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         void *kaddr;
 502         struct nilfs_segment_usage *su;
 503         int ret;
 504
 505         down_write(&NILFS_MDT(sufile)->mi_sem);
 506         ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0, &bh);
 507         if (!ret) {
 508                 mark_buffer_dirty(bh);
 509                 nilfs_mdt_mark_dirty(sufile);
 510                 kaddr = kmap_atomic(bh->b_page);
 511                 su = nilfs_sufile_block_get_segment_usage(sufile, segnum, bh, kaddr);
 512                 nilfs_segment_usage_set_dirty(su);
 513                 kunmap_atomic(kaddr);
 514                 brelse(bh);
 515         }
 516         up_write(&NILFS_MDT(sufile)->mi_sem);
 517         return ret;
 518 }
 519
 520 /**
 521  * nilfs_sufile_set_segment_usage - set usage of a segment
 522  * @sufile: inode of segment usage file
 523  * @segnum: segment number
 524  * @nblocks: number of live blocks in the segment
 525  * @modtime: modification time (option)
 526  */
 527 int nilfs_sufile_set_segment_usage(struct inode *sufile, __u64 segnum,
 528                                    unsigned long nblocks, time64_t modtime)
 529 {
 530         struct buffer_head *bh;
 531         struct nilfs_segment_usage *su;
 532         void *kaddr;
 533         int ret;
 534
 535         down_write(&NILFS_MDT(sufile)->mi_sem);
 536         ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0, &bh);
 537         if (ret < 0)
 538                 goto out_sem;
 539
 540         kaddr = kmap_atomic(bh->b_page);
 541         su = nilfs_sufile_block_get_segment_usage(sufile, segnum, bh, kaddr);
 542         WARN_ON(nilfs_segment_usage_error(su));
 543         if (modtime)
 544                 su->su_lastmod = cpu_to_le64(modtime);
 545         su->su_nblocks = cpu_to_le32(nblocks);
 546         kunmap_atomic(kaddr);
 547
 548         mark_buffer_dirty(bh);
 549         nilfs_mdt_mark_dirty(sufile);
 550         brelse(bh);
 551
 552  out_sem:
 553         up_write(&NILFS_MDT(sufile)->mi_sem);
 554         return ret;
 555 }
 556
 557 /**
 558  * nilfs_sufile_get_stat - get segment usage statistics
 559  * @sufile: inode of segment usage file
 560  * @stat: pointer to a structure of segment usage statistics
 561  *
 562  * Description: nilfs_sufile_get_stat() returns information about segment
 563  * usage.
 564  *
 565  * Return Value: On success, 0 is returned, and segment usage information is
 566  * stored in the place pointed by @stat. On error, one of the following
 567  * negative error codes is returned.
 568  *
 569  * %-EIO - I/O error.
 570  *
 571  * %-ENOMEM - Insufficient amount of memory available.
 572  */
 573 int nilfs_sufile_get_stat(struct inode *sufile, struct nilfs_sustat *sustat)
 574 {
 575         struct buffer_head *header_bh;
 576         struct nilfs_sufile_header *header;
 577         struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
 578         void *kaddr;
 579         int ret;
 580
 581         down_read(&NILFS_MDT(sufile)->mi_sem);
 582
 583         ret = nilfs_sufile_get_header_block(sufile, &header_bh);
 584         if (ret < 0)
 585                 goto out_sem;
 586
 587         kaddr = kmap_atomic(header_bh->b_page);
 588         header = kaddr + bh_offset(header_bh);
 589         sustat->ss_nsegs = nilfs_sufile_get_nsegments(sufile);
 590         sustat->ss_ncleansegs = le64_to_cpu(header->sh_ncleansegs);
 591         sustat->ss_ndirtysegs = le64_to_cpu(header->sh_ndirtysegs);
 592         sustat->ss_ctime = nilfs->ns_ctime;
 593         sustat->ss_nongc_ctime = nilfs->ns_nongc_ctime;
 594         spin_lock(&nilfs->ns_last_segment_lock);
 595         sustat->ss_prot_seq = nilfs->ns_prot_seq;
 596         spin_unlock(&nilfs->ns_last_segment_lock);
 597         kunmap_atomic(kaddr);
 598         brelse(header_bh);
 599
 600  out_sem:
 601         up_read(&NILFS_MDT(sufile)->mi_sem);
 602         return ret;
 603 }
 604
