Linux Audio

Check our new training course

In-person Linux kernel drivers training

Jun 16-20, 2025
Register
Loading...
   1/*
   2 * Copyright (c) International Business Machines Corp., 2006
   3 *
   4 * This program is free software; you can redistribute it and/or modify
   5 * it under the terms of the GNU General Public License as published by
   6 * the Free Software Foundation; either version 2 of the License, or
   7 * (at your option) any later version.
   8 *
   9 * This program is distributed in the hope that it will be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
  12 * the GNU General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public License
  15 * along with this program; if not, write to the Free Software
  16 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
  17 *
  18 * Author: Artem Bityutskiy (Битюцкий Артём)
  19 */
  20
  21/*
  22 * The UBI Eraseblock Association (EBA) sub-system.
  23 *
  24 * This sub-system is responsible for I/O to/from logical eraseblock.
  25 *
  26 * Although in this implementation the EBA table is fully kept and managed in
  27 * RAM, which assumes poor scalability, it might be (partially) maintained on
  28 * flash in future implementations.
  29 *
  30 * The EBA sub-system implements per-logical eraseblock locking. Before
  31 * accessing a logical eraseblock it is locked for reading or writing. The
  32 * per-logical eraseblock locking is implemented by means of the lock tree. The
  33 * lock tree is an RB-tree which refers all the currently locked logical
  34 * eraseblocks. The lock tree elements are &struct ubi_ltree_entry objects.
  35 * They are indexed by (@vol_id, @lnum) pairs.
  36 *
  37 * EBA also maintains the global sequence counter which is incremented each
  38 * time a logical eraseblock is mapped to a physical eraseblock and it is
  39 * stored in the volume identifier header. This means that each VID header has
  40 * a unique sequence number. The sequence number is only increased an we assume
  41 * 64 bits is enough to never overflow.
  42 */
  43
  44#include <linux/slab.h>
  45#include <linux/crc32.h>
  46#include <linux/err.h>
  47#include "ubi.h"
  48
  49/* Number of physical eraseblocks reserved for atomic LEB change operation */
  50#define EBA_RESERVED_PEBS 1
  51
  52/**
  53 * next_sqnum - get next sequence number.
  54 * @ubi: UBI device description object
  55 *
  56 * This function returns next sequence number to use, which is just the current
  57 * global sequence counter value. It also increases the global sequence
  58 * counter.
  59 */
  60unsigned long long ubi_next_sqnum(struct ubi_device *ubi)
  61{
  62	unsigned long long sqnum;
  63
  64	spin_lock(&ubi->ltree_lock);
  65	sqnum = ubi->global_sqnum++;
  66	spin_unlock(&ubi->ltree_lock);
  67
  68	return sqnum;
  69}
  70
  71/**
  72 * ubi_get_compat - get compatibility flags of a volume.
  73 * @ubi: UBI device description object
  74 * @vol_id: volume ID
  75 *
  76 * This function returns compatibility flags for an internal volume. User
  77 * volumes have no compatibility flags, so %0 is returned.
  78 */
  79static int ubi_get_compat(const struct ubi_device *ubi, int vol_id)
  80{
  81	if (vol_id == UBI_LAYOUT_VOLUME_ID)
  82		return UBI_LAYOUT_VOLUME_COMPAT;
  83	return 0;
  84}
  85
  86/**
  87 * ltree_lookup - look up the lock tree.
  88 * @ubi: UBI device description object
  89 * @vol_id: volume ID
  90 * @lnum: logical eraseblock number
  91 *
  92 * This function returns a pointer to the corresponding &struct ubi_ltree_entry
  93 * object if the logical eraseblock is locked and %NULL if it is not.
  94 * @ubi->ltree_lock has to be locked.
  95 */
  96static struct ubi_ltree_entry *ltree_lookup(struct ubi_device *ubi, int vol_id,
  97					    int lnum)
  98{
  99	struct rb_node *p;
 100
 101	p = ubi->ltree.rb_node;
 102	while (p) {
 103		struct ubi_ltree_entry *le;
 104
 105		le = rb_entry(p, struct ubi_ltree_entry, rb);
 106
 107		if (vol_id < le->vol_id)
 108			p = p->rb_left;
 109		else if (vol_id > le->vol_id)
 110			p = p->rb_right;
 111		else {
 112			if (lnum < le->lnum)
 113				p = p->rb_left;
 114			else if (lnum > le->lnum)
 115				p = p->rb_right;
 116			else
 117				return le;
 118		}
 119	}
 120
 121	return NULL;
 122}
 123
 124/**
 125 * ltree_add_entry - add new entry to the lock tree.
 126 * @ubi: UBI device description object
 127 * @vol_id: volume ID
 128 * @lnum: logical eraseblock number
 129 *
 130 * This function adds new entry for logical eraseblock (@vol_id, @lnum) to the
 131 * lock tree. If such entry is already there, its usage counter is increased.
 132 * Returns pointer to the lock tree entry or %-ENOMEM if memory allocation
 133 * failed.
 134 */
 135static struct ubi_ltree_entry *ltree_add_entry(struct ubi_device *ubi,
 136					       int vol_id, int lnum)
 137{
 138	struct ubi_ltree_entry *le, *le1, *le_free;
 139
 140	le = kmalloc(sizeof(struct ubi_ltree_entry), GFP_NOFS);
 141	if (!le)
 142		return ERR_PTR(-ENOMEM);
 143
 144	le->users = 0;
 145	init_rwsem(&le->mutex);
 146	le->vol_id = vol_id;
 147	le->lnum = lnum;
 148
 149	spin_lock(&ubi->ltree_lock);
 150	le1 = ltree_lookup(ubi, vol_id, lnum);
 151
 152	if (le1) {
 153		/*
 154		 * This logical eraseblock is already locked. The newly
 155		 * allocated lock entry is not needed.
 156		 */
 157		le_free = le;
 158		le = le1;
 159	} else {
 160		struct rb_node **p, *parent = NULL;
 161
 162		/*
 163		 * No lock entry, add the newly allocated one to the
 164		 * @ubi->ltree RB-tree.
 165		 */
 166		le_free = NULL;
 167
 168		p = &ubi->ltree.rb_node;
 169		while (*p) {
 170			parent = *p;
 171			le1 = rb_entry(parent, struct ubi_ltree_entry, rb);
 172
 173			if (vol_id < le1->vol_id)
 174				p = &(*p)->rb_left;
 175			else if (vol_id > le1->vol_id)
 176				p = &(*p)->rb_right;
 177			else {
 178				ubi_assert(lnum != le1->lnum);
 179				if (lnum < le1->lnum)
 180					p = &(*p)->rb_left;
 181				else
 182					p = &(*p)->rb_right;
 183			}
 184		}
 185
 186		rb_link_node(&le->rb, parent, p);
 187		rb_insert_color(&le->rb, &ubi->ltree);
 188	}
 189	le->users += 1;
 190	spin_unlock(&ubi->ltree_lock);
 191
 192	kfree(le_free);
 193	return le;
 194}
 195
 196/**
 197 * leb_read_lock - lock logical eraseblock for reading.
 198 * @ubi: UBI device description object
 199 * @vol_id: volume ID
 200 * @lnum: logical eraseblock number
 201 *
 202 * This function locks a logical eraseblock for reading. Returns zero in case
 203 * of success and a negative error code in case of failure.
 204 */
 205static int leb_read_lock(struct ubi_device *ubi, int vol_id, int lnum)
 206{
 207	struct ubi_ltree_entry *le;
 208
 209	le = ltree_add_entry(ubi, vol_id, lnum);
 210	if (IS_ERR(le))
 211		return PTR_ERR(le);
 212	down_read(&le->mutex);
 213	return 0;
 214}
 215
 216/**
 217 * leb_read_unlock - unlock logical eraseblock.
 218 * @ubi: UBI device description object
 219 * @vol_id: volume ID
 220 * @lnum: logical eraseblock number
 221 */
 222static void leb_read_unlock(struct ubi_device *ubi, int vol_id, int lnum)
 223{
 224	struct ubi_ltree_entry *le;
 225
 226	spin_lock(&ubi->ltree_lock);
 227	le = ltree_lookup(ubi, vol_id, lnum);
 228	le->users -= 1;
 229	ubi_assert(le->users >= 0);
 230	up_read(&le->mutex);
 231	if (le->users == 0) {
 232		rb_erase(&le->rb, &ubi->ltree);
 233		kfree(le);
 234	}
 235	spin_unlock(&ubi->ltree_lock);
 236}
 237
 238/**
 239 * leb_write_lock - lock logical eraseblock for writing.
 240 * @ubi: UBI device description object
 241 * @vol_id: volume ID
 242 * @lnum: logical eraseblock number
 243 *
 244 * This function locks a logical eraseblock for writing. Returns zero in case
 245 * of success and a negative error code in case of failure.
 246 */
 247static int leb_write_lock(struct ubi_device *ubi, int vol_id, int lnum)
 248{
 249	struct ubi_ltree_entry *le;
 250
 251	le = ltree_add_entry(ubi, vol_id, lnum);
 252	if (IS_ERR(le))
 253		return PTR_ERR(le);
 254	down_write(&le->mutex);
 255	return 0;
 256}
 257
 258/**
 259 * leb_write_lock - lock logical eraseblock for writing.
 260 * @ubi: UBI device description object
 261 * @vol_id: volume ID
 262 * @lnum: logical eraseblock number
 263 *
 264 * This function locks a logical eraseblock for writing if there is no
 265 * contention and does nothing if there is contention. Returns %0 in case of
 266 * success, %1 in case of contention, and and a negative error code in case of
 267 * failure.
 268 */
 269static int leb_write_trylock(struct ubi_device *ubi, int vol_id, int lnum)
 270{
 271	struct ubi_ltree_entry *le;
 272
 273	le = ltree_add_entry(ubi, vol_id, lnum);
 274	if (IS_ERR(le))
 275		return PTR_ERR(le);
 276	if (down_write_trylock(&le->mutex))
 277		return 0;
 278
 279	/* Contention, cancel */
 280	spin_lock(&ubi->ltree_lock);
 281	le->users -= 1;
 282	ubi_assert(le->users >= 0);
 283	if (le->users == 0) {
 284		rb_erase(&le->rb, &ubi->ltree);
 285		kfree(le);
 286	}
 287	spin_unlock(&ubi->ltree_lock);
 288
 289	return 1;
 290}
 291
 292/**
 293 * leb_write_unlock - unlock logical eraseblock.
 294 * @ubi: UBI device description object
 295 * @vol_id: volume ID
 296 * @lnum: logical eraseblock number
 297 */
 298static void leb_write_unlock(struct ubi_device *ubi, int vol_id, int lnum)
 299{
 300	struct ubi_ltree_entry *le;
 301
 302	spin_lock(&ubi->ltree_lock);
 303	le = ltree_lookup(ubi, vol_id, lnum);
 304	le->users -= 1;
 305	ubi_assert(le->users >= 0);
 306	up_write(&le->mutex);
 307	if (le->users == 0) {
 308		rb_erase(&le->rb, &ubi->ltree);
 309		kfree(le);
 310	}
 311	spin_unlock(&ubi->ltree_lock);
 312}
 313
 314/**
 315 * ubi_eba_unmap_leb - un-map logical eraseblock.
 316 * @ubi: UBI device description object
 317 * @vol: volume description object
 318 * @lnum: logical eraseblock number
 319 *
 320 * This function un-maps logical eraseblock @lnum and schedules corresponding
 321 * physical eraseblock for erasure. Returns zero in case of success and a
 322 * negative error code in case of failure.
 323 */
 324int ubi_eba_unmap_leb(struct ubi_device *ubi, struct ubi_volume *vol,
 325		      int lnum)
 326{
 327	int err, pnum, vol_id = vol->vol_id;
 328
 329	if (ubi->ro_mode)
 330		return -EROFS;
 331
 332	err = leb_write_lock(ubi, vol_id, lnum);
 333	if (err)
 334		return err;
 335
 336	pnum = vol->eba_tbl[lnum];
 337	if (pnum < 0)
 338		/* This logical eraseblock is already unmapped */
 339		goto out_unlock;
 340
 341	dbg_eba("erase LEB %d:%d, PEB %d", vol_id, lnum, pnum);
 342
 343	down_read(&ubi->fm_sem);
 344	vol->eba_tbl[lnum] = UBI_LEB_UNMAPPED;
 345	up_read(&ubi->fm_sem);
 346	err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 0);
 347
 348out_unlock:
 349	leb_write_unlock(ubi, vol_id, lnum);
 350	return err;
 351}
 352
 353/**
 354 * ubi_eba_read_leb - read data.
 355 * @ubi: UBI device description object
 356 * @vol: volume description object
 357 * @lnum: logical eraseblock number
 358 * @buf: buffer to store the read data
 359 * @offset: offset from where to read
 360 * @len: how many bytes to read
 361 * @check: data CRC check flag
 362 *
 363 * If the logical eraseblock @lnum is unmapped, @buf is filled with 0xFF
 364 * bytes. The @check flag only makes sense for static volumes and forces
 365 * eraseblock data CRC checking.
 366 *
 367 * In case of success this function returns zero. In case of a static volume,
 368 * if data CRC mismatches - %-EBADMSG is returned. %-EBADMSG may also be
 369 * returned for any volume type if an ECC error was detected by the MTD device
 370 * driver. Other negative error cored may be returned in case of other errors.
 371 */
 372int ubi_eba_read_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
 373		     void *buf, int offset, int len, int check)
 374{
 375	int err, pnum, scrub = 0, vol_id = vol->vol_id;
 376	struct ubi_vid_hdr *vid_hdr;
 377	uint32_t uninitialized_var(crc);
 378
 379	err = leb_read_lock(ubi, vol_id, lnum);
 380	if (err)
 381		return err;
 382
 383	pnum = vol->eba_tbl[lnum];
 384	if (pnum < 0) {
 385		/*
 386		 * The logical eraseblock is not mapped, fill the whole buffer
 387		 * with 0xFF bytes. The exception is static volumes for which
 388		 * it is an error to read unmapped logical eraseblocks.
