fs/nilfs2/sufile.c

   1 /*
   2  * sufile.c - NILFS segment usage file.
   3  *
   4  * Copyright (C) 2006-2008 Nippon Telegraph and Telephone Corporation.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * Written by Koji Sato.
  17  * Revised by Ryusuke Konishi.
  18  */
  19
  20 #include <linux/kernel.h>
  21 #include <linux/fs.h>
  22 #include <linux/string.h>
  23 #include <linux/buffer_head.h>
  24 #include <linux/errno.h>
  25 #include <linux/nilfs2_fs.h>
  26 #include "mdt.h"
  27 #include "sufile.h"
  28
  29 #include <trace/events/nilfs2.h>
  30
  31 /**
  32  * struct nilfs_sufile_info - on-memory private data of sufile
  33  * @mi: on-memory private data of metadata file
  34  * @ncleansegs: number of clean segments
  35  * @allocmin: lower limit of allocatable segment range
  36  * @allocmax: upper limit of allocatable segment range
  37  */
  38 struct nilfs_sufile_info {
  39         struct nilfs_mdt_info mi;
  40         unsigned long ncleansegs;/* number of clean segments */
  41         __u64 allocmin;         /* lower limit of allocatable segment range */
  42         __u64 allocmax;         /* upper limit of allocatable segment range */
  43 };
  44
  45 static inline struct nilfs_sufile_info *NILFS_SUI(struct inode *sufile)
  46 {
  47         return (struct nilfs_sufile_info *)NILFS_MDT(sufile);
  48 }
  49
  50 static inline unsigned long
  51 nilfs_sufile_segment_usages_per_block(const struct inode *sufile)
  52 {
  53         return NILFS_MDT(sufile)->mi_entries_per_block;
  54 }
  55
  56 static unsigned long
  57 nilfs_sufile_get_blkoff(const struct inode *sufile, __u64 segnum)
  58 {
  59         __u64 t = segnum + NILFS_MDT(sufile)->mi_first_entry_offset;
  60
  61         do_div(t, nilfs_sufile_segment_usages_per_block(sufile));
  62         return (unsigned long)t;
  63 }
  64
  65 static unsigned long
  66 nilfs_sufile_get_offset(const struct inode *sufile, __u64 segnum)
  67 {
  68         __u64 t = segnum + NILFS_MDT(sufile)->mi_first_entry_offset;
  69
  70         return do_div(t, nilfs_sufile_segment_usages_per_block(sufile));
  71 }
  72
  73 static unsigned long
  74 nilfs_sufile_segment_usages_in_block(const struct inode *sufile, __u64 curr,
  75                                      __u64 max)
  76 {
  77         return min_t(unsigned long,
  78                      nilfs_sufile_segment_usages_per_block(sufile) -
  79                      nilfs_sufile_get_offset(sufile, curr),
  80                      max - curr + 1);
  81 }
  82
  83 static struct nilfs_segment_usage *
  84 nilfs_sufile_block_get_segment_usage(const struct inode *sufile, __u64 segnum,
  85                                      struct buffer_head *bh, void *kaddr)
  86 {
  87         return kaddr + bh_offset(bh) +
  88                 nilfs_sufile_get_offset(sufile, segnum) *
  89                 NILFS_MDT(sufile)->mi_entry_size;
  90 }
  91
  92 static inline int nilfs_sufile_get_header_block(struct inode *sufile,
  93                                                 struct buffer_head **bhp)
  94 {
  95         return nilfs_mdt_get_block(sufile, 0, 0, NULL, bhp);
  96 }
  97
  98 static inline int
  99 nilfs_sufile_get_segment_usage_block(struct inode *sufile, __u64 segnum,
 100                                      int create, struct buffer_head **bhp)
 101 {
 102         return nilfs_mdt_get_block(sufile,
 103                                    nilfs_sufile_get_blkoff(sufile, segnum),
 104                                    create, NULL, bhp);
 105 }
 106
 107 static int nilfs_sufile_delete_segment_usage_block(struct inode *sufile,
 108                                                    __u64 segnum)
 109 {
 110         return nilfs_mdt_delete_block(sufile,
 111                                       nilfs_sufile_get_blkoff(sufile, segnum));
 112 }
 113
 114 static void nilfs_sufile_mod_counter(struct buffer_head *header_bh,
 115                                      u64 ncleanadd, u64 ndirtyadd)
 116 {
 117         struct nilfs_sufile_header *header;
 118         void *kaddr;
 119
 120         kaddr = kmap_atomic(header_bh->b_page);
 121         header = kaddr + bh_offset(header_bh);
 122         le64_add_cpu(&header->sh_ncleansegs, ncleanadd);
 123         le64_add_cpu(&header->sh_ndirtysegs, ndirtyadd);
 124         kunmap_atomic(kaddr);
 125
 126         mark_buffer_dirty(header_bh);
 127 }
 128
 129 /**
 130  * nilfs_sufile_get_ncleansegs - return the number of clean segments
 131  * @sufile: inode of segment usage file
 132  */
 133 unsigned long nilfs_sufile_get_ncleansegs(struct inode *sufile)
 134 {
 135         return NILFS_SUI(sufile)->ncleansegs;
 136 }
 137
 138 /**
 139  * nilfs_sufile_updatev - modify multiple segment usages at a time
 140  * @sufile: inode of segment usage file
 141  * @segnumv: array of segment numbers
 142  * @nsegs: size of @segnumv array
 143  * @create: creation flag
 144  * @ndone: place to store number of modified segments on @segnumv
 145  * @dofunc: primitive operation for the update
 146  *
 147  * Description: nilfs_sufile_updatev() repeatedly calls @dofunc
 148  * against the given array of segments.  The @dofunc is called with
 149  * buffers of a header block and the sufile block in which the target
 150  * segment usage entry is contained.  If @ndone is given, the number
 151  * of successfully modified segments from the head is stored in the
 152  * place @ndone points to.
 153  *
 154  * Return Value: On success, zero is returned.  On error, one of the
 155  * following negative error codes is returned.
 156  *
 157  * %-EIO - I/O error.
 158  *
 159  * %-ENOMEM - Insufficient amount of memory available.
