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 "segbuf.h"
  43
  44
  45 /*
  46  * Segment constructor
  47  */
  48 #define SC_N_INODEVEC   16   /* Size of locally allocated inode vector */
  49
  50 #define SC_MAX_SEGDELTA 64   /* Upper limit of the number of segments
  51                                 appended in collection retry loop */
  52
  53 /* Construction mode */
  54 enum {
  55         SC_LSEG_SR = 1, /* Make a logical segment having a super root */
  56         SC_LSEG_DSYNC,  /* Flush data blocks of a given file and make
  57                            a logical segment without a super root */
  58         SC_FLUSH_FILE,  /* Flush data files, leads to segment writes without
  59                            creating a checkpoint */
  60         SC_FLUSH_DAT,   /* Flush DAT file. This also creates segments without
  61                            a checkpoint */
  62 };
  63
  64 /* Stage numbers of dirty block collection */
  65 enum {
  66         NILFS_ST_INIT = 0,
  67         NILFS_ST_GC,            /* Collecting dirty blocks for GC */
  68         NILFS_ST_FILE,
  69         NILFS_ST_IFILE,
  70         NILFS_ST_CPFILE,
  71         NILFS_ST_SUFILE,
  72         NILFS_ST_DAT,
  73         NILFS_ST_SR,            /* Super root */
  74         NILFS_ST_DSYNC,         /* Data sync blocks */
  75         NILFS_ST_DONE,
  76 };
  77
  78 /* State flags of collection */
  79 #define NILFS_CF_NODE           0x0001  /* Collecting node blocks */
  80 #define NILFS_CF_IFILE_STARTED  0x0002  /* IFILE stage has started */
  81 #define NILFS_CF_SUFREED        0x0004  /* segment usages has been freed */
  82 #define NILFS_CF_HISTORY_MASK   (NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
  83
  84 /* Operations depending on the construction mode and file type */
  85 struct nilfs_sc_operations {
  86         int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
  87                             struct inode *);
  88         int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
  89                             struct inode *);
  90         int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
  91                             struct inode *);
  92         void (*write_data_binfo)(struct nilfs_sc_info *,
  93                                  struct nilfs_segsum_pointer *,
  94                                  union nilfs_binfo *);
  95         void (*write_node_binfo)(struct nilfs_sc_info *,
  96                                  struct nilfs_segsum_pointer *,
  97                                  union nilfs_binfo *);
  98 };
  99
 100 /*
 101  * Other definitions
 102  */
 103 static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
 104 static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
 105 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
 106 static void nilfs_dispose_list(struct nilfs_sb_info *, struct list_head *,
 107                                int);
 108
 109 #define nilfs_cnt32_gt(a, b)   \
 110         (typecheck(__u32, a) && typecheck(__u32, b) && \
 111          ((__s32)(b) - (__s32)(a) < 0))
 112 #define nilfs_cnt32_ge(a, b)   \
 113         (typecheck(__u32, a) && typecheck(__u32, b) && \
 114          ((__s32)(a) - (__s32)(b) >= 0))
 115 #define nilfs_cnt32_lt(a, b)  nilfs_cnt32_gt(b, a)
 116 #define nilfs_cnt32_le(a, b)  nilfs_cnt32_ge(b, a)
 117
 118 /*
 119  * Transaction
 120  */
 121 static struct kmem_cache *nilfs_transaction_cachep;
 122
 123 /**
 124  * nilfs_init_transaction_cache - create a cache for nilfs_transaction_info
 125  *
 126  * nilfs_init_transaction_cache() creates a slab cache for the struct
 127  * nilfs_transaction_info.
 128  *
 129  * Return Value: On success, it returns 0. On error, one of the following
 130  * negative error code is returned.
 131  *
 132  * %-ENOMEM - Insufficient memory available.
 133  */
 134 int nilfs_init_transaction_cache(void)
 135 {
 136         nilfs_transaction_cachep =
 137                 kmem_cache_create("nilfs2_transaction_cache",
 138                                   sizeof(struct nilfs_transaction_info),
 139                                   0, SLAB_RECLAIM_ACCOUNT, NULL);
 140         return (nilfs_transaction_cachep == NULL) ? -ENOMEM : 0;
 141 }
 142
 143 /**
 144  * nilfs_detroy_transaction_cache - destroy the cache for transaction info
 145  *
 146  * nilfs_destroy_transaction_cache() frees the slab cache for the struct
 147  * nilfs_transaction_info.
 148  */
 149 void nilfs_destroy_transaction_cache(void)
 150 {
 151         kmem_cache_destroy(nilfs_transaction_cachep);
 152 }
 153
 154 static int nilfs_prepare_segment_lock(struct nilfs_transaction_info *ti)
 155 {
 156         struct nilfs_transaction_info *cur_ti = current->journal_info;
 157         void *save = NULL;
 158
 159         if (cur_ti) {
 160                 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
 161                         return ++cur_ti->ti_count;
 162                 else {
 163                         /*
 164                          * If journal_info field is occupied by other FS,
 165                          * it is saved and will be restored on
 166                          * nilfs_transaction_commit().
 167                          */
 168                         printk(KERN_WARNING
 169                                "NILFS warning: journal info from a different "
 170                                "FS\n");
 171                         save = current->journal_info;
 172                 }
 173         }
 174         if (!ti) {
 175                 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
 176                 if (!ti)
 177                         return -ENOMEM;
 178                 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
 179         } else {
 180                 ti->ti_flags = 0;
 181         }
 182         ti->ti_count = 0;
 183         ti->ti_save = save;
 184         ti->ti_magic = NILFS_TI_MAGIC;
 185         current->journal_info = ti;
 186         return 0;
 187 }
 188
 189 /**
 190  * nilfs_transaction_begin - start indivisible file operations.
 191  * @sb: super block
 192  * @ti: nilfs_transaction_info
 193  * @vacancy_check: flags for vacancy rate checks
 194  *
 195  * nilfs_transaction_begin() acquires a reader/writer semaphore, called
 196  * the segment semaphore, to make a segment construction and write tasks
 197  * exclusive.  The function is used with nilfs_transaction_commit() in pairs.
 198  * The region enclosed by these two functions can be nested.  To avoid a
 199  * deadlock, the semaphore is only acquired or released in the outermost call.
 200  *
 201  * This function allocates a nilfs_transaction_info struct to keep context
 202  * information on it.  It is initialized and hooked onto the current task in
 203  * the outermost call.  If a pre-allocated struct is given to @ti, it is used
 204  * instead; othewise a new struct is assigned from a slab.
 205  *
 206  * When @vacancy_check flag is set, this function will check the amount of
 207  * free space, and will wait for the GC to reclaim disk space if low capacity.
 208  *
 209  * Return Value: On success, 0 is returned. On error, one of the following
 210  * negative error code is returned.
 211  *
 212  * %-ENOMEM - Insufficient memory available.
 213  *
 214  * %-ENOSPC - No space left on device
 215  */
 216 int nilfs_transaction_begin(struct super_block *sb,
 217                             struct nilfs_transaction_info *ti,
 218                             int vacancy_check)
 219 {
 220         struct nilfs_sb_info *sbi;
 221         struct the_nilfs *nilfs;
 222         int ret = nilfs_prepare_segment_lock(ti);
 223
 224         if (unlikely(ret < 0))
 225                 return ret;
 226         if (ret > 0)
 227                 return 0;
 228
 229         sbi = NILFS_SB(sb);
 230         nilfs = sbi->s_nilfs;
 231         down_read(&nilfs->ns_segctor_sem);
 232         if (vacancy_check && nilfs_near_disk_full(nilfs)) {
 233                 up_read(&nilfs->ns_segctor_sem);
 234                 ret = -ENOSPC;
 235                 goto failed;
 236         }
 237         return 0;
 238
 239  failed:
 240         ti = current->journal_info;
 241         current->journal_info = ti->ti_save;
 242         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 243                 kmem_cache_free(nilfs_transaction_cachep, ti);
 244         return ret;
 245 }
 246
 247 /**
 248  * nilfs_transaction_commit - commit indivisible file operations.
 249  * @sb: super block
 250  *
 251  * nilfs_transaction_commit() releases the read semaphore which is
 252  * acquired by nilfs_transaction_begin(). This is only performed
 253  * in outermost call of this function.  If a commit flag is set,
 254  * nilfs_transaction_commit() sets a timer to start the segment
 255  * constructor.  If a sync flag is set, it starts construction
 256  * directly.
 257  */
 258 int nilfs_transaction_commit(struct super_block *sb)
 259 {
 260         struct nilfs_transaction_info *ti = current->journal_info;
 261         struct nilfs_sb_info *sbi;
 262         struct nilfs_sc_info *sci;
 263         int err = 0;
 264
 265         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 266         ti->ti_flags |= NILFS_TI_COMMIT;
 267         if (ti->ti_count > 0) {
 268                 ti->ti_count--;
 269                 return 0;
 270         }
 271         sbi = NILFS_SB(sb);
 272         sci = NILFS_SC(sbi);
 273         if (sci != NULL) {
 274                 if (ti->ti_flags & NILFS_TI_COMMIT)
 275                         nilfs_segctor_start_timer(sci);
 276                 if (atomic_read(&sbi->s_nilfs->ns_ndirtyblks) >
 277                     sci->sc_watermark)
 278                         nilfs_segctor_do_flush(sci, 0);
 279         }
 280         up_read(&sbi->s_nilfs->ns_segctor_sem);
 281         current->journal_info = ti->ti_save;
 282
 283         if (ti->ti_flags & NILFS_TI_SYNC)
 284                 err = nilfs_construct_segment(sb);
 285         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 286                 kmem_cache_free(nilfs_transaction_cachep, ti);
 287         return err;
 288 }
 289
 290 void nilfs_transaction_abort(struct super_block *sb)
 291 {
 292         struct nilfs_transaction_info *ti = current->journal_info;
 293
 294         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 295         if (ti->ti_count > 0) {
 296                 ti->ti_count--;
 297                 return;
 298         }
 299         up_read(&NILFS_SB(sb)->s_nilfs->ns_segctor_sem);
 300
 301         current->journal_info = ti->ti_save;
 302         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 303                 kmem_cache_free(nilfs_transaction_cachep, ti);
 304 }
 305
 306 void nilfs_relax_pressure_in_lock(struct super_block *sb)
 307 {
 308         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 309         struct nilfs_sc_info *sci = NILFS_SC(sbi);
 310         struct the_nilfs *nilfs = sbi->s_nilfs;
 311
 312         if (!sci || !sci->sc_flush_request)
 313                 return;
 314
 315         set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
 316         up_read(&nilfs->ns_segctor_sem);
 317
 318         down_write(&nilfs->ns_segctor_sem);
 319         if (sci->sc_flush_request &&
 320             test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
 321                 struct nilfs_transaction_info *ti = current->journal_info;
 322
 323                 ti->ti_flags |= NILFS_TI_WRITER;
 324                 nilfs_segctor_do_immediate_flush(sci);
 325                 ti->ti_flags &= ~NILFS_TI_WRITER;
 326         }
 327         downgrade_write(&nilfs->ns_segctor_sem);
 328 }
 329
 330 static void nilfs_transaction_lock(struct nilfs_sb_info *sbi,
 331                                    struct nilfs_transaction_info *ti,
 332                                    int gcflag)
 333 {
 334         struct nilfs_transaction_info *cur_ti = current->journal_info;
 335
 336         WARN_ON(cur_ti);
 337         ti->ti_flags = NILFS_TI_WRITER;
 338         ti->ti_count = 0;
 339         ti->ti_save = cur_ti;
 340         ti->ti_magic = NILFS_TI_MAGIC;
 341         INIT_LIST_HEAD(&ti->ti_garbage);
 342         current->journal_info = ti;
 343
 344         for (;;) {
 345                 down_write(&sbi->s_nilfs->ns_segctor_sem);
 346                 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &NILFS_SC(sbi)->sc_flags))
 347                         break;
 348
 349                 nilfs_segctor_do_immediate_flush(NILFS_SC(sbi));
 350
 351                 up_write(&sbi->s_nilfs->ns_segctor_sem);
 352                 yield();
 353         }
 354         if (gcflag)
 355                 ti->ti_flags |= NILFS_TI_GC;
 356 }
 357
 358 static void nilfs_transaction_unlock(struct nilfs_sb_info *sbi)
 359 {
 360         struct nilfs_transaction_info *ti = current->journal_info;
 361
 362         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 363         BUG_ON(ti->ti_count > 0);