 389		 */
 390		dbg_eba("read %d bytes from offset %d of LEB %d:%d (unmapped)",
 391			len, offset, vol_id, lnum);
 392		leb_read_unlock(ubi, vol_id, lnum);
 393		ubi_assert(vol->vol_type != UBI_STATIC_VOLUME);
 394		memset(buf, 0xFF, len);
 395		return 0;
 396	}
 397
 398	dbg_eba("read %d bytes from offset %d of LEB %d:%d, PEB %d",
 399		len, offset, vol_id, lnum, pnum);
 400
 401	if (vol->vol_type == UBI_DYNAMIC_VOLUME)
 402		check = 0;
 403
 404retry:
 405	if (check) {
 406		vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
 407		if (!vid_hdr) {
 408			err = -ENOMEM;
 409			goto out_unlock;
 410		}
 411
 412		err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
 413		if (err && err != UBI_IO_BITFLIPS) {
 414			if (err > 0) {
 415				/*
 416				 * The header is either absent or corrupted.
 417				 * The former case means there is a bug -
 418				 * switch to read-only mode just in case.
 419				 * The latter case means a real corruption - we
 420				 * may try to recover data. FIXME: but this is
 421				 * not implemented.
 422				 */
 423				if (err == UBI_IO_BAD_HDR_EBADMSG ||
 424				    err == UBI_IO_BAD_HDR) {
 425					ubi_warn("corrupted VID header at PEB %d, LEB %d:%d",
 426						 pnum, vol_id, lnum);
 427					err = -EBADMSG;
 428				} else
 429					ubi_ro_mode(ubi);
 430			}
 431			goto out_free;
 432		} else if (err == UBI_IO_BITFLIPS)
 433			scrub = 1;
 434
 435		ubi_assert(lnum < be32_to_cpu(vid_hdr->used_ebs));
 436		ubi_assert(len == be32_to_cpu(vid_hdr->data_size));
 437
 438		crc = be32_to_cpu(vid_hdr->data_crc);
 439		ubi_free_vid_hdr(ubi, vid_hdr);
 440	}
 441
 442	err = ubi_io_read_data(ubi, buf, pnum, offset, len);
 443	if (err) {
 444		if (err == UBI_IO_BITFLIPS) {
 445			scrub = 1;
 446			err = 0;
 447		} else if (mtd_is_eccerr(err)) {
 448			if (vol->vol_type == UBI_DYNAMIC_VOLUME)
 449				goto out_unlock;
 450			scrub = 1;
 451			if (!check) {
 452				ubi_msg("force data checking");
 453				check = 1;
 454				goto retry;
 455			}
 456		} else
 457			goto out_unlock;
 458	}
 459
 460	if (check) {
 461		uint32_t crc1 = crc32(UBI_CRC32_INIT, buf, len);
 462		if (crc1 != crc) {
 463			ubi_warn("CRC error: calculated %#08x, must be %#08x",
 464				 crc1, crc);
 465			err = -EBADMSG;
 466			goto out_unlock;
 467		}
 468	}
 469
 470	if (scrub)
 471		err = ubi_wl_scrub_peb(ubi, pnum);
 472
 473	leb_read_unlock(ubi, vol_id, lnum);
 474	return err;
 475
 476out_free:
 477	ubi_free_vid_hdr(ubi, vid_hdr);
 478out_unlock:
 479	leb_read_unlock(ubi, vol_id, lnum);
 480	return err;
 481}
 482
 483/**
 484 * recover_peb - recover from write failure.
 485 * @ubi: UBI device description object
 486 * @pnum: the physical eraseblock to recover
 487 * @vol_id: volume ID
 488 * @lnum: logical eraseblock number
 489 * @buf: data which was not written because of the write failure
 490 * @offset: offset of the failed write
 491 * @len: how many bytes should have been written
 492 *
 493 * This function is called in case of a write failure and moves all good data
 494 * from the potentially bad physical eraseblock to a good physical eraseblock.
 495 * This function also writes the data which was not written due to the failure.
 496 * Returns new physical eraseblock number in case of success, and a negative
 497 * error code in case of failure.
 498 */
 499static int recover_peb(struct ubi_device *ubi, int pnum, int vol_id, int lnum,
 500		       const void *buf, int offset, int len)
 501{
 502	int err, idx = vol_id2idx(ubi, vol_id), new_pnum, data_size, tries = 0;
 503	struct ubi_volume *vol = ubi->volumes[idx];
 504	struct ubi_vid_hdr *vid_hdr;
 505
 506	vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
 507	if (!vid_hdr)
 508		return -ENOMEM;
 509
 510retry:
 511	new_pnum = ubi_wl_get_peb(ubi);
 512	if (new_pnum < 0) {
 513		ubi_free_vid_hdr(ubi, vid_hdr);
 514		return new_pnum;
 515	}
 516
 517	ubi_msg("recover PEB %d, move data to PEB %d", pnum, new_pnum);
 518
 519	err = ubi_io_read_vid_hdr(ubi, pnum, vid_hdr, 1);
 520	if (err && err != UBI_IO_BITFLIPS) {
 521		if (err > 0)
 522			err = -EIO;
 523		goto out_put;
 524	}
 525
 526	vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
 527	err = ubi_io_write_vid_hdr(ubi, new_pnum, vid_hdr);
 528	if (err)
 529		goto write_error;
 530
 531	data_size = offset + len;
 532	mutex_lock(&ubi->buf_mutex);
 533	memset(ubi->peb_buf + offset, 0xFF, len);
 534
 535	/* Read everything before the area where the write failure happened */
 536	if (offset > 0) {
 537		err = ubi_io_read_data(ubi, ubi->peb_buf, pnum, 0, offset);
 538		if (err && err != UBI_IO_BITFLIPS)
 539			goto out_unlock;
 540	}
 541
 542	memcpy(ubi->peb_buf + offset, buf, len);
 543
 544	err = ubi_io_write_data(ubi, ubi->peb_buf, new_pnum, 0, data_size);
 545	if (err) {
 546		mutex_unlock(&ubi->buf_mutex);
 547		goto write_error;
 548	}
 549
 550	mutex_unlock(&ubi->buf_mutex);
 551	ubi_free_vid_hdr(ubi, vid_hdr);
 552
 553	down_read(&ubi->fm_sem);
 554	vol->eba_tbl[lnum] = new_pnum;
 555	up_read(&ubi->fm_sem);
 556	ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
 557
 558	ubi_msg("data was successfully recovered");
 559	return 0;
 560
 561out_unlock:
 562	mutex_unlock(&ubi->buf_mutex);
 563out_put:
 564	ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
 565	ubi_free_vid_hdr(ubi, vid_hdr);
 566	return err;
 567
 568write_error:
 569	/*
 570	 * Bad luck? This physical eraseblock is bad too? Crud. Let's try to
 571	 * get another one.
 572	 */
 573	ubi_warn("failed to write to PEB %d", new_pnum);
 574	ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
 575	if (++tries > UBI_IO_RETRIES) {
 576		ubi_free_vid_hdr(ubi, vid_hdr);
 577		return err;
 578	}
 579	ubi_msg("try again");
 580	goto retry;
 581}
 582
 583/**
 584 * ubi_eba_write_leb - write data to dynamic volume.
 585 * @ubi: UBI device description object
 586 * @vol: volume description object
 587 * @lnum: logical eraseblock number
 588 * @buf: the data to write
 589 * @offset: offset within the logical eraseblock where to write
 590 * @len: how many bytes to write
 591 *
 592 * This function writes data to logical eraseblock @lnum of a dynamic volume
 593 * @vol. Returns zero in case of success and a negative error code in case
 594 * of failure. In case of error, it is possible that something was still
 595 * written to the flash media, but may be some garbage.
 596 */
 597int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
 598		      const void *buf, int offset, int len)
 599{
 600	int err, pnum, tries = 0, vol_id = vol->vol_id;
 601	struct ubi_vid_hdr *vid_hdr;
 602
 603	if (ubi->ro_mode)
 604		return -EROFS;
 605
 606	err = leb_write_lock(ubi, vol_id, lnum);
 607	if (err)
 608		return err;
 609
 610	pnum = vol->eba_tbl[lnum];
 611	if (pnum >= 0) {
 612		dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
 613			len, offset, vol_id, lnum, pnum);
 614
 615		err = ubi_io_write_data(ubi, buf, pnum, offset, len);
 616		if (err) {
 617			ubi_warn("failed to write data to PEB %d", pnum);
 618			if (err == -EIO && ubi->bad_allowed)
 619				err = recover_peb(ubi, pnum, vol_id, lnum, buf,
 620						  offset, len);
 621			if (err)
 622				ubi_ro_mode(ubi);
 623		}
 624		leb_write_unlock(ubi, vol_id, lnum);
 625		return err;
 626	}
 627
 628	/*
 629	 * The logical eraseblock is not mapped. We have to get a free physical
 630	 * eraseblock and write the volume identifier header there first.
 631	 */
 632	vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
 633	if (!vid_hdr) {
 634		leb_write_unlock(ubi, vol_id, lnum);
 635		return -ENOMEM;
 636	}
 637
 638	vid_hdr->vol_type = UBI_VID_DYNAMIC;
 639	vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
 640	vid_hdr->vol_id = cpu_to_be32(vol_id);
 641	vid_hdr->lnum = cpu_to_be32(lnum);
 642	vid_hdr->compat = ubi_get_compat(ubi, vol_id);
 643	vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
 644
 645retry:
 646	pnum = ubi_wl_get_peb(ubi);
 647	if (pnum < 0) {
 648		ubi_free_vid_hdr(ubi, vid_hdr);
 649		leb_write_unlock(ubi, vol_id, lnum);
 650		return pnum;
 651	}
 652
 653	dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
 654		len, offset, vol_id, lnum, pnum);
 655
 656	err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
 657	if (err) {
 658		ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
 659			 vol_id, lnum, pnum);
 660		goto write_error;
 661	}
 662
 663	if (len) {
 664		err = ubi_io_write_data(ubi, buf, pnum, offset, len);
 665		if (err) {
 666			ubi_warn("failed to write %d bytes at offset %d of LEB %d:%d, PEB %d",
 667				 len, offset, vol_id, lnum, pnum);
 668			goto write_error;
 669		}
 670	}
 671
 672	down_read(&ubi->fm_sem);
 673	vol->eba_tbl[lnum] = pnum;
 674	up_read(&ubi->fm_sem);
 675
 676	leb_write_unlock(ubi, vol_id, lnum);
 677	ubi_free_vid_hdr(ubi, vid_hdr);
 678	return 0;
 679
 680write_error:
 681	if (err != -EIO || !ubi->bad_allowed) {
 682		ubi_ro_mode(ubi);
 683		leb_write_unlock(ubi, vol_id, lnum);
 684		ubi_free_vid_hdr(ubi, vid_hdr);
 685		return err;
 686	}
 687
 688	/*
 689	 * Fortunately, this is the first write operation to this physical
 690	 * eraseblock, so just put it and request a new one. We assume that if
 691	 * this physical eraseblock went bad, the erase code will handle that.
 692	 */
 693	err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
 694	if (err || ++tries > UBI_IO_RETRIES) {
 695		ubi_ro_mode(ubi);
 696		leb_write_unlock(ubi, vol_id, lnum);
 697		ubi_free_vid_hdr(ubi, vid_hdr);
 698		return err;
 699	}
 700
 701	vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
 702	ubi_msg("try another PEB");
 703	goto retry;
 704}
 705
 706/**
 707 * ubi_eba_write_leb_st - write data to static volume.
 708 * @ubi: UBI device description object
 709 * @vol: volume description object
 710 * @lnum: logical eraseblock number
 711 * @buf: data to write
 712 * @len: how many bytes to write
 713 * @used_ebs: how many logical eraseblocks will this volume contain
 714 *
 715 * This function writes data to logical eraseblock @lnum of static volume
 716 * @vol. The @used_ebs argument should contain total number of logical
 717 * eraseblock in this static volume.
 718 *
 719 * When writing to the last logical eraseblock, the @len argument doesn't have
 720 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
 721 * to the real data size, although the @buf buffer has to contain the
 722 * alignment. In all other cases, @len has to be aligned.
 723 *
 724 * It is prohibited to write more than once to logical eraseblocks of static
 725 * volumes. This function returns zero in case of success and a negative error
 726 * code in case of failure.
 727 */
 728int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
 729			 int lnum, const void *buf, int len, int used_ebs)
 730{
 731	int err, pnum, tries = 0, data_size = len, vol_id = vol->vol_id;
 732	struct ubi_vid_hdr *vid_hdr;
 733	uint32_t crc;
 734
 735	if (ubi->ro_mode)
 736		return -EROFS;
 737
 738	if (lnum == used_ebs - 1)
 739		/* If this is the last LEB @len may be unaligned */
 740		len = ALIGN(data_size, ubi->min_io_size);
 741	else
 742		ubi_assert(!(len & (ubi->min_io_size - 1)));
 743
 744	vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
 745	if (!vid_hdr)
 746		return -ENOMEM;
 747
 748	err = leb_write_lock(ubi, vol_id, lnum);
 749	if (err) {
 750		ubi_free_vid_hdr(ubi, vid_hdr);
 751		return err;
 752	}
 753
 754	vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
 755	vid_hdr->vol_id = cpu_to_be32(vol_id);
 756	vid_hdr->lnum = cpu_to_be32(lnum);
 757	vid_hdr->compat = ubi_get_compat(ubi, vol_id);
 758	vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
 759
 760	crc = crc32(UBI_CRC32_INIT, buf, data_size);
 761	vid_hdr->vol_type = UBI_VID_STATIC;
 762	vid_hdr->data_size = cpu_to_be32(data_size);
 763	vid_hdr->used_ebs = cpu_to_be32(used_ebs);
 764	vid_hdr->data_crc = cpu_to_be32(crc);
 765
 766retry:
 767	pnum = ubi_wl_get_peb(ubi);
 768	if (pnum < 0) {
 769		ubi_free_vid_hdr(ubi, vid_hdr);
 770		leb_write_unlock(ubi, vol_id, lnum);
 771		return pnum;
 772	}
 773
 774	dbg_eba("write VID hdr and %d bytes at LEB %d:%d, PEB %d, used_ebs %d",
 775		len, vol_id, lnum, pnum, used_ebs);
 776
 777	err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
 778	if (err) {
 779		ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
 780			 vol_id, lnum, pnum);
 781		goto write_error;
 782	}
 783
 784	err = ubi_io_write_data(ubi, buf, pnum, 0, len);
 785	if (err) {
 786		ubi_warn("failed to write %d bytes of data to PEB %d",
 787			 len, pnum);
 788		goto write_error;
 789	}
 790
 791	ubi_assert(vol->eba_tbl[lnum] < 0);
 792	down_read(&ubi->fm_sem);
 793	vol->eba_tbl[lnum] = pnum;
 794	up_read(&ubi->fm_sem);
 795
 796	leb_write_unlock(ubi, vol_id, lnum);
 797	ubi_free_vid_hdr(ubi, vid_hdr);
 798	return 0;
 799
 800write_error:
 801	if (err != -EIO || !ubi->bad_allowed) {
 802		/*
 803		 * This flash device does not admit of bad eraseblocks or
 804		 * something nasty and unexpected happened. Switch to read-only
 805		 * mode just in case.
