fs/nilfs2/segment.c

   1 /*
   2  * segment.c - NILFS segment constructor.
   3  *
   4  * Copyright (C) 2005-2008 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  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  19  *
  20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
  21  *
  22  */
  23
  24 #include <linux/pagemap.h>
  25 #include <linux/buffer_head.h>
  26 #include <linux/writeback.h>
  27 #include <linux/bio.h>
  28 #include <linux/completion.h>
  29 #include <linux/blkdev.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/freezer.h>
  32 #include <linux/kthread.h>
  33 #include <linux/crc32.h>
  34 #include <linux/pagevec.h>
  35 #include "nilfs.h"
  36 #include "btnode.h"
  37 #include "page.h"
  38 #include "segment.h"
  39 #include "sufile.h"
  40 #include "cpfile.h"
  41 #include "ifile.h"
  42 #include "seglist.h"
  43 #include "segbuf.h"
  44
  45
  46 /*
  47  * Segment constructor
  48  */
  49 #define SC_N_INODEVEC   16   /* Size of locally allocated inode vector */
  50
  51 #define SC_MAX_SEGDELTA 64   /* Upper limit of the number of segments
  52                                 appended in collection retry loop */
  53
  54 /* Construction mode */
  55 enum {
  56         SC_LSEG_SR = 1, /* Make a logical segment having a super root */
  57         SC_LSEG_DSYNC,  /* Flush data blocks of a given file and make
  58                            a logical segment without a super root */
  59         SC_FLUSH_FILE,  /* Flush data files, leads to segment writes without
  60                            creating a checkpoint */
  61         SC_FLUSH_DAT,   /* Flush DAT file. This also creates segments without
  62                            a checkpoint */
  63 };
  64
  65 /* Stage numbers of dirty block collection */
  66 enum {
  67         NILFS_ST_INIT = 0,
  68         NILFS_ST_GC,            /* Collecting dirty blocks for GC */
  69         NILFS_ST_FILE,
  70         NILFS_ST_SKETCH,
  71         NILFS_ST_IFILE,
  72         NILFS_ST_CPFILE,
  73         NILFS_ST_SUFILE,
  74         NILFS_ST_DAT,
  75         NILFS_ST_SR,            /* Super root */
  76         NILFS_ST_DSYNC,         /* Data sync blocks */
  77         NILFS_ST_DONE,
  78 };
  79
  80 /* State flags of collection */
  81 #define NILFS_CF_NODE           0x0001  /* Collecting node blocks */
  82 #define NILFS_CF_IFILE_STARTED  0x0002  /* IFILE stage has started */
  83 #define NILFS_CF_HISTORY_MASK   (NILFS_CF_IFILE_STARTED)
  84
  85 /* Operations depending on the construction mode and file type */
  86 struct nilfs_sc_operations {
  87         int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
  88                             struct inode *);
  89         int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
  90                             struct inode *);
  91         int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
  92                             struct inode *);
  93         void (*write_data_binfo)(struct nilfs_sc_info *,
  94                                  struct nilfs_segsum_pointer *,
  95                                  union nilfs_binfo *);
  96         void (*write_node_binfo)(struct nilfs_sc_info *,
  97                                  struct nilfs_segsum_pointer *,
  98                                  union nilfs_binfo *);
  99 };
 100
 101 /*
 102  * Other definitions
 103  */
 104 static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
 105 static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
 106 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
 107 static void nilfs_dispose_list(struct nilfs_sb_info *, struct list_head *,
 108                                int);
 109
 110 #define nilfs_cnt32_gt(a, b)   \
 111         (typecheck(__u32, a) && typecheck(__u32, b) && \
 112          ((__s32)(b) - (__s32)(a) < 0))
 113 #define nilfs_cnt32_ge(a, b)   \
 114         (typecheck(__u32, a) && typecheck(__u32, b) && \
 115          ((__s32)(a) - (__s32)(b) >= 0))
 116 #define nilfs_cnt32_lt(a, b)  nilfs_cnt32_gt(b, a)
 117 #define nilfs_cnt32_le(a, b)  nilfs_cnt32_ge(b, a)
 118
 119 /*
 120  * Transaction
 121  */
 122 static struct kmem_cache *nilfs_transaction_cachep;
 123
 124 /**
 125  * nilfs_init_transaction_cache - create a cache for nilfs_transaction_info
 126  *
 127  * nilfs_init_transaction_cache() creates a slab cache for the struct
 128  * nilfs_transaction_info.
 129  *
 130  * Return Value: On success, it returns 0. On error, one of the following
 131  * negative error code is returned.
 132  *
 133  * %-ENOMEM - Insufficient memory available.
 134  */
 135 int nilfs_init_transaction_cache(void)
 136 {
 137         nilfs_transaction_cachep =
 138                 kmem_cache_create("nilfs2_transaction_cache",
 139                                   sizeof(struct nilfs_transaction_info),
 140                                   0, SLAB_RECLAIM_ACCOUNT, NULL);
 141         return (nilfs_transaction_cachep == NULL) ? -ENOMEM : 0;
 142 }
 143
 144 /**
 145  * nilfs_detroy_transaction_cache - destroy the cache for transaction info
 146  *
 147  * nilfs_destroy_transaction_cache() frees the slab cache for the struct
 148  * nilfs_transaction_info.
 149  */
 150 void nilfs_destroy_transaction_cache(void)
 151 {
 152         kmem_cache_destroy(nilfs_transaction_cachep);
 153 }
 154
 155 static int nilfs_prepare_segment_lock(struct nilfs_transaction_info *ti)
 156 {
 157         struct nilfs_transaction_info *cur_ti = current->journal_info;
 158         void *save = NULL;
 159
 160         if (cur_ti) {
 161                 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
 162                         return ++cur_ti->ti_count;
 163                 else {
 164                         /*
 165                          * If journal_info field is occupied by other FS,
 166                          * it is saved and will be restored on
 167                          * nilfs_transaction_commit().
 168                          */
 169                         printk(KERN_WARNING
 170                                "NILFS warning: journal info from a different "
 171                                "FS\n");
 172                         save = current->journal_info;
 173                 }
 174         }
 175         if (!ti) {
 176                 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
 177                 if (!ti)
 178                         return -ENOMEM;
 179                 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
 180         } else {
 181                 ti->ti_flags = 0;
 182         }
 183         ti->ti_count = 0;
 184         ti->ti_save = save;
 185         ti->ti_magic = NILFS_TI_MAGIC;
 186         current->journal_info = ti;
 187         return 0;
 188 }
 189
 190 /**
 191  * nilfs_transaction_begin - start indivisible file operations.
 192  * @sb: super block
 193  * @ti: nilfs_transaction_info
 194  * @vacancy_check: flags for vacancy rate checks
 195  *
 196  * nilfs_transaction_begin() acquires a reader/writer semaphore, called
 197  * the segment semaphore, to make a segment construction and write tasks
 198  * exclusive.  The function is used with nilfs_transaction_commit() in pairs.
 199  * The region enclosed by these two functions can be nested.  To avoid a
 200  * deadlock, the semaphore is only acquired or released in the outermost call.
 201  *
 202  * This function allocates a nilfs_transaction_info struct to keep context
 203  * information on it.  It is initialized and hooked onto the current task in
 204  * the outermost call.  If a pre-allocated struct is given to @ti, it is used
 205  * instead; othewise a new struct is assigned from a slab.
 206  *
 207  * When @vacancy_check flag is set, this function will check the amount of
 208  * free space, and will wait for the GC to reclaim disk space if low capacity.
 209  *
 210  * Return Value: On success, 0 is returned. On error, one of the following
 211  * negative error code is returned.
 212  *
 213  * %-ENOMEM - Insufficient memory available.
 214  *
 215  * %-ENOSPC - No space left on device
 216  */
 217 int nilfs_transaction_begin(struct super_block *sb,
 218                             struct nilfs_transaction_info *ti,
 219                             int vacancy_check)
 220 {
 221         struct nilfs_sb_info *sbi;
 222         struct the_nilfs *nilfs;
 223         int ret = nilfs_prepare_segment_lock(ti);
 224
 225         if (unlikely(ret < 0))
 226                 return ret;
 227         if (ret > 0)
 228                 return 0;
 229
 230         sbi = NILFS_SB(sb);
 231         nilfs = sbi->s_nilfs;
 232         down_read(&nilfs->ns_segctor_sem);
 233         if (vacancy_check && nilfs_near_disk_full(nilfs)) {
 234                 up_read(&nilfs->ns_segctor_sem);
 235                 ret = -ENOSPC;
 236                 goto failed;
 237         }
 238         return 0;
 239
 240  failed:
 241         ti = current->journal_info;
 242         current->journal_info = ti->ti_save;
 243         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 244                 kmem_cache_free(nilfs_transaction_cachep, ti);
 245         return ret;
 246 }
 247
 248 /**
 249  * nilfs_transaction_commit - commit indivisible file operations.
 250  * @sb: super block
 251  *
 252  * nilfs_transaction_commit() releases the read semaphore which is
 253  * acquired by nilfs_transaction_begin(). This is only performed
 254  * in outermost call of this function.  If a commit flag is set,
 255  * nilfs_transaction_commit() sets a timer to start the segment
 256  * constructor.  If a sync flag is set, it starts construction
 257  * directly.
 258  */
 259 int nilfs_transaction_commit(struct super_block *sb)
 260 {
 261         struct nilfs_transaction_info *ti = current->journal_info;
 262         struct nilfs_sb_info *sbi;
 263         struct nilfs_sc_info *sci;
 264         int err = 0;
 265
 266         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 267         ti->ti_flags |= NILFS_TI_COMMIT;
 268         if (ti->ti_count > 0) {
 269                 ti->ti_count--;
 270                 return 0;
 271         }
 272         sbi = NILFS_SB(sb);
 273         sci = NILFS_SC(sbi);
 274         if (sci != NULL) {
 275                 if (ti->ti_flags & NILFS_TI_COMMIT)
 276                         nilfs_segctor_start_timer(sci);
 277                 if (atomic_read(&sbi->s_nilfs->ns_ndirtyblks) >
 278                     sci->sc_watermark)
 279                         nilfs_segctor_do_flush(sci, 0);
 280         }
 281         up_read(&sbi->s_nilfs->ns_segctor_sem);
 282         current->journal_info = ti->ti_save;
 283
 284         if (ti->ti_flags & NILFS_TI_SYNC)
 285                 err = nilfs_construct_segment(sb);
 286         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 287                 kmem_cache_free(nilfs_transaction_cachep, ti);
 288         return err;
 289 }
 290
 291 void nilfs_transaction_abort(struct super_block *sb)
 292 {
 293         struct nilfs_transaction_info *ti = current->journal_info;
 294
 295         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 296         if (ti->ti_count > 0) {
 297                 ti->ti_count--;
 298                 return;
 299         }
 300         up_read(&NILFS_SB(sb)->s_nilfs->ns_segctor_sem);
 301
 302         current->journal_info = ti->ti_save;
 303         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 304                 kmem_cache_free(nilfs_transaction_cachep, ti);
 305 }
 306
 307 void nilfs_relax_pressure_in_lock(struct super_block *sb)
 308 {
 309         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 310         struct nilfs_sc_info *sci = NILFS_SC(sbi);
 311         struct the_nilfs *nilfs = sbi->s_nilfs;
 312
 313         if (!sci || !sci->sc_flush_request)
 314                 return;
 315
 316         set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
 317         up_read(&nilfs->ns_segctor_sem);
 318
 319         down_write(&nilfs->ns_segctor_sem);
 320         if (sci->sc_flush_request &&
 321             test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
 322                 struct nilfs_transaction_info *ti = current->journal_info;
 323
 324                 ti->ti_flags |= NILFS_TI_WRITER;
 325                 nilfs_segctor_do_immediate_flush(sci);
 326                 ti->ti_flags &= ~NILFS_TI_WRITER;
 327         }
 328         downgrade_write(&nilfs->ns_segctor_sem);
 329 }
 330
 331 static void nilfs_transaction_lock(struct nilfs_sb_info *sbi,
 332                                    struct nilfs_transaction_info *ti,
 333                                    int gcflag)
 334 {
 335         struct nilfs_transaction_info *cur_ti = current->journal_info;
 336
 337         WARN_ON(cur_ti);
 338         ti->ti_flags = NILFS_TI_WRITER;
 339         ti->ti_count = 0;
 340         ti->ti_save = cur_ti;
 341         ti->ti_magic = NILFS_TI_MAGIC;
 342         INIT_LIST_HEAD(&ti->ti_garbage);
 343         current->journal_info = ti;
 344
 345         for (;;) {
 346                 down_write(&sbi->s_nilfs->ns_segctor_sem);
 347                 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &NILFS_SC(sbi)->sc_flags))
 348                         break;
 349
 350                 nilfs_segctor_do_immediate_flush(NILFS_SC(sbi));
 351
 352                 up_write(&sbi->s_nilfs->ns_segctor_sem);
 353                 yield();
 354         }
 355         if (gcflag)
 356                 ti->ti_flags |= NILFS_TI_GC;
 357 }
 358
 359 static void nilfs_transaction_unlock(struct nilfs_sb_info *sbi)
 360 {
 361         struct nilfs_transaction_info *ti = current->journal_info;
 362
 363         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 364         BUG_ON(ti->ti_count > 0);
 365
 366         up_write(&sbi->s_nilfs->ns_segctor_sem);
 367         current->journal_info = ti->ti_save;
 368         if (!list_empty(&ti->ti_garbage))
 369                 nilfs_dispose_list(sbi, &ti->ti_garbage, 0);
 370 }
 371
 372 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
 373                                             struct nilfs_segsum_pointer *ssp,
 374                                             unsigned bytes)
 375 {
 376         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 377         unsigned blocksize = sci->sc_super->s_blocksize;
 378         void *p;
 379
 380         if (unlikely(ssp->offset + bytes > blocksize)) {
 381                 ssp->offset = 0;
 382                 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
 383                                                &segbuf->sb_segsum_buffers));
 384                 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
 385         }
 386         p = ssp->bh->b_data + ssp->offset;
 387         ssp->offset += bytes;
 388         return p;
 389 }
 390
 391 /**
 392  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
 393  * @sci: nilfs_sc_info
 394  */
 395 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
 396 {
 397         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 398         struct buffer_head *sumbh;
 399         unsigned sumbytes;
 400         unsigned flags = 0;
 401         int err;
 402
 403         if (nilfs_doing_gc())
 404                 flags = NILFS_SS_GC;
 405         err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime);
 406         if (unlikely(err))
 407                 return err;
 408
 409         sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
 410         sumbytes = segbuf->sb_sum.sumbytes;
 411         sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
 412         sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
 413         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 414         return 0;
 415 }
 416
 417 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
 418 {
 419         sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
 420         if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
 421                 return -E2BIG; /* The current segment is filled up
 422                                   (internal code) */
 423         sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
 424         return nilfs_segctor_reset_segment_buffer(sci);
 425 }
 426
 427 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
 428 {
 429         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 430         int err;
 431
 432         if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
 433                 err = nilfs_segctor_feed_segment(sci);
 434                 if (err)
 435                         return err;
 436                 segbuf = sci->sc_curseg;
 437         }
 438         err = nilfs_segbuf_extend_payload(segbuf, &sci->sc_super_root);
 439         if (likely(!err))
 440                 segbuf->sb_sum.flags |= NILFS_SS_SR;
 441         return err;
 442 }
 443
 444 /*
 445  * Functions for making segment summary and payloads
 446  */
 447 static int nilfs_segctor_segsum_block_required(
 448         struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
