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) || buffer_async_write(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                                                 !buffer_async_write(bh)) {
 701                                         get_bh(bh);
 702                                         list_add_tail(&bh->b_assoc_buffers,
 703                                                       listp);
 704                                 }
 705                                 bh = bh->b_this_page;
 706                         } while (bh != head);
 707                 }
 708                 pagevec_release(&pvec);
 709                 cond_resched();
 710         }
 711 }
 712
 713 static void nilfs_dispose_list(struct the_nilfs *nilfs,
 714                                struct list_head *head, int force)
 715 {
 716         struct nilfs_inode_info *ii, *n;
 717         struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
 718         unsigned nv = 0;
 719
 720         while (!list_empty(head)) {
 721                 spin_lock(&nilfs->ns_inode_lock);
 722                 list_for_each_entry_safe(ii, n, head, i_dirty) {
 723                         list_del_init(&ii->i_dirty);
 724                         if (force) {
 725                                 if (unlikely(ii->i_bh)) {
 726                                         brelse(ii->i_bh);
 727                                         ii->i_bh = NULL;
 728                                 }
 729                         } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
 730                                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 731                                 list_add_tail(&ii->i_dirty,
 732                                               &nilfs->ns_dirty_files);
 733                                 continue;
 734                         }
 735                         ivec[nv++] = ii;
 736                         if (nv == SC_N_INODEVEC)
 737                                 break;
 738                 }
 739                 spin_unlock(&nilfs->ns_inode_lock);
 740
 741                 for (pii = ivec; nv > 0; pii++, nv--)
 742                         iput(&(*pii)->vfs_inode);
 743         }
 744 }
 745
 746 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
 747                                      struct nilfs_root *root)
 748 {
 749         int ret = 0;
 750
 751         if (nilfs_mdt_fetch_dirty(root->ifile))
 752                 ret++;
 753         if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
 754                 ret++;
 755         if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
 756                 ret++;
 757         if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
 758                 ret++;
 759         return ret;
 760 }
 761
 762 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
 763 {
 764         return list_empty(&sci->sc_dirty_files) &&
 765                 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
 766                 sci->sc_nfreesegs == 0 &&
 767                 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
 768 }
 769
 770 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
 771 {
 772         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 773         int ret = 0;
 774
 775         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
 776                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
 777
 778         spin_lock(&nilfs->ns_inode_lock);
 779         if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
 780                 ret++;
 781
 782         spin_unlock(&nilfs->ns_inode_lock);
 783         return ret;
 784 }
 785
 786 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
 787 {
 788         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 789
 790         nilfs_mdt_clear_dirty(sci->sc_root->ifile);
 791         nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
 792         nilfs_mdt_clear_dirty(nilfs->ns_sufile);
 793         nilfs_mdt_clear_dirty(nilfs->ns_dat);
 794 }
 795
 796 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
 797 {
 798         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 799         struct buffer_head *bh_cp;
 800         struct nilfs_checkpoint *raw_cp;
 801         int err;
 802
 803         /* XXX: this interface will be changed */
 804         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
 805                                           &raw_cp, &bh_cp);
 806         if (likely(!err)) {
 807                 /* The following code is duplicated with cpfile.  But, it is
 808                    needed to collect the checkpoint even if it was not newly
 809                    created */
 810                 mark_buffer_dirty(bh_cp);
 811                 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
 812                 nilfs_cpfile_put_checkpoint(
 813                         nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 814         } else
 815                 WARN_ON(err == -EINVAL || err == -ENOENT);
 816
 817         return err;
 818 }
 819
 820 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
 821 {
 822         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 823         struct buffer_head *bh_cp;
 824         struct nilfs_checkpoint *raw_cp;
 825         int err;
 826
 827         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
 828                                           &raw_cp, &bh_cp);
 829         if (unlikely(err)) {
 830                 WARN_ON(err == -EINVAL || err == -ENOENT);
 831                 goto failed_ibh;
 832         }
 833         raw_cp->cp_snapshot_list.ssl_next = 0;
 834         raw_cp->cp_snapshot_list.ssl_prev = 0;
 835         raw_cp->cp_inodes_count =
 836                 cpu_to_le64(atomic_read(&sci->sc_root->inodes_count));
 837         raw_cp->cp_blocks_count =
 838                 cpu_to_le64(atomic_read(&sci->sc_root->blocks_count));
 839         raw_cp->cp_nblk_inc =
 840                 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
 841         raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
 842         raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
 843
 844         if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 845                 nilfs_checkpoint_clear_minor(raw_cp);
 846         else
 847                 nilfs_checkpoint_set_minor(raw_cp);
 848
 849         nilfs_write_inode_common(sci->sc_root->ifile,
 850                                  &raw_cp->cp_ifile_inode, 1);
 851         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 852         return 0;
 853
 854  failed_ibh:
 855         return err;
 856 }
 857
 858 static void nilfs_fill_in_file_bmap(struct inode *ifile,
 859                                     struct nilfs_inode_info *ii)
 860
 861 {
 862         struct buffer_head *ibh;
 863         struct nilfs_inode *raw_inode;
 864
 865         if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
 866                 ibh = ii->i_bh;
 867                 BUG_ON(!ibh);
 868                 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
 869                                                   ibh);
 870                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 871                 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
 872         }
 873 }
 874
 875 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
 876 {
 877         struct nilfs_inode_info *ii;
 878
 879         list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
 880                 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
 881                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
 882         }
 883 }
 884
 885 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
 886                                              struct the_nilfs *nilfs)
 887 {
 888         struct buffer_head *bh_sr;
 889         struct nilfs_super_root *raw_sr;
 890         unsigned isz, srsz;
 891
 892         bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
 893         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
 894         isz = nilfs->ns_inode_size;
 895         srsz = NILFS_SR_BYTES(isz);
 896
 897         raw_sr->sr_bytes = cpu_to_le16(srsz);
 898         raw_sr->sr_nongc_ctime
 899                 = cpu_to_le64(nilfs_doing_gc() ?
