SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/logfs/journal.c

   1 /*
   2  * fs/logfs/journal.c   - journal handling code
   3  *
   4  * As should be obvious for Linux kernel code, license is GPLv2
   5  *
   6  * Copyright (c) 2005-2008 Joern Engel <joern@logfs.org>
   7  */
   8 #include "logfs.h"
   9 #include <linux/slab.h>
  10
  11 static void logfs_calc_free(struct super_block *sb)
  12 {
  13         struct logfs_super *super = logfs_super(sb);
  14         u64 reserve, no_segs = super->s_no_segs;
  15         s64 free;
  16         int i;
  17
  18         /* superblock segments */
  19         no_segs -= 2;
  20         super->s_no_journal_segs = 0;
  21         /* journal */
  22         journal_for_each(i)
  23                 if (super->s_journal_seg[i]) {
  24                         no_segs--;
  25                         super->s_no_journal_segs++;
  26                 }
  27
  28         /* open segments plus one extra per level for GC */
  29         no_segs -= 2 * super->s_total_levels;
  30
  31         free = no_segs * (super->s_segsize - LOGFS_SEGMENT_RESERVE);
  32         free -= super->s_used_bytes;
  33         /* just a bit extra */
  34         free -= super->s_total_levels * 4096;
  35
  36         /* Bad blocks are 'paid' for with speed reserve - the filesystem
  37          * simply gets slower as bad blocks accumulate.  Until the bad blocks
  38          * exceed the speed reserve - then the filesystem gets smaller.
  39          */
  40         reserve = super->s_bad_segments + super->s_bad_seg_reserve;
  41         reserve *= super->s_segsize - LOGFS_SEGMENT_RESERVE;
  42         reserve = max(reserve, super->s_speed_reserve);
  43         free -= reserve;
  44         if (free < 0)
  45                 free = 0;
  46
  47         super->s_free_bytes = free;
  48 }
  49
  50 static void reserve_sb_and_journal(struct super_block *sb)
  51 {
  52         struct logfs_super *super = logfs_super(sb);
  53         struct btree_head32 *head = &super->s_reserved_segments;
  54         int i, err;
  55
  56         err = btree_insert32(head, seg_no(sb, super->s_sb_ofs[0]), (void *)1,
  57                         GFP_KERNEL);
  58         BUG_ON(err);
  59
  60         err = btree_insert32(head, seg_no(sb, super->s_sb_ofs[1]), (void *)1,
  61                         GFP_KERNEL);
  62         BUG_ON(err);
  63
  64         journal_for_each(i) {
  65                 if (!super->s_journal_seg[i])
  66                         continue;
  67                 err = btree_insert32(head, super->s_journal_seg[i], (void *)1,
  68                                 GFP_KERNEL);
  69                 BUG_ON(err);
  70         }
  71 }
  72
  73 static void read_dynsb(struct super_block *sb,
  74                 struct logfs_je_dynsb *dynsb)
  75 {
  76         struct logfs_super *super = logfs_super(sb);
  77
  78         super->s_gec            = be64_to_cpu(dynsb->ds_gec);
  79         super->s_sweeper        = be64_to_cpu(dynsb->ds_sweeper);
  80         super->s_victim_ino     = be64_to_cpu(dynsb->ds_victim_ino);
  81         super->s_rename_dir     = be64_to_cpu(dynsb->ds_rename_dir);
  82         super->s_rename_pos     = be64_to_cpu(dynsb->ds_rename_pos);
  83         super->s_used_bytes     = be64_to_cpu(dynsb->ds_used_bytes);
  84         super->s_generation     = be32_to_cpu(dynsb->ds_generation);
  85 }
  86
  87 static void read_anchor(struct super_block *sb,
  88                 struct logfs_je_anchor *da)
  89 {
  90         struct logfs_super *super = logfs_super(sb);
  91         struct inode *inode = super->s_master_inode;
  92         struct logfs_inode *li = logfs_inode(inode);
  93         int i;
  94
  95         super->s_last_ino = be64_to_cpu(da->da_last_ino);
  96         li->li_flags    = 0;
  97         li->li_height   = da->da_height;