 449         unsigned binfo_size)
 450 {
 451         unsigned blocksize = sci->sc_super->s_blocksize;
 452         /* Size of finfo and binfo is enough small against blocksize */
 453
 454         return ssp->offset + binfo_size +
 455                 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
 456                 blocksize;
 457 }
 458
 459 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
 460                                       struct inode *inode)
 461 {
 462         sci->sc_curseg->sb_sum.nfinfo++;
 463         sci->sc_binfo_ptr = sci->sc_finfo_ptr;
 464         nilfs_segctor_map_segsum_entry(
 465                 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
 466         /* skip finfo */
 467 }
 468
 469 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
 470                                     struct inode *inode)
 471 {
 472         struct nilfs_finfo *finfo;
 473         struct nilfs_inode_info *ii;
 474         struct nilfs_segment_buffer *segbuf;
 475
 476         if (sci->sc_blk_cnt == 0)
 477                 return;
 478
 479         ii = NILFS_I(inode);
 480         finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
 481                                                  sizeof(*finfo));
 482         finfo->fi_ino = cpu_to_le64(inode->i_ino);
 483         finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
 484         finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
 485         finfo->fi_cno = cpu_to_le64(ii->i_cno);
 486
 487         segbuf = sci->sc_curseg;
 488         segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
 489                 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
 490         sci->sc_finfo_ptr = sci->sc_binfo_ptr;
 491         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 492 }
 493
 494 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
 495                                         struct buffer_head *bh,
 496                                         struct inode *inode,
 497                                         unsigned binfo_size)
 498 {
 499         struct nilfs_segment_buffer *segbuf;
 500         int required, err = 0;
 501
 502  retry:
 503         segbuf = sci->sc_curseg;
 504         required = nilfs_segctor_segsum_block_required(
 505                 sci, &sci->sc_binfo_ptr, binfo_size);
 506         if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
 507                 nilfs_segctor_end_finfo(sci, inode);
 508                 err = nilfs_segctor_feed_segment(sci);
 509                 if (err)
 510                         return err;
 511                 goto retry;
 512         }
 513         if (unlikely(required)) {
 514                 err = nilfs_segbuf_extend_segsum(segbuf);
 515                 if (unlikely(err))
 516                         goto failed;
 517         }
 518         if (sci->sc_blk_cnt == 0)
 519                 nilfs_segctor_begin_finfo(sci, inode);
 520
 521         nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
 522         /* Substitution to vblocknr is delayed until update_blocknr() */
 523         nilfs_segbuf_add_file_buffer(segbuf, bh);
 524         sci->sc_blk_cnt++;
 525  failed:
 526         return err;
 527 }
 528
 529 static int nilfs_handle_bmap_error(int err, const char *fname,
 530                                    struct inode *inode, struct super_block *sb)
 531 {
 532         if (err == -EINVAL) {
 533                 nilfs_error(sb, fname, "broken bmap (inode=%lu)\n",
 534                             inode->i_ino);
 535                 err = -EIO;
 536         }
 537         return err;
 538 }
 539
 540 /*
 541  * Callback functions that enumerate, mark, and collect dirty blocks
 542  */
 543 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
 544                                    struct buffer_head *bh, struct inode *inode)
 545 {
 546         int err;
 547
 548         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 549         if (unlikely(err < 0))
 550                 return nilfs_handle_bmap_error(err, __func__, inode,
 551                                                sci->sc_super);
 552
 553         err = nilfs_segctor_add_file_block(sci, bh, inode,
 554                                            sizeof(struct nilfs_binfo_v));
 555         if (!err)
 556                 sci->sc_datablk_cnt++;
 557         return err;
 558 }
 559
 560 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
 561                                    struct buffer_head *bh,
 562                                    struct inode *inode)
 563 {
 564         int err;
 565
 566         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 567         if (unlikely(err < 0))
 568                 return nilfs_handle_bmap_error(err, __func__, inode,
 569                                                sci->sc_super);
 570         return 0;
 571 }
 572
 573 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
 574                                    struct buffer_head *bh,
 575                                    struct inode *inode)
 576 {
 577         WARN_ON(!buffer_dirty(bh));
 578         return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 579 }
 580
 581 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
 582                                         struct nilfs_segsum_pointer *ssp,
 583                                         union nilfs_binfo *binfo)
 584 {
 585         struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
 586                 sci, ssp, sizeof(*binfo_v));
 587         *binfo_v = binfo->bi_v;
 588 }
 589
 590 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
 591                                         struct nilfs_segsum_pointer *ssp,
 592                                         union nilfs_binfo *binfo)
 593 {
 594         __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
 595                 sci, ssp, sizeof(*vblocknr));
 596         *vblocknr = binfo->bi_v.bi_vblocknr;
 597 }
 598
 599 struct nilfs_sc_operations nilfs_sc_file_ops = {
 600         .collect_data = nilfs_collect_file_data,
 601         .collect_node = nilfs_collect_file_node,
 602         .collect_bmap = nilfs_collect_file_bmap,
 603         .write_data_binfo = nilfs_write_file_data_binfo,
 604         .write_node_binfo = nilfs_write_file_node_binfo,
 605 };
 606
 607 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
 608                                   struct buffer_head *bh, struct inode *inode)
 609 {
 610         int err;
 611
 612         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 613         if (unlikely(err < 0))
 614                 return nilfs_handle_bmap_error(err, __func__, inode,
 615                                                sci->sc_super);
 616
 617         err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 618         if (!err)
 619                 sci->sc_datablk_cnt++;
 620         return err;
 621 }
 622
 623 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
 624                                   struct buffer_head *bh, struct inode *inode)
 625 {
 626         WARN_ON(!buffer_dirty(bh));
 627         return nilfs_segctor_add_file_block(sci, bh, inode,
 628                                             sizeof(struct nilfs_binfo_dat));
 629 }
 630
 631 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
 632                                        struct nilfs_segsum_pointer *ssp,
 633                                        union nilfs_binfo *binfo)
 634 {
 635         __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
 636                                                           sizeof(*blkoff));
 637         *blkoff = binfo->bi_dat.bi_blkoff;
 638 }
 639
 640 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
 641                                        struct nilfs_segsum_pointer *ssp,
 642                                        union nilfs_binfo *binfo)
 643 {
 644         struct nilfs_binfo_dat *binfo_dat =
 645                 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
 646         *binfo_dat = binfo->bi_dat;
 647 }
 648
 649 struct nilfs_sc_operations nilfs_sc_dat_ops = {
 650         .collect_data = nilfs_collect_dat_data,
 651         .collect_node = nilfs_collect_file_node,
 652         .collect_bmap = nilfs_collect_dat_bmap,
 653         .write_data_binfo = nilfs_write_dat_data_binfo,
 654         .write_node_binfo = nilfs_write_dat_node_binfo,
 655 };
 656
 657 struct nilfs_sc_operations nilfs_sc_dsync_ops = {
 658         .collect_data = nilfs_collect_file_data,
 659         .collect_node = NULL,
 660         .collect_bmap = NULL,
 661         .write_data_binfo = nilfs_write_file_data_binfo,
 662         .write_node_binfo = NULL,
 663 };
 664
 665 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
 666                                               struct list_head *listp,
 667                                               size_t nlimit,
 668                                               loff_t start, loff_t end)
 669 {
 670         struct address_space *mapping = inode->i_mapping;
 671         struct pagevec pvec;
 672         pgoff_t index = 0, last = ULONG_MAX;
 673         size_t ndirties = 0;
 674         int i;
 675
 676         if (unlikely(start != 0 || end != LLONG_MAX)) {
 677                 /*
 678                  * A valid range is given for sync-ing data pages. The
 679                  * range is rounded to per-page; extra dirty buffers
 680                  * may be included if blocksize < pagesize.
 681                  */
 682                 index = start >> PAGE_SHIFT;
 683                 last = end >> PAGE_SHIFT;
 684         }
 685         pagevec_init(&pvec, 0);
 686  repeat:
 687         if (unlikely(index > last) ||
 688             !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 689                                 min_t(pgoff_t, last - index,
 690                                       PAGEVEC_SIZE - 1) + 1))
 691                 return ndirties;
 692
 693         for (i = 0; i < pagevec_count(&pvec); i++) {
 694                 struct buffer_head *bh, *head;
 695                 struct page *page = pvec.pages[i];
 696
 697                 if (unlikely(page->index > last))
 698                         break;
 699
 700                 if (mapping->host) {
 701                         lock_page(page);
 702                         if (!page_has_buffers(page))
 703                                 create_empty_buffers(page,
 704                                                      1 << inode->i_blkbits, 0);
 705                         unlock_page(page);
 706                 }
 707
 708                 bh = head = page_buffers(page);
 709                 do {
 710                         if (!buffer_dirty(bh))
 711                                 continue;
 712                         get_bh(bh);
 713                         list_add_tail(&bh->b_assoc_buffers, listp);
 714                         ndirties++;
 715                         if (unlikely(ndirties >= nlimit)) {
 716                                 pagevec_release(&pvec);
 717                                 cond_resched();
 718                                 return ndirties;
 719                         }
 720                 } while (bh = bh->b_this_page, bh != head);
 721         }
 722         pagevec_release(&pvec);
 723         cond_resched();
 724         goto repeat;
 725 }
 726
 727 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
 728                                             struct list_head *listp)
 729 {
 730         struct nilfs_inode_info *ii = NILFS_I(inode);
 731         struct address_space *mapping = &ii->i_btnode_cache;
 732         struct pagevec pvec;
 733         struct buffer_head *bh, *head;
 734         unsigned int i;
 735         pgoff_t index = 0;
 736
 737         pagevec_init(&pvec, 0);
 738
 739         while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 740                                   PAGEVEC_SIZE)) {
 741                 for (i = 0; i < pagevec_count(&pvec); i++) {
 742                         bh = head = page_buffers(pvec.pages[i]);
 743                         do {
 744                                 if (buffer_dirty(bh)) {
 745                                         get_bh(bh);
 746                                         list_add_tail(&bh->b_assoc_buffers,
 747                                                       listp);
 748                                 }
 749                                 bh = bh->b_this_page;
 750                         } while (bh != head);
 751                 }
 752                 pagevec_release(&pvec);
 753                 cond_resched();
 754         }
 755 }
 756
 757 static void nilfs_dispose_list(struct nilfs_sb_info *sbi,
 758                                struct list_head *head, int force)
 759 {
 760         struct nilfs_inode_info *ii, *n;
 761         struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
 762         unsigned nv = 0;
 763
 764         while (!list_empty(head)) {
 765                 spin_lock(&sbi->s_inode_lock);
 766                 list_for_each_entry_safe(ii, n, head, i_dirty) {
 767                         list_del_init(&ii->i_dirty);
 768                         if (force) {
 769                                 if (unlikely(ii->i_bh)) {
 770                                         brelse(ii->i_bh);
 771                                         ii->i_bh = NULL;
 772                                 }
 773                         } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
 774                                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 775                                 list_add_tail(&ii->i_dirty,
 776                                               &sbi->s_dirty_files);
 777                                 continue;
 778                         }
 779                         ivec[nv++] = ii;
 780                         if (nv == SC_N_INODEVEC)
 781                                 break;
 782                 }
 783                 spin_unlock(&sbi->s_inode_lock);
 784
 785                 for (pii = ivec; nv > 0; pii++, nv--)
 786                         iput(&(*pii)->vfs_inode);
 787         }
 788 }
 789
 790 static int nilfs_test_metadata_dirty(struct nilfs_sb_info *sbi)
 791 {
 792         struct the_nilfs *nilfs = sbi->s_nilfs;
 793         int ret = 0;
 794
 795         if (nilfs_mdt_fetch_dirty(sbi->s_ifile))
 796                 ret++;
 797         if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
 798                 ret++;
 799         if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
 800                 ret++;
 801         if (ret || nilfs_doing_gc())
 802                 if (nilfs_mdt_fetch_dirty(nilfs_dat_inode(nilfs)))
 803                         ret++;
 804         return ret;
 805 }
 806
 807 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
 808 {
 809         return list_empty(&sci->sc_dirty_files) &&
 810                 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
 811                 list_empty(&sci->sc_cleaning_segments) &&
 812                 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
 813 }
 814
 815 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
 816 {
 817         struct nilfs_sb_info *sbi = sci->sc_sbi;
 818         int ret = 0;
 819
 820         if (nilfs_test_metadata_dirty(sbi))
 821                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
 822
 823         spin_lock(&sbi->s_inode_lock);
 824         if (list_empty(&sbi->s_dirty_files) && nilfs_segctor_clean(sci))
 825                 ret++;
 826
 827         spin_unlock(&sbi->s_inode_lock);
 828         return ret;
 829 }
 830
 831 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
 832 {
 833         struct nilfs_sb_info *sbi = sci->sc_sbi;
 834         struct the_nilfs *nilfs = sbi->s_nilfs;
 835
 836         nilfs_mdt_clear_dirty(sbi->s_ifile);
 837         nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
 838         nilfs_mdt_clear_dirty(nilfs->ns_sufile);
 839         nilfs_mdt_clear_dirty(nilfs_dat_inode(nilfs));
 840 }
 841
 842 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
 843 {
 844         struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
 845         struct buffer_head *bh_cp;
 846         struct nilfs_checkpoint *raw_cp;
 847         int err;
 848
 849         /* XXX: this interface will be changed */
 850         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
 851                                           &raw_cp, &bh_cp);
 852         if (likely(!err)) {
 853                 /* The following code is duplicated with cpfile.  But, it is
 854                    needed to collect the checkpoint even if it was not newly
 855                    created */
 856                 nilfs_mdt_mark_buffer_dirty(bh_cp);
 857                 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
 858                 nilfs_cpfile_put_checkpoint(
