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  * Written by Ryusuke Konishi.
  17  *
  18  */
  19
  20 #include <linux/pagemap.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/writeback.h>
  23 #include <linux/bitops.h>
  24 #include <linux/bio.h>
  25 #include <linux/completion.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/freezer.h>
  29 #include <linux/kthread.h>
  30 #include <linux/crc32.h>
  31 #include <linux/pagevec.h>
  32 #include <linux/slab.h>
  33 #include <linux/sched/signal.h>
  34
  35 #include "nilfs.h"
  36 #include "btnode.h"
  37 #include "page.h"
  38 #include "segment.h"
  39 #include "sufile.h"
  40 #include "cpfile.h"
  41 #include "ifile.h"
  42 #include "segbuf.h"
  43
  44
  45 /*
  46  * Segment constructor
  47  */
  48 #define SC_N_INODEVEC   16   /* Size of locally allocated inode vector */
  49
  50 #define SC_MAX_SEGDELTA 64   /*
  51                               * Upper limit of the number of segments
  52                               * appended in collection retry loop
  53                               */
  54
  55 /* Construction mode */
  56 enum {
  57         SC_LSEG_SR = 1, /* Make a logical segment having a super root */
  58         SC_LSEG_DSYNC,  /*
  59                          * Flush data blocks of a given file and make
  60                          * a logical segment without a super root.
  61                          */
  62         SC_FLUSH_FILE,  /*
  63                          * Flush data files, leads to segment writes without
  64                          * creating a checkpoint.
  65                          */
  66         SC_FLUSH_DAT,   /*
  67                          * Flush DAT file.  This also creates segments
  68                          * without a checkpoint.
  69                          */
  70 };
  71
  72 /* Stage numbers of dirty block collection */
  73 enum {
  74         NILFS_ST_INIT = 0,
  75         NILFS_ST_GC,            /* Collecting dirty blocks for GC */
  76         NILFS_ST_FILE,
  77         NILFS_ST_IFILE,
  78         NILFS_ST_CPFILE,
  79         NILFS_ST_SUFILE,
  80         NILFS_ST_DAT,
  81         NILFS_ST_SR,            /* Super root */
  82         NILFS_ST_DSYNC,         /* Data sync blocks */
  83         NILFS_ST_DONE,
  84 };
  85
  86 #define CREATE_TRACE_POINTS
  87 #include <trace/events/nilfs2.h>
  88
  89 /*
  90  * nilfs_sc_cstage_inc(), nilfs_sc_cstage_set(), nilfs_sc_cstage_get() are
  91  * wrapper functions of stage count (nilfs_sc_info->sc_stage.scnt). Users of
  92  * the variable must use them because transition of stage count must involve
  93  * trace events (trace_nilfs2_collection_stage_transition).
  94  *
  95  * nilfs_sc_cstage_get() isn't required for the above purpose because it doesn't
  96  * produce tracepoint events. It is provided just for making the intention
  97  * clear.
  98  */
  99 static inline void nilfs_sc_cstage_inc(struct nilfs_sc_info *sci)
 100 {
 101         sci->sc_stage.scnt++;
 102         trace_nilfs2_collection_stage_transition(sci);
 103 }
 104
 105 static inline void nilfs_sc_cstage_set(struct nilfs_sc_info *sci, int next_scnt)
 106 {
 107         sci->sc_stage.scnt = next_scnt;
 108         trace_nilfs2_collection_stage_transition(sci);
 109 }
 110
 111 static inline int nilfs_sc_cstage_get(struct nilfs_sc_info *sci)
 112 {
 113         return sci->sc_stage.scnt;
 114 }
 115
 116 /* State flags of collection */
 117 #define NILFS_CF_NODE           0x0001  /* Collecting node blocks */
 118 #define NILFS_CF_IFILE_STARTED  0x0002  /* IFILE stage has started */
 119 #define NILFS_CF_SUFREED        0x0004  /* segment usages has been freed */
 120 #define NILFS_CF_HISTORY_MASK   (NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
 121
 122 /* Operations depending on the construction mode and file type */
 123 struct nilfs_sc_operations {
 124         int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
 125                             struct inode *);
 126         int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
 127                             struct inode *);
 128         int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
 129                             struct inode *);
 130         void (*write_data_binfo)(struct nilfs_sc_info *,
 131                                  struct nilfs_segsum_pointer *,
 132                                  union nilfs_binfo *);
 133         void (*write_node_binfo)(struct nilfs_sc_info *,
 134                                  struct nilfs_segsum_pointer *,
 135                                  union nilfs_binfo *);
 136 };
 137
 138 /*
 139  * Other definitions
 140  */
 141 static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
 142 static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
 143 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
 144 static void nilfs_dispose_list(struct the_nilfs *, struct list_head *, int);
 145
 146 #define nilfs_cnt32_gt(a, b)   \
 147         (typecheck(__u32, a) && typecheck(__u32, b) && \
 148          ((__s32)(b) - (__s32)(a) < 0))
 149 #define nilfs_cnt32_ge(a, b)   \
 150         (typecheck(__u32, a) && typecheck(__u32, b) && \
 151          ((__s32)(a) - (__s32)(b) >= 0))
 152 #define nilfs_cnt32_lt(a, b)  nilfs_cnt32_gt(b, a)
 153 #define nilfs_cnt32_le(a, b)  nilfs_cnt32_ge(b, a)
 154
 155 static int nilfs_prepare_segment_lock(struct super_block *sb,
 156                                       struct nilfs_transaction_info *ti)
 157 {
 158         struct nilfs_transaction_info *cur_ti = current->journal_info;
 159         void *save = NULL;
 160
 161         if (cur_ti) {
 162                 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
 163                         return ++cur_ti->ti_count;
 164
 165                 /*
 166                  * If journal_info field is occupied by other FS,
 167                  * it is saved and will be restored on
 168                  * nilfs_transaction_commit().
 169                  */
 170                 nilfs_msg(sb, KERN_WARNING, "journal info from a different FS");
 171                 save = current->journal_info;
 172         }
 173         if (!ti) {
 174                 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
 175                 if (!ti)
 176                         return -ENOMEM;
 177                 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
 178         } else {
 179                 ti->ti_flags = 0;
 180         }
 181         ti->ti_count = 0;
 182         ti->ti_save = save;
 183         ti->ti_magic = NILFS_TI_MAGIC;
 184         current->journal_info = ti;
 185         return 0;
 186 }
 187
 188 /**
 189  * nilfs_transaction_begin - start indivisible file operations.
 190  * @sb: super block
 191  * @ti: nilfs_transaction_info
 192  * @vacancy_check: flags for vacancy rate checks
 193  *
 194  * nilfs_transaction_begin() acquires a reader/writer semaphore, called
 195  * the segment semaphore, to make a segment construction and write tasks
 196  * exclusive.  The function is used with nilfs_transaction_commit() in pairs.
 197  * The region enclosed by these two functions can be nested.  To avoid a
 198  * deadlock, the semaphore is only acquired or released in the outermost call.
 199  *
 200  * This function allocates a nilfs_transaction_info struct to keep context
 201  * information on it.  It is initialized and hooked onto the current task in
 202  * the outermost call.  If a pre-allocated struct is given to @ti, it is used
 203  * instead; otherwise a new struct is assigned from a slab.
 204  *
 205  * When @vacancy_check flag is set, this function will check the amount of
 206  * free space, and will wait for the GC to reclaim disk space if low capacity.
 207  *
 208  * Return Value: On success, 0 is returned. On error, one of the following
 209  * negative error code is returned.
 210  *
 211  * %-ENOMEM - Insufficient memory available.
 212  *
 213  * %-ENOSPC - No space left on device
 214  */
 215 int nilfs_transaction_begin(struct super_block *sb,
 216                             struct nilfs_transaction_info *ti,
 217                             int vacancy_check)
 218 {
 219         struct the_nilfs *nilfs;
 220         int ret = nilfs_prepare_segment_lock(sb, ti);
 221         struct nilfs_transaction_info *trace_ti;
 222
 223         if (unlikely(ret < 0))
 224                 return ret;
 225         if (ret > 0) {
 226                 trace_ti = current->journal_info;
 227
 228                 trace_nilfs2_transaction_transition(sb, trace_ti,
 229                                     trace_ti->ti_count, trace_ti->ti_flags,
 230                                     TRACE_NILFS2_TRANSACTION_BEGIN);
 231                 return 0;
 232         }
 233
 234         sb_start_intwrite(sb);
 235
 236         nilfs = sb->s_fs_info;
 237         down_read(&nilfs->ns_segctor_sem);
 238         if (vacancy_check && nilfs_near_disk_full(nilfs)) {
 239                 up_read(&nilfs->ns_segctor_sem);
 240                 ret = -ENOSPC;
 241                 goto failed;
 242         }
 243
 244         trace_ti = current->journal_info;
 245         trace_nilfs2_transaction_transition(sb, trace_ti, trace_ti->ti_count,
 246                                             trace_ti->ti_flags,
 247                                             TRACE_NILFS2_TRANSACTION_BEGIN);
 248         return 0;
 249
 250  failed:
 251         ti = current->journal_info;
 252         current->journal_info = ti->ti_save;
 253         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 254                 kmem_cache_free(nilfs_transaction_cachep, ti);
 255         sb_end_intwrite(sb);
 256         return ret;
 257 }
 258
 259 /**
 260  * nilfs_transaction_commit - commit indivisible file operations.
 261  * @sb: super block
 262  *
 263  * nilfs_transaction_commit() releases the read semaphore which is
 264  * acquired by nilfs_transaction_begin(). This is only performed
 265  * in outermost call of this function.  If a commit flag is set,
 266  * nilfs_transaction_commit() sets a timer to start the segment
 267  * constructor.  If a sync flag is set, it starts construction
 268  * directly.
