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  * Written by Ryusuke Konishi.
  17  *
  18  */
  19
  20 #include <linux/pagemap.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/writeback.h>
  23 #include <linux/bitops.h>
  24 #include <linux/bio.h>
  25 #include <linux/completion.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/freezer.h>
  29 #include <linux/kthread.h>
  30 #include <linux/crc32.h>
  31 #include <linux/pagevec.h>
  32 #include <linux/slab.h>
  33 #include "nilfs.h"
  34 #include "btnode.h"
  35 #include "page.h"
  36 #include "segment.h"
  37 #include "sufile.h"
  38 #include "cpfile.h"
  39 #include "ifile.h"
  40 #include "segbuf.h"
  41
  42
  43 /*
  44  * Segment constructor
  45  */
  46 #define SC_N_INODEVEC   16   /* Size of locally allocated inode vector */
  47
  48 #define SC_MAX_SEGDELTA 64   /*
  49                               * Upper limit of the number of segments
  50                               * appended in collection retry loop
  51                               */
  52
  53 /* Construction mode */
  54 enum {
  55         SC_LSEG_SR = 1, /* Make a logical segment having a super root */
  56         SC_LSEG_DSYNC,  /*
  57                          * Flush data blocks of a given file and make
  58                          * a logical segment without a super root.
  59                          */
  60         SC_FLUSH_FILE,  /*
  61                          * Flush data files, leads to segment writes without
  62                          * creating a checkpoint.
  63                          */
  64         SC_FLUSH_DAT,   /*
  65                          * Flush DAT file.  This also creates segments
  66                          * without a checkpoint.
  67                          */
  68 };
  69
  70 /* Stage numbers of dirty block collection */
  71 enum {
  72         NILFS_ST_INIT = 0,
  73         NILFS_ST_GC,            /* Collecting dirty blocks for GC */
  74         NILFS_ST_FILE,
  75         NILFS_ST_IFILE,
  76         NILFS_ST_CPFILE,
  77         NILFS_ST_SUFILE,
  78         NILFS_ST_DAT,
  79         NILFS_ST_SR,            /* Super root */
  80         NILFS_ST_DSYNC,         /* Data sync blocks */
  81         NILFS_ST_DONE,
  82 };
  83
  84 #define CREATE_TRACE_POINTS
  85 #include <trace/events/nilfs2.h>
  86
  87 /*
  88  * nilfs_sc_cstage_inc(), nilfs_sc_cstage_set(), nilfs_sc_cstage_get() are
  89  * wrapper functions of stage count (nilfs_sc_info->sc_stage.scnt). Users of
  90  * the variable must use them because transition of stage count must involve
  91  * trace events (trace_nilfs2_collection_stage_transition).
  92  *
  93  * nilfs_sc_cstage_get() isn't required for the above purpose because it doesn't
  94  * produce tracepoint events. It is provided just for making the intention
  95  * clear.
  96  */
  97 static inline void nilfs_sc_cstage_inc(struct nilfs_sc_info *sci)
  98 {
  99         sci->sc_stage.scnt++;
 100         trace_nilfs2_collection_stage_transition(sci);
 101 }
 102
 103 static inline void nilfs_sc_cstage_set(struct nilfs_sc_info *sci, int next_scnt)
 104 {
 105         sci->sc_stage.scnt = next_scnt;
 106         trace_nilfs2_collection_stage_transition(sci);
 107 }
 108
 109 static inline int nilfs_sc_cstage_get(struct nilfs_sc_info *sci)
 110 {
 111         return sci->sc_stage.scnt;
 112 }
 113
 114 /* State flags of collection */
 115 #define NILFS_CF_NODE           0x0001  /* Collecting node blocks */
 116 #define NILFS_CF_IFILE_STARTED  0x0002  /* IFILE stage has started */
 117 #define NILFS_CF_SUFREED        0x0004  /* segment usages has been freed */
 118 #define NILFS_CF_HISTORY_MASK   (NILFS_CF_IFILE_STARTED | NILFS_CF_SUFREED)
 119
 120 /* Operations depending on the construction mode and file type */
 121 struct nilfs_sc_operations {
 122         int (*collect_data)(struct nilfs_sc_info *, struct buffer_head *,
 123                             struct inode *);
 124         int (*collect_node)(struct nilfs_sc_info *, struct buffer_head *,
 125                             struct inode *);
 126         int (*collect_bmap)(struct nilfs_sc_info *, struct buffer_head *,
 127                             struct inode *);
 128         void (*write_data_binfo)(struct nilfs_sc_info *,
 129                                  struct nilfs_segsum_pointer *,
 130                                  union nilfs_binfo *);
 131         void (*write_node_binfo)(struct nilfs_sc_info *,
 132                                  struct nilfs_segsum_pointer *,
 133                                  union nilfs_binfo *);
 134 };
 135
 136 /*
 137  * Other definitions
 138  */
 139 static void nilfs_segctor_start_timer(struct nilfs_sc_info *);
 140 static void nilfs_segctor_do_flush(struct nilfs_sc_info *, int);
 141 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *);
 142 static void nilfs_dispose_list(struct the_nilfs *, struct list_head *, int);
 143
 144 #define nilfs_cnt32_gt(a, b)   \
 145         (typecheck(__u32, a) && typecheck(__u32, b) && \
 146          ((__s32)(b) - (__s32)(a) < 0))
 147 #define nilfs_cnt32_ge(a, b)   \
 148         (typecheck(__u32, a) && typecheck(__u32, b) && \
 149          ((__s32)(a) - (__s32)(b) >= 0))
 150 #define nilfs_cnt32_lt(a, b)  nilfs_cnt32_gt(b, a)
 151 #define nilfs_cnt32_le(a, b)  nilfs_cnt32_ge(b, a)
 152
 153 static int nilfs_prepare_segment_lock(struct super_block *sb,
 154                                       struct nilfs_transaction_info *ti)
 155 {
 156         struct nilfs_transaction_info *cur_ti = current->journal_info;
 157         void *save = NULL;
 158
 159         if (cur_ti) {
 160                 if (cur_ti->ti_magic == NILFS_TI_MAGIC)
 161                         return ++cur_ti->ti_count;
 162
 163                 /*
 164                  * If journal_info field is occupied by other FS,
 165                  * it is saved and will be restored on
 166                  * nilfs_transaction_commit().
 167                  */
 168                 nilfs_msg(sb, KERN_WARNING, "journal info from a different FS");
 169                 save = current->journal_info;
 170         }
 171         if (!ti) {
 172                 ti = kmem_cache_alloc(nilfs_transaction_cachep, GFP_NOFS);
 173                 if (!ti)
 174                         return -ENOMEM;
 175                 ti->ti_flags = NILFS_TI_DYNAMIC_ALLOC;
 176         } else {
 177                 ti->ti_flags = 0;
 178         }
 179         ti->ti_count = 0;
 180         ti->ti_save = save;
 181         ti->ti_magic = NILFS_TI_MAGIC;
 182         current->journal_info = ti;
 183         return 0;
 184 }
 185
 186 /**
 187  * nilfs_transaction_begin - start indivisible file operations.
 188  * @sb: super block
 189  * @ti: nilfs_transaction_info
 190  * @vacancy_check: flags for vacancy rate checks
 191  *
 192  * nilfs_transaction_begin() acquires a reader/writer semaphore, called
 193  * the segment semaphore, to make a segment construction and write tasks
 194  * exclusive.  The function is used with nilfs_transaction_commit() in pairs.
 195  * The region enclosed by these two functions can be nested.  To avoid a
 196  * deadlock, the semaphore is only acquired or released in the outermost call.
 197  *
 198  * This function allocates a nilfs_transaction_info struct to keep context
 199  * information on it.  It is initialized and hooked onto the current task in
 200  * the outermost call.  If a pre-allocated struct is given to @ti, it is used
 201  * instead; otherwise a new struct is assigned from a slab.
 202  *
 203  * When @vacancy_check flag is set, this function will check the amount of
 204  * free space, and will wait for the GC to reclaim disk space if low capacity.
 205  *
 206  * Return Value: On success, 0 is returned. On error, one of the following
 207  * negative error code is returned.
 208  *
 209  * %-ENOMEM - Insufficient memory available.
 210  *
 211  * %-ENOSPC - No space left on device
 212  */
 213 int nilfs_transaction_begin(struct super_block *sb,
 214                             struct nilfs_transaction_info *ti,
 215                             int vacancy_check)
 216 {
 217         struct the_nilfs *nilfs;
 218         int ret = nilfs_prepare_segment_lock(sb, ti);
 219         struct nilfs_transaction_info *trace_ti;
 220
 221         if (unlikely(ret < 0))
 222                 return ret;
 223         if (ret > 0) {
 224                 trace_ti = current->journal_info;
 225
 226                 trace_nilfs2_transaction_transition(sb, trace_ti,
 227                                     trace_ti->ti_count, trace_ti->ti_flags,
 228                                     TRACE_NILFS2_TRANSACTION_BEGIN);
 229                 return 0;
 230         }
 231
 232         sb_start_intwrite(sb);
 233
 234         nilfs = sb->s_fs_info;
 235         down_read(&nilfs->ns_segctor_sem);
 236         if (vacancy_check && nilfs_near_disk_full(nilfs)) {
 237                 up_read(&nilfs->ns_segctor_sem);
 238                 ret = -ENOSPC;
 239                 goto failed;
 240         }
 241
 242         trace_ti = current->journal_info;
 243         trace_nilfs2_transaction_transition(sb, trace_ti, trace_ti->ti_count,
 244                                             trace_ti->ti_flags,
 245                                             TRACE_NILFS2_TRANSACTION_BEGIN);
 246         return 0;
 247
 248  failed:
 249         ti = current->journal_info;
 250         current->journal_info = ti->ti_save;
 251         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 252                 kmem_cache_free(nilfs_transaction_cachep, ti);
 253         sb_end_intwrite(sb);
 254         return ret;
 255 }
 256
 257 /**
 258  * nilfs_transaction_commit - commit indivisible file operations.
 259  * @sb: super block
 260  *
 261  * nilfs_transaction_commit() releases the read semaphore which is
 262  * acquired by nilfs_transaction_begin(). This is only performed
 263  * in outermost call of this function.  If a commit flag is set,
 264  * nilfs_transaction_commit() sets a timer to start the segment
 265  * constructor.  If a sync flag is set, it starts construction
 266  * directly.
