fs/nilfs2/segment.c

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