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 "nilfs.h"
  36 #include "btnode.h"
  37 #include "page.h"
  38 #include "segment.h"
  39 #include "sufile.h"
  40 #include "cpfile.h"
  41 #include "ifile.h"
  42 #include "seglist.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_SKETCH,
  71         NILFS_ST_IFILE,
  72         NILFS_ST_CPFILE,
  73         NILFS_ST_SUFILE,
  74         NILFS_ST_DAT,
  75         NILFS_ST_SR,            /* Super root */
  76         NILFS_ST_DSYNC,         /* Data sync blocks */
  77         NILFS_ST_DONE,
  78 };
  79
  80 /* State flags of collection */
  81 #define NILFS_CF_NODE           0x0001  /* Collecting node blocks */
  82 #define NILFS_CF_IFILE_STARTED  0x0002  /* IFILE stage has started */
  83 #define NILFS_CF_HISTORY_MASK   (NILFS_CF_IFILE_STARTED)
  84
  85 /* Operations depending on the construction mode and file type */
  86 struct nilfs_sc_operations {
  87         int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
  88                             struct inode *);
  89         int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
  90                             struct inode *);
  91         int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
  92                             struct inode *);
  93         void (*write_data_binfo)(struct nilfs_sc_info *,
  94                                  struct nilfs_segsum_pointer *,
  95                                  union nilfs_binfo *);
  96         void (*write_node_binfo)(struct nilfs_sc_info *,
  97                                  struct nilfs_segsum_pointer *,
  98                                  union nilfs_binfo *);
  99 };
 100
 101 /*
 102  * Other definitions
 103  */
 104 static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
 105 static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
 106 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
 107 static void nilfs_dispose_list(struct nilfs_sb_info *, struct list_head *,
 108                                int);
 109
 110 #define nilfs_cnt32_gt(a, b)   \
 111         (typecheck(__u32, a) && typecheck(__u32, b) && \
 112          ((__s32)(b) - (__s32)(a) < 0))
 113 #define nilfs_cnt32_ge(a, b)   \
 114         (typecheck(__u32, a) && typecheck(__u32, b) && \
 115          ((__s32)(a) - (__s32)(b) >= 0))
 116 #define nilfs_cnt32_lt(a, b)  nilfs_cnt32_gt(b, a)
 117 #define nilfs_cnt32_le(a, b)  nilfs_cnt32_ge(b, a)
 118
 119 /*
 120  * Transaction
 121  */
 122 static struct kmem_cache *nilfs_transaction_cachep;
 123
 124 /**
 125  * nilfs_init_transaction_cache - create a cache for nilfs_transaction_info
 126  *
 127  * nilfs_init_transaction_cache() creates a slab cache for the struct
 128  * nilfs_transaction_info.
 129  *
 130  * Return Value: On success, it returns 0. On error, one of the following
 131  * negative error code is returned.
 132  *
 133  * %-ENOMEM - Insufficient memory available.
 134  */
 135 int nilfs_init_transaction_cache(void)
 136 {
 137         nilfs_transaction_cachep =
 138                 kmem_cache_create("nilfs2_transaction_cache",
 139                                   sizeof(struct nilfs_transaction_info),
 140                                   0, SLAB_RECLAIM_ACCOUNT, NULL);
 141         return (nilfs_transaction_cachep == NULL) ? -ENOMEM : 0;
 142 }
 143
 144 /**
 145  * nilfs_detroy_transaction_cache - destroy the cache for transaction info
 146  *
 147  * nilfs_destroy_transaction_cache() frees the slab cache for the struct
 148  * nilfs_transaction_info.
 149  */
 150 void nilfs_destroy_transaction_cache(void)
 151 {
 152         kmem_cache_destroy(nilfs_transaction_cachep);
 153 }
 154
 155 static int nilfs_prepare_segment_lock(struct nilfs_transaction_info *ti)
 156 {
 157         struct nilfs_transaction_info *cur_ti = current->journal_info;
 158         void *save = NULL;
 159
 160         if (cur_ti) {
 161                 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
 162                         return ++cur_ti->ti_count;
 163                 else {
 164                         /*
 165                          * If journal_info field is occupied by other FS,
 166                          * we save it and restore on nilfs_transaction_end().
 167                          * But this should never happen.
 168                          */
 169                         printk(KERN_WARNING
 170                                "NILFS warning: journal info from a different "
 171                                "FS\n");
 172                         save = current->journal_info;
 173                 }
 174         }
 175         if (!ti) {
 176                 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
 177                 if (!ti)
 178                         return -ENOMEM;
 179                 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
 180         } else {
 181                 ti->ti_flags = 0;
 182         }
 183         ti->ti_count = 0;
 184         ti->ti_save = save;
 185         ti->ti_magic = NILFS_TI_MAGIC;
 186         current->journal_info = ti;
 187         return 0;
 188 }
 189
 190 /**
 191  * nilfs_transaction_begin - start indivisible file operations.
 192  * @sb: super block
 193  * @ti: nilfs_transaction_info
 194  * @vacancy_check: flags for vacancy rate checks
 195  *
 196  * nilfs_transaction_begin() acquires a reader/writer semaphore, called
 197  * the segment semaphore, to make a segment construction and write tasks
 198  * exclusive.  The function is used with nilfs_transaction_end() in pairs.
 199  * The region enclosed by these two functions can be nested.  To avoid a
 200  * deadlock, the semaphore is only acquired or released in the outermost call.
 201  *
 202  * This function allocates a nilfs_transaction_info struct to keep context
 203  * information on it.  It is initialized and hooked onto the current task in
 204  * the outermost call.  If a pre-allocated struct is given to @ti, it is used
 205  * instead; othewise a new struct is assigned from a slab.
 206  *
 207  * When @vacancy_check flag is set, this function will check the amount of
 208  * free space, and will wait for the GC to reclaim disk space if low capacity.
 209  *
 210  * Return Value: On success, 0 is returned. On error, one of the following
 211  * negative error code is returned.
 212  *
 213  * %-ENOMEM - Insufficient memory available.
 214  *
 215  * %-ERESTARTSYS - Interrupted
 216  *
 217  * %-ENOSPC - No space left on device
 218  */
 219 int nilfs_transaction_begin(struct super_block *sb,
 220                             struct nilfs_transaction_info *ti,
 221                             int vacancy_check)
 222 {
 223         struct nilfs_sb_info *sbi;
 224         struct the_nilfs *nilfs;
 225         int ret = nilfs_prepare_segment_lock(ti);
 226
 227         if (unlikely(ret < 0))
 228                 return ret;
 229         if (ret > 0)
 230                 return 0;
 231
 232         sbi = NILFS_SB(sb);
 233         nilfs = sbi->s_nilfs;
 234         down_read(&nilfs->ns_segctor_sem);
 235         if (vacancy_check && nilfs_near_disk_full(nilfs)) {
 236                 up_read(&nilfs->ns_segctor_sem);
 237                 ret = -ENOSPC;
 238                 goto failed;
 239         }
 240         return 0;
 241
 242  failed:
 243         ti = current->journal_info;
 244         current->journal_info = ti->ti_save;
 245         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 246                 kmem_cache_free(nilfs_transaction_cachep, ti);
 247         return ret;
 248 }
 249
 250 /**
 251  * nilfs_transaction_end - end indivisible file operations.
 252  * @sb: super block
 253  * @commit: commit flag (0 for no change)
 254  *
 255  * nilfs_transaction_end() releases the read semaphore which is
 256  * acquired by nilfs_transaction_begin(). Its releasing is only done
 257  * in outermost call of this function. If the nilfs_transaction_info
 258  * was allocated dynamically, it is given back to a slab cache.
 259  */
 260 int nilfs_transaction_end(struct super_block *sb, int commit)
 261 {
 262         struct nilfs_transaction_info *ti = current->journal_info;
 263         struct nilfs_sb_info *sbi;
 264         struct nilfs_sc_info *sci;
 265         int err = 0;
 266
 267         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 268
 269         if (commit)
 270                 ti->ti_flags |= NILFS_TI_COMMIT;
 271         if (ti->ti_count > 0) {
 272                 ti->ti_count--;
 273                 return 0;
 274         }
 275         sbi = NILFS_SB(sb);
 276         sci = NILFS_SC(sbi);
 277         if (sci != NULL) {
 278                 if (ti->ti_flags & NILFS_TI_COMMIT)
 279                         nilfs_segctor_start_timer(sci);
 280                 if (atomic_read(&sbi->s_nilfs->ns_ndirtyblks) >
 281                     sci->sc_watermark)
 282                         nilfs_segctor_do_flush(sci, 0);
 283         }
 284         up_read(&sbi->s_nilfs->ns_segctor_sem);
 285         current->journal_info = ti->ti_save;
 286
 287         if (ti->ti_flags & NILFS_TI_SYNC)
 288                 err = nilfs_construct_segment(sb);
 289         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 290                 kmem_cache_free(nilfs_transaction_cachep, ti);
 291         return err;
 292 }
 293
 294 void nilfs_relax_pressure_in_lock(struct super_block *sb)
 295 {
 296         struct nilfs_sb_info *sbi = NILFS_SB(sb);
 297         struct nilfs_sc_info *sci = NILFS_SC(sbi);
 298         struct the_nilfs *nilfs = sbi->s_nilfs;
 299
 300         if (!sci || !sci->sc_flush_request)
 301                 return;
 302
 303         set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
 304         up_read(&nilfs->ns_segctor_sem);
 305
 306         down_write(&nilfs->ns_segctor_sem);
 307         if (sci->sc_flush_request &&
 308             test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
 309                 struct nilfs_transaction_info *ti = current->journal_info;
 310
 311                 ti->ti_flags |= NILFS_TI_WRITER;
 312                 nilfs_segctor_do_immediate_flush(sci);
 313                 ti->ti_flags &= ~NILFS_TI_WRITER;
 314         }
 315         downgrade_write(&nilfs->ns_segctor_sem);
 316 }
 317
 318 static void nilfs_transaction_lock(struct nilfs_sb_info *sbi,
 319                                    struct nilfs_transaction_info *ti,
 320                                    int gcflag)
 321 {
 322         struct nilfs_transaction_info *cur_ti = current->journal_info;
 323
 324         BUG_ON(cur_ti);
 325         BUG_ON(!ti);
 326         ti->ti_flags = NILFS_TI_WRITER;
 327         ti->ti_count = 0;
 328         ti->ti_save = cur_ti;
 329         ti->ti_magic = NILFS_TI_MAGIC;
 330         INIT_LIST_HEAD(&ti->ti_garbage);
 331         current->journal_info = ti;
 332
 333         for (;;) {
 334                 down_write(&sbi->s_nilfs->ns_segctor_sem);
 335                 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &NILFS_SC(sbi)->sc_flags))
 336                         break;
 337
 338                 nilfs_segctor_do_immediate_flush(NILFS_SC(sbi));
 339
 340                 up_write(&sbi->s_nilfs->ns_segctor_sem);
 341                 yield();
 342         }
 343         if (gcflag)
 344                 ti->ti_flags |= NILFS_TI_GC;
 345 }
 346
 347 static void nilfs_transaction_unlock(struct nilfs_sb_info *sbi)
 348 {
 349         struct nilfs_transaction_info *ti = current->journal_info;
 350
 351         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 352         BUG_ON(ti->ti_count > 0);
 353
 354         up_write(&sbi->s_nilfs->ns_segctor_sem);
 355         current->journal_info = ti->ti_save;
 356         if (!list_empty(&ti->ti_garbage))
 357                 nilfs_dispose_list(sbi, &ti->ti_garbage, 0);
 358 }
 359
 360 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
 361                                             struct nilfs_segsum_pointer *ssp,
 362                                             unsigned bytes)
 363 {
 364         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 365         unsigned blocksize = sci->sc_super->s_blocksize;
 366         void *p;
 367
 368         if (unlikely(ssp->offset + bytes > blocksize)) {
 369                 ssp->offset = 0;
 370                 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
 371                                                &segbuf->sb_segsum_buffers));
 372                 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
 373         }
 374         p = ssp->bh->b_data + ssp->offset;
 375         ssp->offset += bytes;
 376         return p;
 377 }
 378
 379 /**
 380  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
 381  * @sci: nilfs_sc_info
 382  */
 383 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
 384 {
 385         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 386         struct buffer_head *sumbh;
 387         unsigned sumbytes;
 388         unsigned flags = 0;
 389         int err;
 390
 391         if (nilfs_doing_gc())
 392                 flags = NILFS_SS_GC;
 393         err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime);
 394         if (unlikely(err))
 395                 return err;
 396
 397         sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
 398         sumbytes = segbuf->sb_sum.sumbytes;
 399         sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
 400         sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
 401         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 402         return 0;
 403 }
 404
 405 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
 406 {
 407         sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
 408         if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
 409                 return -E2BIG; /* The current segment is filled up
 410                                   (internal code) */
 411         sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
 412         return nilfs_segctor_reset_segment_buffer(sci);
 413 }
 414
 415 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
 416 {
 417         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 418         int err;
 419
 420         if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
 421                 err = nilfs_segctor_feed_segment(sci);
 422                 if (err)
 423                         return err;
 424                 segbuf = sci->sc_curseg;
 425         }
 426         err = nilfs_segbuf_extend_payload(segbuf, &sci->sc_super_root);
 427         if (likely(!err))
 428                 segbuf->sb_sum.flags |= NILFS_SS_SR;
 429         return err;
 430 }
 431
 432 /*
 433  * Functions for making segment summary and payloads
 434  */
 435 static int nilfs_segctor_segsum_block_required(
 436         struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
 437         unsigned binfo_size)
 438 {
 439         unsigned blocksize = sci->sc_super->s_blocksize;
 440         /* Size of finfo and binfo is enough small against blocksize */
 441
 442         return ssp->offset + binfo_size +
 443                 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
 444                 blocksize;
 445 }
 446
 447 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
 448                                       struct inode *inode)
 449 {
 450         sci->sc_curseg->sb_sum.nfinfo++;
 451         sci->sc_binfo_ptr = sci->sc_finfo_ptr;
 452         nilfs_segctor_map_segsum_entry(
 453                 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
 454         /* skip finfo */
 455 }
 456
 457 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
 458                                     struct inode *inode)
 459 {
 460         struct nilfs_finfo *finfo;
 461         struct nilfs_inode_info *ii;
 462         struct nilfs_segment_buffer *segbuf;
 463
 464         if (sci->sc_blk_cnt == 0)
 465                 return;
 466
 467         ii = NILFS_I(inode);
 468         finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
 469                                                  sizeof(*finfo));
 470         finfo->fi_ino = cpu_to_le64(inode->i_ino);
 471         finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
 472         finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
 473         finfo->fi_cno = cpu_to_le64(ii->i_cno);
 474
 475         segbuf = sci->sc_curseg;
 476         segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
 477                 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
 478         sci->sc_finfo_ptr = sci->sc_binfo_ptr;
 479         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 480 }
 481
 482 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
 483                                         struct buffer_head *bh,
 484                                         struct inode *inode,
 485                                         unsigned binfo_size)
 486 {
 487         struct nilfs_segment_buffer *segbuf;
 488         int required, err = 0;
 489
 490  retry:
 491         segbuf = sci->sc_curseg;
 492         required = nilfs_segctor_segsum_block_required(
 493                 sci, &sci->sc_binfo_ptr, binfo_size);
 494         if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
 495                 nilfs_segctor_end_finfo(sci, inode);
 496                 err = nilfs_segctor_feed_segment(sci);
 497                 if (err)
 498                         return err;
 499                 goto retry;
 500         }
 501         if (unlikely(required)) {
 502                 err = nilfs_segbuf_extend_segsum(segbuf);
 503                 if (unlikely(err))
 504                         goto failed;
 505         }
 506         if (sci->sc_blk_cnt == 0)
 507                 nilfs_segctor_begin_finfo(sci, inode);
 508
 509         nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
 510         /* Substitution to vblocknr is delayed until update_blocknr() */
 511         nilfs_segbuf_add_file_buffer(segbuf, bh);
 512         sci->sc_blk_cnt++;
 513  failed:
 514         return err;
 515 }
 516
 517 static int nilfs_handle_bmap_error(int err, const char *fname,
 518                                    struct inode *inode, struct super_block *sb)
 519 {
 520         if (err == -EINVAL) {
 521                 nilfs_error(sb, fname, "broken bmap (inode=%lu)\n",
