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                          * we save it and restore on nilfs_transaction_end().
 167                          * But this should never happen.
 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_end() 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  * %-ERESTARTSYS - Interrupted
 216  *
 217  * %-ENOSPC - No space left on device
 218  */
 219 int nilfs_transaction_begin(struct super_block *sb,
 220                             struct nilfs_transaction_info *ti,
 221                             int vacancy_check)
 222 {
 223         struct nilfs_sb_info *sbi;
 224         struct the_nilfs *nilfs;
 225         int ret = nilfs_prepare_segment_lock(ti);
 226
 227         if (unlikely(ret < 0))
 228                 return ret;
 229         if (ret > 0)
 230                 return 0;
 231
 232         sbi = NILFS_SB(sb);
 233         nilfs = sbi->s_nilfs;
 234         down_read(&nilfs->ns_segctor_sem);
 235         if (vacancy_check && nilfs_near_disk_full(nilfs)) {
 236                 up_read(&nilfs->ns_segctor_sem);
 237                 ret = -ENOSPC;
 238                 goto failed;
 239         }
 240         return 0;
 241
 242  failed:
 243         ti = current->journal_info;
 244         current->journal_info = ti->ti_save;
 245         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 246                 kmem_cache_free(nilfs_transaction_cachep, ti);
 247         return ret;
 248 }
 249
 250 /**
 251  * nilfs_transaction_end - end indivisible file operations.
 252  * @sb: super block
 253  * @commit: commit flag (0 for no change)
 254  *
 255  * nilfs_transaction_end() releases the read semaphore which is
 256  * acquired by nilfs_transaction_begin(). Its releasing is only done
 257  * in outermost call of this function. If the nilfs_transaction_info
 258  * was allocated dynamically, it is given back to a slab cache.
 259  */
 260 int nilfs_transaction_end(struct super_block *sb, int commit)
 261 {
 262         struct nilfs_transaction_info *ti = current->journal_info;
 263         struct nilfs_sb_info *sbi;
 264         struct nilfs_sc_info *sci;
 265         int err = 0;
 266
 267         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 268
 269         if (commit)
 270                 ti->ti_flags |= NILFS_TI_COMMIT;
 271         if (ti->ti_count > 0) {
 272                 ti->ti_count--;
 273                 return 0;
 274         }
 275         sbi = NILFS_SB(sb);
 276         sci = NILFS_SC(sbi);
 277         if (sci != NULL) {
 278                 if (ti->ti_flags & NILFS_TI_COMMIT)
 279                         nilfs_segctor_start_timer(sci);
 280                 if (atomic_read(&sbi->s_nilfs->ns_ndirtyblks) >
 281                     sci->sc_watermark)
 282                         nilfs_segctor_do_flush(sci, 0);
 283         }
 284         up_read(&sbi->s_nilfs->ns_segctor_sem);
 285         current->journal_info = ti->ti_save;
 286
 287         if (ti->ti_flags & NILFS_TI_SYNC)
 288                 err = nilfs_construct_segment(sb);
 289         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 290                 kmem_cache_free(nilfs_transaction_cachep, ti);
 291         return err;
 292 }
 293
 294 void nilfs_relax_pressure_in_lock(struct super_block *sb)
 295 {
 296         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 297         struct nilfs_sc_info *sci = NILFS_SC(sbi);
 298         struct the_nilfs *nilfs = sbi->s_nilfs;
 299
 300         if (!sci || !sci->sc_flush_request)
 301                 return;
 302
 303         set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
 304         up_read(&nilfs->ns_segctor_sem);
 305
 306         down_write(&nilfs->ns_segctor_sem);
 307         if (sci->sc_flush_request &&
 308             test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
 309                 struct nilfs_transaction_info *ti = current->journal_info;
 310
 311                 ti->ti_flags |= NILFS_TI_WRITER;
 312                 nilfs_segctor_do_immediate_flush(sci);
 313                 ti->ti_flags &= ~NILFS_TI_WRITER;
 314         }
 315         downgrade_write(&nilfs->ns_segctor_sem);
 316 }
 317
 318 static void nilfs_transaction_lock(struct nilfs_sb_info *sbi,
 319                                    struct nilfs_transaction_info *ti,
 320                                    int gcflag)
 321 {
 322         struct nilfs_transaction_info *cur_ti = current->journal_info;
 323
 324         BUG_ON(cur_ti);
 325         BUG_ON(!ti);
 326         ti->ti_flags = NILFS_TI_WRITER;
 327         ti->ti_count = 0;
 328         ti->ti_save = cur_ti;
 329         ti->ti_magic = NILFS_TI_MAGIC;
 330         INIT_LIST_HEAD(&ti->ti_garbage);
 331         current->journal_info = ti;
 332
 333         for (;;) {
 334                 down_write(&sbi->s_nilfs->ns_segctor_sem);
 335                 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &NILFS_SC(sbi)->sc_flags))
 336                         break;
 337
 338                 nilfs_segctor_do_immediate_flush(NILFS_SC(sbi));
 339
 340                 up_write(&sbi->s_nilfs->ns_segctor_sem);
 341                 yield();
 342         }
 343         if (gcflag)
 344                 ti->ti_flags |= NILFS_TI_GC;
 345 }
 346
 347 static void nilfs_transaction_unlock(struct nilfs_sb_info *sbi)
 348 {
 349         struct nilfs_transaction_info *ti = current->journal_info;
 350
 351         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 352         BUG_ON(ti->ti_count > 0);
 353
 354         up_write(&sbi->s_nilfs->ns_segctor_sem);
 355         current->journal_info = ti->ti_save;
 356         if (!list_empty(&ti->ti_garbage))
 357                 nilfs_dispose_list(sbi, &ti->ti_garbage, 0);
 358 }
 359
 360 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
 361                                             struct nilfs_segsum_pointer *ssp,
 362                                             unsigned bytes)
 363 {
 364         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 365         unsigned blocksize = sci->sc_super->s_blocksize;
 366         void *p;
 367
 368         if (unlikely(ssp->offset + bytes > blocksize)) {
 369                 ssp->offset = 0;
 370                 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
 371                                                &segbuf->sb_segsum_buffers));
 372                 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
 373         }
 374         p = ssp->bh->b_data + ssp->offset;
 375         ssp->offset += bytes;
 376         return p;
 377 }
 378
 379 /**
 380  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
 381  * @sci: nilfs_sc_info
 382  */
 383 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
 384 {
 385         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 386         struct buffer_head *sumbh;
 387         unsigned sumbytes;
 388         unsigned flags = 0;
 389         int err;
 390
 391         if (nilfs_doing_gc())
 392                 flags = NILFS_SS_GC;
 393         err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime);
 394         if (unlikely(err))
 395                 return err;
 396
 397         sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
 398         sumbytes = segbuf->sb_sum.sumbytes;
 399         sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
 400         sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
 401         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 402         return 0;
 403 }
 404
 405 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
 406 {
 407         sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
 408         if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
 409                 return -E2BIG; /* The current segment is filled up
 410                                   (internal code) */
 411         sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
 412         return nilfs_segctor_reset_segment_buffer(sci);
 413 }
 414
 415 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
 416 {
 417         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 418         int err;
 419
 420         if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
 421                 err = nilfs_segctor_feed_segment(sci);
 422                 if (err)
 423                         return err;
 424                 segbuf = sci->sc_curseg;
 425         }
 426         err = nilfs_segbuf_extend_payload(segbuf, &sci->sc_super_root);
 427         if (likely(!err))
 428                 segbuf->sb_sum.flags |= NILFS_SS_SR;
 429         return err;
 430 }
 431
 432 /*
 433  * Functions for making segment summary and payloads
 434  */
 435 static int nilfs_segctor_segsum_block_required(
 436         struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
 437         unsigned binfo_size)
 438 {
 439         unsigned blocksize = sci->sc_super->s_blocksize;
 440         /* Size of finfo and binfo is enough small against blocksize */
 441
 442         return ssp->offset + binfo_size +
 443                 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
 444                 blocksize;
 445 }
 446
 447 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
 448                                       struct inode *inode)
 449 {
 450         sci->sc_curseg->sb_sum.nfinfo++;
 451         sci->sc_binfo_ptr = sci->sc_finfo_ptr;
 452         nilfs_segctor_map_segsum_entry(
 453                 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
 454         /* skip finfo */
 455 }
 456
 457 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
 458                                     struct inode *inode)
 459 {
 460         struct nilfs_finfo *finfo;
 461         struct nilfs_inode_info *ii;
 462         struct nilfs_segment_buffer *segbuf;
 463
 464         if (sci->sc_blk_cnt == 0)
 465                 return;
 466
 467         ii = NILFS_I(inode);
 468         finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
 469                                                  sizeof(*finfo));
 470         finfo->fi_ino = cpu_to_le64(inode->i_ino);
 471         finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
 472         finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
 473         finfo->fi_cno = cpu_to_le64(ii->i_cno);
 474
 475         segbuf = sci->sc_curseg;
 476         segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
 477                 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
 478         sci->sc_finfo_ptr = sci->sc_binfo_ptr;
 479         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 480 }
 481
 482 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
 483                                         struct buffer_head *bh,
 484                                         struct inode *inode,
 485                                         unsigned binfo_size)
 486 {
 487         struct nilfs_segment_buffer *segbuf;
 488         int required, err = 0;
 489
 490  retry:
 491         segbuf = sci->sc_curseg;
 492         required = nilfs_segctor_segsum_block_required(
 493                 sci, &sci->sc_binfo_ptr, binfo_size);
 494         if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
 495                 nilfs_segctor_end_finfo(sci, inode);
 496                 err = nilfs_segctor_feed_segment(sci);
 497                 if (err)
 498                         return err;
 499                 goto retry;
 500         }
 501         if (unlikely(required)) {
 502                 err = nilfs_segbuf_extend_segsum(segbuf);
 503                 if (unlikely(err))
 504                         goto failed;
 505         }
 506         if (sci->sc_blk_cnt == 0)
 507                 nilfs_segctor_begin_finfo(sci, inode);
 508
 509         nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
 510         /* Substitution to vblocknr is delayed until update_blocknr() */
 511         nilfs_segbuf_add_file_buffer(segbuf, bh);
 512         sci->sc_blk_cnt++;
 513  failed:
 514         return err;
 515 }
 516
 517 static int nilfs_handle_bmap_error(int err, const char *fname,
 518                                    struct inode *inode, struct super_block *sb)
 519 {
 520         if (err == -EINVAL) {
 521                 nilfs_error(sb, fname, "broken bmap (inode=%lu)\n",
 522                             inode->i_ino);
 523                 err = -EIO;
 524         }
 525         return err;
 526 }
 527
 528 /*
 529  * Callback functions that enumerate, mark, and collect dirty blocks
 530  */
 531 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
 532                                    struct buffer_head *bh, struct inode *inode)
 533 {
 534         int err;
 535
 536         /* BUG_ON(!buffer_dirty(bh)); */
 537         /* excluded by scan_dirty_data_buffers() */
 538         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 539         if (unlikely(err < 0))
 540                 return nilfs_handle_bmap_error(err, __func__, inode,
 541                                                sci->sc_super);
 542
 543         err = nilfs_segctor_add_file_block(sci, bh, inode,
 544                                            sizeof(struct nilfs_binfo_v));
 545         if (!err)
 546                 sci->sc_datablk_cnt++;
 547         return err;
 548 }
 549
 550 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
 551                                    struct buffer_head *bh,
 552                                    struct inode *inode)
 553 {
 554         int err;
 555
 556         /* BUG_ON(!buffer_dirty(bh)); */
 557         /* excluded by scan_dirty_node_buffers() */
 558         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 559         if (unlikely(err < 0))
 560                 return nilfs_handle_bmap_error(err, __func__, inode,
 561                                                sci->sc_super);
 562         return 0;
 563 }
 564
 565 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
 566                                    struct buffer_head *bh,
 567                                    struct inode *inode)
 568 {
 569         BUG_ON(!buffer_dirty(bh));
 570         return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 571 }
 572
 573 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
 574                                         struct nilfs_segsum_pointer *ssp,
 575                                         union nilfs_binfo *binfo)
 576 {
 577         struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
 578                 sci, ssp, sizeof(*binfo_v));
 579         *binfo_v = binfo->bi_v;
 580 }
 581
 582 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
 583                                         struct nilfs_segsum_pointer *ssp,
 584                                         union nilfs_binfo *binfo)
 585 {
 586         __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
 587                 sci, ssp, sizeof(*vblocknr));
 588         *vblocknr = binfo->bi_v.bi_vblocknr;
 589 }
 590
 591 struct nilfs_sc_operations nilfs_sc_file_ops = {
 592         .collect_data = nilfs_collect_file_data,
