Back to home page

OSCL-LXR

 
 

    


0001 // SPDX-License-Identifier: GPL-2.0-or-later
0002 /*
0003  * Copyright (c) International Business Machines Corp., 2006
0004  *
0005  * Author: Artem Bityutskiy (Битюцкий Артём)
0006  */
0007 
0008 /*
0009  * The UBI Eraseblock Association (EBA) sub-system.
0010  *
0011  * This sub-system is responsible for I/O to/from logical eraseblock.
0012  *
0013  * Although in this implementation the EBA table is fully kept and managed in
0014  * RAM, which assumes poor scalability, it might be (partially) maintained on
0015  * flash in future implementations.
0016  *
0017  * The EBA sub-system implements per-logical eraseblock locking. Before
0018  * accessing a logical eraseblock it is locked for reading or writing. The
0019  * per-logical eraseblock locking is implemented by means of the lock tree. The
0020  * lock tree is an RB-tree which refers all the currently locked logical
0021  * eraseblocks. The lock tree elements are &struct ubi_ltree_entry objects.
0022  * They are indexed by (@vol_id, @lnum) pairs.
0023  *
0024  * EBA also maintains the global sequence counter which is incremented each
0025  * time a logical eraseblock is mapped to a physical eraseblock and it is
0026  * stored in the volume identifier header. This means that each VID header has
0027  * a unique sequence number. The sequence number is only increased an we assume
0028  * 64 bits is enough to never overflow.
0029  */
0030 
0031 #include <linux/slab.h>
0032 #include <linux/crc32.h>
0033 #include <linux/err.h>
0034 #include "ubi.h"
0035 
0036 /* Number of physical eraseblocks reserved for atomic LEB change operation */
0037 #define EBA_RESERVED_PEBS 1
0038 
0039 /**
0040  * struct ubi_eba_entry - structure encoding a single LEB -> PEB association
0041  * @pnum: the physical eraseblock number attached to the LEB
0042  *
0043  * This structure is encoding a LEB -> PEB association. Note that the LEB
0044  * number is not stored here, because it is the index used to access the
0045  * entries table.
0046  */
0047 struct ubi_eba_entry {
0048     int pnum;
0049 };
0050 
0051 /**
0052  * struct ubi_eba_table - LEB -> PEB association information
0053  * @entries: the LEB to PEB mapping (one entry per LEB).
0054  *
0055  * This structure is private to the EBA logic and should be kept here.
0056  * It is encoding the LEB to PEB association table, and is subject to
0057  * changes.
0058  */
0059 struct ubi_eba_table {
0060     struct ubi_eba_entry *entries;
0061 };
0062 
0063 /**
0064  * next_sqnum - get next sequence number.
0065  * @ubi: UBI device description object
0066  *
0067  * This function returns next sequence number to use, which is just the current
0068  * global sequence counter value. It also increases the global sequence
0069  * counter.
0070  */
0071 unsigned long long ubi_next_sqnum(struct ubi_device *ubi)
0072 {
0073     unsigned long long sqnum;
0074 
0075     spin_lock(&ubi->ltree_lock);
0076     sqnum = ubi->global_sqnum++;
0077     spin_unlock(&ubi->ltree_lock);
0078 
0079     return sqnum;
0080 }
0081 
0082 /**
0083  * ubi_get_compat - get compatibility flags of a volume.
0084  * @ubi: UBI device description object
0085  * @vol_id: volume ID
0086  *
0087  * This function returns compatibility flags for an internal volume. User
0088  * volumes have no compatibility flags, so %0 is returned.
0089  */
0090 static int ubi_get_compat(const struct ubi_device *ubi, int vol_id)
0091 {
0092     if (vol_id == UBI_LAYOUT_VOLUME_ID)
0093         return UBI_LAYOUT_VOLUME_COMPAT;
0094     return 0;
0095 }
0096 
0097 /**
0098  * ubi_eba_get_ldesc - get information about a LEB
0099  * @vol: volume description object
0100  * @lnum: logical eraseblock number
0101  * @ldesc: the LEB descriptor to fill
0102  *
0103  * Used to query information about a specific LEB.
0104  * It is currently only returning the physical position of the LEB, but will be
0105  * extended to provide more information.
0106  */
0107 void ubi_eba_get_ldesc(struct ubi_volume *vol, int lnum,
0108                struct ubi_eba_leb_desc *ldesc)
0109 {
0110     ldesc->lnum = lnum;
0111     ldesc->pnum = vol->eba_tbl->entries[lnum].pnum;
0112 }
0113 
0114 /**
0115  * ubi_eba_create_table - allocate a new EBA table and initialize it with all
0116  *            LEBs unmapped
0117  * @vol: volume containing the EBA table to copy
0118  * @nentries: number of entries in the table
0119  *
0120  * Allocate a new EBA table and initialize it with all LEBs unmapped.
0121  * Returns a valid pointer if it succeed, an ERR_PTR() otherwise.
0122  */
0123 struct ubi_eba_table *ubi_eba_create_table(struct ubi_volume *vol,
0124                        int nentries)
0125 {
0126     struct ubi_eba_table *tbl;
0127     int err = -ENOMEM;
0128     int i;
0129 
0130     tbl = kzalloc(sizeof(*tbl), GFP_KERNEL);
0131     if (!tbl)
0132         return ERR_PTR(-ENOMEM);
0133 
0134     tbl->entries = kmalloc_array(nentries, sizeof(*tbl->entries),
0135                      GFP_KERNEL);
0136     if (!tbl->entries)
0137         goto err;
0138 
0139     for (i = 0; i < nentries; i++)
0140         tbl->entries[i].pnum = UBI_LEB_UNMAPPED;
0141 
0142     return tbl;
0143 
0144 err:
0145     kfree(tbl);
0146 
0147     return ERR_PTR(err);
0148 }
0149 
0150 /**
0151  * ubi_eba_destroy_table - destroy an EBA table
0152  * @tbl: the table to destroy
0153  *
0154  * Destroy an EBA table.
0155  */
0156 void ubi_eba_destroy_table(struct ubi_eba_table *tbl)
0157 {
0158     if (!tbl)
0159         return;
0160 
0161     kfree(tbl->entries);
0162     kfree(tbl);
0163 }
0164 
0165 /**
0166  * ubi_eba_copy_table - copy the EBA table attached to vol into another table
0167  * @vol: volume containing the EBA table to copy
0168  * @dst: destination
0169  * @nentries: number of entries to copy
0170  *
0171  * Copy the EBA table stored in vol into the one pointed by dst.
0172  */
0173 void ubi_eba_copy_table(struct ubi_volume *vol, struct ubi_eba_table *dst,
0174             int nentries)
0175 {
0176     struct ubi_eba_table *src;
0177     int i;
0178 
0179     ubi_assert(dst && vol && vol->eba_tbl);
0180 
0181     src = vol->eba_tbl;
0182 
0183     for (i = 0; i < nentries; i++)
0184         dst->entries[i].pnum = src->entries[i].pnum;
0185 }
0186 
0187 /**
0188  * ubi_eba_replace_table - assign a new EBA table to a volume
0189  * @vol: volume containing the EBA table to copy
0190  * @tbl: new EBA table
0191  *
0192  * Assign a new EBA table to the volume and release the old one.