 522                             inode->i_ino);
 523                 err = -EIO;
 524         }
 525         return err;
 526 }
 527
 528 /*
 529  * Callback functions that enumerate, mark, and collect dirty blocks
 530  */
 531 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
 532                                    struct buffer_head *bh, struct inode *inode)
 533 {
 534         int err;
 535
 536         /* BUG_ON(!buffer_dirty(bh)); */
 537         /* excluded by scan_dirty_data_buffers() */
 538         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 539         if (unlikely(err < 0))
 540                 return nilfs_handle_bmap_error(err, __func__, inode,
 541                                                sci->sc_super);
 542
 543         err = nilfs_segctor_add_file_block(sci, bh, inode,
 544                                            sizeof(struct nilfs_binfo_v));
 545         if (!err)
 546                 sci->sc_datablk_cnt++;
 547         return err;
 548 }
 549
 550 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
 551                                    struct buffer_head *bh,
 552                                    struct inode *inode)
 553 {
 554         int err;
 555
 556         /* BUG_ON(!buffer_dirty(bh)); */
 557         /* excluded by scan_dirty_node_buffers() */
 558         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 559         if (unlikely(err < 0))
 560                 return nilfs_handle_bmap_error(err, __func__, inode,
 561                                                sci->sc_super);
 562         return 0;
 563 }
 564
 565 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
 566                                    struct buffer_head *bh,
 567                                    struct inode *inode)
 568 {
 569         BUG_ON(!buffer_dirty(bh));
 570         return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 571 }
 572
 573 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
 574                                         struct nilfs_segsum_pointer *ssp,
 575                                         union nilfs_binfo *binfo)
 576 {
 577         struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
 578                 sci, ssp, sizeof(*binfo_v));
 579         *binfo_v = binfo->bi_v;
 580 }
 581
 582 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
 583                                         struct nilfs_segsum_pointer *ssp,
 584                                         union nilfs_binfo *binfo)
 585 {
 586         __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
 587                 sci, ssp, sizeof(*vblocknr));
 588         *vblocknr = binfo->bi_v.bi_vblocknr;
 589 }
 590
 591 struct nilfs_sc_operations nilfs_sc_file_ops = {
 592         .collect_data = nilfs_collect_file_data,
 593         .collect_node = nilfs_collect_file_node,
 594         .collect_bmap = nilfs_collect_file_bmap,
 595         .write_data_binfo = nilfs_write_file_data_binfo,
 596         .write_node_binfo = nilfs_write_file_node_binfo,
 597 };
 598
 599 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
 600                                   struct buffer_head *bh, struct inode *inode)
 601 {
 602         int err;
 603
 604         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 605         if (unlikely(err < 0))
 606                 return nilfs_handle_bmap_error(err, __func__, inode,
 607                                                sci->sc_super);
 608
 609         err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 610         if (!err)
 611                 sci->sc_datablk_cnt++;
 612         return err;
 613 }
 614
 615 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
 616                                   struct buffer_head *bh, struct inode *inode)
 617 {
 618         BUG_ON(!buffer_dirty(bh));
 619         return nilfs_segctor_add_file_block(sci, bh, inode,
 620                                             sizeof(struct nilfs_binfo_dat));
 621 }
 622
 623 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
 624                                        struct nilfs_segsum_pointer *ssp,
 625                                        union nilfs_binfo *binfo)
 626 {
 627         __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
 628                                                           sizeof(*blkoff));
 629         *blkoff = binfo->bi_dat.bi_blkoff;
 630 }
 631
 632 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
 633                                        struct nilfs_segsum_pointer *ssp,
 634                                        union nilfs_binfo *binfo)
 635 {
 636         struct nilfs_binfo_dat *binfo_dat =
 637                 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
 638         *binfo_dat = binfo->bi_dat;
 639 }
 640
 641 struct nilfs_sc_operations nilfs_sc_dat_ops = {
 642         .collect_data = nilfs_collect_dat_data,
 643         .collect_node = nilfs_collect_file_node,
 644         .collect_bmap = nilfs_collect_dat_bmap,
 645         .write_data_binfo = nilfs_write_dat_data_binfo,
 646         .write_node_binfo = nilfs_write_dat_node_binfo,
 647 };
 648
 649 struct nilfs_sc_operations nilfs_sc_dsync_ops = {
 650         .collect_data = nilfs_collect_file_data,
 651         .collect_node = NULL,
 652         .collect_bmap = NULL,
 653         .write_data_binfo = nilfs_write_file_data_binfo,
 654         .write_node_binfo = NULL,
 655 };
 656
 657 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
 658                                               struct list_head *listp,
 659                                               size_t nlimit,
 660                                               loff_t start, loff_t end)
 661 {
 662         struct address_space *mapping = inode->i_mapping;
 663         struct pagevec pvec;
 664         pgoff_t index = 0, last = ULONG_MAX;
 665         size_t ndirties = 0;
 666         int i;
 667
 668         if (unlikely(start != 0 || end != LLONG_MAX)) {
 669                 /*
 670                  * A valid range is given for sync-ing data pages. The
 671                  * range is rounded to per-page; extra dirty buffers
 672                  * may be included if blocksize < pagesize.
 673                  */
 674                 index = start >> PAGE_SHIFT;
 675                 last = end >> PAGE_SHIFT;
 676         }
 677         pagevec_init(&pvec, 0);
 678  repeat:
 679         if (unlikely(index > last) ||
 680             !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 681                                 min_t(pgoff_t, last - index,
 682                                       PAGEVEC_SIZE - 1) + 1))
 683                 return ndirties;
 684
 685         for (i = 0; i < pagevec_count(&pvec); i++) {
 686                 struct buffer_head *bh, *head;
 687                 struct page *page = pvec.pages[i];
 688
 689                 if (unlikely(page->index > last))
 690                         break;
 691
 692                 if (mapping->host) {
 693                         lock_page(page);
 694                         if (!page_has_buffers(page))
 695                                 create_empty_buffers(page,
 696                                                      1 << inode->i_blkbits, 0);
 697                         unlock_page(page);
 698                 }
 699
 700                 bh = head = page_buffers(page);
 701                 do {
 702                         if (!buffer_dirty(bh))
 703                                 continue;
 704                         get_bh(bh);
 705                         list_add_tail(&bh->b_assoc_buffers, listp);
 706                         ndirties++;
 707                         if (unlikely(ndirties >= nlimit)) {
 708                                 pagevec_release(&pvec);
 709                                 cond_resched();
 710                                 return ndirties;
 711                         }
 712                 } while (bh = bh->b_this_page, bh != head);
 713         }
 714         pagevec_release(&pvec);
 715         cond_resched();
 716         goto repeat;
 717 }
 718
 719 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
 720                                             struct list_head *listp)
 721 {
 722         struct nilfs_inode_info *ii = NILFS_I(inode);
 723         struct address_space *mapping = &ii->i_btnode_cache;
 724         struct pagevec pvec;
 725         struct buffer_head *bh, *head;
 726         unsigned int i;
 727         pgoff_t index = 0;
 728
 729         pagevec_init(&pvec, 0);
 730
 731         while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 732                                   PAGEVEC_SIZE)) {
 733                 for (i = 0; i < pagevec_count(&pvec); i++) {
 734                         bh = head = page_buffers(pvec.pages[i]);
 735                         do {
 736                                 if (buffer_dirty(bh)) {
 737                                         get_bh(bh);
 738                                         list_add_tail(&bh->b_assoc_buffers,
 739                                                       listp);
 740                                 }
 741                                 bh = bh->b_this_page;
 742                         } while (bh != head);
 743                 }
 744                 pagevec_release(&pvec);
 745                 cond_resched();
 746         }
 747 }
 748
 749 static void nilfs_dispose_list(struct nilfs_sb_info *sbi,
 750                                struct list_head *head, int force)
 751 {
 752         struct nilfs_inode_info *ii, *n;
 753         struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
 754         unsigned nv = 0;
 755
 756         while (!list_empty(head)) {
 757                 spin_lock(&sbi->s_inode_lock);
 758                 list_for_each_entry_safe(ii, n, head, i_dirty) {
 759                         list_del_init(&ii->i_dirty);
 760                         if (force) {
 761                                 if (unlikely(ii->i_bh)) {
 762                                         brelse(ii->i_bh);
 763                                         ii->i_bh = NULL;
 764                                 }
 765                         } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
 766                                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 767                                 list_add_tail(&ii->i_dirty,
 768                                               &sbi->s_dirty_files);
 769                                 continue;
 770                         }
 771                         ivec[nv++] = ii;
 772                         if (nv == SC_N_INODEVEC)
 773                                 break;
 774                 }
 775                 spin_unlock(&sbi->s_inode_lock);
 776
 777                 for (pii = ivec; nv > 0; pii++, nv--)
 778                         iput(&(*pii)->vfs_inode);
 779         }
 780 }
 781
 782 static int nilfs_test_metadata_dirty(struct nilfs_sb_info *sbi)
 783 {
 784         struct the_nilfs *nilfs = sbi->s_nilfs;
 785         int ret = 0;
 786
 787         if (nilfs_mdt_fetch_dirty(sbi->s_ifile))
 788                 ret++;
 789         if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
 790                 ret++;
 791         if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
 792                 ret++;
 793         if (ret || nilfs_doing_gc())
 794                 if (nilfs_mdt_fetch_dirty(nilfs_dat_inode(nilfs)))
 795                         ret++;
 796         return ret;
 797 }
 798
 799 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
 800 {
 801         return list_empty(&sci->sc_dirty_files) &&
 802                 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
 803                 list_empty(&sci->sc_cleaning_segments) &&
 804                 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
 805 }
 806
 807 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
 808 {
 809         struct nilfs_sb_info *sbi = sci->sc_sbi;
 810         int ret = 0;
 811
 812         if (nilfs_test_metadata_dirty(sbi))
 813                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
 814
 815         spin_lock(&sbi->s_inode_lock);
 816         if (list_empty(&sbi->s_dirty_files) && nilfs_segctor_clean(sci))
 817                 ret++;
 818
 819         spin_unlock(&sbi->s_inode_lock);
 820         return ret;
 821 }
 822
 823 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
 824 {
 825         struct nilfs_sb_info *sbi = sci->sc_sbi;
 826         struct the_nilfs *nilfs = sbi->s_nilfs;
 827
 828         nilfs_mdt_clear_dirty(sbi->s_ifile);
 829         nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
 830         nilfs_mdt_clear_dirty(nilfs->ns_sufile);
 831         nilfs_mdt_clear_dirty(nilfs_dat_inode(nilfs));
 832 }
 833
 834 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
 835 {
 836         struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
 837         struct buffer_head *bh_cp;
 838         struct nilfs_checkpoint *raw_cp;
 839         int err;
 840
 841         /* XXX: this interface will be changed */
 842         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
 843                                           &raw_cp, &bh_cp);
 844         if (likely(!err)) {
 845                 /* The following code is duplicated with cpfile.  But, it is
 846                    needed to collect the checkpoint even if it was not newly
 847                    created */
 848                 nilfs_mdt_mark_buffer_dirty(bh_cp);
 849                 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
 850                 nilfs_cpfile_put_checkpoint(
 851                         nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 852         } else {
 853                 BUG_ON(err == -EINVAL || err == -ENOENT);
 854         }
 855         return err;
 856 }
 857
 858 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
 859 {
 860         struct nilfs_sb_info *sbi = sci->sc_sbi;
 861         struct the_nilfs *nilfs = sbi->s_nilfs;
 862         struct buffer_head *bh_cp;
 863         struct nilfs_checkpoint *raw_cp;
 864         int err;
 865
 866         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
 867                                           &raw_cp, &bh_cp);
 868         if (unlikely(err)) {
 869                 BUG_ON(err == -EINVAL || err == -ENOENT);
 870                 goto failed_ibh;
 871         }
 872         raw_cp->cp_snapshot_list.ssl_next = 0;
 873         raw_cp->cp_snapshot_list.ssl_prev = 0;
 874         raw_cp->cp_inodes_count =
 875                 cpu_to_le64(atomic_read(&sbi->s_inodes_count));
 876         raw_cp->cp_blocks_count =
 877                 cpu_to_le64(atomic_read(&sbi->s_blocks_count));
 878         raw_cp->cp_nblk_inc =
 879                 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
 880         raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
 881         raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
 882         if (sci->sc_sketch_inode && i_size_read(sci->sc_sketch_inode) > 0)
 883                 nilfs_checkpoint_set_sketch(raw_cp);
 884         nilfs_write_inode_common(sbi->s_ifile, &raw_cp->cp_ifile_inode, 1);
 885         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 886         return 0;
 887
 888  failed_ibh:
 889         return err;
 890 }
 891
 892 static void nilfs_fill_in_file_bmap(struct inode *ifile,
 893                                     struct nilfs_inode_info *ii)
 894
 895 {
 896         struct buffer_head *ibh;
 897         struct nilfs_inode *raw_inode;
 898
 899         if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
 900                 ibh = ii->i_bh;
 901                 BUG_ON(!ibh);
 902                 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
 903                                                   ibh);
 904                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 905                 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
 906         }
 907 }
 908
 909 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci,
 910                                             struct inode *ifile)
 911 {
 912         struct nilfs_inode_info *ii;
 913
 914         list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
 915                 nilfs_fill_in_file_bmap(ifile, ii);
 916                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
 917         }
 918         if (sci->sc_sketch_inode) {
 919                 ii = NILFS_I(sci->sc_sketch_inode);
 920                 if (test_bit(NILFS_I_DIRTY, &ii->i_state))
 921                         nilfs_fill_in_file_bmap(ifile, ii);
 922         }
 923 }
 924
 925 /*
 926  * CRC calculation routines
 927  */
 928 static void nilfs_fill_in_super_root_crc(struct buffer_head *bh_sr, u32 seed)
 929 {
 930         struct nilfs_super_root *raw_sr =
 931                 (struct nilfs_super_root *)bh_sr->b_data;
 932         u32 crc;
 933
 934         BUG_ON(NILFS_SR_BYTES > bh_sr->b_size);
 935         crc = crc32_le(seed,
 936                        (unsigned char *)raw_sr + sizeof(raw_sr->sr_sum),
 937                        NILFS_SR_BYTES - sizeof(raw_sr->sr_sum));
 938         raw_sr->sr_sum = cpu_to_le32(crc);
 939 }
 940
 941 static void nilfs_segctor_fill_in_checksums(struct nilfs_sc_info *sci,
 942                                             u32 seed)
 943 {
 944         struct nilfs_segment_buffer *segbuf;
 945
 946         if (sci->sc_super_root)
 947                 nilfs_fill_in_super_root_crc(sci->sc_super_root, seed);
 948
 949         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
 950                 nilfs_segbuf_fill_in_segsum_crc(segbuf, seed);
 951                 nilfs_segbuf_fill_in_data_crc(segbuf, seed);
 952         }
 953 }
 954
 955 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
 956                                              struct the_nilfs *nilfs)
 957 {
 958         struct buffer_head *bh_sr = sci->sc_super_root;
 959         struct nilfs_super_root *raw_sr =
 960                 (struct nilfs_super_root *)bh_sr->b_data;
 961         unsigned isz = nilfs->ns_inode_size;
 962
 963         raw_sr->sr_bytes = cpu_to_le16(NILFS_SR_BYTES);
 964         raw_sr->sr_nongc_ctime
 965                 = cpu_to_le64(nilfs_doing_gc() ?