 593         .collect_node = nilfs_collect_file_node,
 594         .collect_bmap = nilfs_collect_file_bmap,
 595         .write_data_binfo = nilfs_write_file_data_binfo,
 596         .write_node_binfo = nilfs_write_file_node_binfo,
 597 };
 598
 599 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
 600                                   struct buffer_head *bh, struct inode *inode)
 601 {
 602         int err;
 603
 604         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 605         if (unlikely(err < 0))
 606                 return nilfs_handle_bmap_error(err, __func__, inode,
 607                                                sci->sc_super);
 608
 609         err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 610         if (!err)
 611                 sci->sc_datablk_cnt++;
 612         return err;
 613 }
 614
 615 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
 616                                   struct buffer_head *bh, struct inode *inode)
 617 {
 618         BUG_ON(!buffer_dirty(bh));
 619         return nilfs_segctor_add_file_block(sci, bh, inode,
 620                                             sizeof(struct nilfs_binfo_dat));
 621 }
 622
 623 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
 624                                        struct nilfs_segsum_pointer *ssp,
 625                                        union nilfs_binfo *binfo)
 626 {
 627         __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
 628                                                           sizeof(*blkoff));
 629         *blkoff = binfo->bi_dat.bi_blkoff;
 630 }
 631
 632 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
 633                                        struct nilfs_segsum_pointer *ssp,
 634                                        union nilfs_binfo *binfo)
 635 {
 636         struct nilfs_binfo_dat *binfo_dat =
 637                 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
 638         *binfo_dat = binfo->bi_dat;
 639 }
 640
 641 struct nilfs_sc_operations nilfs_sc_dat_ops = {
 642         .collect_data = nilfs_collect_dat_data,
 643         .collect_node = nilfs_collect_file_node,
 644         .collect_bmap = nilfs_collect_dat_bmap,
 645         .write_data_binfo = nilfs_write_dat_data_binfo,
 646         .write_node_binfo = nilfs_write_dat_node_binfo,
 647 };
 648
 649 struct nilfs_sc_operations nilfs_sc_dsync_ops = {
 650         .collect_data = nilfs_collect_file_data,
 651         .collect_node = NULL,
 652         .collect_bmap = NULL,
 653         .write_data_binfo = nilfs_write_file_data_binfo,
 654         .write_node_binfo = NULL,
 655 };
 656
 657 static int nilfs_lookup_dirty_data_buffers(struct inode *inode,
 658                                            struct list_head *listp,
 659                                            struct nilfs_sc_info *sci)
 660 {
 661         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 662         struct address_space *mapping = inode->i_mapping;
 663         struct pagevec pvec;
 664         unsigned i, ndirties = 0, nlimit;
 665         pgoff_t index = 0;
 666         int err = 0;
 667
 668         nlimit = sci->sc_segbuf_nblocks -
 669                 (sci->sc_nblk_this_inc + segbuf->sb_sum.nblocks);
 670         pagevec_init(&pvec, 0);
 671  repeat:
 672         if (!pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 673                                 PAGEVEC_SIZE))
 674                 return 0;
 675
 676         for (i = 0; i < pagevec_count(&pvec); i++) {
 677                 struct buffer_head *bh, *head;
 678                 struct page *page = pvec.pages[i];
 679
 680                 if (mapping->host) {
 681                         lock_page(page);
 682                         if (!page_has_buffers(page))
 683                                 create_empty_buffers(page,
 684                                                      1 << inode->i_blkbits, 0);
 685                         unlock_page(page);
 686                 }
 687
 688                 bh = head = page_buffers(page);
 689                 do {
 690                         if (buffer_dirty(bh)) {
 691                                 if (ndirties > nlimit) {
 692                                         err = -E2BIG;
 693                                         break;
 694                                 }
 695                                 get_bh(bh);
 696                                 list_add_tail(&bh->b_assoc_buffers, listp);
 697                                 ndirties++;
 698                         }
 699                         bh = bh->b_this_page;
 700                 } while (bh != head);
 701         }
 702         pagevec_release(&pvec);
 703         cond_resched();
 704
 705         if (!err)
 706                 goto repeat;
 707         return err;
 708 }
 709
 710 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
 711                                             struct list_head *listp)
 712 {
 713         struct nilfs_inode_info *ii = NILFS_I(inode);
 714         struct address_space *mapping = &ii->i_btnode_cache;
 715         struct pagevec pvec;
 716         struct buffer_head *bh, *head;
 717         unsigned int i;
 718         pgoff_t index = 0;
 719
 720         pagevec_init(&pvec, 0);
 721
 722         while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 723                                   PAGEVEC_SIZE)) {
 724                 for (i = 0; i < pagevec_count(&pvec); i++) {
 725                         bh = head = page_buffers(pvec.pages[i]);
 726                         do {
 727                                 if (buffer_dirty(bh)) {
 728                                         get_bh(bh);
 729                                         list_add_tail(&bh->b_assoc_buffers,
 730                                                       listp);
 731                                 }
 732                                 bh = bh->b_this_page;
 733                         } while (bh != head);
 734                 }
 735                 pagevec_release(&pvec);
 736                 cond_resched();
 737         }
 738 }
 739
 740 static void nilfs_dispose_list(struct nilfs_sb_info *sbi,
 741                                struct list_head *head, int force)
 742 {
 743         struct nilfs_inode_info *ii, *n;
 744         struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
 745         unsigned nv = 0;
 746
 747         while (!list_empty(head)) {
 748                 spin_lock(&sbi->s_inode_lock);
 749                 list_for_each_entry_safe(ii, n, head, i_dirty) {
 750                         list_del_init(&ii->i_dirty);
 751                         if (force) {
 752                                 if (unlikely(ii->i_bh)) {
 753                                         brelse(ii->i_bh);
 754                                         ii->i_bh = NULL;
 755                                 }
 756                         } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
 757                                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 758                                 list_add_tail(&ii->i_dirty,
 759                                               &sbi->s_dirty_files);
 760                                 continue;
 761                         }
 762                         ivec[nv++] = ii;
 763                         if (nv == SC_N_INODEVEC)
 764                                 break;
 765                 }
 766                 spin_unlock(&sbi->s_inode_lock);
 767
 768                 for (pii = ivec; nv > 0; pii++, nv--)
 769                         iput(&(*pii)->vfs_inode);
 770         }
 771 }
 772
 773 static int nilfs_test_metadata_dirty(struct nilfs_sb_info *sbi)
 774 {
 775         struct the_nilfs *nilfs = sbi->s_nilfs;
 776         int ret = 0;
 777
 778         if (nilfs_mdt_fetch_dirty(sbi->s_ifile))
 779                 ret++;
 780         if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
 781                 ret++;
 782         if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
 783                 ret++;
 784         if (ret || nilfs_doing_gc())
 785                 if (nilfs_mdt_fetch_dirty(nilfs_dat_inode(nilfs)))
 786                         ret++;
 787         return ret;
 788 }
 789
 790 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
 791 {
 792         return list_empty(&sci->sc_dirty_files) &&
 793                 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
 794                 list_empty(&sci->sc_cleaning_segments) &&
 795                 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
 796 }
 797
 798 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
 799 {
 800         struct nilfs_sb_info *sbi = sci->sc_sbi;
 801         int ret = 0;
 802
 803         if (nilfs_test_metadata_dirty(sbi))
 804                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
 805
 806         spin_lock(&sbi->s_inode_lock);
 807         if (list_empty(&sbi->s_dirty_files) && nilfs_segctor_clean(sci))
 808                 ret++;
 809
 810         spin_unlock(&sbi->s_inode_lock);
 811         return ret;
 812 }
 813
 814 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
 815 {
 816         struct nilfs_sb_info *sbi = sci->sc_sbi;
 817         struct the_nilfs *nilfs = sbi->s_nilfs;
 818
 819         nilfs_mdt_clear_dirty(sbi->s_ifile);
 820         nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
 821         nilfs_mdt_clear_dirty(nilfs->ns_sufile);
 822         nilfs_mdt_clear_dirty(nilfs_dat_inode(nilfs));
 823 }
 824
 825 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
 826 {
 827         struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
 828         struct buffer_head *bh_cp;
 829         struct nilfs_checkpoint *raw_cp;
 830         int err;
 831
 832         /* XXX: this interface will be changed */
 833         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
 834                                           &raw_cp, &bh_cp);
 835         if (likely(!err)) {
 836                 /* The following code is duplicated with cpfile.  But, it is
 837                    needed to collect the checkpoint even if it was not newly
 838                    created */
 839                 nilfs_mdt_mark_buffer_dirty(bh_cp);
 840                 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
 841                 nilfs_cpfile_put_checkpoint(
 842                         nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 843         } else {
 844                 BUG_ON(err == -EINVAL || err == -ENOENT);
 845         }
 846         return err;
 847 }
 848
 849 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
 850 {
 851         struct nilfs_sb_info *sbi = sci->sc_sbi;
 852         struct the_nilfs *nilfs = sbi->s_nilfs;
 853         struct buffer_head *bh_cp;
 854         struct nilfs_checkpoint *raw_cp;
 855         int err;
 856
 857         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
 858                                           &raw_cp, &bh_cp);
 859         if (unlikely(err)) {
 860                 BUG_ON(err == -EINVAL || err == -ENOENT);
 861                 goto failed_ibh;
 862         }
 863         raw_cp->cp_snapshot_list.ssl_next = 0;
 864         raw_cp->cp_snapshot_list.ssl_prev = 0;
 865         raw_cp->cp_inodes_count =
 866                 cpu_to_le64(atomic_read(&sbi->s_inodes_count));
 867         raw_cp->cp_blocks_count =
 868                 cpu_to_le64(atomic_read(&sbi->s_blocks_count));
 869         raw_cp->cp_nblk_inc =
 870                 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
 871         raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
 872         raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
 873         if (sci->sc_sketch_inode && i_size_read(sci->sc_sketch_inode) > 0)
 874                 nilfs_checkpoint_set_sketch(raw_cp);
 875         nilfs_write_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode, 1);
 876         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 877         return 0;
 878
 879  failed_ibh:
 880         return err;
 881 }
 882
 883 static void nilfs_fill_in_file_bmap(struct inode *ifile,
 884                                     struct nilfs_inode_info *ii)
 885
 886 {
 887         struct buffer_head *ibh;
 888         struct nilfs_inode *raw_inode;
 889
 890         if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
 891                 ibh = ii->i_bh;
 892                 BUG_ON(!ibh);
 893                 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
 894                                                   ibh);
 895                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 896                 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
 897         }
 898 }
 899
 900 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci,
 901                                             struct inode *ifile)
 902 {
 903         struct nilfs_inode_info *ii;
 904
 905         list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
 906                 nilfs_fill_in_file_bmap(ifile, ii);
 907                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
 908         }
 909         if (sci->sc_sketch_inode) {
 910                 ii = NILFS_I(sci->sc_sketch_inode);
 911                 if (test_bit(NILFS_I_DIRTY, &ii->i_state))
 912                         nilfs_fill_in_file_bmap(ifile, ii);
 913         }
 914 }
 915
 916 /*
 917  * CRC calculation routines
 918  */
 919 static void nilfs_fill_in_super_root_crc(struct buffer_head *bh_sr, u32 seed)
 920 {
 921         struct nilfs_super_root *raw_sr =
 922                 (struct nilfs_super_root *)bh_sr->b_data;
 923         u32 crc;
 924
 925         BUG_ON(NILFS_SR_BYTES > bh_sr->b_size);
 926         crc = crc32_le(seed,
 927                        (unsigned char *)raw_sr + sizeof(raw_sr->sr_sum),
 928                        NILFS_SR_BYTES - sizeof(raw_sr->sr_sum));
 929         raw_sr->sr_sum = cpu_to_le32(crc);
 930 }
 931
 932 static void nilfs_segctor_fill_in_checksums(struct nilfs_sc_info *sci,
 933                                             u32 seed)
 934 {
 935         struct nilfs_segment_buffer *segbuf;
 936
 937         if (sci->sc_super_root)
 938                 nilfs_fill_in_super_root_crc(sci->sc_super_root, seed);
 939
 940         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
 941                 nilfs_segbuf_fill_in_segsum_crc(segbuf, seed);
 942                 nilfs_segbuf_fill_in_data_crc(segbuf, seed);
 943         }
 944 }
 945
 946 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
 947                                              struct the_nilfs *nilfs)
 948 {
 949         struct buffer_head *bh_sr = sci->sc_super_root;
 950         struct nilfs_super_root *raw_sr =
 951                 (struct nilfs_super_root *)bh_sr->b_data;
 952         unsigned isz = nilfs->ns_inode_size;
 953
 954         raw_sr->sr_bytes = cpu_to_le16(NILFS_SR_BYTES);
 955         raw_sr->sr_nongc_ctime
 956                 = cpu_to_le64(nilfs_doing_gc() ?
