9394 format: this statement may fall through
[unleashed.git] / usr / src / cmd / format / analyze.c
blobe887058a2c14cad66b37423b39dee0694dddfc75
1 /*
2 * CDDL HEADER START
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
19 * CDDL HEADER END
22 * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
27 * This file contains routines to analyze the surface of a disk.
29 #include "global.h"
30 #include "analyze.h"
31 #include <stdlib.h>
32 #include <errno.h>
33 #include "misc.h"
34 #include "defect.h"
35 #include "label.h"
36 #include "param.h"
37 #include "checkdev.h"
41 * These global variables control the surface analysis process. They
42 * are set from a command in the defect menu.
44 int scan_entire = 1; /* scan whole disk flag */
45 diskaddr_t scan_lower = 0; /* lower bound */
46 diskaddr_t scan_upper = 0; /* upper bound */
47 int scan_correct = 1; /* correct errors flag */
48 int scan_stop = 0; /* stop after error flag */
49 int scan_loop = 0; /* loop forever flag */
50 int scan_passes = 2; /* number of passes */
51 int scan_random = 0; /* random patterns flag */
52 uint_t scan_size = 0; /* sectors/scan operation */
53 int scan_auto = 1; /* scan after format flag */
54 int scan_restore_defects = 1; /* restore defect list after writing */
55 int scan_restore_label = 1; /* restore label after writing */
58 * These are summary variables to print out info after analysis.
59 * Values less than 0 imply they are invalid.
61 offset_t scan_cur_block = -1; /* current block */
62 int64_t scan_blocks_fixed = -1; /* # blocks repaired */
65 * This variable is used to tell whether the most recent surface
66 * analysis error was caused by a media defect or some other problem.
68 int media_error; /* error was caused by defect */
70 int disk_error; /* disk errors during analysis */
73 * These are the data patterns used if random patterns are not chosen.
74 * They are designed to show pattern dependent errors.
76 static unsigned int scan_patterns[] = {
77 0xc6dec6de,
78 0x6db6db6d,
79 0x00000000,
80 0xffffffff,
81 0xaaaaaaaa,
83 #define NPATTERNS 5 /* number of predefined patterns */
86 * These are the data patterns from the SunFed requirements document.
88 static unsigned int purge_patterns[] = { /* patterns to be written */
89 0xaaaaaaaa, /* 10101010... */
90 0x55555555, /* 01010101... == UUUU... */
91 0xaaaaaaaa, /* 10101010... */
92 0xaaaaaaaa, /* 10101010... */
95 static unsigned int alpha_pattern = 0x40404040; /* 10000000... == @@@@... */
97 /* Function prototypes */
98 #ifdef __STDC__
100 static int scan_repair(diskaddr_t bn, int mode);
101 static int analyze_blocks(int flags, diskaddr_t blkno, uint_t blkcnt,
102 unsigned data, int init, int driver_flags, int *xfercntp);
103 static int handle_error_conditions(void);
104 static int verify_blocks(int flags, diskaddr_t blkno, uint_t blkcnt,
105 unsigned data, int driver_flags, int *xfercntp);
106 #else /* __STDC__ */
108 static int scan_repair();
109 static int analyze_blocks();
110 static int handle_error_conditions();
111 static int verify_blocks();
113 #endif /* __STDC__ */
116 * This routine performs a surface analysis based upon the global
117 * parameters. It is called from several commands in the defect menu,
118 * and from the format command in the command menu (if post-format
119 * analysis is enable).
122 do_scan(flags, mode)
123 int flags, mode;
125 diskaddr_t start, end, curnt;
126 int pass, needinit, data;
127 uint_t size;
128 int status, founderr, i, j;
129 int error = 0;
130 int pattern = 0;
131 int xfercnt;
134 * Check to be sure we aren't correcting without a defect list
135 * if the controller can correct the defect.
137 if (scan_correct && !EMBEDDED_SCSI && (cur_ops->op_repair != NULL) &&
138 (cur_list.list == NULL)) {
139 err_print("Current Defect List must be initialized ");
140 err_print("to do automatic repair.\n");
141 return (-1);
144 * Define the bounds of the scan.
