fs/nilfs2/segment.c

   1 /*
   2  * segment.c - NILFS segment constructor.
   3  *
   4  * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License as published by
   8  * the Free Software Foundation; either version 2 of the License, or
   9  * (at your option) any later version.
  10  *
  11  * This program is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  14  * GNU General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  19  *
  20  * Written by Ryusuke Konishi <ryusuke@osrg.net>
  21  *
  22  */
  23
  24 #include <linux/pagemap.h>
  25 #include <linux/buffer_head.h>
  26 #include <linux/writeback.h>
  27 #include <linux/bio.h>
  28 #include <linux/completion.h>
  29 #include <linux/blkdev.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/freezer.h>
  32 #include <linux/kthread.h>
  33 #include <linux/crc32.h>
  34 #include <linux/pagevec.h>
  35 #include <linux/slab.h>
  36 #include "nilfs.h"
  37 #include "btnode.h"
  38 #include "page.h"
  39 #include "segment.h"
  40 #include "sufile.h"
  41 #include "cpfile.h"
  42 #include "ifile.h"
  43 #include "segbuf.h"
  44
  45
  46 /*
  47  * Segment constructor
  48  */
  49 #define SC_N_INODEVEC   16   /* Size of locally allocated inode vector */
  50
  51 #define SC_MAX_SEGDELTA 64   /* Upper limit of the number of segments
  52                                 appended in collection retry loop */
  53
  54 /* Construction mode */
  55 enum {
  56         SC_LSEG_SR = 1, /* Make a logical segment having a super root */
  57         SC_LSEG_DSYNC,  /* Flush data blocks of a given file and make
  58                            a logical segment without a super root */
  59         SC_FLUSH_FILE,  /* Flush data files, leads to segment writes without
  60                            creating a checkpoint */
  61         SC_FLUSH_DAT,   /* Flush DAT file. This also creates segments without
  62                            a checkpoint */
  63 };
  64
  65 /* Stage numbers of dirty block collection */
  66 enum {
  67         NILFS_ST_INIT = 0,
  68         NILFS_ST_GC,            /* Collecting dirty blocks for GC */
  69         NILFS_ST_FILE,
  70         NILFS_ST_IFILE,
  71         NILFS_ST_CPFILE,
  72         NILFS_ST_SUFILE,
  73         NILFS_ST_DAT,
  74         NILFS_ST_SR,            /* Super root */
  75         NILFS_ST_DSYNC,         /* Data sync blocks */
  76         NILFS_ST_DONE,
  77 };
  78
  79 /* State flags of collection */
  80 #define NILFS_CF_NODE           0x0001  /* Collecting node blocks */
  81 #define NILFS_CF_IFILE_STARTED  0x0002  /* IFILE stage has started */
  82 #define NILFS_CF_SUFREED        0x0004  /* segment usages has been freed */
  83 #define NILFS_CF_HISTORY_MASK   (NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
  84
  85 /* Operations depending on the construction mode and file type */
  86 struct nilfs_sc_operations {
  87         int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
  88                             struct inode *);
  89         int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
  90                             struct inode *);
  91         int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
  92                             struct inode *);
  93         void (*write_data_binfo)(struct nilfs_sc_info *,
  94                                  struct nilfs_segsum_pointer *,
  95                                  union nilfs_binfo *);
  96         void (*write_node_binfo)(struct nilfs_sc_info *,
  97                                  struct nilfs_segsum_pointer *,
  98                                  union nilfs_binfo *);
  99 };
 100
 101 /*
 102  * Other definitions
 103  */
 104 static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
 105 static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
 106 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
 107 static void nilfs_dispose_list(struct nilfs_sb_info *, struct list_head *,
 108                                int);
 109
 110 #define nilfs_cnt32_gt(a, b)   \
 111         (typecheck(__u32, a) && typecheck(__u32, b) && \
 112          ((__s32)(b) - (__s32)(a) < 0))
 113 #define nilfs_cnt32_ge(a, b)   \
 114         (typecheck(__u32, a) && typecheck(__u32, b) && \
 115          ((__s32)(a) - (__s32)(b) >= 0))
 116 #define nilfs_cnt32_lt(a, b)  nilfs_cnt32_gt(b, a)
 117 #define nilfs_cnt32_le(a, b)  nilfs_cnt32_ge(b, a)
 118
 119 /*
 120  * Transaction
 121  */
 122 static struct kmem_cache *nilfs_transaction_cachep;
 123
 124 /**
 125  * nilfs_init_transaction_cache - create a cache for nilfs_transaction_info
 126  *
 127  * nilfs_init_transaction_cache() creates a slab cache for the struct
 128  * nilfs_transaction_info.
 129  *
 130  * Return Value: On success, it returns 0. On error, one of the following
 131  * negative error code is returned.
 132  *
 133  * %-ENOMEM - Insufficient memory available.
 134  */
 135 int nilfs_init_transaction_cache(void)
 136 {
 137         nilfs_transaction_cachep =
 138                 kmem_cache_create("nilfs2_transaction_cache",
 139                                   sizeof(struct nilfs_transaction_info),
 140                                   0, SLAB_RECLAIM_ACCOUNT, NULL);
 141         return (nilfs_transaction_cachep == NULL) ? -ENOMEM : 0;
 142 }
 143
 144 /**
 145  * nilfs_destroy_transaction_cache - destroy the cache for transaction info
 146  *
 147  * nilfs_destroy_transaction_cache() frees the slab cache for the struct
 148  * nilfs_transaction_info.
 149  */
 150 void nilfs_destroy_transaction_cache(void)
 151 {
 152         kmem_cache_destroy(nilfs_transaction_cachep);
 153 }
 154
 155 static int nilfs_prepare_segment_lock(struct nilfs_transaction_info *ti)
 156 {
 157         struct nilfs_transaction_info *cur_ti = current->journal_info;
 158         void *save = NULL;
 159
 160         if (cur_ti) {
 161                 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
 162                         return ++cur_ti->ti_count;
 163                 else {
 164                         /*
 165                          * If journal_info field is occupied by other FS,
 166                          * it is saved and will be restored on
 167                          * nilfs_transaction_commit().
 168                          */
 169                         printk(KERN_WARNING
 170                                "NILFS warning: journal info from a different "
 171                                "FS\n");
 172                         save = current->journal_info;
 173                 }
 174         }
 175         if (!ti) {
 176                 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
 177                 if (!ti)
 178                         return -ENOMEM;
 179                 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
 180         } else {
 181                 ti->ti_flags = 0;
 182         }
 183         ti->ti_count = 0;
 184         ti->ti_save = save;
 185         ti->ti_magic = NILFS_TI_MAGIC;
 186         current->journal_info = ti;
 187         return 0;
 188 }
 189
 190 /**
 191  * nilfs_transaction_begin - start indivisible file operations.
 192  * @sb: super block
 193  * @ti: nilfs_transaction_info
 194  * @vacancy_check: flags for vacancy rate checks
 195  *
 196  * nilfs_transaction_begin() acquires a reader/writer semaphore, called
 197  * the segment semaphore, to make a segment construction and write tasks
 198  * exclusive.  The function is used with nilfs_transaction_commit() in pairs.
 199  * The region enclosed by these two functions can be nested.  To avoid a
 200  * deadlock, the semaphore is only acquired or released in the outermost call.
 201  *
 202  * This function allocates a nilfs_transaction_info struct to keep context
 203  * information on it.  It is initialized and hooked onto the current task in
 204  * the outermost call.  If a pre-allocated struct is given to @ti, it is used
 205  * instead; otherwise a new struct is assigned from a slab.
 206  *
 207  * When @vacancy_check flag is set, this function will check the amount of
 208  * free space, and will wait for the GC to reclaim disk space if low capacity.
 209  *
 210  * Return Value: On success, 0 is returned. On error, one of the following
 211  * negative error code is returned.
 212  *
 213  * %-ENOMEM - Insufficient memory available.
 214  *
 215  * %-ENOSPC - No space left on device
 216  */
 217 int nilfs_transaction_begin(struct super_block *sb,
 218                             struct nilfs_transaction_info *ti,
 219                             int vacancy_check)
 220 {
 221         struct nilfs_sb_info *sbi;
 222         struct the_nilfs *nilfs;
 223         int ret = nilfs_prepare_segment_lock(ti);
 224
 225         if (unlikely(ret < 0))
 226                 return ret;
 227         if (ret > 0)
 228                 return 0;
 229
 230         sbi = NILFS_SB(sb);
 231         nilfs = sbi->s_nilfs;
 232         down_read(&nilfs->ns_segctor_sem);
 233         if (vacancy_check && nilfs_near_disk_full(nilfs)) {
 234                 up_read(&nilfs->ns_segctor_sem);
 235                 ret = -ENOSPC;
 236                 goto failed;
 237         }
 238         return 0;
 239
 240  failed:
 241         ti = current->journal_info;
 242         current->journal_info = ti->ti_save;
 243         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 244                 kmem_cache_free(nilfs_transaction_cachep, ti);
 245         return ret;
 246 }
 247
 248 /**
 249  * nilfs_transaction_commit - commit indivisible file operations.
 250  * @sb: super block
 251  *
 252  * nilfs_transaction_commit() releases the read semaphore which is
 253  * acquired by nilfs_transaction_begin(). This is only performed
 254  * in outermost call of this function.  If a commit flag is set,
 255  * nilfs_transaction_commit() sets a timer to start the segment
 256  * constructor.  If a sync flag is set, it starts construction
 257  * directly.
