fs/nilfs2/sufile.c

   1 // SPDX-License-Identifier: GPL-2.0+
   2 /*
   3  * sufile.c - NILFS segment usage file.
   4  *
   5  * Copyright (C) 2006-2008 Nippon Telegraph and Telephone Corporation.
   6  *
   7  * Written by Koji Sato.
   8  * Revised by Ryusuke Konishi.
   9  */
  10
  11 #include <linux/kernel.h>
  12 #include <linux/fs.h>
  13 #include <linux/string.h>
  14 #include <linux/buffer_head.h>
  15 #include <linux/errno.h>
  16 #include "mdt.h"
  17 #include "sufile.h"
  18
  19 #include <trace/events/nilfs2.h>
  20
  21 /**
  22  * struct nilfs_sufile_info - on-memory private data of sufile
  23  * @mi: on-memory private data of metadata file
  24  * @ncleansegs: number of clean segments
  25  * @allocmin: lower limit of allocatable segment range
  26  * @allocmax: upper limit of allocatable segment range
  27  */
  28 struct nilfs_sufile_info {
  29         struct nilfs_mdt_info mi;
  30         unsigned long ncleansegs;/* number of clean segments */
  31         __u64 allocmin;         /* lower limit of allocatable segment range */
  32         __u64 allocmax;         /* upper limit of allocatable segment range */
  33 };
  34
  35 static inline struct nilfs_sufile_info *NILFS_SUI(struct inode *sufile)
  36 {
  37         return (struct nilfs_sufile_info *)NILFS_MDT(sufile);
  38 }
  39
  40 static inline unsigned long
  41 nilfs_sufile_segment_usages_per_block(const struct inode *sufile)
  42 {
  43         return NILFS_MDT(sufile)->mi_entries_per_block;
  44 }
  45
  46 static unsigned long
  47 nilfs_sufile_get_blkoff(const struct inode *sufile, __u64 segnum)
  48 {
  49         __u64 t = segnum + NILFS_MDT(sufile)->mi_first_entry_offset;
  50
  51         do_div(t, nilfs_sufile_segment_usages_per_block(sufile));
  52         return (unsigned long)t;
  53 }
  54
  55 static unsigned long
  56 nilfs_sufile_get_offset(const struct inode *sufile, __u64 segnum)
  57 {
  58         __u64 t = segnum + NILFS_MDT(sufile)->mi_first_entry_offset;
  59
  60         return do_div(t, nilfs_sufile_segment_usages_per_block(sufile));
  61 }
  62
  63 static unsigned long
  64 nilfs_sufile_segment_usages_in_block(const struct inode *sufile, __u64 curr,
  65                                      __u64 max)
  66 {
  67         return min_t(unsigned long,
  68                      nilfs_sufile_segment_usages_per_block(sufile) -
  69                      nilfs_sufile_get_offset(sufile, curr),
  70                      max - curr + 1);
  71 }
  72
  73 static struct nilfs_segment_usage *
  74 nilfs_sufile_block_get_segment_usage(const struct inode *sufile, __u64 segnum,
  75                                      struct buffer_head *bh, void *kaddr)
  76 {
  77         return kaddr + bh_offset(bh) +
  78                 nilfs_sufile_get_offset(sufile, segnum) *
  79                 NILFS_MDT(sufile)->mi_entry_size;
  80 }
  81
  82 static inline int nilfs_sufile_get_header_block(struct inode *sufile,
  83                                                 struct buffer_head **bhp)
  84 {
  85         return nilfs_mdt_get_block(sufile, 0, 0, NULL, bhp);
  86 }
  87
  88 static inline int
  89 nilfs_sufile_get_segment_usage_block(struct inode *sufile, __u64 segnum,
  90                                      int create, struct buffer_head **bhp)
  91 {
  92         return nilfs_mdt_get_block(sufile,
  93                                    nilfs_sufile_get_blkoff(sufile, segnum),
  94                                    create, NULL, bhp);
  95 }
  96
  97 static int nilfs_sufile_delete_segment_usage_block(struct inode *sufile,
  98                                                    __u64 segnum)
  99 {
 100         return nilfs_mdt_delete_block(sufile,
 101                                       nilfs_sufile_get_blkoff(sufile, segnum));
 102 }
 103
 104 static void nilfs_sufile_mod_counter(struct buffer_head *header_bh,
 105                                      u64 ncleanadd, u64 ndirtyadd)
 106 {
 107         struct nilfs_sufile_header *header;
 108         void *kaddr;
 109
 110         kaddr = kmap_atomic(header_bh->b_page);
 111         header = kaddr + bh_offset(header_bh);
 112         le64_add_cpu(&header->sh_ncleansegs, ncleanadd);
 113         le64_add_cpu(&header->sh_ndirtysegs, ndirtyadd);
 114         kunmap_atomic(kaddr);
 115
 116         mark_buffer_dirty(header_bh);
 117 }
 118
 119 /**
 120  * nilfs_sufile_get_ncleansegs - return the number of clean segments
 121  * @sufile: inode of segment usage file
 122  */
 123 unsigned long nilfs_sufile_get_ncleansegs(struct inode *sufile)
 124 {
 125         return NILFS_SUI(sufile)->ncleansegs;
 126 }
 127
 128 /**
 129  * nilfs_sufile_updatev - modify multiple segment usages at a time
 130  * @sufile: inode of segment usage file
 131  * @segnumv: array of segment numbers
 132  * @nsegs: size of @segnumv array
 133  * @create: creation flag
 134  * @ndone: place to store number of modified segments on @segnumv
 135  * @dofunc: primitive operation for the update
 136  *
 137  * Description: nilfs_sufile_updatev() repeatedly calls @dofunc
 138  * against the given array of segments.  The @dofunc is called with
 139  * buffers of a header block and the sufile block in which the target
 140  * segment usage entry is contained.  If @ndone is given, the number
 141  * of successfully modified segments from the head is stored in the
 142  * place @ndone points to.
 143  *
 144  * Return Value: On success, zero is returned.  On error, one of the
 145  * following negative error codes is returned.
 146  *
 147  * %-EIO - I/O error.
 148  *
 149  * %-ENOMEM - Insufficient amount of memory available.
