape level. | |
611 | * When each tape is opened, a new process is created by forking; this | |
612 | * saves all of the necessary context in the parent. The child | |
613 | * continues the dump; the parent waits around, saving the context. | |
614 | * If the child returns X_REWRITE, then it had problems writing that tape; | |
615 | * this causes the parent to fork again, duplicating the context, and | |
616 | * everything continues as if nothing had happened. | |
617 | */ | |
618 | void | |
ddd2ef55 | 619 | startnewtape(int top) |
1227625a SP |
620 | { |
621 | int parentpid; | |
622 | int childpid; | |
623 | int status; | |
624 | int waitpid; | |
625 | char *p; | |
626 | #ifdef __linux__ | |
ddd2ef55 | 627 | void (*interrupt_save) __P((int signo)); |
1227625a SP |
628 | #else /* __linux__ */ |
629 | #ifdef sunos | |
630 | void (*interrupt_save)(); | |
631 | #else | |
632 | sig_t interrupt_save; | |
633 | #endif | |
634 | #endif /* __linux__ */ | |
635 | ||
636 | interrupt_save = signal(SIGINT, SIG_IGN); | |
637 | parentpid = getpid(); | |
ddd2ef55 | 638 | tapea_volume = spcl.c_tapea; |
8d4197bb SP |
639 | #ifdef __linux__ |
640 | (void)time4(&tstart_volume); | |
641 | #else | |
642 | (void)time((&tstart_volume); | |
643 | #endif | |
1227625a SP |
644 | |
645 | restore_check_point: | |
646 | (void)signal(SIGINT, interrupt_save); | |
647 | /* | |
648 | * All signals are inherited... | |
649 | */ | |
650 | childpid = fork(); | |
651 | if (childpid < 0) { | |
652 | msg("Context save fork fails in parent %d\n", parentpid); | |
653 | Exit(X_ABORT); | |
654 | } | |
655 | if (childpid != 0) { | |
656 | /* | |
657 | * PARENT: | |
658 | * save the context by waiting | |
659 | * until the child doing all of the work returns. | |
660 | * don't catch the interrupt | |
661 | */ | |
662 | signal(SIGINT, SIG_IGN); | |
663 | #ifdef TDEBUG | |
664 | msg("Tape: %d; parent process: %d child process %d\n", | |
665 | tapeno+1, parentpid, childpid); | |
666 | #endif /* TDEBUG */ | |
667 | while ((waitpid = wait(&status)) != childpid) | |
f34aca37 SP |
668 | if (waitpid != rshpid) |
669 | msg("Parent %d waiting for child %d has another child %d return\n", | |
1227625a SP |
670 | parentpid, childpid, waitpid); |
671 | if (status & 0xFF) { | |
672 | msg("Child %d returns LOB status %o\n", | |
673 | childpid, status&0xFF); | |
674 | } | |
675 | status = (status >> 8) & 0xFF; | |
676 | #ifdef TDEBUG | |
677 | switch(status) { | |
678 | case X_FINOK: | |
679 | msg("Child %d finishes X_FINOK\n", childpid); | |
680 | break; | |
b45f51d6 | 681 | case X_ABORT: |
1227625a SP |
682 | msg("Child %d finishes X_ABORT\n", childpid); |
683 | break; | |
684 | case X_REWRITE: | |
685 | msg("Child %d finishes X_REWRITE\n", childpid); | |
686 | break; | |
687 | default: | |
688 | msg("Child %d finishes unknown %d\n", | |
689 | childpid, status); | |
690 | break; | |
691 | } | |
692 | #endif /* TDEBUG */ | |
693 | switch(status) { | |
694 | case X_FINOK: | |
695 | Exit(X_FINOK); | |
696 | case X_ABORT: | |
697 | Exit(X_ABORT); | |
698 | case X_REWRITE: | |
