uboot/drivers/mtd/ubi/eba.c
<<
>>
Prefs
   1/*
   2 * Copyright (c) International Business Machines Corp., 2006
   3 *
   4 * SPDX-License-Identifier:     GPL-2.0+
   5 *
   6 * Author: Artem Bityutskiy (Битюцкий Артём)
   7 */
   8
   9/*
  10 * The UBI Eraseblock Association (EBA) sub-system.
  11 *
  12 * This sub-system is responsible for I/O to/from logical eraseblock.
  13 *
  14 * Although in this implementation the EBA table is fully kept and managed in
  15 * RAM, which assumes poor scalability, it might be (partially) maintained on
  16 * flash in future implementations.
  17 *
  18 * The EBA sub-system implements per-logical eraseblock locking. Before
  19 * accessing a logical eraseblock it is locked for reading or writing. The
  20 * per-logical eraseblock locking is implemented by means of the lock tree. The
  21 * lock tree is an RB-tree which refers all the currently locked logical
  22 * eraseblocks. The lock tree elements are &struct ubi_ltree_entry objects.
  23 * They are indexed by (@vol_id, @lnum) pairs.
  24 *
  25 * EBA also maintains the global sequence counter which is incremented each
  26 * time a logical eraseblock is mapped to a physical eraseblock and it is
  27 * stored in the volume identifier header. This means that each VID header has
  28 * a unique sequence number. The sequence number is only increased an we assume
  29 * 64 bits is enough to never overflow.
  30 */
  31
  32#ifndef __UBOOT__
  33#include <linux/slab.h>
  34#include <linux/crc32.h>
  35#else
  36#include <ubi_uboot.h>
  37#endif
  38
  39#include <linux/err.h>
  40#include "ubi.h"
  41
  42/* Number of physical eraseblocks reserved for atomic LEB change operation */
  43#define EBA_RESERVED_PEBS 1
  44
  45/**
  46 * next_sqnum - get next sequence number.
  47 * @ubi: UBI device description object
  48 *
  49 * This function returns next sequence number to use, which is just the current
  50 * global sequence counter value. It also increases the global sequence
  51 * counter.
  52 */
  53unsigned long long ubi_next_sqnum(struct ubi_device *ubi)
  54{
  55        unsigned long long sqnum;
  56
  57        spin_lock(&ubi->ltree_lock);
  58        sqnum = ubi->global_sqnum++;
  59        spin_unlock(&ubi->ltree_lock);
  60
  61        return sqnum;
  62}
  63
  64/**
  65 * ubi_get_compat - get compatibility flags of a volume.
  66 * @ubi: UBI device description object
  67 * @vol_id: volume ID
  68 *
  69 * This function returns compatibility flags for an internal volume. User
  70 * volumes have no compatibility flags, so %0 is returned.
  71 */
  72static int ubi_get_compat(const struct ubi_device *ubi, int vol_id)
  73{
  74        if (vol_id == UBI_LAYOUT_VOLUME_ID)
  75                return UBI_LAYOUT_VOLUME_COMPAT;
  76        return 0;
  77}
  78
  79/**
  80 * ltree_lookup - look up the lock tree.
  81 * @ubi: UBI device description object
  82 * @vol_id: volume ID
  83 * @lnum: logical eraseblock number
  84 *
  85 * This function returns a pointer to the corresponding &struct ubi_ltree_entry
  86 * object if the logical eraseblock is locked and %NULL if it is not.
  87 * @ubi->ltree_lock has to be locked.
  88 */
  89static struct ubi_ltree_entry *ltree_lookup(struct ubi_device *ubi, int vol_id,
  90                                            int lnum)
  91{
  92        struct rb_node *p;
  93
  94        p = ubi->ltree.rb_node;
  95        while (p) {
  96                struct ubi_ltree_entry *le;
  97
  98                le = rb_entry(p, struct ubi_ltree_entry, rb);
  99
 100                if (vol_id < le->vol_id)
 101                        p = p->rb_left;
 102                else if (vol_id > le->vol_id)
 103                        p = p->rb_right;
 104                else {
 105                        if (lnum < le->lnum)
 106                                p = p->rb_left;
 107                        else if (lnum > le->lnum)
 108                                p = p->rb_right;
 109                        else
 110                                return le;
 111                }
 112        }
 113
 114        return NULL;
 115}
 116
 117/**
 118 * ltree_add_entry - add new entry to the lock tree.
 119 * @ubi: UBI device description object
 120 * @vol_id: volume ID
 121 * @lnum: logical eraseblock number
 122 *
 123 * This function adds new entry for logical eraseblock (@vol_id, @lnum) to the
 124 * lock tree. If such entry is already there, its usage counter is increased.
 125 * Returns pointer to the lock tree entry or %-ENOMEM if memory allocation
 126 * failed.
 127 */
 128static struct ubi_ltree_entry *ltree_add_entry(struct ubi_device *ubi,
 129                                               int vol_id, int lnum)
 130{
 131        struct ubi_ltree_entry *le, *le1, *le_free;
 132
 133        le = kmalloc(sizeof(struct ubi_ltree_entry), GFP_NOFS);
 134        if (!le)
 135                return ERR_PTR(-ENOMEM);
 136
 137        le->users = 0;
 138        init_rwsem(&le->mutex);
 139        le->vol_id = vol_id;
 140        le->lnum = lnum;
 141
 142        spin_lock(&ubi->ltree_lock);
 143        le1 = ltree_lookup(ubi, vol_id, lnum);
 144
 145        if (le1) {
 146                /*
 147                 * This logical eraseblock is already locked. The newly
 148                 * allocated lock entry is not needed.
 149                 */
 150                le_free = le;
 151                le = le1;
 152        } else {
 153                struct rb_node **p, *parent = NULL;
 154
 155                /*
 156                 * No lock entry, add the newly allocated one to the
 157                 * @ubi->ltree RB-tree.
 158                 */
 159                le_free = NULL;
 160
 161                p = &ubi->ltree.rb_node;
 162                while (*p) {
 163                        parent = *p;
 164                        le1 = rb_entry(parent, struct ubi_ltree_entry, rb);
 165
 166                        if (vol_id < le1->vol_id)
 167                                p = &(*p)->rb_left;
 168                        else if (vol_id > le1->vol_id)
 169                                p = &(*p)->rb_right;
 170                        else {
 171                                ubi_assert(lnum != le1->lnum);
 172                                if (lnum < le1->lnum)
 173                                        p = &(*p)->rb_left;
 174                                else
 175                                        p = &(*p)->rb_right;
 176                        }
 177                }
 178
 179                rb_link_node(&le->rb, parent, p);
 180                rb_insert_color(&le->rb, &ubi->ltree);
 181        }
 182        le->users += 1;
 183        spin_unlock(&ubi->ltree_lock);
 184
 185        kfree(le_free);
 186        return le;
 187}
 188
 189/**
 190 * leb_read_lock - lock logical eraseblock for reading.