 900                               nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
 901         raw_sr->sr_flags = 0;
 902
 903         nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
 904                                  NILFS_SR_DAT_OFFSET(isz), 1);
 905         nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
 906                                  NILFS_SR_CPFILE_OFFSET(isz), 1);
 907         nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
 908                                  NILFS_SR_SUFILE_OFFSET(isz), 1);
 909         memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
 910 }
 911
 912 static void nilfs_redirty_inodes(struct list_head *head)
 913 {
 914         struct nilfs_inode_info *ii;
 915
 916         list_for_each_entry(ii, head, i_dirty) {
 917                 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
 918                         clear_bit(NILFS_I_COLLECTED, &ii->i_state);
 919         }
 920 }
 921
 922 static void nilfs_drop_collected_inodes(struct list_head *head)
 923 {
 924         struct nilfs_inode_info *ii;
 925
 926         list_for_each_entry(ii, head, i_dirty) {
 927                 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
 928                         continue;
 929
 930                 clear_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 931                 set_bit(NILFS_I_UPDATED, &ii->i_state);
 932         }
 933 }
 934
 935 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
 936                                        struct inode *inode,
 937                                        struct list_head *listp,
 938                                        int (*collect)(struct nilfs_sc_info *,
 939                                                       struct buffer_head *,
 940                                                       struct inode *))
 941 {
 942         struct buffer_head *bh, *n;
 943         int err = 0;
 944
 945         if (collect) {
 946                 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
 947                         list_del_init(&bh->b_assoc_buffers);
 948                         err = collect(sci, bh, inode);
 949                         brelse(bh);
 950                         if (unlikely(err))
 951                                 goto dispose_buffers;
 952                 }
 953                 return 0;
 954         }
 955
 956  dispose_buffers:
 957         while (!list_empty(listp)) {
 958                 bh = list_first_entry(listp, struct buffer_head,
 959                                       b_assoc_buffers);
 960                 list_del_init(&bh->b_assoc_buffers);
 961                 brelse(bh);
 962         }
 963         return err;
 964 }
 965
 966 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
 967 {
 968         /* Remaining number of blocks within segment buffer */
 969         return sci->sc_segbuf_nblocks -
 970                 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
 971 }
 972
 973 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
 974                                    struct inode *inode,
 975                                    struct nilfs_sc_operations *sc_ops)
 976 {
 977         LIST_HEAD(data_buffers);
 978         LIST_HEAD(node_buffers);
 979         int err;
 980
 981         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
 982                 size_t n, rest = nilfs_segctor_buffer_rest(sci);
 983
 984                 n = nilfs_lookup_dirty_data_buffers(
 985                         inode, &data_buffers, rest + 1, 0, LLONG_MAX);
 986                 if (n > rest) {
 987                         err = nilfs_segctor_apply_buffers(
 988                                 sci, inode, &data_buffers,
 989                                 sc_ops->collect_data);
 990                         BUG_ON(!err); /* always receive -E2BIG or true error */
 991                         goto break_or_fail;
 992                 }
 993         }
 994         nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
 995
 996         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
 997                 err = nilfs_segctor_apply_buffers(
 998                         sci, inode, &data_buffers, sc_ops->collect_data);
 999                 if (unlikely(err)) {
1000                         /* dispose node list */
1001                         nilfs_segctor_apply_buffers(
1002                                 sci, inode, &node_buffers, NULL);
1003                         goto break_or_fail;
1004                 }
1005                 sci->sc_stage.flags |= NILFS_CF_NODE;
1006         }
1007         /* Collect node */
1008         err = nilfs_segctor_apply_buffers(
1009                 sci, inode, &node_buffers, sc_ops->collect_node);
1010         if (unlikely(err))
1011                 goto break_or_fail;
1012
1013         nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1014         err = nilfs_segctor_apply_buffers(
1015                 sci, inode, &node_buffers, sc_ops->collect_bmap);
1016         if (unlikely(err))
1017                 goto break_or_fail;
1018
1019         nilfs_segctor_end_finfo(sci, inode);
1020         sci->sc_stage.flags &= ~NILFS_CF_NODE;
1021
1022  break_or_fail:
1023         return err;
1024 }
1025
1026 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1027                                          struct inode *inode)
1028 {
1029         LIST_HEAD(data_buffers);
1030         size_t n, rest = nilfs_segctor_buffer_rest(sci);
1031         int err;
1032
1033         n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1034                                             sci->sc_dsync_start,
1035                                             sci->sc_dsync_end);
1036
1037         err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1038                                           nilfs_collect_file_data);
1039         if (!err) {
1040                 nilfs_segctor_end_finfo(sci, inode);
1041                 BUG_ON(n > rest);
1042                 /* always receive -E2BIG or true error if n > rest */
1043         }
1044         return err;
1045 }
1046
1047 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1048 {
1049         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1050         struct list_head *head;
1051         struct nilfs_inode_info *ii;
1052         size_t ndone;
1053         int err = 0;
1054
1055         switch (sci->sc_stage.scnt) {
1056         case NILFS_ST_INIT:
1057                 /* Pre-processes */
1058                 sci->sc_stage.flags = 0;
1059
1060                 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1061                         sci->sc_nblk_inc = 0;
1062                         sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1063                         if (mode == SC_LSEG_DSYNC) {
1064                                 sci->sc_stage.scnt = NILFS_ST_DSYNC;
1065                                 goto dsync_mode;
1066                         }
1067                 }
1068
1069                 sci->sc_stage.dirty_file_ptr = NULL;
1070                 sci->sc_stage.gc_inode_ptr = NULL;
1071                 if (mode == SC_FLUSH_DAT) {
1072                         sci->sc_stage.scnt = NILFS_ST_DAT;
1073                         goto dat_stage;
1074                 }
1075                 sci->sc_stage.scnt++;  /* Fall through */
1076         case NILFS_ST_GC:
1077                 if (nilfs_doing_gc()) {
1078                         head = &sci->sc_gc_inodes;
1079                         ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1080                                                 head, i_dirty);
1081                         list_for_each_entry_continue(ii, head, i_dirty) {
1082                                 err = nilfs_segctor_scan_file(
1083                                         sci, &ii->vfs_inode,
1084                                         &nilfs_sc_file_ops);
1085                                 if (unlikely(err)) {
1086                                         sci->sc_stage.gc_inode_ptr = list_entry(
1087                                                 ii->i_dirty.prev,
1088                                                 struct nilfs_inode_info,
1089                                                 i_dirty);
1090                                         goto break_or_fail;
1091                                 }
1092                                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1093                         }
1094                         sci->sc_stage.gc_inode_ptr = NULL;
1095                 }
1096                 sci->sc_stage.scnt++;  /* Fall through */
1097         case NILFS_ST_FILE:
1098                 head = &sci->sc_dirty_files;
1099                 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1100                                         i_dirty);
1101                 list_for_each_entry_continue(ii, head, i_dirty) {
1102                         clear_bit(NILFS_I_DIRTY, &ii->i_state);
1103
1104                         err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1105                                                       &nilfs_sc_file_ops);
1106                         if (unlikely(err)) {
1107                                 sci->sc_stage.dirty_file_ptr =
1108                                         list_entry(ii->i_dirty.prev,
1109                                                    struct nilfs_inode_info,
1110                                                    i_dirty);
1111                                 goto break_or_fail;
1112                         }
1113                         /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1114                         /* XXX: required ? */
1115                 }
1116                 sci->sc_stage.dirty_file_ptr = NULL;
1117                 if (mode == SC_FLUSH_FILE) {
1118                         sci->sc_stage.scnt = NILFS_ST_DONE;
1119                         return 0;
1120                 }
1121                 sci->sc_stage.scnt++;
1122                 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1123                 /* Fall through */
1124         case NILFS_ST_IFILE:
1125                 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1126                                               &nilfs_sc_file_ops);
1127                 if (unlikely(err))
1128                         break;
1129                 sci->sc_stage.scnt++;
1130                 /* Creating a checkpoint */
1131                 err = nilfs_segctor_create_checkpoint(sci);
1132                 if (unlikely(err))
1133                         break;
1134                 /* Fall through */
1135         case NILFS_ST_CPFILE:
1136                 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1137                                               &nilfs_sc_file_ops);
1138                 if (unlikely(err))
1139                         break;
1140                 sci->sc_stage.scnt++;  /* Fall through */
1141         case NILFS_ST_SUFILE:
1142                 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1143                                          sci->sc_nfreesegs, &ndone);
1144                 if (unlikely(err)) {
1145                         nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1146                                                   sci->sc_freesegs, ndone,
1147                                                   NULL);
1148                         break;
1149                 }
1150                 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1151
1152                 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1153                                               &nilfs_sc_file_ops);
1154                 if (unlikely(err))
1155                         break;
1156                 sci->sc_stage.scnt++;  /* Fall through */
1157         case NILFS_ST_DAT:
1158  dat_stage:
1159                 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1160                                               &nilfs_sc_dat_ops);
1161                 if (unlikely(err))
1162                         break;
1163                 if (mode == SC_FLUSH_DAT) {
1164                         sci->sc_stage.scnt = NILFS_ST_DONE;
1165                         return 0;
1166                 }
1167                 sci->sc_stage.scnt++;  /* Fall through */
1168         case NILFS_ST_SR:
1169                 if (mode == SC_LSEG_SR) {
1170                         /* Appending a super root */
1171                         err = nilfs_segctor_add_super_root(sci);