 605 void nilfs_sufile_do_set_error(struct inode *sufile, __u64 segnum,
 606                                struct buffer_head *header_bh,
 607                                struct buffer_head *su_bh)
 608 {
 609         struct nilfs_segment_usage *su;
 610         void *kaddr;
 611         int suclean;
 612
 613         kaddr = kmap_atomic(su_bh->b_page);
 614         su = nilfs_sufile_block_get_segment_usage(sufile, segnum, su_bh, kaddr);
 615         if (nilfs_segment_usage_error(su)) {
 616                 kunmap_atomic(kaddr);
 617                 return;
 618         }
 619         suclean = nilfs_segment_usage_clean(su);
 620         nilfs_segment_usage_set_error(su);
 621         kunmap_atomic(kaddr);
 622
 623         if (suclean) {
 624                 nilfs_sufile_mod_counter(header_bh, -1, 0);
 625                 NILFS_SUI(sufile)->ncleansegs--;
 626         }
 627         mark_buffer_dirty(su_bh);
 628         nilfs_mdt_mark_dirty(sufile);
 629 }
 630
 631 /**
 632  * nilfs_sufile_truncate_range - truncate range of segment array
 633  * @sufile: inode of segment usage file
 634  * @start: start segment number (inclusive)
 635  * @end: end segment number (inclusive)
 636  *
 637  * Return Value: On success, 0 is returned.  On error, one of the
 638  * following negative error codes is returned.
 639  *
 640  * %-EIO - I/O error.
 641  *
 642  * %-ENOMEM - Insufficient amount of memory available.
 643  *
 644  * %-EINVAL - Invalid number of segments specified
 645  *
 646  * %-EBUSY - Dirty or active segments are present in the range
 647  */
 648 static int nilfs_sufile_truncate_range(struct inode *sufile,
 649                                        __u64 start, __u64 end)
 650 {
 651         struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
 652         struct buffer_head *header_bh;
 653         struct buffer_head *su_bh;
 654         struct nilfs_segment_usage *su, *su2;
 655         size_t susz = NILFS_MDT(sufile)->mi_entry_size;
 656         unsigned long segusages_per_block;
 657         unsigned long nsegs, ncleaned;
 658         __u64 segnum;
 659         void *kaddr;
 660         ssize_t n, nc;
 661         int ret;
 662         int j;
 663
 664         nsegs = nilfs_sufile_get_nsegments(sufile);
 665
 666         ret = -EINVAL;
 667         if (start > end || start >= nsegs)
 668                 goto out;
 669
 670         ret = nilfs_sufile_get_header_block(sufile, &header_bh);
 671         if (ret < 0)
 672                 goto out;
 673
 674         segusages_per_block = nilfs_sufile_segment_usages_per_block(sufile);
 675         ncleaned = 0;
 676
 677         for (segnum = start; segnum <= end; segnum += n) {
 678                 n = min_t(unsigned long,
 679                           segusages_per_block -
 680                                   nilfs_sufile_get_offset(sufile, segnum),
 681                           end - segnum + 1);
 682                 ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0,
 683                                                            &su_bh);
 684                 if (ret < 0) {
 685                         if (ret != -ENOENT)
 686                                 goto out_header;
 687                         /* hole */
 688                         continue;
 689                 }
 690                 kaddr = kmap_atomic(su_bh->b_page);
 691                 su = nilfs_sufile_block_get_segment_usage(
 692                         sufile, segnum, su_bh, kaddr);
 693                 su2 = su;
 694                 for (j = 0; j < n; j++, su = (void *)su + susz) {
 695                         if ((le32_to_cpu(su->su_flags) &
 696                              ~BIT(NILFS_SEGMENT_USAGE_ERROR)) ||
 697                             nilfs_segment_is_active(nilfs, segnum + j)) {
 698                                 ret = -EBUSY;
 699                                 kunmap_atomic(kaddr);
 700                                 brelse(su_bh);
 701                                 goto out_header;
 702                         }