 191 * @ubi: UBI device description object
 192 * @vol_id: volume ID
 193 * @lnum: logical eraseblock number
 194 *
 195 * This function locks a logical eraseblock for reading. Returns zero in case
 196 * of success and a negative error code in case of failure.
 197 */
 198static int leb_read_lock(struct ubi_device *ubi, int vol_id, int lnum)
 199{
 200        struct ubi_ltree_entry *le;
 201
 202        le = ltree_add_entry(ubi, vol_id, lnum);
 203        if (IS_ERR(le))
 204                return PTR_ERR(le);
 205        down_read(&le->mutex);
 206        return 0;
 207}
 208
 209/**
 210 * leb_read_unlock - unlock logical eraseblock.
 211 * @ubi: UBI device description object
 212 * @vol_id: volume ID
 213 * @lnum: logical eraseblock number
 214 */
 215static void leb_read_unlock(struct ubi_device *ubi, int vol_id, int lnum)
 216{
 217        struct ubi_ltree_entry *le;
 218
 219        spin_lock(&ubi->ltree_lock);
 220        le = ltree_lookup(ubi, vol_id, lnum);
 221        le->users -= 1;
 222        ubi_assert(le->users >= 0);
 223        up_read(&le->mutex);
 224        if (le->users == 0) {
 225                rb_erase(&le->rb, &ubi->ltree);
 226                kfree(le);
 227        }
 228        spin_unlock(&ubi->ltree_lock);
 229}
 230
 231/**
 232 * leb_write_lock - lock logical eraseblock for writing.
 233 * @ubi: UBI device description object
 234 * @vol_id: volume ID
 235 * @lnum: logical eraseblock number
 236 *
 237 * This function locks a logical eraseblock for writing. Returns zero in case
 238 * of success and a negative error code in case of failure.
 239 */
 240static int leb_write_lock(struct ubi_device *ubi, int vol_id, int lnum)
 241{
 242        struct ubi_ltree_entry *le;
 243
 244        le = ltree_add_entry(ubi, vol_id, lnum);
 245        if (IS_ERR(le))
 246                return PTR_ERR(le);
 247        down_write(&le->mutex);
 248        return 0;
 249}
 250
 251/**
 252 * leb_write_lock - lock logical eraseblock for writing.
 253 * @ubi: UBI device description object
 254 * @vol_id: volume ID
 255 * @lnum: logical eraseblock number
 256 *
 257 * This function locks a logical eraseblock for writing if there is no
 258 * contention and does nothing if there is contention. Returns %0 in case of
 259 * success, %1 in case of contention, and and a negative error code in case of
 260 * failure.
 261 */
 262static int leb_write_trylock(struct ubi_device *ubi, int vol_id, int lnum)
 263{
 264        struct ubi_ltree_entry *le;
 265
 266        le = ltree_add_entry(ubi, vol_id, lnum);
 267        if (IS_ERR(le))
 268                return PTR_ERR(le);
 269        if (down_write_trylock(&le->mutex))
 270                return 0;
 271
 272        /* Contention, cancel */
 273        spin_lock(&ubi->ltree_lock);
 274        le->users -= 1;
 275        ubi_assert(le->users >= 0);
 276        if (le->users == 0) {
 277                rb_erase(&le->rb, &ubi->ltree);
 278                kfree(le);
 279        }
 280        spin_unlock(&ubi->ltree_lock);
 281
 282        return 1;
 283}
 284
 285/**
 286 * leb_write_unlock - unlock logical eraseblock.
 287 * @ubi: UBI device description object
 288 * @vol_id: volume ID
 289 * @lnum: logical eraseblock number
 290 */
 291static void leb_write_unlock(struct ubi_device *ubi, int vol_id, int lnum)
 292{
 293        struct ubi_ltree_entry *le;
 294
 295        spin_lock(&ubi->ltree_lock);
 296        le = ltree_lookup(ubi, vol_id, lnum);
 297        le->users -= 1;
 298        ubi_assert(le->users >= 0);
 299        up_write(&le->mutex);
 300        if (le->users == 0) {
 301                rb_erase(&le->rb, &ubi->ltree);
 302                kfree(le);
 303        }
 304        spin_unlock(&ubi->ltree_lock);
 305}
 306
 307/**
 308 * ubi_eba_unmap_leb - un-map logical eraseblock.
 309 * @ubi: UBI device description object
 310 * @vol: volume description object
 311 * @lnum: logical eraseblock number
 312 *
 313 * This function un-maps logical eraseblock @lnum and schedules corresponding
 314 * physical eraseblock for erasure. Returns zero in case of success and a
 315 * negative error code in case of failure.
 316 */
 317int ubi_eba_unmap_leb(struct ubi_device *ubi, struct ubi_volume *vol,
 318                      int lnum)
 319{
 320        int err, pnum, vol_id = vol->vol_id;
 321
 322        if (ubi->ro_mode)
 323                return -EROFS;
 324
 325        err = leb_write_lock(ubi, vol_id, lnum);
 326        if (err)
 327                return err;
 328
 329        pnum = vol->eba_tbl[lnum];
 330        if (pnum < 0)
 331                /* This logical eraseblock is already unmapped */
 332                goto out_unlock;
 333
 334        dbg_eba("erase LEB %d:%d, PEB %d", vol_id, lnum, pnum);
 335
 336        down_read(&ubi->fm_eba_sem);
 337        vol->eba_tbl[lnum] = UBI_LEB_UNMAPPED;
 338        up_read(&ubi->fm_eba_sem);
 339        err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 0);
 340
 341out_unlock:
 342        leb_write_unlock(ubi, vol_id, lnum);
 343        return err;
 344}
 345
 346/**
 347 * ubi_eba_read_leb - read data.
 348 * @ubi: UBI device description object
 349 * @vol: volume description object
 350 * @lnum: logical eraseblock number
 351 * @buf: buffer to store the read data
 352 * @offset: offset from where to read
 353 * @len: how many bytes to read
 354 * @check: data CRC check flag
 355 *
 356 * If the logical eraseblock @lnum is unmapped, @buf is filled with 0xFF
 357 * bytes. The @check flag only makes sense for static volumes and forces
 358 * eraseblock data CRC checking.
 359 *
 360 * In case of success this function returns zero. In case of a static volume,
 361 * if data CRC mismatches - %-EBADMSG is returned. %-EBADMSG may also be
 362 * returned for any volume type if an ECC error was detected by the MTD device
 363 * driver. Other negative error cored may be returned in case of other errors.
 364 */
 365int ubi_eba_read_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
 366                     void *buf, int offset, int len, int check)
 367{
 368        int err, pnum, scrub = 0, vol_id = vol->vol_id;
 369        struct ubi_vid_hdr *vid_hdr;
 370        uint32_t uninitialized_var(crc);
 371
 372        err = leb_read_lock(ubi, vol_id, lnum);
 373        if (err)
 374                return err;
 375
 376        pnum = vol->eba_tbl[lnum];
 377        if (pnum < 0) {
 378                /*
 379                 * The logical eraseblock is not mapped, fill the whole buffer
 380                 * with 0xFF bytes. The exception is static volumes for which
 381                 * it is an error to read unmapped logical eraseblocks.
