badblocks.c

   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Bad block management
   4 *
   5 * - Heavily based on MD badblocks code from Neil Brown
   6 *
   7 * Copyright (c) 2015, Intel Corporation.
   8 */
   9
  10#include <linux/badblocks.h>
  11#include <linux/seqlock.h>
  12#include <linux/device.h>
  13#include <linux/kernel.h>
  14#include <linux/module.h>
  15#include <linux/stddef.h>
  16#include <linux/types.h>
  17#include <linux/slab.h>
  18
  19/*
  20 * The purpose of badblocks set/clear is to manage bad blocks ranges which are
  21 * identified by LBA addresses.
  22 *
  23 * When the caller of badblocks_set() wants to set a range of bad blocks, the
  24 * setting range can be acked or unacked. And the setting range may merge,
  25 * overwrite, skip the overlapped already set range, depends on who they are
  26 * overlapped or adjacent, and the acknowledgment type of the ranges. It can be
  27 * more complicated when the setting range covers multiple already set bad block
  28 * ranges, with restrictions of maximum length of each bad range and the bad
  29 * table space limitation.
  30 *
  31 * It is difficult and unnecessary to take care of all the possible situations,
  32 * for setting a large range of bad blocks, we can handle it by dividing the
  33 * large range into smaller ones when encounter overlap, max range length or
  34 * bad table full conditions. Every time only a smaller piece of the bad range
  35 * is handled with a limited number of conditions how it is interacted with
  36 * possible overlapped or adjacent already set bad block ranges. Then the hard
  37 * complicated problem can be much simpler to handle in proper way.
  38 *
  39 * When setting a range of bad blocks to the bad table, the simplified situations
  40 * to be considered are, (The already set bad blocks ranges are naming with
  41 *  prefix E, and the setting bad blocks range is naming with prefix S)
  42 *
  43 * 1) A setting range is not overlapped or adjacent to any other already set bad
  44 *    block range.
  45 *                         +--------+
  46 *                         |    S   |
  47 *                         +--------+
  48 *        +-------------+               +-------------+
  49 *        |      E1     |               |      E2     |
  50 *        +-------------+               +-------------+
  51 *    For this situation if the bad blocks table is not full, just allocate a
  52 *    free slot from the bad blocks table to mark the setting range S. The
  53 *    result is,
  54 *        +-------------+  +--------+   +-------------+
  55 *        |      E1     |  |    S   |   |      E2     |
  56 *        +-------------+  +--------+   +-------------+
  57 * 2) A setting range starts exactly at a start LBA of an already set bad blocks
  58 *    range.
  59 * 2.1) The setting range size < already set range size
  60 *        +--------+
  61 *        |    S   |
  62 *        +--------+
  63 *        +-------------+
  64 *        |      E      |
  65 *        +-------------+
  66 * 2.1.1) If S and E are both acked or unacked range, the setting range S can
  67 *    be merged into existing bad range E. The result is,
  68 *        +-------------+
  69 *        |      S      |
  70 *        +-------------+
  71 * 2.1.2) If S is unacked setting and E is acked, the setting will be denied, and
  72 *    the result is,
  73 *        +-------------+
  74 *        |      E      |
  75 *        +-------------+
  76 * 2.1.3) If S is acked setting and E is unacked, range S can overwrite on E.
  77 *    An extra slot from the bad blocks table will be allocated for S, and head
  78 *    of E will move to end of the inserted range S. The result is,
  79 *        +--------+----+
  80 *        |    S   | E  |
  81 *        +--------+----+
  82 * 2.2) The setting range size == already set range size
  83 * 2.2.1) If S and E are both acked or unacked range, the setting range S can
  84 *    be merged into existing bad range E. The result is,
  85 *        +-------------+
  86 *        |      S      |
  87 *        +-------------+
  88 * 2.2.2) If S is unacked setting and E is acked, the setting will be denied, and
  89 *    the result is,
  90 *        +-------------+
  91 *        |      E      |
  92 *        +-------------+
  93 * 2.2.3) If S is acked setting and E is unacked, range S can overwrite all of
  94      bad blocks range E. The result is,
  95 *        +-------------+
  96 *        |      S      |
  97 *        +-------------+
  98 * 2.3) The setting range size > already set range size
  99 *        +-------------------+
 100 *        |          S        |
 101 *        +-------------------+
 102 *        +-------------+
 103 *        |      E      |
 104 *        +-------------+
 105 *    For such situation, the setting range S can be treated as two parts, the
 106 *    first part (S1) is as same size as the already set range E, the second
 107 *    part (S2) is the rest of setting range.
 108 *        +-------------+-----+        +-------------+       +-----+
 109 *        |    S1       | S2  |        |     S1      |       | S2  |
 110 *        +-------------+-----+  ===>  +-------------+       +-----+
 111 *        +-------------+              +-------------+
 112 *        |      E      |              |      E      |
 113 *        +-------------+              +-------------+
 114 *    Now we only focus on how to handle the setting range S1 and already set
 115 *    range E, which are already explained in 2.2), for the rest S2 it will be
 116 *    handled later in next loop.
 117 * 3) A setting range starts before the start LBA of an already set bad blocks
 118 *    range.
 119 *        +-------------+
 120 *        |      S      |
 121 *        +-------------+
 122 *             +-------------+
 123 *             |      E      |
 124 *             +-------------+
 125 *    For this situation, the setting range S can be divided into two parts, the
 126 *    first (S1) ends at the start LBA of already set range E, the second part
 127 *    (S2) starts exactly at a start LBA of the already set range E.
 128 *        +----+---------+             +----+      +---------+
 129 *        | S1 |    S2   |             | S1 |      |    S2   |
 130 *        +----+---------+      ===>   +----+      +---------+
 131 *             +-------------+                     +-------------+
 132 *             |      E      |                     |      E      |
 133 *             +-------------+                     +-------------+
 134 *    Now only the first part S1 should be handled in this loop, which is in
 135 *    similar condition as 1). The rest part S2 has exact same start LBA address
 136 *    of the already set range E, they will be handled in next loop in one of
 137 *    situations in 2).
 138 * 4) A setting range starts after the start LBA of an already set bad blocks
 139 *    range.
 140 * 4.1) If the setting range S exactly matches the tail part of already set bad
 141 *    blocks range E, like the following chart shows,
 142 *            +---------+
 143 *            |   S     |
 144 *            +---------+
 145 *        +-------------+
 146 *        |      E      |
 147 *        +-------------+
 148 * 4.1.1) If range S and E have same acknowledge value (both acked or unacked),
 149 *    they will be merged into one, the result is,
 150 *        +-------------+
 151 *        |      S      |
 152 *        +-------------+
 153 * 4.1.2) If range E is acked and the setting range S is unacked, the setting
 154 *    request of S will be rejected, the result is,
 155 *        +-------------+
 156 *        |      E      |
 157 *        +-------------+
 158 * 4.1.3) If range E is unacked, and the setting range S is acked, then S may
 159 *    overwrite the overlapped range of E, the result is,
 160 *        +---+---------+
 161 *        | E |    S    |
 162 *        +---+---------+
 163 * 4.2) If the setting range S stays in middle of an already set range E, like
 164 *    the following chart shows,
 165 *             +----+
 166 *             | S  |
 167 *             +----+
 168 *        +--------------+
 169 *        |       E      |
 170 *        +--------------+
 171 * 4.2.1) If range S and E have same acknowledge value (both acked or unacked),
 172 *    they will be merged into one, the result is,
 173 *        +--------------+
 174 *        |       S      |
 175 *        +--------------+
 176 * 4.2.2) If range E is acked and the setting range S is unacked, the setting
 177 *    request of S will be rejected, the result is also,
 178 *        +--------------+
 179 *        |       E      |
 180 *        +--------------+
 181 * 4.2.3) If range E is unacked, and the setting range S is acked, then S will
 182 *    inserted into middle of E and split previous range E into two parts (E1
 183 *    and E2), the result is,
 184 *        +----+----+----+
 185 *        | E1 |  S | E2 |
 186 *        +----+----+----+
 187 * 4.3) If the setting bad blocks range S is overlapped with an already set bad
 188 *    blocks range E. The range S starts after the start LBA of range E, and
 189 *    ends after the end LBA of range E, as the following chart shows,
 190 *            +-------------------+
 191 *            |          S        |
 192 *            +-------------------+
 193 *        +-------------+
 194 *        |      E      |
 195 *        +-------------+
 196 *    For this situation the range S can be divided into two parts, the first
 197 *    part (S1) ends at end range E, and the second part (S2) has rest range of
 198 *    origin S.