 267  */
 268 int nilfs_transaction_commit(struct super_block *sb)
 269 {
 270         struct nilfs_transaction_info *ti = current->journal_info;
 271         struct the_nilfs *nilfs = sb->s_fs_info;
 272         int err = 0;
 273
 274         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 275         ti->ti_flags |= NILFS_TI_COMMIT;
 276         if (ti->ti_count > 0) {
 277                 ti->ti_count--;
 278                 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 279                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
 280                 return 0;
 281         }
 282         if (nilfs->ns_writer) {
 283                 struct nilfs_sc_info *sci = nilfs->ns_writer;
 284
 285                 if (ti->ti_flags & NILFS_TI_COMMIT)
 286                         nilfs_segctor_start_timer(sci);
 287                 if (atomic_read(&nilfs->ns_ndirtyblks) > sci->sc_watermark)
 288                         nilfs_segctor_do_flush(sci, 0);
 289         }
 290         up_read(&nilfs->ns_segctor_sem);
 291         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 292                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_COMMIT);
 293
 294         current->journal_info = ti->ti_save;
 295
 296         if (ti->ti_flags & NILFS_TI_SYNC)
 297                 err = nilfs_construct_segment(sb);
 298         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 299                 kmem_cache_free(nilfs_transaction_cachep, ti);
 300         sb_end_intwrite(sb);
 301         return err;
 302 }
 303
 304 void nilfs_transaction_abort(struct super_block *sb)
 305 {
 306         struct nilfs_transaction_info *ti = current->journal_info;
 307         struct the_nilfs *nilfs = sb->s_fs_info;
 308
 309         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 310         if (ti->ti_count > 0) {
 311                 ti->ti_count--;
 312                 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 313                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
 314                 return;
 315         }
 316         up_read(&nilfs->ns_segctor_sem);
 317
 318         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 319                     ti->ti_flags, TRACE_NILFS2_TRANSACTION_ABORT);
 320
 321         current->journal_info = ti->ti_save;
 322         if (ti->ti_flags & NILFS_TI_DYNAMIC_ALLOC)
 323                 kmem_cache_free(nilfs_transaction_cachep, ti);
 324         sb_end_intwrite(sb);
 325 }
 326
 327 void nilfs_relax_pressure_in_lock(struct super_block *sb)
 328 {
 329         struct the_nilfs *nilfs = sb->s_fs_info;
 330         struct nilfs_sc_info *sci = nilfs->ns_writer;
 331
 332         if (!sci || !sci->sc_flush_request)
 333                 return;
 334
 335         set_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
 336         up_read(&nilfs->ns_segctor_sem);
 337
 338         down_write(&nilfs->ns_segctor_sem);
 339         if (sci->sc_flush_request &&
 340             test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags)) {
 341                 struct nilfs_transaction_info *ti = current->journal_info;
 342
 343                 ti->ti_flags |= NILFS_TI_WRITER;
 344                 nilfs_segctor_do_immediate_flush(sci);
 345                 ti->ti_flags &= ~NILFS_TI_WRITER;
 346         }
 347         downgrade_write(&nilfs->ns_segctor_sem);
 348 }
 349
 350 static void nilfs_transaction_lock(struct super_block *sb,
 351                                    struct nilfs_transaction_info *ti,
 352                                    int gcflag)
 353 {
 354         struct nilfs_transaction_info *cur_ti = current->journal_info;
 355         struct the_nilfs *nilfs = sb->s_fs_info;
 356         struct nilfs_sc_info *sci = nilfs->ns_writer;
 357
 358         WARN_ON(cur_ti);
 359         ti->ti_flags = NILFS_TI_WRITER;
 360         ti->ti_count = 0;
 361         ti->ti_save = cur_ti;
 362         ti->ti_magic = NILFS_TI_MAGIC;
 363         current->journal_info = ti;
 364
 365         for (;;) {
 366                 trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 367                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_TRYLOCK);
 368
 369                 down_write(&nilfs->ns_segctor_sem);
 370                 if (!test_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags))
 371                         break;
 372
 373                 nilfs_segctor_do_immediate_flush(sci);
 374
 375                 up_write(&nilfs->ns_segctor_sem);
 376                 cond_resched();
 377         }
 378         if (gcflag)
 379                 ti->ti_flags |= NILFS_TI_GC;
 380
 381         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 382                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_LOCK);
 383 }
 384
 385 static void nilfs_transaction_unlock(struct super_block *sb)
 386 {
 387         struct nilfs_transaction_info *ti = current->journal_info;
 388         struct the_nilfs *nilfs = sb->s_fs_info;
 389
 390         BUG_ON(ti == NULL || ti->ti_magic != NILFS_TI_MAGIC);
 391         BUG_ON(ti->ti_count > 0);
 392
 393         up_write(&nilfs->ns_segctor_sem);
 394         current->journal_info = ti->ti_save;
 395
 396         trace_nilfs2_transaction_transition(sb, ti, ti->ti_count,
 397                             ti->ti_flags, TRACE_NILFS2_TRANSACTION_UNLOCK);
 398 }
 399
 400 static void *nilfs_segctor_map_segsum_entry(struct nilfs_sc_info *sci,
 401                                             struct nilfs_segsum_pointer *ssp,
 402                                             unsigned int bytes)
 403 {
 404         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 405         unsigned int blocksize = sci->sc_super->s_blocksize;
 406         void *p;
 407
 408         if (unlikely(ssp->offset + bytes > blocksize)) {
 409                 ssp->offset = 0;
 410                 BUG_ON(NILFS_SEGBUF_BH_IS_LAST(ssp->bh,
 411                                                &segbuf->sb_segsum_buffers));
 412                 ssp->bh = NILFS_SEGBUF_NEXT_BH(ssp->bh);
 413         }
 414         p = ssp->bh->b_data + ssp->offset;
 415         ssp->offset += bytes;
 416         return p;
 417 }
 418
 419 /**
 420  * nilfs_segctor_reset_segment_buffer - reset the current segment buffer
 421  * @sci: nilfs_sc_info
 422  */
 423 static int nilfs_segctor_reset_segment_buffer(struct nilfs_sc_info *sci)
 424 {
 425         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 426         struct buffer_head *sumbh;
 427         unsigned int sumbytes;
 428         unsigned int flags = 0;
 429         int err;
 430
 431         if (nilfs_doing_gc())
 432                 flags = NILFS_SS_GC;
 433         err = nilfs_segbuf_reset(segbuf, flags, sci->sc_seg_ctime, sci->sc_cno);
 434         if (unlikely(err))
 435                 return err;
 436
 437         sumbh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
 438         sumbytes = segbuf->sb_sum.sumbytes;
 439         sci->sc_finfo_ptr.bh = sumbh;  sci->sc_finfo_ptr.offset = sumbytes;
 440         sci->sc_binfo_ptr.bh = sumbh;  sci->sc_binfo_ptr.offset = sumbytes;
 441         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 442         return 0;
 443 }
 444
 445 static int nilfs_segctor_feed_segment(struct nilfs_sc_info *sci)
 446 {
 447         sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
 448         if (NILFS_SEGBUF_IS_LAST(sci->sc_curseg, &sci->sc_segbufs))
 449                 return -E2BIG; /*
 450                                 * The current segment is filled up
 451                                 * (internal code)
 452                                 */
 453         sci->sc_curseg = NILFS_NEXT_SEGBUF(sci->sc_curseg);
 454         return nilfs_segctor_reset_segment_buffer(sci);
 455 }
 456
 457 static int nilfs_segctor_add_super_root(struct nilfs_sc_info *sci)
 458 {
 459         struct nilfs_segment_buffer *segbuf = sci->sc_curseg;
 460         int err;
 461
 462         if (segbuf->sb_sum.nblocks >= segbuf->sb_rest_blocks) {
 463                 err = nilfs_segctor_feed_segment(sci);
 464                 if (err)
 465                         return err;
 466                 segbuf = sci->sc_curseg;
 467         }
 468         err = nilfs_segbuf_extend_payload(segbuf, &segbuf->sb_super_root);
 469         if (likely(!err))
 470                 segbuf->sb_sum.flags |= NILFS_SS_SR;
 471         return err;
 472 }
 473
 474 /*
 475  * Functions for making segment summary and payloads
 476  */
 477 static int nilfs_segctor_segsum_block_required(
 478         struct nilfs_sc_info *sci, const struct nilfs_segsum_pointer *ssp,
 479         unsigned int binfo_size)
 480 {
 481         unsigned int blocksize = sci->sc_super->s_blocksize;
 482         /* Size of finfo and binfo is enough small against blocksize */
 483
 484         return ssp->offset + binfo_size +
 485                 (!sci->sc_blk_cnt ? sizeof(struct nilfs_finfo) : 0) >
 486                 blocksize;
 487 }
 488
 489 static void nilfs_segctor_begin_finfo(struct nilfs_sc_info *sci,
 490                                       struct inode *inode)
 491 {
 492         sci->sc_curseg->sb_sum.nfinfo++;
 493         sci->sc_binfo_ptr = sci->sc_finfo_ptr;
 494         nilfs_segctor_map_segsum_entry(
 495                 sci, &sci->sc_binfo_ptr, sizeof(struct nilfs_finfo));
 496
 497         if (NILFS_I(inode)->i_root &&
 498             !test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 499                 set_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
 500         /* skip finfo */
 501 }
 502
 503 static void nilfs_segctor_end_finfo(struct nilfs_sc_info *sci,
 504                                     struct inode *inode)
 505 {
 506         struct nilfs_finfo *finfo;
 507         struct nilfs_inode_info *ii;
 508         struct nilfs_segment_buffer *segbuf;
 509         __u64 cno;
 510
 511         if (sci->sc_blk_cnt == 0)
 512                 return;
 513
 514         ii = NILFS_I(inode);
 515
 516         if (test_bit(NILFS_I_GCINODE, &ii->i_state))
 517                 cno = ii->i_cno;
 518         else if (NILFS_ROOT_METADATA_FILE(inode->i_ino))
 519                 cno = 0;
 520         else
 521                 cno = sci->sc_cno;
 522
 523         finfo = nilfs_segctor_map_segsum_entry(sci, &sci->sc_finfo_ptr,
 524                                                  sizeof(*finfo));
 525         finfo->fi_ino = cpu_to_le64(inode->i_ino);
 526         finfo->fi_nblocks = cpu_to_le32(sci->sc_blk_cnt);
 527         finfo->fi_ndatablk = cpu_to_le32(sci->sc_datablk_cnt);
 528         finfo->fi_cno = cpu_to_le64(cno);
 529
 530         segbuf = sci->sc_curseg;
 531         segbuf->sb_sum.sumbytes = sci->sc_binfo_ptr.offset +
 532                 sci->sc_super->s_blocksize * (segbuf->sb_sum.nsumblk - 1);
 533         sci->sc_finfo_ptr = sci->sc_binfo_ptr;
 534         sci->sc_blk_cnt = sci->sc_datablk_cnt = 0;
 535 }
 536
 537 static int nilfs_segctor_add_file_block(struct nilfs_sc_info *sci,
 538                                         struct buffer_head *bh,
 539                                         struct inode *inode,
 540                                         unsigned int binfo_size)
 541 {
 542         struct nilfs_segment_buffer *segbuf;
 543         int required, err = 0;
 544
 545  retry:
 546         segbuf = sci->sc_curseg;
 547         required = nilfs_segctor_segsum_block_required(
 548                 sci, &sci->sc_binfo_ptr, binfo_size);
 549         if (segbuf->sb_sum.nblocks + required + 1 > segbuf->sb_rest_blocks) {
 550                 nilfs_segctor_end_finfo(sci, inode);
 551                 err = nilfs_segctor_feed_segment(sci);
 552                 if (err)
 553                         return err;
 554                 goto retry;
 555         }
 556         if (unlikely(required)) {
 557                 err = nilfs_segbuf_extend_segsum(segbuf);
 558                 if (unlikely(err))
 559                         goto failed;
 560         }
 561         if (sci->sc_blk_cnt == 0)
 562                 nilfs_segctor_begin_finfo(sci, inode);
 563
 564         nilfs_segctor_map_segsum_entry(sci, &sci->sc_binfo_ptr, binfo_size);
 565         /* Substitution to vblocknr is delayed until update_blocknr() */
 566         nilfs_segbuf_add_file_buffer(segbuf, bh);
 567         sci->sc_blk_cnt++;
 568  failed:
 569         return err;
 570 }
 571
 572 /*
 573  * Callback functions that enumerate, mark, and collect dirty blocks
 574  */
 575 static int nilfs_collect_file_data(struct nilfs_sc_info *sci,
 576                                    struct buffer_head *bh, struct inode *inode)
 577 {
 578         int err;
 579
 580         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 581         if (err < 0)
 582                 return err;
 583
 584         err = nilfs_segctor_add_file_block(sci, bh, inode,
 585                                            sizeof(struct nilfs_binfo_v));
 586         if (!err)
 587                 sci->sc_datablk_cnt++;
 588         return err;
 589 }
 590
 591 static int nilfs_collect_file_node(struct nilfs_sc_info *sci,
 592                                    struct buffer_head *bh,
 593                                    struct inode *inode)
 594 {
 595         return nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 596 }
 597
 598 static int nilfs_collect_file_bmap(struct nilfs_sc_info *sci,
 599                                    struct buffer_head *bh,
 600                                    struct inode *inode)
 601 {
 602         WARN_ON(!buffer_dirty(bh));
 603         return nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 604 }
 605
 606 static void nilfs_write_file_data_binfo(struct nilfs_sc_info *sci,
 607                                         struct nilfs_segsum_pointer *ssp,
 608                                         union nilfs_binfo *binfo)
 609 {
 610         struct nilfs_binfo_v *binfo_v = nilfs_segctor_map_segsum_entry(
 611                 sci, ssp, sizeof(*binfo_v));
 612         *binfo_v = binfo->bi_v;
 613 }
 614
 615 static void nilfs_write_file_node_binfo(struct nilfs_sc_info *sci,
 616                                         struct nilfs_segsum_pointer *ssp,
 617                                         union nilfs_binfo *binfo)
 618 {
 619         __le64 *vblocknr = nilfs_segctor_map_segsum_entry(
 620                 sci, ssp, sizeof(*vblocknr));
 621         *vblocknr = binfo->bi_v.bi_vblocknr;
 622 }
 623
 624 static const struct nilfs_sc_operations nilfs_sc_file_ops = {
 625         .collect_data = nilfs_collect_file_data,
 626         .collect_node = nilfs_collect_file_node,
 627         .collect_bmap = nilfs_collect_file_bmap,
 628         .write_data_binfo = nilfs_write_file_data_binfo,
 629         .write_node_binfo = nilfs_write_file_node_binfo,
 630 };
 631
 632 static int nilfs_collect_dat_data(struct nilfs_sc_info *sci,
 633                                   struct buffer_head *bh, struct inode *inode)
 634 {
 635         int err;
 636
 637         err = nilfs_bmap_propagate(NILFS_I(inode)->i_bmap, bh);
 638         if (err < 0)
 639                 return err;
 640
 641         err = nilfs_segctor_add_file_block(sci, bh, inode, sizeof(__le64));
 642         if (!err)
 643                 sci->sc_datablk_cnt++;
 644         return err;
 645 }
 646
 647 static int nilfs_collect_dat_bmap(struct nilfs_sc_info *sci,
 648                                   struct buffer_head *bh, struct inode *inode)
 649 {
 650         WARN_ON(!buffer_dirty(bh));
 651         return nilfs_segctor_add_file_block(sci, bh, inode,
 652                                             sizeof(struct nilfs_binfo_dat));
 653 }
 654
 655 static void nilfs_write_dat_data_binfo(struct nilfs_sc_info *sci,
 656                                        struct nilfs_segsum_pointer *ssp,
 657                                        union nilfs_binfo *binfo)
 658 {
 659         __le64 *blkoff = nilfs_segctor_map_segsum_entry(sci, ssp,
 660                                                           sizeof(*blkoff));
 661         *blkoff = binfo->bi_dat.bi_blkoff;
 662 }
 663
 664 static void nilfs_write_dat_node_binfo(struct nilfs_sc_info *sci,
 665                                        struct nilfs_segsum_pointer *ssp,
 666                                        union nilfs_binfo *binfo)
 667 {
 668         struct nilfs_binfo_dat *binfo_dat =
 669                 nilfs_segctor_map_segsum_entry(sci, ssp, sizeof(*binfo_dat));
 670         *binfo_dat = binfo->bi_dat;
 671 }
 672
 673 static const struct nilfs_sc_operations nilfs_sc_dat_ops = {
 674         .collect_data = nilfs_collect_dat_data,
 675         .collect_node = nilfs_collect_file_node,
 676         .collect_bmap = nilfs_collect_dat_bmap,
 677         .write_data_binfo = nilfs_write_dat_data_binfo,
 678         .write_node_binfo = nilfs_write_dat_node_binfo,
 679 };
 680
 681 static const struct nilfs_sc_operations nilfs_sc_dsync_ops = {
 682         .collect_data = nilfs_collect_file_data,
 683         .collect_node = NULL,
 684         .collect_bmap = NULL,
 685         .write_data_binfo = nilfs_write_file_data_binfo,
 686         .write_node_binfo = NULL,
 687 };
 688
 689 static size_t nilfs_lookup_dirty_data_buffers(struct inode *inode,
 690                                               struct list_head *listp,
 691                                               size_t nlimit,
 692                                               loff_t start, loff_t end)
 693 {
 694         struct address_space *mapping = inode->i_mapping;
 695         struct pagevec pvec;
 696         pgoff_t index = 0, last = ULONG_MAX;
 697         size_t ndirties = 0;
 698         int i;
 699
 700         if (unlikely(start != 0 || end != LLONG_MAX)) {
 701                 /*
 702                  * A valid range is given for sync-ing data pages. The
 703                  * range is rounded to per-page; extra dirty buffers
 704                  * may be included if blocksize < pagesize.