 957                               nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
 958         raw_sr->sr_flags = 0;
 959
 960         nilfs_mdt_write_inode_direct(
 961                 nilfs_dat_inode(nilfs), bh_sr, NILFS_SR_DAT_OFFSET(isz));
 962         nilfs_mdt_write_inode_direct(
 963                 nilfs->ns_cpfile, bh_sr, NILFS_SR_CPFILE_OFFSET(isz));
 964         nilfs_mdt_write_inode_direct(
 965                 nilfs->ns_sufile, bh_sr, NILFS_SR_SUFILE_OFFSET(isz));
 966 }
 967
 968 static void nilfs_redirty_inodes(struct list_head *head)
 969 {
 970         struct nilfs_inode_info *ii;
 971
 972         list_for_each_entry(ii, head, i_dirty) {
 973                 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
 974                         clear_bit(NILFS_I_COLLECTED, &ii->i_state);
 975         }
 976 }
 977
 978 static void nilfs_drop_collected_inodes(struct list_head *head)
 979 {
 980         struct nilfs_inode_info *ii;
 981
 982         list_for_each_entry(ii, head, i_dirty) {
 983                 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
 984                         continue;
 985
 986                 clear_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 987                 set_bit(NILFS_I_UPDATED, &ii->i_state);
 988         }
 989 }
 990
 991 static void nilfs_segctor_cancel_free_segments(struct nilfs_sc_info *sci,
 992                                                struct inode *sufile)
 993
 994 {
 995         struct list_head *head = &sci->sc_cleaning_segments;
 996         struct nilfs_segment_entry *ent;
 997         int err;
 998
 999         list_for_each_entry(ent, head, list) {
1000                 if (!(ent->flags & NILFS_SLH_FREED))
1001                         break;
1002                 err = nilfs_sufile_cancel_free(sufile, ent->segnum);
1003                 BUG_ON(err);
1004
1005                 ent->flags &= ~NILFS_SLH_FREED;
1006         }
1007 }
1008
1009 static int nilfs_segctor_prepare_free_segments(struct nilfs_sc_info *sci,
1010                                                struct inode *sufile)
1011 {
1012         struct list_head *head = &sci->sc_cleaning_segments;
1013         struct nilfs_segment_entry *ent;
1014         int err;
1015
1016         list_for_each_entry(ent, head, list) {
1017                 err = nilfs_sufile_free(sufile, ent->segnum);
1018                 if (unlikely(err))
1019                         return err;
1020                 ent->flags |= NILFS_SLH_FREED;
1021         }
1022         return 0;
1023 }
1024
1025 static void nilfs_segctor_commit_free_segments(struct nilfs_sc_info *sci)
1026 {
1027         nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
1028 }
1029
1030 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1031                                        struct inode *inode,
1032                                        struct list_head *listp,
1033                                        int (*collect)(struct nilfs_sc_info *,
1034                                                       struct buffer_head *,
1035                                                       struct inode *))
1036 {
1037         struct buffer_head *bh, *n;
1038         int err = 0;
1039
1040         if (collect) {
1041                 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1042                         list_del_init(&bh->b_assoc_buffers);
1043                         err = collect(sci, bh, inode);
1044                         brelse(bh);
1045                         if (unlikely(err))
1046                                 goto dispose_buffers;
1047                 }
1048                 return 0;
1049         }
1050
1051  dispose_buffers:
1052         while (!list_empty(listp)) {
1053                 bh = list_entry(listp->next, struct buffer_head,
1054                                 b_assoc_buffers);
1055                 list_del_init(&bh->b_assoc_buffers);
1056                 brelse(bh);
1057         }
1058         return err;
1059 }
1060
1061 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1062                                    struct inode *inode,
1063                                    struct nilfs_sc_operations *sc_ops)
1064 {
1065         LIST_HEAD(data_buffers);
1066         LIST_HEAD(node_buffers);
1067         int err, err2;
1068
1069         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1070                 err = nilfs_lookup_dirty_data_buffers(inode, &data_buffers,
1071                                                       sci);
1072                 if (err) {
1073                         err2 = nilfs_segctor_apply_buffers(
1074                                 sci, inode, &data_buffers,
1075                                 err == -E2BIG ? sc_ops->collect_data : NULL);
1076                         if (err == -E2BIG)
1077                                 err = err2;
1078                         goto break_or_fail;
1079                 }
1080         }
1081         nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1082
1083         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1084                 err = nilfs_segctor_apply_buffers(
1085                         sci, inode, &data_buffers, sc_ops->collect_data);
1086                 if (unlikely(err)) {
1087                         /* dispose node list */
1088                         nilfs_segctor_apply_buffers(
1089                                 sci, inode, &node_buffers, NULL);
1090                         goto break_or_fail;
1091                 }
1092                 sci->sc_stage.flags |= NILFS_CF_NODE;
1093         }
1094         /* Collect node */
1095         err = nilfs_segctor_apply_buffers(
1096                 sci, inode, &node_buffers, sc_ops->collect_node);
1097         if (unlikely(err))
1098                 goto break_or_fail;
1099
1100         nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1101         err = nilfs_segctor_apply_buffers(
1102                 sci, inode, &node_buffers, sc_ops->collect_bmap);
1103         if (unlikely(err))
1104                 goto break_or_fail;
1105
1106         nilfs_segctor_end_finfo(sci, inode);
1107         sci->sc_stage.flags &= ~NILFS_CF_NODE;
1108
1109  break_or_fail:
1110         return err;
1111 }
1112
1113 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1114                                          struct inode *inode)
1115 {
1116         LIST_HEAD(data_buffers);
1117         int err, err2;
1118
1119         err = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, sci);
1120         err2 = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1121                                            (!err || err == -E2BIG) ?
1122                                            nilfs_collect_file_data : NULL);
1123         if (err == -E2BIG)
1124                 err = err2;
1125         if (!err)
1126                 nilfs_segctor_end_finfo(sci, inode);
1127         return err;
1128 }
1129
1130 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1131 {
1132         struct nilfs_sb_info *sbi = sci->sc_sbi;
1133         struct the_nilfs *nilfs = sbi->s_nilfs;
1134         struct list_head *head;
1135         struct nilfs_inode_info *ii;
1136         int err = 0;
1137
1138         switch (sci->sc_stage.scnt) {
1139         case NILFS_ST_INIT:
1140                 /* Pre-processes */
1141                 sci->sc_stage.flags = 0;
1142
1143                 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1144                         sci->sc_nblk_inc = 0;
1145                         sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1146                         if (mode == SC_LSEG_DSYNC) {
1147                                 sci->sc_stage.scnt = NILFS_ST_DSYNC;
1148                                 goto dsync_mode;
1149                         }
1150                 }
1151
1152                 sci->sc_stage.dirty_file_ptr = NULL;
1153                 sci->sc_stage.gc_inode_ptr = NULL;
1154                 if (mode == SC_FLUSH_DAT) {
1155                         sci->sc_stage.scnt = NILFS_ST_DAT;
1156                         goto dat_stage;
1157                 }
1158                 sci->sc_stage.scnt++;  /* Fall through */
1159         case NILFS_ST_GC:
1160                 if (nilfs_doing_gc()) {
1161                         head = &sci->sc_gc_inodes;
1162                         ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1163                                                 head, i_dirty);
1164                         list_for_each_entry_continue(ii, head, i_dirty) {
1165                                 err = nilfs_segctor_scan_file(
1166                                         sci, &ii->vfs_inode,
1167                                         &nilfs_sc_file_ops);
1168                                 if (unlikely(err)) {
1169                                         sci->sc_stage.gc_inode_ptr = list_entry(
1170                                                 ii->i_dirty.prev,
1171                                                 struct nilfs_inode_info,
1172                                                 i_dirty);
1173                                         goto break_or_fail;
1174                                 }
1175                                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1176                         }
1177                         sci->sc_stage.gc_inode_ptr = NULL;
1178                 }
1179                 sci->sc_stage.scnt++;  /* Fall through */
1180         case NILFS_ST_FILE:
1181                 head = &sci->sc_dirty_files;
1182                 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1183                                         i_dirty);
1184                 list_for_each_entry_continue(ii, head, i_dirty) {
1185                         clear_bit(NILFS_I_DIRTY, &ii->i_state);
1186
1187                         err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1188                                                       &nilfs_sc_file_ops);
1189                         if (unlikely(err)) {
1190                                 sci->sc_stage.dirty_file_ptr =
1191                                         list_entry(ii->i_dirty.prev,
1192                                                    struct nilfs_inode_info,
1193                                                    i_dirty);
1194                                 goto break_or_fail;
1195                         }
1196                         /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1197                         /* XXX: required ? */
1198                 }
1199                 sci->sc_stage.dirty_file_ptr = NULL;
1200                 if (mode == SC_FLUSH_FILE) {
1201                         sci->sc_stage.scnt = NILFS_ST_DONE;
1202                         return 0;
1203                 }
1204                 sci->sc_stage.scnt++;  /* Fall through */
1205         case NILFS_ST_SKETCH:
1206                 if (mode == SC_LSEG_SR && sci->sc_sketch_inode) {
1207                         ii = NILFS_I(sci->sc_sketch_inode);
1208                         if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
1209                                 sci->sc_sketch_inode->i_ctime.tv_sec
1210                                         = sci->sc_seg_ctime;
1211                                 sci->sc_sketch_inode->i_mtime.tv_sec
1212                                         = sci->sc_seg_ctime;
1213                                 err = nilfs_mark_inode_dirty(
1214                                         sci->sc_sketch_inode);
1215                                 if (unlikely(err))
1216                                         goto break_or_fail;
1217                         }
1218                         err = nilfs_segctor_scan_file(sci,
1219                                                       sci->sc_sketch_inode,
1220                                                       &nilfs_sc_file_ops);
1221                         if (unlikely(err))
1222                                 goto break_or_fail;
1223                 }
1224                 sci->sc_stage.scnt++;
1225                 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1226                 /* Fall through */
1227         case NILFS_ST_IFILE:
1228                 err = nilfs_segctor_scan_file(sci, sbi->s_ifile,
1229                                               &nilfs_sc_file_ops);
1230                 if (unlikely(err))
1231                         break;
1232                 sci->sc_stage.scnt++;
1233                 /* Creating a checkpoint */
1234                 err = nilfs_segctor_create_checkpoint(sci);
1235                 if (unlikely(err))
1236                         break;
1237                 /* Fall through */
1238         case NILFS_ST_CPFILE:
1239                 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1240                                               &nilfs_sc_file_ops);
1241                 if (unlikely(err))
1242                         break;
1243                 sci->sc_stage.scnt++;  /* Fall through */
1244         case NILFS_ST_SUFILE:
1245                 err = nilfs_segctor_prepare_free_segments(sci,
1246                                                           nilfs->ns_sufile);
1247                 if (unlikely(err))
1248                         break;
1249                 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1250                                               &nilfs_sc_file_ops);
1251                 if (unlikely(err))
1252                         break;
1253                 sci->sc_stage.scnt++;  /* Fall through */
1254         case NILFS_ST_DAT:
1255  dat_stage:
1256                 err = nilfs_segctor_scan_file(sci, nilfs_dat_inode(nilfs),
1257                                               &nilfs_sc_dat_ops);
1258                 if (unlikely(err))
1259                         break;
1260                 if (mode == SC_FLUSH_DAT) {
1261                         sci->sc_stage.scnt = NILFS_ST_DONE;
1262                         return 0;
1263                 }
1264                 sci->sc_stage.scnt++;  /* Fall through */
1265         case NILFS_ST_SR:
1266                 if (mode == SC_LSEG_SR) {
1267                         /* Appending a super root */
1268                         err = nilfs_segctor_add_super_root(sci);
1269                         if (unlikely(err))
1270                                 break;
1271                 }
1272                 /* End of a logical segment */
1273                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1274                 sci->sc_stage.scnt = NILFS_ST_DONE;
1275                 return 0;
1276         case NILFS_ST_DSYNC:
1277  dsync_mode:
1278                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1279                 ii = sci->sc_stage.dirty_file_ptr;
1280                 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1281                         break;
1282
1283                 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1284                 if (unlikely(err))
1285                         break;
1286                 sci->sc_stage.dirty_file_ptr = NULL;
1287                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1288                 sci->sc_stage.scnt = NILFS_ST_DONE;
1289                 return 0;
1290         case NILFS_ST_DONE:
1291                 return 0;
1292         default:
1293                 BUG();
1294         }
1295
1296  break_or_fail:
1297         return err;
1298 }
1299
1300 static int nilfs_segctor_terminate_segment(struct nilfs_sc_info *sci,
1301                                            struct nilfs_segment_buffer *segbuf,
1302                                            struct inode *sufile)
1303 {
1304         struct nilfs_segment_entry *ent = segbuf->sb_segent;
1305         int err;
1306
1307         err = nilfs_open_segment_entry(ent, sufile);
1308         if (unlikely(err))
1309                 return err;
1310         nilfs_mdt_mark_buffer_dirty(ent->bh_su);
1311         nilfs_mdt_mark_dirty(sufile);
1312         nilfs_close_segment_entry(ent, sufile);
1313
1314         list_add_tail(&ent->list, &sci->sc_active_segments);
1315         segbuf->sb_segent = NULL;
1316         return 0;
1317 }
1318
1319 static int nilfs_touch_segusage(struct inode *sufile, __u64 segnum)
1320 {
1321         struct buffer_head *bh_su;
1322         struct nilfs_segment_usage *raw_su;
1323         int err;
1324
1325         err = nilfs_sufile_get_segment_usage(sufile, segnum, &raw_su, &bh_su);
1326         if (unlikely(err))
1327                 return err;
1328         nilfs_mdt_mark_buffer_dirty(bh_su);
1329         nilfs_mdt_mark_dirty(sufile);
1330         nilfs_sufile_put_segment_usage(sufile, segnum, bh_su);
1331         return 0;
1332 }
1333
1334 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1335                                             struct the_nilfs *nilfs)
1336 {
1337         struct nilfs_segment_buffer *segbuf, *n;
1338         struct inode *sufile = nilfs->ns_sufile;
1339         __u64 nextnum;
1340         int err;
1341
1342         if (list_empty(&sci->sc_segbufs)) {
1343                 segbuf = nilfs_segbuf_new(sci->sc_super);
1344                 if (unlikely(!segbuf))
1345                         return -ENOMEM;
1346                 list_add(&segbuf->sb_list, &sci->sc_segbufs);
1347         } else
1348                 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1349
1350         err = nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1351                                nilfs->ns_pseg_offset, nilfs);
1352         if (unlikely(err))
1353                 return err;
1354
1355         if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1356                 err = nilfs_segctor_terminate_segment(sci, segbuf, sufile);
1357                 if (unlikely(err))
1358                         return err;
1359
1360                 nilfs_shift_to_next_segment(nilfs);
1361                 err = nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1362         }
1363         sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1364
1365         err = nilfs_touch_segusage(sufile, segbuf->sb_segnum);
1366         if (unlikely(err))
1367                 return err;
1368
1369         if (nilfs->ns_segnum == nilfs->ns_nextnum) {
1370                 /* Start from the head of a new full segment */
1371                 err = nilfs_sufile_alloc(sufile, &nextnum);
1372                 if (unlikely(err))
1373                         return err;
1374         } else
1375                 nextnum = nilfs->ns_nextnum;
1376
1377         segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1378         nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1379
1380         /* truncating segment buffers */
1381         list_for_each_entry_safe_continue(segbuf, n, &sci->sc_segbufs,
1382                                           sb_list) {
1383                 list_del_init(&segbuf->sb_list);
1384                 nilfs_segbuf_free(segbuf);
1385         }
1386         return err;
1387 }
1388
1389 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1390                                          struct the_nilfs *nilfs, int nadd)
1391 {
1392         struct nilfs_segment_buffer *segbuf, *prev, *n;
1393         struct inode *sufile = nilfs->ns_sufile;
1394         __u64 nextnextnum;
1395         LIST_HEAD(list);
1396         int err, ret, i;
1397
1398         prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1399         /*
1400          * Since the segment specified with nextnum might be allocated during
1401          * the previous construction, the buffer including its segusage may
1402          * not be dirty.  The following call ensures that the buffer is dirty
1403          * and will pin the buffer on memory until the sufile is written.
