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         BUG_ON(cur_ti);
 338         BUG_ON(!ti);
 339         ti->ti_flags = NILFS_TI_WRITER;
 340         ti->ti_count = 0;
 341         ti->ti_save = cur_ti;
 342         ti->ti_magic = NILFS_TI_MAGIC;
 343         INIT_LIST_HEAD(&ti->ti_garbage);
 344         current->journal_info = ti;
 345
 346         for (;;) {
 347                 down_write(&sbi->s_nilfs->ns_segctor_sem);
 348                 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &NILFS_SC(sbi)->sc_flags))
 349                         break;
 350
 351                 nilfs_segctor_do_immediate_flush(NILFS_SC(sbi));
 352
 353                 up_write(&sbi->s_nilfs->ns_segctor_sem);
 354                 yield();
 355         }
 356         if (gcflag)
 357                 ti->ti_flags |= NILFS_TI_GC;
 358 }
 359
 360 static void nilfs_transaction_unlock(struct nilfs_sb_info *sbi)
 361 {
 362         struct nilfs_transaction_info *ti = current->journal_info;
 363
 364         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 365         BUG_ON(ti->ti_count > 0);
 366
 367         up_write(&sbi->s_nilfs->ns_segctor_sem);
 368         current->journal_info = ti->ti_save;
 369         if (!list_empty(&ti->ti_garbage))
 370                 nilfs_dispose_list(sbi, &ti->ti_garbage, 0);
 371 }
 372
 373 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
 374                                             struct nilfs_segsum_pointer *ssp,
 375                                             unsigned bytes)
 376 {
 377         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 378         unsigned blocksize = sci->sc_super->s_blocksize;
 379         void *p;
 380
 381         if (unlikely(ssp->offset + bytes > blocksize)) {
 382                 ssp->offset = 0;
 383                 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
 384                                                &segbuf->sb_segsum_buffers));
 385                 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
 386         }
 387         p = ssp->bh->b_data + ssp->offset;
 388         ssp->offset += bytes;
 389         return p;
 390 }
 391
 392 /**
 393  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
 394  * @sci: nilfs_sc_info
 395  */
 396 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
 397 {
 398         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 399         struct buffer_head *sumbh;
 400         unsigned sumbytes;
 401         unsigned flags = 0;
 402         int err;
 403
 404         if (nilfs_doing_gc())
 405                 flags = NILFS_SS_GC;
 406         err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime);
 407         if (unlikely(err))
 408                 return err;
 409
 410         sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
 411         sumbytes = segbuf->sb_sum.sumbytes;
 412         sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
 413         sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
 414         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 415         return 0;
 416 }
 417
 418 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
 419 {
 420         sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
 421         if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
 422                 return -E2BIG; /* The current segment is filled up
 423                                   (internal code) */
 424         sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
 425         return nilfs_segctor_reset_segment_buffer(sci);
 426 }
 427
 428 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
 429 {
 430         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 431         int err;
 432
 433         if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
 434                 err = nilfs_segctor_feed_segment(sci);
 435                 if (err)
 436                         return err;
 437                 segbuf = sci->sc_curseg;
 438         }
 439         err = nilfs_segbuf_extend_payload(segbuf, &sci->sc_super_root);
 440         if (likely(!err))
 441                 segbuf->sb_sum.flags |= NILFS_SS_SR;
 442         return err;
 443 }
 444
 445 /*
 446  * Functions for making segment summary and payloads
 447  */
 448 static int nilfs_segctor_segsum_block_required(
 449         struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
 450         unsigned binfo_size)
 451 {
 452         unsigned blocksize = sci->sc_super->s_blocksize;
 453         /* Size of finfo and binfo is enough small against blocksize */
 454
 455         return ssp->offset + binfo_size +
 456                 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
 457                 blocksize;
 458 }
 459
 460 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
 461                                       struct inode *inode)
 462 {
 463         sci->sc_curseg->sb_sum.nfinfo++;
 464         sci->sc_binfo_ptr = sci->sc_finfo_ptr;
 465         nilfs_segctor_map_segsum_entry(
 466                 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
 467         /* skip finfo */
 468 }
 469
 470 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
 471                                     struct inode *inode)
 472 {
 473         struct nilfs_finfo *finfo;
 474         struct nilfs_inode_info *ii;
 475         struct nilfs_segment_buffer *segbuf;
 476
 477         if (sci->sc_blk_cnt == 0)
 478                 return;
 479
 480         ii = NILFS_I(inode);
 481         finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
 482                                                  sizeof(*finfo));
 483         finfo->fi_ino = cpu_to_le64(inode->i_ino);
 484         finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
 485         finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
 486         finfo->fi_cno = cpu_to_le64(ii->i_cno);
 487
 488         segbuf = sci->sc_curseg;
 489         segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
 490                 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
 491         sci->sc_finfo_ptr = sci->sc_binfo_ptr;
 492         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 493 }
 494
 495 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
 496                                         struct buffer_head *bh,
 497                                         struct inode *inode,
 498                                         unsigned binfo_size)
 499 {
 500         struct nilfs_segment_buffer *segbuf;
 501         int required, err = 0;
 502
 503  retry:
 504         segbuf = sci->sc_curseg;
 505         required = nilfs_segctor_segsum_block_required(
 506                 sci, &sci->sc_binfo_ptr, binfo_size);
 507         if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
 508                 nilfs_segctor_end_finfo(sci, inode);
 509                 err = nilfs_segctor_feed_segment(sci);
 510                 if (err)
 511                         return err;
 512                 goto retry;
 513         }
 514         if (unlikely(required)) {
 515                 err = nilfs_segbuf_extend_segsum(segbuf);
 516                 if (unlikely(err))
 517                         goto failed;
 518         }
 519         if (sci->sc_blk_cnt == 0)
 520                 nilfs_segctor_begin_finfo(sci, inode);
 521
 522         nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
 523         /* Substitution to vblocknr is delayed until update_blocknr() */
 524         nilfs_segbuf_add_file_buffer(segbuf, bh);
 525         sci->sc_blk_cnt++;
 526  failed:
 527         return err;
 528 }
 529
 530 static int nilfs_handle_bmap_error(int err, const char *fname,
 531                                    struct inode *inode, struct super_block *sb)
 532 {
 533         if (err == -EINVAL) {
 534                 nilfs_error(sb, fname, "broken bmap (inode=%lu)\n",
 535                             inode->i_ino);
 536                 err = -EIO;
 537         }
 538         return err;
 539 }
 540
 541 /*
 542  * Callback functions that enumerate, mark, and collect dirty blocks
 543  */
 544 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
 545                                    struct buffer_head *bh, struct inode *inode)
 546 {
 547         int err;
 548
 549         /* BUG_ON(!buffer_dirty(bh)); */
 550         /* excluded by scan_dirty_data_buffers() */
 551         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 552         if (unlikely(err < 0))
 553                 return nilfs_handle_bmap_error(err, __func__, inode,
 554                                                sci->sc_super);
 555
 556         err = nilfs_segctor_add_file_block(sci, bh, inode,
 557                                            sizeof(struct nilfs_binfo_v));
 558         if (!err)
 559                 sci->sc_datablk_cnt++;
 560         return err;
 561 }
 562
 563 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
 564                                    struct buffer_head *bh,
 565                                    struct inode *inode)
 566 {
 567         int err;
 568
 569         /* BUG_ON(!buffer_dirty(bh)); */
 570         /* excluded by scan_dirty_node_buffers() */
 571         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 572         if (unlikely(err < 0))
 573                 return nilfs_handle_bmap_error(err, __func__, inode,
 574                                                sci->sc_super);
 575         return 0;
 576 }
 577
 578 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
 579                                    struct buffer_head *bh,
 580                                    struct inode *inode)
 581 {
 582         BUG_ON(!buffer_dirty(bh));
 583         return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 584 }
 585
 586 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
 587                                         struct nilfs_segsum_pointer *ssp,
 588                                         union nilfs_binfo *binfo)
 589 {
 590         struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
 591                 sci, ssp, sizeof(*binfo_v));
 592         *binfo_v = binfo->bi_v;
 593 }
 594
 595 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
 596                                         struct nilfs_segsum_pointer *ssp,
 597                                         union nilfs_binfo *binfo)
 598 {
 599         __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
 600                 sci, ssp, sizeof(*vblocknr));
 601         *vblocknr = binfo->bi_v.bi_vblocknr;
 602 }
 603
 604 struct nilfs_sc_operations nilfs_sc_file_ops = {
 605         .collect_data = nilfs_collect_file_data,
 606         .collect_node = nilfs_collect_file_node,
 607         .collect_bmap = nilfs_collect_file_bmap,
 608         .write_data_binfo = nilfs_write_file_data_binfo,
 609         .write_node_binfo = nilfs_write_file_node_binfo,
 610 };
 611
 612 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
 613                                   struct buffer_head *bh, struct inode *inode)
 614 {
 615         int err;
 616
 617         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 618         if (unlikely(err < 0))
 619                 return nilfs_handle_bmap_error(err, __func__, inode,
 620                                                sci->sc_super);
 621
 622         err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 623         if (!err)
 624                 sci->sc_datablk_cnt++;
 625         return err;
 626 }
 627
 628 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
 629                                   struct buffer_head *bh, struct inode *inode)
 630 {
 631         BUG_ON(!buffer_dirty(bh));
 632         return nilfs_segctor_add_file_block(sci, bh, inode,
 633                                             sizeof(struct nilfs_binfo_dat));
 634 }
 635
 636 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
 637                                        struct nilfs_segsum_pointer *ssp,
 638                                        union nilfs_binfo *binfo)
 639 {
 640         __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
 641                                                           sizeof(*blkoff));
 642         *blkoff = binfo->bi_dat.bi_blkoff;
 643 }
 644
 645 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
 646                                        struct nilfs_segsum_pointer *ssp,
 647                                        union nilfs_binfo *binfo)
 648 {
 649         struct nilfs_binfo_dat *binfo_dat =
 650                 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
 651         *binfo_dat = binfo->bi_dat;
 652 }
 653
 654 struct nilfs_sc_operations nilfs_sc_dat_ops = {
 655         .collect_data = nilfs_collect_dat_data,
 656         .collect_node = nilfs_collect_file_node,
 657         .collect_bmap = nilfs_collect_dat_bmap,
 658         .write_data_binfo = nilfs_write_dat_data_binfo,
 659         .write_node_binfo = nilfs_write_dat_node_binfo,
 660 };
 661
 662 struct nilfs_sc_operations nilfs_sc_dsync_ops = {
 663         .collect_data = nilfs_collect_file_data,
 664         .collect_node = NULL,
 665         .collect_bmap = NULL,
 666         .write_data_binfo = nilfs_write_file_data_binfo,
 667         .write_node_binfo = NULL,
 668 };
 669
 670 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
 671                                               struct list_head *listp,
 672                                               size_t nlimit,
 673                                               loff_t start, loff_t end)
 674 {
 675         struct address_space *mapping = inode->i_mapping;
 676         struct pagevec pvec;
 677         pgoff_t index = 0, last = ULONG_MAX;
 678         size_t ndirties = 0;
 679         int i;
 680
 681         if (unlikely(start != 0 || end != LLONG_MAX)) {
 682                 /*
 683                  * A valid range is given for sync-ing data pages. The
 684                  * range is rounded to per-page; extra dirty buffers
 685                  * may be included if blocksize < pagesize.
