minix/fs/mfs/super.c

   1 /* This file manages the super block table and the related data structures,
   2  * namely, the bit maps that keep track of which zones and which inodes are
   3  * allocated and which are free.  When a new inode or zone is needed, the
   4  * appropriate bit map is searched for a free entry.
   5  *
   6  * The entry points into this file are
   7  *   alloc_bit:       somebody wants to allocate a zone or inode; find one
   8  *   free_bit:        indicate that a zone or inode is available for allocation
   9  *   mounted:         tells if file inode is on mounted (or ROOT) file system
  10  *   read_super:      read a superblock
  11  */
  12
  13 #include "fs.h"
  14 #include <string.h>
  15 #include <assert.h>
  16 #include <minix/com.h>
  17 #include <minix/u64.h>
  18 #include <minix/bdev.h>
  19 #include <machine/param.h>
  20 #include <machine/vmparam.h>
  21 #include "buf.h"
  22 #include "inode.h"
  23 #include "super.h"
  24 #include "const.h"
  25
  26 /*===========================================================================*
  27  *                              alloc_bit                                    *
  28  *===========================================================================*/
  29 bit_t alloc_bit(sp, map, origin)
  30 struct super_block *sp;         /* the filesystem to allocate from */
  31 int map;                        /* IMAP (inode map) or ZMAP (zone map) */
  32 bit_t origin;                   /* number of bit to start searching at */
  33 {
  34 /* Allocate a bit from a bit map and return its bit number. */
  35
  36   block_t start_block;          /* first bit block */
  37   block_t block;
  38   bit_t map_bits;               /* how many bits are there in the bit map? */
  39   short bit_blocks;             /* how many blocks are there in the bit map? */
  40   unsigned word, bcount;
  41   struct buf *bp;
  42   bitchunk_t *wptr, *wlim, k;
  43   bit_t i, b;
  44
  45   if (sp->s_rd_only)
  46         panic("can't allocate bit on read-only filesys");
  47
  48   if (map == IMAP) {
  49         start_block = START_BLOCK;
  50         map_bits = (bit_t) (sp->s_ninodes + 1);
  51         bit_blocks = sp->s_imap_blocks;
  52   } else {
  53         start_block = START_BLOCK + sp->s_imap_blocks;
  54         map_bits = (bit_t) (sp->s_zones - (sp->s_firstdatazone - 1));
  55         bit_blocks = sp->s_zmap_blocks;
  56   }
  57
  58   /* Figure out where to start the bit search (depends on 'origin'). */
  59   if (origin >= map_bits) origin = 0;   /* for robustness */
  60
  61   /* Locate the starting place. */
  62   block = (block_t) (origin / FS_BITS_PER_BLOCK(sp->s_block_size));
  63   word = (origin % FS_BITS_PER_BLOCK(sp->s_block_size)) / FS_BITCHUNK_BITS;
  64
  65   /* Iterate over all blocks plus one, because we start in the middle. */
  66   bcount = bit_blocks + 1;
  67   do {
  68         bp = get_block(sp->s_dev, start_block + block, NORMAL);
  69         wlim = &b_bitmap(bp)[FS_BITMAP_CHUNKS(sp->s_block_size)];
  70
  71         /* Iterate over the words in block. */
  72         for (wptr = &b_bitmap(bp)[word]; wptr < wlim; wptr++) {
  73
  74                 /* Does this word contain a free bit? */
  75                 if (*wptr == (bitchunk_t) ~0) continue;
  76
  77                 /* Find and allocate the free bit. */
  78                 k = (bitchunk_t) conv4(sp->s_native, (int) *wptr);
  79                 for (i = 0; (k & (1 << i)) != 0; ++i) {}
  80
  81                 /* Bit number from the start of the bit map. */
  82                 b = ((bit_t) block * FS_BITS_PER_BLOCK(sp->s_block_size))
  83                     + (wptr - &b_bitmap(bp)[0]) * FS_BITCHUNK_BITS