 150  *
 151  * %-ENOENT - Given segment usage is in hole block (may be returned if
 152  *            @create is zero)
 153  *
 154  * %-EINVAL - Invalid segment usage number
 155  */
 156 int nilfs_sufile_updatev(struct inode *sufile, __u64 *segnumv, size_t nsegs,
 157                          int create, size_t *ndone,
 158                          void (*dofunc)(struct inode *, __u64,
 159                                         struct buffer_head *,
 160                                         struct buffer_head *))
 161 {
 162         struct buffer_head *header_bh, *bh;
 163         unsigned long blkoff, prev_blkoff;
 164         __u64 *seg;
 165         size_t nerr = 0, n = 0;
 166         int ret = 0;
 167
 168         if (unlikely(nsegs == 0))
 169                 goto out;
 170
 171         down_write(&NILFS_MDT(sufile)->mi_sem);
 172         for (seg = segnumv; seg < segnumv + nsegs; seg++) {
 173                 if (unlikely(*seg >= nilfs_sufile_get_nsegments(sufile))) {
 174                         nilfs_msg(sufile->i_sb, KERN_WARNING,
 175                                   "%s: invalid segment number: %llu",
 176                                   __func__, (unsigned long long)*seg);
 177                         nerr++;
 178                 }
 179         }
 180         if (nerr > 0) {
 181                 ret = -EINVAL;
 182                 goto out_sem;
 183         }
 184
 185         ret = nilfs_sufile_get_header_block(sufile, &header_bh);
 186         if (ret < 0)
 187                 goto out_sem;
 188
 189         seg = segnumv;
 190         blkoff = nilfs_sufile_get_blkoff(sufile, *seg);
 191         ret = nilfs_mdt_get_block(sufile, blkoff, create, NULL, &bh);
 192         if (ret < 0)
 193                 goto out_header;
 194
 195         for (;;) {
 196                 dofunc(sufile, *seg, header_bh, bh);
 197
 198                 if (++seg >= segnumv + nsegs)
 199                         break;
 200                 prev_blkoff = blkoff;
 201                 blkoff = nilfs_sufile_get_blkoff(sufile, *seg);
 202                 if (blkoff == prev_blkoff)
 203                         continue;
 204
 205                 /* get different block */
 206                 brelse(bh);
 207                 ret = nilfs_mdt_get_block(sufile, blkoff, create, NULL, &bh);
 208                 if (unlikely(ret < 0))
 209                         goto out_header;
 210         }
 211         brelse(bh);
 212
 213  out_header:
 214         n = seg - segnumv;
 215         brelse(header_bh);
 216  out_sem:
 217         up_write(&NILFS_MDT(sufile)->mi_sem);
 218  out:
 219         if (ndone)
 220                 *ndone = n;
 221         return ret;
 222 }
 223
 224 int nilfs_sufile_update(struct inode *sufile, __u64 segnum, int create,
 225                         void (*dofunc)(struct inode *, __u64,
 226                                        struct buffer_head *,
 227                                        struct buffer_head *))
 228 {
 229         struct buffer_head *header_bh, *bh;
 230         int ret;
 231
 232         if (unlikely(segnum >= nilfs_sufile_get_nsegments(sufile))) {
 233                 nilfs_msg(sufile->i_sb, KERN_WARNING,
 234                           "%s: invalid segment number: %llu",
 235                           __func__, (unsigned long long)segnum);
 236                 return -EINVAL;
 237         }
 238         down_write(&NILFS_MDT(sufile)->mi_sem);
 239
 240         ret = nilfs_sufile_get_header_block(sufile, &header_bh);
 241         if (ret < 0)
 242                 goto out_sem;
 243
 244         ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, create, &bh);
 245         if (!ret) {
 246                 dofunc(sufile, segnum, header_bh, bh);
 247                 brelse(bh);
 248         }
 249         brelse(header_bh);
 250
 251  out_sem:
 252         up_write(&NILFS_MDT(sufile)->mi_sem);
 253         return ret;
 254 }
 255
 256 /**
 257  * nilfs_sufile_set_alloc_range - limit range of segment to be allocated
 258  * @sufile: inode of segment usage file
 259  * @start: minimum segment number of allocatable region (inclusive)
 260  * @end: maximum segment number of allocatable region (inclusive)
 261  *
 262  * Return Value: On success, 0 is returned.  On error, one of the
 263  * following negative error codes is returned.
 264  *
 265  * %-ERANGE - invalid segment region
 266  */
 267 int nilfs_sufile_set_alloc_range(struct inode *sufile, __u64 start, __u64 end)
 268 {
 269         struct nilfs_sufile_info *sui = NILFS_SUI(sufile);
 270         __u64 nsegs;
 271         int ret = -ERANGE;
 272
 273         down_write(&NILFS_MDT(sufile)->mi_sem);
 274         nsegs = nilfs_sufile_get_nsegments(sufile);
 275
 276         if (start <= end && end < nsegs) {
 277                 sui->allocmin = start;
 278                 sui->allocmax = end;
 279                 ret = 0;
 280         }
 281         up_write(&NILFS_MDT(sufile)->mi_sem);
 282         return ret;
 283 }
 284
 285 /**
 286  * nilfs_sufile_alloc - allocate a segment
 287  * @sufile: inode of segment usage file
 288  * @segnump: pointer to segment number
 289  *
 290  * Description: nilfs_sufile_alloc() allocates a clean segment.
 291  *
 292  * Return Value: On success, 0 is returned and the segment number of the
 293  * allocated segment is stored in the place pointed by @segnump. On error, one
 294  * of the following negative error codes is returned.
 295  *
 296  * %-EIO - I/O error.
 297  *
 298  * %-ENOMEM - Insufficient amount of memory available.
 299  *
 300  * %-ENOSPC - No clean segment left.