 806		 */
 807		ubi_ro_mode(ubi);
 808		leb_write_unlock(ubi, vol_id, lnum);
 809		ubi_free_vid_hdr(ubi, vid_hdr);
 810		return err;
 811	}
 812
 813	err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
 814	if (err || ++tries > UBI_IO_RETRIES) {
 815		ubi_ro_mode(ubi);
 816		leb_write_unlock(ubi, vol_id, lnum);
 817		ubi_free_vid_hdr(ubi, vid_hdr);
 818		return err;
 819	}
 820
 821	vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
 822	ubi_msg("try another PEB");
 823	goto retry;
 824}
 825
 826/*
 827 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
 828 * @ubi: UBI device description object
 829 * @vol: volume description object
 830 * @lnum: logical eraseblock number
 831 * @buf: data to write
 832 * @len: how many bytes to write
 833 *
 834 * This function changes the contents of a logical eraseblock atomically. @buf
 835 * has to contain new logical eraseblock data, and @len - the length of the
 836 * data, which has to be aligned. This function guarantees that in case of an
 837 * unclean reboot the old contents is preserved. Returns zero in case of
 838 * success and a negative error code in case of failure.
 839 *
 840 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
 841 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
 842 */
 843int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
 844			      int lnum, const void *buf, int len)
 845{
 846	int err, pnum, tries = 0, vol_id = vol->vol_id;
 847	struct ubi_vid_hdr *vid_hdr;
 848	uint32_t crc;
 849
 850	if (ubi->ro_mode)
 851		return -EROFS;
 852
 853	if (len == 0) {
 854		/*
 855		 * Special case when data length is zero. In this case the LEB
 856		 * has to be unmapped and mapped somewhere else.
 857		 */
 858		err = ubi_eba_unmap_leb(ubi, vol, lnum);
 859		if (err)
 860			return err;
 861		return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0);
 862	}
 863
 864	vid_hdr = ubi_zalloc_vid_hdr(ubi, GFP_NOFS);
 865	if (!vid_hdr)
 866		return -ENOMEM;
 867
 868	mutex_lock(&ubi->alc_mutex);
 869	err = leb_write_lock(ubi, vol_id, lnum);
 870	if (err)
 871		goto out_mutex;
 872
 873	vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
 874	vid_hdr->vol_id = cpu_to_be32(vol_id);
 875	vid_hdr->lnum = cpu_to_be32(lnum);
 876	vid_hdr->compat = ubi_get_compat(ubi, vol_id);
 877	vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
 878
 879	crc = crc32(UBI_CRC32_INIT, buf, len);
 880	vid_hdr->vol_type = UBI_VID_DYNAMIC;
 881	vid_hdr->data_size = cpu_to_be32(len);
 882	vid_hdr->copy_flag = 1;
 883	vid_hdr->data_crc = cpu_to_be32(crc);
 884
 885retry:
 886	pnum = ubi_wl_get_peb(ubi);
 887	if (pnum < 0) {
 888		err = pnum;
 889		goto out_leb_unlock;
 890	}
 891
 892	dbg_eba("change LEB %d:%d, PEB %d, write VID hdr to PEB %d",
 893		vol_id, lnum, vol->eba_tbl[lnum], pnum);
 894
 895	err = ubi_io_write_vid_hdr(ubi, pnum, vid_hdr);
 896	if (err) {
 897		ubi_warn("failed to write VID header to LEB %d:%d, PEB %d",
 898			 vol_id, lnum, pnum);
 899		goto write_error;
 900	}
 901
 902	err = ubi_io_write_data(ubi, buf, pnum, 0, len);
 903	if (err) {
 904		ubi_warn("failed to write %d bytes of data to PEB %d",
 905			 len, pnum);
 906		goto write_error;
 907	}
 908
 909	if (vol->eba_tbl[lnum] >= 0) {
 910		err = ubi_wl_put_peb(ubi, vol_id, lnum, vol->eba_tbl[lnum], 0);
 911		if (err)
 912			goto out_leb_unlock;
 913	}
 914
 915	down_read(&ubi->fm_sem);
 916	vol->eba_tbl[lnum] = pnum;
 917	up_read(&ubi->fm_sem);
 918
 919out_leb_unlock:
 920	leb_write_unlock(ubi, vol_id, lnum);
 921out_mutex:
 922	mutex_unlock(&ubi->alc_mutex);
 923	ubi_free_vid_hdr(ubi, vid_hdr);
 924	return err;
 925
 926write_error:
 927	if (err != -EIO || !ubi->bad_allowed) {
 928		/*
 929		 * This flash device does not admit of bad eraseblocks or
 930		 * something nasty and unexpected happened. Switch to read-only
 931		 * mode just in case.
 932		 */
 933		ubi_ro_mode(ubi);
 934		goto out_leb_unlock;
 935	}
 936
 937	err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
 938	if (err || ++tries > UBI_IO_RETRIES) {
 939		ubi_ro_mode(ubi);
 940		goto out_leb_unlock;
 941	}
 942
 943	vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
 944	ubi_msg("try another PEB");
 945	goto retry;
 946}
 947
 948/**
 949 * is_error_sane - check whether a read error is sane.
 950 * @err: code of the error happened during reading
 951 *
 952 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
 953 * cannot read data from the target PEB (an error @err happened). If the error
 954 * code is sane, then we treat this error as non-fatal. Otherwise the error is
 955 * fatal and UBI will be switched to R/O mode later.
 956 *
 957 * The idea is that we try not to switch to R/O mode if the read error is
 958 * something which suggests there was a real read problem. E.g., %-EIO. Or a
 959 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
 960 * mode, simply because we do not know what happened at the MTD level, and we
 961 * cannot handle this. E.g., the underlying driver may have become crazy, and
 962 * it is safer to switch to R/O mode to preserve the data.
 963 *
 964 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
 965 * which we have just written.
 966 */
 967static int is_error_sane(int err)
 968{
 969	if (err == -EIO || err == -ENOMEM || err == UBI_IO_BAD_HDR ||
 970	    err == UBI_IO_BAD_HDR_EBADMSG || err == -ETIMEDOUT)
 971		return 0;
 972	return 1;
 973}
 974
 975/**
 976 * ubi_eba_copy_leb - copy logical eraseblock.
 977 * @ubi: UBI device description object
 978 * @from: physical eraseblock number from where to copy
 979 * @to: physical eraseblock number where to copy
 980 * @vid_hdr: VID header of the @from physical eraseblock
 981 *
 982 * This function copies logical eraseblock from physical eraseblock @from to
 983 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
 984 * function. Returns:
 985 *   o %0 in case of success;
 986 *   o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_TARGET_BITFLIPS, etc;
 987 *   o a negative error code in case of failure.
 988 */
 989int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
 990		     struct ubi_vid_hdr *vid_hdr)
 991{
 992	int err, vol_id, lnum, data_size, aldata_size, idx;
 993	struct ubi_volume *vol;
 994	uint32_t crc;
 995
 996	vol_id = be32_to_cpu(vid_hdr->vol_id);
 997	lnum = be32_to_cpu(vid_hdr->lnum);
 998
 999	dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
1000
1001	if (vid_hdr->vol_type == UBI_VID_STATIC) {
1002		data_size = be32_to_cpu(vid_hdr->data_size);
1003		aldata_size = ALIGN(data_size, ubi->min_io_size);
1004	} else
1005		data_size = aldata_size =
1006			    ubi->leb_size - be32_to_cpu(vid_hdr->data_pad);
1007
1008	idx = vol_id2idx(ubi, vol_id);
1009	spin_lock(&ubi->volumes_lock);
1010	/*
1011	 * Note, we may race with volume deletion, which means that the volume
1012	 * this logical eraseblock belongs to might be being deleted. Since the
1013	 * volume deletion un-maps all the volume's logical eraseblocks, it will
1014	 * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1015	 */
1016	vol = ubi->volumes[idx];
1017	spin_unlock(&ubi->volumes_lock);
1018	if (!vol) {
1019		/* No need to do further work, cancel */
1020		dbg_wl("volume %d is being removed, cancel", vol_id);
1021		return MOVE_CANCEL_RACE;
1022	}
1023
1024	/*
1025	 * We do not want anybody to write to this logical eraseblock while we
1026	 * are moving it, so lock it.
1027	 *
1028	 * Note, we are using non-waiting locking here, because we cannot sleep
1029	 * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1030	 * unmapping the LEB which is mapped to the PEB we are going to move
1031	 * (@from). This task locks the LEB and goes sleep in the
1032	 * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1033	 * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1034	 * LEB is already locked, we just do not move it and return
1035	 * %MOVE_RETRY. Note, we do not return %MOVE_CANCEL_RACE here because
1036	 * we do not know the reasons of the contention - it may be just a
1037	 * normal I/O on this LEB, so we want to re-try.
1038	 */
1039	err = leb_write_trylock(ubi, vol_id, lnum);
1040	if (err) {
1041		dbg_wl("contention on LEB %d:%d, cancel", vol_id, lnum);
1042		return MOVE_RETRY;
1043	}
1044
1045	/*
1046	 * The LEB might have been put meanwhile, and the task which put it is
1047	 * probably waiting on @ubi->move_mutex. No need to continue the work,
1048	 * cancel it.
1049	 */
1050	if (vol->eba_tbl[lnum] != from) {
1051		dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to PEB %d, cancel",
1052		       vol_id, lnum, from, vol->eba_tbl[lnum]);
1053		err = MOVE_CANCEL_RACE;
1054		goto out_unlock_leb;
1055	}
1056
1057	/*
1058	 * OK, now the LEB is locked and we can safely start moving it. Since
1059	 * this function utilizes the @ubi->peb_buf buffer which is shared
1060	 * with some other functions - we lock the buffer by taking the
1061	 * @ubi->buf_mutex.
1062	 */
1063	mutex_lock(&ubi->buf_mutex);
1064	dbg_wl("read %d bytes of data", aldata_size);
1065	err = ubi_io_read_data(ubi, ubi->peb_buf, from, 0, aldata_size);
1066	if (err && err != UBI_IO_BITFLIPS) {
1067		ubi_warn("error %d while reading data from PEB %d",
1068			 err, from);
1069		err = MOVE_SOURCE_RD_ERR;
1070		goto out_unlock_buf;
1071	}
1072
1073	/*
1074	 * Now we have got to calculate how much data we have to copy. In
1075	 * case of a static volume it is fairly easy - the VID header contains
1076	 * the data size. In case of a dynamic volume it is more difficult - we
1077	 * have to read the contents, cut 0xFF bytes from the end and copy only
1078	 * the first part. We must do this to avoid writing 0xFF bytes as it
1079	 * may have some side-effects. And not only this. It is important not
1080	 * to include those 0xFFs to CRC because later the they may be filled
1081	 * by data.
1082	 */
1083	if (vid_hdr->vol_type == UBI_VID_DYNAMIC)
1084		aldata_size = data_size =
1085			ubi_calc_data_len(ubi, ubi->peb_buf, data_size);
1086
1087	cond_resched();
1088	crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
1089	cond_resched();
1090
1091	/*
1092	 * It may turn out to be that the whole @from physical eraseblock
1093	 * contains only 0xFF bytes. Then we have to only write the VID header
1094	 * and do not write any data. This also means we should not set
1095	 * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1096	 */
1097	if (data_size > 0) {
1098		vid_hdr->copy_flag = 1;
1099		vid_hdr->data_size = cpu_to_be32(data_size);
1100		vid_hdr->data_crc = cpu_to_be32(crc);
1101	}
1102	vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1103
1104	err = ubi_io_write_vid_hdr(ubi, to, vid_hdr);
1105	if (err) {
1106		if (err == -EIO)
1107			err = MOVE_TARGET_WR_ERR;
1108		goto out_unlock_buf;
1109	}
1110
1111	cond_resched();
1112
1113	/* Read the VID header back and check if it was written correctly */
1114	err = ubi_io_read_vid_hdr(ubi, to, vid_hdr, 1);
1115	if (err) {
1116		if (err != UBI_IO_BITFLIPS) {
1117			ubi_warn("error %d while reading VID header back from PEB %d",
1118				 err, to);
1119			if (is_error_sane(err))
1120				err = MOVE_TARGET_RD_ERR;
1121		} else
1122			err = MOVE_TARGET_BITFLIPS;
1123		goto out_unlock_buf;
1124	}
1125
1126	if (data_size > 0) {
1127		err = ubi_io_write_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1128		if (err) {
1129			if (err == -EIO)
1130				err = MOVE_TARGET_WR_ERR;
1131			goto out_unlock_buf;
1132		}
1133
1134		cond_resched();
1135
1136		/*
1137		 * We've written the data and are going to read it back to make
1138		 * sure it was written correctly.
1139		 */
1140		memset(ubi->peb_buf, 0xFF, aldata_size);
1141		err = ubi_io_read_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1142		if (err) {
1143			if (err != UBI_IO_BITFLIPS) {
1144				ubi_warn("error %d while reading data back from PEB %d",
1145					 err, to);
1146				if (is_error_sane(err))
1147					err = MOVE_TARGET_RD_ERR;
1148			} else
1149				err = MOVE_TARGET_BITFLIPS;
1150			goto out_unlock_buf;
1151		}
1152
1153		cond_resched();
1154
1155		if (crc != crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size)) {
1156			ubi_warn("read data back from PEB %d and it is different",
1157				 to);
1158			err = -EINVAL;
1159			goto out_unlock_buf;
1160		}
1161	}
1162
1163	ubi_assert(vol->eba_tbl[lnum] == from);
1164	down_read(&ubi->fm_sem);
1165	vol->eba_tbl[lnum] = to;
1166	up_read(&ubi->fm_sem);
1167
1168out_unlock_buf:
1169	mutex_unlock(&ubi->buf_mutex);
1170out_unlock_leb:
1171	leb_write_unlock(ubi, vol_id, lnum);
1172	return err;
1173}
1174
1175/**
1176 * print_rsvd_warning - warn about not having enough reserved PEBs.
