]>
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 | ||
e0d95aac N |
86 | case 600 + ALGORITHM_PARITY_N_6: |
87 | if (block == -2) | |
88 | return raid_disks - 1; | |
89 | if (block == -1) | |
90 | return raid_disks - 2; /* parity block */ | |
91 | return block; | |
92 | case 600 + ALGORITHM_LEFT_ASYMMETRIC_6: | |
93 | if (block == -2) | |
94 | return raid_disks - 1; | |
95 | raid_disks--; | |
96 | pd = (raid_disks-1) - stripe % raid_disks; | |
97 | if (block == -1) return pd; | |
98 | if (block >= pd) | |
99 | block++; | |
100 | return block; | |
101 | ||
102 | case 600 + ALGORITHM_RIGHT_ASYMMETRIC_6: | |
103 | if (block == -2) | |
104 | return raid_disks - 1; | |
105 | raid_disks--; | |
106 | pd = stripe % raid_disks; | |
107 | if (block == -1) return pd; | |
108 | if (block >= pd) | |
109 | block++; | |
110 | return block; | |
111 | ||
112 | case 600 + ALGORITHM_LEFT_SYMMETRIC_6: | |
113 | if (block == -2) | |
114 | return raid_disks - 1; | |
115 | raid_disks--; | |
116 | pd = (raid_disks - 1) - stripe % raid_disks; | |
117 | if (block == -1) return pd; | |
118 | return (pd + 1 + block) % raid_disks; | |
119 | ||
120 | case 600 + ALGORITHM_RIGHT_SYMMETRIC_6: | |
121 | if (block == -2) | |
122 | return raid_disks - 1; | |
123 | raid_disks--; | |
124 | pd = stripe % raid_disks; | |
125 | if (block == -1) return pd; | |
126 | return (pd + 1 + block) % raid_disks; | |
127 | ||
128 | case 600 + ALGORITHM_PARITY_0_6: | |
129 | if (block == -2) | |
130 | return raid_disks - 1; | |
131 | return block + 1; | |
132 | ||
e0d95aac N |
133 | case 600 + ALGORITHM_PARITY_0: |
134 | if (block == -1) | |
135 | return 0; | |
136 | if (block == -2) | |
137 | return 1; | |
138 | return block + 2; | |
139 | ||
e86c9dd6 NB |
140 | case 600 + ALGORITHM_LEFT_ASYMMETRIC: |
141 | pd = raid_disks - 1 - (stripe % raid_disks); | |
142 | if (block == -1) return pd; | |
48327135 | 143 | if (block == -2) return (pd+1) % raid_disks; |
e86c9dd6 NB |
144 | if (pd == raid_disks - 1) |
145 | return block+1; | |
146 | if (block >= pd) | |
147 | return block+2; | |
148 | return block; | |
149 | ||
e0d95aac N |
150 | case 600 + ALGORITHM_ROTATING_ZERO_RESTART: |
151 | /* Different order for calculating Q, otherwize same as ... */ | |
e86c9dd6 NB |
152 | case 600 + ALGORITHM_RIGHT_ASYMMETRIC: |
153 | pd = stripe % raid_disks; | |
154 | if (block == -1) return pd; | |
48327135 | 155 | if (block == -2) return (pd+1) % raid_disks; |
e86c9dd6 NB |
156 | if (pd == raid_disks - 1) |
157 | return block+1; | |
158 | if (block >= pd) | |
159 | return block+2; | |
160 | return block; | |
161 | ||
162 | case 600 + ALGORITHM_LEFT_SYMMETRIC: | |
163 | pd = raid_disks - 1 - (stripe % raid_disks); | |
164 | if (block == -1) return pd; | |
48327135 | 165 | if (block == -2) return (pd+1) % raid_disks; |
e86c9dd6 NB |
166 | return (pd + 2 + block) % raid_disks; |
167 | ||
168 | case 600 + ALGORITHM_RIGHT_SYMMETRIC: | |
169 | pd = stripe % raid_disks; | |
170 | if (block == -1) return pd; | |
48327135 | 171 | if (block == -2) return (pd+1) % raid_disks; |
e86c9dd6 | 172 | return (pd + 2 + block) % raid_disks; |
e0d95aac | 173 | |
e0d95aac N |
174 | case 600 + ALGORITHM_ROTATING_N_RESTART: |
175 | /* Same a left_asymmetric, by first stripe is | |
176 | * D D D P Q rather than | |
177 | * Q D D D P | |
178 | */ | |
179 | pd = raid_disks - 1 - ((stripe + 1) % raid_disks); | |
180 | if (block == -1) return pd; | |
181 | if (block == -2) return (pd+1) % raid_disks; | |
182 | if (pd == raid_disks - 1) | |
183 | return block+1; | |
184 | if (block >= pd) | |
185 | return block+2; | |
186 | return block; | |
187 | ||
188 | case 600 + ALGORITHM_ROTATING_N_CONTINUE: | |
189 | /* Same as left_symmetric but Q is before P */ | |
190 | pd = raid_disks - 1 - (stripe % raid_disks); | |
191 | if (block == -1) return pd; | |
192 | if (block == -2) return (pd+raid_disks-1) % raid_disks; | |
193 | return (pd + 1 + block) % raid_disks; | |
e86c9dd6 NB |