 686                  */
 687                 index = start >> PAGE_SHIFT;
 688                 last = end >> PAGE_SHIFT;
 689         }
 690         pagevec_init(&pvec, 0);
 691  repeat:
 692         if (unlikely(index > last) ||
 693             !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 694                                 min_t(pgoff_t, last - index,
 695                                       PAGEVEC_SIZE - 1) + 1))
 696                 return ndirties;
 697
 698         for (i = 0; i < pagevec_count(&pvec); i++) {
 699                 struct buffer_head *bh, *head;
 700                 struct page *page = pvec.pages[i];
 701
 702                 if (unlikely(page->index > last))
 703                         break;
 704
 705                 if (mapping->host) {
 706                         lock_page(page);
 707                         if (!page_has_buffers(page))
 708                                 create_empty_buffers(page,
 709                                                      1 << inode->i_blkbits, 0);
 710                         unlock_page(page);
 711                 }
 712
 713                 bh = head = page_buffers(page);
 714                 do {
 715                         if (!buffer_dirty(bh))
 716                                 continue;
 717                         get_bh(bh);
 718                         list_add_tail(&bh->b_assoc_buffers, listp);
 719                         ndirties++;
 720                         if (unlikely(ndirties >= nlimit)) {
 721                                 pagevec_release(&pvec);
 722                                 cond_resched();
 723                                 return ndirties;
 724                         }
 725                 } while (bh = bh->b_this_page, bh != head);
 726         }
 727         pagevec_release(&pvec);
 728         cond_resched();
 729         goto repeat;
 730 }
 731
 732 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
 733                                             struct list_head *listp)
 734 {
 735         struct nilfs_inode_info *ii = NILFS_I(inode);
 736         struct address_space *mapping = &ii->i_btnode_cache;
 737         struct pagevec pvec;
 738         struct buffer_head *bh, *head;
 739         unsigned int i;
 740         pgoff_t index = 0;
 741
 742         pagevec_init(&pvec, 0);
 743
 744         while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 745                                   PAGEVEC_SIZE)) {
 746                 for (i = 0; i < pagevec_count(&pvec); i++) {
 747                         bh = head = page_buffers(pvec.pages[i]);
 748                         do {
 749                                 if (buffer_dirty(bh)) {
 750                                         get_bh(bh);
 751                                         list_add_tail(&bh->b_assoc_buffers,
 752                                                       listp);
 753                                 }
 754                                 bh = bh->b_this_page;
 755                         } while (bh != head);
 756                 }
 757                 pagevec_release(&pvec);
 758                 cond_resched();
 759         }
 760 }
 761
 762 static void nilfs_dispose_list(struct nilfs_sb_info *sbi,
 763                                struct list_head *head, int force)
 764 {
 765         struct nilfs_inode_info *ii, *n;
 766         struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
 767         unsigned nv = 0;
 768
 769         while (!list_empty(head)) {
 770                 spin_lock(&sbi->s_inode_lock);
 771                 list_for_each_entry_safe(ii, n, head, i_dirty) {
 772                         list_del_init(&ii->i_dirty);
 773                         if (force) {
 774                                 if (unlikely(ii->i_bh)) {
 775                                         brelse(ii->i_bh);
 776                                         ii->i_bh = NULL;
 777                                 }
 778                         } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
 779                                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 780                                 list_add_tail(&ii->i_dirty,
 781                                               &sbi->s_dirty_files);
 782                                 continue;
 783                         }
 784                         ivec[nv++] = ii;
 785                         if (nv == SC_N_INODEVEC)
 786                                 break;
 787                 }
 788                 spin_unlock(&sbi->s_inode_lock);
 789
 790                 for (pii = ivec; nv > 0; pii++, nv--)
 791                         iput(&(*pii)->vfs_inode);
 792         }
 793 }
 794
 795 static int nilfs_test_metadata_dirty(struct nilfs_sb_info *sbi)
 796 {
 797         struct the_nilfs *nilfs = sbi->s_nilfs;
 798         int ret = 0;
 799
 800         if (nilfs_mdt_fetch_dirty(sbi->s_ifile))
 801                 ret++;
 802         if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
 803                 ret++;
 804         if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
 805                 ret++;
 806         if (ret || nilfs_doing_gc())
 807                 if (nilfs_mdt_fetch_dirty(nilfs_dat_inode(nilfs)))
 808                         ret++;
 809         return ret;
 810 }
 811
 812 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
 813 {
 814         return list_empty(&sci->sc_dirty_files) &&
 815                 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
 816                 list_empty(&sci->sc_cleaning_segments) &&
 817                 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
 818 }
 819
 820 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
 821 {
 822         struct nilfs_sb_info *sbi = sci->sc_sbi;
 823         int ret = 0;
 824
 825         if (nilfs_test_metadata_dirty(sbi))
 826                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
 827
 828         spin_lock(&sbi->s_inode_lock);
 829         if (list_empty(&sbi->s_dirty_files) && nilfs_segctor_clean(sci))
 830                 ret++;
 831
 832         spin_unlock(&sbi->s_inode_lock);
 833         return ret;
 834 }
 835
 836 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
 837 {
 838         struct nilfs_sb_info *sbi = sci->sc_sbi;
 839         struct the_nilfs *nilfs = sbi->s_nilfs;
 840
 841         nilfs_mdt_clear_dirty(sbi->s_ifile);
 842         nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
 843         nilfs_mdt_clear_dirty(nilfs->ns_sufile);
 844         nilfs_mdt_clear_dirty(nilfs_dat_inode(nilfs));
 845 }
 846
 847 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
 848 {
 849         struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
 850         struct buffer_head *bh_cp;
 851         struct nilfs_checkpoint *raw_cp;
 852         int err;
 853
 854         /* XXX: this interface will be changed */
 855         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
 856                                           &raw_cp, &bh_cp);
 857         if (likely(!err)) {
 858                 /* The following code is duplicated with cpfile.  But, it is
 859                    needed to collect the checkpoint even if it was not newly
 860                    created */
 861                 nilfs_mdt_mark_buffer_dirty(bh_cp);
 862                 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
 863                 nilfs_cpfile_put_checkpoint(
 864                         nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 865         } else {
 866                 BUG_ON(err == -EINVAL || err == -ENOENT);
 867         }
 868         return err;
 869 }
 870
 871 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
 872 {
 873         struct nilfs_sb_info *sbi = sci->sc_sbi;
 874         struct the_nilfs *nilfs = sbi->s_nilfs;
 875         struct buffer_head *bh_cp;
 876         struct nilfs_checkpoint *raw_cp;
 877         int err;
 878
 879         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
 880                                           &raw_cp, &bh_cp);
 881         if (unlikely(err)) {
 882                 BUG_ON(err == -EINVAL || err == -ENOENT);
 883                 goto failed_ibh;
 884         }
 885         raw_cp->cp_snapshot_list.ssl_next = 0;
 886         raw_cp->cp_snapshot_list.ssl_prev = 0;
 887         raw_cp->cp_inodes_count =
 888                 cpu_to_le64(atomic_read(&sbi->s_inodes_count));
 889         raw_cp->cp_blocks_count =
 890                 cpu_to_le64(atomic_read(&sbi->s_blocks_count));
 891         raw_cp->cp_nblk_inc =
 892                 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
 893         raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
 894         raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
 895         if (sci->sc_sketch_inode && i_size_read(sci->sc_sketch_inode) > 0)
 896                 nilfs_checkpoint_set_sketch(raw_cp);
 897         nilfs_write_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode, 1);
 898         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 899         return 0;
 900
 901  failed_ibh:
 902         return err;
 903 }
 904
 905 static void nilfs_fill_in_file_bmap(struct inode *ifile,
 906                                     struct nilfs_inode_info *ii)
 907
 908 {
 909         struct buffer_head *ibh;
 910         struct nilfs_inode *raw_inode;
 911
 912         if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
 913                 ibh = ii->i_bh;
 914                 BUG_ON(!ibh);
 915                 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
 916                                                   ibh);
 917                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 918                 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
 919         }
 920 }
 921
 922 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci,
 923                                             struct inode *ifile)
 924 {
 925         struct nilfs_inode_info *ii;
 926
 927         list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
 928                 nilfs_fill_in_file_bmap(ifile, ii);
 929                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
 930         }
 931         if (sci->sc_sketch_inode) {
 932                 ii = NILFS_I(sci->sc_sketch_inode);
 933                 if (test_bit(NILFS_I_DIRTY, &ii->i_state))
 934                         nilfs_fill_in_file_bmap(ifile, ii);
 935         }
 936 }
 937
 938 /*
 939  * CRC calculation routines
 940  */
 941 static void nilfs_fill_in_super_root_crc(struct buffer_head *bh_sr, u32 seed)
 942 {
 943         struct nilfs_super_root *raw_sr =
 944                 (struct nilfs_super_root *)bh_sr->b_data;
 945         u32 crc;
 946
 947         BUG_ON(NILFS_SR_BYTES > bh_sr->b_size);
 948         crc = crc32_le(seed,
 949                        (unsigned char *)raw_sr + sizeof(raw_sr->sr_sum),
 950                        NILFS_SR_BYTES - sizeof(raw_sr->sr_sum));
 951         raw_sr->sr_sum = cpu_to_le32(crc);
 952 }
 953
 954 static void nilfs_segctor_fill_in_checksums(struct nilfs_sc_info *sci,
 955                                             u32 seed)
 956 {
 957         struct nilfs_segment_buffer *segbuf;
 958
 959         if (sci->sc_super_root)
 960                 nilfs_fill_in_super_root_crc(sci->sc_super_root, seed);
 961
 962         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
 963                 nilfs_segbuf_fill_in_segsum_crc(segbuf, seed);
 964                 nilfs_segbuf_fill_in_data_crc(segbuf, seed);
 965         }
 966 }
 967
 968 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
 969                                              struct the_nilfs *nilfs)
 970 {
 971         struct buffer_head *bh_sr = sci->sc_super_root;
 972         struct nilfs_super_root *raw_sr =
 973                 (struct nilfs_super_root *)bh_sr->b_data;
 974         unsigned isz = nilfs->ns_inode_size;
 975
 976         raw_sr->sr_bytes = cpu_to_le16(NILFS_SR_BYTES);
 977         raw_sr->sr_nongc_ctime
 978                 = cpu_to_le64(nilfs_doing_gc() ?