 705                  */
 706                 index = start >> PAGE_SHIFT;
 707                 last = end >> PAGE_SHIFT;
 708         }
 709         pagevec_init(&pvec, 0);
 710  repeat:
 711         if (unlikely(index > last) ||
 712             !pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 713                                 min_t(pgoff_t, last - index,
 714                                       PAGEVEC_SIZE - 1) + 1))
 715                 return ndirties;
 716
 717         for (i = 0; i < pagevec_count(&pvec); i++) {
 718                 struct buffer_head *bh, *head;
 719                 struct page *page = pvec.pages[i];
 720
 721                 if (unlikely(page->index > last))
 722                         break;
 723
 724                 lock_page(page);
 725                 if (!page_has_buffers(page))
 726                         create_empty_buffers(page, i_blocksize(inode), 0);
 727                 unlock_page(page);
 728
 729                 bh = head = page_buffers(page);
 730                 do {
 731                         if (!buffer_dirty(bh) || buffer_async_write(bh))
 732                                 continue;
 733                         get_bh(bh);
 734                         list_add_tail(&bh->b_assoc_buffers, listp);
 735                         ndirties++;
 736                         if (unlikely(ndirties >= nlimit)) {
 737                                 pagevec_release(&pvec);
 738                                 cond_resched();
 739                                 return ndirties;
 740                         }
 741                 } while (bh = bh->b_this_page, bh != head);
 742         }
 743         pagevec_release(&pvec);
 744         cond_resched();
 745         goto repeat;
 746 }
 747
 748 static void nilfs_lookup_dirty_node_buffers(struct inode *inode,
 749                                             struct list_head *listp)
 750 {
 751         struct nilfs_inode_info *ii = NILFS_I(inode);
 752         struct address_space *mapping = &ii->i_btnode_cache;
 753         struct pagevec pvec;
 754         struct buffer_head *bh, *head;
 755         unsigned int i;
 756         pgoff_t index = 0;
 757
 758         pagevec_init(&pvec, 0);
 759
 760         while (pagevec_lookup_tag(&pvec, mapping, &index, PAGECACHE_TAG_DIRTY,
 761                                   PAGEVEC_SIZE)) {
 762                 for (i = 0; i < pagevec_count(&pvec); i++) {
 763                         bh = head = page_buffers(pvec.pages[i]);
 764                         do {
 765                                 if (buffer_dirty(bh) &&
 766                                                 !buffer_async_write(bh)) {
 767                                         get_bh(bh);
 768                                         list_add_tail(&bh->b_assoc_buffers,
 769                                                       listp);
 770                                 }
 771                                 bh = bh->b_this_page;
 772                         } while (bh != head);
 773                 }
 774                 pagevec_release(&pvec);
 775                 cond_resched();
 776         }
 777 }
 778
 779 static void nilfs_dispose_list(struct the_nilfs *nilfs,
 780                                struct list_head *head, int force)
 781 {
 782         struct nilfs_inode_info *ii, *n;
 783         struct nilfs_inode_info *ivec[SC_N_INODEVEC], **pii;
 784         unsigned int nv = 0;
 785
 786         while (!list_empty(head)) {
 787                 spin_lock(&nilfs->ns_inode_lock);
 788                 list_for_each_entry_safe(ii, n, head, i_dirty) {
 789                         list_del_init(&ii->i_dirty);
 790                         if (force) {
 791                                 if (unlikely(ii->i_bh)) {
 792                                         brelse(ii->i_bh);
 793                                         ii->i_bh = NULL;
 794                                 }
 795                         } else if (test_bit(NILFS_I_DIRTY, &ii->i_state)) {
 796                                 set_bit(NILFS_I_QUEUED, &ii->i_state);
 797                                 list_add_tail(&ii->i_dirty,
 798                                               &nilfs->ns_dirty_files);
 799                                 continue;
 800                         }
 801                         ivec[nv++] = ii;
 802                         if (nv == SC_N_INODEVEC)
 803                                 break;
 804                 }
 805                 spin_unlock(&nilfs->ns_inode_lock);
 806
 807                 for (pii = ivec; nv > 0; pii++, nv--)
 808                         iput(&(*pii)->vfs_inode);
 809         }
 810 }
 811
 812 static void nilfs_iput_work_func(struct work_struct *work)
 813 {
 814         struct nilfs_sc_info *sci = container_of(work, struct nilfs_sc_info,
 815                                                  sc_iput_work);
 816         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 817
 818         nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 0);
 819 }
 820
 821 static int nilfs_test_metadata_dirty(struct the_nilfs *nilfs,
 822                                      struct nilfs_root *root)
 823 {
 824         int ret = 0;
 825
 826         if (nilfs_mdt_fetch_dirty(root->ifile))
 827                 ret++;
 828         if (nilfs_mdt_fetch_dirty(nilfs->ns_cpfile))
 829                 ret++;
 830         if (nilfs_mdt_fetch_dirty(nilfs->ns_sufile))
 831                 ret++;
 832         if ((ret || nilfs_doing_gc()) && nilfs_mdt_fetch_dirty(nilfs->ns_dat))
 833                 ret++;
 834         return ret;
 835 }
 836
 837 static int nilfs_segctor_clean(struct nilfs_sc_info *sci)
 838 {
 839         return list_empty(&sci->sc_dirty_files) &&
 840                 !test_bit(NILFS_SC_DIRTY, &sci->sc_flags) &&
 841                 sci->sc_nfreesegs == 0 &&
 842                 (!nilfs_doing_gc() || list_empty(&sci->sc_gc_inodes));
 843 }
 844
 845 static int nilfs_segctor_confirm(struct nilfs_sc_info *sci)
 846 {
 847         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 848         int ret = 0;
 849
 850         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
 851                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
 852
 853         spin_lock(&nilfs->ns_inode_lock);
 854         if (list_empty(&nilfs->ns_dirty_files) && nilfs_segctor_clean(sci))
 855                 ret++;
 856
 857         spin_unlock(&nilfs->ns_inode_lock);
 858         return ret;
 859 }
 860
 861 static void nilfs_segctor_clear_metadata_dirty(struct nilfs_sc_info *sci)
 862 {
 863         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 864
 865         nilfs_mdt_clear_dirty(sci->sc_root->ifile);
 866         nilfs_mdt_clear_dirty(nilfs->ns_cpfile);
 867         nilfs_mdt_clear_dirty(nilfs->ns_sufile);
 868         nilfs_mdt_clear_dirty(nilfs->ns_dat);
 869 }
 870
 871 static int nilfs_segctor_create_checkpoint(struct nilfs_sc_info *sci)
 872 {
 873         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 874         struct buffer_head *bh_cp;
 875         struct nilfs_checkpoint *raw_cp;
 876         int err;
 877
 878         /* XXX: this interface will be changed */
 879         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 1,
 880                                           &raw_cp, &bh_cp);
 881         if (likely(!err)) {
 882                 /*
 883                  * The following code is duplicated with cpfile.  But, it is
 884                  * needed to collect the checkpoint even if it was not newly
 885                  * created.
 886                  */
 887                 mark_buffer_dirty(bh_cp);
 888                 nilfs_mdt_mark_dirty(nilfs->ns_cpfile);
 889                 nilfs_cpfile_put_checkpoint(
 890                         nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 891         } else if (err == -EINVAL || err == -ENOENT) {
 892                 nilfs_error(sci->sc_super,
 893                             "checkpoint creation failed due to metadata corruption.");
 894                 err = -EIO;
 895         }
 896         return err;
 897 }
 898
 899 static int nilfs_segctor_fill_in_checkpoint(struct nilfs_sc_info *sci)
 900 {
 901         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
 902         struct buffer_head *bh_cp;
 903         struct nilfs_checkpoint *raw_cp;
 904         int err;
 905
 906         err = nilfs_cpfile_get_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, 0,
 907                                           &raw_cp, &bh_cp);
 908         if (unlikely(err)) {
 909                 if (err == -EINVAL || err == -ENOENT) {
 910                         nilfs_error(sci->sc_super,
 911                                     "checkpoint finalization failed due to metadata corruption.");
 912                         err = -EIO;
 913                 }
 914                 goto failed_ibh;
 915         }
 916         raw_cp->cp_snapshot_list.ssl_next = 0;
 917         raw_cp->cp_snapshot_list.ssl_prev = 0;
 918         raw_cp->cp_inodes_count =
 919                 cpu_to_le64(atomic64_read(&sci->sc_root->inodes_count));
 920         raw_cp->cp_blocks_count =
 921                 cpu_to_le64(atomic64_read(&sci->sc_root->blocks_count));
 922         raw_cp->cp_nblk_inc =
 923                 cpu_to_le64(sci->sc_nblk_inc + sci->sc_nblk_this_inc);
 924         raw_cp->cp_create = cpu_to_le64(sci->sc_seg_ctime);
 925         raw_cp->cp_cno = cpu_to_le64(nilfs->ns_cno);
 926
 927         if (test_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags))
 928                 nilfs_checkpoint_clear_minor(raw_cp);
 929         else
 930                 nilfs_checkpoint_set_minor(raw_cp);
 931
 932         nilfs_write_inode_common(sci->sc_root->ifile,
 933                                  &raw_cp->cp_ifile_inode, 1);
 934         nilfs_cpfile_put_checkpoint(nilfs->ns_cpfile, nilfs->ns_cno, bh_cp);
 935         return 0;
 936
 937  failed_ibh:
 938         return err;
 939 }
 940
 941 static void nilfs_fill_in_file_bmap(struct inode *ifile,
 942                                     struct nilfs_inode_info *ii)
 943
 944 {
 945         struct buffer_head *ibh;
 946         struct nilfs_inode *raw_inode;
 947
 948         if (test_bit(NILFS_I_BMAP, &ii->i_state)) {
 949                 ibh = ii->i_bh;
 950                 BUG_ON(!ibh);
 951                 raw_inode = nilfs_ifile_map_inode(ifile, ii->vfs_inode.i_ino,
 952                                                   ibh);
 953                 nilfs_bmap_write(ii->i_bmap, raw_inode);
 954                 nilfs_ifile_unmap_inode(ifile, ii->vfs_inode.i_ino, ibh);
 955         }
 956 }
 957
 958 static void nilfs_segctor_fill_in_file_bmap(struct nilfs_sc_info *sci)
 959 {
 960         struct nilfs_inode_info *ii;
 961
 962         list_for_each_entry(ii, &sci->sc_dirty_files, i_dirty) {
 963                 nilfs_fill_in_file_bmap(sci->sc_root->ifile, ii);
 964                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
 965         }
 966 }
 967
 968 static void nilfs_segctor_fill_in_super_root(struct nilfs_sc_info *sci,
 969                                              struct the_nilfs *nilfs)
 970 {
 971         struct buffer_head *bh_sr;
 972         struct nilfs_super_root *raw_sr;
 973         unsigned int isz, srsz;
 974
 975         bh_sr = NILFS_LAST_SEGBUF(&sci->sc_segbufs)->sb_super_root;
 976         raw_sr = (struct nilfs_super_root *)bh_sr->b_data;
 977         isz = nilfs->ns_inode_size;
 978         srsz = NILFS_SR_BYTES(isz);
 979
 980         raw_sr->sr_bytes = cpu_to_le16(srsz);
 981         raw_sr->sr_nongc_ctime
 982                 = cpu_to_le64(nilfs_doing_gc() ?
