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