 301  */
 302 int nilfs_sufile_alloc(struct inode *sufile, __u64 *segnump)
 303 {
 304         struct buffer_head *header_bh, *su_bh;
 305         struct nilfs_sufile_header *header;
 306         struct nilfs_segment_usage *su;
 307         struct nilfs_sufile_info *sui = NILFS_SUI(sufile);
 308         size_t susz = NILFS_MDT(sufile)->mi_entry_size;
 309         __u64 segnum, maxsegnum, last_alloc;
 310         void *kaddr;
 311         unsigned long nsegments, nsus, cnt;
 312         int ret, j;
 313
 314         down_write(&NILFS_MDT(sufile)->mi_sem);
 315
 316         ret = nilfs_sufile_get_header_block(sufile, &header_bh);
 317         if (ret < 0)
 318                 goto out_sem;
 319         kaddr = kmap_atomic(header_bh->b_page);
 320         header = kaddr + bh_offset(header_bh);
 321         last_alloc = le64_to_cpu(header->sh_last_alloc);
 322         kunmap_atomic(kaddr);
 323
 324         nsegments = nilfs_sufile_get_nsegments(sufile);
 325         maxsegnum = sui->allocmax;
 326         segnum = last_alloc + 1;
 327         if (segnum < sui->allocmin || segnum > sui->allocmax)
 328                 segnum = sui->allocmin;
 329
 330         for (cnt = 0; cnt < nsegments; cnt += nsus) {
 331                 if (segnum > maxsegnum) {
 332                         if (cnt < sui->allocmax - sui->allocmin + 1) {
 333                                 /*
 334                                  * wrap around in the limited region.
 335                                  * if allocation started from
 336                                  * sui->allocmin, this never happens.
 337                                  */
 338                                 segnum = sui->allocmin;
 339                                 maxsegnum = last_alloc;
 340                         } else if (segnum > sui->allocmin &&
 341                                    sui->allocmax + 1 < nsegments) {
 342                                 segnum = sui->allocmax + 1;
 343                                 maxsegnum = nsegments - 1;
 344                         } else if (sui->allocmin > 0)  {
 345                                 segnum = 0;
 346                                 maxsegnum = sui->allocmin - 1;
 347                         } else {
 348                                 break; /* never happens */
 349                         }
 350                 }
 351                 trace_nilfs2_segment_usage_check(sufile, segnum, cnt);
 352                 ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 1,
 353                                                            &su_bh);
 354                 if (ret < 0)
 355                         goto out_header;
 356                 kaddr = kmap_atomic(su_bh->b_page);
 357                 su = nilfs_sufile_block_get_segment_usage(
 358                         sufile, segnum, su_bh, kaddr);
 359
 360                 nsus = nilfs_sufile_segment_usages_in_block(
 361                         sufile, segnum, maxsegnum);
 362                 for (j = 0; j < nsus; j++, su = (void *)su + susz, segnum++) {
 363                         if (!nilfs_segment_usage_clean(su))
 364                                 continue;
 365                         /* found a clean segment */
 366                         nilfs_segment_usage_set_dirty(su);
 367                         kunmap_atomic(kaddr);
 368
 369                         kaddr = kmap_atomic(header_bh->b_page);
 370                         header = kaddr + bh_offset(header_bh);
 371                         le64_add_cpu(&header->sh_ncleansegs, -1);
 372                         le64_add_cpu(&header->sh_ndirtysegs, 1);
 373                         header->sh_last_alloc = cpu_to_le64(segnum);
 374                         kunmap_atomic(kaddr);
 375
 376                         sui->ncleansegs--;
 377                         mark_buffer_dirty(header_bh);
 378                         mark_buffer_dirty(su_bh);
 379                         nilfs_mdt_mark_dirty(sufile);
 380                         brelse(su_bh);
 381                         *segnump = segnum;
 382
 383                         trace_nilfs2_segment_usage_allocated(sufile, segnum);
 384
 385                         goto out_header;
 386                 }
 387
 388                 kunmap_atomic(kaddr);
 389                 brelse(su_bh);
 390         }
 391
 392         /* no segments left */
 393         ret = -ENOSPC;
 394
 395  out_header:
 396         brelse(header_bh);
 397
 398  out_sem:
 399         up_write(&NILFS_MDT(sufile)->mi_sem);
 400         return ret;
 401 }
 402
 403 void nilfs_sufile_do_cancel_free(struct inode *sufile, __u64 segnum,
 404                                  struct buffer_head *header_bh,
 405                                  struct buffer_head *su_bh)
 406 {
 407         struct nilfs_segment_usage *su;
 408         void *kaddr;
 409
 410         kaddr = kmap_atomic(su_bh->b_page);
 411         su = nilfs_sufile_block_get_segment_usage(sufile, segnum, su_bh, kaddr);
 412         if (unlikely(!nilfs_segment_usage_clean(su))) {
 413                 nilfs_msg(sufile->i_sb, KERN_WARNING,
 414                           "%s: segment %llu must be clean", __func__,
 415                           (unsigned long long)segnum);
 416                 kunmap_atomic(kaddr);
 417                 return;
 418         }
 419         nilfs_segment_usage_set_dirty(su);
 420         kunmap_atomic(kaddr);
 421
 422         nilfs_sufile_mod_counter(header_bh, -1, 1);
 423         NILFS_SUI(sufile)->ncleansegs--;
 424
 425         mark_buffer_dirty(su_bh);
 426         nilfs_mdt_mark_dirty(sufile);
 427 }
 428
 429 void nilfs_sufile_do_scrap(struct inode *sufile, __u64 segnum,
 430                            struct buffer_head *header_bh,
 431                            struct buffer_head *su_bh)
 432 {
 433         struct nilfs_segment_usage *su;
 434         void *kaddr;
 435         int clean, dirty;
 436
 437         kaddr = kmap_atomic(su_bh->b_page);
 438         su = nilfs_sufile_block_get_segment_usage(sufile, segnum, su_bh, kaddr);
 439         if (su->su_flags == cpu_to_le32(BIT(NILFS_SEGMENT_USAGE_DIRTY)) &&
 440             su->su_nblocks == cpu_to_le32(0)) {
 441                 kunmap_atomic(kaddr);
 442                 return;
 443         }
 444         clean = nilfs_segment_usage_clean(su);
 445         dirty = nilfs_segment_usage_dirty(su);
 446
 447         /* make the segment garbage */
 448         su->su_lastmod = cpu_to_le64(0);
 449         su->su_nblocks = cpu_to_le32(0);
 450         su->su_flags = cpu_to_le32(BIT(NILFS_SEGMENT_USAGE_DIRTY));
 451         kunmap_atomic(kaddr);
 452