 364
 365         up_write(&sbi->s_nilfs->ns_segctor_sem);
 366         current->journal_info = ti->ti_save;
 367         if (!list_empty(&ti->ti_garbage))
 368                 nilfs_dispose_list(sbi, &ti->ti_garbage, 0);
 369 }
 370
 371 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
 372                                             struct nilfs_segsum_pointer *ssp,
 373                                             unsigned bytes)
 374 {
 375         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 376         unsigned blocksize = sci->sc_super->s_blocksize;
 377         void *p;
 378
 379         if (unlikely(ssp->offset + bytes > blocksize)) {
 380                 ssp->offset = 0;
 381                 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
 382                                                &segbuf->sb_segsum_buffers));
 383                 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
 384         }
 385         p = ssp->bh->b_data + ssp->offset;
 386         ssp->offset += bytes;
 387         return p;
 388 }
 389
 390 /**
 391  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
 392  * @sci: nilfs_sc_info
 393  */
 394 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
 395 {
 396         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 397         struct buffer_head *sumbh;
 398         unsigned sumbytes;
 399         unsigned flags = 0;
 400         int err;
 401
 402         if (nilfs_doing_gc())
 403                 flags = NILFS_SS_GC;
 404         err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime);
 405         if (unlikely(err))
 406                 return err;
 407
 408         sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
 409         sumbytes = segbuf->sb_sum.sumbytes;
 410         sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
 411         sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
 412         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 413         return 0;
 414 }
 415
 416 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
 417 {
 418         sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
 419         if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
 420                 return -E2BIG; /* The current segment is filled up
 421                                   (internal code) */
 422         sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
 423         return nilfs_segctor_reset_segment_buffer(sci);
 424 }
 425
 426 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
 427 {
 428         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 429         int err;
 430
 431         if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
 432                 err = nilfs_segctor_feed_segment(sci);
 433                 if (err)
 434                         return err;
 435                 segbuf = sci->sc_curseg;
 436         }
 437         err = nilfs_segbuf_extend_payload(segbuf, &sci->sc_super_root);
 438         if (likely(!err))
 439                 segbuf->sb_sum.flags |= NILFS_SS_SR;
 440         return err;
 441 }
 442
 443 /*
 444  * Functions for making segment summary and payloads
 445  */
 446 static int nilfs_segctor_segsum_block_required(
 447         struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
 448         unsigned binfo_size)
 449 {
 450         unsigned blocksize = sci->sc_super->s_blocksize;
 451         /* Size of finfo and binfo is enough small against blocksize */
 452
 453         return ssp->offset + binfo_size +
 454                 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
 455                 blocksize;
 456 }
 457
 458 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
 459                                       struct inode *inode)
 460 {
 461         sci->sc_curseg->sb_sum.nfinfo++;
 462         sci->sc_binfo_ptr = sci->sc_finfo_ptr;
 463         nilfs_segctor_map_segsum_entry(
 464                 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
 465
 466         if (inode->i_sb && !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 467                 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
 468         /* skip finfo */
 469 }
 470
 471 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
 472                                     struct inode *inode)
 473 {
 474         struct nilfs_finfo *finfo;
 475         struct nilfs_inode_info *ii;
 476         struct nilfs_segment_buffer *segbuf;
 477
 478         if (sci->sc_blk_cnt == 0)
 479                 return;
 480
 481         ii = NILFS_I(inode);
 482         finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
 483                                                  sizeof(*finfo));
 484         finfo->fi_ino = cpu_to_le64(inode->i_ino);
 485         finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
 486         finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
 487         finfo->fi_cno = cpu_to_le64(ii->i_cno);
 488
 489         segbuf = sci->sc_curseg;
 490         segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
 491                 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
 492         sci->sc_finfo_ptr = sci->sc_binfo_ptr;
 493         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 494 }
 495
 496 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
 497                                         struct buffer_head *bh,
 498                                         struct inode *inode,
 499                                         unsigned binfo_size)
 500 {
 501         struct nilfs_segment_buffer *segbuf;
 502         int required, err = 0;
 503
 504  retry:
 505         segbuf = sci->sc_curseg;
 506         required = nilfs_segctor_segsum_block_required(
 507                 sci, &sci->sc_binfo_ptr, binfo_size);
 508         if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
 509                 nilfs_segctor_end_finfo(sci, inode);
 510                 err = nilfs_segctor_feed_segment(sci);
 511                 if (err)
 512                         return err;
 513                 goto retry;
 514         }
 515         if (unlikely(required)) {
 516                 err = nilfs_segbuf_extend_segsum(segbuf);
 517                 if (unlikely(err))
 518                         goto failed;
 519         }
 520         if (sci->sc_blk_cnt == 0)
 521                 nilfs_segctor_begin_finfo(sci, inode);
 522
 523         nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
 524         /* Substitution to vblocknr is delayed until update_blocknr() */
 525         nilfs_segbuf_add_file_buffer(segbuf, bh);
 526         sci->sc_blk_cnt++;
 527  failed:
 528         return err;
 529 }
 530
 531 static int nilfs_handle_bmap_error(int err, const char *fname,
 532                                    struct inode *inode, struct super_block *sb)
 533 {
 534         if (err == -EINVAL) {
 535                 nilfs_error(sb, fname, "broken bmap (inode=%lu)\n",
 536                             inode->i_ino);
 537                 err = -EIO;
 538         }
 539         return err;
 540 }
 541
 542 /*
 543  * Callback functions that enumerate, mark, and collect dirty blocks
 544  */
 545 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
 546                                    struct buffer_head *bh, struct inode *inode)
 547 {
 548         int err;
 549
 550         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 551         if (unlikely(err < 0))
 552                 return nilfs_handle_bmap_error(err, __func__, inode,
 553                                                sci->sc_super);
 554
 555         err = nilfs_segctor_add_file_block(sci, bh, inode,
 556                                            sizeof(struct nilfs_binfo_v));
 557         if (!err)
 558                 sci->sc_datablk_cnt++;
 559         return err;
 560 }
 561
 562 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
 563                                    struct buffer_head *bh,
 564                                    struct inode *inode)
 565 {
 566         int err;
 567
 568         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 569         if (unlikely(err < 0))
 570                 return nilfs_handle_bmap_error(err, __func__, inode,
 571                                                sci->sc_super);
 572         return 0;
 573 }
 574
 575 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
 576                                    struct buffer_head *bh,
 577                                    struct inode *inode)
 578 {
 579         WARN_ON(!buffer_dirty(bh));
 580         return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 581 }
 582
 583 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
 584                                         struct nilfs_segsum_pointer *ssp,
 585                                         union nilfs_binfo *binfo)
 586 {
 587         struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
 588                 sci, ssp, sizeof(*binfo_v));
 589         *binfo_v = binfo->bi_v;
 590 }
 591
 592 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
 593                                         struct nilfs_segsum_pointer *ssp,
 594                                         union nilfs_binfo *binfo)
 595 {
 596         __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
 597                 sci, ssp, sizeof(*vblocknr));
 598         *vblocknr = binfo->bi_v.bi_vblocknr;
 599 }
 600
 601 struct nilfs_sc_operations nilfs_sc_file_ops = {
 602         .collect_data = nilfs_collect_file_data,
 603         .collect_node = nilfs_collect_file_node,
 604         .collect_bmap = nilfs_collect_file_bmap,
 605         .write_data_binfo = nilfs_write_file_data_binfo,
 606         .write_node_binfo = nilfs_write_file_node_binfo,
 607 };
 608
 609 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
 610                                   struct buffer_head *bh, struct inode *inode)
 611 {
 612         int err;
 613
 614         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 615         if (unlikely(err < 0))
 616                 return nilfs_handle_bmap_error(err, __func__, inode,
 617                                                sci->sc_super);
 618
 619         err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 620         if (!err)
 621                 sci->sc_datablk_cnt++;
 622         return err;
 623 }
 624
 625 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
 626                                   struct buffer_head *bh, struct inode *inode)
 627 {
 628         WARN_ON(!buffer_dirty(bh));
 629         return nilfs_segctor_add_file_block(sci, bh, inode,
 630                                             sizeof(struct nilfs_binfo_dat));
 631 }
 632
 633 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
 634                                        struct nilfs_segsum_pointer *ssp,
 635                                        union nilfs_binfo *binfo)
 636 {
 637         __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
 638                                                           sizeof(*blkoff));
 639         *blkoff = binfo->bi_dat.bi_blkoff;
 640 }
 641
 642 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
 643                                        struct nilfs_segsum_pointer *ssp,
 644                                        union nilfs_binfo *binfo)
 645 {
 646         struct nilfs_binfo_dat *binfo_dat =
 647                 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
 648         *binfo_dat = binfo->bi_dat;
 649 }
 650
 651 struct nilfs_sc_operations nilfs_sc_dat_ops = {
 652         .collect_data = nilfs_collect_dat_data,
 653         .collect_node = nilfs_collect_file_node,
 654         .collect_bmap = nilfs_collect_dat_bmap,
 655         .write_data_binfo = nilfs_write_dat_data_binfo,
 656         .write_node_binfo = nilfs_write_dat_node_binfo,
 657 };
 658
 659 struct nilfs_sc_operations nilfs_sc_dsync_ops = {
 660         .collect_data = nilfs_collect_file_data,
 661         .collect_node = NULL,
 662         .collect_bmap = NULL,
 663         .write_data_binfo = nilfs_write_file_data_binfo,
 664         .write_node_binfo = NULL,
 665 };
 666
 667 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
 668                                               struct list_head *listp,
 669                                               size_t nlimit,
 670                                               loff_t start, loff_t end)
 671 {
 672         struct address_space *mapping = inode->i_mapping;
 673         struct pagevec pvec;
 674         pgoff_t index = 0, last = ULONG_MAX;
 675         size_t ndirties = 0;
 676         int i;
 677
 678         if (unlikely(start != 0 || end != LLONG_MAX)) {
 679                 /*
 680                  * A valid range is given for sync-ing data pages. The
 681                  * range is rounded to per-page; extra dirty buffers
 682                  * may be included if blocksize < pagesize.