 966                               nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
 967         raw_sr->sr_flags = 0;
 968
 969         nilfs_mdt_write_inode_direct(
 970                 nilfs_dat_inode(nilfs), bh_sr, NILFS_SR_DAT_OFFSET(isz));
 971         nilfs_mdt_write_inode_direct(
 972                 nilfs->ns_cpfile, bh_sr, NILFS_SR_CPFILE_OFFSET(isz));
 973         nilfs_mdt_write_inode_direct(
 974                 nilfs->ns_sufile, bh_sr, NILFS_SR_SUFILE_OFFSET(isz));
 975 }
 976
 977 static void nilfs_redirty_inodes(struct list_head *head)
 978 {
 979         struct nilfs_inode_info *ii;
 980
 981         list_for_each_entry(ii, head, i_dirty) {
 982                 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
 983                         clear_bit(NILFS_I_COLLECTED, &ii->i_state);
 984         }
 985 }
 986
 987 static void nilfs_drop_collected_inodes(struct list_head *head)
 988 {
 989         struct nilfs_inode_info *ii;
 990
 991         list_for_each_entry(ii, head, i_dirty) {
 992                 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
 993                         continue;
 994
 995                 clear_bit(NILFS_I_INODE_DIRTY, &ii->i_state);
 996                 set_bit(NILFS_I_UPDATED, &ii->i_state);
 997         }
 998 }
 999
1000 static void nilfs_segctor_cancel_free_segments(struct nilfs_sc_info *sci,
1001                                                struct inode *sufile)
1002
1003 {
1004         struct list_head *head = &sci->sc_cleaning_segments;
1005         struct nilfs_segment_entry *ent;
1006         int err;
1007
1008         list_for_each_entry(ent, head, list) {
1009                 if (!(ent->flags & NILFS_SLH_FREED))
1010                         break;
1011                 err = nilfs_sufile_cancel_free(sufile, ent->segnum);
1012                 BUG_ON(err);
1013
1014                 ent->flags &= ~NILFS_SLH_FREED;
1015         }
1016 }
1017
1018 static int nilfs_segctor_prepare_free_segments(struct nilfs_sc_info *sci,
1019                                                struct inode *sufile)
1020 {
1021         struct list_head *head = &sci->sc_cleaning_segments;
1022         struct nilfs_segment_entry *ent;
1023         int err;
1024
1025         list_for_each_entry(ent, head, list) {
1026                 err = nilfs_sufile_free(sufile, ent->segnum);
1027                 if (unlikely(err))
1028                         return err;
1029                 ent->flags |= NILFS_SLH_FREED;
1030         }
1031         return 0;
1032 }
1033
1034 static void nilfs_segctor_commit_free_segments(struct nilfs_sc_info *sci)
1035 {
1036         nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
1037 }
1038
1039 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1040                                        struct inode *inode,
1041                                        struct list_head *listp,
1042                                        int (*collect)(struct nilfs_sc_info *,
1043                                                       struct buffer_head *,
1044                                                       struct inode *))
1045 {
1046         struct buffer_head *bh, *n;
1047         int err = 0;
1048
1049         if (collect) {
1050                 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1051                         list_del_init(&bh->b_assoc_buffers);
1052                         err = collect(sci, bh, inode);
1053                         brelse(bh);
1054                         if (unlikely(err))
1055                                 goto dispose_buffers;
1056                 }
1057                 return 0;
1058         }
1059
1060  dispose_buffers:
1061         while (!list_empty(listp)) {
1062                 bh = list_entry(listp->next, struct buffer_head,
1063                                 b_assoc_buffers);
1064                 list_del_init(&bh->b_assoc_buffers);
1065                 brelse(bh);
1066         }
1067         return err;
1068 }
1069
1070 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1071 {
1072         /* Remaining number of blocks within segment buffer */
1073         return sci->sc_segbuf_nblocks -
1074                 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1075 }
1076
1077 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1078                                    struct inode *inode,
1079                                    struct nilfs_sc_operations *sc_ops)
1080 {
1081         LIST_HEAD(data_buffers);
1082         LIST_HEAD(node_buffers);
1083         int err;
1084
1085         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1086                 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1087
1088                 n = nilfs_lookup_dirty_data_buffers(
1089                         inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1090                 if (n > rest) {
1091                         err = nilfs_segctor_apply_buffers(
1092                                 sci, inode, &data_buffers,
1093                                 sc_ops->collect_data);
1094                         BUG_ON(!err); /* always receive -E2BIG or true error */
1095                         goto break_or_fail;
1096                 }
1097         }
1098         nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1099
1100         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1101                 err = nilfs_segctor_apply_buffers(
1102                         sci, inode, &data_buffers, sc_ops->collect_data);
1103                 if (unlikely(err)) {
1104                         /* dispose node list */
1105                         nilfs_segctor_apply_buffers(
1106                                 sci, inode, &node_buffers, NULL);
1107                         goto break_or_fail;
1108                 }
1109                 sci->sc_stage.flags |= NILFS_CF_NODE;
1110         }
1111         /* Collect node */
1112         err = nilfs_segctor_apply_buffers(
1113                 sci, inode, &node_buffers, sc_ops->collect_node);
1114         if (unlikely(err))
1115                 goto break_or_fail;
1116
1117         nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1118         err = nilfs_segctor_apply_buffers(
1119                 sci, inode, &node_buffers, sc_ops->collect_bmap);
1120         if (unlikely(err))
1121                 goto break_or_fail;
1122
1123         nilfs_segctor_end_finfo(sci, inode);
1124         sci->sc_stage.flags &= ~NILFS_CF_NODE;
1125
1126  break_or_fail:
1127         return err;
1128 }
1129
1130 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1131                                          struct inode *inode)
1132 {
1133         LIST_HEAD(data_buffers);
1134         size_t n, rest = nilfs_segctor_buffer_rest(sci);
1135         int err;
1136
1137         n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1138                                             sci->sc_dsync_start,
1139                                             sci->sc_dsync_end);
1140
1141         err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1142                                           nilfs_collect_file_data);
1143         if (!err) {
1144                 nilfs_segctor_end_finfo(sci, inode);
1145                 BUG_ON(n > rest);
1146                 /* always receive -E2BIG or true error if n > rest */
1147         }
1148         return err;
1149 }
1150
1151 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1152 {
1153         struct nilfs_sb_info *sbi = sci->sc_sbi;
1154         struct the_nilfs *nilfs = sbi->s_nilfs;
1155         struct list_head *head;
1156         struct nilfs_inode_info *ii;
1157         int err = 0;
1158
1159         switch (sci->sc_stage.scnt) {
1160         case NILFS_ST_INIT:
1161                 /* Pre-processes */
1162                 sci->sc_stage.flags = 0;
1163
1164                 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1165                         sci->sc_nblk_inc = 0;
1166                         sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1167                         if (mode == SC_LSEG_DSYNC) {
1168                                 sci->sc_stage.scnt = NILFS_ST_DSYNC;
1169                                 goto dsync_mode;
1170                         }
1171                 }
1172
1173                 sci->sc_stage.dirty_file_ptr = NULL;
1174                 sci->sc_stage.gc_inode_ptr = NULL;
1175                 if (mode == SC_FLUSH_DAT) {
1176                         sci->sc_stage.scnt = NILFS_ST_DAT;
1177                         goto dat_stage;
1178                 }
1179                 sci->sc_stage.scnt++;  /* Fall through */
1180         case NILFS_ST_GC:
1181                 if (nilfs_doing_gc()) {
1182                         head = &sci->sc_gc_inodes;
1183                         ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1184                                                 head, i_dirty);
1185                         list_for_each_entry_continue(ii, head, i_dirty) {
1186                                 err = nilfs_segctor_scan_file(
1187                                         sci, &ii->vfs_inode,
1188                                         &nilfs_sc_file_ops);
1189                                 if (unlikely(err)) {
1190                                         sci->sc_stage.gc_inode_ptr = list_entry(
1191                                                 ii->i_dirty.prev,
1192                                                 struct nilfs_inode_info,
1193                                                 i_dirty);
1194                                         goto break_or_fail;
1195                                 }
1196                                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1197                         }
1198                         sci->sc_stage.gc_inode_ptr = NULL;
1199                 }
1200                 sci->sc_stage.scnt++;  /* Fall through */
1201         case NILFS_ST_FILE:
1202                 head = &sci->sc_dirty_files;
1203                 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1204                                         i_dirty);
1205                 list_for_each_entry_continue(ii, head, i_dirty) {
1206                         clear_bit(NILFS_I_DIRTY, &ii->i_state);
1207
1208                         err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1209                                                       &nilfs_sc_file_ops);
1210                         if (unlikely(err)) {
1211                                 sci->sc_stage.dirty_file_ptr =
1212                                         list_entry(ii->i_dirty.prev,
1213                                                    struct nilfs_inode_info,
1214                                                    i_dirty);
1215                                 goto break_or_fail;
1216                         }
1217                         /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1218                         /* XXX: required ? */
1219                 }
1220                 sci->sc_stage.dirty_file_ptr = NULL;
1221                 if (mode == SC_FLUSH_FILE) {
1222                         sci->sc_stage.scnt = NILFS_ST_DONE;
1223                         return 0;
1224                 }
1225                 sci->sc_stage.scnt++;  /* Fall through */
1226         case NILFS_ST_SKETCH:
1227                 if (mode == SC_LSEG_SR && sci->sc_sketch_inode) {
1228                         ii = NILFS_I(sci->sc_sketch_inode);
1229                         if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
1230                                 sci->sc_sketch_inode->i_ctime.tv_sec
1231                                         = sci->sc_seg_ctime;
1232                                 sci->sc_sketch_inode->i_mtime.tv_sec
1233                                         = sci->sc_seg_ctime;
1234                                 err = nilfs_mark_inode_dirty(
1235                                         sci->sc_sketch_inode);
1236                                 if (unlikely(err))
1237                                         goto break_or_fail;
1238                         }
1239                         err = nilfs_segctor_scan_file(sci,
1240                                                       sci->sc_sketch_inode,
1241                                                       &nilfs_sc_file_ops);
1242                         if (unlikely(err))
1243                                 goto break_or_fail;
1244                 }
1245                 sci->sc_stage.scnt++;
1246                 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1247                 /* Fall through */
1248         case NILFS_ST_IFILE:
1249                 err = nilfs_segctor_scan_file(sci, sbi->s_ifile,
1250                                               &nilfs_sc_file_ops);
1251                 if (unlikely(err))
1252                         break;
1253                 sci->sc_stage.scnt++;
1254                 /* Creating a checkpoint */
1255                 err = nilfs_segctor_create_checkpoint(sci);
1256                 if (unlikely(err))
1257                         break;
1258                 /* Fall through */
1259         case NILFS_ST_CPFILE:
1260                 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1261                                               &nilfs_sc_file_ops);
1262                 if (unlikely(err))
1263                         break;
1264                 sci->sc_stage.scnt++;  /* Fall through */
1265         case NILFS_ST_SUFILE:
1266                 err = nilfs_segctor_prepare_free_segments(sci,
1267                                                           nilfs->ns_sufile);
1268                 if (unlikely(err))
1269                         break;
1270                 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1271                                               &nilfs_sc_file_ops);
1272                 if (unlikely(err))
1273                         break;
1274                 sci->sc_stage.scnt++;  /* Fall through */
1275         case NILFS_ST_DAT:
1276  dat_stage:
1277                 err = nilfs_segctor_scan_file(sci, nilfs_dat_inode(nilfs),
1278                                               &nilfs_sc_dat_ops);
1279                 if (unlikely(err))
1280                         break;
1281                 if (mode == SC_FLUSH_DAT) {
1282                         sci->sc_stage.scnt = NILFS_ST_DONE;
1283                         return 0;
1284                 }
1285                 sci->sc_stage.scnt++;  /* Fall through */
1286         case NILFS_ST_SR:
1287                 if (mode == SC_LSEG_SR) {
1288                         /* Appending a super root */
1289                         err = nilfs_segctor_add_super_root(sci);
1290                         if (unlikely(err))
1291                                 break;
1292                 }
1293                 /* End of a logical segment */
1294                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1295                 sci->sc_stage.scnt = NILFS_ST_DONE;
1296                 return 0;
1297         case NILFS_ST_DSYNC:
1298  dsync_mode:
1299                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1300                 ii = sci->sc_dsync_inode;
1301                 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1302                         break;
1303
1304                 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1305                 if (unlikely(err))
1306                         break;
