fs/nilfs2/segment.c

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