 258  */
 259 int nilfs_transaction_commit(struct super_block *sb)
 260 {
 261         struct nilfs_transaction_info *ti = current->journal_info;
 262         struct nilfs_sb_info *sbi;
 263         struct nilfs_sc_info *sci;
 264         int err = 0;
 265
 266         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 267         ti->ti_flags |= NILFS_TI_COMMIT;
 268         if (ti->ti_count > 0) {
 269                 ti->ti_count--;
 270                 return 0;
 271         }
 272         sbi = NILFS_SB(sb);
 273         sci = NILFS_SC(sbi);
 274         if (sci != NULL) {
 275                 if (ti->ti_flags & NILFS_TI_COMMIT)
 276                         nilfs_segctor_start_timer(sci);
 277                 if (atomic_read(&sbi->s_nilfs->ns_ndirtyblks) >
 278                     sci->sc_watermark)
 279                         nilfs_segctor_do_flush(sci, 0);
 280         }
 281         up_read(&sbi->s_nilfs->ns_segctor_sem);
 282         current->journal_info = ti->ti_save;
 283
 284         if (ti->ti_flags & NILFS_TI_SYNC)
 285                 err = nilfs_construct_segment(sb);
 286         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 287                 kmem_cache_free(nilfs_transaction_cachep, ti);
 288         return err;
 289 }
 290
 291 void nilfs_transaction_abort(struct super_block *sb)
 292 {
 293         struct nilfs_transaction_info *ti = current->journal_info;
 294
 295         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 296         if (ti->ti_count > 0) {
 297                 ti->ti_count--;
 298                 return;
 299         }
 300         up_read(&NILFS_SB(sb)->s_nilfs->ns_segctor_sem);
 301
 302         current->journal_info = ti->ti_save;
 303         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 304                 kmem_cache_free(nilfs_transaction_cachep, ti);
 305 }
 306
 307 void nilfs_relax_pressure_in_lock(struct super_block *sb)
 308 {
 309         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 310         struct nilfs_sc_info *sci = NILFS_SC(sbi);
 311         struct the_nilfs *nilfs = sbi->s_nilfs;
 312
 313         if (!sci || !sci->sc_flush_request)
 314                 return;
 315
 316         set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
 317         up_read(&nilfs->ns_segctor_sem);
 318
 319         down_write(&nilfs->ns_segctor_sem);
 320         if (sci->sc_flush_request &&
 321             test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
 322                 struct nilfs_transaction_info *ti = current->journal_info;
 323
 324                 ti->ti_flags |= NILFS_TI_WRITER;
 325                 nilfs_segctor_do_immediate_flush(sci);
 326                 ti->ti_flags &= ~NILFS_TI_WRITER;
 327         }
 328         downgrade_write(&nilfs->ns_segctor_sem);
 329 }
 330
 331 static void nilfs_transaction_lock(struct nilfs_sb_info *sbi,
 332                                    struct nilfs_transaction_info *ti,
 333                                    int gcflag)
 334 {
 335         struct nilfs_transaction_info *cur_ti = current->journal_info;
 336
 337         WARN_ON(cur_ti);
 338         ti->ti_flags = NILFS_TI_WRITER;
 339         ti->ti_count = 0;
 340         ti->ti_save = cur_ti;
 341         ti->ti_magic = NILFS_TI_MAGIC;
 342         INIT_LIST_HEAD(&ti->ti_garbage);
 343         current->journal_info = ti;
 344
 345         for (;;) {
 346                 down_write(&sbi->s_nilfs->ns_segctor_sem);
 347                 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &NILFS_SC(sbi)->sc_flags))
 348                         break;
 349
 350                 nilfs_segctor_do_immediate_flush(NILFS_SC(sbi));
 351
 352                 up_write(&sbi->s_nilfs->ns_segctor_sem);
 353                 yield();
 354         }
 355         if (gcflag)
 356                 ti->ti_flags |= NILFS_TI_GC;
 357 }
 358
 359 static void nilfs_transaction_unlock(struct nilfs_sb_info *sbi)
 360 {
 361         struct nilfs_transaction_info *ti = current->journal_info;
 362
 363         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 364         BUG_ON(ti->ti_count > 0);
 365
 366         up_write(&sbi->s_nilfs->ns_segctor_sem);
 367         current->journal_info = ti->ti_save;
 368         if (!list_empty(&ti->ti_garbage))
 369                 nilfs_dispose_list(sbi, &ti->ti_garbage, 0);
 370 }
 371
 372 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
 373                                             struct nilfs_segsum_pointer *ssp,
 374                                             unsigned bytes)
 375 {
 376         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 377         unsigned blocksize = sci->sc_super->s_blocksize;
 378         void *p;
 379
 380         if (unlikely(ssp->offset + bytes > blocksize)) {
 381                 ssp->offset = 0;
 382                 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
 383                                                &segbuf->sb_segsum_buffers));
 384                 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
 385         }
 386         p = ssp->bh->b_data + ssp->offset;
 387         ssp->offset += bytes;
 388         return p;
 389 }
 390
 391 /**
 392  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
 393  * @sci: nilfs_sc_info
 394  */
 395 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
 396 {
 397         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 398         struct buffer_head *sumbh;
 399         unsigned sumbytes;
 400         unsigned flags = 0;
 401         int err;
 402
 403         if (nilfs_doing_gc())
 404                 flags = NILFS_SS_GC;
 405         err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime);
 406         if (unlikely(err))
 407                 return err;
 408
 409         sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
 410         sumbytes = segbuf->sb_sum.sumbytes;
 411         sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
 412         sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
 413         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 414         return 0;
 415 }
 416
 417 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
 418 {
 419         sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
 420         if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
 421                 return -E2BIG; /* The current segment is filled up
 422                                   (internal code) */
 423         sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
 424         return nilfs_segctor_reset_segment_buffer(sci);
 425 }
 426
 427 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
 428 {
 429         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 430         int err;
 431
 432         if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
 433                 err = nilfs_segctor_feed_segment(sci);
 434                 if (err)
 435                         return err;
 436                 segbuf = sci->sc_curseg;
 437         }
 438         err = nilfs_segbuf_extend_payload(segbuf, &sci->sc_super_root);
 439         if (likely(!err))
 440                 segbuf->sb_sum.flags |= NILFS_SS_SR;
 441         return err;
 442 }
 443
 444 /*
 445  * Functions for making segment summary and payloads
 446  */
 447 static int nilfs_segctor_segsum_block_required(
 448         struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
 449         unsigned binfo_size)
 450 {
 451         unsigned blocksize = sci->sc_super->s_blocksize;
 452         /* Size of finfo and binfo is enough small against blocksize */
 453
 454         return ssp->offset + binfo_size +
 455                 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
 456                 blocksize;
 457 }
 458
 459 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
 460                                       struct inode *inode)
 461 {
 462         sci->sc_curseg->sb_sum.nfinfo++;
 463         sci->sc_binfo_ptr = sci->sc_finfo_ptr;
 464         nilfs_segctor_map_segsum_entry(
 465                 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
 466
 467         if (inode->i_sb && !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 468                 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
 469         /* skip finfo */
 470 }
 471
 472 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
 473                                     struct inode *inode)
 474 {
 475         struct nilfs_finfo *finfo;
 476         struct nilfs_inode_info *ii;
 477         struct nilfs_segment_buffer *segbuf;
 478
 479         if (sci->sc_blk_cnt == 0)
 480                 return;
 481
 482         ii = NILFS_I(inode);
 483         finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
 484                                                  sizeof(*finfo));
 485         finfo->fi_ino = cpu_to_le64(inode->i_ino);
 486         finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
 487         finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
 488         finfo->fi_cno = cpu_to_le64(ii->i_cno);
 489
 490         segbuf = sci->sc_curseg;
 491         segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
 492                 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
 493         sci->sc_finfo_ptr = sci->sc_binfo_ptr;
 494         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 495 }
 496
 497 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
 498                                         struct buffer_head *bh,
 499                                         struct inode *inode,
 500                                         unsigned binfo_size)
 501 {
 502         struct nilfs_segment_buffer *segbuf;
 503         int required, err = 0;
 504
 505  retry:
 506         segbuf = sci->sc_curseg;
 507         required = nilfs_segctor_segsum_block_required(
 508                 sci, &sci->sc_binfo_ptr, binfo_size);
 509         if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
 510                 nilfs_segctor_end_finfo(sci, inode);
 511                 err = nilfs_segctor_feed_segment(sci);
 512                 if (err)
 513                         return err;
 514                 goto retry;
 515         }
 516         if (unlikely(required)) {
 517                 err = nilfs_segbuf_extend_segsum(segbuf);
 518                 if (unlikely(err))
 519                         goto failed;
 520         }
 521         if (sci->sc_blk_cnt == 0)
 522                 nilfs_segctor_begin_finfo(sci, inode);
 523
 524         nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
 525         /* Substitution to vblocknr is delayed until update_blocknr() */
 526         nilfs_segbuf_add_file_buffer(segbuf, bh);
 527         sci->sc_blk_cnt++;
 528  failed:
 529         return err;
 530 }
 531
 532 static int nilfs_handle_bmap_error(int err, const char *fname,
 533                                    struct inode *inode, struct super_block *sb)
 534 {
 535         if (err == -EINVAL) {
 536                 nilfs_error(sb, fname, "broken bmap (inode=%lu)\n",
 537                             inode->i_ino);
 538                 err = -EIO;
 539         }
 540         return err;
 541 }
 542
 543 /*
 544  * Callback functions that enumerate, mark, and collect dirty blocks
 545  */
 546 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
 547                                    struct buffer_head *bh, struct inode *inode)
 548 {
 549         int err;
 550
 551         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 552         if (unlikely(err < 0))
 553                 return nilfs_handle_bmap_error(err, __func__, inode,
 554                                                sci->sc_super);
 555
 556         err = nilfs_segctor_add_file_block(sci, bh, inode,
 557                                            sizeof(struct nilfs_binfo_v));
 558         if (!err)
 559                 sci->sc_datablk_cnt++;
 560         return err;
 561 }
 562
 563 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
 564                                    struct buffer_head *bh,
 565                                    struct inode *inode)
 566 {
 567         int err;
 568
 569         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 570         if (unlikely(err < 0))
 571                 return nilfs_handle_bmap_error(err, __func__, inode,
 572                                                sci->sc_super);
 573         return 0;
 574 }
 575
 576 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
 577                                    struct buffer_head *bh,
 578                                    struct inode *inode)
 579 {
 580         WARN_ON(!buffer_dirty(bh));
 581         return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 582 }
 583
 584 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
 585                                         struct nilfs_segsum_pointer *ssp,
 586                                         union nilfs_binfo *binfo)
 587 {
 588         struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
 589                 sci, ssp, sizeof(*binfo_v));
 590         *binfo_v = binfo->bi_v;
 591 }
 592
 593 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
 594                                         struct nilfs_segsum_pointer *ssp,
 595                                         union nilfs_binfo *binfo)
 596 {
 597         __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
 598                 sci, ssp, sizeof(*vblocknr));
 599         *vblocknr = binfo->bi_v.bi_vblocknr;
 600 }
 601
 602 struct nilfs_sc_operations nilfs_sc_file_ops = {
 603         .collect_data = nilfs_collect_file_data,
 604         .collect_node = nilfs_collect_file_node,
 605         .collect_bmap = nilfs_collect_file_bmap,
 606         .write_data_binfo = nilfs_write_file_data_binfo,
 607         .write_node_binfo = nilfs_write_file_node_binfo,
 608 };
 609
 610 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
 611                                   struct buffer_head *bh, struct inode *inode)
 612 {
 613         int err;
 614
 615         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 616         if (unlikely(err < 0))
 617                 return nilfs_handle_bmap_error(err, __func__, inode,
 618                                                sci->sc_super);
 619
 620         err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 621         if (!err)
 622                 sci->sc_datablk_cnt++;
 623         return err;
 624 }
 625
 626 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
 627                                   struct buffer_head *bh, struct inode *inode)
 628 {
 629         WARN_ON(!buffer_dirty(bh));
 630         return nilfs_segctor_add_file_block(sci, bh, inode,
 631                                             sizeof(struct nilfs_binfo_dat));
 632 }
 633
 634 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
 635                                        struct nilfs_segsum_pointer *ssp,
 636                                        union nilfs_binfo *binfo)
 637 {
 638         __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
 639                                                           sizeof(*blkoff));
 640         *blkoff = binfo->bi_dat.bi_blkoff;
 641 }
 642
 643 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
 644                                        struct nilfs_segsum_pointer *ssp,
 645                                        union nilfs_binfo *binfo)
 646 {
 647         struct nilfs_binfo_dat *binfo_dat =
 648                 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
 649         *binfo_dat = binfo->bi_dat;
 650 }
 651
 652 struct nilfs_sc_operations nilfs_sc_dat_ops = {
 653         .collect_data = nilfs_collect_dat_data,
 654         .collect_node = nilfs_collect_file_node,
 655         .collect_bmap = nilfs_collect_dat_bmap,
 656         .write_data_binfo = nilfs_write_dat_data_binfo,
 657         .write_node_binfo = nilfs_write_dat_node_binfo,
 658 };
 659
 660 struct nilfs_sc_operations nilfs_sc_dsync_ops = {
 661         .collect_data = nilfs_collect_file_data,
 662         .collect_node = NULL,
 663         .collect_bmap = NULL,
 664         .write_data_binfo = nilfs_write_file_data_binfo,
 665         .write_node_binfo = NULL,
 666 };
 667
 668 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
 669                                               struct list_head *listp,
 670                                               size_t nlimit,
 671                                               loff_t start, loff_t end)
 672 {
 673         struct address_space *mapping = inode->i_mapping;
 674         struct pagevec pvec;
 675         pgoff_t index = 0, last = ULONG_MAX;
 676         size_t ndirties = 0;
 677         int i;
 678
 679         if (unlikely(start != 0 || end != LLONG_MAX)) {
 680                 /*
 681                  * A valid range is given for sync-ing data pages. The
 682                  * range is rounded to per-page; extra dirty buffers
 683                  * may be included if blocksize < pagesize.