 382                 */
 383                dbg_eba("read %d bytes from offset %d of LEB %d:%d (unmapped)",
 384                        len, offset, vol_id, lnum);
 385                leb_read_unlock(ubi, vol_id, lnum);
 386                ubi_assert(vol->vol_type != UBI_STATIC_VOLUME);
 387                memset(buf, 0xFF, len);
 388                return 0;
 389        }
 390
 391        dbg_eba("read %d bytes from offset %d of LEB %d:%d, PEB %d",
 392                len, offset, vol_id, lnum, pnum);
 393
 394        if (vol->vol_type == UBI_DYNAMIC_VOLUME)
 395                check = 0;
 396
 397retry:
 398        if (check) {
 399                vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
 400                if (!vid_hdr) {
 401                        err = -ENOMEM;
 402                        goto out_unlock;
 403                }
 404
 405                err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
 406                if (err && err != UBI_IO_BITFLIPS) {
 407                        if (err > 0) {
 408                                /*
 409                                 * The header is either absent or corrupted.
 410                                 * The former case means there is a bug -
 411                                 * switch to read-only mode just in case.
 412                                 * The latter case means a real corruption - we
 413                                 * may try to recover data. FIXME: but this is
 414                                 * not implemented.
 415                                 */
 416                                if (err == UBI_IO_BAD_HDR_EBADMSG ||
 417                                    err == UBI_IO_BAD_HDR) {
 418                                        ubi_warn(ubi, "corrupted VID header at PEB %d, LEB %d:%d",
 419                                                 pnum, vol_id, lnum);
 420                                        err = -EBADMSG;
 421                                } else {
 422                                        err = -EINVAL;
 423                                        ubi_ro_mode(ubi);
 424                                }
 425                        }
 426                        goto out_free;
 427                } else if (err == UBI_IO_BITFLIPS)
 428                        scrub = 1;
 429
 430                ubi_assert(lnum < be32_to_cpu(vid_hdr->used_ebs));
 431                ubi_assert(len == be32_to_cpu(vid_hdr->data_size));
 432
 433                crc = be32_to_cpu(vid_hdr->data_crc);
 434                ubi_free_vid_hdr(ubi, vid_hdr);
 435        }
 436
 437        err = ubi_io_read_data(ubi, buf, pnum, offset, len);
 438        if (err) {
 439                if (err == UBI_IO_BITFLIPS)
 440                        scrub = 1;
 441                else if (mtd_is_eccerr(err)) {
 442                        if (vol->vol_type == UBI_DYNAMIC_VOLUME)
 443                                goto out_unlock;
 444                        scrub = 1;
 445                        if (!check) {
 446                                ubi_msg(ubi, "force data checking");
 447                                check = 1;
 448                                goto retry;
 449                        }
 450                } else
 451                        goto out_unlock;
 452        }
 453
 454        if (check) {
 455                uint32_t crc1 = crc32(UBI_CRC32_INIT, buf, len);
 456                if (crc1 != crc) {
 457                        ubi_warn(ubi, "CRC error: calculated %#08x, must be %#08x",
 458                                 crc1, crc);
 459                        err = -EBADMSG;
 460                        goto out_unlock;
 461                }
 462        }
 463
 464        if (scrub)
 465                err = ubi_wl_scrub_peb(ubi, pnum);
 466
 467        leb_read_unlock(ubi, vol_id, lnum);
 468        return err;
 469
 470out_free:
 471        ubi_free_vid_hdr(ubi, vid_hdr);
 472out_unlock:
 473        leb_read_unlock(ubi, vol_id, lnum);
 474        return err;
 475}
 476
 477#ifndef __UBOOT__
 478/**
 479 * ubi_eba_read_leb_sg - read data into a scatter gather list.
 480 * @ubi: UBI device description object
 481 * @vol: volume description object
 482 * @lnum: logical eraseblock number
 483 * @sgl: UBI scatter gather list to store the read data
 484 * @offset: offset from where to read
 485 * @len: how many bytes to read
 486 * @check: data CRC check flag
 487 *
 488 * This function works exactly like ubi_eba_read_leb(). But instead of
 489 * storing the read data into a buffer it writes to an UBI scatter gather
 490 * list.
 491 */
 492int ubi_eba_read_leb_sg(struct ubi_device *ubi, struct ubi_volume *vol,
 493                        struct ubi_sgl *sgl, int lnum, int offset, int len,
 494                        int check)
 495{
 496        int to_read;
 497        int ret;
 498        struct scatterlist *sg;
 499
 500        for (;;) {
 501                ubi_assert(sgl->list_pos < UBI_MAX_SG_COUNT);
 502                sg = &sgl->sg[sgl->list_pos];
 503                if (len < sg->length - sgl->page_pos)
 504                        to_read = len;
 505                else
 506                        to_read = sg->length - sgl->page_pos;
 507
 508                ret = ubi_eba_read_leb(ubi, vol, lnum,
 509                                       sg_virt(sg) + sgl->page_pos, offset,
 510                                       to_read, check);
 511                if (ret < 0)
 512                        return ret;
 513
 514                offset += to_read;
 515                len -= to_read;
 516                if (!len) {
 517                        sgl->page_pos += to_read;
 518                        if (sgl->page_pos == sg->length) {
 519                                sgl->list_pos++;
 520                                sgl->page_pos = 0;
 521                        }
 522
 523                        break;
 524                }
 525
 526                sgl->list_pos++;
 527                sgl->page_pos = 0;
 528        }
 529
 530        return ret;
 531}
 532#endif
 533
 534/**
 535 * recover_peb - recover from write failure.
 536 * @ubi: UBI device description object
 537 * @pnum: the physical eraseblock to recover
 538 * @vol_id: volume ID
 539 * @lnum: logical eraseblock number
 540 * @buf: data which was not written because of the write failure
 541 * @offset: offset of the failed write
 542 * @len: how many bytes should have been written
 543 *
 544 * This function is called in case of a write failure and moves all good data
 545 * from the potentially bad physical eraseblock to a good physical eraseblock.
 546 * This function also writes the data which was not written due to the failure.
 547 * Returns new physical eraseblock number in case of success, and a negative
 548 * error code in case of failure.