 199 *            +---------+---------+            +---------+      +---------+
 200 *            |    S1   |    S2   |            |    S1   |      |    S2   |
 201 *            +---------+---------+  ===>      +---------+      +---------+
 202 *        +-------------+                  +-------------+
 203 *        |      E      |                  |      E      |
 204 *        +-------------+                  +-------------+
 205 *     Now in this loop the setting range S1 and already set range E can be
 206 *     handled as the situations 4.1), the rest range S2 will be handled in next
 207 *     loop and ignored in this loop.
 208 * 5) A setting bad blocks range S is adjacent to one or more already set bad
 209 *    blocks range(s), and they are all acked or unacked range.
 210 * 5.1) Front merge: If the already set bad blocks range E is before setting
 211 *    range S and they are adjacent,
 212 *                +------+
 213 *                |  S   |
 214 *                +------+
 215 *        +-------+
 216 *        |   E   |
 217 *        +-------+
 218 * 5.1.1) When total size of range S and E <= BB_MAX_LEN, and their acknowledge
 219 *    values are same, the setting range S can front merges into range E. The
 220 *    result is,
 221 *        +--------------+
 222 *        |       S      |
 223 *        +--------------+
 224 * 5.1.2) Otherwise these two ranges cannot merge, just insert the setting
 225 *    range S right after already set range E into the bad blocks table. The
 226 *    result is,
 227 *        +--------+------+
 228 *        |   E    |   S  |
 229 *        +--------+------+
 230 * 6) Special cases which above conditions cannot handle
 231 * 6.1) Multiple already set ranges may merge into less ones in a full bad table
 232 *        +-------------------------------------------------------+
 233 *        |                           S                           |
 234 *        +-------------------------------------------------------+
 235 *        |<----- BB_MAX_LEN ----->|
 236 *                                 +-----+     +-----+   +-----+
 237 *                                 | E1  |     | E2  |   | E3  |
 238 *                                 +-----+     +-----+   +-----+
 239 *     In the above example, when the bad blocks table is full, inserting the
 240 *     first part of setting range S will fail because no more available slot
 241 *     can be allocated from bad blocks table. In this situation a proper
 242 *     setting method should be go though all the setting bad blocks range and
 243 *     look for chance to merge already set ranges into less ones. When there
 244 *     is available slot from bad blocks table, re-try again to handle more
 245 *     setting bad blocks ranges as many as possible.
 246 *        +------------------------+
 247 *        |          S3            |
 248 *        +------------------------+
 249 *        |<----- BB_MAX_LEN ----->|
 250 *                                 +-----+-----+-----+---+-----+--+
 251 *                                 |       S1        |     S2     |
 252 *                                 +-----+-----+-----+---+-----+--+
 253 *     The above chart shows although the first part (S3) cannot be inserted due
 254 *     to no-space in bad blocks table, but the following E1, E2 and E3 ranges
 255 *     can be merged with rest part of S into less range S1 and S2. Now there is
 256 *     1 free slot in bad blocks table.
 257 *        +------------------------+-----+-----+-----+---+-----+--+
 258 *        |           S3           |       S1        |     S2     |
 259 *        +------------------------+-----+-----+-----+---+-----+--+
 260 *     Since the bad blocks table is not full anymore, re-try again for the
 261 *     origin setting range S. Now the setting range S3 can be inserted into the
 262 *     bad blocks table with previous freed slot from multiple ranges merge.
 263 * 6.2) Front merge after overwrite
 264 *    In the following example, in bad blocks table, E1 is an acked bad blocks
 265 *    range and E2 is an unacked bad blocks range, therefore they are not able
 266 *    to merge into a larger range. The setting bad blocks range S is acked,
 267 *    therefore part of E2 can be overwritten by S.
 268 *                      +--------+
 269 *                      |    S   |                             acknowledged
 270 *                      +--------+                         S:       1
 271 *              +-------+-------------+                   E1:       1
 272 *              |   E1  |    E2       |                   E2:       0
 273 *              +-------+-------------+
 274 *     With previous simplified routines, after overwriting part of E2 with S,
 275 *     the bad blocks table should be (E3 is remaining part of E2 which is not
 276 *     overwritten by S),
 277 *                                                             acknowledged
 278 *              +-------+--------+----+                    S:       1
 279 *              |   E1  |    S   | E3 |                   E1:       1
 280 *              +-------+--------+----+                   E3:       0
 281 *     The above result is correct but not perfect. Range E1 and S in the bad
 282 *     blocks table are all acked, merging them into a larger one range may
 283 *     occupy less bad blocks table space and make badblocks_check() faster.
 284 *     Therefore in such situation, after overwriting range S, the previous range
 285 *     E1 should be checked for possible front combination. Then the ideal
 286 *     result can be,
 287 *              +----------------+----+                        acknowledged
 288 *              |       E1       | E3 |                   E1:       1
 289 *              +----------------+----+                   E3:       0
 290 * 6.3) Behind merge: If the already set bad blocks range E is behind the setting
 291 *    range S and they are adjacent. Normally we don't need to care about this
 292 *    because front merge handles this while going though range S from head to
 293 *    tail, except for the tail part of range S. When the setting range S are
 294 *    fully handled, all the above simplified routine doesn't check whether the
 295 *    tail LBA of range S is adjacent to the next already set range and not
 296 *    merge them even it is possible.
 297 *        +------+
 298 *        |  S   |
 299 *        +------+
 300 *               +-------+
 301 *               |   E   |
 302 *               +-------+
 303 *    For the above special situation, when the setting range S are all handled
 304 *    and the loop ends, an extra check is necessary for whether next already
 305 *    set range E is right after S and mergeable.
 306 * 6.3.1) When total size of range E and S <= BB_MAX_LEN, and their acknowledge
 307 *    values are same, the setting range S can behind merges into range E. The
 308 *    result is,
 309 *        +--------------+
 310 *        |       S      |
 311 *        +--------------+
 312 * 6.3.2) Otherwise these two ranges cannot merge, just insert the setting range
 313 *     S in front of the already set range E in the bad blocks table. The result
 314 *     is,
 315 *        +------+-------+
 316 *        |  S   |   E   |
 317 *        +------+-------+
 318 *
 319 * All the above 5 simplified situations and 3 special cases may cover 99%+ of
 320 * the bad block range setting conditions. Maybe there is some rare corner case
 321 * is not considered and optimized, it won't hurt if badblocks_set() fails due
 322 * to no space, or some ranges are not merged to save bad blocks table space.