 160  *
 161  * %-ENOENT - Given segment usage is in hole block (may be returned if
 162  *            @create is zero)
 163  *
 164  * %-EINVAL - Invalid segment usage number
 165  */
 166 int nilfs_sufile_updatev(struct inode *sufile, __u64 *segnumv, size_t nsegs,
 167                          int create, size_t *ndone,
 168                          void (*dofunc)(struct inode *, __u64,
 169                                         struct buffer_head *,
 170                                         struct buffer_head *))
 171 {
 172         struct buffer_head *header_bh, *bh;
 173         unsigned long blkoff, prev_blkoff;
 174         __u64 *seg;
 175         size_t nerr = 0, n = 0;
 176         int ret = 0;
 177
 178         if (unlikely(nsegs == 0))
 179                 goto out;
 180
 181         down_write(&NILFS_MDT(sufile)->mi_sem);
 182         for (seg = segnumv; seg < segnumv + nsegs; seg++) {
 183                 if (unlikely(*seg >= nilfs_sufile_get_nsegments(sufile))) {
 184                         printk(KERN_WARNING
 185                                "%s: invalid segment number: %llu\n", __func__,
 186                                (unsigned long long)*seg);
 187                         nerr++;
 188                 }
 189         }
 190         if (nerr > 0) {
 191                 ret = -EINVAL;
 192                 goto out_sem;
 193         }
 194
 195         ret = nilfs_sufile_get_header_block(sufile, &header_bh);
 196         if (ret < 0)
 197                 goto out_sem;
 198
 199         seg = segnumv;
 200         blkoff = nilfs_sufile_get_blkoff(sufile, *seg);
 201         ret = nilfs_mdt_get_block(sufile, blkoff, create, NULL, &bh);
 202         if (ret < 0)
 203                 goto out_header;
 204
 205         for (;;) {
 206                 dofunc(sufile, *seg, header_bh, bh);
 207
 208                 if (++seg >= segnumv + nsegs)
 209                         break;
 210                 prev_blkoff = blkoff;
 211                 blkoff = nilfs_sufile_get_blkoff(sufile, *seg);
 212                 if (blkoff == prev_blkoff)
 213                         continue;
 214
 215                 /* get different block */
 216                 brelse(bh);
 217                 ret = nilfs_mdt_get_block(sufile, blkoff, create, NULL, &bh);
 218                 if (unlikely(ret < 0))
 219                         goto out_header;
 220         }
 221         brelse(bh);
 222
 223  out_header:
 224         n = seg - segnumv;
 225         brelse(header_bh);
 226  out_sem:
 227         up_write(&NILFS_MDT(sufile)->mi_sem);
 228  out:
 229         if (ndone)
 230                 *ndone = n;
 231         return ret;
 232 }
 233
 234 int nilfs_sufile_update(struct inode *sufile, __u64 segnum, int create,
 235                         void (*dofunc)(struct inode *, __u64,
 236                                        struct buffer_head *,
 237                                        struct buffer_head *))
 238 {
 239         struct buffer_head *header_bh, *bh;
 240         int ret;
 241
 242         if (unlikely(segnum >= nilfs_sufile_get_nsegments(sufile))) {
 243                 printk(KERN_WARNING "%s: invalid segment number: %llu\n",
 244                        __func__, (unsigned long long)segnum);
 245                 return -EINVAL;
 246         }
 247         down_write(&NILFS_MDT(sufile)->mi_sem);
 248
 249         ret = nilfs_sufile_get_header_block(sufile, &header_bh);
 250         if (ret < 0)
 251                 goto out_sem;
 252
 253         ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, create, &bh);
 254         if (!ret) {
 255                 dofunc(sufile, segnum, header_bh, bh);
 256                 brelse(bh);
 257         }
 258         brelse(header_bh);
 259
 260  out_sem:
 261         up_write(&NILFS_MDT(sufile)->mi_sem);
 262         return ret;
 263 }
 264
 265 /**
 266  * nilfs_sufile_set_alloc_range - limit range of segment to be allocated
 267  * @sufile: inode of segment usage file
 268  * @start: minimum segment number of allocatable region (inclusive)
 269  * @end: maximum segment number of allocatable region (inclusive)
 270  *
 271  * Return Value: On success, 0 is returned.  On error, one of the
 272  * following negative error codes is returned.
 273  *
 274  * %-ERANGE - invalid segment region
 275  */
 276 int nilfs_sufile_set_alloc_range(struct inode *sufile, __u64 start, __u64 end)
 277 {
 278         struct nilfs_sufile_info *sui = NILFS_SUI(sufile);
 279         __u64 nsegs;
 280         int ret = -ERANGE;
 281
 282         down_write(&NILFS_MDT(sufile)->mi_sem);
 283         nsegs = nilfs_sufile_get_nsegments(sufile);
 284
 285         if (start <= end && end < nsegs) {
 286                 sui->allocmin = start;
 287                 sui->allocmax = end;
 288                 ret = 0;
 289         }
 290         up_write(&NILFS_MDT(sufile)->mi_sem);
 291         return ret;
 292 }
 293
 294 /**
 295  * nilfs_sufile_alloc - allocate a segment
 296  * @sufile: inode of segment usage file
 297  * @segnump: pointer to segment number
 298  *
 299  * Description: nilfs_sufile_alloc() allocates a clean segment.
 300  *
 301  * Return Value: On success, 0 is returned and the segment number of the
 302  * allocated segment is stored in the place pointed by @segnump. On error, one
 303  * of the following negative error codes is returned.
 304  *
 305  * %-EIO - I/O error.
 306  *
 307  * %-ENOMEM - Insufficient amount of memory available.
 308  *
 309  * %-ENOSPC - No clean segment left.