1177 * @ubi: UBI device description object
1178 *
1179 * This is a helper function for 'ubi_eba_init()' which is called when UBI
1180 * cannot reserve enough PEBs for bad block handling. This function makes a
1181 * decision whether we have to print a warning or not. The algorithm is as
1182 * follows:
1183 *   o if this is a new UBI image, then just print the warning
1184 *   o if this is an UBI image which has already been used for some time, print
1185 *     a warning only if we can reserve less than 10% of the expected amount of
1186 *     the reserved PEB.
1187 *
1188 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1189 * of PEBs becomes smaller, which is normal and we do not want to scare users
1190 * with a warning every time they attach the MTD device. This was an issue
1191 * reported by real users.
1192 */
1193static void print_rsvd_warning(struct ubi_device *ubi,
1194			       struct ubi_attach_info *ai)
1195{
1196	/*
1197	 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1198	 * large number to distinguish between newly flashed and used images.
1199	 */
1200	if (ai->max_sqnum > (1 << 18)) {
1201		int min = ubi->beb_rsvd_level / 10;
1202
1203		if (!min)
1204			min = 1;
1205		if (ubi->beb_rsvd_pebs > min)
1206			return;
1207	}
1208
1209	ubi_warn("cannot reserve enough PEBs for bad PEB handling, reserved %d, need %d",
1210		 ubi->beb_rsvd_pebs, ubi->beb_rsvd_level);
1211	if (ubi->corr_peb_count)
1212		ubi_warn("%d PEBs are corrupted and not used",
1213			 ubi->corr_peb_count);
1214}
1215
1216/**
1217 * self_check_eba - run a self check on the EBA table constructed by fastmap.
1218 * @ubi: UBI device description object
1219 * @ai_fastmap: UBI attach info object created by fastmap
1220 * @ai_scan: UBI attach info object created by scanning
1221 *
1222 * Returns < 0 in case of an internal error, 0 otherwise.
1223 * If a bad EBA table entry was found it will be printed out and
1224 * ubi_assert() triggers.
1225 */
1226int self_check_eba(struct ubi_device *ubi, struct ubi_attach_info *ai_fastmap,
1227		   struct ubi_attach_info *ai_scan)
1228{
1229	int i, j, num_volumes, ret = 0;
1230	int **scan_eba, **fm_eba;
1231	struct ubi_ainf_volume *av;
1232	struct ubi_volume *vol;
1233	struct ubi_ainf_peb *aeb;
1234	struct rb_node *rb;
1235
1236	num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1237
1238	scan_eba = kmalloc(sizeof(*scan_eba) * num_volumes, GFP_KERNEL);
1239	if (!scan_eba)
1240		return -ENOMEM;
1241
1242	fm_eba = kmalloc(sizeof(*fm_eba) * num_volumes, GFP_KERNEL);
1243	if (!fm_eba) {
1244		kfree(scan_eba);
1245		return -ENOMEM;
1246	}
1247
1248	for (i = 0; i < num_volumes; i++) {
1249		vol = ubi->volumes[i];
1250		if (!vol)
1251			continue;
1252
1253		scan_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**scan_eba),
1254				      GFP_KERNEL);
1255		if (!scan_eba[i]) {
1256			ret = -ENOMEM;
1257			goto out_free;
1258		}
1259
1260		fm_eba[i] = kmalloc(vol->reserved_pebs * sizeof(**fm_eba),
1261				    GFP_KERNEL);
1262		if (!fm_eba[i]) {
1263			ret = -ENOMEM;
1264			goto out_free;
1265		}
1266
1267		for (j = 0; j < vol->reserved_pebs; j++)
1268			scan_eba[i][j] = fm_eba[i][j] = UBI_LEB_UNMAPPED;
1269
1270		av = ubi_find_av(ai_scan, idx2vol_id(ubi, i));
1271		if (!av)
1272			continue;
1273
1274		ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1275			scan_eba[i][aeb->lnum] = aeb->pnum;
1276
1277		av = ubi_find_av(ai_fastmap, idx2vol_id(ubi, i));
1278		if (!av)
1279			continue;
1280
1281		ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1282			fm_eba[i][aeb->lnum] = aeb->pnum;
1283
1284		for (j = 0; j < vol->reserved_pebs; j++) {
1285			if (scan_eba[i][j] != fm_eba[i][j]) {
1286				if (scan_eba[i][j] == UBI_LEB_UNMAPPED ||
1287					fm_eba[i][j] == UBI_LEB_UNMAPPED)
1288					continue;
1289
1290				ubi_err("LEB:%i:%i is PEB:%i instead of %i!",
1291					vol->vol_id, i, fm_eba[i][j],
1292					scan_eba[i][j]);
1293				ubi_assert(0);
1294			}
1295		}
1296	}
1297
1298out_free:
1299	for (i = 0; i < num_volumes; i++) {
1300		if (!ubi->volumes[i])
1301			continue;
1302
1303		kfree(scan_eba[i]);
1304		kfree(fm_eba[i]);
1305	}
1306
1307	kfree(scan_eba);
1308	kfree(fm_eba);
1309	return ret;
1310}
1311
1312/**
1313 * ubi_eba_init - initialize the EBA sub-system using attaching information.
1314 * @ubi: UBI device description object
1315 * @ai: attaching information
1316 *
1317 * This function returns zero in case of success and a negative error code in
1318 * case of failure.
1319 */
1320int ubi_eba_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1321{
1322	int i, j, err, num_volumes;
1323	struct ubi_ainf_volume *av;
1324	struct ubi_volume *vol;
1325	struct ubi_ainf_peb *aeb;
1326	struct rb_node *rb;
1327
1328	dbg_eba("initialize EBA sub-system");
1329
1330	spin_lock_init(&ubi->ltree_lock);
1331	mutex_init(&ubi->alc_mutex);
1332	ubi->ltree = RB_ROOT;
1333
1334	ubi->global_sqnum = ai->max_sqnum + 1;
1335	num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1336
1337	for (i = 0; i < num_volumes; i++) {
1338		vol = ubi->volumes[i];
1339		if (!vol)
1340			continue;
1341
1342		cond_resched();
1343
1344		vol->eba_tbl = kmalloc(vol->reserved_pebs * sizeof(int),
1345				       GFP_KERNEL);
1346		if (!vol->eba_tbl) {
1347			err = -ENOMEM;
1348			goto out_free;
1349		}
1350
1351		for (j = 0; j < vol->reserved_pebs; j++)
1352			vol->eba_tbl[j] = UBI_LEB_UNMAPPED;
1353
1354		av = ubi_find_av(ai, idx2vol_id(ubi, i));
1355		if (!av)
1356			continue;
1357
1358		ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb) {
1359			if (aeb->lnum >= vol->reserved_pebs)
1360				/*
1361				 * This may happen in case of an unclean reboot
1362				 * during re-size.
1363				 */
1364				ubi_move_aeb_to_list(av, aeb, &ai->erase);
1365			vol->eba_tbl[aeb->lnum] = aeb->pnum;
1366		}
1367	}
1368
1369	if (ubi->avail_pebs < EBA_RESERVED_PEBS) {
1370		ubi_err("no enough physical eraseblocks (%d, need %d)",
1371			ubi->avail_pebs, EBA_RESERVED_PEBS);
1372		if (ubi->corr_peb_count)
1373			ubi_err("%d PEBs are corrupted and not used",
1374				ubi->corr_peb_count);
1375		err = -ENOSPC;
1376		goto out_free;
1377	}
1378	ubi->avail_pebs -= EBA_RESERVED_PEBS;
1379	ubi->rsvd_pebs += EBA_RESERVED_PEBS;
1380
1381	if (ubi->bad_allowed) {
1382		ubi_calculate_reserved(ubi);
1383
1384		if (ubi->avail_pebs < ubi->beb_rsvd_level) {
1385			/* No enough free physical eraseblocks */
1386			ubi->beb_rsvd_pebs = ubi->avail_pebs;
1387			print_rsvd_warning(ubi, ai);
1388		} else
1389			ubi->beb_rsvd_pebs = ubi->beb_rsvd_level;
1390
1391		ubi->avail_pebs -= ubi->beb_rsvd_pebs;
1392		ubi->rsvd_pebs  += ubi->beb_rsvd_pebs;
1393	}
1394
1395	dbg_eba("EBA sub-system is initialized");
1396	return 0;
1397
1398out_free:
1399	for (i = 0; i < num_volumes; i++) {
1400		if (!ubi->volumes[i])
1401			continue;
1402		kfree(ubi->volumes[i]->eba_tbl);
1403		ubi->volumes[i]->eba_tbl = NULL;
1404	}
1405	return err;
1406}
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/*
   3 * Copyright (c) International Business Machines Corp., 2006
   4 *
   5 * Author: Artem Bityutskiy (Битюцкий Артём)
   6 */
   7
   8/*
   9 * The UBI Eraseblock Association (EBA) sub-system.
  10 *
  11 * This sub-system is responsible for I/O to/from logical eraseblock.
  12 *
  13 * Although in this implementation the EBA table is fully kept and managed in
  14 * RAM, which assumes poor scalability, it might be (partially) maintained on
  15 * flash in future implementations.
  16 *
  17 * The EBA sub-system implements per-logical eraseblock locking. Before
  18 * accessing a logical eraseblock it is locked for reading or writing. The
  19 * per-logical eraseblock locking is implemented by means of the lock tree. The
  20 * lock tree is an RB-tree which refers all the currently locked logical
  21 * eraseblocks. The lock tree elements are &struct ubi_ltree_entry objects.
  22 * They are indexed by (@vol_id, @lnum) pairs.
  23 *
  24 * EBA also maintains the global sequence counter which is incremented each
  25 * time a logical eraseblock is mapped to a physical eraseblock and it is
  26 * stored in the volume identifier header. This means that each VID header has
  27 * a unique sequence number. The sequence number is only increased an we assume
  28 * 64 bits is enough to never overflow.
  29 */
  30
  31#include <linux/slab.h>
  32#include <linux/crc32.h>
  33#include <linux/err.h>
  34#include "ubi.h"
  35
  36/* Number of physical eraseblocks reserved for atomic LEB change operation */
  37#define EBA_RESERVED_PEBS 1
  38
  39/**
  40 * struct ubi_eba_entry - structure encoding a single LEB -> PEB association
  41 * @pnum: the physical eraseblock number attached to the LEB
  42 *
  43 * This structure is encoding a LEB -> PEB association. Note that the LEB
  44 * number is not stored here, because it is the index used to access the
  45 * entries table.
  46 */
  47struct ubi_eba_entry {
  48	int pnum;
  49};
  50
  51/**
  52 * struct ubi_eba_table - LEB -> PEB association information
  53 * @entries: the LEB to PEB mapping (one entry per LEB).
  54 *
  55 * This structure is private to the EBA logic and should be kept here.
  56 * It is encoding the LEB to PEB association table, and is subject to
  57 * changes.
  58 */
  59struct ubi_eba_table {
  60	struct ubi_eba_entry *entries;
  61};
  62
  63/**
  64 * next_sqnum - get next sequence number.
  65 * @ubi: UBI device description object
  66 *
  67 * This function returns next sequence number to use, which is just the current
  68 * global sequence counter value. It also increases the global sequence
  69 * counter.
  70 */
  71unsigned long long ubi_next_sqnum(struct ubi_device *ubi)
  72{
  73	unsigned long long sqnum;
  74
  75	spin_lock(&ubi->ltree_lock);
  76	sqnum = ubi->global_sqnum++;
  77	spin_unlock(&ubi->ltree_lock);
  78
  79	return sqnum;
  80}
  81
  82/**
  83 * ubi_get_compat - get compatibility flags of a volume.
  84 * @ubi: UBI device description object
  85 * @vol_id: volume ID
  86 *
  87 * This function returns compatibility flags for an internal volume. User
  88 * volumes have no compatibility flags, so %0 is returned.
  89 */
  90static int ubi_get_compat(const struct ubi_device *ubi, int vol_id)
  91{
  92	if (vol_id == UBI_LAYOUT_VOLUME_ID)
  93		return UBI_LAYOUT_VOLUME_COMPAT;
  94	return 0;
  95}
  96
  97/**
  98 * ubi_eba_get_ldesc - get information about a LEB
  99 * @vol: volume description object
 100 * @lnum: logical eraseblock number
 101 * @ldesc: the LEB descriptor to fill
 102 *
 103 * Used to query information about a specific LEB.
 104 * It is currently only returning the physical position of the LEB, but will be
 105 * extended to provide more information.
 106 */
 107void ubi_eba_get_ldesc(struct ubi_volume *vol, int lnum,
 108		       struct ubi_eba_leb_desc *ldesc)
 109{
 110	ldesc->lnum = lnum;
 111	ldesc->pnum = vol->eba_tbl->entries[lnum].pnum;
 112}
 113
 114/**
 115 * ubi_eba_create_table - allocate a new EBA table and initialize it with all
 116 *			  LEBs unmapped
 117 * @vol: volume containing the EBA table to copy
 118 * @nentries: number of entries in the table
 119 *
 120 * Allocate a new EBA table and initialize it with all LEBs unmapped.
 121 * Returns a valid pointer if it succeed, an ERR_PTR() otherwise.
 122 */
 123struct ubi_eba_table *ubi_eba_create_table(struct ubi_volume *vol,
 124					   int nentries)
 125{
 126	struct ubi_eba_table *tbl;
 127	int err = -ENOMEM;
 128	int i;
 129
 130	tbl = kzalloc(sizeof(*tbl), GFP_KERNEL);
 131	if (!tbl)
 132		return ERR_PTR(-ENOMEM);
 133
 134	tbl->entries = kmalloc_array(nentries, sizeof(*tbl->entries),
 135				     GFP_KERNEL);
 136	if (!tbl->entries)
 137		goto err;
 138
 139	for (i = 0; i < nentries; i++)
 140		tbl->entries[i].pnum = UBI_LEB_UNMAPPED;
 141
 142	return tbl;
 143
 144err:
 145	kfree(tbl->entries);
 146	kfree(tbl);
 147
 148	return ERR_PTR(err);
 149}
 150
 151/**
 152 * ubi_eba_destroy_table - destroy an EBA table
 153 * @tbl: the table to destroy
 154 *
 155 * Destroy an EBA table.