699 | goto restore_check_point; | |
700 | default: | |
701 | msg("Bad return code from dump: %d\n", status); | |
702 | Exit(X_ABORT); | |
703 | } | |
704 | /*NOTREACHED*/ | |
705 | } else { /* we are the child; just continue */ | |
706 | #ifdef TDEBUG | |
707 | sleep(4); /* allow time for parent's message to get out */ | |
708 | msg("Child on Tape %d has parent %d, my pid = %d\n", | |
709 | tapeno+1, parentpid, getpid()); | |
710 | #endif /* TDEBUG */ | |
711 | /* | |
712 | * If we have a name like "/dev/rmt0,/dev/rmt1", | |
713 | * use the name before the comma first, and save | |
714 | * the remaining names for subsequent volumes. | |
715 | */ | |
716 | tapeno++; /* current tape sequence */ | |
d1c73b9a | 717 | if (Mflag) { |
92a9bf12 SP |
718 | snprintf(tape, MAXPATHLEN, "%s%03d", tapeprefix, tapeno); |
719 | tape[MAXPATHLEN - 1] = '\0'; | |
d1c73b9a SP |
720 | msg("Dumping volume %d on %s\n", tapeno, tape); |
721 | } | |
722 | else if (nexttape || strchr(tapeprefix, ',')) { | |
1227625a | 723 | if (nexttape && *nexttape) |
d1c73b9a SP |
724 | tapeprefix = nexttape; |
725 | if ((p = strchr(tapeprefix, ',')) != NULL) { | |
1227625a SP |
726 | *p = '\0'; |
727 | nexttape = p + 1; | |
728 | } else | |
729 | nexttape = NULL; | |
92a9bf12 SP |
730 | strncpy(tape, tapeprefix, MAXPATHLEN); |
731 | tape[MAXPATHLEN - 1] = '\0'; | |
1227625a SP |
732 | msg("Dumping volume %d on %s\n", tapeno, tape); |
733 | } | |
734 | #ifdef RDUMP | |
d3393043 SP |
735 | while ((tapefd = (host ? rmtopen(tape, 2) : pipeout ? |
736 | fileno(stdout) : | |
737 | open(tape, O_WRONLY|O_CREAT, 0666))) < 0) | |
1227625a | 738 | #else |
d3393043 | 739 | while ((tapefd = (pipeout ? fileno(stdout) : |
1227625a SP |
740 | open(tape, O_WRONLY|O_CREAT, 0666))) < 0) |
741 | #endif | |
742 | { | |
743 | msg("Cannot open output \"%s\".\n", tape); | |
744 | if (!query("Do you want to retry the open?")) | |
745 | dumpabort(0); | |
746 | } | |
747 | ||
748 | enslave(); /* Share open tape file descriptor with slaves */ | |
749 | ||
750 | asize = 0; | |
751 | blocksthisvol = 0; | |
752 | if (top) | |
753 | newtape++; /* new tape signal */ | |
b45f51d6 | 754 | spcl.c_count = slp->count; |
1227625a SP |
755 | /* |
756 | * measure firstrec in TP_BSIZE units since restore doesn't | |
757 | * know the correct ntrec value... | |
758 | */ | |
759 | spcl.c_firstrec = slp->firstrec; | |
760 | spcl.c_volume++; | |
761 | spcl.c_type = TS_TAPE; | |
762 | spcl.c_flags |= DR_NEWHEADER; | |
763 | writeheader((ino_t)slp->inode); | |
764 | spcl.c_flags &=~ DR_NEWHEADER; | |
8d4197bb SP |
765 | msg("Volume %d started at: %s", tapeno, |
766 | #ifdef __linux__ | |
767 | ctime4(&tstart_volume)); | |
768 | #else | |
769 | ctime(&tstart_volume)); | |
770 | #endif | |
1227625a SP |
771 | if (tapeno > 1) |
772 | msg("Volume %d begins with blocks from inode %d\n", | |
773 | tapeno, slp->inode); | |
774 | } | |
775 | } | |
776 | ||
777 | void | |
ddd2ef55 | 778 | dumpabort(int signo) |
1227625a SP |
779 | { |
780 | ||
781 | if (master != 0 && master != getpid()) | |