0193  */
0194 void ubi_eba_replace_table(struct ubi_volume *vol, struct ubi_eba_table *tbl)
0195 {
0196     ubi_eba_destroy_table(vol->eba_tbl);
0197     vol->eba_tbl = tbl;
0198 }
0199 
0200 /**
0201  * ltree_lookup - look up the lock tree.
0202  * @ubi: UBI device description object
0203  * @vol_id: volume ID
0204  * @lnum: logical eraseblock number
0205  *
0206  * This function returns a pointer to the corresponding &struct ubi_ltree_entry
0207  * object if the logical eraseblock is locked and %NULL if it is not.
0208  * @ubi->ltree_lock has to be locked.
0209  */
0210 static struct ubi_ltree_entry *ltree_lookup(struct ubi_device *ubi, int vol_id,
0211                         int lnum)
0212 {
0213     struct rb_node *p;
0214 
0215     p = ubi->ltree.rb_node;
0216     while (p) {
0217         struct ubi_ltree_entry *le;
0218 
0219         le = rb_entry(p, struct ubi_ltree_entry, rb);
0220 
0221         if (vol_id < le->vol_id)
0222             p = p->rb_left;
0223         else if (vol_id > le->vol_id)
0224             p = p->rb_right;
0225         else {
0226             if (lnum < le->lnum)
0227                 p = p->rb_left;
0228             else if (lnum > le->lnum)
0229                 p = p->rb_right;
0230             else
0231                 return le;
0232         }
0233     }
0234 
0235     return NULL;
0236 }
0237 
0238 /**
0239  * ltree_add_entry - add new entry to the lock tree.
0240  * @ubi: UBI device description object
0241  * @vol_id: volume ID
0242  * @lnum: logical eraseblock number
0243  *
0244  * This function adds new entry for logical eraseblock (@vol_id, @lnum) to the
0245  * lock tree. If such entry is already there, its usage counter is increased.
0246  * Returns pointer to the lock tree entry or %-ENOMEM if memory allocation
0247  * failed.
0248  */
0249 static struct ubi_ltree_entry *ltree_add_entry(struct ubi_device *ubi,
0250                            int vol_id, int lnum)
0251 {
0252     struct ubi_ltree_entry *le, *le1, *le_free;
0253 
0254     le = kmalloc(sizeof(struct ubi_ltree_entry), GFP_NOFS);
0255     if (!le)
0256         return ERR_PTR(-ENOMEM);
0257 
0258     le->users = 0;
0259     init_rwsem(&le->mutex);
0260     le->vol_id = vol_id;
0261     le->lnum = lnum;
0262 
0263     spin_lock(&ubi->ltree_lock);
0264     le1 = ltree_lookup(ubi, vol_id, lnum);
0265 
0266     if (le1) {
0267         /*
0268          * This logical eraseblock is already locked. The newly
0269          * allocated lock entry is not needed.
0270          */
0271         le_free = le;
0272         le = le1;
0273     } else {
0274         struct rb_node **p, *parent = NULL;
0275 
0276         /*
0277          * No lock entry, add the newly allocated one to the
0278          * @ubi->ltree RB-tree.
0279          */
0280         le_free = NULL;
0281 
0282         p = &ubi->ltree.rb_node;
0283         while (*p) {
0284             parent = *p;
0285             le1 = rb_entry(parent, struct ubi_ltree_entry, rb);
0286 
0287             if (vol_id < le1->vol_id)
0288                 p = &(*p)->rb_left;
0289             else if (vol_id > le1->vol_id)
0290                 p = &(*p)->rb_right;
0291             else {
0292                 ubi_assert(lnum != le1->lnum);
0293                 if (lnum < le1->lnum)
0294                     p = &(*p)->rb_left;
0295                 else
0296                     p = &(*p)->rb_right;
0297             }
0298         }
0299 
0300         rb_link_node(&le->rb, parent, p);
0301         rb_insert_color(&le->rb, &ubi->ltree);
0302     }
0303     le->users += 1;
0304     spin_unlock(&ubi->ltree_lock);
0305 
0306     kfree(le_free);
0307     return le;
0308 }
0309 
0310 /**
0311  * leb_read_lock - lock logical eraseblock for reading.
0312  * @ubi: UBI device description object
0313  * @vol_id: volume ID
0314  * @lnum: logical eraseblock number
0315  *
0316  * This function locks a logical eraseblock for reading. Returns zero in case
0317  * of success and a negative error code in case of failure.
0318  */
0319 static int leb_read_lock(struct ubi_device *ubi, int vol_id, int lnum)
0320 {
0321     struct ubi_ltree_entry *le;
0322 
0323     le = ltree_add_entry(ubi, vol_id, lnum);
0324     if (IS_ERR(le))
0325         return PTR_ERR(le);
0326     down_read(&le->mutex);
0327     return 0;
0328 }
0329 
0330 /**
0331  * leb_read_unlock - unlock logical eraseblock.
0332  * @ubi: UBI device description object
0333  * @vol_id: volume ID
0334  * @lnum: logical eraseblock number
0335  */
0336 static void leb_read_unlock(struct ubi_device *ubi, int vol_id, int lnum)
0337 {
0338     struct ubi_ltree_entry *le;
0339 
0340     spin_lock(&ubi->ltree_lock);
0341     le = ltree_lookup(ubi, vol_id, lnum);
0342     le->users -= 1;
0343     ubi_assert(le->users >= 0);
0344     up_read(&le->mutex);
0345     if (le->users == 0) {
0346         rb_erase(&le->rb, &ubi->ltree);
0347         kfree(le);
0348     }
0349     spin_unlock(&ubi->ltree_lock);
0350 }
0351 
0352 /**
0353  * leb_write_lock - lock logical eraseblock for writing.
0354  * @ubi: UBI device description object
0355  * @vol_id: volume ID
0356  * @lnum: logical eraseblock number
0357  *
0358  * This function locks a logical eraseblock for writing. Returns zero in case
0359  * of success and a negative error code in case of failure.
0360  */
0361 static int leb_write_lock(struct ubi_device *ubi, int vol_id, int lnum)
0362 {
0363     struct ubi_ltree_entry *le;
0364 
0365     le = ltree_add_entry(ubi, vol_id, lnum);
0366     if (IS_ERR(le))
0367         return PTR_ERR(le);
0368     down_write(&le->mutex);
0369     return 0;
0370 }
0371 
0372 /**
0373  * leb_write_trylock - try to lock logical eraseblock for writing.
0374  * @ubi: UBI device description object
0375  * @vol_id: volume ID
0376  * @lnum: logical eraseblock number
0377  *
0378  * This function locks a logical eraseblock for writing if there is no
0379  * contention and does nothing if there is contention. Returns %0 in case of
0380  * success, %1 in case of contention, and and a negative error code in case of
0381  * failure.