 310  */
 311 int nilfs_sufile_alloc(struct inode *sufile, __u64 *segnump)
 312 {
 313         struct buffer_head *header_bh, *su_bh;
 314         struct nilfs_sufile_header *header;
 315         struct nilfs_segment_usage *su;
 316         struct nilfs_sufile_info *sui = NILFS_SUI(sufile);
 317         size_t susz = NILFS_MDT(sufile)->mi_entry_size;
 318         __u64 segnum, maxsegnum, last_alloc;
 319         void *kaddr;
 320         unsigned long nsegments, nsus, cnt;
 321         int ret, j;
 322
 323         down_write(&NILFS_MDT(sufile)->mi_sem);
 324
 325         ret = nilfs_sufile_get_header_block(sufile, &header_bh);
 326         if (ret < 0)
 327                 goto out_sem;
 328         kaddr = kmap_atomic(header_bh->b_page);
 329         header = kaddr + bh_offset(header_bh);
 330         last_alloc = le64_to_cpu(header->sh_last_alloc);
 331         kunmap_atomic(kaddr);
 332
 333         nsegments = nilfs_sufile_get_nsegments(sufile);
 334         maxsegnum = sui->allocmax;
 335         segnum = last_alloc + 1;
 336         if (segnum < sui->allocmin || segnum > sui->allocmax)
 337                 segnum = sui->allocmin;
 338
 339         for (cnt = 0; cnt < nsegments; cnt += nsus) {
 340                 if (segnum > maxsegnum) {
 341                         if (cnt < sui->allocmax - sui->allocmin + 1) {
 342                                 /*
 343                                  * wrap around in the limited region.
 344                                  * if allocation started from
 345                                  * sui->allocmin, this never happens.
 346                                  */
 347                                 segnum = sui->allocmin;
 348                                 maxsegnum = last_alloc;
 349                         } else if (segnum > sui->allocmin &&
 350                                    sui->allocmax + 1 < nsegments) {
 351                                 segnum = sui->allocmax + 1;
 352                                 maxsegnum = nsegments - 1;
 353                         } else if (sui->allocmin > 0)  {
 354                                 segnum = 0;
 355                                 maxsegnum = sui->allocmin - 1;
 356                         } else {
 357                                 break; /* never happens */
 358                         }
 359                 }
 360                 trace_nilfs2_segment_usage_check(sufile, segnum, cnt);
 361                 ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 1,
 362                                                            &su_bh);
 363                 if (ret < 0)
 364                         goto out_header;
 365                 kaddr = kmap_atomic(su_bh->b_page);
 366                 su = nilfs_sufile_block_get_segment_usage(
 367                         sufile, segnum, su_bh, kaddr);
 368
 369                 nsus = nilfs_sufile_segment_usages_in_block(
 370                         sufile, segnum, maxsegnum);
 371                 for (j = 0; j < nsus; j++, su = (void *)su + susz, segnum++) {
 372                         if (!nilfs_segment_usage_clean(su))
 373                                 continue;
 374                         /* found a clean segment */
 375                         nilfs_segment_usage_set_dirty(su);
 376                         kunmap_atomic(kaddr);
 377
 378                         kaddr = kmap_atomic(header_bh->b_page);
 379                         header = kaddr + bh_offset(header_bh);
 380                         le64_add_cpu(&header->sh_ncleansegs, -1);
 381                         le64_add_cpu(&header->sh_ndirtysegs, 1);
 382                         header->sh_last_alloc = cpu_to_le64(segnum);
 383                         kunmap_atomic(kaddr);
 384
 385                         sui->ncleansegs--;
 386                         mark_buffer_dirty(header_bh);
 387                         mark_buffer_dirty(su_bh);
 388                         nilfs_mdt_mark_dirty(sufile);
 389                         brelse(su_bh);
 390                         *segnump = segnum;
 391
 392                         trace_nilfs2_segment_usage_allocated(sufile, segnum);
 393
 394                         goto out_header;
 395                 }
 396
 397                 kunmap_atomic(kaddr);
 398                 brelse(su_bh);
 399         }
 400
 401         /* no segments left */
 402         ret = -ENOSPC;
 403
 404  out_header:
 405         brelse(header_bh);
 406
 407  out_sem:
 408         up_write(&NILFS_MDT(sufile)->mi_sem);
 409         return ret;
 410 }
 411
 412 void nilfs_sufile_do_cancel_free(struct inode *sufile, __u64 segnum,
 413                                  struct buffer_head *header_bh,
 414                                  struct buffer_head *su_bh)
 415 {
 416         struct nilfs_segment_usage *su;
 417         void *kaddr;
 418
 419         kaddr = kmap_atomic(su_bh->b_page);
 420         su = nilfs_sufile_block_get_segment_usage(sufile, segnum, su_bh, kaddr);
 421         if (unlikely(!nilfs_segment_usage_clean(su))) {
 422                 printk(KERN_WARNING "%s: segment %llu must be clean\n",
 423                        __func__, (unsigned long long)segnum);
 424                 kunmap_atomic(kaddr);
 425                 return;
 426         }
 427         nilfs_segment_usage_set_dirty(su);
 428         kunmap_atomic(kaddr);
 429
 430         nilfs_sufile_mod_counter(header_bh, -1, 1);
 431         NILFS_SUI(sufile)->ncleansegs--;
 432
 433         mark_buffer_dirty(su_bh);
 434         nilfs_mdt_mark_dirty(sufile);
 435 }
 436
 437 void nilfs_sufile_do_scrap(struct inode *sufile, __u64 segnum,
 438                            struct buffer_head *header_bh,
 439                            struct buffer_head *su_bh)
 440 {
 441         struct nilfs_segment_usage *su;
 442         void *kaddr;
 443         int clean, dirty;
 444
 445         kaddr = kmap_atomic(su_bh->b_page);
 446         su = nilfs_sufile_block_get_segment_usage(sufile, segnum, su_bh, kaddr);
 447         if (su->su_flags == cpu_to_le32(1UL << NILFS_SEGMENT_USAGE_DIRTY) &&
 448             su->su_nblocks == cpu_to_le32(0)) {
 449                 kunmap_atomic(kaddr);
 450                 return;
 451         }
 452         clean = nilfs_segment_usage_clean(su);
 453         dirty = nilfs_segment_usage_dirty(su);
 454
 455         /* make the segment garbage */
 456         su->su_lastmod = cpu_to_le64(0);
 457         su->su_nblocks = cpu_to_le32(0);
 458         su->su_flags = cpu_to_le32(1UL << NILFS_SEGMENT_USAGE_DIRTY);
 459         kunmap_atomic(kaddr);
 460