 453         nilfs_sufile_mod_counter(header_bh, clean ? (u64)-1 : 0, dirty ? 0 : 1);
 454         NILFS_SUI(sufile)->ncleansegs -= clean;
 455
 456         mark_buffer_dirty(su_bh);
 457         nilfs_mdt_mark_dirty(sufile);
 458 }
 459
 460 void nilfs_sufile_do_free(struct inode *sufile, __u64 segnum,
 461                           struct buffer_head *header_bh,
 462                           struct buffer_head *su_bh)
 463 {
 464         struct nilfs_segment_usage *su;
 465         void *kaddr;
 466         int sudirty;
 467
 468         kaddr = kmap_atomic(su_bh->b_page);
 469         su = nilfs_sufile_block_get_segment_usage(sufile, segnum, su_bh, kaddr);
 470         if (nilfs_segment_usage_clean(su)) {
 471                 nilfs_msg(sufile->i_sb, KERN_WARNING,
 472                           "%s: segment %llu is already clean",
 473                           __func__, (unsigned long long)segnum);
 474                 kunmap_atomic(kaddr);
 475                 return;
 476         }
 477         WARN_ON(nilfs_segment_usage_error(su));
 478         WARN_ON(!nilfs_segment_usage_dirty(su));
 479
 480         sudirty = nilfs_segment_usage_dirty(su);
 481         nilfs_segment_usage_set_clean(su);
 482         kunmap_atomic(kaddr);
 483         mark_buffer_dirty(su_bh);
 484
 485         nilfs_sufile_mod_counter(header_bh, 1, sudirty ? (u64)-1 : 0);
 486         NILFS_SUI(sufile)->ncleansegs++;
 487
 488         nilfs_mdt_mark_dirty(sufile);
 489
 490         trace_nilfs2_segment_usage_freed(sufile, segnum);
 491 }
 492
 493 /**
 494  * nilfs_sufile_mark_dirty - mark the buffer having a segment usage dirty
 495  * @sufile: inode of segment usage file
 496  * @segnum: segment number
 497  */
 498 int nilfs_sufile_mark_dirty(struct inode *sufile, __u64 segnum)
 499 {
 500         struct buffer_head *bh;
 501         int ret;
 502
 503         ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0, &bh);
 504         if (!ret) {
 505                 mark_buffer_dirty(bh);
 506                 nilfs_mdt_mark_dirty(sufile);
 507                 brelse(bh);
 508         }
 509         return ret;
 510 }
 511
 512 /**
 513  * nilfs_sufile_set_segment_usage - set usage of a segment
 514  * @sufile: inode of segment usage file
 515  * @segnum: segment number
 516  * @nblocks: number of live blocks in the segment
 517  * @modtime: modification time (option)
 518  */
 519 int nilfs_sufile_set_segment_usage(struct inode *sufile, __u64 segnum,
 520                                    unsigned long nblocks, time64_t modtime)
 521 {
 522         struct buffer_head *bh;
 523         struct nilfs_segment_usage *su;
 524         void *kaddr;
 525         int ret;
 526
 527         down_write(&NILFS_MDT(sufile)->mi_sem);
 528         ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0, &bh);
 529         if (ret < 0)
 530                 goto out_sem;
 531
 532         kaddr = kmap_atomic(bh->b_page);
 533         su = nilfs_sufile_block_get_segment_usage(sufile, segnum, bh, kaddr);
 534         WARN_ON(nilfs_segment_usage_error(su));
 535         if (modtime)
 536                 su->su_lastmod = cpu_to_le64(modtime);
 537         su->su_nblocks = cpu_to_le32(nblocks);
 538         kunmap_atomic(kaddr);
 539
 540         mark_buffer_dirty(bh);
 541         nilfs_mdt_mark_dirty(sufile);
 542         brelse(bh);
 543
 544  out_sem:
 545         up_write(&NILFS_MDT(sufile)->mi_sem);
 546         return ret;
 547 }
 548
 549 /**
 550  * nilfs_sufile_get_stat - get segment usage statistics
 551  * @sufile: inode of segment usage file
 552  * @stat: pointer to a structure of segment usage statistics
 553  *
 554  * Description: nilfs_sufile_get_stat() returns information about segment
 555  * usage.
 556  *
 557  * Return Value: On success, 0 is returned, and segment usage information is
 558  * stored in the place pointed by @stat. On error, one of the following
 559  * negative error codes is returned.
 560  *
 561  * %-EIO - I/O error.
 562  *
 563  * %-ENOMEM - Insufficient amount of memory available.
 564  */
 565 int nilfs_sufile_get_stat(struct inode *sufile, struct nilfs_sustat *sustat)
 566 {
 567         struct buffer_head *header_bh;
 568         struct nilfs_sufile_header *header;
 569         struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
 570         void *kaddr;
 571         int ret;
 572
 573         down_read(&NILFS_MDT(sufile)->mi_sem);
 574
 575         ret = nilfs_sufile_get_header_block(sufile, &header_bh);
 576         if (ret < 0)
 577                 goto out_sem;
 578
 579         kaddr = kmap_atomic(header_bh->b_page);
 580         header = kaddr + bh_offset(header_bh);
 581         sustat->ss_nsegs = nilfs_sufile_get_nsegments(sufile);
 582         sustat->ss_ncleansegs = le64_to_cpu(header->sh_ncleansegs);
 583         sustat->ss_ndirtysegs = le64_to_cpu(header->sh_ndirtysegs);
 584         sustat->ss_ctime = nilfs->ns_ctime;
 585         sustat->ss_nongc_ctime = nilfs->ns_nongc_ctime;
 586         spin_lock(&nilfs->ns_last_segment_lock);
 587         sustat->ss_prot_seq = nilfs->ns_prot_seq;
 588         spin_unlock(&nilfs->ns_last_segment_lock);
 589         kunmap_atomic(kaddr);
 590         brelse(header_bh);
 591
 592  out_sem:
 593         up_read(&NILFS_MDT(sufile)->mi_sem);
 594         return ret;
 595 }
 596
 597 void nilfs_sufile_do_set_error(struct inode *sufile, __u64 segnum,
 598                                struct buffer_head *header_bh,
 599                                struct buffer_head *su_bh)
 600 {
 601         struct nilfs_segment_usage *su;
 602         void *kaddr;
 603         int suclean;
 604
 605         kaddr = kmap_atomic(su_bh->b_page);
 606         su = nilfs_sufile_block_get_segment_usage(sufile, segnum, su_bh, kaddr);
 607         if (nilfs_segment_usage_error(su)) {
 608                 kunmap_atomic(kaddr);
 609                 return;
 610         }
 611         suclean = nilfs_segment_usage_clean(su);
 612         nilfs_segment_usage_set_error(su);
 613         kunmap_atomic(kaddr);
 614
 615         if (suclean) {
 616                 nilfs_sufile_mod_counter(header_bh, -1, 0);
 617                 NILFS_SUI(sufile)->ncleansegs--;
 618         }
 619         mark_buffer_dirty(su_bh);
 620         nilfs_mdt_mark_dirty(sufile);
 621 }
 622
 623 /**
 624  * nilfs_sufile_truncate_range - truncate range of segment array
 625  * @sufile: inode of segment usage file
 626  * @start: start segment number (inclusive)
 627  * @end: end segment number (inclusive)
 628  *
 629  * Return Value: On success, 0 is returned.  On error, one of the
 630  * following negative error codes is returned.