0382  */
0383 static int leb_write_trylock(struct ubi_device *ubi, int vol_id, int lnum)
0384 {
0385     struct ubi_ltree_entry *le;
0386 
0387     le = ltree_add_entry(ubi, vol_id, lnum);
0388     if (IS_ERR(le))
0389         return PTR_ERR(le);
0390     if (down_write_trylock(&le->mutex))
0391         return 0;
0392 
0393     /* Contention, cancel */
0394     spin_lock(&ubi->ltree_lock);
0395     le->users -= 1;
0396     ubi_assert(le->users >= 0);
0397     if (le->users == 0) {
0398         rb_erase(&le->rb, &ubi->ltree);
0399         kfree(le);
0400     }
0401     spin_unlock(&ubi->ltree_lock);
0402 
0403     return 1;
0404 }
0405 
0406 /**
0407  * leb_write_unlock - unlock logical eraseblock.
0408  * @ubi: UBI device description object
0409  * @vol_id: volume ID
0410  * @lnum: logical eraseblock number
0411  */
0412 static void leb_write_unlock(struct ubi_device *ubi, int vol_id, int lnum)
0413 {
0414     struct ubi_ltree_entry *le;
0415 
0416     spin_lock(&ubi->ltree_lock);
0417     le = ltree_lookup(ubi, vol_id, lnum);
0418     le->users -= 1;
0419     ubi_assert(le->users >= 0);
0420     up_write(&le->mutex);
0421     if (le->users == 0) {
0422         rb_erase(&le->rb, &ubi->ltree);
0423         kfree(le);
0424     }
0425     spin_unlock(&ubi->ltree_lock);
0426 }
0427 
0428 /**
0429  * ubi_eba_is_mapped - check if a LEB is mapped.
0430  * @vol: volume description object
0431  * @lnum: logical eraseblock number
0432  *
0433  * This function returns true if the LEB is mapped, false otherwise.
0434  */
0435 bool ubi_eba_is_mapped(struct ubi_volume *vol, int lnum)
0436 {
0437     return vol->eba_tbl->entries[lnum].pnum >= 0;
0438 }
0439 
0440 /**
0441  * ubi_eba_unmap_leb - un-map logical eraseblock.
0442  * @ubi: UBI device description object
0443  * @vol: volume description object
0444  * @lnum: logical eraseblock number
0445  *
0446  * This function un-maps logical eraseblock @lnum and schedules corresponding
0447  * physical eraseblock for erasure. Returns zero in case of success and a
0448  * negative error code in case of failure.
0449  */
0450 int ubi_eba_unmap_leb(struct ubi_device *ubi, struct ubi_volume *vol,
0451               int lnum)
0452 {
0453     int err, pnum, vol_id = vol->vol_id;
0454 
0455     if (ubi->ro_mode)
0456         return -EROFS;
0457 
0458     err = leb_write_lock(ubi, vol_id, lnum);
0459     if (err)
0460         return err;
0461 
0462     pnum = vol->eba_tbl->entries[lnum].pnum;
0463     if (pnum < 0)
0464         /* This logical eraseblock is already unmapped */
0465         goto out_unlock;
0466 
0467     dbg_eba("erase LEB %d:%d, PEB %d", vol_id, lnum, pnum);
0468 
0469     down_read(&ubi->fm_eba_sem);
0470     vol->eba_tbl->entries[lnum].pnum = UBI_LEB_UNMAPPED;
0471     up_read(&ubi->fm_eba_sem);
0472     err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 0);
0473 
0474 out_unlock:
0475     leb_write_unlock(ubi, vol_id, lnum);
0476     return err;
0477 }
0478 
0479 #ifdef CONFIG_MTD_UBI_FASTMAP
0480 /**
0481  * check_mapping - check and fixup a mapping
0482  * @ubi: UBI device description object
0483  * @vol: volume description object
0484  * @lnum: logical eraseblock number
0485  * @pnum: physical eraseblock number
0486  *
0487  * Checks whether a given mapping is valid. Fastmap cannot track LEB unmap
0488  * operations, if such an operation is interrupted the mapping still looks
0489  * good, but upon first read an ECC is reported to the upper layer.
0490  * Normaly during the full-scan at attach time this is fixed, for Fastmap
0491  * we have to deal with it while reading.
0492  * If the PEB behind a LEB shows this symthom we change the mapping to
0493  * %UBI_LEB_UNMAPPED and schedule the PEB for erasure.
0494  *
0495  * Returns 0 on success, negative error code in case of failure.
0496  */
0497 static int check_mapping(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
0498              int *pnum)
0499 {
0500     int err;
0501     struct ubi_vid_io_buf *vidb;
0502     struct ubi_vid_hdr *vid_hdr;
0503 
0504     if (!ubi->fast_attach)
0505         return 0;
0506 
0507     if (!vol->checkmap || test_bit(lnum, vol->checkmap))
0508         return 0;
0509 
0510     vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
0511     if (!vidb)
0512         return -ENOMEM;
0513 
0514     err = ubi_io_read_vid_hdr(ubi, *pnum, vidb, 0);
0515     if (err > 0 && err != UBI_IO_BITFLIPS) {
0516         int torture = 0;
0517 
0518         switch (err) {
0519             case UBI_IO_FF:
0520             case UBI_IO_FF_BITFLIPS:
0521             case UBI_IO_BAD_HDR:
0522             case UBI_IO_BAD_HDR_EBADMSG:
0523                 break;
0524             default:
0525                 ubi_assert(0);
0526         }
0527 
0528         if (err == UBI_IO_BAD_HDR_EBADMSG || err == UBI_IO_FF_BITFLIPS)
0529             torture = 1;
0530 
0531         down_read(&ubi->fm_eba_sem);
0532         vol->eba_tbl->entries[lnum].pnum = UBI_LEB_UNMAPPED;
0533         up_read(&ubi->fm_eba_sem);
0534         ubi_wl_put_peb(ubi, vol->vol_id, lnum, *pnum, torture);
0535 
0536         *pnum = UBI_LEB_UNMAPPED;
0537     } else if (err < 0) {
0538         ubi_err(ubi, "unable to read VID header back from PEB %i: %i",
0539             *pnum, err);
0540 
0541         goto out_free;
0542     } else {
0543         int found_vol_id, found_lnum;
0544 
0545         ubi_assert(err == 0 || err == UBI_IO_BITFLIPS);
0546 
0547         vid_hdr = ubi_get_vid_hdr(vidb);
0548         found_vol_id = be32_to_cpu(vid_hdr->vol_id);
0549         found_lnum = be32_to_cpu(vid_hdr->lnum);
0550 
0551         if (found_lnum != lnum || found_vol_id != vol->vol_id) {
0552             ubi_err(ubi, "EBA mismatch! PEB %i is LEB %i:%i instead of LEB %i:%i",
0553                 *pnum, found_vol_id, found_lnum, vol->vol_id, lnum);
0554             ubi_ro_mode(ubi);
0555             err = -EINVAL;
0556             goto out_free;
0557         }
0558     }
0559 
0560     set_bit(lnum, vol->checkmap);
0561     err = 0;
0562 
0563 out_free:
0564     ubi_free_vid_buf(vidb);
0565 
0566     return err;
0567 }
0568 #else
0569 static int check_mapping(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
0570           int *pnum)
0571 {
0572     return 0;
0573 }
0574 #endif
0575 
0576 /**
0577  * ubi_eba_read_leb - read data.