1172                         if (unlikely(err))
1173                                 break;
1174                 }
1175                 /* End of a logical segment */
1176                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1177                 sci->sc_stage.scnt = NILFS_ST_DONE;
1178                 return 0;
1179         case NILFS_ST_DSYNC:
1180  dsync_mode:
1181                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1182                 ii = sci->sc_dsync_inode;
1183                 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1184                         break;
1185
1186                 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1187                 if (unlikely(err))
1188                         break;
1189                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1190                 sci->sc_stage.scnt = NILFS_ST_DONE;
1191                 return 0;
1192         case NILFS_ST_DONE:
1193                 return 0;
1194         default:
1195                 BUG();
1196         }
1197
1198  break_or_fail:
1199         return err;
1200 }
1201
1202 /**
1203  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1204  * @sci: nilfs_sc_info
1205  * @nilfs: nilfs object
1206  */
1207 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1208                                             struct the_nilfs *nilfs)
1209 {
1210         struct nilfs_segment_buffer *segbuf, *prev;
1211         __u64 nextnum;
1212         int err, alloc = 0;
1213
1214         segbuf = nilfs_segbuf_new(sci->sc_super);
1215         if (unlikely(!segbuf))
1216                 return -ENOMEM;
1217
1218         if (list_empty(&sci->sc_write_logs)) {
1219                 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1220                                  nilfs->ns_pseg_offset, nilfs);
1221                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1222                         nilfs_shift_to_next_segment(nilfs);
1223                         nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1224                 }
1225
1226                 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1227                 nextnum = nilfs->ns_nextnum;
1228
1229                 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1230                         /* Start from the head of a new full segment */
1231                         alloc++;
1232         } else {
1233                 /* Continue logs */
1234                 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1235                 nilfs_segbuf_map_cont(segbuf, prev);
1236                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1237                 nextnum = prev->sb_nextnum;
1238
1239                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1240                         nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1241                         segbuf->sb_sum.seg_seq++;
1242                         alloc++;
1243                 }
1244         }
1245
1246         err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1247         if (err)
1248                 goto failed;
1249
1250         if (alloc) {
1251                 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1252                 if (err)
1253                         goto failed;
1254         }
1255         nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1256
1257         BUG_ON(!list_empty(&sci->sc_segbufs));
1258         list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1259         sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1260         return 0;
1261
1262  failed:
1263         nilfs_segbuf_free(segbuf);
1264         return err;
1265 }
1266
1267 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1268                                          struct the_nilfs *nilfs, int nadd)
1269 {
1270         struct nilfs_segment_buffer *segbuf, *prev;
1271         struct inode *sufile = nilfs->ns_sufile;
1272         __u64 nextnextnum;
1273         LIST_HEAD(list);
1274         int err, ret, i;
1275
1276         prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1277         /*
1278          * Since the segment specified with nextnum might be allocated during
1279          * the previous construction, the buffer including its segusage may
1280          * not be dirty.  The following call ensures that the buffer is dirty
1281          * and will pin the buffer on memory until the sufile is written.
1282          */
1283         err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1284         if (unlikely(err))
1285                 return err;
1286
1287         for (i = 0; i < nadd; i++) {
1288                 /* extend segment info */
1289                 err = -ENOMEM;
1290                 segbuf = nilfs_segbuf_new(sci->sc_super);
1291                 if (unlikely(!segbuf))
1292                         goto failed;
1293
1294                 /* map this buffer to region of segment on-disk */
1295                 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1296                 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1297
1298                 /* allocate the next next full segment */
1299                 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1300                 if (unlikely(err))
1301                         goto failed_segbuf;
1302
1303                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1304                 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1305
1306                 list_add_tail(&segbuf->sb_list, &list);
1307                 prev = segbuf;
1308         }
1309         list_splice_tail(&list, &sci->sc_segbufs);
1310         return 0;
1311
1312  failed_segbuf:
1313         nilfs_segbuf_free(segbuf);
1314  failed:
1315         list_for_each_entry(segbuf, &list, sb_list) {
1316                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1317                 WARN_ON(ret); /* never fails */
1318         }
1319         nilfs_destroy_logs(&list);
1320         return err;
1321 }
1322
1323 static void nilfs_free_incomplete_logs(struct list_head *logs,
1324                                        struct the_nilfs *nilfs)
1325 {
1326         struct nilfs_segment_buffer *segbuf, *prev;
1327         struct inode *sufile = nilfs->ns_sufile;
1328         int ret;
1329
1330         segbuf = NILFS_FIRST_SEGBUF(logs);
1331         if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1332                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1333                 WARN_ON(ret); /* never fails */
1334         }
1335         if (atomic_read(&segbuf->sb_err)) {
1336                 /* Case 1: The first segment failed */
1337                 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1338                         /* Case 1a:  Partial segment appended into an existing
1339                            segment */
1340                         nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1341                                                 segbuf->sb_fseg_end);
1342                 else /* Case 1b:  New full segment */
1343                         set_nilfs_discontinued(nilfs);
1344         }
1345
1346         prev = segbuf;
1347         list_for_each_entry_continue(segbuf, logs, sb_list) {
1348                 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1349                         ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1350                         WARN_ON(ret); /* never fails */
1351                 }
1352                 if (atomic_read(&segbuf->sb_err) &&
1353                     segbuf->sb_segnum != nilfs->ns_nextnum)
1354                         /* Case 2: extended segment (!= next) failed */
1355                         nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1356                 prev = segbuf;
1357         }
1358 }
1359
1360 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1361                                           struct inode *sufile)
1362 {
1363         struct nilfs_segment_buffer *segbuf;
1364         unsigned long live_blocks;
1365         int ret;
1366
1367         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1368                 live_blocks = segbuf->sb_sum.nblocks +
1369                         (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1370                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1371                                                      live_blocks,
1372                                                      sci->sc_seg_ctime);
1373                 WARN_ON(ret); /* always succeed because the segusage is dirty */
1374         }
1375 }
1376
1377 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1378 {
1379         struct nilfs_segment_buffer *segbuf;
1380         int ret;
1381
1382         segbuf = NILFS_FIRST_SEGBUF(logs);
1383         ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1384                                              segbuf->sb_pseg_start -
1385                                              segbuf->sb_fseg_start, 0);
1386         WARN_ON(ret); /* always succeed because the segusage is dirty */
1387
1388         list_for_each_entry_continue(segbuf, logs, sb_list) {
1389                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1390                                                      0, 0);
1391                 WARN_ON(ret); /* always succeed */
1392         }
1393 }
1394
1395 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1396                                             struct nilfs_segment_buffer *last,
1397                                             struct inode *sufile)
1398 {
1399         struct nilfs_segment_buffer *segbuf = last;
1400         int ret;
1401
1402         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1403                 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1404                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1405                 WARN_ON(ret);
1406         }
1407         nilfs_truncate_logs(&sci->sc_segbufs, last);
1408 }
1409
1410
1411 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1412                                  struct the_nilfs *nilfs, int mode)
1413 {
1414         struct nilfs_cstage prev_stage = sci->sc_stage;
1415         int err, nadd = 1;
1416
1417         /* Collection retry loop */
1418         for (;;) {
1419                 sci->sc_nblk_this_inc = 0;
1420                 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1421
1422                 err = nilfs_segctor_reset_segment_buffer(sci);
1423                 if (unlikely(err))
1424                         goto failed;
1425
1426                 err = nilfs_segctor_collect_blocks(sci, mode);
1427                 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1428                 if (!err)
1429                         break;
1430
1431                 if (unlikely(err != -E2BIG))
1432                         goto failed;
1433
1434                 /* The current segment is filled up */
1435                 if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
1436                         break;
1437
1438                 nilfs_clear_logs(&sci->sc_segbufs);
1439
1440                 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1441                 if (unlikely(err))
1442                         return err;
1443
1444                 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1445                         err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1446                                                         sci->sc_freesegs,
1447                                                         sci->sc_nfreesegs,
1448                                                         NULL);
1449                         WARN_ON(err); /* do not happen */
1450                 }
1451                 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1452                 sci->sc_stage = prev_stage;
1453         }
1454         nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1455         return 0;
1456
1457  failed:
1458         return err;
1459 }
1460
1461 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1462                                       struct buffer_head *new_bh)
1463 {
1464         BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1465
1466         list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1467         /* The caller must release old_bh */
1468 }
1469
1470 static int
1471 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1472                                      struct nilfs_segment_buffer *segbuf,