 683                  */
 684                 index = start >> PAGE_SHIFT;
 685                 last = end >> PAGE_SHIFT;
 686         }
 687         pagevec_init(&pvec, 0);
 688  repeat:
 689         if (unlikely(index > last) ||
 690             !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 691                                 min_t(pgoff_t, last - index,
 692                                       PAGEVEC_SIZE - 1) + 1))
 693                 return ndirties;
 694
 695         for (i = 0; i < pagevec_count(&pvec); i++) {
 696                 struct buffer_head *bh, *head;
 697                 struct page *page = pvec.pages[i];
 698
 699                 if (unlikely(page->index > last))
 700                         break;
 701
 702                 if (mapping->host) {
 703                         lock_page(page);
 704                         if (!page_has_buffers(page))
 705                                 create_empty_buffers(page,
 706                                                      1 << inode->i_blkbits, 0);
 707                         unlock_page(page);
 708                 }
 709
 710                 bh = head = page_buffers(page);
 711                 do {
 712                         if (!buffer_dirty(bh))
 713                                 continue;
 714                         get_bh(bh);
 715                         list_add_tail(&bh->b_assoc_buffers, listp);
 716                         ndirties++;
 717                         if (unlikely(ndirties >= nlimit)) {
 718                                 pagevec_release(&pvec);
 719                                 cond_resched();
 720                                 return ndirties;
 721                         }
 722                 } while (bh = bh->b_this_page, bh != head);
 723         }
 724         pagevec_release(&pvec);
 725         cond_resched();
 726         goto repeat;
 727 }
 728
 729 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
 730                                             struct list_head *listp)
 731 {
 732         struct nilfs_inode_info *ii = NILFS_I(inode);
 733         struct address_space *mapping = &ii->i_btnode_cache;
 734         struct pagevec pvec;
 735         struct buffer_head *bh, *head;
 736         unsigned int i;
 737         pgoff_t index = 0;
 738
 739         pagevec_init(&pvec, 0);
 740
 741         while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 742                                   PAGEVEC_SIZE)) {
 743                 for (i = 0; i < pagevec_count(&pvec); i++) {
 744                         bh = head = page_buffers(pvec.pages[i]);
 745                         do {
 746                                 if (buffer_dirty(bh)) {
 747                                         get_bh(bh);
 748                                         list_add_tail(&bh->b_assoc_buffers,
 749                                                       listp);
 750                                 }
 751                                 bh = bh->b_this_page;
 752                         } while (bh != head);
 753                 }
 754                 pagevec_release(&pvec);
 755                 cond_resched();
 756         }
 757 }
 758
 759 static void nilfs_dispose_list(struct nilfs_sb_info *sbi,
 760                                struct list_head *head, int force)
 761 {
 762         struct nilfs_inode_info *ii, *n;
 763         struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
 764         unsigned nv = 0;
 765
 766         while (!list_empty(head)) {
 767                 spin_lock(&sbi->s_inode_lock);
 768                 list_for_each_entry_safe(ii, n, head, i_dirty) {
 769                         list_del_init(&ii->i_dirty);
 770                         if (force) {
 771                                 if (unlikely(ii->i_bh)) {
 772                                         brelse(ii->i_bh);
 773                                         ii->i_bh = NULL;
 774                                 }
 775                         } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
 776                                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 777                                 list_add_tail(&ii->i_dirty,
 778                                               &sbi->s_dirty_files);
 779                                 continue;
 780                         }
 781                         ivec[nv++] = ii;
 782                         if (nv == SC_N_INODEVEC)
 783                                 break;
 784                 }
 785                 spin_unlock(&sbi->s_inode_lock);
 786
 787                 for (pii = ivec; nv > 0; pii++, nv--)
 788                         iput(&(*pii)->vfs_inode);
 789         }
 790 }
 791
 792 static int nilfs_test_metadata_dirty(struct nilfs_sb_info *sbi)
 793 {
 794         struct the_nilfs *nilfs = sbi->s_nilfs;
 795         int ret = 0;
 796
 797         if (nilfs_mdt_fetch_dirty(sbi->s_ifile))
 798                 ret++;
 799         if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
 800                 ret++;
 801         if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
 802                 ret++;
 803         if (ret || nilfs_doing_gc())
 804                 if (nilfs_mdt_fetch_dirty(nilfs_dat_inode(nilfs)))
 805                         ret++;
 806         return ret;
 807 }
 808
 809 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
 810 {
 811         return list_empty(&sci->sc_dirty_files) &&
 812                 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
 813                 sci->sc_nfreesegs == 0 &&
 814                 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
 815 }
 816
 817 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
 818 {
 819         struct nilfs_sb_info *sbi = sci->sc_sbi;
 820         int ret = 0;
 821
 822         if (nilfs_test_metadata_dirty(sbi))
 823                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
 824
 825         spin_lock(&sbi->s_inode_lock);
 826         if (list_empty(&sbi->s_dirty_files) && nilfs_segctor_clean(sci))
 827                 ret++;
 828
 829         spin_unlock(&sbi->s_inode_lock);
 830         return ret;
 831 }
 832
 833 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
 834 {
 835         struct nilfs_sb_info *sbi = sci->sc_sbi;
 836         struct the_nilfs *nilfs = sbi->s_nilfs;
 837
 838         nilfs_mdt_clear_dirty(sbi->s_ifile);
 839         nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
 840         nilfs_mdt_clear_dirty(nilfs->ns_sufile);
 841         nilfs_mdt_clear_dirty(nilfs_dat_inode(nilfs));
 842 }
 843
 844 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
 845 {
 846         struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
 847         struct buffer_head *bh_cp;
 848         struct nilfs_checkpoint *raw_cp;
 849         int err;
 850
 851         /* XXX: this interface will be changed */
 852         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
 853                                           &raw_cp, &bh_cp);
 854         if (likely(!err)) {
 855                 /* The following code is duplicated with cpfile.  But, it is
 856                    needed to collect the checkpoint even if it was not newly
 857                    created */
 858                 nilfs_mdt_mark_buffer_dirty(bh_cp);
 859                 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
 860                 nilfs_cpfile_put_checkpoint(
 861                         nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 862         } else
 863                 WARN_ON(err == -EINVAL || err == -ENOENT);
 864
 865         return err;
 866 }
 867
 868 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
 869 {
 870         struct nilfs_sb_info *sbi = sci->sc_sbi;
 871         struct the_nilfs *nilfs = sbi->s_nilfs;
 872         struct buffer_head *bh_cp;
 873         struct nilfs_checkpoint *raw_cp;
 874         int err;
 875
 876         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
 877                                           &raw_cp, &bh_cp);
 878         if (unlikely(err)) {
 879                 WARN_ON(err == -EINVAL || err == -ENOENT);
 880                 goto failed_ibh;
 881         }
 882         raw_cp->cp_snapshot_list.ssl_next = 0;
 883         raw_cp->cp_snapshot_list.ssl_prev = 0;
 884         raw_cp->cp_inodes_count =
 885                 cpu_to_le64(atomic_read(&sbi->s_inodes_count));
 886         raw_cp->cp_blocks_count =
 887                 cpu_to_le64(atomic_read(&sbi->s_blocks_count));
 888         raw_cp->cp_nblk_inc =
 889                 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
 890         raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
 891         raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
 892
 893         if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 894                 nilfs_checkpoint_clear_minor(raw_cp);
 895         else
 896                 nilfs_checkpoint_set_minor(raw_cp);
 897
 898         nilfs_write_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode, 1);
 899         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 900         return 0;
 901
 902  failed_ibh:
 903         return err;
 904 }
 905
 906 static void nilfs_fill_in_file_bmap(struct inode *ifile,
 907                                     struct nilfs_inode_info *ii)
 908
 909 {
 910         struct buffer_head *ibh;
 911         struct nilfs_inode *raw_inode;
 912
 913         if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
 914                 ibh = ii->i_bh;
 915                 BUG_ON(!ibh);
 916                 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
 917                                                   ibh);
 918                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 919                 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
 920         }
 921 }
 922
 923 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci,
 924                                             struct inode *ifile)
 925 {
 926         struct nilfs_inode_info *ii;
 927
 928         list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
 929                 nilfs_fill_in_file_bmap(ifile, ii);
 930                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
 931         }
 932 }
 933
 934 /*
 935  * CRC calculation routines
 936  */
 937 static void nilfs_fill_in_super_root_crc(struct buffer_head *bh_sr, u32 seed)
 938 {
 939         struct nilfs_super_root *raw_sr =
 940                 (struct nilfs_super_root *)bh_sr->b_data;
 941         u32 crc;
 942
 943         crc = crc32_le(seed,
 944                        (unsigned char *)raw_sr + sizeof(raw_sr->sr_sum),
 945                        NILFS_SR_BYTES - sizeof(raw_sr->sr_sum));
 946         raw_sr->sr_sum = cpu_to_le32(crc);
 947 }
 948
 949 static void nilfs_segctor_fill_in_checksums(struct nilfs_sc_info *sci,
 950                                             u32 seed)
 951 {
 952         struct nilfs_segment_buffer *segbuf;
 953
 954         if (sci->sc_super_root)
 955                 nilfs_fill_in_super_root_crc(sci->sc_super_root, seed);
 956
 957         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
 958                 nilfs_segbuf_fill_in_segsum_crc(segbuf, seed);
 959                 nilfs_segbuf_fill_in_data_crc(segbuf, seed);
 960         }
 961 }
 962
 963 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
 964                                              struct the_nilfs *nilfs)
 965 {
 966         struct buffer_head *bh_sr = sci->sc_super_root;
 967         struct nilfs_super_root *raw_sr =
 968                 (struct nilfs_super_root *)bh_sr->b_data;
 969         unsigned isz = nilfs->ns_inode_size;
 970
 971         raw_sr->sr_bytes = cpu_to_le16(NILFS_SR_BYTES);
 972         raw_sr->sr_nongc_ctime
 973                 = cpu_to_le64(nilfs_doing_gc() ?
 974                               nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
 975         raw_sr->sr_flags = 0;
 976
 977         nilfs_mdt_write_inode_direct(
 978                 nilfs_dat_inode(nilfs), bh_sr, NILFS_SR_DAT_OFFSET(isz));
 979         nilfs_mdt_write_inode_direct(
 980                 nilfs->ns_cpfile, bh_sr, NILFS_SR_CPFILE_OFFSET(isz));
 981         nilfs_mdt_write_inode_direct(
 982                 nilfs->ns_sufile, bh_sr, NILFS_SR_SUFILE_OFFSET(isz));
 983 }
 984
 985 static void nilfs_redirty_inodes(struct list_head *head)
 986 {
 987         struct nilfs_inode_info *ii;
 988
 989         list_for_each_entry(ii, head, i_dirty) {
 990                 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
 991                         clear_bit(NILFS_I_COLLECTED, &ii->i_state);
 992         }
 993 }
 994
 995 static void nilfs_drop_collected_inodes(struct list_head *head)
 996 {
 997         struct nilfs_inode_info *ii;
 998
 999         list_for_each_entry(ii, head, i_dirty) {
1000                 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1001                         continue;
1002
1003                 clear_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
1004                 set_bit(NILFS_I_UPDATED, &ii->i_state);
1005         }
1006 }
1007
1008 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1009                                        struct inode *inode,
1010                                        struct list_head *listp,
1011                                        int (*collect)(struct nilfs_sc_info *,
1012                                                       struct buffer_head *,
1013                                                       struct inode *))
1014 {
1015         struct buffer_head *bh, *n;
1016         int err = 0;
1017
1018         if (collect) {
1019                 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1020                         list_del_init(&bh->b_assoc_buffers);
1021                         err = collect(sci, bh, inode);
1022                         brelse(bh);
1023                         if (unlikely(err))
1024                                 goto dispose_buffers;
1025                 }
1026                 return 0;
1027         }
1028
1029  dispose_buffers:
1030         while (!list_empty(listp)) {
1031                 bh = list_entry(listp->next, struct buffer_head,
1032                                 b_assoc_buffers);
1033                 list_del_init(&bh->b_assoc_buffers);
