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/bitops.h>
  28 #include <linux/bio.h>
  29 #include <linux/completion.h>
  30 #include <linux/blkdev.h>
  31 #include <linux/backing-dev.h>
  32 #include <linux/freezer.h>
  33 #include <linux/kthread.h>
  34 #include <linux/crc32.h>
  35 #include <linux/pagevec.h>
  36 #include <linux/slab.h>
  37 #include "nilfs.h"
  38 #include "btnode.h"
  39 #include "page.h"
  40 #include "segment.h"
  41 #include "sufile.h"
  42 #include "cpfile.h"
  43 #include "ifile.h"
  44 #include "segbuf.h"
  45
  46
  47 /*
  48  * Segment constructor
  49  */
  50 #define SC_N_INODEVEC   16   /* Size of locally allocated inode vector */
  51
  52 #define SC_MAX_SEGDELTA 64   /* Upper limit of the number of segments
  53                                 appended in collection retry loop */
  54
  55 /* Construction mode */
  56 enum {
  57         SC_LSEG_SR = 1, /* Make a logical segment having a super root */
  58         SC_LSEG_DSYNC,  /* Flush data blocks of a given file and make
  59                            a logical segment without a super root */
  60         SC_FLUSH_FILE,  /* Flush data files, leads to segment writes without
  61                            creating a checkpoint */
  62         SC_FLUSH_DAT,   /* Flush DAT file. This also creates segments without
  63                            a checkpoint */
  64 };
  65
  66 /* Stage numbers of dirty block collection */
  67 enum {
  68         NILFS_ST_INIT = 0,
  69         NILFS_ST_GC,            /* Collecting dirty blocks for GC */
  70         NILFS_ST_FILE,
  71         NILFS_ST_IFILE,
  72         NILFS_ST_CPFILE,
  73         NILFS_ST_SUFILE,
  74         NILFS_ST_DAT,
  75         NILFS_ST_SR,            /* Super root */
  76         NILFS_ST_DSYNC,         /* Data sync blocks */
  77         NILFS_ST_DONE,
  78 };
  79
  80 /* State flags of collection */
  81 #define NILFS_CF_NODE           0x0001  /* Collecting node blocks */
  82 #define NILFS_CF_IFILE_STARTED  0x0002  /* IFILE stage has started */
  83 #define NILFS_CF_SUFREED        0x0004  /* segment usages has been freed */
  84 #define NILFS_CF_HISTORY_MASK   (NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
  85
  86 /* Operations depending on the construction mode and file type */
  87 struct nilfs_sc_operations {
  88         int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
  89                             struct inode *);
  90         int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
  91                             struct inode *);
  92         int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
  93                             struct inode *);
  94         void (*write_data_binfo)(struct nilfs_sc_info *,
  95                                  struct nilfs_segsum_pointer *,
  96                                  union nilfs_binfo *);
  97         void (*write_node_binfo)(struct nilfs_sc_info *,
  98                                  struct nilfs_segsum_pointer *,
  99                                  union nilfs_binfo *);
 100 };
 101
 102 /*
 103  * Other definitions
 104  */
 105 static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
 106 static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
 107 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
 108 static void nilfs_dispose_list(struct the_nilfs *, struct list_head *, 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 the_nilfs *nilfs;
 186         int ret = nilfs_prepare_segment_lock(ti);
 187
 188         if (unlikely(ret < 0))
 189                 return ret;
 190         if (ret > 0)
 191                 return 0;
 192
 193         sb_start_intwrite(sb);
 194
 195         nilfs = sb->s_fs_info;
 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         sb_end_intwrite(sb);
 210         return ret;
 211 }
 212
 213 /**
 214  * nilfs_transaction_commit - commit indivisible file operations.
 215  * @sb: super block
 216  *
 217  * nilfs_transaction_commit() releases the read semaphore which is
 218  * acquired by nilfs_transaction_begin(). This is only performed
 219  * in outermost call of this function.  If a commit flag is set,
 220  * nilfs_transaction_commit() sets a timer to start the segment
 221  * constructor.  If a sync flag is set, it starts construction
 222  * directly.
 223  */
 224 int nilfs_transaction_commit(struct super_block *sb)
 225 {
 226         struct nilfs_transaction_info *ti = current->journal_info;
 227         struct the_nilfs *nilfs = sb->s_fs_info;
 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         if (nilfs->ns_writer) {
 237                 struct nilfs_sc_info *sci = nilfs->ns_writer;
 238
 239                 if (ti->ti_flags & NILFS_TI_COMMIT)
 240                         nilfs_segctor_start_timer(sci);
 241                 if (atomic_read(&nilfs->ns_ndirtyblks) > sci->sc_watermark)
 242                         nilfs_segctor_do_flush(sci, 0);
 243         }
 244         up_read(&nilfs->ns_segctor_sem);
 245         current->journal_info = ti->ti_save;
 246
 247         if (ti->ti_flags & NILFS_TI_SYNC)
 248                 err = nilfs_construct_segment(sb);
 249         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 250                 kmem_cache_free(nilfs_transaction_cachep, ti);
 251         sb_end_intwrite(sb);
 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         struct the_nilfs *nilfs = sb->s_fs_info;
 259
 260         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 261         if (ti->ti_count > 0) {
 262                 ti->ti_count--;
 263                 return;
 264         }
 265         up_read(&nilfs->ns_segctor_sem);
 266
 267         current->journal_info = ti->ti_save;
 268         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 269                 kmem_cache_free(nilfs_transaction_cachep, ti);
 270         sb_end_intwrite(sb);
 271 }
 272
 273 void nilfs_relax_pressure_in_lock(struct super_block *sb)
 274 {
 275         struct the_nilfs *nilfs = sb->s_fs_info;
 276         struct nilfs_sc_info *sci = nilfs->ns_writer;
 277
 278         if (!sci || !sci->sc_flush_request)
 279                 return;
 280
 281         set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
 282         up_read(&nilfs->ns_segctor_sem);
 283
 284         down_write(&nilfs->ns_segctor_sem);
 285         if (sci->sc_flush_request &&
 286             test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
 287                 struct nilfs_transaction_info *ti = current->journal_info;
 288
 289                 ti->ti_flags |= NILFS_TI_WRITER;
 290                 nilfs_segctor_do_immediate_flush(sci);
 291                 ti->ti_flags &= ~NILFS_TI_WRITER;
 292         }
 293         downgrade_write(&nilfs->ns_segctor_sem);
 294 }
 295
 296 static void nilfs_transaction_lock(struct super_block *sb,
 297                                    struct nilfs_transaction_info *ti,
 298                                    int gcflag)
 299 {
 300         struct nilfs_transaction_info *cur_ti = current->journal_info;
 301         struct the_nilfs *nilfs = sb->s_fs_info;
 302         struct nilfs_sc_info *sci = nilfs->ns_writer;
 303
 304         WARN_ON(cur_ti);
 305         ti->ti_flags = NILFS_TI_WRITER;
 306         ti->ti_count = 0;
 307         ti->ti_save = cur_ti;
 308         ti->ti_magic = NILFS_TI_MAGIC;
 309         current->journal_info = ti;
 310
 311         for (;;) {
 312                 down_write(&nilfs->ns_segctor_sem);
 313                 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
 314                         break;
 315
 316                 nilfs_segctor_do_immediate_flush(sci);
 317
 318                 up_write(&nilfs->ns_segctor_sem);
 319                 yield();
 320         }
 321         if (gcflag)
 322                 ti->ti_flags |= NILFS_TI_GC;
 323 }
 324
 325 static void nilfs_transaction_unlock(struct super_block *sb)
 326 {
 327         struct nilfs_transaction_info *ti = current->journal_info;
 328         struct the_nilfs *nilfs = sb->s_fs_info;
 329
 330         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 331         BUG_ON(ti->ti_count > 0);
 332
 333         up_write(&nilfs->ns_segctor_sem);
 334         current->journal_info = ti->ti_save;
 335 }
 336
 337 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
 338                                             struct nilfs_segsum_pointer *ssp,
 339                                             unsigned bytes)
 340 {
 341         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 342         unsigned blocksize = sci->sc_super->s_blocksize;
 343         void *p;
 344
 345         if (unlikely(ssp->offset + bytes > blocksize)) {
 346                 ssp->offset = 0;
 347                 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
 348                                                &segbuf->sb_segsum_buffers));
 349                 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
 350         }
 351         p = ssp->bh->b_data + ssp->offset;
 352         ssp->offset += bytes;
 353         return p;
 354 }
 355
 356 /**
 357  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
 358  * @sci: nilfs_sc_info
 359  */
 360 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
 361 {
 362         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 363         struct buffer_head *sumbh;
 364         unsigned sumbytes;
 365         unsigned flags = 0;
 366         int err;
 367
 368         if (nilfs_doing_gc())
 369                 flags = NILFS_SS_GC;
 370         err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_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 (NILFS_I(inode)->i_root &&
 433             !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 434                 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
 435         /* skip finfo */
 436 }
 437
 438 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
 439                                     struct inode *inode)
 440 {
 441         struct nilfs_finfo *finfo;
 442         struct nilfs_inode_info *ii;
 443         struct nilfs_segment_buffer *segbuf;
 444         __u64 cno;
 445
 446         if (sci->sc_blk_cnt == 0)
 447                 return;
 448
 449         ii = NILFS_I(inode);
 450
 451         if (test_bit(NILFS_I_GCINODE, &ii->i_state))
 452                 cno = ii->i_cno;
 453         else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
 454                 cno = 0;
 455         else
 456                 cno = sci->sc_cno;
 457
 458         finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
 459                                                  sizeof(*finfo));
 460         finfo->fi_ino = cpu_to_le64(inode->i_ino);
 461         finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
 462         finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
 463         finfo->fi_cno = cpu_to_le64(cno);
 464
 465         segbuf = sci->sc_curseg;
 466         segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
 467                 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
 468         sci->sc_finfo_ptr = sci->sc_binfo_ptr;
 469         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 470 }
 471
 472 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
 473                                         struct buffer_head *bh,
 474                                         struct inode *inode,
 475                                         unsigned binfo_size)
 476 {
 477         struct nilfs_segment_buffer *segbuf;
 478         int required, err = 0;
 479
 480  retry:
 481         segbuf = sci->sc_curseg;
 482         required = nilfs_segctor_segsum_block_required(
 483                 sci, &sci->sc_binfo_ptr, binfo_size);
 484         if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
 485                 nilfs_segctor_end_finfo(sci, inode);
 486                 err = nilfs_segctor_feed_segment(sci);
 487                 if (err)
 488                         return err;
 489                 goto retry;
 490         }
 491         if (unlikely(required)) {
 492                 err = nilfs_segbuf_extend_segsum(segbuf);
 493                 if (unlikely(err))
 494                         goto failed;
 495         }
 496         if (sci->sc_blk_cnt == 0)
 497                 nilfs_segctor_begin_finfo(sci, inode);
 498
 499         nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
 500         /* Substitution to vblocknr is delayed until update_blocknr() */
 501         nilfs_segbuf_add_file_buffer(segbuf, bh);
 502         sci->sc_blk_cnt++;
 503  failed:
 504         return err;
 505 }
 506
 507 /*
 508  * Callback functions that enumerate, mark, and collect dirty blocks
 509  */
 510 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
 511                                    struct buffer_head *bh, struct inode *inode)
 512 {
 513         int err;
 514
 515         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 516         if (err < 0)
 517                 return err;
 518
 519         err = nilfs_segctor_add_file_block(sci, bh, inode,
 520                                            sizeof(struct nilfs_binfo_v));
 521         if (!err)
 522                 sci->sc_datablk_cnt++;
 523         return err;
 524 }
 525
 526 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
 527                                    struct buffer_head *bh,
 528                                    struct inode *inode)