 979                               nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
 980         raw_sr->sr_flags = 0;
 981
 982         nilfs_mdt_write_inode_direct(
 983                 nilfs_dat_inode(nilfs), bh_sr, NILFS_SR_DAT_OFFSET(isz));
 984         nilfs_mdt_write_inode_direct(
 985                 nilfs->ns_cpfile, bh_sr, NILFS_SR_CPFILE_OFFSET(isz));
 986         nilfs_mdt_write_inode_direct(
 987                 nilfs->ns_sufile, bh_sr, NILFS_SR_SUFILE_OFFSET(isz));
 988 }
 989
 990 static void nilfs_redirty_inodes(struct list_head *head)
 991 {
 992         struct nilfs_inode_info *ii;
 993
 994         list_for_each_entry(ii, head, i_dirty) {
 995                 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
 996                         clear_bit(NILFS_I_COLLECTED, &ii->i_state);
 997         }
 998 }
 999
1000 static void nilfs_drop_collected_inodes(struct list_head *head)
1001 {
1002         struct nilfs_inode_info *ii;
1003
1004         list_for_each_entry(ii, head, i_dirty) {
1005                 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1006                         continue;
1007
1008                 clear_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
1009                 set_bit(NILFS_I_UPDATED, &ii->i_state);
1010         }
1011 }
1012
1013 static void nilfs_segctor_cancel_free_segments(struct nilfs_sc_info *sci,
1014                                                struct inode *sufile)
1015
1016 {
1017         struct list_head *head = &sci->sc_cleaning_segments;
1018         struct nilfs_segment_entry *ent;
1019         int err;
1020
1021         list_for_each_entry(ent, head, list) {
1022                 if (!(ent->flags & NILFS_SLH_FREED))
1023                         break;
1024                 err = nilfs_sufile_cancel_free(sufile, ent->segnum);
1025                 BUG_ON(err);
1026
1027                 ent->flags &= ~NILFS_SLH_FREED;
1028         }
1029 }
1030
1031 static int nilfs_segctor_prepare_free_segments(struct nilfs_sc_info *sci,
1032                                                struct inode *sufile)
1033 {
1034         struct list_head *head = &sci->sc_cleaning_segments;
1035         struct nilfs_segment_entry *ent;
1036         int err;
1037
1038         list_for_each_entry(ent, head, list) {
1039                 err = nilfs_sufile_free(sufile, ent->segnum);
1040                 if (unlikely(err))
1041                         return err;
1042                 ent->flags |= NILFS_SLH_FREED;
1043         }
1044         return 0;
1045 }
1046
1047 static void nilfs_segctor_commit_free_segments(struct nilfs_sc_info *sci)
1048 {
1049         nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
1050 }
1051
1052 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1053                                        struct inode *inode,
1054                                        struct list_head *listp,
1055                                        int (*collect)(struct nilfs_sc_info *,
1056                                                       struct buffer_head *,
1057                                                       struct inode *))
1058 {
1059         struct buffer_head *bh, *n;
1060         int err = 0;
1061
1062         if (collect) {
1063                 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1064                         list_del_init(&bh->b_assoc_buffers);
1065                         err = collect(sci, bh, inode);
1066                         brelse(bh);
1067                         if (unlikely(err))
1068                                 goto dispose_buffers;
1069                 }
1070                 return 0;
1071         }
1072
1073  dispose_buffers:
1074         while (!list_empty(listp)) {
1075                 bh = list_entry(listp->next, struct buffer_head,
1076                                 b_assoc_buffers);
1077                 list_del_init(&bh->b_assoc_buffers);
1078                 brelse(bh);
1079         }
1080         return err;
1081 }
1082
1083 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1084 {
1085         /* Remaining number of blocks within segment buffer */
1086         return sci->sc_segbuf_nblocks -
1087                 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1088 }
1089
1090 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1091                                    struct inode *inode,
1092                                    struct nilfs_sc_operations *sc_ops)
1093 {
1094         LIST_HEAD(data_buffers);
1095         LIST_HEAD(node_buffers);
1096         int err;
1097
1098         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1099                 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1100
1101                 n = nilfs_lookup_dirty_data_buffers(
1102                         inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1103                 if (n > rest) {
1104                         err = nilfs_segctor_apply_buffers(
1105                                 sci, inode, &data_buffers,
1106                                 sc_ops->collect_data);
1107                         BUG_ON(!err); /* always receive -E2BIG or true error */
1108                         goto break_or_fail;
1109                 }
1110         }
1111         nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1112
1113         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1114                 err = nilfs_segctor_apply_buffers(
1115                         sci, inode, &data_buffers, sc_ops->collect_data);
1116                 if (unlikely(err)) {
1117                         /* dispose node list */
1118                         nilfs_segctor_apply_buffers(
1119                                 sci, inode, &node_buffers, NULL);
1120                         goto break_or_fail;
1121                 }
1122                 sci->sc_stage.flags |= NILFS_CF_NODE;
1123         }
1124         /* Collect node */
1125         err = nilfs_segctor_apply_buffers(
1126                 sci, inode, &node_buffers, sc_ops->collect_node);
1127         if (unlikely(err))
1128                 goto break_or_fail;
1129
1130         nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1131         err = nilfs_segctor_apply_buffers(
1132                 sci, inode, &node_buffers, sc_ops->collect_bmap);
1133         if (unlikely(err))
1134                 goto break_or_fail;
1135
1136         nilfs_segctor_end_finfo(sci, inode);
1137         sci->sc_stage.flags &= ~NILFS_CF_NODE;
1138
1139  break_or_fail:
1140         return err;
1141 }
1142
1143 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1144                                          struct inode *inode)
1145 {
1146         LIST_HEAD(data_buffers);
1147         size_t n, rest = nilfs_segctor_buffer_rest(sci);
1148         int err;
1149
1150         n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1151                                             sci->sc_dsync_start,
1152                                             sci->sc_dsync_end);
1153
1154         err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1155                                           nilfs_collect_file_data);
1156         if (!err) {
1157                 nilfs_segctor_end_finfo(sci, inode);
1158                 BUG_ON(n > rest);
1159                 /* always receive -E2BIG or true error if n > rest */
1160         }
1161         return err;
1162 }
1163
1164 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1165 {
1166         struct nilfs_sb_info *sbi = sci->sc_sbi;
1167         struct the_nilfs *nilfs = sbi->s_nilfs;
1168         struct list_head *head;
1169         struct nilfs_inode_info *ii;
1170         int err = 0;
1171
1172         switch (sci->sc_stage.scnt) {
1173         case NILFS_ST_INIT:
1174                 /* Pre-processes */
1175                 sci->sc_stage.flags = 0;
1176
1177                 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1178                         sci->sc_nblk_inc = 0;
1179                         sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1180                         if (mode == SC_LSEG_DSYNC) {
1181                                 sci->sc_stage.scnt = NILFS_ST_DSYNC;
1182                                 goto dsync_mode;
1183                         }
1184                 }
1185
1186                 sci->sc_stage.dirty_file_ptr = NULL;
1187                 sci->sc_stage.gc_inode_ptr = NULL;
1188                 if (mode == SC_FLUSH_DAT) {
1189                         sci->sc_stage.scnt = NILFS_ST_DAT;
1190                         goto dat_stage;
1191                 }
1192                 sci->sc_stage.scnt++;  /* Fall through */
1193         case NILFS_ST_GC:
1194                 if (nilfs_doing_gc()) {
1195                         head = &sci->sc_gc_inodes;
1196                         ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1197                                                 head, i_dirty);
1198                         list_for_each_entry_continue(ii, head, i_dirty) {
1199                                 err = nilfs_segctor_scan_file(
1200                                         sci, &ii->vfs_inode,
1201                                         &nilfs_sc_file_ops);
1202                                 if (unlikely(err)) {
1203                                         sci->sc_stage.gc_inode_ptr = list_entry(
1204                                                 ii->i_dirty.prev,
1205                                                 struct nilfs_inode_info,
1206                                                 i_dirty);
1207                                         goto break_or_fail;
1208                                 }
1209                                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1210                         }
1211                         sci->sc_stage.gc_inode_ptr = NULL;
1212                 }
1213                 sci->sc_stage.scnt++;  /* Fall through */
1214         case NILFS_ST_FILE:
1215                 head = &sci->sc_dirty_files;
1216                 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1217                                         i_dirty);
1218                 list_for_each_entry_continue(ii, head, i_dirty) {
1219                         clear_bit(NILFS_I_DIRTY, &ii->i_state);
1220
1221                         err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1222                                                       &nilfs_sc_file_ops);
1223                         if (unlikely(err)) {
1224                                 sci->sc_stage.dirty_file_ptr =
1225                                         list_entry(ii->i_dirty.prev,
1226                                                    struct nilfs_inode_info,
1227                                                    i_dirty);
1228                                 goto break_or_fail;
1229                         }
1230                         /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1231                         /* XXX: required ? */
1232                 }
1233                 sci->sc_stage.dirty_file_ptr = NULL;
1234                 if (mode == SC_FLUSH_FILE) {
1235                         sci->sc_stage.scnt = NILFS_ST_DONE;
1236                         return 0;
1237                 }
1238                 sci->sc_stage.scnt++;  /* Fall through */
1239         case NILFS_ST_SKETCH:
1240                 if (mode == SC_LSEG_SR && sci->sc_sketch_inode) {
1241                         ii = NILFS_I(sci->sc_sketch_inode);
1242                         if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
1243                                 sci->sc_sketch_inode->i_ctime.tv_sec
1244                                         = sci->sc_seg_ctime;
1245                                 sci->sc_sketch_inode->i_mtime.tv_sec
1246                                         = sci->sc_seg_ctime;
1247                                 err = nilfs_mark_inode_dirty(
1248                                         sci->sc_sketch_inode);
1249                                 if (unlikely(err))
1250                                         goto break_or_fail;
1251                         }
1252                         err = nilfs_segctor_scan_file(sci,
1253                                                       sci->sc_sketch_inode,
1254                                                       &nilfs_sc_file_ops);
1255                         if (unlikely(err))
1256                                 goto break_or_fail;
1257                 }
1258                 sci->sc_stage.scnt++;
1259                 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1260                 /* Fall through */
1261         case NILFS_ST_IFILE:
1262                 err = nilfs_segctor_scan_file(sci, sbi->s_ifile,
1263                                               &nilfs_sc_file_ops);
1264                 if (unlikely(err))
1265                         break;
1266                 sci->sc_stage.scnt++;
1267                 /* Creating a checkpoint */
1268                 err = nilfs_segctor_create_checkpoint(sci);
1269                 if (unlikely(err))
1270                         break;
1271                 /* Fall through */
1272         case NILFS_ST_CPFILE:
1273                 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1274                                               &nilfs_sc_file_ops);
1275                 if (unlikely(err))
1276                         break;
1277                 sci->sc_stage.scnt++;  /* Fall through */
1278         case NILFS_ST_SUFILE:
1279                 err = nilfs_segctor_prepare_free_segments(sci,
1280                                                           nilfs->ns_sufile);
1281                 if (unlikely(err))
1282                         break;
1283                 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1284                                               &nilfs_sc_file_ops);
1285                 if (unlikely(err))
1286                         break;
1287                 sci->sc_stage.scnt++;  /* Fall through */
1288         case NILFS_ST_DAT:
1289  dat_stage:
1290                 err = nilfs_segctor_scan_file(sci, nilfs_dat_inode(nilfs),
1291                                               &nilfs_sc_dat_ops);
1292                 if (unlikely(err))
1293                         break;
1294                 if (mode == SC_FLUSH_DAT) {
1295                         sci->sc_stage.scnt = NILFS_ST_DONE;
1296                         return 0;
1297                 }
1298                 sci->sc_stage.scnt++;  /* Fall through */
1299         case NILFS_ST_SR:
1300                 if (mode == SC_LSEG_SR) {
1301                         /* Appending a super root */
1302                         err = nilfs_segctor_add_super_root(sci);
1303                         if (unlikely(err))
1304                                 break;
1305                 }
1306                 /* End of a logical segment */
1307                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1308                 sci->sc_stage.scnt = NILFS_ST_DONE;
1309                 return 0;
1310         case NILFS_ST_DSYNC:
1311  dsync_mode:
1312                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1313                 ii = sci->sc_dsync_inode;
1314                 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1315                         break;
1316
1317                 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1318                 if (unlikely(err))
1319                         break;
1320                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1321                 sci->sc_stage.scnt = NILFS_ST_DONE;
1322                 return 0;
1323         case NILFS_ST_DONE:
1324                 return 0;
1325         default:
1326                 BUG();
1327         }
1328
1329  break_or_fail:
1330         return err;
1331 }
1332
1333 static int nilfs_segctor_terminate_segment(struct nilfs_sc_info *sci,
1334                                            struct nilfs_segment_buffer *segbuf,
1335                                            struct inode *sufile)
1336 {
1337         struct nilfs_segment_entry *ent = segbuf->sb_segent;
1338         int err;
1339
1340         err = nilfs_open_segment_entry(ent, sufile);
1341         if (unlikely(err))
1342                 return err;
1343         nilfs_mdt_mark_buffer_dirty(ent->bh_su);
1344         nilfs_mdt_mark_dirty(sufile);
1345         nilfs_close_segment_entry(ent, sufile);
1346
1347         list_add_tail(&ent->list, &sci->sc_active_segments);
1348         segbuf->sb_segent = NULL;
1349         return 0;
1350 }
1351
1352 static int nilfs_touch_segusage(struct inode *sufile, __u64 segnum)
1353 {
1354         struct buffer_head *bh_su;
1355         struct nilfs_segment_usage *raw_su;
1356         int err;
1357
1358         err = nilfs_sufile_get_segment_usage(sufile, segnum, &raw_su, &bh_su);
1359         if (unlikely(err))
1360                 return err;
1361         nilfs_mdt_mark_buffer_dirty(bh_su);
1362         nilfs_mdt_mark_dirty(sufile);
1363         nilfs_sufile_put_segment_usage(sufile, segnum, bh_su);
1364         return 0;
1365 }
1366
1367 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1368                                             struct the_nilfs *nilfs)
1369 {
1370         struct nilfs_segment_buffer *segbuf, *n;
1371         struct inode *sufile = nilfs->ns_sufile;
1372         __u64 nextnum;
1373         int err;
1374
1375         if (list_empty(&sci->sc_segbufs)) {
1376                 segbuf = nilfs_segbuf_new(sci->sc_super);
1377                 if (unlikely(!segbuf))
1378                         return -ENOMEM;
1379                 list_add(&segbuf->sb_list, &sci->sc_segbufs);
1380         } else
1381                 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1382
1383         err = nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1384                                nilfs->ns_pseg_offset, nilfs);
1385         if (unlikely(err))
1386                 return err;
1387
1388         if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1389                 err = nilfs_segctor_terminate_segment(sci, segbuf, sufile);
1390                 if (unlikely(err))
1391                         return err;
1392
1393                 nilfs_shift_to_next_segment(nilfs);
1394                 err = nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1395         }
1396         sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1397
1398         err = nilfs_touch_segusage(sufile, segbuf->sb_segnum);
1399         if (unlikely(err))
1400                 return err;
1401
1402         if (nilfs->ns_segnum == nilfs->ns_nextnum) {
1403                 /* Start from the head of a new full segment */
1404                 err = nilfs_sufile_alloc(sufile, &nextnum);
1405                 if (unlikely(err))
1406                         return err;
1407         } else
1408                 nextnum = nilfs->ns_nextnum;
1409
1410         segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1411         nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1412
1413         /* truncating segment buffers */
1414         list_for_each_entry_safe_continue(segbuf, n, &sci->sc_segbufs,
1415                                           sb_list) {
1416                 list_del_init(&segbuf->sb_list);
1417                 nilfs_segbuf_free(segbuf);
1418         }
1419         return err;
1420 }
1421
1422 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1423                                          struct the_nilfs *nilfs, int nadd)
1424 {
1425         struct nilfs_segment_buffer *segbuf, *prev, *n;
1426         struct inode *sufile = nilfs->ns_sufile;
1427         __u64 nextnextnum;
1428         LIST_HEAD(list);
1429         int err, ret, i;
1430
1431         prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1432         /*
1433          * Since the segment specified with nextnum might be allocated during
1434          * the previous construction, the buffer including its segusage may
1435          * not be dirty.  The following call ensures that the buffer is dirty
1436          * and will pin the buffer on memory until the sufile is written.