  84                     + i;
  85
  86                 /* Don't allocate bits beyond the end of the map. */
  87                 if (b >= map_bits) break;
  88
  89                 /* Allocate and return bit number. */
  90                 k |= 1 << i;
  91                 *wptr = (bitchunk_t) conv4(sp->s_native, (int) k);
  92                 MARKDIRTY(bp);
  93                 put_block(bp);
  94                 if(map == ZMAP) {
  95                         used_zones++;
  96                         lmfs_change_blockusage(1);
  97                 }
  98                 return(b);
  99         }
 100         put_block(bp);
 101         if (++block >= (unsigned int) bit_blocks) /* last block, wrap around */
 102                 block = 0;
 103         word = 0;
 104   } while (--bcount > 0);
 105   return(NO_BIT);               /* no bit could be allocated */
 106 }
 107
 108 /*===========================================================================*
 109  *                              free_bit                                     *
 110  *===========================================================================*/
 111 void free_bit(sp, map, bit_returned)
 112 struct super_block *sp;         /* the filesystem to operate on */
 113 int map;                        /* IMAP (inode map) or ZMAP (zone map) */
 114 bit_t bit_returned;             /* number of bit to insert into the map */
 115 {
 116 /* Return a zone or inode by turning off its bitmap bit. */
 117
 118   unsigned block, word, bit;
 119   struct buf *bp;
 120   bitchunk_t k, mask;
 121   block_t start_block;
 122
 123   if (sp->s_rd_only)
 124         panic("can't free bit on read-only filesys");
 125
 126   if (map == IMAP) {
 127         start_block = START_BLOCK;
 128   } else {
 129         start_block = START_BLOCK + sp->s_imap_blocks;
 130   }
 131   block = bit_returned / FS_BITS_PER_BLOCK(sp->s_block_size);
 132   word = (bit_returned % FS_BITS_PER_BLOCK(sp->s_block_size))
 133          / FS_BITCHUNK_BITS;
 134
 135   bit = bit_returned % FS_BITCHUNK_BITS;
 136   mask = 1 << bit;
 137
 138   bp = get_block(sp->s_dev, start_block + block, NORMAL);
 139
 140   k = (bitchunk_t) conv4(sp->s_native, (int) b_bitmap(bp)[word]);
 141   if (!(k & mask)) {
 142         if (map == IMAP) panic("tried to free unused inode");
 143         else panic("tried to free unused block: %u", bit_returned);
 144   }
 145
 146   k &= ~mask;
 147   b_bitmap(bp)[word] = (bitchunk_t) conv4(sp->s_native, (int) k);
 148   MARKDIRTY(bp);
 149
 150   put_block(bp);
 151
 152   if(map == ZMAP) {
 153         used_zones--;
 154         lmfs_change_blockusage(-1);
 155   }
 156 }
 157
 158 /*===========================================================================*
 159  *                              get_block_size                               *
 160  *===========================================================================*/
 161 unsigned int get_block_size(dev_t dev)
 162 {
 163   if (dev == NO_DEV)
 164         panic("request for block size of NO_DEV");
 165
 166   return(lmfs_fs_block_size());
 167 }
 168
 169
 170 /*===========================================================================*
 171  *                              rw_super                                     *
 172  *===========================================================================*/
 173 static int rw_super(struct super_block *sp, int writing)
 174 {
 175 /* Read/write a superblock. */
 176   dev_t save_dev = sp->s_dev;
 177   struct buf *bp;
 178   char *sbbuf;
 179   int r;
 180
 181 /* To keep the 1kb on disk clean, only read/write up to and including
 182  * this field.