 529 {
 530         return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 531 }
 532
 533 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
 534                                    struct buffer_head *bh,
 535                                    struct inode *inode)
 536 {
 537         WARN_ON(!buffer_dirty(bh));
 538         return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 539 }
 540
 541 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
 542                                         struct nilfs_segsum_pointer *ssp,
 543                                         union nilfs_binfo *binfo)
 544 {
 545         struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
 546                 sci, ssp, sizeof(*binfo_v));
 547         *binfo_v = binfo->bi_v;
 548 }
 549
 550 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
 551                                         struct nilfs_segsum_pointer *ssp,
 552                                         union nilfs_binfo *binfo)
 553 {
 554         __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
 555                 sci, ssp, sizeof(*vblocknr));
 556         *vblocknr = binfo->bi_v.bi_vblocknr;
 557 }
 558
 559 static struct nilfs_sc_operations nilfs_sc_file_ops = {
 560         .collect_data = nilfs_collect_file_data,
 561         .collect_node = nilfs_collect_file_node,
 562         .collect_bmap = nilfs_collect_file_bmap,
 563         .write_data_binfo = nilfs_write_file_data_binfo,
 564         .write_node_binfo = nilfs_write_file_node_binfo,
 565 };
 566
 567 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
 568                                   struct buffer_head *bh, struct inode *inode)
 569 {
 570         int err;
 571
 572         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 573         if (err < 0)
 574                 return err;
 575
 576         err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 577         if (!err)
 578                 sci->sc_datablk_cnt++;
 579         return err;
 580 }
 581
 582 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
 583                                   struct buffer_head *bh, struct inode *inode)
 584 {
 585         WARN_ON(!buffer_dirty(bh));
 586         return nilfs_segctor_add_file_block(sci, bh, inode,
 587                                             sizeof(struct nilfs_binfo_dat));
 588 }
 589
 590 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
 591                                        struct nilfs_segsum_pointer *ssp,
 592                                        union nilfs_binfo *binfo)
 593 {
 594         __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
 595                                                           sizeof(*blkoff));
 596         *blkoff = binfo->bi_dat.bi_blkoff;
 597 }
 598
 599 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
 600                                        struct nilfs_segsum_pointer *ssp,
 601                                        union nilfs_binfo *binfo)
 602 {
 603         struct nilfs_binfo_dat *binfo_dat =
 604                 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
 605         *binfo_dat = binfo->bi_dat;
 606 }
 607
 608 static struct nilfs_sc_operations nilfs_sc_dat_ops = {
 609         .collect_data = nilfs_collect_dat_data,
 610         .collect_node = nilfs_collect_file_node,
 611         .collect_bmap = nilfs_collect_dat_bmap,
 612         .write_data_binfo = nilfs_write_dat_data_binfo,
 613         .write_node_binfo = nilfs_write_dat_node_binfo,
 614 };
 615
 616 static struct nilfs_sc_operations nilfs_sc_dsync_ops = {
 617         .collect_data = nilfs_collect_file_data,
 618         .collect_node = NULL,
 619         .collect_bmap = NULL,
 620         .write_data_binfo = nilfs_write_file_data_binfo,
 621         .write_node_binfo = NULL,
 622 };
 623
 624 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
 625                                               struct list_head *listp,
 626                                               size_t nlimit,
 627                                               loff_t start, loff_t end)
 628 {
 629         struct address_space *mapping = inode->i_mapping;
 630         struct pagevec pvec;
 631         pgoff_t index = 0, last = ULONG_MAX;
 632         size_t ndirties = 0;
 633         int i;
 634
 635         if (unlikely(start != 0 || end != LLONG_MAX)) {
 636                 /*
 637                  * A valid range is given for sync-ing data pages. The
 638                  * range is rounded to per-page; extra dirty buffers
 639                  * may be included if blocksize < pagesize.
 640                  */
 641                 index = start >> PAGE_SHIFT;
 642                 last = end >> PAGE_SHIFT;
 643         }
 644         pagevec_init(&pvec, 0);
 645  repeat:
 646         if (unlikely(index > last) ||
 647             !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 648                                 min_t(pgoff_t, last - index,
 649                                       PAGEVEC_SIZE - 1) + 1))
 650                 return ndirties;
 651
 652         for (i = 0; i < pagevec_count(&pvec); i++) {
 653                 struct buffer_head *bh, *head;
 654                 struct page *page = pvec.pages[i];
 655
 656                 if (unlikely(page->index > last))
 657                         break;
 658
 659                 lock_page(page);
 660                 if (!page_has_buffers(page))
 661                         create_empty_buffers(page, 1 << inode->i_blkbits, 0);
 662                 unlock_page(page);
 663
 664                 bh = head = page_buffers(page);
 665                 do {
 666                         if (!buffer_dirty(bh) || buffer_async_write(bh))
 667                                 continue;
 668                         get_bh(bh);
 669                         list_add_tail(&bh->b_assoc_buffers, listp);
 670                         ndirties++;
 671                         if (unlikely(ndirties >= nlimit)) {
 672                                 pagevec_release(&pvec);
 673                                 cond_resched();
 674                                 return ndirties;
 675                         }
 676                 } while (bh = bh->b_this_page, bh != head);
 677         }
 678         pagevec_release(&pvec);
 679         cond_resched();
 680         goto repeat;
 681 }
 682
 683 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
 684                                             struct list_head *listp)
 685 {
 686         struct nilfs_inode_info *ii = NILFS_I(inode);
 687         struct address_space *mapping = &ii->i_btnode_cache;
 688         struct pagevec pvec;
 689         struct buffer_head *bh, *head;
 690         unsigned int i;
 691         pgoff_t index = 0;
 692
 693         pagevec_init(&pvec, 0);
 694
 695         while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 696                                   PAGEVEC_SIZE)) {
 697                 for (i = 0; i < pagevec_count(&pvec); i++) {
 698                         bh = head = page_buffers(pvec.pages[i]);
 699                         do {
 700                                 if (buffer_dirty(bh) &&
 701                                                 !buffer_async_write(bh)) {
 702                                         get_bh(bh);
 703                                         list_add_tail(&bh->b_assoc_buffers,
 704                                                       listp);
 705                                 }
 706                                 bh = bh->b_this_page;
 707                         } while (bh != head);
 708                 }
 709                 pagevec_release(&pvec);
 710                 cond_resched();
 711         }
 712 }
 713
 714 static void nilfs_dispose_list(struct the_nilfs *nilfs,
 715                                struct list_head *head, int force)
 716 {
 717         struct nilfs_inode_info *ii, *n;
 718         struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
 719         unsigned nv = 0;
 720
 721         while (!list_empty(head)) {
 722                 spin_lock(&nilfs->ns_inode_lock);
 723                 list_for_each_entry_safe(ii, n, head, i_dirty) {
 724                         list_del_init(&ii->i_dirty);
 725                         if (force) {
 726                                 if (unlikely(ii->i_bh)) {
 727                                         brelse(ii->i_bh);
 728                                         ii->i_bh = NULL;
 729                                 }
 730                         } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
 731                                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 732                                 list_add_tail(&ii->i_dirty,
 733                                               &nilfs->ns_dirty_files);
 734                                 continue;
 735                         }
 736                         ivec[nv++] = ii;
 737                         if (nv == SC_N_INODEVEC)
 738                                 break;
 739                 }
 740                 spin_unlock(&nilfs->ns_inode_lock);
 741
 742                 for (pii = ivec; nv > 0; pii++, nv--)
 743                         iput(&(*pii)->vfs_inode);
 744         }
 745 }
 746
 747 static void nilfs_iput_work_func(struct work_struct *work)
 748 {
 749         struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
 750                                                  sc_iput_work);
 751         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 752
 753         nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
 754 }
 755
 756 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
 757                                      struct nilfs_root *root)
 758 {
 759         int ret = 0;
 760
 761         if (nilfs_mdt_fetch_dirty(root->ifile))
 762                 ret++;
 763         if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
 764                 ret++;
 765         if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
 766                 ret++;
 767         if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
 768                 ret++;
 769         return ret;
 770 }
 771
 772 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
 773 {
 774         return list_empty(&sci->sc_dirty_files) &&
 775                 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
 776                 sci->sc_nfreesegs == 0 &&
 777                 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
 778 }
 779
 780 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
 781 {
 782         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 783         int ret = 0;
 784
 785         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
 786                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
 787
 788         spin_lock(&nilfs->ns_inode_lock);
 789         if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
 790                 ret++;
 791
 792         spin_unlock(&nilfs->ns_inode_lock);
 793         return ret;
 794 }
 795
 796 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
 797 {
 798         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 799
 800         nilfs_mdt_clear_dirty(sci->sc_root->ifile);
 801         nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
 802         nilfs_mdt_clear_dirty(nilfs->ns_sufile);
 803         nilfs_mdt_clear_dirty(nilfs->ns_dat);
 804 }
 805
 806 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
 807 {
 808         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 809         struct buffer_head *bh_cp;
 810         struct nilfs_checkpoint *raw_cp;
 811         int err;
 812
 813         /* XXX: this interface will be changed */
 814         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
 815                                           &raw_cp, &bh_cp);
 816         if (likely(!err)) {
 817                 /* The following code is duplicated with cpfile.  But, it is
 818                    needed to collect the checkpoint even if it was not newly
 819                    created */
 820                 mark_buffer_dirty(bh_cp);
 821                 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
 822                 nilfs_cpfile_put_checkpoint(
 823                         nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 824         } else
 825                 WARN_ON(err == -EINVAL || err == -ENOENT);
 826
 827         return err;
 828 }
 829
 830 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
 831 {
 832         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 833         struct buffer_head *bh_cp;
 834         struct nilfs_checkpoint *raw_cp;
 835         int err;
 836
 837         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
 838                                           &raw_cp, &bh_cp);
 839         if (unlikely(err)) {
 840                 WARN_ON(err == -EINVAL || err == -ENOENT);
 841                 goto failed_ibh;
 842         }
 843         raw_cp->cp_snapshot_list.ssl_next = 0;
 844         raw_cp->cp_snapshot_list.ssl_prev = 0;
 845         raw_cp->cp_inodes_count =
 846                 cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
 847         raw_cp->cp_blocks_count =
 848                 cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
 849         raw_cp->cp_nblk_inc =
