drivers/mtd/mtdpstore.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #define dev_fmt(fmt) "mtdoops-pstore: " fmt
   4
   5 #include <linux/kernel.h>
   6 #include <linux/module.h>
   7 #include <linux/pstore_blk.h>
   8 #include <linux/mtd/mtd.h>
   9 #include <linux/bitops.h>
  10
  11 static struct mtdpstore_context {
  12         int index;
  13         struct pstore_blk_config info;
  14         struct pstore_device_info dev;
  15         struct mtd_info *mtd;
  16         unsigned long *rmmap;           /* removed bit map */
  17         unsigned long *usedmap;         /* used bit map */
  18         /*
  19          * used for panic write
  20          * As there are no block_isbad for panic case, we should keep this
  21          * status before panic to ensure panic_write not failed.
  22          */
  23         unsigned long *badmap;          /* bad block bit map */
  24 } oops_cxt;
  25
  26 static int mtdpstore_block_isbad(struct mtdpstore_context *cxt, loff_t off)
  27 {
  28         int ret;
  29         struct mtd_info *mtd = cxt->mtd;
  30         u64 blknum;
  31
  32         off = ALIGN_DOWN(off, mtd->erasesize);
  33         blknum = div_u64(off, mtd->erasesize);
  34
  35         if (test_bit(blknum, cxt->badmap))
  36                 return true;
  37         ret = mtd_block_isbad(mtd, off);
  38         if (ret < 0) {
  39                 dev_err(&mtd->dev, "mtd_block_isbad failed, aborting\n");
  40                 return ret;
  41         } else if (ret > 0) {
  42                 set_bit(blknum, cxt->badmap);
  43                 return true;
  44         }
  45         return false;
  46 }
  47
  48 static inline int mtdpstore_panic_block_isbad(struct mtdpstore_context *cxt,
  49                 loff_t off)
  50 {
  51         struct mtd_info *mtd = cxt->mtd;
  52         u64 blknum;
  53
  54         off = ALIGN_DOWN(off, mtd->erasesize);
  55         blknum = div_u64(off, mtd->erasesize);
  56         return test_bit(blknum, cxt->badmap);
  57 }
  58
  59 static inline void mtdpstore_mark_used(struct mtdpstore_context *cxt,
  60                 loff_t off)
  61 {
  62         struct mtd_info *mtd = cxt->mtd;
  63         u64 zonenum = div_u64(off, cxt->info.kmsg_size);
  64
  65         dev_dbg(&mtd->dev, "mark zone %llu used\n", zonenum);
  66         set_bit(zonenum, cxt->usedmap);
  67 }
  68
  69 static inline void mtdpstore_mark_unused(struct mtdpstore_context *cxt,
  70                 loff_t off)
  71 {
  72         struct mtd_info *mtd = cxt->mtd;
  73         u64 zonenum = div_u64(off, cxt->info.kmsg_size);
  74
  75         dev_dbg(&mtd->dev, "mark zone %llu unused\n", zonenum);
  76         clear_bit(zonenum, cxt->usedmap);
  77 }
  78
  79 static inline void mtdpstore_block_mark_unused(struct mtdpstore_context *cxt,
  80                 loff_t off)
  81 {
  82         struct mtd_info *mtd = cxt->mtd;
  83         u32 zonecnt = mtd->erasesize / cxt->info.kmsg_size;
  84         u64 zonenum;
  85
  86         off = ALIGN_DOWN(off, mtd->erasesize);
  87         zonenum = div_u64(off, cxt->info.kmsg_size);
  88         while (zonecnt > 0) {
  89                 dev_dbg(&mtd->dev, "mark zone %llu unused\n", zonenum);
  90                 clear_bit(zonenum, cxt->usedmap);
  91                 zonenum++;
  92                 zonecnt--;
  93         }
  94 }
  95
  96 static inline int mtdpstore_is_used(struct mtdpstore_context *cxt, loff_t off)
  97 {
  98         u64 zonenum = div_u64(off, cxt->info.kmsg_size);
  99         u64 blknum = div_u64(off, cxt->mtd->erasesize);
 100
 101         if (test_bit(blknum, cxt->badmap))