 631  *
 632  * %-EIO - I/O error.
 633  *
 634  * %-ENOMEM - Insufficient amount of memory available.
 635  *
 636  * %-EINVAL - Invalid number of segments specified
 637  *
 638  * %-EBUSY - Dirty or active segments are present in the range
 639  */
 640 static int nilfs_sufile_truncate_range(struct inode *sufile,
 641                                        __u64 start, __u64 end)
 642 {
 643         struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
 644         struct buffer_head *header_bh;
 645         struct buffer_head *su_bh;
 646         struct nilfs_segment_usage *su, *su2;
 647         size_t susz = NILFS_MDT(sufile)->mi_entry_size;
 648         unsigned long segusages_per_block;
 649         unsigned long nsegs, ncleaned;
 650         __u64 segnum;
 651         void *kaddr;
 652         ssize_t n, nc;
 653         int ret;
 654         int j;
 655
 656         nsegs = nilfs_sufile_get_nsegments(sufile);
 657
 658         ret = -EINVAL;
 659         if (start > end || start >= nsegs)
 660                 goto out;
 661
 662         ret = nilfs_sufile_get_header_block(sufile, &header_bh);
 663         if (ret < 0)
 664                 goto out;
 665
 666         segusages_per_block = nilfs_sufile_segment_usages_per_block(sufile);
 667         ncleaned = 0;
 668
 669         for (segnum = start; segnum <= end; segnum += n) {
 670                 n = min_t(unsigned long,
 671                           segusages_per_block -
 672                                   nilfs_sufile_get_offset(sufile, segnum),
 673                           end - segnum + 1);
 674                 ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0,
 675                                                            &su_bh);
 676                 if (ret < 0) {
 677                         if (ret != -ENOENT)
 678                                 goto out_header;
 679                         /* hole */
 680                         continue;
 681                 }
 682                 kaddr = kmap_atomic(su_bh->b_page);
 683                 su = nilfs_sufile_block_get_segment_usage(
 684                         sufile, segnum, su_bh, kaddr);
 685                 su2 = su;
 686                 for (j = 0; j < n; j++, su = (void *)su + susz) {
 687                         if ((le32_to_cpu(su->su_flags) &
 688                              ~BIT(NILFS_SEGMENT_USAGE_ERROR)) ||
 689                             nilfs_segment_is_active(nilfs, segnum + j)) {
 690                                 ret = -EBUSY;
 691                                 kunmap_atomic(kaddr);
 692                                 brelse(su_bh);
 693                                 goto out_header;
 694                         }
 695                 }
 696                 nc = 0;
 697                 for (su = su2, j = 0; j < n; j++, su = (void *)su + susz) {
 698                         if (nilfs_segment_usage_error(su)) {
 699                                 nilfs_segment_usage_set_clean(su);
 700                                 nc++;
 701                         }
 702                 }
 703                 kunmap_atomic(kaddr);
 704                 if (nc > 0) {
 705                         mark_buffer_dirty(su_bh);
 706                         ncleaned += nc;
 707                 }
 708                 brelse(su_bh);
 709
 710                 if (n == segusages_per_block) {
 711                         /* make hole */
 712                         nilfs_sufile_delete_segment_usage_block(sufile, segnum);
 713                 }
 714         }
 715         ret = 0;
 716
 717 out_header:
 718         if (ncleaned > 0) {
 719                 NILFS_SUI(sufile)->ncleansegs += ncleaned;
 720                 nilfs_sufile_mod_counter(header_bh, ncleaned, 0);
 721                 nilfs_mdt_mark_dirty(sufile);
 722         }
 723         brelse(header_bh);
 724 out:
 725         return ret;
 726 }
 727
 728 /**
 729  * nilfs_sufile_resize - resize segment array
 730  * @sufile: inode of segment usage file
 731  * @newnsegs: new number of segments
 732  *
 733  * Return Value: On success, 0 is returned.  On error, one of the
 734  * following negative error codes is returned.
 735  *
 736  * %-EIO - I/O error.
 737  *
 738  * %-ENOMEM - Insufficient amount of memory available.