 323 *
 324 * Inside badblocks_set() each loop starts by jumping to re_insert label, every
 325 * time for the new loop prev_badblocks() is called to find an already set range
 326 * which starts before or at current setting range. Since the setting bad blocks
 327 * range is handled from head to tail, most of the cases it is unnecessary to do
 328 * the binary search inside prev_badblocks(), it is possible to provide a hint
 329 * to prev_badblocks() for a fast path, then the expensive binary search can be
 330 * avoided. In my test with the hint to prev_badblocks(), except for the first
 331 * loop, all rested calls to prev_badblocks() can go into the fast path and
 332 * return correct bad blocks table index immediately.
 333 *
 334 *
 335 * Clearing a bad blocks range from the bad block table has similar idea as
 336 * setting does, but much more simpler. The only thing needs to be noticed is
 337 * when the clearing range hits middle of a bad block range, the existing bad
 338 * block range will split into two, and one more item should be added into the
 339 * bad block table. The simplified situations to be considered are, (The already
 340 * set bad blocks ranges in bad block table are naming with prefix E, and the
 341 * clearing bad blocks range is naming with prefix C)
 342 *
 343 * 1) A clearing range is not overlapped to any already set ranges in bad block
 344 *    table.
 345 *    +-----+         |          +-----+         |          +-----+
 346 *    |  C  |         |          |  C  |         |          |  C  |
 347 *    +-----+         or         +-----+         or         +-----+
 348 *            +---+   |   +----+         +----+  |  +---+
 349 *            | E |   |   | E1 |         | E2 |  |  | E |
 350 *            +---+   |   +----+         +----+  |  +---+
 351 *    For the above situations, no bad block to be cleared and no failure
 352 *    happens, simply returns 0.
 353 * 2) The clearing range hits middle of an already setting bad blocks range in
 354 *    the bad block table.
 355 *            +---+
 356 *            | C |
 357 *            +---+
 358 *     +-----------------+
 359 *     |         E       |
 360 *     +-----------------+
 361 *    In this situation if the bad block table is not full, the range E will be
 362 *    split into two ranges E1 and E2. The result is,
 363 *     +------+   +------+
 364 *     |  E1  |   |  E2  |
 365 *     +------+   +------+
 366 * 3) The clearing range starts exactly at same LBA as an already set bad block range
 367 *    from the bad block table.
 368 * 3.1) Partially covered at head part
 369 *         +------------+
 370 *         |     C      |
 371 *         +------------+
 372 *         +-----------------+
 373 *         |         E       |
 374 *         +-----------------+
 375 *    For this situation, the overlapped already set range will update the
 376 *    start LBA to end of C and shrink the range to BB_LEN(E) - BB_LEN(C). No
 377 *    item deleted from bad block table. The result is,
 378 *                      +----+
 379 *                      | E1 |
 380 *                      +----+
 381 * 3.2) Exact fully covered
 382 *         +-----------------+
 383 *         |         C       |
 384 *         +-----------------+
 385 *         +-----------------+
 386 *         |         E       |
 387 *         +-----------------+
 388 *    For this situation the whole bad blocks range E will be cleared and its
 389 *    corresponded item is deleted from the bad block table.
 390 * 4) The clearing range exactly ends at same LBA as an already set bad block
 391 *    range.
 392 *                   +-------+
 393 *                   |   C   |
 394 *                   +-------+
 395 *         +-----------------+
 396 *         |         E       |
 397 *         +-----------------+
 398 *    For the above situation, the already set range E is updated to shrink its
 399 *    end to the start of C, and reduce its length to BB_LEN(E) - BB_LEN(C).
 400 *    The result is,
 401 *         +---------+
 402 *         |    E    |
 403 *         +---------+
 404 * 5) The clearing range is partially overlapped with an already set bad block
 405 *    range from the bad block table.
 406 * 5.1) The already set bad block range is front overlapped with the clearing
 407 *    range.
 408 *         +----------+
 409 *         |     C    |
 410 *         +----------+
 411 *              +------------+
 412 *              |      E     |
 413 *              +------------+
 414 *   For such situation, the clearing range C can be treated as two parts. The
 415 *   first part ends at the start LBA of range E, and the second part starts at
 416 *   same LBA of range E.
 417 *         +----+-----+               +----+   +-----+
 418 *         | C1 | C2  |               | C1 |   | C2  |
 419 *         +----+-----+         ===>  +----+   +-----+
 420 *              +------------+                 +------------+
 421 *              |      E     |                 |      E     |
 422 *              +------------+                 +------------+
 423 *   Now the first part C1 can be handled as condition 1), and the second part C2 can be
 424 *   handled as condition 3.1) in next loop.
 425 * 5.2) The already set bad block range is behind overlaopped with the clearing
 426 *   range.
 427 *                 +----------+
 428 *                 |     C    |
 429 *                 +----------+
 430 *         +------------+
 431 *         |      E     |
 432 *         +------------+
 433 *   For such situation, the clearing range C can be treated as two parts. The
 434 *   first part C1 ends at same end LBA of range E, and the second part starts
 435 *   at end LBA of range E.
 436 *                 +----+-----+                 +----+    +-----+
 437 *                 | C1 | C2  |                 | C1 |    | C2  |
 438 *                 +----+-----+  ===>           +----+    +-----+
 439 *         +------------+               +------------+
 440 *         |      E     |               |      E     |
 441 *         +------------+               +------------+
 442 *   Now the first part clearing range C1 can be handled as condition 4), and
 443 *   the second part clearing range C2 can be handled as condition 1) in next
 444 *   loop.
 445 *
 446 *   All bad blocks range clearing can be simplified into the above 5 situations
 447 *   by only handling the head part of the clearing range in each run of the
 448 *   while-loop. The idea is similar to bad blocks range setting but much
 449 *   simpler.
 450 */
 451
 452/*
 453 * Find the range starts at-or-before 's' from bad table. The search
 454 * starts from index 'hint' and stops at index 'hint_end' from the bad
 455 * table.
 456 */
 457static int prev_by_hint(struct badblocks *bb, sector_t s, int hint)
 458{
 459	int hint_end = hint + 2;
 460	u64 *p = bb->page;
 461	int ret = -1;
 462
 463	while ((hint < hint_end) && ((hint + 1) <= bb->count) &&
 464	       (BB_OFFSET(p[hint]) <= s)) {
 465		if ((hint + 1) == bb->count || BB_OFFSET(p[hint + 1]) > s) {
 466			ret = hint;
 467			break;
 468		}
 469		hint++;
 470	}
 471
 472	return ret;
 473}
 474
 475/*
 476 * Find the range starts at-or-before bad->start. If 'hint' is provided
 477 * (hint >= 0) then search in the bad table from hint firstly. It is
 478 * very probably the wanted bad range can be found from the hint index,
 479 * then the unnecessary while-loop iteration can be avoided.