 983                               nilfs->ns_nongc_ctime : sci->sc_seg_ctime);
 984         raw_sr->sr_flags = 0;
 985
 986         nilfs_write_inode_common(nilfs->ns_dat, (void *)raw_sr +
 987                                  NILFS_SR_DAT_OFFSET(isz), 1);
 988         nilfs_write_inode_common(nilfs->ns_cpfile, (void *)raw_sr +
 989                                  NILFS_SR_CPFILE_OFFSET(isz), 1);
 990         nilfs_write_inode_common(nilfs->ns_sufile, (void *)raw_sr +
 991                                  NILFS_SR_SUFILE_OFFSET(isz), 1);
 992         memset((void *)raw_sr + srsz, 0, nilfs->ns_blocksize - srsz);
 993 }
 994
 995 static void nilfs_redirty_inodes(struct list_head *head)
 996 {
 997         struct nilfs_inode_info *ii;
 998
 999         list_for_each_entry(ii, head, i_dirty) {
1000                 if (test_bit(NILFS_I_COLLECTED, &ii->i_state))
1001                         clear_bit(NILFS_I_COLLECTED, &ii->i_state);
1002         }
1003 }
1004
1005 static void nilfs_drop_collected_inodes(struct list_head *head)
1006 {
1007         struct nilfs_inode_info *ii;
1008
1009         list_for_each_entry(ii, head, i_dirty) {
1010                 if (!test_and_clear_bit(NILFS_I_COLLECTED, &ii->i_state))
1011                         continue;
1012
1013                 clear_bit(NILFS_I_INODE_SYNC, &ii->i_state);
1014                 set_bit(NILFS_I_UPDATED, &ii->i_state);
1015         }
1016 }
1017
1018 static int nilfs_segctor_apply_buffers(struct nilfs_sc_info *sci,
1019                                        struct inode *inode,
1020                                        struct list_head *listp,
1021                                        int (*collect)(struct nilfs_sc_info *,
1022                                                       struct buffer_head *,
1023                                                       struct inode *))
1024 {
1025         struct buffer_head *bh, *n;
1026         int err = 0;
1027
1028         if (collect) {
1029                 list_for_each_entry_safe(bh, n, listp, b_assoc_buffers) {
1030                         list_del_init(&bh->b_assoc_buffers);
1031                         err = collect(sci, bh, inode);
1032                         brelse(bh);
1033                         if (unlikely(err))
1034                                 goto dispose_buffers;
1035                 }
1036                 return 0;
1037         }
1038
1039  dispose_buffers:
1040         while (!list_empty(listp)) {
1041                 bh = list_first_entry(listp, struct buffer_head,
1042                                       b_assoc_buffers);
1043                 list_del_init(&bh->b_assoc_buffers);
1044                 brelse(bh);
1045         }
1046         return err;
1047 }
1048
1049 static size_t nilfs_segctor_buffer_rest(struct nilfs_sc_info *sci)
1050 {
1051         /* Remaining number of blocks within segment buffer */
1052         return sci->sc_segbuf_nblocks -
1053                 (sci->sc_nblk_this_inc + sci->sc_curseg->sb_sum.nblocks);
1054 }
1055
1056 static int nilfs_segctor_scan_file(struct nilfs_sc_info *sci,
1057                                    struct inode *inode,
1058                                    const struct nilfs_sc_operations *sc_ops)
1059 {
1060         LIST_HEAD(data_buffers);
1061         LIST_HEAD(node_buffers);
1062         int err;
1063
1064         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1065                 size_t n, rest = nilfs_segctor_buffer_rest(sci);
1066
1067                 n = nilfs_lookup_dirty_data_buffers(
1068                         inode, &data_buffers, rest + 1, 0, LLONG_MAX);
1069                 if (n > rest) {
1070                         err = nilfs_segctor_apply_buffers(
1071                                 sci, inode, &data_buffers,
1072                                 sc_ops->collect_data);
1073                         BUG_ON(!err); /* always receive -E2BIG or true error */
1074                         goto break_or_fail;
1075                 }
1076         }
1077         nilfs_lookup_dirty_node_buffers(inode, &node_buffers);
1078
1079         if (!(sci->sc_stage.flags & NILFS_CF_NODE)) {
1080                 err = nilfs_segctor_apply_buffers(
1081                         sci, inode, &data_buffers, sc_ops->collect_data);
1082                 if (unlikely(err)) {
1083                         /* dispose node list */
1084                         nilfs_segctor_apply_buffers(
1085                                 sci, inode, &node_buffers, NULL);
1086                         goto break_or_fail;
1087                 }
1088                 sci->sc_stage.flags |= NILFS_CF_NODE;
1089         }
1090         /* Collect node */
1091         err = nilfs_segctor_apply_buffers(
1092                 sci, inode, &node_buffers, sc_ops->collect_node);
1093         if (unlikely(err))
1094                 goto break_or_fail;
1095
1096         nilfs_bmap_lookup_dirty_buffers(NILFS_I(inode)->i_bmap, &node_buffers);
1097         err = nilfs_segctor_apply_buffers(
1098                 sci, inode, &node_buffers, sc_ops->collect_bmap);
1099         if (unlikely(err))
1100                 goto break_or_fail;
1101
1102         nilfs_segctor_end_finfo(sci, inode);
1103         sci->sc_stage.flags &= ~NILFS_CF_NODE;
1104
1105  break_or_fail:
1106         return err;
1107 }
1108
1109 static int nilfs_segctor_scan_file_dsync(struct nilfs_sc_info *sci,
1110                                          struct inode *inode)
1111 {
1112         LIST_HEAD(data_buffers);
1113         size_t n, rest = nilfs_segctor_buffer_rest(sci);
1114         int err;
1115
1116         n = nilfs_lookup_dirty_data_buffers(inode, &data_buffers, rest + 1,
1117                                             sci->sc_dsync_start,
1118                                             sci->sc_dsync_end);
1119
1120         err = nilfs_segctor_apply_buffers(sci, inode, &data_buffers,
1121                                           nilfs_collect_file_data);
1122         if (!err) {
1123                 nilfs_segctor_end_finfo(sci, inode);
1124                 BUG_ON(n > rest);
1125                 /* always receive -E2BIG or true error if n > rest */
1126         }
1127         return err;
1128 }
1129
1130 static int nilfs_segctor_collect_blocks(struct nilfs_sc_info *sci, int mode)
1131 {
1132         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1133         struct list_head *head;
1134         struct nilfs_inode_info *ii;
1135         size_t ndone;
1136         int err = 0;
1137
1138         switch (nilfs_sc_cstage_get(sci)) {
1139         case NILFS_ST_INIT:
1140                 /* Pre-processes */
1141                 sci->sc_stage.flags = 0;
1142
1143                 if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags)) {
1144                         sci->sc_nblk_inc = 0;
1145                         sci->sc_curseg->sb_sum.flags = NILFS_SS_LOGBGN;
1146                         if (mode == SC_LSEG_DSYNC) {
1147                                 nilfs_sc_cstage_set(sci, NILFS_ST_DSYNC);
1148                                 goto dsync_mode;
1149                         }
1150                 }
1151
1152                 sci->sc_stage.dirty_file_ptr = NULL;
1153                 sci->sc_stage.gc_inode_ptr = NULL;
1154                 if (mode == SC_FLUSH_DAT) {
1155                         nilfs_sc_cstage_set(sci, NILFS_ST_DAT);
1156                         goto dat_stage;
1157                 }
1158                 nilfs_sc_cstage_inc(sci);  /* Fall through */
1159         case NILFS_ST_GC:
1160                 if (nilfs_doing_gc()) {
1161                         head = &sci->sc_gc_inodes;
1162                         ii = list_prepare_entry(sci->sc_stage.gc_inode_ptr,
1163                                                 head, i_dirty);
1164                         list_for_each_entry_continue(ii, head, i_dirty) {
1165                                 err = nilfs_segctor_scan_file(
1166                                         sci, &ii->vfs_inode,
1167                                         &nilfs_sc_file_ops);
1168                                 if (unlikely(err)) {
1169                                         sci->sc_stage.gc_inode_ptr = list_entry(
1170                                                 ii->i_dirty.prev,
1171                                                 struct nilfs_inode_info,
1172                                                 i_dirty);
1173                                         goto break_or_fail;
1174                                 }
1175                                 set_bit(NILFS_I_COLLECTED, &ii->i_state);
1176                         }
1177                         sci->sc_stage.gc_inode_ptr = NULL;
1178                 }
1179                 nilfs_sc_cstage_inc(sci);  /* Fall through */
1180         case NILFS_ST_FILE:
1181                 head = &sci->sc_dirty_files;
1182                 ii = list_prepare_entry(sci->sc_stage.dirty_file_ptr, head,
1183                                         i_dirty);
1184                 list_for_each_entry_continue(ii, head, i_dirty) {
1185                         clear_bit(NILFS_I_DIRTY, &ii->i_state);
1186
1187                         err = nilfs_segctor_scan_file(sci, &ii->vfs_inode,
1188                                                       &nilfs_sc_file_ops);
1189                         if (unlikely(err)) {
1190                                 sci->sc_stage.dirty_file_ptr =
1191                                         list_entry(ii->i_dirty.prev,
1192                                                    struct nilfs_inode_info,
1193                                                    i_dirty);
1194                                 goto break_or_fail;
1195                         }
1196                         /* sci->sc_stage.dirty_file_ptr = NILFS_I(inode); */
1197                         /* XXX: required ? */
1198                 }
1199                 sci->sc_stage.dirty_file_ptr = NULL;
1200                 if (mode == SC_FLUSH_FILE) {
1201                         nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1202                         return 0;
1203                 }
1204                 nilfs_sc_cstage_inc(sci);
1205                 sci->sc_stage.flags |= NILFS_CF_IFILE_STARTED;
1206                 /* Fall through */
1207         case NILFS_ST_IFILE:
1208                 err = nilfs_segctor_scan_file(sci, sci->sc_root->ifile,
1209                                               &nilfs_sc_file_ops);
1210                 if (unlikely(err))
1211                         break;
1212                 nilfs_sc_cstage_inc(sci);
1213                 /* Creating a checkpoint */
1214                 err = nilfs_segctor_create_checkpoint(sci);
1215                 if (unlikely(err))
1216                         break;
1217                 /* Fall through */
1218         case NILFS_ST_CPFILE:
1219                 err = nilfs_segctor_scan_file(sci, nilfs->ns_cpfile,
1220                                               &nilfs_sc_file_ops);
1221                 if (unlikely(err))
1222                         break;
1223                 nilfs_sc_cstage_inc(sci);  /* Fall through */
1224         case NILFS_ST_SUFILE:
1225                 err = nilfs_sufile_freev(nilfs->ns_sufile, sci->sc_freesegs,
1226                                          sci->sc_nfreesegs, &ndone);
1227                 if (unlikely(err)) {
1228                         nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1229                                                   sci->sc_freesegs, ndone,
1230                                                   NULL);
1231                         break;
1232                 }
1233                 sci->sc_stage.flags |= NILFS_CF_SUFREED;
1234
1235                 err = nilfs_segctor_scan_file(sci, nilfs->ns_sufile,
1236                                               &nilfs_sc_file_ops);
1237                 if (unlikely(err))
1238                         break;
1239                 nilfs_sc_cstage_inc(sci);  /* Fall through */
1240         case NILFS_ST_DAT:
1241  dat_stage:
1242                 err = nilfs_segctor_scan_file(sci, nilfs->ns_dat,
1243                                               &nilfs_sc_dat_ops);
1244                 if (unlikely(err))
1245                         break;
1246                 if (mode == SC_FLUSH_DAT) {
1247                         nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1248                         return 0;
1249                 }
1250                 nilfs_sc_cstage_inc(sci);  /* Fall through */
1251         case NILFS_ST_SR:
1252                 if (mode == SC_LSEG_SR) {
1253                         /* Appending a super root */
1254                         err = nilfs_segctor_add_super_root(sci);
1255                         if (unlikely(err))
1256                                 break;
1257                 }
1258                 /* End of a logical segment */
1259                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1260                 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1261                 return 0;
1262         case NILFS_ST_DSYNC:
1263  dsync_mode:
1264                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_SYNDT;
1265                 ii = sci->sc_dsync_inode;
1266                 if (!test_bit(NILFS_I_BUSY, &ii->i_state))
1267                         break;
1268
1269                 err = nilfs_segctor_scan_file_dsync(sci, &ii->vfs_inode);
1270                 if (unlikely(err))
1271                         break;
1272                 sci->sc_curseg->sb_sum.flags |= NILFS_SS_LOGEND;
1273                 nilfs_sc_cstage_set(sci, NILFS_ST_DONE);
1274                 return 0;
1275         case NILFS_ST_DONE:
1276                 return 0;
1277         default:
1278                 BUG();
1279         }
1280
1281  break_or_fail:
1282         return err;
1283 }
1284
1285 /**
1286  * nilfs_segctor_begin_construction - setup segment buffer to make a new log
1287  * @sci: nilfs_sc_info
1288  * @nilfs: nilfs object
1289  */
1290 static int nilfs_segctor_begin_construction(struct nilfs_sc_info *sci,
1291                                             struct the_nilfs *nilfs)
1292 {
1293         struct nilfs_segment_buffer *segbuf, *prev;
1294         __u64 nextnum;
1295         int err, alloc = 0;
1296
1297         segbuf = nilfs_segbuf_new(sci->sc_super);
1298         if (unlikely(!segbuf))
1299                 return -ENOMEM;
1300
1301         if (list_empty(&sci->sc_write_logs)) {
1302                 nilfs_segbuf_map(segbuf, nilfs->ns_segnum,
1303                                  nilfs->ns_pseg_offset, nilfs);
1304                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1305                         nilfs_shift_to_next_segment(nilfs);
1306                         nilfs_segbuf_map(segbuf, nilfs->ns_segnum, 0, nilfs);
1307                 }
1308
1309                 segbuf->sb_sum.seg_seq = nilfs->ns_seg_seq;
1310                 nextnum = nilfs->ns_nextnum;
1311
1312                 if (nilfs->ns_segnum == nilfs->ns_nextnum)
1313                         /* Start from the head of a new full segment */
1314                         alloc++;
1315         } else {
1316                 /* Continue logs */
1317                 prev = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1318                 nilfs_segbuf_map_cont(segbuf, prev);
1319                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq;
1320                 nextnum = prev->sb_nextnum;
1321
1322                 if (segbuf->sb_rest_blocks < NILFS_PSEG_MIN_BLOCKS) {
1323                         nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1324                         segbuf->sb_sum.seg_seq++;
1325                         alloc++;
1326                 }
1327         }
1328
1329         err = nilfs_sufile_mark_dirty(nilfs->ns_sufile, segbuf->sb_segnum);
1330         if (err)
1331                 goto failed;
1332
1333         if (alloc) {
1334                 err = nilfs_sufile_alloc(nilfs->ns_sufile, &nextnum);
1335                 if (err)
1336                         goto failed;
1337         }
1338         nilfs_segbuf_set_next_segnum(segbuf, nextnum, nilfs);
1339
1340         BUG_ON(!list_empty(&sci->sc_segbufs));
1341         list_add_tail(&segbuf->sb_list, &sci->sc_segbufs);
1342         sci->sc_segbuf_nblocks = segbuf->sb_rest_blocks;
1343         return 0;
1344
1345  failed:
1346         nilfs_segbuf_free(segbuf);
1347         return err;
1348 }
1349
1350 static int nilfs_segctor_extend_segments(struct nilfs_sc_info *sci,
1351                                          struct the_nilfs *nilfs, int nadd)
1352 {
1353         struct nilfs_segment_buffer *segbuf, *prev;
1354         struct inode *sufile = nilfs->ns_sufile;
1355         __u64 nextnextnum;
1356         LIST_HEAD(list);
1357         int err, ret, i;
1358
1359         prev = NILFS_LAST_SEGBUF(&sci->sc_segbufs);
1360         /*
1361          * Since the segment specified with nextnum might be allocated during
1362          * the previous construction, the buffer including its segusage may
1363          * not be dirty.  The following call ensures that the buffer is dirty
1364          * and will pin the buffer on memory until the sufile is written.