1437          */
1438         err = nilfs_touch_segusage(sufile, prev->sb_nextnum);
1439         if (unlikely(err))
1440                 return err;
1441
1442         for (i = 0; i < nadd; i++) {
1443                 /* extend segment info */
1444                 err = -ENOMEM;
1445                 segbuf = nilfs_segbuf_new(sci->sc_super);
1446                 if (unlikely(!segbuf))
1447                         goto failed;
1448
1449                 /* map this buffer to region of segment on-disk */
1450                 err = nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1451                 if (unlikely(err))
1452                         goto failed_segbuf;
1453
1454                 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1455
1456                 /* allocate the next next full segment */
1457                 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1458                 if (unlikely(err))
1459                         goto failed_segbuf;
1460
1461                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1462                 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1463
1464                 list_add_tail(&segbuf->sb_list, &list);
1465                 prev = segbuf;
1466         }
1467         list_splice(&list, sci->sc_segbufs.prev);
1468         return 0;
1469
1470  failed_segbuf:
1471         nilfs_segbuf_free(segbuf);
1472  failed:
1473         list_for_each_entry_safe(segbuf, n, &list, sb_list) {
1474                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1475                 BUG_ON(ret);
1476                 list_del_init(&segbuf->sb_list);
1477                 nilfs_segbuf_free(segbuf);
1478         }
1479         return err;
1480 }
1481
1482 static void nilfs_segctor_free_incomplete_segments(struct nilfs_sc_info *sci,
1483                                                    struct the_nilfs *nilfs)
1484 {
1485         struct nilfs_segment_buffer *segbuf;
1486         int ret, done = 0;
1487
1488         segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1489         if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1490                 ret = nilfs_sufile_free(nilfs->ns_sufile, segbuf->sb_nextnum);
1491                 BUG_ON(ret);
1492         }
1493         if (segbuf->sb_io_error) {
1494                 /* Case 1: The first segment failed */
1495                 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1496                         /* Case 1a:  Partial segment appended into an existing
1497                            segment */
1498                         nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1499                                                 segbuf->sb_fseg_end);
1500                 else /* Case 1b:  New full segment */
1501                         set_nilfs_discontinued(nilfs);
1502                 done++;
1503         }
1504
1505         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1506                 ret = nilfs_sufile_free(nilfs->ns_sufile, segbuf->sb_nextnum);
1507                 BUG_ON(ret);
1508                 if (!done && segbuf->sb_io_error) {
1509                         if (segbuf->sb_segnum != nilfs->ns_nextnum)
1510                                 /* Case 2: extended segment (!= next) failed */
1511                                 nilfs_sufile_set_error(nilfs->ns_sufile,
1512                                                        segbuf->sb_segnum);
1513                         done++;
1514                 }
1515         }
1516 }
1517
1518 static void nilfs_segctor_clear_segment_buffers(struct nilfs_sc_info *sci)
1519 {
1520         struct nilfs_segment_buffer *segbuf;
1521
1522         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list)
1523                 nilfs_segbuf_clear(segbuf);
1524         sci->sc_super_root = NULL;
1525 }
1526
1527 static void nilfs_segctor_destroy_segment_buffers(struct nilfs_sc_info *sci)
1528 {
1529         struct nilfs_segment_buffer *segbuf;
1530
1531         while (!list_empty(&sci->sc_segbufs)) {
1532                 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1533                 list_del_init(&segbuf->sb_list);
1534                 nilfs_segbuf_free(segbuf);
1535         }
1536         /* sci->sc_curseg = NULL; */
1537 }
1538
1539 static void nilfs_segctor_end_construction(struct nilfs_sc_info *sci,
1540                                            struct the_nilfs *nilfs, int err)
1541 {
1542         if (unlikely(err)) {
1543                 nilfs_segctor_free_incomplete_segments(sci, nilfs);
1544                 nilfs_segctor_cancel_free_segments(sci, nilfs->ns_sufile);
1545         }
1546         nilfs_segctor_clear_segment_buffers(sci);
1547 }
1548
1549 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1550                                           struct inode *sufile)
1551 {
1552         struct nilfs_segment_buffer *segbuf;
1553         struct buffer_head *bh_su;
1554         struct nilfs_segment_usage *raw_su;
1555         unsigned long live_blocks;
1556         int ret;
1557
1558         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1559                 ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1560                                                      &raw_su, &bh_su);
1561                 BUG_ON(ret); /* always succeed because bh_su is dirty */
1562                 live_blocks = segbuf->sb_sum.nblocks +
1563                         (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1564                 raw_su->su_lastmod = cpu_to_le64(sci->sc_seg_ctime);
1565                 raw_su->su_nblocks = cpu_to_le32(live_blocks);
1566                 nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum,
1567                                                bh_su);
1568         }
1569 }
1570
1571 static void nilfs_segctor_cancel_segusage(struct nilfs_sc_info *sci,
1572                                           struct inode *sufile)
1573 {
1574         struct nilfs_segment_buffer *segbuf;
1575         struct buffer_head *bh_su;
1576         struct nilfs_segment_usage *raw_su;
1577         int ret;
1578
1579         segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1580         ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1581                                              &raw_su, &bh_su);
1582         BUG_ON(ret); /* always succeed because bh_su is dirty */
1583         raw_su->su_nblocks = cpu_to_le32(segbuf->sb_pseg_start -
1584                                          segbuf->sb_fseg_start);
1585         nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum, bh_su);
1586
1587         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1588                 ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1589                                                      &raw_su, &bh_su);
1590                 BUG_ON(ret); /* always succeed */
1591                 raw_su->su_nblocks = 0;
1592                 nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum,
1593                                                bh_su);
1594         }
1595 }
1596
1597 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1598                                             struct nilfs_segment_buffer *last,
1599                                             struct inode *sufile)
1600 {
1601         struct nilfs_segment_buffer *segbuf = last, *n;
1602         int ret;
1603
1604         list_for_each_entry_safe_continue(segbuf, n, &sci->sc_segbufs,
1605                                           sb_list) {
1606                 list_del_init(&segbuf->sb_list);
1607                 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1608                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1609                 BUG_ON(ret);
1610                 nilfs_segbuf_free(segbuf);
1611         }
1612 }
1613
1614
1615 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1616                                  struct the_nilfs *nilfs, int mode)
1617 {
1618         struct nilfs_cstage prev_stage = sci->sc_stage;
1619         int err, nadd = 1;
1620
1621         /* Collection retry loop */
1622         for (;;) {
1623                 sci->sc_super_root = NULL;
1624                 sci->sc_nblk_this_inc = 0;
1625                 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1626
1627                 err = nilfs_segctor_reset_segment_buffer(sci);
1628                 if (unlikely(err))
1629                         goto failed;
1630
1631                 err = nilfs_segctor_collect_blocks(sci, mode);
1632                 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1633                 if (!err)
1634                         break;
1635
1636                 if (unlikely(err != -E2BIG))
1637                         goto failed;
1638
1639                 /* The current segment is filled up */
1640                 if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
1641                         break;
1642
1643                 nilfs_segctor_cancel_free_segments(sci, nilfs->ns_sufile);
1644                 nilfs_segctor_clear_segment_buffers(sci);
1645
1646                 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1647                 if (unlikely(err))
1648                         return err;
1649
1650                 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1651                 sci->sc_stage = prev_stage;
1652         }
1653         nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1654         return 0;
1655
1656  failed:
1657         return err;
1658 }
1659
1660 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1661                                       struct buffer_head *new_bh)
1662 {
1663         BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1664
1665         list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1666         /* The caller must release old_bh */
1667 }
1668
1669 static int
1670 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1671                                      struct nilfs_segment_buffer *segbuf,
1672                                      int mode)
1673 {
1674         struct inode *inode = NULL;
1675         sector_t blocknr;
1676         unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1677         unsigned long nblocks = 0, ndatablk = 0;
1678         struct nilfs_sc_operations *sc_op = NULL;
1679         struct nilfs_segsum_pointer ssp;
1680         struct nilfs_finfo *finfo = NULL;
1681         union nilfs_binfo binfo;
1682         struct buffer_head *bh, *bh_org;
1683         ino_t ino = 0;
1684         int err = 0;
1685
1686         if (!nfinfo)
1687                 goto out;
1688
1689         blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1690         ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1691         ssp.offset = sizeof(struct nilfs_segment_summary);
1692
1693         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1694                 if (bh == sci->sc_super_root)
1695                         break;
1696                 if (!finfo) {
1697                         finfo = nilfs_segctor_map_segsum_entry(
1698                                 sci, &ssp, sizeof(*finfo));
1699                         ino = le64_to_cpu(finfo->fi_ino);
1700                         nblocks = le32_to_cpu(finfo->fi_nblocks);
1701                         ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1702
1703                         if (buffer_nilfs_node(bh))
1704                                 inode = NILFS_BTNC_I(bh->b_page->mapping);
1705                         else
1706                                 inode = NILFS_AS_I(bh->b_page->mapping);
1707
1708                         if (mode == SC_LSEG_DSYNC)
1709                                 sc_op = &nilfs_sc_dsync_ops;
1710                         else if (ino == NILFS_DAT_INO)
1711                                 sc_op = &nilfs_sc_dat_ops;
1712                         else /* file blocks */
1713                                 sc_op = &nilfs_sc_file_ops;
1714                 }
1715                 bh_org = bh;
1716                 get_bh(bh_org);
1717                 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1718                                         &binfo);
1719                 if (bh != bh_org)
1720                         nilfs_list_replace_buffer(bh_org, bh);
1721                 brelse(bh_org);
1722                 if (unlikely(err))
1723                         goto failed_bmap;
1724
1725                 if (ndatablk > 0)
1726                         sc_op->write_data_binfo(sci, &ssp, &binfo);
1727                 else
1728                         sc_op->write_node_binfo(sci, &ssp, &binfo);
1729
1730                 blocknr++;
1731                 if (--nblocks == 0) {
1732                         finfo = NULL;
1733                         if (--nfinfo == 0)
1734                                 break;
1735                 } else if (ndatablk > 0)
1736                         ndatablk--;
1737         }
1738  out:
1739         return 0;
1740
1741  failed_bmap:
1742         err = nilfs_handle_bmap_error(err, __func__, inode, sci->sc_super);
1743         return err;
1744 }
1745
1746 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1747 {
1748         struct nilfs_segment_buffer *segbuf;
1749         int err;
1750
1751         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1752                 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1753                 if (unlikely(err))
1754                         return err;
1755                 nilfs_segbuf_fill_in_segsum(segbuf);
1756         }
1757         return 0;
1758 }
1759
1760 static int
1761 nilfs_copy_replace_page_buffers(struct page *page, struct list_head *out)
1762 {
1763         struct page *clone_page;
1764         struct buffer_head *bh, *head, *bh2;
1765         void *kaddr;
1766
1767         bh = head = page_buffers(page);
1768
1769         clone_page = nilfs_alloc_private_page(bh->b_bdev, bh->b_size, 0);
1770         if (unlikely(!clone_page))
1771                 return -ENOMEM;
1772
1773         bh2 = page_buffers(clone_page);
1774         kaddr = kmap_atomic(page, KM_USER0);
1775         do {