1034                 brelse(bh);
1035         }
1036         return err;
1037 }
1038
1039 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1040 {
1041         /* Remaining number of blocks within segment buffer */
1042         return sci->sc_segbuf_nblocks -
1043                 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1044 }
1045
1046 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1047                                    struct inode *inode,
1048                                    struct nilfs_sc_operations *sc_ops)
1049 {
1050         LIST_HEAD(data_buffers);
1051         LIST_HEAD(node_buffers);
1052         int err;
1053
1054         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1055                 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1056
1057                 n = nilfs_lookup_dirty_data_buffers(
1058                         inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1059                 if (n > rest) {
1060                         err = nilfs_segctor_apply_buffers(
1061                                 sci, inode, &data_buffers,
1062                                 sc_ops->collect_data);
1063                         BUG_ON(!err); /* always receive -E2BIG or true error */
1064                         goto break_or_fail;
1065                 }
1066         }
1067         nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1068
1069         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1070                 err = nilfs_segctor_apply_buffers(
1071                         sci, inode, &data_buffers, sc_ops->collect_data);
1072                 if (unlikely(err)) {
1073                         /* dispose node list */
1074                         nilfs_segctor_apply_buffers(
1075                                 sci, inode, &node_buffers, NULL);
1076                         goto break_or_fail;
1077                 }
1078                 sci->sc_stage.flags |= NILFS_CF_NODE;
1079         }
1080         /* Collect node */
1081         err = nilfs_segctor_apply_buffers(
1082                 sci, inode, &node_buffers, sc_ops->collect_node);
1083         if (unlikely(err))
1084                 goto break_or_fail;
1085
1086         nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1087         err = nilfs_segctor_apply_buffers(
1088                 sci, inode, &node_buffers, sc_ops->collect_bmap);
1089         if (unlikely(err))
1090                 goto break_or_fail;
1091
1092         nilfs_segctor_end_finfo(sci, inode);
1093         sci->sc_stage.flags &= ~NILFS_CF_NODE;
1094
1095  break_or_fail:
1096         return err;
1097 }
1098
1099 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1100                                          struct inode *inode)
1101 {
1102         LIST_HEAD(data_buffers);
1103         size_t n, rest = nilfs_segctor_buffer_rest(sci);
1104         int err;
1105
1106         n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1107                                             sci->sc_dsync_start,
1108                                             sci->sc_dsync_end);
1109
1110         err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1111                                           nilfs_collect_file_data);
1112         if (!err) {
1113                 nilfs_segctor_end_finfo(sci, inode);
1114                 BUG_ON(n > rest);
1115                 /* always receive -E2BIG or true error if n > rest */
1116         }
1117         return err;
1118 }
1119
1120 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1121 {
1122         struct nilfs_sb_info *sbi = sci->sc_sbi;
1123         struct the_nilfs *nilfs = sbi->s_nilfs;
1124         struct list_head *head;
1125         struct nilfs_inode_info *ii;
1126         size_t ndone;
1127         int err = 0;
1128
1129         switch (sci->sc_stage.scnt) {
1130         case NILFS_ST_INIT:
1131                 /* Pre-processes */
1132                 sci->sc_stage.flags = 0;
1133
1134                 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1135                         sci->sc_nblk_inc = 0;
1136                         sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1137                         if (mode == SC_LSEG_DSYNC) {
1138                                 sci->sc_stage.scnt = NILFS_ST_DSYNC;
1139                                 goto dsync_mode;
1140                         }
1141                 }
1142
1143                 sci->sc_stage.dirty_file_ptr = NULL;
1144                 sci->sc_stage.gc_inode_ptr = NULL;
1145                 if (mode == SC_FLUSH_DAT) {
1146                         sci->sc_stage.scnt = NILFS_ST_DAT;
1147                         goto dat_stage;
1148                 }
1149                 sci->sc_stage.scnt++;  /* Fall through */
1150         case NILFS_ST_GC:
1151                 if (nilfs_doing_gc()) {
1152                         head = &sci->sc_gc_inodes;
1153                         ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1154                                                 head, i_dirty);
1155                         list_for_each_entry_continue(ii, head, i_dirty) {
1156                                 err = nilfs_segctor_scan_file(
1157                                         sci, &ii->vfs_inode,
1158                                         &nilfs_sc_file_ops);
1159                                 if (unlikely(err)) {
1160                                         sci->sc_stage.gc_inode_ptr = list_entry(
1161                                                 ii->i_dirty.prev,
1162                                                 struct nilfs_inode_info,
1163                                                 i_dirty);
1164                                         goto break_or_fail;
1165                                 }
1166                                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1167                         }
1168                         sci->sc_stage.gc_inode_ptr = NULL;
1169                 }
1170                 sci->sc_stage.scnt++;  /* Fall through */
1171         case NILFS_ST_FILE:
1172                 head = &sci->sc_dirty_files;
1173                 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1174                                         i_dirty);
1175                 list_for_each_entry_continue(ii, head, i_dirty) {
1176                         clear_bit(NILFS_I_DIRTY, &ii->i_state);
1177
1178                         err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1179                                                       &nilfs_sc_file_ops);
1180                         if (unlikely(err)) {
1181                                 sci->sc_stage.dirty_file_ptr =
1182                                         list_entry(ii->i_dirty.prev,
1183                                                    struct nilfs_inode_info,
1184                                                    i_dirty);
1185                                 goto break_or_fail;
1186                         }
1187                         /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1188                         /* XXX: required ? */
1189                 }
1190                 sci->sc_stage.dirty_file_ptr = NULL;
1191                 if (mode == SC_FLUSH_FILE) {
1192                         sci->sc_stage.scnt = NILFS_ST_DONE;
1193                         return 0;
1194                 }
1195                 sci->sc_stage.scnt++;
1196                 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1197                 /* Fall through */
1198         case NILFS_ST_IFILE:
1199                 err = nilfs_segctor_scan_file(sci, sbi->s_ifile,
1200                                               &nilfs_sc_file_ops);
1201                 if (unlikely(err))
1202                         break;
1203                 sci->sc_stage.scnt++;
1204                 /* Creating a checkpoint */
1205                 err = nilfs_segctor_create_checkpoint(sci);
1206                 if (unlikely(err))
1207                         break;
1208                 /* Fall through */
1209         case NILFS_ST_CPFILE:
1210                 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1211                                               &nilfs_sc_file_ops);
1212                 if (unlikely(err))
1213                         break;
1214                 sci->sc_stage.scnt++;  /* Fall through */
1215         case NILFS_ST_SUFILE:
1216                 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1217                                          sci->sc_nfreesegs, &ndone);
1218                 if (unlikely(err)) {
1219                         nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1220                                                   sci->sc_freesegs, ndone,
1221                                                   NULL);
1222                         break;
1223                 }
1224                 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1225
1226                 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1227                                               &nilfs_sc_file_ops);
1228                 if (unlikely(err))
1229                         break;
1230                 sci->sc_stage.scnt++;  /* Fall through */
1231         case NILFS_ST_DAT:
1232  dat_stage:
1233                 err = nilfs_segctor_scan_file(sci, nilfs_dat_inode(nilfs),
1234                                               &nilfs_sc_dat_ops);
1235                 if (unlikely(err))
1236                         break;
1237                 if (mode == SC_FLUSH_DAT) {
1238                         sci->sc_stage.scnt = NILFS_ST_DONE;
1239                         return 0;
1240                 }
1241                 sci->sc_stage.scnt++;  /* Fall through */
1242         case NILFS_ST_SR:
1243                 if (mode == SC_LSEG_SR) {
1244                         /* Appending a super root */
1245                         err = nilfs_segctor_add_super_root(sci);
1246                         if (unlikely(err))
1247                                 break;
1248                 }
1249                 /* End of a logical segment */
1250                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1251                 sci->sc_stage.scnt = NILFS_ST_DONE;
1252                 return 0;
1253         case NILFS_ST_DSYNC:
1254  dsync_mode:
1255                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1256                 ii = sci->sc_dsync_inode;
1257                 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1258                         break;
1259
1260                 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1261                 if (unlikely(err))
1262                         break;
1263                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1264                 sci->sc_stage.scnt = NILFS_ST_DONE;
1265                 return 0;
1266         case NILFS_ST_DONE:
1267                 return 0;
1268         default:
1269                 BUG();
1270         }
1271
1272  break_or_fail:
1273         return err;
1274 }
1275
1276 static int nilfs_touch_segusage(struct inode *sufile, __u64 segnum)
1277 {
1278         struct buffer_head *bh_su;
1279         struct nilfs_segment_usage *raw_su;
1280         int err;
1281
1282         err = nilfs_sufile_get_segment_usage(sufile, segnum, &raw_su, &bh_su);
1283         if (unlikely(err))
1284                 return err;
1285         nilfs_mdt_mark_buffer_dirty(bh_su);
1286         nilfs_mdt_mark_dirty(sufile);
1287         nilfs_sufile_put_segment_usage(sufile, segnum, bh_su);
1288         return 0;
1289 }
1290
1291 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1292                                             struct the_nilfs *nilfs)
1293 {
1294         struct nilfs_segment_buffer *segbuf, *n;
1295         __u64 nextnum;
1296         int err;
1297
1298         if (list_empty(&sci->sc_segbufs)) {
1299                 segbuf = nilfs_segbuf_new(sci->sc_super);
1300                 if (unlikely(!segbuf))
1301                         return -ENOMEM;
1302                 list_add(&segbuf->sb_list, &sci->sc_segbufs);
1303         } else
1304                 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1305
1306         nilfs_segbuf_map(segbuf, nilfs->ns_segnum, nilfs->ns_pseg_offset,
1307                          nilfs);
1308
1309         if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1310                 nilfs_shift_to_next_segment(nilfs);
1311                 nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1312         }
1313         sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1314
1315         err = nilfs_touch_segusage(nilfs->ns_sufile, segbuf->sb_segnum);
1316         if (unlikely(err))
1317                 return err;
1318
1319         if (nilfs->ns_segnum == nilfs->ns_nextnum) {
1320                 /* Start from the head of a new full segment */
1321                 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1322                 if (unlikely(err))
1323                         return err;
1324         } else
1325                 nextnum = nilfs->ns_nextnum;
1326
1327         segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1328         nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1329
1330         /* truncating segment buffers */
1331         list_for_each_entry_safe_continue(segbuf, n, &sci->sc_segbufs,
1332                                           sb_list) {
1333                 list_del_init(&segbuf->sb_list);
1334                 nilfs_segbuf_free(segbuf);
1335         }
1336         return 0;
1337 }
1338
1339 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1340                                          struct the_nilfs *nilfs, int nadd)
1341 {
1342         struct nilfs_segment_buffer *segbuf, *prev, *n;
1343         struct inode *sufile = nilfs->ns_sufile;
1344         __u64 nextnextnum;
1345         LIST_HEAD(list);
1346         int err, ret, i;
1347
1348         prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1349         /*
1350          * Since the segment specified with nextnum might be allocated during
1351          * the previous construction, the buffer including its segusage may
1352          * not be dirty.  The following call ensures that the buffer is dirty
1353          * and will pin the buffer on memory until the sufile is written.