146 if (scan_entire) {
147 start = 0;
148 if (cur_label == L_TYPE_SOLARIS) {
149 if (cur_ctype->ctype_flags & CF_SCSI)
150 end = datasects() - 1;
151 else
152 end = physsects() - 1;
153 } else if (cur_label == L_TYPE_EFI) {
154 end = cur_parts->etoc->efi_last_lba;
156 } else {
157 start = scan_lower;
158 end = scan_upper;
161 * Make sure the user knows if we are scanning over a mounted
162 * partition.
164 if ((flags & (SCAN_PATTERN | SCAN_WRITE)) &&
165 (checkmount(start, end))) {
166 err_print("Cannot do analysis on a mounted partition.\n");
167 return (-1);
171 * Make sure the user knows if we are scanning over a
172 * partition being used for swapping.
174 if ((flags & (SCAN_PATTERN | SCAN_WRITE)) &&
175 (checkswap(start, end))) {
176 err_print("Cannot do analysis on a partition \
177 which is currently being used for swapping.\n");
178 return (-1);
182 * Check to see if any partitions used for svm, vxvm, ZFS zpool
183 * or live upgrade are on the disk.
185 if ((flags & (SCAN_PATTERN | SCAN_WRITE)) &&
186 (checkdevinuse(cur_disk->disk_name, (diskaddr_t)-1,
187 (diskaddr_t)-1, 0, 0))) {
188 err_print("Cannot do analysis on a partition "
189 "while it in use as described above.\n");
190 return (-1);
194 * If we are scanning destructively over certain sectors,
195 * we mark the defect list and/or label dirty so it will get rewritten.
197 if (flags & (SCAN_PATTERN | SCAN_WRITE)) {
198 if (cur_label == L_TYPE_SOLARIS) {
199 if (start < (diskaddr_t)totalsects() &&
200 end >= (diskaddr_t)datasects()) {
201 if (!EMBEDDED_SCSI) {
202 cur_list.flags |= LIST_DIRTY;
204 if (cur_disk->disk_flags & DSK_LABEL)
205 cur_flags |= LABEL_DIRTY;
208 if (start == 0) {
209 if (cur_disk->disk_flags & DSK_LABEL)
210 cur_flags |= LABEL_DIRTY;
214 * Initialize the summary info on sectors repaired.
216 scan_blocks_fixed = 0;
218 * Loop through the passes of the scan. If required, loop forever.
220 for (pass = 0; pass < scan_passes || scan_loop; pass++) {
222 * Determine the data pattern to use if pattern testing
223 * is to be done.
225 if (flags & SCAN_PATTERN) {
226 if (scan_random)
227 data = (int)mrand48();
228 else
229 data = scan_patterns[pass % NPPATTERNS];
231 if (flags & SCAN_PURGE) {
232 flags &= ~(SCAN_PURGE_READ_PASS
233 | SCAN_PURGE_ALPHA_PASS);
234 switch (pattern % (NPPATTERNS + 1)) {
235 case NPPATTERNS:
236 pattern = 0;
237 if (!error) {
238 fmt_print(
239 "\nThe last %d passes were successful, running alpha pattern pass", NPPATTERNS);
240 flags |= SCAN_PURGE_ALPHA_PASS;
241 data = alpha_pattern;
242 } else {
243 data = purge_patterns[pattern];
244 pattern++;
246 break;
247 case READPATTERN:
248 flags |= SCAN_PURGE_READ_PASS;
249 /* FALLTHROUGH */
250 default:
251 data = purge_patterns[pattern];
252 pattern++;
253 break;
256 fmt_print("\n pass %d", pass);
257 fmt_print(" - pattern = 0x%x", data);
258 } else
259 fmt_print("\n pass %d", pass);
261 fmt_print("\n");
263 * Mark the pattern buffer as corrupt, since it
264 * hasn't been initialized.
266 needinit = 1;
268 * Print the first block number to the log file if
269 * logging is on so there is some record of what
270 * analysis was performed.