 859                         nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 860         } else
 861                 WARN_ON(err == -EINVAL || err == -ENOENT);
 862
 863         return err;
 864 }
 865
 866 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
 867 {
 868         struct nilfs_sb_info *sbi = sci->sc_sbi;
 869         struct the_nilfs *nilfs = sbi->s_nilfs;
 870         struct buffer_head *bh_cp;
 871         struct nilfs_checkpoint *raw_cp;
 872         int err;
 873
 874         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
 875                                           &raw_cp, &bh_cp);
 876         if (unlikely(err)) {
 877                 WARN_ON(err == -EINVAL || err == -ENOENT);
 878                 goto failed_ibh;
 879         }
 880         raw_cp->cp_snapshot_list.ssl_next = 0;
 881         raw_cp->cp_snapshot_list.ssl_prev = 0;
 882         raw_cp->cp_inodes_count =
 883                 cpu_to_le64(atomic_read(&sbi->s_inodes_count));
 884         raw_cp->cp_blocks_count =
 885                 cpu_to_le64(atomic_read(&sbi->s_blocks_count));
 886         raw_cp->cp_nblk_inc =
 887                 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
 888         raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
 889         raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
 890         if (sci->sc_sketch_inode && i_size_read(sci->sc_sketch_inode) > 0)
 891                 nilfs_checkpoint_set_sketch(raw_cp);
 892         nilfs_write_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode, 1);
 893         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 894         return 0;
 895
 896  failed_ibh:
 897         return err;
 898 }
 899
 900 static void nilfs_fill_in_file_bmap(struct inode *ifile,
 901                                     struct nilfs_inode_info *ii)
 902
 903 {
 904         struct buffer_head *ibh;
 905         struct nilfs_inode *raw_inode;
 906
 907         if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
 908                 ibh = ii->i_bh;
 909                 BUG_ON(!ibh);
 910                 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
 911                                                   ibh);
 912                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 913                 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
 914         }
 915 }
 916
 917 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci,
 918                                             struct inode *ifile)
 919 {
 920         struct nilfs_inode_info *ii;
 921
 922         list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
 923                 nilfs_fill_in_file_bmap(ifile, ii);
 924                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
 925         }
 926         if (sci->sc_sketch_inode) {
 927                 ii = NILFS_I(sci->sc_sketch_inode);
 928                 if (test_bit(NILFS_I_DIRTY, &ii->i_state))
 929                         nilfs_fill_in_file_bmap(ifile, ii);
 930         }
 931 }
 932
 933 /*
 934  * CRC calculation routines
 935  */
 936 static void nilfs_fill_in_super_root_crc(struct buffer_head *bh_sr, u32 seed)
 937 {
 938         struct nilfs_super_root *raw_sr =
 939                 (struct nilfs_super_root *)bh_sr->b_data;
 940         u32 crc;
 941
 942         crc = crc32_le(seed,
 943                        (unsigned char *)raw_sr + sizeof(raw_sr->sr_sum),
 944                        NILFS_SR_BYTES - sizeof(raw_sr->sr_sum));
 945         raw_sr->sr_sum = cpu_to_le32(crc);
 946 }
 947
 948 static void nilfs_segctor_fill_in_checksums(struct nilfs_sc_info *sci,
 949                                             u32 seed)
 950 {
 951         struct nilfs_segment_buffer *segbuf;
 952
 953         if (sci->sc_super_root)
 954                 nilfs_fill_in_super_root_crc(sci->sc_super_root, seed);
 955
 956         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
 957                 nilfs_segbuf_fill_in_segsum_crc(segbuf, seed);
 958                 nilfs_segbuf_fill_in_data_crc(segbuf, seed);
 959         }
 960 }
 961
 962 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
 963                                              struct the_nilfs *nilfs)
 964 {
 965         struct buffer_head *bh_sr = sci->sc_super_root;
 966         struct nilfs_super_root *raw_sr =
 967                 (struct nilfs_super_root *)bh_sr->b_data;
 968         unsigned isz = nilfs->ns_inode_size;
 969
 970         raw_sr->sr_bytes = cpu_to_le16(NILFS_SR_BYTES);
 971         raw_sr->sr_nongc_ctime
 972                 = cpu_to_le64(nilfs_doing_gc() ?
 973                               nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
 974         raw_sr->sr_flags = 0;
 975
 976         nilfs_mdt_write_inode_direct(
 977                 nilfs_dat_inode(nilfs), bh_sr, NILFS_SR_DAT_OFFSET(isz));
 978         nilfs_mdt_write_inode_direct(
 979                 nilfs->ns_cpfile, bh_sr, NILFS_SR_CPFILE_OFFSET(isz));
 980         nilfs_mdt_write_inode_direct(
 981                 nilfs->ns_sufile, bh_sr, NILFS_SR_SUFILE_OFFSET(isz));
 982 }
 983
 984 static void nilfs_redirty_inodes(struct list_head *head)
 985 {
 986         struct nilfs_inode_info *ii;
 987
 988         list_for_each_entry(ii, head, i_dirty) {
 989                 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
 990                         clear_bit(NILFS_I_COLLECTED, &ii->i_state);
 991         }
 992 }
 993
 994 static void nilfs_drop_collected_inodes(struct list_head *head)
 995 {
 996         struct nilfs_inode_info *ii;
 997
 998         list_for_each_entry(ii, head, i_dirty) {
 999                 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1000                         continue;
1001
1002                 clear_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
1003                 set_bit(NILFS_I_UPDATED, &ii->i_state);
1004         }
1005 }
1006
1007 static void nilfs_segctor_cancel_free_segments(struct nilfs_sc_info *sci,
1008                                                struct inode *sufile)
1009
1010 {
1011         struct list_head *head = &sci->sc_cleaning_segments;
1012         struct nilfs_segment_entry *ent;
1013         int err;
1014
1015         list_for_each_entry(ent, head, list) {
1016                 if (!(ent->flags & NILFS_SLH_FREED))
1017                         break;
1018                 err = nilfs_sufile_cancel_free(sufile, ent->segnum);
1019                 WARN_ON(err); /* do not happen */
1020                 ent->flags &= ~NILFS_SLH_FREED;
1021         }
1022 }
1023
1024 static int nilfs_segctor_prepare_free_segments(struct nilfs_sc_info *sci,
1025                                                struct inode *sufile)
1026 {
1027         struct list_head *head = &sci->sc_cleaning_segments;
1028         struct nilfs_segment_entry *ent;
1029         int err;
1030
1031         list_for_each_entry(ent, head, list) {
1032                 err = nilfs_sufile_free(sufile, ent->segnum);
1033                 if (unlikely(err))
1034                         return err;
1035                 ent->flags |= NILFS_SLH_FREED;
1036         }
1037         return 0;
1038 }
1039
1040 static void nilfs_segctor_commit_free_segments(struct nilfs_sc_info *sci)
1041 {
1042         nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
1043 }
1044
1045 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1046                                        struct inode *inode,
1047                                        struct list_head *listp,
1048                                        int (*collect)(struct nilfs_sc_info *,
1049                                                       struct buffer_head *,
1050                                                       struct inode *))
1051 {
1052         struct buffer_head *bh, *n;
1053         int err = 0;
1054
1055         if (collect) {
1056                 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1057                         list_del_init(&bh->b_assoc_buffers);
1058                         err = collect(sci, bh, inode);
1059                         brelse(bh);
1060                         if (unlikely(err))
1061                                 goto dispose_buffers;
1062                 }
1063                 return 0;
1064         }
1065
1066  dispose_buffers:
1067         while (!list_empty(listp)) {
1068                 bh = list_entry(listp->next, struct buffer_head,
1069                                 b_assoc_buffers);
1070                 list_del_init(&bh->b_assoc_buffers);
1071                 brelse(bh);
1072         }
1073         return err;
1074 }
1075
1076 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1077 {
1078         /* Remaining number of blocks within segment buffer */
1079         return sci->sc_segbuf_nblocks -
1080                 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1081 }
1082
1083 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1084                                    struct inode *inode,
1085                                    struct nilfs_sc_operations *sc_ops)
1086 {
1087         LIST_HEAD(data_buffers);
1088         LIST_HEAD(node_buffers);
1089         int err;
1090
1091         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1092                 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1093
1094                 n = nilfs_lookup_dirty_data_buffers(
1095                         inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1096                 if (n > rest) {
1097                         err = nilfs_segctor_apply_buffers(
1098                                 sci, inode, &data_buffers,
1099                                 sc_ops->collect_data);
1100                         BUG_ON(!err); /* always receive -E2BIG or true error */
1101                         goto break_or_fail;
1102                 }
1103         }
1104         nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1105
1106         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1107                 err = nilfs_segctor_apply_buffers(
1108                         sci, inode, &data_buffers, sc_ops->collect_data);
1109                 if (unlikely(err)) {
1110                         /* dispose node list */
1111                         nilfs_segctor_apply_buffers(
1112                                 sci, inode, &node_buffers, NULL);
1113                         goto break_or_fail;
1114                 }
1115                 sci->sc_stage.flags |= NILFS_CF_NODE;
1116         }
1117         /* Collect node */
1118         err = nilfs_segctor_apply_buffers(
1119                 sci, inode, &node_buffers, sc_ops->collect_node);
1120         if (unlikely(err))
1121                 goto break_or_fail;
1122
1123         nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1124         err = nilfs_segctor_apply_buffers(
1125                 sci, inode, &node_buffers, sc_ops->collect_bmap);
1126         if (unlikely(err))
1127                 goto break_or_fail;
1128
1129         nilfs_segctor_end_finfo(sci, inode);
1130         sci->sc_stage.flags &= ~NILFS_CF_NODE;
1131
1132  break_or_fail:
1133         return err;
1134 }
1135
1136 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1137                                          struct inode *inode)
1138 {
1139         LIST_HEAD(data_buffers);
1140         size_t n, rest = nilfs_segctor_buffer_rest(sci);
1141         int err;
1142
1143         n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1144                                             sci->sc_dsync_start,
1145                                             sci->sc_dsync_end);
1146
1147         err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1148                                           nilfs_collect_file_data);
1149         if (!err) {
1150                 nilfs_segctor_end_finfo(sci, inode);
1151                 BUG_ON(n > rest);
1152                 /* always receive -E2BIG or true error if n > rest */
1153         }
1154         return err;
1155 }
1156
1157 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1158 {
1159         struct nilfs_sb_info *sbi = sci->sc_sbi;
1160         struct the_nilfs *nilfs = sbi->s_nilfs;
1161         struct list_head *head;
1162         struct nilfs_inode_info *ii;
1163         int err = 0;
1164
1165         switch (sci->sc_stage.scnt) {
1166         case NILFS_ST_INIT:
1167                 /* Pre-processes */
1168                 sci->sc_stage.flags = 0;
1169
1170                 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1171                         sci->sc_nblk_inc = 0;
1172                         sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1173                         if (mode == SC_LSEG_DSYNC) {
1174                                 sci->sc_stage.scnt = NILFS_ST_DSYNC;
1175                                 goto dsync_mode;
1176                         }
1177                 }
1178
1179                 sci->sc_stage.dirty_file_ptr = NULL;
1180                 sci->sc_stage.gc_inode_ptr = NULL;
1181                 if (mode == SC_FLUSH_DAT) {
1182                         sci->sc_stage.scnt = NILFS_ST_DAT;
1183                         goto dat_stage;
1184                 }
1185                 sci->sc_stage.scnt++;  /* Fall through */
1186         case NILFS_ST_GC:
1187                 if (nilfs_doing_gc()) {
1188                         head = &sci->sc_gc_inodes;
1189                         ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1190                                                 head, i_dirty);
1191                         list_for_each_entry_continue(ii, head, i_dirty) {
1192                                 err = nilfs_segctor_scan_file(
1193                                         sci, &ii->vfs_inode,
1194                                         &nilfs_sc_file_ops);
1195                                 if (unlikely(err)) {
1196                                         sci->sc_stage.gc_inode_ptr = list_entry(
1197                                                 ii->i_dirty.prev,
1198                                                 struct nilfs_inode_info,
1199                                                 i_dirty);
1200                                         goto break_or_fail;
1201                                 }
1202                                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1203                         }
1204                         sci->sc_stage.gc_inode_ptr = NULL;
1205                 }
1206                 sci->sc_stage.scnt++;  /* Fall through */
1207         case NILFS_ST_FILE:
1208                 head = &sci->sc_dirty_files;
1209                 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1210                                         i_dirty);
1211                 list_for_each_entry_continue(ii, head, i_dirty) {
1212                         clear_bit(NILFS_I_DIRTY, &ii->i_state);
1213
1214                         err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1215                                                       &nilfs_sc_file_ops);
1216                         if (unlikely(err)) {
1217                                 sci->sc_stage.dirty_file_ptr =
1218                                         list_entry(ii->i_dirty.prev,
1219                                                    struct nilfs_inode_info,
1220                                                    i_dirty);
1221                                 goto break_or_fail;
1222                         }
1223                         /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1224                         /* XXX: required ? */
1225                 }
1226                 sci->sc_stage.dirty_file_ptr = NULL;
1227                 if (mode == SC_FLUSH_FILE) {
1228                         sci->sc_stage.scnt = NILFS_ST_DONE;
1229                         return 0;
1230                 }
1231                 sci->sc_stage.scnt++;  /* Fall through */
1232         case NILFS_ST_SKETCH:
1233                 if (mode == SC_LSEG_SR && sci->sc_sketch_inode) {
1234                         ii = NILFS_I(sci->sc_sketch_inode);
1235                         if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
1236                                 sci->sc_sketch_inode->i_ctime.tv_sec
1237                                         = sci->sc_seg_ctime;
1238                                 sci->sc_sketch_inode->i_mtime.tv_sec
1239                                         = sci->sc_seg_ctime;
1240                                 err = nilfs_mark_inode_dirty(
1241                                         sci->sc_sketch_inode);
1242                                 if (unlikely(err))
1243                                         goto break_or_fail;
1244                         }
1245                         err = nilfs_segctor_scan_file(sci,
1246                                                       sci->sc_sketch_inode,
1247                                                       &nilfs_sc_file_ops);
1248                         if (unlikely(err))