1776                 if (list_empty(&bh->b_assoc_buffers))
1777                         continue;
1778                 get_bh(bh2);
1779                 page_cache_get(clone_page); /* for each bh */
1780                 memcpy(bh2->b_data, kaddr + bh_offset(bh), bh2->b_size);
1781                 bh2->b_blocknr = bh->b_blocknr;
1782                 list_replace(&bh->b_assoc_buffers, &bh2->b_assoc_buffers);
1783                 list_add_tail(&bh->b_assoc_buffers, out);
1784         } while (bh = bh->b_this_page, bh2 = bh2->b_this_page, bh != head);
1785         kunmap_atomic(kaddr, KM_USER0);
1786
1787         if (!TestSetPageWriteback(clone_page))
1788                 inc_zone_page_state(clone_page, NR_WRITEBACK);
1789         unlock_page(clone_page);
1790
1791         return 0;
1792 }
1793
1794 static int nilfs_test_page_to_be_frozen(struct page *page)
1795 {
1796         struct address_space *mapping = page->mapping;
1797
1798         if (!mapping || !mapping->host || S_ISDIR(mapping->host->i_mode))
1799                 return 0;
1800
1801         if (page_mapped(page)) {
1802                 ClearPageChecked(page);
1803                 return 1;
1804         }
1805         return PageChecked(page);
1806 }
1807
1808 static int nilfs_begin_page_io(struct page *page, struct list_head *out)
1809 {
1810         if (!page || PageWriteback(page))
1811                 /* For split b-tree node pages, this function may be called
1812                    twice.  We ignore the 2nd or later calls by this check. */
1813                 return 0;
1814
1815         lock_page(page);
1816         clear_page_dirty_for_io(page);
1817         set_page_writeback(page);
1818         unlock_page(page);
1819
1820         if (nilfs_test_page_to_be_frozen(page)) {
1821                 int err = nilfs_copy_replace_page_buffers(page, out);
1822                 if (unlikely(err))
1823                         return err;
1824         }
1825         return 0;
1826 }
1827
1828 static int nilfs_segctor_prepare_write(struct nilfs_sc_info *sci,
1829                                        struct page **failed_page)
1830 {
1831         struct nilfs_segment_buffer *segbuf;
1832         struct page *bd_page = NULL, *fs_page = NULL;
1833         struct list_head *list = &sci->sc_copied_buffers;
1834         int err;
1835
1836         *failed_page = NULL;
1837         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1838                 struct buffer_head *bh;
1839
1840                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1841                                     b_assoc_buffers) {
1842                         if (bh->b_page != bd_page) {
1843                                 if (bd_page) {
1844                                         lock_page(bd_page);
1845                                         clear_page_dirty_for_io(bd_page);
1846                                         set_page_writeback(bd_page);
1847                                         unlock_page(bd_page);
1848                                 }
1849                                 bd_page = bh->b_page;
1850                         }
1851                 }
1852
1853                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1854                                     b_assoc_buffers) {
1855                         if (bh == sci->sc_super_root) {
1856                                 if (bh->b_page != bd_page) {
1857                                         lock_page(bd_page);
1858                                         clear_page_dirty_for_io(bd_page);
1859                                         set_page_writeback(bd_page);
1860                                         unlock_page(bd_page);
1861                                         bd_page = bh->b_page;
1862                                 }
1863                                 break;
1864                         }
1865                         if (bh->b_page != fs_page) {
1866                                 err = nilfs_begin_page_io(fs_page, list);
1867                                 if (unlikely(err)) {
1868                                         *failed_page = fs_page;
1869                                         goto out;
1870                                 }
1871                                 fs_page = bh->b_page;
1872                         }
1873                 }
1874         }
1875         if (bd_page) {
1876                 lock_page(bd_page);
1877                 clear_page_dirty_for_io(bd_page);
1878                 set_page_writeback(bd_page);
1879                 unlock_page(bd_page);
1880         }
1881         err = nilfs_begin_page_io(fs_page, list);
1882         if (unlikely(err))
1883                 *failed_page = fs_page;
1884  out:
1885         return err;
1886 }
1887
1888 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1889                                struct backing_dev_info *bdi)
1890 {
1891         struct nilfs_segment_buffer *segbuf;
1892         struct nilfs_write_info wi;
1893         int err, res;
1894
1895         wi.sb = sci->sc_super;
1896         wi.bh_sr = sci->sc_super_root;
1897         wi.bdi = bdi;
1898
1899         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1900                 nilfs_segbuf_prepare_write(segbuf, &wi);
1901                 err = nilfs_segbuf_write(segbuf, &wi);
1902
1903                 res = nilfs_segbuf_wait(segbuf, &wi);
1904                 err = unlikely(err) ? : res;
1905                 if (unlikely(err))
1906                         return err;
1907         }
1908         return 0;
1909 }
1910
1911 static int nilfs_page_has_uncleared_buffer(struct page *page)
1912 {
1913         struct buffer_head *head, *bh;
1914
1915         head = bh = page_buffers(page);
1916         do {
1917                 if (buffer_dirty(bh) && !list_empty(&bh->b_assoc_buffers))
1918                         return 1;
1919                 bh = bh->b_this_page;
1920         } while (bh != head);
1921         return 0;
1922 }
1923
1924 static void __nilfs_end_page_io(struct page *page, int err)
1925 {
1926         /* BUG_ON(err > 0); */
1927         if (!err) {
1928                 if (!nilfs_page_buffers_clean(page))
1929                         __set_page_dirty_nobuffers(page);
1930                 ClearPageError(page);
1931         } else {
1932                 __set_page_dirty_nobuffers(page);
1933                 SetPageError(page);
1934         }
1935
1936         if (buffer_nilfs_allocated(page_buffers(page))) {
1937                 if (TestClearPageWriteback(page))
1938                         dec_zone_page_state(page, NR_WRITEBACK);
1939         } else
1940                 end_page_writeback(page);
1941 }
1942
1943 static void nilfs_end_page_io(struct page *page, int err)
1944 {
1945         if (!page)
1946                 return;
1947
1948         if (buffer_nilfs_node(page_buffers(page)) &&
1949             nilfs_page_has_uncleared_buffer(page))
1950                 /* For b-tree node pages, this function may be called twice
1951                    or more because they might be split in a segment.
1952                    This check assures that cleanup has been done for all
1953                    buffers in a split btnode page. */
1954                 return;
1955
1956         __nilfs_end_page_io(page, err);
1957 }
1958
1959 static void nilfs_clear_copied_buffers(struct list_head *list, int err)
1960 {
1961         struct buffer_head *bh, *head;
1962         struct page *page;
1963
1964         while (!list_empty(list)) {
1965                 bh = list_entry(list->next, struct buffer_head,
1966                                 b_assoc_buffers);
1967                 page = bh->b_page;
1968                 page_cache_get(page);
1969                 head = bh = page_buffers(page);
1970                 do {
1971                         if (!list_empty(&bh->b_assoc_buffers)) {
1972                                 list_del_init(&bh->b_assoc_buffers);
1973                                 if (!err) {
1974                                         set_buffer_uptodate(bh);
1975                                         clear_buffer_dirty(bh);
1976                                         clear_buffer_nilfs_volatile(bh);
1977                                 }
1978                                 brelse(bh); /* for b_assoc_buffers */
1979                         }
1980                 } while ((bh = bh->b_this_page) != head);
1981
1982                 __nilfs_end_page_io(page, err);
1983                 page_cache_release(page);
1984         }
1985 }
1986
1987 static void nilfs_segctor_abort_write(struct nilfs_sc_info *sci,
1988                                       struct page *failed_page, int err)
1989 {
1990         struct nilfs_segment_buffer *segbuf;
1991         struct page *bd_page = NULL, *fs_page = NULL;
1992
1993         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1994                 struct buffer_head *bh;
1995
1996                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1997                                     b_assoc_buffers) {
1998                         if (bh->b_page != bd_page) {
1999                                 if (bd_page)
2000                                         end_page_writeback(bd_page);
2001                                 bd_page = bh->b_page;
2002                         }
2003                 }
2004
2005                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
2006                                     b_assoc_buffers) {
2007                         if (bh == sci->sc_super_root) {
2008                                 if (bh->b_page != bd_page) {
2009                                         end_page_writeback(bd_page);
2010                                         bd_page = bh->b_page;
2011                                 }
2012                                 break;
2013                         }
2014                         if (bh->b_page != fs_page) {
2015                                 nilfs_end_page_io(fs_page, err);
2016                                 if (unlikely(fs_page == failed_page))
2017                                         goto done;
2018                                 fs_page = bh->b_page;
2019                         }
2020                 }
2021         }
2022         if (bd_page)
2023                 end_page_writeback(bd_page);
2024
2025         nilfs_end_page_io(fs_page, err);
2026  done:
2027         nilfs_clear_copied_buffers(&sci->sc_copied_buffers, err);
2028 }
2029
2030 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
2031                                    struct nilfs_segment_buffer *segbuf)
2032 {
2033         nilfs->ns_segnum = segbuf->sb_segnum;
2034         nilfs->ns_nextnum = segbuf->sb_nextnum;
2035         nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
2036                 + segbuf->sb_sum.nblocks;
2037         nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
2038         nilfs->ns_ctime = segbuf->sb_sum.ctime;
2039 }
2040
2041 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
2042 {
2043         struct nilfs_segment_buffer *segbuf;
2044         struct page *bd_page = NULL, *fs_page = NULL;
2045         struct nilfs_sb_info *sbi = sci->sc_sbi;
2046         struct the_nilfs *nilfs = sbi->s_nilfs;
2047         int update_sr = (sci->sc_super_root != NULL);
2048
2049         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
2050                 struct buffer_head *bh;
2051
2052                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
2053                                     b_assoc_buffers) {
2054                         set_buffer_uptodate(bh);
2055                         clear_buffer_dirty(bh);
2056                         if (bh->b_page != bd_page) {
2057                                 if (bd_page)
2058                                         end_page_writeback(bd_page);
2059                                 bd_page = bh->b_page;
2060                         }
2061                 }
2062                 /*
2063                  * We assume that the buffers which belong to the same page
2064                  * continue over the buffer list.
2065                  * Under this assumption, the last BHs of pages is
2066                  * identifiable by the discontinuity of bh->b_page
2067                  * (page != fs_page).
2068                  *
2069                  * For B-tree node blocks, however, this assumption is not
2070                  * guaranteed.  The cleanup code of B-tree node pages needs
2071                  * special care.
2072                  */
2073                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
2074                                     b_assoc_buffers) {
2075                         set_buffer_uptodate(bh);
2076                         clear_buffer_dirty(bh);
2077                         clear_buffer_nilfs_volatile(bh);
2078                         if (bh == sci->sc_super_root) {
2079                                 if (bh->b_page != bd_page) {
2080                                         end_page_writeback(bd_page);
2081                                         bd_page = bh->b_page;
2082                                 }
2083                                 break;
2084                         }
2085                         if (bh->b_page != fs_page) {
2086                                 nilfs_end_page_io(fs_page, 0);
2087                                 fs_page = bh->b_page;
2088                         }
2089                 }
2090
2091                 if (!NILFS_SEG_SIMPLEX(&segbuf->sb_sum)) {
2092                         if (NILFS_SEG_LOGBGN(&segbuf->sb_sum)) {
2093                                 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
2094                                 sci->sc_lseg_stime = jiffies;
2095                         }
2096                         if (NILFS_SEG_LOGEND(&segbuf->sb_sum))
2097                                 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
2098                 }
2099         }
2100         /*
2101          * Since pages may continue over multiple segment buffers,
2102          * end of the last page must be checked outside of the loop.