272 if (log_file) {
273 pr_dblock(log_print, start);
274 log_print("\n");
277 * Loop through this pass, each time analyzing an amount
278 * specified by the global parameters.
280 xfercnt = 0;
281 for (curnt = start; curnt <= end; curnt += size) {
282 if ((end - curnt) < scan_size)
283 size = end - curnt + 1;
284 else
285 size = scan_size;
287 * Print out where we are, so we don't look dead.
288 * Also store it in summary info for logging.
290 scan_cur_block = curnt;
291 nolog_print(" ");
292 pr_dblock(nolog_print, curnt);
293 nolog_print(" \015");
294 (void) fflush(stdout);
295 disk_error = 0;
297 * Do the actual analysis.
299 status = analyze_blocks(flags, curnt, size,
300 (unsigned)data, needinit, (F_ALLERRS | F_SILENT),
301 &xfercnt);
303 * If there were no errors, the pattern buffer is
304 * still initialized, and we just loop to next chunk.
306 needinit = 0;
307 if (!status)
308 continue;
310 * There was an error. Check if surface analysis
311 * can be continued.
313 if (handle_error_conditions()) {
314 scan_blocks_fixed = scan_cur_block = -1;
315 return (-1);
318 * There was an error. Mark the pattern buffer
319 * corrupt so it will get reinitialized.
321 needinit = 1;
323 * If it was not a media error, ignore it.
325 if (!media_error)
326 continue;
328 * Loop 5 times through each sector of the chunk,
329 * analyzing them individually.
331 nolog_print(" ");
332 pr_dblock(nolog_print, curnt);
333 nolog_print(" \015");
334 (void) fflush(stdout);
335 founderr = 0;
336 for (j = 0; j < size * 5; j++) {
337 i = j % size;
338 disk_error = 0;
339 status = analyze_blocks(flags, (curnt + i), 1,
340 (unsigned)data, needinit, F_ALLERRS, NULL);
341 needinit = 0;
342 if (!status)
343 continue;
345 * There was an error. Check if surface analysis
346 * can be continued.
348 if (handle_error_conditions()) {
349 scan_blocks_fixed = scan_cur_block = -1;
350 return (-1);
353 * An error occurred. Mark the buffer
354 * corrupt and see if it was media
355 * related.
357 needinit = 1;
358 if (!media_error)
359 continue;
361 * We found a bad sector. Print out a message
362 * and fix it if required.
364 founderr = 1;
365 if (scan_correct && (flags != SCAN_VALID)) {
366 if (scan_repair(curnt+i, mode)) {
367 error = -1;
369 } else
370 err_print("\n");
372 * Stop after the error if required.
374 if (scan_stop)
375 goto out;
378 * Mark the pattern buffer corrupt to be safe.
380 needinit = 1;
382 * We didn't find an individual sector that was bad.
383 * Print out a warning.
385 if (!founderr) {
386 err_print("Warning: unable to pinpoint ");
387 err_print("defective block.\n");
391 * Print the end of each pass to the log file.
393 enter_critical();
394 if (log_file) {
395 pr_dblock(log_print, scan_cur_block);
396 log_print("\n");
398 scan_cur_block = -1;
399 exit_critical();
400 fmt_print("\n");
403 * alternate the read and write for SCAN_VERIFY test
405 if (flags & SCAN_VERIFY) {
406 flags ^= SCAN_VERIFY_READ_PASS;
409 out:
411 * We got here either by giving up after an error or falling
412 * through after all passes were completed.
414 fmt_print("\n");
415 enter_critical();
417 * If the defect list is dirty, write it to disk,
418 * if scan_restore_defects (the default) is true.
420 if (!EMBEDDED_SCSI && (cur_list.flags & LIST_DIRTY) &&
421 (scan_restore_defects)) {
422 cur_list.flags = 0;
423 write_deflist(&cur_list);
426 * If the label is dirty, write it to disk.
427 * if scan_restore_label (the default) is true.