 183  */
 184 #define LAST_ONDISK_FIELD s_disk_version
 185   int ondisk_bytes = (int) ((char *) &sp->LAST_ONDISK_FIELD - (char *) sp)
 186         + sizeof(sp->LAST_ONDISK_FIELD);
 187
 188   assert(ondisk_bytes > 0);
 189   assert(ondisk_bytes < PAGE_SIZE);
 190   assert(ondisk_bytes < sizeof(struct super_block));
 191
 192   if (sp->s_dev == NO_DEV)
 193         panic("request for super_block of NO_DEV");
 194
 195   /* we rely on the cache blocksize, before reading the
 196    * superblock, being big enough that our complete superblock
 197    * is in block 0.
 198    *
 199    * copy between the disk block and the superblock buffer (depending
 200    * on direction). mark the disk block dirty if the copy is into the
 201    * disk block.
 202    */
 203   assert(lmfs_fs_block_size() >= sizeof(struct super_block) + SUPER_BLOCK_BYTES);
 204   assert(SUPER_BLOCK_BYTES >= sizeof(struct super_block));
 205   assert(SUPER_BLOCK_BYTES >= ondisk_bytes);
 206
 207   /* Unlike accessing any other block, failure to read the superblock is a
 208    * somewhat legitimate use case: it may happen when trying to mount a
 209    * zero-sized partition.  In that case, we'd rather faily cleanly than
 210    * crash the MFS service.
 211    */
 212   if ((r = lmfs_get_block(&bp, sp->s_dev, 0, NORMAL)) != OK) {
 213         if (writing)
 214                 panic("get_block of superblock failed: %d", r);
 215         else
 216                 return r;
 217   }
 218
 219   /* sbbuf points to the disk block at the superblock offset */
 220   sbbuf = (char *) b_data(bp) + SUPER_BLOCK_BYTES;
 221
 222   if(writing) {
 223         memset(b_data(bp), 0, lmfs_fs_block_size());
 224         memcpy(sbbuf, sp, ondisk_bytes);
 225         lmfs_markdirty(bp);
 226   } else {
 227         memset(sp, 0, sizeof(*sp));
 228         memcpy(sp, sbbuf, ondisk_bytes);
 229         sp->s_dev = save_dev;
 230   }
 231
 232   put_block(bp);
 233   lmfs_flushall();
 234
 235   return OK;
 236 }
 237
 238 /*===========================================================================*
 239  *                              read_super                                   *
 240  *===========================================================================*/
 241 int read_super(struct super_block *sp)
 242 {
 243   unsigned int magic;
 244   block_t offset;
 245   int version, native, r;
 246
 247   if((r=rw_super(sp, 0)) != OK)
 248         return r;
 249
 250   magic = sp->s_magic;          /* determines file system type */
 251
 252   if(magic == SUPER_V2 || magic == SUPER_MAGIC) {
 253         printf("MFS: only supports V3 filesystems.\n");
 254         return EINVAL;
 255   }
 256
 257   /* Get file system version and type - only support v3. */
 258   if(magic != SUPER_V3) {
 259         return EINVAL;
 260   }
 261   version = V3;
 262   native = 1;
 263
 264   /* If the super block has the wrong byte order, swap the fields; the magic
 265    * number doesn't need conversion. */
 266   sp->s_ninodes =           (ino_t) conv4(native, (int) sp->s_ninodes);
 267   sp->s_nzones =          (zone1_t) conv2(native, (int) sp->s_nzones);
 268   sp->s_imap_blocks =       (short) conv2(native, (int) sp->s_imap_blocks);
 269   sp->s_zmap_blocks =       (short) conv2(native, (int) sp->s_zmap_blocks);
 270   sp->s_firstdatazone_old =(zone1_t)conv2(native,(int)sp->s_firstdatazone_old);
 271   sp->s_log_zone_size =     (short) conv2(native, (int) sp->s_log_zone_size);
 272   sp->s_max_size =          (off_t) conv4(native, sp->s_max_size);
 273   sp->s_zones =             (zone_t)conv4(native, sp->s_zones);
 274
 275   /* Zones consisting of multiple blocks are longer supported, so fail as early
 276    * as possible. There is still a lot of code cleanup to do here, though.
