]>
Commit | Line | Data |
---|---|---|
e86c9dd6 NB |
1 | /* |
2 | * mdadm - manage Linux "md" devices aka RAID arrays. | |
3 | * | |
e736b623 | 4 | * Copyright (C) 2006-2009 Neil Brown <neilb@suse.de> |
e86c9dd6 NB |
5 | * |
6 | * | |
7 | * This program is free software; you can redistribute it and/or modify | |
8 | * it under the terms of the GNU General Public License as published by | |
9 | * the Free Software Foundation; either version 2 of the License, or | |
10 | * (at your option) any later version. | |
11 | * | |
12 | * This program is distributed in the hope that it will be useful, | |
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
15 | * GNU General Public License for more details. | |
16 | * | |
17 | * You should have received a copy of the GNU General Public License | |
18 | * along with this program; if not, write to the Free Software | |
19 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
20 | * | |
21 | * Author: Neil Brown | |
22 | * Email: <neilb@suse.de> | |
23 | */ | |
24 | ||
25 | #include "mdadm.h" | |
a6288483 | 26 | #include <stdint.h> |
e86c9dd6 NB |
27 | |
28 | /* To restripe, we read from old geometry to a buffer, and | |
29 | * read from buffer to new geometry. | |
a6288483 N |
30 | * When reading, we might have missing devices and so could need |
31 | * to reconstruct. | |
32 | * When writing, we need to create correct parity and Q. | |
e86c9dd6 NB |
33 | * |
34 | */ | |
35 | ||
979afcb8 | 36 | int geo_map(int block, unsigned long long stripe, int raid_disks, |
e0d95aac | 37 | int level, int layout) |
e86c9dd6 | 38 | { |
48327135 | 39 | /* On the given stripe, find which disk in the array will have |
e86c9dd6 | 40 | * block numbered 'block'. |
48327135 NB |
41 | * '-1' means the parity block. |
42 | * '-2' means the Q syndrome. | |
e86c9dd6 NB |
43 | */ |
44 | int pd; | |
45 | ||
b6e317c8 AK |
46 | /* layout is not relevant for raid0 and raid4 */ |
47 | if ((level == 0) || | |
48 | (level == 4)) | |
49 | layout = 0; | |
50 | ||
e86c9dd6 NB |
51 | switch(level*100 + layout) { |
52 | case 000: | |
53 | case 400: | |
e0d95aac | 54 | case 500 + ALGORITHM_PARITY_N: |
e86c9dd6 NB |
55 | /* raid 4 isn't messed around by parity blocks */ |
56 | if (block == -1) | |
57 | return raid_disks-1; /* parity block */ | |
58 | return block; | |
59 | case 500 + ALGORITHM_LEFT_ASYMMETRIC: | |
60 | pd = (raid_disks-1) - stripe % raid_disks; | |
61 | if (block == -1) return pd; | |
62 | if (block >= pd) | |
63 | block++; | |
64 | return block; | |
65 | ||
66 | case 500 + ALGORITHM_RIGHT_ASYMMETRIC: | |
67 | pd = stripe % raid_disks; | |
68 | if (block == -1) return pd; | |
69 | if (block >= pd) | |
70 | block++; | |
71 | return block; | |
72 | ||
73 | case 500 + ALGORITHM_LEFT_SYMMETRIC: | |
74 | pd = (raid_disks - 1) - stripe % raid_disks; | |
75 | if (block == -1) return pd; | |
76 | return (pd + 1 + block) % raid_disks; | |
77 | ||
78 | case 500 + ALGORITHM_RIGHT_SYMMETRIC: | |
79 | pd = stripe % raid_disks; | |
80 | if (block == -1) return pd; | |
81 | return (pd + 1 + block) % raid_disks; | |
82 | ||
e0d95aac N |
83 | case 500 + ALGORITHM_PARITY_0: |
84 | return block + 1; | |
85 | ||
86 | ||
87 | case 600 + ALGORITHM_PARITY_N_6: | |
88 | if (block == -2) | |
89 | return raid_disks - 1; | |
90 | if (block == -1) | |
91 | return raid_disks - 2; /* parity block */ | |
92 | return block; | |
93 | case 600 + ALGORITHM_LEFT_ASYMMETRIC_6: | |
94 | if (block == -2) | |
95 | return raid_disks - 1; | |
96 | raid_disks--; | |
97 | pd = (raid_disks-1) - stripe % raid_disks; | |
98 | if (block == -1) return pd; | |
99 | if (block >= pd) | |
100 | block++; | |
101 | return block; | |
102 | ||
103 | case 600 + ALGORITHM_RIGHT_ASYMMETRIC_6: | |
104 | if (block == -2) | |
105 | return raid_disks - 1; | |
106 | raid_disks--; | |
107 | pd = stripe % raid_disks; | |
108 | if (block == -1) return pd; | |
109 | if (block >= pd) | |
110 | block++; | |
111 | return block; | |
112 | ||
113 | case 600 + ALGORITHM_LEFT_SYMMETRIC_6: | |
114 | if (block == -2) | |
115 | return raid_disks - 1; | |
116 | raid_disks--; | |
117 | pd = (raid_disks - 1) - stripe % raid_disks; | |
118 | if (block == -1) return pd; | |
119 | return (pd + 1 + block) % raid_disks; | |
120 | ||
121 | case 600 + ALGORITHM_RIGHT_SYMMETRIC_6: | |
122 | if (block == -2) | |
123 | return raid_disks - 1; | |