 684                  */
 685                 index = start >> PAGE_SHIFT;
 686                 last = end >> PAGE_SHIFT;
 687         }
 688         pagevec_init(&pvec, 0);
 689  repeat:
 690         if (unlikely(index > last) ||
 691             !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 692                                 min_t(pgoff_t, last - index,
 693                                       PAGEVEC_SIZE - 1) + 1))
 694                 return ndirties;
 695
 696         for (i = 0; i < pagevec_count(&pvec); i++) {
 697                 struct buffer_head *bh, *head;
 698                 struct page *page = pvec.pages[i];
 699
 700                 if (unlikely(page->index > last))
 701                         break;
 702
 703                 if (mapping->host) {
 704                         lock_page(page);
 705                         if (!page_has_buffers(page))
 706                                 create_empty_buffers(page,
 707                                                      1 << inode->i_blkbits, 0);
 708                         unlock_page(page);
 709                 }
 710
 711                 bh = head = page_buffers(page);
 712                 do {
 713                         if (!buffer_dirty(bh))
 714                                 continue;
 715                         get_bh(bh);
 716                         list_add_tail(&bh->b_assoc_buffers, listp);
 717                         ndirties++;
 718                         if (unlikely(ndirties >= nlimit)) {
 719                                 pagevec_release(&pvec);
 720                                 cond_resched();
 721                                 return ndirties;
 722                         }
 723                 } while (bh = bh->b_this_page, bh != head);
 724         }
 725         pagevec_release(&pvec);
 726         cond_resched();
 727         goto repeat;
 728 }
 729
 730 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
 731                                             struct list_head *listp)
 732 {
 733         struct nilfs_inode_info *ii = NILFS_I(inode);
 734         struct address_space *mapping = &ii->i_btnode_cache;
 735         struct pagevec pvec;
 736         struct buffer_head *bh, *head;
 737         unsigned int i;
 738         pgoff_t index = 0;
 739
 740         pagevec_init(&pvec, 0);
 741
 742         while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 743                                   PAGEVEC_SIZE)) {
 744                 for (i = 0; i < pagevec_count(&pvec); i++) {
 745                         bh = head = page_buffers(pvec.pages[i]);
 746                         do {
 747                                 if (buffer_dirty(bh)) {
 748                                         get_bh(bh);
 749                                         list_add_tail(&bh->b_assoc_buffers,
 750                                                       listp);
 751                                 }
 752                                 bh = bh->b_this_page;
 753                         } while (bh != head);
 754                 }
 755                 pagevec_release(&pvec);
 756                 cond_resched();
 757         }
 758 }
 759
 760 static void nilfs_dispose_list(struct nilfs_sb_info *sbi,
 761                                struct list_head *head, int force)
 762 {
 763         struct nilfs_inode_info *ii, *n;
 764         struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
 765         unsigned nv = 0;
 766
 767         while (!list_empty(head)) {
 768                 spin_lock(&sbi->s_inode_lock);
 769                 list_for_each_entry_safe(ii, n, head, i_dirty) {
 770                         list_del_init(&ii->i_dirty);
 771                         if (force) {
 772                                 if (unlikely(ii->i_bh)) {
 773                                         brelse(ii->i_bh);
 774                                         ii->i_bh = NULL;
 775                                 }
 776                         } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
 777                                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 778                                 list_add_tail(&ii->i_dirty,
 779                                               &sbi->s_dirty_files);
 780                                 continue;
 781                         }
 782                         ivec[nv++] = ii;
 783                         if (nv == SC_N_INODEVEC)
 784                                 break;
 785                 }
 786                 spin_unlock(&sbi->s_inode_lock);
 787
 788                 for (pii = ivec; nv > 0; pii++, nv--)
 789                         iput(&(*pii)->vfs_inode);
 790         }
 791 }
 792
 793 static int nilfs_test_metadata_dirty(struct nilfs_sb_info *sbi)
 794 {
 795         struct the_nilfs *nilfs = sbi->s_nilfs;
 796         int ret = 0;
 797
 798         if (nilfs_mdt_fetch_dirty(sbi->s_ifile))
 799                 ret++;
 800         if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
 801                 ret++;
 802         if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
 803                 ret++;
 804         if (ret || nilfs_doing_gc())
 805                 if (nilfs_mdt_fetch_dirty(nilfs_dat_inode(nilfs)))
 806                         ret++;
 807         return ret;
 808 }
 809
 810 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
 811 {
 812         return list_empty(&sci->sc_dirty_files) &&
 813                 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
 814                 sci->sc_nfreesegs == 0 &&
 815                 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
 816 }
 817
 818 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
 819 {
 820         struct nilfs_sb_info *sbi = sci->sc_sbi;
 821         int ret = 0;
 822
 823         if (nilfs_test_metadata_dirty(sbi))
 824                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
 825
 826         spin_lock(&sbi->s_inode_lock);
 827         if (list_empty(&sbi->s_dirty_files) && nilfs_segctor_clean(sci))
 828                 ret++;
 829
 830         spin_unlock(&sbi->s_inode_lock);
 831         return ret;
 832 }
 833
 834 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
 835 {
 836         struct nilfs_sb_info *sbi = sci->sc_sbi;
 837         struct the_nilfs *nilfs = sbi->s_nilfs;
 838
 839         nilfs_mdt_clear_dirty(sbi->s_ifile);
 840         nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
 841         nilfs_mdt_clear_dirty(nilfs->ns_sufile);
 842         nilfs_mdt_clear_dirty(nilfs_dat_inode(nilfs));
 843 }
 844
 845 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
 846 {
 847         struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
 848         struct buffer_head *bh_cp;
 849         struct nilfs_checkpoint *raw_cp;
 850         int err;
 851
 852         /* XXX: this interface will be changed */
 853         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
 854                                           &raw_cp, &bh_cp);
 855         if (likely(!err)) {
 856                 /* The following code is duplicated with cpfile.  But, it is
 857                    needed to collect the checkpoint even if it was not newly
 858                    created */
 859                 nilfs_mdt_mark_buffer_dirty(bh_cp);
 860                 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
 861                 nilfs_cpfile_put_checkpoint(
 862                         nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 863         } else
 864                 WARN_ON(err == -EINVAL || err == -ENOENT);
 865
 866         return err;
 867 }
 868
 869 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
 870 {
 871         struct nilfs_sb_info *sbi = sci->sc_sbi;
 872         struct the_nilfs *nilfs = sbi->s_nilfs;
 873         struct buffer_head *bh_cp;
 874         struct nilfs_checkpoint *raw_cp;
 875         int err;
 876
 877         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
 878                                           &raw_cp, &bh_cp);
 879         if (unlikely(err)) {
 880                 WARN_ON(err == -EINVAL || err == -ENOENT);
 881                 goto failed_ibh;
 882         }
 883         raw_cp->cp_snapshot_list.ssl_next = 0;
 884         raw_cp->cp_snapshot_list.ssl_prev = 0;
 885         raw_cp->cp_inodes_count =
 886                 cpu_to_le64(atomic_read(&sbi->s_inodes_count));
 887         raw_cp->cp_blocks_count =
 888                 cpu_to_le64(atomic_read(&sbi->s_blocks_count));
 889         raw_cp->cp_nblk_inc =
 890                 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
 891         raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
 892         raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
 893
 894         if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 895                 nilfs_checkpoint_clear_minor(raw_cp);
 896         else
 897                 nilfs_checkpoint_set_minor(raw_cp);
 898
 899         nilfs_write_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode, 1);
 900         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 901         return 0;
 902
 903  failed_ibh:
 904         return err;
 905 }
 906
 907 static void nilfs_fill_in_file_bmap(struct inode *ifile,
 908                                     struct nilfs_inode_info *ii)
 909
 910 {
 911         struct buffer_head *ibh;
 912         struct nilfs_inode *raw_inode;
 913
 914         if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
 915                 ibh = ii->i_bh;
 916                 BUG_ON(!ibh);
 917                 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
 918                                                   ibh);
 919                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 920                 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
 921         }
 922 }
 923
 924 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci,
 925                                             struct inode *ifile)
 926 {
 927         struct nilfs_inode_info *ii;
 928
 929         list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
 930                 nilfs_fill_in_file_bmap(ifile, ii);
 931                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
 932         }
 933 }
 934
 935 /*
 936  * CRC calculation routines
 937  */
 938 static void nilfs_fill_in_super_root_crc(struct buffer_head *bh_sr, u32 seed)
 939 {
 940         struct nilfs_super_root *raw_sr =
 941                 (struct nilfs_super_root *)bh_sr->b_data;
 942         u32 crc;
 943
 944         crc = crc32_le(seed,
 945                        (unsigned char *)raw_sr + sizeof(raw_sr->sr_sum),
 946                        NILFS_SR_BYTES - sizeof(raw_sr->sr_sum));
 947         raw_sr->sr_sum = cpu_to_le32(crc);
 948 }
 949
 950 static void nilfs_segctor_fill_in_checksums(struct nilfs_sc_info *sci,
 951                                             u32 seed)
 952 {
 953         struct nilfs_segment_buffer *segbuf;
 954
 955         if (sci->sc_super_root)
 956                 nilfs_fill_in_super_root_crc(sci->sc_super_root, seed);
 957
 958         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
 959                 nilfs_segbuf_fill_in_segsum_crc(segbuf, seed);
 960                 nilfs_segbuf_fill_in_data_crc(segbuf, seed);
 961         }
 962 }
 963
 964 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
 965                                              struct the_nilfs *nilfs)
 966 {
 967         struct buffer_head *bh_sr = sci->sc_super_root;
 968         struct nilfs_super_root *raw_sr =
 969                 (struct nilfs_super_root *)bh_sr->b_data;
 970         unsigned isz = nilfs->ns_inode_size;
 971
 972         raw_sr->sr_bytes = cpu_to_le16(NILFS_SR_BYTES);
 973         raw_sr->sr_nongc_ctime
 974                 = cpu_to_le64(nilfs_doing_gc() ?