  98         i_size_write(inode, be64_to_cpu(da->da_size));
  99         li->li_used_bytes = be64_to_cpu(da->da_used_bytes);
 100
 101         for (i = 0; i < LOGFS_EMBEDDED_FIELDS; i++)
 102                 li->li_data[i] = be64_to_cpu(da->da_data[i]);
 103 }
 104
 105 static void read_erasecount(struct super_block *sb,
 106                 struct logfs_je_journal_ec *ec)
 107 {
 108         struct logfs_super *super = logfs_super(sb);
 109         int i;
 110
 111         journal_for_each(i)
 112                 super->s_journal_ec[i] = be32_to_cpu(ec->ec[i]);
 113 }
 114
 115 static int read_area(struct super_block *sb, struct logfs_je_area *a)
 116 {
 117         struct logfs_super *super = logfs_super(sb);
 118         struct logfs_area *area = super->s_area[a->gc_level];
 119         u64 ofs;
 120         u32 writemask = ~(super->s_writesize - 1);
 121
 122         if (a->gc_level >= LOGFS_NO_AREAS)
 123                 return -EIO;
 124         if (a->vim != VIM_DEFAULT)
 125                 return -EIO; /* TODO: close area and continue */
 126
 127         area->a_used_bytes = be32_to_cpu(a->used_bytes);
 128         area->a_written_bytes = area->a_used_bytes & writemask;
 129         area->a_segno = be32_to_cpu(a->segno);
 130         if (area->a_segno)
 131                 area->a_is_open = 1;
 132
 133         ofs = dev_ofs(sb, area->a_segno, area->a_written_bytes);
 134         if (super->s_writesize > 1)
 135                 logfs_buf_recover(area, ofs, a + 1, super->s_writesize);
 136         else
 137                 logfs_buf_recover(area, ofs, NULL, 0);
 138         return 0;
 139 }
 140
 141 static void *unpack(void *from, void *to)
 142 {
 143         struct logfs_journal_header *jh = from;
 144         void *data = from + sizeof(struct logfs_journal_header);
 145         int err;
 146         size_t inlen, outlen;
 147
 148         inlen = be16_to_cpu(jh->h_len);
 149         outlen = be16_to_cpu(jh->h_datalen);
 150
 151         if (jh->h_compr == COMPR_NONE)
 152                 memcpy(to, data, inlen);
 153         else {
 154                 err = logfs_uncompress(data, to, inlen, outlen);
 155                 BUG_ON(err);
 156         }
 157         return to;
 158 }
 159
 160 static int __read_je_header(struct super_block *sb, u64 ofs,
 161                 struct logfs_journal_header *jh)
 162 {
 163         struct logfs_super *super = logfs_super(sb);
 164         size_t bufsize = max_t(size_t, sb->s_blocksize, super->s_writesize)
 165                 + MAX_JOURNAL_HEADER;
 166         u16 type, len, datalen;
 167         int err;
 168
 169         /* read header only */
 170         err = wbuf_read(sb, ofs, sizeof(*jh), jh);
 171         if (err)
 172                 return err;
 173         type = be16_to_cpu(jh->h_type);
 174         len = be16_to_cpu(jh->h_len);
 175         datalen = be16_to_cpu(jh->h_datalen);
 176         if (len > sb->s_blocksize)
 177                 return -EIO;
 178         if ((type < JE_FIRST) || (type > JE_LAST))
 179                 return -EIO;
 180         if (datalen > bufsize)
 181                 return -EIO;
 182         return 0;
 183 }
 184
 185 static int __read_je_payload(struct super_block *sb, u64 ofs,
 186                 struct logfs_journal_header *jh)
 187 {
 188         u16 len;
 189         int err;
 190
 191         len = be16_to_cpu(jh->h_len);
 192         err = wbuf_read(sb, ofs + sizeof(*jh), len, jh + 1);
 193         if (err)
 194                 return err;
 195         if (jh->h_crc != logfs_crc32(jh, len + sizeof(*jh), 4)) {
 196                 /* Old code was confused.  It forgot about the header length
 197                  * and stopped calculating the crc 16 bytes before the end
 198                  * of data - ick!
 199                  * FIXME: Remove this hack once the old code is fixed.