782 | /* Signals master to call dumpabort */ | |
783 | (void) kill(master, SIGTERM); | |
784 | else { | |
785 | killall(); | |
786 | msg("The ENTIRE dump is aborted.\n"); | |
787 | } | |
788 | #ifdef RDUMP | |
789 | rmtclose(); | |
790 | #endif | |
791 | Exit(X_ABORT); | |
792 | } | |
793 | ||
b45f51d6 | 794 | void |
ddd2ef55 | 795 | Exit(int status) |
1227625a SP |
796 | { |
797 | ||
798 | #ifdef TDEBUG | |
799 | msg("pid = %d exits with status %d\n", getpid(), status); | |
800 | #endif /* TDEBUG */ | |
801 | exit(status); | |
802 | } | |
803 | ||
804 | /* | |
805 | * proceed - handler for SIGUSR2, used to synchronize IO between the slaves. | |
806 | */ | |
ddd2ef55 SP |
807 | static void |
808 | proceed(int signo) | |
1227625a SP |
809 | { |
810 | ||
811 | if (ready) | |
ddd2ef55 | 812 | siglongjmp(jmpbuf, 1); |
1227625a SP |
813 | caught++; |
814 | } | |
815 | ||
816 | void | |
ddd2ef55 | 817 | enslave(void) |
1227625a SP |
818 | { |
819 | int cmd[2]; | |
820 | #ifdef LINUX_FORK_BUG | |
821 | int i, j; | |
822 | #else | |
823 | register int i, j; | |
824 | #endif | |
825 | ||
826 | master = getpid(); | |
827 | ||
828 | signal(SIGTERM, dumpabort); /* Slave sends SIGTERM on dumpabort() */ | |
829 | signal(SIGPIPE, sigpipe); | |
830 | signal(SIGUSR1, tperror); /* Slave sends SIGUSR1 on tape errors */ | |
831 | signal(SIGUSR2, proceed); /* Slave sends SIGUSR2 to next slave */ | |
832 | ||
833 | for (i = 0; i < SLAVES; i++) { | |
834 | if (i == slp - &slaves[0]) { | |
835 | caught = 1; | |
836 | } else { | |
837 | caught = 0; | |
838 | } | |
839 | ||
840 | if (socketpair(AF_UNIX, SOCK_STREAM, 0, cmd) < 0 || | |
841 | (slaves[i].pid = fork()) < 0) | |
842 | quit("too many slaves, %d (recompile smaller): %s\n", | |
843 | i, strerror(errno)); | |
844 | ||
845 | slaves[i].fd = cmd[1]; | |
846 | slaves[i].sent = 0; | |
847 | if (slaves[i].pid == 0) { /* Slave starts up here */ | |
848 | for (j = 0; j <= i; j++) | |
849 | (void) close(slaves[j].fd); | |
850 | signal(SIGINT, SIG_IGN); /* Master handles this */ | |
ddd2ef55 SP |
851 | #if defined(SIGINFO) |
852 | signal(SIGINFO, SIG_IGN); | |
853 | #endif | |
854 | ||
1227625a | 855 | #ifdef LINUX_FORK_BUG |
ddd2ef55 | 856 | if (atomic_write( cmd[0], (char *) &i, sizeof i) |
1227625a SP |
857 | != sizeof i) |
858 | quit("master/slave protocol botched 3\n"); | |
859 | #endif | |
860 | doslave(cmd[0], i); | |
861 | Exit(X_FINOK); | |
862 | } | |
863 | } | |
864 | ||
865 | #ifdef LINUX_FORK_BUG | |
866 | /* | |
867 | * Wait for all slaves to _actually_ start to circumvent a bug in | |
868 | * Linux kernels >= 2.1.3 where a signal sent to a child that hasn't | |
869 | * returned from fork() causes a SEGV in the child process | |
870 | */ | |
871 | for (i = 0; i < SLAVES; i++) | |
ddd2ef55 | 872 | if (atomic_read( slaves[i].fd, (char *) &j, sizeof j) != sizeof j) |
1227625a SP |
873 | quit("master/slave protocol botched 4\n"); |
874 | #endif | |
875 | ||
1227625a | 876 | for (i = 0; i < SLAVES; i++) |
ddd2ef55 | 877 | (void) atomic_write( slaves[i].fd, |
1227625a SP |
