kernel/fs/nilfs2/segment.c

   1 /*
   2  * segment.c - NILFS segment constructor.
   3  *
   4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  19  *
  20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
  21  *
  22  */
  23
  24 #include <linux/pagemap.h>
  25 #include <linux/buffer_head.h>
  26 #include <linux/writeback.h>
  27 #include <linux/bitops.h>
  28 #include <linux/bio.h>
  29 #include <linux/completion.h>
  30 #include <linux/blkdev.h>
  31 #include <linux/backing-dev.h>
  32 #include <linux/freezer.h>
  33 #include <linux/kthread.h>
  34 #include <linux/crc32.h>
  35 #include <linux/pagevec.h>
  36 #include <linux/slab.h>
  37 #include "nilfs.h"
  38 #include "btnode.h"
  39 #include "page.h"
  40 #include "segment.h"
  41 #include "sufile.h"
  42 #include "cpfile.h"
  43 #include "ifile.h"
  44 #include "segbuf.h"
  45
  46
  47 /*
  48  * Segment constructor
  49  */
  50 #define SC_N_INODEVEC   16   /* Size of locally allocated inode vector */
  51
  52 #define SC_MAX_SEGDELTA 64   /* Upper limit of the number of segments
  53                                 appended in collection retry loop */
  54
  55 /* Construction mode */
  56 enum {
  57         SC_LSEG_SR = 1, /* Make a logical segment having a super root */
  58         SC_LSEG_DSYNC,  /* Flush data blocks of a given file and make
  59                            a logical segment without a super root */
  60         SC_FLUSH_FILE,  /* Flush data files, leads to segment writes without
  61                            creating a checkpoint */
  62         SC_FLUSH_DAT,   /* Flush DAT file. This also creates segments without
  63                            a checkpoint */
  64 };
  65
  66 /* Stage numbers of dirty block collection */
  67 enum {
  68         NILFS_ST_INIT = 0,
  69         NILFS_ST_GC,            /* Collecting dirty blocks for GC */
  70         NILFS_ST_FILE,
  71         NILFS_ST_IFILE,
  72         NILFS_ST_CPFILE,
  73         NILFS_ST_SUFILE,
  74         NILFS_ST_DAT,
  75         NILFS_ST_SR,            /* Super root */
  76         NILFS_ST_DSYNC,         /* Data sync blocks */
  77         NILFS_ST_DONE,
  78 };
  79
  80 #define CREATE_TRACE_POINTS
  81 #include <trace/events/nilfs2.h>
  82
  83 /*
  84  * nilfs_sc_cstage_inc(), nilfs_sc_cstage_set(), nilfs_sc_cstage_get() are
  85  * wrapper functions of stage count (nilfs_sc_info->sc_stage.scnt). Users of
  86  * the variable must use them because transition of stage count must involve
  87  * trace events (trace_nilfs2_collection_stage_transition).
  88  *
  89  * nilfs_sc_cstage_get() isn't required for the above purpose because it doesn't
  90  * produce tracepoint events. It is provided just for making the intention
  91  * clear.
  92  */
  93 static inline void nilfs_sc_cstage_inc(struct nilfs_sc_info *sci)
  94 {
  95         sci->sc_stage.scnt++;
  96         trace_nilfs2_collection_stage_transition(sci);
  97 }
  98
  99 static inline void nilfs_sc_cstage_set(struct nilfs_sc_info *sci, int next_scnt)
 100 {
 101         sci->sc_stage.scnt = next_scnt;
 102         trace_nilfs2_collection_stage_transition(sci);
 103 }
 104
 105 static inline int nilfs_sc_cstage_get(struct nilfs_sc_info *sci)
 106 {
 107         return sci->sc_stage.scnt;
 108 }
 109
 110 /* State flags of collection */
 111 #define NILFS_CF_NODE           0x0001  /* Collecting node blocks */
 112 #define NILFS_CF_IFILE_STARTED  0x0002  /* IFILE stage has started */
 113 #define NILFS_CF_SUFREED        0x0004  /* segment usages has been freed */
 114 #define NILFS_CF_HISTORY_MASK   (NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
 115
 116 /* Operations depending on the construction mode and file type */
 117 struct nilfs_sc_operations {
 118         int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
 119                             struct inode *);
 120         int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
 121                             struct inode *);
 122         int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
 123                             struct inode *);
 124         void (*write_data_binfo)(struct nilfs_sc_info *,
 125                                  struct nilfs_segsum_pointer *,
 126                                  union nilfs_binfo *);
 127         void (*write_node_binfo)(struct nilfs_sc_info *,
 128                                  struct nilfs_segsum_pointer *,
 129                                  union nilfs_binfo *);
 130 };
 131
 132 /*
 133  * Other definitions
 134  */
 135 static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
 136 static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
 137 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
 138 static void nilfs_dispose_list(struct the_nilfs *, struct list_head *, int);
 139
 140 #define nilfs_cnt32_gt(a, b)   \
 141         (typecheck(__u32, a) && typecheck(__u32, b) && \
 142          ((__s32)(b) - (__s32)(a) < 0))
 143 #define nilfs_cnt32_ge(a, b)   \
 144         (typecheck(__u32, a) && typecheck(__u32, b) && \
 145          ((__s32)(a) - (__s32)(b) >= 0))
 146 #define nilfs_cnt32_lt(a, b)  nilfs_cnt32_gt(b, a)
 147 #define nilfs_cnt32_le(a, b)  nilfs_cnt32_ge(b, a)
 148
 149 static int nilfs_prepare_segment_lock(struct nilfs_transaction_info *ti)
 150 {
 151         struct nilfs_transaction_info *cur_ti = current->journal_info;
 152         void *save = NULL;
 153
 154         if (cur_ti) {
 155                 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
 156                         return ++cur_ti->ti_count;
 157                 else {
 158                         /*
 159                          * If journal_info field is occupied by other FS,
 160                          * it is saved and will be restored on
 161                          * nilfs_transaction_commit().
 162                          */
 163                         printk(KERN_WARNING
 164                                "NILFS warning: journal info from a different "
 165                                "FS\n");
 166                         save = current->journal_info;
 167                 }
 168         }
 169         if (!ti) {
 170                 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
 171                 if (!ti)
 172                         return -ENOMEM;
 173                 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
 174         } else {
 175                 ti->ti_flags = 0;
 176         }
 177         ti->ti_count = 0;
 178         ti->ti_save = save;
 179         ti->ti_magic = NILFS_TI_MAGIC;
 180         current->journal_info = ti;
 181         return 0;
 182 }
 183
 184 /**
 185  * nilfs_transaction_begin - start indivisible file operations.
 186  * @sb: super block
 187  * @ti: nilfs_transaction_info
 188  * @vacancy_check: flags for vacancy rate checks
 189  *
 190  * nilfs_transaction_begin() acquires a reader/writer semaphore, called
 191  * the segment semaphore, to make a segment construction and write tasks
 192  * exclusive.  The function is used with nilfs_transaction_commit() in pairs.
 193  * The region enclosed by these two functions can be nested.  To avoid a
 194  * deadlock, the semaphore is only acquired or released in the outermost call.
 195  *
 196  * This function allocates a nilfs_transaction_info struct to keep context
 197  * information on it.  It is initialized and hooked onto the current task in
 198  * the outermost call.  If a pre-allocated struct is given to @ti, it is used
 199  * instead; otherwise a new struct is assigned from a slab.
 200  *
 201  * When @vacancy_check flag is set, this function will check the amount of
 202  * free space, and will wait for the GC to reclaim disk space if low capacity.
 203  *
 204  * Return Value: On success, 0 is returned. On error, one of the following
 205  * negative error code is returned.
 206  *
 207  * %-ENOMEM - Insufficient memory available.
 208  *
 209  * %-ENOSPC - No space left on device
 210  */
 211 int nilfs_transaction_begin(struct super_block *sb,
 212                             struct nilfs_transaction_info *ti,
 213                             int vacancy_check)
 214 {
 215         struct the_nilfs *nilfs;
 216         int ret = nilfs_prepare_segment_lock(ti);
 217         struct nilfs_transaction_info *trace_ti;
 218
 219         if (unlikely(ret < 0))
 220                 return ret;
 221         if (ret > 0) {
 222                 trace_ti = current->journal_info;
 223
 224                 trace_nilfs2_transaction_transition(sb, trace_ti,
 225                                     trace_ti->ti_count, trace_ti->ti_flags,
 226                                     TRACE_NILFS2_TRANSACTION_BEGIN);
 227                 return 0;
 228         }
 229
 230         sb_start_intwrite(sb);
 231
 232         nilfs = sb->s_fs_info;
 233         down_read(&nilfs->ns_segctor_sem);
 234         if (vacancy_check && nilfs_near_disk_full(nilfs)) {
 235                 up_read(&nilfs->ns_segctor_sem);
 236                 ret = -ENOSPC;
 237                 goto failed;
 238         }
 239
 240         trace_ti = current->journal_info;
 241         trace_nilfs2_transaction_transition(sb, trace_ti, trace_ti->ti_count,
 242                                             trace_ti->ti_flags,
 243                                             TRACE_NILFS2_TRANSACTION_BEGIN);
 244         return 0;
 245
 246  failed:
 247         ti = current->journal_info;
 248         current->journal_info = ti->ti_save;
 249         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 250                 kmem_cache_free(nilfs_transaction_cachep, ti);
 251         sb_end_intwrite(sb);
 252         return ret;
 253 }
 254
 255 /**
 256  * nilfs_transaction_commit - commit indivisible file operations.
 257  * @sb: super block
 258  *
 259  * nilfs_transaction_commit() releases the read semaphore which is
 260  * acquired by nilfs_transaction_begin(). This is only performed
 261  * in outermost call of this function.  If a commit flag is set,
 262  * nilfs_transaction_commit() sets a timer to start the segment
 263  * constructor.  If a sync flag is set, it starts construction
 264  * directly.