1404          */
1405         err = nilfs_touch_segusage(sufile, prev->sb_nextnum);
1406         if (unlikely(err))
1407                 return err;
1408
1409         for (i = 0; i < nadd; i++) {
1410                 /* extend segment info */
1411                 err = -ENOMEM;
1412                 segbuf = nilfs_segbuf_new(sci->sc_super);
1413                 if (unlikely(!segbuf))
1414                         goto failed;
1415
1416                 /* map this buffer to region of segment on-disk */
1417                 err = nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1418                 if (unlikely(err))
1419                         goto failed_segbuf;
1420
1421                 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1422
1423                 /* allocate the next next full segment */
1424                 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1425                 if (unlikely(err))
1426                         goto failed_segbuf;
1427
1428                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1429                 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1430
1431                 list_add_tail(&segbuf->sb_list, &list);
1432                 prev = segbuf;
1433         }
1434         list_splice(&list, sci->sc_segbufs.prev);
1435         return 0;
1436
1437  failed_segbuf:
1438         nilfs_segbuf_free(segbuf);
1439  failed:
1440         list_for_each_entry_safe(segbuf, n, &list, sb_list) {
1441                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1442                 BUG_ON(ret);
1443                 list_del_init(&segbuf->sb_list);
1444                 nilfs_segbuf_free(segbuf);
1445         }
1446         return err;
1447 }
1448
1449 static void nilfs_segctor_free_incomplete_segments(struct nilfs_sc_info *sci,
1450                                                    struct the_nilfs *nilfs)
1451 {
1452         struct nilfs_segment_buffer *segbuf;
1453         int ret, done = 0;
1454
1455         segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1456         if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1457                 ret = nilfs_sufile_free(nilfs->ns_sufile, segbuf->sb_nextnum);
1458                 BUG_ON(ret);
1459         }
1460         if (segbuf->sb_io_error) {
1461                 /* Case 1: The first segment failed */
1462                 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1463                         /* Case 1a:  Partial segment appended into an existing
1464                            segment */
1465                         nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1466                                                 segbuf->sb_fseg_end);
1467                 else /* Case 1b:  New full segment */
1468                         set_nilfs_discontinued(nilfs);
1469                 done++;
1470         }
1471
1472         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1473                 ret = nilfs_sufile_free(nilfs->ns_sufile, segbuf->sb_nextnum);
1474                 BUG_ON(ret);
1475                 if (!done && segbuf->sb_io_error) {
1476                         if (segbuf->sb_segnum != nilfs->ns_nextnum)
1477                                 /* Case 2: extended segment (!= next) failed */
1478                                 nilfs_sufile_set_error(nilfs->ns_sufile,
1479                                                        segbuf->sb_segnum);
1480                         done++;
1481                 }
1482         }
1483 }
1484
1485 static void nilfs_segctor_clear_segment_buffers(struct nilfs_sc_info *sci)
1486 {
1487         struct nilfs_segment_buffer *segbuf;
1488
1489         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list)
1490                 nilfs_segbuf_clear(segbuf);
1491         sci->sc_super_root = NULL;
1492 }
1493
1494 static void nilfs_segctor_destroy_segment_buffers(struct nilfs_sc_info *sci)
1495 {
1496         struct nilfs_segment_buffer *segbuf;
1497
1498         while (!list_empty(&sci->sc_segbufs)) {
1499                 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1500                 list_del_init(&segbuf->sb_list);
1501                 nilfs_segbuf_free(segbuf);
1502         }
1503         /* sci->sc_curseg = NULL; */
1504 }
1505
1506 static void nilfs_segctor_end_construction(struct nilfs_sc_info *sci,
1507                                            struct the_nilfs *nilfs, int err)
1508 {
1509         if (unlikely(err)) {
1510                 nilfs_segctor_free_incomplete_segments(sci, nilfs);
1511                 nilfs_segctor_cancel_free_segments(sci, nilfs->ns_sufile);
1512         }
1513         nilfs_segctor_clear_segment_buffers(sci);
1514 }
1515
1516 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1517                                           struct inode *sufile)
1518 {
1519         struct nilfs_segment_buffer *segbuf;
1520         struct buffer_head *bh_su;
1521         struct nilfs_segment_usage *raw_su;
1522         unsigned long live_blocks;
1523         int ret;
1524
1525         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1526                 ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1527                                                      &raw_su, &bh_su);
1528                 BUG_ON(ret); /* always succeed because bh_su is dirty */
1529                 live_blocks = segbuf->sb_sum.nblocks +
1530                         (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1531                 raw_su->su_lastmod = cpu_to_le64(sci->sc_seg_ctime);
1532                 raw_su->su_nblocks = cpu_to_le32(live_blocks);
1533                 nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum,
1534                                                bh_su);
1535         }
1536 }
1537
1538 static void nilfs_segctor_cancel_segusage(struct nilfs_sc_info *sci,
1539                                           struct inode *sufile)
1540 {
1541         struct nilfs_segment_buffer *segbuf;
1542         struct buffer_head *bh_su;
1543         struct nilfs_segment_usage *raw_su;
1544         int ret;
1545
1546         segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1547         ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1548                                              &raw_su, &bh_su);
1549         BUG_ON(ret); /* always succeed because bh_su is dirty */
1550         raw_su->su_nblocks = cpu_to_le32(segbuf->sb_pseg_start -
1551                                          segbuf->sb_fseg_start);
1552         nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum, bh_su);
1553
1554         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1555                 ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1556                                                      &raw_su, &bh_su);
1557                 BUG_ON(ret); /* always succeed */
1558                 raw_su->su_nblocks = 0;
1559                 nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum,
1560                                                bh_su);
1561         }
1562 }
1563
1564 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1565                                             struct nilfs_segment_buffer *last,
1566                                             struct inode *sufile)
1567 {
1568         struct nilfs_segment_buffer *segbuf = last, *n;
1569         int ret;
1570
1571         list_for_each_entry_safe_continue(segbuf, n, &sci->sc_segbufs,
1572                                           sb_list) {
1573                 list_del_init(&segbuf->sb_list);
1574                 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1575                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1576                 BUG_ON(ret);
1577                 nilfs_segbuf_free(segbuf);
1578         }
1579 }
1580
1581
1582 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1583                                  struct the_nilfs *nilfs, int mode)
1584 {
1585         struct nilfs_cstage prev_stage = sci->sc_stage;
1586         int err, nadd = 1;
1587
1588         /* Collection retry loop */
1589         for (;;) {
1590                 sci->sc_super_root = NULL;
1591                 sci->sc_nblk_this_inc = 0;
1592                 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1593
1594                 err = nilfs_segctor_reset_segment_buffer(sci);
1595                 if (unlikely(err))
1596                         goto failed;
1597
1598                 err = nilfs_segctor_collect_blocks(sci, mode);
1599                 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1600                 if (!err)
1601                         break;
1602
1603                 if (unlikely(err != -E2BIG))
1604                         goto failed;
1605
1606                 /* The current segment is filled up */
1607                 if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
1608                         break;
1609
1610                 nilfs_segctor_cancel_free_segments(sci, nilfs->ns_sufile);
1611                 nilfs_segctor_clear_segment_buffers(sci);
1612
1613                 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1614                 if (unlikely(err))
1615                         return err;
1616
1617                 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1618                 sci->sc_stage = prev_stage;
1619         }
1620         nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1621         return 0;
1622
1623  failed:
1624         return err;
1625 }
1626
1627 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1628                                       struct buffer_head *new_bh)
1629 {
1630         BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1631
1632         list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1633         /* The caller must release old_bh */
1634 }
1635
1636 static int
1637 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1638                                      struct nilfs_segment_buffer *segbuf,
1639                                      int mode)
1640 {
1641         struct inode *inode = NULL;
1642         sector_t blocknr;
1643         unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1644         unsigned long nblocks = 0, ndatablk = 0;
1645         struct nilfs_sc_operations *sc_op = NULL;
1646         struct nilfs_segsum_pointer ssp;
1647         struct nilfs_finfo *finfo = NULL;
1648         union nilfs_binfo binfo;
1649         struct buffer_head *bh, *bh_org;
1650         ino_t ino = 0;
1651         int err = 0;
1652
1653         if (!nfinfo)
1654                 goto out;
1655
1656         blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1657         ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1658         ssp.offset = sizeof(struct nilfs_segment_summary);
1659
1660         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1661                 if (bh == sci->sc_super_root)
1662                         break;
1663                 if (!finfo) {
1664                         finfo = nilfs_segctor_map_segsum_entry(
1665                                 sci, &ssp, sizeof(*finfo));
1666                         ino = le64_to_cpu(finfo->fi_ino);
1667                         nblocks = le32_to_cpu(finfo->fi_nblocks);
1668                         ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1669
1670                         if (buffer_nilfs_node(bh))
1671                                 inode = NILFS_BTNC_I(bh->b_page->mapping);
1672                         else
1673                                 inode = NILFS_AS_I(bh->b_page->mapping);
1674
1675                         if (mode == SC_LSEG_DSYNC)
1676                                 sc_op = &nilfs_sc_dsync_ops;
1677                         else if (ino == NILFS_DAT_INO)
1678                                 sc_op = &nilfs_sc_dat_ops;
1679                         else /* file blocks */
1680                                 sc_op = &nilfs_sc_file_ops;
1681                 }
1682                 bh_org = bh;
1683                 get_bh(bh_org);
1684                 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1685                                         &binfo);
1686                 if (bh != bh_org)
1687                         nilfs_list_replace_buffer(bh_org, bh);
1688                 brelse(bh_org);
1689                 if (unlikely(err))
1690                         goto failed_bmap;
1691
1692                 if (ndatablk > 0)
1693                         sc_op->write_data_binfo(sci, &ssp, &binfo);
1694                 else
1695                         sc_op->write_node_binfo(sci, &ssp, &binfo);
1696
1697                 blocknr++;
1698                 if (--nblocks == 0) {
1699                         finfo = NULL;
1700                         if (--nfinfo == 0)
1701                                 break;
1702                 } else if (ndatablk > 0)
1703                         ndatablk--;
1704         }
1705  out:
1706         return 0;
1707
1708  failed_bmap:
1709         err = nilfs_handle_bmap_error(err, __func__, inode, sci->sc_super);
1710         return err;
1711 }
1712
1713 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1714 {
1715         struct nilfs_segment_buffer *segbuf;
1716         int err;
1717
1718         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1719                 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1720                 if (unlikely(err))
1721                         return err;
1722                 nilfs_segbuf_fill_in_segsum(segbuf);
1723         }
1724         return 0;
1725 }
1726
1727 static int
1728 nilfs_copy_replace_page_buffers(struct page *page, struct list_head *out)
1729 {
1730         struct page *clone_page;
1731         struct buffer_head *bh, *head, *bh2;
1732         void *kaddr;
1733
1734         bh = head = page_buffers(page);
1735
1736         clone_page = nilfs_alloc_private_page(bh->b_bdev, bh->b_size, 0);
1737         if (unlikely(!clone_page))
1738                 return -ENOMEM;
1739
1740         bh2 = page_buffers(clone_page);
1741         kaddr = kmap_atomic(page, KM_USER0);
1742         do {