124 | raid_disks--; | |
125 | pd = stripe % raid_disks; | |
126 | if (block == -1) return pd; | |
127 | return (pd + 1 + block) % raid_disks; | |
128 | ||
129 | case 600 + ALGORITHM_PARITY_0_6: | |
130 | if (block == -2) | |
131 | return raid_disks - 1; | |
132 | return block + 1; | |
133 | ||
134 | ||
135 | case 600 + ALGORITHM_PARITY_0: | |
136 | if (block == -1) | |
137 | return 0; | |
138 | if (block == -2) | |
139 | return 1; | |
140 | return block + 2; | |
141 | ||
e86c9dd6 NB |
142 | case 600 + ALGORITHM_LEFT_ASYMMETRIC: |
143 | pd = raid_disks - 1 - (stripe % raid_disks); | |
144 | if (block == -1) return pd; | |
48327135 | 145 | if (block == -2) return (pd+1) % raid_disks; |
e86c9dd6 NB |
146 | if (pd == raid_disks - 1) |
147 | return block+1; | |
148 | if (block >= pd) | |
149 | return block+2; | |
150 | return block; | |
151 | ||
e0d95aac N |
152 | case 600 + ALGORITHM_ROTATING_ZERO_RESTART: |
153 | /* Different order for calculating Q, otherwize same as ... */ | |
e86c9dd6 NB |
154 | case 600 + ALGORITHM_RIGHT_ASYMMETRIC: |
155 | pd = stripe % raid_disks; | |
156 | if (block == -1) return pd; | |
48327135 | 157 | if (block == -2) return (pd+1) % raid_disks; |
e86c9dd6 NB |
158 | if (pd == raid_disks - 1) |
159 | return block+1; | |
160 | if (block >= pd) | |
161 | return block+2; | |
162 | return block; | |
163 | ||
164 | case 600 + ALGORITHM_LEFT_SYMMETRIC: | |
165 | pd = raid_disks - 1 - (stripe % raid_disks); | |
166 | if (block == -1) return pd; | |
48327135 | 167 | if (block == -2) return (pd+1) % raid_disks; |
e86c9dd6 NB |
168 | return (pd + 2 + block) % raid_disks; |
169 | ||
170 | case 600 + ALGORITHM_RIGHT_SYMMETRIC: | |
171 | pd = stripe % raid_disks; | |
172 | if (block == -1) return pd; | |
48327135 | 173 | if (block == -2) return (pd+1) % raid_disks; |
e86c9dd6 | 174 | return (pd + 2 + block) % raid_disks; |
e0d95aac N |
175 | |
176 | ||
177 | case 600 + ALGORITHM_ROTATING_N_RESTART: | |
178 | /* Same a left_asymmetric, by first stripe is | |
179 | * D D D P Q rather than | |
180 | * Q D D D P | |
181 | */ | |
182 | pd = raid_disks - 1 - ((stripe + 1) % raid_disks); | |
183 | if (block == -1) return pd; | |
184 | if (block == -2) return (pd+1) % raid_disks; | |
185 | if (pd == raid_disks - 1) | |
186 | return block+1; | |
187 | if (block >= pd) | |
188 | return block+2; | |
189 | return block; | |
190 | ||
191 | case 600 + ALGORITHM_ROTATING_N_CONTINUE: | |
192 | /* Same as left_symmetric but Q is before P */ | |
193 | pd = raid_disks - 1 - (stripe % raid_disks); | |
194 | if (block == -1) return pd; | |
195 | if (block == -2) return (pd+raid_disks-1) % raid_disks; | |
196 | return (pd + 1 + block) % raid_disks; | |
e86c9dd6 NB |
197 | } |
198 | return -1; | |
199 | } | |
e0d95aac N |
200 | static int is_ddf(int layout) |
201 | { | |
202 | switch (layout) | |
203 | { | |
204 | default: | |
205 | return 0; | |
206 | case ALGORITHM_ROTATING_N_CONTINUE: | |
207 | case ALGORITHM_ROTATING_N_RESTART: | |
208 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
209 | return 1; | |
210 | } | |
211 | } | |
e86c9dd6 NB |
212 | |
213 | ||
214 | static void xor_blocks(char *target, char **sources, int disks, int size) | |
215 | { | |
216 | int i, j; | |
217 | /* Amazingly inefficient... */ | |
218 | for (i=0; i<size; i++) { | |
219 | char c = 0; | |
220 | for (j=0 ; j<disks; j++) | |
221 | c ^= sources[j][i]; | |
222 | target[i] = c; | |
223 | } | |
224 | } | |
225 | ||
979afcb8 | 226 | void qsyndrome(uint8_t *p, uint8_t *q, uint8_t **sources, int disks, int size) |
48327135 NB |
227 | { |
228 | int d, z; | |
a6288483 | 229 | uint8_t wq0, wp0, wd0, w10, w20; |
48327135 NB |
230 | for ( d = 0; d < size; d++) { |
231 | wq0 = wp0 = sources[disks-1][d]; | |
232 | for ( z = disks-2 ; z >= 0 ; z-- ) { | |
233 | wd0 = sources[z][d]; | |
234 | wp0 ^= wd0; | |
235 | w20 = (wq0&0x80) ? 0xff : 0x00; | |
236 | w10 = (wq0 << 1) & 0xff; | |
237 | w20 &= 0x1d; | |
238 | w10 ^= w20; | |
239 | wq0 = w10 ^ wd0; | |
240 | } | |
241 | p[d] = wp0; | |
242 | q[d] = wq0; | |
243 | } | |
244 | } | |
245 | ||
a6288483 N |
246 | |
247 | /* | |
248 | * The following was taken from linux/drivers/md/mktables.c, and modified | |
249 | * to create in-memory tables rather than C code | |
250 | */ | |
251 | static uint8_t gfmul(uint8_t a, uint8_t b) | |
252 | { | |