 975                               nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
 976         raw_sr->sr_flags = 0;
 977
 978         nilfs_write_inode_common(nilfs_dat_inode(nilfs), (void *)raw_sr +
 979                                  NILFS_SR_DAT_OFFSET(isz), 1);
 980         nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
 981                                  NILFS_SR_CPFILE_OFFSET(isz), 1);
 982         nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
 983                                  NILFS_SR_SUFILE_OFFSET(isz), 1);
 984 }
 985
 986 static void nilfs_redirty_inodes(struct list_head *head)
 987 {
 988         struct nilfs_inode_info *ii;
 989
 990         list_for_each_entry(ii, head, i_dirty) {
 991                 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
 992                         clear_bit(NILFS_I_COLLECTED, &ii->i_state);
 993         }
 994 }
 995
 996 static void nilfs_drop_collected_inodes(struct list_head *head)
 997 {
 998         struct nilfs_inode_info *ii;
 999
1000         list_for_each_entry(ii, head, i_dirty) {
1001                 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1002                         continue;
1003
1004                 clear_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
1005                 set_bit(NILFS_I_UPDATED, &ii->i_state);
1006         }
1007 }
1008
1009 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1010                                        struct inode *inode,
1011                                        struct list_head *listp,
1012                                        int (*collect)(struct nilfs_sc_info *,
1013                                                       struct buffer_head *,
1014                                                       struct inode *))
1015 {
1016         struct buffer_head *bh, *n;
1017         int err = 0;
1018
1019         if (collect) {
1020                 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1021                         list_del_init(&bh->b_assoc_buffers);
1022                         err = collect(sci, bh, inode);
1023                         brelse(bh);
1024                         if (unlikely(err))
1025                                 goto dispose_buffers;
1026                 }
1027                 return 0;
1028         }
1029
1030  dispose_buffers:
1031         while (!list_empty(listp)) {
1032                 bh = list_entry(listp->next, struct buffer_head,
1033                                 b_assoc_buffers);
1034                 list_del_init(&bh->b_assoc_buffers);
1035                 brelse(bh);
1036         }
1037         return err;
1038 }
1039
1040 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1041 {
1042         /* Remaining number of blocks within segment buffer */
1043         return sci->sc_segbuf_nblocks -
1044                 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1045 }
1046
1047 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1048                                    struct inode *inode,
1049                                    struct nilfs_sc_operations *sc_ops)
1050 {
1051         LIST_HEAD(data_buffers);
1052         LIST_HEAD(node_buffers);
1053         int err;
1054
1055         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1056                 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1057
1058                 n = nilfs_lookup_dirty_data_buffers(
1059                         inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1060                 if (n > rest) {
1061                         err = nilfs_segctor_apply_buffers(
1062                                 sci, inode, &data_buffers,
1063                                 sc_ops->collect_data);
1064                         BUG_ON(!err); /* always receive -E2BIG or true error */
1065                         goto break_or_fail;
1066                 }
1067         }
1068         nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1069
1070         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1071                 err = nilfs_segctor_apply_buffers(
1072                         sci, inode, &data_buffers, sc_ops->collect_data);
1073                 if (unlikely(err)) {
1074                         /* dispose node list */
1075                         nilfs_segctor_apply_buffers(
1076                                 sci, inode, &node_buffers, NULL);
1077                         goto break_or_fail;
1078                 }
1079                 sci->sc_stage.flags |= NILFS_CF_NODE;
1080         }
1081         /* Collect node */
1082         err = nilfs_segctor_apply_buffers(
1083                 sci, inode, &node_buffers, sc_ops->collect_node);
1084         if (unlikely(err))
1085                 goto break_or_fail;
1086
1087         nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1088         err = nilfs_segctor_apply_buffers(
1089                 sci, inode, &node_buffers, sc_ops->collect_bmap);
1090         if (unlikely(err))
1091                 goto break_or_fail;
1092
1093         nilfs_segctor_end_finfo(sci, inode);
1094         sci->sc_stage.flags &= ~NILFS_CF_NODE;
1095
1096  break_or_fail:
1097         return err;
1098 }
1099
1100 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1101                                          struct inode *inode)
1102 {
1103         LIST_HEAD(data_buffers);
1104         size_t n, rest = nilfs_segctor_buffer_rest(sci);
1105         int err;
1106
1107         n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1108                                             sci->sc_dsync_start,
1109                                             sci->sc_dsync_end);
1110
1111         err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1112                                           nilfs_collect_file_data);
1113         if (!err) {
1114                 nilfs_segctor_end_finfo(sci, inode);
1115                 BUG_ON(n > rest);
1116                 /* always receive -E2BIG or true error if n > rest */
1117         }
1118         return err;
1119 }
1120
1121 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1122 {
1123         struct nilfs_sb_info *sbi = sci->sc_sbi;
1124         struct the_nilfs *nilfs = sbi->s_nilfs;
1125         struct list_head *head;
1126         struct nilfs_inode_info *ii;
1127         size_t ndone;
1128         int err = 0;
1129
1130         switch (sci->sc_stage.scnt) {
1131         case NILFS_ST_INIT:
1132                 /* Pre-processes */
1133                 sci->sc_stage.flags = 0;
1134
1135                 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1136                         sci->sc_nblk_inc = 0;
1137                         sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1138                         if (mode == SC_LSEG_DSYNC) {
1139                                 sci->sc_stage.scnt = NILFS_ST_DSYNC;
1140                                 goto dsync_mode;
1141                         }
1142                 }
1143
1144                 sci->sc_stage.dirty_file_ptr = NULL;
1145                 sci->sc_stage.gc_inode_ptr = NULL;
1146                 if (mode == SC_FLUSH_DAT) {
1147                         sci->sc_stage.scnt = NILFS_ST_DAT;
1148                         goto dat_stage;
1149                 }
1150                 sci->sc_stage.scnt++;  /* Fall through */
1151         case NILFS_ST_GC:
1152                 if (nilfs_doing_gc()) {
1153                         head = &sci->sc_gc_inodes;
1154                         ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1155                                                 head, i_dirty);
1156                         list_for_each_entry_continue(ii, head, i_dirty) {
1157                                 err = nilfs_segctor_scan_file(
1158                                         sci, &ii->vfs_inode,
1159                                         &nilfs_sc_file_ops);
1160                                 if (unlikely(err)) {
1161                                         sci->sc_stage.gc_inode_ptr = list_entry(
1162                                                 ii->i_dirty.prev,
1163                                                 struct nilfs_inode_info,
1164                                                 i_dirty);
1165                                         goto break_or_fail;
1166                                 }
1167                                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1168                         }
1169                         sci->sc_stage.gc_inode_ptr = NULL;
1170                 }
1171                 sci->sc_stage.scnt++;  /* Fall through */
1172         case NILFS_ST_FILE:
1173                 head = &sci->sc_dirty_files;
1174                 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1175                                         i_dirty);
1176                 list_for_each_entry_continue(ii, head, i_dirty) {
1177                         clear_bit(NILFS_I_DIRTY, &ii->i_state);
1178
1179                         err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1180                                                       &nilfs_sc_file_ops);
1181                         if (unlikely(err)) {
1182                                 sci->sc_stage.dirty_file_ptr =
1183                                         list_entry(ii->i_dirty.prev,
1184                                                    struct nilfs_inode_info,
1185                                                    i_dirty);
1186                                 goto break_or_fail;
1187                         }
1188                         /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1189                         /* XXX: required ? */
1190                 }
1191                 sci->sc_stage.dirty_file_ptr = NULL;
1192                 if (mode == SC_FLUSH_FILE) {
1193                         sci->sc_stage.scnt = NILFS_ST_DONE;
1194                         return 0;
1195                 }
1196                 sci->sc_stage.scnt++;
1197                 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1198                 /* Fall through */
1199         case NILFS_ST_IFILE:
1200                 err = nilfs_segctor_scan_file(sci, sbi->s_ifile,
1201                                               &nilfs_sc_file_ops);
1202                 if (unlikely(err))
1203                         break;
1204                 sci->sc_stage.scnt++;
1205                 /* Creating a checkpoint */
1206                 err = nilfs_segctor_create_checkpoint(sci);
1207                 if (unlikely(err))
1208                         break;
1209                 /* Fall through */
1210         case NILFS_ST_CPFILE:
1211                 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1212                                               &nilfs_sc_file_ops);
1213                 if (unlikely(err))
1214                         break;
1215                 sci->sc_stage.scnt++;  /* Fall through */
1216         case NILFS_ST_SUFILE:
1217                 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1218                                          sci->sc_nfreesegs, &ndone);
1219                 if (unlikely(err)) {
1220                         nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1221                                                   sci->sc_freesegs, ndone,
1222                                                   NULL);
1223                         break;
1224                 }
1225                 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1226
1227                 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1228                                               &nilfs_sc_file_ops);
1229                 if (unlikely(err))
1230                         break;
1231                 sci->sc_stage.scnt++;  /* Fall through */
1232         case NILFS_ST_DAT:
1233  dat_stage:
1234                 err = nilfs_segctor_scan_file(sci, nilfs_dat_inode(nilfs),
1235                                               &nilfs_sc_dat_ops);
1236                 if (unlikely(err))
1237                         break;
1238                 if (mode == SC_FLUSH_DAT) {
1239                         sci->sc_stage.scnt = NILFS_ST_DONE;
1240                         return 0;
1241                 }
1242                 sci->sc_stage.scnt++;  /* Fall through */
1243         case NILFS_ST_SR:
1244                 if (mode == SC_LSEG_SR) {
1245                         /* Appending a super root */
1246                         err = nilfs_segctor_add_super_root(sci);
1247                         if (unlikely(err))
1248                                 break;
1249                 }
1250                 /* End of a logical segment */
1251                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1252                 sci->sc_stage.scnt = NILFS_ST_DONE;
1253                 return 0;
1254         case NILFS_ST_DSYNC:
1255  dsync_mode:
1256                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1257                 ii = sci->sc_dsync_inode;
1258                 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1259                         break;
1260
1261                 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1262                 if (unlikely(err))
1263                         break;
1264                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1265                 sci->sc_stage.scnt = NILFS_ST_DONE;
1266                 return 0;
1267         case NILFS_ST_DONE:
1268                 return 0;
1269         default:
1270                 BUG();
1271         }
1272
1273  break_or_fail:
1274         return err;
1275 }
1276
1277 /**
1278  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1279  * @sci: nilfs_sc_info
1280  * @nilfs: nilfs object
1281  */
1282 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1283                                             struct the_nilfs *nilfs)
1284 {
1285         struct nilfs_segment_buffer *segbuf, *prev;
1286         __u64 nextnum;
1287         int err, alloc = 0;
1288
1289         segbuf = nilfs_segbuf_new(sci->sc_super);
1290         if (unlikely(!segbuf))
1291                 return -ENOMEM;
1292
1293         if (list_empty(&sci->sc_write_logs)) {
1294                 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1295                                  nilfs->ns_pseg_offset, nilfs);
1296                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1297                         nilfs_shift_to_next_segment(nilfs);
1298                         nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1299                 }
1300
1301                 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1302                 nextnum = nilfs->ns_nextnum;
1303
1304                 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1305                         /* Start from the head of a new full segment */
1306                         alloc++;
1307         } else {
1308                 /* Continue logs */
1309                 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1310                 nilfs_segbuf_map_cont(segbuf, prev);
1311                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1312                 nextnum = prev->sb_nextnum;
1313
1314                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1315                         nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1316                         segbuf->sb_sum.seg_seq++;
1317                         alloc++;
1318                 }
1319         }
1320
1321         err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1322         if (err)
1323                 goto failed;
1324
1325         if (alloc) {
1326                 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1327                 if (err)
1328                         goto failed;
1329         }
1330         nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1331
1332         BUG_ON(!list_empty(&sci->sc_segbufs));
1333         list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1334         sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1335         return 0;
1336
1337  failed:
1338         nilfs_segbuf_free(segbuf);
1339         return err;
1340 }
1341
1342 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1343                                          struct the_nilfs *nilfs, int nadd)
1344 {
1345         struct nilfs_segment_buffer *segbuf, *prev;
1346         struct inode *sufile = nilfs->ns_sufile;
1347         __u64 nextnextnum;
1348         LIST_HEAD(list);
1349         int err, ret, i;
1350
1351         prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1352         /*
1353          * Since the segment specified with nextnum might be allocated during
1354          * the previous construction, the buffer including its segusage may
1355          * not be dirty.  The following call ensures that the buffer is dirty
1356          * and will pin the buffer on memory until the sufile is written.