 200                  */
 201                 if (jh->h_crc == logfs_crc32(jh, len, 4))
 202                         WARN_ON_ONCE(1);
 203                 else
 204                         return -EIO;
 205         }
 206         return 0;
 207 }
 208
 209 /*
 210  * jh needs to be large enough to hold the complete entry, not just the header
 211  */
 212 static int __read_je(struct super_block *sb, u64 ofs,
 213                 struct logfs_journal_header *jh)
 214 {
 215         int err;
 216
 217         err = __read_je_header(sb, ofs, jh);
 218         if (err)
 219                 return err;
 220         return __read_je_payload(sb, ofs, jh);
 221 }
 222
 223 static int read_je(struct super_block *sb, u64 ofs)
 224 {
 225         struct logfs_super *super = logfs_super(sb);
 226         struct logfs_journal_header *jh = super->s_compressed_je;
 227         void *scratch = super->s_je;
 228         u16 type, datalen;
 229         int err;
 230
 231         err = __read_je(sb, ofs, jh);
 232         if (err)
 233                 return err;
 234         type = be16_to_cpu(jh->h_type);
 235         datalen = be16_to_cpu(jh->h_datalen);
 236
 237         switch (type) {
 238         case JE_DYNSB:
 239                 read_dynsb(sb, unpack(jh, scratch));
 240                 break;
 241         case JE_ANCHOR:
 242                 read_anchor(sb, unpack(jh, scratch));
 243                 break;
 244         case JE_ERASECOUNT:
 245                 read_erasecount(sb, unpack(jh, scratch));
 246                 break;
 247         case JE_AREA:
 248                 read_area(sb, unpack(jh, scratch));
 249                 break;
 250         case JE_OBJ_ALIAS:
 251                 err = logfs_load_object_aliases(sb, unpack(jh, scratch),
 252                                 datalen);
 253                 break;
 254         default:
 255                 WARN_ON_ONCE(1);
 256                 return -EIO;
 257         }
 258         return err;
 259 }
 260
 261 static int logfs_read_segment(struct super_block *sb, u32 segno)
 262 {
 263         struct logfs_super *super = logfs_super(sb);
 264         struct logfs_journal_header *jh = super->s_compressed_je;
 265         u64 ofs, seg_ofs = dev_ofs(sb, segno, 0);
 266         u32 h_ofs, last_ofs = 0;
 267         u16 len, datalen, last_len = 0;
 268         int i, err;
 269
 270         /* search for most recent commit */
 271         for (h_ofs = 0; h_ofs < super->s_segsize; h_ofs += sizeof(*jh)) {
 272                 ofs = seg_ofs + h_ofs;
 273                 err = __read_je_header(sb, ofs, jh);
 274                 if (err)
 275                         continue;
 276                 if (jh->h_type != cpu_to_be16(JE_COMMIT))
 277                         continue;
 278                 err = __read_je_payload(sb, ofs, jh);
 279                 if (err)
 280                         continue;
 281                 len = be16_to_cpu(jh->h_len);
 282                 datalen = be16_to_cpu(jh->h_datalen);
 283                 if ((datalen > sizeof(super->s_je_array)) ||
 284                                 (datalen % sizeof(__be64)))
 285                         continue;
 286                 last_ofs = h_ofs;
 287                 last_len = datalen;
 288                 h_ofs += ALIGN(len, sizeof(*jh)) - sizeof(*jh);
 289         }
 290         /* read commit */
 291         if (last_ofs == 0)
 292                 return -ENOENT;
 293         ofs = seg_ofs + last_ofs;
 294         log_journal("Read commit from %llx\n", ofs);
 295         err = __read_je(sb, ofs, jh);
 296         BUG_ON(err); /* We should have caught it in the scan loop already */
 297         if (err)
 298                 return err;
 299         /* uncompress */
 300         unpack(jh, super->s_je_array);
 301         super->s_no_je = last_len / sizeof(__be64);
 302         /* iterate over array */
 303         for (i = 0; i < super->s_no_je; i++) {
 304                 err = read_je(sb, be64_to_cpu(super->s_je_array[i]));
 305                 if (err)
 306                         return err;
 307         }
 308         super->s_journal_area->a_segno = segno;
 309         return 0;
 310 }
 311
 312 static u64 read_gec(struct super_block *sb, u32 segno)
 313 {
 314         struct logfs_segment_header sh;
 315         __be32 crc;
 316         int err;
 317
 318         if (!segno)
 319                 return 0;
 320         err = wbuf_read(sb, dev_ofs(sb, segno, 0), sizeof(sh), &sh);
 321         if (err)
 322                 return 0;
 323         crc = logfs_crc32(&sh, sizeof(sh), 4);
 324         if (crc != sh.crc) {
 325                 WARN_ON(sh.gec != cpu_to_be64(0xffffffffffffffffull));
 326                 /* Most likely it was just erased */
 327                 return 0;
 328         }
 329         return be64_to_cpu(sh.gec);
 330 }
 331
 332 static int logfs_read_journal(struct super_block *sb)
 333 {
 334         struct logfs_super *super = logfs_super(sb);
 335         u64 gec[LOGFS_JOURNAL_SEGS], max;
 336         u32 segno;
 337         int i, max_i;
 338
 339         max = 0;
 340         max_i = -1;
 341         journal_for_each(i) {
 342                 segno = super->s_journal_seg[i];
 343                 gec[i] = read_gec(sb, super->s_journal_seg[i]);
 344                 if (gec[i] > max) {
 345                         max = gec[i];
 346                         max_i = i;
 347                 }
 348         }
 349         if (max_i == -1)
 350                 return -EIO;
 351         /* FIXME: Try older segments in case of error */
 352         return logfs_read_segment(sb, super->s_journal_seg[max_i]);
 353 }
 354
 355 /*
 356  * First search the current segment (outer loop), then pick the next segment
 357  * in the array, skipping any zero entries (inner loop).