1354          */
1355         err = nilfs_touch_segusage(sufile, prev->sb_nextnum);
1356         if (unlikely(err))
1357                 return err;
1358
1359         for (i = 0; i < nadd; i++) {
1360                 /* extend segment info */
1361                 err = -ENOMEM;
1362                 segbuf = nilfs_segbuf_new(sci->sc_super);
1363                 if (unlikely(!segbuf))
1364                         goto failed;
1365
1366                 /* map this buffer to region of segment on-disk */
1367                 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1368                 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1369
1370                 /* allocate the next next full segment */
1371                 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1372                 if (unlikely(err))
1373                         goto failed_segbuf;
1374
1375                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1376                 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1377
1378                 list_add_tail(&segbuf->sb_list, &list);
1379                 prev = segbuf;
1380         }
1381         list_splice(&list, sci->sc_segbufs.prev);
1382         return 0;
1383
1384  failed_segbuf:
1385         nilfs_segbuf_free(segbuf);
1386  failed:
1387         list_for_each_entry_safe(segbuf, n, &list, sb_list) {
1388                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1389                 WARN_ON(ret); /* never fails */
1390                 list_del_init(&segbuf->sb_list);
1391                 nilfs_segbuf_free(segbuf);
1392         }
1393         return err;
1394 }
1395
1396 static void nilfs_segctor_free_incomplete_segments(struct nilfs_sc_info *sci,
1397                                                    struct the_nilfs *nilfs)
1398 {
1399         struct nilfs_segment_buffer *segbuf;
1400         int ret, done = 0;
1401
1402         segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1403         if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1404                 ret = nilfs_sufile_free(nilfs->ns_sufile, segbuf->sb_nextnum);
1405                 WARN_ON(ret); /* never fails */
1406         }
1407         if (segbuf->sb_io_error) {
1408                 /* Case 1: The first segment failed */
1409                 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1410                         /* Case 1a:  Partial segment appended into an existing
1411                            segment */
1412                         nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1413                                                 segbuf->sb_fseg_end);
1414                 else /* Case 1b:  New full segment */
1415                         set_nilfs_discontinued(nilfs);
1416                 done++;
1417         }
1418
1419         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1420                 ret = nilfs_sufile_free(nilfs->ns_sufile, segbuf->sb_nextnum);
1421                 WARN_ON(ret); /* never fails */
1422                 if (!done && segbuf->sb_io_error) {
1423                         if (segbuf->sb_segnum != nilfs->ns_nextnum)
1424                                 /* Case 2: extended segment (!= next) failed */
1425                                 nilfs_sufile_set_error(nilfs->ns_sufile,
1426                                                        segbuf->sb_segnum);
1427                         done++;
1428                 }
1429         }
1430 }
1431
1432 static void nilfs_segctor_clear_segment_buffers(struct nilfs_sc_info *sci)
1433 {
1434         struct nilfs_segment_buffer *segbuf;
1435
1436         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list)
1437                 nilfs_segbuf_clear(segbuf);
1438         sci->sc_super_root = NULL;
1439 }
1440
1441 static void nilfs_segctor_destroy_segment_buffers(struct nilfs_sc_info *sci)
1442 {
1443         struct nilfs_segment_buffer *segbuf;
1444
1445         while (!list_empty(&sci->sc_segbufs)) {
1446                 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1447                 list_del_init(&segbuf->sb_list);
1448                 nilfs_segbuf_free(segbuf);
1449         }
1450         /* sci->sc_curseg = NULL; */
1451 }
1452
1453 static void nilfs_segctor_end_construction(struct nilfs_sc_info *sci,
1454                                            struct the_nilfs *nilfs, int err)
1455 {
1456         if (unlikely(err)) {
1457                 nilfs_segctor_free_incomplete_segments(sci, nilfs);
1458                 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1459                         int ret;
1460
1461                         ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1462                                                         sci->sc_freesegs,
1463                                                         sci->sc_nfreesegs,
1464                                                         NULL);
1465                         WARN_ON(ret); /* do not happen */
1466                 }
1467         }
1468         nilfs_segctor_clear_segment_buffers(sci);
1469 }
1470
1471 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1472                                           struct inode *sufile)
1473 {
1474         struct nilfs_segment_buffer *segbuf;
1475         struct buffer_head *bh_su;
1476         struct nilfs_segment_usage *raw_su;
1477         unsigned long live_blocks;
1478         int ret;
1479
1480         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1481                 ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1482                                                      &raw_su, &bh_su);
1483                 WARN_ON(ret); /* always succeed because bh_su is dirty */
1484                 live_blocks = segbuf->sb_sum.nblocks +
1485                         (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1486                 raw_su->su_lastmod = cpu_to_le64(sci->sc_seg_ctime);
1487                 raw_su->su_nblocks = cpu_to_le32(live_blocks);
1488                 nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum,
1489                                                bh_su);
1490         }
1491 }
1492
1493 static void nilfs_segctor_cancel_segusage(struct nilfs_sc_info *sci,
1494                                           struct inode *sufile)
1495 {
1496         struct nilfs_segment_buffer *segbuf;
1497         struct buffer_head *bh_su;
1498         struct nilfs_segment_usage *raw_su;
1499         int ret;
1500
1501         segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1502         ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1503                                              &raw_su, &bh_su);
1504         WARN_ON(ret); /* always succeed because bh_su is dirty */
1505         raw_su->su_nblocks = cpu_to_le32(segbuf->sb_pseg_start -
1506                                          segbuf->sb_fseg_start);
1507         nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum, bh_su);
1508
1509         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1510                 ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1511                                                      &raw_su, &bh_su);
1512                 WARN_ON(ret); /* always succeed */
1513                 raw_su->su_nblocks = 0;
1514                 nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum,
1515                                                bh_su);
1516         }
1517 }
1518
1519 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1520                                             struct nilfs_segment_buffer *last,
1521                                             struct inode *sufile)
1522 {
1523         struct nilfs_segment_buffer *segbuf = last, *n;
1524         int ret;
1525
1526         list_for_each_entry_safe_continue(segbuf, n, &sci->sc_segbufs,
1527                                           sb_list) {
1528                 list_del_init(&segbuf->sb_list);
1529                 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1530                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1531                 WARN_ON(ret);
1532                 nilfs_segbuf_free(segbuf);
1533         }
1534 }
1535
1536
1537 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1538                                  struct the_nilfs *nilfs, int mode)
1539 {
1540         struct nilfs_cstage prev_stage = sci->sc_stage;
1541         int err, nadd = 1;
1542
1543         /* Collection retry loop */
1544         for (;;) {
1545                 sci->sc_super_root = NULL;
1546                 sci->sc_nblk_this_inc = 0;
1547                 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1548
1549                 err = nilfs_segctor_reset_segment_buffer(sci);
1550                 if (unlikely(err))
1551                         goto failed;
1552
1553                 err = nilfs_segctor_collect_blocks(sci, mode);
1554                 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1555                 if (!err)
1556                         break;
1557
1558                 if (unlikely(err != -E2BIG))
1559                         goto failed;
1560
1561                 /* The current segment is filled up */
1562                 if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
1563                         break;
1564
1565                 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1566                         err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1567                                                         sci->sc_freesegs,
1568                                                         sci->sc_nfreesegs,
1569                                                         NULL);
1570                         WARN_ON(err); /* do not happen */
1571                 }
1572                 nilfs_segctor_clear_segment_buffers(sci);
1573
1574                 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1575                 if (unlikely(err))
1576                         return err;
1577
1578                 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1579                 sci->sc_stage = prev_stage;
1580         }
1581         nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1582         return 0;
1583
1584  failed:
1585         return err;
1586 }
1587
1588 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1589                                       struct buffer_head *new_bh)
1590 {
1591         BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1592
1593         list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1594         /* The caller must release old_bh */
1595 }
1596
1597 static int
1598 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1599                                      struct nilfs_segment_buffer *segbuf,
1600                                      int mode)
1601 {
1602         struct inode *inode = NULL;
1603         sector_t blocknr;
1604         unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1605         unsigned long nblocks = 0, ndatablk = 0;
1606         struct nilfs_sc_operations *sc_op = NULL;
1607         struct nilfs_segsum_pointer ssp;
1608         struct nilfs_finfo *finfo = NULL;
1609         union nilfs_binfo binfo;
1610         struct buffer_head *bh, *bh_org;
1611         ino_t ino = 0;
1612         int err = 0;
1613
1614         if (!nfinfo)
1615                 goto out;
1616
1617         blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1618         ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1619         ssp.offset = sizeof(struct nilfs_segment_summary);
1620
1621         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1622                 if (bh == sci->sc_super_root)
1623                         break;
1624                 if (!finfo) {
1625                         finfo = nilfs_segctor_map_segsum_entry(
1626                                 sci, &ssp, sizeof(*finfo));
1627                         ino = le64_to_cpu(finfo->fi_ino);
1628                         nblocks = le32_to_cpu(finfo->fi_nblocks);
1629                         ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1630
1631                         if (buffer_nilfs_node(bh))
1632                                 inode = NILFS_BTNC_I(bh->b_page->mapping);
1633                         else
1634                                 inode = NILFS_AS_I(bh->b_page->mapping);
1635
1636                         if (mode == SC_LSEG_DSYNC)
1637                                 sc_op = &nilfs_sc_dsync_ops;
1638                         else if (ino == NILFS_DAT_INO)
1639                                 sc_op = &nilfs_sc_dat_ops;
1640                         else /* file blocks */
1641                                 sc_op = &nilfs_sc_file_ops;
1642                 }
1643                 bh_org = bh;
1644                 get_bh(bh_org);
1645                 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1646                                         &binfo);
1647                 if (bh != bh_org)
1648                         nilfs_list_replace_buffer(bh_org, bh);
1649                 brelse(bh_org);
1650                 if (unlikely(err))
1651                         goto failed_bmap;
1652
1653                 if (ndatablk > 0)
1654                         sc_op->write_data_binfo(sci, &ssp, &binfo);
1655                 else
1656                         sc_op->write_node_binfo(sci, &ssp, &binfo);
1657
1658                 blocknr++;
1659                 if (--nblocks == 0) {
1660                         finfo = NULL;
1661                         if (--nfinfo == 0)
1662                                 break;
1663                 } else if (ndatablk > 0)
1664                         ndatablk--;
1665         }
1666  out:
1667         return 0;
1668
1669  failed_bmap:
1670         err = nilfs_handle_bmap_error(err, __func__, inode, sci->sc_super);
1671         return err;
1672 }
1673
1674 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1675 {
1676         struct nilfs_segment_buffer *segbuf;
1677         int err;
1678
1679         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1680                 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1681                 if (unlikely(err))
1682                         return err;
1683                 nilfs_segbuf_fill_in_segsum(segbuf);
1684         }
1685         return 0;
1686 }
1687
1688 static int
1689 nilfs_copy_replace_page_buffers(struct page *page, struct list_head *out)
1690 {
1691         struct page *clone_page;
1692         struct buffer_head *bh, *head, *bh2;
1693         void *kaddr;
1694
1695         bh = head = page_buffers(page);
1696
1697         clone_page = nilfs_alloc_private_page(bh->b_bdev, bh->b_size, 0);
1698         if (unlikely(!clone_page))
1699                 return -ENOMEM;
1700
1701         bh2 = page_buffers(clone_page);
1702         kaddr = kmap_atomic(page, KM_USER0);
1703         do {
1704                 if (list_empty(&bh->b_assoc_buffers))
1705                         continue;
1706                 get_bh(bh2);
1707                 page_cache_get(clone_page); /* for each bh */
1708                 memcpy(bh2->b_data, kaddr + bh_offset(bh), bh2->b_size);
1709                 bh2->b_blocknr = bh->b_blocknr;
1710                 list_replace(&bh->b_assoc_buffers, &bh2->b_assoc_buffers);
1711                 list_add_tail(&bh->b_assoc_buffers, out);
1712         } while (bh = bh->b_this_page, bh2 = bh2->b_this_page, bh != head);
1713         kunmap_atomic(kaddr, KM_USER0);
1714
1715         if (!TestSetPageWriteback(clone_page))
1716                 inc_zone_page_state(clone_page, NR_WRITEBACK);
1717         unlock_page(clone_page);
1718
1719         return 0;
1720 }
1721
1722 static int nilfs_test_page_to_be_frozen(struct page *page)
1723 {
1724         struct address_space *mapping = page->mapping;
1725
1726         if (!mapping || !mapping->host || S_ISDIR(mapping->host->i_mode))
1727                 return 0;
1728
1729         if (page_mapped(page)) {
1730                 ClearPageChecked(page);
1731                 return 1;
1732         }
1733         return PageChecked(page);
1734 }
1735
1736 static int nilfs_begin_page_io(struct page *page, struct list_head *out)
1737 {
1738         if (!page || PageWriteback(page))
1739                 /* For split b-tree node pages, this function may be called
1740                    twice.  We ignore the 2nd or later calls by this check. */
1741                 return 0;
1742
1743         lock_page(page);
1744         clear_page_dirty_for_io(page);
1745         set_page_writeback(page);
1746         unlock_page(page);
1747
1748         if (nilfs_test_page_to_be_frozen(page)) {
1749                 int err = nilfs_copy_replace_page_buffers(page, out);
1750                 if (unlikely(err))
1751                         return err;
1752         }
1753         return 0;
1754 }
1755
1756 static int nilfs_segctor_prepare_write(struct nilfs_sc_info *sci,
1757                                        struct page **failed_page)
1758 {
1759         struct nilfs_segment_buffer *segbuf;
1760         struct page *bd_page = NULL, *fs_page = NULL;
1761         struct list_head *list = &sci->sc_copied_buffers;
1762         int err;
1763
1764         *failed_page = NULL;
1765         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1766                 struct buffer_head *bh;
1767
1768                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1769                                     b_assoc_buffers) {
1770                         if (bh->b_page != bd_page) {
1771                                 if (bd_page) {
1772                                         lock_page(bd_page);
1773                                         clear_page_dirty_for_io(bd_page);
1774                                         set_page_writeback(bd_page);
1775                                         unlock_page(bd_page);
1776                                 }
1777                                 bd_page = bh->b_page;
1778                         }
1779                 }
1780
1781                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1782                                     b_assoc_buffers) {
1783                         if (bh == sci->sc_super_root) {
1784                                 if (bh->b_page != bd_page) {
1785                                         lock_page(bd_page);
1786                                         clear_page_dirty_for_io(bd_page);
1787                                         set_page_writeback(bd_page);
1788                                         unlock_page(bd_page);
1789                                         bd_page = bh->b_page;
1790                                 }
1791                                 break;
1792                         }
1793                         if (bh->b_page != fs_page) {
1794                                 err = nilfs_begin_page_io(fs_page, list);
1795                                 if (unlikely(err)) {
1796                                         *failed_page = fs_page;
1797                                         goto out;
1798                                 }
1799                                 fs_page = bh->b_page;
1800                         }
1801                 }
1802         }
1803         if (bd_page) {
1804                 lock_page(bd_page);
1805                 clear_page_dirty_for_io(bd_page);
1806                 set_page_writeback(bd_page);
1807                 unlock_page(bd_page);
1808         }
1809         err = nilfs_begin_page_io(fs_page, list);
1810         if (unlikely(err))
1811                 *failed_page = fs_page;
1812  out:
1813         return err;
1814 }
1815
1816 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1817                                struct backing_dev_info *bdi)
1818 {
1819         struct nilfs_segment_buffer *segbuf;
1820         struct nilfs_write_info wi;
1821         int err, res;
1822
1823         wi.sb = sci->sc_super;
1824         wi.bh_sr = sci->sc_super_root;
1825         wi.bdi = bdi;
1826
1827         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1828                 nilfs_segbuf_prepare_write(segbuf, &wi);
1829                 err = nilfs_segbuf_write(segbuf, &wi);
1830
1831                 res = nilfs_segbuf_wait(segbuf, &wi);
1832                 err = err ? : res;
1833                 if (err)
1834                         return err;
1835         }
1836         return 0;
1837 }
1838
1839 static void __nilfs_end_page_io(struct page *page, int err)
1840 {
1841         if (!err) {
1842                 if (!nilfs_page_buffers_clean(page))
1843                         __set_page_dirty_nobuffers(page);
1844                 ClearPageError(page);
1845         } else {
1846                 __set_page_dirty_nobuffers(page);
1847                 SetPageError(page);
1848         }
1849
1850         if (buffer_nilfs_allocated(page_buffers(page))) {
1851                 if (TestClearPageWriteback(page))
1852                         dec_zone_page_state(page, NR_WRITEBACK);
1853         } else
1854                 end_page_writeback(page);
1855 }
1856
1857 static void nilfs_end_page_io(struct page *page, int err)
1858 {
1859         if (!page)
1860                 return;
1861
1862         if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1863                 /*
1864                  * For b-tree node pages, this function may be called twice
1865                  * or more because they might be split in a segment.