 480 */
 481static int prev_badblocks(struct badblocks *bb, struct badblocks_context *bad,
 482			  int hint)
 483{
 484	sector_t s = bad->start;
 485	int ret = -1;
 486	int lo, hi;
 487	u64 *p;
 488
 489	if (!bb->count)
 490		goto out;
 491
 492	if (hint >= 0) {
 493		ret = prev_by_hint(bb, s, hint);
 494		if (ret >= 0)
 495			goto out;
 496	}
 497
 498	lo = 0;
 499	hi = bb->count;
 500	p = bb->page;
 501
 502	/* The following bisect search might be unnecessary */
 503	if (BB_OFFSET(p[lo]) > s)
 504		return -1;
 505	if (BB_OFFSET(p[hi - 1]) <= s)
 506		return hi - 1;
 507
 508	/* Do bisect search in bad table */
 509	while (hi - lo > 1) {
 510		int mid = (lo + hi)/2;
 511		sector_t a = BB_OFFSET(p[mid]);
 512
 513		if (a == s) {
 514			ret = mid;
 515			goto out;
 516		}
 517
 518		if (a < s)
 519			lo = mid;
 520		else
 521			hi = mid;
 522	}
 523
 524	if (BB_OFFSET(p[lo]) <= s)
 525		ret = lo;
 526out:
 527	return ret;
 528}
 529
 530/*
 531 * Return 'true' if the range indicated by 'bad' can be forward
 532 * merged with the bad range (from the bad table) indexed by 'prev'.
 533 */
 534static bool can_merge_front(struct badblocks *bb, int prev,
 535			    struct badblocks_context *bad)
 536{
 537	sector_t s = bad->start;
 538	u64 *p = bb->page;
 539
 540	if (BB_ACK(p[prev]) == bad->ack &&
 541	    (s < BB_END(p[prev]) ||
 542	     (s == BB_END(p[prev]) && (BB_LEN(p[prev]) < BB_MAX_LEN))))
 543		return true;
 544	return false;
 545}
 546
 547/*
 548 * Do forward merge for range indicated by 'bad' and the bad range
 549 * (from bad table) indexed by 'prev'. The return value is sectors
 550 * merged from bad->len.
 551 */
 552static int front_merge(struct badblocks *bb, int prev, struct badblocks_context *bad)
 553{
 554	sector_t sectors = bad->len;
 555	sector_t s = bad->start;
 556	u64 *p = bb->page;
 557	int merged = 0;
 558
 559	WARN_ON(s > BB_END(p[prev]));
 560
 561	if (s < BB_END(p[prev])) {
 562		merged = min_t(sector_t, sectors, BB_END(p[prev]) - s);
 563	} else {
 564		merged = min_t(sector_t, sectors, BB_MAX_LEN - BB_LEN(p[prev]));
 565		if ((prev + 1) < bb->count &&
 566		    merged > (BB_OFFSET(p[prev + 1]) - BB_END(p[prev]))) {
 567			merged = BB_OFFSET(p[prev + 1]) - BB_END(p[prev]);
 568		}
 569
 570		p[prev] = BB_MAKE(BB_OFFSET(p[prev]),
 571				  BB_LEN(p[prev]) + merged, bad->ack);
 572	}
 573
 574	return merged;
 575}
 576
 577/*
 578 * 'Combine' is a special case which can_merge_front() is not able to
 579 * handle: If a bad range (indexed by 'prev' from bad table) exactly
 580 * starts as bad->start, and the bad range ahead of 'prev' (indexed by
 581 * 'prev - 1' from bad table) exactly ends at where 'prev' starts, and
 582 * the sum of their lengths does not exceed BB_MAX_LEN limitation, then
 583 * these two bad range (from bad table) can be combined.
 584 *
 585 * Return 'true' if bad ranges indexed by 'prev' and 'prev - 1' from bad
 586 * table can be combined.
 587 */
 588static bool can_combine_front(struct badblocks *bb, int prev,
 589			      struct badblocks_context *bad)
 590{
 591	u64 *p = bb->page;
 592
 593	if ((prev > 0) &&
 594	    (BB_OFFSET(p[prev]) == bad->start) &&
 595	    (BB_END(p[prev - 1]) == BB_OFFSET(p[prev])) &&
 596	    (BB_LEN(p[prev - 1]) + BB_LEN(p[prev]) <= BB_MAX_LEN) &&
 597	    (BB_ACK(p[prev - 1]) == BB_ACK(p[prev])))
 598		return true;
 599	return false;
 600}
 601
 602/*
 603 * Combine the bad ranges indexed by 'prev' and 'prev - 1' (from bad
 604 * table) into one larger bad range, and the new range is indexed by
 605 * 'prev - 1'.
 606 * The caller of front_combine() will decrease bb->count, therefore
 607 * it is unnecessary to clear p[perv] after front merge.
 608 */
 609static void front_combine(struct badblocks *bb, int prev)
 610{
 611	u64 *p = bb->page;
 612
 613	p[prev - 1] = BB_MAKE(BB_OFFSET(p[prev - 1]),
 614			      BB_LEN(p[prev - 1]) + BB_LEN(p[prev]),
 615			      BB_ACK(p[prev]));
 616	if ((prev + 1) < bb->count)
 617		memmove(p + prev, p + prev + 1, (bb->count - prev - 1) * 8);
 618}
 619
 620/*
 621 * Return 'true' if the range indicated by 'bad' is exactly forward
 622 * overlapped with the bad range (from bad table) indexed by 'front'.
 623 * Exactly forward overlap means the bad range (from bad table) indexed
 624 * by 'prev' does not cover the whole range indicated by 'bad'.
 625 */
 626static bool overlap_front(struct badblocks *bb, int front,
 627			  struct badblocks_context *bad)
 628{
 629	u64 *p = bb->page;
 630
 631	if (bad->start >= BB_OFFSET(p[front]) &&
 632	    bad->start < BB_END(p[front]))
 633		return true;
 634	return false;
 635}
 636
 637/*
 638 * Return 'true' if the range indicated by 'bad' is exactly backward
 639 * overlapped with the bad range (from bad table) indexed by 'behind'.
 640 */
 641static bool overlap_behind(struct badblocks *bb, struct badblocks_context *bad,
 642			   int behind)
 643{
 644	u64 *p = bb->page;
 645
 646	if (bad->start < BB_OFFSET(p[behind]) &&
 647	    (bad->start + bad->len) > BB_OFFSET(p[behind]))
 648		return true;
 649	return false;
 650}
 651
 652/*
 653 * Return 'true' if the range indicated by 'bad' can overwrite the bad
 654 * range (from bad table) indexed by 'prev'.
 655 *
 656 * The range indicated by 'bad' can overwrite the bad range indexed by
 657 * 'prev' when,
 658 * 1) The whole range indicated by 'bad' can cover partial or whole bad
 659 *    range (from bad table) indexed by 'prev'.
 660 * 2) The ack value of 'bad' is larger or equal to the ack value of bad
 661 *    range 'prev'.
 662 *
 663 * If the overwriting doesn't cover the whole bad range (from bad table)
 664 * indexed by 'prev', new range might be split from existing bad range,
 665 * 1) The overwrite covers head or tail part of existing bad range, 1
 666 *    extra bad range will be split and added into the bad table.
 667 * 2) The overwrite covers middle of existing bad range, 2 extra bad
 668 *    ranges will be split (ahead and after the overwritten range) and
 669 *    added into the bad table.
 670 * The number of extra split ranges of the overwriting is stored in
 671 * 'extra' and returned for the caller.