2103          */
2104         if (bd_page)
2105                 end_page_writeback(bd_page);
2106
2107         nilfs_end_page_io(fs_page, 0);
2108
2109         nilfs_clear_copied_buffers(&sci->sc_copied_buffers, 0);
2110
2111         nilfs_drop_collected_inodes(&sci->sc_dirty_files);
2112
2113         if (nilfs_doing_gc()) {
2114                 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
2115                 if (update_sr)
2116                         nilfs_commit_gcdat_inode(nilfs);
2117         } else
2118                 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
2119
2120         sci->sc_nblk_inc += sci->sc_nblk_this_inc;
2121
2122         segbuf = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2123         nilfs_set_next_segment(nilfs, segbuf);
2124
2125         if (update_sr) {
2126                 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
2127                                        segbuf->sb_sum.seg_seq, nilfs->ns_cno);
2128
2129                 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2130                 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
2131         } else
2132                 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
2133 }
2134
2135 static int nilfs_segctor_check_in_files(struct nilfs_sc_info *sci,
2136                                         struct nilfs_sb_info *sbi)
2137 {
2138         struct nilfs_inode_info *ii, *n;
2139         __u64 cno = sbi->s_nilfs->ns_cno;
2140
2141         spin_lock(&sbi->s_inode_lock);
2142  retry:
2143         list_for_each_entry_safe(ii, n, &sbi->s_dirty_files, i_dirty) {
2144                 if (!ii->i_bh) {
2145                         struct buffer_head *ibh;
2146                         int err;
2147
2148                         spin_unlock(&sbi->s_inode_lock);
2149                         err = nilfs_ifile_get_inode_block(
2150                                 sbi->s_ifile, ii->vfs_inode.i_ino, &ibh);
2151                         if (unlikely(err)) {
2152                                 nilfs_warning(sbi->s_super, __func__,
2153                                               "failed to get inode block.\n");
2154                                 return err;
2155                         }
2156                         nilfs_mdt_mark_buffer_dirty(ibh);
2157                         nilfs_mdt_mark_dirty(sbi->s_ifile);
2158                         spin_lock(&sbi->s_inode_lock);
2159                         if (likely(!ii->i_bh))
2160                                 ii->i_bh = ibh;
2161                         else
2162                                 brelse(ibh);
2163                         goto retry;
2164                 }
2165                 ii->i_cno = cno;
2166
2167                 clear_bit(NILFS_I_QUEUED, &ii->i_state);
2168                 set_bit(NILFS_I_BUSY, &ii->i_state);
2169                 list_del(&ii->i_dirty);
2170                 list_add_tail(&ii->i_dirty, &sci->sc_dirty_files);
2171         }
2172         spin_unlock(&sbi->s_inode_lock);
2173
2174         NILFS_I(sbi->s_ifile)->i_cno = cno;
2175
2176         return 0;
2177 }
2178
2179 static void nilfs_segctor_check_out_files(struct nilfs_sc_info *sci,
2180                                           struct nilfs_sb_info *sbi)
2181 {
2182         struct nilfs_transaction_info *ti = current->journal_info;
2183         struct nilfs_inode_info *ii, *n;
2184         __u64 cno = sbi->s_nilfs->ns_cno;
2185
2186         spin_lock(&sbi->s_inode_lock);
2187         list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2188                 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2189                     test_bit(NILFS_I_DIRTY, &ii->i_state)) {
2190                         /* The current checkpoint number (=nilfs->ns_cno) is
2191                            changed between check-in and check-out only if the
2192                            super root is written out.  So, we can update i_cno
2193                            for the inodes that remain in the dirty list. */
2194                         ii->i_cno = cno;
2195                         continue;
2196                 }
2197                 clear_bit(NILFS_I_BUSY, &ii->i_state);
2198                 brelse(ii->i_bh);
2199                 ii->i_bh = NULL;
2200                 list_del(&ii->i_dirty);
2201                 list_add_tail(&ii->i_dirty, &ti->ti_garbage);
2202         }
2203         spin_unlock(&sbi->s_inode_lock);
2204 }
2205
2206 /*
2207  * Nasty routines to manipulate active flags on sufile.
2208  * These would be removed in a future release.
2209  */
2210 static void nilfs_segctor_reactivate_segments(struct nilfs_sc_info *sci,
2211                                               struct the_nilfs *nilfs)
2212 {
2213         struct nilfs_segment_buffer *segbuf, *last;
2214         struct nilfs_segment_entry *ent, *n;
2215         struct inode *sufile = nilfs->ns_sufile;
2216         struct list_head *head;
2217
2218         last = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2219         nilfs_for_each_segbuf_before(segbuf, last, &sci->sc_segbufs) {
2220                 ent = segbuf->sb_segent;
2221                 if (!ent)
2222                         break; /* ignore unmapped segments (should check it?)*/
2223                 nilfs_segment_usage_set_active(ent->raw_su);
2224                 nilfs_close_segment_entry(ent, sufile);
2225         }
2226
2227         head = &sci->sc_active_segments;
2228         list_for_each_entry_safe(ent, n, head, list) {
2229                 nilfs_segment_usage_set_active(ent->raw_su);
2230                 nilfs_close_segment_entry(ent, sufile);
2231         }
2232 }
2233
2234 static int nilfs_segctor_deactivate_segments(struct nilfs_sc_info *sci,
2235                                              struct the_nilfs *nilfs)
2236 {
2237         struct nilfs_segment_buffer *segbuf, *last;
2238         struct nilfs_segment_entry *ent;
2239         struct inode *sufile = nilfs->ns_sufile;
2240         int err;
2241
2242         last = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2243         nilfs_for_each_segbuf_before(segbuf, last, &sci->sc_segbufs) {
2244                 /*
2245                  * Deactivate ongoing full segments.  The last segment is kept
2246                  * active because it is a start point of recovery, and is not
2247                  * relocatable until the super block points to a newer
2248                  * checkpoint.
2249                  */
2250                 ent = segbuf->sb_segent;
2251                 if (!ent)
2252                         break; /* ignore unmapped segments (should check it?)*/
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                 BUG_ON(!buffer_dirty(ent->bh_su));
2258         }
2259
2260         list_for_each_entry(ent, &sci->sc_active_segments, list) {
2261                 err = nilfs_open_segment_entry(ent, sufile);
2262                 if (unlikely(err))
2263                         goto failed;
2264                 nilfs_segment_usage_clear_active(ent->raw_su);
2265                 BUG_ON(!buffer_dirty(ent->bh_su));
2266         }
2267         return 0;
2268
2269  failed:
2270         nilfs_segctor_reactivate_segments(sci, nilfs);
2271         return err;
2272 }
2273
2274 static void nilfs_segctor_bead_completed_segments(struct nilfs_sc_info *sci)
2275 {
2276         struct nilfs_segment_buffer *segbuf, *last;
2277         struct nilfs_segment_entry *ent;
2278
2279         /* move each segbuf->sb_segent to the list of used active segments */
2280         last = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2281         nilfs_for_each_segbuf_before(segbuf, last, &sci->sc_segbufs) {
2282                 ent = segbuf->sb_segent;
2283                 if (!ent)
2284                         break; /* ignore unmapped segments (should check it?)*/
2285                 list_add_tail(&ent->list, &sci->sc_active_segments);
2286                 segbuf->sb_segent = NULL;
2287         }
2288 }
2289
2290 static void nilfs_segctor_commit_deactivate_segments(struct nilfs_sc_info *sci,
2291                                                      struct the_nilfs *nilfs)
2292 {
2293         struct nilfs_segment_entry *ent, *n;
2294
2295         list_for_each_entry_safe(ent, n, &sci->sc_active_segments, list) {
2296                 list_del(&ent->list);
2297                 nilfs_close_segment_entry(ent, nilfs->ns_sufile);
2298                 nilfs_free_segment_entry(ent);
2299         }
2300 }
2301
2302 /*
2303  * Main procedure of segment constructor
2304  */
2305 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2306 {
2307         struct nilfs_sb_info *sbi = sci->sc_sbi;
2308         struct the_nilfs *nilfs = sbi->s_nilfs;
2309         struct page *failed_page;
2310         int err, has_sr = 0;
2311
2312         sci->sc_stage.scnt = NILFS_ST_INIT;
2313
2314         err = nilfs_segctor_check_in_files(sci, sbi);
2315         if (unlikely(err))
2316                 goto out;
2317
2318         if (nilfs_test_metadata_dirty(sbi))
2319                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2320
2321         if (nilfs_segctor_clean(sci))
2322                 goto out;
2323
2324         do {
2325                 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2326
2327                 err = nilfs_segctor_begin_construction(sci, nilfs);
2328                 if (unlikely(err))
2329                         goto out;
2330
2331                 /* Update time stamp */
2332                 sci->sc_seg_ctime = get_seconds();
2333
2334                 err = nilfs_segctor_collect(sci, nilfs, mode);
2335                 if (unlikely(err))
2336                         goto failed;
2337
2338                 has_sr = (sci->sc_super_root != NULL);
2339
2340                 /* Avoid empty segment */
2341                 if (sci->sc_stage.scnt == NILFS_ST_DONE &&
2342                     NILFS_SEG_EMPTY(&sci->sc_curseg->sb_sum)) {
2343                         BUG_ON(mode == SC_LSEG_SR);
2344                         nilfs_segctor_end_construction(sci, nilfs, 1);
2345                         goto out;
2346                 }
2347
2348                 err = nilfs_segctor_assign(sci, mode);
2349                 if (unlikely(err))
2350                         goto failed;
2351
2352                 if (has_sr) {
2353                         err = nilfs_segctor_deactivate_segments(sci, nilfs);
2354                         if (unlikely(err))
2355                                 goto failed;
2356                 }
2357                 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2358                         nilfs_segctor_fill_in_file_bmap(sci, sbi->s_ifile);
2359
2360                 if (has_sr) {
2361                         err = nilfs_segctor_fill_in_checkpoint(sci);
2362                         if (unlikely(err))
2363                                 goto failed_to_make_up;
2364
2365                         nilfs_segctor_fill_in_super_root(sci, nilfs);
2366                 }
2367                 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2368
2369                 /* Write partial segments */
2370                 err = nilfs_segctor_prepare_write(sci, &failed_page);
2371                 if (unlikely(err))
2372                         goto failed_to_write;
2373
2374                 nilfs_segctor_fill_in_checksums(sci, nilfs->ns_crc_seed);
2375
2376                 err = nilfs_segctor_write(sci, nilfs->ns_bdi);
2377                 if (unlikely(err))
2378                         goto failed_to_write;
2379
2380                 nilfs_segctor_complete_write(sci);
2381
2382                 /* Commit segments */
2383                 nilfs_segctor_bead_completed_segments(sci);
2384                 if (has_sr) {
2385                         down_write(&nilfs->ns_sem);
2386                         nilfs_update_last_segment(sbi, 1);
2387                         up_write(&nilfs->ns_sem);
2388                         nilfs_segctor_commit_deactivate_segments(sci, nilfs);
2389                         nilfs_segctor_commit_free_segments(sci);
2390                         nilfs_segctor_clear_metadata_dirty(sci);
2391                 }
2392
2393                 nilfs_segctor_end_construction(sci, nilfs, 0);
2394
2395         } while (sci->sc_stage.scnt != NILFS_ST_DONE);
2396
2397         /* Clearing sketch data */
2398         if (has_sr && sci->sc_sketch_inode) {
2399                 if (i_size_read(sci->sc_sketch_inode) == 0)
2400                         clear_bit(NILFS_I_DIRTY,
2401                                   &NILFS_I(sci->sc_sketch_inode)->i_state);
2402                 i_size_write(sci->sc_sketch_inode, 0);
2403         }
2404  out:
2405         nilfs_segctor_destroy_segment_buffers(sci);
2406         nilfs_segctor_check_out_files(sci, sbi);
2407         return err;
2408
2409  failed_to_write:
2410         nilfs_segctor_abort_write(sci, failed_page, err);
2411         nilfs_segctor_cancel_segusage(sci, nilfs->ns_sufile);
2412
2413  failed_to_make_up:
2414         if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2415                 nilfs_redirty_inodes(&sci->sc_dirty_files);
2416         if (has_sr)
2417                 nilfs_segctor_reactivate_segments(sci, nilfs);
2418
2419  failed:
2420         if (nilfs_doing_gc())
2421                 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2422         nilfs_segctor_end_construction(sci, nilfs, err);
2423         goto out;
2424 }
2425
2426 /**
2427  * nilfs_secgtor_start_timer - set timer of background write
2428  * @sci: nilfs_sc_info
2429  *
2430  * If the timer has already been set, it ignores the new request.
2431  * This function MUST be called within a section locking the segment
2432  * semaphore.
2433  */
2434 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2435 {
2436         spin_lock(&sci->sc_state_lock);
2437         if (sci->sc_timer && !(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2438                 sci->sc_timer->expires = jiffies + sci->sc_interval;
2439                 add_timer(sci->sc_timer);
2440                 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2441         }
2442         spin_unlock(&sci->sc_state_lock);
2443 }
2444
2445 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2446 {
2447         spin_lock(&sci->sc_state_lock);
2448         if (!(sci->sc_flush_request & (1 << bn))) {
2449                 unsigned long prev_req = sci->sc_flush_request;
2450
2451                 sci->sc_flush_request |= (1 << bn);
2452                 if (!prev_req)
2453                         wake_up(&sci->sc_wait_daemon);
2454         }
2455         spin_unlock(&sci->sc_state_lock);
2456 }
2457
2458 /**
2459  * nilfs_flush_segment - trigger a segment construction for resource control
2460  * @sb: super block
2461  * @ino: inode number of the file to be flushed out.