0578  * @ubi: UBI device description object
0579  * @vol: volume description object
0580  * @lnum: logical eraseblock number
0581  * @buf: buffer to store the read data
0582  * @offset: offset from where to read
0583  * @len: how many bytes to read
0584  * @check: data CRC check flag
0585  *
0586  * If the logical eraseblock @lnum is unmapped, @buf is filled with 0xFF
0587  * bytes. The @check flag only makes sense for static volumes and forces
0588  * eraseblock data CRC checking.
0589  *
0590  * In case of success this function returns zero. In case of a static volume,
0591  * if data CRC mismatches - %-EBADMSG is returned. %-EBADMSG may also be
0592  * returned for any volume type if an ECC error was detected by the MTD device
0593  * driver. Other negative error cored may be returned in case of other errors.
0594  */
0595 int ubi_eba_read_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
0596              void *buf, int offset, int len, int check)
0597 {
0598     int err, pnum, scrub = 0, vol_id = vol->vol_id;
0599     struct ubi_vid_io_buf *vidb;
0600     struct ubi_vid_hdr *vid_hdr;
0601     uint32_t crc;
0602 
0603     err = leb_read_lock(ubi, vol_id, lnum);
0604     if (err)
0605         return err;
0606 
0607     pnum = vol->eba_tbl->entries[lnum].pnum;
0608     if (pnum >= 0) {
0609         err = check_mapping(ubi, vol, lnum, &pnum);
0610         if (err < 0)
0611             goto out_unlock;
0612     }
0613 
0614     if (pnum == UBI_LEB_UNMAPPED) {
0615         /*
0616          * The logical eraseblock is not mapped, fill the whole buffer
0617          * with 0xFF bytes. The exception is static volumes for which
0618          * it is an error to read unmapped logical eraseblocks.
0619          */
0620         dbg_eba("read %d bytes from offset %d of LEB %d:%d (unmapped)",
0621             len, offset, vol_id, lnum);
0622         leb_read_unlock(ubi, vol_id, lnum);
0623         ubi_assert(vol->vol_type != UBI_STATIC_VOLUME);
0624         memset(buf, 0xFF, len);
0625         return 0;
0626     }
0627 
0628     dbg_eba("read %d bytes from offset %d of LEB %d:%d, PEB %d",
0629         len, offset, vol_id, lnum, pnum);
0630 
0631     if (vol->vol_type == UBI_DYNAMIC_VOLUME)
0632         check = 0;
0633 
0634 retry:
0635     if (check) {
0636         vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
0637         if (!vidb) {
0638             err = -ENOMEM;
0639             goto out_unlock;
0640         }
0641 
0642         vid_hdr = ubi_get_vid_hdr(vidb);
0643 
0644         err = ubi_io_read_vid_hdr(ubi, pnum, vidb, 1);
0645         if (err && err != UBI_IO_BITFLIPS) {
0646             if (err > 0) {
0647                 /*
0648                  * The header is either absent or corrupted.
0649                  * The former case means there is a bug -
0650                  * switch to read-only mode just in case.
0651                  * The latter case means a real corruption - we
0652                  * may try to recover data. FIXME: but this is
0653                  * not implemented.
0654                  */
0655                 if (err == UBI_IO_BAD_HDR_EBADMSG ||
0656                     err == UBI_IO_BAD_HDR) {
0657                     ubi_warn(ubi, "corrupted VID header at PEB %d, LEB %d:%d",
0658                          pnum, vol_id, lnum);
0659                     err = -EBADMSG;
0660                 } else {
0661                     /*
0662                      * Ending up here in the non-Fastmap case
0663                      * is a clear bug as the VID header had to
0664                      * be present at scan time to have it referenced.
0665                      * With fastmap the story is more complicated.
0666                      * Fastmap has the mapping info without the need
0667                      * of a full scan. So the LEB could have been
0668                      * unmapped, Fastmap cannot know this and keeps
0669                      * the LEB referenced.
0670                      * This is valid and works as the layer above UBI
0671                      * has to do bookkeeping about used/referenced
0672                      * LEBs in any case.
0673                      */
0674                     if (ubi->fast_attach) {
0675                         err = -EBADMSG;
0676                     } else {
0677                         err = -EINVAL;
0678                         ubi_ro_mode(ubi);
0679                     }
0680                 }
0681             }
0682             goto out_free;
0683         } else if (err == UBI_IO_BITFLIPS)
0684             scrub = 1;
0685 
0686         ubi_assert(lnum < be32_to_cpu(vid_hdr->used_ebs));
0687         ubi_assert(len == be32_to_cpu(vid_hdr->data_size));
0688 
0689         crc = be32_to_cpu(vid_hdr->data_crc);
0690         ubi_free_vid_buf(vidb);
0691     }
0692 
0693     err = ubi_io_read_data(ubi, buf, pnum, offset, len);
0694     if (err) {
0695         if (err == UBI_IO_BITFLIPS)
0696             scrub = 1;
0697         else if (mtd_is_eccerr(err)) {
0698             if (vol->vol_type == UBI_DYNAMIC_VOLUME)
0699                 goto out_unlock;
0700             scrub = 1;
0701             if (!check) {
0702                 ubi_msg(ubi, "force data checking");
0703                 check = 1;
0704                 goto retry;
0705             }
0706         } else
0707             goto out_unlock;
0708     }
0709 
0710     if (check) {
0711         uint32_t crc1 = crc32(UBI_CRC32_INIT, buf, len);
0712         if (crc1 != crc) {
0713             ubi_warn(ubi, "CRC error: calculated %#08x, must be %#08x",
0714                  crc1, crc);
0715             err = -EBADMSG;
0716             goto out_unlock;
0717         }
0718     }
0719 
0720     if (scrub)
0721         err = ubi_wl_scrub_peb(ubi, pnum);
0722 
0723     leb_read_unlock(ubi, vol_id, lnum);
0724     return err;
0725 
0726 out_free:
0727     ubi_free_vid_buf(vidb);
0728 out_unlock:
0729     leb_read_unlock(ubi, vol_id, lnum);
0730     return err;
0731 }
0732 
0733 /**
0734  * ubi_eba_read_leb_sg - read data into a scatter gather list.
0735  * @ubi: UBI device description object
0736  * @vol: volume description object
0737  * @lnum: logical eraseblock number
0738  * @sgl: UBI scatter gather list to store the read data
0739  * @offset: offset from where to read
0740  * @len: how many bytes to read
0741  * @check: data CRC check flag
0742  *
0743  * This function works exactly like ubi_eba_read_leb(). But instead of
0744  * storing the read data into a buffer it writes to an UBI scatter gather
0745  * list.
0746  */
0747 int ubi_eba_read_leb_sg(struct ubi_device *ubi, struct ubi_volume *vol,
0748             struct ubi_sgl *sgl, int lnum, int offset, int len,
0749             int check)
0750 {
0751     int to_read;
0752     int ret;
0753     struct scatterlist *sg;
0754 
0755     for (;;) {
0756         ubi_assert(sgl->list_pos < UBI_MAX_SG_COUNT);
0757         sg = &sgl->sg[sgl->list_pos];
0758         if (len < sg->length - sgl->page_pos)
0759             to_read = len;
0760         else
0761             to_read = sg->length - sgl->page_pos;
0762 
0763         ret = ubi_eba_read_leb(ubi, vol, lnum,
0764                        sg_virt(sg) + sgl->page_pos, offset,
0765                        to_read, check);
0766         if (ret < 0)
0767             return ret;
0768 
0769         offset += to_read;
0770         len -= to_read;
0771         if (!len) {
0772             sgl->page_pos += to_read;
0773             if (sgl->page_pos == sg->length) {
0774                 sgl->list_pos++;
0775                 sgl->page_pos = 0;
0776             }
0777 
0778             break;
0779         }
0780 
0781         sgl->list_pos++;
0782         sgl->page_pos = 0;
0783     }
0784 
0785     return ret;
0786 }
0787 
0788 /**
0789  * try_recover_peb - try to recover from write failure.