 358  */
 359 static void journal_get_free_segment(struct logfs_area *area)
 360 {
 361         struct logfs_super *super = logfs_super(area->a_sb);
 362         int i;
 363
 364         journal_for_each(i) {
 365                 if (area->a_segno != super->s_journal_seg[i])
 366                         continue;
 367
 368                 do {
 369                         i++;
 370                         if (i == LOGFS_JOURNAL_SEGS)
 371                                 i = 0;
 372                 } while (!super->s_journal_seg[i]);
 373
 374                 area->a_segno = super->s_journal_seg[i];
 375                 area->a_erase_count = ++(super->s_journal_ec[i]);
 376                 log_journal("Journal now at %x (ec %x)\n", area->a_segno,
 377                                 area->a_erase_count);
 378                 return;
 379         }
 380         BUG();
 381 }
 382
 383 static void journal_get_erase_count(struct logfs_area *area)
 384 {
 385         /* erase count is stored globally and incremented in
 386          * journal_get_free_segment() - nothing to do here */
 387 }
 388
 389 static int journal_erase_segment(struct logfs_area *area)
 390 {
 391         struct super_block *sb = area->a_sb;
 392         struct logfs_segment_header sh;
 393         u64 ofs;
 394         int err;
 395
 396         err = logfs_erase_segment(sb, area->a_segno, 1);
 397         if (err)
 398                 return err;
 399
 400         sh.pad = 0;
 401         sh.type = SEG_JOURNAL;
 402         sh.level = 0;
 403         sh.segno = cpu_to_be32(area->a_segno);
 404         sh.ec = cpu_to_be32(area->a_erase_count);
 405         sh.gec = cpu_to_be64(logfs_super(sb)->s_gec);
 406         sh.crc = logfs_crc32(&sh, sizeof(sh), 4);
 407
 408         /* This causes a bug in segment.c.  Not yet. */
 409         //logfs_set_segment_erased(sb, area->a_segno, area->a_erase_count, 0);
 410
 411         ofs = dev_ofs(sb, area->a_segno, 0);
 412         area->a_used_bytes = ALIGN(sizeof(sh), 16);
 413         logfs_buf_write(area, ofs, &sh, sizeof(sh));
 414         return 0;
 415 }
 416
 417 static size_t __logfs_write_header(struct logfs_super *super,
 418                 struct logfs_journal_header *jh, size_t len, size_t datalen,
 419                 u16 type, u8 compr)
 420 {
 421         jh->h_len       = cpu_to_be16(len);
 422         jh->h_type      = cpu_to_be16(type);
 423         jh->h_datalen   = cpu_to_be16(datalen);
 424         jh->h_compr     = compr;
 425         jh->h_pad[0]    = 'H';
 426         jh->h_pad[1]    = 'E';
 427         jh->h_pad[2]    = 'A';
 428         jh->h_pad[3]    = 'D';
 429         jh->h_pad[4]    = 'R';
 430         jh->h_crc       = logfs_crc32(jh, len + sizeof(*jh), 4);
 431         return ALIGN(len, 16) + sizeof(*jh);
 432 }
 433
 434 static size_t logfs_write_header(struct logfs_super *super,
 435                 struct logfs_journal_header *jh, size_t datalen, u16 type)
 436 {
 437         size_t len = datalen;
 438
 439         return __logfs_write_header(super, jh, len, datalen, type, COMPR_NONE);
 440 }
 441
 442 static inline size_t logfs_journal_erasecount_size(struct logfs_super *super)
 443 {
 444         return LOGFS_JOURNAL_SEGS * sizeof(__be32);
 445 }
 446
 447 static void *logfs_write_erasecount(struct super_block *sb, void *_ec,
 448                 u16 *type, size_t *len)
 449 {
 450         struct logfs_super *super = logfs_super(sb);
 451         struct logfs_je_journal_ec *ec = _ec;
 452         int i;
 453
 454         journal_for_each(i)
 455                 ec->ec[i] = cpu_to_be32(super->s_journal_ec[i]);
 456         *type = JE_ERASECOUNT;
 457         *len = logfs_journal_erasecount_size(super);
 458         return ec;
 459 }
 460
 461 static void account_shadow(void *_shadow, unsigned long _sb, u64 ignore,
 462                 size_t ignore2)
 463 {
 464         struct logfs_shadow *shadow = _shadow;
 465         struct super_block *sb = (void *)_sb;
 466         struct logfs_super *super = logfs_super(sb);
 467
 468         /* consume new space */
 469         super->s_free_bytes       -= shadow->new_len;
 470         super->s_used_bytes       += shadow->new_len;