 102                 return true;
 103         return test_bit(zonenum, cxt->usedmap);
 104 }
 105
 106 static int mtdpstore_block_is_used(struct mtdpstore_context *cxt,
 107                 loff_t off)
 108 {
 109         struct mtd_info *mtd = cxt->mtd;
 110         u32 zonecnt = mtd->erasesize / cxt->info.kmsg_size;
 111         u64 zonenum;
 112
 113         off = ALIGN_DOWN(off, mtd->erasesize);
 114         zonenum = div_u64(off, cxt->info.kmsg_size);
 115         while (zonecnt > 0) {
 116                 if (test_bit(zonenum, cxt->usedmap))
 117                         return true;
 118                 zonenum++;
 119                 zonecnt--;
 120         }
 121         return false;
 122 }
 123
 124 static int mtdpstore_is_empty(struct mtdpstore_context *cxt, char *buf,
 125                 size_t size)
 126 {
 127         struct mtd_info *mtd = cxt->mtd;
 128         size_t sz;
 129         int i;
 130
 131         sz = min_t(uint32_t, size, mtd->writesize / 4);
 132         for (i = 0; i < sz; i++) {
 133                 if (buf[i] != (char)0xFF)
 134                         return false;
 135         }
 136         return true;
 137 }
 138
 139 static void mtdpstore_mark_removed(struct mtdpstore_context *cxt, loff_t off)
 140 {
 141         struct mtd_info *mtd = cxt->mtd;
 142         u64 zonenum = div_u64(off, cxt->info.kmsg_size);
 143
 144         dev_dbg(&mtd->dev, "mark zone %llu removed\n", zonenum);
 145         set_bit(zonenum, cxt->rmmap);
 146 }
 147
 148 static void mtdpstore_block_clear_removed(struct mtdpstore_context *cxt,
 149                 loff_t off)
 150 {
 151         struct mtd_info *mtd = cxt->mtd;
 152         u32 zonecnt = mtd->erasesize / cxt->info.kmsg_size;
 153         u64 zonenum;
 154
 155         off = ALIGN_DOWN(off, mtd->erasesize);
 156         zonenum = div_u64(off, cxt->info.kmsg_size);
 157         while (zonecnt > 0) {
 158                 clear_bit(zonenum, cxt->rmmap);
 159                 zonenum++;
 160                 zonecnt--;
 161         }
 162 }
 163
 164 static int mtdpstore_block_is_removed(struct mtdpstore_context *cxt,
 165                 loff_t off)
 166 {
 167         struct mtd_info *mtd = cxt->mtd;
 168         u32 zonecnt = mtd->erasesize / cxt->info.kmsg_size;
 169         u64 zonenum;
 170
 171         off = ALIGN_DOWN(off, mtd->erasesize);
 172         zonenum = div_u64(off, cxt->info.kmsg_size);
 173         while (zonecnt > 0) {
 174                 if (test_bit(zonenum, cxt->rmmap))
 175                         return true;
 176                 zonenum++;
 177                 zonecnt--;
 178         }
 179         return false;
 180 }
 181
 182 static int mtdpstore_erase_do(struct mtdpstore_context *cxt, loff_t off)
 183 {
 184         struct mtd_info *mtd = cxt->mtd;
 185         struct erase_info erase;
 186         int ret;
 187
 188         off = ALIGN_DOWN(off, cxt->mtd->erasesize);
 189         dev_dbg(&mtd->dev, "try to erase off 0x%llx\n", off);
 190         erase.len = cxt->mtd->erasesize;
 191         erase.addr = off;
 192         ret = mtd_erase(cxt->mtd, &erase);
 193         if (!ret)
 194                 mtdpstore_block_clear_removed(cxt, off);
 195         else
 196                 dev_err(&mtd->dev, "erase of region [0x%llx, 0x%llx] on \"%s\" failed\n",
 197                        (unsigned long long)erase.addr,
 198                        (unsigned long long)erase.len, cxt->info.device);
 199         return ret;
 200 }
 201
 202 /*
 203  * called while removing file
 204  *
 205  * Avoiding over erasing, do erase block only when the whole block is unused.