 265  */
 266 int nilfs_transaction_commit(struct super_block *sb)
 267 {
 268         struct nilfs_transaction_info *ti = current->journal_info;
 269         struct the_nilfs *nilfs = sb->s_fs_info;
 270         int err = 0;
 271
 272         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 273         ti->ti_flags |= NILFS_TI_COMMIT;
 274         if (ti->ti_count > 0) {
 275                 ti->ti_count--;
 276                 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 277                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
 278                 return 0;
 279         }
 280         if (nilfs->ns_writer) {
 281                 struct nilfs_sc_info *sci = nilfs->ns_writer;
 282
 283                 if (ti->ti_flags & NILFS_TI_COMMIT)
 284                         nilfs_segctor_start_timer(sci);
 285                 if (atomic_read(&nilfs->ns_ndirtyblks) > sci->sc_watermark)
 286                         nilfs_segctor_do_flush(sci, 0);
 287         }
 288         up_read(&nilfs->ns_segctor_sem);
 289         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 290                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
 291
 292         current->journal_info = ti->ti_save;
 293
 294         if (ti->ti_flags & NILFS_TI_SYNC)
 295                 err = nilfs_construct_segment(sb);
 296         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 297                 kmem_cache_free(nilfs_transaction_cachep, ti);
 298         sb_end_intwrite(sb);
 299         return err;
 300 }
 301
 302 void nilfs_transaction_abort(struct super_block *sb)
 303 {
 304         struct nilfs_transaction_info *ti = current->journal_info;
 305         struct the_nilfs *nilfs = sb->s_fs_info;
 306
 307         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 308         if (ti->ti_count > 0) {
 309                 ti->ti_count--;
 310                 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 311                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
 312                 return;
 313         }
 314         up_read(&nilfs->ns_segctor_sem);
 315
 316         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 317                     ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
 318
 319         current->journal_info = ti->ti_save;
 320         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 321                 kmem_cache_free(nilfs_transaction_cachep, ti);
 322         sb_end_intwrite(sb);
 323 }
 324
 325 void nilfs_relax_pressure_in_lock(struct super_block *sb)
 326 {
 327         struct the_nilfs *nilfs = sb->s_fs_info;
 328         struct nilfs_sc_info *sci = nilfs->ns_writer;
 329
 330         if (!sci || !sci->sc_flush_request)
 331                 return;
 332
 333         set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
 334         up_read(&nilfs->ns_segctor_sem);
 335
 336         down_write(&nilfs->ns_segctor_sem);
 337         if (sci->sc_flush_request &&
 338             test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
 339                 struct nilfs_transaction_info *ti = current->journal_info;
 340
 341                 ti->ti_flags |= NILFS_TI_WRITER;
 342                 nilfs_segctor_do_immediate_flush(sci);
 343                 ti->ti_flags &= ~NILFS_TI_WRITER;
 344         }
 345         downgrade_write(&nilfs->ns_segctor_sem);
 346 }
 347
 348 static void nilfs_transaction_lock(struct super_block *sb,
 349                                    struct nilfs_transaction_info *ti,
 350                                    int gcflag)
 351 {
 352         struct nilfs_transaction_info *cur_ti = current->journal_info;
 353         struct the_nilfs *nilfs = sb->s_fs_info;
 354         struct nilfs_sc_info *sci = nilfs->ns_writer;
 355
 356         WARN_ON(cur_ti);
 357         ti->ti_flags = NILFS_TI_WRITER;
 358         ti->ti_count = 0;
 359         ti->ti_save = cur_ti;
 360         ti->ti_magic = NILFS_TI_MAGIC;
 361         current->journal_info = ti;
 362
 363         for (;;) {
 364                 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 365                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_TRYLOCK);
 366
 367                 down_write(&nilfs->ns_segctor_sem);
 368                 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
 369                         break;
 370
 371                 nilfs_segctor_do_immediate_flush(sci);
 372
 373                 up_write(&nilfs->ns_segctor_sem);
 374                 yield();
 375         }
 376         if (gcflag)
 377                 ti->ti_flags |= NILFS_TI_GC;
 378
 379         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 380                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_LOCK);
 381 }
 382
 383 static void nilfs_transaction_unlock(struct super_block *sb)
 384 {
 385         struct nilfs_transaction_info *ti = current->journal_info;
 386         struct the_nilfs *nilfs = sb->s_fs_info;
 387
 388         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 389         BUG_ON(ti->ti_count > 0);
 390
 391         up_write(&nilfs->ns_segctor_sem);
 392         current->journal_info = ti->ti_save;
 393
 394         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 395                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_UNLOCK);
 396 }
 397
 398 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
 399                                             struct nilfs_segsum_pointer *ssp,
 400                                             unsigned bytes)
 401 {
 402         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 403         unsigned blocksize = sci->sc_super->s_blocksize;
 404         void *p;
 405
 406         if (unlikely(ssp->offset + bytes > blocksize)) {
 407                 ssp->offset = 0;
 408                 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
 409                                                &segbuf->sb_segsum_buffers));
 410                 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
 411         }
 412         p = ssp->bh->b_data + ssp->offset;
 413         ssp->offset += bytes;
 414         return p;
 415 }
 416
 417 /**
 418  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
 419  * @sci: nilfs_sc_info
 420  */
 421 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
 422 {
 423         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 424         struct buffer_head *sumbh;
 425         unsigned sumbytes;
 426         unsigned flags = 0;
 427         int err;
 428
 429         if (nilfs_doing_gc())
 430                 flags = NILFS_SS_GC;
 431         err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
 432         if (unlikely(err))
 433                 return err;
 434
 435         sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
 436         sumbytes = segbuf->sb_sum.sumbytes;
 437         sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
 438         sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
 439         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 440         return 0;
 441 }
 442
 443 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
 444 {
 445         sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
 446         if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
 447                 return -E2BIG; /* The current segment is filled up
 448                                   (internal code) */
 449         sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
 450         return nilfs_segctor_reset_segment_buffer(sci);
 451 }
 452
 453 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
 454 {
 455         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 456         int err;
 457
 458         if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
 459                 err = nilfs_segctor_feed_segment(sci);
 460                 if (err)
 461                         return err;
 462                 segbuf = sci->sc_curseg;
 463         }
 464         err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
 465         if (likely(!err))
 466                 segbuf->sb_sum.flags |= NILFS_SS_SR;
 467         return err;
 468 }
 469
 470 /*
 471  * Functions for making segment summary and payloads
 472  */
 473 static int nilfs_segctor_segsum_block_required(
 474         struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
 475         unsigned binfo_size)
 476 {
 477         unsigned blocksize = sci->sc_super->s_blocksize;
 478         /* Size of finfo and binfo is enough small against blocksize */
 479
 480         return ssp->offset + binfo_size +
 481                 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
 482                 blocksize;
 483 }
 484
 485 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
 486                                       struct inode *inode)
 487 {
 488         sci->sc_curseg->sb_sum.nfinfo++;
 489         sci->sc_binfo_ptr = sci->sc_finfo_ptr;
 490         nilfs_segctor_map_segsum_entry(
 491                 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
 492
 493         if (NILFS_I(inode)->i_root &&
 494             !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 495                 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
 496         /* skip finfo */
 497 }
 498
 499 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
 500                                     struct inode *inode)
 501 {
 502         struct nilfs_finfo *finfo;
 503         struct nilfs_inode_info *ii;
 504         struct nilfs_segment_buffer *segbuf;
 505         __u64 cno;
 506
 507         if (sci->sc_blk_cnt == 0)
 508                 return;
 509
 510         ii = NILFS_I(inode);
 511
 512         if (test_bit(NILFS_I_GCINODE, &ii->i_state))
 513                 cno = ii->i_cno;
 514         else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
 515                 cno = 0;
 516         else
 517                 cno = sci->sc_cno;
 518
 519         finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
 520                                                  sizeof(*finfo));
 521         finfo->fi_ino = cpu_to_le64(inode->i_ino);
 522         finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
 523         finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
 524         finfo->fi_cno = cpu_to_le64(cno);
 525
 526         segbuf = sci->sc_curseg;
 527         segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
 528                 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
 529         sci->sc_finfo_ptr = sci->sc_binfo_ptr;
 530         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 531 }
 532
 533 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
 534                                         struct buffer_head *bh,
 535                                         struct inode *inode,
 536                                         unsigned binfo_size)
 537 {
 538         struct nilfs_segment_buffer *segbuf;
 539         int required, err = 0;
 540
 541  retry:
 542         segbuf = sci->sc_curseg;
 543         required = nilfs_segctor_segsum_block_required(
 544                 sci, &sci->sc_binfo_ptr, binfo_size);
 545         if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
 546                 nilfs_segctor_end_finfo(sci, inode);
 547                 err = nilfs_segctor_feed_segment(sci);
 548                 if (err)
 549                         return err;
 550                 goto retry;
 551         }
 552         if (unlikely(required)) {
 553                 err = nilfs_segbuf_extend_segsum(segbuf);
 554                 if (unlikely(err))
 555                         goto failed;
 556         }
 557         if (sci->sc_blk_cnt == 0)
 558                 nilfs_segctor_begin_finfo(sci, inode);
 559
 560         nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
 561         /* Substitution to vblocknr is delayed until update_blocknr() */
 562         nilfs_segbuf_add_file_buffer(segbuf, bh);
 563         sci->sc_blk_cnt++;
 564  failed:
 565         return err;
 566 }
 567
 568 /*
 569  * Callback functions that enumerate, mark, and collect dirty blocks
 570  */
 571 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
 572                                    struct buffer_head *bh, struct inode *inode)
 573 {
 574         int err;
 575
 576         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 577         if (err < 0)
 578                 return err;
 579
 580         err = nilfs_segctor_add_file_block(sci, bh, inode,
 581                                            sizeof(struct nilfs_binfo_v));
 582         if (!err)
 583                 sci->sc_datablk_cnt++;
 584         return err;
 585 }
 586
 587 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
 588                                    struct buffer_head *bh,
 589                                    struct inode *inode)
 590 {
 591         return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 592 }
 593
 594 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
 595                                    struct buffer_head *bh,
 596                                    struct inode *inode)
 597 {
 598         WARN_ON(!buffer_dirty(bh));
 599         return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 600 }
 601
 602 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
 603                                         struct nilfs_segsum_pointer *ssp,
 604                                         union nilfs_binfo *binfo)
 605 {
 606         struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
 607                 sci, ssp, sizeof(*binfo_v));
 608         *binfo_v = binfo->bi_v;
 609 }
 610
 611 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
 612                                         struct nilfs_segsum_pointer *ssp,
 613                                         union nilfs_binfo *binfo)
 614 {
 615         __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
 616                 sci, ssp, sizeof(*vblocknr));
 617         *vblocknr = binfo->bi_v.bi_vblocknr;
 618 }
 619
 620 static struct nilfs_sc_operations nilfs_sc_file_ops = {
 621         .collect_data = nilfs_collect_file_data,
 622         .collect_node = nilfs_collect_file_node,
 623         .collect_bmap = nilfs_collect_file_bmap,
 624         .write_data_binfo = nilfs_write_file_data_binfo,
 625         .write_node_binfo = nilfs_write_file_node_binfo,
 626 };
 627
 628 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
 629                                   struct buffer_head *bh, struct inode *inode)
 630 {
 631         int err;
 632
 633         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 634         if (err < 0)
 635                 return err;
 636
 637         err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 638         if (!err)
 639                 sci->sc_datablk_cnt++;
 640         return err;
 641 }
 642
 643 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
 644                                   struct buffer_head *bh, struct inode *inode)
 645 {
 646         WARN_ON(!buffer_dirty(bh));
 647         return nilfs_segctor_add_file_block(sci, bh, inode,
 648                                             sizeof(struct nilfs_binfo_dat));
 649 }
 650
 651 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
 652                                        struct nilfs_segsum_pointer *ssp,
 653                                        union nilfs_binfo *binfo)
 654 {
 655         __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
 656                                                           sizeof(*blkoff));
 657         *blkoff = binfo->bi_dat.bi_blkoff;
 658 }
 659
 660 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
 661                                        struct nilfs_segsum_pointer *ssp,
 662                                        union nilfs_binfo *binfo)
 663 {
 664         struct nilfs_binfo_dat *binfo_dat =
 665                 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
 666         *binfo_dat = binfo->bi_dat;
 667 }
 668
 669 static struct nilfs_sc_operations nilfs_sc_dat_ops = {
 670         .collect_data = nilfs_collect_dat_data,
 671         .collect_node = nilfs_collect_file_node,
 672         .collect_bmap = nilfs_collect_dat_bmap,
 673         .write_data_binfo = nilfs_write_dat_data_binfo,
 674         .write_node_binfo = nilfs_write_dat_node_binfo,
 675 };
 676
 677 static struct nilfs_sc_operations nilfs_sc_dsync_ops = {
 678         .collect_data = nilfs_collect_file_data,
 679         .collect_node = NULL,
 680         .collect_bmap = NULL,
 681         .write_data_binfo = nilfs_write_file_data_binfo,
 682         .write_node_binfo = NULL,
 683 };
 684
 685 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
 686                                               struct list_head *listp,
 687                                               size_t nlimit,
 688                                               loff_t start, loff_t end)
 689 {
 690         struct address_space *mapping = inode->i_mapping;
 691         struct pagevec pvec;
 692         pgoff_t index = 0, last = ULONG_MAX;
 693         size_t ndirties = 0;
 694         int i;
 695
 696         if (unlikely(start != 0 || end != LLONG_MAX)) {
 697                 /*
 698                  * A valid range is given for sync-ing data pages. The
 699                  * range is rounded to per-page; extra dirty buffers
 700                  * may be included if blocksize < pagesize.