 850                 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
 851         raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
 852         raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
 853
 854         if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 855                 nilfs_checkpoint_clear_minor(raw_cp);
 856         else
 857                 nilfs_checkpoint_set_minor(raw_cp);
 858
 859         nilfs_write_inode_common(sci->sc_root->ifile,
 860                                  &raw_cp->cp_ifile_inode, 1);
 861         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 862         return 0;
 863
 864  failed_ibh:
 865         return err;
 866 }
 867
 868 static void nilfs_fill_in_file_bmap(struct inode *ifile,
 869                                     struct nilfs_inode_info *ii)
 870
 871 {
 872         struct buffer_head *ibh;
 873         struct nilfs_inode *raw_inode;
 874
 875         if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
 876                 ibh = ii->i_bh;
 877                 BUG_ON(!ibh);
 878                 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
 879                                                   ibh);
 880                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 881                 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
 882         }
 883 }
 884
 885 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
 886 {
 887         struct nilfs_inode_info *ii;
 888
 889         list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
 890                 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
 891                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
 892         }
 893 }
 894
 895 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
 896                                              struct the_nilfs *nilfs)
 897 {
 898         struct buffer_head *bh_sr;
 899         struct nilfs_super_root *raw_sr;
 900         unsigned isz, srsz;
 901
 902         bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
 903         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
 904         isz = nilfs->ns_inode_size;
 905         srsz = NILFS_SR_BYTES(isz);
 906
 907         raw_sr->sr_bytes = cpu_to_le16(srsz);
 908         raw_sr->sr_nongc_ctime
 909                 = cpu_to_le64(nilfs_doing_gc() ?
 910                               nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
 911         raw_sr->sr_flags = 0;
 912
 913         nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
 914                                  NILFS_SR_DAT_OFFSET(isz), 1);
 915         nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
 916                                  NILFS_SR_CPFILE_OFFSET(isz), 1);
 917         nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
 918                                  NILFS_SR_SUFILE_OFFSET(isz), 1);
 919         memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
 920 }
 921
 922 static void nilfs_redirty_inodes(struct list_head *head)
 923 {
 924         struct nilfs_inode_info *ii;
 925
 926         list_for_each_entry(ii, head, i_dirty) {
 927                 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
 928                         clear_bit(NILFS_I_COLLECTED, &ii->i_state);
 929         }
 930 }
 931
 932 static void nilfs_drop_collected_inodes(struct list_head *head)
 933 {
 934         struct nilfs_inode_info *ii;
 935
 936         list_for_each_entry(ii, head, i_dirty) {
 937                 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
 938                         continue;
 939
 940                 clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
 941                 set_bit(NILFS_I_UPDATED, &ii->i_state);
 942         }
 943 }
 944
 945 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
 946                                        struct inode *inode,
 947                                        struct list_head *listp,
 948                                        int (*collect)(struct nilfs_sc_info *,
 949                                                       struct buffer_head *,
 950                                                       struct inode *))
 951 {
 952         struct buffer_head *bh, *n;
 953         int err = 0;
 954
 955         if (collect) {
 956                 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
 957                         list_del_init(&bh->b_assoc_buffers);
 958                         err = collect(sci, bh, inode);
 959                         brelse(bh);
 960                         if (unlikely(err))
 961                                 goto dispose_buffers;
 962                 }
 963                 return 0;
 964         }
 965
 966  dispose_buffers:
 967         while (!list_empty(listp)) {
 968                 bh = list_first_entry(listp, struct buffer_head,
 969                                       b_assoc_buffers);
 970                 list_del_init(&bh->b_assoc_buffers);
 971                 brelse(bh);
 972         }
 973         return err;
 974 }
 975
 976 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
 977 {
 978         /* Remaining number of blocks within segment buffer */
 979         return sci->sc_segbuf_nblocks -
 980                 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
 981 }
 982
 983 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
 984                                    struct inode *inode,
 985                                    struct nilfs_sc_operations *sc_ops)
 986 {
 987         LIST_HEAD(data_buffers);
 988         LIST_HEAD(node_buffers);
 989         int err;
 990
 991         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
 992                 size_t n, rest = nilfs_segctor_buffer_rest(sci);
 993
 994                 n = nilfs_lookup_dirty_data_buffers(
 995                         inode, &data_buffers, rest + 1, 0, LLONG_MAX);
 996                 if (n > rest) {
 997                         err = nilfs_segctor_apply_buffers(
 998                                 sci, inode, &data_buffers,
 999                                 sc_ops->collect_data);
1000                         BUG_ON(!err); /* always receive -E2BIG or true error */
1001                         goto break_or_fail;
1002                 }
1003         }
1004         nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1005
1006         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1007                 err = nilfs_segctor_apply_buffers(
1008                         sci, inode, &data_buffers, sc_ops->collect_data);
1009                 if (unlikely(err)) {
1010                         /* dispose node list */
1011                         nilfs_segctor_apply_buffers(
1012                                 sci, inode, &node_buffers, NULL);
1013                         goto break_or_fail;
1014                 }
1015                 sci->sc_stage.flags |= NILFS_CF_NODE;
1016         }
1017         /* Collect node */
1018         err = nilfs_segctor_apply_buffers(
1019                 sci, inode, &node_buffers, sc_ops->collect_node);
1020         if (unlikely(err))
1021                 goto break_or_fail;
1022
1023         nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1024         err = nilfs_segctor_apply_buffers(
1025                 sci, inode, &node_buffers, sc_ops->collect_bmap);
1026         if (unlikely(err))
1027                 goto break_or_fail;
1028
1029         nilfs_segctor_end_finfo(sci, inode);
1030         sci->sc_stage.flags &= ~NILFS_CF_NODE;
1031
1032  break_or_fail:
1033         return err;
1034 }
1035
1036 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1037                                          struct inode *inode)
1038 {
1039         LIST_HEAD(data_buffers);
1040         size_t n, rest = nilfs_segctor_buffer_rest(sci);
1041         int err;
1042
1043         n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1044                                             sci->sc_dsync_start,
1045                                             sci->sc_dsync_end);
1046
1047         err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1048                                           nilfs_collect_file_data);
1049         if (!err) {
1050                 nilfs_segctor_end_finfo(sci, inode);
1051                 BUG_ON(n > rest);
1052                 /* always receive -E2BIG or true error if n > rest */
1053         }
1054         return err;
1055 }
1056
1057 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1058 {
1059         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1060         struct list_head *head;
1061         struct nilfs_inode_info *ii;
1062         size_t ndone;
1063         int err = 0;
1064
1065         switch (sci->sc_stage.scnt) {
1066         case NILFS_ST_INIT:
1067                 /* Pre-processes */
1068                 sci->sc_stage.flags = 0;
1069
1070                 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1071                         sci->sc_nblk_inc = 0;
1072                         sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1073                         if (mode == SC_LSEG_DSYNC) {
1074                                 sci->sc_stage.scnt = NILFS_ST_DSYNC;
1075                                 goto dsync_mode;
1076                         }
1077                 }
1078
1079                 sci->sc_stage.dirty_file_ptr = NULL;
1080                 sci->sc_stage.gc_inode_ptr = NULL;
1081                 if (mode == SC_FLUSH_DAT) {
1082                         sci->sc_stage.scnt = NILFS_ST_DAT;
1083                         goto dat_stage;
1084                 }
1085                 sci->sc_stage.scnt++;  /* Fall through */
1086         case NILFS_ST_GC:
1087                 if (nilfs_doing_gc()) {
1088                         head = &sci->sc_gc_inodes;
1089                         ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1090                                                 head, i_dirty);
1091                         list_for_each_entry_continue(ii, head, i_dirty) {
1092                                 err = nilfs_segctor_scan_file(
1093                                         sci, &ii->vfs_inode,
1094                                         &nilfs_sc_file_ops);
1095                                 if (unlikely(err)) {
1096                                         sci->sc_stage.gc_inode_ptr = list_entry(
1097                                                 ii->i_dirty.prev,
1098                                                 struct nilfs_inode_info,
1099                                                 i_dirty);
1100                                         goto break_or_fail;
1101                                 }
1102                                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1103                         }
1104                         sci->sc_stage.gc_inode_ptr = NULL;
1105                 }
1106                 sci->sc_stage.scnt++;  /* Fall through */
1107         case NILFS_ST_FILE:
1108                 head = &sci->sc_dirty_files;
1109                 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1110                                         i_dirty);
1111                 list_for_each_entry_continue(ii, head, i_dirty) {
1112                         clear_bit(NILFS_I_DIRTY, &ii->i_state);
1113
1114                         err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1115                                                       &nilfs_sc_file_ops);
1116                         if (unlikely(err)) {
1117                                 sci->sc_stage.dirty_file_ptr =
1118                                         list_entry(ii->i_dirty.prev,
1119                                                    struct nilfs_inode_info,
1120                                                    i_dirty);
1121                                 goto break_or_fail;
1122                         }
1123                         /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1124                         /* XXX: required ? */
1125                 }
1126                 sci->sc_stage.dirty_file_ptr = NULL;
1127                 if (mode == SC_FLUSH_FILE) {
1128                         sci->sc_stage.scnt = NILFS_ST_DONE;
1129                         return 0;
1130                 }
1131                 sci->sc_stage.scnt++;
1132                 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1133                 /* Fall through */
1134         case NILFS_ST_IFILE:
1135                 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1136                                               &nilfs_sc_file_ops);
1137                 if (unlikely(err))
1138                         break;
1139                 sci->sc_stage.scnt++;
1140                 /* Creating a checkpoint */
1141                 err = nilfs_segctor_create_checkpoint(sci);
1142                 if (unlikely(err))
1143                         break;
1144                 /* Fall through */
1145         case NILFS_ST_CPFILE:
1146                 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1147                                               &nilfs_sc_file_ops);
1148                 if (unlikely(err))
1149                         break;
1150                 sci->sc_stage.scnt++;  /* Fall through */
1151         case NILFS_ST_SUFILE:
1152                 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1153                                          sci->sc_nfreesegs, &ndone);