1473                                      int mode)
1474 {
1475         struct inode *inode = NULL;
1476         sector_t blocknr;
1477         unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1478         unsigned long nblocks = 0, ndatablk = 0;
1479         struct nilfs_sc_operations *sc_op = NULL;
1480         struct nilfs_segsum_pointer ssp;
1481         struct nilfs_finfo *finfo = NULL;
1482         union nilfs_binfo binfo;
1483         struct buffer_head *bh, *bh_org;
1484         ino_t ino = 0;
1485         int err = 0;
1486
1487         if (!nfinfo)
1488                 goto out;
1489
1490         blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1491         ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1492         ssp.offset = sizeof(struct nilfs_segment_summary);
1493
1494         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1495                 if (bh == segbuf->sb_super_root)
1496                         break;
1497                 if (!finfo) {
1498                         finfo = nilfs_segctor_map_segsum_entry(
1499                                 sci, &ssp, sizeof(*finfo));
1500                         ino = le64_to_cpu(finfo->fi_ino);
1501                         nblocks = le32_to_cpu(finfo->fi_nblocks);
1502                         ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1503
1504                         inode = bh->b_page->mapping->host;
1505
1506                         if (mode == SC_LSEG_DSYNC)
1507                                 sc_op = &nilfs_sc_dsync_ops;
1508                         else if (ino == NILFS_DAT_INO)
1509                                 sc_op = &nilfs_sc_dat_ops;
1510                         else /* file blocks */
1511                                 sc_op = &nilfs_sc_file_ops;
1512                 }
1513                 bh_org = bh;
1514                 get_bh(bh_org);
1515                 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1516                                         &binfo);
1517                 if (bh != bh_org)
1518                         nilfs_list_replace_buffer(bh_org, bh);
1519                 brelse(bh_org);
1520                 if (unlikely(err))
1521                         goto failed_bmap;
1522
1523                 if (ndatablk > 0)
1524                         sc_op->write_data_binfo(sci, &ssp, &binfo);
1525                 else
1526                         sc_op->write_node_binfo(sci, &ssp, &binfo);
1527
1528                 blocknr++;
1529                 if (--nblocks == 0) {
1530                         finfo = NULL;
1531                         if (--nfinfo == 0)
1532                                 break;
1533                 } else if (ndatablk > 0)
1534                         ndatablk--;
1535         }
1536  out:
1537         return 0;
1538
1539  failed_bmap:
1540         return err;
1541 }
1542
1543 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1544 {
1545         struct nilfs_segment_buffer *segbuf;
1546         int err;
1547
1548         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1549                 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1550                 if (unlikely(err))
1551                         return err;
1552                 nilfs_segbuf_fill_in_segsum(segbuf);
1553         }
1554         return 0;
1555 }
1556
1557 static void nilfs_begin_page_io(struct page *page)
1558 {
1559         if (!page || PageWriteback(page))
1560                 /* For split b-tree node pages, this function may be called
1561                    twice.  We ignore the 2nd or later calls by this check. */
1562                 return;
1563
1564         lock_page(page);
1565         clear_page_dirty_for_io(page);
1566         set_page_writeback(page);
1567         unlock_page(page);
1568 }
1569
1570 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1571 {
1572         struct nilfs_segment_buffer *segbuf;
1573         struct page *bd_page = NULL, *fs_page = NULL;
1574
1575         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1576                 struct buffer_head *bh;
1577
1578                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1579                                     b_assoc_buffers) {
1580                         set_buffer_async_write(bh);
1581                         if (bh->b_page != bd_page) {
1582                                 if (bd_page) {
1583                                         lock_page(bd_page);
1584                                         clear_page_dirty_for_io(bd_page);
1585                                         set_page_writeback(bd_page);
1586                                         unlock_page(bd_page);
1587                                 }
1588                                 bd_page = bh->b_page;
1589                         }
1590                 }
1591
1592                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1593                                     b_assoc_buffers) {
1594                         set_buffer_async_write(bh);
1595                         if (bh == segbuf->sb_super_root) {
1596                                 if (bh->b_page != bd_page) {
1597                                         lock_page(bd_page);
1598                                         clear_page_dirty_for_io(bd_page);
1599                                         set_page_writeback(bd_page);
1600                                         unlock_page(bd_page);
1601                                         bd_page = bh->b_page;
1602                                 }
1603                                 break;
1604                         }
1605                         if (bh->b_page != fs_page) {
1606                                 nilfs_begin_page_io(fs_page);
1607                                 fs_page = bh->b_page;
1608                         }
1609                 }
1610         }
1611         if (bd_page) {
1612                 lock_page(bd_page);
1613                 clear_page_dirty_for_io(bd_page);
1614                 set_page_writeback(bd_page);
1615                 unlock_page(bd_page);
1616         }
1617         nilfs_begin_page_io(fs_page);
1618 }
1619
1620 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1621                                struct the_nilfs *nilfs)
1622 {
1623         int ret;
1624
1625         ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1626         list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1627         return ret;
1628 }
1629
1630 static void nilfs_end_page_io(struct page *page, int err)
1631 {
1632         if (!page)
1633                 return;
1634
1635         if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1636                 /*
1637                  * For b-tree node pages, this function may be called twice
1638                  * or more because they might be split in a segment.
1639                  */
1640                 if (PageDirty(page)) {
1641                         /*
1642                          * For pages holding split b-tree node buffers, dirty
1643                          * flag on the buffers may be cleared discretely.
1644                          * In that case, the page is once redirtied for
1645                          * remaining buffers, and it must be cancelled if
1646                          * all the buffers get cleaned later.
1647                          */
1648                         lock_page(page);
1649                         if (nilfs_page_buffers_clean(page))
1650                                 __nilfs_clear_page_dirty(page);
1651                         unlock_page(page);
1652                 }
1653                 return;
1654         }
1655
1656         if (!err) {
1657                 if (!nilfs_page_buffers_clean(page))
1658                         __set_page_dirty_nobuffers(page);
1659                 ClearPageError(page);
1660         } else {
1661                 __set_page_dirty_nobuffers(page);
1662                 SetPageError(page);
1663         }
1664
1665         end_page_writeback(page);
1666 }
1667
1668 static void nilfs_abort_logs(struct list_head *logs, int err)
1669 {
1670         struct nilfs_segment_buffer *segbuf;
1671         struct page *bd_page = NULL, *fs_page = NULL;
1672         struct buffer_head *bh;
1673
1674         if (list_empty(logs))
1675                 return;
1676
1677         list_for_each_entry(segbuf, logs, sb_list) {
1678                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1679                                     b_assoc_buffers) {
1680                         clear_buffer_async_write(bh);
1681                         if (bh->b_page != bd_page) {
1682                                 if (bd_page)
1683                                         end_page_writeback(bd_page);
1684                                 bd_page = bh->b_page;
1685                         }
1686                 }
1687
1688                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1689                                     b_assoc_buffers) {
1690                         clear_buffer_async_write(bh);
1691                         if (bh == segbuf->sb_super_root) {
1692                                 if (bh->b_page != bd_page) {
1693                                         end_page_writeback(bd_page);
1694                                         bd_page = bh->b_page;
1695                                 }
1696                                 break;
1697                         }
1698                         if (bh->b_page != fs_page) {
1699                                 nilfs_end_page_io(fs_page, err);
1700                                 fs_page = bh->b_page;
1701                         }
1702                 }
1703         }
1704         if (bd_page)
1705                 end_page_writeback(bd_page);
1706
1707         nilfs_end_page_io(fs_page, err);
1708 }
1709
1710 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1711                                              struct the_nilfs *nilfs, int err)
1712 {
1713         LIST_HEAD(logs);
1714         int ret;
1715
1716         list_splice_tail_init(&sci->sc_write_logs, &logs);
1717         ret = nilfs_wait_on_logs(&logs);
1718         nilfs_abort_logs(&logs, ret ? : err);
1719
1720         list_splice_tail_init(&sci->sc_segbufs, &logs);
1721         nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1722         nilfs_free_incomplete_logs(&logs, nilfs);
1723
1724         if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1725                 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1726                                                 sci->sc_freesegs,
1727                                                 sci->sc_nfreesegs,
1728                                                 NULL);
1729                 WARN_ON(ret); /* do not happen */
1730         }
1731
1732         nilfs_destroy_logs(&logs);
1733 }
1734
1735 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1736                                    struct nilfs_segment_buffer *segbuf)
1737 {
1738         nilfs->ns_segnum = segbuf->sb_segnum;
1739         nilfs->ns_nextnum = segbuf->sb_nextnum;
1740         nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1741                 + segbuf->sb_sum.nblocks;
1742         nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1743         nilfs->ns_ctime = segbuf->sb_sum.ctime;
1744 }
1745
1746 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1747 {
1748         struct nilfs_segment_buffer *segbuf;
1749         struct page *bd_page = NULL, *fs_page = NULL;
1750         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1751         int update_sr = false;
1752
1753         list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1754                 struct buffer_head *bh;
1755
1756                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1757                                     b_assoc_buffers) {
1758                         set_buffer_uptodate(bh);
1759                         clear_buffer_dirty(bh);
1760                         clear_buffer_async_write(bh);
1761                         if (bh->b_page != bd_page) {
1762                                 if (bd_page)
1763                                         end_page_writeback(bd_page);
1764                                 bd_page = bh->b_page;
1765                         }
1766                 }
1767                 /*
1768                  * We assume that the buffers which belong to the same page
1769                  * continue over the buffer list.