 471         super->s_dirty_used_bytes -= shadow->new_len;
 472
 473         /* free up old space */
 474         super->s_free_bytes       += shadow->old_len;
 475         super->s_used_bytes       -= shadow->old_len;
 476         super->s_dirty_free_bytes -= shadow->old_len;
 477
 478         logfs_set_segment_used(sb, shadow->old_ofs, -shadow->old_len);
 479         logfs_set_segment_used(sb, shadow->new_ofs, shadow->new_len);
 480
 481         log_journal("account_shadow(%llx, %llx, %x) %llx->%llx %x->%x\n",
 482                         shadow->ino, shadow->bix, shadow->gc_level,
 483                         shadow->old_ofs, shadow->new_ofs,
 484                         shadow->old_len, shadow->new_len);
 485         mempool_free(shadow, super->s_shadow_pool);
 486 }
 487
 488 static void account_shadows(struct super_block *sb)
 489 {
 490         struct logfs_super *super = logfs_super(sb);
 491         struct inode *inode = super->s_master_inode;
 492         struct logfs_inode *li = logfs_inode(inode);
 493         struct shadow_tree *tree = &super->s_shadow_tree;
 494
 495         btree_grim_visitor64(&tree->new, (unsigned long)sb, account_shadow);
 496         btree_grim_visitor64(&tree->old, (unsigned long)sb, account_shadow);
 497
 498         if (li->li_block) {
 499                 /*
 500                  * We never actually use the structure, when attached to the
 501                  * master inode.  But it is easier to always free it here than
 502                  * to have checks in several places elsewhere when allocating
 503                  * it.
 504                  */
 505                 li->li_block->ops->free_block(sb, li->li_block);
 506         }
 507         BUG_ON((s64)li->li_used_bytes < 0);
 508 }
 509
 510 static void *__logfs_write_anchor(struct super_block *sb, void *_da,
 511                 u16 *type, size_t *len)
 512 {
 513         struct logfs_super *super = logfs_super(sb);
 514         struct logfs_je_anchor *da = _da;
 515         struct inode *inode = super->s_master_inode;
 516         struct logfs_inode *li = logfs_inode(inode);
 517         int i;
 518
 519         da->da_height   = li->li_height;
 520         da->da_last_ino = cpu_to_be64(super->s_last_ino);
 521         da->da_size     = cpu_to_be64(i_size_read(inode));
 522         da->da_used_bytes = cpu_to_be64(li->li_used_bytes);
 523         for (i = 0; i < LOGFS_EMBEDDED_FIELDS; i++)
 524                 da->da_data[i] = cpu_to_be64(li->li_data[i]);
 525         *type = JE_ANCHOR;
 526         *len = sizeof(*da);
 527         return da;
 528 }
 529
 530 static void *logfs_write_dynsb(struct super_block *sb, void *_dynsb,
 531                 u16 *type, size_t *len)
 532 {
 533         struct logfs_super *super = logfs_super(sb);
 534         struct logfs_je_dynsb *dynsb = _dynsb;
 535
 536         dynsb->ds_gec           = cpu_to_be64(super->s_gec);
 537         dynsb->ds_sweeper       = cpu_to_be64(super->s_sweeper);
 538         dynsb->ds_victim_ino    = cpu_to_be64(super->s_victim_ino);
 539         dynsb->ds_rename_dir    = cpu_to_be64(super->s_rename_dir);
 540         dynsb->ds_rename_pos    = cpu_to_be64(super->s_rename_pos);
 541         dynsb->ds_used_bytes    = cpu_to_be64(super->s_used_bytes);
 542         dynsb->ds_generation    = cpu_to_be32(super->s_generation);
 543         *type = JE_DYNSB;
 544         *len = sizeof(*dynsb);
 545         return dynsb;
 546 }
 547
 548 static void write_wbuf(struct super_block *sb, struct logfs_area *area,
 549                 void *wbuf)
 550 {
 551         struct logfs_super *super = logfs_super(sb);
 552         struct address_space *mapping = super->s_mapping_inode->i_mapping;
 553         u64 ofs;
 554         pgoff_t index;
 555         int page_ofs;
 556         struct page *page;
 557
 558         ofs = dev_ofs(sb, area->a_segno,
 559                         area->a_used_bytes & ~(super->s_writesize - 1));
 560         index = ofs >> PAGE_SHIFT;
 561         page_ofs = ofs & (PAGE_SIZE - 1);
 562
 563         page = find_lock_page(mapping, index);
 564         BUG_ON(!page);
 565         memcpy(wbuf, page_address(page) + page_ofs, super->s_writesize);