1307                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1308                 sci->sc_stage.scnt = NILFS_ST_DONE;
1309                 return 0;
1310         case NILFS_ST_DONE:
1311                 return 0;
1312         default:
1313                 BUG();
1314         }
1315
1316  break_or_fail:
1317         return err;
1318 }
1319
1320 static int nilfs_segctor_terminate_segment(struct nilfs_sc_info *sci,
1321                                            struct nilfs_segment_buffer *segbuf,
1322                                            struct inode *sufile)
1323 {
1324         struct nilfs_segment_entry *ent = segbuf->sb_segent;
1325         int err;
1326
1327         err = nilfs_open_segment_entry(ent, sufile);
1328         if (unlikely(err))
1329                 return err;
1330         nilfs_mdt_mark_buffer_dirty(ent->bh_su);
1331         nilfs_mdt_mark_dirty(sufile);
1332         nilfs_close_segment_entry(ent, sufile);
1333
1334         list_add_tail(&ent->list, &sci->sc_active_segments);
1335         segbuf->sb_segent = NULL;
1336         return 0;
1337 }
1338
1339 static int nilfs_touch_segusage(struct inode *sufile, __u64 segnum)
1340 {
1341         struct buffer_head *bh_su;
1342         struct nilfs_segment_usage *raw_su;
1343         int err;
1344
1345         err = nilfs_sufile_get_segment_usage(sufile, segnum, &raw_su, &bh_su);
1346         if (unlikely(err))
1347                 return err;
1348         nilfs_mdt_mark_buffer_dirty(bh_su);
1349         nilfs_mdt_mark_dirty(sufile);
1350         nilfs_sufile_put_segment_usage(sufile, segnum, bh_su);
1351         return 0;
1352 }
1353
1354 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1355                                             struct the_nilfs *nilfs)
1356 {
1357         struct nilfs_segment_buffer *segbuf, *n;
1358         struct inode *sufile = nilfs->ns_sufile;
1359         __u64 nextnum;
1360         int err;
1361
1362         if (list_empty(&sci->sc_segbufs)) {
1363                 segbuf = nilfs_segbuf_new(sci->sc_super);
1364                 if (unlikely(!segbuf))
1365                         return -ENOMEM;
1366                 list_add(&segbuf->sb_list, &sci->sc_segbufs);
1367         } else
1368                 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1369
1370         err = nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1371                                nilfs->ns_pseg_offset, nilfs);
1372         if (unlikely(err))
1373                 return err;
1374
1375         if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1376                 err = nilfs_segctor_terminate_segment(sci, segbuf, sufile);
1377                 if (unlikely(err))
1378                         return err;
1379
1380                 nilfs_shift_to_next_segment(nilfs);
1381                 err = nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1382         }
1383         sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1384
1385         err = nilfs_touch_segusage(sufile, segbuf->sb_segnum);
1386         if (unlikely(err))
1387                 return err;
1388
1389         if (nilfs->ns_segnum == nilfs->ns_nextnum) {
1390                 /* Start from the head of a new full segment */
1391                 err = nilfs_sufile_alloc(sufile, &nextnum);
1392                 if (unlikely(err))
1393                         return err;
1394         } else
1395                 nextnum = nilfs->ns_nextnum;
1396
1397         segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1398         nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1399
1400         /* truncating segment buffers */
1401         list_for_each_entry_safe_continue(segbuf, n, &sci->sc_segbufs,
1402                                           sb_list) {
1403                 list_del_init(&segbuf->sb_list);
1404                 nilfs_segbuf_free(segbuf);
1405         }
1406         return err;
1407 }
1408
1409 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1410                                          struct the_nilfs *nilfs, int nadd)
1411 {
1412         struct nilfs_segment_buffer *segbuf, *prev, *n;
1413         struct inode *sufile = nilfs->ns_sufile;
1414         __u64 nextnextnum;
1415         LIST_HEAD(list);
1416         int err, ret, i;
1417
1418         prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1419         /*
1420          * Since the segment specified with nextnum might be allocated during
1421          * the previous construction, the buffer including its segusage may
1422          * not be dirty.  The following call ensures that the buffer is dirty
1423          * and will pin the buffer on memory until the sufile is written.
1424          */
1425         err = nilfs_touch_segusage(sufile, prev->sb_nextnum);
1426         if (unlikely(err))
1427                 return err;
1428
1429         for (i = 0; i < nadd; i++) {
1430                 /* extend segment info */
1431                 err = -ENOMEM;
1432                 segbuf = nilfs_segbuf_new(sci->sc_super);
1433                 if (unlikely(!segbuf))
1434                         goto failed;
1435
1436                 /* map this buffer to region of segment on-disk */
1437                 err = nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1438                 if (unlikely(err))
1439                         goto failed_segbuf;
1440
1441                 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1442
1443                 /* allocate the next next full segment */
1444                 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1445                 if (unlikely(err))
1446                         goto failed_segbuf;
1447
1448                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1449                 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1450
1451                 list_add_tail(&segbuf->sb_list, &list);
1452                 prev = segbuf;
1453         }
1454         list_splice(&list, sci->sc_segbufs.prev);
1455         return 0;
1456
1457  failed_segbuf:
1458         nilfs_segbuf_free(segbuf);
1459  failed:
1460         list_for_each_entry_safe(segbuf, n, &list, sb_list) {
1461                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1462                 BUG_ON(ret);
1463                 list_del_init(&segbuf->sb_list);
1464                 nilfs_segbuf_free(segbuf);
1465         }
1466         return err;
1467 }
1468
1469 static void nilfs_segctor_free_incomplete_segments(struct nilfs_sc_info *sci,
1470                                                    struct the_nilfs *nilfs)
1471 {
1472         struct nilfs_segment_buffer *segbuf;
1473         int ret, done = 0;
1474
1475         segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1476         if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1477                 ret = nilfs_sufile_free(nilfs->ns_sufile, segbuf->sb_nextnum);
1478                 BUG_ON(ret);
1479         }
1480         if (segbuf->sb_io_error) {
1481                 /* Case 1: The first segment failed */
1482                 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1483                         /* Case 1a:  Partial segment appended into an existing
1484                            segment */
1485                         nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1486                                                 segbuf->sb_fseg_end);
1487                 else /* Case 1b:  New full segment */
1488                         set_nilfs_discontinued(nilfs);
1489                 done++;
1490         }
1491
1492         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1493                 ret = nilfs_sufile_free(nilfs->ns_sufile, segbuf->sb_nextnum);
1494                 BUG_ON(ret);
1495                 if (!done && segbuf->sb_io_error) {
1496                         if (segbuf->sb_segnum != nilfs->ns_nextnum)
1497                                 /* Case 2: extended segment (!= next) failed */
1498                                 nilfs_sufile_set_error(nilfs->ns_sufile,
1499                                                        segbuf->sb_segnum);
1500                         done++;
1501                 }
1502         }
1503 }
1504
1505 static void nilfs_segctor_clear_segment_buffers(struct nilfs_sc_info *sci)
1506 {
1507         struct nilfs_segment_buffer *segbuf;
1508
1509         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list)
1510                 nilfs_segbuf_clear(segbuf);
1511         sci->sc_super_root = NULL;
1512 }
1513
1514 static void nilfs_segctor_destroy_segment_buffers(struct nilfs_sc_info *sci)
1515 {
1516         struct nilfs_segment_buffer *segbuf;
1517
1518         while (!list_empty(&sci->sc_segbufs)) {
1519                 segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1520                 list_del_init(&segbuf->sb_list);
1521                 nilfs_segbuf_free(segbuf);
1522         }
1523         /* sci->sc_curseg = NULL; */
1524 }
1525
1526 static void nilfs_segctor_end_construction(struct nilfs_sc_info *sci,
1527                                            struct the_nilfs *nilfs, int err)
1528 {
1529         if (unlikely(err)) {
1530                 nilfs_segctor_free_incomplete_segments(sci, nilfs);
1531                 nilfs_segctor_cancel_free_segments(sci, nilfs->ns_sufile);
1532         }
1533         nilfs_segctor_clear_segment_buffers(sci);
1534 }
1535
1536 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1537                                           struct inode *sufile)
1538 {
1539         struct nilfs_segment_buffer *segbuf;
1540         struct buffer_head *bh_su;
1541         struct nilfs_segment_usage *raw_su;
1542         unsigned long live_blocks;
1543         int ret;
1544
1545         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1546                 ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1547                                                      &raw_su, &bh_su);
1548                 BUG_ON(ret); /* always succeed because bh_su is dirty */
1549                 live_blocks = segbuf->sb_sum.nblocks +
1550                         (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1551                 raw_su->su_lastmod = cpu_to_le64(sci->sc_seg_ctime);
1552                 raw_su->su_nblocks = cpu_to_le32(live_blocks);
1553                 nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum,
1554                                                bh_su);
1555         }
1556 }
1557
1558 static void nilfs_segctor_cancel_segusage(struct nilfs_sc_info *sci,
1559                                           struct inode *sufile)
1560 {
1561         struct nilfs_segment_buffer *segbuf;
1562         struct buffer_head *bh_su;
1563         struct nilfs_segment_usage *raw_su;
1564         int ret;
1565
1566         segbuf = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1567         ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1568                                              &raw_su, &bh_su);
1569         BUG_ON(ret); /* always succeed because bh_su is dirty */
1570         raw_su->su_nblocks = cpu_to_le32(segbuf->sb_pseg_start -
1571                                          segbuf->sb_fseg_start);
1572         nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum, bh_su);
1573
1574         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1575                 ret = nilfs_sufile_get_segment_usage(sufile, segbuf->sb_segnum,
1576                                                      &raw_su, &bh_su);
1577                 BUG_ON(ret); /* always succeed */
1578                 raw_su->su_nblocks = 0;
1579                 nilfs_sufile_put_segment_usage(sufile, segbuf->sb_segnum,
1580                                                bh_su);
1581         }
1582 }
1583
1584 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1585                                             struct nilfs_segment_buffer *last,
1586                                             struct inode *sufile)
1587 {
1588         struct nilfs_segment_buffer *segbuf = last, *n;
1589         int ret;
1590
1591         list_for_each_entry_safe_continue(segbuf, n, &sci->sc_segbufs,
1592                                           sb_list) {
1593                 list_del_init(&segbuf->sb_list);
1594                 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1595                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1596                 BUG_ON(ret);
1597                 nilfs_segbuf_free(segbuf);
1598         }
1599 }
1600
1601
1602 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1603                                  struct the_nilfs *nilfs, int mode)
1604 {
1605         struct nilfs_cstage prev_stage = sci->sc_stage;
1606         int err, nadd = 1;
1607
1608         /* Collection retry loop */
1609         for (;;) {
1610                 sci->sc_super_root = NULL;
1611                 sci->sc_nblk_this_inc = 0;
1612                 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1613
1614                 err = nilfs_segctor_reset_segment_buffer(sci);
1615                 if (unlikely(err))
1616                         goto failed;
1617
1618                 err = nilfs_segctor_collect_blocks(sci, mode);
1619                 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1620                 if (!err)
1621                         break;
1622
1623                 if (unlikely(err != -E2BIG))
1624                         goto failed;
1625
1626                 /* The current segment is filled up */
1627                 if (mode != SC_LSEG_SR || sci->sc_stage.scnt < NILFS_ST_CPFILE)
1628                         break;
1629
1630                 nilfs_segctor_cancel_free_segments(sci, nilfs->ns_sufile);
1631                 nilfs_segctor_clear_segment_buffers(sci);
1632
1633                 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1634                 if (unlikely(err))
1635                         return err;
1636
1637                 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1638                 sci->sc_stage = prev_stage;
1639         }
1640         nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1641         return 0;
1642
1643  failed:
1644         return err;
1645 }
1646
1647 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1648                                       struct buffer_head *new_bh)
1649 {
1650         BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1651
1652         list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1653         /* The caller must release old_bh */
1654 }
1655
1656 static int
1657 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1658                                      struct nilfs_segment_buffer *segbuf,
1659                                      int mode)
1660 {
1661         struct inode *inode = NULL;
1662         sector_t blocknr;
1663         unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1664         unsigned long nblocks = 0, ndatablk = 0;
1665         struct nilfs_sc_operations *sc_op = NULL;
1666         struct nilfs_segsum_pointer ssp;
1667         struct nilfs_finfo *finfo = NULL;
1668         union nilfs_binfo binfo;
1669         struct buffer_head *bh, *bh_org;
1670         ino_t ino = 0;
1671         int err = 0;
1672
1673         if (!nfinfo)
1674                 goto out;
1675
1676         blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1677         ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1678         ssp.offset = sizeof(struct nilfs_segment_summary);
