fs/nilfs2/segment.c

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