1357          */
1358         err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1359         if (unlikely(err))
1360                 return err;
1361
1362         for (i = 0; i < nadd; i++) {
1363                 /* extend segment info */
1364                 err = -ENOMEM;
1365                 segbuf = nilfs_segbuf_new(sci->sc_super);
1366                 if (unlikely(!segbuf))
1367                         goto failed;
1368
1369                 /* map this buffer to region of segment on-disk */
1370                 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1371                 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1372
1373                 /* allocate the next next full segment */
1374                 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1375                 if (unlikely(err))
1376                         goto failed_segbuf;
1377
1378                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1379                 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1380
1381                 list_add_tail(&segbuf->sb_list, &list);
1382                 prev = segbuf;
1383         }
1384         list_splice_tail(&list, &sci->sc_segbufs);
1385         return 0;
1386
1387  failed_segbuf:
1388         nilfs_segbuf_free(segbuf);
1389  failed:
1390         list_for_each_entry(segbuf, &list, sb_list) {
1391                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1392                 WARN_ON(ret); /* never fails */
1393         }
1394         nilfs_destroy_logs(&list);
1395         return err;
1396 }
1397
1398 static void nilfs_free_incomplete_logs(struct list_head *logs,
1399                                        struct the_nilfs *nilfs)
1400 {
1401         struct nilfs_segment_buffer *segbuf, *prev;
1402         struct inode *sufile = nilfs->ns_sufile;
1403         int ret;
1404
1405         segbuf = NILFS_FIRST_SEGBUF(logs);
1406         if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1407                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1408                 WARN_ON(ret); /* never fails */
1409         }
1410         if (atomic_read(&segbuf->sb_err)) {
1411                 /* Case 1: The first segment failed */
1412                 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1413                         /* Case 1a:  Partial segment appended into an existing
1414                            segment */
1415                         nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1416                                                 segbuf->sb_fseg_end);
1417                 else /* Case 1b:  New full segment */
1418                         set_nilfs_discontinued(nilfs);
1419         }
1420
1421         prev = segbuf;
1422         list_for_each_entry_continue(segbuf, logs, sb_list) {
1423                 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1424                         ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1425                         WARN_ON(ret); /* never fails */
1426                 }
1427                 if (atomic_read(&segbuf->sb_err) &&
1428                     segbuf->sb_segnum != nilfs->ns_nextnum)
1429                         /* Case 2: extended segment (!= next) failed */
1430                         nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1431                 prev = segbuf;
1432         }
1433 }
1434
1435 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1436                                           struct inode *sufile)
1437 {
1438         struct nilfs_segment_buffer *segbuf;
1439         unsigned long live_blocks;
1440         int ret;
1441
1442         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1443                 live_blocks = segbuf->sb_sum.nblocks +
1444                         (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1445                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1446                                                      live_blocks,
1447                                                      sci->sc_seg_ctime);
1448                 WARN_ON(ret); /* always succeed because the segusage is dirty */
1449         }
1450 }
1451
1452 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1453 {
1454         struct nilfs_segment_buffer *segbuf;
1455         int ret;
1456
1457         segbuf = NILFS_FIRST_SEGBUF(logs);
1458         ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1459                                              segbuf->sb_pseg_start -
1460                                              segbuf->sb_fseg_start, 0);
1461         WARN_ON(ret); /* always succeed because the segusage is dirty */
1462
1463         list_for_each_entry_continue(segbuf, logs, sb_list) {
1464                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1465                                                      0, 0);
1466                 WARN_ON(ret); /* always succeed */
1467         }
1468 }
1469
1470 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1471                                             struct nilfs_segment_buffer *last,
1472                                             struct inode *sufile)
1473 {
1474         struct nilfs_segment_buffer *segbuf = last;
1475         int ret;
1476
1477         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1478                 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1479                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1480                 WARN_ON(ret);
1481         }
1482         nilfs_truncate_logs(&sci->sc_segbufs, last);
1483 }
1484
1485
1486 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1487                                  struct the_nilfs *nilfs, int mode)
1488 {
1489         struct nilfs_cstage prev_stage = sci->sc_stage;
1490         int err, nadd = 1;
1491
1492         /* Collection retry loop */
1493         for (;;) {
1494                 sci->sc_super_root = NULL;
1495                 sci->sc_nblk_this_inc = 0;
1496                 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1497
1498                 err = nilfs_segctor_reset_segment_buffer(sci);
1499                 if (unlikely(err))
1500                         goto failed;
1501
1502                 err = nilfs_segctor_collect_blocks(sci, mode);
1503                 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1504                 if (!err)
1505                         break;
1506
1507                 if (unlikely(err != -E2BIG))
1508                         goto failed;
1509
1510                 /* The current segment is filled up */
1511                 if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
1512                         break;
1513
1514                 nilfs_clear_logs(&sci->sc_segbufs);
1515
1516                 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1517                 if (unlikely(err))
1518                         return err;
1519
1520                 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1521                         err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1522                                                         sci->sc_freesegs,
1523                                                         sci->sc_nfreesegs,
1524                                                         NULL);
1525                         WARN_ON(err); /* do not happen */
1526                 }
1527                 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1528                 sci->sc_stage = prev_stage;
1529         }
1530         nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1531         return 0;
1532
1533  failed:
1534         return err;
1535 }
1536
1537 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1538                                       struct buffer_head *new_bh)
1539 {
1540         BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1541
1542         list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1543         /* The caller must release old_bh */
1544 }
1545
1546 static int
1547 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1548                                      struct nilfs_segment_buffer *segbuf,
1549                                      int mode)
1550 {
1551         struct inode *inode = NULL;
1552         sector_t blocknr;
1553         unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1554         unsigned long nblocks = 0, ndatablk = 0;
1555         struct nilfs_sc_operations *sc_op = NULL;
1556         struct nilfs_segsum_pointer ssp;
1557         struct nilfs_finfo *finfo = NULL;
1558         union nilfs_binfo binfo;
1559         struct buffer_head *bh, *bh_org;
1560         ino_t ino = 0;
1561         int err = 0;
1562
1563         if (!nfinfo)
1564                 goto out;
1565
1566         blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1567         ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1568         ssp.offset = sizeof(struct nilfs_segment_summary);
1569
1570         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1571                 if (bh == sci->sc_super_root)
1572                         break;
1573                 if (!finfo) {
1574                         finfo = nilfs_segctor_map_segsum_entry(
1575                                 sci, &ssp, sizeof(*finfo));
1576                         ino = le64_to_cpu(finfo->fi_ino);
1577                         nblocks = le32_to_cpu(finfo->fi_nblocks);
1578                         ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1579
1580                         if (buffer_nilfs_node(bh))
1581                                 inode = NILFS_BTNC_I(bh->b_page->mapping);
1582                         else
1583                                 inode = NILFS_AS_I(bh->b_page->mapping);
1584
1585                         if (mode == SC_LSEG_DSYNC)
1586                                 sc_op = &nilfs_sc_dsync_ops;
1587                         else if (ino == NILFS_DAT_INO)
1588                                 sc_op = &nilfs_sc_dat_ops;
1589                         else /* file blocks */
1590                                 sc_op = &nilfs_sc_file_ops;
1591                 }
1592                 bh_org = bh;
1593                 get_bh(bh_org);
1594                 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1595                                         &binfo);
1596                 if (bh != bh_org)
1597                         nilfs_list_replace_buffer(bh_org, bh);
1598                 brelse(bh_org);
1599                 if (unlikely(err))
1600                         goto failed_bmap;
1601
1602                 if (ndatablk > 0)
1603                         sc_op->write_data_binfo(sci, &ssp, &binfo);
1604                 else
1605                         sc_op->write_node_binfo(sci, &ssp, &binfo);
1606
1607                 blocknr++;
1608                 if (--nblocks == 0) {
1609                         finfo = NULL;
1610                         if (--nfinfo == 0)
1611                                 break;
1612                 } else if (ndatablk > 0)
1613                         ndatablk--;
1614         }
1615  out:
1616         return 0;
1617
1618  failed_bmap:
1619         err = nilfs_handle_bmap_error(err, __func__, inode, sci->sc_super);
1620         return err;
1621 }
1622
1623 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1624 {
1625         struct nilfs_segment_buffer *segbuf;
1626         int err;
1627
1628         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1629                 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1630                 if (unlikely(err))
1631                         return err;
1632                 nilfs_segbuf_fill_in_segsum(segbuf);
1633         }
1634         return 0;
1635 }
1636
1637 static int
1638 nilfs_copy_replace_page_buffers(struct page *page, struct list_head *out)
1639 {
1640         struct page *clone_page;
1641         struct buffer_head *bh, *head, *bh2;
1642         void *kaddr;
1643
1644         bh = head = page_buffers(page);
1645
1646         clone_page = nilfs_alloc_private_page(bh->b_bdev, bh->b_size, 0);
1647         if (unlikely(!clone_page))
1648                 return -ENOMEM;
1649
1650         bh2 = page_buffers(clone_page);
1651         kaddr = kmap_atomic(page, KM_USER0);
1652         do {
1653                 if (list_empty(&bh->b_assoc_buffers))
1654                         continue;
1655                 get_bh(bh2);
1656                 page_cache_get(clone_page); /* for each bh */
1657                 memcpy(bh2->b_data, kaddr + bh_offset(bh), bh2->b_size);
1658                 bh2->b_blocknr = bh->b_blocknr;
1659                 list_replace(&bh->b_assoc_buffers, &bh2->b_assoc_buffers);
1660                 list_add_tail(&bh->b_assoc_buffers, out);
1661         } while (bh = bh->b_this_page, bh2 = bh2->b_this_page, bh != head);
1662         kunmap_atomic(kaddr, KM_USER0);
1663
1664         if (!TestSetPageWriteback(clone_page))
1665                 inc_zone_page_state(clone_page, NR_WRITEBACK);
1666         unlock_page(clone_page);
1667
1668         return 0;
1669 }
1670
1671 static int nilfs_test_page_to_be_frozen(struct page *page)
1672 {
1673         struct address_space *mapping = page->mapping;
1674
1675         if (!mapping || !mapping->host || S_ISDIR(mapping->host->i_mode))
1676                 return 0;
1677
1678         if (page_mapped(page)) {
1679                 ClearPageChecked(page);
1680                 return 1;
1681         }
1682         return PageChecked(page);
1683 }
1684
1685 static int nilfs_begin_page_io(struct page *page, struct list_head *out)
1686 {
1687         if (!page || PageWriteback(page))
1688                 /* For split b-tree node pages, this function may be called
1689                    twice.  We ignore the 2nd or later calls by this check. */
1690                 return 0;
1691
1692         lock_page(page);
1693         clear_page_dirty_for_io(page);
1694         set_page_writeback(page);
1695         unlock_page(page);
1696
1697         if (nilfs_test_page_to_be_frozen(page)) {
1698                 int err = nilfs_copy_replace_page_buffers(page, out);
1699                 if (unlikely(err))
1700                         return err;
1701         }
1702         return 0;
1703 }
1704
1705 static int nilfs_segctor_prepare_write(struct nilfs_sc_info *sci,
1706                                        struct page **failed_page)
1707 {
1708         struct nilfs_segment_buffer *segbuf;
1709         struct page *bd_page = NULL, *fs_page = NULL;
1710         struct list_head *list = &sci->sc_copied_buffers;
1711         int err;
1712
1713         *failed_page = NULL;
1714         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1715                 struct buffer_head *bh;
1716
1717                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1718                                     b_assoc_buffers) {
1719                         if (bh->b_page != bd_page) {
1720                                 if (bd_page) {
1721                                         lock_page(bd_page);
1722                                         clear_page_dirty_for_io(bd_page);
1723                                         set_page_writeback(bd_page);
1724                                         unlock_page(bd_page);
1725                                 }
1726                                 bd_page = bh->b_page;
1727                         }
1728                 }
1729
1730                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1731                                     b_assoc_buffers) {
1732                         if (bh == sci->sc_super_root) {
1733                                 if (bh->b_page != bd_page) {
1734                                         lock_page(bd_page);
1735                                         clear_page_dirty_for_io(bd_page);
1736                                         set_page_writeback(bd_page);
1737                                         unlock_page(bd_page);
1738                                         bd_page = bh->b_page;
1739                                 }
1740                                 break;
1741                         }
1742                         if (bh->b_page != fs_page) {
1743                                 err = nilfs_begin_page_io(fs_page, list);
1744                                 if (unlikely(err)) {
1745                                         *failed_page = fs_page;
1746                                         goto out;
1747                                 }
1748                                 fs_page = bh->b_page;
1749                         }
1750                 }
1751         }
1752         if (bd_page) {
1753                 lock_page(bd_page);
1754                 clear_page_dirty_for_io(bd_page);
1755                 set_page_writeback(bd_page);
1756                 unlock_page(bd_page);
1757         }
1758         err = nilfs_begin_page_io(fs_page, list);
1759         if (unlikely(err))
1760                 *failed_page = fs_page;
1761  out:
1762         return err;
1763 }
1764
1765 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1766                                struct the_nilfs *nilfs)
1767 {
1768         int ret;
1769
1770         ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1771         list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1772         return ret;
1773 }
1774
1775 static void __nilfs_end_page_io(struct page *page, int err)
1776 {
1777         if (!err) {
1778                 if (!nilfs_page_buffers_clean(page))
1779                         __set_page_dirty_nobuffers(page);
1780                 ClearPageError(page);
1781         } else {
1782                 __set_page_dirty_nobuffers(page);
1783                 SetPageError(page);
1784         }
1785
1786         if (buffer_nilfs_allocated(page_buffers(page))) {
1787                 if (TestClearPageWriteback(page))
1788                         dec_zone_page_state(page, NR_WRITEBACK);
1789         } else
1790                 end_page_writeback(page);
1791 }
1792
1793 static void nilfs_end_page_io(struct page *page, int err)
1794 {
1795         if (!page)
1796                 return;
1797
1798         if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1799                 /*
1800                  * For b-tree node pages, this function may be called twice
1801                  * or more because they might be split in a segment.
1802                  */
1803                 if (PageDirty(page)) {
1804                         /*
1805                          * For pages holding split b-tree node buffers, dirty
1806                          * flag on the buffers may be cleared discretely.
1807                          * In that case, the page is once redirtied for
1808                          * remaining buffers, and it must be cancelled if
1809                          * all the buffers get cleaned later.