 549 */
 550static int recover_peb(struct ubi_device *ubi, int pnum, int vol_id, int lnum,
 551                       const void *buf, int offset, int len)
 552{
 553        int err, idx = vol_id2idx(ubi, vol_id), new_pnum, data_size, tries = 0;
 554        struct ubi_volume *vol = ubi->volumes[idx];
 555        struct ubi_vid_hdr *vid_hdr;
 556
 557        vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
 558        if (!vid_hdr)
 559                return -ENOMEM;
 560
 561retry:
 562        new_pnum = ubi_wl_get_peb(ubi);
 563        if (new_pnum < 0) {
 564                ubi_free_vid_hdr(ubi, vid_hdr);
 565                up_read(&ubi->fm_eba_sem);
 566                return new_pnum;
 567        }
 568
 569        ubi_msg(ubi, "recover PEB %d, move data to PEB %d",
 570                pnum, new_pnum);
 571
 572        err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
 573        if (err && err != UBI_IO_BITFLIPS) {
 574                if (err > 0)
 575                        err = -EIO;
 576                up_read(&ubi->fm_eba_sem);
 577                goto out_put;
 578        }
 579
 580        vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
 581        err = ubi_io_write_vid_hdr(ubi, new_pnum, vid_hdr);
 582        if (err) {
 583                up_read(&ubi->fm_eba_sem);
 584                goto write_error;
 585        }
 586
 587        data_size = offset + len;
 588        mutex_lock(&ubi->buf_mutex);
 589        memset(ubi->peb_buf + offset, 0xFF, len);
 590
 591        /* Read everything before the area where the write failure happened */
 592        if (offset > 0) {
 593                err = ubi_io_read_data(ubi, ubi->peb_buf, pnum, 0, offset);
 594                if (err && err != UBI_IO_BITFLIPS) {
 595                        up_read(&ubi->fm_eba_sem);
 596                        goto out_unlock;
 597                }
 598        }
 599
 600        memcpy(ubi->peb_buf + offset, buf, len);
 601
 602        err = ubi_io_write_data(ubi, ubi->peb_buf, new_pnum, 0, data_size);
 603        if (err) {
 604                mutex_unlock(&ubi->buf_mutex);
 605                up_read(&ubi->fm_eba_sem);
 606                goto write_error;
 607        }
 608
 609        mutex_unlock(&ubi->buf_mutex);
 610        ubi_free_vid_hdr(ubi, vid_hdr);
 611
 612        vol->eba_tbl[lnum] = new_pnum;
 613        up_read(&ubi->fm_eba_sem);
 614        ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
 615
 616        ubi_msg(ubi, "data was successfully recovered");
 617        return 0;
 618
 619out_unlock:
 620        mutex_unlock(&ubi->buf_mutex);
 621out_put:
 622        ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
 623        ubi_free_vid_hdr(ubi, vid_hdr);
 624        return err;
 625
 626write_error:
 627        /*
 628         * Bad luck? This physical eraseblock is bad too? Crud. Let's try to
 629         * get another one.
 630         */
 631        ubi_warn(ubi, "failed to write to PEB %d", new_pnum);
 632        ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
 633        if (++tries > UBI_IO_RETRIES) {
 634                ubi_free_vid_hdr(ubi, vid_hdr);
 635                return err;
 636        }
 637        ubi_msg(ubi, "try again");
 638        goto retry;
 639}
 640
 641/**
 642 * ubi_eba_write_leb - write data to dynamic volume.
 643 * @ubi: UBI device description object
 644 * @vol: volume description object
 645 * @lnum: logical eraseblock number
 646 * @buf: the data to write
 647 * @offset: offset within the logical eraseblock where to write
 648 * @len: how many bytes to write
 649 *
 650 * This function writes data to logical eraseblock @lnum of a dynamic volume
 651 * @vol. Returns zero in case of success and a negative error code in case
 652 * of failure. In case of error, it is possible that something was still
 653 * written to the flash media, but may be some garbage.
 654 */
 655int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
 656                      const void *buf, int offset, int len)
 657{
 658        int err, pnum, tries = 0, vol_id = vol->vol_id;
 659        struct ubi_vid_hdr *vid_hdr;
 660
 661        if (ubi->ro_mode)
 662                return -EROFS;
 663
 664        err = leb_write_lock(ubi, vol_id, lnum);
 665        if (err)
 666                return err;
 667
 668        pnum = vol->eba_tbl[lnum];
 669        if (pnum >= 0) {
 670                dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
 671                        len, offset, vol_id, lnum, pnum);
 672
 673                err = ubi_io_write_data(ubi, buf, pnum, offset, len);
 674                if (err) {
 675                        ubi_warn(ubi, "failed to write data to PEB %d", pnum);
 676                        if (err == -EIO && ubi->bad_allowed)
 677                                err = recover_peb(ubi, pnum, vol_id, lnum, buf,
 678                                                  offset, len);
 679                        if (err)
 680                                ubi_ro_mode(ubi);
 681                }
 682                leb_write_unlock(ubi, vol_id, lnum);
 683                return err;
 684        }
 685
 686        /*
 687         * The logical eraseblock is not mapped. We have to get a free physical
 688         * eraseblock and write the volume identifier header there first.
 689         */
 690        vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
 691        if (!vid_hdr) {
 692                leb_write_unlock(ubi, vol_id, lnum);
 693                return -ENOMEM;
 694        }
 695
 696        vid_hdr->vol_type = UBI_VID_DYNAMIC;
 697        vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
 698        vid_hdr->vol_id = cpu_to_be32(vol_id);
 699        vid_hdr->lnum = cpu_to_be32(lnum);
 700        vid_hdr->compat = ubi_get_compat(ubi, vol_id);
 701        vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
 702
 703retry:
 704        pnum = ubi_wl_get_peb(ubi);
 705        if (pnum < 0) {
 706                ubi_free_vid_hdr(ubi, vid_hdr);
 707                leb_write_unlock(ubi, vol_id, lnum);
 708                up_read(&ubi->fm_eba_sem);
 709                return pnum;
 710        }
 711
 712        dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
 713                len, offset, vol_id, lnum, pnum);
 714
 715        err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
 716        if (err) {
 717                ubi_warn(ubi, "failed to write VID header to LEB %d:%d, PEB %d",
 718                         vol_id, lnum, pnum);
 719                up_read(&ubi->fm_eba_sem);
 720                goto write_error;
 721        }
 722
 723        if (len) {
 724                err = ubi_io_write_data(ubi, buf, pnum, offset, len);
 725                if (err) {
 726                        ubi_warn(ubi, "failed to write %d bytes at offset %d of LEB %d:%d, PEB %d",
 727                                 len, offset, vol_id, lnum, pnum);
 728                        up_read(&ubi->fm_eba_sem);
 729                        goto write_error;
 730                }
 731        }
 732
 733        vol->eba_tbl[lnum] = pnum;
 734        up_read(&ubi->fm_eba_sem);
 735
 736        leb_write_unlock(ubi, vol_id, lnum);
 737        ubi_free_vid_hdr(ubi, vid_hdr);
 738        return 0;
 739
 740write_error:
 741        if (err != -EIO || !ubi->bad_allowed) {
 742                ubi_ro_mode(ubi);
 743                leb_write_unlock(ubi, vol_id, lnum);
 744                ubi_free_vid_hdr(ubi, vid_hdr);
 745                return err;
 746        }
 747
 748        /*
 749         * Fortunately, this is the first write operation to this physical
 750         * eraseblock, so just put it and request a new one. We assume that if
 751         * this physical eraseblock went bad, the erase code will handle that.