 703                 }
 704                 nc = 0;
 705                 for (su = su2, j = 0; j < n; j++, su = (void *)su + susz) {
 706                         if (nilfs_segment_usage_error(su)) {
 707                                 nilfs_segment_usage_set_clean(su);
 708                                 nc++;
 709                         }
 710                 }
 711                 kunmap_atomic(kaddr);
 712                 if (nc > 0) {
 713                         mark_buffer_dirty(su_bh);
 714                         ncleaned += nc;
 715                 }
 716                 brelse(su_bh);
 717
 718                 if (n == segusages_per_block) {
 719                         /* make hole */
 720                         nilfs_sufile_delete_segment_usage_block(sufile, segnum);
 721                 }
 722         }
 723         ret = 0;
 724
 725 out_header:
 726         if (ncleaned > 0) {
 727                 NILFS_SUI(sufile)->ncleansegs += ncleaned;
 728                 nilfs_sufile_mod_counter(header_bh, ncleaned, 0);
 729                 nilfs_mdt_mark_dirty(sufile);
 730         }
 731         brelse(header_bh);
 732 out:
 733         return ret;
 734 }
 735
 736 /**
 737  * nilfs_sufile_resize - resize segment array
 738  * @sufile: inode of segment usage file
 739  * @newnsegs: new number of segments
 740  *
 741  * Return Value: On success, 0 is returned.  On error, one of the
 742  * following negative error codes is returned.
 743  *
 744  * %-EIO - I/O error.
 745  *
 746  * %-ENOMEM - Insufficient amount of memory available.
 747  *
 748  * %-ENOSPC - Enough free space is not left for shrinking
 749  *
 750  * %-EBUSY - Dirty or active segments exist in the region to be truncated
 751  */
 752 int nilfs_sufile_resize(struct inode *sufile, __u64 newnsegs)
 753 {
 754         struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
 755         struct buffer_head *header_bh;
 756         struct nilfs_sufile_header *header;
 757         struct nilfs_sufile_info *sui = NILFS_SUI(sufile);
 758         void *kaddr;
 759         unsigned long nsegs, nrsvsegs;
 760         int ret = 0;
 761
 762         down_write(&NILFS_MDT(sufile)->mi_sem);
 763
 764         nsegs = nilfs_sufile_get_nsegments(sufile);
 765         if (nsegs == newnsegs)
 766                 goto out;
 767
 768         ret = -ENOSPC;
 769         nrsvsegs = nilfs_nrsvsegs(nilfs, newnsegs);
 770         if (newnsegs < nsegs && nsegs - newnsegs + nrsvsegs > sui->ncleansegs)
 771                 goto out;
 772
 773         ret = nilfs_sufile_get_header_block(sufile, &header_bh);
 774         if (ret < 0)
 775                 goto out;
 776
 777         if (newnsegs > nsegs) {
 778                 sui->ncleansegs += newnsegs - nsegs;
 779         } else /* newnsegs < nsegs */ {
 780                 ret = nilfs_sufile_truncate_range(sufile, newnsegs, nsegs - 1);
 781                 if (ret < 0)
 782                         goto out_header;
 783
 784                 sui->ncleansegs -= nsegs - newnsegs;
 785         }
 786
 787         kaddr = kmap_atomic(header_bh->b_page);
 788         header = kaddr + bh_offset(header_bh);
 789         header->sh_ncleansegs = cpu_to_le64(sui->ncleansegs);
 790         kunmap_atomic(kaddr);
 791
 792         mark_buffer_dirty(header_bh);
 793         nilfs_mdt_mark_dirty(sufile);
 794         nilfs_set_nsegments(nilfs, newnsegs);
 795
 796 out_header:
 797         brelse(header_bh);
 798 out:
 799         up_write(&NILFS_MDT(sufile)->mi_sem);
 800         return ret;
 801 }
 802
 803 /**
 804  * nilfs_sufile_get_suinfo -
 805  * @sufile: inode of segment usage file
 806  * @segnum: segment number to start looking
 807  * @buf: array of suinfo
 808  * @sisz: byte size of suinfo
 809  * @nsi: size of suinfo array
 810  *
 811  * Description:
 812  *
 813  * Return Value: On success, 0 is returned and .... On error, one of the
 814  * following negative error codes is returned.
 815  *
 816  * %-EIO - I/O error.