 672 */
 673static bool can_front_overwrite(struct badblocks *bb, int prev,
 674				struct badblocks_context *bad, int *extra)
 675{
 676	u64 *p = bb->page;
 677	int len;
 678
 679	WARN_ON(!overlap_front(bb, prev, bad));
 680
 681	if (BB_ACK(p[prev]) >= bad->ack)
 682		return false;
 683
 684	if (BB_END(p[prev]) <= (bad->start + bad->len)) {
 685		len = BB_END(p[prev]) - bad->start;
 686		if (BB_OFFSET(p[prev]) == bad->start)
 687			*extra = 0;
 688		else
 689			*extra = 1;
 690
 691		bad->len = len;
 692	} else {
 693		if (BB_OFFSET(p[prev]) == bad->start)
 694			*extra = 1;
 695		else
 696		/*
 697		 * prev range will be split into two, beside the overwritten
 698		 * one, an extra slot needed from bad table.
 699		 */
 700			*extra = 2;
 701	}
 702
 703	if ((bb->count + (*extra)) > MAX_BADBLOCKS)
 704		return false;
 705
 706	return true;
 707}
 708
 709/*
 710 * Do the overwrite from the range indicated by 'bad' to the bad range
 711 * (from bad table) indexed by 'prev'.
 712 * The previously called can_front_overwrite() will provide how many
 713 * extra bad range(s) might be split and added into the bad table. All
 714 * the splitting cases in the bad table will be handled here.
 715 */
 716static int front_overwrite(struct badblocks *bb, int prev,
 717			   struct badblocks_context *bad, int extra)
 718{
 719	u64 *p = bb->page;
 720	sector_t orig_end = BB_END(p[prev]);
 721	int orig_ack = BB_ACK(p[prev]);
 722
 723	switch (extra) {
 724	case 0:
 725		p[prev] = BB_MAKE(BB_OFFSET(p[prev]), BB_LEN(p[prev]),
 726				  bad->ack);
 727		break;
 728	case 1:
 729		if (BB_OFFSET(p[prev]) == bad->start) {
 730			p[prev] = BB_MAKE(BB_OFFSET(p[prev]),
 731					  bad->len, bad->ack);
 732			memmove(p + prev + 2, p + prev + 1,
 733				(bb->count - prev - 1) * 8);
 734			p[prev + 1] = BB_MAKE(bad->start + bad->len,
 735					      orig_end - BB_END(p[prev]),
 736					      orig_ack);
 737		} else {
 738			p[prev] = BB_MAKE(BB_OFFSET(p[prev]),
 739					  bad->start - BB_OFFSET(p[prev]),
 740					  orig_ack);
 741			/*
 742			 * prev +2 -> prev + 1 + 1, which is for,
 743			 * 1) prev + 1: the slot index of the previous one
 744			 * 2) + 1: one more slot for extra being 1.
 745			 */
 746			memmove(p + prev + 2, p + prev + 1,
 747				(bb->count - prev - 1) * 8);
 748			p[prev + 1] = BB_MAKE(bad->start, bad->len, bad->ack);
 749		}
 750		break;
 751	case 2:
 752		p[prev] = BB_MAKE(BB_OFFSET(p[prev]),
 753				  bad->start - BB_OFFSET(p[prev]),
 754				  orig_ack);
 755		/*
 756		 * prev + 3 -> prev + 1 + 2, which is for,
 757		 * 1) prev + 1: the slot index of the previous one
 758		 * 2) + 2: two more slots for extra being 2.
 759		 */
 760		memmove(p + prev + 3, p + prev + 1,
 761			(bb->count - prev - 1) * 8);
 762		p[prev + 1] = BB_MAKE(bad->start, bad->len, bad->ack);
 763		p[prev + 2] = BB_MAKE(BB_END(p[prev + 1]),
 764				      orig_end - BB_END(p[prev + 1]),
 765				      orig_ack);
 766		break;
 767	default:
 768		break;
 769	}
 770
 771	return bad->len;
 772}
 773
 774/*
 775 * Explicitly insert a range indicated by 'bad' to the bad table, where
 776 * the location is indexed by 'at'.
 777 */
 778static int insert_at(struct badblocks *bb, int at, struct badblocks_context *bad)
 779{
 780	u64 *p = bb->page;
 781	int len;
 782
 783	WARN_ON(badblocks_full(bb));
 784
 785	len = min_t(sector_t, bad->len, BB_MAX_LEN);
 786	if (at < bb->count)
 787		memmove(p + at + 1, p + at, (bb->count - at) * 8);
 788	p[at] = BB_MAKE(bad->start, len, bad->ack);
 789
 790	return len;
 791}
 792
 793static void badblocks_update_acked(struct badblocks *bb)
 794{
 795	bool unacked = false;
 796	u64 *p = bb->page;
 797	int i;
 798
 799	if (!bb->unacked_exist)
 800		return;
 801
 802	for (i = 0; i < bb->count ; i++) {
 803		if (!BB_ACK(p[i])) {
 804			unacked = true;
 805			break;
 806		}
 807	}
 808
 809	if (!unacked)
 810		bb->unacked_exist = 0;
 811}
 812
 813/*
 814 * Return 'true' if the range indicated by 'bad' is exactly backward
 815 * overlapped with the bad range (from bad table) indexed by 'behind'.
 816 */
 817static bool try_adjacent_combine(struct badblocks *bb, int prev)
 818{
 819	u64 *p = bb->page;
 820
 821	if (prev >= 0 && (prev + 1) < bb->count &&
 822	    BB_END(p[prev]) == BB_OFFSET(p[prev + 1]) &&
 823	    (BB_LEN(p[prev]) + BB_LEN(p[prev + 1])) <= BB_MAX_LEN &&
 824	    BB_ACK(p[prev]) == BB_ACK(p[prev + 1])) {
 825		p[prev] = BB_MAKE(BB_OFFSET(p[prev]),
 826				  BB_LEN(p[prev]) + BB_LEN(p[prev + 1]),
 827				  BB_ACK(p[prev]));
 828
 829		if ((prev + 2) < bb->count)
 830			memmove(p + prev + 1, p + prev + 2,
 831				(bb->count -  (prev + 2)) * 8);
 832		bb->count--;
 833		return true;
 834	}
 835	return false;
 836}
 837
 838/* Do exact work to set bad block range into the bad block table */
 839static bool _badblocks_set(struct badblocks *bb, sector_t s, sector_t sectors,
 840			   int acknowledged)
 841{
 842	int len = 0, added = 0;
 843	struct badblocks_context bad;
 844	int prev = -1, hint = -1;
 845	unsigned long flags;
 846	u64 *p;
 847
 848	if (bb->shift < 0)
 849		/* badblocks are disabled */
 850		return false;
 851
 852	if (sectors == 0)
 853		/* Invalid sectors number */
 854		return false;
 855
 856	if (bb->shift) {
 857		/* round the start down, and the end up */
 858		sector_t next = s + sectors;
 859
 860		rounddown(s, 1 << bb->shift);
 861		roundup(next, 1 << bb->shift);
 862		sectors = next - s;
 863	}
 864
 865	write_seqlock_irqsave(&bb->lock, flags);
 866
 867	bad.ack = acknowledged;
 868	p = bb->page;
 869
 870re_insert:
 871	bad.start = s;
 872	bad.len = sectors;
 873	len = 0;
 874
 875	if (badblocks_full(bb))
 876		goto out;
 877
 878	if (badblocks_empty(bb)) {
 879		len = insert_at(bb, 0, &bad);
 880		bb->count++;
 881		added++;
 882		goto update_sectors;
 883	}
 884
 885	prev = prev_badblocks(bb, &bad, hint);
 886
 887	/* start before all badblocks */
 888	if (prev < 0) {
 889		/* insert on the first */
 890		if (bad.len > (BB_OFFSET(p[0]) - bad.start))
 891			bad.len = BB_OFFSET(p[0]) - bad.start;
 892		len = insert_at(bb, 0, &bad);
 893		bb->count++;
 894		added++;
 895		hint = ++prev;
 896		goto update_sectors;
 897	}
 898
 899	/* in case p[prev-1] can be merged with p[prev] */
 900	if (can_combine_front(bb, prev, &bad)) {
 901		front_combine(bb, prev);
 902		bb->count--;
 903		added++;
 904		hint = prev;
 905		goto update_sectors;
 906	}
 907
 908	if (can_merge_front(bb, prev, &bad)) {
 909		len = front_merge(bb, prev, &bad);
 910		added++;
 911		hint = prev;
 912		goto update_sectors;
 913	}
 914
 915	if (overlap_front(bb, prev, &bad)) {
 916		int extra = 0;
 917
 918		if (!can_front_overwrite(bb, prev, &bad, &extra)) {
 919			if (extra > 0)
 920				goto out;
 921
 922			len = min_t(sector_t,
 923				    BB_END(p[prev]) - s, sectors);
 924			hint = prev;
 925			goto update_sectors;
 926		}
 927
 928		len = front_overwrite(bb, prev, &bad, extra);
 929		added++;
 930		bb->count += extra;
 931
 932		if (can_combine_front(bb, prev, &bad)) {
 933			front_combine(bb, prev);
 934			bb->count--;
 935		}
 936
 937		hint = prev;
 938		goto update_sectors;
 939	}
 940
 941	/* cannot merge and there is space in bad table */
 942	if ((prev + 1) < bb->count &&
 943	    overlap_behind(bb, &bad, prev + 1))
 944		bad.len = min_t(sector_t,
 945				bad.len, BB_OFFSET(p[prev + 1]) - bad.start);
 946
 947	len = insert_at(bb, prev + 1, &bad);
 948	bb->count++;
 949	added++;
 950	hint = ++prev;
 951
 952update_sectors:
 953	s += len;
 954	sectors -= len;
 955
 956	if (sectors > 0)
 957		goto re_insert;
 958
 959	/*
 960	 * Check whether the following already set range can be
 961	 * merged. (prev < 0) condition is not handled here,
 962	 * because it's already complicated enough.
 963	 */
 964	try_adjacent_combine(bb, prev);
 965
 966out:
 967	if (added) {
 968		set_changed(bb);
 969
 970		if (!acknowledged)
 971			bb->unacked_exist = 1;
 972		else
 973			badblocks_update_acked(bb);
 974	}
 975
 976	write_sequnlock_irqrestore(&bb->lock, flags);
 977
 978	return sectors == 0;
 979}
 980
 981/*
 982 * Clear the bad block range from bad block table which is front overlapped
 983 * with the clearing range. The return value is how many sectors from an
 984 * already set bad block range are cleared. If the whole bad block range is
 985 * covered by the clearing range and fully cleared, 'delete' is set as 1 for
 986 * the caller to reduce bb->count.
 987 */
 988static int front_clear(struct badblocks *bb, int prev,
 989		       struct badblocks_context *bad, int *deleted)
 990{
 991	sector_t sectors = bad->len;
 992	sector_t s = bad->start;
 993	u64 *p = bb->page;
 994	int cleared = 0;
 995
 996	*deleted = 0;
 997	if (s == BB_OFFSET(p[prev])) {
 998		if (BB_LEN(p[prev]) > sectors) {
 999			p[prev] = BB_MAKE(BB_OFFSET(p[prev]) + sectors,
1000					  BB_LEN(p[prev]) - sectors,
1001					  BB_ACK(p[prev]));
1002			cleared = sectors;
1003		} else {
1004			/* BB_LEN(p[prev]) <= sectors */
1005			cleared = BB_LEN(p[prev]);
1006			if ((prev + 1) < bb->count)
1007				memmove(p + prev, p + prev + 1,
1008				       (bb->count - prev - 1) * 8);
1009			*deleted = 1;
1010		}
1011	} else if (s > BB_OFFSET(p[prev])) {
1012		if (BB_END(p[prev]) <= (s + sectors)) {
1013			cleared = BB_END(p[prev]) - s;
1014			p[prev] = BB_MAKE(BB_OFFSET(p[prev]),
1015					  s - BB_OFFSET(p[prev]),
1016					  BB_ACK(p[prev]));
1017		} else {
1018			/* Splitting is handled in front_splitting_clear() */
1019			BUG();
1020		}
1021	}
1022
1023	return cleared;
1024}
1025
1026/*
1027 * Handle the condition that the clearing range hits middle of an already set
1028 * bad block range from bad block table. In this condition the existing bad
1029 * block range is split into two after the middle part is cleared.
1030 */
1031static int front_splitting_clear(struct badblocks *bb, int prev,
1032				  struct badblocks_context *bad)
1033{
1034	u64 *p = bb->page;
1035	u64 end = BB_END(p[prev]);
1036	int ack = BB_ACK(p[prev]);
1037	sector_t sectors = bad->len;
1038	sector_t s = bad->start;
1039
1040	p[prev] = BB_MAKE(BB_OFFSET(p[prev]),
1041			  s - BB_OFFSET(p[prev]),
1042			  ack);
1043	memmove(p + prev + 2, p + prev + 1, (bb->count - prev - 1) * 8);
1044	p[prev + 1] = BB_MAKE(s + sectors, end - s - sectors, ack);
1045	return sectors;
1046}
1047
1048/* Do the exact work to clear bad block range from the bad block table */
1049static bool _badblocks_clear(struct badblocks *bb, sector_t s, sector_t sectors)
1050{
1051	struct badblocks_context bad;
1052	int prev = -1, hint = -1;
1053	int len = 0, cleared = 0;
1054	u64 *p;
1055
1056	if (bb->shift < 0)
1057		/* badblocks are disabled */
1058		return false;
1059
1060	if (sectors == 0)
1061		/* Invalid sectors number */
1062		return false;
1063
1064	if (bb->shift) {
1065		sector_t target;
1066
1067		/* When clearing we round the start up and the end down.
1068		 * This should not matter as the shift should align with
1069		 * the block size and no rounding should ever be needed.
1070		 * However it is better the think a block is bad when it
1071		 * isn't than to think a block is not bad when it is.