 269  */
 270 int nilfs_transaction_commit(struct super_block *sb)
 271 {
 272         struct nilfs_transaction_info *ti = current->journal_info;
 273         struct the_nilfs *nilfs = sb->s_fs_info;
 274         int err = 0;
 275
 276         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 277         ti->ti_flags |= NILFS_TI_COMMIT;
 278         if (ti->ti_count > 0) {
 279                 ti->ti_count--;
 280                 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 281                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
 282                 return 0;
 283         }
 284         if (nilfs->ns_writer) {
 285                 struct nilfs_sc_info *sci = nilfs->ns_writer;
 286
 287                 if (ti->ti_flags & NILFS_TI_COMMIT)
 288                         nilfs_segctor_start_timer(sci);
 289                 if (atomic_read(&nilfs->ns_ndirtyblks) > sci->sc_watermark)
 290                         nilfs_segctor_do_flush(sci, 0);
 291         }
 292         up_read(&nilfs->ns_segctor_sem);
 293         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 294                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
 295
 296         current->journal_info = ti->ti_save;
 297
 298         if (ti->ti_flags & NILFS_TI_SYNC)
 299                 err = nilfs_construct_segment(sb);
 300         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 301                 kmem_cache_free(nilfs_transaction_cachep, ti);
 302         sb_end_intwrite(sb);
 303         return err;
 304 }
 305
 306 void nilfs_transaction_abort(struct super_block *sb)
 307 {
 308         struct nilfs_transaction_info *ti = current->journal_info;
 309         struct the_nilfs *nilfs = sb->s_fs_info;
 310
 311         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 312         if (ti->ti_count > 0) {
 313                 ti->ti_count--;
 314                 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 315                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
 316                 return;
 317         }
 318         up_read(&nilfs->ns_segctor_sem);
 319
 320         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 321                     ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
 322
 323         current->journal_info = ti->ti_save;
 324         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 325                 kmem_cache_free(nilfs_transaction_cachep, ti);
 326         sb_end_intwrite(sb);
 327 }
 328
 329 void nilfs_relax_pressure_in_lock(struct super_block *sb)
 330 {
 331         struct the_nilfs *nilfs = sb->s_fs_info;
 332         struct nilfs_sc_info *sci = nilfs->ns_writer;
 333
 334         if (!sci || !sci->sc_flush_request)
 335                 return;
 336
 337         set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
 338         up_read(&nilfs->ns_segctor_sem);
 339
 340         down_write(&nilfs->ns_segctor_sem);
 341         if (sci->sc_flush_request &&
 342             test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
 343                 struct nilfs_transaction_info *ti = current->journal_info;
 344
 345                 ti->ti_flags |= NILFS_TI_WRITER;
 346                 nilfs_segctor_do_immediate_flush(sci);
 347                 ti->ti_flags &= ~NILFS_TI_WRITER;
 348         }
 349         downgrade_write(&nilfs->ns_segctor_sem);
 350 }
 351
 352 static void nilfs_transaction_lock(struct super_block *sb,
 353                                    struct nilfs_transaction_info *ti,
 354                                    int gcflag)
 355 {
 356         struct nilfs_transaction_info *cur_ti = current->journal_info;
 357         struct the_nilfs *nilfs = sb->s_fs_info;
 358         struct nilfs_sc_info *sci = nilfs->ns_writer;
 359
 360         WARN_ON(cur_ti);
 361         ti->ti_flags = NILFS_TI_WRITER;
 362         ti->ti_count = 0;
 363         ti->ti_save = cur_ti;
 364         ti->ti_magic = NILFS_TI_MAGIC;
 365         current->journal_info = ti;
 366
 367         for (;;) {
 368                 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 369                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_TRYLOCK);
 370
 371                 down_write(&nilfs->ns_segctor_sem);
 372                 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
 373                         break;
 374
 375                 nilfs_segctor_do_immediate_flush(sci);
 376
 377                 up_write(&nilfs->ns_segctor_sem);
 378                 cond_resched();
 379         }
 380         if (gcflag)
 381                 ti->ti_flags |= NILFS_TI_GC;
 382
 383         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 384                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_LOCK);
 385 }
 386
 387 static void nilfs_transaction_unlock(struct super_block *sb)
 388 {
 389         struct nilfs_transaction_info *ti = current->journal_info;
 390         struct the_nilfs *nilfs = sb->s_fs_info;
 391
 392         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 393         BUG_ON(ti->ti_count > 0);
 394
 395         up_write(&nilfs->ns_segctor_sem);
 396         current->journal_info = ti->ti_save;
 397
 398         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 399                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_UNLOCK);
 400 }
 401
 402 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
 403                                             struct nilfs_segsum_pointer *ssp,
 404                                             unsigned int bytes)
 405 {
 406         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 407         unsigned int blocksize = sci->sc_super->s_blocksize;
 408         void *p;
 409
 410         if (unlikely(ssp->offset + bytes > blocksize)) {
 411                 ssp->offset = 0;
 412                 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
 413                                                &segbuf->sb_segsum_buffers));
 414                 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
 415         }
 416         p = ssp->bh->b_data + ssp->offset;
 417         ssp->offset += bytes;
 418         return p;
 419 }
 420
 421 /**
 422  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
 423  * @sci: nilfs_sc_info
 424  */
 425 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
 426 {
 427         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 428         struct buffer_head *sumbh;
 429         unsigned int sumbytes;
 430         unsigned int flags = 0;
 431         int err;
 432
 433         if (nilfs_doing_gc())
 434                 flags = NILFS_SS_GC;
 435         err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
 436         if (unlikely(err))
 437                 return err;
 438
 439         sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
 440         sumbytes = segbuf->sb_sum.sumbytes;
 441         sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
 442         sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
 443         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 444         return 0;
 445 }
 446
 447 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
 448 {
 449         sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
 450         if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
 451                 return -E2BIG; /*
 452                                 * The current segment is filled up
 453                                 * (internal code)
 454                                 */
 455         sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
 456         return nilfs_segctor_reset_segment_buffer(sci);
 457 }
 458
 459 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
 460 {
 461         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 462         int err;
 463
 464         if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
 465                 err = nilfs_segctor_feed_segment(sci);
 466                 if (err)
 467                         return err;
 468                 segbuf = sci->sc_curseg;
 469         }
 470         err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
 471         if (likely(!err))
 472                 segbuf->sb_sum.flags |= NILFS_SS_SR;
 473         return err;
 474 }
 475
 476 /*
 477  * Functions for making segment summary and payloads
 478  */
 479 static int nilfs_segctor_segsum_block_required(
 480         struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
 481         unsigned int binfo_size)
 482 {
 483         unsigned int blocksize = sci->sc_super->s_blocksize;
 484         /* Size of finfo and binfo is enough small against blocksize */
 485
 486         return ssp->offset + binfo_size +
 487                 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
 488                 blocksize;
 489 }
 490
 491 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
 492                                       struct inode *inode)
 493 {
 494         sci->sc_curseg->sb_sum.nfinfo++;
 495         sci->sc_binfo_ptr = sci->sc_finfo_ptr;
 496         nilfs_segctor_map_segsum_entry(
 497                 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
 498
 499         if (NILFS_I(inode)->i_root &&
 500             !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 501                 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
 502         /* skip finfo */
 503 }
 504
 505 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
 506                                     struct inode *inode)
 507 {
 508         struct nilfs_finfo *finfo;
 509         struct nilfs_inode_info *ii;
 510         struct nilfs_segment_buffer *segbuf;
 511         __u64 cno;
 512
 513         if (sci->sc_blk_cnt == 0)
 514                 return;
 515
 516         ii = NILFS_I(inode);
 517
 518         if (test_bit(NILFS_I_GCINODE, &ii->i_state))
 519                 cno = ii->i_cno;
 520         else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
 521                 cno = 0;
 522         else
 523                 cno = sci->sc_cno;
 524
 525         finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
 526                                                  sizeof(*finfo));
 527         finfo->fi_ino = cpu_to_le64(inode->i_ino);
 528         finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
 529         finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
 530         finfo->fi_cno = cpu_to_le64(cno);
 531
 532         segbuf = sci->sc_curseg;
 533         segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
 534                 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
 535         sci->sc_finfo_ptr = sci->sc_binfo_ptr;
 536         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 537 }
 538
 539 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
 540                                         struct buffer_head *bh,
 541                                         struct inode *inode,
 542                                         unsigned int binfo_size)
 543 {
 544         struct nilfs_segment_buffer *segbuf;
 545         int required, err = 0;
 546
 547  retry:
 548         segbuf = sci->sc_curseg;
 549         required = nilfs_segctor_segsum_block_required(
 550                 sci, &sci->sc_binfo_ptr, binfo_size);
 551         if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
 552                 nilfs_segctor_end_finfo(sci, inode);
 553                 err = nilfs_segctor_feed_segment(sci);
 554                 if (err)
 555                         return err;
 556                 goto retry;
 557         }
 558         if (unlikely(required)) {
 559                 err = nilfs_segbuf_extend_segsum(segbuf);
 560                 if (unlikely(err))
 561                         goto failed;
 562         }
 563         if (sci->sc_blk_cnt == 0)
 564                 nilfs_segctor_begin_finfo(sci, inode);
 565
 566         nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
 567         /* Substitution to vblocknr is delayed until update_blocknr() */
 568         nilfs_segbuf_add_file_buffer(segbuf, bh);
 569         sci->sc_blk_cnt++;
 570  failed:
 571         return err;
 572 }
 573
 574 /*
 575  * Callback functions that enumerate, mark, and collect dirty blocks
 576  */
 577 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
 578                                    struct buffer_head *bh, struct inode *inode)
 579 {
 580         int err;
 581
 582         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 583         if (err < 0)
 584                 return err;
 585
 586         err = nilfs_segctor_add_file_block(sci, bh, inode,
 587                                            sizeof(struct nilfs_binfo_v));
 588         if (!err)
 589                 sci->sc_datablk_cnt++;
 590         return err;
 591 }
 592
 593 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
 594                                    struct buffer_head *bh,
 595                                    struct inode *inode)
 596 {
 597         return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 598 }
 599
 600 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
 601                                    struct buffer_head *bh,
 602                                    struct inode *inode)
 603 {
 604         WARN_ON(!buffer_dirty(bh));
 605         return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 606 }
 607
 608 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
 609                                         struct nilfs_segsum_pointer *ssp,
 610                                         union nilfs_binfo *binfo)
 611 {
 612         struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
 613                 sci, ssp, sizeof(*binfo_v));
 614         *binfo_v = binfo->bi_v;
 615 }
 616
 617 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
 618                                         struct nilfs_segsum_pointer *ssp,
 619                                         union nilfs_binfo *binfo)
 620 {
 621         __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
 622                 sci, ssp, sizeof(*vblocknr));
 623         *vblocknr = binfo->bi_v.bi_vblocknr;
 624 }
 625
 626 static const struct nilfs_sc_operations nilfs_sc_file_ops = {
 627         .collect_data = nilfs_collect_file_data,
 628         .collect_node = nilfs_collect_file_node,
 629         .collect_bmap = nilfs_collect_file_bmap,
 630         .write_data_binfo = nilfs_write_file_data_binfo,
 631         .write_node_binfo = nilfs_write_file_node_binfo,
 632 };
 633
 634 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
 635                                   struct buffer_head *bh, struct inode *inode)
 636 {
 637         int err;
 638
 639         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 640         if (err < 0)
 641                 return err;
 642
 643         err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 644         if (!err)
 645                 sci->sc_datablk_cnt++;
 646         return err;
 647 }
 648
 649 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
 650                                   struct buffer_head *bh, struct inode *inode)
 651 {
 652         WARN_ON(!buffer_dirty(bh));
 653         return nilfs_segctor_add_file_block(sci, bh, inode,
 654                                             sizeof(struct nilfs_binfo_dat));
 655 }
 656
 657 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
 658                                        struct nilfs_segsum_pointer *ssp,
 659                                        union nilfs_binfo *binfo)
 660 {
 661         __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
 662                                                           sizeof(*blkoff));
 663         *blkoff = binfo->bi_dat.bi_blkoff;
 664 }
 665
 666 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
 667                                        struct nilfs_segsum_pointer *ssp,
 668                                        union nilfs_binfo *binfo)
 669 {
 670         struct nilfs_binfo_dat *binfo_dat =
 671                 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
 672         *binfo_dat = binfo->bi_dat;
 673 }
 674
 675 static const struct nilfs_sc_operations nilfs_sc_dat_ops = {
 676         .collect_data = nilfs_collect_dat_data,
 677         .collect_node = nilfs_collect_file_node,
 678         .collect_bmap = nilfs_collect_dat_bmap,
 679         .write_data_binfo = nilfs_write_dat_data_binfo,
 680         .write_node_binfo = nilfs_write_dat_node_binfo,
 681 };
 682
 683 static const struct nilfs_sc_operations nilfs_sc_dsync_ops = {
 684         .collect_data = nilfs_collect_file_data,
 685         .collect_node = NULL,
 686         .collect_bmap = NULL,
 687         .write_data_binfo = nilfs_write_file_data_binfo,
 688         .write_node_binfo = NULL,
 689 };
 690
 691 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
 692                                               struct list_head *listp,
 693                                               size_t nlimit,
 694                                               loff_t start, loff_t end)
 695 {
 696         struct address_space *mapping = inode->i_mapping;
 697         struct pagevec pvec;
 698         pgoff_t index = 0, last = ULONG_MAX;
 699         size_t ndirties = 0;
 700         int i;
 701
 702         if (unlikely(start != 0 || end != LLONG_MAX)) {
 703                 /*
 704                  * A valid range is given for sync-ing data pages. The
 705                  * range is rounded to per-page; extra dirty buffers
 706                  * may be included if blocksize < pagesize.