429 if ((cur_flags & LABEL_DIRTY) && (scan_restore_label)) {
430 cur_flags &= ~LABEL_DIRTY;
431 (void) write_label();
434 * If we dropped down to here after an error, we need to write
435 * the final block number to the log file for record keeping.
437 if (log_file && scan_cur_block >= 0) {
438 pr_dblock(log_print, scan_cur_block);
439 log_print("\n");
441 fmt_print("Total of %lld defective blocks repaired.\n",
442 scan_blocks_fixed);
444 * Reinitialize the logging variables so they don't get used
445 * when they are not really valid.
447 scan_blocks_fixed = scan_cur_block = -1;
448 exit_critical();
449 return (error);
454 * This routine is called to repair a bad block discovered
455 * during a scan operation. Return 0 for success, 1 for failure.
456 * (This has been extracted out of do_scan(), to simplify it.)
458 static int
459 scan_repair(bn, mode)
460 diskaddr_t bn;
461 int mode;
463 int status;
464 int result = 1;
465 char *buf;
466 int buf_is_good;
467 int i;
469 if (cur_ops->op_repair == NULL) {
470 err_print("Warning: Controller does ");
471 err_print("not support repairing.\n\n");
472 return (result);
475 buf = malloc(cur_blksz);
476 if (buf == NULL) {
477 err_print("Warning: no memory.\n\n");
478 return (result);
480 enter_critical();
483 * Determine if the error appears to be hard or soft. We
484 * already assume there's an error. If we can get any
485 * good data out of the sector, write that data back
486 * after the repair.
488 buf_is_good = 0;
489 for (i = 0; i < 5; i++) {
490 status = (*cur_ops->op_rdwr)(DIR_READ, cur_file, bn, 1,
491 buf, F_SILENT, NULL);
492 if (status == 0) {
493 buf_is_good = 1;
494 break;
498 fmt_print("Repairing %s error on %llu (",
499 buf_is_good ? "soft" : "hard", bn);
500 pr_dblock(fmt_print, bn);
501 fmt_print(")...");
503 status = (*cur_ops->op_repair)(bn, mode);
504 if (status) {
506 * If the repair failed, we note it and will return the
507 * failure. However, the analysis goes on.
509 fmt_print("failed.\n\n");
510 } else {
512 * The repair worked. Write the good data we could
513 * recover from the failed block, if possible.
514 * If not, zero the block. In doing so, try to
515 * determine if the new block appears ok.
517 if (!buf_is_good) {
518 bzero(buf, cur_blksz);
519 fmt_print("Warning: Block %llu zero-filled.\n", bn);
520 } else {
521 fmt_print("ok.\n");
523 status = (*cur_ops->op_rdwr)(DIR_WRITE, cur_file, bn,
524 1, buf, (F_SILENT | F_ALLERRS), NULL);
525 if (status == 0) {
526 status = (*cur_ops->op_rdwr)(DIR_READ, cur_file, bn,
527 1, buf, (F_SILENT | F_ALLERRS), NULL);
529 if (status) {
530 fmt_print("The new block also appears defective.\n");
532 fmt_print("\n");
534 * add the defect to the list and write the list out.
535 * Also, kill the working list so it will get resynced
536 * with the current list.
538 * For embedded scsi, we don't require a defect list.
539 * However, if we have one, add the defect if the
540 * list includes the grown list. If not, kill it
541 * to force a resync if we need the list later.
543 if (EMBEDDED_SCSI) {
544 if (cur_list.list != NULL) {
545 if (cur_list.flags & LIST_PGLIST) {
546 add_ldef(bn, &cur_list);
547 } else {
548 kill_deflist(&cur_list);
552 * The next "if" statement reflects the fix for
553 * bug id 1026096 where format keeps adding the
554 * same defect to the defect list.