1072		 */
1073		target = s + sectors;
1074		roundup(s, 1 << bb->shift);
1075		rounddown(target, 1 << bb->shift);
1076		sectors = target - s;
1077	}
1078
1079	write_seqlock_irq(&bb->lock);
1080
1081	bad.ack = true;
1082	p = bb->page;
1083
1084re_clear:
1085	bad.start = s;
1086	bad.len = sectors;
1087
1088	if (badblocks_empty(bb)) {
1089		len = sectors;
1090		cleared++;
1091		goto update_sectors;
1092	}
1093
1094
1095	prev = prev_badblocks(bb, &bad, hint);
1096
1097	/* Start before all badblocks */
1098	if (prev < 0) {
1099		if (overlap_behind(bb, &bad, 0)) {
1100			len = BB_OFFSET(p[0]) - s;
1101			hint = 0;
1102		} else {
1103			len = sectors;
1104		}
1105		/*
1106		 * Both situations are to clear non-bad range,
1107		 * should be treated as successful
1108		 */
1109		cleared++;
1110		goto update_sectors;
1111	}
1112
1113	/* Start after all badblocks */
1114	if ((prev + 1) >= bb->count && !overlap_front(bb, prev, &bad)) {
1115		len = sectors;
1116		cleared++;
1117		goto update_sectors;
1118	}
1119
1120	/* Clear will split a bad record but the table is full */
1121	if (badblocks_full(bb) && (BB_OFFSET(p[prev]) < bad.start) &&
1122	    (BB_END(p[prev]) > (bad.start + sectors))) {
1123		len = sectors;
1124		goto update_sectors;
1125	}
1126
1127	if (overlap_front(bb, prev, &bad)) {
1128		if ((BB_OFFSET(p[prev]) < bad.start) &&
1129		    (BB_END(p[prev]) > (bad.start + bad.len))) {
1130			/* Splitting */
1131			if ((bb->count + 1) <= MAX_BADBLOCKS) {
1132				len = front_splitting_clear(bb, prev, &bad);
1133				bb->count += 1;
1134				cleared++;
1135			} else {
1136				/* No space to split, give up */
1137				len = sectors;
1138			}
1139		} else {
1140			int deleted = 0;
1141
1142			len = front_clear(bb, prev, &bad, &deleted);
1143			bb->count -= deleted;
1144			cleared++;
1145			hint = prev;
1146		}
1147
1148		goto update_sectors;
1149	}
1150
1151	/* Not front overlap, but behind overlap */
1152	if ((prev + 1) < bb->count && overlap_behind(bb, &bad, prev + 1)) {
1153		len = BB_OFFSET(p[prev + 1]) - bad.start;
1154		hint = prev + 1;
1155		/* Clear non-bad range should be treated as successful */
1156		cleared++;
1157		goto update_sectors;
1158	}
1159
1160	/* Not cover any badblocks range in the table */
1161	len = sectors;
1162	/* Clear non-bad range should be treated as successful */
1163	cleared++;
1164
1165update_sectors:
1166	s += len;
1167	sectors -= len;
1168
1169	if (sectors > 0)
1170		goto re_clear;
1171
1172	if (cleared) {
1173		badblocks_update_acked(bb);
1174		set_changed(bb);
1175	}
1176
1177	write_sequnlock_irq(&bb->lock);
1178
1179	if (!cleared)
1180		return false;
1181
1182	return true;
1183}
1184
1185/* Do the exact work to check bad blocks range from the bad block table */
1186static int _badblocks_check(struct badblocks *bb, sector_t s, sector_t sectors,
1187			    sector_t *first_bad, sector_t *bad_sectors)
1188{
1189	int prev = -1, hint = -1, set = 0;
1190	struct badblocks_context bad;
1191	int unacked_badblocks = 0;
1192	int acked_badblocks = 0;
1193	u64 *p = bb->page;
1194	int len, rv;
1195
1196re_check:
1197	bad.start = s;
1198	bad.len = sectors;
1199
1200	if (badblocks_empty(bb)) {
1201		len = sectors;
1202		goto update_sectors;
1203	}
1204
1205	prev = prev_badblocks(bb, &bad, hint);
1206
1207	/* start after all badblocks */
1208	if ((prev >= 0) &&
1209	    ((prev + 1) >= bb->count) && !overlap_front(bb, prev, &bad)) {
1210		len = sectors;
1211		goto update_sectors;
1212	}
1213
1214	/* Overlapped with front badblocks record */
1215	if ((prev >= 0) && overlap_front(bb, prev, &bad)) {
1216		if (BB_ACK(p[prev]))
1217			acked_badblocks++;
1218		else
1219			unacked_badblocks++;
1220
1221		if (BB_END(p[prev]) >= (s + sectors))
1222			len = sectors;
1223		else
1224			len = BB_END(p[prev]) - s;
1225
1226		if (set == 0) {
1227			*first_bad = BB_OFFSET(p[prev]);
1228			*bad_sectors = BB_LEN(p[prev]);
1229			set = 1;
1230		}
1231		goto update_sectors;
1232	}
1233
1234	/* Not front overlap, but behind overlap */
1235	if ((prev + 1) < bb->count && overlap_behind(bb, &bad, prev + 1)) {
1236		len = BB_OFFSET(p[prev + 1]) - bad.start;
1237		hint = prev + 1;
1238		goto update_sectors;
1239	}
1240
1241	/* not cover any badblocks range in the table */
1242	len = sectors;
1243
1244update_sectors:
1245	/* This situation should never happen */
1246	WARN_ON(sectors < len);
1247
1248	s += len;
1249	sectors -= len;
1250
1251	if (sectors > 0)
1252		goto re_check;
1253
1254	if (unacked_badblocks > 0)
1255		rv = -1;
1256	else if (acked_badblocks > 0)
1257		rv = 1;
1258	else
1259		rv = 0;
1260
1261	return rv;
1262}
1263
1264/**
1265 * badblocks_check() - check a given range for bad sectors
1266 * @bb:		the badblocks structure that holds all badblock information
1267 * @s:		sector (start) at which to check for badblocks
1268 * @sectors:	number of sectors to check for badblocks
1269 * @first_bad:	pointer to store location of the first badblock
1270 * @bad_sectors: pointer to store number of badblocks after @first_bad
1271 *
1272 * We can record which blocks on each device are 'bad' and so just
1273 * fail those blocks, or that stripe, rather than the whole device.
1274 * Entries in the bad-block table are 64bits wide.  This comprises:
1275 * Length of bad-range, in sectors: 0-511 for lengths 1-512
1276 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
1277 *  A 'shift' can be set so that larger blocks are tracked and
1278 *  consequently larger devices can be covered.
1279 * 'Acknowledged' flag - 1 bit. - the most significant bit.
1280 *
1281 * Locking of the bad-block table uses a seqlock so badblocks_check
1282 * might need to retry if it is very unlucky.
1283 * We will sometimes want to check for bad blocks in a bi_end_io function,
1284 * so we use the write_seqlock_irq variant.
1285 *
1286 * When looking for a bad block we specify a range and want to
1287 * know if any block in the range is bad.  So we binary-search
1288 * to the last range that starts at-or-before the given endpoint,
1289 * (or "before the sector after the target range")
1290 * then see if it ends after the given start.
1291 *
1292 * Return:
1293 *  0: there are no known bad blocks in the range
1294 *  1: there are known bad block which are all acknowledged
1295 * -1: there are bad blocks which have not yet been acknowledged in metadata.
1296 * plus the start/length of the first bad section we overlap.
1297 */
1298int badblocks_check(struct badblocks *bb, sector_t s, sector_t sectors,
1299			sector_t *first_bad, sector_t *bad_sectors)
1300{
1301	unsigned int seq;
1302	int rv;
1303
1304	WARN_ON(bb->shift < 0 || sectors == 0);
1305
1306	if (bb->shift > 0) {
1307		/* round the start down, and the end up */
1308		sector_t target = s + sectors;
1309
1310		rounddown(s, 1 << bb->shift);
1311		roundup(target, 1 << bb->shift);
1312		sectors = target - s;
1313	}
1314
1315retry:
1316	seq = read_seqbegin(&bb->lock);
1317	rv = _badblocks_check(bb, s, sectors, first_bad, bad_sectors);
1318	if (read_seqretry(&bb->lock, seq))
1319		goto retry;
1320
1321	return rv;
1322}
1323EXPORT_SYMBOL_GPL(badblocks_check);
1324
1325/**
1326 * badblocks_set() - Add a range of bad blocks to the table.