 707                  */
 708                 index = start >> PAGE_SHIFT;
 709                 last = end >> PAGE_SHIFT;
 710         }
 711         pagevec_init(&pvec, 0);
 712  repeat:
 713         if (unlikely(index > last) ||
 714             !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 715                                 min_t(pgoff_t, last - index,
 716                                       PAGEVEC_SIZE - 1) + 1))
 717                 return ndirties;
 718
 719         for (i = 0; i < pagevec_count(&pvec); i++) {
 720                 struct buffer_head *bh, *head;
 721                 struct page *page = pvec.pages[i];
 722
 723                 if (unlikely(page->index > last))
 724                         break;
 725
 726                 lock_page(page);
 727                 if (!page_has_buffers(page))
 728                         create_empty_buffers(page, i_blocksize(inode), 0);
 729                 unlock_page(page);
 730
 731                 bh = head = page_buffers(page);
 732                 do {
 733                         if (!buffer_dirty(bh) || buffer_async_write(bh))
 734                                 continue;
 735                         get_bh(bh);
 736                         list_add_tail(&bh->b_assoc_buffers, listp);
 737                         ndirties++;
 738                         if (unlikely(ndirties >= nlimit)) {
 739                                 pagevec_release(&pvec);
 740                                 cond_resched();
 741                                 return ndirties;
 742                         }
 743                 } while (bh = bh->b_this_page, bh != head);
 744         }
 745         pagevec_release(&pvec);
 746         cond_resched();
 747         goto repeat;
 748 }
 749
 750 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
 751                                             struct list_head *listp)
 752 {
 753         struct nilfs_inode_info *ii = NILFS_I(inode);
 754         struct address_space *mapping = &ii->i_btnode_cache;
 755         struct pagevec pvec;
 756         struct buffer_head *bh, *head;
 757         unsigned int i;
 758         pgoff_t index = 0;
 759
 760         pagevec_init(&pvec, 0);
 761
 762         while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 763                                   PAGEVEC_SIZE)) {
 764                 for (i = 0; i < pagevec_count(&pvec); i++) {
 765                         bh = head = page_buffers(pvec.pages[i]);
 766                         do {
 767                                 if (buffer_dirty(bh) &&
 768                                                 !buffer_async_write(bh)) {
 769                                         get_bh(bh);
 770                                         list_add_tail(&bh->b_assoc_buffers,
 771                                                       listp);
 772                                 }
 773                                 bh = bh->b_this_page;
 774                         } while (bh != head);
 775                 }
 776                 pagevec_release(&pvec);
 777                 cond_resched();
 778         }
 779 }
 780
 781 static void nilfs_dispose_list(struct the_nilfs *nilfs,
 782                                struct list_head *head, int force)
 783 {
 784         struct nilfs_inode_info *ii, *n;
 785         struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
 786         unsigned int nv = 0;
 787
 788         while (!list_empty(head)) {
 789                 spin_lock(&nilfs->ns_inode_lock);
 790                 list_for_each_entry_safe(ii, n, head, i_dirty) {
 791                         list_del_init(&ii->i_dirty);
 792                         if (force) {
 793                                 if (unlikely(ii->i_bh)) {
 794                                         brelse(ii->i_bh);
 795                                         ii->i_bh = NULL;
 796                                 }
 797                         } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
 798                                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 799                                 list_add_tail(&ii->i_dirty,
 800                                               &nilfs->ns_dirty_files);
 801                                 continue;
 802                         }
 803                         ivec[nv++] = ii;
 804                         if (nv == SC_N_INODEVEC)
 805                                 break;
 806                 }
 807                 spin_unlock(&nilfs->ns_inode_lock);
 808
 809                 for (pii = ivec; nv > 0; pii++, nv--)
 810                         iput(&(*pii)->vfs_inode);
 811         }
 812 }
 813
 814 static void nilfs_iput_work_func(struct work_struct *work)
 815 {
 816         struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
 817                                                  sc_iput_work);
 818         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 819
 820         nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
 821 }
 822
 823 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
 824                                      struct nilfs_root *root)
 825 {
 826         int ret = 0;
 827
 828         if (nilfs_mdt_fetch_dirty(root->ifile))
 829                 ret++;
 830         if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
 831                 ret++;
 832         if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
 833                 ret++;
 834         if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
 835                 ret++;
 836         return ret;
 837 }
 838
 839 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
 840 {
 841         return list_empty(&sci->sc_dirty_files) &&
 842                 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
 843                 sci->sc_nfreesegs == 0 &&
 844                 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
 845 }
 846
 847 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
 848 {
 849         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 850         int ret = 0;
 851
 852         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
 853                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
 854
 855         spin_lock(&nilfs->ns_inode_lock);
 856         if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
 857                 ret++;
 858
 859         spin_unlock(&nilfs->ns_inode_lock);
 860         return ret;
 861 }
 862
 863 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
 864 {
 865         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 866
 867         nilfs_mdt_clear_dirty(sci->sc_root->ifile);
 868         nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
 869         nilfs_mdt_clear_dirty(nilfs->ns_sufile);
 870         nilfs_mdt_clear_dirty(nilfs->ns_dat);
 871 }
 872
 873 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
 874 {
 875         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 876         struct buffer_head *bh_cp;
 877         struct nilfs_checkpoint *raw_cp;
 878         int err;
 879
 880         /* XXX: this interface will be changed */
 881         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
 882                                           &raw_cp, &bh_cp);
 883         if (likely(!err)) {
 884                 /*
 885                  * The following code is duplicated with cpfile.  But, it is
 886                  * needed to collect the checkpoint even if it was not newly
 887                  * created.
 888                  */
 889                 mark_buffer_dirty(bh_cp);
 890                 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
 891                 nilfs_cpfile_put_checkpoint(
 892                         nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 893         } else
 894                 WARN_ON(err == -EINVAL || err == -ENOENT);
 895
 896         return err;
 897 }
 898
 899 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
 900 {
 901         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 902         struct buffer_head *bh_cp;
 903         struct nilfs_checkpoint *raw_cp;
 904         int err;
 905
 906         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
 907                                           &raw_cp, &bh_cp);
 908         if (unlikely(err)) {
 909                 WARN_ON(err == -EINVAL || err == -ENOENT);
 910                 goto failed_ibh;
 911         }
 912         raw_cp->cp_snapshot_list.ssl_next = 0;
 913         raw_cp->cp_snapshot_list.ssl_prev = 0;
 914         raw_cp->cp_inodes_count =
 915                 cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
 916         raw_cp->cp_blocks_count =
 917                 cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
 918         raw_cp->cp_nblk_inc =
 919                 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
 920         raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
 921         raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
 922
 923         if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 924                 nilfs_checkpoint_clear_minor(raw_cp);
 925         else
 926                 nilfs_checkpoint_set_minor(raw_cp);
 927
 928         nilfs_write_inode_common(sci->sc_root->ifile,
 929                                  &raw_cp->cp_ifile_inode, 1);
 930         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 931         return 0;
 932
 933  failed_ibh:
 934         return err;
 935 }
 936
 937 static void nilfs_fill_in_file_bmap(struct inode *ifile,
 938                                     struct nilfs_inode_info *ii)
 939
 940 {
 941         struct buffer_head *ibh;
 942         struct nilfs_inode *raw_inode;
 943
 944         if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
 945                 ibh = ii->i_bh;
 946                 BUG_ON(!ibh);
 947                 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
 948                                                   ibh);
 949                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 950                 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
 951         }
 952 }
 953
 954 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
 955 {
 956         struct nilfs_inode_info *ii;
 957
 958         list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
 959                 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
 960                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
 961         }
 962 }
 963
 964 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
 965                                              struct the_nilfs *nilfs)
 966 {
 967         struct buffer_head *bh_sr;
 968         struct nilfs_super_root *raw_sr;
 969         unsigned int isz, srsz;
 970
 971         bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
 972         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
 973         isz = nilfs->ns_inode_size;
 974         srsz = NILFS_SR_BYTES(isz);
 975
 976         raw_sr->sr_bytes = cpu_to_le16(srsz);
 977         raw_sr->sr_nongc_ctime
 978                 = cpu_to_le64(nilfs_doing_gc() ?