1249                                 goto break_or_fail;
1250                 }
1251                 sci->sc_stage.scnt++;
1252                 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1253                 /* Fall through */
1254         case NILFS_ST_IFILE:
1255                 err = nilfs_segctor_scan_file(sci, sbi->s_ifile,
1256                                               &nilfs_sc_file_ops);
1257                 if (unlikely(err))
1258                         break;
1259                 sci->sc_stage.scnt++;
1260                 /* Creating a checkpoint */
1261                 err = nilfs_segctor_create_checkpoint(sci);
1262                 if (unlikely(err))
1263                         break;
1264                 /* Fall through */
1265         case NILFS_ST_CPFILE:
1266                 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1267                                               &nilfs_sc_file_ops);
1268                 if (unlikely(err))
1269                         break;
1270                 sci->sc_stage.scnt++;  /* Fall through */
1271         case NILFS_ST_SUFILE:
1272                 err = nilfs_segctor_prepare_free_segments(sci,
1273                                                           nilfs->ns_sufile);
1274                 if (unlikely(err))
1275                         break;
1276                 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1277                                               &nilfs_sc_file_ops);
1278                 if (unlikely(err))
1279                         break;
1280                 sci->sc_stage.scnt++;  /* Fall through */
1281         case NILFS_ST_DAT:
1282  dat_stage:
1283                 err = nilfs_segctor_scan_file(sci, nilfs_dat_inode(nilfs),
1284                                               &nilfs_sc_dat_ops);
1285                 if (unlikely(err))
1286                         break;
1287                 if (mode == SC_FLUSH_DAT) {
1288                         sci->sc_stage.scnt = NILFS_ST_DONE;
1289                         return 0;
1290                 }
1291                 sci->sc_stage.scnt++;  /* Fall through */
1292         case NILFS_ST_SR:
1293                 if (mode == SC_LSEG_SR) {
1294                         /* Appending a super root */
1295                         err = nilfs_segctor_add_super_root(sci);
1296                         if (unlikely(err))
1297                                 break;
1298                 }
1299                 /* End of a logical segment */
1300                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1301                 sci->sc_stage.scnt = NILFS_ST_DONE;
1302                 return 0;
1303         case NILFS_ST_DSYNC:
1304  dsync_mode:
1305                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1306                 ii = sci->sc_dsync_inode;
1307                 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1308                         break;
1309
1310                 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1311                 if (unlikely(err))
1312                         break;
1313                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1314                 sci->sc_stage.scnt = NILFS_ST_DONE;
1315                 return 0;
1316         case NILFS_ST_DONE:
1317                 return 0;
1318         default:
1319                 BUG();
1320         }
1321
1322  break_or_fail:
1323         return err;
1324 }
1325
1326 static int nilfs_segctor_terminate_segment(struct nilfs_sc_info *sci,
1327                                            struct nilfs_segment_buffer *segbuf,
1328                                            struct inode *sufile)
1329 {
1330         struct nilfs_segment_entry *ent = segbuf->sb_segent;
1331         int err;
1332
1333         err = nilfs_open_segment_entry(ent, sufile);
1334         if (unlikely(err))
1335                 return err;
1336         nilfs_mdt_mark_buffer_dirty(ent->bh_su);
1337         nilfs_mdt_mark_dirty(sufile);
1338         nilfs_close_segment_entry(ent, sufile);
1339
1340         list_add_tail(&ent->list, &sci->sc_active_segments);
1341         segbuf->sb_segent = NULL;
1342         return 0;
1343 }
1344
1345 static int nilfs_touch_segusage(struct inode *sufile, __u64 segnum)
1346 {
1347         struct buffer_head *bh_su;
1348         struct nilfs_segment_usage *raw_su;
1349         int err;
1350
1351         err = nilfs_sufile_get_segment_usage(sufile, segnum, &raw_su, &bh_su);
1352         if (unlikely(err))
1353                 return err;
1354         nilfs_mdt_mark_buffer_dirty(bh_su);
1355         nilfs_mdt_mark_dirty(sufile);
1356         nilfs_sufile_put_segment_usage(sufile, segnum, bh_su);
1357         return 0;
1358 }
1359
1360 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1361                                             struct the_nilfs *nilfs)
1362 {
1363         struct nilfs_segment_buffer *segbuf, *n;
1364         struct inode *sufile = nilfs->ns_sufile;
1365         __u64 nextnum;
1366         int err;
1367
1368         if (list_empty(&sci->sc_segbufs)) {
1369                 segbuf = nilfs_segbuf_new(sci->sc_super);
1370                 if (unlikely(!segbuf))
1371                         return -ENOMEM;
1372                 list_add(&segbuf->sb_list, &sci->sc_segbufs);
1373         } else
1374                 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1375
1376         err = nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1377                                nilfs->ns_pseg_offset, nilfs);
1378         if (unlikely(err))
1379                 return err;
1380
1381         if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1382                 err = nilfs_segctor_terminate_segment(sci, segbuf, sufile);
1383                 if (unlikely(err))
1384                         return err;
1385
1386                 nilfs_shift_to_next_segment(nilfs);
1387                 err = nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1388         }
1389         sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1390
1391         err = nilfs_touch_segusage(sufile, segbuf->sb_segnum);
1392         if (unlikely(err))
1393                 return err;
1394
1395         if (nilfs->ns_segnum == nilfs->ns_nextnum) {
1396                 /* Start from the head of a new full segment */
1397                 err = nilfs_sufile_alloc(sufile, &nextnum);
1398                 if (unlikely(err))
1399                         return err;
1400         } else
1401                 nextnum = nilfs->ns_nextnum;
1402
1403         segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1404         nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1405
1406         /* truncating segment buffers */
1407         list_for_each_entry_safe_continue(segbuf, n, &sci->sc_segbufs,
1408                                           sb_list) {
1409                 list_del_init(&segbuf->sb_list);
1410                 nilfs_segbuf_free(segbuf);
1411         }
1412         return err;
1413 }
1414
1415 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1416                                          struct the_nilfs *nilfs, int nadd)
1417 {
1418         struct nilfs_segment_buffer *segbuf, *prev, *n;
1419         struct inode *sufile = nilfs->ns_sufile;
1420         __u64 nextnextnum;
1421         LIST_HEAD(list);
1422         int err, ret, i;
1423
1424         prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1425         /*
1426          * Since the segment specified with nextnum might be allocated during
1427          * the previous construction, the buffer including its segusage may
1428          * not be dirty.  The following call ensures that the buffer is dirty
1429          * and will pin the buffer on memory until the sufile is written.
1430          */
1431         err = nilfs_touch_segusage(sufile, prev->sb_nextnum);
1432         if (unlikely(err))
1433                 return err;
1434
1435         for (i = 0; i < nadd; i++) {
1436                 /* extend segment info */
1437                 err = -ENOMEM;
1438                 segbuf = nilfs_segbuf_new(sci->sc_super);
1439                 if (unlikely(!segbuf))
1440                         goto failed;
1441
1442                 /* map this buffer to region of segment on-disk */
1443                 err = nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1444                 if (unlikely(err))
1445                         goto failed_segbuf;
1446
1447                 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1448
1449                 /* allocate the next next full segment */
1450                 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1451                 if (unlikely(err))
1452                         goto failed_segbuf;
1453
1454                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1455                 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1456
1457                 list_add_tail(&segbuf->sb_list, &list);
1458                 prev = segbuf;
1459         }
1460         list_splice(&list, sci->sc_segbufs.prev);
1461         return 0;
1462
1463  failed_segbuf:
1464         nilfs_segbuf_free(segbuf);
1465  failed:
1466         list_for_each_entry_safe(segbuf, n, &list, sb_list) {
1467                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1468                 WARN_ON(ret); /* never fails */
1469                 list_del_init(&segbuf->sb_list);
1470                 nilfs_segbuf_free(segbuf);
1471         }
1472         return err;
1473 }
1474
1475 static void nilfs_segctor_free_incomplete_segments(struct nilfs_sc_info *sci,
1476                                                    struct the_nilfs *nilfs)
1477 {
1478         struct nilfs_segment_buffer *segbuf;
1479         int ret, done = 0;
1480
1481         segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1482         if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1483                 ret = nilfs_sufile_free(nilfs->ns_sufile, segbuf->sb_nextnum);
1484                 WARN_ON(ret); /* never fails */
1485         }
1486         if (segbuf->sb_io_error) {
1487                 /* Case 1: The first segment failed */
1488                 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1489                         /* Case 1a:  Partial segment appended into an existing
1490                            segment */
1491                         nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1492                                                 segbuf->sb_fseg_end);
1493                 else /* Case 1b:  New full segment */
1494                         set_nilfs_discontinued(nilfs);
1495                 done++;
1496         }
1497
1498         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1499                 ret = nilfs_sufile_free(nilfs->ns_sufile, segbuf->sb_nextnum);
1500                 WARN_ON(ret); /* never fails */
1501                 if (!done && segbuf->sb_io_error) {
1502                         if (segbuf->sb_segnum != nilfs->ns_nextnum)
1503                                 /* Case 2: extended segment (!= next) failed */
1504                                 nilfs_sufile_set_error(nilfs->ns_sufile,
1505                                                        segbuf->sb_segnum);
1506                         done++;
1507                 }
1508         }
1509 }
1510
1511 static void nilfs_segctor_clear_segment_buffers(struct nilfs_sc_info *sci)
1512 {
1513         struct nilfs_segment_buffer *segbuf;
1514
1515         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list)
1516                 nilfs_segbuf_clear(segbuf);
1517         sci->sc_super_root = NULL;
1518 }
1519
1520 static void nilfs_segctor_destroy_segment_buffers(struct nilfs_sc_info *sci)
1521 {
1522         struct nilfs_segment_buffer *segbuf;
1523
1524         while (!list_empty(&sci->sc_segbufs)) {
1525                 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1526                 list_del_init(&segbuf->sb_list);
1527                 nilfs_segbuf_free(segbuf);
1528         }
1529         /* sci->sc_curseg = NULL; */
1530 }
1531
1532 static void nilfs_segctor_end_construction(struct nilfs_sc_info *sci,
1533                                            struct the_nilfs *nilfs, int err)
1534 {
1535         if (unlikely(err)) {
1536                 nilfs_segctor_free_incomplete_segments(sci, nilfs);
1537                 nilfs_segctor_cancel_free_segments(sci, nilfs->ns_sufile);
1538         }
1539         nilfs_segctor_clear_segment_buffers(sci);
1540 }
1541
1542 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1543                                           struct inode *sufile)
1544 {
1545         struct nilfs_segment_buffer *segbuf;
1546         struct buffer_head *bh_su;
1547         struct nilfs_segment_usage *raw_su;
1548         unsigned long live_blocks;
1549         int ret;
1550
1551         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1552                 ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1553                                                      &raw_su, &bh_su);
1554                 WARN_ON(ret); /* always succeed because bh_su is dirty */
1555                 live_blocks = segbuf->sb_sum.nblocks +
1556                         (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1557                 raw_su->su_lastmod = cpu_to_le64(sci->sc_seg_ctime);
1558                 raw_su->su_nblocks = cpu_to_le32(live_blocks);
1559                 nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum,
1560                                                bh_su);
1561         }
1562 }
1563
1564 static void nilfs_segctor_cancel_segusage(struct nilfs_sc_info *sci,
1565                                           struct inode *sufile)
1566 {
1567         struct nilfs_segment_buffer *segbuf;
1568         struct buffer_head *bh_su;
1569         struct nilfs_segment_usage *raw_su;
1570         int ret;
1571
1572         segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1573         ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1574                                              &raw_su, &bh_su);
1575         WARN_ON(ret); /* always succeed because bh_su is dirty */
1576         raw_su->su_nblocks = cpu_to_le32(segbuf->sb_pseg_start -
1577                                          segbuf->sb_fseg_start);
1578         nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum, bh_su);
1579
1580         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1581                 ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1582                                                      &raw_su, &bh_su);
1583                 WARN_ON(ret); /* always succeed */
1584                 raw_su->su_nblocks = 0;
1585                 nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum,
1586                                                bh_su);
1587         }
1588 }
1589
1590 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1591                                             struct nilfs_segment_buffer *last,
1592                                             struct inode *sufile)
1593 {
1594         struct nilfs_segment_buffer *segbuf = last, *n;
1595         int ret;
1596
1597         list_for_each_entry_safe_continue(segbuf, n, &sci->sc_segbufs,
1598                                           sb_list) {
1599                 list_del_init(&segbuf->sb_list);
1600                 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1601                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1602                 WARN_ON(ret);
1603                 nilfs_segbuf_free(segbuf);
1604         }
1605 }
1606
1607
1608 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1609                                  struct the_nilfs *nilfs, int mode)
1610 {
1611         struct nilfs_cstage prev_stage = sci->sc_stage;
1612         int err, nadd = 1;
1613
1614         /* Collection retry loop */
1615         for (;;) {
1616                 sci->sc_super_root = NULL;
1617                 sci->sc_nblk_this_inc = 0;
1618                 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1619
1620                 err = nilfs_segctor_reset_segment_buffer(sci);
1621                 if (unlikely(err))
1622                         goto failed;
1623
1624                 err = nilfs_segctor_collect_blocks(sci, mode);
1625                 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1626                 if (!err)
1627                         break;
1628
1629                 if (unlikely(err != -E2BIG))
1630                         goto failed;
1631
1632                 /* The current segment is filled up */
1633                 if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
1634                         break;
1635
1636                 nilfs_segctor_cancel_free_segments(sci, nilfs->ns_sufile);
1637                 nilfs_segctor_clear_segment_buffers(sci);
1638
1639                 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1640                 if (unlikely(err))