 739  *
 740  * %-ENOSPC - Enough free space is not left for shrinking
 741  *
 742  * %-EBUSY - Dirty or active segments exist in the region to be truncated
 743  */
 744 int nilfs_sufile_resize(struct inode *sufile, __u64 newnsegs)
 745 {
 746         struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
 747         struct buffer_head *header_bh;
 748         struct nilfs_sufile_header *header;
 749         struct nilfs_sufile_info *sui = NILFS_SUI(sufile);
 750         void *kaddr;
 751         unsigned long nsegs, nrsvsegs;
 752         int ret = 0;
 753
 754         down_write(&NILFS_MDT(sufile)->mi_sem);
 755
 756         nsegs = nilfs_sufile_get_nsegments(sufile);
 757         if (nsegs == newnsegs)
 758                 goto out;
 759
 760         ret = -ENOSPC;
 761         nrsvsegs = nilfs_nrsvsegs(nilfs, newnsegs);
 762         if (newnsegs < nsegs && nsegs - newnsegs + nrsvsegs > sui->ncleansegs)
 763                 goto out;
 764
 765         ret = nilfs_sufile_get_header_block(sufile, &header_bh);
 766         if (ret < 0)
 767                 goto out;
 768
 769         if (newnsegs > nsegs) {
 770                 sui->ncleansegs += newnsegs - nsegs;
 771         } else /* newnsegs < nsegs */ {
 772                 ret = nilfs_sufile_truncate_range(sufile, newnsegs, nsegs - 1);
 773                 if (ret < 0)
 774                         goto out_header;
 775
 776                 sui->ncleansegs -= nsegs - newnsegs;
 777         }
 778
 779         kaddr = kmap_atomic(header_bh->b_page);
 780         header = kaddr + bh_offset(header_bh);
 781         header->sh_ncleansegs = cpu_to_le64(sui->ncleansegs);
 782         kunmap_atomic(kaddr);
 783
 784         mark_buffer_dirty(header_bh);
 785         nilfs_mdt_mark_dirty(sufile);
 786         nilfs_set_nsegments(nilfs, newnsegs);
 787
 788 out_header:
 789         brelse(header_bh);
 790 out:
 791         up_write(&NILFS_MDT(sufile)->mi_sem);
 792         return ret;
 793 }
 794
 795 /**
 796  * nilfs_sufile_get_suinfo -
 797  * @sufile: inode of segment usage file
 798  * @segnum: segment number to start looking
 799  * @buf: array of suinfo
 800  * @sisz: byte size of suinfo
 801  * @nsi: size of suinfo array
 802  *
 803  * Description:
 804  *
 805  * Return Value: On success, 0 is returned and .... On error, one of the
 806  * following negative error codes is returned.
 807  *
 808  * %-EIO - I/O error.
 809  *
 810  * %-ENOMEM - Insufficient amount of memory available.
 811  */
 812 ssize_t nilfs_sufile_get_suinfo(struct inode *sufile, __u64 segnum, void *buf,
 813                                 unsigned int sisz, size_t nsi)
 814 {
 815         struct buffer_head *su_bh;
 816         struct nilfs_segment_usage *su;
 817         struct nilfs_suinfo *si = buf;
 818         size_t susz = NILFS_MDT(sufile)->mi_entry_size;
 819         struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
 820         void *kaddr;
 821         unsigned long nsegs, segusages_per_block;
 822         ssize_t n;
 823         int ret, i, j;
 824
 825         down_read(&NILFS_MDT(sufile)->mi_sem);
 826
 827         segusages_per_block = nilfs_sufile_segment_usages_per_block(sufile);
 828         nsegs = min_t(unsigned long,
 829                       nilfs_sufile_get_nsegments(sufile) - segnum,
 830                       nsi);
 831         for (i = 0; i < nsegs; i += n, segnum += n) {
 832                 n = min_t(unsigned long,
 833                           segusages_per_block -
 834                                   nilfs_sufile_get_offset(sufile, segnum),
 835                           nsegs - i);
 836                 ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0,
 837                                                            &su_bh);
 838                 if (ret < 0) {
 839                         if (ret != -ENOENT)
 840                                 goto out;
 841                         /* hole */
 842                         memset(si, 0, sisz * n);
 843                         si = (void *)si + sisz * n;
 844                         continue;
 845                 }
 846
 847                 kaddr = kmap_atomic(su_bh->b_page);
 848                 su = nilfs_sufile_block_get_segment_usage(
 849                         sufile, segnum, su_bh, kaddr);
 850                 for (j = 0; j < n;
 851                      j++, su = (void *)su + susz, si = (void *)si + sisz) {
 852                         si->sui_lastmod = le64_to_cpu(su->su_lastmod);
 853                         si->sui_nblocks = le32_to_cpu(su->su_nblocks);
 854                         si->sui_flags = le32_to_cpu(su->su_flags) &
 855                                 ~BIT(NILFS_SEGMENT_USAGE_ACTIVE);
 856                         if (nilfs_segment_is_active(nilfs, segnum + j))
 857                                 si->sui_flags |=
 858                                         BIT(NILFS_SEGMENT_USAGE_ACTIVE);
 859                 }
 860                 kunmap_atomic(kaddr);
 861                 brelse(su_bh);
 862         }
 863         ret = nsegs;
 864
 865  out:
 866         up_read(&NILFS_MDT(sufile)->mi_sem);
 867         return ret;
 868 }
 869
 870 /**
 871  * nilfs_sufile_set_suinfo - sets segment usage info
 872  * @sufile: inode of segment usage file
 873  * @buf: array of suinfo_update
 874  * @supsz: byte size of suinfo_update
 875  * @nsup: size of suinfo_update array
 876  *
 877  * Description: Takes an array of nilfs_suinfo_update structs and updates
 878  * segment usage accordingly. Only the fields indicated by the sup_flags
 879  * are updated.
 880  *
 881  * Return Value: On success, 0 is returned. On error, one of the
 882  * following negative error codes is returned.
 883  *
 884  * %-EIO - I/O error.
 885  *
 886  * %-ENOMEM - Insufficient amount of memory available.