253 | uint8_t v = 0; | |
254 | ||
255 | while (b) { | |
256 | if (b & 1) | |
257 | v ^= a; | |
258 | a = (a << 1) ^ (a & 0x80 ? 0x1d : 0); | |
259 | b >>= 1; | |
260 | } | |
261 | ||
262 | return v; | |
263 | } | |
264 | ||
265 | static uint8_t gfpow(uint8_t a, int b) | |
266 | { | |
267 | uint8_t v = 1; | |
268 | ||
269 | b %= 255; | |
270 | if (b < 0) | |
271 | b += 255; | |
272 | ||
273 | while (b) { | |
274 | if (b & 1) | |
275 | v = gfmul(v, a); | |
276 | a = gfmul(a, a); | |
277 | b >>= 1; | |
278 | } | |
279 | ||
280 | return v; | |
281 | } | |
282 | ||
283 | int tables_ready = 0; | |
284 | uint8_t raid6_gfmul[256][256]; | |
285 | uint8_t raid6_gfexp[256]; | |
286 | uint8_t raid6_gfinv[256]; | |
287 | uint8_t raid6_gfexi[256]; | |
9d0e7840 PS |
288 | uint8_t raid6_gflog[256]; |
289 | uint8_t raid6_gfilog[256]; | |
a6288483 N |
290 | void make_tables(void) |
291 | { | |
292 | int i, j; | |
293 | uint8_t v; | |
9d0e7840 | 294 | uint32_t b, log; |
a6288483 N |
295 | |
296 | /* Compute multiplication table */ | |
297 | for (i = 0; i < 256; i++) | |
298 | for (j = 0; j < 256; j++) | |
299 | raid6_gfmul[i][j] = gfmul(i, j); | |
300 | ||
301 | /* Compute power-of-2 table (exponent) */ | |
302 | v = 1; | |
303 | for (i = 0; i < 256; i++) { | |
304 | raid6_gfexp[i] = v; | |
305 | v = gfmul(v, 2); | |
306 | if (v == 1) | |
307 | v = 0; /* For entry 255, not a real entry */ | |
308 | } | |
309 | ||
310 | /* Compute inverse table x^-1 == x^254 */ | |
311 | for (i = 0; i < 256; i++) | |
312 | raid6_gfinv[i] = gfpow(i, 254); | |
313 | ||
314 | /* Compute inv(2^x + 1) (exponent-xor-inverse) table */ | |
315 | for (i = 0; i < 256; i ++) | |
316 | raid6_gfexi[i] = raid6_gfinv[raid6_gfexp[i] ^ 1]; | |
317 | ||
9d0e7840 PS |
318 | /* Compute log and inverse log */ |
319 | /* Modified code from: | |
320 | * http://web.eecs.utk.edu/~plank/plank/papers/CS-96-332.html | |
321 | */ | |
322 | b = 1; | |
323 | raid6_gflog[0] = 0; | |
324 | raid6_gfilog[255] = 0; | |
325 | ||
326 | for (log = 0; log < 255; log++) { | |
327 | raid6_gflog[b] = (uint8_t) log; | |
328 | raid6_gfilog[log] = (uint8_t) b; | |
329 | b = b << 1; | |
330 | if (b & 256) b = b ^ 0435; | |
331 | } | |
332 | ||
a6288483 N |
333 | tables_ready = 1; |
334 | } | |
335 | ||
336 | uint8_t *zero; | |
d47a2925 | 337 | int zero_size; |
a6288483 N |
338 | /* Following was taken from linux/drivers/md/raid6recov.c */ |
339 | ||
340 | /* Recover two failed data blocks. */ | |
341 | void raid6_2data_recov(int disks, size_t bytes, int faila, int failb, | |
342 | uint8_t **ptrs) | |
343 | { | |
344 | uint8_t *p, *q, *dp, *dq; | |
345 | uint8_t px, qx, db; | |
346 | const uint8_t *pbmul; /* P multiplier table for B data */ | |
347 | const uint8_t *qmul; /* Q multiplier table (for both) */ | |
348 | ||
349 | p = ptrs[disks-2]; | |
350 | q = ptrs[disks-1]; | |
351 | ||
352 | /* Compute syndrome with zero for the missing data pages | |
353 | Use the dead data pages as temporary storage for | |
354 | delta p and delta q */ | |
355 | dp = ptrs[faila]; | |
356 | ptrs[faila] = zero; | |
357 | dq = ptrs[failb]; | |
358 | ptrs[failb] = zero; | |
359 | ||
360 | qsyndrome(dp, dq, ptrs, disks-2, bytes); | |
361 | ||
362 | /* Restore pointer table */ | |
363 | ptrs[faila] = dp; | |
364 | ptrs[failb] = dq; | |
365 | ||
366 | /* Now, pick the proper data tables */ | |
367 | pbmul = raid6_gfmul[raid6_gfexi[failb-faila]]; | |
368 | qmul = raid6_gfmul[raid6_gfinv[raid6_gfexp[faila]^raid6_gfexp[failb]]]; | |
369 | ||
370 | /* Now do it... */ | |
371 | while ( bytes-- ) { | |
372 | px = *p ^ *dp; | |
373 | qx = qmul[*q ^ *dq]; | |
374 | *dq++ = db = pbmul[px] ^ qx; /* Reconstructed B */ | |
375 | *dp++ = db ^ px; /* Reconstructed A */ | |
376 | p++; q++; | |
377 | } | |
378 | } | |
379 | ||
380 | /* Recover failure of one data block plus the P block */ | |
381 | void raid6_datap_recov(int disks, size_t bytes, int faila, uint8_t **ptrs) | |
382 | { | |
383 | uint8_t *p, *q, *dq; | |
384 | const uint8_t *qmul; /* Q multiplier table */ | |
385 | ||
386 | p = ptrs[disks-2]; | |
387 | q = ptrs[disks-1]; | |
388 | ||
389 | /* Compute syndrome with zero for the missing data page | |
390 | Use the dead data page as temporary storage for delta q */ | |
391 | dq = ptrs[faila]; | |
392 | ptrs[faila] = zero; | |
393 | ||
394 | qsyndrome(p, dq, ptrs, disks-2, bytes); | |