1365          */
1366         err = nilfs_sufile_mark_dirty(sufile, prev->sb_nextnum);
1367         if (unlikely(err))
1368                 return err;
1369
1370         for (i = 0; i < nadd; i++) {
1371                 /* extend segment info */
1372                 err = -ENOMEM;
1373                 segbuf = nilfs_segbuf_new(sci->sc_super);
1374                 if (unlikely(!segbuf))
1375                         goto failed;
1376
1377                 /* map this buffer to region of segment on-disk */
1378                 nilfs_segbuf_map(segbuf, prev->sb_nextnum, 0, nilfs);
1379                 sci->sc_segbuf_nblocks += segbuf->sb_rest_blocks;
1380
1381                 /* allocate the next next full segment */
1382                 err = nilfs_sufile_alloc(sufile, &nextnextnum);
1383                 if (unlikely(err))
1384                         goto failed_segbuf;
1385
1386                 segbuf->sb_sum.seg_seq = prev->sb_sum.seg_seq + 1;
1387                 nilfs_segbuf_set_next_segnum(segbuf, nextnextnum, nilfs);
1388
1389                 list_add_tail(&segbuf->sb_list, &list);
1390                 prev = segbuf;
1391         }
1392         list_splice_tail(&list, &sci->sc_segbufs);
1393         return 0;
1394
1395  failed_segbuf:
1396         nilfs_segbuf_free(segbuf);
1397  failed:
1398         list_for_each_entry(segbuf, &list, sb_list) {
1399                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1400                 WARN_ON(ret); /* never fails */
1401         }
1402         nilfs_destroy_logs(&list);
1403         return err;
1404 }
1405
1406 static void nilfs_free_incomplete_logs(struct list_head *logs,
1407                                        struct the_nilfs *nilfs)
1408 {
1409         struct nilfs_segment_buffer *segbuf, *prev;
1410         struct inode *sufile = nilfs->ns_sufile;
1411         int ret;
1412
1413         segbuf = NILFS_FIRST_SEGBUF(logs);
1414         if (nilfs->ns_nextnum != segbuf->sb_nextnum) {
1415                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1416                 WARN_ON(ret); /* never fails */
1417         }
1418         if (atomic_read(&segbuf->sb_err)) {
1419                 /* Case 1: The first segment failed */
1420                 if (segbuf->sb_pseg_start != segbuf->sb_fseg_start)
1421                         /*
1422                          * Case 1a:  Partial segment appended into an existing
1423                          * segment
1424                          */
1425                         nilfs_terminate_segment(nilfs, segbuf->sb_fseg_start,
1426                                                 segbuf->sb_fseg_end);
1427                 else /* Case 1b:  New full segment */
1428                         set_nilfs_discontinued(nilfs);
1429         }
1430
1431         prev = segbuf;
1432         list_for_each_entry_continue(segbuf, logs, sb_list) {
1433                 if (prev->sb_nextnum != segbuf->sb_nextnum) {
1434                         ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1435                         WARN_ON(ret); /* never fails */
1436                 }
1437                 if (atomic_read(&segbuf->sb_err) &&
1438                     segbuf->sb_segnum != nilfs->ns_nextnum)
1439                         /* Case 2: extended segment (!= next) failed */
1440                         nilfs_sufile_set_error(sufile, segbuf->sb_segnum);
1441                 prev = segbuf;
1442         }
1443 }
1444
1445 static void nilfs_segctor_update_segusage(struct nilfs_sc_info *sci,
1446                                           struct inode *sufile)
1447 {
1448         struct nilfs_segment_buffer *segbuf;
1449         unsigned long live_blocks;
1450         int ret;
1451
1452         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1453                 live_blocks = segbuf->sb_sum.nblocks +
1454                         (segbuf->sb_pseg_start - segbuf->sb_fseg_start);
1455                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1456                                                      live_blocks,
1457                                                      sci->sc_seg_ctime);
1458                 WARN_ON(ret); /* always succeed because the segusage is dirty */
1459         }
1460 }
1461
1462 static void nilfs_cancel_segusage(struct list_head *logs, struct inode *sufile)
1463 {
1464         struct nilfs_segment_buffer *segbuf;
1465         int ret;
1466
1467         segbuf = NILFS_FIRST_SEGBUF(logs);
1468         ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1469                                              segbuf->sb_pseg_start -
1470                                              segbuf->sb_fseg_start, 0);
1471         WARN_ON(ret); /* always succeed because the segusage is dirty */
1472
1473         list_for_each_entry_continue(segbuf, logs, sb_list) {
1474                 ret = nilfs_sufile_set_segment_usage(sufile, segbuf->sb_segnum,
1475                                                      0, 0);
1476                 WARN_ON(ret); /* always succeed */
1477         }
1478 }
1479
1480 static void nilfs_segctor_truncate_segments(struct nilfs_sc_info *sci,
1481                                             struct nilfs_segment_buffer *last,
1482                                             struct inode *sufile)
1483 {
1484         struct nilfs_segment_buffer *segbuf = last;
1485         int ret;
1486
1487         list_for_each_entry_continue(segbuf, &sci->sc_segbufs, sb_list) {
1488                 sci->sc_segbuf_nblocks -= segbuf->sb_rest_blocks;
1489                 ret = nilfs_sufile_free(sufile, segbuf->sb_nextnum);
1490                 WARN_ON(ret);
1491         }
1492         nilfs_truncate_logs(&sci->sc_segbufs, last);
1493 }
1494
1495
1496 static int nilfs_segctor_collect(struct nilfs_sc_info *sci,
1497                                  struct the_nilfs *nilfs, int mode)
1498 {
1499         struct nilfs_cstage prev_stage = sci->sc_stage;
1500         int err, nadd = 1;
1501
1502         /* Collection retry loop */
1503         for (;;) {
1504                 sci->sc_nblk_this_inc = 0;
1505                 sci->sc_curseg = NILFS_FIRST_SEGBUF(&sci->sc_segbufs);
1506
1507                 err = nilfs_segctor_reset_segment_buffer(sci);
1508                 if (unlikely(err))
1509                         goto failed;
1510
1511                 err = nilfs_segctor_collect_blocks(sci, mode);
1512                 sci->sc_nblk_this_inc += sci->sc_curseg->sb_sum.nblocks;
1513                 if (!err)
1514                         break;
1515
1516                 if (unlikely(err != -E2BIG))
1517                         goto failed;
1518
1519                 /* The current segment is filled up */
1520                 if (mode != SC_LSEG_SR ||
1521                     nilfs_sc_cstage_get(sci) < NILFS_ST_CPFILE)
1522                         break;
1523
1524                 nilfs_clear_logs(&sci->sc_segbufs);
1525
1526                 if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1527                         err = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1528                                                         sci->sc_freesegs,
1529                                                         sci->sc_nfreesegs,
1530                                                         NULL);
1531                         WARN_ON(err); /* do not happen */
1532                         sci->sc_stage.flags &= ~NILFS_CF_SUFREED;
1533                 }
1534
1535                 err = nilfs_segctor_extend_segments(sci, nilfs, nadd);
1536                 if (unlikely(err))
1537                         return err;
1538
1539                 nadd = min_t(int, nadd << 1, SC_MAX_SEGDELTA);
1540                 sci->sc_stage = prev_stage;
1541         }
1542         nilfs_segctor_truncate_segments(sci, sci->sc_curseg, nilfs->ns_sufile);
1543         return 0;
1544
1545  failed:
1546         return err;
1547 }
1548
1549 static void nilfs_list_replace_buffer(struct buffer_head *old_bh,
1550                                       struct buffer_head *new_bh)
1551 {
1552         BUG_ON(!list_empty(&new_bh->b_assoc_buffers));
1553
1554         list_replace_init(&old_bh->b_assoc_buffers, &new_bh->b_assoc_buffers);
1555         /* The caller must release old_bh */
1556 }
1557
1558 static int
1559 nilfs_segctor_update_payload_blocknr(struct nilfs_sc_info *sci,
1560                                      struct nilfs_segment_buffer *segbuf,
1561                                      int mode)
1562 {
1563         struct inode *inode = NULL;
1564         sector_t blocknr;
1565         unsigned long nfinfo = segbuf->sb_sum.nfinfo;
1566         unsigned long nblocks = 0, ndatablk = 0;
1567         const struct nilfs_sc_operations *sc_op = NULL;
1568         struct nilfs_segsum_pointer ssp;
1569         struct nilfs_finfo *finfo = NULL;
1570         union nilfs_binfo binfo;
1571         struct buffer_head *bh, *bh_org;
1572         ino_t ino = 0;
1573         int err = 0;
1574
1575         if (!nfinfo)
1576                 goto out;
1577
1578         blocknr = segbuf->sb_pseg_start + segbuf->sb_sum.nsumblk;
1579         ssp.bh = NILFS_SEGBUF_FIRST_BH(&segbuf->sb_segsum_buffers);
1580         ssp.offset = sizeof(struct nilfs_segment_summary);
1581
1582         list_for_each_entry(bh, &segbuf->sb_payload_buffers, b_assoc_buffers) {
1583                 if (bh == segbuf->sb_super_root)
1584                         break;
1585                 if (!finfo) {
1586                         finfo = nilfs_segctor_map_segsum_entry(
1587                                 sci, &ssp, sizeof(*finfo));
1588                         ino = le64_to_cpu(finfo->fi_ino);
1589                         nblocks = le32_to_cpu(finfo->fi_nblocks);
1590                         ndatablk = le32_to_cpu(finfo->fi_ndatablk);
1591
1592                         inode = bh->b_page->mapping->host;
1593
1594                         if (mode == SC_LSEG_DSYNC)
1595                                 sc_op = &nilfs_sc_dsync_ops;
1596                         else if (ino == NILFS_DAT_INO)
1597                                 sc_op = &nilfs_sc_dat_ops;
1598                         else /* file blocks */
1599                                 sc_op = &nilfs_sc_file_ops;
1600                 }
1601                 bh_org = bh;
1602                 get_bh(bh_org);
1603                 err = nilfs_bmap_assign(NILFS_I(inode)->i_bmap, &bh, blocknr,
1604                                         &binfo);
1605                 if (bh != bh_org)
1606                         nilfs_list_replace_buffer(bh_org, bh);
1607                 brelse(bh_org);
1608                 if (unlikely(err))
1609                         goto failed_bmap;
1610
1611                 if (ndatablk > 0)
1612                         sc_op->write_data_binfo(sci, &ssp, &binfo);
1613                 else
1614                         sc_op->write_node_binfo(sci, &ssp, &binfo);
1615
1616                 blocknr++;
1617                 if (--nblocks == 0) {
1618                         finfo = NULL;
1619                         if (--nfinfo == 0)
1620                                 break;
1621                 } else if (ndatablk > 0)
1622                         ndatablk--;
1623         }
1624  out:
1625         return 0;
1626
1627  failed_bmap:
1628         return err;
1629 }
1630
1631 static int nilfs_segctor_assign(struct nilfs_sc_info *sci, int mode)
1632 {
1633         struct nilfs_segment_buffer *segbuf;
1634         int err;
1635
1636         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1637                 err = nilfs_segctor_update_payload_blocknr(sci, segbuf, mode);
1638                 if (unlikely(err))
1639                         return err;
1640                 nilfs_segbuf_fill_in_segsum(segbuf);
1641         }
1642         return 0;
1643 }
1644
1645 static void nilfs_begin_page_io(struct page *page)
1646 {
1647         if (!page || PageWriteback(page))
1648                 /*
1649                  * For split b-tree node pages, this function may be called
1650                  * twice.  We ignore the 2nd or later calls by this check.
1651                  */
1652                 return;
1653
1654         lock_page(page);
1655         clear_page_dirty_for_io(page);
1656         set_page_writeback(page);
1657         unlock_page(page);
1658 }
1659
1660 static void nilfs_segctor_prepare_write(struct nilfs_sc_info *sci)
1661 {
1662         struct nilfs_segment_buffer *segbuf;
1663         struct page *bd_page = NULL, *fs_page = NULL;
1664
1665         list_for_each_entry(segbuf, &sci->sc_segbufs, sb_list) {
1666                 struct buffer_head *bh;
1667
1668                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1669                                     b_assoc_buffers) {
1670                         if (bh->b_page != bd_page) {
1671                                 if (bd_page) {
1672                                         lock_page(bd_page);
1673                                         clear_page_dirty_for_io(bd_page);
1674                                         set_page_writeback(bd_page);
1675                                         unlock_page(bd_page);
1676                                 }
1677                                 bd_page = bh->b_page;
1678                         }
1679                 }
1680
1681                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1682                                     b_assoc_buffers) {
1683                         set_buffer_async_write(bh);
1684                         if (bh == segbuf->sb_super_root) {
1685                                 if (bh->b_page != bd_page) {
1686                                         lock_page(bd_page);
1687                                         clear_page_dirty_for_io(bd_page);
1688                                         set_page_writeback(bd_page);
1689                                         unlock_page(bd_page);
1690                                         bd_page = bh->b_page;
1691                                 }
1692                                 break;
1693                         }
1694                         if (bh->b_page != fs_page) {
1695                                 nilfs_begin_page_io(fs_page);
1696                                 fs_page = bh->b_page;
1697                         }
1698                 }
1699         }
1700         if (bd_page) {
1701                 lock_page(bd_page);
1702                 clear_page_dirty_for_io(bd_page);
1703                 set_page_writeback(bd_page);
1704                 unlock_page(bd_page);
1705         }
1706         nilfs_begin_page_io(fs_page);
1707 }
1708
1709 static int nilfs_segctor_write(struct nilfs_sc_info *sci,
1710                                struct the_nilfs *nilfs)
1711 {
1712         int ret;
1713
1714         ret = nilfs_write_logs(&sci->sc_segbufs, nilfs);
1715         list_splice_tail_init(&sci->sc_segbufs, &sci->sc_write_logs);
1716         return ret;
1717 }
1718
1719 static void nilfs_end_page_io(struct page *page, int err)
1720 {
1721         if (!page)
1722                 return;
1723
1724         if (buffer_nilfs_node(page_buffers(page)) && !PageWriteback(page)) {
1725                 /*
1726                  * For b-tree node pages, this function may be called twice
1727                  * or more because they might be split in a segment.
1728                  */
1729                 if (PageDirty(page)) {
1730                         /*
1731                          * For pages holding split b-tree node buffers, dirty
1732                          * flag on the buffers may be cleared discretely.
1733                          * In that case, the page is once redirtied for
1734                          * remaining buffers, and it must be cancelled if
1735                          * all the buffers get cleaned later.