1770                  * Under this assumption, the last BHs of pages is
1771                  * identifiable by the discontinuity of bh->b_page
1772                  * (page != fs_page).
1773                  *
1774                  * For B-tree node blocks, however, this assumption is not
1775                  * guaranteed.  The cleanup code of B-tree node pages needs
1776                  * special care.
1777                  */
1778                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1779                                     b_assoc_buffers) {
1780                         set_buffer_uptodate(bh);
1781                         clear_buffer_dirty(bh);
1782                         clear_buffer_async_write(bh);
1783                         clear_buffer_delay(bh);
1784                         clear_buffer_nilfs_volatile(bh);
1785                         clear_buffer_nilfs_redirected(bh);
1786                         if (bh == segbuf->sb_super_root) {
1787                                 if (bh->b_page != bd_page) {
1788                                         end_page_writeback(bd_page);
1789                                         bd_page = bh->b_page;
1790                                 }
1791                                 update_sr = true;
1792                                 break;
1793                         }
1794                         if (bh->b_page != fs_page) {
1795                                 nilfs_end_page_io(fs_page, 0);
1796                                 fs_page = bh->b_page;
1797                         }
1798                 }
1799
1800                 if (!nilfs_segbuf_simplex(segbuf)) {
1801                         if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1802                                 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1803                                 sci->sc_lseg_stime = jiffies;
1804                         }
1805                         if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1806                                 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1807                 }
1808         }
1809         /*
1810          * Since pages may continue over multiple segment buffers,
1811          * end of the last page must be checked outside of the loop.
1812          */
1813         if (bd_page)
1814                 end_page_writeback(bd_page);
1815
1816         nilfs_end_page_io(fs_page, 0);
1817
1818         nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1819
1820         if (nilfs_doing_gc())
1821                 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1822         else
1823                 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1824
1825         sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1826
1827         segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1828         nilfs_set_next_segment(nilfs, segbuf);
1829
1830         if (update_sr) {
1831                 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1832                                        segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1833
1834                 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1835                 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1836                 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1837                 nilfs_segctor_clear_metadata_dirty(sci);
1838         } else
1839                 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1840 }
1841
1842 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1843 {
1844         int ret;
1845
1846         ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1847         if (!ret) {
1848                 nilfs_segctor_complete_write(sci);
1849                 nilfs_destroy_logs(&sci->sc_write_logs);
1850         }
1851         return ret;
1852 }
1853
1854 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1855                                              struct the_nilfs *nilfs)
1856 {
1857         struct nilfs_inode_info *ii, *n;
1858         struct inode *ifile = sci->sc_root->ifile;
1859
1860         spin_lock(&nilfs->ns_inode_lock);
1861  retry:
1862         list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1863                 if (!ii->i_bh) {
1864                         struct buffer_head *ibh;
1865                         int err;
1866
1867                         spin_unlock(&nilfs->ns_inode_lock);
1868                         err = nilfs_ifile_get_inode_block(
1869                                 ifile, ii->vfs_inode.i_ino, &ibh);
1870                         if (unlikely(err)) {
1871                                 nilfs_warning(sci->sc_super, __func__,
1872                                               "failed to get inode block.\n");
1873                                 return err;
1874                         }
1875                         mark_buffer_dirty(ibh);
1876                         nilfs_mdt_mark_dirty(ifile);
1877                         spin_lock(&nilfs->ns_inode_lock);
1878                         if (likely(!ii->i_bh))
1879                                 ii->i_bh = ibh;
1880                         else
1881                                 brelse(ibh);
1882                         goto retry;
1883                 }
1884
1885                 clear_bit(NILFS_I_QUEUED, &ii->i_state);
1886                 set_bit(NILFS_I_BUSY, &ii->i_state);
1887                 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
1888         }
1889         spin_unlock(&nilfs->ns_inode_lock);
1890
1891         return 0;
1892 }
1893
1894 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
1895                                              struct the_nilfs *nilfs)
1896 {
1897         struct nilfs_transaction_info *ti = current->journal_info;
1898         struct nilfs_inode_info *ii, *n;
1899
1900         spin_lock(&nilfs->ns_inode_lock);
1901         list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
1902                 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
1903                     test_bit(NILFS_I_DIRTY, &ii->i_state))
1904                         continue;
1905
1906                 clear_bit(NILFS_I_BUSY, &ii->i_state);
1907                 brelse(ii->i_bh);
1908                 ii->i_bh = NULL;
1909                 list_move_tail(&ii->i_dirty, &ti->ti_garbage);
1910         }
1911         spin_unlock(&nilfs->ns_inode_lock);
1912 }
1913
1914 /*
1915  * Main procedure of segment constructor
1916  */
1917 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
1918 {
1919         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1920         int err;
1921
1922         sci->sc_stage.scnt = NILFS_ST_INIT;
1923         sci->sc_cno = nilfs->ns_cno;
1924
1925         err = nilfs_segctor_collect_dirty_files(sci, nilfs);
1926         if (unlikely(err))
1927                 goto out;
1928
1929         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
1930                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1931
1932         if (nilfs_segctor_clean(sci))
1933                 goto out;
1934
1935         do {
1936                 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
1937
1938                 err = nilfs_segctor_begin_construction(sci, nilfs);
1939                 if (unlikely(err))
1940                         goto out;
1941
1942                 /* Update time stamp */
1943                 sci->sc_seg_ctime = get_seconds();
1944
1945                 err = nilfs_segctor_collect(sci, nilfs, mode);
1946                 if (unlikely(err))
1947                         goto failed;
1948
1949                 /* Avoid empty segment */
1950                 if (sci->sc_stage.scnt == NILFS_ST_DONE &&
1951                     nilfs_segbuf_empty(sci->sc_curseg)) {
1952                         nilfs_segctor_abort_construction(sci, nilfs, 1);
1953                         goto out;
1954                 }
1955
1956                 err = nilfs_segctor_assign(sci, mode);
1957                 if (unlikely(err))
1958                         goto failed;
1959
1960                 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
1961                         nilfs_segctor_fill_in_file_bmap(sci);
1962
1963                 if (mode == SC_LSEG_SR &&
1964                     sci->sc_stage.scnt >= NILFS_ST_CPFILE) {
1965                         err = nilfs_segctor_fill_in_checkpoint(sci);
1966                         if (unlikely(err))
1967                                 goto failed_to_write;
1968
1969                         nilfs_segctor_fill_in_super_root(sci, nilfs);
1970                 }
1971                 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
1972
1973                 /* Write partial segments */
1974                 nilfs_segctor_prepare_write(sci);
1975
1976                 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
1977                                             nilfs->ns_crc_seed);
1978
1979                 err = nilfs_segctor_write(sci, nilfs);
1980                 if (unlikely(err))
1981                         goto failed_to_write;
1982
1983                 if (sci->sc_stage.scnt == NILFS_ST_DONE ||
1984                     nilfs->ns_blocksize_bits != PAGE_CACHE_SHIFT) {
1985                         /*
1986                          * At this point, we avoid double buffering
1987                          * for blocksize < pagesize because page dirty
1988                          * flag is turned off during write and dirty
1989                          * buffers are not properly collected for
1990                          * pages crossing over segments.
