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