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