 461         nilfs_sufile_mod_counter(header_bh, clean ? (u64)-1 : 0, dirty ? 0 : 1);
 462         NILFS_SUI(sufile)->ncleansegs -= clean;
 463
 464         mark_buffer_dirty(su_bh);
 465         nilfs_mdt_mark_dirty(sufile);
 466 }
 467
 468 void nilfs_sufile_do_free(struct inode *sufile, __u64 segnum,
 469                           struct buffer_head *header_bh,
 470                           struct buffer_head *su_bh)
 471 {
 472         struct nilfs_segment_usage *su;
 473         void *kaddr;
 474         int sudirty;
 475
 476         kaddr = kmap_atomic(su_bh->b_page);
 477         su = nilfs_sufile_block_get_segment_usage(sufile, segnum, su_bh, kaddr);
 478         if (nilfs_segment_usage_clean(su)) {
 479                 printk(KERN_WARNING "%s: segment %llu is already clean\n",
 480                        __func__, (unsigned long long)segnum);
 481                 kunmap_atomic(kaddr);
 482                 return;
 483         }
 484         WARN_ON(nilfs_segment_usage_error(su));
 485         WARN_ON(!nilfs_segment_usage_dirty(su));
 486
 487         sudirty = nilfs_segment_usage_dirty(su);
 488         nilfs_segment_usage_set_clean(su);
 489         kunmap_atomic(kaddr);
 490         mark_buffer_dirty(su_bh);
 491
 492         nilfs_sufile_mod_counter(header_bh, 1, sudirty ? (u64)-1 : 0);
 493         NILFS_SUI(sufile)->ncleansegs++;
 494
 495         nilfs_mdt_mark_dirty(sufile);
 496
 497         trace_nilfs2_segment_usage_freed(sufile, segnum);
 498 }
 499
 500 /**
 501  * nilfs_sufile_mark_dirty - mark the buffer having a segment usage dirty
 502  * @sufile: inode of segment usage file
 503  * @segnum: segment number
 504  */
 505 int nilfs_sufile_mark_dirty(struct inode *sufile, __u64 segnum)
 506 {
 507         struct buffer_head *bh;
 508         int ret;
 509
 510         ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0, &bh);
 511         if (!ret) {
 512                 mark_buffer_dirty(bh);
 513                 nilfs_mdt_mark_dirty(sufile);
 514                 brelse(bh);
 515         }
 516         return ret;
 517 }
 518
 519 /**
 520  * nilfs_sufile_set_segment_usage - set usage of a segment
 521  * @sufile: inode of segment usage file
 522  * @segnum: segment number
 523  * @nblocks: number of live blocks in the segment
 524  * @modtime: modification time (option)
 525  */
 526 int nilfs_sufile_set_segment_usage(struct inode *sufile, __u64 segnum,
 527                                    unsigned long nblocks, time_t modtime)
 528 {
 529         struct buffer_head *bh;
 530         struct nilfs_segment_usage *su;
 531         void *kaddr;
 532         int ret;
 533
 534         down_write(&NILFS_MDT(sufile)->mi_sem);
 535         ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0, &bh);
 536         if (ret < 0)
 537                 goto out_sem;
 538
 539         kaddr = kmap_atomic(bh->b_page);
 540         su = nilfs_sufile_block_get_segment_usage(sufile, segnum, bh, kaddr);
 541         WARN_ON(nilfs_segment_usage_error(su));
 542         if (modtime)
 543                 su->su_lastmod = cpu_to_le64(modtime);
 544         su->su_nblocks = cpu_to_le32(nblocks);
 545         kunmap_atomic(kaddr);
 546
 547         mark_buffer_dirty(bh);
 548         nilfs_mdt_mark_dirty(sufile);
 549         brelse(bh);
 550
 551  out_sem:
 552         up_write(&NILFS_MDT(sufile)->mi_sem);
 553         return ret;
 554 }
 555
 556 /**
 557  * nilfs_sufile_get_stat - get segment usage statistics
 558  * @sufile: inode of segment usage file
 559  * @stat: pointer to a structure of segment usage statistics
 560  *
 561  * Description: nilfs_sufile_get_stat() returns information about segment
 562  * usage.
 563  *
 564  * Return Value: On success, 0 is returned, and segment usage information is
 565  * stored in the place pointed by @stat. On error, one of the following
 566  * negative error codes is returned.
 567  *
 568  * %-EIO - I/O error.
 569  *
 570  * %-ENOMEM - Insufficient amount of memory available.
 571  */
 572 int nilfs_sufile_get_stat(struct inode *sufile, struct nilfs_sustat *sustat)
 573 {
 574         struct buffer_head *header_bh;
 575         struct nilfs_sufile_header *header;
 576         struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
 577         void *kaddr;
 578         int ret;
 579
 580         down_read(&NILFS_MDT(sufile)->mi_sem);
 581
 582         ret = nilfs_sufile_get_header_block(sufile, &header_bh);
 583         if (ret < 0)
 584                 goto out_sem;
 585
 586         kaddr = kmap_atomic(header_bh->b_page);
 587         header = kaddr + bh_offset(header_bh);
 588         sustat->ss_nsegs = nilfs_sufile_get_nsegments(sufile);
 589         sustat->ss_ncleansegs = le64_to_cpu(header->sh_ncleansegs);
 590         sustat->ss_ndirtysegs = le64_to_cpu(header->sh_ndirtysegs);
 591         sustat->ss_ctime = nilfs->ns_ctime;
 592         sustat->ss_nongc_ctime = nilfs->ns_nongc_ctime;
 593         spin_lock(&nilfs->ns_last_segment_lock);
 594         sustat->ss_prot_seq = nilfs->ns_prot_seq;
 595         spin_unlock(&nilfs->ns_last_segment_lock);
 596         kunmap_atomic(kaddr);
 597         brelse(header_bh);
 598
 599  out_sem:
 600         up_read(&NILFS_MDT(sufile)->mi_sem);
 601         return ret;
 602 }
 603
 604 void nilfs_sufile_do_set_error(struct inode *sufile, __u64 segnum,
 605                                struct buffer_head *header_bh,
 606                                struct buffer_head *su_bh)
 607 {
 608         struct nilfs_segment_usage *su;
 609         void *kaddr;
 610         int suclean;
 611
 612         kaddr = kmap_atomic(su_bh->b_page);
 613         su = nilfs_sufile_block_get_segment_usage(sufile, segnum, su_bh, kaddr);
 614         if (nilfs_segment_usage_error(su)) {
 615                 kunmap_atomic(kaddr);
 616                 return;
 617         }
 618         suclean = nilfs_segment_usage_clean(su);
 619         nilfs_segment_usage_set_error(su);
 620         kunmap_atomic(kaddr);
 621
 622         if (suclean) {
 623                 nilfs_sufile_mod_counter(header_bh, -1, 0);
 624                 NILFS_SUI(sufile)->ncleansegs--;
 625         }
 626         mark_buffer_dirty(su_bh);
 627         nilfs_mdt_mark_dirty(sufile);
 628 }
 629
 630 /**
 631   * nilfs_sufile_truncate_range - truncate range of segment array
 632   * @sufile: inode of segment usage file
 633   * @start: start segment number (inclusive)
 634   * @end: end segment number (inclusive)
 635   *
 636   * Return Value: On success, 0 is returned.  On error, one of the
 637   * following negative error codes is returned.