 887  *
 888  * %-EINVAL - Invalid values in input (segment number, flags or nblocks)
 889  */
 890 ssize_t nilfs_sufile_set_suinfo(struct inode *sufile, void *buf,
 891                                 unsigned int supsz, size_t nsup)
 892 {
 893         struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
 894         struct buffer_head *header_bh, *bh;
 895         struct nilfs_suinfo_update *sup, *supend = buf + supsz * nsup;
 896         struct nilfs_segment_usage *su;
 897         void *kaddr;
 898         unsigned long blkoff, prev_blkoff;
 899         int cleansi, cleansu, dirtysi, dirtysu;
 900         long ncleaned = 0, ndirtied = 0;
 901         int ret = 0;
 902
 903         if (unlikely(nsup == 0))
 904                 return ret;
 905
 906         for (sup = buf; sup < supend; sup = (void *)sup + supsz) {
 907                 if (sup->sup_segnum >= nilfs->ns_nsegments
 908                         || (sup->sup_flags &
 909                                 (~0UL << __NR_NILFS_SUINFO_UPDATE_FIELDS))
 910                         || (nilfs_suinfo_update_nblocks(sup) &&
 911                                 sup->sup_sui.sui_nblocks >
 912                                 nilfs->ns_blocks_per_segment))
 913                         return -EINVAL;
 914         }
 915
 916         down_write(&NILFS_MDT(sufile)->mi_sem);
 917
 918         ret = nilfs_sufile_get_header_block(sufile, &header_bh);
 919         if (ret < 0)
 920                 goto out_sem;
 921
 922         sup = buf;
 923         blkoff = nilfs_sufile_get_blkoff(sufile, sup->sup_segnum);
 924         ret = nilfs_mdt_get_block(sufile, blkoff, 1, NULL, &bh);
 925         if (ret < 0)
 926                 goto out_header;
 927
 928         for (;;) {
 929                 kaddr = kmap_atomic(bh->b_page);
 930                 su = nilfs_sufile_block_get_segment_usage(
 931                         sufile, sup->sup_segnum, bh, kaddr);
 932
 933                 if (nilfs_suinfo_update_lastmod(sup))
 934                         su->su_lastmod = cpu_to_le64(sup->sup_sui.sui_lastmod);
 935
 936                 if (nilfs_suinfo_update_nblocks(sup))
 937                         su->su_nblocks = cpu_to_le32(sup->sup_sui.sui_nblocks);
 938
 939                 if (nilfs_suinfo_update_flags(sup)) {
 940                         /*
 941                          * Active flag is a virtual flag projected by running
 942                          * nilfs kernel code - drop it not to write it to
 943                          * disk.
 944                          */
 945                         sup->sup_sui.sui_flags &=
 946                                         ~BIT(NILFS_SEGMENT_USAGE_ACTIVE);
 947
 948                         cleansi = nilfs_suinfo_clean(&sup->sup_sui);
 949                         cleansu = nilfs_segment_usage_clean(su);
 950                         dirtysi = nilfs_suinfo_dirty(&sup->sup_sui);
 951                         dirtysu = nilfs_segment_usage_dirty(su);
 952
 953                         if (cleansi && !cleansu)
 954                                 ++ncleaned;
 955                         else if (!cleansi && cleansu)
 956                                 --ncleaned;
 957
 958                         if (dirtysi && !dirtysu)
 959                                 ++ndirtied;
 960                         else if (!dirtysi && dirtysu)
 961                                 --ndirtied;
 962
 963                         su->su_flags = cpu_to_le32(sup->sup_sui.sui_flags);
 964                 }
 965
 966                 kunmap_atomic(kaddr);
 967
 968                 sup = (void *)sup + supsz;
 969                 if (sup >= supend)
 970                         break;
 971
 972                 prev_blkoff = blkoff;
 973                 blkoff = nilfs_sufile_get_blkoff(sufile, sup->sup_segnum);
 974                 if (blkoff == prev_blkoff)
 975                         continue;
 976
 977                 /* get different block */
 978                 mark_buffer_dirty(bh);
 979                 put_bh(bh);
 980                 ret = nilfs_mdt_get_block(sufile, blkoff, 1, NULL, &bh);
 981                 if (unlikely(ret < 0))
 982                         goto out_mark;
 983         }
 984         mark_buffer_dirty(bh);
 985         put_bh(bh);
 986
 987  out_mark:
 988         if (ncleaned || ndirtied) {
 989                 nilfs_sufile_mod_counter(header_bh, (u64)ncleaned,
 990                                 (u64)ndirtied);
 991                 NILFS_SUI(sufile)->ncleansegs += ncleaned;
 992         }
 993         nilfs_mdt_mark_dirty(sufile);
 994  out_header:
 995         put_bh(header_bh);
 996  out_sem:
 997         up_write(&NILFS_MDT(sufile)->mi_sem);
 998         return ret;
 999 }
1000
1001 /**
1002  * nilfs_sufile_trim_fs() - trim ioctl handle function
1003  * @sufile: inode of segment usage file
1004  * @range: fstrim_range structure
1005  *
1006  * start:       First Byte to trim
1007  * len:         number of Bytes to trim from start
1008  * minlen:      minimum extent length in Bytes
1009  *
1010  * Decription: nilfs_sufile_trim_fs goes through all segments containing bytes
1011  * from start to start+len. start is rounded up to the next block boundary
1012  * and start+len is rounded down. For each clean segment blkdev_issue_discard
1013  * function is invoked.
1014  *
1015  * Return Value: On success, 0 is returned or negative error code, otherwise.
1016  */
1017 int nilfs_sufile_trim_fs(struct inode *sufile, struct fstrim_range *range)
1018 {
1019         struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
1020         struct buffer_head *su_bh;
1021         struct nilfs_segment_usage *su;
1022         void *kaddr;
1023         size_t n, i, susz = NILFS_MDT(sufile)->mi_entry_size;
1024         sector_t seg_start, seg_end, start_block, end_block;
1025         sector_t start = 0, nblocks = 0;
1026         u64 segnum, segnum_end, minlen, len, max_blocks, ndiscarded = 0;
1027         int ret = 0;
1028         unsigned int sects_per_block;
1029
1030         sects_per_block = (1 << nilfs->ns_blocksize_bits) /
1031                         bdev_logical_block_size(nilfs->ns_bdev);
1032         len = range->len >> nilfs->ns_blocksize_bits;
1033         minlen = range->minlen >> nilfs->ns_blocksize_bits;
1034         max_blocks = ((u64)nilfs->ns_nsegments * nilfs->ns_blocks_per_segment);
1035
1036         if (!len || range->start >= max_blocks << nilfs->ns_blocksize_bits)
1037                 return -EINVAL;
1038
1039         start_block = (range->start + nilfs->ns_blocksize - 1) >>
1040                         nilfs->ns_blocksize_bits;
1041
1042         /*
1043          * range->len can be very large (actually, it is set to
1044          * ULLONG_MAX by default) - truncate upper end of the range
1045          * carefully so as not to overflow.