194 | } |
195 | return -1; | |
196 | } | |
e0d95aac N |
197 | static int is_ddf(int layout) |
198 | { | |
199 | switch (layout) | |
200 | { | |
201 | default: | |
202 | return 0; | |
203 | case ALGORITHM_ROTATING_N_CONTINUE: | |
204 | case ALGORITHM_ROTATING_N_RESTART: | |
205 | case ALGORITHM_ROTATING_ZERO_RESTART: | |
206 | return 1; | |
207 | } | |
208 | } | |
e86c9dd6 | 209 | |
59679536 | 210 | void xor_blocks(char *target, char **sources, int disks, int size) |
e86c9dd6 NB |
211 | { |
212 | int i, j; | |
213 | /* Amazingly inefficient... */ | |
214 | for (i=0; i<size; i++) { | |
215 | char c = 0; | |
216 | for (j=0 ; j<disks; j++) | |
217 | c ^= sources[j][i]; | |
218 | target[i] = c; | |
219 | } | |
220 | } | |
221 | ||
979afcb8 | 222 | void qsyndrome(uint8_t *p, uint8_t *q, uint8_t **sources, int disks, int size) |
48327135 NB |
223 | { |
224 | int d, z; | |
a6288483 | 225 | uint8_t wq0, wp0, wd0, w10, w20; |
48327135 NB |
226 | for ( d = 0; d < size; d++) { |
227 | wq0 = wp0 = sources[disks-1][d]; | |
228 | for ( z = disks-2 ; z >= 0 ; z-- ) { | |
229 | wd0 = sources[z][d]; | |
230 | wp0 ^= wd0; | |
231 | w20 = (wq0&0x80) ? 0xff : 0x00; | |
232 | w10 = (wq0 << 1) & 0xff; | |
233 | w20 &= 0x1d; | |
234 | w10 ^= w20; | |
235 | wq0 = w10 ^ wd0; | |
236 | } | |
237 | p[d] = wp0; | |
238 | q[d] = wq0; | |
239 | } | |
240 | } | |
241 | ||
a6288483 N |
242 | /* |
243 | * The following was taken from linux/drivers/md/mktables.c, and modified | |
244 | * to create in-memory tables rather than C code | |
245 | */ | |
246 | static uint8_t gfmul(uint8_t a, uint8_t b) | |
247 | { | |
248 | uint8_t v = 0; | |
249 | ||
250 | while (b) { | |
251 | if (b & 1) | |
252 | v ^= a; | |
253 | a = (a << 1) ^ (a & 0x80 ? 0x1d : 0); | |
254 | b >>= 1; | |
255 | } | |
256 | ||
257 | return v; | |
258 | } | |
259 | ||
260 | static uint8_t gfpow(uint8_t a, int b) | |
261 | { | |
262 | uint8_t v = 1; | |
263 | ||
264 | b %= 255; | |
265 | if (b < 0) | |
266 | b += 255; | |
267 | ||
268 | while (b) { | |
269 | if (b & 1) | |
270 | v = gfmul(v, a); | |
271 | a = gfmul(a, a); | |
272 | b >>= 1; | |
273 | } | |
274 | ||
275 | return v; | |
276 | } | |
277 | ||
278 | int tables_ready = 0; | |
279 | uint8_t raid6_gfmul[256][256]; | |
280 | uint8_t raid6_gfexp[256]; | |
281 | uint8_t raid6_gfinv[256]; | |
282 | uint8_t raid6_gfexi[256]; | |
9d0e7840 PS |
283 | uint8_t raid6_gflog[256]; |
284 | uint8_t raid6_gfilog[256]; | |
a6288483 N |
285 | void make_tables(void) |
286 | { | |
287 | int i, j; | |
288 | uint8_t v; | |
9d0e7840 | 289 | uint32_t b, log; |
a6288483 N |
290 | |
291 | /* Compute multiplication table */ | |
292 | for (i = 0; i < 256; i++) | |
293 | for (j = 0; j < 256; j++) | |
294 | raid6_gfmul[i][j] = gfmul(i, j); | |
295 | ||
296 | /* Compute power-of-2 table (exponent) */ | |
297 | v = 1; | |
298 | for (i = 0; i < 256; i++) { | |
299 | raid6_gfexp[i] = v; | |
300 | v = gfmul(v, 2); | |
301 | if (v == 1) | |
302 | v = 0; /* For entry 255, not a real entry */ | |
303 | } | |
304 | ||
305 | /* Compute inverse table x^-1 == x^254 */ | |
306 | for (i = 0; i < 256; i++) | |
307 | raid6_gfinv[i] = gfpow(i, 254); | |
308 | ||
309 | /* Compute inv(2^x + 1) (exponent-xor-inverse) table */ | |
310 | for (i = 0; i < 256; i ++) | |
311 | raid6_gfexi[i] = raid6_gfinv[raid6_gfexp[i] ^ 1]; | |
312 | ||
9d0e7840 PS |
313 | /* Compute log and inverse log */ |
314 | /* Modified code from: | |
315 | * http://web.eecs.utk.edu/~plank/plank/papers/CS-96-332.html | |
316 | */ | |
317 | b = 1; | |
318 | raid6_gflog[0] = 0; | |
319 | raid6_gfilog[255] = 0; | |
320 | ||
321 | for (log = 0; log < 255; log++) { | |
322 | raid6_gflog[b] = (uint8_t) log; | |
323 | raid6_gfilog[log] = (uint8_t) b; | |
324 | b = b << 1; | |
325 | if (b & 256) b = b ^ 0435; | |
326 | } | |
327 | ||
a6288483 N |
328 | tables_ready = 1; |
329 | } | |
330 | ||
331 | uint8_t *zero; | |
d47a2925 | 332 | int zero_size; |
59679536 RB |
333 | |