1679
1680         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1681                 if (bh == sci->sc_super_root)
1682                         break;
1683                 if (!finfo) {
1684                         finfo = nilfs_segctor_map_segsum_entry(
1685                                 sci, &ssp, sizeof(*finfo));
1686                         ino = le64_to_cpu(finfo->fi_ino);
1687                         nblocks = le32_to_cpu(finfo->fi_nblocks);
1688                         ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1689
1690                         if (buffer_nilfs_node(bh))
1691                                 inode = NILFS_BTNC_I(bh->b_page->mapping);
1692                         else
1693                                 inode = NILFS_AS_I(bh->b_page->mapping);
1694
1695                         if (mode == SC_LSEG_DSYNC)
1696                                 sc_op = &nilfs_sc_dsync_ops;
1697                         else if (ino == NILFS_DAT_INO)
1698                                 sc_op = &nilfs_sc_dat_ops;
1699                         else /* file blocks */
1700                                 sc_op = &nilfs_sc_file_ops;
1701                 }
1702                 bh_org = bh;
1703                 get_bh(bh_org);
1704                 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1705                                         &binfo);
1706                 if (bh != bh_org)
1707                         nilfs_list_replace_buffer(bh_org, bh);
1708                 brelse(bh_org);
1709                 if (unlikely(err))
1710                         goto failed_bmap;
1711
1712                 if (ndatablk > 0)
1713                         sc_op->write_data_binfo(sci, &ssp, &binfo);
1714                 else
1715                         sc_op->write_node_binfo(sci, &ssp, &binfo);
1716
1717                 blocknr++;
1718                 if (--nblocks == 0) {
1719                         finfo = NULL;
1720                         if (--nfinfo == 0)
1721                                 break;
1722                 } else if (ndatablk > 0)
1723                         ndatablk--;
1724         }
1725  out:
1726         return 0;
1727
1728  failed_bmap:
1729         err = nilfs_handle_bmap_error(err, __func__, inode, sci->sc_super);
1730         return err;
1731 }
1732
1733 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1734 {
1735         struct nilfs_segment_buffer *segbuf;
1736         int err;
1737
1738         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1739                 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1740                 if (unlikely(err))
1741                         return err;
1742                 nilfs_segbuf_fill_in_segsum(segbuf);
1743         }
1744         return 0;
1745 }
1746
1747 static int
1748 nilfs_copy_replace_page_buffers(struct page *page, struct list_head *out)
1749 {
1750         struct page *clone_page;
1751         struct buffer_head *bh, *head, *bh2;
1752         void *kaddr;
1753
1754         bh = head = page_buffers(page);
1755
1756         clone_page = nilfs_alloc_private_page(bh->b_bdev, bh->b_size, 0);
1757         if (unlikely(!clone_page))
1758                 return -ENOMEM;
1759
1760         bh2 = page_buffers(clone_page);
1761         kaddr = kmap_atomic(page, KM_USER0);
1762         do {
1763                 if (list_empty(&bh->b_assoc_buffers))
1764                         continue;
1765                 get_bh(bh2);
1766                 page_cache_get(clone_page); /* for each bh */
1767                 memcpy(bh2->b_data, kaddr + bh_offset(bh), bh2->b_size);
1768                 bh2->b_blocknr = bh->b_blocknr;
1769                 list_replace(&bh->b_assoc_buffers, &bh2->b_assoc_buffers);
1770                 list_add_tail(&bh->b_assoc_buffers, out);
1771         } while (bh = bh->b_this_page, bh2 = bh2->b_this_page, bh != head);
1772         kunmap_atomic(kaddr, KM_USER0);
1773
1774         if (!TestSetPageWriteback(clone_page))
1775                 inc_zone_page_state(clone_page, NR_WRITEBACK);
1776         unlock_page(clone_page);
1777
1778         return 0;
1779 }
1780
1781 static int nilfs_test_page_to_be_frozen(struct page *page)
1782 {
1783         struct address_space *mapping = page->mapping;
1784
1785         if (!mapping || !mapping->host || S_ISDIR(mapping->host->i_mode))
1786                 return 0;
1787
1788         if (page_mapped(page)) {
1789                 ClearPageChecked(page);
1790                 return 1;
1791         }
1792         return PageChecked(page);
1793 }
1794
1795 static int nilfs_begin_page_io(struct page *page, struct list_head *out)
1796 {
1797         if (!page || PageWriteback(page))
1798                 /* For split b-tree node pages, this function may be called
1799                    twice.  We ignore the 2nd or later calls by this check. */
1800                 return 0;
1801
1802         lock_page(page);
1803         clear_page_dirty_for_io(page);
1804         set_page_writeback(page);
1805         unlock_page(page);
1806
1807         if (nilfs_test_page_to_be_frozen(page)) {
1808                 int err = nilfs_copy_replace_page_buffers(page, out);
1809                 if (unlikely(err))
1810                         return err;
1811         }
1812         return 0;
1813 }
1814
1815 static int nilfs_segctor_prepare_write(struct nilfs_sc_info *sci,
1816                                        struct page **failed_page)
1817 {
1818         struct nilfs_segment_buffer *segbuf;
1819         struct page *bd_page = NULL, *fs_page = NULL;
1820         struct list_head *list = &sci->sc_copied_buffers;
1821         int err;
1822
1823         *failed_page = NULL;
1824         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1825                 struct buffer_head *bh;
1826
1827                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1828                                     b_assoc_buffers) {
1829                         if (bh->b_page != bd_page) {
1830                                 if (bd_page) {
1831                                         lock_page(bd_page);
1832                                         clear_page_dirty_for_io(bd_page);
1833                                         set_page_writeback(bd_page);
1834                                         unlock_page(bd_page);
1835                                 }
1836                                 bd_page = bh->b_page;
1837                         }
1838                 }
1839
1840                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1841                                     b_assoc_buffers) {
1842                         if (bh == sci->sc_super_root) {
1843                                 if (bh->b_page != bd_page) {
1844                                         lock_page(bd_page);
1845                                         clear_page_dirty_for_io(bd_page);
1846                                         set_page_writeback(bd_page);
1847                                         unlock_page(bd_page);
1848                                         bd_page = bh->b_page;
1849                                 }
1850                                 break;
1851                         }
1852                         if (bh->b_page != fs_page) {
1853                                 err = nilfs_begin_page_io(fs_page, list);
1854                                 if (unlikely(err)) {
1855                                         *failed_page = fs_page;
1856                                         goto out;
1857                                 }
1858                                 fs_page = bh->b_page;
1859                         }
1860                 }
1861         }
1862         if (bd_page) {
1863                 lock_page(bd_page);
1864                 clear_page_dirty_for_io(bd_page);
1865                 set_page_writeback(bd_page);
1866                 unlock_page(bd_page);
1867         }
1868         err = nilfs_begin_page_io(fs_page, list);
1869         if (unlikely(err))
1870                 *failed_page = fs_page;
1871  out:
1872         return err;
1873 }
1874
1875 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1876                                struct backing_dev_info *bdi)
1877 {
1878         struct nilfs_segment_buffer *segbuf;
1879         struct nilfs_write_info wi;
1880         int err, res;
1881
1882         wi.sb = sci->sc_super;
1883         wi.bh_sr = sci->sc_super_root;
1884         wi.bdi = bdi;
1885
1886         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1887                 nilfs_segbuf_prepare_write(segbuf, &wi);
1888                 err = nilfs_segbuf_write(segbuf, &wi);
1889
1890                 res = nilfs_segbuf_wait(segbuf, &wi);
1891                 err = unlikely(err) ? : res;
1892                 if (unlikely(err))
1893                         return err;
1894         }
1895         return 0;
1896 }
1897
1898 static int nilfs_page_has_uncleared_buffer(struct page *page)
1899 {
1900         struct buffer_head *head, *bh;
1901
1902         head = bh = page_buffers(page);
1903         do {
1904                 if (buffer_dirty(bh) && !list_empty(&bh->b_assoc_buffers))
1905                         return 1;
1906                 bh = bh->b_this_page;
1907         } while (bh != head);
1908         return 0;
1909 }
1910
1911 static void __nilfs_end_page_io(struct page *page, int err)
1912 {
1913         /* BUG_ON(err > 0); */
1914         if (!err) {
1915                 if (!nilfs_page_buffers_clean(page))
1916                         __set_page_dirty_nobuffers(page);
1917                 ClearPageError(page);
1918         } else {
1919                 __set_page_dirty_nobuffers(page);
1920                 SetPageError(page);
1921         }
1922
1923         if (buffer_nilfs_allocated(page_buffers(page))) {
1924                 if (TestClearPageWriteback(page))
1925                         dec_zone_page_state(page, NR_WRITEBACK);
1926         } else
1927                 end_page_writeback(page);
1928 }
1929
1930 static void nilfs_end_page_io(struct page *page, int err)
1931 {
1932         if (!page)
1933                 return;
1934
1935         if (buffer_nilfs_node(page_buffers(page)) &&
1936             nilfs_page_has_uncleared_buffer(page))
1937                 /* For b-tree node pages, this function may be called twice
1938                    or more because they might be split in a segment.
1939                    This check assures that cleanup has been done for all
1940                    buffers in a split btnode page. */
1941                 return;
1942
1943         __nilfs_end_page_io(page, err);
1944 }
1945
1946 static void nilfs_clear_copied_buffers(struct list_head *list, int err)
1947 {
1948         struct buffer_head *bh, *head;
1949         struct page *page;
1950
1951         while (!list_empty(list)) {
1952                 bh = list_entry(list->next, struct buffer_head,
1953                                 b_assoc_buffers);
1954                 page = bh->b_page;
1955                 page_cache_get(page);
1956                 head = bh = page_buffers(page);
1957                 do {
1958                         if (!list_empty(&bh->b_assoc_buffers)) {
1959                                 list_del_init(&bh->b_assoc_buffers);
1960                                 if (!err) {
1961                                         set_buffer_uptodate(bh);
1962                                         clear_buffer_dirty(bh);
1963                                         clear_buffer_nilfs_volatile(bh);
1964                                 }
1965                                 brelse(bh); /* for b_assoc_buffers */
1966                         }
1967                 } while ((bh = bh->b_this_page) != head);
1968
1969                 __nilfs_end_page_io(page, err);
1970                 page_cache_release(page);
1971         }
1972 }
1973
1974 static void nilfs_segctor_abort_write(struct nilfs_sc_info *sci,
1975                                       struct page *failed_page, int err)
1976 {
1977         struct nilfs_segment_buffer *segbuf;
1978         struct page *bd_page = NULL, *fs_page = NULL;
1979
1980         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1981                 struct buffer_head *bh;
1982
1983                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1984                                     b_assoc_buffers) {
1985                         if (bh->b_page != bd_page) {
1986                                 if (bd_page)
1987                                         end_page_writeback(bd_page);
1988                                 bd_page = bh->b_page;
1989                         }
1990                 }
1991
1992                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1993                                     b_assoc_buffers) {
1994                         if (bh == sci->sc_super_root) {
1995                                 if (bh->b_page != bd_page) {
1996                                         end_page_writeback(bd_page);
1997                                         bd_page = bh->b_page;
1998                                 }
1999                                 break;
2000                         }
2001                         if (bh->b_page != fs_page) {
2002                                 nilfs_end_page_io(fs_page, err);
2003                                 if (unlikely(fs_page == failed_page))
2004                                         goto done;
2005                                 fs_page = bh->b_page;
2006                         }
2007                 }
2008         }
2009         if (bd_page)
2010                 end_page_writeback(bd_page);
2011
2012         nilfs_end_page_io(fs_page, err);
2013  done:
2014         nilfs_clear_copied_buffers(&sci->sc_copied_buffers, err);
2015 }
2016
2017 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
2018                                    struct nilfs_segment_buffer *segbuf)
2019 {
2020         nilfs->ns_segnum = segbuf->sb_segnum;
2021         nilfs->ns_nextnum = segbuf->sb_nextnum;
2022         nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
2023                 + segbuf->sb_sum.nblocks;
2024         nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
2025         nilfs->ns_ctime = segbuf->sb_sum.ctime;
2026 }
2027
2028 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
2029 {
2030         struct nilfs_segment_buffer *segbuf;
2031         struct page *bd_page = NULL, *fs_page = NULL;
2032         struct nilfs_sb_info *sbi = sci->sc_sbi;
2033         struct the_nilfs *nilfs = sbi->s_nilfs;
2034         int update_sr = (sci->sc_super_root != NULL);
2035
2036         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
2037                 struct buffer_head *bh;
2038
2039                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
2040                                     b_assoc_buffers) {
2041                         set_buffer_uptodate(bh);
2042                         clear_buffer_dirty(bh);
2043                         if (bh->b_page != bd_page) {
2044                                 if (bd_page)
2045                                         end_page_writeback(bd_page);
2046                                 bd_page = bh->b_page;
2047                         }
2048                 }
2049                 /*
2050                  * We assume that the buffers which belong to the same page
2051                  * continue over the buffer list.