1810                          */
1811                         lock_page(page);
1812                         if (nilfs_page_buffers_clean(page))
1813                                 __nilfs_clear_page_dirty(page);
1814                         unlock_page(page);
1815                 }
1816                 return;
1817         }
1818
1819         __nilfs_end_page_io(page, err);
1820 }
1821
1822 static void nilfs_clear_copied_buffers(struct list_head *list, int err)
1823 {
1824         struct buffer_head *bh, *head;
1825         struct page *page;
1826
1827         while (!list_empty(list)) {
1828                 bh = list_entry(list->next, struct buffer_head,
1829                                 b_assoc_buffers);
1830                 page = bh->b_page;
1831                 page_cache_get(page);
1832                 head = bh = page_buffers(page);
1833                 do {
1834                         if (!list_empty(&bh->b_assoc_buffers)) {
1835                                 list_del_init(&bh->b_assoc_buffers);
1836                                 if (!err) {
1837                                         set_buffer_uptodate(bh);
1838                                         clear_buffer_dirty(bh);
1839                                         clear_buffer_nilfs_volatile(bh);
1840                                 }
1841                                 brelse(bh); /* for b_assoc_buffers */
1842                         }
1843                 } while ((bh = bh->b_this_page) != head);
1844
1845                 __nilfs_end_page_io(page, err);
1846                 page_cache_release(page);
1847         }
1848 }
1849
1850 static void nilfs_abort_logs(struct list_head *logs, struct page *failed_page,
1851                              struct buffer_head *bh_sr, int err)
1852 {
1853         struct nilfs_segment_buffer *segbuf;
1854         struct page *bd_page = NULL, *fs_page = NULL;
1855         struct buffer_head *bh;
1856
1857         if (list_empty(logs))
1858                 return;
1859
1860         list_for_each_entry(segbuf, logs, sb_list) {
1861                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1862                                     b_assoc_buffers) {
1863                         if (bh->b_page != bd_page) {
1864                                 if (bd_page)
1865                                         end_page_writeback(bd_page);
1866                                 bd_page = bh->b_page;
1867                         }
1868                 }
1869
1870                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1871                                     b_assoc_buffers) {
1872                         if (bh == bh_sr) {
1873                                 if (bh->b_page != bd_page) {
1874                                         end_page_writeback(bd_page);
1875                                         bd_page = bh->b_page;
1876                                 }
1877                                 break;
1878                         }
1879                         if (bh->b_page != fs_page) {
1880                                 nilfs_end_page_io(fs_page, err);
1881                                 if (fs_page && fs_page == failed_page)
1882                                         return;
1883                                 fs_page = bh->b_page;
1884                         }
1885                 }
1886         }
1887         if (bd_page)
1888                 end_page_writeback(bd_page);
1889
1890         nilfs_end_page_io(fs_page, err);
1891 }
1892
1893 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1894                                              struct the_nilfs *nilfs, int err)
1895 {
1896         LIST_HEAD(logs);
1897         int ret;
1898
1899         list_splice_tail_init(&sci->sc_write_logs, &logs);
1900         ret = nilfs_wait_on_logs(&logs);
1901         nilfs_abort_logs(&logs, NULL, sci->sc_super_root, ret ? : err);
1902
1903         list_splice_tail_init(&sci->sc_segbufs, &logs);
1904         nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1905         nilfs_free_incomplete_logs(&logs, nilfs);
1906         nilfs_clear_copied_buffers(&sci->sc_copied_buffers, err);
1907
1908         if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1909                 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1910                                                 sci->sc_freesegs,
1911                                                 sci->sc_nfreesegs,
1912                                                 NULL);
1913                 WARN_ON(ret); /* do not happen */
1914         }
1915
1916         nilfs_destroy_logs(&logs);
1917         sci->sc_super_root = NULL;
1918 }
1919
1920 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1921                                    struct nilfs_segment_buffer *segbuf)
1922 {
1923         nilfs->ns_segnum = segbuf->sb_segnum;
1924         nilfs->ns_nextnum = segbuf->sb_nextnum;
1925         nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1926                 + segbuf->sb_sum.nblocks;
1927         nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1928         nilfs->ns_ctime = segbuf->sb_sum.ctime;
1929 }
1930
1931 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1932 {
1933         struct nilfs_segment_buffer *segbuf;
1934         struct page *bd_page = NULL, *fs_page = NULL;
1935         struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
1936         int update_sr = (sci->sc_super_root != NULL);
1937
1938         list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1939                 struct buffer_head *bh;
1940
1941                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1942                                     b_assoc_buffers) {
1943                         set_buffer_uptodate(bh);
1944                         clear_buffer_dirty(bh);
1945                         if (bh->b_page != bd_page) {
1946                                 if (bd_page)
1947                                         end_page_writeback(bd_page);
1948                                 bd_page = bh->b_page;
1949                         }
1950                 }
1951                 /*
1952                  * We assume that the buffers which belong to the same page
1953                  * continue over the buffer list.
1954                  * Under this assumption, the last BHs of pages is
1955                  * identifiable by the discontinuity of bh->b_page
1956                  * (page != fs_page).
1957                  *
1958                  * For B-tree node blocks, however, this assumption is not
1959                  * guaranteed.  The cleanup code of B-tree node pages needs
1960                  * special care.
1961                  */
1962                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1963                                     b_assoc_buffers) {
1964                         set_buffer_uptodate(bh);
1965                         clear_buffer_dirty(bh);
1966                         clear_buffer_nilfs_volatile(bh);
1967                         if (bh == sci->sc_super_root) {
1968                                 if (bh->b_page != bd_page) {
1969                                         end_page_writeback(bd_page);
1970                                         bd_page = bh->b_page;
1971                                 }
1972                                 break;
1973                         }
1974                         if (bh->b_page != fs_page) {
1975                                 nilfs_end_page_io(fs_page, 0);
1976                                 fs_page = bh->b_page;
1977                         }
1978                 }
1979
1980                 if (!NILFS_SEG_SIMPLEX(&segbuf->sb_sum)) {
1981                         if (NILFS_SEG_LOGBGN(&segbuf->sb_sum)) {
1982                                 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1983                                 sci->sc_lseg_stime = jiffies;
1984                         }
1985                         if (NILFS_SEG_LOGEND(&segbuf->sb_sum))
1986                                 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1987                 }
1988         }
1989         /*
1990          * Since pages may continue over multiple segment buffers,
1991          * end of the last page must be checked outside of the loop.
1992          */
1993         if (bd_page)
1994                 end_page_writeback(bd_page);
1995
1996         nilfs_end_page_io(fs_page, 0);
1997
1998         nilfs_clear_copied_buffers(&sci->sc_copied_buffers, 0);
1999
2000         nilfs_drop_collected_inodes(&sci->sc_dirty_files);
2001
2002         if (nilfs_doing_gc()) {
2003                 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
2004                 if (update_sr)
2005                         nilfs_commit_gcdat_inode(nilfs);
2006         } else
2007                 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
2008
2009         sci->sc_nblk_inc += sci->sc_nblk_this_inc;
2010
2011         segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
2012         nilfs_set_next_segment(nilfs, segbuf);
2013
2014         if (update_sr) {
2015                 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
2016                                        segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
2017                 set_nilfs_sb_dirty(nilfs);
2018
2019                 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
2020                 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2021                 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
2022                 nilfs_segctor_clear_metadata_dirty(sci);
2023         } else
2024                 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
2025 }
2026
2027 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
2028 {
2029         int ret;
2030
2031         ret = nilfs_wait_on_logs(&sci->sc_write_logs);
2032         if (!ret) {
2033                 nilfs_segctor_complete_write(sci);
2034                 nilfs_destroy_logs(&sci->sc_write_logs);
2035         }
2036         return ret;
2037 }
2038
2039 static int nilfs_segctor_check_in_files(struct nilfs_sc_info *sci,
2040                                         struct nilfs_sb_info *sbi)
2041 {
2042         struct nilfs_inode_info *ii, *n;
2043         __u64 cno = sbi->s_nilfs->ns_cno;
2044
2045         spin_lock(&sbi->s_inode_lock);
2046  retry:
2047         list_for_each_entry_safe(ii, n, &sbi->s_dirty_files, i_dirty) {
2048                 if (!ii->i_bh) {
2049                         struct buffer_head *ibh;
2050                         int err;
2051
2052                         spin_unlock(&sbi->s_inode_lock);
2053                         err = nilfs_ifile_get_inode_block(
2054                                 sbi->s_ifile, ii->vfs_inode.i_ino, &ibh);
2055                         if (unlikely(err)) {
2056                                 nilfs_warning(sbi->s_super, __func__,
2057                                               "failed to get inode block.\n");
2058                                 return err;
2059                         }
2060                         nilfs_mdt_mark_buffer_dirty(ibh);
2061                         nilfs_mdt_mark_dirty(sbi->s_ifile);
2062                         spin_lock(&sbi->s_inode_lock);
2063                         if (likely(!ii->i_bh))
2064                                 ii->i_bh = ibh;
2065                         else
2066                                 brelse(ibh);
2067                         goto retry;
2068                 }
2069                 ii->i_cno = cno;
2070
2071                 clear_bit(NILFS_I_QUEUED, &ii->i_state);
2072                 set_bit(NILFS_I_BUSY, &ii->i_state);
2073                 list_del(&ii->i_dirty);
2074                 list_add_tail(&ii->i_dirty, &sci->sc_dirty_files);
2075         }
2076         spin_unlock(&sbi->s_inode_lock);
2077
2078         NILFS_I(sbi->s_ifile)->i_cno = cno;
2079
2080         return 0;
2081 }
2082
2083 static void nilfs_segctor_check_out_files(struct nilfs_sc_info *sci,
2084                                           struct nilfs_sb_info *sbi)
2085 {
2086         struct nilfs_transaction_info *ti = current->journal_info;
2087         struct nilfs_inode_info *ii, *n;
2088         __u64 cno = sbi->s_nilfs->ns_cno;
2089
2090         spin_lock(&sbi->s_inode_lock);
2091         list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2092                 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2093                     test_bit(NILFS_I_DIRTY, &ii->i_state)) {
2094                         /* The current checkpoint number (=nilfs->ns_cno) is
2095                            changed between check-in and check-out only if the
2096                            super root is written out.  So, we can update i_cno
2097                            for the inodes that remain in the dirty list. */
2098                         ii->i_cno = cno;
2099                         continue;
2100                 }
2101                 clear_bit(NILFS_I_BUSY, &ii->i_state);
2102                 brelse(ii->i_bh);
2103                 ii->i_bh = NULL;
2104                 list_del(&ii->i_dirty);
2105                 list_add_tail(&ii->i_dirty, &ti->ti_garbage);
2106         }
2107         spin_unlock(&sbi->s_inode_lock);
2108 }
2109
2110 /*
2111  * Main procedure of segment constructor
2112  */
2113 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2114 {
2115         struct nilfs_sb_info *sbi = sci->sc_sbi;
2116         struct the_nilfs *nilfs = sbi->s_nilfs;
2117         struct page *failed_page;
2118         int err, has_sr = 0;
2119
2120         sci->sc_stage.scnt = NILFS_ST_INIT;
2121
2122         err = nilfs_segctor_check_in_files(sci, sbi);
2123         if (unlikely(err))
2124                 goto out;
2125
2126         if (nilfs_test_metadata_dirty(sbi))
2127                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2128
2129         if (nilfs_segctor_clean(sci))
2130                 goto out;
2131
2132         do {
2133                 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2134
2135                 err = nilfs_segctor_begin_construction(sci, nilfs);
2136                 if (unlikely(err))
2137                         goto out;
2138
2139                 /* Update time stamp */
2140                 sci->sc_seg_ctime = get_seconds();
2141
2142                 err = nilfs_segctor_collect(sci, nilfs, mode);
2143                 if (unlikely(err))
2144                         goto failed;
2145
2146                 has_sr = (sci->sc_super_root != NULL);
2147
2148                 /* Avoid empty segment */
2149                 if (sci->sc_stage.scnt == NILFS_ST_DONE &&
2150                     NILFS_SEG_EMPTY(&sci->sc_curseg->sb_sum)) {
2151                         nilfs_segctor_abort_construction(sci, nilfs, 1);
2152                         goto out;
2153                 }
2154
2155                 err = nilfs_segctor_assign(sci, mode);
2156                 if (unlikely(err))
2157                         goto failed;
2158
2159                 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2160                         nilfs_segctor_fill_in_file_bmap(sci, sbi->s_ifile);
2161
2162                 if (has_sr) {
2163                         err = nilfs_segctor_fill_in_checkpoint(sci);
2164                         if (unlikely(err))
2165                                 goto failed_to_write;
2166
2167                         nilfs_segctor_fill_in_super_root(sci, nilfs);
2168                 }
2169                 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2170
2171                 /* Write partial segments */
2172                 err = nilfs_segctor_prepare_write(sci, &failed_page);
2173                 if (err) {
2174                         nilfs_abort_logs(&sci->sc_segbufs, failed_page,
2175                                          sci->sc_super_root, err);
2176                         goto failed_to_write;
2177                 }
2178                 nilfs_segctor_fill_in_checksums(sci, nilfs->ns_crc_seed);
2179
2180                 err = nilfs_segctor_write(sci, nilfs);
2181                 if (unlikely(err))
2182                         goto failed_to_write;
2183
2184                 if (sci->sc_stage.scnt == NILFS_ST_DONE ||
2185                     nilfs->ns_blocksize_bits != PAGE_CACHE_SHIFT) {
2186                         /*
2187                          * At this point, we avoid double buffering
2188                          * for blocksize < pagesize because page dirty
2189                          * flag is turned off during write and dirty
2190                          * buffers are not properly collected for
2191                          * pages crossing over segments.