1743                 if (list_empty(&bh->b_assoc_buffers))
1744                         continue;
1745                 get_bh(bh2);
1746                 page_cache_get(clone_page); /* for each bh */
1747                 memcpy(bh2->b_data, kaddr + bh_offset(bh), bh2->b_size);
1748                 bh2->b_blocknr = bh->b_blocknr;
1749                 list_replace(&bh->b_assoc_buffers, &bh2->b_assoc_buffers);
1750                 list_add_tail(&bh->b_assoc_buffers, out);
1751         } while (bh = bh->b_this_page, bh2 = bh2->b_this_page, bh != head);
1752         kunmap_atomic(kaddr, KM_USER0);
1753
1754         if (!TestSetPageWriteback(clone_page))
1755                 inc_zone_page_state(clone_page, NR_WRITEBACK);
1756         unlock_page(clone_page);
1757
1758         return 0;
1759 }
1760
1761 static int nilfs_test_page_to_be_frozen(struct page *page)
1762 {
1763         struct address_space *mapping = page->mapping;
1764
1765         if (!mapping || !mapping->host || S_ISDIR(mapping->host->i_mode))
1766                 return 0;
1767
1768         if (page_mapped(page)) {
1769                 ClearPageChecked(page);
1770                 return 1;
1771         }
1772         return PageChecked(page);
1773 }
1774
1775 static int nilfs_begin_page_io(struct page *page, struct list_head *out)
1776 {
1777         if (!page || PageWriteback(page))
1778                 /* For split b-tree node pages, this function may be called
1779                    twice.  We ignore the 2nd or later calls by this check. */
1780                 return 0;
1781
1782         lock_page(page);
1783         clear_page_dirty_for_io(page);
1784         set_page_writeback(page);
1785         unlock_page(page);
1786
1787         if (nilfs_test_page_to_be_frozen(page)) {
1788                 int err = nilfs_copy_replace_page_buffers(page, out);
1789                 if (unlikely(err))
1790                         return err;
1791         }
1792         return 0;
1793 }
1794
1795 static int nilfs_segctor_prepare_write(struct nilfs_sc_info *sci,
1796                                        struct page **failed_page)
1797 {
1798         struct nilfs_segment_buffer *segbuf;
1799         struct page *bd_page = NULL, *fs_page = NULL;
1800         struct list_head *list = &sci->sc_copied_buffers;
1801         int err;
1802
1803         *failed_page = NULL;
1804         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1805                 struct buffer_head *bh;
1806
1807                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1808                                     b_assoc_buffers) {
1809                         if (bh->b_page != bd_page) {
1810                                 if (bd_page) {
1811                                         lock_page(bd_page);
1812                                         clear_page_dirty_for_io(bd_page);
1813                                         set_page_writeback(bd_page);
1814                                         unlock_page(bd_page);
1815                                 }
1816                                 bd_page = bh->b_page;
1817                         }
1818                 }
1819
1820                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1821                                     b_assoc_buffers) {
1822                         if (bh == sci->sc_super_root) {
1823                                 if (bh->b_page != bd_page) {
1824                                         lock_page(bd_page);
1825                                         clear_page_dirty_for_io(bd_page);
1826                                         set_page_writeback(bd_page);
1827                                         unlock_page(bd_page);
1828                                         bd_page = bh->b_page;
1829                                 }
1830                                 break;
1831                         }
1832                         if (bh->b_page != fs_page) {
1833                                 err = nilfs_begin_page_io(fs_page, list);
1834                                 if (unlikely(err)) {
1835                                         *failed_page = fs_page;
1836                                         goto out;
1837                                 }
1838                                 fs_page = bh->b_page;
1839                         }
1840                 }
1841         }
1842         if (bd_page) {
1843                 lock_page(bd_page);
1844                 clear_page_dirty_for_io(bd_page);
1845                 set_page_writeback(bd_page);
1846                 unlock_page(bd_page);
1847         }
1848         err = nilfs_begin_page_io(fs_page, list);
1849         if (unlikely(err))
1850                 *failed_page = fs_page;
1851  out:
1852         return err;
1853 }
1854
1855 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1856                                struct backing_dev_info *bdi)
1857 {
1858         struct nilfs_segment_buffer *segbuf;
1859         struct nilfs_write_info wi;
1860         int err, res;
1861
1862         wi.sb = sci->sc_super;
1863         wi.bh_sr = sci->sc_super_root;
1864         wi.bdi = bdi;
1865
1866         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1867                 nilfs_segbuf_prepare_write(segbuf, &wi);
1868                 err = nilfs_segbuf_write(segbuf, &wi);
1869
1870                 res = nilfs_segbuf_wait(segbuf, &wi);
1871                 err = unlikely(err) ? : res;
1872                 if (unlikely(err))
1873                         return err;
1874         }
1875         return 0;
1876 }
1877
1878 static int nilfs_page_has_uncleared_buffer(struct page *page)
1879 {
1880         struct buffer_head *head, *bh;
1881
1882         head = bh = page_buffers(page);
1883         do {
1884                 if (buffer_dirty(bh) && !list_empty(&bh->b_assoc_buffers))
1885                         return 1;
1886                 bh = bh->b_this_page;
1887         } while (bh != head);
1888         return 0;
1889 }
1890
1891 static void __nilfs_end_page_io(struct page *page, int err)
1892 {
1893         /* BUG_ON(err > 0); */
1894         if (!err) {
1895                 if (!nilfs_page_buffers_clean(page))
1896                         __set_page_dirty_nobuffers(page);
1897                 ClearPageError(page);
1898         } else {
1899                 __set_page_dirty_nobuffers(page);
1900                 SetPageError(page);
1901         }
1902
1903         if (buffer_nilfs_allocated(page_buffers(page))) {
1904                 if (TestClearPageWriteback(page))
1905                         dec_zone_page_state(page, NR_WRITEBACK);
1906         } else
1907                 end_page_writeback(page);
1908 }
1909
1910 static void nilfs_end_page_io(struct page *page, int err)
1911 {
1912         if (!page)
1913                 return;
1914
1915         if (buffer_nilfs_node(page_buffers(page)) &&
1916             nilfs_page_has_uncleared_buffer(page))
1917                 /* For b-tree node pages, this function may be called twice
1918                    or more because they might be split in a segment.
1919                    This check assures that cleanup has been done for all
1920                    buffers in a split btnode page. */
1921                 return;
1922
1923         __nilfs_end_page_io(page, err);
1924 }
1925
1926 static void nilfs_clear_copied_buffers(struct list_head *list, int err)
1927 {
1928         struct buffer_head *bh, *head;
1929         struct page *page;
1930
1931         while (!list_empty(list)) {
1932                 bh = list_entry(list->next, struct buffer_head,
1933                                 b_assoc_buffers);
1934                 page = bh->b_page;
1935                 page_cache_get(page);
1936                 head = bh = page_buffers(page);
1937                 do {
1938                         if (!list_empty(&bh->b_assoc_buffers)) {
1939                                 list_del_init(&bh->b_assoc_buffers);
1940                                 if (!err) {
1941                                         set_buffer_uptodate(bh);
1942                                         clear_buffer_dirty(bh);
1943                                         clear_buffer_nilfs_volatile(bh);
1944                                 }
1945                                 brelse(bh); /* for b_assoc_buffers */
1946                         }
1947                 } while ((bh = bh->b_this_page) != head);
1948
1949                 __nilfs_end_page_io(page, err);
1950                 page_cache_release(page);
1951         }
1952 }
1953
1954 static void nilfs_segctor_abort_write(struct nilfs_sc_info *sci,
1955                                       struct page *failed_page, int err)
1956 {
1957         struct nilfs_segment_buffer *segbuf;
1958         struct page *bd_page = NULL, *fs_page = NULL;
1959
1960         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1961                 struct buffer_head *bh;
1962
1963                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1964                                     b_assoc_buffers) {
1965                         if (bh->b_page != bd_page) {
1966                                 if (bd_page)
1967                                         end_page_writeback(bd_page);
1968                                 bd_page = bh->b_page;
1969                         }
1970                 }
1971
1972                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1973                                     b_assoc_buffers) {
1974                         if (bh == sci->sc_super_root) {
1975                                 if (bh->b_page != bd_page) {
1976                                         end_page_writeback(bd_page);
1977                                         bd_page = bh->b_page;
1978                                 }
1979                                 break;
1980                         }
1981                         if (bh->b_page != fs_page) {
1982                                 nilfs_end_page_io(fs_page, err);
1983                                 if (unlikely(fs_page == failed_page))
1984                                         goto done;
1985                                 fs_page = bh->b_page;
1986                         }
1987                 }
1988         }
1989         if (bd_page)
1990                 end_page_writeback(bd_page);
1991
1992         nilfs_end_page_io(fs_page, err);
1993  done:
1994         nilfs_clear_copied_buffers(&sci->sc_copied_buffers, err);
1995 }
1996
1997 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1998                                    struct nilfs_segment_buffer *segbuf)
1999 {
2000         nilfs->ns_segnum = segbuf->sb_segnum;
2001         nilfs->ns_nextnum = segbuf->sb_nextnum;
2002         nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
2003                 + segbuf->sb_sum.nblocks;
2004         nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
2005         nilfs->ns_ctime = segbuf->sb_sum.ctime;
2006 }
2007
2008 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
2009 {
2010         struct nilfs_segment_buffer *segbuf;
2011         struct page *bd_page = NULL, *fs_page = NULL;
2012         struct nilfs_sb_info *sbi = sci->sc_sbi;
2013         struct the_nilfs *nilfs = sbi->s_nilfs;
2014         int update_sr = (sci->sc_super_root != NULL);
2015
2016         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
2017                 struct buffer_head *bh;
2018
2019                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
2020                                     b_assoc_buffers) {
2021                         set_buffer_uptodate(bh);
2022                         clear_buffer_dirty(bh);
2023                         if (bh->b_page != bd_page) {
2024                                 if (bd_page)
2025                                         end_page_writeback(bd_page);
2026                                 bd_page = bh->b_page;
2027                         }
2028                 }
2029                 /*
2030                  * We assume that the buffers which belong to the same page
2031                  * continue over the buffer list.
2032                  * Under this assumption, the last BHs of pages is
2033                  * identifiable by the discontinuity of bh->b_page
2034                  * (page != fs_page).
2035                  *
2036                  * For B-tree node blocks, however, this assumption is not
2037                  * guaranteed.  The cleanup code of B-tree node pages needs
2038                  * special care.
2039                  */
2040                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
2041                                     b_assoc_buffers) {
2042                         set_buffer_uptodate(bh);
2043                         clear_buffer_dirty(bh);
2044                         clear_buffer_nilfs_volatile(bh);
2045                         if (bh == sci->sc_super_root) {
2046                                 if (bh->b_page != bd_page) {
2047                                         end_page_writeback(bd_page);
2048                                         bd_page = bh->b_page;
2049                                 }
2050                                 break;
2051                         }
2052                         if (bh->b_page != fs_page) {
2053                                 nilfs_end_page_io(fs_page, 0);
2054                                 fs_page = bh->b_page;
2055                         }
2056                 }
2057
2058                 if (!NILFS_SEG_SIMPLEX(&segbuf->sb_sum)) {
2059                         if (NILFS_SEG_LOGBGN(&segbuf->sb_sum)) {
2060                                 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
2061                                 sci->sc_lseg_stime = jiffies;
2062                         }
2063                         if (NILFS_SEG_LOGEND(&segbuf->sb_sum))
2064                                 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
2065                 }
2066         }
2067         /*
2068          * Since pages may continue over multiple segment buffers,
2069          * end of the last page must be checked outside of the loop.