395 | ||
396 | /* Restore pointer table */ | |
397 | ptrs[faila] = dq; | |
398 | ||
399 | /* Now, pick the proper data tables */ | |
400 | qmul = raid6_gfmul[raid6_gfinv[raid6_gfexp[faila]]]; | |
401 | ||
402 | /* Now do it... */ | |
403 | while ( bytes-- ) { | |
404 | *p++ ^= *dq = qmul[*q ^ *dq]; | |
405 | q++; dq++; | |
406 | } | |
407 | } | |
408 | ||
9d0e7840 PS |
409 | /* Try to find out if a specific disk has a problem */ |
410 | int raid6_check_disks(int data_disks, int start, int chunk_size, | |
411 | int level, int layout, int diskP, int diskQ, | |
412 | char *p, char *q, char **stripes) | |
413 | { | |
414 | int i; | |
415 | int data_id, diskD; | |
416 | uint8_t Px, Qx; | |
417 | int curr_broken_disk = -1; | |
418 | int prev_broken_disk = -1; | |
419 | int broken_status = 0; | |
420 | ||
421 | for(i = 0; i < chunk_size; i++) { | |
422 | Px = (uint8_t)stripes[diskP][i] ^ (uint8_t)p[i]; | |
423 | Qx = (uint8_t)stripes[diskQ][i] ^ (uint8_t)q[i]; | |
424 | ||
425 | if((Px != 0) && (Qx == 0)) | |
426 | curr_broken_disk = diskP; | |
427 | ||
428 | ||
429 | if((Px == 0) && (Qx != 0)) | |
430 | curr_broken_disk = diskQ; | |
431 | ||
432 | ||
433 | if((Px != 0) && (Qx != 0)) { | |
c4db5301 PS |
434 | data_id = (raid6_gflog[Qx] - raid6_gflog[Px]); |
435 | if(data_id < 0) data_id += 255; | |
9d0e7840 PS |
436 | diskD = geo_map(data_id, start/chunk_size, |
437 | data_disks + 2, level, layout); | |
438 | curr_broken_disk = diskD; | |
439 | } | |
440 | ||
441 | if((Px == 0) && (Qx == 0)) | |
442 | curr_broken_disk = curr_broken_disk; | |
443 | ||
c4db5301 PS |
444 | if(curr_broken_disk >= data_disks + 2) |
445 | broken_status = 2; | |
446 | ||
9d0e7840 PS |
447 | switch(broken_status) { |
448 | case 0: | |
449 | if(curr_broken_disk != -1) { | |
450 | prev_broken_disk = curr_broken_disk; | |
451 | broken_status = 1; | |
452 | } | |
453 | break; | |
454 | ||
455 | case 1: | |
456 | if(curr_broken_disk != prev_broken_disk) | |
457 | broken_status = 2; | |
9d0e7840 PS |
458 | break; |
459 | ||
460 | case 2: | |
461 | default: | |
462 | curr_broken_disk = prev_broken_disk = -2; | |
463 | break; | |
464 | } | |
465 | } | |
466 | ||
467 | return curr_broken_disk; | |
468 | } | |
469 | ||
2fcb75ae AK |
470 | /******************************************************************************* |
471 | * Function: save_stripes | |
472 | * Description: | |
473 | * Function reads data (only data without P and Q) from array and writes | |
474 | * it to buf and opcjonaly to backup files | |
475 | * Parameters: | |
476 | * source : A list of 'fds' of the active disks. | |
477 | * Some may be absent | |
478 | * offsets : A list of offsets on disk belonging | |
479 | * to the array [bytes] | |
480 | * raid_disks : geometry: number of disks in the array | |
481 | * chunk_size : geometry: chunk size [bytes] | |
482 | * level : geometry: RAID level | |
483 | * layout : geometry: layout | |
484 | * nwrites : number of backup files | |
485 | * dest : A list of 'fds' for mirrored targets | |
486 | * (e.g. backup files). They are already seeked to right | |
487 | * (write) location. If NULL, data will be wrote | |
488 | * to the buf only | |
489 | * start : start address of data to read (must be stripe-aligned) | |
490 | * [bytes] | |
491 | * length - : length of data to read (must be stripe-aligned) | |
492 | * [bytes] | |
493 | * buf : buffer for data. It is large enough to hold | |
494 | * one stripe. It is stripe aligned | |
495 | * Returns: | |
496 | * 0 : success | |
497 | * -1 : fail | |
498 | ******************************************************************************/ | |
e86c9dd6 NB |
499 | int save_stripes(int *source, unsigned long long *offsets, |
500 | int raid_disks, int chunk_size, int level, int layout, | |
501 | int nwrites, int *dest, | |
a6288483 N |
502 | unsigned long long start, unsigned long long length, |
503 | char *buf) | |
e86c9dd6 | 504 | { |
e86c9dd6 NB |
505 | int len; |
506 | int data_disks = raid_disks - (level == 0 ? 0 : level <=5 ? 1 : 2); | |
507 | int disk; | |
a6288483 | 508 | int i; |
2fcb75ae | 509 | unsigned long long length_test; |
e86c9dd6 | 510 | |
a6288483 N |
511 | if (!tables_ready) |
512 | make_tables(); | |
513 | ||
d47a2925 N |
514 | if (zero == NULL || chunk_size > zero_size) { |
515 | if (zero) | |
516 | free(zero); | |