556 } else if (cur_ctype->ctype_flags & CF_WLIST) {
557 kill_deflist(&cur_list);
558 (*cur_ops->op_ex_cur)(&cur_list);
559 fmt_print("Current list updated\n");
560 } else {
561 add_ldef(bn, &cur_list);
562 write_deflist(&cur_list);
564 kill_deflist(&work_list);
566 /* Log the repair. */
567 scan_blocks_fixed++;
569 /* return ok */
570 result = 0;
573 exit_critical();
574 free(buf);
575 return (result);
580 * This routine analyzes a set of sectors on the disk. It simply returns
581 * an error if a defect is found. It is called by do_scan().
583 static int
584 analyze_blocks(flags, blkno, blkcnt, data, init, driver_flags, xfercntp)
585 int flags, driver_flags, init;
586 uint_t blkcnt;
587 register unsigned data;
588 diskaddr_t blkno;
589 int *xfercntp;
591 int corrupt = 0;
592 int status;
593 register diskaddr_t i, nints;
594 register unsigned *ptr = (uint_t *)pattern_buf;
596 media_error = 0;
597 if (flags & SCAN_VERIFY) {
598 return (verify_blocks(flags, blkno, blkcnt, data,
599 driver_flags, xfercntp));
603 * Initialize the pattern buffer if necessary.
605 nints = (diskaddr_t)blkcnt * cur_blksz / sizeof (int);
606 if ((flags & SCAN_PATTERN) && init) {
607 for (i = 0; i < nints; i++)
608 *((int *)((int *)pattern_buf + i)) = data;
611 * Lock out interrupts so we can insure valid data will get
612 * restored. This is necessary because there are modes
613 * of scanning that corrupt the disk data then restore it at
614 * the end of the analysis.
616 enter_critical();
618 * If the disk data is valid, read it into the data buffer.
620 if (flags & SCAN_VALID) {
621 status = (*cur_ops->op_rdwr)(DIR_READ, cur_file, blkno,
622 blkcnt, (caddr_t)cur_buf, driver_flags, xfercntp);
623 if (status)
624 goto bad;
627 * If we are doing pattern testing, write and read the pattern
628 * from the pattern buffer.
630 if (flags & SCAN_PATTERN) {
632 * If the disk data was valid, mark it corrupt so we know
633 * to restore it later.
635 if (flags & SCAN_VALID)
636 corrupt++;
638 * Only write if we're not on the read pass of SCAN_PURGE.
640 if (!(flags & SCAN_PURGE_READ_PASS)) {
641 status = (*cur_ops->op_rdwr)(DIR_WRITE, cur_file, blkno,
642 blkcnt, (caddr_t)pattern_buf, driver_flags,
643 xfercntp);
644 if (status)
645 goto bad;
648 * Only read if we are on the read pass of SCAN_PURGE, if we
649 * are purging.
651 if ((!(flags & SCAN_PURGE)) || (flags & SCAN_PURGE_READ_PASS)) {
652 status = (*cur_ops->op_rdwr)(DIR_READ, cur_file, blkno,
653 blkcnt, (caddr_t)pattern_buf, driver_flags,
654 xfercntp);
655 if (status)
656 goto bad;
660 * If we are doing a data compare, make sure the pattern
661 * came back intact.
662 * Only compare if we are on the read pass of SCAN_PURGE, or
663 * we wrote random data instead of the expected data pattern.
665 if ((flags & SCAN_COMPARE) || (flags & SCAN_PURGE_READ_PASS)) {
666 for (i = nints, ptr = (uint_t *)pattern_buf; i; i--)
667 if (*ptr++ != data) {
668 err_print("Data miscompare error (expecting ");
669 err_print("0x%x, got 0x%x) at ", data,
670 *((int *)((int *)pattern_buf +
671 (nints - i))));
672 pr_dblock(err_print, blkno);
673 err_print(", offset = 0x%llx.\n",
674 (nints - i) * sizeof (int));
675 goto bad;
679 * If we are supposed to write data out, do so.
681 if (flags & SCAN_WRITE) {
682 status = (*cur_ops->op_rdwr)(DIR_WRITE, cur_file, blkno,
683 blkcnt, (caddr_t)cur_buf, driver_flags, xfercntp);
684 if (status)
685 goto bad;
687 exit_critical();
689 * No errors occurred, return ok.
