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