 206  * If the block contains valid log, do erase lazily on flush_removed() when
 207  * unregister.
 208  */
 209 static ssize_t mtdpstore_erase(size_t size, loff_t off)
 210 {
 211         struct mtdpstore_context *cxt = &oops_cxt;
 212
 213         if (mtdpstore_block_isbad(cxt, off))
 214                 return -EIO;
 215
 216         mtdpstore_mark_unused(cxt, off);
 217
 218         /* If the block still has valid data, mtdpstore do erase lazily */
 219         if (likely(mtdpstore_block_is_used(cxt, off))) {
 220                 mtdpstore_mark_removed(cxt, off);
 221                 return 0;
 222         }
 223
 224         /* all zones are unused, erase it */
 225         return mtdpstore_erase_do(cxt, off);
 226 }
 227
 228 /*
 229  * What is security for mtdpstore?
 230  * As there is no erase for panic case, we should ensure at least one zone
 231  * is writable. Otherwise, panic write will fail.
 232  * If zone is used, write operation will return -ENOMSG, which means that
 233  * pstore/blk will try one by one until gets an empty zone. So, it is not
 234  * needed to ensure the next zone is empty, but at least one.
 235  */
 236 static int mtdpstore_security(struct mtdpstore_context *cxt, loff_t off)
 237 {
 238         int ret = 0, i;
 239         struct mtd_info *mtd = cxt->mtd;
 240         u32 zonenum = (u32)div_u64(off, cxt->info.kmsg_size);
 241         u32 zonecnt = (u32)div_u64(cxt->mtd->size, cxt->info.kmsg_size);
 242         u32 blkcnt = (u32)div_u64(cxt->mtd->size, cxt->mtd->erasesize);
 243         u32 erasesize = cxt->mtd->erasesize;
 244
 245         for (i = 0; i < zonecnt; i++) {
 246                 u32 num = (zonenum + i) % zonecnt;
 247
 248                 /* found empty zone */
 249                 if (!test_bit(num, cxt->usedmap))
 250                         return 0;
 251         }
 252
 253         /* If there is no any empty zone, we have no way but to do erase */
 254         while (blkcnt--) {
 255                 div64_u64_rem(off + erasesize, cxt->mtd->size, (u64 *)&off);
 256
 257                 if (mtdpstore_block_isbad(cxt, off))
 258                         continue;
 259
 260                 ret = mtdpstore_erase_do(cxt, off);
 261                 if (!ret) {
 262                         mtdpstore_block_mark_unused(cxt, off);
 263                         break;
 264                 }
 265         }
 266
 267         if (ret)
 268                 dev_err(&mtd->dev, "all blocks bad!\n");
 269         dev_dbg(&mtd->dev, "end security\n");
 270         return ret;
 271 }
 272
 273 static ssize_t mtdpstore_write(const char *buf, size_t size, loff_t off)
 274 {
 275         struct mtdpstore_context *cxt = &oops_cxt;
 276         struct mtd_info *mtd = cxt->mtd;
 277         size_t retlen;
 278         int ret;
 279
 280         if (mtdpstore_block_isbad(cxt, off))
 281                 return -ENOMSG;
 282
 283         /* zone is used, please try next one */
 284         if (mtdpstore_is_used(cxt, off))
 285                 return -ENOMSG;
 286
 287         dev_dbg(&mtd->dev, "try to write off 0x%llx size %zu\n", off, size);
 288         ret = mtd_write(cxt->mtd, off, size, &retlen, (u_char *)buf);
 289         if (ret < 0 || retlen != size) {
 290                 dev_err(&mtd->dev, "write failure at %lld (%zu of %zu written), err %d\n",
 291                                 off, retlen, size, ret);
 292                 return -EIO;
 293         }
 294         mtdpstore_mark_used(cxt, off);
 295
 296         mtdpstore_security(cxt, off);
 297         return retlen;
 298 }
 299
 300 static inline bool mtdpstore_is_io_error(int ret)
 301 {
 302         return ret < 0 && !mtd_is_bitflip(ret) && !mtd_is_eccerr(ret);
 303 }
 304
 305 /*
 306  * All zones will be read as pstore/blk will read zone one by one when do
 307  * recover.