 979                               nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
 980         raw_sr->sr_flags = 0;
 981
 982         nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
 983                                  NILFS_SR_DAT_OFFSET(isz), 1);
 984         nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
 985                                  NILFS_SR_CPFILE_OFFSET(isz), 1);
 986         nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
 987                                  NILFS_SR_SUFILE_OFFSET(isz), 1);
 988         memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
 989 }
 990
 991 static void nilfs_redirty_inodes(struct list_head *head)
 992 {
 993         struct nilfs_inode_info *ii;
 994
 995         list_for_each_entry(ii, head, i_dirty) {
 996                 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
 997                         clear_bit(NILFS_I_COLLECTED, &ii->i_state);
 998         }
 999 }
1000
1001 static void nilfs_drop_collected_inodes(struct list_head *head)
1002 {
1003         struct nilfs_inode_info *ii;
1004
1005         list_for_each_entry(ii, head, i_dirty) {
1006                 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1007                         continue;
1008
1009                 clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
1010                 set_bit(NILFS_I_UPDATED, &ii->i_state);
1011         }
1012 }
1013
1014 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1015                                        struct inode *inode,
1016                                        struct list_head *listp,
1017                                        int (*collect)(struct nilfs_sc_info *,
1018                                                       struct buffer_head *,
1019                                                       struct inode *))
1020 {
1021         struct buffer_head *bh, *n;
1022         int err = 0;
1023
1024         if (collect) {
1025                 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1026                         list_del_init(&bh->b_assoc_buffers);
1027                         err = collect(sci, bh, inode);
1028                         brelse(bh);
1029                         if (unlikely(err))
1030                                 goto dispose_buffers;
1031                 }
1032                 return 0;
1033         }
1034
1035  dispose_buffers:
1036         while (!list_empty(listp)) {
1037                 bh = list_first_entry(listp, struct buffer_head,
1038                                       b_assoc_buffers);
1039                 list_del_init(&bh->b_assoc_buffers);
1040                 brelse(bh);
1041         }
1042         return err;
1043 }
1044
1045 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1046 {
1047         /* Remaining number of blocks within segment buffer */
1048         return sci->sc_segbuf_nblocks -
1049                 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1050 }
1051
1052 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1053                                    struct inode *inode,
1054                                    const struct nilfs_sc_operations *sc_ops)
1055 {
1056         LIST_HEAD(data_buffers);
1057         LIST_HEAD(node_buffers);
1058         int err;
1059
1060         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1061                 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1062
1063                 n = nilfs_lookup_dirty_data_buffers(
1064                         inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1065                 if (n > rest) {
1066                         err = nilfs_segctor_apply_buffers(
1067                                 sci, inode, &data_buffers,
1068                                 sc_ops->collect_data);
1069                         BUG_ON(!err); /* always receive -E2BIG or true error */
1070                         goto break_or_fail;
1071                 }
1072         }
1073         nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1074
1075         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1076                 err = nilfs_segctor_apply_buffers(
1077                         sci, inode, &data_buffers, sc_ops->collect_data);
1078                 if (unlikely(err)) {
1079                         /* dispose node list */
1080                         nilfs_segctor_apply_buffers(
1081                                 sci, inode, &node_buffers, NULL);
1082                         goto break_or_fail;
1083                 }
1084                 sci->sc_stage.flags |= NILFS_CF_NODE;
1085         }
1086         /* Collect node */
1087         err = nilfs_segctor_apply_buffers(
1088                 sci, inode, &node_buffers, sc_ops->collect_node);
1089         if (unlikely(err))
1090                 goto break_or_fail;
1091
1092         nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1093         err = nilfs_segctor_apply_buffers(
1094                 sci, inode, &node_buffers, sc_ops->collect_bmap);
1095         if (unlikely(err))
1096                 goto break_or_fail;
1097
1098         nilfs_segctor_end_finfo(sci, inode);
1099         sci->sc_stage.flags &= ~NILFS_CF_NODE;
1100
1101  break_or_fail:
1102         return err;
1103 }
1104
1105 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1106                                          struct inode *inode)
1107 {
1108         LIST_HEAD(data_buffers);
1109         size_t n, rest = nilfs_segctor_buffer_rest(sci);
1110         int err;
1111
1112         n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1113                                             sci->sc_dsync_start,
1114                                             sci->sc_dsync_end);
1115
1116         err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1117                                           nilfs_collect_file_data);
1118         if (!err) {
1119                 nilfs_segctor_end_finfo(sci, inode);
1120                 BUG_ON(n > rest);
1121                 /* always receive -E2BIG or true error if n > rest */
1122         }
1123         return err;
1124 }
1125
1126 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1127 {
1128         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1129         struct list_head *head;
1130         struct nilfs_inode_info *ii;
1131         size_t ndone;
1132         int err = 0;
1133
1134         switch (nilfs_sc_cstage_get(sci)) {
1135         case NILFS_ST_INIT:
1136                 /* Pre-processes */
1137                 sci->sc_stage.flags = 0;
1138
1139                 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1140                         sci->sc_nblk_inc = 0;
1141                         sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1142                         if (mode == SC_LSEG_DSYNC) {
1143                                 nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1144                                 goto dsync_mode;
1145                         }
1146                 }
1147
1148                 sci->sc_stage.dirty_file_ptr = NULL;
1149                 sci->sc_stage.gc_inode_ptr = NULL;
1150                 if (mode == SC_FLUSH_DAT) {
1151                         nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1152                         goto dat_stage;
1153                 }
1154                 nilfs_sc_cstage_inc(sci);  /* Fall through */
1155         case NILFS_ST_GC:
1156                 if (nilfs_doing_gc()) {
1157                         head = &sci->sc_gc_inodes;
1158                         ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1159                                                 head, i_dirty);
1160                         list_for_each_entry_continue(ii, head, i_dirty) {
1161                                 err = nilfs_segctor_scan_file(
1162                                         sci, &ii->vfs_inode,
1163                                         &nilfs_sc_file_ops);
1164                                 if (unlikely(err)) {
1165                                         sci->sc_stage.gc_inode_ptr = list_entry(
1166                                                 ii->i_dirty.prev,
1167                                                 struct nilfs_inode_info,
1168                                                 i_dirty);
1169                                         goto break_or_fail;
1170                                 }
1171                                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1172                         }
1173                         sci->sc_stage.gc_inode_ptr = NULL;
1174                 }
1175                 nilfs_sc_cstage_inc(sci);  /* Fall through */
1176         case NILFS_ST_FILE:
1177                 head = &sci->sc_dirty_files;
1178                 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1179                                         i_dirty);
1180                 list_for_each_entry_continue(ii, head, i_dirty) {
1181                         clear_bit(NILFS_I_DIRTY, &ii->i_state);
1182
1183                         err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1184                                                       &nilfs_sc_file_ops);
1185                         if (unlikely(err)) {
1186                                 sci->sc_stage.dirty_file_ptr =
1187                                         list_entry(ii->i_dirty.prev,
1188                                                    struct nilfs_inode_info,
1189                                                    i_dirty);
1190                                 goto break_or_fail;
1191                         }
1192                         /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1193                         /* XXX: required ? */
1194                 }
1195                 sci->sc_stage.dirty_file_ptr = NULL;
1196                 if (mode == SC_FLUSH_FILE) {
1197                         nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1198                         return 0;
1199                 }
1200                 nilfs_sc_cstage_inc(sci);
1201                 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1202                 /* Fall through */
1203         case NILFS_ST_IFILE:
1204                 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1205                                               &nilfs_sc_file_ops);
1206                 if (unlikely(err))
1207                         break;
1208                 nilfs_sc_cstage_inc(sci);
1209                 /* Creating a checkpoint */
1210                 err = nilfs_segctor_create_checkpoint(sci);
1211                 if (unlikely(err))
1212                         break;
1213                 /* Fall through */
1214         case NILFS_ST_CPFILE:
1215                 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1216                                               &nilfs_sc_file_ops);
1217                 if (unlikely(err))
1218                         break;
1219                 nilfs_sc_cstage_inc(sci);  /* Fall through */
1220         case NILFS_ST_SUFILE:
1221                 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1222                                          sci->sc_nfreesegs, &ndone);
1223                 if (unlikely(err)) {
1224                         nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1225                                                   sci->sc_freesegs, ndone,
1226                                                   NULL);
1227                         break;
1228                 }
1229                 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1230
1231                 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1232                                               &nilfs_sc_file_ops);
1233                 if (unlikely(err))
1234                         break;
1235                 nilfs_sc_cstage_inc(sci);  /* Fall through */
1236         case NILFS_ST_DAT:
1237  dat_stage:
1238                 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1239                                               &nilfs_sc_dat_ops);
1240                 if (unlikely(err))
1241                         break;
1242                 if (mode == SC_FLUSH_DAT) {
1243                         nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1244                         return 0;
1245                 }
1246                 nilfs_sc_cstage_inc(sci);  /* Fall through */
1247         case NILFS_ST_SR:
1248                 if (mode == SC_LSEG_SR) {
1249                         /* Appending a super root */
1250                         err = nilfs_segctor_add_super_root(sci);
1251                         if (unlikely(err))
1252                                 break;
1253                 }
1254                 /* End of a logical segment */
1255                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1256                 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1257                 return 0;
1258         case NILFS_ST_DSYNC:
1259  dsync_mode:
1260                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1261                 ii = sci->sc_dsync_inode;
1262                 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1263                         break;
1264
1265                 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1266                 if (unlikely(err))
1267                         break;
1268                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1269                 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1270                 return 0;
1271         case NILFS_ST_DONE:
1272                 return 0;
1273         default:
1274                 BUG();
1275         }
1276
1277  break_or_fail:
1278         return err;
1279 }
1280
1281 /**
1282  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1283  * @sci: nilfs_sc_info
1284  * @nilfs: nilfs object
1285  */
1286 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1287                                             struct the_nilfs *nilfs)
1288 {
1289         struct nilfs_segment_buffer *segbuf, *prev;
1290         __u64 nextnum;
1291         int err, alloc = 0;
1292
1293         segbuf = nilfs_segbuf_new(sci->sc_super);
1294         if (unlikely(!segbuf))
1295                 return -ENOMEM;
1296
1297         if (list_empty(&sci->sc_write_logs)) {
1298                 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1299                                  nilfs->ns_pseg_offset, nilfs);
1300                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1301                         nilfs_shift_to_next_segment(nilfs);
1302                         nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1303                 }
1304
1305                 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1306                 nextnum = nilfs->ns_nextnum;
1307
1308                 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1309                         /* Start from the head of a new full segment */
1310                         alloc++;
1311         } else {
1312                 /* Continue logs */
1313                 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1314                 nilfs_segbuf_map_cont(segbuf, prev);
1315                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1316                 nextnum = prev->sb_nextnum;
1317
1318                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1319                         nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1320                         segbuf->sb_sum.seg_seq++;
1321                         alloc++;
1322                 }
1323         }
1324
1325         err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1326         if (err)
1327                 goto failed;
1328
1329         if (alloc) {
1330                 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1331                 if (err)
1332                         goto failed;
1333         }
1334         nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1335
1336         BUG_ON(!list_empty(&sci->sc_segbufs));
1337         list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1338         sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1339         return 0;
1340
1341  failed:
1342         nilfs_segbuf_free(segbuf);
1343         return err;
1344 }
1345
1346 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1347                                          struct the_nilfs *nilfs, int nadd)
1348 {
1349         struct nilfs_segment_buffer *segbuf, *prev;
1350         struct inode *sufile = nilfs->ns_sufile;
1351         __u64 nextnextnum;
1352         LIST_HEAD(list);
1353         int err, ret, i;
1354
1355         prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1356         /*
1357          * Since the segment specified with nextnum might be allocated during
1358          * the previous construction, the buffer including its segusage may
1359          * not be dirty.  The following call ensures that the buffer is dirty
1360          * and will pin the buffer on memory until the sufile is written.