a6288483 | 517 | zero = malloc(chunk_size); |
d47a2925 N |
518 | if (zero) |
519 | memset(zero, 0, chunk_size); | |
520 | zero_size = chunk_size; | |
a6288483 N |
521 | } |
522 | ||
523 | len = data_disks * chunk_size; | |
2fcb75ae AK |
524 | length_test = length / len; |
525 | length_test *= len; | |
526 | ||
527 | if (length != length_test) { | |
528 | dprintf("Error: save_stripes(): Data are not alligned. EXIT\n"); | |
529 | dprintf("\tArea for saving stripes (length) = %llu\n", length); | |
530 | dprintf("\tWork step (len) = %i\n", len); | |
531 | dprintf("\tExpected save area (length_test) = %llu\n", | |
532 | length_test); | |
533 | abort(); | |
534 | } | |
535 | ||
e86c9dd6 | 536 | while (length > 0) { |
a6288483 N |
537 | int failed = 0; |
538 | int fdisk[3], fblock[3]; | |
539 | for (disk = 0; disk < raid_disks ; disk++) { | |
540 | unsigned long long offset; | |
541 | int dnum; | |
a6288483 N |
542 | |
543 | offset = (start/chunk_size/data_disks)*chunk_size; | |
544 | dnum = geo_map(disk < data_disks ? disk : data_disks - disk - 1, | |
545 | start/chunk_size/data_disks, | |
546 | raid_disks, level, layout); | |
7236ee7a | 547 | if (dnum < 0) abort(); |
a6288483 | 548 | if (source[dnum] < 0 || |
cc50ccdc | 549 | lseek64(source[dnum], offsets[dnum]+offset, 0) < 0 || |
7236ee7a N |
550 | read(source[dnum], buf+disk * chunk_size, chunk_size) |
551 | != chunk_size) | |
a6288483 N |
552 | if (failed <= 2) { |
553 | fdisk[failed] = dnum; | |
554 | fblock[failed] = disk; | |
555 | failed++; | |
556 | } | |
557 | } | |
558 | if (failed == 0 || fblock[0] >= data_disks) | |
559 | /* all data disks are good */ | |
560 | ; | |
561 | else if (failed == 1 || fblock[1] >= data_disks+1) { | |
562 | /* one failed data disk and good parity */ | |
563 | char *bufs[data_disks]; | |
564 | for (i=0; i < data_disks; i++) | |
565 | if (fblock[0] == i) | |
566 | bufs[i] = buf + data_disks*chunk_size; | |
567 | else | |
568 | bufs[i] = buf + i*chunk_size; | |
569 | ||
570 | xor_blocks(buf + fblock[0]*chunk_size, | |
571 | bufs, data_disks, chunk_size); | |
572 | } else if (failed > 2 || level != 6) | |
573 | /* too much failure */ | |
e86c9dd6 | 574 | return -1; |
a6288483 N |
575 | else { |
576 | /* RAID6 computations needed. */ | |
577 | uint8_t *bufs[data_disks+4]; | |
578 | int qdisk; | |
579 | int syndrome_disks; | |
580 | disk = geo_map(-1, start/chunk_size/data_disks, | |
581 | raid_disks, level, layout); | |
582 | qdisk = geo_map(-2, start/chunk_size/data_disks, | |
583 | raid_disks, level, layout); | |
584 | if (is_ddf(layout)) { | |
585 | /* q over 'raid_disks' blocks, in device order. | |
586 | * 'p' and 'q' get to be all zero | |
587 | */ | |
588 | for (i = 0; i < raid_disks; i++) | |
cc50ccdc N |
589 | bufs[i] = zero; |
590 | for (i = 0; i < data_disks; i++) { | |
591 | int dnum = geo_map(i, | |
592 | start/chunk_size/data_disks, | |
593 | raid_disks, level, layout); | |
594 | int snum; | |
595 | /* i is the logical block number, so is index to 'buf'. | |
596 | * dnum is physical disk number | |
597 | * and thus the syndrome number. | |
598 | */ | |
599 | snum = dnum; | |
600 | bufs[snum] = (uint8_t*)buf + chunk_size * i; | |
601 | } | |
a6288483 N |
602 | syndrome_disks = raid_disks; |
603 | } else { | |
604 | /* for md, q is over 'data_disks' blocks, | |
605 | * starting immediately after 'q' | |
1eac9f84 N |
606 | * Note that for the '_6' variety, the p block |
607 | * makes a hole that we need to be careful of. | |
a6288483 | 608 | */ |
1eac9f84 N |
609 | int j; |
610 | int snum = 0; | |
611 | for (j = 0; j < raid_disks; j++) { | |
612 | int dnum = (qdisk + 1 + j) % raid_disks; | |
613 | if (dnum == disk || dnum == qdisk) | |
614 | continue; | |
615 | for (i = 0; i < data_disks; i++) | |
616 | if (geo_map(i, | |
617 | start/chunk_size/data_disks, | |
618 | raid_disks, level, layout) == dnum) | |
619 | break; | |
cc50ccdc N |
620 | /* i is the logical block number, so is index to 'buf'. |
621 | * dnum is physical disk number | |
622 | * snum is syndrome disk for which 0 is immediately after Q | |
623 | */ | |
cc50ccdc | 624 | bufs[snum] = (uint8_t*)buf + chunk_size * i; |
1eac9f84 N |
625 | |
626 | if (fblock[0] == i) | |
627 | fdisk[0] = snum; | |
628 | if (fblock[1] == i) | |
629 | fdisk[1] = snum; | |
630 | snum++; | |
cc50ccdc | 631 | } |
a6288483 | 632 | |
a6288483 N |