 638   *
 639   * %-EIO - I/O error.
 640   *
 641   * %-ENOMEM - Insufficient amount of memory available.
 642   *
 643   * %-EINVAL - Invalid number of segments specified
 644   *
 645   * %-EBUSY - Dirty or active segments are present in the range
 646   */
 647 static int nilfs_sufile_truncate_range(struct inode *sufile,
 648                                        __u64 start, __u64 end)
 649 {
 650         struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
 651         struct buffer_head *header_bh;
 652         struct buffer_head *su_bh;
 653         struct nilfs_segment_usage *su, *su2;
 654         size_t susz = NILFS_MDT(sufile)->mi_entry_size;
 655         unsigned long segusages_per_block;
 656         unsigned long nsegs, ncleaned;
 657         __u64 segnum;
 658         void *kaddr;
 659         ssize_t n, nc;
 660         int ret;
 661         int j;
 662
 663         nsegs = nilfs_sufile_get_nsegments(sufile);
 664
 665         ret = -EINVAL;
 666         if (start > end || start >= nsegs)
 667                 goto out;
 668
 669         ret = nilfs_sufile_get_header_block(sufile, &header_bh);
 670         if (ret < 0)
 671                 goto out;
 672
 673         segusages_per_block = nilfs_sufile_segment_usages_per_block(sufile);
 674         ncleaned = 0;
 675
 676         for (segnum = start; segnum <= end; segnum += n) {
 677                 n = min_t(unsigned long,
 678                           segusages_per_block -
 679                                   nilfs_sufile_get_offset(sufile, segnum),
 680                           end - segnum + 1);
 681                 ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0,
 682                                                            &su_bh);
 683                 if (ret < 0) {
 684                         if (ret != -ENOENT)
 685                                 goto out_header;
 686                         /* hole */
 687                         continue;
 688                 }
 689                 kaddr = kmap_atomic(su_bh->b_page);
 690                 su = nilfs_sufile_block_get_segment_usage(
 691                         sufile, segnum, su_bh, kaddr);
 692                 su2 = su;
 693                 for (j = 0; j < n; j++, su = (void *)su + susz) {
 694                         if ((le32_to_cpu(su->su_flags) &
 695                              ~(1UL << NILFS_SEGMENT_USAGE_ERROR)) ||
 696                             nilfs_segment_is_active(nilfs, segnum + j)) {
 697                                 ret = -EBUSY;
 698                                 kunmap_atomic(kaddr);
 699                                 brelse(su_bh);
 700                                 goto out_header;
 701                         }
 702                 }
 703                 nc = 0;
 704                 for (su = su2, j = 0; j < n; j++, su = (void *)su + susz) {
 705                         if (nilfs_segment_usage_error(su)) {
 706                                 nilfs_segment_usage_set_clean(su);
 707                                 nc++;
 708                         }
 709                 }
 710                 kunmap_atomic(kaddr);
 711                 if (nc > 0) {
 712                         mark_buffer_dirty(su_bh);
 713                         ncleaned += nc;
 714                 }
 715                 brelse(su_bh);
 716
 717                 if (n == segusages_per_block) {
 718                         /* make hole */
 719                         nilfs_sufile_delete_segment_usage_block(sufile, segnum);
 720                 }
 721         }
 722         ret = 0;
 723
 724 out_header:
 725         if (ncleaned > 0) {
 726                 NILFS_SUI(sufile)->ncleansegs += ncleaned;
 727                 nilfs_sufile_mod_counter(header_bh, ncleaned, 0);
 728                 nilfs_mdt_mark_dirty(sufile);
 729         }
 730         brelse(header_bh);
 731 out:
 732         return ret;
 733 }
 734
 735 /**
 736  * nilfs_sufile_resize - resize segment array
 737  * @sufile: inode of segment usage file
 738  * @newnsegs: new number of segments
 739  *
 740  * Return Value: On success, 0 is returned.  On error, one of the
 741  * following negative error codes is returned.
 742  *
 743  * %-EIO - I/O error.
 744  *
 745  * %-ENOMEM - Insufficient amount of memory available.