2462  */
2463 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2464 {
2465         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2466         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2467
2468         if (!sci || nilfs_doing_construction())
2469                 return;
2470         nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2471                                         /* assign bit 0 to data files */
2472 }
2473
2474 int nilfs_segctor_add_segments_to_be_freed(struct nilfs_sc_info *sci,
2475                                            __u64 *segnum, size_t nsegs)
2476 {
2477         struct nilfs_segment_entry *ent;
2478         struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
2479         struct inode *sufile = nilfs->ns_sufile;
2480         LIST_HEAD(list);
2481         __u64 *pnum;
2482         const char *flag_name;
2483         size_t i;
2484         int err, err2 = 0;
2485
2486         for (pnum = segnum, i = 0; i < nsegs; pnum++, i++) {
2487                 ent = nilfs_alloc_segment_entry(*pnum);
2488                 if (unlikely(!ent)) {
2489                         err = -ENOMEM;
2490                         goto failed;
2491                 }
2492                 list_add_tail(&ent->list, &list);
2493
2494                 err = nilfs_open_segment_entry(ent, sufile);
2495                 if (unlikely(err))
2496                         goto failed;
2497
2498                 if (unlikely(le32_to_cpu(ent->raw_su->su_flags) !=
2499                              (1UL << NILFS_SEGMENT_USAGE_DIRTY))) {
2500                         if (nilfs_segment_usage_clean(ent->raw_su))
2501                                 flag_name = "clean";
2502                         else if (nilfs_segment_usage_active(ent->raw_su))
2503                                 flag_name = "active";
2504                         else if (nilfs_segment_usage_volatile_active(
2505                                          ent->raw_su))
2506                                 flag_name = "volatile active";
2507                         else if (!nilfs_segment_usage_dirty(ent->raw_su))
2508                                 flag_name = "non-dirty";
2509                         else
2510                                 flag_name = "erroneous";
2511
2512                         printk(KERN_ERR
2513                                "NILFS: %s segment is requested to be cleaned "
2514                                "(segnum=%llu)\n",
2515                                flag_name, (unsigned long long)ent->segnum);
2516                         err2 = -EINVAL;
2517                 }
2518                 nilfs_close_segment_entry(ent, sufile);
2519         }
2520         if (unlikely(err2)) {
2521                 err = err2;
2522                 goto failed;
2523         }
2524         list_splice(&list, sci->sc_cleaning_segments.prev);
2525         return 0;
2526
2527  failed:
2528         nilfs_dispose_segment_list(&list);
2529         return err;
2530 }
2531
2532 void nilfs_segctor_clear_segments_to_be_freed(struct nilfs_sc_info *sci)
2533 {
2534         nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
2535 }
2536
2537 struct nilfs_segctor_wait_request {
2538         wait_queue_t    wq;
2539         __u32           seq;
2540         int             err;
2541         atomic_t        done;
2542 };
2543
2544 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2545 {
2546         struct nilfs_segctor_wait_request wait_req;
2547         int err = 0;
2548
2549         spin_lock(&sci->sc_state_lock);
2550         init_wait(&wait_req.wq);
2551         wait_req.err = 0;
2552         atomic_set(&wait_req.done, 0);
2553         wait_req.seq = ++sci->sc_seq_request;
2554         spin_unlock(&sci->sc_state_lock);
2555
2556         init_waitqueue_entry(&wait_req.wq, current);
2557         add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2558         set_current_state(TASK_INTERRUPTIBLE);
2559         wake_up(&sci->sc_wait_daemon);
2560
2561         for (;;) {
2562                 if (atomic_read(&wait_req.done)) {
2563                         err = wait_req.err;
2564                         break;
2565                 }
2566                 if (!signal_pending(current)) {
2567                         schedule();
2568                         continue;
2569                 }
2570                 err = -ERESTARTSYS;
2571                 break;
2572         }
2573         finish_wait(&sci->sc_wait_request, &wait_req.wq);
2574         return err;
2575 }
2576
2577 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2578 {
2579         struct nilfs_segctor_wait_request *wrq, *n;
2580         unsigned long flags;
2581
2582         spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2583         list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2584                                  wq.task_list) {
2585                 if (!atomic_read(&wrq->done) &&
2586                     nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2587                         wrq->err = err;
2588                         atomic_set(&wrq->done, 1);
2589                 }
2590                 if (atomic_read(&wrq->done)) {
2591                         wrq->wq.func(&wrq->wq,
2592                                      TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2593                                      0, NULL);
2594                 }
2595         }
2596         spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2597 }
2598
2599 /**
2600  * nilfs_construct_segment - construct a logical segment
2601  * @sb: super block
2602  *
2603  * Return Value: On success, 0 is retured. On errors, one of the following
2604  * negative error code is returned.
2605  *
2606  * %-EROFS - Read only filesystem.
2607  *
2608  * %-EIO - I/O error
2609  *
2610  * %-ENOSPC - No space left on device (only in a panic state).
2611  *
2612  * %-ERESTARTSYS - Interrupted.
2613  *
2614  * %-ENOMEM - Insufficient memory available.
2615  */
2616 int nilfs_construct_segment(struct super_block *sb)
2617 {
2618         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2619         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2620         struct nilfs_transaction_info *ti;
2621         int err;
2622
2623         if (!sci)
2624                 return -EROFS;
2625
2626         /* A call inside transactions causes a deadlock. */
2627         BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2628
2629         err = nilfs_segctor_sync(sci);
2630         return err;
2631 }
2632
2633 /**
2634  * nilfs_construct_dsync_segment - construct a data-only logical segment
2635  * @sb: super block
2636  * @inode: inode whose data blocks should be written out
2637  * @start: start byte offset
2638  * @end: end byte offset (inclusive)
2639  *
2640  * Return Value: On success, 0 is retured. On errors, one of the following
2641  * negative error code is returned.
2642  *
2643  * %-EROFS - Read only filesystem.
2644  *
2645  * %-EIO - I/O error
2646  *
2647  * %-ENOSPC - No space left on device (only in a panic state).
2648  *
2649  * %-ERESTARTSYS - Interrupted.
2650  *
2651  * %-ENOMEM - Insufficient memory available.
2652  */
2653 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2654                                   loff_t start, loff_t end)
2655 {
2656         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2657         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2658         struct nilfs_inode_info *ii;
2659         struct nilfs_transaction_info ti;
2660         int err = 0;
2661
2662         if (!sci)
2663                 return -EROFS;
2664
2665         nilfs_transaction_lock(sbi, &ti, 0);
2666
2667         ii = NILFS_I(inode);
2668         if (test_bit(NILFS_I_INODE_DIRTY, &ii->i_state) ||
2669             nilfs_test_opt(sbi, STRICT_ORDER) ||
2670             test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2671             nilfs_discontinued(sbi->s_nilfs)) {
2672                 nilfs_transaction_unlock(sbi);
2673                 err = nilfs_segctor_sync(sci);
2674                 return err;
2675         }
2676
2677         spin_lock(&sbi->s_inode_lock);
2678         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2679             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2680                 spin_unlock(&sbi->s_inode_lock);
2681                 nilfs_transaction_unlock(sbi);
2682                 return 0;
2683         }
2684         spin_unlock(&sbi->s_inode_lock);
2685         sci->sc_dsync_inode = ii;
2686         sci->sc_dsync_start = start;
2687         sci->sc_dsync_end = end;
2688
2689         err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2690
2691         nilfs_transaction_unlock(sbi);
2692         return err;
2693 }
2694
2695 struct nilfs_segctor_req {
2696         int mode;
2697         __u32 seq_accepted;
2698         int sc_err;  /* construction failure */
2699         int sb_err;  /* super block writeback failure */
2700 };
2701
2702 #define FLUSH_FILE_BIT  (0x1) /* data file only */
2703 #define FLUSH_DAT_BIT   (1 << NILFS_DAT_INO) /* DAT only */
2704
2705 static void nilfs_segctor_accept(struct nilfs_sc_info *sci,
2706                                  struct nilfs_segctor_req *req)
2707 {
2708         BUG_ON(!sci);
2709
2710         req->sc_err = req->sb_err = 0;
2711         spin_lock(&sci->sc_state_lock);
2712         req->seq_accepted = sci->sc_seq_request;
2713         spin_unlock(&sci->sc_state_lock);
2714
2715         if (sci->sc_timer)
2716                 del_timer_sync(sci->sc_timer);
2717 }
2718
2719 static void nilfs_segctor_notify(struct nilfs_sc_info *sci,
2720                                  struct nilfs_segctor_req *req)
2721 {
2722         /* Clear requests (even when the construction failed) */
2723         spin_lock(&sci->sc_state_lock);
2724
2725         sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2726
2727         if (req->mode == SC_LSEG_SR) {
2728                 sci->sc_seq_done = req->seq_accepted;
2729                 nilfs_segctor_wakeup(sci, req->sc_err ? : req->sb_err);
2730                 sci->sc_flush_request = 0;
2731         } else if (req->mode == SC_FLUSH_FILE)
2732                 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2733         else if (req->mode == SC_FLUSH_DAT)
2734                 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2735
2736         spin_unlock(&sci->sc_state_lock);
2737 }
2738
2739 static int nilfs_segctor_construct(struct nilfs_sc_info *sci,
2740                                    struct nilfs_segctor_req *req)
2741 {
2742         struct nilfs_sb_info *sbi = sci->sc_sbi;
2743         struct the_nilfs *nilfs = sbi->s_nilfs;
2744         int err = 0;
2745
2746         if (nilfs_discontinued(nilfs))
2747                 req->mode = SC_LSEG_SR;
2748         if (!nilfs_segctor_confirm(sci)) {
2749                 err = nilfs_segctor_do_construct(sci, req->mode);
2750                 req->sc_err = err;
2751         }
2752         if (likely(!err)) {
2753                 if (req->mode != SC_FLUSH_DAT)
2754                         atomic_set(&nilfs->ns_ndirtyblks, 0);
2755                 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2756                     nilfs_discontinued(nilfs)) {
2757                         down_write(&nilfs->ns_sem);
2758                         req->sb_err = nilfs_commit_super(sbi);
2759                         up_write(&nilfs->ns_sem);
2760                 }
2761         }
2762         return err;
2763 }
2764
2765 static void nilfs_construction_timeout(unsigned long data)
2766 {
2767         struct task_struct *p = (struct task_struct *)data;
2768         wake_up_process(p);
2769 }
2770
2771 static void
2772 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2773 {
2774         struct nilfs_inode_info *ii, *n;
2775
2776         list_for_each_entry_safe(ii, n, head, i_dirty) {
2777                 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2778                         continue;
2779                 hlist_del_init(&ii->vfs_inode.i_hash);
2780                 list_del_init(&ii->i_dirty);
2781                 nilfs_clear_gcinode(&ii->vfs_inode);
2782         }
2783 }
2784
2785 int nilfs_clean_segments(struct super_block *sb, void __user *argp)
2786 {
2787         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2788         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2789         struct the_nilfs *nilfs = sbi->s_nilfs;
2790         struct nilfs_transaction_info ti;
2791         struct nilfs_segctor_req req = { .mode = SC_LSEG_SR };
2792         int err;
2793
2794         if (unlikely(!sci))
2795                 return -EROFS;
2796
2797         nilfs_transaction_lock(sbi, &ti, 1);
2798
2799         err = nilfs_init_gcdat_inode(nilfs);
2800         if (unlikely(err))
2801                 goto out_unlock;
2802         err = nilfs_ioctl_prepare_clean_segments(nilfs, argp);
2803         if (unlikely(err))
2804                 goto out_unlock;
2805
2806         list_splice_init(&nilfs->ns_gc_inodes, sci->sc_gc_inodes.prev);
2807
2808         for (;;) {
2809                 nilfs_segctor_accept(sci, &req);
2810                 err = nilfs_segctor_construct(sci, &req);
2811                 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2812                 nilfs_segctor_notify(sci, &req);
2813
2814                 if (likely(!err))
2815                         break;
2816
2817                 nilfs_warning(sb, __func__,
2818                               "segment construction failed. (err=%d)", err);
2819                 set_current_state(TASK_INTERRUPTIBLE);
2820                 schedule_timeout(sci->sc_interval);
2821         }
2822
2823  out_unlock:
2824         nilfs_clear_gcdat_inode(nilfs);
2825         nilfs_transaction_unlock(sbi);
2826         return err;
2827 }
2828
2829 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2830 {
2831         struct nilfs_sb_info *sbi = sci->sc_sbi;
2832         struct nilfs_transaction_info ti;
2833         struct nilfs_segctor_req req = { .mode = mode };
2834
2835         nilfs_transaction_lock(sbi, &ti, 0);
2836
2837         nilfs_segctor_accept(sci, &req);
2838         nilfs_segctor_construct(sci, &req);
2839         nilfs_segctor_notify(sci, &req);
2840
2841         /*
2842          * Unclosed segment should be retried.  We do this using sc_timer.