 752         */
 753        err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
 754        if (err || ++tries > UBI_IO_RETRIES) {
 755                ubi_ro_mode(ubi);
 756                leb_write_unlock(ubi, vol_id, lnum);
 757                ubi_free_vid_hdr(ubi, vid_hdr);
 758                return err;
 759        }
 760
 761        vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
 762        ubi_msg(ubi, "try another PEB");
 763        goto retry;
 764}
 765
 766/**
 767 * ubi_eba_write_leb_st - write data to static volume.
 768 * @ubi: UBI device description object
 769 * @vol: volume description object
 770 * @lnum: logical eraseblock number
 771 * @buf: data to write
 772 * @len: how many bytes to write
 773 * @used_ebs: how many logical eraseblocks will this volume contain
 774 *
 775 * This function writes data to logical eraseblock @lnum of static volume
 776 * @vol. The @used_ebs argument should contain total number of logical
 777 * eraseblock in this static volume.
 778 *
 779 * When writing to the last logical eraseblock, the @len argument doesn't have
 780 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
 781 * to the real data size, although the @buf buffer has to contain the
 782 * alignment. In all other cases, @len has to be aligned.
 783 *
 784 * It is prohibited to write more than once to logical eraseblocks of static
 785 * volumes. This function returns zero in case of success and a negative error
 786 * code in case of failure.
 787 */
 788int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
 789                         int lnum, const void *buf, int len, int used_ebs)
 790{
 791        int err, pnum, tries = 0, data_size = len, vol_id = vol->vol_id;
 792        struct ubi_vid_hdr *vid_hdr;
 793        uint32_t crc;
 794
 795        if (ubi->ro_mode)
 796                return -EROFS;
 797
 798        if (lnum == used_ebs - 1)
 799                /* If this is the last LEB @len may be unaligned */
 800                len = ALIGN(data_size, ubi->min_io_size);
 801        else
 802                ubi_assert(!(len & (ubi->min_io_size - 1)));
 803
 804        vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
 805        if (!vid_hdr)
 806                return -ENOMEM;
 807
 808        err = leb_write_lock(ubi, vol_id, lnum);
 809        if (err) {
 810                ubi_free_vid_hdr(ubi, vid_hdr);
 811                return err;
 812        }
 813
 814        vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
 815        vid_hdr->vol_id = cpu_to_be32(vol_id);
 816        vid_hdr->lnum = cpu_to_be32(lnum);
 817        vid_hdr->compat = ubi_get_compat(ubi, vol_id);
 818        vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
 819
 820        crc = crc32(UBI_CRC32_INIT, buf, data_size);
 821        vid_hdr->vol_type = UBI_VID_STATIC;
 822        vid_hdr->data_size = cpu_to_be32(data_size);
 823        vid_hdr->used_ebs = cpu_to_be32(used_ebs);
 824        vid_hdr->data_crc = cpu_to_be32(crc);
 825
 826retry:
 827        pnum = ubi_wl_get_peb(ubi);
 828        if (pnum < 0) {
 829                ubi_free_vid_hdr(ubi, vid_hdr);
 830                leb_write_unlock(ubi, vol_id, lnum);
 831                up_read(&ubi->fm_eba_sem);
 832                return pnum;
 833        }
 834
 835        dbg_eba("write VID hdr and %d bytes at LEB %d:%d, PEB %d, used_ebs %d",
 836                len, vol_id, lnum, pnum, used_ebs);
 837
 838        err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
 839        if (err) {
 840                ubi_warn(ubi, "failed to write VID header to LEB %d:%d, PEB %d",
 841                         vol_id, lnum, pnum);
 842                up_read(&ubi->fm_eba_sem);
 843                goto write_error;
 844        }
 845
 846        err = ubi_io_write_data(ubi, buf, pnum, 0, len);
 847        if (err) {
 848                ubi_warn(ubi, "failed to write %d bytes of data to PEB %d",
 849                         len, pnum);
 850                up_read(&ubi->fm_eba_sem);
 851                goto write_error;
 852        }
 853
 854        ubi_assert(vol->eba_tbl[lnum] < 0);
 855        vol->eba_tbl[lnum] = pnum;
 856        up_read(&ubi->fm_eba_sem);
 857
 858        leb_write_unlock(ubi, vol_id, lnum);
 859        ubi_free_vid_hdr(ubi, vid_hdr);
 860        return 0;
 861
 862write_error:
 863        if (err != -EIO || !ubi->bad_allowed) {
 864                /*
 865                 * This flash device does not admit of bad eraseblocks or
 866                 * something nasty and unexpected happened. Switch to read-only
 867                 * mode just in case.
 868                 */
 869                ubi_ro_mode(ubi);
 870                leb_write_unlock(ubi, vol_id, lnum);
 871                ubi_free_vid_hdr(ubi, vid_hdr);
 872                return err;
 873        }
 874
 875        err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
 876        if (err || ++tries > UBI_IO_RETRIES) {
 877                ubi_ro_mode(ubi);
 878                leb_write_unlock(ubi, vol_id, lnum);
 879                ubi_free_vid_hdr(ubi, vid_hdr);
 880                return err;
 881        }
 882
 883        vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
 884        ubi_msg(ubi, "try another PEB");
 885        goto retry;
 886}
 887
 888/*
 889 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
 890 * @ubi: UBI device description object
 891 * @vol: volume description object
 892 * @lnum: logical eraseblock number
 893 * @buf: data to write
 894 * @len: how many bytes to write
 895 *
 896 * This function changes the contents of a logical eraseblock atomically. @buf
 897 * has to contain new logical eraseblock data, and @len - the length of the
 898 * data, which has to be aligned. This function guarantees that in case of an
 899 * unclean reboot the old contents is preserved. Returns zero in case of
 900 * success and a negative error code in case of failure.
 901 *
 902 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
 903 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
 904 */
 905int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
 906                              int lnum, const void *buf, int len)
 907{
 908        int err, pnum, old_pnum, tries = 0, vol_id = vol->vol_id;
 909        struct ubi_vid_hdr *vid_hdr;
 910        uint32_t crc;
 911
 912        if (ubi->ro_mode)
 913                return -EROFS;
 914
 915        if (len == 0) {
 916                /*
 917                 * Special case when data length is zero. In this case the LEB
 918                 * has to be unmapped and mapped somewhere else.