2070          */
2071         if (bd_page)
2072                 end_page_writeback(bd_page);
2073
2074         nilfs_end_page_io(fs_page, 0);
2075
2076         nilfs_clear_copied_buffers(&sci->sc_copied_buffers, 0);
2077
2078         nilfs_drop_collected_inodes(&sci->sc_dirty_files);
2079
2080         if (nilfs_doing_gc()) {
2081                 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
2082                 if (update_sr)
2083                         nilfs_commit_gcdat_inode(nilfs);
2084         } else {
2085                 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
2086                 set_nilfs_cond_nongc_write(nilfs);
2087                 wake_up(&nilfs->ns_cleanerd_wq);
2088         }
2089
2090         sci->sc_nblk_inc += sci->sc_nblk_this_inc;
2091
2092         segbuf = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2093         nilfs_set_next_segment(nilfs, segbuf);
2094
2095         if (update_sr) {
2096                 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
2097                                        segbuf->sb_sum.seg_seq, nilfs->ns_cno);
2098
2099                 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2100                 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
2101         } else
2102                 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
2103 }
2104
2105 static int nilfs_segctor_check_in_files(struct nilfs_sc_info *sci,
2106                                         struct nilfs_sb_info *sbi)
2107 {
2108         struct nilfs_inode_info *ii, *n;
2109         __u64 cno = sbi->s_nilfs->ns_cno;
2110
2111         spin_lock(&sbi->s_inode_lock);
2112  retry:
2113         list_for_each_entry_safe(ii, n, &sbi->s_dirty_files, i_dirty) {
2114                 if (!ii->i_bh) {
2115                         struct buffer_head *ibh;
2116                         int err;
2117
2118                         spin_unlock(&sbi->s_inode_lock);
2119                         err = nilfs_ifile_get_inode_block(
2120                                 sbi->s_ifile, ii->vfs_inode.i_ino, &ibh);
2121                         if (unlikely(err)) {
2122                                 nilfs_warning(sbi->s_super, __func__,
2123                                               "failed to get inode block.\n");
2124                                 return err;
2125                         }
2126                         nilfs_mdt_mark_buffer_dirty(ibh);
2127                         nilfs_mdt_mark_dirty(sbi->s_ifile);
2128                         spin_lock(&sbi->s_inode_lock);
2129                         if (likely(!ii->i_bh))
2130                                 ii->i_bh = ibh;
2131                         else
2132                                 brelse(ibh);
2133                         goto retry;
2134                 }
2135                 ii->i_cno = cno;
2136
2137                 clear_bit(NILFS_I_QUEUED, &ii->i_state);
2138                 set_bit(NILFS_I_BUSY, &ii->i_state);
2139                 list_del(&ii->i_dirty);
2140                 list_add_tail(&ii->i_dirty, &sci->sc_dirty_files);
2141         }
2142         spin_unlock(&sbi->s_inode_lock);
2143
2144         NILFS_I(sbi->s_ifile)->i_cno = cno;
2145
2146         return 0;
2147 }
2148
2149 static void nilfs_segctor_check_out_files(struct nilfs_sc_info *sci,
2150                                           struct nilfs_sb_info *sbi)
2151 {
2152         struct nilfs_transaction_info *ti = current->journal_info;
2153         struct nilfs_inode_info *ii, *n;
2154         __u64 cno = sbi->s_nilfs->ns_cno;
2155
2156         spin_lock(&sbi->s_inode_lock);
2157         list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2158                 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2159                     test_bit(NILFS_I_DIRTY, &ii->i_state)) {
2160                         /* The current checkpoint number (=nilfs->ns_cno) is
2161                            changed between check-in and check-out only if the
2162                            super root is written out.  So, we can update i_cno
2163                            for the inodes that remain in the dirty list. */
2164                         ii->i_cno = cno;
2165                         continue;
2166                 }
2167                 clear_bit(NILFS_I_BUSY, &ii->i_state);
2168                 brelse(ii->i_bh);
2169                 ii->i_bh = NULL;
2170                 list_del(&ii->i_dirty);
2171                 list_add_tail(&ii->i_dirty, &ti->ti_garbage);
2172         }
2173         spin_unlock(&sbi->s_inode_lock);
2174 }
2175
2176 /*
2177  * Nasty routines to manipulate active flags on sufile.
2178  * These would be removed in a future release.
2179  */
2180 static void nilfs_segctor_reactivate_segments(struct nilfs_sc_info *sci,
2181                                               struct the_nilfs *nilfs)
2182 {
2183         struct nilfs_segment_buffer *segbuf, *last;
2184         struct nilfs_segment_entry *ent, *n;
2185         struct inode *sufile = nilfs->ns_sufile;
2186         struct list_head *head;
2187
2188         last = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2189         nilfs_for_each_segbuf_before(segbuf, last, &sci->sc_segbufs) {
2190                 ent = segbuf->sb_segent;
2191                 if (!ent)
2192                         break; /* ignore unmapped segments (should check it?)*/
2193                 nilfs_segment_usage_set_active(ent->raw_su);
2194                 nilfs_close_segment_entry(ent, sufile);
2195         }
2196
2197         head = &sci->sc_active_segments;
2198         list_for_each_entry_safe(ent, n, head, list) {
2199                 nilfs_segment_usage_set_active(ent->raw_su);
2200                 nilfs_close_segment_entry(ent, sufile);
2201         }
2202
2203         down_write(&nilfs->ns_sem);
2204         head = &nilfs->ns_used_segments;
2205         list_for_each_entry(ent, head, list) {
2206                 nilfs_segment_usage_set_volatile_active(ent->raw_su);
2207         }
2208         up_write(&nilfs->ns_sem);
2209 }
2210
2211 static int nilfs_segctor_deactivate_segments(struct nilfs_sc_info *sci,
2212                                              struct the_nilfs *nilfs)
2213 {
2214         struct nilfs_segment_buffer *segbuf, *last;
2215         struct nilfs_segment_entry *ent;
2216         struct inode *sufile = nilfs->ns_sufile;
2217         struct list_head *head;
2218         int err;
2219
2220         last = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2221         nilfs_for_each_segbuf_before(segbuf, last, &sci->sc_segbufs) {
2222                 /*
2223                  * Deactivate ongoing full segments.  The last segment is kept
2224                  * active because it is a start point of recovery, and is not
2225                  * relocatable until the super block points to a newer
2226                  * checkpoint.
2227                  */
2228                 ent = segbuf->sb_segent;
2229                 if (!ent)
2230                         break; /* ignore unmapped segments (should check it?)*/
2231                 err = nilfs_open_segment_entry(ent, sufile);
2232                 if (unlikely(err))
2233                         goto failed;
2234                 nilfs_segment_usage_clear_active(ent->raw_su);
2235                 BUG_ON(!buffer_dirty(ent->bh_su));
2236         }
2237
2238         head = &sci->sc_active_segments;
2239         list_for_each_entry(ent, head, list) {
2240                 err = nilfs_open_segment_entry(ent, sufile);
2241                 if (unlikely(err))
2242                         goto failed;
2243                 nilfs_segment_usage_clear_active(ent->raw_su);
2244                 BUG_ON(!buffer_dirty(ent->bh_su));
2245         }
2246
2247         down_write(&nilfs->ns_sem);
2248         head = &nilfs->ns_used_segments;
2249         list_for_each_entry(ent, head, list) {
2250                 /* clear volatile active for segments of older generations */
2251                 nilfs_segment_usage_clear_volatile_active(ent->raw_su);
2252         }
2253         up_write(&nilfs->ns_sem);
2254         return 0;
2255
2256  failed:
2257         nilfs_segctor_reactivate_segments(sci, nilfs);
2258         return err;
2259 }
2260
2261 static void nilfs_segctor_bead_completed_segments(struct nilfs_sc_info *sci)
2262 {
2263         struct nilfs_segment_buffer *segbuf, *last;
2264         struct nilfs_segment_entry *ent;
2265
2266         /* move each segbuf->sb_segent to the list of used active segments */
2267         last = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2268         nilfs_for_each_segbuf_before(segbuf, last, &sci->sc_segbufs) {
2269                 ent = segbuf->sb_segent;
2270                 if (!ent)
2271                         break; /* ignore unmapped segments (should check it?)*/
2272                 list_add_tail(&ent->list, &sci->sc_active_segments);
2273                 segbuf->sb_segent = NULL;
2274         }
2275 }
2276
2277 static void
2278 __nilfs_segctor_commit_deactivate_segments(struct nilfs_sc_info *sci,
2279                                            struct the_nilfs *nilfs)
2280
2281 {
2282         struct nilfs_segment_entry *ent;
2283
2284         list_splice_init(&sci->sc_active_segments,
2285                          nilfs->ns_used_segments.prev);
2286
2287         list_for_each_entry(ent, &nilfs->ns_used_segments, list) {
2288                 nilfs_segment_usage_set_volatile_active(ent->raw_su);
2289                 /* These segments are kept open */
2290         }
2291 }
2292
2293 /*
2294  * Main procedure of segment constructor
2295  */
2296 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2297 {
2298         struct nilfs_sb_info *sbi = sci->sc_sbi;
2299         struct the_nilfs *nilfs = sbi->s_nilfs;
2300         struct page *failed_page;
2301         int err, has_sr = 0;
2302
2303         sci->sc_stage.scnt = NILFS_ST_INIT;
2304
2305         err = nilfs_segctor_check_in_files(sci, sbi);
2306         if (unlikely(err))
2307                 goto out;
2308
2309         if (nilfs_test_metadata_dirty(sbi))
2310                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2311
2312         if (nilfs_segctor_clean(sci))
2313                 goto out;
2314
2315         do {
2316                 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2317
2318                 err = nilfs_segctor_begin_construction(sci, nilfs);
2319                 if (unlikely(err))
2320                         goto out;
2321
2322                 /* Update time stamp */
2323                 sci->sc_seg_ctime = get_seconds();
2324
2325                 err = nilfs_segctor_collect(sci, nilfs, mode);
2326                 if (unlikely(err))
2327                         goto failed;
2328
2329                 has_sr = (sci->sc_super_root != NULL);
2330
2331                 /* Avoid empty segment */
2332                 if (sci->sc_stage.scnt == NILFS_ST_DONE &&
2333                     NILFS_SEG_EMPTY(&sci->sc_curseg->sb_sum)) {
2334                         BUG_ON(mode == SC_LSEG_SR);
2335                         nilfs_segctor_end_construction(sci, nilfs, 1);
2336                         goto out;
2337                 }
2338
2339                 err = nilfs_segctor_assign(sci, mode);
2340                 if (unlikely(err))
2341                         goto failed;
2342
2343                 if (has_sr) {
2344                         err = nilfs_segctor_deactivate_segments(sci, nilfs);
2345                         if (unlikely(err))
2346                                 goto failed;
2347                 }
2348                 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2349                         nilfs_segctor_fill_in_file_bmap(sci, sbi->s_ifile);
2350
2351                 if (has_sr) {
2352                         err = nilfs_segctor_fill_in_checkpoint(sci);
2353                         if (unlikely(err))
2354                                 goto failed_to_make_up;
2355
2356                         nilfs_segctor_fill_in_super_root(sci, nilfs);
2357                 }
2358                 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2359
2360                 /* Write partial segments */
2361                 err = nilfs_segctor_prepare_write(sci, &failed_page);
2362                 if (unlikely(err))
2363                         goto failed_to_write;
2364
2365                 nilfs_segctor_fill_in_checksums(sci, nilfs->ns_crc_seed);
2366
2367                 err = nilfs_segctor_write(sci, nilfs->ns_bdi);
2368                 if (unlikely(err))
2369                         goto failed_to_write;
2370
2371                 nilfs_segctor_complete_write(sci);
2372
2373                 /* Commit segments */
2374                 nilfs_segctor_bead_completed_segments(sci);
2375                 if (has_sr) {
2376                         down_write(&nilfs->ns_sem);
2377                         nilfs_update_last_segment(sbi, 1);
2378                         __nilfs_segctor_commit_deactivate_segments(sci, nilfs);
2379                         up_write(&nilfs->ns_sem);
2380                         nilfs_segctor_commit_free_segments(sci);
2381                         nilfs_segctor_clear_metadata_dirty(sci);
2382                 }
2383
2384                 nilfs_segctor_end_construction(sci, nilfs, 0);
2385
2386         } while (sci->sc_stage.scnt != NILFS_ST_DONE);
2387
2388         /* Clearing sketch data */
2389         if (has_sr && sci->sc_sketch_inode) {
2390                 if (i_size_read(sci->sc_sketch_inode) == 0)
2391                         clear_bit(NILFS_I_DIRTY,
2392                                   &NILFS_I(sci->sc_sketch_inode)->i_state);
2393                 i_size_write(sci->sc_sketch_inode, 0);
2394         }
2395  out:
2396         nilfs_segctor_destroy_segment_buffers(sci);
2397         nilfs_segctor_check_out_files(sci, sbi);
2398         return err;
2399
2400  failed_to_write:
2401         nilfs_segctor_abort_write(sci, failed_page, err);
2402         nilfs_segctor_cancel_segusage(sci, nilfs->ns_sufile);
2403
2404  failed_to_make_up:
2405         if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2406                 nilfs_redirty_inodes(&sci->sc_dirty_files);
2407         if (has_sr)
2408                 nilfs_segctor_reactivate_segments(sci, nilfs);
2409
2410  failed:
2411         if (nilfs_doing_gc())
2412                 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2413         nilfs_segctor_end_construction(sci, nilfs, err);
2414         goto out;
2415 }
2416
2417 /**
2418  * nilfs_secgtor_start_timer - set timer of background write
2419  * @sci: nilfs_sc_info
2420  *
2421  * If the timer has already been set, it ignores the new request.
2422  * This function MUST be called within a section locking the segment
2423  * semaphore.