1154                 if (unlikely(err)) {
1155                         nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1156                                                   sci->sc_freesegs, ndone,
1157                                                   NULL);
1158                         break;
1159                 }
1160                 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1161
1162                 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1163                                               &nilfs_sc_file_ops);
1164                 if (unlikely(err))
1165                         break;
1166                 sci->sc_stage.scnt++;  /* Fall through */
1167         case NILFS_ST_DAT:
1168  dat_stage:
1169                 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1170                                               &nilfs_sc_dat_ops);
1171                 if (unlikely(err))
1172                         break;
1173                 if (mode == SC_FLUSH_DAT) {
1174                         sci->sc_stage.scnt = NILFS_ST_DONE;
1175                         return 0;
1176                 }
1177                 sci->sc_stage.scnt++;  /* Fall through */
1178         case NILFS_ST_SR:
1179                 if (mode == SC_LSEG_SR) {
1180                         /* Appending a super root */
1181                         err = nilfs_segctor_add_super_root(sci);
1182                         if (unlikely(err))
1183                                 break;
1184                 }
1185                 /* End of a logical segment */
1186                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1187                 sci->sc_stage.scnt = NILFS_ST_DONE;
1188                 return 0;
1189         case NILFS_ST_DSYNC:
1190  dsync_mode:
1191                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1192                 ii = sci->sc_dsync_inode;
1193                 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1194                         break;
1195
1196                 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1197                 if (unlikely(err))
1198                         break;
1199                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1200                 sci->sc_stage.scnt = NILFS_ST_DONE;
1201                 return 0;
1202         case NILFS_ST_DONE:
1203                 return 0;
1204         default:
1205                 BUG();
1206         }
1207
1208  break_or_fail:
1209         return err;
1210 }
1211
1212 /**
1213  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1214  * @sci: nilfs_sc_info
1215  * @nilfs: nilfs object
1216  */
1217 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1218                                             struct the_nilfs *nilfs)
1219 {
1220         struct nilfs_segment_buffer *segbuf, *prev;
1221         __u64 nextnum;
1222         int err, alloc = 0;
1223
1224         segbuf = nilfs_segbuf_new(sci->sc_super);
1225         if (unlikely(!segbuf))
1226                 return -ENOMEM;
1227
1228         if (list_empty(&sci->sc_write_logs)) {
1229                 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1230                                  nilfs->ns_pseg_offset, nilfs);
1231                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1232                         nilfs_shift_to_next_segment(nilfs);
1233                         nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1234                 }
1235
1236                 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1237                 nextnum = nilfs->ns_nextnum;
1238
1239                 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1240                         /* Start from the head of a new full segment */
1241                         alloc++;
1242         } else {
1243                 /* Continue logs */
1244                 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1245                 nilfs_segbuf_map_cont(segbuf, prev);
1246                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1247                 nextnum = prev->sb_nextnum;
1248
1249                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1250                         nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1251                         segbuf->sb_sum.seg_seq++;
1252                         alloc++;
1253                 }
1254         }
1255
1256         err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1257         if (err)
1258                 goto failed;
1259
1260         if (alloc) {
1261                 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1262                 if (err)
1263                         goto failed;
1264         }
1265         nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1266
1267         BUG_ON(!list_empty(&sci->sc_segbufs));
1268         list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1269         sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1270         return 0;
1271
1272  failed:
1273         nilfs_segbuf_free(segbuf);
1274         return err;
1275 }
1276
1277 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1278                                          struct the_nilfs *nilfs, int nadd)
1279 {
1280         struct nilfs_segment_buffer *segbuf, *prev;
1281         struct inode *sufile = nilfs->ns_sufile;
1282         __u64 nextnextnum;
1283         LIST_HEAD(list);
1284         int err, ret, i;
1285
1286         prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1287         /*
1288          * Since the segment specified with nextnum might be allocated during
1289          * the previous construction, the buffer including its segusage may
1290          * not be dirty.  The following call ensures that the buffer is dirty
1291          * and will pin the buffer on memory until the sufile is written.
1292          */
1293         err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1294         if (unlikely(err))
1295                 return err;
1296
1297         for (i = 0; i < nadd; i++) {
1298                 /* extend segment info */
1299                 err = -ENOMEM;
1300                 segbuf = nilfs_segbuf_new(sci->sc_super);
1301                 if (unlikely(!segbuf))
1302                         goto failed;
1303
1304                 /* map this buffer to region of segment on-disk */
1305                 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1306                 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1307
1308                 /* allocate the next next full segment */
1309                 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1310                 if (unlikely(err))
1311                         goto failed_segbuf;
1312
1313                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1314                 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1315
1316                 list_add_tail(&segbuf->sb_list, &list);
1317                 prev = segbuf;
1318         }
1319         list_splice_tail(&list, &sci->sc_segbufs);
1320         return 0;
1321
1322  failed_segbuf:
1323         nilfs_segbuf_free(segbuf);
1324  failed:
1325         list_for_each_entry(segbuf, &list, sb_list) {
1326                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1327                 WARN_ON(ret); /* never fails */
1328         }
1329         nilfs_destroy_logs(&list);
1330         return err;
1331 }
1332
1333 static void nilfs_free_incomplete_logs(struct list_head *logs,
1334                                        struct the_nilfs *nilfs)
1335 {
1336         struct nilfs_segment_buffer *segbuf, *prev;
1337         struct inode *sufile = nilfs->ns_sufile;
1338         int ret;
1339
1340         segbuf = NILFS_FIRST_SEGBUF(logs);
1341         if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1342                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1343                 WARN_ON(ret); /* never fails */
1344         }
1345         if (atomic_read(&segbuf->sb_err)) {
1346                 /* Case 1: The first segment failed */
1347                 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1348                         /* Case 1a:  Partial segment appended into an existing
1349                            segment */
1350                         nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1351                                                 segbuf->sb_fseg_end);
1352                 else /* Case 1b:  New full segment */
1353                         set_nilfs_discontinued(nilfs);
1354         }
1355
1356         prev = segbuf;
1357         list_for_each_entry_continue(segbuf, logs, sb_list) {
1358                 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1359                         ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1360                         WARN_ON(ret); /* never fails */
1361                 }
1362                 if (atomic_read(&segbuf->sb_err) &&
1363                     segbuf->sb_segnum != nilfs->ns_nextnum)
1364                         /* Case 2: extended segment (!= next) failed */
1365                         nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1366                 prev = segbuf;
1367         }
1368 }
1369
1370 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1371                                           struct inode *sufile)
1372 {
1373         struct nilfs_segment_buffer *segbuf;
1374         unsigned long live_blocks;
1375         int ret;
1376
1377         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1378                 live_blocks = segbuf->sb_sum.nblocks +
1379                         (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1380                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1381                                                      live_blocks,
1382                                                      sci->sc_seg_ctime);
1383                 WARN_ON(ret); /* always succeed because the segusage is dirty */
1384         }
1385 }
1386
1387 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1388 {
1389         struct nilfs_segment_buffer *segbuf;
1390         int ret;
1391
1392         segbuf = NILFS_FIRST_SEGBUF(logs);
1393         ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1394                                              segbuf->sb_pseg_start -
1395                                              segbuf->sb_fseg_start, 0);
1396         WARN_ON(ret); /* always succeed because the segusage is dirty */
1397
1398         list_for_each_entry_continue(segbuf, logs, sb_list) {
1399                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1400                                                      0, 0);
1401                 WARN_ON(ret); /* always succeed */
1402         }
1403 }
1404
1405 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1406                                             struct nilfs_segment_buffer *last,
1407                                             struct inode *sufile)
1408 {
1409         struct nilfs_segment_buffer *segbuf = last;
1410         int ret;
1411
1412         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1413                 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1414                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1415                 WARN_ON(ret);
1416         }
1417         nilfs_truncate_logs(&sci->sc_segbufs, last);
1418 }
1419
1420
1421 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1422                                  struct the_nilfs *nilfs, int mode)
1423 {
1424         struct nilfs_cstage prev_stage = sci->sc_stage;
1425         int err, nadd = 1;
1426
1427         /* Collection retry loop */
1428         for (;;) {
1429                 sci->sc_nblk_this_inc = 0;
1430                 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1431
1432                 err = nilfs_segctor_reset_segment_buffer(sci);
1433                 if (unlikely(err))
1434                         goto failed;
1435
1436                 err = nilfs_segctor_collect_blocks(sci, mode);
1437                 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1438                 if (!err)
1439                         break;
1440
1441                 if (unlikely(err != -E2BIG))
1442                         goto failed;
1443
1444                 /* The current segment is filled up */
1445                 if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
1446                         break;
1447
1448                 nilfs_clear_logs(&sci->sc_segbufs);
1449
1450                 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1451                         err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1452                                                         sci->sc_freesegs,
1453                                                         sci->sc_nfreesegs,
1454                                                         NULL);
1455                         WARN_ON(err); /* do not happen */
1456                         sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1457                 }
1458
1459                 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1460                 if (unlikely(err))
1461                         return err;
1462
1463                 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1464                 sci->sc_stage = prev_stage;