 746  *
 747  * %-ENOSPC - Enough free space is not left for shrinking
 748  *
 749  * %-EBUSY - Dirty or active segments exist in the region to be truncated
 750  */
 751 int nilfs_sufile_resize(struct inode *sufile, __u64 newnsegs)
 752 {
 753         struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
 754         struct buffer_head *header_bh;
 755         struct nilfs_sufile_header *header;
 756         struct nilfs_sufile_info *sui = NILFS_SUI(sufile);
 757         void *kaddr;
 758         unsigned long nsegs, nrsvsegs;
 759         int ret = 0;
 760
 761         down_write(&NILFS_MDT(sufile)->mi_sem);
 762
 763         nsegs = nilfs_sufile_get_nsegments(sufile);
 764         if (nsegs == newnsegs)
 765                 goto out;
 766
 767         ret = -ENOSPC;
 768         nrsvsegs = nilfs_nrsvsegs(nilfs, newnsegs);
 769         if (newnsegs < nsegs && nsegs - newnsegs + nrsvsegs > sui->ncleansegs)
 770                 goto out;
 771
 772         ret = nilfs_sufile_get_header_block(sufile, &header_bh);
 773         if (ret < 0)
 774                 goto out;
 775
 776         if (newnsegs > nsegs) {
 777                 sui->ncleansegs += newnsegs - nsegs;
 778         } else /* newnsegs < nsegs */ {
 779                 ret = nilfs_sufile_truncate_range(sufile, newnsegs, nsegs - 1);
 780                 if (ret < 0)
 781                         goto out_header;
 782
 783                 sui->ncleansegs -= nsegs - newnsegs;
 784         }
 785
 786         kaddr = kmap_atomic(header_bh->b_page);
 787         header = kaddr + bh_offset(header_bh);
 788         header->sh_ncleansegs = cpu_to_le64(sui->ncleansegs);
 789         kunmap_atomic(kaddr);
 790
 791         mark_buffer_dirty(header_bh);
 792         nilfs_mdt_mark_dirty(sufile);
 793         nilfs_set_nsegments(nilfs, newnsegs);
 794
 795 out_header:
 796         brelse(header_bh);
 797 out:
 798         up_write(&NILFS_MDT(sufile)->mi_sem);
 799         return ret;
 800 }
 801
 802 /**
 803  * nilfs_sufile_get_suinfo -
 804  * @sufile: inode of segment usage file
 805  * @segnum: segment number to start looking
 806  * @buf: array of suinfo
 807  * @sisz: byte size of suinfo
 808  * @nsi: size of suinfo array
 809  *
 810  * Description:
 811  *
 812  * Return Value: On success, 0 is returned and .... On error, one of the
 813  * following negative error codes is returned.
 814  *
 815  * %-EIO - I/O error.
 816  *
 817  * %-ENOMEM - Insufficient amount of memory available.
 818  */
 819 ssize_t nilfs_sufile_get_suinfo(struct inode *sufile, __u64 segnum, void *buf,
 820                                 unsigned int sisz, size_t nsi)
 821 {
 822         struct buffer_head *su_bh;
 823         struct nilfs_segment_usage *su;
 824         struct nilfs_suinfo *si = buf;
 825         size_t susz = NILFS_MDT(sufile)->mi_entry_size;
 826         struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
 827         void *kaddr;
 828         unsigned long nsegs, segusages_per_block;
 829         ssize_t n;
 830         int ret, i, j;
 831
 832         down_read(&NILFS_MDT(sufile)->mi_sem);
 833
 834         segusages_per_block = nilfs_sufile_segment_usages_per_block(sufile);
 835         nsegs = min_t(unsigned long,
 836                       nilfs_sufile_get_nsegments(sufile) - segnum,
 837                       nsi);
 838         for (i = 0; i < nsegs; i += n, segnum += n) {
 839                 n = min_t(unsigned long,
 840                           segusages_per_block -
 841                                   nilfs_sufile_get_offset(sufile, segnum),
 842                           nsegs - i);
 843                 ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0,
 844                                                            &su_bh);
 845                 if (ret < 0) {
 846                         if (ret != -ENOENT)
 847                                 goto out;
 848                         /* hole */
 849                         memset(si, 0, sisz * n);
 850                         si = (void *)si + sisz * n;
 851                         continue;
 852                 }
 853
 854                 kaddr = kmap_atomic(su_bh->b_page);
 855                 su = nilfs_sufile_block_get_segment_usage(
 856                         sufile, segnum, su_bh, kaddr);
 857                 for (j = 0; j < n;
 858                      j++, su = (void *)su + susz, si = (void *)si + sisz) {
 859                         si->sui_lastmod = le64_to_cpu(su->su_lastmod);
 860                         si->sui_nblocks = le32_to_cpu(su->su_nblocks);
 861                         si->sui_flags = le32_to_cpu(su->su_flags) &
 862                                 ~(1UL << NILFS_SEGMENT_USAGE_ACTIVE);
 863                         if (nilfs_segment_is_active(nilfs, segnum + j))
 864                                 si->sui_flags |=
 865                                         (1UL << NILFS_SEGMENT_USAGE_ACTIVE);
 866                 }
 867                 kunmap_atomic(kaddr);
 868                 brelse(su_bh);
 869         }
 870         ret = nsegs;
 871
 872  out:
 873         up_read(&NILFS_MDT(sufile)->mi_sem);
 874         return ret;
 875 }
 876
 877 /**
 878  * nilfs_sufile_set_suinfo - sets segment usage info
 879  * @sufile: inode of segment usage file
 880  * @buf: array of suinfo_update
 881  * @supsz: byte size of suinfo_update
 882  * @nsup: size of suinfo_update array
 883  *
 884  * Description: Takes an array of nilfs_suinfo_update structs and updates
 885  * segment usage accordingly. Only the fields indicated by the sup_flags
 886  * are updated.
 887  *
 888  * Return Value: On success, 0 is returned. On error, one of the
 889  * following negative error codes is returned.
 890  *
 891  * %-EIO - I/O error.
 892  *
 893  * %-ENOMEM - Insufficient amount of memory available.