334 | void ensure_zero_has_size(int chunk_size) | |
335 | { | |
336 | if (zero == NULL || chunk_size > zero_size) { | |
337 | if (zero) | |
338 | free(zero); | |
339 | zero = xcalloc(1, chunk_size); | |
340 | zero_size = chunk_size; | |
341 | } | |
342 | } | |
343 | ||
a6288483 N |
344 | /* Following was taken from linux/drivers/md/raid6recov.c */ |
345 | ||
346 | /* Recover two failed data blocks. */ | |
347 | void raid6_2data_recov(int disks, size_t bytes, int faila, int failb, | |
348 | uint8_t **ptrs) | |
349 | { | |
350 | uint8_t *p, *q, *dp, *dq; | |
351 | uint8_t px, qx, db; | |
352 | const uint8_t *pbmul; /* P multiplier table for B data */ | |
353 | const uint8_t *qmul; /* Q multiplier table (for both) */ | |
354 | ||
355 | p = ptrs[disks-2]; | |
356 | q = ptrs[disks-1]; | |
357 | ||
358 | /* Compute syndrome with zero for the missing data pages | |
359 | Use the dead data pages as temporary storage for | |
360 | delta p and delta q */ | |
361 | dp = ptrs[faila]; | |
362 | ptrs[faila] = zero; | |
363 | dq = ptrs[failb]; | |
364 | ptrs[failb] = zero; | |
365 | ||
366 | qsyndrome(dp, dq, ptrs, disks-2, bytes); | |
367 | ||
368 | /* Restore pointer table */ | |
369 | ptrs[faila] = dp; | |
370 | ptrs[failb] = dq; | |
371 | ||
372 | /* Now, pick the proper data tables */ | |
373 | pbmul = raid6_gfmul[raid6_gfexi[failb-faila]]; | |
374 | qmul = raid6_gfmul[raid6_gfinv[raid6_gfexp[faila]^raid6_gfexp[failb]]]; | |
375 | ||
376 | /* Now do it... */ | |
377 | while ( bytes-- ) { | |
378 | px = *p ^ *dp; | |
379 | qx = qmul[*q ^ *dq]; | |
380 | *dq++ = db = pbmul[px] ^ qx; /* Reconstructed B */ | |
381 | *dp++ = db ^ px; /* Reconstructed A */ | |
382 | p++; q++; | |
383 | } | |
384 | } | |
385 | ||
386 | /* Recover failure of one data block plus the P block */ | |
387 | void raid6_datap_recov(int disks, size_t bytes, int faila, uint8_t **ptrs) | |
388 | { | |
389 | uint8_t *p, *q, *dq; | |
390 | const uint8_t *qmul; /* Q multiplier table */ | |
391 | ||
392 | p = ptrs[disks-2]; | |
393 | q = ptrs[disks-1]; | |
394 | ||
395 | /* Compute syndrome with zero for the missing data page | |
396 | Use the dead data page as temporary storage for delta q */ | |
397 | dq = ptrs[faila]; | |
398 | ptrs[faila] = zero; | |
399 | ||
400 | qsyndrome(p, dq, ptrs, disks-2, bytes); | |
401 | ||
402 | /* Restore pointer table */ | |
403 | ptrs[faila] = dq; | |
404 | ||
405 | /* Now, pick the proper data tables */ | |
406 | qmul = raid6_gfmul[raid6_gfinv[raid6_gfexp[faila]]]; | |
407 | ||
408 | /* Now do it... */ | |
409 | while ( bytes-- ) { | |
410 | *p++ ^= *dq = qmul[*q ^ *dq]; | |
411 | q++; dq++; | |
412 | } | |
413 | } | |
414 | ||
9d0e7840 PS |
415 | /* Try to find out if a specific disk has a problem */ |
416 | int raid6_check_disks(int data_disks, int start, int chunk_size, | |
417 | int level, int layout, int diskP, int diskQ, | |
418 | char *p, char *q, char **stripes) | |
419 | { | |
420 | int i; | |
421 | int data_id, diskD; | |
422 | uint8_t Px, Qx; | |
423 | int curr_broken_disk = -1; | |
424 | int prev_broken_disk = -1; | |
425 | int broken_status = 0; | |
426 | ||
427 | for(i = 0; i < chunk_size; i++) { | |
428 | Px = (uint8_t)stripes[diskP][i] ^ (uint8_t)p[i]; | |
429 | Qx = (uint8_t)stripes[diskQ][i] ^ (uint8_t)q[i]; | |
430 | ||
431 | if((Px != 0) && (Qx == 0)) | |
432 | curr_broken_disk = diskP; | |
433 | ||
9d0e7840 PS |
434 | if((Px == 0) && (Qx != 0)) |
435 | curr_broken_disk = diskQ; | |
436 | ||
9d0e7840 | 437 | if((Px != 0) && (Qx != 0)) { |
c4db5301 PS |
438 | data_id = (raid6_gflog[Qx] - raid6_gflog[Px]); |
439 | if(data_id < 0) data_id += 255; | |
9d0e7840 PS |
440 | diskD = geo_map(data_id, start/chunk_size, |
441 | data_disks + 2, level, layout); | |
442 | curr_broken_disk = diskD; | |
443 | } | |
444 | ||
445 | if((Px == 0) && (Qx == 0)) | |
446 | curr_broken_disk = curr_broken_disk; | |
447 | ||
c4db5301 PS |
448 | if(curr_broken_disk >= data_disks + 2) |
449 | broken_status = 2; | |
450 | ||