1991                          */
1992                         err = nilfs_segctor_wait(sci);
1993                         if (err)
1994                                 goto failed_to_write;
1995                 }
1996         } while (sci->sc_stage.scnt != NILFS_ST_DONE);
1997
1998  out:
1999         nilfs_segctor_drop_written_files(sci, nilfs);
2000         return err;
2001
2002  failed_to_write:
2003         if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2004                 nilfs_redirty_inodes(&sci->sc_dirty_files);
2005
2006  failed:
2007         if (nilfs_doing_gc())
2008                 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2009         nilfs_segctor_abort_construction(sci, nilfs, err);
2010         goto out;
2011 }
2012
2013 /**
2014  * nilfs_segctor_start_timer - set timer of background write
2015  * @sci: nilfs_sc_info
2016  *
2017  * If the timer has already been set, it ignores the new request.
2018  * This function MUST be called within a section locking the segment
2019  * semaphore.
2020  */
2021 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2022 {
2023         spin_lock(&sci->sc_state_lock);
2024         if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2025                 sci->sc_timer.expires = jiffies + sci->sc_interval;
2026                 add_timer(&sci->sc_timer);
2027                 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2028         }
2029         spin_unlock(&sci->sc_state_lock);
2030 }
2031
2032 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2033 {
2034         spin_lock(&sci->sc_state_lock);
2035         if (!(sci->sc_flush_request & (1 << bn))) {
2036                 unsigned long prev_req = sci->sc_flush_request;
2037
2038                 sci->sc_flush_request |= (1 << bn);
2039                 if (!prev_req)
2040                         wake_up(&sci->sc_wait_daemon);
2041         }
2042         spin_unlock(&sci->sc_state_lock);
2043 }
2044
2045 /**
2046  * nilfs_flush_segment - trigger a segment construction for resource control
2047  * @sb: super block
2048  * @ino: inode number of the file to be flushed out.
2049  */
2050 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2051 {
2052         struct the_nilfs *nilfs = sb->s_fs_info;
2053         struct nilfs_sc_info *sci = nilfs->ns_writer;
2054
2055         if (!sci || nilfs_doing_construction())
2056                 return;
2057         nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2058                                         /* assign bit 0 to data files */
2059 }
2060
2061 struct nilfs_segctor_wait_request {
2062         wait_queue_t    wq;
2063         __u32           seq;
2064         int             err;
2065         atomic_t        done;
2066 };
2067
2068 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2069 {
2070         struct nilfs_segctor_wait_request wait_req;
2071         int err = 0;
2072
2073         spin_lock(&sci->sc_state_lock);
2074         init_wait(&wait_req.wq);
2075         wait_req.err = 0;
2076         atomic_set(&wait_req.done, 0);
2077         wait_req.seq = ++sci->sc_seq_request;
2078         spin_unlock(&sci->sc_state_lock);
2079
2080         init_waitqueue_entry(&wait_req.wq, current);
2081         add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2082         set_current_state(TASK_INTERRUPTIBLE);
2083         wake_up(&sci->sc_wait_daemon);
2084
2085         for (;;) {
2086                 if (atomic_read(&wait_req.done)) {
2087                         err = wait_req.err;
2088                         break;
2089                 }
2090                 if (!signal_pending(current)) {
2091                         schedule();
2092                         continue;
2093                 }
2094                 err = -ERESTARTSYS;
2095                 break;
2096         }
2097         finish_wait(&sci->sc_wait_request, &wait_req.wq);
2098         return err;
2099 }
2100
2101 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2102 {
2103         struct nilfs_segctor_wait_request *wrq, *n;
2104         unsigned long flags;
2105
2106         spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2107         list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2108                                  wq.task_list) {
2109                 if (!atomic_read(&wrq->done) &&
2110                     nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2111                         wrq->err = err;
2112                         atomic_set(&wrq->done, 1);
2113                 }
2114                 if (atomic_read(&wrq->done)) {
2115                         wrq->wq.func(&wrq->wq,
2116                                      TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2117                                      0, NULL);
2118                 }
2119         }
2120         spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2121 }
2122
2123 /**
2124  * nilfs_construct_segment - construct a logical segment
2125  * @sb: super block
2126  *
2127  * Return Value: On success, 0 is retured. On errors, one of the following
2128  * negative error code is returned.
2129  *
2130  * %-EROFS - Read only filesystem.
2131  *
2132  * %-EIO - I/O error
2133  *
2134  * %-ENOSPC - No space left on device (only in a panic state).
2135  *
2136  * %-ERESTARTSYS - Interrupted.
2137  *
2138  * %-ENOMEM - Insufficient memory available.
2139  */
2140 int nilfs_construct_segment(struct super_block *sb)
2141 {
2142         struct the_nilfs *nilfs = sb->s_fs_info;
2143         struct nilfs_sc_info *sci = nilfs->ns_writer;
2144         struct nilfs_transaction_info *ti;
2145         int err;
2146
2147         if (!sci)
2148                 return -EROFS;
2149
2150         /* A call inside transactions causes a deadlock. */
2151         BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2152
2153         err = nilfs_segctor_sync(sci);
2154         return err;
2155 }
2156
2157 /**
2158  * nilfs_construct_dsync_segment - construct a data-only logical segment
2159  * @sb: super block
2160  * @inode: inode whose data blocks should be written out
2161  * @start: start byte offset
2162  * @end: end byte offset (inclusive)
2163  *
2164  * Return Value: On success, 0 is retured. On errors, one of the following
2165  * negative error code is returned.
2166  *
2167  * %-EROFS - Read only filesystem.
2168  *
2169  * %-EIO - I/O error
2170  *
2171  * %-ENOSPC - No space left on device (only in a panic state).
2172  *
2173  * %-ERESTARTSYS - Interrupted.
2174  *
2175  * %-ENOMEM - Insufficient memory available.