2424  */
2425 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2426 {
2427         spin_lock(&sci->sc_state_lock);
2428         if (sci->sc_timer && !(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2429                 sci->sc_timer->expires = jiffies + sci->sc_interval;
2430                 add_timer(sci->sc_timer);
2431                 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2432         }
2433         spin_unlock(&sci->sc_state_lock);
2434 }
2435
2436 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2437 {
2438         spin_lock(&sci->sc_state_lock);
2439         if (!(sci->sc_flush_request & (1 << bn))) {
2440                 unsigned long prev_req = sci->sc_flush_request;
2441
2442                 sci->sc_flush_request |= (1 << bn);
2443                 if (!prev_req)
2444                         wake_up(&sci->sc_wait_daemon);
2445         }
2446         spin_unlock(&sci->sc_state_lock);
2447 }
2448
2449 /**
2450  * nilfs_flush_segment - trigger a segment construction for resource control
2451  * @sb: super block
2452  * @ino: inode number of the file to be flushed out.
2453  */
2454 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2455 {
2456         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2457         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2458
2459         if (!sci || nilfs_doing_construction())
2460                 return;
2461         nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2462                                         /* assign bit 0 to data files */
2463 }
2464
2465 int nilfs_segctor_add_segments_to_be_freed(struct nilfs_sc_info *sci,
2466                                            __u64 *segnum, size_t nsegs)
2467 {
2468         struct nilfs_segment_entry *ent;
2469         struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
2470         struct inode *sufile = nilfs->ns_sufile;
2471         LIST_HEAD(list);
2472         __u64 *pnum;
2473         const char *flag_name;
2474         size_t i;
2475         int err, err2 = 0;
2476
2477         for (pnum = segnum, i = 0; i < nsegs; pnum++, i++) {
2478                 ent = nilfs_alloc_segment_entry(*pnum);
2479                 if (unlikely(!ent)) {
2480                         err = -ENOMEM;
2481                         goto failed;
2482                 }
2483                 list_add_tail(&ent->list, &list);
2484
2485                 err = nilfs_open_segment_entry(ent, sufile);
2486                 if (unlikely(err))
2487                         goto failed;
2488
2489                 if (unlikely(le32_to_cpu(ent->raw_su->su_flags) !=
2490                              (1UL << NILFS_SEGMENT_USAGE_DIRTY))) {
2491                         if (nilfs_segment_usage_clean(ent->raw_su))
2492                                 flag_name = "clean";
2493                         else if (nilfs_segment_usage_active(ent->raw_su))
2494                                 flag_name = "active";
2495                         else if (nilfs_segment_usage_volatile_active(
2496                                          ent->raw_su))
2497                                 flag_name = "volatile active";
2498                         else if (!nilfs_segment_usage_dirty(ent->raw_su))
2499                                 flag_name = "non-dirty";
2500                         else
2501                                 flag_name = "erroneous";
2502
2503                         printk(KERN_ERR
2504                                "NILFS: %s segment is requested to be cleaned "
2505                                "(segnum=%llu)\n",
2506                                flag_name, (unsigned long long)ent->segnum);
2507                         err2 = -EINVAL;
2508                 }
2509                 nilfs_close_segment_entry(ent, sufile);
2510         }
2511         if (unlikely(err2)) {
2512                 err = err2;
2513                 goto failed;
2514         }
2515         list_splice(&list, sci->sc_cleaning_segments.prev);
2516         return 0;
2517
2518  failed:
2519         nilfs_dispose_segment_list(&list);
2520         return err;
2521 }
2522
2523 void nilfs_segctor_clear_segments_to_be_freed(struct nilfs_sc_info *sci)
2524 {
2525         nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
2526 }
2527
2528 struct nilfs_segctor_wait_request {
2529         wait_queue_t    wq;
2530         __u32           seq;
2531         int             err;
2532         atomic_t        done;
2533 };
2534
2535 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2536 {
2537         struct nilfs_segctor_wait_request wait_req;
2538         int err = 0;
2539
2540         spin_lock(&sci->sc_state_lock);
2541         init_wait(&wait_req.wq);
2542         wait_req.err = 0;
2543         atomic_set(&wait_req.done, 0);
2544         wait_req.seq = ++sci->sc_seq_request;
2545         spin_unlock(&sci->sc_state_lock);
2546
2547         init_waitqueue_entry(&wait_req.wq, current);
2548         add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2549         set_current_state(TASK_INTERRUPTIBLE);
2550         wake_up(&sci->sc_wait_daemon);
2551
2552         for (;;) {
2553                 if (atomic_read(&wait_req.done)) {
2554                         err = wait_req.err;
2555                         break;
2556                 }
2557                 if (!signal_pending(current)) {
2558                         schedule();
2559                         continue;
2560                 }
2561                 err = -ERESTARTSYS;
2562                 break;
2563         }
2564         finish_wait(&sci->sc_wait_request, &wait_req.wq);
2565         return err;
2566 }
2567
2568 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2569 {
2570         struct nilfs_segctor_wait_request *wrq, *n;
2571         unsigned long flags;
2572
2573         spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2574         list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2575                                  wq.task_list) {
2576                 if (!atomic_read(&wrq->done) &&
2577                     nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2578                         wrq->err = err;
2579                         atomic_set(&wrq->done, 1);
2580                 }
2581                 if (atomic_read(&wrq->done)) {
2582                         wrq->wq.func(&wrq->wq,
2583                                      TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2584                                      0, NULL);
2585                 }
2586         }
2587         spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2588 }
2589
2590 /**
2591  * nilfs_construct_segment - construct a logical segment
2592  * @sb: super block
2593  *
2594  * Return Value: On success, 0 is retured. On errors, one of the following
2595  * negative error code is returned.
2596  *
2597  * %-EROFS - Read only filesystem.
2598  *
2599  * %-EIO - I/O error
2600  *
2601  * %-ENOSPC - No space left on device (only in a panic state).
2602  *
2603  * %-ERESTARTSYS - Interrupted.
2604  *
2605  * %-ENOMEM - Insufficient memory available.
2606  */
2607 int nilfs_construct_segment(struct super_block *sb)
2608 {
2609         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2610         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2611         struct nilfs_transaction_info *ti;
2612         int err;
2613
2614         if (!sci)
2615                 return -EROFS;
2616
2617         /* A call inside transactions causes a deadlock. */
2618         BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2619
2620         err = nilfs_segctor_sync(sci);
2621         return err;
2622 }
2623
2624 /**
2625  * nilfs_construct_dsync_segment - construct a data-only logical segment
2626  * @sb: super block
2627  * @inode: the inode whose data blocks should be written out
2628  *
2629  * Return Value: On success, 0 is retured. On errors, one of the following
2630  * negative error code is returned.
2631  *
2632  * %-EROFS - Read only filesystem.
2633  *
2634  * %-EIO - I/O error
2635  *
2636  * %-ENOSPC - No space left on device (only in a panic state).
2637  *
2638  * %-ERESTARTSYS - Interrupted.
2639  *
2640  * %-ENOMEM - Insufficient memory available.
2641  */
2642 int nilfs_construct_dsync_segment(struct super_block *sb,
2643                                   struct inode *inode)
2644 {
2645         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2646         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2647         struct nilfs_inode_info *ii;
2648         struct nilfs_transaction_info ti;
2649         int err = 0;
2650
2651         if (!sci)
2652                 return -EROFS;
2653
2654         nilfs_transaction_lock(sbi, &ti, 0);
2655
2656         ii = NILFS_I(inode);
2657         if (test_bit(NILFS_I_INODE_DIRTY, &ii->i_state) ||
2658             nilfs_test_opt(sbi, STRICT_ORDER) ||
2659             test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2660             nilfs_discontinued(sbi->s_nilfs)) {
2661                 nilfs_transaction_unlock(sbi);
2662                 err = nilfs_segctor_sync(sci);
2663                 return err;
2664         }
2665
2666         spin_lock(&sbi->s_inode_lock);
2667         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2668             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2669                 spin_unlock(&sbi->s_inode_lock);
2670                 nilfs_transaction_unlock(sbi);
2671                 return 0;
2672         }
2673         spin_unlock(&sbi->s_inode_lock);
2674         sci->sc_stage.dirty_file_ptr = ii;
2675
2676         err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2677
2678         nilfs_transaction_unlock(sbi);
2679         return err;
2680 }
2681
2682 struct nilfs_segctor_req {
2683         int mode;
2684         __u32 seq_accepted;
2685         int sc_err;  /* construction failure */
2686         int sb_err;  /* super block writeback failure */
2687 };
2688
2689 #define FLUSH_FILE_BIT  (0x1) /* data file only */
2690 #define FLUSH_DAT_BIT   (1 << NILFS_DAT_INO) /* DAT only */
2691
2692 static void nilfs_segctor_accept(struct nilfs_sc_info *sci,
2693                                  struct nilfs_segctor_req *req)
2694 {
2695         BUG_ON(!sci);
2696
2697         req->sc_err = req->sb_err = 0;
2698         spin_lock(&sci->sc_state_lock);
2699         req->seq_accepted = sci->sc_seq_request;
2700         spin_unlock(&sci->sc_state_lock);
2701
2702         if (sci->sc_timer)
2703                 del_timer_sync(sci->sc_timer);
2704 }
2705
2706 static void nilfs_segctor_notify(struct nilfs_sc_info *sci,
2707                                  struct nilfs_segctor_req *req)
2708 {
2709         /* Clear requests (even when the construction failed) */
2710         spin_lock(&sci->sc_state_lock);
2711
2712         sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2713
2714         if (req->mode == SC_LSEG_SR) {
2715                 sci->sc_seq_done = req->seq_accepted;
2716                 nilfs_segctor_wakeup(sci, req->sc_err ? : req->sb_err);
2717                 sci->sc_flush_request = 0;
2718         } else if (req->mode == SC_FLUSH_FILE)
2719                 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2720         else if (req->mode == SC_FLUSH_DAT)
2721                 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2722
2723         spin_unlock(&sci->sc_state_lock);
2724 }
2725
2726 static int nilfs_segctor_construct(struct nilfs_sc_info *sci,
2727                                    struct nilfs_segctor_req *req)
2728 {
2729         struct nilfs_sb_info *sbi = sci->sc_sbi;
2730         struct the_nilfs *nilfs = sbi->s_nilfs;
2731         int err = 0;
2732
2733         if (nilfs_discontinued(nilfs))
2734                 req->mode = SC_LSEG_SR;
2735         if (!nilfs_segctor_confirm(sci)) {
2736                 err = nilfs_segctor_do_construct(sci, req->mode);
2737                 req->sc_err = err;
2738         }
2739         if (likely(!err)) {
2740                 if (req->mode != SC_FLUSH_DAT)
2741                         atomic_set(&nilfs->ns_ndirtyblks, 0);
2742                 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2743                     nilfs_discontinued(nilfs)) {
2744                         down_write(&nilfs->ns_sem);
2745                         req->sb_err = nilfs_commit_super(sbi);
2746                         up_write(&nilfs->ns_sem);
2747                 }
2748         }
2749         return err;
2750 }
2751
2752 static void nilfs_construction_timeout(unsigned long data)
2753 {
2754         struct task_struct *p = (struct task_struct *)data;
2755         wake_up_process(p);
2756 }
2757
2758 static void
2759 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2760 {
2761         struct nilfs_inode_info *ii, *n;
2762
2763         list_for_each_entry_safe(ii, n, head, i_dirty) {
2764                 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2765                         continue;
2766                 hlist_del_init(&ii->vfs_inode.i_hash);
2767                 list_del_init(&ii->i_dirty);
2768                 nilfs_clear_gcinode(&ii->vfs_inode);
2769         }
2770 }
2771
2772 int nilfs_clean_segments(struct super_block *sb, void __user *argp)
2773 {
2774         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2775         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2776         struct the_nilfs *nilfs = sbi->s_nilfs;
2777         struct nilfs_transaction_info ti;
2778         struct nilfs_segctor_req req = { .mode = SC_LSEG_SR };
2779         int err;
2780
2781         if (unlikely(!sci))
2782                 return -EROFS;
2783
2784         nilfs_transaction_lock(sbi, &ti, 1);
2785
2786         err = nilfs_init_gcdat_inode(nilfs);
2787         if (unlikely(err))
2788                 goto out_unlock;
2789         err = nilfs_ioctl_prepare_clean_segments(nilfs, argp);
2790         if (unlikely(err))
2791                 goto out_unlock;
2792
2793         list_splice_init(&nilfs->ns_gc_inodes, sci->sc_gc_inodes.prev);
2794
2795         for (;;) {
2796                 nilfs_segctor_accept(sci, &req);
2797                 err = nilfs_segctor_construct(sci, &req);
2798                 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2799                 nilfs_segctor_notify(sci, &req);
2800
2801                 if (likely(!err))
2802                         break;
2803
2804                 nilfs_warning(sb, __func__,
2805                               "segment construction failed. (err=%d)", err);
2806                 set_current_state(TASK_INTERRUPTIBLE);
2807                 schedule_timeout(sci->sc_interval);
2808         }
2809
2810  out_unlock:
2811         nilfs_clear_gcdat_inode(nilfs);
2812         nilfs_transaction_unlock(sbi);
2813         return err;
2814 }
2815
2816 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2817 {
2818         struct nilfs_sb_info *sbi = sci->sc_sbi;
2819         struct nilfs_transaction_info ti;
2820         struct nilfs_segctor_req req = { .mode = mode };
2821
2822         nilfs_transaction_lock(sbi, &ti, 0);
2823
2824         nilfs_segctor_accept(sci, &req);
2825         nilfs_segctor_construct(sci, &req);
2826         nilfs_segctor_notify(sci, &req);
2827
2828         /*
2829          * Unclosed segment should be retried.  We do this using sc_timer.