 817  *
 818  * %-ENOMEM - Insufficient amount of memory available.
 819  */
 820 ssize_t nilfs_sufile_get_suinfo(struct inode *sufile, __u64 segnum, void *buf,
 821                                 unsigned int sisz, size_t nsi)
 822 {
 823         struct buffer_head *su_bh;
 824         struct nilfs_segment_usage *su;
 825         struct nilfs_suinfo *si = buf;
 826         size_t susz = NILFS_MDT(sufile)->mi_entry_size;
 827         struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
 828         void *kaddr;
 829         unsigned long nsegs, segusages_per_block;
 830         ssize_t n;
 831         int ret, i, j;
 832
 833         down_read(&NILFS_MDT(sufile)->mi_sem);
 834
 835         segusages_per_block = nilfs_sufile_segment_usages_per_block(sufile);
 836         nsegs = min_t(unsigned long,
 837                       nilfs_sufile_get_nsegments(sufile) - segnum,
 838                       nsi);
 839         for (i = 0; i < nsegs; i += n, segnum += n) {
 840                 n = min_t(unsigned long,
 841                           segusages_per_block -
 842                                   nilfs_sufile_get_offset(sufile, segnum),
 843                           nsegs - i);
 844                 ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0,
 845                                                            &su_bh);
 846                 if (ret < 0) {
 847                         if (ret != -ENOENT)
 848                                 goto out;
 849                         /* hole */
 850                         memset(si, 0, sisz * n);
 851                         si = (void *)si + sisz * n;
 852                         continue;
 853                 }
 854
 855                 kaddr = kmap_atomic(su_bh->b_page);
 856                 su = nilfs_sufile_block_get_segment_usage(
 857                         sufile, segnum, su_bh, kaddr);
 858                 for (j = 0; j < n;
 859                      j++, su = (void *)su + susz, si = (void *)si + sisz) {
 860                         si->sui_lastmod = le64_to_cpu(su->su_lastmod);
 861                         si->sui_nblocks = le32_to_cpu(su->su_nblocks);
 862                         si->sui_flags = le32_to_cpu(su->su_flags) &
 863                                 ~BIT(NILFS_SEGMENT_USAGE_ACTIVE);
 864                         if (nilfs_segment_is_active(nilfs, segnum + j))
 865                                 si->sui_flags |=
 866                                         BIT(NILFS_SEGMENT_USAGE_ACTIVE);
 867                 }
 868                 kunmap_atomic(kaddr);
 869                 brelse(su_bh);
 870         }
 871         ret = nsegs;
 872
 873  out:
 874         up_read(&NILFS_MDT(sufile)->mi_sem);
 875         return ret;
 876 }
 877
 878 /**
 879  * nilfs_sufile_set_suinfo - sets segment usage info
 880  * @sufile: inode of segment usage file
 881  * @buf: array of suinfo_update
 882  * @supsz: byte size of suinfo_update
 883  * @nsup: size of suinfo_update array
 884  *
 885  * Description: Takes an array of nilfs_suinfo_update structs and updates
 886  * segment usage accordingly. Only the fields indicated by the sup_flags
 887  * are updated.
 888  *
 889  * Return Value: On success, 0 is returned. On error, one of the
 890  * following negative error codes is returned.
 891  *
 892  * %-EIO - I/O error.
 893  *
 894  * %-ENOMEM - Insufficient amount of memory available.
