fs/nilfs2/segment.c

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