691 return (0);
692 bad:
694 * There was an error. If the data was corrupted, we write it
695 * out from the data buffer to restore it.
697 if (corrupt) {
698 if ((*cur_ops->op_rdwr)(DIR_WRITE, cur_file, blkno,
699 blkcnt, (caddr_t)cur_buf, F_NORMAL, xfercntp))
700 err_print("Warning: unable to restore original data.\n");
702 exit_critical();
704 * Return the error.
706 return (-1);
711 * This routine analyzes a set of sectors on the disk. It simply returns
712 * an error if a defect is found. It is called by analyze_blocks().
713 * For simplicity, this is done as a separate function instead of
714 * making the analyze_block routine complex.
716 * This routine implements the 'verify' command. It writes the disk
717 * by writing unique data for each block; after the write pass, it
718 * reads the data and verifies for correctness. Note that the entire
719 * disk (or the range of disk) is fully written first and then read.
720 * This should eliminate any caching effect on the drives.
722 static int
723 verify_blocks(int flags,
724 diskaddr_t blkno,
725 uint_t blkcnt,
726 unsigned data,
727 int driver_flags,
728 int *xfercntp)
730 int status, i, nints;
731 unsigned *ptr = (uint_t *)pattern_buf;
733 nints = cur_blksz / sizeof (int);
736 * Initialize the pattern buffer if we are in write pass.
737 * Use the block number itself as data, each block has unique
738 * buffer data that way.
740 if (!(flags & SCAN_VERIFY_READ_PASS)) {
741 for (data = blkno; data < blkno + blkcnt; data++) {
742 for (i = 0; i < nints; i++) {
743 *ptr++ = data;
746 ptr = (uint_t *)pattern_buf;
750 * Only write if we're not on the read pass of SCAN_VERIFY.
752 if (!(flags & SCAN_VERIFY_READ_PASS)) {
753 status = (*cur_ops->op_rdwr)(DIR_WRITE, cur_file, blkno,
754 blkcnt, (caddr_t)pattern_buf, driver_flags, xfercntp);
755 if (status)
756 goto bad;
757 } else {
759 * Only read if we are on the read pass of SCAN_VERIFY
761 status = (*cur_ops->op_rdwr)(DIR_READ, cur_file, blkno,
762 blkcnt, (caddr_t)pattern_buf, driver_flags, xfercntp);
763 if (status)
764 goto bad;
766 * compare and make sure the pattern came back intact.
768 for (data = blkno; data < blkno + blkcnt; data++) {
769 for (i = 0; i < nints; i++) {
770 if (*ptr++ != data) {
771 ptr--;
772 err_print("Data miscompare error "
773 "(expecting 0x%x, got 0x%x) at ",
774 data, *ptr);
775 pr_dblock(err_print, blkno);
776 err_print(", offset = 0x%x.\n",
777 (ptr - (uint_t *)pattern_buf) *
778 sizeof (int));
779 goto bad;
785 * No errors occurred, return ok.
787 return (0);
788 bad:
789 return (-1);
793 static int
794 handle_error_conditions()
798 * Check if the errno is ENXIO.
800 if (errno == ENXIO) {
801 fmt_print("\n\nWarning:Cannot access drive, ");
802 fmt_print("aborting surface analysis.\n");
803 return (-1);
806 * check for disk errors
808 switch (disk_error) {
809 case DISK_STAT_RESERVED:
810 case DISK_STAT_UNAVAILABLE:
811 fmt_print("\n\nWarning:Drive may be reserved ");
812 fmt_print("or has been removed, ");
813 fmt_print("aborting surface analysis.\n");
814 return (-1);
815 case DISK_STAT_NOTREADY:
816 fmt_print("\n\nWarning: Drive not ready, ");
817 fmt_print("aborting surface analysis.\n");
818 return (-1);
819 case DISK_STAT_DATA_PROTECT:
820 fmt_print("\n\nWarning: Drive is write protected, ");
821 fmt_print("aborting surface analysis.\n");
822 return (-1);
823 default:
824 break;
826 return (0);