0790  * @vol: volume description object
0791  * @pnum: the physical eraseblock to recover
0792  * @lnum: logical eraseblock number
0793  * @buf: data which was not written because of the write failure
0794  * @offset: offset of the failed write
0795  * @len: how many bytes should have been written
0796  * @vidb: VID buffer
0797  * @retry: whether the caller should retry in case of failure
0798  *
0799  * This function is called in case of a write failure and moves all good data
0800  * from the potentially bad physical eraseblock to a good physical eraseblock.
0801  * This function also writes the data which was not written due to the failure.
0802  * Returns 0 in case of success, and a negative error code in case of failure.
0803  * In case of failure, the %retry parameter is set to false if this is a fatal
0804  * error (retrying won't help), and true otherwise.
0805  */
0806 static int try_recover_peb(struct ubi_volume *vol, int pnum, int lnum,
0807                const void *buf, int offset, int len,
0808                struct ubi_vid_io_buf *vidb, bool *retry)
0809 {
0810     struct ubi_device *ubi = vol->ubi;
0811     struct ubi_vid_hdr *vid_hdr;
0812     int new_pnum, err, vol_id = vol->vol_id, data_size;
0813     uint32_t crc;
0814 
0815     *retry = false;
0816 
0817     new_pnum = ubi_wl_get_peb(ubi);
0818     if (new_pnum < 0) {
0819         err = new_pnum;
0820         goto out_put;
0821     }
0822 
0823     ubi_msg(ubi, "recover PEB %d, move data to PEB %d",
0824         pnum, new_pnum);
0825 
0826     err = ubi_io_read_vid_hdr(ubi, pnum, vidb, 1);
0827     if (err && err != UBI_IO_BITFLIPS) {
0828         if (err > 0)
0829             err = -EIO;
0830         goto out_put;
0831     }
0832 
0833     vid_hdr = ubi_get_vid_hdr(vidb);
0834     ubi_assert(vid_hdr->vol_type == UBI_VID_DYNAMIC);
0835 
0836     mutex_lock(&ubi->buf_mutex);
0837     memset(ubi->peb_buf + offset, 0xFF, len);
0838 
0839     /* Read everything before the area where the write failure happened */
0840     if (offset > 0) {
0841         err = ubi_io_read_data(ubi, ubi->peb_buf, pnum, 0, offset);
0842         if (err && err != UBI_IO_BITFLIPS)
0843             goto out_unlock;
0844     }
0845 
0846     *retry = true;
0847 
0848     memcpy(ubi->peb_buf + offset, buf, len);
0849 
0850     data_size = offset + len;
0851     crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
0852     vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
0853     vid_hdr->copy_flag = 1;
0854     vid_hdr->data_size = cpu_to_be32(data_size);
0855     vid_hdr->data_crc = cpu_to_be32(crc);
0856     err = ubi_io_write_vid_hdr(ubi, new_pnum, vidb);
0857     if (err)
0858         goto out_unlock;
0859 
0860     err = ubi_io_write_data(ubi, ubi->peb_buf, new_pnum, 0, data_size);
0861 
0862 out_unlock:
0863     mutex_unlock(&ubi->buf_mutex);
0864 
0865     if (!err)
0866         vol->eba_tbl->entries[lnum].pnum = new_pnum;
0867 
0868 out_put:
0869     up_read(&ubi->fm_eba_sem);
0870 
0871     if (!err) {
0872         ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
0873         ubi_msg(ubi, "data was successfully recovered");
0874     } else if (new_pnum >= 0) {
0875         /*
0876          * Bad luck? This physical eraseblock is bad too? Crud. Let's
0877          * try to get another one.
0878          */
0879         ubi_wl_put_peb(ubi, vol_id, lnum, new_pnum, 1);
0880         ubi_warn(ubi, "failed to write to PEB %d", new_pnum);
0881     }
0882 
0883     return err;
0884 }
0885 
0886 /**
0887  * recover_peb - recover from write failure.
0888  * @ubi: UBI device description object
0889  * @pnum: the physical eraseblock to recover
0890  * @vol_id: volume ID
0891  * @lnum: logical eraseblock number
0892  * @buf: data which was not written because of the write failure
0893  * @offset: offset of the failed write
0894  * @len: how many bytes should have been written
0895  *
0896  * This function is called in case of a write failure and moves all good data
0897  * from the potentially bad physical eraseblock to a good physical eraseblock.
0898  * This function also writes the data which was not written due to the failure.
0899  * Returns 0 in case of success, and a negative error code in case of failure.
0900  * This function tries %UBI_IO_RETRIES before giving up.
0901  */
0902 static int recover_peb(struct ubi_device *ubi, int pnum, int vol_id, int lnum,
0903                const void *buf, int offset, int len)
0904 {
0905     int err, idx = vol_id2idx(ubi, vol_id), tries;
0906     struct ubi_volume *vol = ubi->volumes[idx];
0907     struct ubi_vid_io_buf *vidb;
0908 
0909     vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
0910     if (!vidb)
0911         return -ENOMEM;
0912 
0913     for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
0914         bool retry;
0915 
0916         err = try_recover_peb(vol, pnum, lnum, buf, offset, len, vidb,
0917                       &retry);
0918         if (!err || !retry)
0919             break;
0920 
0921         ubi_msg(ubi, "try again");
0922     }
0923 
0924     ubi_free_vid_buf(vidb);
0925 
0926     return err;
0927 }
0928 
0929 /**
0930  * try_write_vid_and_data - try to write VID header and data to a new PEB.
0931  * @vol: volume description object
0932  * @lnum: logical eraseblock number
0933  * @vidb: the VID buffer to write
0934  * @buf: buffer containing the data
0935  * @offset: where to start writing data
0936  * @len: how many bytes should be written
0937  *
0938  * This function tries to write VID header and data belonging to logical
0939  * eraseblock @lnum of volume @vol to a new physical eraseblock. Returns zero
0940  * in case of success and a negative error code in case of failure.
0941  * In case of error, it is possible that something was still written to the
0942  * flash media, but may be some garbage.