9d0e7840 PS |
451 | switch(broken_status) { |
452 | case 0: | |
453 | if(curr_broken_disk != -1) { | |
454 | prev_broken_disk = curr_broken_disk; | |
455 | broken_status = 1; | |
456 | } | |
457 | break; | |
458 | ||
459 | case 1: | |
460 | if(curr_broken_disk != prev_broken_disk) | |
461 | broken_status = 2; | |
9d0e7840 PS |
462 | break; |
463 | ||
464 | case 2: | |
465 | default: | |
466 | curr_broken_disk = prev_broken_disk = -2; | |
467 | break; | |
468 | } | |
469 | } | |
470 | ||
471 | return curr_broken_disk; | |
472 | } | |
473 | ||
2fcb75ae AK |
474 | /******************************************************************************* |
475 | * Function: save_stripes | |
476 | * Description: | |
477 | * Function reads data (only data without P and Q) from array and writes | |
478 | * it to buf and opcjonaly to backup files | |
479 | * Parameters: | |
480 | * source : A list of 'fds' of the active disks. | |
481 | * Some may be absent | |
482 | * offsets : A list of offsets on disk belonging | |
483 | * to the array [bytes] | |
484 | * raid_disks : geometry: number of disks in the array | |
485 | * chunk_size : geometry: chunk size [bytes] | |
486 | * level : geometry: RAID level | |
487 | * layout : geometry: layout | |
488 | * nwrites : number of backup files | |
489 | * dest : A list of 'fds' for mirrored targets | |
490 | * (e.g. backup files). They are already seeked to right | |
491 | * (write) location. If NULL, data will be wrote | |
492 | * to the buf only | |
493 | * start : start address of data to read (must be stripe-aligned) | |
494 | * [bytes] | |
495 | * length - : length of data to read (must be stripe-aligned) | |
496 | * [bytes] | |
497 | * buf : buffer for data. It is large enough to hold | |
498 | * one stripe. It is stripe aligned | |
499 | * Returns: | |
500 | * 0 : success | |
501 | * -1 : fail | |
502 | ******************************************************************************/ | |
e86c9dd6 NB |
503 | int save_stripes(int *source, unsigned long long *offsets, |
504 | int raid_disks, int chunk_size, int level, int layout, | |
505 | int nwrites, int *dest, | |
a6288483 N |
506 | unsigned long long start, unsigned long long length, |
507 | char *buf) | |
e86c9dd6 | 508 | { |
e86c9dd6 NB |
509 | int len; |
510 | int data_disks = raid_disks - (level == 0 ? 0 : level <=5 ? 1 : 2); | |
511 | int disk; | |
a6288483 | 512 | int i; |
2fcb75ae | 513 | unsigned long long length_test; |
e86c9dd6 | 514 | |
a6288483 N |
515 | if (!tables_ready) |
516 | make_tables(); | |
59679536 | 517 | ensure_zero_has_size(chunk_size); |
a6288483 N |
518 | |
519 | len = data_disks * chunk_size; | |
2fcb75ae AK |
520 | length_test = length / len; |
521 | length_test *= len; | |
522 | ||
523 | if (length != length_test) { | |
524 | dprintf("Error: save_stripes(): Data are not alligned. EXIT\n"); | |
525 | dprintf("\tArea for saving stripes (length) = %llu\n", length); | |
526 | dprintf("\tWork step (len) = %i\n", len); | |
527 | dprintf("\tExpected save area (length_test) = %llu\n", | |
528 | length_test); | |
529 | abort(); | |
530 | } | |
531 | ||
e86c9dd6 | 532 | while (length > 0) { |
a6288483 N |
533 | int failed = 0; |
534 | int fdisk[3], fblock[3]; | |
535 | for (disk = 0; disk < raid_disks ; disk++) { | |
536 | unsigned long long offset; | |
537 | int dnum; | |
a6288483 N |
538 | |
539 | offset = (start/chunk_size/data_disks)*chunk_size; | |
540 | dnum = geo_map(disk < data_disks ? disk : data_disks - disk - 1, | |
541 | start/chunk_size/data_disks, | |
542 | raid_disks, level, layout); | |
7236ee7a | 543 | if (dnum < 0) abort(); |
a6288483 | 544 | if (source[dnum] < 0 || |
cc50ccdc | 545 | lseek64(source[dnum], offsets[dnum]+offset, 0) < 0 || |
7236ee7a N |
546 | read(source[dnum], buf+disk * chunk_size, chunk_size) |
547 | != chunk_size) | |
a6288483 N |
548 | if (failed <= 2) { |
549 | fdisk[failed] = dnum; | |
550 | fblock[failed] = disk; | |
551 | failed++; | |
552 | } | |