1866                  */
1867                 if (PageDirty(page)) {
1868                         /*
1869                          * For pages holding split b-tree node buffers, dirty
1870                          * flag on the buffers may be cleared discretely.
1871                          * In that case, the page is once redirtied for
1872                          * remaining buffers, and it must be cancelled if
1873                          * all the buffers get cleaned later.
1874                          */
1875                         lock_page(page);
1876                         if (nilfs_page_buffers_clean(page))
1877                                 __nilfs_clear_page_dirty(page);
1878                         unlock_page(page);
1879                 }
1880                 return;
1881         }
1882
1883         __nilfs_end_page_io(page, err);
1884 }
1885
1886 static void nilfs_clear_copied_buffers(struct list_head *list, int err)
1887 {
1888         struct buffer_head *bh, *head;
1889         struct page *page;
1890
1891         while (!list_empty(list)) {
1892                 bh = list_entry(list->next, struct buffer_head,
1893                                 b_assoc_buffers);
1894                 page = bh->b_page;
1895                 page_cache_get(page);
1896                 head = bh = page_buffers(page);
1897                 do {
1898                         if (!list_empty(&bh->b_assoc_buffers)) {
1899                                 list_del_init(&bh->b_assoc_buffers);
1900                                 if (!err) {
1901                                         set_buffer_uptodate(bh);
1902                                         clear_buffer_dirty(bh);
1903                                         clear_buffer_nilfs_volatile(bh);
1904                                 }
1905                                 brelse(bh); /* for b_assoc_buffers */
1906                         }
1907                 } while ((bh = bh->b_this_page) != head);
1908
1909                 __nilfs_end_page_io(page, err);
1910                 page_cache_release(page);
1911         }
1912 }
1913
1914 static void nilfs_segctor_abort_write(struct nilfs_sc_info *sci,
1915                                       struct page *failed_page, int err)
1916 {
1917         struct nilfs_segment_buffer *segbuf;
1918         struct page *bd_page = NULL, *fs_page = NULL;
1919
1920         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1921                 struct buffer_head *bh;
1922
1923                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1924                                     b_assoc_buffers) {
1925                         if (bh->b_page != bd_page) {
1926                                 if (bd_page)
1927                                         end_page_writeback(bd_page);
1928                                 bd_page = bh->b_page;
1929                         }
1930                 }
1931
1932                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1933                                     b_assoc_buffers) {
1934                         if (bh == sci->sc_super_root) {
1935                                 if (bh->b_page != bd_page) {
1936                                         end_page_writeback(bd_page);
1937                                         bd_page = bh->b_page;
1938                                 }
1939                                 break;
1940                         }
1941                         if (bh->b_page != fs_page) {
1942                                 nilfs_end_page_io(fs_page, err);
1943                                 if (fs_page && fs_page == failed_page)
1944                                         goto done;
1945                                 fs_page = bh->b_page;
1946                         }
1947                 }
1948         }
1949         if (bd_page)
1950                 end_page_writeback(bd_page);
1951
1952         nilfs_end_page_io(fs_page, err);
1953  done:
1954         nilfs_clear_copied_buffers(&sci->sc_copied_buffers, err);
1955 }
1956
1957 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1958                                    struct nilfs_segment_buffer *segbuf)
1959 {
1960         nilfs->ns_segnum = segbuf->sb_segnum;
1961         nilfs->ns_nextnum = segbuf->sb_nextnum;
1962         nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1963                 + segbuf->sb_sum.nblocks;
1964         nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1965         nilfs->ns_ctime = segbuf->sb_sum.ctime;
1966 }
1967
1968 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1969 {
1970         struct nilfs_segment_buffer *segbuf;
1971         struct page *bd_page = NULL, *fs_page = NULL;
1972         struct nilfs_sb_info *sbi = sci->sc_sbi;
1973         struct the_nilfs *nilfs = sbi->s_nilfs;
1974         int update_sr = (sci->sc_super_root != NULL);
1975
1976         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1977                 struct buffer_head *bh;
1978
1979                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1980                                     b_assoc_buffers) {
1981                         set_buffer_uptodate(bh);
1982                         clear_buffer_dirty(bh);
1983                         if (bh->b_page != bd_page) {
1984                                 if (bd_page)
1985                                         end_page_writeback(bd_page);
1986                                 bd_page = bh->b_page;
1987                         }
1988                 }
1989                 /*
1990                  * We assume that the buffers which belong to the same page
1991                  * continue over the buffer list.
1992                  * Under this assumption, the last BHs of pages is
1993                  * identifiable by the discontinuity of bh->b_page
1994                  * (page != fs_page).
1995                  *
1996                  * For B-tree node blocks, however, this assumption is not
1997                  * guaranteed.  The cleanup code of B-tree node pages needs
1998                  * special care.
1999                  */
2000                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
2001                                     b_assoc_buffers) {
2002                         set_buffer_uptodate(bh);
2003                         clear_buffer_dirty(bh);
2004                         clear_buffer_nilfs_volatile(bh);
2005                         if (bh == sci->sc_super_root) {
2006                                 if (bh->b_page != bd_page) {
2007                                         end_page_writeback(bd_page);
2008                                         bd_page = bh->b_page;
2009                                 }
2010                                 break;
2011                         }
2012                         if (bh->b_page != fs_page) {
2013                                 nilfs_end_page_io(fs_page, 0);
2014                                 fs_page = bh->b_page;
2015                         }
2016                 }
2017
2018                 if (!NILFS_SEG_SIMPLEX(&segbuf->sb_sum)) {
2019                         if (NILFS_SEG_LOGBGN(&segbuf->sb_sum)) {
2020                                 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
2021                                 sci->sc_lseg_stime = jiffies;
2022                         }
2023                         if (NILFS_SEG_LOGEND(&segbuf->sb_sum))
2024                                 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
2025                 }
2026         }
2027         /*
2028          * Since pages may continue over multiple segment buffers,
2029          * end of the last page must be checked outside of the loop.
2030          */
2031         if (bd_page)
2032                 end_page_writeback(bd_page);
2033
2034         nilfs_end_page_io(fs_page, 0);
2035
2036         nilfs_clear_copied_buffers(&sci->sc_copied_buffers, 0);
2037
2038         nilfs_drop_collected_inodes(&sci->sc_dirty_files);
2039
2040         if (nilfs_doing_gc()) {
2041                 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
2042                 if (update_sr)
2043                         nilfs_commit_gcdat_inode(nilfs);
2044         } else
2045                 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
2046
2047         sci->sc_nblk_inc += sci->sc_nblk_this_inc;
2048
2049         segbuf = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2050         nilfs_set_next_segment(nilfs, segbuf);
2051
2052         if (update_sr) {
2053                 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
2054                                        segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
2055                 sbi->s_super->s_dirt = 1;
2056
2057                 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
2058                 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2059                 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
2060         } else
2061                 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
2062 }
2063
2064 static int nilfs_segctor_check_in_files(struct nilfs_sc_info *sci,
2065                                         struct nilfs_sb_info *sbi)
2066 {
2067         struct nilfs_inode_info *ii, *n;
2068         __u64 cno = sbi->s_nilfs->ns_cno;
2069
2070         spin_lock(&sbi->s_inode_lock);
2071  retry:
2072         list_for_each_entry_safe(ii, n, &sbi->s_dirty_files, i_dirty) {
2073                 if (!ii->i_bh) {
2074                         struct buffer_head *ibh;
2075                         int err;
2076
2077                         spin_unlock(&sbi->s_inode_lock);
2078                         err = nilfs_ifile_get_inode_block(
2079                                 sbi->s_ifile, ii->vfs_inode.i_ino, &ibh);
2080                         if (unlikely(err)) {
2081                                 nilfs_warning(sbi->s_super, __func__,
2082                                               "failed to get inode block.\n");
2083                                 return err;
2084                         }
2085                         nilfs_mdt_mark_buffer_dirty(ibh);
2086                         nilfs_mdt_mark_dirty(sbi->s_ifile);
2087                         spin_lock(&sbi->s_inode_lock);
2088                         if (likely(!ii->i_bh))
2089                                 ii->i_bh = ibh;
2090                         else
2091                                 brelse(ibh);
2092                         goto retry;
2093                 }
2094                 ii->i_cno = cno;
2095
2096                 clear_bit(NILFS_I_QUEUED, &ii->i_state);
2097                 set_bit(NILFS_I_BUSY, &ii->i_state);
2098                 list_del(&ii->i_dirty);
2099                 list_add_tail(&ii->i_dirty, &sci->sc_dirty_files);
2100         }
2101         spin_unlock(&sbi->s_inode_lock);
2102
2103         NILFS_I(sbi->s_ifile)->i_cno = cno;
2104
2105         return 0;
2106 }
2107
2108 static void nilfs_segctor_check_out_files(struct nilfs_sc_info *sci,
2109                                           struct nilfs_sb_info *sbi)
2110 {
2111         struct nilfs_transaction_info *ti = current->journal_info;
2112         struct nilfs_inode_info *ii, *n;
2113         __u64 cno = sbi->s_nilfs->ns_cno;
2114
2115         spin_lock(&sbi->s_inode_lock);
2116         list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2117                 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2118                     test_bit(NILFS_I_DIRTY, &ii->i_state)) {
2119                         /* The current checkpoint number (=nilfs->ns_cno) is
2120                            changed between check-in and check-out only if the
2121                            super root is written out.  So, we can update i_cno
2122                            for the inodes that remain in the dirty list. */
2123                         ii->i_cno = cno;
2124                         continue;
2125                 }
2126                 clear_bit(NILFS_I_BUSY, &ii->i_state);
2127                 brelse(ii->i_bh);
2128                 ii->i_bh = NULL;
2129                 list_del(&ii->i_dirty);
2130                 list_add_tail(&ii->i_dirty, &ti->ti_garbage);
2131         }
2132         spin_unlock(&sbi->s_inode_lock);
2133 }
2134
2135 /*
2136  * Main procedure of segment constructor
2137  */
2138 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2139 {
2140         struct nilfs_sb_info *sbi = sci->sc_sbi;
2141         struct the_nilfs *nilfs = sbi->s_nilfs;
2142         struct page *failed_page;
2143         int err, has_sr = 0;
2144
2145         sci->sc_stage.scnt = NILFS_ST_INIT;
2146
2147         err = nilfs_segctor_check_in_files(sci, sbi);
2148         if (unlikely(err))
2149                 goto out;
2150
2151         if (nilfs_test_metadata_dirty(sbi))
2152                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2153
2154         if (nilfs_segctor_clean(sci))
2155                 goto out;
2156
2157         do {
2158                 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2159
2160                 err = nilfs_segctor_begin_construction(sci, nilfs);
2161                 if (unlikely(err))
2162                         goto out;
2163
2164                 /* Update time stamp */
2165                 sci->sc_seg_ctime = get_seconds();
2166
2167                 err = nilfs_segctor_collect(sci, nilfs, mode);
2168                 if (unlikely(err))
2169                         goto failed;
2170
2171                 has_sr = (sci->sc_super_root != NULL);
2172
2173                 /* Avoid empty segment */
2174                 if (sci->sc_stage.scnt == NILFS_ST_DONE &&
2175                     NILFS_SEG_EMPTY(&sci->sc_curseg->sb_sum)) {
2176                         nilfs_segctor_end_construction(sci, nilfs, 1);
2177                         goto out;
2178                 }
2179
2180                 err = nilfs_segctor_assign(sci, mode);
2181                 if (unlikely(err))
2182                         goto failed;
2183
2184                 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2185                         nilfs_segctor_fill_in_file_bmap(sci, sbi->s_ifile);
2186
2187                 if (has_sr) {
2188                         err = nilfs_segctor_fill_in_checkpoint(sci);
2189                         if (unlikely(err))
2190                                 goto failed_to_make_up;
2191
2192                         nilfs_segctor_fill_in_super_root(sci, nilfs);
2193                 }
2194                 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2195
2196                 /* Write partial segments */
2197                 err = nilfs_segctor_prepare_write(sci, &failed_page);
2198                 if (unlikely(err))
2199                         goto failed_to_write;
2200
2201                 nilfs_segctor_fill_in_checksums(sci, nilfs->ns_crc_seed);
2202
2203                 err = nilfs_segctor_write(sci, nilfs->ns_bdi);
2204                 if (unlikely(err))
2205                         goto failed_to_write;
2206
2207                 nilfs_segctor_complete_write(sci);
2208
2209                 /* Commit segments */
2210                 if (has_sr)
2211                         nilfs_segctor_clear_metadata_dirty(sci);
2212
2213                 nilfs_segctor_end_construction(sci, nilfs, 0);
2214
2215         } while (sci->sc_stage.scnt != NILFS_ST_DONE);
2216
2217  out:
2218         nilfs_segctor_destroy_segment_buffers(sci);
2219         nilfs_segctor_check_out_files(sci, sbi);
2220         return err;
2221
2222  failed_to_write:
2223         nilfs_segctor_abort_write(sci, failed_page, err);
2224         nilfs_segctor_cancel_segusage(sci, nilfs->ns_sufile);
2225
2226  failed_to_make_up:
2227         if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2228                 nilfs_redirty_inodes(&sci->sc_dirty_files);
2229
2230  failed:
2231         if (nilfs_doing_gc())
2232                 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2233         nilfs_segctor_end_construction(sci, nilfs, err);
2234         goto out;
2235 }
2236
2237 /**
2238  * nilfs_secgtor_start_timer - set timer of background write
2239  * @sci: nilfs_sc_info
2240  *
2241  * If the timer has already been set, it ignores the new request.