2052                  * Under this assumption, the last BHs of pages is
2053                  * identifiable by the discontinuity of bh->b_page
2054                  * (page != fs_page).
2055                  *
2056                  * For B-tree node blocks, however, this assumption is not
2057                  * guaranteed.  The cleanup code of B-tree node pages needs
2058                  * special care.
2059                  */
2060                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
2061                                     b_assoc_buffers) {
2062                         set_buffer_uptodate(bh);
2063                         clear_buffer_dirty(bh);
2064                         clear_buffer_nilfs_volatile(bh);
2065                         if (bh == sci->sc_super_root) {
2066                                 if (bh->b_page != bd_page) {
2067                                         end_page_writeback(bd_page);
2068                                         bd_page = bh->b_page;
2069                                 }
2070                                 break;
2071                         }
2072                         if (bh->b_page != fs_page) {
2073                                 nilfs_end_page_io(fs_page, 0);
2074                                 fs_page = bh->b_page;
2075                         }
2076                 }
2077
2078                 if (!NILFS_SEG_SIMPLEX(&segbuf->sb_sum)) {
2079                         if (NILFS_SEG_LOGBGN(&segbuf->sb_sum)) {
2080                                 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
2081                                 sci->sc_lseg_stime = jiffies;
2082                         }
2083                         if (NILFS_SEG_LOGEND(&segbuf->sb_sum))
2084                                 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
2085                 }
2086         }
2087         /*
2088          * Since pages may continue over multiple segment buffers,
2089          * end of the last page must be checked outside of the loop.
2090          */
2091         if (bd_page)
2092                 end_page_writeback(bd_page);
2093
2094         nilfs_end_page_io(fs_page, 0);
2095
2096         nilfs_clear_copied_buffers(&sci->sc_copied_buffers, 0);
2097
2098         nilfs_drop_collected_inodes(&sci->sc_dirty_files);
2099
2100         if (nilfs_doing_gc()) {
2101                 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
2102                 if (update_sr)
2103                         nilfs_commit_gcdat_inode(nilfs);
2104         } else {
2105                 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
2106                 set_nilfs_cond_nongc_write(nilfs);
2107                 wake_up(&nilfs->ns_cleanerd_wq);
2108         }
2109
2110         sci->sc_nblk_inc += sci->sc_nblk_this_inc;
2111
2112         segbuf = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2113         nilfs_set_next_segment(nilfs, segbuf);
2114
2115         if (update_sr) {
2116                 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
2117                                        segbuf->sb_sum.seg_seq, nilfs->ns_cno);
2118
2119                 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2120                 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
2121         } else
2122                 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
2123 }
2124
2125 static int nilfs_segctor_check_in_files(struct nilfs_sc_info *sci,
2126                                         struct nilfs_sb_info *sbi)
2127 {
2128         struct nilfs_inode_info *ii, *n;
2129         __u64 cno = sbi->s_nilfs->ns_cno;
2130
2131         spin_lock(&sbi->s_inode_lock);
2132  retry:
2133         list_for_each_entry_safe(ii, n, &sbi->s_dirty_files, i_dirty) {
2134                 if (!ii->i_bh) {
2135                         struct buffer_head *ibh;
2136                         int err;
2137
2138                         spin_unlock(&sbi->s_inode_lock);
2139                         err = nilfs_ifile_get_inode_block(
2140                                 sbi->s_ifile, ii->vfs_inode.i_ino, &ibh);
2141                         if (unlikely(err)) {
2142                                 nilfs_warning(sbi->s_super, __func__,
2143                                               "failed to get inode block.\n");
2144                                 return err;
2145                         }
2146                         nilfs_mdt_mark_buffer_dirty(ibh);
2147                         nilfs_mdt_mark_dirty(sbi->s_ifile);
2148                         spin_lock(&sbi->s_inode_lock);
2149                         if (likely(!ii->i_bh))
2150                                 ii->i_bh = ibh;
2151                         else
2152                                 brelse(ibh);
2153                         goto retry;
2154                 }
2155                 ii->i_cno = cno;
2156
2157                 clear_bit(NILFS_I_QUEUED, &ii->i_state);
2158                 set_bit(NILFS_I_BUSY, &ii->i_state);
2159                 list_del(&ii->i_dirty);
2160                 list_add_tail(&ii->i_dirty, &sci->sc_dirty_files);
2161         }
2162         spin_unlock(&sbi->s_inode_lock);
2163
2164         NILFS_I(sbi->s_ifile)->i_cno = cno;
2165
2166         return 0;
2167 }
2168
2169 static void nilfs_segctor_check_out_files(struct nilfs_sc_info *sci,
2170                                           struct nilfs_sb_info *sbi)
2171 {
2172         struct nilfs_transaction_info *ti = current->journal_info;
2173         struct nilfs_inode_info *ii, *n;
2174         __u64 cno = sbi->s_nilfs->ns_cno;
2175
2176         spin_lock(&sbi->s_inode_lock);
2177         list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
2178                 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
2179                     test_bit(NILFS_I_DIRTY, &ii->i_state)) {
2180                         /* The current checkpoint number (=nilfs->ns_cno) is
2181                            changed between check-in and check-out only if the
2182                            super root is written out.  So, we can update i_cno
2183                            for the inodes that remain in the dirty list. */
2184                         ii->i_cno = cno;
2185                         continue;
2186                 }
2187                 clear_bit(NILFS_I_BUSY, &ii->i_state);
2188                 brelse(ii->i_bh);
2189                 ii->i_bh = NULL;
2190                 list_del(&ii->i_dirty);
2191                 list_add_tail(&ii->i_dirty, &ti->ti_garbage);
2192         }
2193         spin_unlock(&sbi->s_inode_lock);
2194 }
2195
2196 /*
2197  * Nasty routines to manipulate active flags on sufile.
2198  * These would be removed in a future release.
2199  */
2200 static void nilfs_segctor_reactivate_segments(struct nilfs_sc_info *sci,
2201                                               struct the_nilfs *nilfs)
2202 {
2203         struct nilfs_segment_buffer *segbuf, *last;
2204         struct nilfs_segment_entry *ent, *n;
2205         struct inode *sufile = nilfs->ns_sufile;
2206         struct list_head *head;
2207
2208         last = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2209         nilfs_for_each_segbuf_before(segbuf, last, &sci->sc_segbufs) {
2210                 ent = segbuf->sb_segent;
2211                 if (!ent)
2212                         break; /* ignore unmapped segments (should check it?)*/
2213                 nilfs_segment_usage_set_active(ent->raw_su);
2214                 nilfs_close_segment_entry(ent, sufile);
2215         }
2216
2217         head = &sci->sc_active_segments;
2218         list_for_each_entry_safe(ent, n, head, list) {
2219                 nilfs_segment_usage_set_active(ent->raw_su);
2220                 nilfs_close_segment_entry(ent, sufile);
2221         }
2222
2223         down_write(&nilfs->ns_sem);
2224         head = &nilfs->ns_used_segments;
2225         list_for_each_entry(ent, head, list) {
2226                 nilfs_segment_usage_set_volatile_active(ent->raw_su);
2227         }
2228         up_write(&nilfs->ns_sem);
2229 }
2230
2231 static int nilfs_segctor_deactivate_segments(struct nilfs_sc_info *sci,
2232                                              struct the_nilfs *nilfs)
2233 {
2234         struct nilfs_segment_buffer *segbuf, *last;
2235         struct nilfs_segment_entry *ent;
2236         struct inode *sufile = nilfs->ns_sufile;
2237         struct list_head *head;
2238         int err;
2239
2240         last = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2241         nilfs_for_each_segbuf_before(segbuf, last, &sci->sc_segbufs) {
2242                 /*
2243                  * Deactivate ongoing full segments.  The last segment is kept
2244                  * active because it is a start point of recovery, and is not
2245                  * relocatable until the super block points to a newer
2246                  * checkpoint.
2247                  */
2248                 ent = segbuf->sb_segent;
2249                 if (!ent)
2250                         break; /* ignore unmapped segments (should check it?)*/
2251                 err = nilfs_open_segment_entry(ent, sufile);
2252                 if (unlikely(err))
2253                         goto failed;
2254                 nilfs_segment_usage_clear_active(ent->raw_su);
2255                 BUG_ON(!buffer_dirty(ent->bh_su));
2256         }
2257
2258         head = &sci->sc_active_segments;
2259         list_for_each_entry(ent, head, list) {
2260                 err = nilfs_open_segment_entry(ent, sufile);
2261                 if (unlikely(err))
2262                         goto failed;
2263                 nilfs_segment_usage_clear_active(ent->raw_su);
2264                 BUG_ON(!buffer_dirty(ent->bh_su));
2265         }
2266
2267         down_write(&nilfs->ns_sem);
2268         head = &nilfs->ns_used_segments;
2269         list_for_each_entry(ent, head, list) {
2270                 /* clear volatile active for segments of older generations */
2271                 nilfs_segment_usage_clear_volatile_active(ent->raw_su);
2272         }
2273         up_write(&nilfs->ns_sem);
2274         return 0;
2275
2276  failed:
2277         nilfs_segctor_reactivate_segments(sci, nilfs);
2278         return err;
2279 }
2280
2281 static void nilfs_segctor_bead_completed_segments(struct nilfs_sc_info *sci)
2282 {
2283         struct nilfs_segment_buffer *segbuf, *last;
2284         struct nilfs_segment_entry *ent;
2285
2286         /* move each segbuf->sb_segent to the list of used active segments */
2287         last = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
2288         nilfs_for_each_segbuf_before(segbuf, last, &sci->sc_segbufs) {
2289                 ent = segbuf->sb_segent;
2290                 if (!ent)
2291                         break; /* ignore unmapped segments (should check it?)*/
2292                 list_add_tail(&ent->list, &sci->sc_active_segments);
2293                 segbuf->sb_segent = NULL;
2294         }
2295 }
2296
2297 static void
2298 __nilfs_segctor_commit_deactivate_segments(struct nilfs_sc_info *sci,
2299                                            struct the_nilfs *nilfs)
2300
2301 {
2302         struct nilfs_segment_entry *ent;
2303
2304         list_splice_init(&sci->sc_active_segments,
2305                          nilfs->ns_used_segments.prev);
2306
2307         list_for_each_entry(ent, &nilfs->ns_used_segments, list) {
2308                 nilfs_segment_usage_set_volatile_active(ent->raw_su);
2309                 /* These segments are kept open */
2310         }
2311 }
2312
2313 /*
2314  * Main procedure of segment constructor
2315  */
2316 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2317 {
2318         struct nilfs_sb_info *sbi = sci->sc_sbi;
2319         struct the_nilfs *nilfs = sbi->s_nilfs;
2320         struct page *failed_page;
2321         int err, has_sr = 0;
2322
2323         sci->sc_stage.scnt = NILFS_ST_INIT;
2324
2325         err = nilfs_segctor_check_in_files(sci, sbi);
2326         if (unlikely(err))
2327                 goto out;
2328
2329         if (nilfs_test_metadata_dirty(sbi))
2330                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2331
2332         if (nilfs_segctor_clean(sci))
2333                 goto out;
2334
2335         do {
2336                 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2337
2338                 err = nilfs_segctor_begin_construction(sci, nilfs);
2339                 if (unlikely(err))
2340                         goto out;
2341
2342                 /* Update time stamp */
2343                 sci->sc_seg_ctime = get_seconds();
2344
2345                 err = nilfs_segctor_collect(sci, nilfs, mode);
2346                 if (unlikely(err))
2347                         goto failed;
2348
2349                 has_sr = (sci->sc_super_root != NULL);
2350
2351                 /* Avoid empty segment */
2352                 if (sci->sc_stage.scnt == NILFS_ST_DONE &&
2353                     NILFS_SEG_EMPTY(&sci->sc_curseg->sb_sum)) {
2354                         BUG_ON(mode == SC_LSEG_SR);
2355                         nilfs_segctor_end_construction(sci, nilfs, 1);
2356                         goto out;
2357                 }
2358
2359                 err = nilfs_segctor_assign(sci, mode);
2360                 if (unlikely(err))
2361                         goto failed;
2362
2363                 if (has_sr) {
2364                         err = nilfs_segctor_deactivate_segments(sci, nilfs);
2365                         if (unlikely(err))
2366                                 goto failed;
2367                 }
2368                 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2369                         nilfs_segctor_fill_in_file_bmap(sci, sbi->s_ifile);
2370
2371                 if (has_sr) {
2372                         err = nilfs_segctor_fill_in_checkpoint(sci);
2373                         if (unlikely(err))
2374                                 goto failed_to_make_up;
2375
2376                         nilfs_segctor_fill_in_super_root(sci, nilfs);
2377                 }
2378                 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2379
2380                 /* Write partial segments */
2381                 err = nilfs_segctor_prepare_write(sci, &failed_page);
2382                 if (unlikely(err))
2383                         goto failed_to_write;
2384
2385                 nilfs_segctor_fill_in_checksums(sci, nilfs->ns_crc_seed);
2386
2387                 err = nilfs_segctor_write(sci, nilfs->ns_bdi);
2388                 if (unlikely(err))
2389                         goto failed_to_write;
2390
2391                 nilfs_segctor_complete_write(sci);
2392
2393                 /* Commit segments */
2394                 nilfs_segctor_bead_completed_segments(sci);
2395                 if (has_sr) {
2396                         down_write(&nilfs->ns_sem);
2397                         nilfs_update_last_segment(sbi, 1);
2398                         __nilfs_segctor_commit_deactivate_segments(sci, nilfs);
2399                         up_write(&nilfs->ns_sem);
2400                         nilfs_segctor_commit_free_segments(sci);
2401                         nilfs_segctor_clear_metadata_dirty(sci);
2402                 }
2403
2404                 nilfs_segctor_end_construction(sci, nilfs, 0);
2405
2406         } while (sci->sc_stage.scnt != NILFS_ST_DONE);
2407
2408         /* Clearing sketch data */
2409         if (has_sr && sci->sc_sketch_inode) {
2410                 if (i_size_read(sci->sc_sketch_inode) == 0)
2411                         clear_bit(NILFS_I_DIRTY,
2412                                   &NILFS_I(sci->sc_sketch_inode)->i_state);
2413                 i_size_write(sci->sc_sketch_inode, 0);
2414         }
2415  out:
2416         nilfs_segctor_destroy_segment_buffers(sci);
2417         nilfs_segctor_check_out_files(sci, sbi);
2418         return err;
2419
2420  failed_to_write:
2421         nilfs_segctor_abort_write(sci, failed_page, err);
2422         nilfs_segctor_cancel_segusage(sci, nilfs->ns_sufile);
2423
2424  failed_to_make_up:
2425         if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2426                 nilfs_redirty_inodes(&sci->sc_dirty_files);
2427         if (has_sr)
2428                 nilfs_segctor_reactivate_segments(sci, nilfs);
2429
2430  failed:
2431         if (nilfs_doing_gc())
2432                 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2433         nilfs_segctor_end_construction(sci, nilfs, err);
2434         goto out;
2435 }
2436
2437 /**
2438  * nilfs_secgtor_start_timer - set timer of background write
2439  * @sci: nilfs_sc_info
2440  *
2441  * If the timer has already been set, it ignores the new request.