 566         unlock_page(page);
 567 }
 568
 569 static void *logfs_write_area(struct super_block *sb, void *_a,
 570                 u16 *type, size_t *len)
 571 {
 572         struct logfs_super *super = logfs_super(sb);
 573         struct logfs_area *area = super->s_area[super->s_sum_index];
 574         struct logfs_je_area *a = _a;
 575
 576         a->vim = VIM_DEFAULT;
 577         a->gc_level = super->s_sum_index;
 578         a->used_bytes = cpu_to_be32(area->a_used_bytes);
 579         a->segno = cpu_to_be32(area->a_segno);
 580         if (super->s_writesize > 1)
 581                 write_wbuf(sb, area, a + 1);
 582
 583         *type = JE_AREA;
 584         *len = sizeof(*a) + super->s_writesize;
 585         return a;
 586 }
 587
 588 static void *logfs_write_commit(struct super_block *sb, void *h,
 589                 u16 *type, size_t *len)
 590 {
 591         struct logfs_super *super = logfs_super(sb);
 592
 593         *type = JE_COMMIT;
 594         *len = super->s_no_je * sizeof(__be64);
 595         return super->s_je_array;
 596 }
 597
 598 static size_t __logfs_write_je(struct super_block *sb, void *buf, u16 type,
 599                 size_t len)
 600 {
 601         struct logfs_super *super = logfs_super(sb);
 602         void *header = super->s_compressed_je;
 603         void *data = header + sizeof(struct logfs_journal_header);
 604         ssize_t compr_len, pad_len;
 605         u8 compr = COMPR_ZLIB;
 606
 607         if (len == 0)
 608                 return logfs_write_header(super, header, 0, type);
 609
 610         compr_len = logfs_compress(buf, data, len, sb->s_blocksize);
 611         if (compr_len < 0 || type == JE_ANCHOR) {
 612                 BUG_ON(len > sb->s_blocksize);
 613                 memcpy(data, buf, len);
 614                 compr_len = len;
 615                 compr = COMPR_NONE;
 616         }
 617
 618         pad_len = ALIGN(compr_len, 16);
 619         memset(data + compr_len, 0, pad_len - compr_len);
 620
 621         return __logfs_write_header(super, header, compr_len, len, type, compr);
 622 }
 623
 624 static s64 logfs_get_free_bytes(struct logfs_area *area, size_t *bytes,
 625                 int must_pad)
 626 {
 627         u32 writesize = logfs_super(area->a_sb)->s_writesize;
 628         s32 ofs;
 629         int ret;
 630
 631         ret = logfs_open_area(area, *bytes);
 632         if (ret)
 633                 return -EAGAIN;
 634
 635         ofs = area->a_used_bytes;
 636         area->a_used_bytes += *bytes;
 637
 638         if (must_pad) {
 639                 area->a_used_bytes = ALIGN(area->a_used_bytes, writesize);
 640                 *bytes = area->a_used_bytes - ofs;
 641         }
 642
 643         return dev_ofs(area->a_sb, area->a_segno, ofs);
 644 }
 645
 646 static int logfs_write_je_buf(struct super_block *sb, void *buf, u16 type,
 647                 size_t buf_len)
 648 {
 649         struct logfs_super *super = logfs_super(sb);
 650         struct logfs_area *area = super->s_journal_area;
 651         struct logfs_journal_header *jh = super->s_compressed_je;
 652         size_t len;
 653         int must_pad = 0;
 654         s64 ofs;
 655
 656         len = __logfs_write_je(sb, buf, type, buf_len);
 657         if (jh->h_type == cpu_to_be16(JE_COMMIT))
 658                 must_pad = 1;
 659
 660         ofs = logfs_get_free_bytes(area, &len, must_pad);
 661         if (ofs < 0)
 662                 return ofs;
 663         logfs_buf_write(area, ofs, super->s_compressed_je, len);
 664         super->s_je_array[super->s_no_je++] = cpu_to_be64(ofs);
 665         return 0;
 666 }
 667
 668 static int logfs_write_je(struct super_block *sb,
 669                 void* (*write)(struct super_block *sb, void *scratch,
 670                         u16 *type, size_t *len))
 671 {
 672         void *buf;
 673         size_t len;
 674         u16 type;
 675
 676         buf = write(sb, logfs_super(sb)->s_je, &type, &len);
 677         return logfs_write_je_buf(sb, buf, type, len);
 678 }
 679
 680 int write_alias_journal(struct super_block *sb, u64 ino, u64 bix,
 681                 level_t level, int child_no, __be64 val)
 682 {