 894  *
 895  * %-EINVAL - Invalid values in input (segment number, flags or nblocks)
 896  */
 897 ssize_t nilfs_sufile_set_suinfo(struct inode *sufile, void *buf,
 898                                 unsigned int supsz, size_t nsup)
 899 {
 900         struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
 901         struct buffer_head *header_bh, *bh;
 902         struct nilfs_suinfo_update *sup, *supend = buf + supsz * nsup;
 903         struct nilfs_segment_usage *su;
 904         void *kaddr;
 905         unsigned long blkoff, prev_blkoff;
 906         int cleansi, cleansu, dirtysi, dirtysu;
 907         long ncleaned = 0, ndirtied = 0;
 908         int ret = 0;
 909
 910         if (unlikely(nsup == 0))
 911                 return ret;
 912
 913         for (sup = buf; sup < supend; sup = (void *)sup + supsz) {
 914                 if (sup->sup_segnum >= nilfs->ns_nsegments
 915                         || (sup->sup_flags &
 916                                 (~0UL << __NR_NILFS_SUINFO_UPDATE_FIELDS))
 917                         || (nilfs_suinfo_update_nblocks(sup) &&
 918                                 sup->sup_sui.sui_nblocks >
 919                                 nilfs->ns_blocks_per_segment))
 920                         return -EINVAL;
 921         }
 922
 923         down_write(&NILFS_MDT(sufile)->mi_sem);
 924
 925         ret = nilfs_sufile_get_header_block(sufile, &header_bh);
 926         if (ret < 0)
 927                 goto out_sem;
 928
 929         sup = buf;
 930         blkoff = nilfs_sufile_get_blkoff(sufile, sup->sup_segnum);
 931         ret = nilfs_mdt_get_block(sufile, blkoff, 1, NULL, &bh);
 932         if (ret < 0)
 933                 goto out_header;
 934
 935         for (;;) {
 936                 kaddr = kmap_atomic(bh->b_page);
 937                 su = nilfs_sufile_block_get_segment_usage(
 938                         sufile, sup->sup_segnum, bh, kaddr);
 939
 940                 if (nilfs_suinfo_update_lastmod(sup))
 941                         su->su_lastmod = cpu_to_le64(sup->sup_sui.sui_lastmod);
 942
 943                 if (nilfs_suinfo_update_nblocks(sup))
 944                         su->su_nblocks = cpu_to_le32(sup->sup_sui.sui_nblocks);
 945
 946                 if (nilfs_suinfo_update_flags(sup)) {
 947                         /*
 948                          * Active flag is a virtual flag projected by running
 949                          * nilfs kernel code - drop it not to write it to
 950                          * disk.
 951                          */
 952                         sup->sup_sui.sui_flags &=
 953                                         ~(1UL << NILFS_SEGMENT_USAGE_ACTIVE);
 954
 955                         cleansi = nilfs_suinfo_clean(&sup->sup_sui);
 956                         cleansu = nilfs_segment_usage_clean(su);
 957                         dirtysi = nilfs_suinfo_dirty(&sup->sup_sui);
 958                         dirtysu = nilfs_segment_usage_dirty(su);
 959
 960                         if (cleansi && !cleansu)
 961                                 ++ncleaned;
 962                         else if (!cleansi && cleansu)
 963                                 --ncleaned;
 964
 965                         if (dirtysi && !dirtysu)
 966                                 ++ndirtied;
 967                         else if (!dirtysi && dirtysu)
 968                                 --ndirtied;
 969
 970                         su->su_flags = cpu_to_le32(sup->sup_sui.sui_flags);
 971                 }
 972
 973                 kunmap_atomic(kaddr);
 974
 975                 sup = (void *)sup + supsz;
 976                 if (sup >= supend)
 977                         break;
 978
 979                 prev_blkoff = blkoff;
 980                 blkoff = nilfs_sufile_get_blkoff(sufile, sup->sup_segnum);
 981                 if (blkoff == prev_blkoff)
 982                         continue;
 983
 984                 /* get different block */
 985                 mark_buffer_dirty(bh);
 986                 put_bh(bh);
 987                 ret = nilfs_mdt_get_block(sufile, blkoff, 1, NULL, &bh);
 988                 if (unlikely(ret < 0))
 989                         goto out_mark;
 990         }
 991         mark_buffer_dirty(bh);
 992         put_bh(bh);
 993
 994  out_mark:
 995         if (ncleaned || ndirtied) {
 996                 nilfs_sufile_mod_counter(header_bh, (u64)ncleaned,
 997                                 (u64)ndirtied);
 998                 NILFS_SUI(sufile)->ncleansegs += ncleaned;
 999         }
1000         nilfs_mdt_mark_dirty(sufile);
1001  out_header:
1002         put_bh(header_bh);
1003  out_sem:
1004         up_write(&NILFS_MDT(sufile)->mi_sem);
1005         return ret;
1006 }
1007
1008 /**
1009  * nilfs_sufile_trim_fs() - trim ioctl handle function
1010  * @sufile: inode of segment usage file
1011  * @range: fstrim_range structure
1012  *
1013  * start:       First Byte to trim
1014  * len:         number of Bytes to trim from start
1015  * minlen:      minimum extent length in Bytes
1016  *
1017  * Decription: nilfs_sufile_trim_fs goes through all segments containing bytes
1018  * from start to start+len. start is rounded up to the next block boundary
1019  * and start+len is rounded down. For each clean segment blkdev_issue_discard
1020  * function is invoked.
1021  *
1022  * Return Value: On success, 0 is returned or negative error code, otherwise.
1023  */
1024 int nilfs_sufile_trim_fs(struct inode *sufile, struct fstrim_range *range)
1025 {
1026         struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
1027         struct buffer_head *su_bh;
1028         struct nilfs_segment_usage *su;
1029         void *kaddr;
1030         size_t n, i, susz = NILFS_MDT(sufile)->mi_entry_size;
1031         sector_t seg_start, seg_end, start_block, end_block;
1032         sector_t start = 0, nblocks = 0;
1033         u64 segnum, segnum_end, minlen, len, max_blocks, ndiscarded = 0;
1034         int ret = 0;
1035         unsigned int sects_per_block;
1036
1037         sects_per_block = (1 << nilfs->ns_blocksize_bits) /
1038                         bdev_logical_block_size(nilfs->ns_bdev);
1039         len = range->len >> nilfs->ns_blocksize_bits;
1040         minlen = range->minlen >> nilfs->ns_blocksize_bits;
1041         max_blocks = ((u64)nilfs->ns_nsegments * nilfs->ns_blocks_per_segment);
1042
1043         if (!len || range->start >= max_blocks << nilfs->ns_blocksize_bits)
1044                 return -EINVAL;
1045
1046         start_block = (range->start + nilfs->ns_blocksize - 1) >>
1047                         nilfs->ns_blocksize_bits;
1048
1049         /*
1050          * range->len can be very large (actually, it is set to
1051          * ULLONG_MAX by default) - truncate upper end of the range
1052          * carefully so as not to overflow.