2442  * This function MUST be called within a section locking the segment
2443  * semaphore.
2444  */
2445 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2446 {
2447         spin_lock(&sci->sc_state_lock);
2448         if (sci->sc_timer && !(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2449                 sci->sc_timer->expires = jiffies + sci->sc_interval;
2450                 add_timer(sci->sc_timer);
2451                 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2452         }
2453         spin_unlock(&sci->sc_state_lock);
2454 }
2455
2456 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2457 {
2458         spin_lock(&sci->sc_state_lock);
2459         if (!(sci->sc_flush_request & (1 << bn))) {
2460                 unsigned long prev_req = sci->sc_flush_request;
2461
2462                 sci->sc_flush_request |= (1 << bn);
2463                 if (!prev_req)
2464                         wake_up(&sci->sc_wait_daemon);
2465         }
2466         spin_unlock(&sci->sc_state_lock);
2467 }
2468
2469 /**
2470  * nilfs_flush_segment - trigger a segment construction for resource control
2471  * @sb: super block
2472  * @ino: inode number of the file to be flushed out.
2473  */
2474 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2475 {
2476         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2477         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2478
2479         if (!sci || nilfs_doing_construction())
2480                 return;
2481         nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2482                                         /* assign bit 0 to data files */
2483 }
2484
2485 int nilfs_segctor_add_segments_to_be_freed(struct nilfs_sc_info *sci,
2486                                            __u64 *segnum, size_t nsegs)
2487 {
2488         struct nilfs_segment_entry *ent;
2489         struct the_nilfs *nilfs = sci->sc_sbi->s_nilfs;
2490         struct inode *sufile = nilfs->ns_sufile;
2491         LIST_HEAD(list);
2492         __u64 *pnum;
2493         const char *flag_name;
2494         size_t i;
2495         int err, err2 = 0;
2496
2497         for (pnum = segnum, i = 0; i < nsegs; pnum++, i++) {
2498                 ent = nilfs_alloc_segment_entry(*pnum);
2499                 if (unlikely(!ent)) {
2500                         err = -ENOMEM;
2501                         goto failed;
2502                 }
2503                 list_add_tail(&ent->list, &list);
2504
2505                 err = nilfs_open_segment_entry(ent, sufile);
2506                 if (unlikely(err))
2507                         goto failed;
2508
2509                 if (unlikely(le32_to_cpu(ent->raw_su->su_flags) !=
2510                              (1UL << NILFS_SEGMENT_USAGE_DIRTY))) {
2511                         if (nilfs_segment_usage_clean(ent->raw_su))
2512                                 flag_name = "clean";
2513                         else if (nilfs_segment_usage_active(ent->raw_su))
2514                                 flag_name = "active";
2515                         else if (nilfs_segment_usage_volatile_active(
2516                                          ent->raw_su))
2517                                 flag_name = "volatile active";
2518                         else if (!nilfs_segment_usage_dirty(ent->raw_su))
2519                                 flag_name = "non-dirty";
2520                         else
2521                                 flag_name = "erroneous";
2522
2523                         printk(KERN_ERR
2524                                "NILFS: %s segment is requested to be cleaned "
2525                                "(segnum=%llu)\n",
2526                                flag_name, (unsigned long long)ent->segnum);
2527                         err2 = -EINVAL;
2528                 }
2529                 nilfs_close_segment_entry(ent, sufile);
2530         }
2531         if (unlikely(err2)) {
2532                 err = err2;
2533                 goto failed;
2534         }
2535         list_splice(&list, sci->sc_cleaning_segments.prev);
2536         return 0;
2537
2538  failed:
2539         nilfs_dispose_segment_list(&list);
2540         return err;
2541 }
2542
2543 void nilfs_segctor_clear_segments_to_be_freed(struct nilfs_sc_info *sci)
2544 {
2545         nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
2546 }
2547
2548 struct nilfs_segctor_wait_request {
2549         wait_queue_t    wq;
2550         __u32           seq;
2551         int             err;
2552         atomic_t        done;
2553 };
2554
2555 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2556 {
2557         struct nilfs_segctor_wait_request wait_req;
2558         int err = 0;
2559
2560         spin_lock(&sci->sc_state_lock);
2561         init_wait(&wait_req.wq);
2562         wait_req.err = 0;
2563         atomic_set(&wait_req.done, 0);
2564         wait_req.seq = ++sci->sc_seq_request;
2565         spin_unlock(&sci->sc_state_lock);
2566
2567         init_waitqueue_entry(&wait_req.wq, current);
2568         add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2569         set_current_state(TASK_INTERRUPTIBLE);
2570         wake_up(&sci->sc_wait_daemon);
2571
2572         for (;;) {
2573                 if (atomic_read(&wait_req.done)) {
2574                         err = wait_req.err;
2575                         break;
2576                 }
2577                 if (!signal_pending(current)) {
2578                         schedule();
2579                         continue;
2580                 }
2581                 err = -ERESTARTSYS;
2582                 break;
2583         }
2584         finish_wait(&sci->sc_wait_request, &wait_req.wq);
2585         return err;
2586 }
2587
2588 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2589 {
2590         struct nilfs_segctor_wait_request *wrq, *n;
2591         unsigned long flags;
2592
2593         spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2594         list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2595                                  wq.task_list) {
2596                 if (!atomic_read(&wrq->done) &&
2597                     nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2598                         wrq->err = err;
2599                         atomic_set(&wrq->done, 1);
2600                 }
2601                 if (atomic_read(&wrq->done)) {
2602                         wrq->wq.func(&wrq->wq,
2603                                      TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2604                                      0, NULL);
2605                 }
2606         }
2607         spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2608 }
2609
2610 /**
2611  * nilfs_construct_segment - construct a logical segment
2612  * @sb: super block
2613  *
2614  * Return Value: On success, 0 is retured. On errors, one of the following
2615  * negative error code is returned.
2616  *
2617  * %-EROFS - Read only filesystem.
2618  *
2619  * %-EIO - I/O error
2620  *
2621  * %-ENOSPC - No space left on device (only in a panic state).
2622  *
2623  * %-ERESTARTSYS - Interrupted.
2624  *
2625  * %-ENOMEM - Insufficient memory available.
2626  */
2627 int nilfs_construct_segment(struct super_block *sb)
2628 {
2629         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2630         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2631         struct nilfs_transaction_info *ti;
2632         int err;
2633
2634         if (!sci)
2635                 return -EROFS;
2636
2637         /* A call inside transactions causes a deadlock. */
2638         BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2639
2640         err = nilfs_segctor_sync(sci);
2641         return err;
2642 }
2643
2644 /**
2645  * nilfs_construct_dsync_segment - construct a data-only logical segment
2646  * @sb: super block
2647  * @inode: inode whose data blocks should be written out
2648  * @start: start byte offset
2649  * @end: end byte offset (inclusive)
2650  *
2651  * Return Value: On success, 0 is retured. On errors, one of the following
2652  * negative error code is returned.
2653  *
2654  * %-EROFS - Read only filesystem.
2655  *
2656  * %-EIO - I/O error
2657  *
2658  * %-ENOSPC - No space left on device (only in a panic state).
2659  *
2660  * %-ERESTARTSYS - Interrupted.
2661  *
2662  * %-ENOMEM - Insufficient memory available.
2663  */
2664 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2665                                   loff_t start, loff_t end)
2666 {
2667         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2668         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2669         struct nilfs_inode_info *ii;
2670         struct nilfs_transaction_info ti;
2671         int err = 0;
2672
2673         if (!sci)
2674                 return -EROFS;
2675
2676         nilfs_transaction_lock(sbi, &ti, 0);
2677
2678         ii = NILFS_I(inode);
2679         if (test_bit(NILFS_I_INODE_DIRTY, &ii->i_state) ||
2680             nilfs_test_opt(sbi, STRICT_ORDER) ||
2681             test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2682             nilfs_discontinued(sbi->s_nilfs)) {
2683                 nilfs_transaction_unlock(sbi);
2684                 err = nilfs_segctor_sync(sci);
2685                 return err;
2686         }
2687
2688         spin_lock(&sbi->s_inode_lock);
2689         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2690             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2691                 spin_unlock(&sbi->s_inode_lock);
2692                 nilfs_transaction_unlock(sbi);
2693                 return 0;
2694         }
2695         spin_unlock(&sbi->s_inode_lock);
2696         sci->sc_dsync_inode = ii;
2697         sci->sc_dsync_start = start;
2698         sci->sc_dsync_end = end;
2699
2700         err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2701
2702         nilfs_transaction_unlock(sbi);
2703         return err;
2704 }
2705
2706 struct nilfs_segctor_req {
2707         int mode;
2708         __u32 seq_accepted;
2709         int sc_err;  /* construction failure */
2710         int sb_err;  /* super block writeback failure */
2711 };
2712
2713 #define FLUSH_FILE_BIT  (0x1) /* data file only */
2714 #define FLUSH_DAT_BIT   (1 << NILFS_DAT_INO) /* DAT only */
2715
2716 static void nilfs_segctor_accept(struct nilfs_sc_info *sci,
2717                                  struct nilfs_segctor_req *req)
2718 {
2719         BUG_ON(!sci);
2720
2721         req->sc_err = req->sb_err = 0;
2722         spin_lock(&sci->sc_state_lock);
2723         req->seq_accepted = sci->sc_seq_request;
2724         spin_unlock(&sci->sc_state_lock);
2725
2726         if (sci->sc_timer)
2727                 del_timer_sync(sci->sc_timer);
2728 }
2729
2730 static void nilfs_segctor_notify(struct nilfs_sc_info *sci,
2731                                  struct nilfs_segctor_req *req)
2732 {
2733         /* Clear requests (even when the construction failed) */
2734         spin_lock(&sci->sc_state_lock);
2735
2736         sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2737
2738         if (req->mode == SC_LSEG_SR) {
2739                 sci->sc_seq_done = req->seq_accepted;
2740                 nilfs_segctor_wakeup(sci, req->sc_err ? : req->sb_err);
2741                 sci->sc_flush_request = 0;
2742         } else if (req->mode == SC_FLUSH_FILE)
2743                 sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2744         else if (req->mode == SC_FLUSH_DAT)
2745                 sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2746
2747         spin_unlock(&sci->sc_state_lock);
2748 }
2749
2750 static int nilfs_segctor_construct(struct nilfs_sc_info *sci,
2751                                    struct nilfs_segctor_req *req)
2752 {
2753         struct nilfs_sb_info *sbi = sci->sc_sbi;
2754         struct the_nilfs *nilfs = sbi->s_nilfs;
2755         int err = 0;
2756
2757         if (nilfs_discontinued(nilfs))
2758                 req->mode = SC_LSEG_SR;
2759         if (!nilfs_segctor_confirm(sci)) {
2760                 err = nilfs_segctor_do_construct(sci, req->mode);
2761                 req->sc_err = err;
2762         }
2763         if (likely(!err)) {
2764                 if (req->mode != SC_FLUSH_DAT)
2765                         atomic_set(&nilfs->ns_ndirtyblks, 0);
2766                 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2767                     nilfs_discontinued(nilfs)) {
2768                         down_write(&nilfs->ns_sem);
2769                         req->sb_err = nilfs_commit_super(sbi);
2770                         up_write(&nilfs->ns_sem);
2771                 }
2772         }
2773         return err;
2774 }
2775
2776 static void nilfs_construction_timeout(unsigned long data)
2777 {
2778         struct task_struct *p = (struct task_struct *)data;
2779         wake_up_process(p);
2780 }
2781
2782 static void
2783 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2784 {
2785         struct nilfs_inode_info *ii, *n;
2786
2787         list_for_each_entry_safe(ii, n, head, i_dirty) {
2788                 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2789                         continue;
2790                 hlist_del_init(&ii->vfs_inode.i_hash);
2791                 list_del_init(&ii->i_dirty);
2792                 nilfs_clear_gcinode(&ii->vfs_inode);
2793         }
2794 }
2795
2796 int nilfs_clean_segments(struct super_block *sb, void __user *argp)
2797 {
2798         struct nilfs_sb_info *sbi = NILFS_SB(sb);
2799         struct nilfs_sc_info *sci = NILFS_SC(sbi);
2800         struct the_nilfs *nilfs = sbi->s_nilfs;
2801         struct nilfs_transaction_info ti;
2802         struct nilfs_segctor_req req = { .mode = SC_LSEG_SR };
2803         int err;
2804
2805         if (unlikely(!sci))
2806                 return -EROFS;
2807
2808         nilfs_transaction_lock(sbi, &ti, 1);
2809
2810         err = nilfs_init_gcdat_inode(nilfs);
2811         if (unlikely(err))
2812                 goto out_unlock;
2813         err = nilfs_ioctl_prepare_clean_segments(nilfs, argp);
2814         if (unlikely(err))
2815                 goto out_unlock;
2816
2817         list_splice_init(&nilfs->ns_gc_inodes, sci->sc_gc_inodes.prev);
2818
2819         for (;;) {
2820                 nilfs_segctor_accept(sci, &req);
2821                 err = nilfs_segctor_construct(sci, &req);
2822                 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2823                 nilfs_segctor_notify(sci, &req);
2824
2825                 if (likely(!err))
2826                         break;
2827
2828                 nilfs_warning(sb, __func__,
2829                               "segment construction failed. (err=%d)", err);
2830                 set_current_state(TASK_INTERRUPTIBLE);
2831                 schedule_timeout(sci->sc_interval);
2832         }
2833
2834  out_unlock:
2835         nilfs_clear_gcdat_inode(nilfs);
2836         nilfs_transaction_unlock(sbi);
2837         return err;
2838 }
2839
2840 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2841 {
2842         struct nilfs_sb_info *sbi = sci->sc_sbi;
2843         struct nilfs_transaction_info ti;
2844         struct nilfs_segctor_req req = { .mode = mode };
2845
2846         nilfs_transaction_lock(sbi, &ti, 0);
2847
2848         nilfs_segctor_accept(sci, &req);
2849         nilfs_segctor_construct(sci, &req);
2850         nilfs_segctor_notify(sci, &req);
2851
2852         /*
2853          * Unclosed segment should be retried.  We do this using sc_timer.