 683         struct logfs_super *super = logfs_super(sb);
 684         struct logfs_obj_alias *oa = super->s_je;
 685         int err = 0, fill = super->s_je_fill;
 686
 687         log_aliases("logfs_write_obj_aliases #%x(%llx, %llx, %x, %x) %llx\n",
 688                         fill, ino, bix, level, child_no, be64_to_cpu(val));
 689         oa[fill].ino = cpu_to_be64(ino);
 690         oa[fill].bix = cpu_to_be64(bix);
 691         oa[fill].val = val;
 692         oa[fill].level = (__force u8)level;
 693         oa[fill].child_no = cpu_to_be16(child_no);
 694         fill++;
 695         if (fill >= sb->s_blocksize / sizeof(*oa)) {
 696                 err = logfs_write_je_buf(sb, oa, JE_OBJ_ALIAS, sb->s_blocksize);
 697                 fill = 0;
 698         }
 699
 700         super->s_je_fill = fill;
 701         return err;
 702 }
 703
 704 static int logfs_write_obj_aliases(struct super_block *sb)
 705 {
 706         struct logfs_super *super = logfs_super(sb);
 707         int err;
 708
 709         log_journal("logfs_write_obj_aliases: %d aliases to write\n",
 710                         super->s_no_object_aliases);
 711         super->s_je_fill = 0;
 712         err = logfs_write_obj_aliases_pagecache(sb);
 713         if (err)
 714                 return err;
 715
 716         if (super->s_je_fill)
 717                 err = logfs_write_je_buf(sb, super->s_je, JE_OBJ_ALIAS,
 718                                 super->s_je_fill
 719                                 * sizeof(struct logfs_obj_alias));
 720         return err;
 721 }
 722
 723 /*
 724  * Write all journal entries.  The goto logic ensures that all journal entries
 725  * are written whenever a new segment is used.  It is ugly and potentially a
 726  * bit wasteful, but robustness is more important.  With this we can *always*
 727  * erase all journal segments except the one containing the most recent commit.
 728  */
 729 void logfs_write_anchor(struct super_block *sb)
 730 {
 731         struct logfs_super *super = logfs_super(sb);
 732         struct logfs_area *area = super->s_journal_area;
 733         int i, err;
 734
 735         if (!(super->s_flags & LOGFS_SB_FLAG_DIRTY))
 736                 return;
 737         super->s_flags &= ~LOGFS_SB_FLAG_DIRTY;
 738
 739         BUG_ON(super->s_flags & LOGFS_SB_FLAG_SHUTDOWN);
 740         mutex_lock(&super->s_journal_mutex);
 741
 742         /* Do this first or suffer corruption */
 743         logfs_sync_segments(sb);
 744         account_shadows(sb);
 745
 746 again:
 747         super->s_no_je = 0;
 748         for_each_area(i) {
 749                 if (!super->s_area[i]->a_is_open)
 750                         continue;
 751                 super->s_sum_index = i;
 752                 err = logfs_write_je(sb, logfs_write_area);
 753                 if (err)
 754                         goto again;
 755         }
 756         err = logfs_write_obj_aliases(sb);
 757         if (err)
 758                 goto again;
 759         err = logfs_write_je(sb, logfs_write_erasecount);
 760         if (err)
 761                 goto again;
 762         err = logfs_write_je(sb, __logfs_write_anchor);
 763         if (err)
 764                 goto again;
 765         err = logfs_write_je(sb, logfs_write_dynsb);
 766         if (err)
 767                 goto again;
 768         /*
 769          * Order is imperative.  First we sync all writes, including the
 770          * non-committed journal writes.  Then we write the final commit and
 771          * sync the current journal segment.
 772          * There is a theoretical bug here.  Syncing the journal segment will
 773          * write a number of journal entries and the final commit.  All these
 774          * are written in a single operation.  If the device layer writes the
 775          * data back-to-front, the commit will precede the other journal
 776          * entries, leaving a race window.
 777          * Two fixes are possible.  Preferred is to fix the device layer to
 778          * ensure writes happen front-to-back.  Alternatively we can insert
 779          * another logfs_sync_area() super->s_devops->sync() combo before
 780          * writing the commit.