1736                          */
1737                         lock_page(page);
1738                         if (nilfs_page_buffers_clean(page))
1739                                 __nilfs_clear_page_dirty(page);
1740                         unlock_page(page);
1741                 }
1742                 return;
1743         }
1744
1745         if (!err) {
1746                 if (!nilfs_page_buffers_clean(page))
1747                         __set_page_dirty_nobuffers(page);
1748                 ClearPageError(page);
1749         } else {
1750                 __set_page_dirty_nobuffers(page);
1751                 SetPageError(page);
1752         }
1753
1754         end_page_writeback(page);
1755 }
1756
1757 static void nilfs_abort_logs(struct list_head *logs, int err)
1758 {
1759         struct nilfs_segment_buffer *segbuf;
1760         struct page *bd_page = NULL, *fs_page = NULL;
1761         struct buffer_head *bh;
1762
1763         if (list_empty(logs))
1764                 return;
1765
1766         list_for_each_entry(segbuf, logs, sb_list) {
1767                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1768                                     b_assoc_buffers) {
1769                         if (bh->b_page != bd_page) {
1770                                 if (bd_page)
1771                                         end_page_writeback(bd_page);
1772                                 bd_page = bh->b_page;
1773                         }
1774                 }
1775
1776                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1777                                     b_assoc_buffers) {
1778                         clear_buffer_async_write(bh);
1779                         if (bh == segbuf->sb_super_root) {
1780                                 if (bh->b_page != bd_page) {
1781                                         end_page_writeback(bd_page);
1782                                         bd_page = bh->b_page;
1783                                 }
1784                                 break;
1785                         }
1786                         if (bh->b_page != fs_page) {
1787                                 nilfs_end_page_io(fs_page, err);
1788                                 fs_page = bh->b_page;
1789                         }
1790                 }
1791         }
1792         if (bd_page)
1793                 end_page_writeback(bd_page);
1794
1795         nilfs_end_page_io(fs_page, err);
1796 }
1797
1798 static void nilfs_segctor_abort_construction(struct nilfs_sc_info *sci,
1799                                              struct the_nilfs *nilfs, int err)
1800 {
1801         LIST_HEAD(logs);
1802         int ret;
1803
1804         list_splice_tail_init(&sci->sc_write_logs, &logs);
1805         ret = nilfs_wait_on_logs(&logs);
1806         nilfs_abort_logs(&logs, ret ? : err);
1807
1808         list_splice_tail_init(&sci->sc_segbufs, &logs);
1809         nilfs_cancel_segusage(&logs, nilfs->ns_sufile);
1810         nilfs_free_incomplete_logs(&logs, nilfs);
1811
1812         if (sci->sc_stage.flags & NILFS_CF_SUFREED) {
1813                 ret = nilfs_sufile_cancel_freev(nilfs->ns_sufile,
1814                                                 sci->sc_freesegs,
1815                                                 sci->sc_nfreesegs,
1816                                                 NULL);
1817                 WARN_ON(ret); /* do not happen */
1818         }
1819
1820         nilfs_destroy_logs(&logs);
1821 }
1822
1823 static void nilfs_set_next_segment(struct the_nilfs *nilfs,
1824                                    struct nilfs_segment_buffer *segbuf)
1825 {
1826         nilfs->ns_segnum = segbuf->sb_segnum;
1827         nilfs->ns_nextnum = segbuf->sb_nextnum;
1828         nilfs->ns_pseg_offset = segbuf->sb_pseg_start - segbuf->sb_fseg_start
1829                 + segbuf->sb_sum.nblocks;
1830         nilfs->ns_seg_seq = segbuf->sb_sum.seg_seq;
1831         nilfs->ns_ctime = segbuf->sb_sum.ctime;
1832 }
1833
1834 static void nilfs_segctor_complete_write(struct nilfs_sc_info *sci)
1835 {
1836         struct nilfs_segment_buffer *segbuf;
1837         struct page *bd_page = NULL, *fs_page = NULL;
1838         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
1839         int update_sr = false;
1840
1841         list_for_each_entry(segbuf, &sci->sc_write_logs, sb_list) {
1842                 struct buffer_head *bh;
1843
1844                 list_for_each_entry(bh, &segbuf->sb_segsum_buffers,
1845                                     b_assoc_buffers) {
1846                         set_buffer_uptodate(bh);
1847                         clear_buffer_dirty(bh);
1848                         if (bh->b_page != bd_page) {
1849                                 if (bd_page)
1850                                         end_page_writeback(bd_page);
1851                                 bd_page = bh->b_page;
1852                         }
1853                 }
1854                 /*
1855                  * We assume that the buffers which belong to the same page
1856                  * continue over the buffer list.
1857                  * Under this assumption, the last BHs of pages is
1858                  * identifiable by the discontinuity of bh->b_page
1859                  * (page != fs_page).
1860                  *
1861                  * For B-tree node blocks, however, this assumption is not
1862                  * guaranteed.  The cleanup code of B-tree node pages needs
1863                  * special care.
1864                  */
1865                 list_for_each_entry(bh, &segbuf->sb_payload_buffers,
1866                                     b_assoc_buffers) {
1867                         const unsigned long set_bits = BIT(BH_Uptodate);
1868                         const unsigned long clear_bits =
1869                                 (BIT(BH_Dirty) | BIT(BH_Async_Write) |
1870                                  BIT(BH_Delay) | BIT(BH_NILFS_Volatile) |
1871                                  BIT(BH_NILFS_Redirected));
1872
1873                         set_mask_bits(&bh->b_state, clear_bits, set_bits);
1874                         if (bh == segbuf->sb_super_root) {
1875                                 if (bh->b_page != bd_page) {
1876                                         end_page_writeback(bd_page);
1877                                         bd_page = bh->b_page;
1878                                 }
1879                                 update_sr = true;
1880                                 break;
1881                         }
1882                         if (bh->b_page != fs_page) {
1883                                 nilfs_end_page_io(fs_page, 0);
1884                                 fs_page = bh->b_page;
1885                         }
1886                 }
1887
1888                 if (!nilfs_segbuf_simplex(segbuf)) {
1889                         if (segbuf->sb_sum.flags & NILFS_SS_LOGBGN) {
1890                                 set_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1891                                 sci->sc_lseg_stime = jiffies;
1892                         }
1893                         if (segbuf->sb_sum.flags & NILFS_SS_LOGEND)
1894                                 clear_bit(NILFS_SC_UNCLOSED, &sci->sc_flags);
1895                 }
1896         }
1897         /*
1898          * Since pages may continue over multiple segment buffers,
1899          * end of the last page must be checked outside of the loop.
1900          */
1901         if (bd_page)
1902                 end_page_writeback(bd_page);
1903
1904         nilfs_end_page_io(fs_page, 0);
1905
1906         nilfs_drop_collected_inodes(&sci->sc_dirty_files);
1907
1908         if (nilfs_doing_gc())
1909                 nilfs_drop_collected_inodes(&sci->sc_gc_inodes);
1910         else
1911                 nilfs->ns_nongc_ctime = sci->sc_seg_ctime;
1912
1913         sci->sc_nblk_inc += sci->sc_nblk_this_inc;
1914
1915         segbuf = NILFS_LAST_SEGBUF(&sci->sc_write_logs);
1916         nilfs_set_next_segment(nilfs, segbuf);
1917
1918         if (update_sr) {
1919                 nilfs->ns_flushed_device = 0;
1920                 nilfs_set_last_segment(nilfs, segbuf->sb_pseg_start,
1921                                        segbuf->sb_sum.seg_seq, nilfs->ns_cno++);
1922
1923                 clear_bit(NILFS_SC_HAVE_DELTA, &sci->sc_flags);
1924                 clear_bit(NILFS_SC_DIRTY, &sci->sc_flags);
1925                 set_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1926                 nilfs_segctor_clear_metadata_dirty(sci);
1927         } else
1928                 clear_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags);
1929 }
1930
1931 static int nilfs_segctor_wait(struct nilfs_sc_info *sci)
1932 {
1933         int ret;
1934
1935         ret = nilfs_wait_on_logs(&sci->sc_write_logs);
1936         if (!ret) {
1937                 nilfs_segctor_complete_write(sci);
1938                 nilfs_destroy_logs(&sci->sc_write_logs);
1939         }
1940         return ret;
1941 }
1942
1943 static int nilfs_segctor_collect_dirty_files(struct nilfs_sc_info *sci,
1944                                              struct the_nilfs *nilfs)
1945 {
1946         struct nilfs_inode_info *ii, *n;
1947         struct inode *ifile = sci->sc_root->ifile;
1948
1949         spin_lock(&nilfs->ns_inode_lock);
1950  retry:
1951         list_for_each_entry_safe(ii, n, &nilfs->ns_dirty_files, i_dirty) {
1952                 if (!ii->i_bh) {
1953                         struct buffer_head *ibh;
1954                         int err;
1955
1956                         spin_unlock(&nilfs->ns_inode_lock);
1957                         err = nilfs_ifile_get_inode_block(
1958                                 ifile, ii->vfs_inode.i_ino, &ibh);
1959                         if (unlikely(err)) {
1960                                 nilfs_msg(sci->sc_super, KERN_WARNING,
1961                                           "log writer: error %d getting inode block (ino=%lu)",
1962                                           err, ii->vfs_inode.i_ino);
1963                                 return err;
1964                         }
1965                         spin_lock(&nilfs->ns_inode_lock);
1966                         if (likely(!ii->i_bh))
1967                                 ii->i_bh = ibh;
1968                         else
1969                                 brelse(ibh);
1970                         goto retry;
1971                 }
1972
1973                 // Always redirty the buffer to avoid race condition
1974                 mark_buffer_dirty(ii->i_bh);
1975                 nilfs_mdt_mark_dirty(ifile);
1976
1977                 clear_bit(NILFS_I_QUEUED, &ii->i_state);
1978                 set_bit(NILFS_I_BUSY, &ii->i_state);
1979                 list_move_tail(&ii->i_dirty, &sci->sc_dirty_files);
1980         }
1981         spin_unlock(&nilfs->ns_inode_lock);
1982
1983         return 0;
1984 }
1985
1986 static void nilfs_segctor_drop_written_files(struct nilfs_sc_info *sci,
1987                                              struct the_nilfs *nilfs)
1988 {
1989         struct nilfs_inode_info *ii, *n;
1990         int during_mount = !(sci->sc_super->s_flags & MS_ACTIVE);
1991         int defer_iput = false;
1992
1993         spin_lock(&nilfs->ns_inode_lock);
1994         list_for_each_entry_safe(ii, n, &sci->sc_dirty_files, i_dirty) {
1995                 if (!test_and_clear_bit(NILFS_I_UPDATED, &ii->i_state) ||
1996                     test_bit(NILFS_I_DIRTY, &ii->i_state))
1997                         continue;
1998
1999                 clear_bit(NILFS_I_BUSY, &ii->i_state);
2000                 brelse(ii->i_bh);
2001                 ii->i_bh = NULL;
2002                 list_del_init(&ii->i_dirty);
2003                 if (!ii->vfs_inode.i_nlink || during_mount) {
2004                         /*
2005                          * Defer calling iput() to avoid deadlocks if
2006                          * i_nlink == 0 or mount is not yet finished.
2007                          */
2008                         list_add_tail(&ii->i_dirty, &sci->sc_iput_queue);
2009                         defer_iput = true;
2010                 } else {
2011                         spin_unlock(&nilfs->ns_inode_lock);
2012                         iput(&ii->vfs_inode);
2013                         spin_lock(&nilfs->ns_inode_lock);
2014                 }
2015         }
2016         spin_unlock(&nilfs->ns_inode_lock);
2017
2018         if (defer_iput)
2019                 schedule_work(&sci->sc_iput_work);
2020 }
2021
2022 /*
2023  * Main procedure of segment constructor
2024  */
2025 static int nilfs_segctor_do_construct(struct nilfs_sc_info *sci, int mode)
2026 {
2027         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2028         int err;
2029
2030         nilfs_sc_cstage_set(sci, NILFS_ST_INIT);
2031         sci->sc_cno = nilfs->ns_cno;
2032
2033         err = nilfs_segctor_collect_dirty_files(sci, nilfs);
2034         if (unlikely(err))
2035                 goto out;
2036
2037         if (nilfs_test_metadata_dirty(nilfs, sci->sc_root))
2038                 set_bit(NILFS_SC_DIRTY, &sci->sc_flags);
2039
2040         if (nilfs_segctor_clean(sci))
2041                 goto out;
2042
2043         do {
2044                 sci->sc_stage.flags &= ~NILFS_CF_HISTORY_MASK;
2045
2046                 err = nilfs_segctor_begin_construction(sci, nilfs);
2047                 if (unlikely(err))
2048                         goto out;
2049
2050                 /* Update time stamp */
2051                 sci->sc_seg_ctime = get_seconds();
2052
2053                 err = nilfs_segctor_collect(sci, nilfs, mode);
2054                 if (unlikely(err))
2055                         goto failed;
2056
2057                 /* Avoid empty segment */
2058                 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE &&
2059                     nilfs_segbuf_empty(sci->sc_curseg)) {
2060                         nilfs_segctor_abort_construction(sci, nilfs, 1);
2061                         goto out;
2062                 }
2063
2064                 err = nilfs_segctor_assign(sci, mode);
2065                 if (unlikely(err))
2066                         goto failed;
2067
2068                 if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2069                         nilfs_segctor_fill_in_file_bmap(sci);
2070
2071                 if (mode == SC_LSEG_SR &&
2072                     nilfs_sc_cstage_get(sci) >= NILFS_ST_CPFILE) {
2073                         err = nilfs_segctor_fill_in_checkpoint(sci);
2074                         if (unlikely(err))
2075                                 goto failed_to_write;
2076
2077                         nilfs_segctor_fill_in_super_root(sci, nilfs);
2078                 }
2079                 nilfs_segctor_update_segusage(sci, nilfs->ns_sufile);
2080
2081                 /* Write partial segments */
2082                 nilfs_segctor_prepare_write(sci);
2083
2084                 nilfs_add_checksums_on_logs(&sci->sc_segbufs,
2085                                             nilfs->ns_crc_seed);
2086
2087                 err = nilfs_segctor_write(sci, nilfs);
2088                 if (unlikely(err))
2089                         goto failed_to_write;
2090
2091                 if (nilfs_sc_cstage_get(sci) == NILFS_ST_DONE ||
2092                     nilfs->ns_blocksize_bits != PAGE_SHIFT) {
2093                         /*
2094                          * At this point, we avoid double buffering
2095                          * for blocksize < pagesize because page dirty
2096                          * flag is turned off during write and dirty
2097                          * buffers are not properly collected for
2098                          * pages crossing over segments.
2099                          */
2100                         err = nilfs_segctor_wait(sci);
2101                         if (err)
2102                                 goto failed_to_write;
2103                 }
2104         } while (nilfs_sc_cstage_get(sci) != NILFS_ST_DONE);
2105
2106  out:
2107         nilfs_segctor_drop_written_files(sci, nilfs);
2108         return err;
2109
2110  failed_to_write:
2111         if (sci->sc_stage.flags & NILFS_CF_IFILE_STARTED)
2112                 nilfs_redirty_inodes(&sci->sc_dirty_files);
2113
2114  failed:
2115         if (nilfs_doing_gc())
2116                 nilfs_redirty_inodes(&sci->sc_gc_inodes);
2117         nilfs_segctor_abort_construction(sci, nilfs, err);
2118         goto out;
2119 }
2120
2121 /**
2122  * nilfs_segctor_start_timer - set timer of background write
2123  * @sci: nilfs_sc_info
2124  *
2125  * If the timer has already been set, it ignores the new request.
2126  * This function MUST be called within a section locking the segment
2127  * semaphore.