 701                  */
 702                 index = start >> PAGE_SHIFT;
 703                 last = end >> PAGE_SHIFT;
 704         }
 705         pagevec_init(&pvec, 0);
 706  repeat:
 707         if (unlikely(index > last) ||
 708             !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 709                                 min_t(pgoff_t, last - index,
 710                                       PAGEVEC_SIZE - 1) + 1))
 711                 return ndirties;
 712
 713         for (i = 0; i < pagevec_count(&pvec); i++) {
 714                 struct buffer_head *bh, *head;
 715                 struct page *page = pvec.pages[i];
 716
 717                 if (unlikely(page->index > last))
 718                         break;
 719
 720                 lock_page(page);
 721                 if (!page_has_buffers(page))
 722                         create_empty_buffers(page, 1 << inode->i_blkbits, 0);
 723                 unlock_page(page);
 724
 725                 bh = head = page_buffers(page);
 726                 do {
 727                         if (!buffer_dirty(bh) || buffer_async_write(bh))
 728                                 continue;
 729                         get_bh(bh);
 730                         list_add_tail(&bh->b_assoc_buffers, listp);
 731                         ndirties++;
 732                         if (unlikely(ndirties >= nlimit)) {
 733                                 pagevec_release(&pvec);
 734                                 cond_resched();
 735                                 return ndirties;
 736                         }
 737                 } while (bh = bh->b_this_page, bh != head);
 738         }
 739         pagevec_release(&pvec);
 740         cond_resched();
 741         goto repeat;
 742 }
 743
 744 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
 745                                             struct list_head *listp)
 746 {
 747         struct nilfs_inode_info *ii = NILFS_I(inode);
 748         struct address_space *mapping = &ii->i_btnode_cache;
 749         struct pagevec pvec;
 750         struct buffer_head *bh, *head;
 751         unsigned int i;
 752         pgoff_t index = 0;
 753
 754         pagevec_init(&pvec, 0);
 755
 756         while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 757                                   PAGEVEC_SIZE)) {
 758                 for (i = 0; i < pagevec_count(&pvec); i++) {
 759                         bh = head = page_buffers(pvec.pages[i]);
 760                         do {
 761                                 if (buffer_dirty(bh) &&
 762                                                 !buffer_async_write(bh)) {
 763                                         get_bh(bh);
 764                                         list_add_tail(&bh->b_assoc_buffers,
 765                                                       listp);
 766                                 }
 767                                 bh = bh->b_this_page;
 768                         } while (bh != head);
 769                 }
 770                 pagevec_release(&pvec);
 771                 cond_resched();
 772         }
 773 }
 774
 775 static void nilfs_dispose_list(struct the_nilfs *nilfs,
 776                                struct list_head *head, int force)
 777 {
 778         struct nilfs_inode_info *ii, *n;
 779         struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
 780         unsigned nv = 0;
 781
 782         while (!list_empty(head)) {
 783                 spin_lock(&nilfs->ns_inode_lock);
 784                 list_for_each_entry_safe(ii, n, head, i_dirty) {
 785                         list_del_init(&ii->i_dirty);
 786                         if (force) {
 787                                 if (unlikely(ii->i_bh)) {
 788                                         brelse(ii->i_bh);
 789                                         ii->i_bh = NULL;
 790                                 }
 791                         } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
 792                                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 793                                 list_add_tail(&ii->i_dirty,
 794                                               &nilfs->ns_dirty_files);
 795                                 continue;
 796                         }
 797                         ivec[nv++] = ii;
 798                         if (nv == SC_N_INODEVEC)
 799                                 break;
 800                 }
 801                 spin_unlock(&nilfs->ns_inode_lock);
 802
 803                 for (pii = ivec; nv > 0; pii++, nv--)
 804                         iput(&(*pii)->vfs_inode);
 805         }
 806 }
 807
 808 static void nilfs_iput_work_func(struct work_struct *work)
 809 {
 810         struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
 811                                                  sc_iput_work);
 812         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 813
 814         nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
 815 }
 816
 817 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
 818                                      struct nilfs_root *root)
 819 {
 820         int ret = 0;
 821
 822         if (nilfs_mdt_fetch_dirty(root->ifile))
 823                 ret++;
 824         if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
 825                 ret++;
 826         if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
 827                 ret++;
 828         if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
 829                 ret++;
 830         return ret;
 831 }
 832
 833 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
 834 {
 835         return list_empty(&sci->sc_dirty_files) &&
 836                 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
 837                 sci->sc_nfreesegs == 0 &&
 838                 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
 839 }
 840
 841 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
 842 {
 843         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 844         int ret = 0;
 845
 846         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
 847                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
 848
 849         spin_lock(&nilfs->ns_inode_lock);
 850         if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
 851                 ret++;
 852
 853         spin_unlock(&nilfs->ns_inode_lock);
 854         return ret;
 855 }
 856
 857 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
 858 {
 859         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 860
 861         nilfs_mdt_clear_dirty(sci->sc_root->ifile);
 862         nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
 863         nilfs_mdt_clear_dirty(nilfs->ns_sufile);
 864         nilfs_mdt_clear_dirty(nilfs->ns_dat);
 865 }
 866
 867 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
 868 {
 869         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 870         struct buffer_head *bh_cp;
 871         struct nilfs_checkpoint *raw_cp;
 872         int err;
 873
 874         /* XXX: this interface will be changed */
 875         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
 876                                           &raw_cp, &bh_cp);
 877         if (likely(!err)) {
 878                 /* The following code is duplicated with cpfile.  But, it is
 879                    needed to collect the checkpoint even if it was not newly
 880                    created */
 881                 mark_buffer_dirty(bh_cp);
 882                 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
 883                 nilfs_cpfile_put_checkpoint(
 884                         nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 885         } else
 886                 WARN_ON(err == -EINVAL || err == -ENOENT);
 887
 888         return err;
 889 }
 890
 891 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
 892 {
 893         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 894         struct buffer_head *bh_cp;
 895         struct nilfs_checkpoint *raw_cp;
 896         int err;
 897
 898         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
 899                                           &raw_cp, &bh_cp);
 900         if (unlikely(err)) {
 901                 WARN_ON(err == -EINVAL || err == -ENOENT);
 902                 goto failed_ibh;
 903         }
 904         raw_cp->cp_snapshot_list.ssl_next = 0;
 905         raw_cp->cp_snapshot_list.ssl_prev = 0;
 906         raw_cp->cp_inodes_count =
 907                 cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
 908         raw_cp->cp_blocks_count =
 909                 cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
 910         raw_cp->cp_nblk_inc =
 911                 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
 912         raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
 913         raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
 914
 915         if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 916                 nilfs_checkpoint_clear_minor(raw_cp);
 917         else
 918                 nilfs_checkpoint_set_minor(raw_cp);
 919
 920         nilfs_write_inode_common(sci->sc_root->ifile,
 921                                  &raw_cp->cp_ifile_inode, 1);
 922         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 923         return 0;
 924
 925  failed_ibh:
 926         return err;
 927 }
 928
 929 static void nilfs_fill_in_file_bmap(struct inode *ifile,
 930                                     struct nilfs_inode_info *ii)
 931
 932 {
 933         struct buffer_head *ibh;
 934         struct nilfs_inode *raw_inode;
 935
 936         if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
 937                 ibh = ii->i_bh;
 938                 BUG_ON(!ibh);
 939                 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
 940                                                   ibh);
 941                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 942                 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
 943         }
 944 }
 945
 946 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
 947 {
 948         struct nilfs_inode_info *ii;
 949
 950         list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
 951                 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
 952                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
 953         }
 954 }
 955
 956 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
 957                                              struct the_nilfs *nilfs)
 958 {
 959         struct buffer_head *bh_sr;
 960         struct nilfs_super_root *raw_sr;
 961         unsigned isz, srsz;
 962
 963         bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
 964         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
 965         isz = nilfs->ns_inode_size;
 966         srsz = NILFS_SR_BYTES(isz);
 967
 968         raw_sr->sr_bytes = cpu_to_le16(srsz);
 969         raw_sr->sr_nongc_ctime
 970                 = cpu_to_le64(nilfs_doing_gc() ?