 308  */
 309 static ssize_t mtdpstore_read(char *buf, size_t size, loff_t off)
 310 {
 311         struct mtdpstore_context *cxt = &oops_cxt;
 312         struct mtd_info *mtd = cxt->mtd;
 313         size_t retlen, done;
 314         int ret;
 315
 316         if (mtdpstore_block_isbad(cxt, off))
 317                 return -ENOMSG;
 318
 319         dev_dbg(&mtd->dev, "try to read off 0x%llx size %zu\n", off, size);
 320         for (done = 0, retlen = 0; done < size; done += retlen) {
 321                 retlen = 0;
 322
 323                 ret = mtd_read(cxt->mtd, off + done, size - done, &retlen,
 324                                 (u_char *)buf + done);
 325                 if (mtdpstore_is_io_error(ret)) {
 326                         dev_err(&mtd->dev, "read failure at %lld (%zu of %zu read), err %d\n",
 327                                         off + done, retlen, size - done, ret);
 328                         /* the zone may be broken, try next one */
 329                         return -ENOMSG;
 330                 }
 331
 332                 /*
 333                  * ECC error. The impact on log data is so small. Maybe we can
 334                  * still read it and try to understand. So mtdpstore just hands
 335                  * over what it gets and user can judge whether the data is
 336                  * valid or not.
 337                  */
 338                 if (mtd_is_eccerr(ret)) {
 339                         dev_err(&mtd->dev, "ecc error at %lld (%zu of %zu read), err %d\n",
 340                                         off + done, retlen, size - done, ret);
 341                         /* driver may not set retlen when ecc error */
 342                         retlen = retlen == 0 ? size - done : retlen;
 343                 }
 344         }
 345
 346         if (mtdpstore_is_empty(cxt, buf, size))
 347                 mtdpstore_mark_unused(cxt, off);
 348         else
 349                 mtdpstore_mark_used(cxt, off);
 350
 351         mtdpstore_security(cxt, off);
 352         return retlen;
 353 }
 354
 355 static ssize_t mtdpstore_panic_write(const char *buf, size_t size, loff_t off)
 356 {
 357         struct mtdpstore_context *cxt = &oops_cxt;
 358         struct mtd_info *mtd = cxt->mtd;
 359         size_t retlen;
 360         int ret;
 361
 362         if (mtdpstore_panic_block_isbad(cxt, off))
 363                 return -ENOMSG;
 364
 365         /* zone is used, please try next one */
 366         if (mtdpstore_is_used(cxt, off))
 367                 return -ENOMSG;
 368
 369         ret = mtd_panic_write(cxt->mtd, off, size, &retlen, (u_char *)buf);
 370         if (ret < 0 || size != retlen) {
 371                 dev_err(&mtd->dev, "panic write failure at %lld (%zu of %zu read), err %d\n",
 372                                 off, retlen, size, ret);
 373                 return -EIO;
 374         }
 375         mtdpstore_mark_used(cxt, off);
 376
 377         return retlen;
 378 }
 379
 380 static void mtdpstore_notify_add(struct mtd_info *mtd)
 381 {
 382         int ret;
 383         struct mtdpstore_context *cxt = &oops_cxt;
 384         struct pstore_blk_config *info = &cxt->info;
 385         unsigned long longcnt;
 386
 387         if (!strcmp(mtd->name, info->device))
 388                 cxt->index = mtd->index;
 389
 390         if (mtd->index != cxt->index || cxt->index < 0)
 391                 return;
 392
 393         dev_dbg(&mtd->dev, "found matching MTD device %s\n", mtd->name);
 394
 395         if (mtd->size < info->kmsg_size * 2) {
 396                 dev_err(&mtd->dev, "MTD partition %d not big enough\n",
 397                                 mtd->index);
 398                 return;
 399         }
 400         /*
 401          * kmsg_size must be aligned to 4096 Bytes, which is limited by
 402          * psblk. The default value of kmsg_size is 64KB. If kmsg_size
 403          * is larger than erasesize, some errors will occur since mtdpsotre
 404          * is designed on it.