2176  */
2177 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2178                                   loff_t start, loff_t end)
2179 {
2180         struct the_nilfs *nilfs = sb->s_fs_info;
2181         struct nilfs_sc_info *sci = nilfs->ns_writer;
2182         struct nilfs_inode_info *ii;
2183         struct nilfs_transaction_info ti;
2184         int err = 0;
2185
2186         if (!sci)
2187                 return -EROFS;
2188
2189         nilfs_transaction_lock(sb, &ti, 0);
2190
2191         ii = NILFS_I(inode);
2192         if (test_bit(NILFS_I_INODE_DIRTY, &ii->i_state) ||
2193             nilfs_test_opt(nilfs, STRICT_ORDER) ||
2194             test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2195             nilfs_discontinued(nilfs)) {
2196                 nilfs_transaction_unlock(sb);
2197                 err = nilfs_segctor_sync(sci);
2198                 return err;
2199         }
2200
2201         spin_lock(&nilfs->ns_inode_lock);
2202         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2203             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2204                 spin_unlock(&nilfs->ns_inode_lock);
2205                 nilfs_transaction_unlock(sb);
2206                 return 0;
2207         }
2208         spin_unlock(&nilfs->ns_inode_lock);
2209         sci->sc_dsync_inode = ii;
2210         sci->sc_dsync_start = start;
2211         sci->sc_dsync_end = end;
2212
2213         err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2214
2215         nilfs_transaction_unlock(sb);
2216         return err;
2217 }
2218
2219 #define FLUSH_FILE_BIT  (0x1) /* data file only */
2220 #define FLUSH_DAT_BIT   (1 << NILFS_DAT_INO) /* DAT only */
2221
2222 /**
2223  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2224  * @sci: segment constructor object
2225  */
2226 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2227 {
2228         spin_lock(&sci->sc_state_lock);
2229         sci->sc_seq_accepted = sci->sc_seq_request;
2230         spin_unlock(&sci->sc_state_lock);
2231         del_timer_sync(&sci->sc_timer);
2232 }
2233
2234 /**
2235  * nilfs_segctor_notify - notify the result of request to caller threads
2236  * @sci: segment constructor object
2237  * @mode: mode of log forming
2238  * @err: error code to be notified
2239  */
2240 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2241 {
2242         /* Clear requests (even when the construction failed) */
2243         spin_lock(&sci->sc_state_lock);
2244
2245         if (mode == SC_LSEG_SR) {
2246                 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2247                 sci->sc_seq_done = sci->sc_seq_accepted;
2248                 nilfs_segctor_wakeup(sci, err);
2249                 sci->sc_flush_request = 0;
2250         } else {
2251                 if (mode == SC_FLUSH_FILE)
2252                         sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2253                 else if (mode == SC_FLUSH_DAT)
2254                         sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2255
2256                 /* re-enable timer if checkpoint creation was not done */
2257                 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2258                     time_before(jiffies, sci->sc_timer.expires))
2259                         add_timer(&sci->sc_timer);
2260         }
2261         spin_unlock(&sci->sc_state_lock);
2262 }
2263
2264 /**
2265  * nilfs_segctor_construct - form logs and write them to disk
2266  * @sci: segment constructor object
2267  * @mode: mode of log forming
2268  */
2269 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2270 {
2271         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2272         struct nilfs_super_block **sbp;
2273         int err = 0;
2274
2275         nilfs_segctor_accept(sci);
2276
2277         if (nilfs_discontinued(nilfs))
2278                 mode = SC_LSEG_SR;
2279         if (!nilfs_segctor_confirm(sci))
2280                 err = nilfs_segctor_do_construct(sci, mode);
2281
2282         if (likely(!err)) {
2283                 if (mode != SC_FLUSH_DAT)
2284                         atomic_set(&nilfs->ns_ndirtyblks, 0);
2285                 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2286                     nilfs_discontinued(nilfs)) {
2287                         down_write(&nilfs->ns_sem);
2288                         err = -EIO;
2289                         sbp = nilfs_prepare_super(sci->sc_super,
2290                                                   nilfs_sb_will_flip(nilfs));
2291                         if (likely(sbp)) {
2292                                 nilfs_set_log_cursor(sbp[0], nilfs);
2293                                 err = nilfs_commit_super(sci->sc_super,
2294                                                          NILFS_SB_COMMIT);
2295                         }
2296                         up_write(&nilfs->ns_sem);
2297                 }
2298         }
2299
2300         nilfs_segctor_notify(sci, mode, err);
2301         return err;
2302 }
2303
2304 static void nilfs_construction_timeout(unsigned long data)
2305 {
2306         struct task_struct *p = (struct task_struct *)data;
2307         wake_up_process(p);
2308 }
2309
2310 static void
2311 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2312 {
2313         struct nilfs_inode_info *ii, *n;
2314
2315         list_for_each_entry_safe(ii, n, head, i_dirty) {
2316                 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2317                         continue;
2318                 list_del_init(&ii->i_dirty);
2319                 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2320                 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
2321                 iput(&ii->vfs_inode);
2322         }
2323 }
2324
2325 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2326                          void **kbufs)
2327 {
2328         struct the_nilfs *nilfs = sb->s_fs_info;
2329         struct nilfs_sc_info *sci = nilfs->ns_writer;
2330         struct nilfs_transaction_info ti;
2331         int err;
2332
2333         if (unlikely(!sci))
2334                 return -EROFS;
2335
2336         nilfs_transaction_lock(sb, &ti, 1);
2337
2338         err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2339         if (unlikely(err))
2340                 goto out_unlock;
2341
2342         err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2343         if (unlikely(err)) {
2344                 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2345                 goto out_unlock;
2346         }
2347
2348         sci->sc_freesegs = kbufs[4];
2349         sci->sc_nfreesegs = argv[4].v_nmembs;
2350         list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2351
2352         for (;;) {
2353                 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2354                 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2355
2356                 if (likely(!err))
2357                         break;
2358
2359                 nilfs_warning(sb, __func__,
2360                               "segment construction failed. (err=%d)", err);
2361                 set_current_state(TASK_INTERRUPTIBLE);
2362                 schedule_timeout(sci->sc_interval);
2363         }
2364         if (nilfs_test_opt(nilfs, DISCARD)) {
2365                 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2366                                                  sci->sc_nfreesegs);
2367                 if (ret) {
2368                         printk(KERN_WARNING
2369                                "NILFS warning: error %d on discard request, "
2370                                "turning discards off for the device\n", ret);
2371                         nilfs_clear_opt(nilfs, DISCARD);
2372                 }
2373         }
2374
2375  out_unlock:
2376         sci->sc_freesegs = NULL;
2377         sci->sc_nfreesegs = 0;
2378         nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2379         nilfs_transaction_unlock(sb);
2380         return err;
2381 }
2382
2383 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2384 {
2385         struct nilfs_transaction_info ti;
2386
2387         nilfs_transaction_lock(sci->sc_super, &ti, 0);
2388         nilfs_segctor_construct(sci, mode);
2389
2390         /*
2391          * Unclosed segment should be retried.  We do this using sc_timer.
2392          * Timeout of sc_timer will invoke complete construction which leads
2393          * to close the current logical segment.
2394          */
2395         if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2396                 nilfs_segctor_start_timer(sci);
2397
2398         nilfs_transaction_unlock(sci->sc_super);
2399 }
2400
2401 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2402 {
2403         int mode = 0;
2404         int err;
2405
2406         spin_lock(&sci->sc_state_lock);
2407         mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2408                 SC_FLUSH_DAT : SC_FLUSH_FILE;
2409         spin_unlock(&sci->sc_state_lock);
2410
2411         if (mode) {
2412                 err = nilfs_segctor_do_construct(sci, mode);
2413
2414                 spin_lock(&sci->sc_state_lock);
2415                 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2416                         ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2417                 spin_unlock(&sci->sc_state_lock);
2418         }
2419         clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2420 }
2421
2422 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2423 {
2424         if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2425             time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2426                 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2427                         return SC_FLUSH_FILE;
2428                 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2429                         return SC_FLUSH_DAT;
2430         }
2431         return SC_LSEG_SR;
2432 }
2433
2434 /**
2435  * nilfs_segctor_thread - main loop of the segment constructor thread.
2436  * @arg: pointer to a struct nilfs_sc_info.
2437  *
2438  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2439  * to execute segment constructions.