553 | } | |
554 | if (failed == 0 || fblock[0] >= data_disks) | |
555 | /* all data disks are good */ | |
556 | ; | |
557 | else if (failed == 1 || fblock[1] >= data_disks+1) { | |
558 | /* one failed data disk and good parity */ | |
559 | char *bufs[data_disks]; | |
560 | for (i=0; i < data_disks; i++) | |
561 | if (fblock[0] == i) | |
562 | bufs[i] = buf + data_disks*chunk_size; | |
563 | else | |
564 | bufs[i] = buf + i*chunk_size; | |
565 | ||
566 | xor_blocks(buf + fblock[0]*chunk_size, | |
567 | bufs, data_disks, chunk_size); | |
568 | } else if (failed > 2 || level != 6) | |
569 | /* too much failure */ | |
e86c9dd6 | 570 | return -1; |
a6288483 N |
571 | else { |
572 | /* RAID6 computations needed. */ | |
573 | uint8_t *bufs[data_disks+4]; | |
574 | int qdisk; | |
575 | int syndrome_disks; | |
576 | disk = geo_map(-1, start/chunk_size/data_disks, | |
577 | raid_disks, level, layout); | |
578 | qdisk = geo_map(-2, start/chunk_size/data_disks, | |
579 | raid_disks, level, layout); | |
580 | if (is_ddf(layout)) { | |
581 | /* q over 'raid_disks' blocks, in device order. | |
582 | * 'p' and 'q' get to be all zero | |
583 | */ | |
584 | for (i = 0; i < raid_disks; i++) | |
cc50ccdc N |
585 | bufs[i] = zero; |
586 | for (i = 0; i < data_disks; i++) { | |
587 | int dnum = geo_map(i, | |
588 | start/chunk_size/data_disks, | |
589 | raid_disks, level, layout); | |
590 | int snum; | |
591 | /* i is the logical block number, so is index to 'buf'. | |
592 | * dnum is physical disk number | |
593 | * and thus the syndrome number. | |
594 | */ | |
595 | snum = dnum; | |
596 | bufs[snum] = (uint8_t*)buf + chunk_size * i; | |
597 | } | |
a6288483 N |
598 | syndrome_disks = raid_disks; |
599 | } else { | |
600 | /* for md, q is over 'data_disks' blocks, | |
601 | * starting immediately after 'q' | |
1eac9f84 N |
602 | * Note that for the '_6' variety, the p block |
603 | * makes a hole that we need to be careful of. | |
a6288483 | 604 | */ |
1eac9f84 N |
605 | int j; |
606 | int snum = 0; | |
607 | for (j = 0; j < raid_disks; j++) { | |
608 | int dnum = (qdisk + 1 + j) % raid_disks; | |
609 | if (dnum == disk || dnum == qdisk) | |
610 | continue; | |
611 | for (i = 0; i < data_disks; i++) | |
612 | if (geo_map(i, | |
613 | start/chunk_size/data_disks, | |
614 | raid_disks, level, layout) == dnum) | |
615 | break; | |
cc50ccdc N |
616 | /* i is the logical block number, so is index to 'buf'. |
617 | * dnum is physical disk number | |
618 | * snum is syndrome disk for which 0 is immediately after Q | |
619 | */ | |
cc50ccdc | 620 | bufs[snum] = (uint8_t*)buf + chunk_size * i; |
1eac9f84 N |
621 | |
622 | if (fblock[0] == i) | |
623 | fdisk[0] = snum; | |
624 | if (fblock[1] == i) | |
625 | fdisk[1] = snum; | |
626 | snum++; | |
cc50ccdc | 627 | } |
a6288483 | 628 | |
a6288483 N |
629 | syndrome_disks = data_disks; |
630 | } | |
cc50ccdc N |
631 | |
632 | /* Place P and Q blocks at end of bufs */ | |
633 | bufs[syndrome_disks] = (uint8_t*)buf + chunk_size * data_disks; | |
634 | bufs[syndrome_disks+1] = (uint8_t*)buf + chunk_size * (data_disks+1); | |
635 | ||
a6288483 N |
636 | if (fblock[1] == data_disks) |
637 | /* One data failed, and parity failed */ | |
638 | raid6_datap_recov(syndrome_disks+2, chunk_size, | |
639 | fdisk[0], bufs); | |
cc50ccdc N |
640 | else { |
641 | if (fdisk[0] > fdisk[1]) { | |
642 | int t = fdisk[0]; | |
643 | fdisk[0] = fdisk[1]; | |
644 | fdisk[1] = t; | |
645 | } | |
a6288483 N |
646 | /* Two data blocks failed, P,Q OK */ |
647 | raid6_2data_recov(syndrome_disks+2, chunk_size, | |
648 | fdisk[0], fdisk[1], bufs); | |
cc50ccdc | 649 | } |
a6288483 | 650 | } |
ccced3dc | 651 | if (dest) { |
2fcb75ae AK |
652 | for (i = 0; i < nwrites; i++) |
653 | if (write(dest[i], buf, len) != len) | |
654 | return -1; | |
ccced3dc AK |
655 | } else { |
656 | /* build next stripe in buffer */ | |
657 | buf += len; | |
658 | } | |
e86c9dd6 NB |
659 | length -= len; |
660 | start += len; | |