 156 */
 157void ubi_eba_destroy_table(struct ubi_eba_table *tbl)
 158{
 159	if (!tbl)
 160		return;
 161
 162	kfree(tbl->entries);
 163	kfree(tbl);
 164}
 165
 166/**
 167 * ubi_eba_copy_table - copy the EBA table attached to vol into another table
 168 * @vol: volume containing the EBA table to copy
 169 * @dst: destination
 170 * @nentries: number of entries to copy
 171 *
 172 * Copy the EBA table stored in vol into the one pointed by dst.
 173 */
 174void ubi_eba_copy_table(struct ubi_volume *vol, struct ubi_eba_table *dst,
 175			int nentries)
 176{
 177	struct ubi_eba_table *src;
 178	int i;
 179
 180	ubi_assert(dst && vol && vol->eba_tbl);
 181
 182	src = vol->eba_tbl;
 183
 184	for (i = 0; i < nentries; i++)
 185		dst->entries[i].pnum = src->entries[i].pnum;
 186}
 187
 188/**
 189 * ubi_eba_replace_table - assign a new EBA table to a volume
 190 * @vol: volume containing the EBA table to copy
 191 * @tbl: new EBA table
 192 *
 193 * Assign a new EBA table to the volume and release the old one.
 194 */
 195void ubi_eba_replace_table(struct ubi_volume *vol, struct ubi_eba_table *tbl)
 196{
 197	ubi_eba_destroy_table(vol->eba_tbl);
 198	vol->eba_tbl = tbl;
 199}
 200
 201/**
 202 * ltree_lookup - look up the lock tree.
 203 * @ubi: UBI device description object
 204 * @vol_id: volume ID
 205 * @lnum: logical eraseblock number
 206 *
 207 * This function returns a pointer to the corresponding &struct ubi_ltree_entry
 208 * object if the logical eraseblock is locked and %NULL if it is not.
 209 * @ubi->ltree_lock has to be locked.
 210 */
 211static struct ubi_ltree_entry *ltree_lookup(struct ubi_device *ubi, int vol_id,
 212					    int lnum)
 213{
 214	struct rb_node *p;
 215
 216	p = ubi->ltree.rb_node;
 217	while (p) {
 218		struct ubi_ltree_entry *le;
 219
 220		le = rb_entry(p, struct ubi_ltree_entry, rb);
 221
 222		if (vol_id < le->vol_id)
 223			p = p->rb_left;
 224		else if (vol_id > le->vol_id)
 225			p = p->rb_right;
 226		else {
 227			if (lnum < le->lnum)
 228				p = p->rb_left;
 229			else if (lnum > le->lnum)
 230				p = p->rb_right;
 231			else
 232				return le;
 233		}
 234	}
 235
 236	return NULL;
 237}
 238
 239/**
 240 * ltree_add_entry - add new entry to the lock tree.
 241 * @ubi: UBI device description object
 242 * @vol_id: volume ID
 243 * @lnum: logical eraseblock number
 244 *
 245 * This function adds new entry for logical eraseblock (@vol_id, @lnum) to the
 246 * lock tree. If such entry is already there, its usage counter is increased.
 247 * Returns pointer to the lock tree entry or %-ENOMEM if memory allocation
 248 * failed.
 249 */
 250static struct ubi_ltree_entry *ltree_add_entry(struct ubi_device *ubi,
 251					       int vol_id, int lnum)
 252{
 253	struct ubi_ltree_entry *le, *le1, *le_free;
 254
 255	le = kmalloc(sizeof(struct ubi_ltree_entry), GFP_NOFS);
 256	if (!le)
 257		return ERR_PTR(-ENOMEM);
 258
 259	le->users = 0;
 260	init_rwsem(&le->mutex);
 261	le->vol_id = vol_id;
 262	le->lnum = lnum;
 263
 264	spin_lock(&ubi->ltree_lock);
 265	le1 = ltree_lookup(ubi, vol_id, lnum);
 266
 267	if (le1) {
 268		/*
 269		 * This logical eraseblock is already locked. The newly
 270		 * allocated lock entry is not needed.
 271		 */
 272		le_free = le;
 273		le = le1;
 274	} else {
 275		struct rb_node **p, *parent = NULL;
 276
 277		/*
 278		 * No lock entry, add the newly allocated one to the
 279		 * @ubi->ltree RB-tree.
 280		 */
 281		le_free = NULL;
 282
 283		p = &ubi->ltree.rb_node;
 284		while (*p) {
 285			parent = *p;
 286			le1 = rb_entry(parent, struct ubi_ltree_entry, rb);
 287
 288			if (vol_id < le1->vol_id)
 289				p = &(*p)->rb_left;
 290			else if (vol_id > le1->vol_id)
 291				p = &(*p)->rb_right;
 292			else {
 293				ubi_assert(lnum != le1->lnum);
 294				if (lnum < le1->lnum)
 295					p = &(*p)->rb_left;
 296				else
 297					p = &(*p)->rb_right;
 298			}
 299		}
 300
 301		rb_link_node(&le->rb, parent, p);
 302		rb_insert_color(&le->rb, &ubi->ltree);
 303	}
 304	le->users += 1;
 305	spin_unlock(&ubi->ltree_lock);
 306
 307	kfree(le_free);
 308	return le;
 309}
 310
 311/**
 312 * leb_read_lock - lock logical eraseblock for reading.
 313 * @ubi: UBI device description object
 314 * @vol_id: volume ID
 315 * @lnum: logical eraseblock number
 316 *
 317 * This function locks a logical eraseblock for reading. Returns zero in case
 318 * of success and a negative error code in case of failure.
 319 */
 320static int leb_read_lock(struct ubi_device *ubi, int vol_id, int lnum)
 321{
 322	struct ubi_ltree_entry *le;
 323
 324	le = ltree_add_entry(ubi, vol_id, lnum);
 325	if (IS_ERR(le))
 326		return PTR_ERR(le);
 327	down_read(&le->mutex);
 328	return 0;
 329}
 330
 331/**
 332 * leb_read_unlock - unlock logical eraseblock.
 333 * @ubi: UBI device description object
 334 * @vol_id: volume ID
 335 * @lnum: logical eraseblock number
 336 */
 337static void leb_read_unlock(struct ubi_device *ubi, int vol_id, int lnum)
 338{
 339	struct ubi_ltree_entry *le;
 340
 341	spin_lock(&ubi->ltree_lock);
 342	le = ltree_lookup(ubi, vol_id, lnum);
 343	le->users -= 1;
 344	ubi_assert(le->users >= 0);
 345	up_read(&le->mutex);
 346	if (le->users == 0) {
 347		rb_erase(&le->rb, &ubi->ltree);
 348		kfree(le);
 349	}
 350	spin_unlock(&ubi->ltree_lock);
 351}
 352
 353/**
 354 * leb_write_lock - lock logical eraseblock for writing.
 355 * @ubi: UBI device description object
 356 * @vol_id: volume ID
 357 * @lnum: logical eraseblock number
 358 *
 359 * This function locks a logical eraseblock for writing. Returns zero in case
 360 * of success and a negative error code in case of failure.
 361 */
 362static int leb_write_lock(struct ubi_device *ubi, int vol_id, int lnum)
 363{
 364	struct ubi_ltree_entry *le;
 365
 366	le = ltree_add_entry(ubi, vol_id, lnum);
 367	if (IS_ERR(le))
 368		return PTR_ERR(le);
 369	down_write(&le->mutex);
 370	return 0;
 371}
 372
 373/**
 374 * leb_write_trylock - try to lock logical eraseblock for writing.
 375 * @ubi: UBI device description object
 376 * @vol_id: volume ID
 377 * @lnum: logical eraseblock number
 378 *
 379 * This function locks a logical eraseblock for writing if there is no
 380 * contention and does nothing if there is contention. Returns %0 in case of
 381 * success, %1 in case of contention, and and a negative error code in case of
 382 * failure.
 383 */
 384static int leb_write_trylock(struct ubi_device *ubi, int vol_id, int lnum)
 385{
 386	struct ubi_ltree_entry *le;
 387
 388	le = ltree_add_entry(ubi, vol_id, lnum);
 389	if (IS_ERR(le))
 390		return PTR_ERR(le);
 391	if (down_write_trylock(&le->mutex))
 392		return 0;
 393
 394	/* Contention, cancel */
 395	spin_lock(&ubi->ltree_lock);
 396	le->users -= 1;
 397	ubi_assert(le->users >= 0);
 398	if (le->users == 0) {
 399		rb_erase(&le->rb, &ubi->ltree);
 400		kfree(le);
 401	}
 402	spin_unlock(&ubi->ltree_lock);
 403
 404	return 1;
 405}
 406
 407/**
 408 * leb_write_unlock - unlock logical eraseblock.
 409 * @ubi: UBI device description object
 410 * @vol_id: volume ID
 411 * @lnum: logical eraseblock number
 412 */
 413static void leb_write_unlock(struct ubi_device *ubi, int vol_id, int lnum)
 414{
 415	struct ubi_ltree_entry *le;
 416
 417	spin_lock(&ubi->ltree_lock);
 418	le = ltree_lookup(ubi, vol_id, lnum);
 419	le->users -= 1;
 420	ubi_assert(le->users >= 0);
 421	up_write(&le->mutex);
 422	if (le->users == 0) {
 423		rb_erase(&le->rb, &ubi->ltree);
 424		kfree(le);
 425	}
 426	spin_unlock(&ubi->ltree_lock);
 427}
 428
 429/**
 430 * ubi_eba_is_mapped - check if a LEB is mapped.
 431 * @vol: volume description object
 432 * @lnum: logical eraseblock number
 433 *
 434 * This function returns true if the LEB is mapped, false otherwise.
 435 */
 436bool ubi_eba_is_mapped(struct ubi_volume *vol, int lnum)
 437{
 438	return vol->eba_tbl->entries[lnum].pnum >= 0;
 439}
 440
 441/**
 442 * ubi_eba_unmap_leb - un-map logical eraseblock.
 443 * @ubi: UBI device description object
 444 * @vol: volume description object
 445 * @lnum: logical eraseblock number
 446 *
 447 * This function un-maps logical eraseblock @lnum and schedules corresponding
 448 * physical eraseblock for erasure. Returns zero in case of success and a
 449 * negative error code in case of failure.
 450 */
 451int ubi_eba_unmap_leb(struct ubi_device *ubi, struct ubi_volume *vol,
 452		      int lnum)
 453{
 454	int err, pnum, vol_id = vol->vol_id;
 455
 456	if (ubi->ro_mode)
 457		return -EROFS;
 458
 459	err = leb_write_lock(ubi, vol_id, lnum);
 460	if (err)
 461		return err;
 462
 463	pnum = vol->eba_tbl->entries[lnum].pnum;
 464	if (pnum < 0)
 465		/* This logical eraseblock is already unmapped */
 466		goto out_unlock;
 467
 468	dbg_eba("erase LEB %d:%d, PEB %d", vol_id, lnum, pnum);
 469
 470	down_read(&ubi->fm_eba_sem);
 471	vol->eba_tbl->entries[lnum].pnum = UBI_LEB_UNMAPPED;
 472	up_read(&ubi->fm_eba_sem);
 473	err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 0);
 474
 475out_unlock:
 476	leb_write_unlock(ubi, vol_id, lnum);
 477	return err;
 478}
 479
 480#ifdef CONFIG_MTD_UBI_FASTMAP
 481/**
 482 * check_mapping - check and fixup a mapping
 483 * @ubi: UBI device description object
 484 * @vol: volume description object
 485 * @lnum: logical eraseblock number
 486 * @pnum: physical eraseblock number
 487 *
 488 * Checks whether a given mapping is valid. Fastmap cannot track LEB unmap
 489 * operations, if such an operation is interrupted the mapping still looks
 490 * good, but upon first read an ECC is reported to the upper layer.
 491 * Normaly during the full-scan at attach time this is fixed, for Fastmap
 492 * we have to deal with it while reading.
 493 * If the PEB behind a LEB shows this symthom we change the mapping to
 494 * %UBI_LEB_UNMAPPED and schedule the PEB for erasure.
 495 *
 496 * Returns 0 on success, negative error code in case of failure.
 497 */
 498static int check_mapping(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
 499			 int *pnum)
 500{
 501	int err;
 502	struct ubi_vid_io_buf *vidb;
 503	struct ubi_vid_hdr *vid_hdr;
 504
 505	if (!ubi->fast_attach)
 506		return 0;
 507
 508	if (!vol->checkmap || test_bit(lnum, vol->checkmap))
 509		return 0;
 510
 511	vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
 512	if (!vidb)
 513		return -ENOMEM;
 514
 515	err = ubi_io_read_vid_hdr(ubi, *pnum, vidb, 0);
 516	if (err > 0 && err != UBI_IO_BITFLIPS) {
 517		int torture = 0;
 518
 519		switch (err) {
 520			case UBI_IO_FF:
 521			case UBI_IO_FF_BITFLIPS:
 522			case UBI_IO_BAD_HDR:
 523			case UBI_IO_BAD_HDR_EBADMSG:
 524				break;
 525			default:
 526				ubi_assert(0);
 527		}
 528
 529		if (err == UBI_IO_BAD_HDR_EBADMSG || err == UBI_IO_FF_BITFLIPS)
 530			torture = 1;
 531
 532		down_read(&ubi->fm_eba_sem);
 533		vol->eba_tbl->entries[lnum].pnum = UBI_LEB_UNMAPPED;
 534		up_read(&ubi->fm_eba_sem);
 535		ubi_wl_put_peb(ubi, vol->vol_id, lnum, *pnum, torture);
 536
 537		*pnum = UBI_LEB_UNMAPPED;
 538	} else if (err < 0) {
 539		ubi_err(ubi, "unable to read VID header back from PEB %i: %i",
 540			*pnum, err);
 541
 542		goto out_free;
 543	} else {
 544		int found_vol_id, found_lnum;
 545
 546		ubi_assert(err == 0 || err == UBI_IO_BITFLIPS);
 547
 548		vid_hdr = ubi_get_vid_hdr(vidb);
 549		found_vol_id = be32_to_cpu(vid_hdr->vol_id);
 550		found_lnum = be32_to_cpu(vid_hdr->lnum);
 551
 552		if (found_lnum != lnum || found_vol_id != vol->vol_id) {
 553			ubi_err(ubi, "EBA mismatch! PEB %i is LEB %i:%i instead of LEB %i:%i",
 554				*pnum, found_vol_id, found_lnum, vol->vol_id, lnum);
 555			ubi_ro_mode(ubi);
 556			err = -EINVAL;
 557			goto out_free;
 558		}
 559	}
 560
 561	set_bit(lnum, vol->checkmap);
 562	err = 0;
 563
 564out_free:
 565	ubi_free_vid_buf(vidb);
 566
 567	return err;
 568}
 569#else
 570static int check_mapping(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
 571		  int *pnum)
 572{
 573	return 0;
 574}
 575#endif
 576
 577/**
 578 * ubi_eba_read_leb - read data.