2440  */
2441 static int nilfs_segctor_thread(void *arg)
2442 {
2443         struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2444         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2445         int timeout = 0;
2446
2447         sci->sc_timer.data = (unsigned long)current;
2448         sci->sc_timer.function = nilfs_construction_timeout;
2449
2450         /* start sync. */
2451         sci->sc_task = current;
2452         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2453         printk(KERN_INFO
2454                "segctord starting. Construction interval = %lu seconds, "
2455                "CP frequency < %lu seconds\n",
2456                sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2457
2458         spin_lock(&sci->sc_state_lock);
2459  loop:
2460         for (;;) {
2461                 int mode;
2462
2463                 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2464                         goto end_thread;
2465
2466                 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2467                         mode = SC_LSEG_SR;
2468                 else if (!sci->sc_flush_request)
2469                         break;
2470                 else
2471                         mode = nilfs_segctor_flush_mode(sci);
2472
2473                 spin_unlock(&sci->sc_state_lock);
2474                 nilfs_segctor_thread_construct(sci, mode);
2475                 spin_lock(&sci->sc_state_lock);
2476                 timeout = 0;
2477         }
2478
2479
2480         if (freezing(current)) {
2481                 spin_unlock(&sci->sc_state_lock);
2482                 refrigerator();
2483                 spin_lock(&sci->sc_state_lock);
2484         } else {
2485                 DEFINE_WAIT(wait);
2486                 int should_sleep = 1;
2487
2488                 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2489                                 TASK_INTERRUPTIBLE);
2490
2491                 if (sci->sc_seq_request != sci->sc_seq_done)
2492                         should_sleep = 0;
2493                 else if (sci->sc_flush_request)
2494                         should_sleep = 0;
2495                 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2496                         should_sleep = time_before(jiffies,
2497                                         sci->sc_timer.expires);
2498
2499                 if (should_sleep) {
2500                         spin_unlock(&sci->sc_state_lock);
2501                         schedule();
2502                         spin_lock(&sci->sc_state_lock);
2503                 }
2504                 finish_wait(&sci->sc_wait_daemon, &wait);
2505                 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2506                            time_after_eq(jiffies, sci->sc_timer.expires));
2507
2508                 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2509                         set_nilfs_discontinued(nilfs);
2510         }
2511         goto loop;
2512
2513  end_thread:
2514         spin_unlock(&sci->sc_state_lock);
2515
2516         /* end sync. */
2517         sci->sc_task = NULL;
2518         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2519         return 0;
2520 }
2521
2522 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2523 {
2524         struct task_struct *t;
2525
2526         t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2527         if (IS_ERR(t)) {
2528                 int err = PTR_ERR(t);
2529
2530                 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2531                        err);
2532                 return err;
2533         }
2534         wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2535         return 0;
2536 }
2537
2538 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2539         __acquires(&sci->sc_state_lock)
2540         __releases(&sci->sc_state_lock)
2541 {
2542         sci->sc_state |= NILFS_SEGCTOR_QUIT;
2543
2544         while (sci->sc_task) {
2545                 wake_up(&sci->sc_wait_daemon);
2546                 spin_unlock(&sci->sc_state_lock);
2547                 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2548                 spin_lock(&sci->sc_state_lock);
2549         }
2550 }
2551
2552 /*
2553  * Setup & clean-up functions
2554  */
2555 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2556                                                struct nilfs_root *root)
2557 {
2558         struct the_nilfs *nilfs = sb->s_fs_info;
2559         struct nilfs_sc_info *sci;
2560
2561         sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2562         if (!sci)
2563                 return NULL;
2564
2565         sci->sc_super = sb;
2566
2567         nilfs_get_root(root);
2568         sci->sc_root = root;
2569
2570         init_waitqueue_head(&sci->sc_wait_request);
2571         init_waitqueue_head(&sci->sc_wait_daemon);
2572         init_waitqueue_head(&sci->sc_wait_task);
2573         spin_lock_init(&sci->sc_state_lock);
2574         INIT_LIST_HEAD(&sci->sc_dirty_files);
2575         INIT_LIST_HEAD(&sci->sc_segbufs);
2576         INIT_LIST_HEAD(&sci->sc_write_logs);
2577         INIT_LIST_HEAD(&sci->sc_gc_inodes);
2578         init_timer(&sci->sc_timer);
2579
2580         sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2581         sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2582         sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2583
2584         if (nilfs->ns_interval)
2585                 sci->sc_interval = HZ * nilfs->ns_interval;
2586         if (nilfs->ns_watermark)
2587                 sci->sc_watermark = nilfs->ns_watermark;
2588         return sci;
2589 }
2590
2591 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2592 {
2593         int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2594
2595         /* The segctord thread was stopped and its timer was removed.
2596            But some tasks remain. */
2597         do {
2598                 struct nilfs_transaction_info ti;
2599
2600                 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2601                 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2602                 nilfs_transaction_unlock(sci->sc_super);
2603
2604         } while (ret && retrycount-- > 0);
2605 }
2606
2607 /**
2608  * nilfs_segctor_destroy - destroy the segment constructor.
2609  * @sci: nilfs_sc_info
2610  *
2611  * nilfs_segctor_destroy() kills the segctord thread and frees
2612  * the nilfs_sc_info struct.
2613  * Caller must hold the segment semaphore.
2614  */
2615 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2616 {
2617         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2618         int flag;
2619
2620         up_write(&nilfs->ns_segctor_sem);
2621
2622         spin_lock(&sci->sc_state_lock);
2623         nilfs_segctor_kill_thread(sci);
2624         flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2625                 || sci->sc_seq_request != sci->sc_seq_done);
2626         spin_unlock(&sci->sc_state_lock);
2627
2628         if (flag || !nilfs_segctor_confirm(sci))
2629                 nilfs_segctor_write_out(sci);
2630
2631         if (!list_empty(&sci->sc_dirty_files)) {
2632                 nilfs_warning(sci->sc_super, __func__,
2633                               "dirty file(s) after the final construction\n");
2634                 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2635         }
2636
2637         WARN_ON(!list_empty(&sci->sc_segbufs));
2638         WARN_ON(!list_empty(&sci->sc_write_logs));
2639
2640         nilfs_put_root(sci->sc_root);
2641
2642         down_write(&nilfs->ns_segctor_sem);
2643
2644         del_timer_sync(&sci->sc_timer);
2645         kfree(sci);
2646 }
2647
2648 /**
2649  * nilfs_attach_log_writer - attach log writer
2650  * @sb: super block instance
2651  * @root: root object of the current filesystem tree
2652  *
2653  * This allocates a log writer object, initializes it, and starts the
2654  * log writer.
2655  *
2656  * Return Value: On success, 0 is returned. On error, one of the following
2657  * negative error code is returned.
2658  *
2659  * %-ENOMEM - Insufficient memory available.
2660  */
2661 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2662 {
2663         struct the_nilfs *nilfs = sb->s_fs_info;
2664         int err;
2665
2666         if (nilfs->ns_writer) {
2667                 /*
2668                  * This happens if the filesystem was remounted
2669                  * read/write after nilfs_error degenerated it into a
2670                  * read-only mount.
2671                  */
2672                 nilfs_detach_log_writer(sb);
2673         }
2674
2675         nilfs->ns_writer = nilfs_segctor_new(sb, root);
2676         if (!nilfs->ns_writer)
2677                 return -ENOMEM;
2678
2679         err = nilfs_segctor_start_thread(nilfs->ns_writer);
2680         if (err) {
2681                 kfree(nilfs->ns_writer);
2682                 nilfs->ns_writer = NULL;
2683         }
2684         return err;
2685 }
2686
2687 /**
2688  * nilfs_detach_log_writer - destroy log writer
2689  * @sb: super block instance
2690  *
2691  * This kills log writer daemon, frees the log writer object, and
2692  * destroys list of dirty files.
2693  */
2694 void nilfs_detach_log_writer(struct super_block *sb)
2695 {
2696         struct the_nilfs *nilfs = sb->s_fs_info;
2697         LIST_HEAD(garbage_list);
2698
2699         down_write(&nilfs->ns_segctor_sem);
2700         if (nilfs->ns_writer) {
2701                 nilfs_segctor_destroy(nilfs->ns_writer);
2702                 nilfs->ns_writer = NULL;
2703         }
2704
2705         /* Force to free the list of dirty files */
2706         spin_lock(&nilfs->ns_inode_lock);
2707         if (!list_empty(&nilfs->ns_dirty_files)) {
2708                 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2709                 nilfs_warning(sb, __func__,
2710                               "Hit dirty file after stopped log writer\n");
2711         }
2712         spin_unlock(&nilfs->ns_inode_lock);
2713         up_write(&nilfs->ns_segctor_sem);
2714
2715         nilfs_dispose_list(nilfs, &garbage_list, 1);
2716 }