 971                               nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
 972         raw_sr->sr_flags = 0;
 973
 974         nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
 975                                  NILFS_SR_DAT_OFFSET(isz), 1);
 976         nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
 977                                  NILFS_SR_CPFILE_OFFSET(isz), 1);
 978         nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
 979                                  NILFS_SR_SUFILE_OFFSET(isz), 1);
 980         memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
 981 }
 982
 983 static void nilfs_redirty_inodes(struct list_head *head)
 984 {
 985         struct nilfs_inode_info *ii;
 986
 987         list_for_each_entry(ii, head, i_dirty) {
 988                 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
 989                         clear_bit(NILFS_I_COLLECTED, &ii->i_state);
 990         }
 991 }
 992
 993 static void nilfs_drop_collected_inodes(struct list_head *head)
 994 {
 995         struct nilfs_inode_info *ii;
 996
 997         list_for_each_entry(ii, head, i_dirty) {
 998                 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
 999                         continue;
1000
1001                 clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
1002                 set_bit(NILFS_I_UPDATED, &ii->i_state);
1003         }
1004 }
1005
1006 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1007                                        struct inode *inode,
1008                                        struct list_head *listp,
1009                                        int (*collect)(struct nilfs_sc_info *,
1010                                                       struct buffer_head *,
1011                                                       struct inode *))
1012 {
1013         struct buffer_head *bh, *n;
1014         int err = 0;
1015
1016         if (collect) {
1017                 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1018                         list_del_init(&bh->b_assoc_buffers);
1019                         err = collect(sci, bh, inode);
1020                         brelse(bh);
1021                         if (unlikely(err))
1022                                 goto dispose_buffers;
1023                 }
1024                 return 0;
1025         }
1026
1027  dispose_buffers:
1028         while (!list_empty(listp)) {
1029                 bh = list_first_entry(listp, struct buffer_head,
1030                                       b_assoc_buffers);
1031                 list_del_init(&bh->b_assoc_buffers);
1032                 brelse(bh);
1033         }
1034         return err;
1035 }
1036
1037 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1038 {
1039         /* Remaining number of blocks within segment buffer */
1040         return sci->sc_segbuf_nblocks -
1041                 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1042 }
1043
1044 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1045                                    struct inode *inode,
1046                                    struct nilfs_sc_operations *sc_ops)
1047 {
1048         LIST_HEAD(data_buffers);
1049         LIST_HEAD(node_buffers);
1050         int err;
1051
1052         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1053                 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1054
1055                 n = nilfs_lookup_dirty_data_buffers(
1056                         inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1057                 if (n > rest) {
1058                         err = nilfs_segctor_apply_buffers(
1059                                 sci, inode, &data_buffers,
1060                                 sc_ops->collect_data);
1061                         BUG_ON(!err); /* always receive -E2BIG or true error */
1062                         goto break_or_fail;
1063                 }
1064         }
1065         nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1066
1067         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1068                 err = nilfs_segctor_apply_buffers(
1069                         sci, inode, &data_buffers, sc_ops->collect_data);
1070                 if (unlikely(err)) {
1071                         /* dispose node list */
1072                         nilfs_segctor_apply_buffers(
1073                                 sci, inode, &node_buffers, NULL);
1074                         goto break_or_fail;
1075                 }
1076                 sci->sc_stage.flags |= NILFS_CF_NODE;
1077         }
1078         /* Collect node */
1079         err = nilfs_segctor_apply_buffers(
1080                 sci, inode, &node_buffers, sc_ops->collect_node);
1081         if (unlikely(err))
1082                 goto break_or_fail;
1083
1084         nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1085         err = nilfs_segctor_apply_buffers(
1086                 sci, inode, &node_buffers, sc_ops->collect_bmap);
1087         if (unlikely(err))
1088                 goto break_or_fail;
1089
1090         nilfs_segctor_end_finfo(sci, inode);
1091         sci->sc_stage.flags &= ~NILFS_CF_NODE;
1092
1093  break_or_fail:
1094         return err;
1095 }
1096
1097 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1098                                          struct inode *inode)
1099 {
1100         LIST_HEAD(data_buffers);
1101         size_t n, rest = nilfs_segctor_buffer_rest(sci);
1102         int err;
1103
1104         n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1105                                             sci->sc_dsync_start,
1106                                             sci->sc_dsync_end);
1107
1108         err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1109                                           nilfs_collect_file_data);
1110         if (!err) {
1111                 nilfs_segctor_end_finfo(sci, inode);
1112                 BUG_ON(n > rest);
1113                 /* always receive -E2BIG or true error if n > rest */
1114         }
1115         return err;
1116 }
1117
1118 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1119 {
1120         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1121         struct list_head *head;
1122         struct nilfs_inode_info *ii;
1123         size_t ndone;
1124         int err = 0;
1125
1126         switch (nilfs_sc_cstage_get(sci)) {
1127         case NILFS_ST_INIT:
1128                 /* Pre-processes */
1129                 sci->sc_stage.flags = 0;
1130
1131                 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1132                         sci->sc_nblk_inc = 0;
1133                         sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1134                         if (mode == SC_LSEG_DSYNC) {
1135                                 nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1136                                 goto dsync_mode;
1137                         }
1138                 }
1139
1140                 sci->sc_stage.dirty_file_ptr = NULL;
1141                 sci->sc_stage.gc_inode_ptr = NULL;
1142                 if (mode == SC_FLUSH_DAT) {
1143                         nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1144                         goto dat_stage;
1145                 }
1146                 nilfs_sc_cstage_inc(sci);  /* Fall through */
1147         case NILFS_ST_GC:
1148                 if (nilfs_doing_gc()) {
1149                         head = &sci->sc_gc_inodes;
1150                         ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1151                                                 head, i_dirty);
1152                         list_for_each_entry_continue(ii, head, i_dirty) {
1153                                 err = nilfs_segctor_scan_file(
1154                                         sci, &ii->vfs_inode,
1155                                         &nilfs_sc_file_ops);
1156                                 if (unlikely(err)) {
1157                                         sci->sc_stage.gc_inode_ptr = list_entry(
1158                                                 ii->i_dirty.prev,
1159                                                 struct nilfs_inode_info,
1160                                                 i_dirty);
1161                                         goto break_or_fail;
1162                                 }
1163                                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1164                         }
1165                         sci->sc_stage.gc_inode_ptr = NULL;
1166                 }
1167                 nilfs_sc_cstage_inc(sci);  /* Fall through */
1168         case NILFS_ST_FILE:
1169                 head = &sci->sc_dirty_files;
1170                 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1171                                         i_dirty);
1172                 list_for_each_entry_continue(ii, head, i_dirty) {
1173                         clear_bit(NILFS_I_DIRTY, &ii->i_state);
1174
1175                         err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1176                                                       &nilfs_sc_file_ops);
1177                         if (unlikely(err)) {
1178                                 sci->sc_stage.dirty_file_ptr =
1179                                         list_entry(ii->i_dirty.prev,
1180                                                    struct nilfs_inode_info,
1181                                                    i_dirty);
1182                                 goto break_or_fail;
1183                         }
1184                         /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1185                         /* XXX: required ? */
1186                 }
1187                 sci->sc_stage.dirty_file_ptr = NULL;
1188                 if (mode == SC_FLUSH_FILE) {
1189                         nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1190                         return 0;
1191                 }
1192                 nilfs_sc_cstage_inc(sci);
1193                 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1194                 /* Fall through */
1195         case NILFS_ST_IFILE:
1196                 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1197                                               &nilfs_sc_file_ops);
1198                 if (unlikely(err))
1199                         break;
1200                 nilfs_sc_cstage_inc(sci);
1201                 /* Creating a checkpoint */
1202                 err = nilfs_segctor_create_checkpoint(sci);
1203                 if (unlikely(err))
1204                         break;
1205                 /* Fall through */
1206         case NILFS_ST_CPFILE:
1207                 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1208                                               &nilfs_sc_file_ops);
1209                 if (unlikely(err))
1210                         break;
1211                 nilfs_sc_cstage_inc(sci);  /* Fall through */
1212         case NILFS_ST_SUFILE:
1213                 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1214                                          sci->sc_nfreesegs, &ndone);
1215                 if (unlikely(err)) {
1216                         nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1217                                                   sci->sc_freesegs, ndone,
1218                                                   NULL);
1219                         break;
1220                 }
1221                 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1222
1223                 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1224                                               &nilfs_sc_file_ops);
1225                 if (unlikely(err))
1226                         break;
1227                 nilfs_sc_cstage_inc(sci);  /* Fall through */
1228         case NILFS_ST_DAT:
1229  dat_stage:
1230                 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1231                                               &nilfs_sc_dat_ops);
1232                 if (unlikely(err))
1233                         break;
1234                 if (mode == SC_FLUSH_DAT) {
1235                         nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1236                         return 0;
1237                 }
1238                 nilfs_sc_cstage_inc(sci);  /* Fall through */
1239         case NILFS_ST_SR:
1240                 if (mode == SC_LSEG_SR) {
1241                         /* Appending a super root */
1242                         err = nilfs_segctor_add_super_root(sci);
1243                         if (unlikely(err))
1244                                 break;
1245                 }
1246                 /* End of a logical segment */
1247                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1248                 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1249                 return 0;
1250         case NILFS_ST_DSYNC:
1251  dsync_mode:
1252                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1253                 ii = sci->sc_dsync_inode;
1254                 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1255                         break;
1256
1257                 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1258                 if (unlikely(err))
1259                         break;
1260                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1261                 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1262                 return 0;
1263         case NILFS_ST_DONE:
1264                 return 0;
1265         default:
1266                 BUG();
1267         }
1268
1269  break_or_fail:
1270         return err;
1271 }
1272
1273 /**
1274  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1275  * @sci: nilfs_sc_info
1276  * @nilfs: nilfs object
1277  */
1278 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1279                                             struct the_nilfs *nilfs)
1280 {
1281         struct nilfs_segment_buffer *segbuf, *prev;
1282         __u64 nextnum;
1283         int err, alloc = 0;
1284
1285         segbuf = nilfs_segbuf_new(sci->sc_super);
1286         if (unlikely(!segbuf))
1287                 return -ENOMEM;
1288
1289         if (list_empty(&sci->sc_write_logs)) {
1290                 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1291                                  nilfs->ns_pseg_offset, nilfs);
1292                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1293                         nilfs_shift_to_next_segment(nilfs);
1294                         nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1295                 }
1296
1297                 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1298                 nextnum = nilfs->ns_nextnum;
1299
1300                 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1301                         /* Start from the head of a new full segment */
1302                         alloc++;
1303         } else {
1304                 /* Continue logs */
1305                 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1306                 nilfs_segbuf_map_cont(segbuf, prev);
1307                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1308                 nextnum = prev->sb_nextnum;
1309
1310                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1311                         nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1312                         segbuf->sb_sum.seg_seq++;
1313                         alloc++;
1314                 }
1315         }
1316
1317         err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1318         if (err)
1319                 goto failed;
1320
1321         if (alloc) {
1322                 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1323                 if (err)
1324                         goto failed;
1325         }
1326         nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1327
1328         BUG_ON(!list_empty(&sci->sc_segbufs));
1329         list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1330         sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1331         return 0;
1332
1333  failed:
1334         nilfs_segbuf_free(segbuf);
1335         return err;
1336 }
1337
1338 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1339                                          struct the_nilfs *nilfs, int nadd)
1340 {
1341         struct nilfs_segment_buffer *segbuf, *prev;
1342         struct inode *sufile = nilfs->ns_sufile;
1343         __u64 nextnextnum;
1344         LIST_HEAD(list);
1345         int err, ret, i;
1346
1347         prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1348         /*
1349          * Since the segment specified with nextnum might be allocated during
1350          * the previous construction, the buffer including its segusage may
1351          * not be dirty.  The following call ensures that the buffer is dirty
1352          * and will pin the buffer on memory until the sufile is written.