1641                         return err;
1642
1643                 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1644                 sci->sc_stage = prev_stage;
1645         }
1646         nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1647         return 0;
1648
1649  failed:
1650         return err;
1651 }
1652
1653 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1654                                       struct buffer_head *new_bh)
1655 {
1656         BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1657
1658         list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1659         /* The caller must release old_bh */
1660 }
1661
1662 static int
1663 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1664                                      struct nilfs_segment_buffer *segbuf,
1665                                      int mode)
1666 {
1667         struct inode *inode = NULL;
1668         sector_t blocknr;
1669         unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1670         unsigned long nblocks = 0, ndatablk = 0;
1671         struct nilfs_sc_operations *sc_op = NULL;
1672         struct nilfs_segsum_pointer ssp;
1673         struct nilfs_finfo *finfo = NULL;
1674         union nilfs_binfo binfo;
1675         struct buffer_head *bh, *bh_org;
1676         ino_t ino = 0;
1677         int err = 0;
1678
1679         if (!nfinfo)
1680                 goto out;
1681
1682         blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1683         ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1684         ssp.offset = sizeof(struct nilfs_segment_summary);
1685
1686         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1687                 if (bh == sci->sc_super_root)
1688                         break;
1689                 if (!finfo) {
1690                         finfo = nilfs_segctor_map_segsum_entry(
1691                                 sci, &ssp, sizeof(*finfo));
1692                         ino = le64_to_cpu(finfo->fi_ino);
1693                         nblocks = le32_to_cpu(finfo->fi_nblocks);
1694                         ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1695
1696                         if (buffer_nilfs_node(bh))
1697                                 inode = NILFS_BTNC_I(bh->b_page->mapping);
1698                         else
1699                                 inode = NILFS_AS_I(bh->b_page->mapping);
1700
1701                         if (mode == SC_LSEG_DSYNC)
1702                                 sc_op = &nilfs_sc_dsync_ops;
1703                         else if (ino == NILFS_DAT_INO)
1704                                 sc_op = &nilfs_sc_dat_ops;
1705                         else /* file blocks */
1706                                 sc_op = &nilfs_sc_file_ops;
1707                 }
1708                 bh_org = bh;
1709                 get_bh(bh_org);
1710                 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1711                                         &binfo);
1712                 if (bh != bh_org)
1713                         nilfs_list_replace_buffer(bh_org, bh);
1714                 brelse(bh_org);
1715                 if (unlikely(err))
1716                         goto failed_bmap;
1717
1718                 if (ndatablk > 0)
1719                         sc_op->write_data_binfo(sci, &ssp, &binfo);
1720                 else
1721                         sc_op->write_node_binfo(sci, &ssp, &binfo);
1722
1723                 blocknr++;
1724                 if (--nblocks == 0) {
1725                         finfo = NULL;
1726                         if (--nfinfo == 0)
1727                                 break;
1728                 } else if (ndatablk > 0)
1729                         ndatablk--;
1730         }
1731  out:
1732         return 0;
1733
1734  failed_bmap:
1735         err = nilfs_handle_bmap_error(err, __func__, inode, sci->sc_super);
1736         return err;
1737 }
1738
1739 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1740 {
1741         struct nilfs_segment_buffer *segbuf;
1742         int err;
1743
1744         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1745                 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1746                 if (unlikely(err))
1747                         return err;
1748                 nilfs_segbuf_fill_in_segsum(segbuf);
1749         }
1750         return 0;
1751 }
1752
1753 static int
1754 nilfs_copy_replace_page_buffers(struct page *page, struct list_head *out)
1755 {
1756         struct page *clone_page;
1757         struct buffer_head *bh, *head, *bh2;
1758         void *kaddr;
1759
1760         bh = head = page_buffers(page);
1761
1762         clone_page = nilfs_alloc_private_page(bh->b_bdev, bh->b_size, 0);
1763         if (unlikely(!clone_page))
1764                 return -ENOMEM;
1765
1766         bh2 = page_buffers(clone_page);
1767         kaddr = kmap_atomic(page, KM_USER0);
1768         do {
1769                 if (list_empty(&bh->b_assoc_buffers))
1770                         continue;
1771                 get_bh(bh2);
1772                 page_cache_get(clone_page); /* for each bh */
1773                 memcpy(bh2->b_data, kaddr + bh_offset(bh), bh2->b_size);
1774                 bh2->b_blocknr = bh->b_blocknr;
1775                 list_replace(&bh->b_assoc_buffers, &bh2->b_assoc_buffers);
1776                 list_add_tail(&bh->b_assoc_buffers, out);
1777         } while (bh = bh->b_this_page, bh2 = bh2->b_this_page, bh != head);
1778         kunmap_atomic(kaddr, KM_USER0);
1779
1780         if (!TestSetPageWriteback(clone_page))
1781                 inc_zone_page_state(clone_page, NR_WRITEBACK);
1782         unlock_page(clone_page);
1783
1784         return 0;
1785 }
1786
1787 static int nilfs_test_page_to_be_frozen(struct page *page)
1788 {
1789         struct address_space *mapping = page->mapping;
1790
1791         if (!mapping || !mapping->host || S_ISDIR(mapping->host->i_mode))
1792                 return 0;
1793
1794         if (page_mapped(page)) {
1795                 ClearPageChecked(page);
1796                 return 1;
1797         }
1798         return PageChecked(page);
1799 }
1800
1801 static int nilfs_begin_page_io(struct page *page, struct list_head *out)
1802 {
1803         if (!page || PageWriteback(page))
1804                 /* For split b-tree node pages, this function may be called
1805                    twice.  We ignore the 2nd or later calls by this check. */
1806                 return 0;
1807
1808         lock_page(page);
1809         clear_page_dirty_for_io(page);
1810         set_page_writeback(page);
1811         unlock_page(page);
1812
1813         if (nilfs_test_page_to_be_frozen(page)) {
1814                 int err = nilfs_copy_replace_page_buffers(page, out);
1815                 if (unlikely(err))
1816                         return err;
1817         }
1818         return 0;
1819 }
1820
1821 static int nilfs_segctor_prepare_write(struct nilfs_sc_info *sci,
1822                                        struct page **failed_page)
1823 {
1824         struct nilfs_segment_buffer *segbuf;
1825         struct page *bd_page = NULL, *fs_page = NULL;
1826         struct list_head *list = &sci->sc_copied_buffers;
1827         int err;
1828
1829         *failed_page = NULL;
1830         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1831                 struct buffer_head *bh;
1832
1833                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1834                                     b_assoc_buffers) {
1835                         if (bh->b_page != bd_page) {
1836                                 if (bd_page) {
1837                                         lock_page(bd_page);
1838                                         clear_page_dirty_for_io(bd_page);
1839                                         set_page_writeback(bd_page);
1840                                         unlock_page(bd_page);
1841                                 }
1842                                 bd_page = bh->b_page;
1843                         }
1844                 }
1845
1846                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1847                                     b_assoc_buffers) {
1848                         if (bh == sci->sc_super_root) {
1849                                 if (bh->b_page != bd_page) {
1850                                         lock_page(bd_page);
1851                                         clear_page_dirty_for_io(bd_page);
1852                                         set_page_writeback(bd_page);
1853                                         unlock_page(bd_page);
1854                                         bd_page = bh->b_page;
1855                                 }
1856                                 break;
1857                         }
1858                         if (bh->b_page != fs_page) {
1859                                 err = nilfs_begin_page_io(fs_page, list);
1860                                 if (unlikely(err)) {
1861                                         *failed_page = fs_page;
1862                                         goto out;
1863                                 }
1864                                 fs_page = bh->b_page;
1865                         }
1866                 }
1867         }
1868         if (bd_page) {
1869                 lock_page(bd_page);
1870                 clear_page_dirty_for_io(bd_page);
1871                 set_page_writeback(bd_page);
1872                 unlock_page(bd_page);
1873         }
1874         err = nilfs_begin_page_io(fs_page, list);
1875         if (unlikely(err))
1876                 *failed_page = fs_page;
1877  out:
1878         return err;
1879 }
1880
1881 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1882                                struct backing_dev_info *bdi)
1883 {
1884         struct nilfs_segment_buffer *segbuf;
1885         struct nilfs_write_info wi;
1886         int err, res;
1887
1888         wi.sb = sci->sc_super;
1889         wi.bh_sr = sci->sc_super_root;
1890         wi.bdi = bdi;
1891
1892         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1893                 nilfs_segbuf_prepare_write(segbuf, &wi);
1894                 err = nilfs_segbuf_write(segbuf, &wi);
1895
1896                 res = nilfs_segbuf_wait(segbuf, &wi);
1897                 err = unlikely(err) ? : res;
1898                 if (unlikely(err))
1899                         return err;
1900         }
1901         return 0;
1902 }
1903
1904 static int nilfs_page_has_uncleared_buffer(struct page *page)
1905 {
1906         struct buffer_head *head, *bh;
1907
1908         head = bh = page_buffers(page);
1909         do {
1910                 if (buffer_dirty(bh) && !list_empty(&bh->b_assoc_buffers))
1911                         return 1;
1912                 bh = bh->b_this_page;
1913         } while (bh != head);
1914         return 0;
1915 }
1916
1917 static void __nilfs_end_page_io(struct page *page, int err)
1918 {
1919         if (!err) {
1920                 if (!nilfs_page_buffers_clean(page))
1921                         __set_page_dirty_nobuffers(page);
1922                 ClearPageError(page);
1923         } else {
1924                 __set_page_dirty_nobuffers(page);
1925                 SetPageError(page);
1926         }
1927
1928         if (buffer_nilfs_allocated(page_buffers(page))) {
1929                 if (TestClearPageWriteback(page))
1930                         dec_zone_page_state(page, NR_WRITEBACK);
1931         } else
1932                 end_page_writeback(page);
1933 }
1934
1935 static void nilfs_end_page_io(struct page *page, int err)
1936 {
1937         if (!page)
1938                 return;
1939
1940         if (buffer_nilfs_node(page_buffers(page)) &&
1941             nilfs_page_has_uncleared_buffer(page))
1942                 /* For b-tree node pages, this function may be called twice
1943                    or more because they might be split in a segment.
1944                    This check assures that cleanup has been done for all
1945                    buffers in a split btnode page. */
1946                 return;
1947
1948         __nilfs_end_page_io(page, err);
1949 }
1950
1951 static void nilfs_clear_copied_buffers(struct list_head *list, int err)
1952 {
1953         struct buffer_head *bh, *head;
1954         struct page *page;
1955
1956         while (!list_empty(list)) {
1957                 bh = list_entry(list->next, struct buffer_head,
1958                                 b_assoc_buffers);
1959                 page = bh->b_page;
1960                 page_cache_get(page);
1961                 head = bh = page_buffers(page);
1962                 do {
1963                         if (!list_empty(&bh->b_assoc_buffers)) {
1964                                 list_del_init(&bh->b_assoc_buffers);
1965                                 if (!err) {
1966                                         set_buffer_uptodate(bh);
1967                                         clear_buffer_dirty(bh);
1968                                         clear_buffer_nilfs_volatile(bh);
1969                                 }
1970                                 brelse(bh); /* for b_assoc_buffers */
1971                         }
1972                 } while ((bh = bh->b_this_page) != head);
1973
1974                 __nilfs_end_page_io(page, err);
1975                 page_cache_release(page);
1976         }
1977 }
1978
1979 static void nilfs_segctor_abort_write(struct nilfs_sc_info *sci,
1980                                       struct page *failed_page, int err)
1981 {
1982         struct nilfs_segment_buffer *segbuf;
1983         struct page *bd_page = NULL, *fs_page = NULL;
1984
1985         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1986                 struct buffer_head *bh;
1987
1988                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1989                                     b_assoc_buffers) {
1990                         if (bh->b_page != bd_page) {
1991                                 if (bd_page)
1992                                         end_page_writeback(bd_page);
1993                                 bd_page = bh->b_page;
1994                         }
1995                 }
1996
1997                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1998                                     b_assoc_buffers) {
1999                         if (bh == sci->sc_super_root) {
2000                                 if (bh->b_page != bd_page) {
2001                                         end_page_writeback(bd_page);
2002                                         bd_page = bh->b_page;
2003                                 }
2004                                 break;
2005                         }
2006                         if (bh->b_page != fs_page) {
2007                                 nilfs_end_page_io(fs_page, err);
2008                                 if (unlikely(fs_page == failed_page))
2009                                         goto done;
2010                                 fs_page = bh->b_page;
2011                         }
2012                 }
2013         }
2014         if (bd_page)
2015                 end_page_writeback(bd_page);
2016
2017         nilfs_end_page_io(fs_page, err);
2018  done:
2019         nilfs_clear_copied_buffers(&sci->sc_copied_buffers, err);
2020 }
2021
2022 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
2023                                    struct nilfs_segment_buffer *segbuf)
2024 {
2025         nilfs->ns_segnum = segbuf->sb_segnum;
2026         nilfs->ns_nextnum = segbuf->sb_nextnum;
2027         nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
2028                 + segbuf->sb_sum.nblocks;
2029         nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
2030         nilfs->ns_ctime = segbuf->sb_sum.ctime;
2031 }
2032
2033 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
2034 {
2035         struct nilfs_segment_buffer *segbuf;
2036         struct page *bd_page = NULL, *fs_page = NULL;
2037         struct nilfs_sb_info *sbi = sci->sc_sbi;
2038         struct the_nilfs *nilfs = sbi->s_nilfs;
2039         int update_sr = (sci->sc_super_root != NULL);
2040
2041         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
2042                 struct buffer_head *bh;
2043
2044                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
2045                                     b_assoc_buffers) {
2046                         set_buffer_uptodate(bh);
2047                         clear_buffer_dirty(bh);
2048                         if (bh->b_page != bd_page) {
2049                                 if (bd_page)
2050                                         end_page_writeback(bd_page);
2051                                 bd_page = bh->b_page;
2052                         }
2053                 }
2054                 /*
2055                  * We assume that the buffers which belong to the same page
2056                  * continue over the buffer list.