 405          */
 406         if (mtd->erasesize < info->kmsg_size) {
 407                 dev_err(&mtd->dev, "eraseblock size of MTD partition %d too small\n",
 408                                 mtd->index);
 409                 return;
 410         }
 411         if (unlikely(info->kmsg_size % mtd->writesize)) {
 412                 dev_err(&mtd->dev, "record size %lu KB must align to write size %d KB\n",
 413                                 info->kmsg_size / 1024,
 414                                 mtd->writesize / 1024);
 415                 return;
 416         }
 417
 418         longcnt = BITS_TO_LONGS(div_u64(mtd->size, info->kmsg_size));
 419         cxt->rmmap = kcalloc(longcnt, sizeof(long), GFP_KERNEL);
 420         cxt->usedmap = kcalloc(longcnt, sizeof(long), GFP_KERNEL);
 421
 422         longcnt = BITS_TO_LONGS(div_u64(mtd->size, mtd->erasesize));
 423         cxt->badmap = kcalloc(longcnt, sizeof(long), GFP_KERNEL);
 424
 425         cxt->dev.total_size = mtd->size;
 426         /* just support dmesg right now */
 427         cxt->dev.flags = PSTORE_FLAGS_DMESG;
 428         cxt->dev.read = mtdpstore_read;
 429         cxt->dev.write = mtdpstore_write;
 430         cxt->dev.erase = mtdpstore_erase;
 431         cxt->dev.panic_write = mtdpstore_panic_write;
 432
 433         ret = register_pstore_device(&cxt->dev);
 434         if (ret) {
 435                 dev_err(&mtd->dev, "mtd%d register to psblk failed\n",
 436                                 mtd->index);
 437                 return;
 438         }
 439         cxt->mtd = mtd;
 440         dev_info(&mtd->dev, "Attached to MTD device %d\n", mtd->index);
 441 }
 442
 443 static int mtdpstore_flush_removed_do(struct mtdpstore_context *cxt,
 444                 loff_t off, size_t size)
 445 {
 446         struct mtd_info *mtd = cxt->mtd;
 447         u_char *buf;
 448         int ret;
 449         size_t retlen;
 450         struct erase_info erase;
 451
 452         buf = kmalloc(mtd->erasesize, GFP_KERNEL);
 453         if (!buf)
 454                 return -ENOMEM;
 455
 456         /* 1st. read to cache */
 457         ret = mtd_read(mtd, off, mtd->erasesize, &retlen, buf);
 458         if (mtdpstore_is_io_error(ret))
 459                 goto free;
 460
 461         /* 2nd. erase block */
 462         erase.len = mtd->erasesize;
 463         erase.addr = off;
 464         ret = mtd_erase(mtd, &erase);
 465         if (ret)
 466                 goto free;
 467
 468         /* 3rd. write back */
 469         while (size) {
 470                 unsigned int zonesize = cxt->info.kmsg_size;
 471
 472                 /* there is valid data on block, write back */
 473                 if (mtdpstore_is_used(cxt, off)) {
 474                         ret = mtd_write(mtd, off, zonesize, &retlen, buf);
 475                         if (ret)
 476                                 dev_err(&mtd->dev, "write failure at %lld (%zu of %u written), err %d\n",
 477                                                 off, retlen, zonesize, ret);
 478                 }
 479
 480                 off += zonesize;
 481                 size -= min_t(unsigned int, zonesize, size);
 482         }
 483
 484 free:
 485         kfree(buf);
 486         return ret;
 487 }
 488
 489 /*
 490  * What does mtdpstore_flush_removed() do?