2192                          */
2193                         err = nilfs_segctor_wait(sci);
2194                         if (err)
2195                                 goto failed_to_write;
2196                 }
2197         } while (sci->sc_stage.scnt != NILFS_ST_DONE);
2198
2199         sci->sc_super_root = NULL;
2200
2201  out:
2202         nilfs_segctor_check_out_files(sci, sbi);
2203         return err;
2204
2205  failed_to_write:
2206         if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2207                 nilfs_redirty_inodes(&sci->sc_dirty_files);
2208
2209  failed:
2210         if (nilfs_doing_gc())
2211                 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2212         nilfs_segctor_abort_construction(sci, nilfs, err);
2213         goto out;
2214 }
2215
2216 /**
2217  * nilfs_segctor_start_timer - set timer of background write
2218  * @sci: nilfs_sc_info
2219  *
2220  * If the timer has already been set, it ignores the new request.
2221  * This function MUST be called within a section locking the segment
2222  * semaphore.
2223  */
2224 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2225 {
2226         spin_lock(&sci->sc_state_lock);
2227         if (sci->sc_timer && !(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2228                 sci->sc_timer->expires = jiffies + sci->sc_interval;
2229                 add_timer(sci->sc_timer);
2230                 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2231         }
2232         spin_unlock(&sci->sc_state_lock);
2233 }
2234
2235 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2236 {
2237         spin_lock(&sci->sc_state_lock);
2238         if (!(sci->sc_flush_request & (1 << bn))) {
2239                 unsigned long prev_req = sci->sc_flush_request;
2240
2241                 sci->sc_flush_request |= (1 << bn);
2242                 if (!prev_req)
2243                         wake_up(&sci->sc_wait_daemon);
2244         }
2245         spin_unlock(&sci->sc_state_lock);
2246 }
2247
2248 /**
2249  * nilfs_flush_segment - trigger a segment construction for resource control
2250  * @sb: super block
2251  * @ino: inode number of the file to be flushed out.
2252  */
2253 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2254 {
2255         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2256         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2257
2258         if (!sci || nilfs_doing_construction())
2259                 return;
2260         nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2261                                         /* assign bit 0 to data files */
2262 }
2263
2264 struct nilfs_segctor_wait_request {
2265         wait_queue_t    wq;
2266         __u32           seq;
2267         int             err;
2268         atomic_t        done;
2269 };
2270
2271 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2272 {
2273         struct nilfs_segctor_wait_request wait_req;
2274         int err = 0;
2275
2276         spin_lock(&sci->sc_state_lock);
2277         init_wait(&wait_req.wq);
2278         wait_req.err = 0;
2279         atomic_set(&wait_req.done, 0);
2280         wait_req.seq = ++sci->sc_seq_request;
2281         spin_unlock(&sci->sc_state_lock);
2282
2283         init_waitqueue_entry(&wait_req.wq, current);
2284         add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2285         set_current_state(TASK_INTERRUPTIBLE);
2286         wake_up(&sci->sc_wait_daemon);
2287
2288         for (;;) {
2289                 if (atomic_read(&wait_req.done)) {
2290                         err = wait_req.err;
2291                         break;
2292                 }
2293                 if (!signal_pending(current)) {
2294                         schedule();
2295                         continue;
2296                 }
2297                 err = -ERESTARTSYS;
2298                 break;
2299         }
2300         finish_wait(&sci->sc_wait_request, &wait_req.wq);
2301         return err;
2302 }
2303
2304 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2305 {
2306         struct nilfs_segctor_wait_request *wrq, *n;
2307         unsigned long flags;
2308
2309         spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2310         list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2311                                  wq.task_list) {
2312                 if (!atomic_read(&wrq->done) &&
2313                     nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2314                         wrq->err = err;
2315                         atomic_set(&wrq->done, 1);
2316                 }
2317                 if (atomic_read(&wrq->done)) {
2318                         wrq->wq.func(&wrq->wq,
2319                                      TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2320                                      0, NULL);
2321                 }
2322         }
2323         spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2324 }
2325
2326 /**
2327  * nilfs_construct_segment - construct a logical segment
2328  * @sb: super block
2329  *
2330  * Return Value: On success, 0 is retured. On errors, one of the following
2331  * negative error code is returned.
2332  *
2333  * %-EROFS - Read only filesystem.
2334  *
2335  * %-EIO - I/O error
2336  *
2337  * %-ENOSPC - No space left on device (only in a panic state).
2338  *
2339  * %-ERESTARTSYS - Interrupted.
2340  *
2341  * %-ENOMEM - Insufficient memory available.
2342  */
2343 int nilfs_construct_segment(struct super_block *sb)
2344 {
2345         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2346         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2347         struct nilfs_transaction_info *ti;
2348         int err;
2349
2350         if (!sci)
2351                 return -EROFS;
2352
2353         /* A call inside transactions causes a deadlock. */
2354         BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2355
2356         err = nilfs_segctor_sync(sci);
2357         return err;
2358 }
2359
2360 /**
2361  * nilfs_construct_dsync_segment - construct a data-only logical segment
2362  * @sb: super block
2363  * @inode: inode whose data blocks should be written out
2364  * @start: start byte offset
2365  * @end: end byte offset (inclusive)
2366  *
2367  * Return Value: On success, 0 is retured. On errors, one of the following
2368  * negative error code is returned.
2369  *
2370  * %-EROFS - Read only filesystem.
2371  *
2372  * %-EIO - I/O error
2373  *
2374  * %-ENOSPC - No space left on device (only in a panic state).
2375  *
2376  * %-ERESTARTSYS - Interrupted.
2377  *
2378  * %-ENOMEM - Insufficient memory available.
2379  */
2380 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2381                                   loff_t start, loff_t end)
2382 {
2383         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2384         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2385         struct nilfs_inode_info *ii;
2386         struct nilfs_transaction_info ti;
2387         int err = 0;
2388
2389         if (!sci)
2390                 return -EROFS;
2391
2392         nilfs_transaction_lock(sbi, &ti, 0);
2393
2394         ii = NILFS_I(inode);
2395         if (test_bit(NILFS_I_INODE_DIRTY, &ii->i_state) ||
2396             nilfs_test_opt(sbi, STRICT_ORDER) ||
2397             test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2398             nilfs_discontinued(sbi->s_nilfs)) {
2399                 nilfs_transaction_unlock(sbi);
2400                 err = nilfs_segctor_sync(sci);
2401                 return err;
2402         }
2403
2404         spin_lock(&sbi->s_inode_lock);
2405         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2406             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2407                 spin_unlock(&sbi->s_inode_lock);
2408                 nilfs_transaction_unlock(sbi);
2409                 return 0;
2410         }
2411         spin_unlock(&sbi->s_inode_lock);
2412         sci->sc_dsync_inode = ii;
2413         sci->sc_dsync_start = start;
2414         sci->sc_dsync_end = end;
2415
2416         err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2417
2418         nilfs_transaction_unlock(sbi);
2419         return err;
2420 }
2421
2422 #define FLUSH_FILE_BIT  (0x1) /* data file only */
2423 #define FLUSH_DAT_BIT   (1 << NILFS_DAT_INO) /* DAT only */
2424
2425 /**
2426  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2427  * @sci: segment constructor object
2428  */
2429 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2430 {
2431         spin_lock(&sci->sc_state_lock);
2432         sci->sc_seq_accepted = sci->sc_seq_request;
2433         spin_unlock(&sci->sc_state_lock);
2434
2435         if (sci->sc_timer)
2436                 del_timer_sync(sci->sc_timer);
2437 }
2438
2439 /**
2440  * nilfs_segctor_notify - notify the result of request to caller threads
2441  * @sci: segment constructor object
2442  * @mode: mode of log forming
2443  * @err: error code to be notified
2444  */
2445 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2446 {
2447         /* Clear requests (even when the construction failed) */
2448         spin_lock(&sci->sc_state_lock);
2449
2450         if (mode == SC_LSEG_SR) {
2451                 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2452                 sci->sc_seq_done = sci->sc_seq_accepted;
2453                 nilfs_segctor_wakeup(sci, err);
2454                 sci->sc_flush_request = 0;
2455         } else {
2456                 if (mode == SC_FLUSH_FILE)
2457                         sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2458                 else if (mode == SC_FLUSH_DAT)
2459                         sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2460
2461                 /* re-enable timer if checkpoint creation was not done */
2462                 if (sci->sc_timer && (sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2463                     time_before(jiffies, sci->sc_timer->expires))
2464                         add_timer(sci->sc_timer);
2465         }
2466         spin_unlock(&sci->sc_state_lock);
2467 }
2468
2469 /**
2470  * nilfs_segctor_construct - form logs and write them to disk
2471  * @sci: segment constructor object
2472  * @mode: mode of log forming
2473  */
2474 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2475 {
2476         struct nilfs_sb_info *sbi = sci->sc_sbi;
2477         struct the_nilfs *nilfs = sbi->s_nilfs;
2478         int err = 0;
2479
2480         nilfs_segctor_accept(sci);
2481
2482         if (nilfs_discontinued(nilfs))
2483                 mode = SC_LSEG_SR;
2484         if (!nilfs_segctor_confirm(sci))
2485                 err = nilfs_segctor_do_construct(sci, mode);
2486
2487         if (likely(!err)) {
2488                 if (mode != SC_FLUSH_DAT)
2489                         atomic_set(&nilfs->ns_ndirtyblks, 0);
2490                 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2491                     nilfs_discontinued(nilfs)) {
2492                         down_write(&nilfs->ns_sem);
2493                         err = nilfs_commit_super(
2494                                 sbi, nilfs_altsb_need_update(nilfs));
2495                         up_write(&nilfs->ns_sem);
2496                 }
2497         }
2498
2499         nilfs_segctor_notify(sci, mode, err);
2500         return err;
2501 }
2502
2503 static void nilfs_construction_timeout(unsigned long data)
2504 {
2505         struct task_struct *p = (struct task_struct *)data;
2506         wake_up_process(p);
2507 }
2508
2509 static void
2510 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2511 {
2512         struct nilfs_inode_info *ii, *n;
2513
2514         list_for_each_entry_safe(ii, n, head, i_dirty) {
2515                 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2516                         continue;
2517                 hlist_del_init(&ii->vfs_inode.i_hash);
2518                 list_del_init(&ii->i_dirty);
2519                 nilfs_clear_gcinode(&ii->vfs_inode);
2520         }
2521 }
2522
2523 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2524                          void **kbufs)
2525 {
2526         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2527         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2528         struct the_nilfs *nilfs = sbi->s_nilfs;
2529         struct nilfs_transaction_info ti;
2530         int err;
2531
2532         if (unlikely(!sci))
2533                 return -EROFS;
2534
2535         nilfs_transaction_lock(sbi, &ti, 1);
2536
2537         err = nilfs_init_gcdat_inode(nilfs);
2538         if (unlikely(err))
2539                 goto out_unlock;
2540
2541         err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2542         if (unlikely(err))
2543                 goto out_unlock;
2544
2545         sci->sc_freesegs = kbufs[4];
2546         sci->sc_nfreesegs = argv[4].v_nmembs;
2547         list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2548
2549         for (;;) {
2550                 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2551                 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2552
2553                 if (likely(!err))
2554                         break;
2555
2556                 nilfs_warning(sb, __func__,
2557                               "segment construction failed. (err=%d)", err);
2558                 set_current_state(TASK_INTERRUPTIBLE);
2559                 schedule_timeout(sci->sc_interval);
2560         }
2561         if (nilfs_test_opt(sbi, DISCARD)) {
2562                 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2563                                                  sci->sc_nfreesegs);
2564                 if (ret) {
2565                         printk(KERN_WARNING
2566                                "NILFS warning: error %d on discard request, "
2567                                "turning discards off for the device\n", ret);
2568                         nilfs_clear_opt(sbi, DISCARD);
2569                 }
2570         }
2571
2572  out_unlock:
2573         sci->sc_freesegs = NULL;
2574         sci->sc_nfreesegs = 0;
2575         nilfs_clear_gcdat_inode(nilfs);
2576         nilfs_transaction_unlock(sbi);
2577         return err;
2578 }
2579
2580 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2581 {
2582         struct nilfs_sb_info *sbi = sci->sc_sbi;
2583         struct nilfs_transaction_info ti;
2584
2585         nilfs_transaction_lock(sbi, &ti, 0);
2586         nilfs_segctor_construct(sci, mode);
2587
2588         /*
2589          * Unclosed segment should be retried.  We do this using sc_timer.