e86c9dd6 NB |
661 | } |
662 | return 0; | |
663 | } | |
664 | ||
665 | /* Restore data: | |
666 | * We are given: | |
667 | * A list of 'fds' of the active disks. Some may be '-1' for not-available. | |
353632d9 | 668 | * A geometry: raid_disks, chunk_size, level, layout |
e86c9dd6 NB |
669 | * An 'fd' to read from. It is already seeked to the right (Read) location. |
670 | * A start and length. | |
671 | * The length must be a multiple of the stripe size. | |
672 | * | |
673 | * We build a full stripe in memory and then write it out. | |
674 | * We assume that there are enough working devices. | |
675 | */ | |
676 | int restore_stripes(int *dest, unsigned long long *offsets, | |
677 | int raid_disks, int chunk_size, int level, int layout, | |
353632d9 | 678 | int source, unsigned long long read_offset, |
2fcb75ae AK |
679 | unsigned long long start, unsigned long long length, |
680 | char *src_buf) | |
e86c9dd6 | 681 | { |
e9e43ec3 | 682 | char *stripe_buf; |
503975b9 N |
683 | char **stripes = xmalloc(raid_disks * sizeof(char*)); |
684 | char **blocks = xmalloc(raid_disks * sizeof(char*)); | |
e86c9dd6 | 685 | int i; |
758be4f1 | 686 | int rv; |
e86c9dd6 | 687 | |
a6288483 | 688 | int data_disks = raid_disks - (level == 0 ? 0 : level <= 5 ? 1 : 2); |
e86c9dd6 | 689 | |
fcf57625 N |
690 | if (posix_memalign((void**)&stripe_buf, 4096, raid_disks * chunk_size)) |
691 | stripe_buf = NULL; | |
d47a2925 N |
692 | |
693 | if (zero == NULL || chunk_size > zero_size) { | |
694 | if (zero) | |
695 | free(zero); | |
503975b9 | 696 | zero = xcalloc(1, chunk_size); |
d47a2925 | 697 | zero_size = chunk_size; |
a6288483 | 698 | } |
d47a2925 | 699 | |
e0d95aac N |
700 | if (stripe_buf == NULL || stripes == NULL || blocks == NULL |
701 | || zero == NULL) { | |
758be4f1 LD |
702 | rv = -2; |
703 | goto abort; | |
e86c9dd6 | 704 | } |
2fcb75ae | 705 | for (i = 0; i < raid_disks; i++) |
e86c9dd6 NB |
706 | stripes[i] = stripe_buf + i * chunk_size; |
707 | while (length > 0) { | |
f21e18ca | 708 | unsigned int len = data_disks * chunk_size; |
e86c9dd6 | 709 | unsigned long long offset; |
48327135 | 710 | int disk, qdisk; |
a6288483 | 711 | int syndrome_disks; |
758be4f1 LD |
712 | if (length < len) { |
713 | rv = -3; | |
714 | goto abort; | |
715 | } | |
2fcb75ae | 716 | for (i = 0; i < data_disks; i++) { |
e86c9dd6 NB |
717 | int disk = geo_map(i, start/chunk_size/data_disks, |
718 | raid_disks, level, layout); | |
2fcb75ae AK |
719 | if (src_buf == NULL) { |
720 | /* read from file */ | |
758be4f1 LD |
721 | if (lseek64(source, read_offset, 0) != |
722 | (off64_t)read_offset) { | |
723 | rv = -1; | |
724 | goto abort; | |
725 | } | |
2fcb75ae AK |
726 | if (read(source, |
727 | stripes[disk], | |
758be4f1 LD |
728 | chunk_size) != chunk_size) { |
729 | rv = -1; | |
730 | goto abort; | |
731 | } | |
2fcb75ae AK |
732 | } else { |
733 | /* read from input buffer */ | |
734 | memcpy(stripes[disk], | |
735 | src_buf + read_offset, | |
736 | chunk_size); | |
737 | } | |
353632d9 | 738 | read_offset += chunk_size; |
e86c9dd6 NB |
739 | } |
740 | /* We have the data, now do the parity */ | |
741 | offset = (start/chunk_size/data_disks) * chunk_size; | |
48327135 NB |
742 | switch (level) { |
743 | case 4: | |
744 | case 5: | |
745 | disk = geo_map(-1, start/chunk_size/data_disks, | |
e86c9dd6 | 746 | raid_disks, level, layout); |
e0d95aac N |
747 | for (i = 0; i < data_disks; i++) |
748 | blocks[i] = stripes[(disk+1+i) % raid_disks]; | |
e86c9dd6 | 749 | xor_blocks(stripes[disk], blocks, data_disks, chunk_size); |
48327135 NB |
750 | break; |
751 | case 6: | |
752 | disk = geo_map(-1, start/chunk_size/data_disks, | |
753 | raid_disks, level, layout); | |
754 | qdisk = geo_map(-2, start/chunk_size/data_disks, | |
755 | raid_disks, level, layout); | |
e0d95aac N |
756 | if (is_ddf(layout)) { |
757 | /* q over 'raid_disks' blocks, in device order. | |
758 | * 'p' and 'q' get to be all zero | |
759 | */ | |