1465         }
1466         nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1467         return 0;
1468
1469  failed:
1470         return err;
1471 }
1472
1473 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1474                                       struct buffer_head *new_bh)
1475 {
1476         BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1477
1478         list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1479         /* The caller must release old_bh */
1480 }
1481
1482 static int
1483 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1484                                      struct nilfs_segment_buffer *segbuf,
1485                                      int mode)
1486 {
1487         struct inode *inode = NULL;
1488         sector_t blocknr;
1489         unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1490         unsigned long nblocks = 0, ndatablk = 0;
1491         struct nilfs_sc_operations *sc_op = NULL;
1492         struct nilfs_segsum_pointer ssp;
1493         struct nilfs_finfo *finfo = NULL;
1494         union nilfs_binfo binfo;
1495         struct buffer_head *bh, *bh_org;
1496         ino_t ino = 0;
1497         int err = 0;
1498
1499         if (!nfinfo)
1500                 goto out;
1501
1502         blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1503         ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1504         ssp.offset = sizeof(struct nilfs_segment_summary);
1505
1506         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1507                 if (bh == segbuf->sb_super_root)
1508                         break;
1509                 if (!finfo) {
1510                         finfo = nilfs_segctor_map_segsum_entry(
1511                                 sci, &ssp, sizeof(*finfo));
1512                         ino = le64_to_cpu(finfo->fi_ino);
1513                         nblocks = le32_to_cpu(finfo->fi_nblocks);
1514                         ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1515
1516                         inode = bh->b_page->mapping->host;
1517
1518                         if (mode == SC_LSEG_DSYNC)
1519                                 sc_op = &nilfs_sc_dsync_ops;
1520                         else if (ino == NILFS_DAT_INO)
1521                                 sc_op = &nilfs_sc_dat_ops;
1522                         else /* file blocks */
1523                                 sc_op = &nilfs_sc_file_ops;
1524                 }
1525                 bh_org = bh;
1526                 get_bh(bh_org);
1527                 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1528                                         &binfo);
1529                 if (bh != bh_org)
1530                         nilfs_list_replace_buffer(bh_org, bh);
1531                 brelse(bh_org);
1532                 if (unlikely(err))
1533                         goto failed_bmap;
1534
1535                 if (ndatablk > 0)
1536                         sc_op->write_data_binfo(sci, &ssp, &binfo);
1537                 else
1538                         sc_op->write_node_binfo(sci, &ssp, &binfo);
1539
1540                 blocknr++;
1541                 if (--nblocks == 0) {
1542                         finfo = NULL;
1543                         if (--nfinfo == 0)
1544                                 break;
1545                 } else if (ndatablk > 0)
1546                         ndatablk--;
1547         }
1548  out:
1549         return 0;
1550
1551  failed_bmap:
1552         return err;
1553 }
1554
1555 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1556 {
1557         struct nilfs_segment_buffer *segbuf;
1558         int err;
1559
1560         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1561                 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1562                 if (unlikely(err))
1563                         return err;
1564                 nilfs_segbuf_fill_in_segsum(segbuf);
1565         }
1566         return 0;
1567 }
1568
1569 static void nilfs_begin_page_io(struct page *page)
1570 {
1571         if (!page || PageWriteback(page))
1572                 /* For split b-tree node pages, this function may be called
1573                    twice.  We ignore the 2nd or later calls by this check. */
1574                 return;
1575
1576         lock_page(page);
1577         clear_page_dirty_for_io(page);
1578         set_page_writeback(page);
1579         unlock_page(page);
1580 }
1581
1582 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1583 {
1584         struct nilfs_segment_buffer *segbuf;
1585         struct page *bd_page = NULL, *fs_page = NULL;
1586
1587         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1588                 struct buffer_head *bh;
1589
1590                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1591                                     b_assoc_buffers) {
1592                         if (bh->b_page != bd_page) {
1593                                 if (bd_page) {
1594                                         lock_page(bd_page);
1595                                         clear_page_dirty_for_io(bd_page);
1596                                         set_page_writeback(bd_page);
1597                                         unlock_page(bd_page);
1598                                 }
1599                                 bd_page = bh->b_page;
1600                         }
1601                 }
1602
1603                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1604                                     b_assoc_buffers) {
1605                         set_buffer_async_write(bh);
1606                         if (bh == segbuf->sb_super_root) {
1607                                 if (bh->b_page != bd_page) {
1608                                         lock_page(bd_page);
1609                                         clear_page_dirty_for_io(bd_page);
1610                                         set_page_writeback(bd_page);
1611                                         unlock_page(bd_page);
1612                                         bd_page = bh->b_page;
1613                                 }
1614                                 break;
1615                         }
1616                         if (bh->b_page != fs_page) {
1617                                 nilfs_begin_page_io(fs_page);
1618                                 fs_page = bh->b_page;
1619                         }
1620                 }
1621         }
1622         if (bd_page) {
1623                 lock_page(bd_page);
1624                 clear_page_dirty_for_io(bd_page);
1625                 set_page_writeback(bd_page);
1626                 unlock_page(bd_page);
1627         }
1628         nilfs_begin_page_io(fs_page);
1629 }
1630
1631 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1632                                struct the_nilfs *nilfs)
1633 {
1634         int ret;
1635
1636         ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1637         list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1638         return ret;
1639 }
1640
1641 static void nilfs_end_page_io(struct page *page, int err)
1642 {
1643         if (!page)
1644                 return;
1645
1646         if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1647                 /*
1648                  * For b-tree node pages, this function may be called twice
1649                  * or more because they might be split in a segment.
1650                  */
1651                 if (PageDirty(page)) {
1652                         /*
1653                          * For pages holding split b-tree node buffers, dirty
1654                          * flag on the buffers may be cleared discretely.
1655                          * In that case, the page is once redirtied for
1656                          * remaining buffers, and it must be cancelled if
1657                          * all the buffers get cleaned later.
1658                          */
1659                         lock_page(page);
1660                         if (nilfs_page_buffers_clean(page))
1661                                 __nilfs_clear_page_dirty(page);
1662                         unlock_page(page);
1663                 }
1664                 return;
1665         }
1666
1667         if (!err) {
1668                 if (!nilfs_page_buffers_clean(page))
1669                         __set_page_dirty_nobuffers(page);
1670                 ClearPageError(page);
1671         } else {
1672                 __set_page_dirty_nobuffers(page);
1673                 SetPageError(page);
1674         }
1675
1676         end_page_writeback(page);
1677 }
1678
1679 static void nilfs_abort_logs(struct list_head *logs, int err)
1680 {
1681         struct nilfs_segment_buffer *segbuf;
1682         struct page *bd_page = NULL, *fs_page = NULL;
1683         struct buffer_head *bh;
1684
1685         if (list_empty(logs))
1686                 return;
1687
1688         list_for_each_entry(segbuf, logs, sb_list) {
1689                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1690                                     b_assoc_buffers) {
1691                         if (bh->b_page != bd_page) {
1692                                 if (bd_page)
1693                                         end_page_writeback(bd_page);
1694                                 bd_page = bh->b_page;
1695                         }
1696                 }
1697
1698                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1699                                     b_assoc_buffers) {
1700                         clear_buffer_async_write(bh);
1701                         if (bh == segbuf->sb_super_root) {
1702                                 if (bh->b_page != bd_page) {
1703                                         end_page_writeback(bd_page);
1704                                         bd_page = bh->b_page;
1705                                 }
1706                                 break;
1707                         }
1708                         if (bh->b_page != fs_page) {
1709                                 nilfs_end_page_io(fs_page, err);
1710                                 fs_page = bh->b_page;
1711                         }
1712                 }
1713         }
1714         if (bd_page)
1715                 end_page_writeback(bd_page);
1716
1717         nilfs_end_page_io(fs_page, err);
1718 }
1719
1720 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1721                                              struct the_nilfs *nilfs, int err)
1722 {
1723         LIST_HEAD(logs);
1724         int ret;
1725
1726         list_splice_tail_init(&sci->sc_write_logs, &logs);
1727         ret = nilfs_wait_on_logs(&logs);
1728         nilfs_abort_logs(&logs, ret ? : err);
1729
1730         list_splice_tail_init(&sci->sc_segbufs, &logs);
1731         nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1732         nilfs_free_incomplete_logs(&logs, nilfs);
1733
1734         if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1735                 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1736                                                 sci->sc_freesegs,
1737                                                 sci->sc_nfreesegs,
1738                                                 NULL);
1739                 WARN_ON(ret); /* do not happen */
1740         }
1741
1742         nilfs_destroy_logs(&logs);
1743 }
1744
1745 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1746                                    struct nilfs_segment_buffer *segbuf)
1747 {
1748         nilfs->ns_segnum = segbuf->sb_segnum;
1749         nilfs->ns_nextnum = segbuf->sb_nextnum;
1750         nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1751                 + segbuf->sb_sum.nblocks;
1752         nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1753         nilfs->ns_ctime = segbuf->sb_sum.ctime;
1754 }
1755
1756 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1757 {
1758         struct nilfs_segment_buffer *segbuf;
1759         struct page *bd_page = NULL, *fs_page = NULL;
1760         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1761         int update_sr = false;
1762
1763         list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1764                 struct buffer_head *bh;
1765
1766                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1767                                     b_assoc_buffers) {
1768                         set_buffer_uptodate(bh);
1769                         clear_buffer_dirty(bh);
1770                         if (bh->b_page != bd_page) {
1771                                 if (bd_page)
1772                                         end_page_writeback(bd_page);
1773                                 bd_page = bh->b_page;
1774                         }
1775                 }
1776                 /*
1777                  * We assume that the buffers which belong to the same page
1778                  * continue over the buffer list.
1779                  * Under this assumption, the last BHs of pages is
1780                  * identifiable by the discontinuity of bh->b_page
1781                  * (page != fs_page).
1782                  *
1783                  * For B-tree node blocks, however, this assumption is not
1784                  * guaranteed.  The cleanup code of B-tree node pages needs
1785                  * special care.