2590          * Timeout of sc_timer will invoke complete construction which leads
2591          * to close the current logical segment.
2592          */
2593         if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2594                 nilfs_segctor_start_timer(sci);
2595
2596         nilfs_transaction_unlock(sbi);
2597 }
2598
2599 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2600 {
2601         int mode = 0;
2602         int err;
2603
2604         spin_lock(&sci->sc_state_lock);
2605         mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2606                 SC_FLUSH_DAT : SC_FLUSH_FILE;
2607         spin_unlock(&sci->sc_state_lock);
2608
2609         if (mode) {
2610                 err = nilfs_segctor_do_construct(sci, mode);
2611
2612                 spin_lock(&sci->sc_state_lock);
2613                 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2614                         ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2615                 spin_unlock(&sci->sc_state_lock);
2616         }
2617         clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2618 }
2619
2620 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2621 {
2622         if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2623             time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2624                 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2625                         return SC_FLUSH_FILE;
2626                 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2627                         return SC_FLUSH_DAT;
2628         }
2629         return SC_LSEG_SR;
2630 }
2631
2632 /**
2633  * nilfs_segctor_thread - main loop of the segment constructor thread.
2634  * @arg: pointer to a struct nilfs_sc_info.
2635  *
2636  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2637  * to execute segment constructions.
2638  */
2639 static int nilfs_segctor_thread(void *arg)
2640 {
2641         struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2642         struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
2643         struct timer_list timer;
2644         int timeout = 0;
2645
2646         init_timer(&timer);
2647         timer.data = (unsigned long)current;
2648         timer.function = nilfs_construction_timeout;
2649         sci->sc_timer = &timer;
2650
2651         /* start sync. */
2652         sci->sc_task = current;
2653         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2654         printk(KERN_INFO
2655                "segctord starting. Construction interval = %lu seconds, "
2656                "CP frequency < %lu seconds\n",
2657                sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2658
2659         spin_lock(&sci->sc_state_lock);
2660  loop:
2661         for (;;) {
2662                 int mode;
2663
2664                 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2665                         goto end_thread;
2666
2667                 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2668                         mode = SC_LSEG_SR;
2669                 else if (!sci->sc_flush_request)
2670                         break;
2671                 else
2672                         mode = nilfs_segctor_flush_mode(sci);
2673
2674                 spin_unlock(&sci->sc_state_lock);
2675                 nilfs_segctor_thread_construct(sci, mode);
2676                 spin_lock(&sci->sc_state_lock);
2677                 timeout = 0;
2678         }
2679
2680
2681         if (freezing(current)) {
2682                 spin_unlock(&sci->sc_state_lock);
2683                 refrigerator();
2684                 spin_lock(&sci->sc_state_lock);
2685         } else {
2686                 DEFINE_WAIT(wait);
2687                 int should_sleep = 1;
2688
2689                 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2690                                 TASK_INTERRUPTIBLE);
2691
2692                 if (sci->sc_seq_request != sci->sc_seq_done)
2693                         should_sleep = 0;
2694                 else if (sci->sc_flush_request)
2695                         should_sleep = 0;
2696                 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2697                         should_sleep = time_before(jiffies,
2698                                                    sci->sc_timer->expires);
2699
2700                 if (should_sleep) {
2701                         spin_unlock(&sci->sc_state_lock);
2702                         schedule();
2703                         spin_lock(&sci->sc_state_lock);
2704                 }
2705                 finish_wait(&sci->sc_wait_daemon, &wait);
2706                 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2707                            time_after_eq(jiffies, sci->sc_timer->expires));
2708
2709                 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2710                         set_nilfs_discontinued(nilfs);
2711         }
2712         goto loop;
2713
2714  end_thread:
2715         spin_unlock(&sci->sc_state_lock);
2716         del_timer_sync(sci->sc_timer);
2717         sci->sc_timer = NULL;
2718
2719         /* end sync. */
2720         sci->sc_task = NULL;
2721         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2722         return 0;
2723 }
2724
2725 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2726 {
2727         struct task_struct *t;
2728
2729         t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2730         if (IS_ERR(t)) {
2731                 int err = PTR_ERR(t);
2732
2733                 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2734                        err);
2735                 return err;
2736         }
2737         wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2738         return 0;
2739 }
2740
2741 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2742 {
2743         sci->sc_state |= NILFS_SEGCTOR_QUIT;
2744
2745         while (sci->sc_task) {
2746                 wake_up(&sci->sc_wait_daemon);
2747                 spin_unlock(&sci->sc_state_lock);
2748                 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2749                 spin_lock(&sci->sc_state_lock);
2750         }
2751 }
2752
2753 static int nilfs_segctor_init(struct nilfs_sc_info *sci)
2754 {
2755         sci->sc_seq_done = sci->sc_seq_request;
2756
2757         return nilfs_segctor_start_thread(sci);
2758 }
2759
2760 /*
2761  * Setup & clean-up functions
2762  */
2763 static struct nilfs_sc_info *nilfs_segctor_new(struct nilfs_sb_info *sbi)
2764 {
2765         struct nilfs_sc_info *sci;
2766
2767         sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2768         if (!sci)
2769                 return NULL;
2770
2771         sci->sc_sbi = sbi;
2772         sci->sc_super = sbi->s_super;
2773
2774         init_waitqueue_head(&sci->sc_wait_request);
2775         init_waitqueue_head(&sci->sc_wait_daemon);
2776         init_waitqueue_head(&sci->sc_wait_task);
2777         spin_lock_init(&sci->sc_state_lock);
2778         INIT_LIST_HEAD(&sci->sc_dirty_files);
2779         INIT_LIST_HEAD(&sci->sc_segbufs);
2780         INIT_LIST_HEAD(&sci->sc_write_logs);
2781         INIT_LIST_HEAD(&sci->sc_gc_inodes);
2782         INIT_LIST_HEAD(&sci->sc_copied_buffers);
2783
2784         sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2785         sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2786         sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2787
2788         if (sbi->s_interval)
2789                 sci->sc_interval = sbi->s_interval;
2790         if (sbi->s_watermark)
2791                 sci->sc_watermark = sbi->s_watermark;
2792         return sci;
2793 }
2794
2795 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2796 {
2797         int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2798
2799         /* The segctord thread was stopped and its timer was removed.
2800            But some tasks remain. */
2801         do {
2802                 struct nilfs_sb_info *sbi = sci->sc_sbi;
2803                 struct nilfs_transaction_info ti;
2804
2805                 nilfs_transaction_lock(sbi, &ti, 0);
2806                 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2807                 nilfs_transaction_unlock(sbi);
2808
2809         } while (ret && retrycount-- > 0);
2810 }
2811
2812 /**
2813  * nilfs_segctor_destroy - destroy the segment constructor.
2814  * @sci: nilfs_sc_info
2815  *
2816  * nilfs_segctor_destroy() kills the segctord thread and frees
2817  * the nilfs_sc_info struct.
2818  * Caller must hold the segment semaphore.
2819  */
2820 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2821 {
2822         struct nilfs_sb_info *sbi = sci->sc_sbi;
2823         int flag;
2824
2825         up_write(&sbi->s_nilfs->ns_segctor_sem);
2826
2827         spin_lock(&sci->sc_state_lock);
2828         nilfs_segctor_kill_thread(sci);
2829         flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2830                 || sci->sc_seq_request != sci->sc_seq_done);
2831         spin_unlock(&sci->sc_state_lock);
2832
2833         if (flag || !nilfs_segctor_confirm(sci))
2834                 nilfs_segctor_write_out(sci);
2835
2836         WARN_ON(!list_empty(&sci->sc_copied_buffers));
2837
2838         if (!list_empty(&sci->sc_dirty_files)) {
2839                 nilfs_warning(sbi->s_super, __func__,
2840                               "dirty file(s) after the final construction\n");
2841                 nilfs_dispose_list(sbi, &sci->sc_dirty_files, 1);
2842         }
2843
2844         WARN_ON(!list_empty(&sci->sc_segbufs));
2845         WARN_ON(!list_empty(&sci->sc_write_logs));
2846
2847         down_write(&sbi->s_nilfs->ns_segctor_sem);
2848
2849         kfree(sci);
2850 }
2851
2852 /**
2853  * nilfs_attach_segment_constructor - attach a segment constructor
2854  * @sbi: nilfs_sb_info
2855  *
2856  * nilfs_attach_segment_constructor() allocates a struct nilfs_sc_info,
2857  * initializes it, and starts the segment constructor.
2858  *
2859  * Return Value: On success, 0 is returned. On error, one of the following
2860  * negative error code is returned.
2861  *
2862  * %-ENOMEM - Insufficient memory available.
2863  */
2864 int nilfs_attach_segment_constructor(struct nilfs_sb_info *sbi)
2865 {
2866         struct the_nilfs *nilfs = sbi->s_nilfs;
2867         int err;
2868
2869         if (NILFS_SC(sbi)) {
2870                 /*
2871                  * This happens if the filesystem was remounted
2872                  * read/write after nilfs_error degenerated it into a
2873                  * read-only mount.
2874                  */
2875                 nilfs_detach_segment_constructor(sbi);
2876         }
2877
2878         sbi->s_sc_info = nilfs_segctor_new(sbi);
2879         if (!sbi->s_sc_info)
2880                 return -ENOMEM;
2881
2882         nilfs_attach_writer(nilfs, sbi);
2883         err = nilfs_segctor_init(NILFS_SC(sbi));
2884         if (err) {
2885                 nilfs_detach_writer(nilfs, sbi);
2886                 kfree(sbi->s_sc_info);
2887                 sbi->s_sc_info = NULL;
2888         }
2889         return err;
2890 }
2891
2892 /**
2893  * nilfs_detach_segment_constructor - destroy the segment constructor
2894  * @sbi: nilfs_sb_info
2895  *
2896  * nilfs_detach_segment_constructor() kills the segment constructor daemon,
2897  * frees the struct nilfs_sc_info, and destroy the dirty file list.
2898  */
2899 void nilfs_detach_segment_constructor(struct nilfs_sb_info *sbi)
2900 {
2901         struct the_nilfs *nilfs = sbi->s_nilfs;
2902         LIST_HEAD(garbage_list);
2903
2904         down_write(&nilfs->ns_segctor_sem);
2905         if (NILFS_SC(sbi)) {
2906                 nilfs_segctor_destroy(NILFS_SC(sbi));
2907                 sbi->s_sc_info = NULL;
2908         }
2909
2910         /* Force to free the list of dirty files */
2911         spin_lock(&sbi->s_inode_lock);
2912         if (!list_empty(&sbi->s_dirty_files)) {
2913                 list_splice_init(&sbi->s_dirty_files, &garbage_list);
2914                 nilfs_warning(sbi->s_super, __func__,
2915                               "Non empty dirty list after the last "
2916                               "segment construction\n");
2917         }
2918         spin_unlock(&sbi->s_inode_lock);
2919         up_write(&nilfs->ns_segctor_sem);
2920
2921         nilfs_dispose_list(sbi, &garbage_list, 1);
2922         nilfs_detach_writer(nilfs, sbi);
2923 }