 919                 */
 920                err = ubi_eba_unmap_leb(ubi, vol, lnum);
 921                if (err)
 922                        return err;
 923                return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0);
 924        }
 925
 926        vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
 927        if (!vid_hdr)
 928                return -ENOMEM;
 929
 930        mutex_lock(&ubi->alc_mutex);
 931        err = leb_write_lock(ubi, vol_id, lnum);
 932        if (err)
 933                goto out_mutex;
 934
 935        vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
 936        vid_hdr->vol_id = cpu_to_be32(vol_id);
 937        vid_hdr->lnum = cpu_to_be32(lnum);
 938        vid_hdr->compat = ubi_get_compat(ubi, vol_id);
 939        vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
 940
 941        crc = crc32(UBI_CRC32_INIT, buf, len);
 942        vid_hdr->vol_type = UBI_VID_DYNAMIC;
 943        vid_hdr->data_size = cpu_to_be32(len);
 944        vid_hdr->copy_flag = 1;
 945        vid_hdr->data_crc = cpu_to_be32(crc);
 946
 947retry:
 948        pnum = ubi_wl_get_peb(ubi);
 949        if (pnum < 0) {
 950                err = pnum;
 951                up_read(&ubi->fm_eba_sem);
 952                goto out_leb_unlock;
 953        }
 954
 955        dbg_eba("change LEB %d:%d, PEB %d, write VID hdr to PEB %d",
 956                vol_id, lnum, vol->eba_tbl[lnum], pnum);
 957
 958        err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
 959        if (err) {
 960                ubi_warn(ubi, "failed to write VID header to LEB %d:%d, PEB %d",
 961                         vol_id, lnum, pnum);
 962                up_read(&ubi->fm_eba_sem);
 963                goto write_error;
 964        }
 965
 966        err = ubi_io_write_data(ubi, buf, pnum, 0, len);
 967        if (err) {
 968                ubi_warn(ubi, "failed to write %d bytes of data to PEB %d",
 969                         len, pnum);
 970                up_read(&ubi->fm_eba_sem);
 971                goto write_error;
 972        }
 973
 974        old_pnum = vol->eba_tbl[lnum];
 975        vol->eba_tbl[lnum] = pnum;
 976        up_read(&ubi->fm_eba_sem);
 977
 978        if (old_pnum >= 0) {
 979                err = ubi_wl_put_peb(ubi, vol_id, lnum, old_pnum, 0);
 980                if (err)
 981                        goto out_leb_unlock;
 982        }
 983
 984out_leb_unlock:
 985        leb_write_unlock(ubi, vol_id, lnum);
 986out_mutex:
 987        mutex_unlock(&ubi->alc_mutex);
 988        ubi_free_vid_hdr(ubi, vid_hdr);
 989        return err;
 990
 991write_error:
 992        if (err != -EIO || !ubi->bad_allowed) {
 993                /*
 994                 * This flash device does not admit of bad eraseblocks or
 995                 * something nasty and unexpected happened. Switch to read-only
 996                 * mode just in case.
 997                 */
 998                ubi_ro_mode(ubi);
 999                goto out_leb_unlock;
1000        }
1001
1002        err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
1003        if (err || ++tries > UBI_IO_RETRIES) {
1004                ubi_ro_mode(ubi);
1005                goto out_leb_unlock;
1006        }
1007
1008        vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1009        ubi_msg(ubi, "try another PEB");
1010        goto retry;
1011}
1012
1013/**
1014 * is_error_sane - check whether a read error is sane.
1015 * @err: code of the error happened during reading
1016 *
1017 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
1018 * cannot read data from the target PEB (an error @err happened). If the error
1019 * code is sane, then we treat this error as non-fatal. Otherwise the error is
1020 * fatal and UBI will be switched to R/O mode later.
1021 *
1022 * The idea is that we try not to switch to R/O mode if the read error is
1023 * something which suggests there was a real read problem. E.g., %-EIO. Or a
1024 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
1025 * mode, simply because we do not know what happened at the MTD level, and we
1026 * cannot handle this. E.g., the underlying driver may have become crazy, and
1027 * it is safer to switch to R/O mode to preserve the data.
1028 *
1029 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
1030 * which we have just written.
1031 */
1032static int is_error_sane(int err)
1033{
1034        if (err == -EIO || err == -ENOMEM || err == UBI_IO_BAD_HDR ||
1035            err == UBI_IO_BAD_HDR_EBADMSG || err == -ETIMEDOUT)
1036                return 0;
1037        return 1;
1038}
1039
1040/**
1041 * ubi_eba_copy_leb - copy logical eraseblock.
1042 * @ubi: UBI device description object
1043 * @from: physical eraseblock number from where to copy
1044 * @to: physical eraseblock number where to copy
1045 * @vid_hdr: VID header of the @from physical eraseblock
1046 *
1047 * This function copies logical eraseblock from physical eraseblock @from to
1048 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
1049 * function. Returns:
1050 *   o %0 in case of success;
1051 *   o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_TARGET_BITFLIPS, etc;
1052 *   o a negative error code in case of failure.
1053 */
1054int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
1055                     struct ubi_vid_hdr *vid_hdr)
1056{
1057        int err, vol_id, lnum, data_size, aldata_size, idx;
1058        struct ubi_volume *vol;
1059        uint32_t crc;
1060
1061        vol_id = be32_to_cpu(vid_hdr->vol_id);
1062        lnum = be32_to_cpu(vid_hdr->lnum);
1063
1064        dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
1065
1066        if (vid_hdr->vol_type == UBI_VID_STATIC) {
1067                data_size = be32_to_cpu(vid_hdr->data_size);
1068                aldata_size = ALIGN(data_size, ubi->min_io_size);
1069        } else
1070                data_size = aldata_size =
1071                            ubi->leb_size - be32_to_cpu(vid_hdr->data_pad);
1072
1073        idx = vol_id2idx(ubi, vol_id);
1074        spin_lock(&ubi->volumes_lock);
1075        /*
1076         * Note, we may race with volume deletion, which means that the volume
1077         * this logical eraseblock belongs to might be being deleted. Since the
1078         * volume deletion un-maps all the volume's logical eraseblocks, it will
1079         * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1080         */
1081        vol = ubi->volumes[idx];
1082        spin_unlock(&ubi->volumes_lock);
1083        if (!vol) {
1084                /* No need to do further work, cancel */
1085                dbg_wl("volume %d is being removed, cancel", vol_id);
1086                return MOVE_CANCEL_RACE;
1087        }
1088
1089        /*
1090         * We do not want anybody to write to this logical eraseblock while we
1091         * are moving it, so lock it.
1092         *
1093         * Note, we are using non-waiting locking here, because we cannot sleep
1094         * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1095         * unmapping the LEB which is mapped to the PEB we are going to move
1096         * (@from). This task locks the LEB and goes sleep in the
1097         * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1098         * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1099         * LEB is already locked, we just do not move it and return
1100         * %MOVE_RETRY. Note, we do not return %MOVE_CANCEL_RACE here because
1101         * we do not know the reasons of the contention - it may be just a
1102         * normal I/O on this LEB, so we want to re-try.
1103         */
1104        err = leb_write_trylock(ubi, vol_id, lnum);
1105        if (err) {
1106                dbg_wl("contention on LEB %d:%d, cancel", vol_id, lnum);
1107                return MOVE_RETRY;
1108        }
1109
1110        /*
1111         * The LEB might have been put meanwhile, and the task which put it is
1112         * probably waiting on @ubi->move_mutex. No need to continue the work,
1113         * cancel it.