 579 * @ubi: UBI device description object
 580 * @vol: volume description object
 581 * @lnum: logical eraseblock number
 582 * @buf: buffer to store the read data
 583 * @offset: offset from where to read
 584 * @len: how many bytes to read
 585 * @check: data CRC check flag
 586 *
 587 * If the logical eraseblock @lnum is unmapped, @buf is filled with 0xFF
 588 * bytes. The @check flag only makes sense for static volumes and forces
 589 * eraseblock data CRC checking.
 590 *
 591 * In case of success this function returns zero. In case of a static volume,
 592 * if data CRC mismatches - %-EBADMSG is returned. %-EBADMSG may also be
 593 * returned for any volume type if an ECC error was detected by the MTD device
 594 * driver. Other negative error cored may be returned in case of other errors.
 595 */
 596int ubi_eba_read_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
 597		     void *buf, int offset, int len, int check)
 598{
 599	int err, pnum, scrub = 0, vol_id = vol->vol_id;
 600	struct ubi_vid_io_buf *vidb;
 601	struct ubi_vid_hdr *vid_hdr;
 602	uint32_t crc;
 603
 604	err = leb_read_lock(ubi, vol_id, lnum);
 605	if (err)
 606		return err;
 607
 608	pnum = vol->eba_tbl->entries[lnum].pnum;
 609	if (pnum >= 0) {
 610		err = check_mapping(ubi, vol, lnum, &pnum);
 611		if (err < 0)
 612			goto out_unlock;
 613	}
 614
 615	if (pnum == UBI_LEB_UNMAPPED) {
 616		/*
 617		 * The logical eraseblock is not mapped, fill the whole buffer
 618		 * with 0xFF bytes. The exception is static volumes for which
 619		 * it is an error to read unmapped logical eraseblocks.
 620		 */
 621		dbg_eba("read %d bytes from offset %d of LEB %d:%d (unmapped)",
 622			len, offset, vol_id, lnum);
 623		leb_read_unlock(ubi, vol_id, lnum);
 624		ubi_assert(vol->vol_type != UBI_STATIC_VOLUME);
 625		memset(buf, 0xFF, len);
 626		return 0;
 627	}
 628
 629	dbg_eba("read %d bytes from offset %d of LEB %d:%d, PEB %d",
 630		len, offset, vol_id, lnum, pnum);
 631
 632	if (vol->vol_type == UBI_DYNAMIC_VOLUME)
 633		check = 0;
 634
 635retry:
 636	if (check) {
 637		vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
 638		if (!vidb) {
 639			err = -ENOMEM;
 640			goto out_unlock;
 641		}
 642
 643		vid_hdr = ubi_get_vid_hdr(vidb);
 644
 645		err = ubi_io_read_vid_hdr(ubi, pnum, vidb, 1);
 646		if (err && err != UBI_IO_BITFLIPS) {
 647			if (err > 0) {
 648				/*
 649				 * The header is either absent or corrupted.
 650				 * The former case means there is a bug -
 651				 * switch to read-only mode just in case.
 652				 * The latter case means a real corruption - we
 653				 * may try to recover data. FIXME: but this is
 654				 * not implemented.
 655				 */
 656				if (err == UBI_IO_BAD_HDR_EBADMSG ||
 657				    err == UBI_IO_BAD_HDR) {
 658					ubi_warn(ubi, "corrupted VID header at PEB %d, LEB %d:%d",
 659						 pnum, vol_id, lnum);
 660					err = -EBADMSG;
 661				} else {
 662					/*
 663					 * Ending up here in the non-Fastmap case
 664					 * is a clear bug as the VID header had to
 665					 * be present at scan time to have it referenced.
 666					 * With fastmap the story is more complicated.
 667					 * Fastmap has the mapping info without the need
 668					 * of a full scan. So the LEB could have been
 669					 * unmapped, Fastmap cannot know this and keeps
 670					 * the LEB referenced.
 671					 * This is valid and works as the layer above UBI
 672					 * has to do bookkeeping about used/referenced
 673					 * LEBs in any case.
 674					 */
 675					if (ubi->fast_attach) {
 676						err = -EBADMSG;
 677					} else {
 678						err = -EINVAL;
 679						ubi_ro_mode(ubi);
 680					}
 681				}
 682			}
 683			goto out_free;
 684		} else if (err == UBI_IO_BITFLIPS)
 685			scrub = 1;
 686
 687		ubi_assert(lnum < be32_to_cpu(vid_hdr->used_ebs));
 688		ubi_assert(len == be32_to_cpu(vid_hdr->data_size));
 689
 690		crc = be32_to_cpu(vid_hdr->data_crc);
 691		ubi_free_vid_buf(vidb);
 692	}
 693
 694	err = ubi_io_read_data(ubi, buf, pnum, offset, len);
 695	if (err) {
 696		if (err == UBI_IO_BITFLIPS)
 697			scrub = 1;
 698		else if (mtd_is_eccerr(err)) {
 699			if (vol->vol_type == UBI_DYNAMIC_VOLUME)
 700				goto out_unlock;
 701			scrub = 1;
 702			if (!check) {
 703				ubi_msg(ubi, "force data checking");
 704				check = 1;
 705				goto retry;
 706			}
 707		} else
 708			goto out_unlock;
 709	}
 710
 711	if (check) {
 712		uint32_t crc1 = crc32(UBI_CRC32_INIT, buf, len);
 713		if (crc1 != crc) {
 714			ubi_warn(ubi, "CRC error: calculated %#08x, must be %#08x",
 715				 crc1, crc);
 716			err = -EBADMSG;
 717			goto out_unlock;
 718		}
 719	}
 720
 721	if (scrub)
 722		err = ubi_wl_scrub_peb(ubi, pnum);
 723
 724	leb_read_unlock(ubi, vol_id, lnum);
 725	return err;
 726
 727out_free:
 728	ubi_free_vid_buf(vidb);
 729out_unlock:
 730	leb_read_unlock(ubi, vol_id, lnum);
 731	return err;
 732}
 733
 734/**
 735 * ubi_eba_read_leb_sg - read data into a scatter gather list.
 736 * @ubi: UBI device description object
 737 * @vol: volume description object
 738 * @lnum: logical eraseblock number
 739 * @sgl: UBI scatter gather list to store the read data
 740 * @offset: offset from where to read
 741 * @len: how many bytes to read
 742 * @check: data CRC check flag
 743 *
 744 * This function works exactly like ubi_eba_read_leb(). But instead of
 745 * storing the read data into a buffer it writes to an UBI scatter gather
 746 * list.
 747 */
 748int ubi_eba_read_leb_sg(struct ubi_device *ubi, struct ubi_volume *vol,
 749			struct ubi_sgl *sgl, int lnum, int offset, int len,
 750			int check)
 751{
 752	int to_read;
 753	int ret;
 754	struct scatterlist *sg;
 755
 756	for (;;) {
 757		ubi_assert(sgl->list_pos < UBI_MAX_SG_COUNT);
 758		sg = &sgl->sg[sgl->list_pos];
 759		if (len < sg->length - sgl->page_pos)
 760			to_read = len;
 761		else
 762			to_read = sg->length - sgl->page_pos;
 763
 764		ret = ubi_eba_read_leb(ubi, vol, lnum,
 765				       sg_virt(sg) + sgl->page_pos, offset,
 766				       to_read, check);
 767		if (ret < 0)
 768			return ret;
 769
 770		offset += to_read;
 771		len -= to_read;
 772		if (!len) {
 773			sgl->page_pos += to_read;
 774			if (sgl->page_pos == sg->length) {
 775				sgl->list_pos++;
 776				sgl->page_pos = 0;
 777			}
 778
 779			break;
 780		}
 781
 782		sgl->list_pos++;
 783		sgl->page_pos = 0;
 784	}
 785
 786	return ret;
 787}
 788
 789/**
 790 * try_recover_peb - try to recover from write failure.
 791 * @vol: volume description object
 792 * @pnum: the physical eraseblock to recover
 793 * @lnum: logical eraseblock number
 794 * @buf: data which was not written because of the write failure
 795 * @offset: offset of the failed write
 796 * @len: how many bytes should have been written
 797 * @vidb: VID buffer
 798 * @retry: whether the caller should retry in case of failure
 799 *
 800 * This function is called in case of a write failure and moves all good data
 801 * from the potentially bad physical eraseblock to a good physical eraseblock.
 802 * This function also writes the data which was not written due to the failure.
 803 * Returns 0 in case of success, and a negative error code in case of failure.
 804 * In case of failure, the %retry parameter is set to false if this is a fatal
 805 * error (retrying won't help), and true otherwise.
 806 */
 807static int try_recover_peb(struct ubi_volume *vol, int pnum, int lnum,
 808			   const void *buf, int offset, int len,
 809			   struct ubi_vid_io_buf *vidb, bool *retry)
 810{
 811	struct ubi_device *ubi = vol->ubi;
 812	struct ubi_vid_hdr *vid_hdr;
 813	int new_pnum, err, vol_id = vol->vol_id, data_size;
 814	uint32_t crc;
 815
 816	*retry = false;
 817
 818	new_pnum = ubi_wl_get_peb(ubi);
 819	if (new_pnum < 0) {
 820		err = new_pnum;
 821		goto out_put;
 822	}
 823
 824	ubi_msg(ubi, "recover PEB %d, move data to PEB %d",
 825		pnum, new_pnum);
 826
 827	err = ubi_io_read_vid_hdr(ubi, pnum, vidb, 1);
 828	if (err && err != UBI_IO_BITFLIPS) {
 829		if (err > 0)
 830			err = -EIO;
 831		goto out_put;
 832	}
 833
 834	vid_hdr = ubi_get_vid_hdr(vidb);
 835	ubi_assert(vid_hdr->vol_type == UBI_VID_DYNAMIC);
 836
 837	mutex_lock(&ubi->buf_mutex);
 838	memset(ubi->peb_buf + offset, 0xFF, len);
 839
 840	/* Read everything before the area where the write failure happened */
 841	if (offset > 0) {
 842		err = ubi_io_read_data(ubi, ubi->peb_buf, pnum, 0, offset);
 843		if (err && err != UBI_IO_BITFLIPS)
 844			goto out_unlock;
 845	}
 846
 847	*retry = true;
 848
 849	memcpy(ubi->peb_buf + offset, buf, len);
 850
 851	data_size = offset + len;
 852	crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
 853	vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
 854	vid_hdr->copy_flag = 1;
 855	vid_hdr->data_size = cpu_to_be32(data_size);
 856	vid_hdr->data_crc = cpu_to_be32(crc);
 857	err = ubi_io_write_vid_hdr(ubi, new_pnum, vidb);
 858	if (err)
 859		goto out_unlock;
 860
 861	err = ubi_io_write_data(ubi, ubi->peb_buf, new_pnum, 0, data_size);
 862
 863out_unlock:
 864	mutex_unlock(&ubi->buf_mutex);
 865
 866	if (!err)
 867		vol->eba_tbl->entries[lnum].pnum = new_pnum;
 868
 869out_put:
 870	up_read(&ubi->fm_eba_sem);
 871
 872	if (!err) {
 873		ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
 874		ubi_msg(ubi, "data was successfully recovered");
 875	} else if (new_pnum >= 0) {
 876		/*
 877		 * Bad luck? This physical eraseblock is bad too? Crud. Let's
 878		 * try to get another one.
 879		 */
 880		ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
 881		ubi_warn(ubi, "failed to write to PEB %d", new_pnum);
 882	}
 883
 884	return err;
 885}
 886
 887/**
 888 * recover_peb - recover from write failure.
 889 * @ubi: UBI device description object
 890 * @pnum: the physical eraseblock to recover
 891 * @vol_id: volume ID
 892 * @lnum: logical eraseblock number
 893 * @buf: data which was not written because of the write failure
 894 * @offset: offset of the failed write
 895 * @len: how many bytes should have been written
 896 *
 897 * This function is called in case of a write failure and moves all good data
 898 * from the potentially bad physical eraseblock to a good physical eraseblock.
 899 * This function also writes the data which was not written due to the failure.
 900 * Returns 0 in case of success, and a negative error code in case of failure.
 901 * This function tries %UBI_IO_RETRIES before giving up.
 902 */
 903static int recover_peb(struct ubi_device *ubi, int pnum, int vol_id, int lnum,
 904		       const void *buf, int offset, int len)
 905{
 906	int err, idx = vol_id2idx(ubi, vol_id), tries;
 907	struct ubi_volume *vol = ubi->volumes[idx];
 908	struct ubi_vid_io_buf *vidb;
 909
 910	vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
 911	if (!vidb)
 912		return -ENOMEM;
 913
 914	for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
 915		bool retry;
 916
 917		err = try_recover_peb(vol, pnum, lnum, buf, offset, len, vidb,
 918				      &retry);
 919		if (!err || !retry)
 920			break;
 921
 922		ubi_msg(ubi, "try again");
 923	}
 924
 925	ubi_free_vid_buf(vidb);
 926
 927	return err;
 928}
 929
 930/**
 931 * try_write_vid_and_data - try to write VID header and data to a new PEB.
 932 * @vol: volume description object
 933 * @lnum: logical eraseblock number
 934 * @vidb: the VID buffer to write
 935 * @buf: buffer containing the data
 936 * @offset: where to start writing data
 937 * @len: how many bytes should be written
 938 *
 939 * This function tries to write VID header and data belonging to logical
 940 * eraseblock @lnum of volume @vol to a new physical eraseblock. Returns zero
 941 * in case of success and a negative error code in case of failure.