 491  * When user remove any log file on pstore filesystem, mtdpstore should do
 492  * something to ensure log file removed. If the whole block is no longer used,
 493  * it's nice to erase the block. However if the block still contains valid log,
 494  * what mtdpstore can do is to erase and write the valid log back.
 495  */
 496 static int mtdpstore_flush_removed(struct mtdpstore_context *cxt)
 497 {
 498         struct mtd_info *mtd = cxt->mtd;
 499         int ret;
 500         loff_t off;
 501         u32 blkcnt = (u32)div_u64(mtd->size, mtd->erasesize);
 502
 503         for (off = 0; blkcnt > 0; blkcnt--, off += mtd->erasesize) {
 504                 ret = mtdpstore_block_isbad(cxt, off);
 505                 if (ret)
 506                         continue;
 507
 508                 ret = mtdpstore_block_is_removed(cxt, off);
 509                 if (!ret)
 510                         continue;
 511
 512                 ret = mtdpstore_flush_removed_do(cxt, off, mtd->erasesize);
 513                 if (ret)
 514                         return ret;
 515         }
 516         return 0;
 517 }
 518
 519 static void mtdpstore_notify_remove(struct mtd_info *mtd)
 520 {
 521         struct mtdpstore_context *cxt = &oops_cxt;
 522
 523         if (mtd->index != cxt->index || cxt->index < 0)
 524                 return;
 525
 526         mtdpstore_flush_removed(cxt);
 527
 528         unregister_pstore_device(&cxt->dev);
 529         kfree(cxt->badmap);
 530         kfree(cxt->usedmap);
 531         kfree(cxt->rmmap);
 532         cxt->mtd = NULL;
 533         cxt->index = -1;
 534 }
 535
 536 static struct mtd_notifier mtdpstore_notifier = {
 537         .add    = mtdpstore_notify_add,
 538         .remove = mtdpstore_notify_remove,
 539 };
 540
 541 static int __init mtdpstore_init(void)
 542 {
 543         int ret;
 544         struct mtdpstore_context *cxt = &oops_cxt;
 545         struct pstore_blk_config *info = &cxt->info;
 546
 547         ret = pstore_blk_get_config(info);
 548         if (unlikely(ret))
 549                 return ret;
 550
 551         if (strlen(info->device) == 0) {
 552                 pr_err("mtd device must be supplied (device name is empty)\n");
 553                 return -EINVAL;
 554         }
 555         if (!info->kmsg_size) {
 556                 pr_err("no backend enabled (kmsg_size is 0)\n");
 557                 return -EINVAL;
 558         }
 559
 560         /* Setup the MTD device to use */
 561         ret = kstrtoint((char *)info->device, 0, &cxt->index);
 562         if (ret)
 563                 cxt->index = -1;
 564
 565         register_mtd_user(&mtdpstore_notifier);
 566         return 0;
 567 }
 568 module_init(mtdpstore_init);
 569
 570 static void __exit mtdpstore_exit(void)
 571 {
 572         unregister_mtd_user(&mtdpstore_notifier);
 573 }
 574 module_exit(mtdpstore_exit);
 575
 576 MODULE_LICENSE("GPL");
 577 MODULE_AUTHOR("WeiXiong Liao <liaoweixiong@allwinnertech.com>");
 578 MODULE_DESCRIPTION("MTD backend for pstore/blk");