1786                  */
1787                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1788                                     b_assoc_buffers) {
1789                         const unsigned long set_bits = (1 << BH_Uptodate);
1790                         const unsigned long clear_bits =
1791                                 (1 << BH_Dirty | 1 << BH_Async_Write |
1792                                  1 << BH_Delay | 1 << BH_NILFS_Volatile |
1793                                  1 << BH_NILFS_Redirected);
1794
1795                         set_mask_bits(&bh->b_state, clear_bits, set_bits);
1796                         if (bh == segbuf->sb_super_root) {
1797                                 if (bh->b_page != bd_page) {
1798                                         end_page_writeback(bd_page);
1799                                         bd_page = bh->b_page;
1800                                 }
1801                                 update_sr = true;
1802                                 break;
1803                         }
1804                         if (bh->b_page != fs_page) {
1805                                 nilfs_end_page_io(fs_page, 0);
1806                                 fs_page = bh->b_page;
1807                         }
1808                 }
1809
1810                 if (!nilfs_segbuf_simplex(segbuf)) {
1811                         if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1812                                 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1813                                 sci->sc_lseg_stime = jiffies;
1814                         }
1815                         if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1816                                 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1817                 }
1818         }
1819         /*
1820          * Since pages may continue over multiple segment buffers,
1821          * end of the last page must be checked outside of the loop.
1822          */
1823         if (bd_page)
1824                 end_page_writeback(bd_page);
1825
1826         nilfs_end_page_io(fs_page, 0);
1827
1828         nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1829
1830         if (nilfs_doing_gc())
1831                 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1832         else
1833                 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1834
1835         sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1836
1837         segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1838         nilfs_set_next_segment(nilfs, segbuf);
1839
1840         if (update_sr) {
1841                 nilfs->ns_flushed_device = 0;
1842                 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1843                                        segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1844
1845                 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1846                 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1847                 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1848                 nilfs_segctor_clear_metadata_dirty(sci);
1849         } else
1850                 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1851 }
1852
1853 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1854 {
1855         int ret;
1856
1857         ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1858         if (!ret) {
1859                 nilfs_segctor_complete_write(sci);
1860                 nilfs_destroy_logs(&sci->sc_write_logs);
1861         }
1862         return ret;
1863 }
1864
1865 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1866                                              struct the_nilfs *nilfs)
1867 {
1868         struct nilfs_inode_info *ii, *n;
1869         struct inode *ifile = sci->sc_root->ifile;
1870
1871         spin_lock(&nilfs->ns_inode_lock);
1872  retry:
1873         list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1874                 if (!ii->i_bh) {
1875                         struct buffer_head *ibh;
1876                         int err;
1877
1878                         spin_unlock(&nilfs->ns_inode_lock);
1879                         err = nilfs_ifile_get_inode_block(
1880                                 ifile, ii->vfs_inode.i_ino, &ibh);
1881                         if (unlikely(err)) {
1882                                 nilfs_warning(sci->sc_super, __func__,
1883                                               "failed to get inode block.\n");
1884                                 return err;
1885                         }
1886                         mark_buffer_dirty(ibh);
1887                         nilfs_mdt_mark_dirty(ifile);
1888                         spin_lock(&nilfs->ns_inode_lock);
1889                         if (likely(!ii->i_bh))
1890                                 ii->i_bh = ibh;
1891                         else
1892                                 brelse(ibh);
1893                         goto retry;
1894                 }
1895
1896                 clear_bit(NILFS_I_QUEUED, &ii->i_state);
1897                 set_bit(NILFS_I_BUSY, &ii->i_state);
1898                 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
1899         }
1900         spin_unlock(&nilfs->ns_inode_lock);
1901
1902         return 0;
1903 }
1904
1905 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
1906                                              struct the_nilfs *nilfs)
1907 {
1908         struct nilfs_inode_info *ii, *n;
1909         int during_mount = !(sci->sc_super->s_flags & MS_ACTIVE);
1910         int defer_iput = false;
1911
1912         spin_lock(&nilfs->ns_inode_lock);
1913         list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
1914                 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
1915                     test_bit(NILFS_I_DIRTY, &ii->i_state))
1916                         continue;
1917
1918                 clear_bit(NILFS_I_BUSY, &ii->i_state);
1919                 brelse(ii->i_bh);
1920                 ii->i_bh = NULL;
1921                 list_del_init(&ii->i_dirty);
1922                 if (!ii->vfs_inode.i_nlink || during_mount) {
1923                         /*
1924                          * Defer calling iput() to avoid deadlocks if
1925                          * i_nlink == 0 or mount is not yet finished.
1926                          */
1927                         list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
1928                         defer_iput = true;
1929                 } else {
1930                         spin_unlock(&nilfs->ns_inode_lock);
1931                         iput(&ii->vfs_inode);
1932                         spin_lock(&nilfs->ns_inode_lock);
1933                 }
1934         }
1935         spin_unlock(&nilfs->ns_inode_lock);
1936
1937         if (defer_iput)
1938                 schedule_work(&sci->sc_iput_work);
1939 }
1940
1941 /*
1942  * Main procedure of segment constructor
1943  */
1944 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
1945 {
1946         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1947         int err;
1948
1949         sci->sc_stage.scnt = NILFS_ST_INIT;
1950         sci->sc_cno = nilfs->ns_cno;
1951
1952         err = nilfs_segctor_collect_dirty_files(sci, nilfs);
1953         if (unlikely(err))
1954                 goto out;
1955
1956         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
1957                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1958
1959         if (nilfs_segctor_clean(sci))
1960                 goto out;
1961
1962         do {
1963                 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
1964
1965                 err = nilfs_segctor_begin_construction(sci, nilfs);
1966                 if (unlikely(err))
1967                         goto out;
1968
1969                 /* Update time stamp */
1970                 sci->sc_seg_ctime = get_seconds();
1971
1972                 err = nilfs_segctor_collect(sci, nilfs, mode);
1973                 if (unlikely(err))
1974                         goto failed;
1975
1976                 /* Avoid empty segment */
1977                 if (sci->sc_stage.scnt == NILFS_ST_DONE &&
1978                     nilfs_segbuf_empty(sci->sc_curseg)) {
1979                         nilfs_segctor_abort_construction(sci, nilfs, 1);
1980                         goto out;
1981                 }
1982
1983                 err = nilfs_segctor_assign(sci, mode);
1984                 if (unlikely(err))
1985                         goto failed;
1986
1987                 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
1988                         nilfs_segctor_fill_in_file_bmap(sci);
1989
1990                 if (mode == SC_LSEG_SR &&
1991                     sci->sc_stage.scnt >= NILFS_ST_CPFILE) {
1992                         err = nilfs_segctor_fill_in_checkpoint(sci);
1993                         if (unlikely(err))
1994                                 goto failed_to_write;
1995
1996                         nilfs_segctor_fill_in_super_root(sci, nilfs);
1997                 }
1998                 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
1999
2000                 /* Write partial segments */
2001                 nilfs_segctor_prepare_write(sci);
2002
2003                 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2004                                             nilfs->ns_crc_seed);
2005
2006                 err = nilfs_segctor_write(sci, nilfs);
2007                 if (unlikely(err))
2008                         goto failed_to_write;
2009
2010                 if (sci->sc_stage.scnt == NILFS_ST_DONE ||
2011                     nilfs->ns_blocksize_bits != PAGE_CACHE_SHIFT) {
2012                         /*
2013                          * At this point, we avoid double buffering
2014                          * for blocksize < pagesize because page dirty
2015                          * flag is turned off during write and dirty
2016                          * buffers are not properly collected for
2017                          * pages crossing over segments.
2018                          */
2019                         err = nilfs_segctor_wait(sci);
2020                         if (err)
2021                                 goto failed_to_write;
2022                 }
2023         } while (sci->sc_stage.scnt != NILFS_ST_DONE);
2024
2025  out:
2026         nilfs_segctor_drop_written_files(sci, nilfs);
2027         return err;
2028
2029  failed_to_write:
2030         if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2031                 nilfs_redirty_inodes(&sci->sc_dirty_files);
2032
2033  failed:
2034         if (nilfs_doing_gc())
2035                 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2036         nilfs_segctor_abort_construction(sci, nilfs, err);
2037         goto out;
2038 }
2039
2040 /**
2041  * nilfs_segctor_start_timer - set timer of background write
2042  * @sci: nilfs_sc_info
2043  *
2044  * If the timer has already been set, it ignores the new request.
2045  * This function MUST be called within a section locking the segment
2046  * semaphore.
2047  */
2048 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2049 {
2050         spin_lock(&sci->sc_state_lock);
2051         if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2052                 sci->sc_timer.expires = jiffies + sci->sc_interval;
2053                 add_timer(&sci->sc_timer);
2054                 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2055         }
2056         spin_unlock(&sci->sc_state_lock);
2057 }
2058
2059 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2060 {
2061         spin_lock(&sci->sc_state_lock);
2062         if (!(sci->sc_flush_request & (1 << bn))) {
2063                 unsigned long prev_req = sci->sc_flush_request;
2064
2065                 sci->sc_flush_request |= (1 << bn);
2066                 if (!prev_req)
2067                         wake_up(&sci->sc_wait_daemon);
2068         }
2069         spin_unlock(&sci->sc_state_lock);
2070 }
2071
2072 /**
2073  * nilfs_flush_segment - trigger a segment construction for resource control
2074  * @sb: super block
2075  * @ino: inode number of the file to be flushed out.
