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