1353          */
1354         err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1355         if (unlikely(err))
1356                 return err;
1357
1358         for (i = 0; i < nadd; i++) {
1359                 /* extend segment info */
1360                 err = -ENOMEM;
1361                 segbuf = nilfs_segbuf_new(sci->sc_super);
1362                 if (unlikely(!segbuf))
1363                         goto failed;
1364
1365                 /* map this buffer to region of segment on-disk */
1366                 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1367                 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1368
1369                 /* allocate the next next full segment */
1370                 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1371                 if (unlikely(err))
1372                         goto failed_segbuf;
1373
1374                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1375                 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1376
1377                 list_add_tail(&segbuf->sb_list, &list);
1378                 prev = segbuf;
1379         }
1380         list_splice_tail(&list, &sci->sc_segbufs);
1381         return 0;
1382
1383  failed_segbuf:
1384         nilfs_segbuf_free(segbuf);
1385  failed:
1386         list_for_each_entry(segbuf, &list, sb_list) {
1387                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1388                 WARN_ON(ret); /* never fails */
1389         }
1390         nilfs_destroy_logs(&list);
1391         return err;
1392 }
1393
1394 static void nilfs_free_incomplete_logs(struct list_head *logs,
1395                                        struct the_nilfs *nilfs)
1396 {
1397         struct nilfs_segment_buffer *segbuf, *prev;
1398         struct inode *sufile = nilfs->ns_sufile;
1399         int ret;
1400
1401         segbuf = NILFS_FIRST_SEGBUF(logs);
1402         if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1403                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1404                 WARN_ON(ret); /* never fails */
1405         }
1406         if (atomic_read(&segbuf->sb_err)) {
1407                 /* Case 1: The first segment failed */
1408                 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1409                         /* Case 1a:  Partial segment appended into an existing
1410                            segment */
1411                         nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1412                                                 segbuf->sb_fseg_end);
1413                 else /* Case 1b:  New full segment */
1414                         set_nilfs_discontinued(nilfs);
1415         }
1416
1417         prev = segbuf;
1418         list_for_each_entry_continue(segbuf, logs, sb_list) {
1419                 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1420                         ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1421                         WARN_ON(ret); /* never fails */
1422                 }
1423                 if (atomic_read(&segbuf->sb_err) &&
1424                     segbuf->sb_segnum != nilfs->ns_nextnum)
1425                         /* Case 2: extended segment (!= next) failed */
1426                         nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1427                 prev = segbuf;
1428         }
1429 }
1430
1431 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1432                                           struct inode *sufile)
1433 {
1434         struct nilfs_segment_buffer *segbuf;
1435         unsigned long live_blocks;
1436         int ret;
1437
1438         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1439                 live_blocks = segbuf->sb_sum.nblocks +
1440                         (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1441                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1442                                                      live_blocks,
1443                                                      sci->sc_seg_ctime);
1444                 WARN_ON(ret); /* always succeed because the segusage is dirty */
1445         }
1446 }
1447
1448 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1449 {
1450         struct nilfs_segment_buffer *segbuf;
1451         int ret;
1452
1453         segbuf = NILFS_FIRST_SEGBUF(logs);
1454         ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1455                                              segbuf->sb_pseg_start -
1456                                              segbuf->sb_fseg_start, 0);
1457         WARN_ON(ret); /* always succeed because the segusage is dirty */
1458
1459         list_for_each_entry_continue(segbuf, logs, sb_list) {
1460                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1461                                                      0, 0);
1462                 WARN_ON(ret); /* always succeed */
1463         }
1464 }
1465
1466 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1467                                             struct nilfs_segment_buffer *last,
1468                                             struct inode *sufile)
1469 {
1470         struct nilfs_segment_buffer *segbuf = last;
1471         int ret;
1472
1473         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1474                 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1475                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1476                 WARN_ON(ret);
1477         }
1478         nilfs_truncate_logs(&sci->sc_segbufs, last);
1479 }
1480
1481
1482 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1483                                  struct the_nilfs *nilfs, int mode)
1484 {
1485         struct nilfs_cstage prev_stage = sci->sc_stage;
1486         int err, nadd = 1;
1487
1488         /* Collection retry loop */
1489         for (;;) {
1490                 sci->sc_nblk_this_inc = 0;
1491                 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1492
1493                 err = nilfs_segctor_reset_segment_buffer(sci);
1494                 if (unlikely(err))
1495                         goto failed;
1496
1497                 err = nilfs_segctor_collect_blocks(sci, mode);
1498                 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1499                 if (!err)
1500                         break;
1501
1502                 if (unlikely(err != -E2BIG))
1503                         goto failed;
1504
1505                 /* The current segment is filled up */
1506                 if (mode != SC_LSEG_SR ||
1507                     nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1508                         break;
1509
1510                 nilfs_clear_logs(&sci->sc_segbufs);
1511
1512                 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1513                         err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1514                                                         sci->sc_freesegs,
1515                                                         sci->sc_nfreesegs,
1516                                                         NULL);
1517                         WARN_ON(err); /* do not happen */
1518                         sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1519                 }
1520
1521                 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1522                 if (unlikely(err))
1523                         return err;
1524
1525                 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1526                 sci->sc_stage = prev_stage;
1527         }
1528         nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1529         return 0;
1530
1531  failed:
1532         return err;
1533 }
1534
1535 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1536                                       struct buffer_head *new_bh)
1537 {
1538         BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1539
1540         list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1541         /* The caller must release old_bh */
1542 }
1543
1544 static int
1545 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1546                                      struct nilfs_segment_buffer *segbuf,
1547                                      int mode)
1548 {
1549         struct inode *inode = NULL;
1550         sector_t blocknr;
1551         unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1552         unsigned long nblocks = 0, ndatablk = 0;
1553         struct nilfs_sc_operations *sc_op = NULL;
1554         struct nilfs_segsum_pointer ssp;
1555         struct nilfs_finfo *finfo = NULL;
1556         union nilfs_binfo binfo;
1557         struct buffer_head *bh, *bh_org;
1558         ino_t ino = 0;
1559         int err = 0;
1560
1561         if (!nfinfo)
1562                 goto out;
1563
1564         blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1565         ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1566         ssp.offset = sizeof(struct nilfs_segment_summary);
1567
1568         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1569                 if (bh == segbuf->sb_super_root)
1570                         break;
1571                 if (!finfo) {
1572                         finfo = nilfs_segctor_map_segsum_entry(
1573                                 sci, &ssp, sizeof(*finfo));
1574                         ino = le64_to_cpu(finfo->fi_ino);
1575                         nblocks = le32_to_cpu(finfo->fi_nblocks);
1576                         ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1577
1578                         inode = bh->b_page->mapping->host;
1579
1580                         if (mode == SC_LSEG_DSYNC)
1581                                 sc_op = &nilfs_sc_dsync_ops;
1582                         else if (ino == NILFS_DAT_INO)
1583                                 sc_op = &nilfs_sc_dat_ops;
1584                         else /* file blocks */
1585                                 sc_op = &nilfs_sc_file_ops;
1586                 }
1587                 bh_org = bh;
1588                 get_bh(bh_org);
1589                 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1590                                         &binfo);
1591                 if (bh != bh_org)
1592                         nilfs_list_replace_buffer(bh_org, bh);
1593                 brelse(bh_org);
1594                 if (unlikely(err))
1595                         goto failed_bmap;
1596
1597                 if (ndatablk > 0)
1598                         sc_op->write_data_binfo(sci, &ssp, &binfo);
1599                 else
1600                         sc_op->write_node_binfo(sci, &ssp, &binfo);
1601
1602                 blocknr++;
1603                 if (--nblocks == 0) {
1604                         finfo = NULL;
1605                         if (--nfinfo == 0)
1606                                 break;
1607                 } else if (ndatablk > 0)
1608                         ndatablk--;
1609         }
1610  out:
1611         return 0;
1612
1613  failed_bmap:
1614         return err;
1615 }
1616
1617 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1618 {
1619         struct nilfs_segment_buffer *segbuf;
1620         int err;
1621
1622         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1623                 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1624                 if (unlikely(err))
1625                         return err;
1626                 nilfs_segbuf_fill_in_segsum(segbuf);
1627         }
1628         return 0;
1629 }
1630
1631 static void nilfs_begin_page_io(struct page *page)
1632 {
1633         if (!page || PageWriteback(page))
1634                 /* For split b-tree node pages, this function may be called
1635                    twice.  We ignore the 2nd or later calls by this check. */
1636                 return;
1637
1638         lock_page(page);
1639         clear_page_dirty_for_io(page);
1640         set_page_writeback(page);
1641         unlock_page(page);
1642 }
1643
1644 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1645 {
1646         struct nilfs_segment_buffer *segbuf;
1647         struct page *bd_page = NULL, *fs_page = NULL;
1648
1649         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1650                 struct buffer_head *bh;
1651
1652                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1653                                     b_assoc_buffers) {
1654                         if (bh->b_page != bd_page) {
1655                                 if (bd_page) {
1656                                         lock_page(bd_page);
1657                                         clear_page_dirty_for_io(bd_page);
1658                                         set_page_writeback(bd_page);
1659                                         unlock_page(bd_page);
1660                                 }
1661                                 bd_page = bh->b_page;
1662                         }
1663                 }
1664
1665                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1666                                     b_assoc_buffers) {
1667                         set_buffer_async_write(bh);
1668                         if (bh == segbuf->sb_super_root) {
1669                                 if (bh->b_page != bd_page) {
1670                                         lock_page(bd_page);
1671                                         clear_page_dirty_for_io(bd_page);
1672                                         set_page_writeback(bd_page);
1673                                         unlock_page(bd_page);
1674                                         bd_page = bh->b_page;
1675                                 }
1676                                 break;
1677                         }
1678                         if (bh->b_page != fs_page) {
1679                                 nilfs_begin_page_io(fs_page);
1680                                 fs_page = bh->b_page;
1681                         }
1682                 }
1683         }
1684         if (bd_page) {
1685                 lock_page(bd_page);
1686                 clear_page_dirty_for_io(bd_page);
1687                 set_page_writeback(bd_page);
1688                 unlock_page(bd_page);
1689         }
1690         nilfs_begin_page_io(fs_page);
1691 }
1692
1693 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1694                                struct the_nilfs *nilfs)
1695 {
1696         int ret;
1697
1698         ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1699         list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1700         return ret;
1701 }
1702
1703 static void nilfs_end_page_io(struct page *page, int err)
1704 {
1705         if (!page)
1706                 return;
1707
1708         if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1709                 /*
1710                  * For b-tree node pages, this function may be called twice
1711                  * or more because they might be split in a segment.
1712                  */
1713                 if (PageDirty(page)) {
1714                         /*
1715                          * For pages holding split b-tree node buffers, dirty
1716                          * flag on the buffers may be cleared discretely.
1717                          * In that case, the page is once redirtied for
1718                          * remaining buffers, and it must be cancelled if
1719                          * all the buffers get cleaned later.