2843          * Timeout of sc_timer will invoke complete construction which leads
2844          * to close the current logical segment.
2845          */
2846         if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2847                 nilfs_segctor_start_timer(sci);
2848
2849         nilfs_transaction_unlock(sbi);
2850 }
2851
2852 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2853 {
2854         int mode = 0;
2855         int err;
2856
2857         spin_lock(&sci->sc_state_lock);
2858         mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2859                 SC_FLUSH_DAT : SC_FLUSH_FILE;
2860         spin_unlock(&sci->sc_state_lock);
2861
2862         if (mode) {
2863                 err = nilfs_segctor_do_construct(sci, mode);
2864
2865                 spin_lock(&sci->sc_state_lock);
2866                 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2867                         ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2868                 spin_unlock(&sci->sc_state_lock);
2869         }
2870         clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2871 }
2872
2873 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2874 {
2875         if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2876             time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2877                 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2878                         return SC_FLUSH_FILE;
2879                 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2880                         return SC_FLUSH_DAT;
2881         }
2882         return SC_LSEG_SR;
2883 }
2884
2885 /**
2886  * nilfs_segctor_thread - main loop of the segment constructor thread.
2887  * @arg: pointer to a struct nilfs_sc_info.
2888  *
2889  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2890  * to execute segment constructions.
2891  */
2892 static int nilfs_segctor_thread(void *arg)
2893 {
2894         struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2895         struct timer_list timer;
2896         int timeout = 0;
2897
2898         init_timer(&timer);
2899         timer.data = (unsigned long)current;
2900         timer.function = nilfs_construction_timeout;
2901         sci->sc_timer = &timer;
2902
2903         /* start sync. */
2904         sci->sc_task = current;
2905         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2906         printk(KERN_INFO
2907                "segctord starting. Construction interval = %lu seconds, "
2908                "CP frequency < %lu seconds\n",
2909                sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2910
2911         spin_lock(&sci->sc_state_lock);
2912  loop:
2913         for (;;) {
2914                 int mode;
2915
2916                 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2917                         goto end_thread;
2918
2919                 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2920                         mode = SC_LSEG_SR;
2921                 else if (!sci->sc_flush_request)
2922                         break;
2923                 else
2924                         mode = nilfs_segctor_flush_mode(sci);
2925
2926                 spin_unlock(&sci->sc_state_lock);
2927                 nilfs_segctor_thread_construct(sci, mode);
2928                 spin_lock(&sci->sc_state_lock);
2929                 timeout = 0;
2930         }
2931
2932
2933         if (freezing(current)) {
2934                 spin_unlock(&sci->sc_state_lock);
2935                 refrigerator();
2936                 spin_lock(&sci->sc_state_lock);
2937         } else {
2938                 DEFINE_WAIT(wait);
2939                 int should_sleep = 1;
2940
2941                 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2942                                 TASK_INTERRUPTIBLE);
2943
2944                 if (sci->sc_seq_request != sci->sc_seq_done)
2945                         should_sleep = 0;
2946                 else if (sci->sc_flush_request)
2947                         should_sleep = 0;
2948                 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2949                         should_sleep = time_before(jiffies,
2950                                                    sci->sc_timer->expires);
2951
2952                 if (should_sleep) {
2953                         spin_unlock(&sci->sc_state_lock);
2954                         schedule();
2955                         spin_lock(&sci->sc_state_lock);
2956                 }
2957                 finish_wait(&sci->sc_wait_daemon, &wait);
2958                 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2959                            time_after_eq(jiffies, sci->sc_timer->expires));
2960         }
2961         goto loop;
2962
2963  end_thread:
2964         spin_unlock(&sci->sc_state_lock);
2965         del_timer_sync(sci->sc_timer);
2966         sci->sc_timer = NULL;
2967
2968         /* end sync. */
2969         sci->sc_task = NULL;
2970         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2971         return 0;
2972 }
2973
2974 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2975 {
2976         struct task_struct *t;
2977
2978         t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2979         if (IS_ERR(t)) {
2980                 int err = PTR_ERR(t);
2981
2982                 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2983                        err);
2984                 return err;
2985         }
2986         wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2987         return 0;
2988 }
2989
2990 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2991 {
2992         sci->sc_state |= NILFS_SEGCTOR_QUIT;
2993
2994         while (sci->sc_task) {
2995                 wake_up(&sci->sc_wait_daemon);
2996                 spin_unlock(&sci->sc_state_lock);
2997                 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2998                 spin_lock(&sci->sc_state_lock);
2999         }
3000 }
3001
3002 static int nilfs_segctor_init(struct nilfs_sc_info *sci,
3003                               struct nilfs_recovery_info *ri)
3004 {
3005         int err;
3006         struct inode *inode = nilfs_iget(sci->sc_super, NILFS_SKETCH_INO);
3007
3008         sci->sc_sketch_inode = IS_ERR(inode) ? NULL : inode;
3009         if (sci->sc_sketch_inode)
3010                 i_size_write(sci->sc_sketch_inode, 0);
3011
3012         sci->sc_seq_done = sci->sc_seq_request;
3013         if (ri)
3014                 list_splice_init(&ri->ri_used_segments,
3015                                  sci->sc_active_segments.prev);
3016
3017         err = nilfs_segctor_start_thread(sci);
3018         if (err) {
3019                 if (ri)
3020                         list_splice_init(&sci->sc_active_segments,
3021                                          ri->ri_used_segments.prev);
3022                 if (sci->sc_sketch_inode) {
3023                         iput(sci->sc_sketch_inode);
3024                         sci->sc_sketch_inode = NULL;
3025                 }
3026         }
3027         return err;
3028 }
3029
3030 /*
3031  * Setup & clean-up functions
3032  */
3033 static struct nilfs_sc_info *nilfs_segctor_new(struct nilfs_sb_info *sbi)
3034 {
3035         struct nilfs_sc_info *sci;
3036
3037         sci = kzalloc(sizeof(*sci), GFP_KERNEL);
3038         if (!sci)
3039                 return NULL;
3040
3041         sci->sc_sbi = sbi;
3042         sci->sc_super = sbi->s_super;
3043
3044         init_waitqueue_head(&sci->sc_wait_request);
3045         init_waitqueue_head(&sci->sc_wait_daemon);
3046         init_waitqueue_head(&sci->sc_wait_task);
3047         spin_lock_init(&sci->sc_state_lock);
3048         INIT_LIST_HEAD(&sci->sc_dirty_files);
3049         INIT_LIST_HEAD(&sci->sc_segbufs);
3050         INIT_LIST_HEAD(&sci->sc_gc_inodes);
3051         INIT_LIST_HEAD(&sci->sc_active_segments);
3052         INIT_LIST_HEAD(&sci->sc_cleaning_segments);
3053         INIT_LIST_HEAD(&sci->sc_copied_buffers);
3054
3055         sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
3056         sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
3057         sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
3058
3059         if (sbi->s_interval)
3060                 sci->sc_interval = sbi->s_interval;
3061         if (sbi->s_watermark)
3062                 sci->sc_watermark = sbi->s_watermark;
3063         return sci;
3064 }
3065
3066 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
3067 {
3068         int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
3069
3070         /* The segctord thread was stopped and its timer was removed.
3071            But some tasks remain. */
3072         do {
3073                 struct nilfs_sb_info *sbi = sci->sc_sbi;
3074                 struct nilfs_transaction_info ti;
3075                 struct nilfs_segctor_req req = { .mode = SC_LSEG_SR };
3076
3077                 nilfs_transaction_lock(sbi, &ti, 0);
3078                 nilfs_segctor_accept(sci, &req);
3079                 ret = nilfs_segctor_construct(sci, &req);
3080                 nilfs_segctor_notify(sci, &req);
3081                 nilfs_transaction_unlock(sbi);
3082
3083         } while (ret && retrycount-- > 0);
3084 }
3085
3086 /**
3087  * nilfs_segctor_destroy - destroy the segment constructor.
3088  * @sci: nilfs_sc_info
3089  *
3090  * nilfs_segctor_destroy() kills the segctord thread and frees
3091  * the nilfs_sc_info struct.
3092  * Caller must hold the segment semaphore.
3093  */
3094 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
3095 {
3096         struct nilfs_sb_info *sbi = sci->sc_sbi;
3097         int flag;
3098
3099         up_write(&sbi->s_nilfs->ns_segctor_sem);
3100
3101         spin_lock(&sci->sc_state_lock);
3102         nilfs_segctor_kill_thread(sci);
3103         flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
3104                 || sci->sc_seq_request != sci->sc_seq_done);
3105         spin_unlock(&sci->sc_state_lock);
3106
3107         if (flag || nilfs_segctor_confirm(sci))
3108                 nilfs_segctor_write_out(sci);
3109
3110         BUG_ON(!list_empty(&sci->sc_copied_buffers));
3111
3112         if (!list_empty(&sci->sc_dirty_files)) {
3113                 nilfs_warning(sbi->s_super, __func__,
3114                               "dirty file(s) after the final construction\n");
3115                 nilfs_dispose_list(sbi, &sci->sc_dirty_files, 1);
3116         }
3117         if (!list_empty(&sci->sc_active_segments))
3118                 nilfs_dispose_segment_list(&sci->sc_active_segments);
3119
3120         if (!list_empty(&sci->sc_cleaning_segments))
3121                 nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
3122
3123         BUG_ON(!list_empty(&sci->sc_segbufs));
3124
3125         if (sci->sc_sketch_inode) {
3126                 iput(sci->sc_sketch_inode);
3127                 sci->sc_sketch_inode = NULL;
3128         }
3129         down_write(&sbi->s_nilfs->ns_segctor_sem);
3130
3131         kfree(sci);
3132 }
3133
3134 /**
3135  * nilfs_attach_segment_constructor - attach a segment constructor
3136  * @sbi: nilfs_sb_info
3137  * @ri: nilfs_recovery_info
3138  *
3139  * nilfs_attach_segment_constructor() allocates a struct nilfs_sc_info,
3140  * initilizes it, and starts the segment constructor.
3141  *
3142  * Return Value: On success, 0 is returned. On error, one of the following
3143  * negative error code is returned.
3144  *
3145  * %-ENOMEM - Insufficient memory available.
3146  */
3147 int nilfs_attach_segment_constructor(struct nilfs_sb_info *sbi,
3148                                      struct nilfs_recovery_info *ri)
3149 {
3150         struct the_nilfs *nilfs = sbi->s_nilfs;
3151         int err;
3152
3153         /* Each field of nilfs_segctor is cleared through the initialization
3154            of super-block info */
3155         sbi->s_sc_info = nilfs_segctor_new(sbi);
3156         if (!sbi->s_sc_info)
3157                 return -ENOMEM;
3158
3159         nilfs_attach_writer(nilfs, sbi);
3160         err = nilfs_segctor_init(NILFS_SC(sbi), ri);
3161         if (err) {
3162                 nilfs_detach_writer(nilfs, sbi);
3163                 kfree(sbi->s_sc_info);
3164                 sbi->s_sc_info = NULL;
3165         }
3166         return err;
3167 }
3168
3169 /**
3170  * nilfs_detach_segment_constructor - destroy the segment constructor
3171  * @sbi: nilfs_sb_info
3172  *
3173  * nilfs_detach_segment_constructor() kills the segment constructor daemon,
3174  * frees the struct nilfs_sc_info, and destroy the dirty file list.
3175  */
3176 void nilfs_detach_segment_constructor(struct nilfs_sb_info *sbi)
3177 {
3178         struct the_nilfs *nilfs = sbi->s_nilfs;
3179         LIST_HEAD(garbage_list);
3180
3181         down_write(&nilfs->ns_segctor_sem);
3182         if (NILFS_SC(sbi)) {
3183                 nilfs_segctor_destroy(NILFS_SC(sbi));
3184                 sbi->s_sc_info = NULL;
3185         }
3186
3187         /* Force to free the list of dirty files */
3188         spin_lock(&sbi->s_inode_lock);
3189         if (!list_empty(&sbi->s_dirty_files)) {
3190                 list_splice_init(&sbi->s_dirty_files, &garbage_list);
3191                 nilfs_warning(sbi->s_super, __func__,
3192                               "Non empty dirty list after the last "
3193                               "segment construction\n");
3194         }
3195         spin_unlock(&sbi->s_inode_lock);
3196         up_write(&nilfs->ns_segctor_sem);
3197
3198         nilfs_dispose_list(sbi, &garbage_list, 1);
3199         nilfs_detach_writer(nilfs, sbi);
3200 }