1361          */
1362         err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1363         if (unlikely(err))
1364                 return err;
1365
1366         for (i = 0; i < nadd; i++) {
1367                 /* extend segment info */
1368                 err = -ENOMEM;
1369                 segbuf = nilfs_segbuf_new(sci->sc_super);
1370                 if (unlikely(!segbuf))
1371                         goto failed;
1372
1373                 /* map this buffer to region of segment on-disk */
1374                 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1375                 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1376
1377                 /* allocate the next next full segment */
1378                 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1379                 if (unlikely(err))
1380                         goto failed_segbuf;
1381
1382                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1383                 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1384
1385                 list_add_tail(&segbuf->sb_list, &list);
1386                 prev = segbuf;
1387         }
1388         list_splice_tail(&list, &sci->sc_segbufs);
1389         return 0;
1390
1391  failed_segbuf:
1392         nilfs_segbuf_free(segbuf);
1393  failed:
1394         list_for_each_entry(segbuf, &list, sb_list) {
1395                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1396                 WARN_ON(ret); /* never fails */
1397         }
1398         nilfs_destroy_logs(&list);
1399         return err;
1400 }
1401
1402 static void nilfs_free_incomplete_logs(struct list_head *logs,
1403                                        struct the_nilfs *nilfs)
1404 {
1405         struct nilfs_segment_buffer *segbuf, *prev;
1406         struct inode *sufile = nilfs->ns_sufile;
1407         int ret;
1408
1409         segbuf = NILFS_FIRST_SEGBUF(logs);
1410         if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1411                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1412                 WARN_ON(ret); /* never fails */
1413         }
1414         if (atomic_read(&segbuf->sb_err)) {
1415                 /* Case 1: The first segment failed */
1416                 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1417                         /*
1418                          * Case 1a:  Partial segment appended into an existing
1419                          * segment
1420                          */
1421                         nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1422                                                 segbuf->sb_fseg_end);
1423                 else /* Case 1b:  New full segment */
1424                         set_nilfs_discontinued(nilfs);
1425         }
1426
1427         prev = segbuf;
1428         list_for_each_entry_continue(segbuf, logs, sb_list) {
1429                 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1430                         ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1431                         WARN_ON(ret); /* never fails */
1432                 }
1433                 if (atomic_read(&segbuf->sb_err) &&
1434                     segbuf->sb_segnum != nilfs->ns_nextnum)
1435                         /* Case 2: extended segment (!= next) failed */
1436                         nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1437                 prev = segbuf;
1438         }
1439 }
1440
1441 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1442                                           struct inode *sufile)
1443 {
1444         struct nilfs_segment_buffer *segbuf;
1445         unsigned long live_blocks;
1446         int ret;
1447
1448         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1449                 live_blocks = segbuf->sb_sum.nblocks +
1450                         (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1451                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1452                                                      live_blocks,
1453                                                      sci->sc_seg_ctime);
1454                 WARN_ON(ret); /* always succeed because the segusage is dirty */
1455         }
1456 }
1457
1458 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1459 {
1460         struct nilfs_segment_buffer *segbuf;
1461         int ret;
1462
1463         segbuf = NILFS_FIRST_SEGBUF(logs);
1464         ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1465                                              segbuf->sb_pseg_start -
1466                                              segbuf->sb_fseg_start, 0);
1467         WARN_ON(ret); /* always succeed because the segusage is dirty */
1468
1469         list_for_each_entry_continue(segbuf, logs, sb_list) {
1470                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1471                                                      0, 0);
1472                 WARN_ON(ret); /* always succeed */
1473         }
1474 }
1475
1476 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1477                                             struct nilfs_segment_buffer *last,
1478                                             struct inode *sufile)
1479 {
1480         struct nilfs_segment_buffer *segbuf = last;
1481         int ret;
1482
1483         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1484                 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1485                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1486                 WARN_ON(ret);
1487         }
1488         nilfs_truncate_logs(&sci->sc_segbufs, last);
1489 }
1490
1491
1492 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1493                                  struct the_nilfs *nilfs, int mode)
1494 {
1495         struct nilfs_cstage prev_stage = sci->sc_stage;
1496         int err, nadd = 1;
1497
1498         /* Collection retry loop */
1499         for (;;) {
1500                 sci->sc_nblk_this_inc = 0;
1501                 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1502
1503                 err = nilfs_segctor_reset_segment_buffer(sci);
1504                 if (unlikely(err))
1505                         goto failed;
1506
1507                 err = nilfs_segctor_collect_blocks(sci, mode);
1508                 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1509                 if (!err)
1510                         break;
1511
1512                 if (unlikely(err != -E2BIG))
1513                         goto failed;
1514
1515                 /* The current segment is filled up */
1516                 if (mode != SC_LSEG_SR ||
1517                     nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1518                         break;
1519
1520                 nilfs_clear_logs(&sci->sc_segbufs);
1521
1522                 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1523                         err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1524                                                         sci->sc_freesegs,
1525                                                         sci->sc_nfreesegs,
1526                                                         NULL);
1527                         WARN_ON(err); /* do not happen */
1528                         sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1529                 }
1530
1531                 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1532                 if (unlikely(err))
1533                         return err;
1534
1535                 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1536                 sci->sc_stage = prev_stage;
1537         }
1538         nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1539         return 0;
1540
1541  failed:
1542         return err;
1543 }
1544
1545 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1546                                       struct buffer_head *new_bh)
1547 {
1548         BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1549
1550         list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1551         /* The caller must release old_bh */
1552 }
1553
1554 static int
1555 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1556                                      struct nilfs_segment_buffer *segbuf,
1557                                      int mode)
1558 {
1559         struct inode *inode = NULL;
1560         sector_t blocknr;
1561         unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1562         unsigned long nblocks = 0, ndatablk = 0;
1563         const struct nilfs_sc_operations *sc_op = NULL;
1564         struct nilfs_segsum_pointer ssp;
1565         struct nilfs_finfo *finfo = NULL;
1566         union nilfs_binfo binfo;
1567         struct buffer_head *bh, *bh_org;
1568         ino_t ino = 0;
1569         int err = 0;
1570
1571         if (!nfinfo)
1572                 goto out;
1573
1574         blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1575         ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1576         ssp.offset = sizeof(struct nilfs_segment_summary);
1577
1578         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1579                 if (bh == segbuf->sb_super_root)
1580                         break;
1581                 if (!finfo) {
1582                         finfo = nilfs_segctor_map_segsum_entry(
1583                                 sci, &ssp, sizeof(*finfo));
1584                         ino = le64_to_cpu(finfo->fi_ino);
1585                         nblocks = le32_to_cpu(finfo->fi_nblocks);
1586                         ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1587
1588                         inode = bh->b_page->mapping->host;
1589
1590                         if (mode == SC_LSEG_DSYNC)
1591                                 sc_op = &nilfs_sc_dsync_ops;
1592                         else if (ino == NILFS_DAT_INO)
1593                                 sc_op = &nilfs_sc_dat_ops;
1594                         else /* file blocks */
1595                                 sc_op = &nilfs_sc_file_ops;
1596                 }
1597                 bh_org = bh;
1598                 get_bh(bh_org);
1599                 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1600                                         &binfo);
1601                 if (bh != bh_org)
1602                         nilfs_list_replace_buffer(bh_org, bh);
1603                 brelse(bh_org);
1604                 if (unlikely(err))
1605                         goto failed_bmap;
1606
1607                 if (ndatablk > 0)
1608                         sc_op->write_data_binfo(sci, &ssp, &binfo);
1609                 else
1610                         sc_op->write_node_binfo(sci, &ssp, &binfo);
1611
1612                 blocknr++;
1613                 if (--nblocks == 0) {
1614                         finfo = NULL;
1615                         if (--nfinfo == 0)
1616                                 break;
1617                 } else if (ndatablk > 0)
1618                         ndatablk--;
1619         }
1620  out:
1621         return 0;
1622
1623  failed_bmap:
1624         return err;
1625 }
1626
1627 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1628 {
1629         struct nilfs_segment_buffer *segbuf;
1630         int err;
1631
1632         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1633                 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1634                 if (unlikely(err))
1635                         return err;
1636                 nilfs_segbuf_fill_in_segsum(segbuf);
1637         }
1638         return 0;
1639 }
1640
1641 static void nilfs_begin_page_io(struct page *page)
1642 {
1643         if (!page || PageWriteback(page))
1644                 /*
1645                  * For split b-tree node pages, this function may be called
1646                  * twice.  We ignore the 2nd or later calls by this check.
1647                  */
1648                 return;
1649
1650         lock_page(page);
1651         clear_page_dirty_for_io(page);
1652         set_page_writeback(page);
1653         unlock_page(page);
1654 }
1655
1656 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1657 {
1658         struct nilfs_segment_buffer *segbuf;
1659         struct page *bd_page = NULL, *fs_page = NULL;
1660
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                         set_buffer_async_write(bh);
1680                         if (bh == segbuf->sb_super_root) {
1681                                 if (bh->b_page != bd_page) {
1682                                         lock_page(bd_page);
1683                                         clear_page_dirty_for_io(bd_page);
1684                                         set_page_writeback(bd_page);
1685                                         unlock_page(bd_page);
1686                                         bd_page = bh->b_page;
1687                                 }
1688                                 break;
1689                         }
1690                         if (bh->b_page != fs_page) {
1691                                 nilfs_begin_page_io(fs_page);
1692                                 fs_page = bh->b_page;
1693                         }
1694                 }
1695         }
1696         if (bd_page) {
1697                 lock_page(bd_page);
1698                 clear_page_dirty_for_io(bd_page);
1699                 set_page_writeback(bd_page);
1700                 unlock_page(bd_page);
1701         }
1702         nilfs_begin_page_io(fs_page);
1703 }
1704
1705 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1706                                struct the_nilfs *nilfs)
1707 {
1708         int ret;
1709
1710         ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1711         list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1712         return ret;
1713 }
1714
1715 static void nilfs_end_page_io(struct page *page, int err)
1716 {
1717         if (!page)
1718                 return;
1719
1720         if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1721                 /*
1722                  * For b-tree node pages, this function may be called twice
1723                  * or more because they might be split in a segment.
1724                  */
1725                 if (PageDirty(page)) {
1726                         /*
1727                          * For pages holding split b-tree node buffers, dirty
1728                          * flag on the buffers may be cleared discretely.
1729                          * In that case, the page is once redirtied for
1730                          * remaining buffers, and it must be cancelled if
1731                          * all the buffers get cleaned later.