1720                          */
1721                         lock_page(page);
1722                         if (nilfs_page_buffers_clean(page))
1723                                 __nilfs_clear_page_dirty(page);
1724                         unlock_page(page);
1725                 }
1726                 return;
1727         }
1728
1729         if (!err) {
1730                 if (!nilfs_page_buffers_clean(page))
1731                         __set_page_dirty_nobuffers(page);
1732                 ClearPageError(page);
1733         } else {
1734                 __set_page_dirty_nobuffers(page);
1735                 SetPageError(page);
1736         }
1737
1738         end_page_writeback(page);
1739 }
1740
1741 static void nilfs_abort_logs(struct list_head *logs, int err)
1742 {
1743         struct nilfs_segment_buffer *segbuf;
1744         struct page *bd_page = NULL, *fs_page = NULL;
1745         struct buffer_head *bh;
1746
1747         if (list_empty(logs))
1748                 return;
1749
1750         list_for_each_entry(segbuf, logs, sb_list) {
1751                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1752                                     b_assoc_buffers) {
1753                         if (bh->b_page != bd_page) {
1754                                 if (bd_page)
1755                                         end_page_writeback(bd_page);
1756                                 bd_page = bh->b_page;
1757                         }
1758                 }
1759
1760                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1761                                     b_assoc_buffers) {
1762                         clear_buffer_async_write(bh);
1763                         if (bh == segbuf->sb_super_root) {
1764                                 if (bh->b_page != bd_page) {
1765                                         end_page_writeback(bd_page);
1766                                         bd_page = bh->b_page;
1767                                 }
1768                                 break;
1769                         }
1770                         if (bh->b_page != fs_page) {
1771                                 nilfs_end_page_io(fs_page, err);
1772                                 fs_page = bh->b_page;
1773                         }
1774                 }
1775         }
1776         if (bd_page)
1777                 end_page_writeback(bd_page);
1778
1779         nilfs_end_page_io(fs_page, err);
1780 }
1781
1782 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1783                                              struct the_nilfs *nilfs, int err)
1784 {
1785         LIST_HEAD(logs);
1786         int ret;
1787
1788         list_splice_tail_init(&sci->sc_write_logs, &logs);
1789         ret = nilfs_wait_on_logs(&logs);
1790         nilfs_abort_logs(&logs, ret ? : err);
1791
1792         list_splice_tail_init(&sci->sc_segbufs, &logs);
1793         nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1794         nilfs_free_incomplete_logs(&logs, nilfs);
1795
1796         if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1797                 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1798                                                 sci->sc_freesegs,
1799                                                 sci->sc_nfreesegs,
1800                                                 NULL);
1801                 WARN_ON(ret); /* do not happen */
1802         }
1803
1804         nilfs_destroy_logs(&logs);
1805 }
1806
1807 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1808                                    struct nilfs_segment_buffer *segbuf)
1809 {
1810         nilfs->ns_segnum = segbuf->sb_segnum;
1811         nilfs->ns_nextnum = segbuf->sb_nextnum;
1812         nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1813                 + segbuf->sb_sum.nblocks;
1814         nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1815         nilfs->ns_ctime = segbuf->sb_sum.ctime;
1816 }
1817
1818 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1819 {
1820         struct nilfs_segment_buffer *segbuf;
1821         struct page *bd_page = NULL, *fs_page = NULL;
1822         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1823         int update_sr = false;
1824
1825         list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1826                 struct buffer_head *bh;
1827
1828                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1829                                     b_assoc_buffers) {
1830                         set_buffer_uptodate(bh);
1831                         clear_buffer_dirty(bh);
1832                         if (bh->b_page != bd_page) {
1833                                 if (bd_page)
1834                                         end_page_writeback(bd_page);
1835                                 bd_page = bh->b_page;
1836                         }
1837                 }
1838                 /*
1839                  * We assume that the buffers which belong to the same page
1840                  * continue over the buffer list.
1841                  * Under this assumption, the last BHs of pages is
1842                  * identifiable by the discontinuity of bh->b_page
1843                  * (page != fs_page).
1844                  *
1845                  * For B-tree node blocks, however, this assumption is not
1846                  * guaranteed.  The cleanup code of B-tree node pages needs
1847                  * special care.
1848                  */
1849                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1850                                     b_assoc_buffers) {
1851                         const unsigned long set_bits = (1 << BH_Uptodate);
1852                         const unsigned long clear_bits =
1853                                 (1 << BH_Dirty | 1 << BH_Async_Write |
1854                                  1 << BH_Delay | 1 << BH_NILFS_Volatile |
1855                                  1 << BH_NILFS_Redirected);
1856
1857                         set_mask_bits(&bh->b_state, clear_bits, set_bits);
1858                         if (bh == segbuf->sb_super_root) {
1859                                 if (bh->b_page != bd_page) {
1860                                         end_page_writeback(bd_page);
1861                                         bd_page = bh->b_page;
1862                                 }
1863                                 update_sr = true;
1864                                 break;
1865                         }
1866                         if (bh->b_page != fs_page) {
1867                                 nilfs_end_page_io(fs_page, 0);
1868                                 fs_page = bh->b_page;
1869                         }
1870                 }
1871
1872                 if (!nilfs_segbuf_simplex(segbuf)) {
1873                         if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1874                                 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1875                                 sci->sc_lseg_stime = jiffies;
1876                         }
1877                         if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1878                                 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1879                 }
1880         }
1881         /*
1882          * Since pages may continue over multiple segment buffers,
1883          * end of the last page must be checked outside of the loop.
1884          */
1885         if (bd_page)
1886                 end_page_writeback(bd_page);
1887
1888         nilfs_end_page_io(fs_page, 0);
1889
1890         nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1891
1892         if (nilfs_doing_gc())
1893                 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1894         else
1895                 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1896
1897         sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1898
1899         segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1900         nilfs_set_next_segment(nilfs, segbuf);
1901
1902         if (update_sr) {
1903                 nilfs->ns_flushed_device = 0;
1904                 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1905                                        segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1906
1907                 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1908                 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1909                 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1910                 nilfs_segctor_clear_metadata_dirty(sci);
1911         } else
1912                 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1913 }
1914
1915 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1916 {
1917         int ret;
1918
1919         ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1920         if (!ret) {
1921                 nilfs_segctor_complete_write(sci);
1922                 nilfs_destroy_logs(&sci->sc_write_logs);
1923         }
1924         return ret;
1925 }
1926
1927 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1928                                              struct the_nilfs *nilfs)
1929 {
1930         struct nilfs_inode_info *ii, *n;
1931         struct inode *ifile = sci->sc_root->ifile;
1932
1933         spin_lock(&nilfs->ns_inode_lock);
1934  retry:
1935         list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1936                 if (!ii->i_bh) {
1937                         struct buffer_head *ibh;
1938                         int err;
1939
1940                         spin_unlock(&nilfs->ns_inode_lock);
1941                         err = nilfs_ifile_get_inode_block(
1942                                 ifile, ii->vfs_inode.i_ino, &ibh);
1943                         if (unlikely(err)) {
1944                                 nilfs_warning(sci->sc_super, __func__,
1945                                               "failed to get inode block.\n");
1946                                 return err;
1947                         }
1948                         mark_buffer_dirty(ibh);
1949                         nilfs_mdt_mark_dirty(ifile);
1950                         spin_lock(&nilfs->ns_inode_lock);
1951                         if (likely(!ii->i_bh))
1952                                 ii->i_bh = ibh;
1953                         else
1954                                 brelse(ibh);
1955                         goto retry;
1956                 }
1957
1958                 clear_bit(NILFS_I_QUEUED, &ii->i_state);
1959                 set_bit(NILFS_I_BUSY, &ii->i_state);
1960                 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
1961         }
1962         spin_unlock(&nilfs->ns_inode_lock);
1963
1964         return 0;
1965 }
1966
1967 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
1968                                              struct the_nilfs *nilfs)
1969 {
1970         struct nilfs_inode_info *ii, *n;
1971         int during_mount = !(sci->sc_super->s_flags & MS_ACTIVE);
1972         int defer_iput = false;
1973
1974         spin_lock(&nilfs->ns_inode_lock);
1975         list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
1976                 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
1977                     test_bit(NILFS_I_DIRTY, &ii->i_state))
1978                         continue;
1979
1980                 clear_bit(NILFS_I_BUSY, &ii->i_state);
1981                 brelse(ii->i_bh);
1982                 ii->i_bh = NULL;
1983                 list_del_init(&ii->i_dirty);
1984                 if (!ii->vfs_inode.i_nlink || during_mount) {
1985                         /*
1986                          * Defer calling iput() to avoid deadlocks if
1987                          * i_nlink == 0 or mount is not yet finished.
1988                          */
1989                         list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
1990                         defer_iput = true;
1991                 } else {
1992                         spin_unlock(&nilfs->ns_inode_lock);
1993                         iput(&ii->vfs_inode);
1994                         spin_lock(&nilfs->ns_inode_lock);
1995                 }
1996         }
1997         spin_unlock(&nilfs->ns_inode_lock);
1998
1999         if (defer_iput)
2000                 schedule_work(&sci->sc_iput_work);
2001 }
2002
2003 /*
2004  * Main procedure of segment constructor
2005  */
2006 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2007 {
2008         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2009         int err;
2010
2011         nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2012         sci->sc_cno = nilfs->ns_cno;
2013
2014         err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2015         if (unlikely(err))
2016                 goto out;
2017
2018         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2019                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2020
2021         if (nilfs_segctor_clean(sci))
2022                 goto out;
2023
2024         do {
2025                 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2026
2027                 err = nilfs_segctor_begin_construction(sci, nilfs);
2028                 if (unlikely(err))
2029                         goto out;
2030
2031                 /* Update time stamp */
2032                 sci->sc_seg_ctime = get_seconds();
2033
2034                 err = nilfs_segctor_collect(sci, nilfs, mode);
2035                 if (unlikely(err))
2036                         goto failed;
2037
2038                 /* Avoid empty segment */
2039                 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2040                     nilfs_segbuf_empty(sci->sc_curseg)) {
2041                         nilfs_segctor_abort_construction(sci, nilfs, 1);
2042                         goto out;
2043                 }
2044
2045                 err = nilfs_segctor_assign(sci, mode);
2046                 if (unlikely(err))
2047                         goto failed;
2048
2049                 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2050                         nilfs_segctor_fill_in_file_bmap(sci);
2051
2052                 if (mode == SC_LSEG_SR &&
2053                     nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2054                         err = nilfs_segctor_fill_in_checkpoint(sci);
2055                         if (unlikely(err))
2056                                 goto failed_to_write;
2057
2058                         nilfs_segctor_fill_in_super_root(sci, nilfs);
2059                 }
2060                 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2061
2062                 /* Write partial segments */
2063                 nilfs_segctor_prepare_write(sci);
2064
2065                 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2066                                             nilfs->ns_crc_seed);
2067
2068                 err = nilfs_segctor_write(sci, nilfs);
2069                 if (unlikely(err))
2070                         goto failed_to_write;
2071
2072                 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2073                     nilfs->ns_blocksize_bits != PAGE_CACHE_SHIFT) {
2074                         /*
2075                          * At this point, we avoid double buffering
2076                          * for blocksize < pagesize because page dirty
2077                          * flag is turned off during write and dirty
2078                          * buffers are not properly collected for
2079                          * pages crossing over segments.
2080                          */
2081                         err = nilfs_segctor_wait(sci);
2082                         if (err)
2083                                 goto failed_to_write;
2084                 }
2085         } while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2086
2087  out:
2088         nilfs_segctor_drop_written_files(sci, nilfs);
2089         return err;
2090
2091  failed_to_write:
2092         if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2093                 nilfs_redirty_inodes(&sci->sc_dirty_files);
2094
2095  failed:
2096         if (nilfs_doing_gc())
2097                 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2098         nilfs_segctor_abort_construction(sci, nilfs, err);
2099         goto out;
2100 }
2101
2102 /**
2103  * nilfs_segctor_start_timer - set timer of background write
2104  * @sci: nilfs_sc_info
2105  *
2106  * If the timer has already been set, it ignores the new request.
2107  * This function MUST be called within a section locking the segment
2108  * semaphore.