2854          * Timeout of sc_timer will invoke complete construction which leads
2855          * to close the current logical segment.
2856          */
2857         if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2858                 nilfs_segctor_start_timer(sci);
2859
2860         nilfs_transaction_unlock(sbi);
2861 }
2862
2863 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2864 {
2865         int mode = 0;
2866         int err;
2867
2868         spin_lock(&sci->sc_state_lock);
2869         mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2870                 SC_FLUSH_DAT : SC_FLUSH_FILE;
2871         spin_unlock(&sci->sc_state_lock);
2872
2873         if (mode) {
2874                 err = nilfs_segctor_do_construct(sci, mode);
2875
2876                 spin_lock(&sci->sc_state_lock);
2877                 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2878                         ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2879                 spin_unlock(&sci->sc_state_lock);
2880         }
2881         clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2882 }
2883
2884 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2885 {
2886         if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2887             time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2888                 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2889                         return SC_FLUSH_FILE;
2890                 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2891                         return SC_FLUSH_DAT;
2892         }
2893         return SC_LSEG_SR;
2894 }
2895
2896 /**
2897  * nilfs_segctor_thread - main loop of the segment constructor thread.
2898  * @arg: pointer to a struct nilfs_sc_info.
2899  *
2900  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2901  * to execute segment constructions.
2902  */
2903 static int nilfs_segctor_thread(void *arg)
2904 {
2905         struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2906         struct timer_list timer;
2907         int timeout = 0;
2908
2909         init_timer(&timer);
2910         timer.data = (unsigned long)current;
2911         timer.function = nilfs_construction_timeout;
2912         sci->sc_timer = &timer;
2913
2914         /* start sync. */
2915         sci->sc_task = current;
2916         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2917         printk(KERN_INFO
2918                "segctord starting. Construction interval = %lu seconds, "
2919                "CP frequency < %lu seconds\n",
2920                sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2921
2922         spin_lock(&sci->sc_state_lock);
2923  loop:
2924         for (;;) {
2925                 int mode;
2926
2927                 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2928                         goto end_thread;
2929
2930                 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2931                         mode = SC_LSEG_SR;
2932                 else if (!sci->sc_flush_request)
2933                         break;
2934                 else
2935                         mode = nilfs_segctor_flush_mode(sci);
2936
2937                 spin_unlock(&sci->sc_state_lock);
2938                 nilfs_segctor_thread_construct(sci, mode);
2939                 spin_lock(&sci->sc_state_lock);
2940                 timeout = 0;
2941         }
2942
2943
2944         if (freezing(current)) {
2945                 spin_unlock(&sci->sc_state_lock);
2946                 refrigerator();
2947                 spin_lock(&sci->sc_state_lock);
2948         } else {
2949                 DEFINE_WAIT(wait);
2950                 int should_sleep = 1;
2951
2952                 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2953                                 TASK_INTERRUPTIBLE);
2954
2955                 if (sci->sc_seq_request != sci->sc_seq_done)
2956                         should_sleep = 0;
2957                 else if (sci->sc_flush_request)
2958                         should_sleep = 0;
2959                 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2960                         should_sleep = time_before(jiffies,
2961                                                    sci->sc_timer->expires);
2962
2963                 if (should_sleep) {
2964                         spin_unlock(&sci->sc_state_lock);
2965                         schedule();
2966                         spin_lock(&sci->sc_state_lock);
2967                 }
2968                 finish_wait(&sci->sc_wait_daemon, &wait);
2969                 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2970                            time_after_eq(jiffies, sci->sc_timer->expires));
2971         }
2972         goto loop;
2973
2974  end_thread:
2975         spin_unlock(&sci->sc_state_lock);
2976         del_timer_sync(sci->sc_timer);
2977         sci->sc_timer = NULL;
2978
2979         /* end sync. */
2980         sci->sc_task = NULL;
2981         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2982         return 0;
2983 }
2984
2985 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2986 {
2987         struct task_struct *t;
2988
2989         t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2990         if (IS_ERR(t)) {
2991                 int err = PTR_ERR(t);
2992
2993                 printk(KERN_ERR "NILFS: error %d creating segctord thread\n",
2994                        err);
2995                 return err;
2996         }
2997         wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2998         return 0;
2999 }
3000
3001 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
3002 {
3003         sci->sc_state |= NILFS_SEGCTOR_QUIT;
3004
3005         while (sci->sc_task) {
3006                 wake_up(&sci->sc_wait_daemon);
3007                 spin_unlock(&sci->sc_state_lock);
3008                 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
3009                 spin_lock(&sci->sc_state_lock);
3010         }
3011 }
3012
3013 static int nilfs_segctor_init(struct nilfs_sc_info *sci,
3014                               struct nilfs_recovery_info *ri)
3015 {
3016         int err;
3017         struct inode *inode = nilfs_iget(sci->sc_super, NILFS_SKETCH_INO);
3018
3019         sci->sc_sketch_inode = IS_ERR(inode) ? NULL : inode;
3020         if (sci->sc_sketch_inode)
3021                 i_size_write(sci->sc_sketch_inode, 0);
3022
3023         sci->sc_seq_done = sci->sc_seq_request;
3024         if (ri)
3025                 list_splice_init(&ri->ri_used_segments,
3026                                  sci->sc_active_segments.prev);
3027
3028         err = nilfs_segctor_start_thread(sci);
3029         if (err) {
3030                 if (ri)
3031                         list_splice_init(&sci->sc_active_segments,
3032                                          ri->ri_used_segments.prev);
3033                 if (sci->sc_sketch_inode) {
3034                         iput(sci->sc_sketch_inode);
3035                         sci->sc_sketch_inode = NULL;
3036                 }
3037         }
3038         return err;
3039 }
3040
3041 /*
3042  * Setup & clean-up functions
3043  */
3044 static struct nilfs_sc_info *nilfs_segctor_new(struct nilfs_sb_info *sbi)
3045 {
3046         struct nilfs_sc_info *sci;
3047
3048         sci = kzalloc(sizeof(*sci), GFP_KERNEL);
3049         if (!sci)
3050                 return NULL;
3051
3052         sci->sc_sbi = sbi;
3053         sci->sc_super = sbi->s_super;
3054
3055         init_waitqueue_head(&sci->sc_wait_request);
3056         init_waitqueue_head(&sci->sc_wait_daemon);
3057         init_waitqueue_head(&sci->sc_wait_task);
3058         spin_lock_init(&sci->sc_state_lock);
3059         INIT_LIST_HEAD(&sci->sc_dirty_files);
3060         INIT_LIST_HEAD(&sci->sc_segbufs);
3061         INIT_LIST_HEAD(&sci->sc_gc_inodes);
3062         INIT_LIST_HEAD(&sci->sc_active_segments);
3063         INIT_LIST_HEAD(&sci->sc_cleaning_segments);
3064         INIT_LIST_HEAD(&sci->sc_copied_buffers);
3065
3066         sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
3067         sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
3068         sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
3069
3070         if (sbi->s_interval)
3071                 sci->sc_interval = sbi->s_interval;
3072         if (sbi->s_watermark)
3073                 sci->sc_watermark = sbi->s_watermark;
3074         return sci;
3075 }
3076
3077 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
3078 {
3079         int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
3080
3081         /* The segctord thread was stopped and its timer was removed.
3082            But some tasks remain. */
3083         do {
3084                 struct nilfs_sb_info *sbi = sci->sc_sbi;
3085                 struct nilfs_transaction_info ti;
3086                 struct nilfs_segctor_req req = { .mode = SC_LSEG_SR };
3087
3088                 nilfs_transaction_lock(sbi, &ti, 0);
3089                 nilfs_segctor_accept(sci, &req);
3090                 ret = nilfs_segctor_construct(sci, &req);
3091                 nilfs_segctor_notify(sci, &req);
3092                 nilfs_transaction_unlock(sbi);
3093
3094         } while (ret && retrycount-- > 0);
3095 }
3096
3097 /**
3098  * nilfs_segctor_destroy - destroy the segment constructor.
3099  * @sci: nilfs_sc_info
3100  *
3101  * nilfs_segctor_destroy() kills the segctord thread and frees
3102  * the nilfs_sc_info struct.
3103  * Caller must hold the segment semaphore.
3104  */
3105 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
3106 {
3107         struct nilfs_sb_info *sbi = sci->sc_sbi;
3108         int flag;
3109
3110         up_write(&sbi->s_nilfs->ns_segctor_sem);
3111
3112         spin_lock(&sci->sc_state_lock);
3113         nilfs_segctor_kill_thread(sci);
3114         flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
3115                 || sci->sc_seq_request != sci->sc_seq_done);
3116         spin_unlock(&sci->sc_state_lock);
3117
3118         if (flag || nilfs_segctor_confirm(sci))
3119                 nilfs_segctor_write_out(sci);
3120
3121         BUG_ON(!list_empty(&sci->sc_copied_buffers));
3122
3123         if (!list_empty(&sci->sc_dirty_files)) {
3124                 nilfs_warning(sbi->s_super, __func__,
3125                               "dirty file(s) after the final construction\n");
3126                 nilfs_dispose_list(sbi, &sci->sc_dirty_files, 1);
3127         }
3128         if (!list_empty(&sci->sc_active_segments))
3129                 nilfs_dispose_segment_list(&sci->sc_active_segments);
3130
3131         if (!list_empty(&sci->sc_cleaning_segments))
3132                 nilfs_dispose_segment_list(&sci->sc_cleaning_segments);
3133
3134         BUG_ON(!list_empty(&sci->sc_segbufs));
3135
3136         if (sci->sc_sketch_inode) {
3137                 iput(sci->sc_sketch_inode);
3138                 sci->sc_sketch_inode = NULL;
3139         }
3140         down_write(&sbi->s_nilfs->ns_segctor_sem);
3141
3142         kfree(sci);
3143 }
3144
3145 /**
3146  * nilfs_attach_segment_constructor - attach a segment constructor
3147  * @sbi: nilfs_sb_info
3148  * @ri: nilfs_recovery_info
3149  *
3150  * nilfs_attach_segment_constructor() allocates a struct nilfs_sc_info,
3151  * initilizes it, and starts the segment constructor.
3152  *
3153  * Return Value: On success, 0 is returned. On error, one of the following
3154  * negative error code is returned.
3155  *
3156  * %-ENOMEM - Insufficient memory available.
3157  */
3158 int nilfs_attach_segment_constructor(struct nilfs_sb_info *sbi,
3159                                      struct nilfs_recovery_info *ri)
3160 {
3161         struct the_nilfs *nilfs = sbi->s_nilfs;
3162         int err;
3163
3164         /* Each field of nilfs_segctor is cleared through the initialization
3165            of super-block info */
3166         sbi->s_sc_info = nilfs_segctor_new(sbi);
3167         if (!sbi->s_sc_info)
3168                 return -ENOMEM;
3169
3170         nilfs_attach_writer(nilfs, sbi);
3171         err = nilfs_segctor_init(NILFS_SC(sbi), ri);
3172         if (err) {
3173                 nilfs_detach_writer(nilfs, sbi);
3174                 kfree(sbi->s_sc_info);
3175                 sbi->s_sc_info = NULL;
3176         }
3177         return err;
3178 }
3179
3180 /**
3181  * nilfs_detach_segment_constructor - destroy the segment constructor
3182  * @sbi: nilfs_sb_info
3183  *
3184  * nilfs_detach_segment_constructor() kills the segment constructor daemon,
3185  * frees the struct nilfs_sc_info, and destroy the dirty file list.
3186  */
3187 void nilfs_detach_segment_constructor(struct nilfs_sb_info *sbi)
3188 {
3189         struct the_nilfs *nilfs = sbi->s_nilfs;
3190         LIST_HEAD(garbage_list);
3191
3192         down_write(&nilfs->ns_segctor_sem);
3193         if (NILFS_SC(sbi)) {
3194                 nilfs_segctor_destroy(NILFS_SC(sbi));
3195                 sbi->s_sc_info = NULL;
3196         }
3197
3198         /* Force to free the list of dirty files */
3199         spin_lock(&sbi->s_inode_lock);
3200         if (!list_empty(&sbi->s_dirty_files)) {
3201                 list_splice_init(&sbi->s_dirty_files, &garbage_list);
3202                 nilfs_warning(sbi->s_super, __func__,
3203                               "Non empty dirty list after the last "
3204                               "segment construction\n");
3205         }
3206         spin_unlock(&sbi->s_inode_lock);
3207         up_write(&nilfs->ns_segctor_sem);
3208
3209         nilfs_dispose_list(sbi, &garbage_list, 1);
3210         nilfs_detach_writer(nilfs, sbi);
3211 }