633 | syndrome_disks = data_disks; |
634 | } | |
cc50ccdc N |
635 | |
636 | /* Place P and Q blocks at end of bufs */ | |
637 | bufs[syndrome_disks] = (uint8_t*)buf + chunk_size * data_disks; | |
638 | bufs[syndrome_disks+1] = (uint8_t*)buf + chunk_size * (data_disks+1); | |
639 | ||
a6288483 N |
640 | if (fblock[1] == data_disks) |
641 | /* One data failed, and parity failed */ | |
642 | raid6_datap_recov(syndrome_disks+2, chunk_size, | |
643 | fdisk[0], bufs); | |
cc50ccdc N |
644 | else { |
645 | if (fdisk[0] > fdisk[1]) { | |
646 | int t = fdisk[0]; | |
647 | fdisk[0] = fdisk[1]; | |
648 | fdisk[1] = t; | |
649 | } | |
a6288483 N |
650 | /* Two data blocks failed, P,Q OK */ |
651 | raid6_2data_recov(syndrome_disks+2, chunk_size, | |
652 | fdisk[0], fdisk[1], bufs); | |
cc50ccdc | 653 | } |
a6288483 | 654 | } |
ccced3dc | 655 | if (dest) { |
2fcb75ae AK |
656 | for (i = 0; i < nwrites; i++) |
657 | if (write(dest[i], buf, len) != len) | |
658 | return -1; | |
ccced3dc AK |
659 | } else { |
660 | /* build next stripe in buffer */ | |
661 | buf += len; | |
662 | } | |
e86c9dd6 NB |
663 | length -= len; |
664 | start += len; | |
e86c9dd6 NB |
665 | } |
666 | return 0; | |
667 | } | |
668 | ||
669 | /* Restore data: | |
670 | * We are given: | |
671 | * A list of 'fds' of the active disks. Some may be '-1' for not-available. | |
353632d9 | 672 | * A geometry: raid_disks, chunk_size, level, layout |
e86c9dd6 NB |
673 | * An 'fd' to read from. It is already seeked to the right (Read) location. |
674 | * A start and length. | |
675 | * The length must be a multiple of the stripe size. | |
676 | * | |
677 | * We build a full stripe in memory and then write it out. | |
678 | * We assume that there are enough working devices. | |
679 | */ | |
680 | int restore_stripes(int *dest, unsigned long long *offsets, | |
681 | int raid_disks, int chunk_size, int level, int layout, | |
353632d9 | 682 | int source, unsigned long long read_offset, |
2fcb75ae AK |
683 | unsigned long long start, unsigned long long length, |
684 | char *src_buf) | |
e86c9dd6 | 685 | { |
e9e43ec3 | 686 | char *stripe_buf; |
e86c9dd6 NB |
687 | char **stripes = malloc(raid_disks * sizeof(char*)); |
688 | char **blocks = malloc(raid_disks * sizeof(char*)); | |
689 | int i; | |
758be4f1 | 690 | int rv; |
e86c9dd6 | 691 | |
a6288483 | 692 | int data_disks = raid_disks - (level == 0 ? 0 : level <= 5 ? 1 : 2); |
e86c9dd6 | 693 | |
fcf57625 N |
694 | if (posix_memalign((void**)&stripe_buf, 4096, raid_disks * chunk_size)) |
695 | stripe_buf = NULL; | |
d47a2925 N |
696 | |
697 | if (zero == NULL || chunk_size > zero_size) { | |
698 | if (zero) | |
699 | free(zero); | |
a6288483 N |
700 | zero = malloc(chunk_size); |
701 | if (zero) | |
702 | memset(zero, 0, chunk_size); | |
d47a2925 | 703 | zero_size = chunk_size; |
a6288483 | 704 | } |
d47a2925 | 705 | |
e0d95aac N |
706 | if (stripe_buf == NULL || stripes == NULL || blocks == NULL |
707 | || zero == NULL) { | |
758be4f1 LD |
708 | rv = -2; |
709 | goto abort; | |
e86c9dd6 | 710 | } |
2fcb75ae | 711 | for (i = 0; i < raid_disks; i++) |
e86c9dd6 NB |
712 | stripes[i] = stripe_buf + i * chunk_size; |
713 | while (length > 0) { | |
f21e18ca | 714 | unsigned int len = data_disks * chunk_size; |
e86c9dd6 | 715 | unsigned long long offset; |
48327135 | 716 | int disk, qdisk; |
a6288483 | 717 | int syndrome_disks; |
758be4f1 LD |
718 | if (length < len) { |
719 | rv = -3; | |
720 | goto abort; | |
721 | } | |
2fcb75ae | 722 | for (i = 0; i < data_disks; i++) { |
e86c9dd6 NB |
723 | int disk = geo_map(i, start/chunk_size/data_disks, |
724 | raid_disks, level, layout); | |
2fcb75ae AK |
725 | if (src_buf == NULL) { |
726 | /* read from file */ | |
758be4f1 LD |
727 | if (lseek64(source, read_offset, 0) != |
728 | (off64_t)read_offset) { | |
729 | rv = -1; | |
730 | goto abort; | |
731 | } | |
2fcb75ae AK |
732 | if (read(source, |
733 | stripes[disk], | |
758be4f1 LD |
734 | chunk_size) != chunk_size) { |
735 | rv = -1; | |
736 | goto abort; | |
737 | } | |
2fcb75ae AK |
738 | } else { |
739 | /* read from input buffer */ | |
740 | memcpy(stripes[disk], | |
741 | src_buf + read_offset, | |
742 | chunk_size); | |
743 | } | |
353632d9 | 744 | read_offset += chunk_size; |
e86c9dd6 NB |
745 | } |
746 | /* We have the data, now do the parity */ | |
747 | offset = (start/chunk_size/data_disks) * chunk_size; | |