0943  */
0944 static int try_write_vid_and_data(struct ubi_volume *vol, int lnum,
0945                   struct ubi_vid_io_buf *vidb, const void *buf,
0946                   int offset, int len)
0947 {
0948     struct ubi_device *ubi = vol->ubi;
0949     int pnum, opnum, err, vol_id = vol->vol_id;
0950 
0951     pnum = ubi_wl_get_peb(ubi);
0952     if (pnum < 0) {
0953         err = pnum;
0954         goto out_put;
0955     }
0956 
0957     opnum = vol->eba_tbl->entries[lnum].pnum;
0958 
0959     dbg_eba("write VID hdr and %d bytes at offset %d of LEB %d:%d, PEB %d",
0960         len, offset, vol_id, lnum, pnum);
0961 
0962     err = ubi_io_write_vid_hdr(ubi, pnum, vidb);
0963     if (err) {
0964         ubi_warn(ubi, "failed to write VID header to LEB %d:%d, PEB %d",
0965              vol_id, lnum, pnum);
0966         goto out_put;
0967     }
0968 
0969     if (len) {
0970         err = ubi_io_write_data(ubi, buf, pnum, offset, len);
0971         if (err) {
0972             ubi_warn(ubi,
0973                  "failed to write %d bytes at offset %d of LEB %d:%d, PEB %d",
0974                  len, offset, vol_id, lnum, pnum);
0975             goto out_put;
0976         }
0977     }
0978 
0979     vol->eba_tbl->entries[lnum].pnum = pnum;
0980 
0981 out_put:
0982     up_read(&ubi->fm_eba_sem);
0983 
0984     if (err && pnum >= 0)
0985         err = ubi_wl_put_peb(ubi, vol_id, lnum, pnum, 1);
0986     else if (!err && opnum >= 0)
0987         err = ubi_wl_put_peb(ubi, vol_id, lnum, opnum, 0);
0988 
0989     return err;
0990 }
0991 
0992 /**
0993  * ubi_eba_write_leb - write data to dynamic volume.
0994  * @ubi: UBI device description object
0995  * @vol: volume description object
0996  * @lnum: logical eraseblock number
0997  * @buf: the data to write
0998  * @offset: offset within the logical eraseblock where to write
0999  * @len: how many bytes to write
1000  *
1001  * This function writes data to logical eraseblock @lnum of a dynamic volume
1002  * @vol. Returns zero in case of success and a negative error code in case
1003  * of failure. In case of error, it is possible that something was still
1004  * written to the flash media, but may be some garbage.
1005  * This function retries %UBI_IO_RETRIES times before giving up.
1006  */
1007 int ubi_eba_write_leb(struct ubi_device *ubi, struct ubi_volume *vol, int lnum,
1008               const void *buf, int offset, int len)
1009 {
1010     int err, pnum, tries, vol_id = vol->vol_id;
1011     struct ubi_vid_io_buf *vidb;
1012     struct ubi_vid_hdr *vid_hdr;
1013 
1014     if (ubi->ro_mode)
1015         return -EROFS;
1016 
1017     err = leb_write_lock(ubi, vol_id, lnum);
1018     if (err)
1019         return err;
1020 
1021     pnum = vol->eba_tbl->entries[lnum].pnum;
1022     if (pnum >= 0) {
1023         err = check_mapping(ubi, vol, lnum, &pnum);
1024         if (err < 0)
1025             goto out;
1026     }
1027 
1028     if (pnum >= 0) {
1029         dbg_eba("write %d bytes at offset %d of LEB %d:%d, PEB %d",
1030             len, offset, vol_id, lnum, pnum);
1031 
1032         err = ubi_io_write_data(ubi, buf, pnum, offset, len);
1033         if (err) {
1034             ubi_warn(ubi, "failed to write data to PEB %d", pnum);
1035             if (err == -EIO && ubi->bad_allowed)
1036                 err = recover_peb(ubi, pnum, vol_id, lnum, buf,
1037                           offset, len);
1038         }
1039 
1040         goto out;
1041     }
1042 
1043     /*
1044      * The logical eraseblock is not mapped. We have to get a free physical
1045      * eraseblock and write the volume identifier header there first.
1046      */
1047     vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
1048     if (!vidb) {
1049         leb_write_unlock(ubi, vol_id, lnum);
1050         return -ENOMEM;
1051     }
1052 
1053     vid_hdr = ubi_get_vid_hdr(vidb);
1054 
1055     vid_hdr->vol_type = UBI_VID_DYNAMIC;
1056     vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1057     vid_hdr->vol_id = cpu_to_be32(vol_id);
1058     vid_hdr->lnum = cpu_to_be32(lnum);
1059     vid_hdr->compat = ubi_get_compat(ubi, vol_id);
1060     vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
1061 
1062     for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
1063         err = try_write_vid_and_data(vol, lnum, vidb, buf, offset, len);
1064         if (err != -EIO || !ubi->bad_allowed)
1065             break;
1066 
1067         /*
1068          * Fortunately, this is the first write operation to this
1069          * physical eraseblock, so just put it and request a new one.
1070          * We assume that if this physical eraseblock went bad, the
1071          * erase code will handle that.
1072          */
1073         vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1074         ubi_msg(ubi, "try another PEB");
1075     }
1076 
1077     ubi_free_vid_buf(vidb);
1078 
1079 out:
1080     if (err)
1081         ubi_ro_mode(ubi);
1082 
1083     leb_write_unlock(ubi, vol_id, lnum);
1084 
1085     return err;
1086 }
1087 
1088 /**
1089  * ubi_eba_write_leb_st - write data to static volume.
1090  * @ubi: UBI device description object
1091  * @vol: volume description object
1092  * @lnum: logical eraseblock number
1093  * @buf: data to write
1094  * @len: how many bytes to write
1095  * @used_ebs: how many logical eraseblocks will this volume contain
1096  *
1097  * This function writes data to logical eraseblock @lnum of static volume
1098  * @vol. The @used_ebs argument should contain total number of logical
1099  * eraseblock in this static volume.
1100  *
1101  * When writing to the last logical eraseblock, the @len argument doesn't have
1102  * to be aligned to the minimal I/O unit size. Instead, it has to be equivalent
1103  * to the real data size, although the @buf buffer has to contain the
1104  * alignment. In all other cases, @len has to be aligned.
1105  *
1106  * It is prohibited to write more than once to logical eraseblocks of static
1107  * volumes. This function returns zero in case of success and a negative error
1108  * code in case of failure.
1109  */
1110 int ubi_eba_write_leb_st(struct ubi_device *ubi, struct ubi_volume *vol,
1111              int lnum, const void *buf, int len, int used_ebs)
1112 {
1113     int err, tries, data_size = len, vol_id = vol->vol_id;
1114     struct ubi_vid_io_buf *vidb;
1115     struct ubi_vid_hdr *vid_hdr;
1116     uint32_t crc;
1117 
1118     if (ubi->ro_mode)
1119         return -EROFS;
1120 
1121     if (lnum == used_ebs - 1)
1122         /* If this is the last LEB @len may be unaligned */
1123         len = ALIGN(data_size, ubi->min_io_size);
1124     else
1125         ubi_assert(!(len & (ubi->min_io_size - 1)));
1126 
1127     vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
1128     if (!vidb)
1129         return -ENOMEM;
1130 
1131     vid_hdr = ubi_get_vid_hdr(vidb);
1132 
1133     err = leb_write_lock(ubi, vol_id, lnum);
1134     if (err)
1135         goto out;
1136 
1137     vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1138     vid_hdr->vol_id = cpu_to_be32(vol_id);
1139     vid_hdr->lnum = cpu_to_be32(lnum);
1140     vid_hdr->compat = ubi_get_compat(ubi, vol_id);
1141     vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
1142 
1143     crc = crc32(UBI_CRC32_INIT, buf, data_size);
1144     vid_hdr->vol_type = UBI_VID_STATIC;
1145     vid_hdr->data_size = cpu_to_be32(data_size);
1146     vid_hdr->used_ebs = cpu_to_be32(used_ebs);
1147     vid_hdr->data_crc = cpu_to_be32(crc);
1148 
1149     ubi_assert(vol->eba_tbl->entries[lnum].pnum < 0);
1150 
1151     for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
1152         err = try_write_vid_and_data(vol, lnum, vidb, buf, 0, len);
1153         if (err != -EIO || !ubi->bad_allowed)
1154             break;
1155 
1156         vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1157         ubi_msg(ubi, "try another PEB");
1158     }
1159 
1160     if (err)
1161         ubi_ro_mode(ubi);
1162 
1163     leb_write_unlock(ubi, vol_id, lnum);
1164 
1165 out:
1166     ubi_free_vid_buf(vidb);
1167 
1168     return err;
1169 }
1170 
1171 /*
1172  * ubi_eba_atomic_leb_change - change logical eraseblock atomically.