760 | for (i = 0; i < raid_disks; i++) | |
761 | if (i == disk || i == qdisk) | |
a6288483 | 762 | blocks[i] = (char*)zero; |
e0d95aac N |
763 | else |
764 | blocks[i] = stripes[i]; | |
a6288483 | 765 | syndrome_disks = raid_disks; |
e0d95aac | 766 | } else { |
a6288483 | 767 | /* for md, q is over 'data_disks' blocks, |
e0d95aac N |
768 | * starting immediately after 'q' |
769 | */ | |
770 | for (i = 0; i < data_disks; i++) | |
771 | blocks[i] = stripes[(qdisk+1+i) % raid_disks]; | |
48327135 | 772 | |
a6288483 | 773 | syndrome_disks = data_disks; |
e0d95aac | 774 | } |
a6288483 | 775 | qsyndrome((uint8_t*)stripes[disk], |
1011e834 | 776 | (uint8_t*)stripes[qdisk], |
a6288483 N |
777 | (uint8_t**)blocks, |
778 | syndrome_disks, chunk_size); | |
48327135 | 779 | break; |
e86c9dd6 NB |
780 | } |
781 | for (i=0; i < raid_disks ; i++) | |
782 | if (dest[i] >= 0) { | |
758be4f1 LD |
783 | if (lseek64(dest[i], |
784 | offsets[i]+offset, 0) < 0) { | |
785 | rv = -1; | |
786 | goto abort; | |
787 | } | |
788 | if (write(dest[i], stripes[i], | |
789 | chunk_size) != chunk_size) { | |
790 | rv = -1; | |
791 | goto abort; | |
792 | } | |
e86c9dd6 NB |
793 | } |
794 | length -= len; | |
795 | start += len; | |
796 | } | |
758be4f1 LD |
797 | rv = 0; |
798 | ||
799 | abort: | |
800 | free(stripe_buf); | |
801 | free(stripes); | |
802 | free(blocks); | |
803 | return rv; | |
e86c9dd6 NB |
804 | } |
805 | ||
806 | #ifdef MAIN | |
807 | ||
48327135 NB |
808 | int test_stripes(int *source, unsigned long long *offsets, |
809 | int raid_disks, int chunk_size, int level, int layout, | |
810 | unsigned long long start, unsigned long long length) | |
811 | { | |
812 | /* ready the data and p (and q) blocks, and check we got them right */ | |
503975b9 N |
813 | char *stripe_buf = xmalloc(raid_disks * chunk_size); |
814 | char **stripes = xmalloc(raid_disks * sizeof(char*)); | |
815 | char **blocks = xmalloc(raid_disks * sizeof(char*)); | |
816 | char *p = xmalloc(chunk_size); | |
817 | char *q = xmalloc(chunk_size); | |
48327135 NB |
818 | |
819 | int i; | |
9d0e7840 | 820 | int diskP, diskQ; |
48327135 | 821 | int data_disks = raid_disks - (level == 5 ? 1: 2); |
9d0e7840 PS |
822 | |
823 | if (!tables_ready) | |
824 | make_tables(); | |
825 | ||
48327135 NB |
826 | for ( i = 0 ; i < raid_disks ; i++) |
827 | stripes[i] = stripe_buf + i * chunk_size; | |
828 | ||
829 | while (length > 0) { | |
830 | int disk; | |
831 | ||
832 | for (i = 0 ; i < raid_disks ; i++) { | |
833 | lseek64(source[i], offsets[i]+start, 0); | |
834 | read(source[i], stripes[i], chunk_size); | |
835 | } | |
836 | for (i = 0 ; i < data_disks ; i++) { | |
837 | int disk = geo_map(i, start/chunk_size, raid_disks, | |
838 | level, layout); | |
839 | blocks[i] = stripes[disk]; | |
840 | printf("%d->%d\n", i, disk); | |
841 | } | |
842 | switch(level) { | |
843 | case 6: | |
521f349c | 844 | qsyndrome(p, q, (uint8_t**)blocks, data_disks, chunk_size); |
9d0e7840 | 845 | diskP = geo_map(-1, start/chunk_size, raid_disks, |
48327135 | 846 | level, layout); |
9d0e7840 PS |
847 | if (memcmp(p, stripes[diskP], chunk_size) != 0) { |
848 | printf("P(%d) wrong at %llu\n", diskP, | |
48327135 NB |
849 | start / chunk_size); |
850 | } | |
9d0e7840 | 851 | diskQ = geo_map(-2, start/chunk_size, raid_disks, |
48327135 | 852 | level, layout); |
9d0e7840 PS |
853 | if (memcmp(q, stripes[diskQ], chunk_size) != 0) { |
854 | printf("Q(%d) wrong at %llu\n", diskQ, | |
48327135 NB |
855 | start / chunk_size); |
856 | } | |
9d0e7840 PS |
857 | disk = raid6_check_disks(data_disks, start, chunk_size, |
858 | level, layout, diskP, diskQ, | |
859 | p, q, stripes); | |
860 | if(disk >= 0) { | |
861 | printf("Possible failed disk: %d\n", disk); | |
862 | } | |
863 | if(disk == -2) { | |
864 | printf("Failure detected, but disk unknown\n"); | |
865 | } | |
48327135 NB |
866 | break; |
867 | } | |
868 | length -= chunk_size; | |
869 | start += chunk_size; | |
870 | } | |