2057                  * Under this assumption, the last BHs of pages is
2058                  * identifiable by the discontinuity of bh->b_page
2059                  * (page != fs_page).
2060                  *
2061                  * For B-tree node blocks, however, this assumption is not
2062                  * guaranteed.  The cleanup code of B-tree node pages needs
2063                  * special care.
2064                  */
2065                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
2066                                     b_assoc_buffers) {
2067                         set_buffer_uptodate(bh);
2068                         clear_buffer_dirty(bh);
2069                         clear_buffer_nilfs_volatile(bh);
2070                         if (bh == sci->sc_super_root) {
2071                                 if (bh->b_page != bd_page) {
2072                                         end_page_writeback(bd_page);
2073                                         bd_page = bh->b_page;
2074                                 }
2075                                 break;
2076                         }
2077                         if (bh->b_page != fs_page) {
2078                                 nilfs_end_page_io(fs_page, 0);
2079                                 fs_page = bh->b_page;
2080                         }
2081                 }
2082
2083                 if (!NILFS_SEG_SIMPLEX(&segbuf->sb_sum)) {
2084                         if (NILFS_SEG_LOGBGN(&segbuf->sb_sum)) {
2085                                 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
2086                                 sci->sc_lseg_stime = jiffies;
2087                         }
2088                         if (NILFS_SEG_LOGEND(&segbuf->sb_sum))
2089                                 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
2090                 }
2091         }
2092         /*
2093          * Since pages may continue over multiple segment buffers,
2094          * end of the last page must be checked outside of the loop.
2095          */
2096         if (bd_page)
2097                 end_page_writeback(bd_page);
2098
2099         nilfs_end_page_io(fs_page, 0);
2100
2101         nilfs_clear_copied_buffers(&sci->sc_copied_buffers, 0);
2102
2103         nilfs_drop_collected_inodes(&sci->sc_dirty_files);
2104
2105         if (nilfs_doing_gc()) {
2106                 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
2107                 if (update_sr)
2108                         nilfs_commit_gcdat_inode(nilfs);
2109         } else
2110                 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
2111
2112         sci->sc_nblk_inc += sci->sc_nblk_this_inc;
2113
2114         segbuf = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2115         nilfs_set_next_segment(nilfs, segbuf);
2116
2117         if (update_sr) {
2118                 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
2119                                        segbuf->sb_sum.seg_seq, nilfs->ns_cno);
2120
2121                 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2122                 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
2123         } else
2124                 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
2125 }
2126
2127 static int nilfs_segctor_check_in_files(struct nilfs_sc_info *sci,
2128                                         struct nilfs_sb_info *sbi)
2129 {
2130         struct nilfs_inode_info *ii, *n;
2131         __u64 cno = sbi->s_nilfs->ns_cno;
2132
2133         spin_lock(&sbi->s_inode_lock);
2134  retry:
2135         list_for_each_entry_safe(ii, n, &sbi->s_dirty_files, i_dirty) {
2136                 if (!ii->i_bh) {
2137                         struct buffer_head *ibh;
2138                         int err;
2139
2140                         spin_unlock(&sbi->s_inode_lock);
2141                         err = nilfs_ifile_get_inode_block(
2142                                 sbi->s_ifile, ii->vfs_inode.i_ino, &ibh);
2143                         if (unlikely(err)) {
2144                                 nilfs_warning(sbi->s_super, __func__,
2145                                               "failed to get inode block.\n");
2146                                 return err;
2147                         }
2148                         nilfs_mdt_mark_buffer_dirty(ibh);
2149                         nilfs_mdt_mark_dirty(sbi->s_ifile);
2150                         spin_lock(&sbi->s_inode_lock);
2151                         if (likely(!ii->i_bh))
2152                                 ii->i_bh = ibh;
2153                         else
2154                                 brelse(ibh);
2155                         goto retry;
2156                 }
2157                 ii->i_cno = cno;
2158
2159                 clear_bit(NILFS_I_QUEUED, &ii->i_state);
2160                 set_bit(NILFS_I_BUSY, &ii->i_state);
2161                 list_del(&ii->i_dirty);
2162                 list_add_tail(&ii->i_dirty, &sci->sc_dirty_files);
2163         }
2164         spin_unlock(&sbi->s_inode_lock);
2165
2166         NILFS_I(sbi->s_ifile)->i_cno = cno;
2167
2168         return 0;
2169 }
2170
2171 static void nilfs_segctor_check_out_files(struct nilfs_sc_info *sci,
2172                                           struct nilfs_sb_info *sbi)
2173 {
2174         struct nilfs_transaction_info *ti = current->journal_info;
2175         struct nilfs_inode_info *ii, *n;
2176         __u64 cno = sbi->s_nilfs->ns_cno;
2177
2178         spin_lock(&sbi->s_inode_lock);
2179         list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2180                 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2181                     test_bit(NILFS_I_DIRTY, &ii->i_state)) {
2182                         /* The current checkpoint number (=nilfs->ns_cno) is
2183                            changed between check-in and check-out only if the
2184                            super root is written out.  So, we can update i_cno
2185                            for the inodes that remain in the dirty list. */
2186                         ii->i_cno = cno;
2187                         continue;
2188                 }
2189                 clear_bit(NILFS_I_BUSY, &ii->i_state);
2190                 brelse(ii->i_bh);
2191                 ii->i_bh = NULL;
2192                 list_del(&ii->i_dirty);
2193                 list_add_tail(&ii->i_dirty, &ti->ti_garbage);
2194         }
2195         spin_unlock(&sbi->s_inode_lock);
2196 }
2197
2198 /*
2199  * Nasty routines to manipulate active flags on sufile.
2200  * These would be removed in a future release.
2201  */
2202 static void nilfs_segctor_reactivate_segments(struct nilfs_sc_info *sci,
2203                                               struct the_nilfs *nilfs)
2204 {
2205         struct nilfs_segment_buffer *segbuf, *last;
2206         struct nilfs_segment_entry *ent, *n;
2207         struct inode *sufile = nilfs->ns_sufile;
2208         struct list_head *head;
2209
2210         last = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2211         nilfs_for_each_segbuf_before(segbuf, last, &sci->sc_segbufs) {
2212                 ent = segbuf->sb_segent;
2213                 if (!ent)
2214                         break; /* ignore unmapped segments (should check it?)*/
2215                 nilfs_segment_usage_set_active(ent->raw_su);
2216                 nilfs_close_segment_entry(ent, sufile);
2217         }
2218
2219         head = &sci->sc_active_segments;
2220         list_for_each_entry_safe(ent, n, head, list) {
2221                 nilfs_segment_usage_set_active(ent->raw_su);
2222                 nilfs_close_segment_entry(ent, sufile);
2223         }
2224 }
2225
2226 static int nilfs_segctor_deactivate_segments(struct nilfs_sc_info *sci,
2227                                              struct the_nilfs *nilfs)
2228 {
2229         struct nilfs_segment_buffer *segbuf, *last;
2230         struct nilfs_segment_entry *ent;
2231         struct inode *sufile = nilfs->ns_sufile;
2232         int err;
2233
2234         last = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2235         nilfs_for_each_segbuf_before(segbuf, last, &sci->sc_segbufs) {
2236                 /*
2237                  * Deactivate ongoing full segments.  The last segment is kept
2238                  * active because it is a start point of recovery, and is not
2239                  * relocatable until the super block points to a newer
2240                  * checkpoint.
2241                  */
2242                 ent = segbuf->sb_segent;
2243                 if (!ent)
2244                         break; /* ignore unmapped segments (should check it?)*/
2245                 err = nilfs_open_segment_entry(ent, sufile);
2246                 if (unlikely(err))
2247                         goto failed;
2248                 nilfs_segment_usage_clear_active(ent->raw_su);
2249                 BUG_ON(!buffer_dirty(ent->bh_su));
2250         }
2251
2252         list_for_each_entry(ent, &sci->sc_active_segments, list) {
2253                 err = nilfs_open_segment_entry(ent, sufile);
2254                 if (unlikely(err))
2255                         goto failed;
2256                 nilfs_segment_usage_clear_active(ent->raw_su);
2257                 WARN_ON(!buffer_dirty(ent->bh_su));
2258         }
2259         return 0;
2260
2261  failed:
2262         nilfs_segctor_reactivate_segments(sci, nilfs);
2263         return err;
2264 }
2265
2266 static void nilfs_segctor_bead_completed_segments(struct nilfs_sc_info *sci)
2267 {
2268         struct nilfs_segment_buffer *segbuf, *last;
2269         struct nilfs_segment_entry *ent;
2270
2271         /* move each segbuf->sb_segent to the list of used active segments */
2272         last = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2273         nilfs_for_each_segbuf_before(segbuf, last, &sci->sc_segbufs) {
2274                 ent = segbuf->sb_segent;
2275                 if (!ent)
2276                         break; /* ignore unmapped segments (should check it?)*/
2277                 list_add_tail(&ent->list, &sci->sc_active_segments);
2278                 segbuf->sb_segent = NULL;
2279         }
2280 }
2281
2282 static void nilfs_segctor_commit_deactivate_segments(struct nilfs_sc_info *sci,
2283                                                      struct the_nilfs *nilfs)
2284 {
2285         struct nilfs_segment_entry *ent, *n;
2286
2287         list_for_each_entry_safe(ent, n, &sci->sc_active_segments, list) {
2288                 list_del(&ent->list);
2289                 nilfs_close_segment_entry(ent, nilfs->ns_sufile);
2290                 nilfs_free_segment_entry(ent);
2291         }
2292 }
2293
2294 /*
2295  * Main procedure of segment constructor
2296  */
2297 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2298 {
2299         struct nilfs_sb_info *sbi = sci->sc_sbi;
2300         struct the_nilfs *nilfs = sbi->s_nilfs;
2301         struct page *failed_page;
2302         int err, has_sr = 0;
2303
2304         sci->sc_stage.scnt = NILFS_ST_INIT;
2305
2306         err = nilfs_segctor_check_in_files(sci, sbi);
2307         if (unlikely(err))
2308                 goto out;
2309
2310         if (nilfs_test_metadata_dirty(sbi))
2311                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2312
2313         if (nilfs_segctor_clean(sci))
2314                 goto out;
2315
2316         do {
2317                 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2318
2319                 err = nilfs_segctor_begin_construction(sci, nilfs);
2320                 if (unlikely(err))
2321                         goto out;
2322
2323                 /* Update time stamp */
2324                 sci->sc_seg_ctime = get_seconds();
2325
2326                 err = nilfs_segctor_collect(sci, nilfs, mode);
2327                 if (unlikely(err))
2328                         goto failed;
2329
2330                 has_sr = (sci->sc_super_root != NULL);
2331
2332                 /* Avoid empty segment */
2333                 if (sci->sc_stage.scnt == NILFS_ST_DONE &&
2334                     NILFS_SEG_EMPTY(&sci->sc_curseg->sb_sum)) {
2335                         nilfs_segctor_end_construction(sci, nilfs, 1);
2336                         goto out;
2337                 }
2338
2339                 err = nilfs_segctor_assign(sci, mode);
2340                 if (unlikely(err))
2341                         goto failed;
2342
2343                 if (has_sr) {
2344                         err = nilfs_segctor_deactivate_segments(sci, nilfs);
2345                         if (unlikely(err))
2346                                 goto failed;
2347                 }
2348                 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2349                         nilfs_segctor_fill_in_file_bmap(sci, sbi->s_ifile);
2350
2351                 if (has_sr) {
2352                         err = nilfs_segctor_fill_in_checkpoint(sci);
2353                         if (unlikely(err))
2354                                 goto failed_to_make_up;
2355
2356                         nilfs_segctor_fill_in_super_root(sci, nilfs);
2357                 }
2358                 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2359
2360                 /* Write partial segments */
2361                 err = nilfs_segctor_prepare_write(sci, &failed_page);
2362                 if (unlikely(err))
2363                         goto failed_to_write;
2364
2365                 nilfs_segctor_fill_in_checksums(sci, nilfs->ns_crc_seed);
2366
2367                 err = nilfs_segctor_write(sci, nilfs->ns_bdi);
2368                 if (unlikely(err))
2369                         goto failed_to_write;
2370
2371                 nilfs_segctor_complete_write(sci);
2372
2373                 /* Commit segments */
2374                 nilfs_segctor_bead_completed_segments(sci);
2375                 if (has_sr) {
2376                         down_write(&nilfs->ns_sem);
2377                         nilfs_update_last_segment(sbi, 1);
2378                         up_write(&nilfs->ns_sem);
2379                         nilfs_segctor_commit_deactivate_segments(sci, nilfs);
2380                         nilfs_segctor_commit_free_segments(sci);
2381                         nilfs_segctor_clear_metadata_dirty(sci);
2382                 }
2383
2384                 nilfs_segctor_end_construction(sci, nilfs, 0);
2385
2386         } while (sci->sc_stage.scnt != NILFS_ST_DONE);
2387
2388         /* Clearing sketch data */
2389         if (has_sr && sci->sc_sketch_inode) {
2390                 if (i_size_read(sci->sc_sketch_inode) == 0)
2391                         clear_bit(NILFS_I_DIRTY,
2392                                   &NILFS_I(sci->sc_sketch_inode)->i_state);
2393                 i_size_write(sci->sc_sketch_inode, 0);
2394         }
2395  out:
2396         nilfs_segctor_destroy_segment_buffers(sci);
2397         nilfs_segctor_check_out_files(sci, sbi);
2398         return err;
2399
2400  failed_to_write:
2401         nilfs_segctor_abort_write(sci, failed_page, err);
2402         nilfs_segctor_cancel_segusage(sci, nilfs->ns_sufile);
2403
2404  failed_to_make_up:
2405         if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2406                 nilfs_redirty_inodes(&sci->sc_dirty_files);
2407         if (has_sr)
2408                 nilfs_segctor_reactivate_segments(sci, nilfs);
2409
2410  failed:
2411         if (nilfs_doing_gc())
2412                 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2413         nilfs_segctor_end_construction(sci, nilfs, err);
2414         goto out;
2415 }
2416
2417 /**
2418  * nilfs_secgtor_start_timer - set timer of background write
2419  * @sci: nilfs_sc_info
2420  *
2421  * If the timer has already been set, it ignores the new request.