1327 * @bb:		the badblocks structure that holds all badblock information
1328 * @s:		first sector to mark as bad
1329 * @sectors:	number of sectors to mark as bad
1330 * @acknowledged: weather to mark the bad sectors as acknowledged
1331 *
1332 * This might extend the table, or might contract it if two adjacent ranges
1333 * can be merged. We binary-search to find the 'insertion' point, then
1334 * decide how best to handle it.
1335 *
1336 * Return:
1337 *  true: success
1338 *  false: failed to set badblocks (out of space). Parital setting will be
1339 *  treated as failure.
1340 */
1341bool badblocks_set(struct badblocks *bb, sector_t s, sector_t sectors,
1342		   int acknowledged)
1343{
1344	return _badblocks_set(bb, s, sectors, acknowledged);
1345}
1346EXPORT_SYMBOL_GPL(badblocks_set);
1347
1348/**
1349 * badblocks_clear() - Remove a range of bad blocks to the table.
1350 * @bb:		the badblocks structure that holds all badblock information
1351 * @s:		first sector to mark as bad
1352 * @sectors:	number of sectors to mark as bad
1353 *
1354 * This may involve extending the table if we spilt a region,
1355 * but it must not fail.  So if the table becomes full, we just
1356 * drop the remove request.
1357 *
1358 * Return:
1359 *  true: success
1360 *  false: failed to clear badblocks
1361 */
1362bool badblocks_clear(struct badblocks *bb, sector_t s, sector_t sectors)
1363{
1364	return _badblocks_clear(bb, s, sectors);
1365}
1366EXPORT_SYMBOL_GPL(badblocks_clear);
1367
1368/**
1369 * ack_all_badblocks() - Acknowledge all bad blocks in a list.
1370 * @bb:		the badblocks structure that holds all badblock information
1371 *
1372 * This only succeeds if ->changed is clear.  It is used by
1373 * in-kernel metadata updates
1374 */
1375void ack_all_badblocks(struct badblocks *bb)
1376{
1377	if (bb->page == NULL || bb->changed)
1378		/* no point even trying */
1379		return;
1380	write_seqlock_irq(&bb->lock);
1381
1382	if (bb->changed == 0 && bb->unacked_exist) {
1383		u64 *p = bb->page;
1384		int i;
1385
1386		for (i = 0; i < bb->count ; i++) {
1387			if (!BB_ACK(p[i])) {
1388				sector_t start = BB_OFFSET(p[i]);
1389				int len = BB_LEN(p[i]);
1390
1391				p[i] = BB_MAKE(start, len, 1);
1392			}
1393		}
1394
1395		for (i = 0; i < bb->count ; i++)
1396			while (try_adjacent_combine(bb, i))
1397				;
1398
1399		bb->unacked_exist = 0;
1400	}
1401	write_sequnlock_irq(&bb->lock);
1402}
1403EXPORT_SYMBOL_GPL(ack_all_badblocks);
1404
1405/**
1406 * badblocks_show() - sysfs access to bad-blocks list
1407 * @bb:		the badblocks structure that holds all badblock information
1408 * @page:	buffer received from sysfs
1409 * @unack:	weather to show unacknowledged badblocks
1410 *
1411 * Return:
1412 *  Length of returned data
1413 */
1414ssize_t badblocks_show(struct badblocks *bb, char *page, int unack)
1415{
1416	size_t len;
1417	int i;
1418	u64 *p = bb->page;
1419	unsigned seq;
1420
1421	if (bb->shift < 0)
1422		return 0;
1423
1424retry:
1425	seq = read_seqbegin(&bb->lock);
1426
1427	len = 0;
1428	i = 0;
1429
1430	while (len < PAGE_SIZE && i < bb->count) {
1431		sector_t s = BB_OFFSET(p[i]);
1432		unsigned int length = BB_LEN(p[i]);
1433		int ack = BB_ACK(p[i]);
1434
1435		i++;
1436
1437		if (unack && ack)
1438			continue;
1439
1440		len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
1441				(unsigned long long)s << bb->shift,
1442				length << bb->shift);
1443	}
1444	if (unack && len == 0)
1445		bb->unacked_exist = 0;
1446
1447	if (read_seqretry(&bb->lock, seq))
1448		goto retry;
1449
1450	return len;
1451}
1452EXPORT_SYMBOL_GPL(badblocks_show);
1453
1454/**
1455 * badblocks_store() - sysfs access to bad-blocks list
1456 * @bb:		the badblocks structure that holds all badblock information
1457 * @page:	buffer received from sysfs
1458 * @len:	length of data received from sysfs
1459 * @unack:	weather to show unacknowledged badblocks
1460 *
1461 * Return:
1462 *  Length of the buffer processed or -ve error.
1463 */
1464ssize_t badblocks_store(struct badblocks *bb, const char *page, size_t len,
1465			int unack)
1466{
1467	unsigned long long sector;
1468	int length;
1469	char newline;
1470
1471	switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
1472	case 3:
1473		if (newline != '\n')
1474			return -EINVAL;
1475		fallthrough;
1476	case 2:
1477		if (length <= 0)
1478			return -EINVAL;
1479		break;
1480	default:
1481		return -EINVAL;
1482	}
1483
1484	if (!badblocks_set(bb, sector, length, !unack))
1485		return -ENOSPC;
1486
1487	return len;
1488}
1489EXPORT_SYMBOL_GPL(badblocks_store);
1490
1491static int __badblocks_init(struct device *dev, struct badblocks *bb,
1492		int enable)
1493{
1494	bb->dev = dev;
1495	bb->count = 0;
1496	if (enable)
1497		bb->shift = 0;
1498	else
1499		bb->shift = -1;
1500	if (dev)
1501		bb->page = devm_kzalloc(dev, PAGE_SIZE, GFP_KERNEL);
1502	else
1503		bb->page = kzalloc(PAGE_SIZE, GFP_KERNEL);
1504	if (!bb->page) {
1505		bb->shift = -1;
1506		return -ENOMEM;
1507	}
1508	seqlock_init(&bb->lock);
1509
1510	return 0;
1511}
1512
1513/**
1514 * badblocks_init() - initialize the badblocks structure
1515 * @bb:		the badblocks structure that holds all badblock information
1516 * @enable:	weather to enable badblocks accounting
1517 *
1518 * Return:
1519 *  0: success
1520 *  -ve errno: on error
1521 */
1522int badblocks_init(struct badblocks *bb, int enable)
1523{
1524	return __badblocks_init(NULL, bb, enable);
1525}
1526EXPORT_SYMBOL_GPL(badblocks_init);
1527
1528int devm_init_badblocks(struct device *dev, struct badblocks *bb)
1529{
1530	if (!bb)
1531		return -EINVAL;
1532	return __badblocks_init(dev, bb, 1);
1533}
1534EXPORT_SYMBOL_GPL(devm_init_badblocks);
1535
1536/**
1537 * badblocks_exit() - free the badblocks structure
1538 * @bb:		the badblocks structure that holds all badblock information
1539 */
1540void badblocks_exit(struct badblocks *bb)
1541{
1542	if (!bb)
1543		return;
1544	if (bb->dev)
1545		devm_kfree(bb->dev, bb->page);
1546	else
1547		kfree(bb->page);
1548	bb->page = NULL;
1549}
1550EXPORT_SYMBOL_GPL(badblocks_exit);