 895  *
 896  * %-EINVAL - Invalid values in input (segment number, flags or nblocks)
 897  */
 898 ssize_t nilfs_sufile_set_suinfo(struct inode *sufile, void *buf,
 899                                 unsigned int supsz, size_t nsup)
 900 {
 901         struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
 902         struct buffer_head *header_bh, *bh;
 903         struct nilfs_suinfo_update *sup, *supend = buf + supsz * nsup;
 904         struct nilfs_segment_usage *su;
 905         void *kaddr;
 906         unsigned long blkoff, prev_blkoff;
 907         int cleansi, cleansu, dirtysi, dirtysu;
 908         long ncleaned = 0, ndirtied = 0;
 909         int ret = 0;
 910
 911         if (unlikely(nsup == 0))
 912                 return ret;
 913
 914         for (sup = buf; sup < supend; sup = (void *)sup + supsz) {
 915                 if (sup->sup_segnum >= nilfs->ns_nsegments
 916                         || (sup->sup_flags &
 917                                 (~0UL << __NR_NILFS_SUINFO_UPDATE_FIELDS))
 918                         || (nilfs_suinfo_update_nblocks(sup) &&
 919                                 sup->sup_sui.sui_nblocks >
 920                                 nilfs->ns_blocks_per_segment))
 921                         return -EINVAL;
 922         }
 923
 924         down_write(&NILFS_MDT(sufile)->mi_sem);
 925
 926         ret = nilfs_sufile_get_header_block(sufile, &header_bh);
 927         if (ret < 0)
 928                 goto out_sem;
 929
 930         sup = buf;
 931         blkoff = nilfs_sufile_get_blkoff(sufile, sup->sup_segnum);
 932         ret = nilfs_mdt_get_block(sufile, blkoff, 1, NULL, &bh);
 933         if (ret < 0)
 934                 goto out_header;
 935
 936         for (;;) {
 937                 kaddr = kmap_atomic(bh->b_page);
 938                 su = nilfs_sufile_block_get_segment_usage(
 939                         sufile, sup->sup_segnum, bh, kaddr);
 940
 941                 if (nilfs_suinfo_update_lastmod(sup))
 942                         su->su_lastmod = cpu_to_le64(sup->sup_sui.sui_lastmod);
 943
 944                 if (nilfs_suinfo_update_nblocks(sup))
 945                         su->su_nblocks = cpu_to_le32(sup->sup_sui.sui_nblocks);
 946
 947                 if (nilfs_suinfo_update_flags(sup)) {
 948                         /*
 949                          * Active flag is a virtual flag projected by running
 950                          * nilfs kernel code - drop it not to write it to
 951                          * disk.
 952                          */
 953                         sup->sup_sui.sui_flags &=
 954                                         ~BIT(NILFS_SEGMENT_USAGE_ACTIVE);
 955
 956                         cleansi = nilfs_suinfo_clean(&sup->sup_sui);
 957                         cleansu = nilfs_segment_usage_clean(su);
 958                         dirtysi = nilfs_suinfo_dirty(&sup->sup_sui);
 959                         dirtysu = nilfs_segment_usage_dirty(su);
 960
 961                         if (cleansi && !cleansu)
 962                                 ++ncleaned;
 963                         else if (!cleansi && cleansu)
 964                                 --ncleaned;
 965
 966                         if (dirtysi && !dirtysu)
 967                                 ++ndirtied;
 968                         else if (!dirtysi && dirtysu)
 969                                 --ndirtied;
 970
 971                         su->su_flags = cpu_to_le32(sup->sup_sui.sui_flags);
 972                 }
 973
 974                 kunmap_atomic(kaddr);
 975
 976                 sup = (void *)sup + supsz;
 977                 if (sup >= supend)
 978                         break;
 979
 980                 prev_blkoff = blkoff;
 981                 blkoff = nilfs_sufile_get_blkoff(sufile, sup->sup_segnum);
 982                 if (blkoff == prev_blkoff)
 983                         continue;
 984
 985                 /* get different block */
 986                 mark_buffer_dirty(bh);
 987                 put_bh(bh);
 988                 ret = nilfs_mdt_get_block(sufile, blkoff, 1, NULL, &bh);
 989                 if (unlikely(ret < 0))
 990                         goto out_mark;
 991         }
 992         mark_buffer_dirty(bh);
 993         put_bh(bh);
 994
 995  out_mark:
 996         if (ncleaned || ndirtied) {
 997                 nilfs_sufile_mod_counter(header_bh, (u64)ncleaned,
 998                                 (u64)ndirtied);
 999                 NILFS_SUI(sufile)->ncleansegs += ncleaned;
1000         }
1001         nilfs_mdt_mark_dirty(sufile);
1002  out_header:
1003         put_bh(header_bh);
1004  out_sem:
1005         up_write(&NILFS_MDT(sufile)->mi_sem);
1006         return ret;
1007 }
1008
1009 /**
1010  * nilfs_sufile_trim_fs() - trim ioctl handle function
1011  * @sufile: inode of segment usage file
1012  * @range: fstrim_range structure
1013  *
1014  * start:       First Byte to trim
1015  * len:         number of Bytes to trim from start
1016  * minlen:      minimum extent length in Bytes
1017  *
1018  * Decription: nilfs_sufile_trim_fs goes through all segments containing bytes
1019  * from start to start+len. start is rounded up to the next block boundary
1020  * and start+len is rounded down. For each clean segment blkdev_issue_discard
1021  * function is invoked.