1114         */
1115        if (vol->eba_tbl[lnum] != from) {
1116                dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to PEB %d, cancel",
1117                       vol_id, lnum, from, vol->eba_tbl[lnum]);
1118                err = MOVE_CANCEL_RACE;
1119                goto out_unlock_leb;
1120        }
1121
1122        /*
1123         * OK, now the LEB is locked and we can safely start moving it. Since
1124         * this function utilizes the @ubi->peb_buf buffer which is shared
1125         * with some other functions - we lock the buffer by taking the
1126         * @ubi->buf_mutex.
1127         */
1128        mutex_lock(&ubi->buf_mutex);
1129        dbg_wl("read %d bytes of data", aldata_size);
1130        err = ubi_io_read_data(ubi, ubi->peb_buf, from, 0, aldata_size);
1131        if (err && err != UBI_IO_BITFLIPS) {
1132                ubi_warn(ubi, "error %d while reading data from PEB %d",
1133                         err, from);
1134                err = MOVE_SOURCE_RD_ERR;
1135                goto out_unlock_buf;
1136        }
1137
1138        /*
1139         * Now we have got to calculate how much data we have to copy. In
1140         * case of a static volume it is fairly easy - the VID header contains
1141         * the data size. In case of a dynamic volume it is more difficult - we
1142         * have to read the contents, cut 0xFF bytes from the end and copy only
1143         * the first part. We must do this to avoid writing 0xFF bytes as it
1144         * may have some side-effects. And not only this. It is important not
1145         * to include those 0xFFs to CRC because later the they may be filled
1146         * by data.
1147         */
1148        if (vid_hdr->vol_type == UBI_VID_DYNAMIC)
1149                aldata_size = data_size =
1150                        ubi_calc_data_len(ubi, ubi->peb_buf, data_size);
1151
1152        cond_resched();
1153        crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
1154        cond_resched();
1155
1156        /*
1157         * It may turn out to be that the whole @from physical eraseblock
1158         * contains only 0xFF bytes. Then we have to only write the VID header
1159         * and do not write any data. This also means we should not set
1160         * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1161         */
1162        if (data_size > 0) {
1163                vid_hdr->copy_flag = 1;
1164                vid_hdr->data_size = cpu_to_be32(data_size);
1165                vid_hdr->data_crc = cpu_to_be32(crc);
1166        }
1167        vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1168
1169        err = ubi_io_write_vid_hdr(ubi, to, vid_hdr);
1170        if (err) {
1171                if (err == -EIO)
1172                        err = MOVE_TARGET_WR_ERR;
1173                goto out_unlock_buf;
1174        }
1175
1176        cond_resched();
1177
1178        /* Read the VID header back and check if it was written correctly */
1179        err = ubi_io_read_vid_hdr(ubi, to, vid_hdr, 1);
1180        if (err) {
1181                if (err != UBI_IO_BITFLIPS) {
1182                        ubi_warn(ubi, "error %d while reading VID header back from PEB %d",
1183                                 err, to);
1184                        if (is_error_sane(err))
1185                                err = MOVE_TARGET_RD_ERR;
1186                } else
1187                        err = MOVE_TARGET_BITFLIPS;
1188                goto out_unlock_buf;
1189        }
1190
1191        if (data_size > 0) {
1192                err = ubi_io_write_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1193                if (err) {
1194                        if (err == -EIO)
1195                                err = MOVE_TARGET_WR_ERR;
1196                        goto out_unlock_buf;
1197                }
1198
1199                cond_resched();
1200
1201                /*
1202                 * We've written the data and are going to read it back to make
1203                 * sure it was written correctly.
1204                 */
1205                memset(ubi->peb_buf, 0xFF, aldata_size);
1206                err = ubi_io_read_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1207                if (err) {
1208                        if (err != UBI_IO_BITFLIPS) {
1209                                ubi_warn(ubi, "error %d while reading data back from PEB %d",
1210                                         err, to);
1211                                if (is_error_sane(err))
1212                                        err = MOVE_TARGET_RD_ERR;
1213                        } else
1214                                err = MOVE_TARGET_BITFLIPS;
1215                        goto out_unlock_buf;
1216                }
1217
1218                cond_resched();
1219
1220                if (crc != crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size)) {
1221                        ubi_warn(ubi, "read data back from PEB %d and it is different",
1222                                 to);
1223                        err = -EINVAL;
1224                        goto out_unlock_buf;
1225                }
1226        }
1227
1228        ubi_assert(vol->eba_tbl[lnum] == from);
1229        down_read(&ubi->fm_eba_sem);
1230        vol->eba_tbl[lnum] = to;
1231        up_read(&ubi->fm_eba_sem);
1232
1233out_unlock_buf:
1234        mutex_unlock(&ubi->buf_mutex);
1235out_unlock_leb:
1236        leb_write_unlock(ubi, vol_id, lnum);
1237        return err;
1238}
1239
1240/**
1241 * print_rsvd_warning - warn about not having enough reserved PEBs.
1242 * @ubi: UBI device description object
1243 *
1244 * This is a helper function for 'ubi_eba_init()' which is called when UBI
1245 * cannot reserve enough PEBs for bad block handling. This function makes a
1246 * decision whether we have to print a warning or not. The algorithm is as
1247 * follows:
1248 *   o if this is a new UBI image, then just print the warning
1249 *   o if this is an UBI image which has already been used for some time, print
1250 *     a warning only if we can reserve less than 10% of the expected amount of
1251 *     the reserved PEB.
1252 *
1253 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1254 * of PEBs becomes smaller, which is normal and we do not want to scare users
1255 * with a warning every time they attach the MTD device. This was an issue
1256 * reported by real users.
1257 */
1258static void print_rsvd_warning(struct ubi_device *ubi,
1259                               struct ubi_attach_info *ai)
1260{
1261        /*
1262         * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1263         * large number to distinguish between newly flashed and used images.
1264         */
1265        if (ai->max_sqnum > (1 << 18)) {
1266                int min = ubi->beb_rsvd_level / 10;
1267
1268                if (!min)
1269                        min = 1;
1270                if (ubi->beb_rsvd_pebs > min)
1271                        return;
1272        }
1273
1274        ubi_warn(ubi, "cannot reserve enough PEBs for bad PEB handling, reserved %d, need %d",
1275                 ubi->beb_rsvd_pebs, ubi->beb_rsvd_level);
1276        if (ubi->corr_peb_count)
1277                ubi_warn(ubi, "%d PEBs are corrupted and not used",
1278                         ubi->corr_peb_count);
1279}
1280
1281/**
1282 * self_check_eba - run a self check on the EBA table constructed by fastmap.
1283 * @ubi: UBI device description object
1284 * @ai_fastmap: UBI attach info object created by fastmap
1285 * @ai_scan: UBI attach info object created by scanning
1286 *
1287 * Returns < 0 in case of an internal error, 0 otherwise.