1173  * @ubi: UBI device description object
1174  * @vol: volume description object
1175  * @lnum: logical eraseblock number
1176  * @buf: data to write
1177  * @len: how many bytes to write
1178  *
1179  * This function changes the contents of a logical eraseblock atomically. @buf
1180  * has to contain new logical eraseblock data, and @len - the length of the
1181  * data, which has to be aligned. This function guarantees that in case of an
1182  * unclean reboot the old contents is preserved. Returns zero in case of
1183  * success and a negative error code in case of failure.
1184  *
1185  * UBI reserves one LEB for the "atomic LEB change" operation, so only one
1186  * LEB change may be done at a time. This is ensured by @ubi->alc_mutex.
1187  */
1188 int ubi_eba_atomic_leb_change(struct ubi_device *ubi, struct ubi_volume *vol,
1189                   int lnum, const void *buf, int len)
1190 {
1191     int err, tries, vol_id = vol->vol_id;
1192     struct ubi_vid_io_buf *vidb;
1193     struct ubi_vid_hdr *vid_hdr;
1194     uint32_t crc;
1195 
1196     if (ubi->ro_mode)
1197         return -EROFS;
1198 
1199     if (len == 0) {
1200         /*
1201          * Special case when data length is zero. In this case the LEB
1202          * has to be unmapped and mapped somewhere else.
1203          */
1204         err = ubi_eba_unmap_leb(ubi, vol, lnum);
1205         if (err)
1206             return err;
1207         return ubi_eba_write_leb(ubi, vol, lnum, NULL, 0, 0);
1208     }
1209 
1210     vidb = ubi_alloc_vid_buf(ubi, GFP_NOFS);
1211     if (!vidb)
1212         return -ENOMEM;
1213 
1214     vid_hdr = ubi_get_vid_hdr(vidb);
1215 
1216     mutex_lock(&ubi->alc_mutex);
1217     err = leb_write_lock(ubi, vol_id, lnum);
1218     if (err)
1219         goto out_mutex;
1220 
1221     vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1222     vid_hdr->vol_id = cpu_to_be32(vol_id);
1223     vid_hdr->lnum = cpu_to_be32(lnum);
1224     vid_hdr->compat = ubi_get_compat(ubi, vol_id);
1225     vid_hdr->data_pad = cpu_to_be32(vol->data_pad);
1226 
1227     crc = crc32(UBI_CRC32_INIT, buf, len);
1228     vid_hdr->vol_type = UBI_VID_DYNAMIC;
1229     vid_hdr->data_size = cpu_to_be32(len);
1230     vid_hdr->copy_flag = 1;
1231     vid_hdr->data_crc = cpu_to_be32(crc);
1232 
1233     dbg_eba("change LEB %d:%d", vol_id, lnum);
1234 
1235     for (tries = 0; tries <= UBI_IO_RETRIES; tries++) {
1236         err = try_write_vid_and_data(vol, lnum, vidb, buf, 0, len);
1237         if (err != -EIO || !ubi->bad_allowed)
1238             break;
1239 
1240         vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1241         ubi_msg(ubi, "try another PEB");
1242     }
1243 
1244     /*
1245      * This flash device does not admit of bad eraseblocks or
1246      * something nasty and unexpected happened. Switch to read-only
1247      * mode just in case.
1248      */
1249     if (err)
1250         ubi_ro_mode(ubi);
1251 
1252     leb_write_unlock(ubi, vol_id, lnum);
1253 
1254 out_mutex:
1255     mutex_unlock(&ubi->alc_mutex);
1256     ubi_free_vid_buf(vidb);
1257     return err;
1258 }
1259 
1260 /**
1261  * is_error_sane - check whether a read error is sane.
1262  * @err: code of the error happened during reading
1263  *
1264  * This is a helper function for 'ubi_eba_copy_leb()' which is called when we
1265  * cannot read data from the target PEB (an error @err happened). If the error
1266  * code is sane, then we treat this error as non-fatal. Otherwise the error is
1267  * fatal and UBI will be switched to R/O mode later.
1268  *
1269  * The idea is that we try not to switch to R/O mode if the read error is
1270  * something which suggests there was a real read problem. E.g., %-EIO. Or a
1271  * memory allocation failed (-%ENOMEM). Otherwise, it is safer to switch to R/O
1272  * mode, simply because we do not know what happened at the MTD level, and we
1273  * cannot handle this. E.g., the underlying driver may have become crazy, and
1274  * it is safer to switch to R/O mode to preserve the data.
1275  *
1276  * And bear in mind, this is about reading from the target PEB, i.e. the PEB
1277  * which we have just written.
1278  */
1279 static int is_error_sane(int err)
1280 {
1281     if (err == -EIO || err == -ENOMEM || err == UBI_IO_BAD_HDR ||
1282         err == UBI_IO_BAD_HDR_EBADMSG || err == -ETIMEDOUT)
1283         return 0;
1284     return 1;
1285 }
1286 
1287 /**
1288  * ubi_eba_copy_leb - copy logical eraseblock.
1289  * @ubi: UBI device description object
1290  * @from: physical eraseblock number from where to copy
1291  * @to: physical eraseblock number where to copy
1292  * @vidb: data structure from where the VID header is derived
1293  *
1294  * This function copies logical eraseblock from physical eraseblock @from to
1295  * physical eraseblock @to. The @vid_hdr buffer may be changed by this
1296  * function. Returns:
1297  *   o %0 in case of success;
1298  *   o %MOVE_CANCEL_RACE, %MOVE_TARGET_WR_ERR, %MOVE_TARGET_BITFLIPS, etc;
1299  *   o a negative error code in case of failure.