1046          */
1047         if (max_blocks - start_block < len)
1048                 end_block = max_blocks - 1;
1049         else
1050                 end_block = start_block + len - 1;
1051
1052         segnum = nilfs_get_segnum_of_block(nilfs, start_block);
1053         segnum_end = nilfs_get_segnum_of_block(nilfs, end_block);
1054
1055         down_read(&NILFS_MDT(sufile)->mi_sem);
1056
1057         while (segnum <= segnum_end) {
1058                 n = nilfs_sufile_segment_usages_in_block(sufile, segnum,
1059                                 segnum_end);
1060
1061                 ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0,
1062                                                            &su_bh);
1063                 if (ret < 0) {
1064                         if (ret != -ENOENT)
1065                                 goto out_sem;
1066                         /* hole */
1067                         segnum += n;
1068                         continue;
1069                 }
1070
1071                 kaddr = kmap_atomic(su_bh->b_page);
1072                 su = nilfs_sufile_block_get_segment_usage(sufile, segnum,
1073                                 su_bh, kaddr);
1074                 for (i = 0; i < n; ++i, ++segnum, su = (void *)su + susz) {
1075                         if (!nilfs_segment_usage_clean(su))
1076                                 continue;
1077
1078                         nilfs_get_segment_range(nilfs, segnum, &seg_start,
1079                                                 &seg_end);
1080
1081                         if (!nblocks) {
1082                                 /* start new extent */
1083                                 start = seg_start;
1084                                 nblocks = seg_end - seg_start + 1;
1085                                 continue;
1086                         }
1087
1088                         if (start + nblocks == seg_start) {
1089                                 /* add to previous extent */
1090                                 nblocks += seg_end - seg_start + 1;
1091                                 continue;
1092                         }
1093
1094                         /* discard previous extent */
1095                         if (start < start_block) {
1096                                 nblocks -= start_block - start;
1097                                 start = start_block;
1098                         }
1099
1100                         if (nblocks >= minlen) {
1101                                 kunmap_atomic(kaddr);
1102
1103                                 ret = blkdev_issue_discard(nilfs->ns_bdev,
1104                                                 start * sects_per_block,
1105                                                 nblocks * sects_per_block,
1106                                                 GFP_NOFS, 0);
1107                                 if (ret < 0) {
1108                                         put_bh(su_bh);
1109                                         goto out_sem;
1110                                 }
1111
1112                                 ndiscarded += nblocks;
1113                                 kaddr = kmap_atomic(su_bh->b_page);
1114                                 su = nilfs_sufile_block_get_segment_usage(
1115                                         sufile, segnum, su_bh, kaddr);
1116                         }
1117
1118                         /* start new extent */
1119                         start = seg_start;
1120                         nblocks = seg_end - seg_start + 1;
1121                 }
1122                 kunmap_atomic(kaddr);
1123                 put_bh(su_bh);
1124         }
1125
1126
1127         if (nblocks) {
1128                 /* discard last extent */
1129                 if (start < start_block) {
1130                         nblocks -= start_block - start;
1131                         start = start_block;
1132                 }
1133                 if (start + nblocks > end_block + 1)
1134                         nblocks = end_block - start + 1;
1135
1136                 if (nblocks >= minlen) {
1137                         ret = blkdev_issue_discard(nilfs->ns_bdev,
1138                                         start * sects_per_block,
1139                                         nblocks * sects_per_block,
1140                                         GFP_NOFS, 0);
1141                         if (!ret)
1142                                 ndiscarded += nblocks;
1143                 }
1144         }
1145
1146 out_sem:
1147         up_read(&NILFS_MDT(sufile)->mi_sem);
1148
1149         range->len = ndiscarded << nilfs->ns_blocksize_bits;
1150         return ret;
1151 }
1152
1153 /**
1154  * nilfs_sufile_read - read or get sufile inode
1155  * @sb: super block instance
1156  * @susize: size of a segment usage entry
1157  * @raw_inode: on-disk sufile inode
1158  * @inodep: buffer to store the inode
1159  */
1160 int nilfs_sufile_read(struct super_block *sb, size_t susize,
1161                       struct nilfs_inode *raw_inode, struct inode **inodep)
1162 {
1163         struct inode *sufile;
1164         struct nilfs_sufile_info *sui;
1165         struct buffer_head *header_bh;
1166         struct nilfs_sufile_header *header;
1167         void *kaddr;
1168         int err;
1169
1170         if (susize > sb->s_blocksize) {
1171                 nilfs_msg(sb, KERN_ERR,
1172                           "too large segment usage size: %zu bytes", susize);
1173                 return -EINVAL;
1174         } else if (susize < NILFS_MIN_SEGMENT_USAGE_SIZE) {
1175                 nilfs_msg(sb, KERN_ERR,
1176                           "too small segment usage size: %zu bytes", susize);
1177                 return -EINVAL;
1178         }
1179
1180         sufile = nilfs_iget_locked(sb, NULL, NILFS_SUFILE_INO);
1181         if (unlikely(!sufile))
1182                 return -ENOMEM;
1183         if (!(sufile->i_state & I_NEW))
1184                 goto out;
1185
1186         err = nilfs_mdt_init(sufile, NILFS_MDT_GFP, sizeof(*sui));
1187         if (err)
1188                 goto failed;
1189
1190         nilfs_mdt_set_entry_size(sufile, susize,
1191                                  sizeof(struct nilfs_sufile_header));
1192
1193         err = nilfs_read_inode_common(sufile, raw_inode);
1194         if (err)
1195                 goto failed;
1196
1197         err = nilfs_sufile_get_header_block(sufile, &header_bh);
1198         if (err)
1199                 goto failed;
1200
1201         sui = NILFS_SUI(sufile);
1202         kaddr = kmap_atomic(header_bh->b_page);
1203         header = kaddr + bh_offset(header_bh);
1204         sui->ncleansegs = le64_to_cpu(header->sh_ncleansegs);
1205         kunmap_atomic(kaddr);
1206         brelse(header_bh);
1207
1208         sui->allocmax = nilfs_sufile_get_nsegments(sufile) - 1;
1209         sui->allocmin = 0;
1210
1211         unlock_new_inode(sufile);
1212  out:
1213         *inodep = sufile;
1214         return 0;
1215  failed:
1216         iget_failed(sufile);
1217         return err;
1218 }