1022  *
1023  * Return Value: On success, 0 is returned or negative error code, otherwise.
1024  */
1025 int nilfs_sufile_trim_fs(struct inode *sufile, struct fstrim_range *range)
1026 {
1027         struct the_nilfs *nilfs = sufile->i_sb->s_fs_info;
1028         struct buffer_head *su_bh;
1029         struct nilfs_segment_usage *su;
1030         void *kaddr;
1031         size_t n, i, susz = NILFS_MDT(sufile)->mi_entry_size;
1032         sector_t seg_start, seg_end, start_block, end_block;
1033         sector_t start = 0, nblocks = 0;
1034         u64 segnum, segnum_end, minlen, len, max_blocks, ndiscarded = 0;
1035         int ret = 0;
1036         unsigned int sects_per_block;
1037
1038         sects_per_block = (1 << nilfs->ns_blocksize_bits) /
1039                         bdev_logical_block_size(nilfs->ns_bdev);
1040         len = range->len >> nilfs->ns_blocksize_bits;
1041         minlen = range->minlen >> nilfs->ns_blocksize_bits;
1042         max_blocks = ((u64)nilfs->ns_nsegments * nilfs->ns_blocks_per_segment);
1043
1044         if (!len || range->start >= max_blocks << nilfs->ns_blocksize_bits)
1045                 return -EINVAL;
1046
1047         start_block = (range->start + nilfs->ns_blocksize - 1) >>
1048                         nilfs->ns_blocksize_bits;
1049
1050         /*
1051          * range->len can be very large (actually, it is set to
1052          * ULLONG_MAX by default) - truncate upper end of the range
1053          * carefully so as not to overflow.
1054          */
1055         if (max_blocks - start_block < len)
1056                 end_block = max_blocks - 1;
1057         else
1058                 end_block = start_block + len - 1;
1059
1060         segnum = nilfs_get_segnum_of_block(nilfs, start_block);
1061         segnum_end = nilfs_get_segnum_of_block(nilfs, end_block);
1062
1063         down_read(&NILFS_MDT(sufile)->mi_sem);
1064
1065         while (segnum <= segnum_end) {
1066                 n = nilfs_sufile_segment_usages_in_block(sufile, segnum,
1067                                 segnum_end);
1068
1069                 ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0,
1070                                                            &su_bh);
1071                 if (ret < 0) {
1072                         if (ret != -ENOENT)
1073                                 goto out_sem;
1074                         /* hole */
1075                         segnum += n;
1076                         continue;
1077                 }
1078
1079                 kaddr = kmap_atomic(su_bh->b_page);
1080                 su = nilfs_sufile_block_get_segment_usage(sufile, segnum,
1081                                 su_bh, kaddr);
1082                 for (i = 0; i < n; ++i, ++segnum, su = (void *)su + susz) {
1083                         if (!nilfs_segment_usage_clean(su))
1084                                 continue;
1085
1086                         nilfs_get_segment_range(nilfs, segnum, &seg_start,
1087                                                 &seg_end);
1088
1089                         if (!nblocks) {
1090                                 /* start new extent */
1091                                 start = seg_start;
1092                                 nblocks = seg_end - seg_start + 1;
1093                                 continue;
1094                         }
1095
1096                         if (start + nblocks == seg_start) {
1097                                 /* add to previous extent */
1098                                 nblocks += seg_end - seg_start + 1;
1099                                 continue;
1100                         }
1101
1102                         /* discard previous extent */
1103                         if (start < start_block) {
1104                                 nblocks -= start_block - start;
1105                                 start = start_block;
1106                         }
1107
1108                         if (nblocks >= minlen) {
1109                                 kunmap_atomic(kaddr);
1110
1111                                 ret = blkdev_issue_discard(nilfs->ns_bdev,
1112                                                 start * sects_per_block,
1113                                                 nblocks * sects_per_block,