2128  */
2129 static void nilfs_segctor_start_timer(struct nilfs_sc_info *sci)
2130 {
2131         spin_lock(&sci->sc_state_lock);
2132         if (!(sci->sc_state & NILFS_SEGCTOR_COMMIT)) {
2133                 sci->sc_timer.expires = jiffies + sci->sc_interval;
2134                 add_timer(&sci->sc_timer);
2135                 sci->sc_state |= NILFS_SEGCTOR_COMMIT;
2136         }
2137         spin_unlock(&sci->sc_state_lock);
2138 }
2139
2140 static void nilfs_segctor_do_flush(struct nilfs_sc_info *sci, int bn)
2141 {
2142         spin_lock(&sci->sc_state_lock);
2143         if (!(sci->sc_flush_request & BIT(bn))) {
2144                 unsigned long prev_req = sci->sc_flush_request;
2145
2146                 sci->sc_flush_request |= BIT(bn);
2147                 if (!prev_req)
2148                         wake_up(&sci->sc_wait_daemon);
2149         }
2150         spin_unlock(&sci->sc_state_lock);
2151 }
2152
2153 /**
2154  * nilfs_flush_segment - trigger a segment construction for resource control
2155  * @sb: super block
2156  * @ino: inode number of the file to be flushed out.
2157  */
2158 void nilfs_flush_segment(struct super_block *sb, ino_t ino)
2159 {
2160         struct the_nilfs *nilfs = sb->s_fs_info;
2161         struct nilfs_sc_info *sci = nilfs->ns_writer;
2162
2163         if (!sci || nilfs_doing_construction())
2164                 return;
2165         nilfs_segctor_do_flush(sci, NILFS_MDT_INODE(sb, ino) ? ino : 0);
2166                                         /* assign bit 0 to data files */
2167 }
2168
2169 struct nilfs_segctor_wait_request {
2170         wait_queue_t    wq;
2171         __u32           seq;
2172         int             err;
2173         atomic_t        done;
2174 };
2175
2176 static int nilfs_segctor_sync(struct nilfs_sc_info *sci)
2177 {
2178         struct nilfs_segctor_wait_request wait_req;
2179         int err = 0;
2180
2181         spin_lock(&sci->sc_state_lock);
2182         init_wait(&wait_req.wq);
2183         wait_req.err = 0;
2184         atomic_set(&wait_req.done, 0);
2185         wait_req.seq = ++sci->sc_seq_request;
2186         spin_unlock(&sci->sc_state_lock);
2187
2188         init_waitqueue_entry(&wait_req.wq, current);
2189         add_wait_queue(&sci->sc_wait_request, &wait_req.wq);
2190         set_current_state(TASK_INTERRUPTIBLE);
2191         wake_up(&sci->sc_wait_daemon);
2192
2193         for (;;) {
2194                 if (atomic_read(&wait_req.done)) {
2195                         err = wait_req.err;
2196                         break;
2197                 }
2198                 if (!signal_pending(current)) {
2199                         schedule();
2200                         continue;
2201                 }
2202                 err = -ERESTARTSYS;
2203                 break;
2204         }
2205         finish_wait(&sci->sc_wait_request, &wait_req.wq);
2206         return err;
2207 }
2208
2209 static void nilfs_segctor_wakeup(struct nilfs_sc_info *sci, int err)
2210 {
2211         struct nilfs_segctor_wait_request *wrq, *n;
2212         unsigned long flags;
2213
2214         spin_lock_irqsave(&sci->sc_wait_request.lock, flags);
2215         list_for_each_entry_safe(wrq, n, &sci->sc_wait_request.task_list,
2216                                  wq.task_list) {
2217                 if (!atomic_read(&wrq->done) &&
2218                     nilfs_cnt32_ge(sci->sc_seq_done, wrq->seq)) {
2219                         wrq->err = err;
2220                         atomic_set(&wrq->done, 1);
2221                 }
2222                 if (atomic_read(&wrq->done)) {
2223                         wrq->wq.func(&wrq->wq,
2224                                      TASK_UNINTERRUPTIBLE | TASK_INTERRUPTIBLE,
2225                                      0, NULL);
2226                 }
2227         }
2228         spin_unlock_irqrestore(&sci->sc_wait_request.lock, flags);
2229 }
2230
2231 /**
2232  * nilfs_construct_segment - construct a logical segment
2233  * @sb: super block
2234  *
2235  * Return Value: On success, 0 is retured. On errors, one of the following
2236  * negative error code is returned.
2237  *
2238  * %-EROFS - Read only filesystem.
2239  *
2240  * %-EIO - I/O error
2241  *
2242  * %-ENOSPC - No space left on device (only in a panic state).
2243  *
2244  * %-ERESTARTSYS - Interrupted.
2245  *
2246  * %-ENOMEM - Insufficient memory available.
2247  */
2248 int nilfs_construct_segment(struct super_block *sb)
2249 {
2250         struct the_nilfs *nilfs = sb->s_fs_info;
2251         struct nilfs_sc_info *sci = nilfs->ns_writer;
2252         struct nilfs_transaction_info *ti;
2253         int err;
2254
2255         if (!sci)
2256                 return -EROFS;
2257
2258         /* A call inside transactions causes a deadlock. */
2259         BUG_ON((ti = current->journal_info) && ti->ti_magic == NILFS_TI_MAGIC);
2260
2261         err = nilfs_segctor_sync(sci);
2262         return err;
2263 }
2264
2265 /**
2266  * nilfs_construct_dsync_segment - construct a data-only logical segment
2267  * @sb: super block
2268  * @inode: inode whose data blocks should be written out
2269  * @start: start byte offset
2270  * @end: end byte offset (inclusive)
2271  *
2272  * Return Value: On success, 0 is retured. On errors, one of the following
2273  * negative error code is returned.
2274  *
2275  * %-EROFS - Read only filesystem.
2276  *
2277  * %-EIO - I/O error
2278  *
2279  * %-ENOSPC - No space left on device (only in a panic state).
2280  *
2281  * %-ERESTARTSYS - Interrupted.
2282  *
2283  * %-ENOMEM - Insufficient memory available.
2284  */
2285 int nilfs_construct_dsync_segment(struct super_block *sb, struct inode *inode,
2286                                   loff_t start, loff_t end)
2287 {
2288         struct the_nilfs *nilfs = sb->s_fs_info;
2289         struct nilfs_sc_info *sci = nilfs->ns_writer;
2290         struct nilfs_inode_info *ii;
2291         struct nilfs_transaction_info ti;
2292         int err = 0;
2293
2294         if (!sci)
2295                 return -EROFS;
2296
2297         nilfs_transaction_lock(sb, &ti, 0);
2298
2299         ii = NILFS_I(inode);
2300         if (test_bit(NILFS_I_INODE_SYNC, &ii->i_state) ||
2301             nilfs_test_opt(nilfs, STRICT_ORDER) ||
2302             test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2303             nilfs_discontinued(nilfs)) {
2304                 nilfs_transaction_unlock(sb);
2305                 err = nilfs_segctor_sync(sci);
2306                 return err;
2307         }
2308
2309         spin_lock(&nilfs->ns_inode_lock);
2310         if (!test_bit(NILFS_I_QUEUED, &ii->i_state) &&
2311             !test_bit(NILFS_I_BUSY, &ii->i_state)) {
2312                 spin_unlock(&nilfs->ns_inode_lock);
2313                 nilfs_transaction_unlock(sb);
2314                 return 0;
2315         }
2316         spin_unlock(&nilfs->ns_inode_lock);
2317         sci->sc_dsync_inode = ii;
2318         sci->sc_dsync_start = start;
2319         sci->sc_dsync_end = end;
2320
2321         err = nilfs_segctor_do_construct(sci, SC_LSEG_DSYNC);
2322         if (!err)
2323                 nilfs->ns_flushed_device = 0;
2324
2325         nilfs_transaction_unlock(sb);
2326         return err;
2327 }
2328
2329 #define FLUSH_FILE_BIT  (0x1) /* data file only */
2330 #define FLUSH_DAT_BIT   BIT(NILFS_DAT_INO) /* DAT only */
2331
2332 /**
2333  * nilfs_segctor_accept - record accepted sequence count of log-write requests
2334  * @sci: segment constructor object
2335  */
2336 static void nilfs_segctor_accept(struct nilfs_sc_info *sci)
2337 {
2338         spin_lock(&sci->sc_state_lock);
2339         sci->sc_seq_accepted = sci->sc_seq_request;
2340         spin_unlock(&sci->sc_state_lock);
2341         del_timer_sync(&sci->sc_timer);
2342 }
2343
2344 /**
2345  * nilfs_segctor_notify - notify the result of request to caller threads
2346  * @sci: segment constructor object
2347  * @mode: mode of log forming
2348  * @err: error code to be notified
2349  */
2350 static void nilfs_segctor_notify(struct nilfs_sc_info *sci, int mode, int err)
2351 {
2352         /* Clear requests (even when the construction failed) */
2353         spin_lock(&sci->sc_state_lock);
2354
2355         if (mode == SC_LSEG_SR) {
2356                 sci->sc_state &= ~NILFS_SEGCTOR_COMMIT;
2357                 sci->sc_seq_done = sci->sc_seq_accepted;
2358                 nilfs_segctor_wakeup(sci, err);
2359                 sci->sc_flush_request = 0;
2360         } else {
2361                 if (mode == SC_FLUSH_FILE)
2362                         sci->sc_flush_request &= ~FLUSH_FILE_BIT;
2363                 else if (mode == SC_FLUSH_DAT)
2364                         sci->sc_flush_request &= ~FLUSH_DAT_BIT;
2365
2366                 /* re-enable timer if checkpoint creation was not done */
2367                 if ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2368                     time_before(jiffies, sci->sc_timer.expires))
2369                         add_timer(&sci->sc_timer);
2370         }
2371         spin_unlock(&sci->sc_state_lock);
2372 }
2373
2374 /**
2375  * nilfs_segctor_construct - form logs and write them to disk
2376  * @sci: segment constructor object
2377  * @mode: mode of log forming
2378  */
2379 static int nilfs_segctor_construct(struct nilfs_sc_info *sci, int mode)
2380 {
2381         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2382         struct nilfs_super_block **sbp;
2383         int err = 0;
2384
2385         nilfs_segctor_accept(sci);
2386
2387         if (nilfs_discontinued(nilfs))
2388                 mode = SC_LSEG_SR;
2389         if (!nilfs_segctor_confirm(sci))
2390                 err = nilfs_segctor_do_construct(sci, mode);
2391
2392         if (likely(!err)) {
2393                 if (mode != SC_FLUSH_DAT)
2394                         atomic_set(&nilfs->ns_ndirtyblks, 0);
2395                 if (test_bit(NILFS_SC_SUPER_ROOT, &sci->sc_flags) &&
2396                     nilfs_discontinued(nilfs)) {
2397                         down_write(&nilfs->ns_sem);
2398                         err = -EIO;
2399                         sbp = nilfs_prepare_super(sci->sc_super,
2400                                                   nilfs_sb_will_flip(nilfs));
2401                         if (likely(sbp)) {
2402                                 nilfs_set_log_cursor(sbp[0], nilfs);
2403                                 err = nilfs_commit_super(sci->sc_super,
2404                                                          NILFS_SB_COMMIT);
2405                         }
2406                         up_write(&nilfs->ns_sem);
2407                 }
2408         }
2409
2410         nilfs_segctor_notify(sci, mode, err);
2411         return err;
2412 }
2413
2414 static void nilfs_construction_timeout(unsigned long data)
2415 {
2416         struct task_struct *p = (struct task_struct *)data;
2417
2418         wake_up_process(p);
2419 }
2420
2421 static void
2422 nilfs_remove_written_gcinodes(struct the_nilfs *nilfs, struct list_head *head)
2423 {
2424         struct nilfs_inode_info *ii, *n;
2425
2426         list_for_each_entry_safe(ii, n, head, i_dirty) {
2427                 if (!test_bit(NILFS_I_UPDATED, &ii->i_state))
2428                         continue;
2429                 list_del_init(&ii->i_dirty);
2430                 truncate_inode_pages(&ii->vfs_inode.i_data, 0);
2431                 nilfs_btnode_cache_clear(&ii->i_btnode_cache);
2432                 iput(&ii->vfs_inode);
2433         }
2434 }
2435
2436 int nilfs_clean_segments(struct super_block *sb, struct nilfs_argv *argv,
2437                          void **kbufs)
2438 {
2439         struct the_nilfs *nilfs = sb->s_fs_info;
2440         struct nilfs_sc_info *sci = nilfs->ns_writer;
2441         struct nilfs_transaction_info ti;
2442         int err;
2443
2444         if (unlikely(!sci))
2445                 return -EROFS;
2446
2447         nilfs_transaction_lock(sb, &ti, 1);
2448
2449         err = nilfs_mdt_save_to_shadow_map(nilfs->ns_dat);
2450         if (unlikely(err))
2451                 goto out_unlock;
2452
2453         err = nilfs_ioctl_prepare_clean_segments(nilfs, argv, kbufs);
2454         if (unlikely(err)) {
2455                 nilfs_mdt_restore_from_shadow_map(nilfs->ns_dat);
2456                 goto out_unlock;
2457         }
2458
2459         sci->sc_freesegs = kbufs[4];
2460         sci->sc_nfreesegs = argv[4].v_nmembs;
2461         list_splice_tail_init(&nilfs->ns_gc_inodes, &sci->sc_gc_inodes);
2462
2463         for (;;) {
2464                 err = nilfs_segctor_construct(sci, SC_LSEG_SR);
2465                 nilfs_remove_written_gcinodes(nilfs, &sci->sc_gc_inodes);
2466
2467                 if (likely(!err))
2468                         break;
2469
2470                 nilfs_msg(sb, KERN_WARNING, "error %d cleaning segments", err);
2471                 set_current_state(TASK_INTERRUPTIBLE);
2472                 schedule_timeout(sci->sc_interval);
2473         }
2474         if (nilfs_test_opt(nilfs, DISCARD)) {
2475                 int ret = nilfs_discard_segments(nilfs, sci->sc_freesegs,
2476                                                  sci->sc_nfreesegs);
2477                 if (ret) {
2478                         nilfs_msg(sb, KERN_WARNING,
2479                                   "error %d on discard request, turning discards off for the device",
2480                                   ret);
2481                         nilfs_clear_opt(nilfs, DISCARD);
2482                 }
2483         }
2484
2485  out_unlock:
2486         sci->sc_freesegs = NULL;
2487         sci->sc_nfreesegs = 0;
2488         nilfs_mdt_clear_shadow_map(nilfs->ns_dat);
2489         nilfs_transaction_unlock(sb);
2490         return err;
2491 }
2492
2493 static void nilfs_segctor_thread_construct(struct nilfs_sc_info *sci, int mode)
2494 {
2495         struct nilfs_transaction_info ti;
2496
2497         nilfs_transaction_lock(sci->sc_super, &ti, 0);
2498         nilfs_segctor_construct(sci, mode);
2499
2500         /*
2501          * Unclosed segment should be retried.  We do this using sc_timer.
