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