2422  * This function MUST be called within a section locking the segment
2423  * semaphore.
2424  */
2425 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2426 {
2427         spin_lock(&sci->sc_state_lock);
2428         if (sci->sc_timer && !(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2429                 sci->sc_timer->expires = jiffies + sci->sc_interval;
2430                 add_timer(sci->sc_timer);
2431                 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2432         }
2433         spin_unlock(&sci->sc_state_lock);
2434 }
2435
2436 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2437 {
2438         spin_lock(&sci->sc_state_lock);
2439         if (!(sci->sc_flush_request & (1 << bn))) {
2440                 unsigned long prev_req = sci->sc_flush_request;
2441
2442                 sci->sc_flush_request |= (1 << bn);
2443                 if (!prev_req)
2444                         wake_up(&sci->sc_wait_daemon);
2445         }
2446         spin_unlock(&sci->sc_state_lock);
2447 }
2448
2449 /**
2450  * nilfs_flush_segment - trigger a segment construction for resource control
2451  * @sb: super block
2452  * @ino: inode number of the file to be flushed out.
2453  */
2454 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2455 {
2456         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2457         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2458
2459         if (!sci || nilfs_doing_construction())
2460                 return;
2461         nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2462                                         /* assign bit 0 to data files */
2463 }
2464
2465 int nilfs_segctor_add_segments_to_be_freed(struct nilfs_sc_info *sci,
2466                                            __u64 *segnum, size_t nsegs)
2467 {
2468         struct nilfs_segment_entry *ent;
2469         struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
2470         struct inode *sufile = nilfs->ns_sufile;
2471         LIST_HEAD(list);
2472         __u64 *pnum;
2473         size_t i;
2474         int err;
2475
2476         for (pnum = segnum, i = 0; i < nsegs; pnum++, i++) {
2477                 ent = nilfs_alloc_segment_entry(*pnum);
2478                 if (unlikely(!ent)) {
2479                         err = -ENOMEM;
2480                         goto failed;
2481                 }
2482                 list_add_tail(&ent->list, &list);
2483
2484                 err = nilfs_open_segment_entry(ent, sufile);
2485                 if (unlikely(err))
2486                         goto failed;
2487
2488                 if (unlikely(!nilfs_segment_usage_dirty(ent->raw_su)))
2489                         printk(KERN_WARNING "NILFS: unused segment is "
2490                                "requested to be cleaned (segnum=%llu)\n",
2491                                (unsigned long long)ent->segnum);
2492                 nilfs_close_segment_entry(ent, sufile);
2493         }
2494         list_splice(&list, sci->sc_cleaning_segments.prev);
2495         return 0;
2496
2497  failed:
2498         nilfs_dispose_segment_list(&list);
2499         return err;
2500 }
2501
2502 void nilfs_segctor_clear_segments_to_be_freed(struct nilfs_sc_info *sci)
2503 {
2504         nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
2505 }
2506
2507 struct nilfs_segctor_wait_request {
2508         wait_queue_t    wq;
2509         __u32           seq;
2510         int             err;
2511         atomic_t        done;
2512 };
2513
2514 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2515 {
2516         struct nilfs_segctor_wait_request wait_req;
2517         int err = 0;
2518
2519         spin_lock(&sci->sc_state_lock);
2520         init_wait(&wait_req.wq);
2521         wait_req.err = 0;
2522         atomic_set(&wait_req.done, 0);
2523         wait_req.seq = ++sci->sc_seq_request;
2524         spin_unlock(&sci->sc_state_lock);
2525
2526         init_waitqueue_entry(&wait_req.wq, current);
2527         add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2528         set_current_state(TASK_INTERRUPTIBLE);
2529         wake_up(&sci->sc_wait_daemon);
2530
2531         for (;;) {
2532                 if (atomic_read(&wait_req.done)) {
2533                         err = wait_req.err;
2534                         break;
2535                 }
2536                 if (!signal_pending(current)) {
2537                         schedule();
2538                         continue;
2539                 }
2540                 err = -ERESTARTSYS;
2541                 break;
2542         }
2543         finish_wait(&sci->sc_wait_request, &wait_req.wq);
2544         return err;
2545 }
2546
2547 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2548 {
2549         struct nilfs_segctor_wait_request *wrq, *n;
2550         unsigned long flags;
2551
2552         spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2553         list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2554                                  wq.task_list) {
2555                 if (!atomic_read(&wrq->done) &&
2556                     nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2557                         wrq->err = err;
2558                         atomic_set(&wrq->done, 1);
2559                 }
2560                 if (atomic_read(&wrq->done)) {
2561                         wrq->wq.func(&wrq->wq,
2562                                      TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2563                                      0, NULL);
2564                 }
2565         }
2566         spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2567 }
2568
2569 /**
2570  * nilfs_construct_segment - construct a logical segment
2571  * @sb: super block
2572  *
2573  * Return Value: On success, 0 is retured. On errors, one of the following
2574  * negative error code is returned.
2575  *
2576  * %-EROFS - Read only filesystem.
2577  *
2578  * %-EIO - I/O error
2579  *
2580  * %-ENOSPC - No space left on device (only in a panic state).
2581  *
2582  * %-ERESTARTSYS - Interrupted.
2583  *
2584  * %-ENOMEM - Insufficient memory available.
2585  */
2586 int nilfs_construct_segment(struct super_block *sb)
2587 {
2588         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2589         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2590         struct nilfs_transaction_info *ti;
2591         int err;
2592
2593         if (!sci)
2594                 return -EROFS;
2595
2596         /* A call inside transactions causes a deadlock. */
2597         BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2598
2599         err = nilfs_segctor_sync(sci);
2600         return err;
2601 }
2602
2603 /**
2604  * nilfs_construct_dsync_segment - construct a data-only logical segment
2605  * @sb: super block
2606  * @inode: inode whose data blocks should be written out
2607  * @start: start byte offset
2608  * @end: end byte offset (inclusive)
2609  *
2610  * Return Value: On success, 0 is retured. On errors, one of the following
2611  * negative error code is returned.
2612  *
2613  * %-EROFS - Read only filesystem.
2614  *
2615  * %-EIO - I/O error
2616  *
2617  * %-ENOSPC - No space left on device (only in a panic state).
2618  *
2619  * %-ERESTARTSYS - Interrupted.
2620  *
2621  * %-ENOMEM - Insufficient memory available.
2622  */
2623 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2624                                   loff_t start, loff_t end)
2625 {
2626         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2627         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2628         struct nilfs_inode_info *ii;
2629         struct nilfs_transaction_info ti;
2630         int err = 0;
2631
2632         if (!sci)
2633                 return -EROFS;
2634
2635         nilfs_transaction_lock(sbi, &ti, 0);
2636
2637         ii = NILFS_I(inode);
2638         if (test_bit(NILFS_I_INODE_DIRTY, &ii->i_state) ||
2639             nilfs_test_opt(sbi, STRICT_ORDER) ||
2640             test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2641             nilfs_discontinued(sbi->s_nilfs)) {
2642                 nilfs_transaction_unlock(sbi);
2643                 err = nilfs_segctor_sync(sci);
2644                 return err;
2645         }
2646
2647         spin_lock(&sbi->s_inode_lock);
2648         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2649             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2650                 spin_unlock(&sbi->s_inode_lock);
2651                 nilfs_transaction_unlock(sbi);
2652                 return 0;
2653         }
2654         spin_unlock(&sbi->s_inode_lock);
2655         sci->sc_dsync_inode = ii;
2656         sci->sc_dsync_start = start;
2657         sci->sc_dsync_end = end;
2658
2659         err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2660
2661         nilfs_transaction_unlock(sbi);
2662         return err;
2663 }
2664
2665 struct nilfs_segctor_req {
2666         int mode;
2667         __u32 seq_accepted;
2668         int sc_err;  /* construction failure */
2669         int sb_err;  /* super block writeback failure */
2670 };
2671
2672 #define FLUSH_FILE_BIT  (0x1) /* data file only */
2673 #define FLUSH_DAT_BIT   (1 << NILFS_DAT_INO) /* DAT only */
2674
2675 static void nilfs_segctor_accept(struct nilfs_sc_info *sci,
2676                                  struct nilfs_segctor_req *req)
2677 {
2678         req->sc_err = req->sb_err = 0;
2679         spin_lock(&sci->sc_state_lock);
2680         req->seq_accepted = sci->sc_seq_request;
2681         spin_unlock(&sci->sc_state_lock);
2682
2683         if (sci->sc_timer)
2684                 del_timer_sync(sci->sc_timer);
2685 }
2686
2687 static void nilfs_segctor_notify(struct nilfs_sc_info *sci,
2688                                  struct nilfs_segctor_req *req)
2689 {
2690         /* Clear requests (even when the construction failed) */
2691         spin_lock(&sci->sc_state_lock);
2692
2693         sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2694
2695         if (req->mode == SC_LSEG_SR) {
2696                 sci->sc_seq_done = req->seq_accepted;
2697                 nilfs_segctor_wakeup(sci, req->sc_err ? : req->sb_err);
2698                 sci->sc_flush_request = 0;
2699         } else if (req->mode == SC_FLUSH_FILE)
2700                 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2701         else if (req->mode == SC_FLUSH_DAT)
2702                 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2703
2704         spin_unlock(&sci->sc_state_lock);
2705 }
2706
2707 static int nilfs_segctor_construct(struct nilfs_sc_info *sci,
2708                                    struct nilfs_segctor_req *req)
2709 {
2710         struct nilfs_sb_info *sbi = sci->sc_sbi;
2711         struct the_nilfs *nilfs = sbi->s_nilfs;
2712         int err = 0;
2713
2714         if (nilfs_discontinued(nilfs))
2715                 req->mode = SC_LSEG_SR;
2716         if (!nilfs_segctor_confirm(sci)) {
2717                 err = nilfs_segctor_do_construct(sci, req->mode);
2718                 req->sc_err = err;
2719         }
2720         if (likely(!err)) {
2721                 if (req->mode != SC_FLUSH_DAT)
2722                         atomic_set(&nilfs->ns_ndirtyblks, 0);
2723                 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2724                     nilfs_discontinued(nilfs)) {
2725                         down_write(&nilfs->ns_sem);
2726                         req->sb_err = nilfs_commit_super(sbi);
2727                         up_write(&nilfs->ns_sem);
2728                 }
2729         }
2730         return err;
2731 }
2732
2733 static void nilfs_construction_timeout(unsigned long data)
2734 {
2735         struct task_struct *p = (struct task_struct *)data;
2736         wake_up_process(p);
2737 }
2738
2739 static void
2740 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2741 {
2742         struct nilfs_inode_info *ii, *n;
2743
2744         list_for_each_entry_safe(ii, n, head, i_dirty) {
2745                 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2746                         continue;
2747                 hlist_del_init(&ii->vfs_inode.i_hash);
2748                 list_del_init(&ii->i_dirty);
2749                 nilfs_clear_gcinode(&ii->vfs_inode);
2750         }
2751 }
2752
2753 int nilfs_clean_segments(struct super_block *sb, void __user *argp)
2754 {
2755         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2756         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2757         struct the_nilfs *nilfs = sbi->s_nilfs;
2758         struct nilfs_transaction_info ti;
2759         struct nilfs_segctor_req req = { .mode = SC_LSEG_SR };
2760         int err;
2761
2762         if (unlikely(!sci))
2763                 return -EROFS;
2764
2765         nilfs_transaction_lock(sbi, &ti, 1);
2766
2767         err = nilfs_init_gcdat_inode(nilfs);
2768         if (unlikely(err))
2769                 goto out_unlock;
2770         err = nilfs_ioctl_prepare_clean_segments(nilfs, argp);
2771         if (unlikely(err))
2772                 goto out_unlock;
2773
2774         list_splice_init(&nilfs->ns_gc_inodes, sci->sc_gc_inodes.prev);
2775
2776         for (;;) {
2777                 nilfs_segctor_accept(sci, &req);
2778                 err = nilfs_segctor_construct(sci, &req);
2779                 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2780                 nilfs_segctor_notify(sci, &req);
2781
2782                 if (likely(!err))
2783                         break;
2784
2785                 nilfs_warning(sb, __func__,
2786                               "segment construction failed. (err=%d)", err);
2787                 set_current_state(TASK_INTERRUPTIBLE);
2788                 schedule_timeout(sci->sc_interval);
2789         }
2790
2791  out_unlock:
2792         nilfs_clear_gcdat_inode(nilfs);
2793         nilfs_transaction_unlock(sbi);
2794         return err;
2795 }
2796
2797 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2798 {
2799         struct nilfs_sb_info *sbi = sci->sc_sbi;
2800         struct nilfs_transaction_info ti;
2801         struct nilfs_segctor_req req = { .mode = mode };
2802
2803         nilfs_transaction_lock(sbi, &ti, 0);
2804
2805         nilfs_segctor_accept(sci, &req);
2806         nilfs_segctor_construct(sci, &req);
2807         nilfs_segctor_notify(sci, &req);
2808
2809         /*
2810          * Unclosed segment should be retried.  We do this using sc_timer.