1732                          */
1733                         lock_page(page);
1734                         if (nilfs_page_buffers_clean(page))
1735                                 __nilfs_clear_page_dirty(page);
1736                         unlock_page(page);
1737                 }
1738                 return;
1739         }
1740
1741         if (!err) {
1742                 if (!nilfs_page_buffers_clean(page))
1743                         __set_page_dirty_nobuffers(page);
1744                 ClearPageError(page);
1745         } else {
1746                 __set_page_dirty_nobuffers(page);
1747                 SetPageError(page);
1748         }
1749
1750         end_page_writeback(page);
1751 }
1752
1753 static void nilfs_abort_logs(struct list_head *logs, int err)
1754 {
1755         struct nilfs_segment_buffer *segbuf;
1756         struct page *bd_page = NULL, *fs_page = NULL;
1757         struct buffer_head *bh;
1758
1759         if (list_empty(logs))
1760                 return;
1761
1762         list_for_each_entry(segbuf, logs, sb_list) {
1763                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1764                                     b_assoc_buffers) {
1765                         if (bh->b_page != bd_page) {
1766                                 if (bd_page)
1767                                         end_page_writeback(bd_page);
1768                                 bd_page = bh->b_page;
1769                         }
1770                 }
1771
1772                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1773                                     b_assoc_buffers) {
1774                         clear_buffer_async_write(bh);
1775                         if (bh == segbuf->sb_super_root) {
1776                                 if (bh->b_page != bd_page) {
1777                                         end_page_writeback(bd_page);
1778                                         bd_page = bh->b_page;
1779                                 }
1780                                 break;
1781                         }
1782                         if (bh->b_page != fs_page) {
1783                                 nilfs_end_page_io(fs_page, err);
1784                                 fs_page = bh->b_page;
1785                         }
1786                 }
1787         }
1788         if (bd_page)
1789                 end_page_writeback(bd_page);
1790
1791         nilfs_end_page_io(fs_page, err);
1792 }
1793
1794 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1795                                              struct the_nilfs *nilfs, int err)
1796 {
1797         LIST_HEAD(logs);
1798         int ret;
1799
1800         list_splice_tail_init(&sci->sc_write_logs, &logs);
1801         ret = nilfs_wait_on_logs(&logs);
1802         nilfs_abort_logs(&logs, ret ? : err);
1803
1804         list_splice_tail_init(&sci->sc_segbufs, &logs);
1805         nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1806         nilfs_free_incomplete_logs(&logs, nilfs);
1807
1808         if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1809                 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1810                                                 sci->sc_freesegs,
1811                                                 sci->sc_nfreesegs,
1812                                                 NULL);
1813                 WARN_ON(ret); /* do not happen */
1814         }
1815
1816         nilfs_destroy_logs(&logs);
1817 }
1818
1819 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1820                                    struct nilfs_segment_buffer *segbuf)
1821 {
1822         nilfs->ns_segnum = segbuf->sb_segnum;
1823         nilfs->ns_nextnum = segbuf->sb_nextnum;
1824         nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1825                 + segbuf->sb_sum.nblocks;
1826         nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1827         nilfs->ns_ctime = segbuf->sb_sum.ctime;
1828 }
1829
1830 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1831 {
1832         struct nilfs_segment_buffer *segbuf;
1833         struct page *bd_page = NULL, *fs_page = NULL;
1834         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1835         int update_sr = false;
1836
1837         list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1838                 struct buffer_head *bh;
1839
1840                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1841                                     b_assoc_buffers) {
1842                         set_buffer_uptodate(bh);
1843                         clear_buffer_dirty(bh);
1844                         if (bh->b_page != bd_page) {
1845                                 if (bd_page)
1846                                         end_page_writeback(bd_page);
1847                                 bd_page = bh->b_page;
1848                         }
1849                 }
1850                 /*
1851                  * We assume that the buffers which belong to the same page
1852                  * continue over the buffer list.
1853                  * Under this assumption, the last BHs of pages is
1854                  * identifiable by the discontinuity of bh->b_page
1855                  * (page != fs_page).
1856                  *
1857                  * For B-tree node blocks, however, this assumption is not
1858                  * guaranteed.  The cleanup code of B-tree node pages needs
1859                  * special care.
1860                  */
1861                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1862                                     b_assoc_buffers) {
1863                         const unsigned long set_bits = BIT(BH_Uptodate);
1864                         const unsigned long clear_bits =
1865                                 (BIT(BH_Dirty) | BIT(BH_Async_Write) |
1866                                  BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1867                                  BIT(BH_NILFS_Redirected));
1868
1869                         set_mask_bits(&bh->b_state, clear_bits, set_bits);
1870                         if (bh == segbuf->sb_super_root) {
1871                                 if (bh->b_page != bd_page) {
1872                                         end_page_writeback(bd_page);
1873                                         bd_page = bh->b_page;
1874                                 }
1875                                 update_sr = true;
1876                                 break;
1877                         }
1878                         if (bh->b_page != fs_page) {
1879                                 nilfs_end_page_io(fs_page, 0);
1880                                 fs_page = bh->b_page;
1881                         }
1882                 }
1883
1884                 if (!nilfs_segbuf_simplex(segbuf)) {
1885                         if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1886                                 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1887                                 sci->sc_lseg_stime = jiffies;
1888                         }
1889                         if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1890                                 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1891                 }
1892         }
1893         /*
1894          * Since pages may continue over multiple segment buffers,
1895          * end of the last page must be checked outside of the loop.
1896          */
1897         if (bd_page)
1898                 end_page_writeback(bd_page);
1899
1900         nilfs_end_page_io(fs_page, 0);
1901
1902         nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1903
1904         if (nilfs_doing_gc())
1905                 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1906         else
1907                 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1908
1909         sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1910
1911         segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1912         nilfs_set_next_segment(nilfs, segbuf);
1913
1914         if (update_sr) {
1915                 nilfs->ns_flushed_device = 0;
1916                 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1917                                        segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1918
1919                 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1920                 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1921                 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1922                 nilfs_segctor_clear_metadata_dirty(sci);
1923         } else
1924                 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1925 }
1926
1927 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1928 {
1929         int ret;
1930
1931         ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1932         if (!ret) {
1933                 nilfs_segctor_complete_write(sci);
1934                 nilfs_destroy_logs(&sci->sc_write_logs);
1935         }
1936         return ret;
1937 }
1938
1939 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1940                                              struct the_nilfs *nilfs)
1941 {
1942         struct nilfs_inode_info *ii, *n;
1943         struct inode *ifile = sci->sc_root->ifile;
1944
1945         spin_lock(&nilfs->ns_inode_lock);
1946  retry:
1947         list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1948                 if (!ii->i_bh) {
1949                         struct buffer_head *ibh;
1950                         int err;
1951
1952                         spin_unlock(&nilfs->ns_inode_lock);
1953                         err = nilfs_ifile_get_inode_block(
1954                                 ifile, ii->vfs_inode.i_ino, &ibh);
1955                         if (unlikely(err)) {
1956                                 nilfs_msg(sci->sc_super, KERN_WARNING,
1957                                           "log writer: error %d getting inode block (ino=%lu)",
1958                                           err, ii->vfs_inode.i_ino);
1959                                 return err;
1960                         }
1961                         mark_buffer_dirty(ibh);
1962                         nilfs_mdt_mark_dirty(ifile);
1963                         spin_lock(&nilfs->ns_inode_lock);
1964                         if (likely(!ii->i_bh))
1965                                 ii->i_bh = ibh;
1966                         else
1967                                 brelse(ibh);
1968                         goto retry;
1969                 }
1970
1971                 clear_bit(NILFS_I_QUEUED, &ii->i_state);
1972                 set_bit(NILFS_I_BUSY, &ii->i_state);
1973                 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
1974         }
1975         spin_unlock(&nilfs->ns_inode_lock);
1976
1977         return 0;
1978 }
1979
1980 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
1981                                              struct the_nilfs *nilfs)
1982 {
1983         struct nilfs_inode_info *ii, *n;
1984         int during_mount = !(sci->sc_super->s_flags & MS_ACTIVE);
1985         int defer_iput = false;
1986
1987         spin_lock(&nilfs->ns_inode_lock);
1988         list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
1989                 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
1990                     test_bit(NILFS_I_DIRTY, &ii->i_state))
1991                         continue;
1992
1993                 clear_bit(NILFS_I_BUSY, &ii->i_state);
1994                 brelse(ii->i_bh);
1995                 ii->i_bh = NULL;
1996                 list_del_init(&ii->i_dirty);
1997                 if (!ii->vfs_inode.i_nlink || during_mount) {
1998                         /*
1999                          * Defer calling iput() to avoid deadlocks if
2000                          * i_nlink == 0 or mount is not yet finished.
2001                          */
2002                         list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
2003                         defer_iput = true;
2004                 } else {
2005                         spin_unlock(&nilfs->ns_inode_lock);
2006                         iput(&ii->vfs_inode);
2007                         spin_lock(&nilfs->ns_inode_lock);
2008                 }
2009         }
2010         spin_unlock(&nilfs->ns_inode_lock);
2011
2012         if (defer_iput)
2013                 schedule_work(&sci->sc_iput_work);
2014 }
2015
2016 /*
2017  * Main procedure of segment constructor
2018  */
2019 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2020 {
2021         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2022         int err;
2023
2024         nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2025         sci->sc_cno = nilfs->ns_cno;
2026
2027         err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2028         if (unlikely(err))
2029                 goto out;
2030
2031         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2032                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2033
2034         if (nilfs_segctor_clean(sci))
2035                 goto out;
2036
2037         do {
2038                 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2039
2040                 err = nilfs_segctor_begin_construction(sci, nilfs);
2041                 if (unlikely(err))
2042                         goto out;
2043
2044                 /* Update time stamp */
2045                 sci->sc_seg_ctime = get_seconds();
2046
2047                 err = nilfs_segctor_collect(sci, nilfs, mode);
2048                 if (unlikely(err))
2049                         goto failed;
2050
2051                 /* Avoid empty segment */
2052                 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2053                     nilfs_segbuf_empty(sci->sc_curseg)) {
2054                         nilfs_segctor_abort_construction(sci, nilfs, 1);
2055                         goto out;
2056                 }
2057
2058                 err = nilfs_segctor_assign(sci, mode);
2059                 if (unlikely(err))
2060                         goto failed;
2061
2062                 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2063                         nilfs_segctor_fill_in_file_bmap(sci);
2064
2065                 if (mode == SC_LSEG_SR &&
2066                     nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2067                         err = nilfs_segctor_fill_in_checkpoint(sci);
2068                         if (unlikely(err))
2069                                 goto failed_to_write;
2070
2071                         nilfs_segctor_fill_in_super_root(sci, nilfs);
2072                 }
2073                 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2074
2075                 /* Write partial segments */
2076                 nilfs_segctor_prepare_write(sci);
2077
2078                 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2079                                             nilfs->ns_crc_seed);
2080
2081                 err = nilfs_segctor_write(sci, nilfs);
2082                 if (unlikely(err))
2083                         goto failed_to_write;
2084
2085                 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2086                     nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2087                         /*
2088                          * At this point, we avoid double buffering
2089                          * for blocksize < pagesize because page dirty
2090                          * flag is turned off during write and dirty
2091                          * buffers are not properly collected for
2092                          * pages crossing over segments.
2093                          */
2094                         err = nilfs_segctor_wait(sci);
2095                         if (err)
2096                                 goto failed_to_write;
2097                 }
2098         } while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2099
2100  out:
2101         nilfs_segctor_drop_written_files(sci, nilfs);
2102         return err;
2103
2104  failed_to_write:
2105         if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2106                 nilfs_redirty_inodes(&sci->sc_dirty_files);
2107
2108  failed:
2109         if (nilfs_doing_gc())
2110                 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2111         nilfs_segctor_abort_construction(sci, nilfs, err);
2112         goto out;
2113 }
2114
2115 /**
2116  * nilfs_segctor_start_timer - set timer of background write
2117  * @sci: nilfs_sc_info
2118  *
2119  * If the timer has already been set, it ignores the new request.
2120  * This function MUST be called within a section locking the segment
2121  * semaphore.