 942 * In case of error, it is possible that something was still written to the
 943 * flash media, but may be some garbage.
 944 */
 945static int try_write_vid_and_data(struct ubi_volume *vol, int lnum,
 946				  struct ubi_vid_io_buf *vidb, const void *buf,
 947				  int offset, int len)
 948{
 949	struct ubi_device *ubi = vol->ubi;
 950	int pnum, opnum, err, vol_id = vol->vol_id;
 951
 952	pnum = ubi_wl_get_peb(ubi);
 953	if (pnum < 0) {
 954		err = pnum;
 955		goto out_put;
 956	}
 957
 958	opnum = vol->eba_tbl->entries[lnum].pnum;
 959
 960	dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
 961		len, offset, vol_id, lnum, pnum);
 962
 963	err = ubi_io_write_vid_hdr(ubi, pnum, vidb);
 964	if (err) {
 965		ubi_warn(ubi, "failed to write VID header to LEB %d:%d, PEB %d",
 966			 vol_id, lnum, pnum);
 967		goto out_put;
 968	}
 969
 970	if (len) {
 971		err = ubi_io_write_data(ubi, buf, pnum, offset, len);
 972		if (err) {
 973			ubi_warn(ubi,
 974				 "failed to write %d bytes at offset %d of LEB %d:%d, PEB %d",
 975				 len, offset, vol_id, lnum, pnum);
 976			goto out_put;
 977		}
 978	}
 979
 980	vol->eba_tbl->entries[lnum].pnum = pnum;
 981
 982out_put:
 983	up_read(&ubi->fm_eba_sem);
 984
 985	if (err && pnum >= 0)
 986		err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
 987	else if (!err && opnum >= 0)
 988		err = ubi_wl_put_peb(ubi, vol_id, lnum, opnum, 0);
 989
 990	return err;
 991}
 992
 993/**
 994 * ubi_eba_write_leb - write data to dynamic volume.
 995 * @ubi: UBI device description object
 996 * @vol: volume description object
 997 * @lnum: logical eraseblock number
 998 * @buf: the data to write
 999 * @offset: offset within the logical eraseblock where to write
1000 * @len: how many bytes to write
1001 *
1002 * This function writes data to logical eraseblock @lnum of a dynamic volume
1003 * @vol. Returns zero in case of success and a negative error code in case
1004 * of failure. In case of error, it is possible that something was still
1005 * written to the flash media, but may be some garbage.
1006 * This function retries %UBI_IO_RETRIES times before giving up.
1007 */
1008int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
1009		      const void *buf, int offset, int len)
1010{
1011	int err, pnum, tries, vol_id = vol->vol_id;
1012	struct ubi_vid_io_buf *vidb;
1013	struct ubi_vid_hdr *vid_hdr;
1014
1015	if (ubi->ro_mode)
1016		return -EROFS;
1017
1018	err = leb_write_lock(ubi, vol_id, lnum);
1019	if (err)
1020		return err;
1021
1022	pnum = vol->eba_tbl->entries[lnum].pnum;
1023	if (pnum >= 0) {
1024		err = check_mapping(ubi, vol, lnum, &pnum);
1025		if (err < 0)
1026			goto out;
1027	}
1028
1029	if (pnum >= 0) {
1030		dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
1031			len, offset, vol_id, lnum, pnum);
1032
1033		err = ubi_io_write_data(ubi, buf, pnum, offset, len);
1034		if (err) {
1035			ubi_warn(ubi, "failed to write data to PEB %d", pnum);
1036			if (err == -EIO && ubi->bad_allowed)
1037				err = recover_peb(ubi, pnum, vol_id, lnum, buf,
1038						  offset, len);
1039		}
1040
1041		goto out;
1042	}
1043
1044	/*
1045	 * The logical eraseblock is not mapped. We have to get a free physical
1046	 * eraseblock and write the volume identifier header there first.
1047	 */
1048	vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
1049	if (!vidb) {
1050		leb_write_unlock(ubi, vol_id, lnum);
1051		return -ENOMEM;
1052	}
1053
1054	vid_hdr = ubi_get_vid_hdr(vidb);
1055
1056	vid_hdr->vol_type = UBI_VID_DYNAMIC;
1057	vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1058	vid_hdr->vol_id = cpu_to_be32(vol_id);
1059	vid_hdr->lnum = cpu_to_be32(lnum);
1060	vid_hdr->compat = ubi_get_compat(ubi, vol_id);
1061	vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
1062
1063	for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
1064		err = try_write_vid_and_data(vol, lnum, vidb, buf, offset, len);
1065		if (err != -EIO || !ubi->bad_allowed)
1066			break;
1067
1068		/*
1069		 * Fortunately, this is the first write operation to this
1070		 * physical eraseblock, so just put it and request a new one.
1071		 * We assume that if this physical eraseblock went bad, the
1072		 * erase code will handle that.
1073		 */
1074		vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1075		ubi_msg(ubi, "try another PEB");
1076	}
1077
1078	ubi_free_vid_buf(vidb);
1079
1080out:
1081	if (err)
1082		ubi_ro_mode(ubi);
1083
1084	leb_write_unlock(ubi, vol_id, lnum);
1085
1086	return err;
1087}
1088
1089/**
1090 * ubi_eba_write_leb_st - write data to static volume.
1091 * @ubi: UBI device description object
1092 * @vol: volume description object
1093 * @lnum: logical eraseblock number
1094 * @buf: data to write
1095 * @len: how many bytes to write
1096 * @used_ebs: how many logical eraseblocks will this volume contain
1097 *
1098 * This function writes data to logical eraseblock @lnum of static volume
1099 * @vol. The @used_ebs argument should contain total number of logical
1100 * eraseblock in this static volume.
1101 *
1102 * When writing to the last logical eraseblock, the @len argument doesn't have
1103 * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
1104 * to the real data size, although the @buf buffer has to contain the
1105 * alignment. In all other cases, @len has to be aligned.
1106 *
1107 * It is prohibited to write more than once to logical eraseblocks of static
1108 * volumes. This function returns zero in case of success and a negative error
1109 * code in case of failure.
1110 */
1111int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
1112			 int lnum, const void *buf, int len, int used_ebs)
1113{
1114	int err, tries, data_size = len, vol_id = vol->vol_id;
1115	struct ubi_vid_io_buf *vidb;
1116	struct ubi_vid_hdr *vid_hdr;
1117	uint32_t crc;
1118
1119	if (ubi->ro_mode)
1120		return -EROFS;
1121
1122	if (lnum == used_ebs - 1)
1123		/* If this is the last LEB @len may be unaligned */
1124		len = ALIGN(data_size, ubi->min_io_size);
1125	else
1126		ubi_assert(!(len & (ubi->min_io_size - 1)));
1127
1128	vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
1129	if (!vidb)
1130		return -ENOMEM;
1131
1132	vid_hdr = ubi_get_vid_hdr(vidb);
1133
1134	err = leb_write_lock(ubi, vol_id, lnum);
1135	if (err)
1136		goto out;
1137
1138	vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1139	vid_hdr->vol_id = cpu_to_be32(vol_id);
1140	vid_hdr->lnum = cpu_to_be32(lnum);
1141	vid_hdr->compat = ubi_get_compat(ubi, vol_id);
1142	vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
1143
1144	crc = crc32(UBI_CRC32_INIT, buf, data_size);
1145	vid_hdr->vol_type = UBI_VID_STATIC;
1146	vid_hdr->data_size = cpu_to_be32(data_size);
1147	vid_hdr->used_ebs = cpu_to_be32(used_ebs);
1148	vid_hdr->data_crc = cpu_to_be32(crc);
1149
1150	ubi_assert(vol->eba_tbl->entries[lnum].pnum < 0);
1151
1152	for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
1153		err = try_write_vid_and_data(vol, lnum, vidb, buf, 0, len);
1154		if (err != -EIO || !ubi->bad_allowed)
1155			break;
1156
1157		vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1158		ubi_msg(ubi, "try another PEB");
1159	}
1160
1161	if (err)
1162		ubi_ro_mode(ubi);
1163
1164	leb_write_unlock(ubi, vol_id, lnum);
1165
1166out:
1167	ubi_free_vid_buf(vidb);
1168
1169	return err;
1170}
1171
1172/*
1173 * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
1174 * @ubi: UBI device description object
1175 * @vol: volume description object
1176 * @lnum: logical eraseblock number
1177 * @buf: data to write
1178 * @len: how many bytes to write
1179 *
1180 * This function changes the contents of a logical eraseblock atomically. @buf
1181 * has to contain new logical eraseblock data, and @len - the length of the
1182 * data, which has to be aligned. This function guarantees that in case of an
1183 * unclean reboot the old contents is preserved. Returns zero in case of
1184 * success and a negative error code in case of failure.
1185 *
1186 * UBI reserves one LEB for the "atomic LEB change" operation, so only one
1187 * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
1188 */
1189int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
1190			      int lnum, const void *buf, int len)
1191{
1192	int err, tries, vol_id = vol->vol_id;
1193	struct ubi_vid_io_buf *vidb;
1194	struct ubi_vid_hdr *vid_hdr;
1195	uint32_t crc;
1196
1197	if (ubi->ro_mode)
1198		return -EROFS;
1199
1200	if (len == 0) {
1201		/*
1202		 * Special case when data length is zero. In this case the LEB
1203		 * has to be unmapped and mapped somewhere else.
1204		 */
1205		err = ubi_eba_unmap_leb(ubi, vol, lnum);
1206		if (err)
1207			return err;
1208		return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0);
1209	}
1210
1211	vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
1212	if (!vidb)
1213		return -ENOMEM;
1214
1215	vid_hdr = ubi_get_vid_hdr(vidb);
1216
1217	mutex_lock(&ubi->alc_mutex);
1218	err = leb_write_lock(ubi, vol_id, lnum);
1219	if (err)
1220		goto out_mutex;
1221
1222	vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1223	vid_hdr->vol_id = cpu_to_be32(vol_id);
1224	vid_hdr->lnum = cpu_to_be32(lnum);
1225	vid_hdr->compat = ubi_get_compat(ubi, vol_id);
1226	vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
1227
1228	crc = crc32(UBI_CRC32_INIT, buf, len);
1229	vid_hdr->vol_type = UBI_VID_DYNAMIC;
1230	vid_hdr->data_size = cpu_to_be32(len);
1231	vid_hdr->copy_flag = 1;
1232	vid_hdr->data_crc = cpu_to_be32(crc);
1233
1234	dbg_eba("change LEB %d:%d", vol_id, lnum);
1235
1236	for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
1237		err = try_write_vid_and_data(vol, lnum, vidb, buf, 0, len);
1238		if (err != -EIO || !ubi->bad_allowed)
1239			break;
1240
1241		vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1242		ubi_msg(ubi, "try another PEB");
1243	}
1244
1245	/*
1246	 * This flash device does not admit of bad eraseblocks or
1247	 * something nasty and unexpected happened. Switch to read-only
1248	 * mode just in case.
1249	 */
1250	if (err)
1251		ubi_ro_mode(ubi);
1252
1253	leb_write_unlock(ubi, vol_id, lnum);
1254
1255out_mutex:
1256	mutex_unlock(&ubi->alc_mutex);
1257	ubi_free_vid_buf(vidb);
1258	return err;
1259}
1260
1261/**
1262 * is_error_sane - check whether a read error is sane.
1263 * @err: code of the error happened during reading
1264 *
1265 * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
1266 * cannot read data from the target PEB (an error @err happened). If the error
1267 * code is sane, then we treat this error as non-fatal. Otherwise the error is
1268 * fatal and UBI will be switched to R/O mode later.
1269 *
1270 * The idea is that we try not to switch to R/O mode if the read error is
1271 * something which suggests there was a real read problem. E.g., %-EIO. Or a
1272 * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
1273 * mode, simply because we do not know what happened at the MTD level, and we
1274 * cannot handle this. E.g., the underlying driver may have become crazy, and
1275 * it is safer to switch to R/O mode to preserve the data.
1276 *
1277 * And bear in mind, this is about reading from the target PEB, i.e. the PEB
1278 * which we have just written.
1279 */
1280static int is_error_sane(int err)
1281{
1282	if (err == -EIO || err == -ENOMEM || err == UBI_IO_BAD_HDR ||
1283	    err == UBI_IO_BAD_HDR_EBADMSG || err == -ETIMEDOUT)
1284		return 0;
1285	return 1;
1286}
1287
1288/**
1289 * ubi_eba_copy_leb - copy logical eraseblock.
1290 * @ubi: UBI device description object
1291 * @from: physical eraseblock number from where to copy
1292 * @to: physical eraseblock number where to copy
1293 * @vid_hdr: VID header of the @from physical eraseblock
1294 *
1295 * This function copies logical eraseblock from physical eraseblock @from to
1296 * physical eraseblock @to. The @vid_hdr buffer may be changed by this
1297 * function. Returns:
1298 *   o %0 in case of success;
1299 *   o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_TARGET_BITFLIPS, etc;
1300 *   o a negative error code in case of failure.
1301 */
1302int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
1303		     struct ubi_vid_io_buf *vidb)
1304{
1305	int err, vol_id, lnum, data_size, aldata_size, idx;
1306	struct ubi_vid_hdr *vid_hdr = ubi_get_vid_hdr(vidb);
1307	struct ubi_volume *vol;
1308	uint32_t crc;
1309
1310	ubi_assert(rwsem_is_locked(&ubi->fm_eba_sem));
1311
1312	vol_id = be32_to_cpu(vid_hdr->vol_id);
1313	lnum = be32_to_cpu(vid_hdr->lnum);
1314
1315	dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
1316
1317	if (vid_hdr->vol_type == UBI_VID_STATIC) {
1318		data_size = be32_to_cpu(vid_hdr->data_size);
1319		aldata_size = ALIGN(data_size, ubi->min_io_size);
1320	} else
1321		data_size = aldata_size =
1322			    ubi->leb_size - be32_to_cpu(vid_hdr->data_pad);
1323
1324	idx = vol_id2idx(ubi, vol_id);
1325	spin_lock(&ubi->volumes_lock);
1326	/*
1327	 * Note, we may race with volume deletion, which means that the volume
1328	 * this logical eraseblock belongs to might be being deleted. Since the
1329	 * volume deletion un-maps all the volume's logical eraseblocks, it will
1330	 * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1331	 */
1332	vol = ubi->volumes[idx];
1333	spin_unlock(&ubi->volumes_lock);
1334	if (!vol) {
1335		/* No need to do further work, cancel */
1336		dbg_wl("volume %d is being removed, cancel", vol_id);
1337		return MOVE_CANCEL_RACE;
1338	}
1339
1340	/*
1341	 * We do not want anybody to write to this logical eraseblock while we
1342	 * are moving it, so lock it.