2076  */
2077 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2078 {
2079         struct the_nilfs *nilfs = sb->s_fs_info;
2080         struct nilfs_sc_info *sci = nilfs->ns_writer;
2081
2082         if (!sci || nilfs_doing_construction())
2083                 return;
2084         nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2085                                         /* assign bit 0 to data files */
2086 }
2087
2088 struct nilfs_segctor_wait_request {
2089         wait_queue_t    wq;
2090         __u32           seq;
2091         int             err;
2092         atomic_t        done;
2093 };
2094
2095 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2096 {
2097         struct nilfs_segctor_wait_request wait_req;
2098         int err = 0;
2099
2100         spin_lock(&sci->sc_state_lock);
2101         init_wait(&wait_req.wq);
2102         wait_req.err = 0;
2103         atomic_set(&wait_req.done, 0);
2104         wait_req.seq = ++sci->sc_seq_request;
2105         spin_unlock(&sci->sc_state_lock);
2106
2107         init_waitqueue_entry(&wait_req.wq, current);
2108         add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2109         set_current_state(TASK_INTERRUPTIBLE);
2110         wake_up(&sci->sc_wait_daemon);
2111
2112         for (;;) {
2113                 if (atomic_read(&wait_req.done)) {
2114                         err = wait_req.err;
2115                         break;
2116                 }
2117                 if (!signal_pending(current)) {
2118                         schedule();
2119                         continue;
2120                 }
2121                 err = -ERESTARTSYS;
2122                 break;
2123         }
2124         finish_wait(&sci->sc_wait_request, &wait_req.wq);
2125         return err;
2126 }
2127
2128 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2129 {
2130         struct nilfs_segctor_wait_request *wrq, *n;
2131         unsigned long flags;
2132
2133         spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2134         list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2135                                  wq.task_list) {
2136                 if (!atomic_read(&wrq->done) &&
2137                     nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2138                         wrq->err = err;
2139                         atomic_set(&wrq->done, 1);
2140                 }
2141                 if (atomic_read(&wrq->done)) {
2142                         wrq->wq.func(&wrq->wq,
2143                                      TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2144                                      0, NULL);
2145                 }
2146         }
2147         spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2148 }
2149
2150 /**
2151  * nilfs_construct_segment - construct a logical segment
2152  * @sb: super block
2153  *
2154  * Return Value: On success, 0 is retured. On errors, one of the following
2155  * negative error code is returned.
2156  *
2157  * %-EROFS - Read only filesystem.
2158  *
2159  * %-EIO - I/O error
2160  *
2161  * %-ENOSPC - No space left on device (only in a panic state).
2162  *
2163  * %-ERESTARTSYS - Interrupted.
2164  *
2165  * %-ENOMEM - Insufficient memory available.
2166  */
2167 int nilfs_construct_segment(struct super_block *sb)
2168 {
2169         struct the_nilfs *nilfs = sb->s_fs_info;
2170         struct nilfs_sc_info *sci = nilfs->ns_writer;
2171         struct nilfs_transaction_info *ti;
2172         int err;
2173
2174         if (!sci)
2175                 return -EROFS;
2176
2177         /* A call inside transactions causes a deadlock. */
2178         BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2179
2180         err = nilfs_segctor_sync(sci);
2181         return err;
2182 }
2183
2184 /**
2185  * nilfs_construct_dsync_segment - construct a data-only logical segment
2186  * @sb: super block
2187  * @inode: inode whose data blocks should be written out
2188  * @start: start byte offset
2189  * @end: end byte offset (inclusive)
2190  *
2191  * Return Value: On success, 0 is retured. On errors, one of the following
2192  * negative error code is returned.
2193  *
2194  * %-EROFS - Read only filesystem.
2195  *
2196  * %-EIO - I/O error
2197  *
2198  * %-ENOSPC - No space left on device (only in a panic state).
2199  *
2200  * %-ERESTARTSYS - Interrupted.
2201  *
2202  * %-ENOMEM - Insufficient memory available.
2203  */
2204 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2205                                   loff_t start, loff_t end)
2206 {
2207         struct the_nilfs *nilfs = sb->s_fs_info;
2208         struct nilfs_sc_info *sci = nilfs->ns_writer;
2209         struct nilfs_inode_info *ii;
2210         struct nilfs_transaction_info ti;
2211         int err = 0;
2212
2213         if (!sci)
2214                 return -EROFS;
2215
2216         nilfs_transaction_lock(sb, &ti, 0);
2217
2218         ii = NILFS_I(inode);
2219         if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2220             nilfs_test_opt(nilfs, STRICT_ORDER) ||
2221             test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2222             nilfs_discontinued(nilfs)) {
2223                 nilfs_transaction_unlock(sb);
2224                 err = nilfs_segctor_sync(sci);
2225                 return err;
2226         }
2227
2228         spin_lock(&nilfs->ns_inode_lock);
2229         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2230             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2231                 spin_unlock(&nilfs->ns_inode_lock);
2232                 nilfs_transaction_unlock(sb);
2233                 return 0;
2234         }
2235         spin_unlock(&nilfs->ns_inode_lock);
2236         sci->sc_dsync_inode = ii;
2237         sci->sc_dsync_start = start;
2238         sci->sc_dsync_end = end;
2239
2240         err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2241         if (!err)
2242                 nilfs->ns_flushed_device = 0;
2243
2244         nilfs_transaction_unlock(sb);
2245         return err;
2246 }
2247
2248 #define FLUSH_FILE_BIT  (0x1) /* data file only */
2249 #define FLUSH_DAT_BIT   (1 << NILFS_DAT_INO) /* DAT only */
2250
2251 /**
2252  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2253  * @sci: segment constructor object
2254  */
2255 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2256 {
2257         spin_lock(&sci->sc_state_lock);
2258         sci->sc_seq_accepted = sci->sc_seq_request;
2259         spin_unlock(&sci->sc_state_lock);
2260         del_timer_sync(&sci->sc_timer);
2261 }
2262
2263 /**
2264  * nilfs_segctor_notify - notify the result of request to caller threads
2265  * @sci: segment constructor object
2266  * @mode: mode of log forming
2267  * @err: error code to be notified
2268  */
2269 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2270 {
2271         /* Clear requests (even when the construction failed) */
2272         spin_lock(&sci->sc_state_lock);
2273
2274         if (mode == SC_LSEG_SR) {
2275                 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2276                 sci->sc_seq_done = sci->sc_seq_accepted;
2277                 nilfs_segctor_wakeup(sci, err);
2278                 sci->sc_flush_request = 0;
2279         } else {
2280                 if (mode == SC_FLUSH_FILE)
2281                         sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2282                 else if (mode == SC_FLUSH_DAT)
2283                         sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2284
2285                 /* re-enable timer if checkpoint creation was not done */
2286                 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2287                     time_before(jiffies, sci->sc_timer.expires))
2288                         add_timer(&sci->sc_timer);
2289         }
2290         spin_unlock(&sci->sc_state_lock);
2291 }
2292
2293 /**
2294  * nilfs_segctor_construct - form logs and write them to disk
2295  * @sci: segment constructor object
2296  * @mode: mode of log forming
2297  */
2298 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2299 {
2300         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2301         struct nilfs_super_block **sbp;
2302         int err = 0;
2303
2304         nilfs_segctor_accept(sci);
2305
2306         if (nilfs_discontinued(nilfs))
2307                 mode = SC_LSEG_SR;
2308         if (!nilfs_segctor_confirm(sci))
2309                 err = nilfs_segctor_do_construct(sci, mode);
2310
2311         if (likely(!err)) {
2312                 if (mode != SC_FLUSH_DAT)
2313                         atomic_set(&nilfs->ns_ndirtyblks, 0);
2314                 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2315                     nilfs_discontinued(nilfs)) {
2316                         down_write(&nilfs->ns_sem);
2317                         err = -EIO;
2318                         sbp = nilfs_prepare_super(sci->sc_super,
2319                                                   nilfs_sb_will_flip(nilfs));
2320                         if (likely(sbp)) {
2321                                 nilfs_set_log_cursor(sbp[0], nilfs);
2322                                 err = nilfs_commit_super(sci->sc_super,
2323                                                          NILFS_SB_COMMIT);
2324                         }
2325                         up_write(&nilfs->ns_sem);
2326                 }
2327         }
2328
2329         nilfs_segctor_notify(sci, mode, err);
2330         return err;
2331 }
2332
2333 static void nilfs_construction_timeout(unsigned long data)
2334 {
2335         struct task_struct *p = (struct task_struct *)data;
2336         wake_up_process(p);
2337 }
2338
2339 static void
2340 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2341 {
2342         struct nilfs_inode_info *ii, *n;
2343
2344         list_for_each_entry_safe(ii, n, head, i_dirty) {
2345                 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2346                         continue;
2347                 list_del_init(&ii->i_dirty);
2348                 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2349                 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
2350                 iput(&ii->vfs_inode);
2351         }
2352 }
2353
2354 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2355                          void **kbufs)
2356 {
2357         struct the_nilfs *nilfs = sb->s_fs_info;
2358         struct nilfs_sc_info *sci = nilfs->ns_writer;
2359         struct nilfs_transaction_info ti;
2360         int err;
2361
2362         if (unlikely(!sci))
2363                 return -EROFS;
2364
2365         nilfs_transaction_lock(sb, &ti, 1);
2366
2367         err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2368         if (unlikely(err))
2369                 goto out_unlock;
2370
2371         err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2372         if (unlikely(err)) {
2373                 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2374                 goto out_unlock;
2375         }
2376
2377         sci->sc_freesegs = kbufs[4];
2378         sci->sc_nfreesegs = argv[4].v_nmembs;
2379         list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2380
2381         for (;;) {
2382                 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2383                 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2384
2385                 if (likely(!err))
2386                         break;
2387
2388                 nilfs_warning(sb, __func__,
2389                               "segment construction failed. (err=%d)", err);
2390                 set_current_state(TASK_INTERRUPTIBLE);
2391                 schedule_timeout(sci->sc_interval);
2392         }
2393         if (nilfs_test_opt(nilfs, DISCARD)) {
2394                 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2395                                                  sci->sc_nfreesegs);
2396                 if (ret) {
2397                         printk(KERN_WARNING
2398                                "NILFS warning: error %d on discard request, "
2399                                "turning discards off for the device\n", ret);
2400                         nilfs_clear_opt(nilfs, DISCARD);
2401                 }
2402         }
2403
2404  out_unlock:
2405         sci->sc_freesegs = NULL;
2406         sci->sc_nfreesegs = 0;
2407         nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2408         nilfs_transaction_unlock(sb);
2409         return err;
2410 }
2411
2412 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2413 {
2414         struct nilfs_transaction_info ti;
2415
2416         nilfs_transaction_lock(sci->sc_super, &ti, 0);
2417         nilfs_segctor_construct(sci, mode);
2418
2419         /*
2420          * Unclosed segment should be retried.  We do this using sc_timer.
2421          * Timeout of sc_timer will invoke complete construction which leads
2422          * to close the current logical segment.