48327135 NB |
748 | switch (level) { |
749 | case 4: | |
750 | case 5: | |
751 | disk = geo_map(-1, start/chunk_size/data_disks, | |
e86c9dd6 | 752 | raid_disks, level, layout); |
e0d95aac N |
753 | for (i = 0; i < data_disks; i++) |
754 | blocks[i] = stripes[(disk+1+i) % raid_disks]; | |
e86c9dd6 | 755 | xor_blocks(stripes[disk], blocks, data_disks, chunk_size); |
48327135 NB |
756 | break; |
757 | case 6: | |
758 | disk = geo_map(-1, start/chunk_size/data_disks, | |
759 | raid_disks, level, layout); | |
760 | qdisk = geo_map(-2, start/chunk_size/data_disks, | |
761 | raid_disks, level, layout); | |
e0d95aac N |
762 | if (is_ddf(layout)) { |
763 | /* q over 'raid_disks' blocks, in device order. | |
764 | * 'p' and 'q' get to be all zero | |
765 | */ | |
766 | for (i = 0; i < raid_disks; i++) | |
767 | if (i == disk || i == qdisk) | |
a6288483 | 768 | blocks[i] = (char*)zero; |
e0d95aac N |
769 | else |
770 | blocks[i] = stripes[i]; | |
a6288483 | 771 | syndrome_disks = raid_disks; |
e0d95aac | 772 | } else { |
a6288483 | 773 | /* for md, q is over 'data_disks' blocks, |
e0d95aac N |
774 | * starting immediately after 'q' |
775 | */ | |
776 | for (i = 0; i < data_disks; i++) | |
777 | blocks[i] = stripes[(qdisk+1+i) % raid_disks]; | |
48327135 | 778 | |
a6288483 | 779 | syndrome_disks = data_disks; |
e0d95aac | 780 | } |
a6288483 N |
781 | qsyndrome((uint8_t*)stripes[disk], |
782 | (uint8_t*)stripes[qdisk], | |
783 | (uint8_t**)blocks, | |
784 | syndrome_disks, chunk_size); | |
48327135 | 785 | break; |
e86c9dd6 NB |
786 | } |
787 | for (i=0; i < raid_disks ; i++) | |
788 | if (dest[i] >= 0) { | |
758be4f1 LD |
789 | if (lseek64(dest[i], |
790 | offsets[i]+offset, 0) < 0) { | |
791 | rv = -1; | |
792 | goto abort; | |
793 | } | |
794 | if (write(dest[i], stripes[i], | |
795 | chunk_size) != chunk_size) { | |
796 | rv = -1; | |
797 | goto abort; | |
798 | } | |
e86c9dd6 NB |
799 | } |
800 | length -= len; | |
801 | start += len; | |
802 | } | |
758be4f1 LD |
803 | rv = 0; |
804 | ||
805 | abort: | |
806 | free(stripe_buf); | |
807 | free(stripes); | |
808 | free(blocks); | |
809 | return rv; | |
e86c9dd6 NB |
810 | } |
811 | ||
812 | #ifdef MAIN | |
813 | ||
48327135 NB |
814 | int test_stripes(int *source, unsigned long long *offsets, |
815 | int raid_disks, int chunk_size, int level, int layout, | |
816 | unsigned long long start, unsigned long long length) | |
817 | { | |
818 | /* ready the data and p (and q) blocks, and check we got them right */ | |
819 | char *stripe_buf = malloc(raid_disks * chunk_size); | |
820 | char **stripes = malloc(raid_disks * sizeof(char*)); | |
821 | char **blocks = malloc(raid_disks * sizeof(char*)); | |
822 | char *p = malloc(chunk_size); | |
823 | char *q = malloc(chunk_size); | |
824 | ||
825 | int i; | |
9d0e7840 | 826 | int diskP, diskQ; |
48327135 | 827 | int data_disks = raid_disks - (level == 5 ? 1: 2); |
9d0e7840 PS |
828 | |
829 | if (!tables_ready) | |
830 | make_tables(); | |
831 | ||
48327135 NB |
832 | for ( i = 0 ; i < raid_disks ; i++) |
833 | stripes[i] = stripe_buf + i * chunk_size; | |
834 | ||
835 | while (length > 0) { | |
836 | int disk; | |
837 | ||
838 | for (i = 0 ; i < raid_disks ; i++) { | |
839 | lseek64(source[i], offsets[i]+start, 0); | |
840 | read(source[i], stripes[i], chunk_size); | |
841 | } | |
842 | for (i = 0 ; i < data_disks ; i++) { | |
843 | int disk = geo_map(i, start/chunk_size, raid_disks, | |
844 | level, layout); | |
845 | blocks[i] = stripes[disk]; | |
846 | printf("%d->%d\n", i, disk); | |
847 | } | |
848 | switch(level) { | |
849 | case 6: | |
521f349c | 850 | qsyndrome(p, q, (uint8_t**)blocks, data_disks, chunk_size); |
9d0e7840 | 851 | diskP = geo_map(-1, start/chunk_size, raid_disks, |
48327135 | 852 | level, layout); |
9d0e7840 PS |
853 | if (memcmp(p, stripes[diskP], chunk_size) != 0) { |
854 | printf("P(%d) wrong at %llu\n", diskP, | |
48327135 NB |
855 | start / chunk_size); |
856 | } | |
9d0e7840 | 857 | diskQ = geo_map(-2, start/chunk_size, raid_disks, |
48327135 | 858 | level, layout); |
9d0e7840 PS |
859 | if (memcmp(q, stripes[diskQ], chunk_size) != 0) { |
860 | printf("Q(%d) wrong at %llu\n", diskQ, | |
48327135 NB |
861 | start / chunk_size); |
862 | } | |
9d0e7840 PS |
863 | disk = raid6_check_disks(data_disks, start, chunk_size, |