1343	 *
1344	 * Note, we are using non-waiting locking here, because we cannot sleep
1345	 * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1346	 * unmapping the LEB which is mapped to the PEB we are going to move
1347	 * (@from). This task locks the LEB and goes sleep in the
1348	 * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1349	 * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1350	 * LEB is already locked, we just do not move it and return
1351	 * %MOVE_RETRY. Note, we do not return %MOVE_CANCEL_RACE here because
1352	 * we do not know the reasons of the contention - it may be just a
1353	 * normal I/O on this LEB, so we want to re-try.
1354	 */
1355	err = leb_write_trylock(ubi, vol_id, lnum);
1356	if (err) {
1357		dbg_wl("contention on LEB %d:%d, cancel", vol_id, lnum);
1358		return MOVE_RETRY;
1359	}
1360
1361	/*
1362	 * The LEB might have been put meanwhile, and the task which put it is
1363	 * probably waiting on @ubi->move_mutex. No need to continue the work,
1364	 * cancel it.
1365	 */
1366	if (vol->eba_tbl->entries[lnum].pnum != from) {
1367		dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to PEB %d, cancel",
1368		       vol_id, lnum, from, vol->eba_tbl->entries[lnum].pnum);
1369		err = MOVE_CANCEL_RACE;
1370		goto out_unlock_leb;
1371	}
1372
1373	/*
1374	 * OK, now the LEB is locked and we can safely start moving it. Since
1375	 * this function utilizes the @ubi->peb_buf buffer which is shared
1376	 * with some other functions - we lock the buffer by taking the
1377	 * @ubi->buf_mutex.
1378	 */
1379	mutex_lock(&ubi->buf_mutex);
1380	dbg_wl("read %d bytes of data", aldata_size);
1381	err = ubi_io_read_data(ubi, ubi->peb_buf, from, 0, aldata_size);
1382	if (err && err != UBI_IO_BITFLIPS) {
1383		ubi_warn(ubi, "error %d while reading data from PEB %d",
1384			 err, from);
1385		err = MOVE_SOURCE_RD_ERR;
1386		goto out_unlock_buf;
1387	}
1388
1389	/*
1390	 * Now we have got to calculate how much data we have to copy. In
1391	 * case of a static volume it is fairly easy - the VID header contains
1392	 * the data size. In case of a dynamic volume it is more difficult - we
1393	 * have to read the contents, cut 0xFF bytes from the end and copy only
1394	 * the first part. We must do this to avoid writing 0xFF bytes as it
1395	 * may have some side-effects. And not only this. It is important not
1396	 * to include those 0xFFs to CRC because later the they may be filled
1397	 * by data.
1398	 */
1399	if (vid_hdr->vol_type == UBI_VID_DYNAMIC)
1400		aldata_size = data_size =
1401			ubi_calc_data_len(ubi, ubi->peb_buf, data_size);
1402
1403	cond_resched();
1404	crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
1405	cond_resched();
1406
1407	/*
1408	 * It may turn out to be that the whole @from physical eraseblock
1409	 * contains only 0xFF bytes. Then we have to only write the VID header
1410	 * and do not write any data. This also means we should not set
1411	 * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1412	 */
1413	if (data_size > 0) {
1414		vid_hdr->copy_flag = 1;
1415		vid_hdr->data_size = cpu_to_be32(data_size);
1416		vid_hdr->data_crc = cpu_to_be32(crc);
1417	}
1418	vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1419
1420	err = ubi_io_write_vid_hdr(ubi, to, vidb);
1421	if (err) {
1422		if (err == -EIO)
1423			err = MOVE_TARGET_WR_ERR;
1424		goto out_unlock_buf;
1425	}
1426
1427	cond_resched();
1428
1429	/* Read the VID header back and check if it was written correctly */
1430	err = ubi_io_read_vid_hdr(ubi, to, vidb, 1);
1431	if (err) {
1432		if (err != UBI_IO_BITFLIPS) {
1433			ubi_warn(ubi, "error %d while reading VID header back from PEB %d",
1434				 err, to);
1435			if (is_error_sane(err))
1436				err = MOVE_TARGET_RD_ERR;
1437		} else
1438			err = MOVE_TARGET_BITFLIPS;
1439		goto out_unlock_buf;
1440	}
1441
1442	if (data_size > 0) {
1443		err = ubi_io_write_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1444		if (err) {
1445			if (err == -EIO)
1446				err = MOVE_TARGET_WR_ERR;
1447			goto out_unlock_buf;
1448		}
1449
1450		cond_resched();
1451	}
1452
1453	ubi_assert(vol->eba_tbl->entries[lnum].pnum == from);
1454	vol->eba_tbl->entries[lnum].pnum = to;
1455
1456out_unlock_buf:
1457	mutex_unlock(&ubi->buf_mutex);
1458out_unlock_leb:
1459	leb_write_unlock(ubi, vol_id, lnum);
1460	return err;
1461}
1462
1463/**
1464 * print_rsvd_warning - warn about not having enough reserved PEBs.
1465 * @ubi: UBI device description object
1466 *
1467 * This is a helper function for 'ubi_eba_init()' which is called when UBI
1468 * cannot reserve enough PEBs for bad block handling. This function makes a
1469 * decision whether we have to print a warning or not. The algorithm is as
1470 * follows:
1471 *   o if this is a new UBI image, then just print the warning
1472 *   o if this is an UBI image which has already been used for some time, print
1473 *     a warning only if we can reserve less than 10% of the expected amount of
1474 *     the reserved PEB.
1475 *
1476 * The idea is that when UBI is used, PEBs become bad, and the reserved pool
1477 * of PEBs becomes smaller, which is normal and we do not want to scare users
1478 * with a warning every time they attach the MTD device. This was an issue
1479 * reported by real users.
1480 */
1481static void print_rsvd_warning(struct ubi_device *ubi,
1482			       struct ubi_attach_info *ai)
1483{
1484	/*
1485	 * The 1 << 18 (256KiB) number is picked randomly, just a reasonably
1486	 * large number to distinguish between newly flashed and used images.
1487	 */
1488	if (ai->max_sqnum > (1 << 18)) {
1489		int min = ubi->beb_rsvd_level / 10;
1490
1491		if (!min)
1492			min = 1;
1493		if (ubi->beb_rsvd_pebs > min)
1494			return;
1495	}
1496
1497	ubi_warn(ubi, "cannot reserve enough PEBs for bad PEB handling, reserved %d, need %d",
1498		 ubi->beb_rsvd_pebs, ubi->beb_rsvd_level);
1499	if (ubi->corr_peb_count)
1500		ubi_warn(ubi, "%d PEBs are corrupted and not used",
1501			 ubi->corr_peb_count);
1502}
1503
1504/**
1505 * self_check_eba - run a self check on the EBA table constructed by fastmap.
1506 * @ubi: UBI device description object
1507 * @ai_fastmap: UBI attach info object created by fastmap
1508 * @ai_scan: UBI attach info object created by scanning
1509 *
1510 * Returns < 0 in case of an internal error, 0 otherwise.
1511 * If a bad EBA table entry was found it will be printed out and
1512 * ubi_assert() triggers.
1513 */
1514int self_check_eba(struct ubi_device *ubi, struct ubi_attach_info *ai_fastmap,
1515		   struct ubi_attach_info *ai_scan)
1516{
1517	int i, j, num_volumes, ret = 0;
1518	int **scan_eba, **fm_eba;
1519	struct ubi_ainf_volume *av;
1520	struct ubi_volume *vol;
1521	struct ubi_ainf_peb *aeb;
1522	struct rb_node *rb;
1523
1524	num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1525
1526	scan_eba = kmalloc_array(num_volumes, sizeof(*scan_eba), GFP_KERNEL);
1527	if (!scan_eba)
1528		return -ENOMEM;
1529
1530	fm_eba = kmalloc_array(num_volumes, sizeof(*fm_eba), GFP_KERNEL);
1531	if (!fm_eba) {
1532		kfree(scan_eba);
1533		return -ENOMEM;
1534	}
1535
1536	for (i = 0; i < num_volumes; i++) {
1537		vol = ubi->volumes[i];
1538		if (!vol)
1539			continue;
1540
1541		scan_eba[i] = kmalloc_array(vol->reserved_pebs,
1542					    sizeof(**scan_eba),
1543					    GFP_KERNEL);
1544		if (!scan_eba[i]) {
1545			ret = -ENOMEM;
1546			goto out_free;
1547		}
1548
1549		fm_eba[i] = kmalloc_array(vol->reserved_pebs,
1550					  sizeof(**fm_eba),
1551					  GFP_KERNEL);
1552		if (!fm_eba[i]) {
1553			ret = -ENOMEM;
1554			goto out_free;
1555		}
1556
1557		for (j = 0; j < vol->reserved_pebs; j++)
1558			scan_eba[i][j] = fm_eba[i][j] = UBI_LEB_UNMAPPED;
1559
1560		av = ubi_find_av(ai_scan, idx2vol_id(ubi, i));
1561		if (!av)
1562			continue;
1563
1564		ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1565			scan_eba[i][aeb->lnum] = aeb->pnum;
1566
1567		av = ubi_find_av(ai_fastmap, idx2vol_id(ubi, i));
1568		if (!av)
1569			continue;
1570
1571		ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb)
1572			fm_eba[i][aeb->lnum] = aeb->pnum;
1573
1574		for (j = 0; j < vol->reserved_pebs; j++) {
1575			if (scan_eba[i][j] != fm_eba[i][j]) {
1576				if (scan_eba[i][j] == UBI_LEB_UNMAPPED ||
1577					fm_eba[i][j] == UBI_LEB_UNMAPPED)
1578					continue;
1579
1580				ubi_err(ubi, "LEB:%i:%i is PEB:%i instead of %i!",
1581					vol->vol_id, j, fm_eba[i][j],
1582					scan_eba[i][j]);
1583				ubi_assert(0);
1584			}
1585		}
1586	}
1587
1588out_free:
1589	for (i = 0; i < num_volumes; i++) {
1590		if (!ubi->volumes[i])
1591			continue;
1592
1593		kfree(scan_eba[i]);
1594		kfree(fm_eba[i]);
1595	}
1596
1597	kfree(scan_eba);
1598	kfree(fm_eba);
1599	return ret;
1600}
1601
1602/**
1603 * ubi_eba_init - initialize the EBA sub-system using attaching information.
1604 * @ubi: UBI device description object
1605 * @ai: attaching information
1606 *
1607 * This function returns zero in case of success and a negative error code in
1608 * case of failure.
1609 */
1610int ubi_eba_init(struct ubi_device *ubi, struct ubi_attach_info *ai)
1611{
1612	int i, err, num_volumes;
1613	struct ubi_ainf_volume *av;
1614	struct ubi_volume *vol;
1615	struct ubi_ainf_peb *aeb;
1616	struct rb_node *rb;
1617
1618	dbg_eba("initialize EBA sub-system");
1619
1620	spin_lock_init(&ubi->ltree_lock);
1621	mutex_init(&ubi->alc_mutex);
1622	ubi->ltree = RB_ROOT;
1623
1624	ubi->global_sqnum = ai->max_sqnum + 1;
1625	num_volumes = ubi->vtbl_slots + UBI_INT_VOL_COUNT;
1626
1627	for (i = 0; i < num_volumes; i++) {
1628		struct ubi_eba_table *tbl;
1629
1630		vol = ubi->volumes[i];
1631		if (!vol)
1632			continue;
1633
1634		cond_resched();
1635
1636		tbl = ubi_eba_create_table(vol, vol->reserved_pebs);
1637		if (IS_ERR(tbl)) {
1638			err = PTR_ERR(tbl);
1639			goto out_free;
1640		}
1641
1642		ubi_eba_replace_table(vol, tbl);
1643
1644		av = ubi_find_av(ai, idx2vol_id(ubi, i));
1645		if (!av)
1646			continue;
1647
1648		ubi_rb_for_each_entry(rb, aeb, &av->root, u.rb) {
1649			if (aeb->lnum >= vol->reserved_pebs) {
1650				/*
1651				 * This may happen in case of an unclean reboot
1652				 * during re-size.
1653				 */
1654				ubi_move_aeb_to_list(av, aeb, &ai->erase);
1655			} else {
1656				struct ubi_eba_entry *entry;
1657
1658				entry = &vol->eba_tbl->entries[aeb->lnum];
1659				entry->pnum = aeb->pnum;
1660			}
1661		}
1662	}
1663
1664	if (ubi->avail_pebs < EBA_RESERVED_PEBS) {
1665		ubi_err(ubi, "no enough physical eraseblocks (%d, need %d)",
1666			ubi->avail_pebs, EBA_RESERVED_PEBS);
1667		if (ubi->corr_peb_count)
1668			ubi_err(ubi, "%d PEBs are corrupted and not used",
1669				ubi->corr_peb_count);
1670		err = -ENOSPC;
1671		goto out_free;
1672	}
1673	ubi->avail_pebs -= EBA_RESERVED_PEBS;
1674	ubi->rsvd_pebs += EBA_RESERVED_PEBS;
1675
1676	if (ubi->bad_allowed) {
1677		ubi_calculate_reserved(ubi);
1678
1679		if (ubi->avail_pebs < ubi->beb_rsvd_level) {
1680			/* No enough free physical eraseblocks */
1681			ubi->beb_rsvd_pebs = ubi->avail_pebs;
1682			print_rsvd_warning(ubi, ai);
1683		} else
1684			ubi->beb_rsvd_pebs = ubi->beb_rsvd_level;
1685
1686		ubi->avail_pebs -= ubi->beb_rsvd_pebs;
1687		ubi->rsvd_pebs  += ubi->beb_rsvd_pebs;
1688	}
1689
1690	dbg_eba("EBA sub-system is initialized");
1691	return 0;
1692
1693out_free:
1694	for (i = 0; i < num_volumes; i++) {
1695		if (!ubi->volumes[i])
1696			continue;
1697		ubi_eba_replace_table(ubi->volumes[i], NULL);
1698	}
1699	return err;
1700}