2423          */
2424         if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2425                 nilfs_segctor_start_timer(sci);
2426
2427         nilfs_transaction_unlock(sci->sc_super);
2428 }
2429
2430 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2431 {
2432         int mode = 0;
2433         int err;
2434
2435         spin_lock(&sci->sc_state_lock);
2436         mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2437                 SC_FLUSH_DAT : SC_FLUSH_FILE;
2438         spin_unlock(&sci->sc_state_lock);
2439
2440         if (mode) {
2441                 err = nilfs_segctor_do_construct(sci, mode);
2442
2443                 spin_lock(&sci->sc_state_lock);
2444                 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2445                         ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2446                 spin_unlock(&sci->sc_state_lock);
2447         }
2448         clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2449 }
2450
2451 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2452 {
2453         if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2454             time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2455                 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2456                         return SC_FLUSH_FILE;
2457                 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2458                         return SC_FLUSH_DAT;
2459         }
2460         return SC_LSEG_SR;
2461 }
2462
2463 /**
2464  * nilfs_segctor_thread - main loop of the segment constructor thread.
2465  * @arg: pointer to a struct nilfs_sc_info.
2466  *
2467  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2468  * to execute segment constructions.
2469  */
2470 static int nilfs_segctor_thread(void *arg)
2471 {
2472         struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2473         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2474         int timeout = 0;
2475
2476         sci->sc_timer.data = (unsigned long)current;
2477         sci->sc_timer.function = nilfs_construction_timeout;
2478
2479         /* start sync. */
2480         sci->sc_task = current;
2481         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2482         printk(KERN_INFO
2483                "segctord starting. Construction interval = %lu seconds, "
2484                "CP frequency < %lu seconds\n",
2485                sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2486
2487         spin_lock(&sci->sc_state_lock);
2488  loop:
2489         for (;;) {
2490                 int mode;
2491
2492                 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2493                         goto end_thread;
2494
2495                 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2496                         mode = SC_LSEG_SR;
2497                 else if (!sci->sc_flush_request)
2498                         break;
2499                 else
2500                         mode = nilfs_segctor_flush_mode(sci);
2501
2502                 spin_unlock(&sci->sc_state_lock);
2503                 nilfs_segctor_thread_construct(sci, mode);
2504                 spin_lock(&sci->sc_state_lock);
2505                 timeout = 0;
2506         }
2507
2508
2509         if (freezing(current)) {
2510                 spin_unlock(&sci->sc_state_lock);
2511                 try_to_freeze();
2512                 spin_lock(&sci->sc_state_lock);
2513         } else {
2514                 DEFINE_WAIT(wait);
2515                 int should_sleep = 1;
2516
2517                 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2518                                 TASK_INTERRUPTIBLE);
2519
2520                 if (sci->sc_seq_request != sci->sc_seq_done)
2521                         should_sleep = 0;
2522                 else if (sci->sc_flush_request)
2523                         should_sleep = 0;
2524                 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2525                         should_sleep = time_before(jiffies,
2526                                         sci->sc_timer.expires);
2527
2528                 if (should_sleep) {
2529                         spin_unlock(&sci->sc_state_lock);
2530                         schedule();
2531                         spin_lock(&sci->sc_state_lock);
2532                 }
2533                 finish_wait(&sci->sc_wait_daemon, &wait);
2534                 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2535                            time_after_eq(jiffies, sci->sc_timer.expires));
2536
2537                 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2538                         set_nilfs_discontinued(nilfs);
2539         }
2540         goto loop;
2541
2542  end_thread:
2543         spin_unlock(&sci->sc_state_lock);
2544
2545         /* end sync. */
2546         sci->sc_task = NULL;
2547         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2548         return 0;
2549 }
2550
2551 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2552 {
2553         struct task_struct *t;
2554
2555         t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2556         if (IS_ERR(t)) {
2557                 int err = PTR_ERR(t);
2558
2559                 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2560                        err);
2561                 return err;
2562         }
2563         wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2564         return 0;
2565 }
2566
2567 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2568         __acquires(&sci->sc_state_lock)
2569         __releases(&sci->sc_state_lock)
2570 {
2571         sci->sc_state |= NILFS_SEGCTOR_QUIT;
2572
2573         while (sci->sc_task) {
2574                 wake_up(&sci->sc_wait_daemon);
2575                 spin_unlock(&sci->sc_state_lock);
2576                 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2577                 spin_lock(&sci->sc_state_lock);
2578         }
2579 }
2580
2581 /*
2582  * Setup & clean-up functions
2583  */
2584 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2585                                                struct nilfs_root *root)
2586 {
2587         struct the_nilfs *nilfs = sb->s_fs_info;
2588         struct nilfs_sc_info *sci;
2589
2590         sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2591         if (!sci)
2592                 return NULL;
2593
2594         sci->sc_super = sb;
2595
2596         nilfs_get_root(root);
2597         sci->sc_root = root;
2598
2599         init_waitqueue_head(&sci->sc_wait_request);
2600         init_waitqueue_head(&sci->sc_wait_daemon);
2601         init_waitqueue_head(&sci->sc_wait_task);
2602         spin_lock_init(&sci->sc_state_lock);
2603         INIT_LIST_HEAD(&sci->sc_dirty_files);
2604         INIT_LIST_HEAD(&sci->sc_segbufs);
2605         INIT_LIST_HEAD(&sci->sc_write_logs);
2606         INIT_LIST_HEAD(&sci->sc_gc_inodes);
2607         INIT_LIST_HEAD(&sci->sc_iput_queue);
2608         INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2609         init_timer(&sci->sc_timer);
2610
2611         sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2612         sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2613         sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2614
2615         if (nilfs->ns_interval)
2616                 sci->sc_interval = HZ * nilfs->ns_interval;
2617         if (nilfs->ns_watermark)
2618                 sci->sc_watermark = nilfs->ns_watermark;
2619         return sci;
2620 }
2621
2622 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2623 {
2624         int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2625
2626         /* The segctord thread was stopped and its timer was removed.
2627            But some tasks remain. */
2628         do {
2629                 struct nilfs_transaction_info ti;
2630
2631                 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2632                 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2633                 nilfs_transaction_unlock(sci->sc_super);
2634
2635                 flush_work(&sci->sc_iput_work);
2636
2637         } while (ret && retrycount-- > 0);
2638 }
2639
2640 /**
2641  * nilfs_segctor_destroy - destroy the segment constructor.
2642  * @sci: nilfs_sc_info
2643  *
2644  * nilfs_segctor_destroy() kills the segctord thread and frees
2645  * the nilfs_sc_info struct.
2646  * Caller must hold the segment semaphore.
2647  */
2648 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2649 {
2650         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2651         int flag;
2652
2653         up_write(&nilfs->ns_segctor_sem);
2654
2655         spin_lock(&sci->sc_state_lock);
2656         nilfs_segctor_kill_thread(sci);
2657         flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2658                 || sci->sc_seq_request != sci->sc_seq_done);
2659         spin_unlock(&sci->sc_state_lock);
2660
2661         if (flush_work(&sci->sc_iput_work))
2662                 flag = true;
2663
2664         if (flag || !nilfs_segctor_confirm(sci))
2665                 nilfs_segctor_write_out(sci);
2666
2667         if (!list_empty(&sci->sc_dirty_files)) {
2668                 nilfs_warning(sci->sc_super, __func__,
2669                               "dirty file(s) after the final construction\n");
2670                 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2671         }
2672
2673         if (!list_empty(&sci->sc_iput_queue)) {
2674                 nilfs_warning(sci->sc_super, __func__,
2675                               "iput queue is not empty\n");
2676                 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2677         }
2678
2679         WARN_ON(!list_empty(&sci->sc_segbufs));
2680         WARN_ON(!list_empty(&sci->sc_write_logs));
2681
2682         nilfs_put_root(sci->sc_root);
2683
2684         down_write(&nilfs->ns_segctor_sem);
2685
2686         del_timer_sync(&sci->sc_timer);
2687         kfree(sci);
2688 }
2689
2690 /**
2691  * nilfs_attach_log_writer - attach log writer
2692  * @sb: super block instance
2693  * @root: root object of the current filesystem tree
2694  *
2695  * This allocates a log writer object, initializes it, and starts the
2696  * log writer.
2697  *
2698  * Return Value: On success, 0 is returned. On error, one of the following
2699  * negative error code is returned.
2700  *
2701  * %-ENOMEM - Insufficient memory available.
2702  */
2703 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2704 {
2705         struct the_nilfs *nilfs = sb->s_fs_info;
2706         int err;
2707
2708         if (nilfs->ns_writer) {
2709                 /*
2710                  * This happens if the filesystem was remounted
2711                  * read/write after nilfs_error degenerated it into a
2712                  * read-only mount.
2713                  */
2714                 nilfs_detach_log_writer(sb);
2715         }
2716
2717         nilfs->ns_writer = nilfs_segctor_new(sb, root);
2718         if (!nilfs->ns_writer)
2719                 return -ENOMEM;
2720
2721         err = nilfs_segctor_start_thread(nilfs->ns_writer);
2722         if (err) {
2723                 kfree(nilfs->ns_writer);
2724                 nilfs->ns_writer = NULL;
2725         }
2726         return err;
2727 }
2728
2729 /**
2730  * nilfs_detach_log_writer - destroy log writer
2731  * @sb: super block instance
2732  *
2733  * This kills log writer daemon, frees the log writer object, and
2734  * destroys list of dirty files.
2735  */
2736 void nilfs_detach_log_writer(struct super_block *sb)
2737 {
2738         struct the_nilfs *nilfs = sb->s_fs_info;
2739         LIST_HEAD(garbage_list);
2740
2741         down_write(&nilfs->ns_segctor_sem);
2742         if (nilfs->ns_writer) {
2743                 nilfs_segctor_destroy(nilfs->ns_writer);
2744                 nilfs->ns_writer = NULL;
2745         }
2746
2747         /* Force to free the list of dirty files */
2748         spin_lock(&nilfs->ns_inode_lock);
2749         if (!list_empty(&nilfs->ns_dirty_files)) {
2750                 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2751                 nilfs_warning(sb, __func__,
2752                               "Hit dirty file after stopped log writer\n");
2753         }
2754         spin_unlock(&nilfs->ns_inode_lock);
2755         up_write(&nilfs->ns_segctor_sem);
2756
2757         nilfs_dispose_list(nilfs, &garbage_list, 1);
2758 }