1300  */
1301 int ubi_eba_copy_leb(struct ubi_device *ubi, int from, int to,
1302              struct ubi_vid_io_buf *vidb)
1303 {
1304     int err, vol_id, lnum, data_size, aldata_size, idx;
1305     struct ubi_vid_hdr *vid_hdr = ubi_get_vid_hdr(vidb);
1306     struct ubi_volume *vol;
1307     uint32_t crc;
1308 
1309     ubi_assert(rwsem_is_locked(&ubi->fm_eba_sem));
1310 
1311     vol_id = be32_to_cpu(vid_hdr->vol_id);
1312     lnum = be32_to_cpu(vid_hdr->lnum);
1313 
1314     dbg_wl("copy LEB %d:%d, PEB %d to PEB %d", vol_id, lnum, from, to);
1315 
1316     if (vid_hdr->vol_type == UBI_VID_STATIC) {
1317         data_size = be32_to_cpu(vid_hdr->data_size);
1318         aldata_size = ALIGN(data_size, ubi->min_io_size);
1319     } else
1320         data_size = aldata_size =
1321                 ubi->leb_size - be32_to_cpu(vid_hdr->data_pad);
1322 
1323     idx = vol_id2idx(ubi, vol_id);
1324     spin_lock(&ubi->volumes_lock);
1325     /*
1326      * Note, we may race with volume deletion, which means that the volume
1327      * this logical eraseblock belongs to might be being deleted. Since the
1328      * volume deletion un-maps all the volume's logical eraseblocks, it will
1329      * be locked in 'ubi_wl_put_peb()' and wait for the WL worker to finish.
1330      */
1331     vol = ubi->volumes[idx];
1332     spin_unlock(&ubi->volumes_lock);
1333     if (!vol) {
1334         /* No need to do further work, cancel */
1335         dbg_wl("volume %d is being removed, cancel", vol_id);
1336         return MOVE_CANCEL_RACE;
1337     }
1338 
1339     /*
1340      * We do not want anybody to write to this logical eraseblock while we
1341      * are moving it, so lock it.
1342      *
1343      * Note, we are using non-waiting locking here, because we cannot sleep
1344      * on the LEB, since it may cause deadlocks. Indeed, imagine a task is
1345      * unmapping the LEB which is mapped to the PEB we are going to move
1346      * (@from). This task locks the LEB and goes sleep in the
1347      * 'ubi_wl_put_peb()' function on the @ubi->move_mutex. In turn, we are
1348      * holding @ubi->move_mutex and go sleep on the LEB lock. So, if the
1349      * LEB is already locked, we just do not move it and return
1350      * %MOVE_RETRY. Note, we do not return %MOVE_CANCEL_RACE here because
1351      * we do not know the reasons of the contention - it may be just a
1352      * normal I/O on this LEB, so we want to re-try.
1353      */
1354     err = leb_write_trylock(ubi, vol_id, lnum);
1355     if (err) {
1356         dbg_wl("contention on LEB %d:%d, cancel", vol_id, lnum);
1357         return MOVE_RETRY;
1358     }
1359 
1360     /*
1361      * The LEB might have been put meanwhile, and the task which put it is
1362      * probably waiting on @ubi->move_mutex. No need to continue the work,
1363      * cancel it.
1364      */
1365     if (vol->eba_tbl->entries[lnum].pnum != from) {
1366         dbg_wl("LEB %d:%d is no longer mapped to PEB %d, mapped to PEB %d, cancel",
1367                vol_id, lnum, from, vol->eba_tbl->entries[lnum].pnum);
1368         err = MOVE_CANCEL_RACE;
1369         goto out_unlock_leb;
1370     }
1371 
1372     /*
1373      * OK, now the LEB is locked and we can safely start moving it. Since
1374      * this function utilizes the @ubi->peb_buf buffer which is shared
1375      * with some other functions - we lock the buffer by taking the
1376      * @ubi->buf_mutex.
1377      */
1378     mutex_lock(&ubi->buf_mutex);
1379     dbg_wl("read %d bytes of data", aldata_size);
1380     err = ubi_io_read_data(ubi, ubi->peb_buf, from, 0, aldata_size);
1381     if (err && err != UBI_IO_BITFLIPS) {
1382         ubi_warn(ubi, "error %d while reading data from PEB %d",
1383              err, from);
1384         err = MOVE_SOURCE_RD_ERR;
1385         goto out_unlock_buf;
1386     }
1387 
1388     /*
1389      * Now we have got to calculate how much data we have to copy. In
1390      * case of a static volume it is fairly easy - the VID header contains
1391      * the data size. In case of a dynamic volume it is more difficult - we
1392      * have to read the contents, cut 0xFF bytes from the end and copy only
1393      * the first part. We must do this to avoid writing 0xFF bytes as it
1394      * may have some side-effects. And not only this. It is important not
1395      * to include those 0xFFs to CRC because later the they may be filled
1396      * by data.
1397      */
1398     if (vid_hdr->vol_type == UBI_VID_DYNAMIC)
1399         aldata_size = data_size =
1400             ubi_calc_data_len(ubi, ubi->peb_buf, data_size);
1401 
1402     cond_resched();
1403     crc = crc32(UBI_CRC32_INIT, ubi->peb_buf, data_size);
1404     cond_resched();
1405 
1406     /*
1407      * It may turn out to be that the whole @from physical eraseblock
1408      * contains only 0xFF bytes. Then we have to only write the VID header
1409      * and do not write any data. This also means we should not set
1410      * @vid_hdr->copy_flag, @vid_hdr->data_size, and @vid_hdr->data_crc.
1411      */
1412     if (data_size > 0) {
1413         vid_hdr->copy_flag = 1;
1414         vid_hdr->data_size = cpu_to_be32(data_size);
1415         vid_hdr->data_crc = cpu_to_be32(crc);
1416     }
1417     vid_hdr->sqnum = cpu_to_be64(ubi_next_sqnum(ubi));
1418 
1419     err = ubi_io_write_vid_hdr(ubi, to, vidb);
1420     if (err) {
1421         if (err == -EIO)
1422             err = MOVE_TARGET_WR_ERR;
1423         goto out_unlock_buf;
1424     }
1425 
1426     cond_resched();
1427 
1428     /* Read the VID header back and check if it was written correctly */
1429     err = ubi_io_read_vid_hdr(ubi, to, vidb, 1);
1430     if (err) {
1431         if (err != UBI_IO_BITFLIPS) {
1432             ubi_warn(ubi, "error %d while reading VID header back from PEB %d",
1433                  err, to);
1434             if (is_error_sane(err))
1435                 err = MOVE_TARGET_RD_ERR;
1436         } else
1437             err = MOVE_TARGET_BITFLIPS;
1438         goto out_unlock_buf;
1439     }
1440 
1441     if (data_size > 0) {
1442         err = ubi_io_write_data(ubi, ubi->peb_buf, to, 0, aldata_size);
1443         if (err) {
1444             if (err == -EIO)
1445                 err = MOVE_TARGET_WR_ERR;
1446             goto out_unlock_buf;
1447         }
1448 
1449         cond_resched();
1450     }
1451 
1452     ubi_assert(vol->eba_tbl->entries[lnum].pnum == from);
1453     vol->eba_tbl->entries[lnum].pnum = to;
1454 
1455 out_unlock_buf:
1456     mutex_unlock(&ubi->buf_mutex);
1457 out_unlock_leb:
1458     leb_write_unlock(ubi, vol_id, lnum);
1459     return err;
1460 }
1461 
1462 /**
1463  * print_rsvd_warning - warn about not having enough reserved PEBs.
1464  * @ubi: UBI device description object
1465  * @ai: UBI attach info 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  */
1481 static 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  */
1514 int 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 
1588 out_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  */
1610 int 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 
1693 out_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 }