864 | level, layout, diskP, diskQ, | |
865 | p, q, stripes); | |
866 | if(disk >= 0) { | |
867 | printf("Possible failed disk: %d\n", disk); | |
868 | } | |
869 | if(disk == -2) { | |
870 | printf("Failure detected, but disk unknown\n"); | |
871 | } | |
48327135 NB |
872 | break; |
873 | } | |
874 | length -= chunk_size; | |
875 | start += chunk_size; | |
876 | } | |
877 | return 0; | |
878 | } | |
879 | ||
e86c9dd6 NB |
880 | unsigned long long getnum(char *str, char **err) |
881 | { | |
882 | char *e; | |
883 | unsigned long long rv = strtoull(str, &e, 10); | |
884 | if (e==str || *e) { | |
885 | *err = str; | |
886 | return 0; | |
887 | } | |
888 | return rv; | |
889 | } | |
890 | ||
891 | main(int argc, char *argv[]) | |
892 | { | |
893 | /* save/restore file raid_disks chunk_size level layout start length devices... | |
894 | */ | |
895 | int save; | |
896 | int *fds; | |
897 | char *file; | |
a6288483 | 898 | char *buf; |
e86c9dd6 NB |
899 | int storefd; |
900 | unsigned long long *offsets; | |
901 | int raid_disks, chunk_size, level, layout; | |
902 | unsigned long long start, length; | |
903 | int i; | |
904 | ||
905 | char *err = NULL; | |
906 | if (argc < 10) { | |
907 | fprintf(stderr, "Usage: test_stripe save/restore file raid_disks" | |
908 | " chunk_size level layout start length devices...\n"); | |
909 | exit(1); | |
910 | } | |
911 | if (strcmp(argv[1], "save")==0) | |
912 | save = 1; | |
913 | else if (strcmp(argv[1], "restore") == 0) | |
914 | save = 0; | |
48327135 NB |
915 | else if (strcmp(argv[1], "test") == 0) |
916 | save = 2; | |
e86c9dd6 NB |
917 | else { |
918 | fprintf(stderr, "test_stripe: must give 'save' or 'restore'.\n"); | |
919 | exit(2); | |
920 | } | |
921 | ||
922 | file = argv[2]; | |
923 | raid_disks = getnum(argv[3], &err); | |
924 | chunk_size = getnum(argv[4], &err); | |
925 | level = getnum(argv[5], &err); | |
926 | layout = getnum(argv[6], &err); | |
927 | start = getnum(argv[7], &err); | |
928 | length = getnum(argv[8], &err); | |
929 | if (err) { | |
930 | fprintf(stderr, "test_stripe: Bad number: %s\n", err); | |
931 | exit(2); | |
932 | } | |
933 | if (argc != raid_disks + 9) { | |
934 | fprintf(stderr, "test_stripe: wrong number of devices: want %d found %d\n", | |
935 | raid_disks, argc-9); | |
936 | exit(2); | |
937 | } | |
938 | fds = malloc(raid_disks * sizeof(*fds)); | |
939 | offsets = malloc(raid_disks * sizeof(*offsets)); | |
940 | memset(offsets, 0, raid_disks * sizeof(*offsets)); | |
941 | ||
942 | storefd = open(file, O_RDWR); | |
943 | if (storefd < 0) { | |
944 | perror(file); | |
945 | fprintf(stderr, "test_stripe: could not open %s.\n", file); | |
946 | exit(3); | |
947 | } | |
948 | for (i=0; i<raid_disks; i++) { | |
6f38d7ae PS |
949 | char *p; |
950 | p = strchr(argv[9+i], ':'); | |
951 | ||
952 | if(p != NULL) { | |
953 | *p++ = '\0'; | |
954 | offsets[i] = atoll(p) * 512; | |
955 | } | |
956 | ||
e86c9dd6 NB |
957 | fds[i] = open(argv[9+i], O_RDWR); |
958 | if (fds[i] < 0) { | |
959 | perror(argv[9+i]); | |
960 | fprintf(stderr,"test_stripe: cannot open %s.\n", argv[9+i]); | |
961 | exit(3); | |
962 | } | |
963 | } | |
964 | ||
a6288483 N |
965 | buf = malloc(raid_disks * chunk_size); |
966 | ||
48327135 | 967 | if (save == 1) { |
e86c9dd6 NB |
968 | int rv = save_stripes(fds, offsets, |
969 | raid_disks, chunk_size, level, layout, | |
970 | 1, &storefd, | |
a6288483 | 971 | start, length, buf); |
e86c9dd6 | 972 | if (rv != 0) { |
48327135 NB |
973 | fprintf(stderr, |
974 | "test_stripe: save_stripes returned %d\n", rv); | |
975 | exit(1); | |
976 | } | |
977 | } else if (save == 2) { | |
978 | int rv = test_stripes(fds, offsets, | |
979 | raid_disks, chunk_size, level, layout, | |
980 | start, length); | |
981 | if (rv != 0) { | |
982 | fprintf(stderr, | |
983 | "test_stripe: test_stripes returned %d\n", rv); | |
e86c9dd6 NB |
984 | exit(1); |
985 | } | |
986 | } else { | |
987 | int rv = restore_stripes(fds, offsets, | |
988 | raid_disks, chunk_size, level, layout, | |
353632d9 | 989 | storefd, 0ULL, |
c071a1cd | 990 | start, length, NULL); |
e86c9dd6 | 991 | if (rv != 0) { |
48327135 NB |
992 | fprintf(stderr, |
993 | "test_stripe: restore_stripes returned %d\n", | |
994 | rv); | |
e86c9dd6 NB |
995 | exit(1); |
996 | } | |
997 | } | |
998 | exit(0); | |
999 | } | |
1000 | ||
1001 | #endif /* MAIN */ |