878 | (char *) &slaves[(i + 1) % SLAVES].pid, |
879 | sizeof slaves[0].pid); | |
880 | ||
881 | master = 0; | |
882 | } | |
883 | ||
884 | void | |
ddd2ef55 | 885 | killall(void) |
1227625a SP |
886 | { |
887 | register int i; | |
888 | ||
889 | for (i = 0; i < SLAVES; i++) | |
b45f51d6 | 890 | if (slaves[i].pid > 0) { |
1227625a | 891 | (void) kill(slaves[i].pid, SIGKILL); |
b45f51d6 SP |
892 | slaves[i].sent = 0; |
893 | } | |
1227625a SP |
894 | } |
895 | ||
896 | /* | |
897 | * Synchronization - each process has a lockfile, and shares file | |
898 | * descriptors to the following process's lockfile. When our write | |
899 | * completes, we release our lock on the following process's lock- | |
900 | * file, allowing the following process to lock it and proceed. We | |
901 | * get the lock back for the next cycle by swapping descriptors. | |
902 | */ | |
903 | static void | |
ddd2ef55 | 904 | doslave(int cmd, int slave_number) |
1227625a SP |
905 | { |
906 | register int nread; | |
ddd2ef55 SP |
907 | int nextslave, size, eot_count; |
908 | volatile int wrote = 0; | |
909 | sigset_t sigset; | |
1227625a SP |
910 | #ifdef __linux__ |
911 | errcode_t retval; | |
912 | #endif | |
913 | ||
1227625a SP |
914 | /* |
915 | * Need our own seek pointer. | |
916 | */ | |
917 | (void) close(diskfd); | |
918 | if ((diskfd = open(disk, O_RDONLY)) < 0) | |
919 | quit("slave couldn't reopen disk: %s\n", strerror(errno)); | |
920 | #ifdef __linux__ | |
921 | ext2fs_close(fs); | |
c930abff | 922 | retval = dump_fs_open(disk, &fs); |
1227625a | 923 | if (retval) |
c930abff | 924 | quit("slave couldn't reopen disk: %s\n", error_message(retval)); |
1227625a SP |
925 | #endif /* __linux__ */ |
926 | ||
927 | /* | |
928 | * Need the pid of the next slave in the loop... | |
929 | */ | |
ddd2ef55 | 930 | if ((nread = atomic_read( cmd, (char *)&nextslave, sizeof nextslave)) |
1227625a SP |
931 | != sizeof nextslave) { |
932 | quit("master/slave protocol botched - didn't get pid of next slave.\n"); | |
933 | } | |
934 | ||
935 | /* | |
936 | * Get list of blocks to dump, read the blocks into tape buffer | |
937 | */ | |
ddd2ef55 | 938 | while ((nread = atomic_read( cmd, (char *)slp->req, reqsiz)) == reqsiz) { |
1227625a SP |
939 | register struct req *p = slp->req; |
940 | ||
941 | for (trecno = 0; trecno < ntrec; | |
942 | trecno += p->count, p += p->count) { | |
943 | if (p->dblk) { | |
944 | bread(p->dblk, slp->tblock[trecno], | |
945 | p->count * TP_BSIZE); | |
946 | } else { | |
ddd2ef55 | 947 | if (p->count != 1 || atomic_read( cmd, |
b45f51d6 | 948 | (char *)slp->tblock[trecno], |
1227625a SP |
949 | TP_BSIZE) != TP_BSIZE) |
950 | quit("master/slave protocol botched.\n"); | |
951 | } | |
952 | } | |
953 | if (setjmp(jmpbuf) == 0) { | |
954 | ready = 1; | |
955 | if (!caught) | |
956 | (void) pause(); | |
957 | } | |
958 | ready = 0; | |
959 | caught = 0; | |
960 | ||
961 | /* Try to write the data... */ | |
b45f51d6 | 962 | wrote = 0; |
1227625a SP |
963 | eot_count = 0; |
964 | size = 0; | |
965 | ||
966 | while (eot_count < 10 && size < writesize) { | |