2109  */
2110 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2111 {
2112         spin_lock(&sci->sc_state_lock);
2113         if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2114                 sci->sc_timer.expires = jiffies + sci->sc_interval;
2115                 add_timer(&sci->sc_timer);
2116                 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2117         }
2118         spin_unlock(&sci->sc_state_lock);
2119 }
2120
2121 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2122 {
2123         spin_lock(&sci->sc_state_lock);
2124         if (!(sci->sc_flush_request & (1 << bn))) {
2125                 unsigned long prev_req = sci->sc_flush_request;
2126
2127                 sci->sc_flush_request |= (1 << bn);
2128                 if (!prev_req)
2129                         wake_up(&sci->sc_wait_daemon);
2130         }
2131         spin_unlock(&sci->sc_state_lock);
2132 }
2133
2134 /**
2135  * nilfs_flush_segment - trigger a segment construction for resource control
2136  * @sb: super block
2137  * @ino: inode number of the file to be flushed out.
2138  */
2139 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2140 {
2141         struct the_nilfs *nilfs = sb->s_fs_info;
2142         struct nilfs_sc_info *sci = nilfs->ns_writer;
2143
2144         if (!sci || nilfs_doing_construction())
2145                 return;
2146         nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2147                                         /* assign bit 0 to data files */
2148 }
2149
2150 struct nilfs_segctor_wait_request {
2151         wait_queue_t    wq;
2152         __u32           seq;
2153         int             err;
2154         atomic_t        done;
2155 };
2156
2157 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2158 {
2159         struct nilfs_segctor_wait_request wait_req;
2160         int err = 0;
2161
2162         spin_lock(&sci->sc_state_lock);
2163         init_wait(&wait_req.wq);
2164         wait_req.err = 0;
2165         atomic_set(&wait_req.done, 0);
2166         wait_req.seq = ++sci->sc_seq_request;
2167         spin_unlock(&sci->sc_state_lock);
2168
2169         init_waitqueue_entry(&wait_req.wq, current);
2170         add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2171         set_current_state(TASK_INTERRUPTIBLE);
2172         wake_up(&sci->sc_wait_daemon);
2173
2174         for (;;) {
2175                 if (atomic_read(&wait_req.done)) {
2176                         err = wait_req.err;
2177                         break;
2178                 }
2179                 if (!signal_pending(current)) {
2180                         schedule();
2181                         continue;
2182                 }
2183                 err = -ERESTARTSYS;
2184                 break;
2185         }
2186         finish_wait(&sci->sc_wait_request, &wait_req.wq);
2187         return err;
2188 }
2189
2190 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2191 {
2192         struct nilfs_segctor_wait_request *wrq, *n;
2193         unsigned long flags;
2194
2195         spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2196         list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2197                                  wq.task_list) {
2198                 if (!atomic_read(&wrq->done) &&
2199                     nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2200                         wrq->err = err;
2201                         atomic_set(&wrq->done, 1);
2202                 }
2203                 if (atomic_read(&wrq->done)) {
2204                         wrq->wq.func(&wrq->wq,
2205                                      TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2206                                      0, NULL);
2207                 }
2208         }
2209         spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2210 }
2211
2212 /**
2213  * nilfs_construct_segment - construct a logical segment
2214  * @sb: super block
2215  *
2216  * Return Value: On success, 0 is retured. On errors, one of the following
2217  * negative error code is returned.
2218  *
2219  * %-EROFS - Read only filesystem.
2220  *
2221  * %-EIO - I/O error
2222  *
2223  * %-ENOSPC - No space left on device (only in a panic state).
2224  *
2225  * %-ERESTARTSYS - Interrupted.
2226  *
2227  * %-ENOMEM - Insufficient memory available.
2228  */
2229 int nilfs_construct_segment(struct super_block *sb)
2230 {
2231         struct the_nilfs *nilfs = sb->s_fs_info;
2232         struct nilfs_sc_info *sci = nilfs->ns_writer;
2233         struct nilfs_transaction_info *ti;
2234         int err;
2235
2236         if (!sci)
2237                 return -EROFS;
2238
2239         /* A call inside transactions causes a deadlock. */
2240         BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2241
2242         err = nilfs_segctor_sync(sci);
2243         return err;
2244 }
2245
2246 /**
2247  * nilfs_construct_dsync_segment - construct a data-only logical segment
2248  * @sb: super block
2249  * @inode: inode whose data blocks should be written out
2250  * @start: start byte offset
2251  * @end: end byte offset (inclusive)
2252  *
2253  * Return Value: On success, 0 is retured. On errors, one of the following
2254  * negative error code is returned.
2255  *
2256  * %-EROFS - Read only filesystem.
2257  *
2258  * %-EIO - I/O error
2259  *
2260  * %-ENOSPC - No space left on device (only in a panic state).
2261  *
2262  * %-ERESTARTSYS - Interrupted.
2263  *
2264  * %-ENOMEM - Insufficient memory available.
2265  */
2266 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2267                                   loff_t start, loff_t end)
2268 {
2269         struct the_nilfs *nilfs = sb->s_fs_info;
2270         struct nilfs_sc_info *sci = nilfs->ns_writer;
2271         struct nilfs_inode_info *ii;
2272         struct nilfs_transaction_info ti;
2273         int err = 0;
2274
2275         if (!sci)
2276                 return -EROFS;
2277
2278         nilfs_transaction_lock(sb, &ti, 0);
2279
2280         ii = NILFS_I(inode);
2281         if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2282             nilfs_test_opt(nilfs, STRICT_ORDER) ||
2283             test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2284             nilfs_discontinued(nilfs)) {
2285                 nilfs_transaction_unlock(sb);
2286                 err = nilfs_segctor_sync(sci);
2287                 return err;
2288         }
2289
2290         spin_lock(&nilfs->ns_inode_lock);
2291         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2292             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2293                 spin_unlock(&nilfs->ns_inode_lock);
2294                 nilfs_transaction_unlock(sb);
2295                 return 0;
2296         }
2297         spin_unlock(&nilfs->ns_inode_lock);
2298         sci->sc_dsync_inode = ii;
2299         sci->sc_dsync_start = start;
2300         sci->sc_dsync_end = end;
2301
2302         err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2303         if (!err)
2304                 nilfs->ns_flushed_device = 0;
2305
2306         nilfs_transaction_unlock(sb);
2307         return err;
2308 }
2309
2310 #define FLUSH_FILE_BIT  (0x1) /* data file only */
2311 #define FLUSH_DAT_BIT   (1 << NILFS_DAT_INO) /* DAT only */
2312
2313 /**
2314  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2315  * @sci: segment constructor object
2316  */
2317 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2318 {
2319         spin_lock(&sci->sc_state_lock);
2320         sci->sc_seq_accepted = sci->sc_seq_request;
2321         spin_unlock(&sci->sc_state_lock);
2322         del_timer_sync(&sci->sc_timer);
2323 }
2324
2325 /**
2326  * nilfs_segctor_notify - notify the result of request to caller threads
2327  * @sci: segment constructor object
2328  * @mode: mode of log forming
2329  * @err: error code to be notified
2330  */
2331 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2332 {
2333         /* Clear requests (even when the construction failed) */
2334         spin_lock(&sci->sc_state_lock);
2335
2336         if (mode == SC_LSEG_SR) {
2337                 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2338                 sci->sc_seq_done = sci->sc_seq_accepted;
2339                 nilfs_segctor_wakeup(sci, err);
2340                 sci->sc_flush_request = 0;
2341         } else {
2342                 if (mode == SC_FLUSH_FILE)
2343                         sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2344                 else if (mode == SC_FLUSH_DAT)
2345                         sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2346
2347                 /* re-enable timer if checkpoint creation was not done */
2348                 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2349                     time_before(jiffies, sci->sc_timer.expires))
2350                         add_timer(&sci->sc_timer);
2351         }
2352         spin_unlock(&sci->sc_state_lock);
2353 }
2354
2355 /**
2356  * nilfs_segctor_construct - form logs and write them to disk
2357  * @sci: segment constructor object
2358  * @mode: mode of log forming
2359  */
2360 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2361 {
2362         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2363         struct nilfs_super_block **sbp;
2364         int err = 0;
2365
2366         nilfs_segctor_accept(sci);
2367
2368         if (nilfs_discontinued(nilfs))
2369                 mode = SC_LSEG_SR;
2370         if (!nilfs_segctor_confirm(sci))
2371                 err = nilfs_segctor_do_construct(sci, mode);
2372
2373         if (likely(!err)) {
2374                 if (mode != SC_FLUSH_DAT)
2375                         atomic_set(&nilfs->ns_ndirtyblks, 0);
2376                 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2377                     nilfs_discontinued(nilfs)) {
2378                         down_write(&nilfs->ns_sem);
2379                         err = -EIO;
2380                         sbp = nilfs_prepare_super(sci->sc_super,
2381                                                   nilfs_sb_will_flip(nilfs));
2382                         if (likely(sbp)) {
2383                                 nilfs_set_log_cursor(sbp[0], nilfs);
2384                                 err = nilfs_commit_super(sci->sc_super,
2385                                                          NILFS_SB_COMMIT);
2386                         }
2387                         up_write(&nilfs->ns_sem);
2388                 }
2389         }
2390
2391         nilfs_segctor_notify(sci, mode, err);
2392         return err;
2393 }
2394
2395 static void nilfs_construction_timeout(unsigned long data)
2396 {
2397         struct task_struct *p = (struct task_struct *)data;
2398         wake_up_process(p);
2399 }
2400
2401 static void
2402 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2403 {
2404         struct nilfs_inode_info *ii, *n;
2405
2406         list_for_each_entry_safe(ii, n, head, i_dirty) {
2407                 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2408                         continue;
2409                 list_del_init(&ii->i_dirty);
2410                 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2411                 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
2412                 iput(&ii->vfs_inode);
2413         }
2414 }
2415
2416 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2417                          void **kbufs)
2418 {
2419         struct the_nilfs *nilfs = sb->s_fs_info;
2420         struct nilfs_sc_info *sci = nilfs->ns_writer;
2421         struct nilfs_transaction_info ti;
2422         int err;
2423
2424         if (unlikely(!sci))
2425                 return -EROFS;
2426
2427         nilfs_transaction_lock(sb, &ti, 1);
2428
2429         err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2430         if (unlikely(err))
2431                 goto out_unlock;
2432
2433         err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2434         if (unlikely(err)) {
2435                 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2436                 goto out_unlock;
2437         }
2438
2439         sci->sc_freesegs = kbufs[4];
2440         sci->sc_nfreesegs = argv[4].v_nmembs;
2441         list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2442
2443         for (;;) {
2444                 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2445                 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2446
2447                 if (likely(!err))
2448                         break;
2449
2450                 nilfs_warning(sb, __func__,
2451                               "segment construction failed. (err=%d)", err);
2452                 set_current_state(TASK_INTERRUPTIBLE);
2453                 schedule_timeout(sci->sc_interval);
2454         }
2455         if (nilfs_test_opt(nilfs, DISCARD)) {
2456                 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2457                                                  sci->sc_nfreesegs);
2458                 if (ret) {
2459                         printk(KERN_WARNING
2460                                "NILFS warning: error %d on discard request, "
2461                                "turning discards off for the device\n", ret);
2462                         nilfs_clear_opt(nilfs, DISCARD);
2463                 }
2464         }
2465
2466  out_unlock:
2467         sci->sc_freesegs = NULL;
2468         sci->sc_nfreesegs = 0;
2469         nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2470         nilfs_transaction_unlock(sb);
2471         return err;
2472 }
2473
2474 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2475 {
2476         struct nilfs_transaction_info ti;
2477
2478         nilfs_transaction_lock(sci->sc_super, &ti, 0);
2479         nilfs_segctor_construct(sci, mode);
2480
2481         /*
2482          * Unclosed segment should be retried.  We do this using sc_timer.