 781          */
 782         /*
 783          * On another subject, super->s_devops->sync is usually not necessary.
 784          * Unless called from sys_sync or friends, a barrier would suffice.
 785          */
 786         super->s_devops->sync(sb);
 787         err = logfs_write_je(sb, logfs_write_commit);
 788         if (err)
 789                 goto again;
 790         log_journal("Write commit to %llx\n",
 791                         be64_to_cpu(super->s_je_array[super->s_no_je - 1]));
 792         logfs_sync_area(area);
 793         BUG_ON(area->a_used_bytes != area->a_written_bytes);
 794         super->s_devops->sync(sb);
 795
 796         mutex_unlock(&super->s_journal_mutex);
 797         return;
 798 }
 799
 800 void do_logfs_journal_wl_pass(struct super_block *sb)
 801 {
 802         struct logfs_super *super = logfs_super(sb);
 803         struct logfs_area *area = super->s_journal_area;
 804         struct btree_head32 *head = &super->s_reserved_segments;
 805         u32 segno, ec;
 806         int i, err;
 807
 808         log_journal("Journal requires wear-leveling.\n");
 809         /* Drop old segments */
 810         journal_for_each(i)
 811                 if (super->s_journal_seg[i]) {
 812                         btree_remove32(head, super->s_journal_seg[i]);
 813                         logfs_set_segment_unreserved(sb,
 814                                         super->s_journal_seg[i],
 815                                         super->s_journal_ec[i]);
 816                         super->s_journal_seg[i] = 0;
 817                         super->s_journal_ec[i] = 0;
 818                 }
 819         /* Get new segments */
 820         for (i = 0; i < super->s_no_journal_segs; i++) {
 821                 segno = get_best_cand(sb, &super->s_reserve_list, &ec);
 822                 super->s_journal_seg[i] = segno;
 823                 super->s_journal_ec[i] = ec;
 824                 logfs_set_segment_reserved(sb, segno);
 825                 err = btree_insert32(head, segno, (void *)1, GFP_KERNEL);
 826                 BUG_ON(err); /* mempool should prevent this */
 827                 err = logfs_erase_segment(sb, segno, 1);
 828                 BUG_ON(err); /* FIXME: remount-ro would be nicer */
 829         }
 830         /* Manually move journal_area */
 831         freeseg(sb, area->a_segno);
 832         area->a_segno = super->s_journal_seg[0];
 833         area->a_is_open = 0;
 834         area->a_used_bytes = 0;
 835         /* Write journal */
 836         logfs_write_anchor(sb);
 837         /* Write superblocks */
 838         err = logfs_write_sb(sb);
 839         BUG_ON(err);
 840 }
 841
 842 static const struct logfs_area_ops journal_area_ops = {
 843         .get_free_segment       = journal_get_free_segment,
 844         .get_erase_count        = journal_get_erase_count,
 845         .erase_segment          = journal_erase_segment,
 846 };
 847
 848 int logfs_init_journal(struct super_block *sb)
 849 {
 850         struct logfs_super *super = logfs_super(sb);
 851         size_t bufsize = max_t(size_t, sb->s_blocksize, super->s_writesize)
 852                 + MAX_JOURNAL_HEADER;
 853         int ret = -ENOMEM;
 854
 855         mutex_init(&super->s_journal_mutex);
 856         btree_init_mempool32(&super->s_reserved_segments, super->s_btree_pool);
 857
 858         super->s_je = kzalloc(bufsize, GFP_KERNEL);
 859         if (!super->s_je)
 860                 return ret;
 861
 862         super->s_compressed_je = kzalloc(bufsize, GFP_KERNEL);
 863         if (!super->s_compressed_je)
 864                 return ret;
 865
 866         super->s_master_inode = logfs_new_meta_inode(sb, LOGFS_INO_MASTER);
 867         if (IS_ERR(super->s_master_inode))
 868                 return PTR_ERR(super->s_master_inode);
 869
 870         ret = logfs_read_journal(sb);
 871         if (ret)
 872                 return -EIO;
 873
 874         reserve_sb_and_journal(sb);
 875         logfs_calc_free(sb);
 876
 877         super->s_journal_area->a_ops = &journal_area_ops;
 878         return 0;
 879 }
 880
 881 void logfs_cleanup_journal(struct super_block *sb)
 882 {
 883         struct logfs_super *super = logfs_super(sb);
 884
 885         btree_grim_visitor32(&super->s_reserved_segments, 0, NULL);
 886         destroy_meta_inode(super->s_master_inode);
 887         super->s_master_inode = NULL;
 888
 889         kfree(super->s_compressed_je);
 890         kfree(super->s_je);
 891 }