2811          * Timeout of sc_timer will invoke complete construction which leads
2812          * to close the current logical segment.
2813          */
2814         if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2815                 nilfs_segctor_start_timer(sci);
2816
2817         nilfs_transaction_unlock(sbi);
2818 }
2819
2820 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2821 {
2822         int mode = 0;
2823         int err;
2824
2825         spin_lock(&sci->sc_state_lock);
2826         mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2827                 SC_FLUSH_DAT : SC_FLUSH_FILE;
2828         spin_unlock(&sci->sc_state_lock);
2829
2830         if (mode) {
2831                 err = nilfs_segctor_do_construct(sci, mode);
2832
2833                 spin_lock(&sci->sc_state_lock);
2834                 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2835                         ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2836                 spin_unlock(&sci->sc_state_lock);
2837         }
2838         clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2839 }
2840
2841 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2842 {
2843         if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2844             time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2845                 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2846                         return SC_FLUSH_FILE;
2847                 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2848                         return SC_FLUSH_DAT;
2849         }
2850         return SC_LSEG_SR;
2851 }
2852
2853 /**
2854  * nilfs_segctor_thread - main loop of the segment constructor thread.
2855  * @arg: pointer to a struct nilfs_sc_info.
2856  *
2857  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2858  * to execute segment constructions.
2859  */
2860 static int nilfs_segctor_thread(void *arg)
2861 {
2862         struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2863         struct timer_list timer;
2864         int timeout = 0;
2865
2866         init_timer(&timer);
2867         timer.data = (unsigned long)current;
2868         timer.function = nilfs_construction_timeout;
2869         sci->sc_timer = &timer;
2870
2871         /* start sync. */
2872         sci->sc_task = current;
2873         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2874         printk(KERN_INFO
2875                "segctord starting. Construction interval = %lu seconds, "
2876                "CP frequency < %lu seconds\n",
2877                sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2878
2879         spin_lock(&sci->sc_state_lock);
2880  loop:
2881         for (;;) {
2882                 int mode;
2883
2884                 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2885                         goto end_thread;
2886
2887                 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2888                         mode = SC_LSEG_SR;
2889                 else if (!sci->sc_flush_request)
2890                         break;
2891                 else
2892                         mode = nilfs_segctor_flush_mode(sci);
2893
2894                 spin_unlock(&sci->sc_state_lock);
2895                 nilfs_segctor_thread_construct(sci, mode);
2896                 spin_lock(&sci->sc_state_lock);
2897                 timeout = 0;
2898         }
2899
2900
2901         if (freezing(current)) {
2902                 spin_unlock(&sci->sc_state_lock);
2903                 refrigerator();
2904                 spin_lock(&sci->sc_state_lock);
2905         } else {
2906                 DEFINE_WAIT(wait);
2907                 int should_sleep = 1;
2908
2909                 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2910                                 TASK_INTERRUPTIBLE);
2911
2912                 if (sci->sc_seq_request != sci->sc_seq_done)
2913                         should_sleep = 0;
2914                 else if (sci->sc_flush_request)
2915                         should_sleep = 0;
2916                 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2917                         should_sleep = time_before(jiffies,
2918                                                    sci->sc_timer->expires);
2919
2920                 if (should_sleep) {
2921                         spin_unlock(&sci->sc_state_lock);
2922                         schedule();
2923                         spin_lock(&sci->sc_state_lock);
2924                 }
2925                 finish_wait(&sci->sc_wait_daemon, &wait);
2926                 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2927                            time_after_eq(jiffies, sci->sc_timer->expires));
2928         }
2929         goto loop;
2930
2931  end_thread:
2932         spin_unlock(&sci->sc_state_lock);
2933         del_timer_sync(sci->sc_timer);
2934         sci->sc_timer = NULL;
2935
2936         /* end sync. */
2937         sci->sc_task = NULL;
2938         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2939         return 0;
2940 }
2941
2942 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2943 {
2944         struct task_struct *t;
2945
2946         t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2947         if (IS_ERR(t)) {
2948                 int err = PTR_ERR(t);
2949
2950                 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2951                        err);
2952                 return err;
2953         }
2954         wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2955         return 0;
2956 }
2957
2958 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2959 {
2960         sci->sc_state |= NILFS_SEGCTOR_QUIT;
2961
2962         while (sci->sc_task) {
2963                 wake_up(&sci->sc_wait_daemon);
2964                 spin_unlock(&sci->sc_state_lock);
2965                 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2966                 spin_lock(&sci->sc_state_lock);
2967         }
2968 }
2969
2970 static int nilfs_segctor_init(struct nilfs_sc_info *sci,
2971                               struct nilfs_recovery_info *ri)
2972 {
2973         int err;
2974         struct inode *inode = nilfs_iget(sci->sc_super, NILFS_SKETCH_INO);
2975
2976         sci->sc_sketch_inode = IS_ERR(inode) ? NULL : inode;
2977         if (sci->sc_sketch_inode)
2978                 i_size_write(sci->sc_sketch_inode, 0);
2979
2980         sci->sc_seq_done = sci->sc_seq_request;
2981         if (ri)
2982                 list_splice_init(&ri->ri_used_segments,
2983                                  sci->sc_active_segments.prev);
2984
2985         err = nilfs_segctor_start_thread(sci);
2986         if (err) {
2987                 if (ri)
2988                         list_splice_init(&sci->sc_active_segments,
2989                                          ri->ri_used_segments.prev);
2990                 if (sci->sc_sketch_inode) {
2991                         iput(sci->sc_sketch_inode);
2992                         sci->sc_sketch_inode = NULL;
2993                 }
2994         }
2995         return err;
2996 }
2997
2998 /*
2999  * Setup & clean-up functions
3000  */
3001 static struct nilfs_sc_info *nilfs_segctor_new(struct nilfs_sb_info *sbi)
3002 {
3003         struct nilfs_sc_info *sci;
3004
3005         sci = kzalloc(sizeof(*sci), GFP_KERNEL);
3006         if (!sci)
3007                 return NULL;
3008
3009         sci->sc_sbi = sbi;
3010         sci->sc_super = sbi->s_super;
3011
3012         init_waitqueue_head(&sci->sc_wait_request);
3013         init_waitqueue_head(&sci->sc_wait_daemon);
3014         init_waitqueue_head(&sci->sc_wait_task);
3015         spin_lock_init(&sci->sc_state_lock);
3016         INIT_LIST_HEAD(&sci->sc_dirty_files);
3017         INIT_LIST_HEAD(&sci->sc_segbufs);
3018         INIT_LIST_HEAD(&sci->sc_gc_inodes);
3019         INIT_LIST_HEAD(&sci->sc_active_segments);
3020         INIT_LIST_HEAD(&sci->sc_cleaning_segments);
3021         INIT_LIST_HEAD(&sci->sc_copied_buffers);
3022
3023         sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
3024         sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
3025         sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
3026
3027         if (sbi->s_interval)
3028                 sci->sc_interval = sbi->s_interval;
3029         if (sbi->s_watermark)
3030                 sci->sc_watermark = sbi->s_watermark;
3031         return sci;
3032 }
3033
3034 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
3035 {
3036         int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
3037
3038         /* The segctord thread was stopped and its timer was removed.
3039            But some tasks remain. */
3040         do {
3041                 struct nilfs_sb_info *sbi = sci->sc_sbi;
3042                 struct nilfs_transaction_info ti;
3043                 struct nilfs_segctor_req req = { .mode = SC_LSEG_SR };
3044
3045                 nilfs_transaction_lock(sbi, &ti, 0);
3046                 nilfs_segctor_accept(sci, &req);
3047                 ret = nilfs_segctor_construct(sci, &req);
3048                 nilfs_segctor_notify(sci, &req);
3049                 nilfs_transaction_unlock(sbi);
3050
3051         } while (ret && retrycount-- > 0);
3052 }
3053
3054 /**
3055  * nilfs_segctor_destroy - destroy the segment constructor.
3056  * @sci: nilfs_sc_info
3057  *
3058  * nilfs_segctor_destroy() kills the segctord thread and frees
3059  * the nilfs_sc_info struct.
3060  * Caller must hold the segment semaphore.
3061  */
3062 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
3063 {
3064         struct nilfs_sb_info *sbi = sci->sc_sbi;
3065         int flag;
3066
3067         up_write(&sbi->s_nilfs->ns_segctor_sem);
3068
3069         spin_lock(&sci->sc_state_lock);
3070         nilfs_segctor_kill_thread(sci);
3071         flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
3072                 || sci->sc_seq_request != sci->sc_seq_done);
3073         spin_unlock(&sci->sc_state_lock);
3074
3075         if (flag || nilfs_segctor_confirm(sci))
3076                 nilfs_segctor_write_out(sci);
3077
3078         WARN_ON(!list_empty(&sci->sc_copied_buffers));
3079
3080         if (!list_empty(&sci->sc_dirty_files)) {
3081                 nilfs_warning(sbi->s_super, __func__,
3082                               "dirty file(s) after the final construction\n");
3083                 nilfs_dispose_list(sbi, &sci->sc_dirty_files, 1);
3084         }
3085         if (!list_empty(&sci->sc_active_segments))
3086                 nilfs_dispose_segment_list(&sci->sc_active_segments);
3087
3088         if (!list_empty(&sci->sc_cleaning_segments))
3089                 nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
3090
3091         WARN_ON(!list_empty(&sci->sc_segbufs));
3092
3093         if (sci->sc_sketch_inode) {
3094                 iput(sci->sc_sketch_inode);
3095                 sci->sc_sketch_inode = NULL;
3096         }
3097         down_write(&sbi->s_nilfs->ns_segctor_sem);
3098
3099         kfree(sci);
3100 }
3101
3102 /**
3103  * nilfs_attach_segment_constructor - attach a segment constructor
3104  * @sbi: nilfs_sb_info
3105  * @ri: nilfs_recovery_info
3106  *
3107  * nilfs_attach_segment_constructor() allocates a struct nilfs_sc_info,
3108  * initilizes it, and starts the segment constructor.
3109  *
3110  * Return Value: On success, 0 is returned. On error, one of the following
3111  * negative error code is returned.
3112  *
3113  * %-ENOMEM - Insufficient memory available.
3114  */
3115 int nilfs_attach_segment_constructor(struct nilfs_sb_info *sbi,
3116                                      struct nilfs_recovery_info *ri)
3117 {
3118         struct the_nilfs *nilfs = sbi->s_nilfs;
3119         int err;
3120
3121         /* Each field of nilfs_segctor is cleared through the initialization
3122            of super-block info */
3123         sbi->s_sc_info = nilfs_segctor_new(sbi);
3124         if (!sbi->s_sc_info)
3125                 return -ENOMEM;
3126
3127         nilfs_attach_writer(nilfs, sbi);
3128         err = nilfs_segctor_init(NILFS_SC(sbi), ri);
3129         if (err) {
3130                 nilfs_detach_writer(nilfs, sbi);
3131                 kfree(sbi->s_sc_info);
3132                 sbi->s_sc_info = NULL;
3133         }
3134         return err;
3135 }
3136
3137 /**
3138  * nilfs_detach_segment_constructor - destroy the segment constructor
3139  * @sbi: nilfs_sb_info
3140  *
3141  * nilfs_detach_segment_constructor() kills the segment constructor daemon,
3142  * frees the struct nilfs_sc_info, and destroy the dirty file list.
3143  */
3144 void nilfs_detach_segment_constructor(struct nilfs_sb_info *sbi)
3145 {
3146         struct the_nilfs *nilfs = sbi->s_nilfs;
3147         LIST_HEAD(garbage_list);
3148
3149         down_write(&nilfs->ns_segctor_sem);
3150         if (NILFS_SC(sbi)) {
3151                 nilfs_segctor_destroy(NILFS_SC(sbi));
3152                 sbi->s_sc_info = NULL;
3153         }
3154
3155         /* Force to free the list of dirty files */
3156         spin_lock(&sbi->s_inode_lock);
3157         if (!list_empty(&sbi->s_dirty_files)) {
3158                 list_splice_init(&sbi->s_dirty_files, &garbage_list);
3159                 nilfs_warning(sbi->s_super, __func__,
3160                               "Non empty dirty list after the last "
3161                               "segment construction\n");
3162         }
3163         spin_unlock(&sbi->s_inode_lock);
3164         up_write(&nilfs->ns_segctor_sem);
3165
3166         nilfs_dispose_list(sbi, &garbage_list, 1);
3167         nilfs_detach_writer(nilfs, sbi);
3168 }