2502          * Timeout of sc_timer will invoke complete construction which leads
2503          * to close the current logical segment.
2504          */
2505         if (test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags))
2506                 nilfs_segctor_start_timer(sci);
2507
2508         nilfs_transaction_unlock(sci->sc_super);
2509 }
2510
2511 static void nilfs_segctor_do_immediate_flush(struct nilfs_sc_info *sci)
2512 {
2513         int mode = 0;
2514
2515         spin_lock(&sci->sc_state_lock);
2516         mode = (sci->sc_flush_request & FLUSH_DAT_BIT) ?
2517                 SC_FLUSH_DAT : SC_FLUSH_FILE;
2518         spin_unlock(&sci->sc_state_lock);
2519
2520         if (mode) {
2521                 nilfs_segctor_do_construct(sci, mode);
2522
2523                 spin_lock(&sci->sc_state_lock);
2524                 sci->sc_flush_request &= (mode == SC_FLUSH_FILE) ?
2525                         ~FLUSH_FILE_BIT : ~FLUSH_DAT_BIT;
2526                 spin_unlock(&sci->sc_state_lock);
2527         }
2528         clear_bit(NILFS_SC_PRIOR_FLUSH, &sci->sc_flags);
2529 }
2530
2531 static int nilfs_segctor_flush_mode(struct nilfs_sc_info *sci)
2532 {
2533         if (!test_bit(NILFS_SC_UNCLOSED, &sci->sc_flags) ||
2534             time_before(jiffies, sci->sc_lseg_stime + sci->sc_mjcp_freq)) {
2535                 if (!(sci->sc_flush_request & ~FLUSH_FILE_BIT))
2536                         return SC_FLUSH_FILE;
2537                 else if (!(sci->sc_flush_request & ~FLUSH_DAT_BIT))
2538                         return SC_FLUSH_DAT;
2539         }
2540         return SC_LSEG_SR;
2541 }
2542
2543 /**
2544  * nilfs_segctor_thread - main loop of the segment constructor thread.
2545  * @arg: pointer to a struct nilfs_sc_info.
2546  *
2547  * nilfs_segctor_thread() initializes a timer and serves as a daemon
2548  * to execute segment constructions.
2549  */
2550 static int nilfs_segctor_thread(void *arg)
2551 {
2552         struct nilfs_sc_info *sci = (struct nilfs_sc_info *)arg;
2553         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2554         int timeout = 0;
2555
2556         sci->sc_timer.data = (unsigned long)current;
2557         sci->sc_timer.function = nilfs_construction_timeout;
2558
2559         /* start sync. */
2560         sci->sc_task = current;
2561         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_start_thread() */
2562         nilfs_msg(sci->sc_super, KERN_INFO,
2563                   "segctord starting. Construction interval = %lu seconds, CP frequency < %lu seconds",
2564                   sci->sc_interval / HZ, sci->sc_mjcp_freq / HZ);
2565
2566         spin_lock(&sci->sc_state_lock);
2567  loop:
2568         for (;;) {
2569                 int mode;
2570
2571                 if (sci->sc_state & NILFS_SEGCTOR_QUIT)
2572                         goto end_thread;
2573
2574                 if (timeout || sci->sc_seq_request != sci->sc_seq_done)
2575                         mode = SC_LSEG_SR;
2576                 else if (sci->sc_flush_request)
2577                         mode = nilfs_segctor_flush_mode(sci);
2578                 else
2579                         break;
2580
2581                 spin_unlock(&sci->sc_state_lock);
2582                 nilfs_segctor_thread_construct(sci, mode);
2583                 spin_lock(&sci->sc_state_lock);
2584                 timeout = 0;
2585         }
2586
2587
2588         if (freezing(current)) {
2589                 spin_unlock(&sci->sc_state_lock);
2590                 try_to_freeze();
2591                 spin_lock(&sci->sc_state_lock);
2592         } else {
2593                 DEFINE_WAIT(wait);
2594                 int should_sleep = 1;
2595
2596                 prepare_to_wait(&sci->sc_wait_daemon, &wait,
2597                                 TASK_INTERRUPTIBLE);
2598
2599                 if (sci->sc_seq_request != sci->sc_seq_done)
2600                         should_sleep = 0;
2601                 else if (sci->sc_flush_request)
2602                         should_sleep = 0;
2603                 else if (sci->sc_state & NILFS_SEGCTOR_COMMIT)
2604                         should_sleep = time_before(jiffies,
2605                                         sci->sc_timer.expires);
2606
2607                 if (should_sleep) {
2608                         spin_unlock(&sci->sc_state_lock);
2609                         schedule();
2610                         spin_lock(&sci->sc_state_lock);
2611                 }
2612                 finish_wait(&sci->sc_wait_daemon, &wait);
2613                 timeout = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) &&
2614                            time_after_eq(jiffies, sci->sc_timer.expires));
2615
2616                 if (nilfs_sb_dirty(nilfs) && nilfs_sb_need_update(nilfs))
2617                         set_nilfs_discontinued(nilfs);
2618         }
2619         goto loop;
2620
2621  end_thread:
2622         spin_unlock(&sci->sc_state_lock);
2623
2624         /* end sync. */
2625         sci->sc_task = NULL;
2626         wake_up(&sci->sc_wait_task); /* for nilfs_segctor_kill_thread() */
2627         return 0;
2628 }
2629
2630 static int nilfs_segctor_start_thread(struct nilfs_sc_info *sci)
2631 {
2632         struct task_struct *t;
2633
2634         t = kthread_run(nilfs_segctor_thread, sci, "segctord");
2635         if (IS_ERR(t)) {
2636                 int err = PTR_ERR(t);
2637
2638                 nilfs_msg(sci->sc_super, KERN_ERR,
2639                           "error %d creating segctord thread", err);
2640                 return err;
2641         }
2642         wait_event(sci->sc_wait_task, sci->sc_task != NULL);
2643         return 0;
2644 }
2645
2646 static void nilfs_segctor_kill_thread(struct nilfs_sc_info *sci)
2647         __acquires(&sci->sc_state_lock)
2648         __releases(&sci->sc_state_lock)
2649 {
2650         sci->sc_state |= NILFS_SEGCTOR_QUIT;
2651
2652         while (sci->sc_task) {
2653                 wake_up(&sci->sc_wait_daemon);
2654                 spin_unlock(&sci->sc_state_lock);
2655                 wait_event(sci->sc_wait_task, sci->sc_task == NULL);
2656                 spin_lock(&sci->sc_state_lock);
2657         }
2658 }
2659
2660 /*
2661  * Setup & clean-up functions
2662  */
2663 static struct nilfs_sc_info *nilfs_segctor_new(struct super_block *sb,
2664                                                struct nilfs_root *root)
2665 {
2666         struct the_nilfs *nilfs = sb->s_fs_info;
2667         struct nilfs_sc_info *sci;
2668
2669         sci = kzalloc(sizeof(*sci), GFP_KERNEL);
2670         if (!sci)
2671                 return NULL;
2672
2673         sci->sc_super = sb;
2674
2675         nilfs_get_root(root);
2676         sci->sc_root = root;
2677
2678         init_waitqueue_head(&sci->sc_wait_request);
2679         init_waitqueue_head(&sci->sc_wait_daemon);
2680         init_waitqueue_head(&sci->sc_wait_task);
2681         spin_lock_init(&sci->sc_state_lock);
2682         INIT_LIST_HEAD(&sci->sc_dirty_files);
2683         INIT_LIST_HEAD(&sci->sc_segbufs);
2684         INIT_LIST_HEAD(&sci->sc_write_logs);
2685         INIT_LIST_HEAD(&sci->sc_gc_inodes);
2686         INIT_LIST_HEAD(&sci->sc_iput_queue);
2687         INIT_WORK(&sci->sc_iput_work, nilfs_iput_work_func);
2688         init_timer(&sci->sc_timer);
2689
2690         sci->sc_interval = HZ * NILFS_SC_DEFAULT_TIMEOUT;
2691         sci->sc_mjcp_freq = HZ * NILFS_SC_DEFAULT_SR_FREQ;
2692         sci->sc_watermark = NILFS_SC_DEFAULT_WATERMARK;
2693
2694         if (nilfs->ns_interval)
2695                 sci->sc_interval = HZ * nilfs->ns_interval;
2696         if (nilfs->ns_watermark)
2697                 sci->sc_watermark = nilfs->ns_watermark;
2698         return sci;
2699 }
2700
2701 static void nilfs_segctor_write_out(struct nilfs_sc_info *sci)
2702 {
2703         int ret, retrycount = NILFS_SC_CLEANUP_RETRY;
2704
2705         /*
2706          * The segctord thread was stopped and its timer was removed.
2707          * But some tasks remain.
2708          */
2709         do {
2710                 struct nilfs_transaction_info ti;
2711
2712                 nilfs_transaction_lock(sci->sc_super, &ti, 0);
2713                 ret = nilfs_segctor_construct(sci, SC_LSEG_SR);
2714                 nilfs_transaction_unlock(sci->sc_super);
2715
2716                 flush_work(&sci->sc_iput_work);
2717
2718         } while (ret && retrycount-- > 0);
2719 }
2720
2721 /**
2722  * nilfs_segctor_destroy - destroy the segment constructor.
2723  * @sci: nilfs_sc_info
2724  *
2725  * nilfs_segctor_destroy() kills the segctord thread and frees
2726  * the nilfs_sc_info struct.
2727  * Caller must hold the segment semaphore.
2728  */
2729 static void nilfs_segctor_destroy(struct nilfs_sc_info *sci)
2730 {
2731         struct the_nilfs *nilfs = sci->sc_super->s_fs_info;
2732         int flag;
2733
2734         up_write(&nilfs->ns_segctor_sem);
2735
2736         spin_lock(&sci->sc_state_lock);
2737         nilfs_segctor_kill_thread(sci);
2738         flag = ((sci->sc_state & NILFS_SEGCTOR_COMMIT) || sci->sc_flush_request
2739                 || sci->sc_seq_request != sci->sc_seq_done);
2740         spin_unlock(&sci->sc_state_lock);
2741
2742         if (flush_work(&sci->sc_iput_work))
2743                 flag = true;
2744
2745         if (flag || !nilfs_segctor_confirm(sci))
2746                 nilfs_segctor_write_out(sci);
2747
2748         if (!list_empty(&sci->sc_dirty_files)) {
2749                 nilfs_msg(sci->sc_super, KERN_WARNING,
2750                           "disposed unprocessed dirty file(s) when stopping log writer");
2751                 nilfs_dispose_list(nilfs, &sci->sc_dirty_files, 1);
2752         }
2753
2754         if (!list_empty(&sci->sc_iput_queue)) {
2755                 nilfs_msg(sci->sc_super, KERN_WARNING,
2756                           "disposed unprocessed inode(s) in iput queue when stopping log writer");
2757                 nilfs_dispose_list(nilfs, &sci->sc_iput_queue, 1);
2758         }
2759
2760         WARN_ON(!list_empty(&sci->sc_segbufs));
2761         WARN_ON(!list_empty(&sci->sc_write_logs));
2762
2763         nilfs_put_root(sci->sc_root);
2764
2765         down_write(&nilfs->ns_segctor_sem);
2766
2767         del_timer_sync(&sci->sc_timer);
2768         kfree(sci);
2769 }
2770
2771 /**
2772  * nilfs_attach_log_writer - attach log writer
2773  * @sb: super block instance
2774  * @root: root object of the current filesystem tree
2775  *
2776  * This allocates a log writer object, initializes it, and starts the
2777  * log writer.
2778  *
2779  * Return Value: On success, 0 is returned. On error, one of the following
2780  * negative error code is returned.
2781  *
2782  * %-ENOMEM - Insufficient memory available.
2783  */
2784 int nilfs_attach_log_writer(struct super_block *sb, struct nilfs_root *root)
2785 {
2786         struct the_nilfs *nilfs = sb->s_fs_info;
2787         int err;
2788
2789         if (nilfs->ns_writer) {
2790                 /*
2791                  * This happens if the filesystem was remounted
2792                  * read/write after nilfs_error degenerated it into a
2793                  * read-only mount.
2794                  */
2795                 nilfs_detach_log_writer(sb);
2796         }
2797
2798         nilfs->ns_writer = nilfs_segctor_new(sb, root);
2799         if (!nilfs->ns_writer)
2800                 return -ENOMEM;
2801
2802         inode_attach_wb(nilfs->ns_bdev->bd_inode, NULL);
2803
2804         err = nilfs_segctor_start_thread(nilfs->ns_writer);
2805         if (unlikely(err))
2806                 nilfs_detach_log_writer(sb);
2807
2808         return err;
2809 }
2810
2811 /**
2812  * nilfs_detach_log_writer - destroy log writer
2813  * @sb: super block instance
2814  *
2815  * This kills log writer daemon, frees the log writer object, and
2816  * destroys list of dirty files.
2817  */
2818 void nilfs_detach_log_writer(struct super_block *sb)
2819 {
2820         struct the_nilfs *nilfs = sb->s_fs_info;
2821         LIST_HEAD(garbage_list);
2822
2823         down_write(&nilfs->ns_segctor_sem);
2824         if (nilfs->ns_writer) {
2825                 nilfs_segctor_destroy(nilfs->ns_writer);
2826                 nilfs->ns_writer = NULL;
2827         }
2828
2829         /* Force to free the list of dirty files */
2830         spin_lock(&nilfs->ns_inode_lock);
2831         if (!list_empty(&nilfs->ns_dirty_files)) {
2832                 list_splice_init(&nilfs->ns_dirty_files, &garbage_list);
2833                 nilfs_msg(sb, KERN_WARNING,
2834                           "disposed unprocessed dirty file(s) when detaching log writer");
2835         }
2836         spin_unlock(&nilfs->ns_inode_lock);
2837         up_write(&nilfs->ns_segctor_sem);
2838
2839         nilfs_dispose_list(nilfs, &garbage_list, 1);
2840 }