fs/nilfs2/segment.c

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