2242  * This function MUST be called within a section locking the segment
2243  * semaphore.
2244  */
2245 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2246 {
2247         spin_lock(&sci->sc_state_lock);
2248         if (sci->sc_timer && !(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2249                 sci->sc_timer->expires = jiffies + sci->sc_interval;
2250                 add_timer(sci->sc_timer);
2251                 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2252         }
2253         spin_unlock(&sci->sc_state_lock);
2254 }
2255
2256 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2257 {
2258         spin_lock(&sci->sc_state_lock);
2259         if (!(sci->sc_flush_request & (1 << bn))) {
2260                 unsigned long prev_req = sci->sc_flush_request;
2261
2262                 sci->sc_flush_request |= (1 << bn);
2263                 if (!prev_req)
2264                         wake_up(&sci->sc_wait_daemon);
2265         }
2266         spin_unlock(&sci->sc_state_lock);
2267 }
2268
2269 /**
2270  * nilfs_flush_segment - trigger a segment construction for resource control
2271  * @sb: super block
2272  * @ino: inode number of the file to be flushed out.
2273  */
2274 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2275 {
2276         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2277         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2278
2279         if (!sci || nilfs_doing_construction())
2280                 return;
2281         nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2282                                         /* assign bit 0 to data files */
2283 }
2284
2285 struct nilfs_segctor_wait_request {
2286         wait_queue_t    wq;
2287         __u32           seq;
2288         int             err;
2289         atomic_t        done;
2290 };
2291
2292 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2293 {
2294         struct nilfs_segctor_wait_request wait_req;
2295         int err = 0;
2296
2297         spin_lock(&sci->sc_state_lock);
2298         init_wait(&wait_req.wq);
2299         wait_req.err = 0;
2300         atomic_set(&wait_req.done, 0);
2301         wait_req.seq = ++sci->sc_seq_request;
2302         spin_unlock(&sci->sc_state_lock);
2303
2304         init_waitqueue_entry(&wait_req.wq, current);
2305         add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2306         set_current_state(TASK_INTERRUPTIBLE);
2307         wake_up(&sci->sc_wait_daemon);
2308
2309         for (;;) {
2310                 if (atomic_read(&wait_req.done)) {
2311                         err = wait_req.err;
2312                         break;
2313                 }
2314                 if (!signal_pending(current)) {
2315                         schedule();
2316                         continue;
2317                 }
2318                 err = -ERESTARTSYS;
2319                 break;
2320         }
2321         finish_wait(&sci->sc_wait_request, &wait_req.wq);
2322         return err;
2323 }
2324
2325 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2326 {
2327         struct nilfs_segctor_wait_request *wrq, *n;
2328         unsigned long flags;
2329
2330         spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2331         list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2332                                  wq.task_list) {
2333                 if (!atomic_read(&wrq->done) &&
2334                     nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2335                         wrq->err = err;
2336                         atomic_set(&wrq->done, 1);
2337                 }
2338                 if (atomic_read(&wrq->done)) {
2339                         wrq->wq.func(&wrq->wq,
2340                                      TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2341                                      0, NULL);
2342                 }
2343         }
2344         spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2345 }
2346
2347 /**
2348  * nilfs_construct_segment - construct a logical segment
2349  * @sb: super block
2350  *
2351  * Return Value: On success, 0 is retured. On errors, one of the following
2352  * negative error code is returned.
2353  *
2354  * %-EROFS - Read only filesystem.
2355  *
2356  * %-EIO - I/O error
2357  *
2358  * %-ENOSPC - No space left on device (only in a panic state).
2359  *
2360  * %-ERESTARTSYS - Interrupted.
2361  *
2362  * %-ENOMEM - Insufficient memory available.
2363  */
2364 int nilfs_construct_segment(struct super_block *sb)
2365 {
2366         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2367         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2368         struct nilfs_transaction_info *ti;
2369         int err;
2370
2371         if (!sci)
2372                 return -EROFS;
2373
2374         /* A call inside transactions causes a deadlock. */
2375         BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2376
2377         err = nilfs_segctor_sync(sci);
2378         return err;
2379 }
2380
2381 /**
2382  * nilfs_construct_dsync_segment - construct a data-only logical segment
2383  * @sb: super block
2384  * @inode: inode whose data blocks should be written out
2385  * @start: start byte offset
2386  * @end: end byte offset (inclusive)
2387  *
2388  * Return Value: On success, 0 is retured. On errors, one of the following
2389  * negative error code is returned.
2390  *
2391  * %-EROFS - Read only filesystem.
2392  *
2393  * %-EIO - I/O error
2394  *
2395  * %-ENOSPC - No space left on device (only in a panic state).
2396  *
2397  * %-ERESTARTSYS - Interrupted.
2398  *
2399  * %-ENOMEM - Insufficient memory available.
2400  */
2401 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2402                                   loff_t start, loff_t end)
2403 {
2404         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2405         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2406         struct nilfs_inode_info *ii;
2407         struct nilfs_transaction_info ti;
2408         int err = 0;
2409
2410         if (!sci)
2411                 return -EROFS;
2412
2413         nilfs_transaction_lock(sbi, &ti, 0);
2414
2415         ii = NILFS_I(inode);
2416         if (test_bit(NILFS_I_INODE_DIRTY, &ii->i_state) ||
2417             nilfs_test_opt(sbi, STRICT_ORDER) ||
2418             test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2419             nilfs_discontinued(sbi->s_nilfs)) {
2420                 nilfs_transaction_unlock(sbi);
2421                 err = nilfs_segctor_sync(sci);
2422                 return err;
2423         }
2424
2425         spin_lock(&sbi->s_inode_lock);
2426         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2427             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2428                 spin_unlock(&sbi->s_inode_lock);
2429                 nilfs_transaction_unlock(sbi);
2430                 return 0;
2431         }
2432         spin_unlock(&sbi->s_inode_lock);
2433         sci->sc_dsync_inode = ii;
2434         sci->sc_dsync_start = start;
2435         sci->sc_dsync_end = end;
2436
2437         err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2438
2439         nilfs_transaction_unlock(sbi);
2440         return err;
2441 }
2442
2443 struct nilfs_segctor_req {
2444         int mode;
2445         __u32 seq_accepted;
2446         int sc_err;  /* construction failure */
2447         int sb_err;  /* super block writeback failure */
2448 };
2449
2450 #define FLUSH_FILE_BIT  (0x1) /* data file only */
2451 #define FLUSH_DAT_BIT   (1 << NILFS_DAT_INO) /* DAT only */
2452
2453 static void nilfs_segctor_accept(struct nilfs_sc_info *sci,
2454                                  struct nilfs_segctor_req *req)
2455 {
2456         req->sc_err = req->sb_err = 0;
2457         spin_lock(&sci->sc_state_lock);
2458         req->seq_accepted = sci->sc_seq_request;
2459         spin_unlock(&sci->sc_state_lock);
2460
2461         if (sci->sc_timer)
2462                 del_timer_sync(sci->sc_timer);
2463 }
2464
2465 static void nilfs_segctor_notify(struct nilfs_sc_info *sci,
2466                                  struct nilfs_segctor_req *req)
2467 {
2468         /* Clear requests (even when the construction failed) */
2469         spin_lock(&sci->sc_state_lock);
2470
2471         sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2472
2473         if (req->mode == SC_LSEG_SR) {
2474                 sci->sc_seq_done = req->seq_accepted;
2475                 nilfs_segctor_wakeup(sci, req->sc_err ? : req->sb_err);
2476                 sci->sc_flush_request = 0;
2477         } else if (req->mode == SC_FLUSH_FILE)
2478                 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2479         else if (req->mode == SC_FLUSH_DAT)
2480                 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2481
2482         spin_unlock(&sci->sc_state_lock);
2483 }
2484
2485 static int nilfs_segctor_construct(struct nilfs_sc_info *sci,
2486                                    struct nilfs_segctor_req *req)
2487 {
2488         struct nilfs_sb_info *sbi = sci->sc_sbi;
2489         struct the_nilfs *nilfs = sbi->s_nilfs;
2490         int err = 0;
2491
2492         if (nilfs_discontinued(nilfs))
2493                 req->mode = SC_LSEG_SR;
2494         if (!nilfs_segctor_confirm(sci)) {
2495                 err = nilfs_segctor_do_construct(sci, req->mode);
2496                 req->sc_err = err;
2497         }
2498         if (likely(!err)) {
2499                 if (req->mode != SC_FLUSH_DAT)
2500                         atomic_set(&nilfs->ns_ndirtyblks, 0);
2501                 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2502                     nilfs_discontinued(nilfs)) {
2503                         down_write(&nilfs->ns_sem);
2504                         req->sb_err = nilfs_commit_super(sbi, 0);
2505                         up_write(&nilfs->ns_sem);
2506                 }
2507         }
2508         return err;
2509 }
2510
2511 static void nilfs_construction_timeout(unsigned long data)
2512 {
2513         struct task_struct *p = (struct task_struct *)data;
2514         wake_up_process(p);
2515 }
2516
2517 static void
2518 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2519 {
2520         struct nilfs_inode_info *ii, *n;
2521
2522         list_for_each_entry_safe(ii, n, head, i_dirty) {
2523                 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2524                         continue;
2525                 hlist_del_init(&ii->vfs_inode.i_hash);
2526                 list_del_init(&ii->i_dirty);
2527                 nilfs_clear_gcinode(&ii->vfs_inode);
2528         }
2529 }
2530
2531 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2532                          void **kbufs)
2533 {
2534         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2535         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2536         struct the_nilfs *nilfs = sbi->s_nilfs;
2537         struct nilfs_transaction_info ti;
2538         struct nilfs_segctor_req req = { .mode = SC_LSEG_SR };
2539         int err;
2540
2541         if (unlikely(!sci))
2542                 return -EROFS;
2543
2544         nilfs_transaction_lock(sbi, &ti, 1);
2545
2546         err = nilfs_init_gcdat_inode(nilfs);
2547         if (unlikely(err))
2548                 goto out_unlock;
2549
2550         err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2551         if (unlikely(err))
2552                 goto out_unlock;
2553
2554         sci->sc_freesegs = kbufs[4];
2555         sci->sc_nfreesegs = argv[4].v_nmembs;
2556         list_splice_init(&nilfs->ns_gc_inodes, sci->sc_gc_inodes.prev);
2557
2558         for (;;) {
2559                 nilfs_segctor_accept(sci, &req);
2560                 err = nilfs_segctor_construct(sci, &req);
2561                 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2562                 nilfs_segctor_notify(sci, &req);
2563
2564                 if (likely(!err))
2565                         break;
2566
2567                 nilfs_warning(sb, __func__,
2568                               "segment construction failed. (err=%d)", err);
2569                 set_current_state(TASK_INTERRUPTIBLE);
2570                 schedule_timeout(sci->sc_interval);
2571         }
2572
2573  out_unlock:
2574         sci->sc_freesegs = NULL;
2575         sci->sc_nfreesegs = 0;
2576         nilfs_clear_gcdat_inode(nilfs);
2577         nilfs_transaction_unlock(sbi);
2578         return err;
2579 }
2580
2581 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2582 {
2583         struct nilfs_sb_info *sbi = sci->sc_sbi;
2584         struct nilfs_transaction_info ti;
2585         struct nilfs_segctor_req req = { .mode = mode };
2586
2587         nilfs_transaction_lock(sbi, &ti, 0);
2588
2589         nilfs_segctor_accept(sci, &req);
2590         nilfs_segctor_construct(sci, &req);
2591         nilfs_segctor_notify(sci, &req);
2592
2593         /*
2594          * Unclosed segment should be retried.  We do this using sc_timer.