2483          * Timeout of sc_timer will invoke complete construction which leads
2484          * to close the current logical segment.
2485          */
2486         if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2487                 nilfs_segctor_start_timer(sci);
2488
2489         nilfs_transaction_unlock(sci->sc_super);
2490 }
2491
2492 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2493 {
2494         int mode = 0;
2495
2496         spin_lock(&sci->sc_state_lock);
2497         mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2498                 SC_FLUSH_DAT : SC_FLUSH_FILE;
2499         spin_unlock(&sci->sc_state_lock);
2500
2501         if (mode) {
2502                 nilfs_segctor_do_construct(sci, mode);
2503
2504                 spin_lock(&sci->sc_state_lock);
2505                 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2506                         ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2507                 spin_unlock(&sci->sc_state_lock);
2508         }
2509         clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2510 }
2511
2512 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2513 {
2514         if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2515             time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2516                 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2517                         return SC_FLUSH_FILE;
2518                 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2519                         return SC_FLUSH_DAT;
2520         }
2521         return SC_LSEG_SR;
2522 }
2523
2524 /**
2525  * nilfs_segctor_thread - main loop of the segment constructor thread.
2526  * @arg: pointer to a struct nilfs_sc_info.
2527  *
2528  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2529  * to execute segment constructions.
2530  */
2531 static int nilfs_segctor_thread(void *arg)
2532 {
2533         struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2534         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2535         int timeout = 0;
2536
2537         sci->sc_timer.data = (unsigned long)current;
2538         sci->sc_timer.function = nilfs_construction_timeout;
2539
2540         /* start sync. */
2541         sci->sc_task = current;
2542         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2543         printk(KERN_INFO
2544                "segctord starting. Construction interval = %lu seconds, "
2545                "CP frequency < %lu seconds\n",
2546                sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2547
2548         spin_lock(&sci->sc_state_lock);
2549  loop:
2550         for (;;) {
2551                 int mode;
2552
2553                 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2554                         goto end_thread;
2555
2556                 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2557                         mode = SC_LSEG_SR;
2558                 else if (!sci->sc_flush_request)
2559                         break;
2560                 else
2561                         mode = nilfs_segctor_flush_mode(sci);
2562
2563                 spin_unlock(&sci->sc_state_lock);
2564                 nilfs_segctor_thread_construct(sci, mode);
2565                 spin_lock(&sci->sc_state_lock);
2566                 timeout = 0;
2567         }
2568
2569
2570         if (freezing(current)) {
2571                 spin_unlock(&sci->sc_state_lock);
2572                 try_to_freeze();
2573                 spin_lock(&sci->sc_state_lock);
2574         } else {
2575                 DEFINE_WAIT(wait);
2576                 int should_sleep = 1;
2577
2578                 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2579                                 TASK_INTERRUPTIBLE);
2580
2581                 if (sci->sc_seq_request != sci->sc_seq_done)
2582                         should_sleep = 0;
2583                 else if (sci->sc_flush_request)
2584                         should_sleep = 0;
2585                 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2586                         should_sleep = time_before(jiffies,
2587                                         sci->sc_timer.expires);
2588
2589                 if (should_sleep) {
2590                         spin_unlock(&sci->sc_state_lock);
2591                         schedule();
2592                         spin_lock(&sci->sc_state_lock);
2593                 }
2594                 finish_wait(&sci->sc_wait_daemon, &wait);
2595                 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2596                            time_after_eq(jiffies, sci->sc_timer.expires));
2597
2598                 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2599                         set_nilfs_discontinued(nilfs);
2600         }
2601         goto loop;
2602
2603  end_thread:
2604         spin_unlock(&sci->sc_state_lock);
2605
2606         /* end sync. */
2607         sci->sc_task = NULL;
2608         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2609         return 0;
2610 }
2611
2612 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2613 {
2614         struct task_struct *t;
2615
2616         t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2617         if (IS_ERR(t)) {
2618                 int err = PTR_ERR(t);
2619
2620                 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2621                        err);
2622                 return err;
2623         }
2624         wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2625         return 0;
2626 }
2627
2628 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2629         __acquires(&sci->sc_state_lock)
2630         __releases(&sci->sc_state_lock)
2631 {
2632         sci->sc_state |= NILFS_SEGCTOR_QUIT;
2633
2634         while (sci->sc_task) {
2635                 wake_up(&sci->sc_wait_daemon);
2636                 spin_unlock(&sci->sc_state_lock);
2637                 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2638                 spin_lock(&sci->sc_state_lock);
2639         }
2640 }
2641
2642 /*
2643  * Setup & clean-up functions
2644  */
2645 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2646                                                struct nilfs_root *root)
2647 {
2648         struct the_nilfs *nilfs = sb->s_fs_info;
2649         struct nilfs_sc_info *sci;
2650
2651         sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2652         if (!sci)
2653                 return NULL;
2654
2655         sci->sc_super = sb;
2656
2657         nilfs_get_root(root);
2658         sci->sc_root = root;
2659
2660         init_waitqueue_head(&sci->sc_wait_request);
2661         init_waitqueue_head(&sci->sc_wait_daemon);
2662         init_waitqueue_head(&sci->sc_wait_task);
2663         spin_lock_init(&sci->sc_state_lock);
2664         INIT_LIST_HEAD(&sci->sc_dirty_files);
2665         INIT_LIST_HEAD(&sci->sc_segbufs);
2666         INIT_LIST_HEAD(&sci->sc_write_logs);
2667         INIT_LIST_HEAD(&sci->sc_gc_inodes);
2668         INIT_LIST_HEAD(&sci->sc_iput_queue);
2669         INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2670         init_timer(&sci->sc_timer);
2671
2672         sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2673         sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2674         sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2675
2676         if (nilfs->ns_interval)
2677                 sci->sc_interval = HZ * nilfs->ns_interval;
2678         if (nilfs->ns_watermark)
2679                 sci->sc_watermark = nilfs->ns_watermark;
2680         return sci;
2681 }
2682
2683 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2684 {
2685         int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2686
2687         /* The segctord thread was stopped and its timer was removed.
2688            But some tasks remain. */
2689         do {
2690                 struct nilfs_transaction_info ti;
2691
2692                 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2693                 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2694                 nilfs_transaction_unlock(sci->sc_super);
2695
2696                 flush_work(&sci->sc_iput_work);
2697
2698         } while (ret && retrycount-- > 0);
2699 }
2700
2701 /**
2702  * nilfs_segctor_destroy - destroy the segment constructor.
2703  * @sci: nilfs_sc_info
2704  *
2705  * nilfs_segctor_destroy() kills the segctord thread and frees
2706  * the nilfs_sc_info struct.
2707  * Caller must hold the segment semaphore.
2708  */
2709 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2710 {
2711         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2712         int flag;
2713
2714         up_write(&nilfs->ns_segctor_sem);
2715
2716         spin_lock(&sci->sc_state_lock);
2717         nilfs_segctor_kill_thread(sci);
2718         flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2719                 || sci->sc_seq_request != sci->sc_seq_done);
2720         spin_unlock(&sci->sc_state_lock);
2721
2722         if (flush_work(&sci->sc_iput_work))
2723                 flag = true;
2724
2725         if (flag || !nilfs_segctor_confirm(sci))
2726                 nilfs_segctor_write_out(sci);
2727
2728         if (!list_empty(&sci->sc_dirty_files)) {
2729                 nilfs_warning(sci->sc_super, __func__,
2730                               "dirty file(s) after the final construction\n");
2731                 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2732         }
2733
2734         if (!list_empty(&sci->sc_iput_queue)) {
2735                 nilfs_warning(sci->sc_super, __func__,
2736                               "iput queue is not empty\n");
2737                 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2738         }
2739
2740         WARN_ON(!list_empty(&sci->sc_segbufs));
2741         WARN_ON(!list_empty(&sci->sc_write_logs));
2742
2743         nilfs_put_root(sci->sc_root);
2744
2745         down_write(&nilfs->ns_segctor_sem);
2746
2747         del_timer_sync(&sci->sc_timer);
2748         kfree(sci);
2749 }
2750
2751 /**
2752  * nilfs_attach_log_writer - attach log writer
2753  * @sb: super block instance
2754  * @root: root object of the current filesystem tree
2755  *
2756  * This allocates a log writer object, initializes it, and starts the
2757  * log writer.
2758  *
2759  * Return Value: On success, 0 is returned. On error, one of the following
2760  * negative error code is returned.
2761  *
2762  * %-ENOMEM - Insufficient memory available.
2763  */
2764 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2765 {
2766         struct the_nilfs *nilfs = sb->s_fs_info;
2767         int err;
2768
2769         if (nilfs->ns_writer) {
2770                 /*
2771                  * This happens if the filesystem was remounted
2772                  * read/write after nilfs_error degenerated it into a
2773                  * read-only mount.
2774                  */
2775                 nilfs_detach_log_writer(sb);
2776         }
2777
2778         nilfs->ns_writer = nilfs_segctor_new(sb, root);
2779         if (!nilfs->ns_writer)
2780                 return -ENOMEM;
2781
2782         err = nilfs_segctor_start_thread(nilfs->ns_writer);
2783         if (err) {
2784                 kfree(nilfs->ns_writer);
2785                 nilfs->ns_writer = NULL;
2786         }
2787         return err;
2788 }
2789
2790 /**
2791  * nilfs_detach_log_writer - destroy log writer
2792  * @sb: super block instance
2793  *
2794  * This kills log writer daemon, frees the log writer object, and
2795  * destroys list of dirty files.
2796  */
2797 void nilfs_detach_log_writer(struct super_block *sb)
2798 {
2799         struct the_nilfs *nilfs = sb->s_fs_info;
2800         LIST_HEAD(garbage_list);
2801
2802         down_write(&nilfs->ns_segctor_sem);
2803         if (nilfs->ns_writer) {
2804                 nilfs_segctor_destroy(nilfs->ns_writer);
2805                 nilfs->ns_writer = NULL;
2806         }
2807
2808         /* Force to free the list of dirty files */
2809         spin_lock(&nilfs->ns_inode_lock);
2810         if (!list_empty(&nilfs->ns_dirty_files)) {
2811                 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2812                 nilfs_warning(sb, __func__,
2813                               "Hit dirty file after stopped log writer\n");
2814         }
2815         spin_unlock(&nilfs->ns_inode_lock);
2816         up_write(&nilfs->ns_segctor_sem);
2817
2818         nilfs_dispose_list(nilfs, &garbage_list, 1);
2819 }