 277    */
 278   if (sp->s_log_zone_size != 0) {
 279         printf("MFS: block and zone sizes are different\n");
 280         return EINVAL;
 281   }
 282
 283   /* Calculate some other numbers that depend on the version here too, to
 284    * hide some of the differences.
 285    */
 286   assert(version == V3);
 287   sp->s_block_size = (unsigned short) conv2(native,(int) sp->s_block_size);
 288   if (sp->s_block_size < PAGE_SIZE) {
 289         return EINVAL;
 290   }
 291   sp->s_inodes_per_block = V2_INODES_PER_BLOCK(sp->s_block_size);
 292   sp->s_ndzones = V2_NR_DZONES;
 293   sp->s_nindirs = V2_INDIRECTS(sp->s_block_size);
 294
 295   /* For even larger disks, a similar problem occurs with s_firstdatazone.
 296    * If the on-disk field contains zero, we assume that the value was too
 297    * large to fit, and compute it on the fly.
 298    */
 299   if (sp->s_firstdatazone_old == 0) {
 300         offset = START_BLOCK + sp->s_imap_blocks + sp->s_zmap_blocks;
 301         offset += (sp->s_ninodes + sp->s_inodes_per_block - 1) /
 302                 sp->s_inodes_per_block;
 303
 304         sp->s_firstdatazone = (offset + (1 << sp->s_log_zone_size) - 1) >>
 305                 sp->s_log_zone_size;
 306   } else {
 307         sp->s_firstdatazone = (zone_t) sp->s_firstdatazone_old;
 308   }
 309
 310   if (sp->s_block_size < PAGE_SIZE)
 311         return(EINVAL);
 312
 313   if ((sp->s_block_size % 512) != 0)
 314         return(EINVAL);
 315
 316   if (SUPER_SIZE > sp->s_block_size)
 317         return(EINVAL);
 318
 319   if ((sp->s_block_size % V2_INODE_SIZE) != 0) {
 320         return(EINVAL);
 321   }
 322
 323   /* Limit s_max_size to LONG_MAX */
 324   if ((unsigned long)sp->s_max_size > LONG_MAX)
 325         sp->s_max_size = LONG_MAX;
 326
 327   sp->s_isearch = 0;            /* inode searches initially start at 0 */
 328   sp->s_zsearch = 0;            /* zone searches initially start at 0 */
 329   sp->s_version = version;
 330   sp->s_native  = native;
 331
 332   /* Make a few basic checks to see if super block looks reasonable. */
 333   if (sp->s_imap_blocks < 1 || sp->s_zmap_blocks < 1
 334                                 || sp->s_ninodes < 1 || sp->s_zones < 1
 335                                 || sp->s_firstdatazone <= 4
 336                                 || sp->s_firstdatazone >= sp->s_zones
 337                                 || (unsigned) sp->s_log_zone_size > 4) {
 338         printf("not enough imap or zone map blocks, \n");
 339         printf("or not enough inodes, or not enough zones, \n"
 340                 "or invalid first data zone, or zone size too large\n");
 341         return(EINVAL);
 342   }
 343
 344
 345   /* Check any flags we don't understand but are required to. Currently
 346    * these don't exist so all such unknown bits are fatal.
 347    */
 348   if(sp->s_flags & MFSFLAG_MANDATORY_MASK) {
 349         printf("MFS: unsupported feature flags on this FS.\n"
 350                 "Please use a newer MFS to mount it.\n");
 351         return(EINVAL);
 352   }
 353
 354   return(OK);
 355 }
 356
 357 /*===========================================================================*
 358  *                              write_super                                  *
 359  *===========================================================================*/
 360 int write_super(struct super_block *sp)
 361 {
 362   if(sp->s_rd_only)
 363         panic("can't write superblock of readonly filesystem");
 364   return rw_super(sp, 1);
 365 }