fs/nilfs2/sufile.c

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