1288 * If a bad EBA table entry was found it will be printed out and
1289 * ubi_assert() triggers.
1290 */
1291int self_check_eba(struct ubi_device *ubi, struct ubi_attach_info *ai_fastmap,
1292                   struct ubi_attach_info *ai_scan)
1293{
1294        int i, j, num_volumes, ret = 0;
1295        int **scan_eba, **fm_eba;
1296        struct ubi_ainf_volume *av;
1297        struct ubi_volume *vol;
1298        struct ubi_ainf_peb *aeb;
1299        struct rb_node *rb;
1300
1301        num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1302
1303        scan_eba = kmalloc(sizeof(*scan_eba) * num_volumes, GFP_KERNEL);
1304        if (!scan_eba)
1305                return -ENOMEM;
1306
1307        fm_eba = kmalloc(sizeof(*fm_eba) * num_volumes, GFP_KERNEL);
1308        if (!fm_eba) {
1309                kfree(scan_eba);
1310                return -ENOMEM;
1311        }
1312
1313        for (i = 0; i < num_volumes; i++) {
1314                vol = ubi->volumes[i];
1315                if (!vol)
1316                        continue;
1317
1318                scan_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**scan_eba),
1319                                      GFP_KERNEL);
1320                if (!scan_eba[i]) {
1321                        ret = -ENOMEM;
1322                        goto out_free;
1323                }
1324
1325                fm_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**fm_eba),
1326                                    GFP_KERNEL);
1327                if (!fm_eba[i]) {
1328                        ret = -ENOMEM;
1329                        goto out_free;
1330                }
1331
1332                for (j = 0; j < vol->reserved_pebs; j++)
1333                        scan_eba[i][j] = fm_eba[i][j] = UBI_LEB_UNMAPPED;
1334
1335                av = ubi_find_av(ai_scan, idx2vol_id(ubi, i));
1336                if (!av)
1337                        continue;
1338
1339                ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1340                        scan_eba[i][aeb->lnum] = aeb->pnum;
1341
1342                av = ubi_find_av(ai_fastmap, idx2vol_id(ubi, i));
1343                if (!av)
1344                        continue;
1345
1346                ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1347                        fm_eba[i][aeb->lnum] = aeb->pnum;
1348
1349                for (j = 0; j < vol->reserved_pebs; j++) {
1350                        if (scan_eba[i][j] != fm_eba[i][j]) {
1351                                if (scan_eba[i][j] == UBI_LEB_UNMAPPED ||
1352                                        fm_eba[i][j] == UBI_LEB_UNMAPPED)
1353                                        continue;
1354
1355                                ubi_err(ubi, "LEB:%i:%i is PEB:%i instead of %i!",
1356                                        vol->vol_id, i, fm_eba[i][j],
1357                                        scan_eba[i][j]);
1358                                ubi_assert(0);
1359                        }
1360                }
1361        }
1362
1363out_free:
1364        for (i = 0; i < num_volumes; i++) {
1365                if (!ubi->volumes[i])
1366                        continue;
1367
1368                kfree(scan_eba[i]);
1369                kfree(fm_eba[i]);
1370        }
1371
1372        kfree(scan_eba);
1373        kfree(fm_eba);
1374        return ret;
1375}
1376
1377/**
1378 * ubi_eba_init - initialize the EBA sub-system using attaching information.
1379 * @ubi: UBI device description object
1380 * @ai: attaching information
1381 *
1382 * This function returns zero in case of success and a negative error code in
1383 * case of failure.
1384 */
1385int ubi_eba_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1386{
1387        int i, j, err, num_volumes;
1388        struct ubi_ainf_volume *av;
1389        struct ubi_volume *vol;
1390        struct ubi_ainf_peb *aeb;
1391        struct rb_node *rb;
1392
1393        dbg_eba("initialize EBA sub-system");
1394
1395        spin_lock_init(&ubi->ltree_lock);
1396        mutex_init(&ubi->alc_mutex);
1397        ubi->ltree = RB_ROOT;
1398
1399        ubi->global_sqnum = ai->max_sqnum + 1;
1400        num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1401
1402        for (i = 0; i < num_volumes; i++) {
1403                vol = ubi->volumes[i];
1404                if (!vol)
1405                        continue;
1406
1407                cond_resched();
1408
1409                vol->eba_tbl = kmalloc(vol->reserved_pebs * sizeof(int),
1410                                       GFP_KERNEL);
1411                if (!vol->eba_tbl) {
1412                        err = -ENOMEM;
1413                        goto out_free;
1414                }
1415
1416                for (j = 0; j < vol->reserved_pebs; j++)
1417                        vol->eba_tbl[j] = UBI_LEB_UNMAPPED;
1418
1419                av = ubi_find_av(ai, idx2vol_id(ubi, i));
1420                if (!av)
1421                        continue;
1422
1423                ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb) {
1424                        if (aeb->lnum >= vol->reserved_pebs)
1425                                /*
1426                                 * This may happen in case of an unclean reboot
1427                                 * during re-size.
1428                                 */
1429                                ubi_move_aeb_to_list(av, aeb, &ai->erase);
1430                        else
1431                                vol->eba_tbl[aeb->lnum] = aeb->pnum;
1432                }
1433        }
1434
1435        if (ubi->avail_pebs < EBA_RESERVED_PEBS) {
1436                ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1437                        ubi->avail_pebs, EBA_RESERVED_PEBS);
1438                if (ubi->corr_peb_count)
1439                        ubi_err(ubi, "%d PEBs are corrupted and not used",
1440                                ubi->corr_peb_count);
1441                err = -ENOSPC;
1442                goto out_free;
1443        }
1444        ubi->avail_pebs -= EBA_RESERVED_PEBS;
1445        ubi->rsvd_pebs += EBA_RESERVED_PEBS;
1446
1447        if (ubi->bad_allowed) {
1448                ubi_calculate_reserved(ubi);
1449
1450                if (ubi->avail_pebs < ubi->beb_rsvd_level) {
1451                        /* No enough free physical eraseblocks */
1452                        ubi->beb_rsvd_pebs = ubi->avail_pebs;
1453                        print_rsvd_warning(ubi, ai);
1454                } else
1455                        ubi->beb_rsvd_pebs = ubi->beb_rsvd_level;
1456
1457                ubi->avail_pebs -= ubi->beb_rsvd_pebs;
1458                ubi->rsvd_pebs  += ubi->beb_rsvd_pebs;
1459        }
1460
1461        dbg_eba("EBA sub-system is initialized");
1462        return 0;
1463
1464out_free:
1465        for (i = 0; i < num_volumes; i++) {
1466                if (!ubi->volumes[i])
1467                        continue;
1468                kfree(ubi->volumes[i]->eba_tbl);
1469                ubi->volumes[i]->eba_tbl = NULL;
1470        }
1471        return err;
1472}
1473