1053          */
1054         if (max_blocks - start_block < len)
1055                 end_block = max_blocks - 1;
1056         else
1057                 end_block = start_block + len - 1;
1058
1059         segnum = nilfs_get_segnum_of_block(nilfs, start_block);
1060         segnum_end = nilfs_get_segnum_of_block(nilfs, end_block);
1061
1062         down_read(&NILFS_MDT(sufile)->mi_sem);
1063
1064         while (segnum <= segnum_end) {
1065                 n = nilfs_sufile_segment_usages_in_block(sufile, segnum,
1066                                 segnum_end);
1067
1068                 ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0,
1069                                                            &su_bh);
1070                 if (ret < 0) {
1071                         if (ret != -ENOENT)
1072                                 goto out_sem;
1073                         /* hole */
1074                         segnum += n;
1075                         continue;
1076                 }
1077
1078                 kaddr = kmap_atomic(su_bh->b_page);
1079                 su = nilfs_sufile_block_get_segment_usage(sufile, segnum,
1080                                 su_bh, kaddr);
1081                 for (i = 0; i < n; ++i, ++segnum, su = (void *)su + susz) {
1082                         if (!nilfs_segment_usage_clean(su))
1083                                 continue;
1084
1085                         nilfs_get_segment_range(nilfs, segnum, &seg_start,
1086                                                 &seg_end);
1087
1088                         if (!nblocks) {
1089                                 /* start new extent */
1090                                 start = seg_start;
1091                                 nblocks = seg_end - seg_start + 1;
1092                                 continue;
1093                         }
1094
1095                         if (start + nblocks == seg_start) {
1096                                 /* add to previous extent */
1097                                 nblocks += seg_end - seg_start + 1;
1098                                 continue;
1099                         }
1100
1101                         /* discard previous extent */
1102                         if (start < start_block) {
1103                                 nblocks -= start_block - start;
1104                                 start = start_block;
1105                         }
1106
1107                         if (nblocks >= minlen) {
1108                                 kunmap_atomic(kaddr);
1109
1110                                 ret = blkdev_issue_discard(nilfs->ns_bdev,
1111                                                 start * sects_per_block,
1112                                                 nblocks * sects_per_block,
1113                                                 GFP_NOFS, 0);
1114                                 if (ret < 0) {
1115                                         put_bh(su_bh);
1116                                         goto out_sem;
1117                                 }
1118
1119                                 ndiscarded += nblocks;
1120                                 kaddr = kmap_atomic(su_bh->b_page);
1121                                 su = nilfs_sufile_block_get_segment_usage(
1122                                         sufile, segnum, su_bh, kaddr);
1123                         }
1124
1125                         /* start new extent */
1126                         start = seg_start;
1127                         nblocks = seg_end - seg_start + 1;
1128                 }
1129                 kunmap_atomic(kaddr);
1130                 put_bh(su_bh);
1131         }
1132
1133
1134         if (nblocks) {
1135                 /* discard last extent */
1136                 if (start < start_block) {
1137                         nblocks -= start_block - start;
1138                         start = start_block;
1139                 }
1140                 if (start + nblocks > end_block + 1)
1141                         nblocks = end_block - start + 1;
1142
1143                 if (nblocks >= minlen) {
1144                         ret = blkdev_issue_discard(nilfs->ns_bdev,
1145                                         start * sects_per_block,
1146                                         nblocks * sects_per_block,
1147                                         GFP_NOFS, 0);
1148                         if (!ret)
1149                                 ndiscarded += nblocks;
1150                 }
1151         }
1152
1153 out_sem:
1154         up_read(&NILFS_MDT(sufile)->mi_sem);
1155
1156         range->len = ndiscarded << nilfs->ns_blocksize_bits;
1157         return ret;
1158 }
1159
1160 /**
1161  * nilfs_sufile_read - read or get sufile inode
1162  * @sb: super block instance
1163  * @susize: size of a segment usage entry
1164  * @raw_inode: on-disk sufile inode
1165  * @inodep: buffer to store the inode
1166  */
1167 int nilfs_sufile_read(struct super_block *sb, size_t susize,
1168                       struct nilfs_inode *raw_inode, struct inode **inodep)
1169 {
1170         struct inode *sufile;
1171         struct nilfs_sufile_info *sui;
1172         struct buffer_head *header_bh;
1173         struct nilfs_sufile_header *header;
1174         void *kaddr;
1175         int err;
1176
1177         if (susize > sb->s_blocksize) {
1178                 printk(KERN_ERR
1179                        "NILFS: too large segment usage size: %zu bytes.\n",
1180                        susize);
1181                 return -EINVAL;
1182         } else if (susize < NILFS_MIN_SEGMENT_USAGE_SIZE) {
1183                 printk(KERN_ERR
1184                        "NILFS: too small segment usage size: %zu bytes.\n",
1185                        susize);
1186                 return -EINVAL;
1187         }
1188
1189         sufile = nilfs_iget_locked(sb, NULL, NILFS_SUFILE_INO);
1190         if (unlikely(!sufile))
1191                 return -ENOMEM;
1192         if (!(sufile->i_state & I_NEW))
1193                 goto out;
1194
1195         err = nilfs_mdt_init(sufile, NILFS_MDT_GFP, sizeof(*sui));
1196         if (err)
1197                 goto failed;
1198
1199         nilfs_mdt_set_entry_size(sufile, susize,
1200                                  sizeof(struct nilfs_sufile_header));
1201
1202         err = nilfs_read_inode_common(sufile, raw_inode);
1203         if (err)
1204                 goto failed;
1205
1206         err = nilfs_sufile_get_header_block(sufile, &header_bh);
1207         if (err)
1208                 goto failed;
1209
1210         sui = NILFS_SUI(sufile);
1211         kaddr = kmap_atomic(header_bh->b_page);
1212         header = kaddr + bh_offset(header_bh);
1213         sui->ncleansegs = le64_to_cpu(header->sh_ncleansegs);
1214         kunmap_atomic(kaddr);
1215         brelse(header_bh);
1216
1217         sui->allocmax = nilfs_sufile_get_nsegments(sufile) - 1;
1218         sui->allocmin = 0;
1219
1220         unlock_new_inode(sufile);
1221  out:
1222         *inodep = sufile;
1223         return 0;
1224  failed:
1225         iget_failed(sufile);
1226         return err;
1227 }