2830          * Timeout of sc_timer will invoke complete construction which leads
2831          * to close the current logical segment.
2832          */
2833         if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2834                 nilfs_segctor_start_timer(sci);
2835
2836         nilfs_transaction_unlock(sbi);
2837 }
2838
2839 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2840 {
2841         int mode = 0;
2842         int err;
2843
2844         spin_lock(&sci->sc_state_lock);
2845         mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2846                 SC_FLUSH_DAT : SC_FLUSH_FILE;
2847         spin_unlock(&sci->sc_state_lock);
2848
2849         if (mode) {
2850                 err = nilfs_segctor_do_construct(sci, mode);
2851
2852                 spin_lock(&sci->sc_state_lock);
2853                 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2854                         ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2855                 spin_unlock(&sci->sc_state_lock);
2856         }
2857         clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2858 }
2859
2860 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2861 {
2862         if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2863             time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2864                 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2865                         return SC_FLUSH_FILE;
2866                 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2867                         return SC_FLUSH_DAT;
2868         }
2869         return SC_LSEG_SR;
2870 }
2871
2872 /**
2873  * nilfs_segctor_thread - main loop of the segment constructor thread.
2874  * @arg: pointer to a struct nilfs_sc_info.
2875  *
2876  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2877  * to execute segment constructions.
2878  */
2879 static int nilfs_segctor_thread(void *arg)
2880 {
2881         struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2882         struct timer_list timer;
2883         int timeout = 0;
2884
2885         init_timer(&timer);
2886         timer.data = (unsigned long)current;
2887         timer.function = nilfs_construction_timeout;
2888         sci->sc_timer = &timer;
2889
2890         /* start sync. */
2891         sci->sc_task = current;
2892         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2893         printk(KERN_INFO
2894                "segctord starting. Construction interval = %lu seconds, "
2895                "CP frequency < %lu seconds\n",
2896                sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2897
2898         spin_lock(&sci->sc_state_lock);
2899  loop:
2900         for (;;) {
2901                 int mode;
2902
2903                 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2904                         goto end_thread;
2905
2906                 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2907                         mode = SC_LSEG_SR;
2908                 else if (!sci->sc_flush_request)
2909                         break;
2910                 else
2911                         mode = nilfs_segctor_flush_mode(sci);
2912
2913                 spin_unlock(&sci->sc_state_lock);
2914                 nilfs_segctor_thread_construct(sci, mode);
2915                 spin_lock(&sci->sc_state_lock);
2916                 timeout = 0;
2917         }
2918
2919
2920         if (freezing(current)) {
2921                 spin_unlock(&sci->sc_state_lock);
2922                 refrigerator();
2923                 spin_lock(&sci->sc_state_lock);
2924         } else {
2925                 DEFINE_WAIT(wait);
2926                 int should_sleep = 1;
2927
2928                 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2929                                 TASK_INTERRUPTIBLE);
2930
2931                 if (sci->sc_seq_request != sci->sc_seq_done)
2932                         should_sleep = 0;
2933                 else if (sci->sc_flush_request)
2934                         should_sleep = 0;
2935                 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2936                         should_sleep = time_before(jiffies,
2937                                                    sci->sc_timer->expires);
2938
2939                 if (should_sleep) {
2940                         spin_unlock(&sci->sc_state_lock);
2941                         schedule();
2942                         spin_lock(&sci->sc_state_lock);
2943                 }
2944                 finish_wait(&sci->sc_wait_daemon, &wait);
2945                 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2946                            time_after_eq(jiffies, sci->sc_timer->expires));
2947         }
2948         goto loop;
2949
2950  end_thread:
2951         spin_unlock(&sci->sc_state_lock);
2952         del_timer_sync(sci->sc_timer);
2953         sci->sc_timer = NULL;
2954
2955         /* end sync. */
2956         sci->sc_task = NULL;
2957         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2958         return 0;
2959 }
2960
2961 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2962 {
2963         struct task_struct *t;
2964
2965         t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2966         if (IS_ERR(t)) {
2967                 int err = PTR_ERR(t);
2968
2969                 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2970                        err);
2971                 return err;
2972         }
2973         wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2974         return 0;
2975 }
2976
2977 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2978 {
2979         sci->sc_state |= NILFS_SEGCTOR_QUIT;
2980
2981         while (sci->sc_task) {
2982                 wake_up(&sci->sc_wait_daemon);
2983                 spin_unlock(&sci->sc_state_lock);
2984                 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2985                 spin_lock(&sci->sc_state_lock);
2986         }
2987 }
2988
2989 static int nilfs_segctor_init(struct nilfs_sc_info *sci,
2990                               struct nilfs_recovery_info *ri)
2991 {
2992         int err;
2993         struct inode *inode = nilfs_iget(sci->sc_super, NILFS_SKETCH_INO);
2994
2995         sci->sc_sketch_inode = IS_ERR(inode) ? NULL : inode;
2996         if (sci->sc_sketch_inode)
2997                 i_size_write(sci->sc_sketch_inode, 0);
2998
2999         sci->sc_seq_done = sci->sc_seq_request;
3000         if (ri)
3001                 list_splice_init(&ri->ri_used_segments,
3002                                  sci->sc_active_segments.prev);
3003
3004         err = nilfs_segctor_start_thread(sci);
3005         if (err) {
3006                 if (ri)
3007                         list_splice_init(&sci->sc_active_segments,
3008                                          ri->ri_used_segments.prev);
3009                 if (sci->sc_sketch_inode) {
3010                         iput(sci->sc_sketch_inode);
3011                         sci->sc_sketch_inode = NULL;
3012                 }
3013         }
3014         return err;
3015 }
3016
3017 /*
3018  * Setup & clean-up functions
3019  */
3020 static struct nilfs_sc_info *nilfs_segctor_new(struct nilfs_sb_info *sbi)
3021 {
3022         struct nilfs_sc_info *sci;
3023
3024         sci = kzalloc(sizeof(*sci), GFP_KERNEL);
3025         if (!sci)
3026                 return NULL;
3027
3028         sci->sc_sbi = sbi;
3029         sci->sc_super = sbi->s_super;
3030
3031         init_waitqueue_head(&sci->sc_wait_request);
3032         init_waitqueue_head(&sci->sc_wait_daemon);
3033         init_waitqueue_head(&sci->sc_wait_task);
3034         spin_lock_init(&sci->sc_state_lock);
3035         INIT_LIST_HEAD(&sci->sc_dirty_files);
3036         INIT_LIST_HEAD(&sci->sc_segbufs);
3037         INIT_LIST_HEAD(&sci->sc_gc_inodes);
3038         INIT_LIST_HEAD(&sci->sc_active_segments);
3039         INIT_LIST_HEAD(&sci->sc_cleaning_segments);
3040         INIT_LIST_HEAD(&sci->sc_copied_buffers);
3041
3042         sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
3043         sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
3044         sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
3045
3046         if (sbi->s_interval)
3047                 sci->sc_interval = sbi->s_interval;
3048         if (sbi->s_watermark)
3049                 sci->sc_watermark = sbi->s_watermark;
3050         return sci;
3051 }
3052
3053 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
3054 {
3055         int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
3056
3057         /* The segctord thread was stopped and its timer was removed.
3058            But some tasks remain. */
3059         do {
3060                 struct nilfs_sb_info *sbi = sci->sc_sbi;
3061                 struct nilfs_transaction_info ti;
3062                 struct nilfs_segctor_req req = { .mode = SC_LSEG_SR };
3063
3064                 nilfs_transaction_lock(sbi, &ti, 0);
3065                 nilfs_segctor_accept(sci, &req);
3066                 ret = nilfs_segctor_construct(sci, &req);
3067                 nilfs_segctor_notify(sci, &req);
3068                 nilfs_transaction_unlock(sbi);
3069
3070         } while (ret && retrycount-- > 0);
3071 }
3072
3073 /**
3074  * nilfs_segctor_destroy - destroy the segment constructor.
3075  * @sci: nilfs_sc_info
3076  *
3077  * nilfs_segctor_destroy() kills the segctord thread and frees
3078  * the nilfs_sc_info struct.
3079  * Caller must hold the segment semaphore.
3080  */
3081 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
3082 {
3083         struct nilfs_sb_info *sbi = sci->sc_sbi;
3084         int flag;
3085
3086         up_write(&sbi->s_nilfs->ns_segctor_sem);
3087
3088         spin_lock(&sci->sc_state_lock);
3089         nilfs_segctor_kill_thread(sci);
3090         flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
3091                 || sci->sc_seq_request != sci->sc_seq_done);
3092         spin_unlock(&sci->sc_state_lock);
3093
3094         if (flag || nilfs_segctor_confirm(sci))
3095                 nilfs_segctor_write_out(sci);
3096
3097         BUG_ON(!list_empty(&sci->sc_copied_buffers));
3098
3099         if (!list_empty(&sci->sc_dirty_files)) {
3100                 nilfs_warning(sbi->s_super, __func__,
3101                               "dirty file(s) after the final construction\n");
3102                 nilfs_dispose_list(sbi, &sci->sc_dirty_files, 1);
3103         }
3104         if (!list_empty(&sci->sc_active_segments))
3105                 nilfs_dispose_segment_list(&sci->sc_active_segments);
3106
3107         if (!list_empty(&sci->sc_cleaning_segments))
3108                 nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
3109
3110         BUG_ON(!list_empty(&sci->sc_segbufs));
3111
3112         if (sci->sc_sketch_inode) {
3113                 iput(sci->sc_sketch_inode);
3114                 sci->sc_sketch_inode = NULL;
3115         }
3116         down_write(&sbi->s_nilfs->ns_segctor_sem);
3117
3118         kfree(sci);
3119 }
3120
3121 /**
3122  * nilfs_attach_segment_constructor - attach a segment constructor
3123  * @sbi: nilfs_sb_info
3124  * @ri: nilfs_recovery_info
3125  *
3126  * nilfs_attach_segment_constructor() allocates a struct nilfs_sc_info,
3127  * initilizes it, and starts the segment constructor.
3128  *
3129  * Return Value: On success, 0 is returned. On error, one of the following
3130  * negative error code is returned.
3131  *
3132  * %-ENOMEM - Insufficient memory available.
3133  */
3134 int nilfs_attach_segment_constructor(struct nilfs_sb_info *sbi,
3135                                      struct nilfs_recovery_info *ri)
3136 {
3137         struct the_nilfs *nilfs = sbi->s_nilfs;
3138         int err;
3139
3140         /* Each field of nilfs_segctor is cleared through the initialization
3141            of super-block info */
3142         sbi->s_sc_info = nilfs_segctor_new(sbi);
3143         if (!sbi->s_sc_info)
3144                 return -ENOMEM;
3145
3146         nilfs_attach_writer(nilfs, sbi);
3147         err = nilfs_segctor_init(NILFS_SC(sbi), ri);
3148         if (err) {
3149                 nilfs_detach_writer(nilfs, sbi);
3150                 kfree(sbi->s_sc_info);
3151                 sbi->s_sc_info = NULL;
3152         }
3153         return err;
3154 }
3155
3156 /**
3157  * nilfs_detach_segment_constructor - destroy the segment constructor
3158  * @sbi: nilfs_sb_info
3159  *
3160  * nilfs_detach_segment_constructor() kills the segment constructor daemon,
3161  * frees the struct nilfs_sc_info, and destroy the dirty file list.
3162  */
3163 void nilfs_detach_segment_constructor(struct nilfs_sb_info *sbi)
3164 {
3165         struct the_nilfs *nilfs = sbi->s_nilfs;
3166         LIST_HEAD(garbage_list);
3167
3168         down_write(&nilfs->ns_segctor_sem);
3169         if (NILFS_SC(sbi)) {
3170                 nilfs_segctor_destroy(NILFS_SC(sbi));
3171                 sbi->s_sc_info = NULL;
3172         }
3173
3174         /* Force to free the list of dirty files */
3175         spin_lock(&sbi->s_inode_lock);
3176         if (!list_empty(&sbi->s_dirty_files)) {
3177                 list_splice_init(&sbi->s_dirty_files, &garbage_list);
3178                 nilfs_warning(sbi->s_super, __func__,
3179                               "Non empty dirty list after the last "
3180                               "segment construction\n");
3181         }
3182         spin_unlock(&sbi->s_inode_lock);
3183         up_write(&nilfs->ns_segctor_sem);
3184
3185         nilfs_dispose_list(sbi, &garbage_list, 1);
3186         nilfs_detach_writer(nilfs, sbi);
3187 }