1114                                                 GFP_NOFS, 0);
1115                                 if (ret < 0) {
1116                                         put_bh(su_bh);
1117                                         goto out_sem;
1118                                 }
1119
1120                                 ndiscarded += nblocks;
1121                                 kaddr = kmap_atomic(su_bh->b_page);
1122                                 su = nilfs_sufile_block_get_segment_usage(
1123                                         sufile, segnum, su_bh, kaddr);
1124                         }
1125
1126                         /* start new extent */
1127                         start = seg_start;
1128                         nblocks = seg_end - seg_start + 1;
1129                 }
1130                 kunmap_atomic(kaddr);
1131                 put_bh(su_bh);
1132         }
1133
1134
1135         if (nblocks) {
1136                 /* discard last extent */
1137                 if (start < start_block) {
1138                         nblocks -= start_block - start;
1139                         start = start_block;
1140                 }
1141                 if (start + nblocks > end_block + 1)
1142                         nblocks = end_block - start + 1;
1143
1144                 if (nblocks >= minlen) {
1145                         ret = blkdev_issue_discard(nilfs->ns_bdev,
1146                                         start * sects_per_block,
1147                                         nblocks * sects_per_block,
1148                                         GFP_NOFS, 0);
1149                         if (!ret)
1150                                 ndiscarded += nblocks;
1151                 }
1152         }
1153
1154 out_sem:
1155         up_read(&NILFS_MDT(sufile)->mi_sem);
1156
1157         range->len = ndiscarded << nilfs->ns_blocksize_bits;
1158         return ret;
1159 }
1160
1161 /**
1162  * nilfs_sufile_read - read or get sufile inode
1163  * @sb: super block instance
1164  * @susize: size of a segment usage entry
1165  * @raw_inode: on-disk sufile inode
1166  * @inodep: buffer to store the inode
1167  */
1168 int nilfs_sufile_read(struct super_block *sb, size_t susize,
1169                       struct nilfs_inode *raw_inode, struct inode **inodep)
1170 {
1171         struct inode *sufile;
1172         struct nilfs_sufile_info *sui;
1173         struct buffer_head *header_bh;
1174         struct nilfs_sufile_header *header;
1175         void *kaddr;
1176         int err;
1177
1178         if (susize > sb->s_blocksize) {
1179                 nilfs_msg(sb, KERN_ERR,
1180                           "too large segment usage size: %zu bytes", susize);
1181                 return -EINVAL;
1182         } else if (susize < NILFS_MIN_SEGMENT_USAGE_SIZE) {
1183                 nilfs_msg(sb, KERN_ERR,
1184                           "too small segment usage size: %zu bytes", susize);
1185                 return -EINVAL;
1186         }
1187
1188         sufile = nilfs_iget_locked(sb, NULL, NILFS_SUFILE_INO);
1189         if (unlikely(!sufile))
1190                 return -ENOMEM;
1191         if (!(sufile->i_state & I_NEW))
1192                 goto out;
1193
1194         err = nilfs_mdt_init(sufile, NILFS_MDT_GFP, sizeof(*sui));
1195         if (err)
1196                 goto failed;
1197
1198         nilfs_mdt_set_entry_size(sufile, susize,
1199                                  sizeof(struct nilfs_sufile_header));
1200
1201         err = nilfs_read_inode_common(sufile, raw_inode);
1202         if (err)
1203                 goto failed;
1204
1205         err = nilfs_sufile_get_header_block(sufile, &header_bh);
1206         if (err)
1207                 goto failed;
1208
1209         sui = NILFS_SUI(sufile);
1210         kaddr = kmap_atomic(header_bh->b_page);
1211         header = kaddr + bh_offset(header_bh);
1212         sui->ncleansegs = le64_to_cpu(header->sh_ncleansegs);
1213         kunmap_atomic(kaddr);
1214         brelse(header_bh);
1215
1216         sui->allocmax = nilfs_sufile_get_nsegments(sufile) - 1;
1217         sui->allocmin = 0;
1218
1219         unlock_new_inode(sufile);
1220  out:
1221         *inodep = sufile;
1222         return 0;
1223  failed:
1224         iget_failed(sufile);
1225         return err;
1226 }