967 | #ifdef RDUMP | |
968 | if (host) | |
969 | wrote = rmtwrite(slp->tblock[0]+size, | |
970 | writesize-size); | |
971 | else | |
972 | #endif | |
973 | wrote = write(tapefd, slp->tblock[0]+size, | |
974 | writesize-size); | |
975 | #ifdef WRITEDEBUG | |
b45f51d6 | 976 | printf("slave %d wrote %d\n", slave_number, wrote); |
1227625a | 977 | #endif |
b45f51d6 | 978 | if (wrote < 0) |
1227625a SP |
979 | break; |
980 | if (wrote == 0) | |
981 | eot_count++; | |
982 | size += wrote; | |
983 | } | |
984 | ||
985 | #ifdef WRITEDEBUG | |
b45f51d6 SP |
986 | if (size != writesize) |
987 | printf("slave %d only wrote %d out of %d bytes and gave up.\n", | |
1227625a SP |
988 | slave_number, size, writesize); |
989 | #endif | |
990 | ||
b45f51d6 SP |
991 | /* |
992 | * Handle ENOSPC as an EOT condition. | |
993 | */ | |
994 | if (wrote < 0 && errno == ENOSPC) { | |
995 | wrote = 0; | |
996 | eot_count++; | |
997 | } | |
998 | ||
1227625a SP |
999 | if (eot_count > 0) |
1000 | size = 0; | |
1001 | ||
b45f51d6 | 1002 | if (wrote < 0) { |
1227625a | 1003 | (void) kill(master, SIGUSR1); |
ddd2ef55 | 1004 | sigemptyset(&sigset); |
1227625a | 1005 | for (;;) |
ddd2ef55 | 1006 | sigsuspend(&sigset); |
1227625a SP |
1007 | } else { |
1008 | /* | |
1009 | * pass size of write back to master | |
1010 | * (for EOT handling) | |
1011 | */ | |
ddd2ef55 | 1012 | (void) atomic_write( cmd, (char *)&size, sizeof size); |
b45f51d6 | 1013 | } |
1227625a SP |
1014 | |
1015 | /* | |
1016 | * If partial write, don't want next slave to go. | |
1017 | * Also jolts him awake. | |
1018 | */ | |
1019 | (void) kill(nextslave, SIGUSR2); | |
1020 | } | |
1021 | if (nread != 0) | |
1022 | quit("error reading command pipe: %s\n", strerror(errno)); | |
1023 | } | |
1024 | ||
1025 | /* | |
1026 | * Since a read from a pipe may not return all we asked for, | |
1027 | * or a write may not write all we ask if we get a signal, | |
1028 | * loop until the count is satisfied (or error). | |
1029 | */ | |
ddd2ef55 SP |
1030 | static ssize_t |
1031 | atomic_read(int fd, void *buf, size_t count) | |
1032 | { | |
1033 | int got, need = count; | |
1034 | ||
2e682895 SP |
1035 | do { |
1036 | while ((got = read(fd, buf, need)) > 0 && (need -= got) > 0) | |
1037 | (char *)buf += got; | |
1038 | } while (got == -1 && errno == EINTR); | |
ddd2ef55 SP |
1039 | return (got < 0 ? got : count - need); |
1040 | } | |
1041 | ||
1042 | /* | |
1043 | * Since a read from a pipe may not return all we asked for, | |
1044 | * or a write may not write all we ask if we get a signal, | |
1045 | * loop until the count is satisfied (or error). | |
1046 | */ | |
1047 | static ssize_t | |
1048 | atomic_write(int fd, const void *buf, size_t count) | |
1227625a SP |
1049 | { |
1050 | int got, need = count; | |
1051 | ||
2e682895 SP |
1052 | do { |
1053 | while ((got = write(fd, buf, need)) > 0 && (need -= got) > 0) | |
1054 | (char *)buf += got; | |
1055 | } while (got == -1 && errno == EINTR); | |
1227625a SP |
1056 | return (got < 0 ? got : count - need); |
1057 | } |