2595          * Timeout of sc_timer will invoke complete construction which leads
2596          * to close the current logical segment.
2597          */
2598         if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2599                 nilfs_segctor_start_timer(sci);
2600
2601         nilfs_transaction_unlock(sbi);
2602 }
2603
2604 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2605 {
2606         int mode = 0;
2607         int err;
2608
2609         spin_lock(&sci->sc_state_lock);
2610         mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2611                 SC_FLUSH_DAT : SC_FLUSH_FILE;
2612         spin_unlock(&sci->sc_state_lock);
2613
2614         if (mode) {
2615                 err = nilfs_segctor_do_construct(sci, mode);
2616
2617                 spin_lock(&sci->sc_state_lock);
2618                 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2619                         ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2620                 spin_unlock(&sci->sc_state_lock);
2621         }
2622         clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2623 }
2624
2625 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2626 {
2627         if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2628             time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2629                 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2630                         return SC_FLUSH_FILE;
2631                 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2632                         return SC_FLUSH_DAT;
2633         }
2634         return SC_LSEG_SR;
2635 }
2636
2637 /**
2638  * nilfs_segctor_thread - main loop of the segment constructor thread.
2639  * @arg: pointer to a struct nilfs_sc_info.
2640  *
2641  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2642  * to execute segment constructions.
2643  */
2644 static int nilfs_segctor_thread(void *arg)
2645 {
2646         struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2647         struct timer_list timer;
2648         int timeout = 0;
2649
2650         init_timer(&timer);
2651         timer.data = (unsigned long)current;
2652         timer.function = nilfs_construction_timeout;
2653         sci->sc_timer = &timer;
2654
2655         /* start sync. */
2656         sci->sc_task = current;
2657         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2658         printk(KERN_INFO
2659                "segctord starting. Construction interval = %lu seconds, "
2660                "CP frequency < %lu seconds\n",
2661                sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2662
2663         spin_lock(&sci->sc_state_lock);
2664  loop:
2665         for (;;) {
2666                 int mode;
2667
2668                 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2669                         goto end_thread;
2670
2671                 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2672                         mode = SC_LSEG_SR;
2673                 else if (!sci->sc_flush_request)
2674                         break;
2675                 else
2676                         mode = nilfs_segctor_flush_mode(sci);
2677
2678                 spin_unlock(&sci->sc_state_lock);
2679                 nilfs_segctor_thread_construct(sci, mode);
2680                 spin_lock(&sci->sc_state_lock);
2681                 timeout = 0;
2682         }
2683
2684
2685         if (freezing(current)) {
2686                 spin_unlock(&sci->sc_state_lock);
2687                 refrigerator();
2688                 spin_lock(&sci->sc_state_lock);
2689         } else {
2690                 DEFINE_WAIT(wait);
2691                 int should_sleep = 1;
2692
2693                 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2694                                 TASK_INTERRUPTIBLE);
2695
2696                 if (sci->sc_seq_request != sci->sc_seq_done)
2697                         should_sleep = 0;
2698                 else if (sci->sc_flush_request)
2699                         should_sleep = 0;
2700                 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2701                         should_sleep = time_before(jiffies,
2702                                                    sci->sc_timer->expires);
2703
2704                 if (should_sleep) {
2705                         spin_unlock(&sci->sc_state_lock);
2706                         schedule();
2707                         spin_lock(&sci->sc_state_lock);
2708                 }
2709                 finish_wait(&sci->sc_wait_daemon, &wait);
2710                 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2711                            time_after_eq(jiffies, sci->sc_timer->expires));
2712         }
2713         goto loop;
2714
2715  end_thread:
2716         spin_unlock(&sci->sc_state_lock);
2717         del_timer_sync(sci->sc_timer);
2718         sci->sc_timer = NULL;
2719
2720         /* end sync. */
2721         sci->sc_task = NULL;
2722         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2723         return 0;
2724 }
2725
2726 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2727 {
2728         struct task_struct *t;
2729
2730         t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2731         if (IS_ERR(t)) {
2732                 int err = PTR_ERR(t);
2733
2734                 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2735                        err);
2736                 return err;
2737         }
2738         wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2739         return 0;
2740 }
2741
2742 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2743 {
2744         sci->sc_state |= NILFS_SEGCTOR_QUIT;
2745
2746         while (sci->sc_task) {
2747                 wake_up(&sci->sc_wait_daemon);
2748                 spin_unlock(&sci->sc_state_lock);
2749                 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2750                 spin_lock(&sci->sc_state_lock);
2751         }
2752 }
2753
2754 static int nilfs_segctor_init(struct nilfs_sc_info *sci)
2755 {
2756         sci->sc_seq_done = sci->sc_seq_request;
2757
2758         return nilfs_segctor_start_thread(sci);
2759 }
2760
2761 /*
2762  * Setup & clean-up functions
2763  */
2764 static struct nilfs_sc_info *nilfs_segctor_new(struct nilfs_sb_info *sbi)
2765 {
2766         struct nilfs_sc_info *sci;
2767
2768         sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2769         if (!sci)
2770                 return NULL;
2771
2772         sci->sc_sbi = sbi;
2773         sci->sc_super = sbi->s_super;
2774
2775         init_waitqueue_head(&sci->sc_wait_request);
2776         init_waitqueue_head(&sci->sc_wait_daemon);
2777         init_waitqueue_head(&sci->sc_wait_task);
2778         spin_lock_init(&sci->sc_state_lock);
2779         INIT_LIST_HEAD(&sci->sc_dirty_files);
2780         INIT_LIST_HEAD(&sci->sc_segbufs);
2781         INIT_LIST_HEAD(&sci->sc_gc_inodes);
2782         INIT_LIST_HEAD(&sci->sc_copied_buffers);
2783
2784         sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2785         sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2786         sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2787
2788         if (sbi->s_interval)
2789                 sci->sc_interval = sbi->s_interval;
2790         if (sbi->s_watermark)
2791                 sci->sc_watermark = sbi->s_watermark;
2792         return sci;
2793 }
2794
2795 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2796 {
2797         int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2798
2799         /* The segctord thread was stopped and its timer was removed.
2800            But some tasks remain. */
2801         do {
2802                 struct nilfs_sb_info *sbi = sci->sc_sbi;
2803                 struct nilfs_transaction_info ti;
2804                 struct nilfs_segctor_req req = { .mode = SC_LSEG_SR };
2805
2806                 nilfs_transaction_lock(sbi, &ti, 0);
2807                 nilfs_segctor_accept(sci, &req);
2808                 ret = nilfs_segctor_construct(sci, &req);
2809                 nilfs_segctor_notify(sci, &req);
2810                 nilfs_transaction_unlock(sbi);
2811
2812         } while (ret && retrycount-- > 0);
2813 }
2814
2815 /**
2816  * nilfs_segctor_destroy - destroy the segment constructor.
2817  * @sci: nilfs_sc_info
2818  *
2819  * nilfs_segctor_destroy() kills the segctord thread and frees
2820  * the nilfs_sc_info struct.
2821  * Caller must hold the segment semaphore.
2822  */
2823 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2824 {
2825         struct nilfs_sb_info *sbi = sci->sc_sbi;
2826         int flag;
2827
2828         up_write(&sbi->s_nilfs->ns_segctor_sem);
2829
2830         spin_lock(&sci->sc_state_lock);
2831         nilfs_segctor_kill_thread(sci);
2832         flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2833                 || sci->sc_seq_request != sci->sc_seq_done);
2834         spin_unlock(&sci->sc_state_lock);
2835
2836         if (flag || nilfs_segctor_confirm(sci))
2837                 nilfs_segctor_write_out(sci);
2838
2839         WARN_ON(!list_empty(&sci->sc_copied_buffers));
2840
2841         if (!list_empty(&sci->sc_dirty_files)) {
2842                 nilfs_warning(sbi->s_super, __func__,
2843                               "dirty file(s) after the final construction\n");
2844                 nilfs_dispose_list(sbi, &sci->sc_dirty_files, 1);
2845         }
2846
2847         WARN_ON(!list_empty(&sci->sc_segbufs));
2848
2849         down_write(&sbi->s_nilfs->ns_segctor_sem);
2850
2851         kfree(sci);
2852 }
2853
2854 /**
2855  * nilfs_attach_segment_constructor - attach a segment constructor
2856  * @sbi: nilfs_sb_info
2857  *
2858  * nilfs_attach_segment_constructor() allocates a struct nilfs_sc_info,
2859  * initilizes it, and starts the segment constructor.
2860  *
2861  * Return Value: On success, 0 is returned. On error, one of the following
2862  * negative error code is returned.
2863  *
2864  * %-ENOMEM - Insufficient memory available.
2865  */
2866 int nilfs_attach_segment_constructor(struct nilfs_sb_info *sbi)
2867 {
2868         struct the_nilfs *nilfs = sbi->s_nilfs;
2869         int err;
2870
2871         /* Each field of nilfs_segctor is cleared through the initialization
2872            of super-block info */
2873         sbi->s_sc_info = nilfs_segctor_new(sbi);
2874         if (!sbi->s_sc_info)
2875                 return -ENOMEM;
2876
2877         nilfs_attach_writer(nilfs, sbi);
2878         err = nilfs_segctor_init(NILFS_SC(sbi));
2879         if (err) {
2880                 nilfs_detach_writer(nilfs, sbi);
2881                 kfree(sbi->s_sc_info);
2882                 sbi->s_sc_info = NULL;
2883         }
2884         return err;
2885 }
2886
2887 /**
2888  * nilfs_detach_segment_constructor - destroy the segment constructor
2889  * @sbi: nilfs_sb_info
2890  *
2891  * nilfs_detach_segment_constructor() kills the segment constructor daemon,
2892  * frees the struct nilfs_sc_info, and destroy the dirty file list.
2893  */
2894 void nilfs_detach_segment_constructor(struct nilfs_sb_info *sbi)
2895 {
2896         struct the_nilfs *nilfs = sbi->s_nilfs;
2897         LIST_HEAD(garbage_list);
2898
2899         down_write(&nilfs->ns_segctor_sem);
2900         if (NILFS_SC(sbi)) {
2901                 nilfs_segctor_destroy(NILFS_SC(sbi));
2902                 sbi->s_sc_info = NULL;
2903         }
2904
2905         /* Force to free the list of dirty files */
2906         spin_lock(&sbi->s_inode_lock);
2907         if (!list_empty(&sbi->s_dirty_files)) {
2908                 list_splice_init(&sbi->s_dirty_files, &garbage_list);
2909                 nilfs_warning(sbi->s_super, __func__,
2910                               "Non empty dirty list after the last "
2911                               "segment construction\n");
2912         }
2913         spin_unlock(&sbi->s_inode_lock);
2914         up_write(&nilfs->ns_segctor_sem);
2915
2916         nilfs_dispose_list(sbi, &garbage_list, 1);
2917         nilfs_detach_writer(nilfs, sbi);
2918 }