2122  */
2123 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2124 {
2125         spin_lock(&sci->sc_state_lock);
2126         if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2127                 sci->sc_timer.expires = jiffies + sci->sc_interval;
2128                 add_timer(&sci->sc_timer);
2129                 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2130         }
2131         spin_unlock(&sci->sc_state_lock);
2132 }
2133
2134 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2135 {
2136         spin_lock(&sci->sc_state_lock);
2137         if (!(sci->sc_flush_request & BIT(bn))) {
2138                 unsigned long prev_req = sci->sc_flush_request;
2139
2140                 sci->sc_flush_request |= BIT(bn);
2141                 if (!prev_req)
2142                         wake_up(&sci->sc_wait_daemon);
2143         }
2144         spin_unlock(&sci->sc_state_lock);
2145 }
2146
2147 /**
2148  * nilfs_flush_segment - trigger a segment construction for resource control
2149  * @sb: super block
2150  * @ino: inode number of the file to be flushed out.
2151  */
2152 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2153 {
2154         struct the_nilfs *nilfs = sb->s_fs_info;
2155         struct nilfs_sc_info *sci = nilfs->ns_writer;
2156
2157         if (!sci || nilfs_doing_construction())
2158                 return;
2159         nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2160                                         /* assign bit 0 to data files */
2161 }
2162
2163 struct nilfs_segctor_wait_request {
2164         wait_queue_entry_t      wq;
2165         __u32           seq;
2166         int             err;
2167         atomic_t        done;
2168 };
2169
2170 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2171 {
2172         struct nilfs_segctor_wait_request wait_req;
2173         int err = 0;
2174
2175         spin_lock(&sci->sc_state_lock);
2176         init_wait(&wait_req.wq);
2177         wait_req.err = 0;
2178         atomic_set(&wait_req.done, 0);
2179         wait_req.seq = ++sci->sc_seq_request;
2180         spin_unlock(&sci->sc_state_lock);
2181
2182         init_waitqueue_entry(&wait_req.wq, current);
2183         add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2184         set_current_state(TASK_INTERRUPTIBLE);
2185         wake_up(&sci->sc_wait_daemon);
2186
2187         for (;;) {
2188                 if (atomic_read(&wait_req.done)) {
2189                         err = wait_req.err;
2190                         break;
2191                 }
2192                 if (!signal_pending(current)) {
2193                         schedule();
2194                         continue;
2195                 }
2196                 err = -ERESTARTSYS;
2197                 break;
2198         }
2199         finish_wait(&sci->sc_wait_request, &wait_req.wq);
2200         return err;
2201 }
2202
2203 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2204 {
2205         struct nilfs_segctor_wait_request *wrq, *n;
2206         unsigned long flags;
2207
2208         spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2209         list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.head, wq.entry) {
2210                 if (!atomic_read(&wrq->done) &&
2211                     nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2212                         wrq->err = err;
2213                         atomic_set(&wrq->done, 1);
2214                 }
2215                 if (atomic_read(&wrq->done)) {
2216                         wrq->wq.func(&wrq->wq,
2217                                      TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2218                                      0, NULL);
2219                 }
2220         }
2221         spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2222 }
2223
2224 /**
2225  * nilfs_construct_segment - construct a logical segment
2226  * @sb: super block
2227  *
2228  * Return Value: On success, 0 is retured. On errors, one of the following
2229  * negative error code is returned.
2230  *
2231  * %-EROFS - Read only filesystem.
2232  *
2233  * %-EIO - I/O error
2234  *
2235  * %-ENOSPC - No space left on device (only in a panic state).
2236  *
2237  * %-ERESTARTSYS - Interrupted.
2238  *
2239  * %-ENOMEM - Insufficient memory available.
2240  */
2241 int nilfs_construct_segment(struct super_block *sb)
2242 {
2243         struct the_nilfs *nilfs = sb->s_fs_info;
2244         struct nilfs_sc_info *sci = nilfs->ns_writer;
2245         struct nilfs_transaction_info *ti;
2246         int err;
2247
2248         if (!sci)
2249                 return -EROFS;
2250
2251         /* A call inside transactions causes a deadlock. */
2252         BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2253
2254         err = nilfs_segctor_sync(sci);
2255         return err;
2256 }
2257
2258 /**
2259  * nilfs_construct_dsync_segment - construct a data-only logical segment
2260  * @sb: super block
2261  * @inode: inode whose data blocks should be written out
2262  * @start: start byte offset
2263  * @end: end byte offset (inclusive)
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_dsync_segment(struct super_block *sb, struct inode *inode,
2279                                   loff_t start, loff_t end)
2280 {
2281         struct the_nilfs *nilfs = sb->s_fs_info;
2282         struct nilfs_sc_info *sci = nilfs->ns_writer;
2283         struct nilfs_inode_info *ii;
2284         struct nilfs_transaction_info ti;
2285         int err = 0;
2286
2287         if (!sci)
2288                 return -EROFS;
2289
2290         nilfs_transaction_lock(sb, &ti, 0);
2291
2292         ii = NILFS_I(inode);
2293         if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2294             nilfs_test_opt(nilfs, STRICT_ORDER) ||
2295             test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2296             nilfs_discontinued(nilfs)) {
2297                 nilfs_transaction_unlock(sb);
2298                 err = nilfs_segctor_sync(sci);
2299                 return err;
2300         }
2301
2302         spin_lock(&nilfs->ns_inode_lock);
2303         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2304             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2305                 spin_unlock(&nilfs->ns_inode_lock);
2306                 nilfs_transaction_unlock(sb);
2307                 return 0;
2308         }
2309         spin_unlock(&nilfs->ns_inode_lock);
2310         sci->sc_dsync_inode = ii;
2311         sci->sc_dsync_start = start;
2312         sci->sc_dsync_end = end;
2313
2314         err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2315         if (!err)
2316                 nilfs->ns_flushed_device = 0;
2317
2318         nilfs_transaction_unlock(sb);
2319         return err;
2320 }
2321
2322 #define FLUSH_FILE_BIT  (0x1) /* data file only */
2323 #define FLUSH_DAT_BIT   BIT(NILFS_DAT_INO) /* DAT only */
2324
2325 /**
2326  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2327  * @sci: segment constructor object
2328  */
2329 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2330 {
2331         spin_lock(&sci->sc_state_lock);
2332         sci->sc_seq_accepted = sci->sc_seq_request;
2333         spin_unlock(&sci->sc_state_lock);
2334         del_timer_sync(&sci->sc_timer);
2335 }
2336
2337 /**
2338  * nilfs_segctor_notify - notify the result of request to caller threads
2339  * @sci: segment constructor object
2340  * @mode: mode of log forming
2341  * @err: error code to be notified
2342  */
2343 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2344 {
2345         /* Clear requests (even when the construction failed) */
2346         spin_lock(&sci->sc_state_lock);
2347
2348         if (mode == SC_LSEG_SR) {
2349                 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2350                 sci->sc_seq_done = sci->sc_seq_accepted;
2351                 nilfs_segctor_wakeup(sci, err);
2352                 sci->sc_flush_request = 0;
2353         } else {
2354                 if (mode == SC_FLUSH_FILE)
2355                         sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2356                 else if (mode == SC_FLUSH_DAT)
2357                         sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2358
2359                 /* re-enable timer if checkpoint creation was not done */
2360                 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2361                     time_before(jiffies, sci->sc_timer.expires))
2362                         add_timer(&sci->sc_timer);
2363         }
2364         spin_unlock(&sci->sc_state_lock);
2365 }
2366
2367 /**
2368  * nilfs_segctor_construct - form logs and write them to disk
2369  * @sci: segment constructor object
2370  * @mode: mode of log forming
2371  */
2372 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2373 {
2374         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2375         struct nilfs_super_block **sbp;
2376         int err = 0;
2377
2378         nilfs_segctor_accept(sci);
2379
2380         if (nilfs_discontinued(nilfs))
2381                 mode = SC_LSEG_SR;
2382         if (!nilfs_segctor_confirm(sci))
2383                 err = nilfs_segctor_do_construct(sci, mode);
2384
2385         if (likely(!err)) {
2386                 if (mode != SC_FLUSH_DAT)
2387                         atomic_set(&nilfs->ns_ndirtyblks, 0);
2388                 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2389                     nilfs_discontinued(nilfs)) {
2390                         down_write(&nilfs->ns_sem);
2391                         err = -EIO;
2392                         sbp = nilfs_prepare_super(sci->sc_super,
2393                                                   nilfs_sb_will_flip(nilfs));
2394                         if (likely(sbp)) {
2395                                 nilfs_set_log_cursor(sbp[0], nilfs);
2396                                 err = nilfs_commit_super(sci->sc_super,
2397                                                          NILFS_SB_COMMIT);
2398                         }
2399                         up_write(&nilfs->ns_sem);
2400                 }
2401         }
2402
2403         nilfs_segctor_notify(sci, mode, err);
2404         return err;
2405 }
2406
2407 static void nilfs_construction_timeout(unsigned long data)
2408 {
2409         struct task_struct *p = (struct task_struct *)data;
2410
2411         wake_up_process(p);
2412 }
2413
2414 static void
2415 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2416 {
2417         struct nilfs_inode_info *ii, *n;
2418
2419         list_for_each_entry_safe(ii, n, head, i_dirty) {
2420                 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2421                         continue;
2422                 list_del_init(&ii->i_dirty);
2423                 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2424                 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
2425                 iput(&ii->vfs_inode);
2426         }
2427 }
2428
2429 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2430                          void **kbufs)
2431 {
2432         struct the_nilfs *nilfs = sb->s_fs_info;
2433         struct nilfs_sc_info *sci = nilfs->ns_writer;
2434         struct nilfs_transaction_info ti;
2435         int err;
2436
2437         if (unlikely(!sci))
2438                 return -EROFS;
2439
2440         nilfs_transaction_lock(sb, &ti, 1);
2441
2442         err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2443         if (unlikely(err))
2444                 goto out_unlock;
2445
2446         err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2447         if (unlikely(err)) {
2448                 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2449                 goto out_unlock;
2450         }
2451
2452         sci->sc_freesegs = kbufs[4];
2453         sci->sc_nfreesegs = argv[4].v_nmembs;
2454         list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2455
2456         for (;;) {
2457                 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2458                 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2459
2460                 if (likely(!err))
2461                         break;
2462
2463                 nilfs_msg(sb, KERN_WARNING, "error %d cleaning segments", err);
2464                 set_current_state(TASK_INTERRUPTIBLE);
2465                 schedule_timeout(sci->sc_interval);
2466         }
2467         if (nilfs_test_opt(nilfs, DISCARD)) {
2468                 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2469                                                  sci->sc_nfreesegs);
2470                 if (ret) {
2471                         nilfs_msg(sb, KERN_WARNING,
2472                                   "error %d on discard request, turning discards off for the device",
2473                                   ret);
2474                         nilfs_clear_opt(nilfs, DISCARD);
2475                 }
2476         }
2477
2478  out_unlock:
2479         sci->sc_freesegs = NULL;
2480         sci->sc_nfreesegs = 0;
2481         nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2482         nilfs_transaction_unlock(sb);
2483         return err;
2484 }
2485
2486 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2487 {
2488         struct nilfs_transaction_info ti;
2489
2490         nilfs_transaction_lock(sci->sc_super, &ti, 0);
2491         nilfs_segctor_construct(sci, mode);
2492
2493         /*
2494          * Unclosed segment should be retried.  We do this using sc_timer.