871 | return 0; | |
872 | } | |
873 | ||
e86c9dd6 NB |
874 | unsigned long long getnum(char *str, char **err) |
875 | { | |
876 | char *e; | |
877 | unsigned long long rv = strtoull(str, &e, 10); | |
878 | if (e==str || *e) { | |
879 | *err = str; | |
880 | return 0; | |
881 | } | |
882 | return rv; | |
883 | } | |
884 | ||
885 | main(int argc, char *argv[]) | |
886 | { | |
887 | /* save/restore file raid_disks chunk_size level layout start length devices... | |
888 | */ | |
889 | int save; | |
890 | int *fds; | |
891 | char *file; | |
a6288483 | 892 | char *buf; |
e86c9dd6 NB |
893 | int storefd; |
894 | unsigned long long *offsets; | |
895 | int raid_disks, chunk_size, level, layout; | |
896 | unsigned long long start, length; | |
897 | int i; | |
898 | ||
899 | char *err = NULL; | |
900 | if (argc < 10) { | |
901 | fprintf(stderr, "Usage: test_stripe save/restore file raid_disks" | |
902 | " chunk_size level layout start length devices...\n"); | |
903 | exit(1); | |
904 | } | |
905 | if (strcmp(argv[1], "save")==0) | |
906 | save = 1; | |
907 | else if (strcmp(argv[1], "restore") == 0) | |
908 | save = 0; | |
48327135 NB |
909 | else if (strcmp(argv[1], "test") == 0) |
910 | save = 2; | |
e86c9dd6 NB |
911 | else { |
912 | fprintf(stderr, "test_stripe: must give 'save' or 'restore'.\n"); | |
913 | exit(2); | |
914 | } | |
915 | ||
916 | file = argv[2]; | |
917 | raid_disks = getnum(argv[3], &err); | |
918 | chunk_size = getnum(argv[4], &err); | |
919 | level = getnum(argv[5], &err); | |
920 | layout = getnum(argv[6], &err); | |
921 | start = getnum(argv[7], &err); | |
922 | length = getnum(argv[8], &err); | |
923 | if (err) { | |
924 | fprintf(stderr, "test_stripe: Bad number: %s\n", err); | |
925 | exit(2); | |
926 | } | |
927 | if (argc != raid_disks + 9) { | |
928 | fprintf(stderr, "test_stripe: wrong number of devices: want %d found %d\n", | |
929 | raid_disks, argc-9); | |
930 | exit(2); | |
931 | } | |
503975b9 N |
932 | fds = xmalloc(raid_disks * sizeof(*fds)); |
933 | offsets = xcalloc(raid_disks, sizeof(*offsets)); | |
e86c9dd6 NB |
934 | |
935 | storefd = open(file, O_RDWR); | |
936 | if (storefd < 0) { | |
937 | perror(file); | |
938 | fprintf(stderr, "test_stripe: could not open %s.\n", file); | |
939 | exit(3); | |
940 | } | |
941 | for (i=0; i<raid_disks; i++) { | |
6f38d7ae PS |
942 | char *p; |
943 | p = strchr(argv[9+i], ':'); | |
944 | ||
945 | if(p != NULL) { | |
946 | *p++ = '\0'; | |
947 | offsets[i] = atoll(p) * 512; | |
948 | } | |
1011e834 | 949 | |
e86c9dd6 NB |
950 | fds[i] = open(argv[9+i], O_RDWR); |
951 | if (fds[i] < 0) { | |
952 | perror(argv[9+i]); | |
953 | fprintf(stderr,"test_stripe: cannot open %s.\n", argv[9+i]); | |
954 | exit(3); | |
955 | } | |
956 | } | |
957 | ||
503975b9 | 958 | buf = xmalloc(raid_disks * chunk_size); |
a6288483 | 959 | |
48327135 | 960 | if (save == 1) { |
e86c9dd6 NB |
961 | int rv = save_stripes(fds, offsets, |
962 | raid_disks, chunk_size, level, layout, | |
963 | 1, &storefd, | |
a6288483 | 964 | start, length, buf); |
e86c9dd6 | 965 | if (rv != 0) { |
48327135 NB |
966 | fprintf(stderr, |
967 | "test_stripe: save_stripes returned %d\n", rv); | |
968 | exit(1); | |
969 | } | |
970 | } else if (save == 2) { | |
971 | int rv = test_stripes(fds, offsets, | |
972 | raid_disks, chunk_size, level, layout, | |
973 | start, length); | |
974 | if (rv != 0) { | |
975 | fprintf(stderr, | |
976 | "test_stripe: test_stripes returned %d\n", rv); | |
e86c9dd6 NB |
977 | exit(1); |
978 | } | |
979 | } else { | |
980 | int rv = restore_stripes(fds, offsets, | |
981 | raid_disks, chunk_size, level, layout, | |
353632d9 | 982 | storefd, 0ULL, |
c071a1cd | 983 | start, length, NULL); |
e86c9dd6 | 984 | if (rv != 0) { |
48327135 NB |
985 | fprintf(stderr, |
986 | "test_stripe: restore_stripes returned %d\n", | |
987 | rv); | |
e86c9dd6 NB |
988 | exit(1); |
989 | } | |
990 | } | |
991 | exit(0); | |
992 | } | |
993 | ||
994 | #endif /* MAIN */ |