2495          * Timeout of sc_timer will invoke complete construction which leads
2496          * to close the current logical segment.
2497          */
2498         if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2499                 nilfs_segctor_start_timer(sci);
2500
2501         nilfs_transaction_unlock(sci->sc_super);
2502 }
2503
2504 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2505 {
2506         int mode = 0;
2507
2508         spin_lock(&sci->sc_state_lock);
2509         mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2510                 SC_FLUSH_DAT : SC_FLUSH_FILE;
2511         spin_unlock(&sci->sc_state_lock);
2512
2513         if (mode) {
2514                 nilfs_segctor_do_construct(sci, mode);
2515
2516                 spin_lock(&sci->sc_state_lock);
2517                 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2518                         ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2519                 spin_unlock(&sci->sc_state_lock);
2520         }
2521         clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2522 }
2523
2524 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2525 {
2526         if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2527             time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2528                 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2529                         return SC_FLUSH_FILE;
2530                 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2531                         return SC_FLUSH_DAT;
2532         }
2533         return SC_LSEG_SR;
2534 }
2535
2536 /**
2537  * nilfs_segctor_thread - main loop of the segment constructor thread.
2538  * @arg: pointer to a struct nilfs_sc_info.
2539  *
2540  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2541  * to execute segment constructions.
2542  */
2543 static int nilfs_segctor_thread(void *arg)
2544 {
2545         struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2546         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2547         int timeout = 0;
2548
2549         sci->sc_timer.data = (unsigned long)current;
2550         sci->sc_timer.function = nilfs_construction_timeout;
2551
2552         /* start sync. */
2553         sci->sc_task = current;
2554         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2555         nilfs_msg(sci->sc_super, KERN_INFO,
2556                   "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2557                   sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2558
2559         spin_lock(&sci->sc_state_lock);
2560  loop:
2561         for (;;) {
2562                 int mode;
2563
2564                 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2565                         goto end_thread;
2566
2567                 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2568                         mode = SC_LSEG_SR;
2569                 else if (sci->sc_flush_request)
2570                         mode = nilfs_segctor_flush_mode(sci);
2571                 else
2572                         break;
2573
2574                 spin_unlock(&sci->sc_state_lock);
2575                 nilfs_segctor_thread_construct(sci, mode);
2576                 spin_lock(&sci->sc_state_lock);
2577                 timeout = 0;
2578         }
2579
2580
2581         if (freezing(current)) {
2582                 spin_unlock(&sci->sc_state_lock);
2583                 try_to_freeze();
2584                 spin_lock(&sci->sc_state_lock);
2585         } else {
2586                 DEFINE_WAIT(wait);
2587                 int should_sleep = 1;
2588
2589                 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2590                                 TASK_INTERRUPTIBLE);
2591
2592                 if (sci->sc_seq_request != sci->sc_seq_done)
2593                         should_sleep = 0;
2594                 else if (sci->sc_flush_request)
2595                         should_sleep = 0;
2596                 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2597                         should_sleep = time_before(jiffies,
2598                                         sci->sc_timer.expires);
2599
2600                 if (should_sleep) {
2601                         spin_unlock(&sci->sc_state_lock);
2602                         schedule();
2603                         spin_lock(&sci->sc_state_lock);
2604                 }
2605                 finish_wait(&sci->sc_wait_daemon, &wait);
2606                 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2607                            time_after_eq(jiffies, sci->sc_timer.expires));
2608
2609                 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2610                         set_nilfs_discontinued(nilfs);
2611         }
2612         goto loop;
2613
2614  end_thread:
2615         spin_unlock(&sci->sc_state_lock);
2616
2617         /* end sync. */
2618         sci->sc_task = NULL;
2619         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2620         return 0;
2621 }
2622
2623 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2624 {
2625         struct task_struct *t;
2626
2627         t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2628         if (IS_ERR(t)) {
2629                 int err = PTR_ERR(t);
2630
2631                 nilfs_msg(sci->sc_super, KERN_ERR,
2632                           "error %d creating segctord thread", err);
2633                 return err;
2634         }
2635         wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2636         return 0;
2637 }
2638
2639 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2640         __acquires(&sci->sc_state_lock)
2641         __releases(&sci->sc_state_lock)
2642 {
2643         sci->sc_state |= NILFS_SEGCTOR_QUIT;
2644
2645         while (sci->sc_task) {
2646                 wake_up(&sci->sc_wait_daemon);
2647                 spin_unlock(&sci->sc_state_lock);
2648                 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2649                 spin_lock(&sci->sc_state_lock);
2650         }
2651 }
2652
2653 /*
2654  * Setup & clean-up functions
2655  */
2656 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2657                                                struct nilfs_root *root)
2658 {
2659         struct the_nilfs *nilfs = sb->s_fs_info;
2660         struct nilfs_sc_info *sci;
2661
2662         sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2663         if (!sci)
2664                 return NULL;
2665
2666         sci->sc_super = sb;
2667
2668         nilfs_get_root(root);
2669         sci->sc_root = root;
2670
2671         init_waitqueue_head(&sci->sc_wait_request);
2672         init_waitqueue_head(&sci->sc_wait_daemon);
2673         init_waitqueue_head(&sci->sc_wait_task);
2674         spin_lock_init(&sci->sc_state_lock);
2675         INIT_LIST_HEAD(&sci->sc_dirty_files);
2676         INIT_LIST_HEAD(&sci->sc_segbufs);
2677         INIT_LIST_HEAD(&sci->sc_write_logs);
2678         INIT_LIST_HEAD(&sci->sc_gc_inodes);
2679         INIT_LIST_HEAD(&sci->sc_iput_queue);
2680         INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2681         init_timer(&sci->sc_timer);
2682
2683         sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2684         sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2685         sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2686
2687         if (nilfs->ns_interval)
2688                 sci->sc_interval = HZ * nilfs->ns_interval;
2689         if (nilfs->ns_watermark)
2690                 sci->sc_watermark = nilfs->ns_watermark;
2691         return sci;
2692 }
2693
2694 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2695 {
2696         int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2697
2698         /*
2699          * The segctord thread was stopped and its timer was removed.
2700          * But some tasks remain.
2701          */
2702         do {
2703                 struct nilfs_transaction_info ti;
2704
2705                 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2706                 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2707                 nilfs_transaction_unlock(sci->sc_super);
2708
2709                 flush_work(&sci->sc_iput_work);
2710
2711         } while (ret && retrycount-- > 0);
2712 }
2713
2714 /**
2715  * nilfs_segctor_destroy - destroy the segment constructor.
2716  * @sci: nilfs_sc_info
2717  *
2718  * nilfs_segctor_destroy() kills the segctord thread and frees
2719  * the nilfs_sc_info struct.
2720  * Caller must hold the segment semaphore.
2721  */
2722 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2723 {
2724         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2725         int flag;
2726
2727         up_write(&nilfs->ns_segctor_sem);
2728
2729         spin_lock(&sci->sc_state_lock);
2730         nilfs_segctor_kill_thread(sci);
2731         flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2732                 || sci->sc_seq_request != sci->sc_seq_done);
2733         spin_unlock(&sci->sc_state_lock);
2734
2735         if (flush_work(&sci->sc_iput_work))
2736                 flag = true;
2737
2738         if (flag || !nilfs_segctor_confirm(sci))
2739                 nilfs_segctor_write_out(sci);
2740
2741         if (!list_empty(&sci->sc_dirty_files)) {
2742                 nilfs_msg(sci->sc_super, KERN_WARNING,
2743                           "disposed unprocessed dirty file(s) when stopping log writer");
2744                 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2745         }
2746
2747         if (!list_empty(&sci->sc_iput_queue)) {
2748                 nilfs_msg(sci->sc_super, KERN_WARNING,
2749                           "disposed unprocessed inode(s) in iput queue when stopping log writer");
2750                 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2751         }
2752
2753         WARN_ON(!list_empty(&sci->sc_segbufs));
2754         WARN_ON(!list_empty(&sci->sc_write_logs));
2755
2756         nilfs_put_root(sci->sc_root);
2757
2758         down_write(&nilfs->ns_segctor_sem);
2759
2760         del_timer_sync(&sci->sc_timer);
2761         kfree(sci);
2762 }
2763
2764 /**
2765  * nilfs_attach_log_writer - attach log writer
2766  * @sb: super block instance
2767  * @root: root object of the current filesystem tree
2768  *
2769  * This allocates a log writer object, initializes it, and starts the
2770  * log writer.
2771  *
2772  * Return Value: On success, 0 is returned. On error, one of the following
2773  * negative error code is returned.
2774  *
2775  * %-ENOMEM - Insufficient memory available.
2776  */
2777 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2778 {
2779         struct the_nilfs *nilfs = sb->s_fs_info;
2780         int err;
2781
2782         if (nilfs->ns_writer) {
2783                 /*
2784                  * This happens if the filesystem was remounted
2785                  * read/write after nilfs_error degenerated it into a
2786                  * read-only mount.
2787                  */
2788                 nilfs_detach_log_writer(sb);
2789         }
2790
2791         nilfs->ns_writer = nilfs_segctor_new(sb, root);
2792         if (!nilfs->ns_writer)
2793                 return -ENOMEM;
2794
2795         err = nilfs_segctor_start_thread(nilfs->ns_writer);
2796         if (err) {
2797                 kfree(nilfs->ns_writer);
2798                 nilfs->ns_writer = NULL;
2799         }
2800         return err;
2801 }
2802
2803 /**
2804  * nilfs_detach_log_writer - destroy log writer
2805  * @sb: super block instance
2806  *
2807  * This kills log writer daemon, frees the log writer object, and
2808  * destroys list of dirty files.
2809  */
2810 void nilfs_detach_log_writer(struct super_block *sb)
2811 {
2812         struct the_nilfs *nilfs = sb->s_fs_info;
2813         LIST_HEAD(garbage_list);
2814
2815         down_write(&nilfs->ns_segctor_sem);
2816         if (nilfs->ns_writer) {
2817                 nilfs_segctor_destroy(nilfs->ns_writer);
2818                 nilfs->ns_writer = NULL;
2819         }
2820
2821         /* Force to free the list of dirty files */
2822         spin_lock(&nilfs->ns_inode_lock);
2823         if (!list_empty(&nilfs->ns_dirty_files)) {
2824                 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2825                 nilfs_msg(sb, KERN_WARNING,
2826                           "disposed unprocessed dirty file(s) when detaching log writer");
2827         }
2828         spin_unlock(&nilfs->ns_inode_lock);
2829         up_write(&nilfs->ns_segctor_sem);
2830
2831         nilfs_dispose_list(nilfs, &garbage_list, 1);
2832 }